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/mutex.h>
  16 #include <linux/workqueue.h>
  17 #include <linux/rcupdate.h>
  18 #include <linux/file.h>
  19 #include <linux/slab.h>
  20
  21 #include <linux/net.h>
  22 #include <linux/if_packet.h>
  23 #include <linux/if_arp.h>
  24 #include <linux/if_tun.h>
  25 #include <linux/if_macvlan.h>
  26
  27 #include <net/sock.h>
  28
  29 #include "vhost.h"
  30
  31 /* Max number of bytes transferred before requeueing the job.
  32  * Using this limit prevents one virtqueue from starving others. */
  33 #define VHOST_NET_WEIGHT 0x80000
  34
  35 enum {
  36         VHOST_NET_VQ_RX = 0,
  37         VHOST_NET_VQ_TX = 1,
  38         VHOST_NET_VQ_MAX = 2,
  39 };
  40
  41 enum vhost_net_poll_state {
  42         VHOST_NET_POLL_DISABLED = 0,
  43         VHOST_NET_POLL_STARTED = 1,
  44         VHOST_NET_POLL_STOPPED = 2,
  45 };
  46
  47 struct vhost_net {
  48         struct vhost_dev dev;
  49         struct vhost_virtqueue vqs[VHOST_NET_VQ_MAX];
  50         struct vhost_poll poll[VHOST_NET_VQ_MAX];
  51         /* Tells us whether we are polling a socket for TX.
  52          * We only do this when socket buffer fills up.
  53          * Protected by tx vq lock. */
  54         enum vhost_net_poll_state tx_poll_state;
  55 };
  56
  57 /* Pop first len bytes from iovec. Return number of segments used. */
  58 static int move_iovec_hdr(struct iovec *from, struct iovec *to,
  59                           size_t len, int iov_count)
  60 {
  61         int seg = 0;
  62         size_t size;
  63
  64         while (len && seg < iov_count) {
  65                 size = min(from->iov_len, len);
  66                 to->iov_base = from->iov_base;
  67                 to->iov_len = size;
  68                 from->iov_len -= size;
  69                 from->iov_base += size;
  70                 len -= size;
  71                 ++from;
  72                 ++to;
  73                 ++seg;
  74         }
  75         return seg;
  76 }
  77 /* Copy iovec entries for len bytes from iovec. */
  78 static void copy_iovec_hdr(const struct iovec *from, struct iovec *to,
  79                            size_t len, int iovcount)
  80 {
  81         int seg = 0;
  82         size_t size;
  83
  84         while (len && seg < iovcount) {
  85                 size = min(from->iov_len, len);
  86                 to->iov_base = from->iov_base;
  87                 to->iov_len = size;
  88                 len -= size;
  89                 ++from;
  90                 ++to;
  91                 ++seg;
  92         }
  93 }
  94
  95 /* Caller must have TX VQ lock */
  96 static void tx_poll_stop(struct vhost_net *net)
  97 {
  98         if (likely(net->tx_poll_state != VHOST_NET_POLL_STARTED))
  99                 return;
 100         vhost_poll_stop(net->poll + VHOST_NET_VQ_TX);
 101         net->tx_poll_state = VHOST_NET_POLL_STOPPED;
 102 }
 103
 104 /* Caller must have TX VQ lock */
 105 static void tx_poll_start(struct vhost_net *net, struct socket *sock)
 106 {
 107         if (unlikely(net->tx_poll_state != VHOST_NET_POLL_STOPPED))
 108                 return;
 109         vhost_poll_start(net->poll + VHOST_NET_VQ_TX, sock->file);
 110         net->tx_poll_state = VHOST_NET_POLL_STARTED;
 111 }
 112
 113 /* Expects to be always run from workqueue - which acts as
 114  * read-size critical section for our kind of RCU. */
 115 static void handle_tx(struct vhost_net *net)
 116 {
 117         struct vhost_virtqueue *vq = &net->dev.vqs[VHOST_NET_VQ_TX];
 118         unsigned out, in, s;
 119         int head;
 120         struct msghdr msg = {
 121                 .msg_name = NULL,
 122                 .msg_namelen = 0,
 123                 .msg_control = NULL,
 124                 .msg_controllen = 0,
 125                 .msg_iov = vq->iov,
 126                 .msg_flags = MSG_DONTWAIT,
 127         };
 128         size_t len, total_len = 0;
 129         int err, wmem;
 130         size_t hdr_size;
 131         struct socket *sock;
 132
 133         /* TODO: check that we are running from vhost_worker? */
 134         sock = rcu_dereference_check(vq->private_data, 1);
 135         if (!sock)
 136                 return;
 137
 138         wmem = atomic_read(&sock->sk->sk_wmem_alloc);
 139         if (wmem >= sock->sk->sk_sndbuf) {
 140                 mutex_lock(&vq->mutex);
 141                 tx_poll_start(net, sock);
 142                 mutex_unlock(&vq->mutex);
 143                 return;
 144         }
 145
 146         mutex_lock(&vq->mutex);
 147         vhost_disable_notify(&net->dev, vq);
 148
 149         if (wmem < sock->sk->sk_sndbuf / 2)
 150                 tx_poll_stop(net);
 151         hdr_size = vq->vhost_hlen;
 152
 153         for (;;) {
 154                 head = vhost_get_vq_desc(&net->dev, vq, vq->iov,
 155                                          ARRAY_SIZE(vq->iov),
 156                                          &out, &in,
 157                                          NULL, NULL);
 158                 /* On error, stop handling until the next kick. */
 159                 if (unlikely(head < 0))
 160                         break;
 161                 /* Nothing new?  Wait for eventfd to tell us they refilled. */
 162                 if (head == vq->num) {
 163                         wmem = atomic_read(&sock->sk->sk_wmem_alloc);
 164                         if (wmem >= sock->sk->sk_sndbuf * 3 / 4) {
 165                                 tx_poll_start(net, sock);
 166                                 set_bit(SOCK_ASYNC_NOSPACE, &sock->flags);
 167                                 break;
 168                         }
 169                         if (unlikely(vhost_enable_notify(&net->dev, vq))) {
 170                                 vhost_disable_notify(&net->dev, vq);
 171                                 continue;
 172                         }
 173                         break;
 174                 }
 175                 if (in) {
 176                         vq_err(vq, "Unexpected descriptor format for TX: "
 177                                "out %d, int %d\n", out, in);
 178                         break;
 179                 }
 180                 /* Skip header. TODO: support TSO. */
 181                 s = move_iovec_hdr(vq->iov, vq->hdr, hdr_size, out);
 182                 msg.msg_iovlen = out;
 183                 len = iov_length(vq->iov, out);
 184                 /* Sanity check */
 185                 if (!len) {
 186                         vq_err(vq, "Unexpected header len for TX: "
 187                                "%zd expected %zd\n",
 188                                iov_length(vq->hdr, s), hdr_size);
 189                         break;
 190                 }
 191                 /* TODO: Check specific error and bomb out unless ENOBUFS? */
 192                 err = sock->ops->sendmsg(NULL, sock, &msg, len);
 193                 if (unlikely(err < 0)) {
 194                         vhost_discard_vq_desc(vq, 1);
 195                         tx_poll_start(net, sock);
 196                         break;
 197                 }
 198                 if (err != len)
 199                         pr_debug("Truncated TX packet: "
 200                                  " len %d != %zd\n", err, len);
 201                 vhost_add_used_and_signal(&net->dev, vq, head, 0);
 202                 total_len += len;
 203                 if (unlikely(total_len >= VHOST_NET_WEIGHT)) {
 204                         vhost_poll_queue(&vq->poll);
 205                         break;
 206                 }
 207         }
 208
 209         mutex_unlock(&vq->mutex);
 210 }
 211
 212 static int peek_head_len(struct sock *sk)
 213 {
 214         struct sk_buff *head;
 215         int len = 0;
 216         unsigned long flags;
 217
 218         spin_lock_irqsave(&sk->sk_receive_queue.lock, flags);
 219         head = skb_peek(&sk->sk_receive_queue);
 220         if (likely(head))
 221                 len = head->len;
 222         spin_unlock_irqrestore(&sk->sk_receive_queue.lock, flags);
 223         return len;
 224 }
 225
 226 /* This is a multi-buffer version of vhost_get_desc, that works if
 227  *      vq has read descriptors only.
 228  * @vq          - the relevant virtqueue
 229  * @datalen     - data length we'll be reading
 230  * @iovcount    - returned count of io vectors we fill
 231  * @log         - vhost log
 232  * @log_num     - log offset
 233  * @quota       - headcount quota, 1 for big buffer
 234  *      returns number of buffer heads allocated, negative on error
 235  */
 236 static int get_rx_bufs(struct vhost_virtqueue *vq,
 237                        struct vring_used_elem *heads,
 238                        int datalen,
 239                        unsigned *iovcount,
 240                        struct vhost_log *log,
 241                        unsigned *log_num,
 242                        unsigned int quota)
 243 {
 244         unsigned int out, in;
 245         int seg = 0;
 246         int headcount = 0;
 247         unsigned d;
 248         int r, nlogs = 0;
 249
 250         while (datalen > 0 && headcount < quota) {
 251                 if (unlikely(seg >= UIO_MAXIOV)) {
 252                         r = -ENOBUFS;
 253                         goto err;
 254                 }
 255                 d = vhost_get_vq_desc(vq->dev, vq, vq->iov + seg,
 256                                       ARRAY_SIZE(vq->iov) - seg, &out,
 257                                       &in, log, log_num);
 258                 if (d == vq->num) {
 259                         r = 0;
 260                         goto err;
 261                 }
 262                 if (unlikely(out || in <= 0)) {
 263                         vq_err(vq, "unexpected descriptor format for RX: "
 264                                 "out %d, in %d\n", out, in);
 265                         r = -EINVAL;
 266                         goto err;
 267                 }
 268                 if (unlikely(log)) {
 269                         nlogs += *log_num;
 270                         log += *log_num;
 271                 }
 272                 heads[headcount].id = d;
 273                 heads[headcount].len = iov_length(vq->iov + seg, in);
 274                 datalen -= heads[headcount].len;
 275                 ++headcount;
 276                 seg += in;
 277         }
 278         heads[headcount - 1].len += datalen;
 279         *iovcount = seg;
 280         if (unlikely(log))
 281                 *log_num = nlogs;
 282         return headcount;
 283 err:
 284         vhost_discard_vq_desc(vq, headcount);
 285         return r;
 286 }
 287
 288 /* Expects to be always run from workqueue - which acts as
 289  * read-size critical section for our kind of RCU. */
 290 static void handle_rx(struct vhost_net *net)
 291 {
 292         struct vhost_virtqueue *vq = &net->dev.vqs[VHOST_NET_VQ_RX];
 293         unsigned uninitialized_var(in), log;
 294         struct vhost_log *vq_log;
 295         struct msghdr msg = {
 296                 .msg_name = NULL,
 297                 .msg_namelen = 0,
 298                 .msg_control = NULL, /* FIXME: get and handle RX aux data. */
 299                 .msg_controllen = 0,
 300                 .msg_iov = vq->iov,
 301                 .msg_flags = MSG_DONTWAIT,
 302         };
 303         struct virtio_net_hdr_mrg_rxbuf hdr = {
 304                 .hdr.flags = 0,
 305                 .hdr.gso_type = VIRTIO_NET_HDR_GSO_NONE
 306         };
 307         size_t total_len = 0;
 308         int err, headcount, mergeable;
 309         size_t vhost_hlen, sock_hlen;
 310         size_t vhost_len, sock_len;
 311         /* TODO: check that we are running from vhost_worker? */
 312         struct socket *sock = rcu_dereference_check(vq->private_data, 1);
 313
 314         if (!sock)
 315                 return;
 316
 317         mutex_lock(&vq->mutex);
 318         vhost_disable_notify(&net->dev, vq);
 319         vhost_hlen = vq->vhost_hlen;
 320         sock_hlen = vq->sock_hlen;
 321
 322         vq_log = unlikely(vhost_has_feature(&net->dev, VHOST_F_LOG_ALL)) ?
 323                 vq->log : NULL;
 324         mergeable = vhost_has_feature(&net->dev, VIRTIO_NET_F_MRG_RXBUF);
 325
 326         while ((sock_len = peek_head_len(sock->sk))) {
 327                 sock_len += sock_hlen;
 328                 vhost_len = sock_len + vhost_hlen;
 329                 headcount = get_rx_bufs(vq, vq->heads, vhost_len,
 330                                         &in, vq_log, &log,
 331                                         likely(mergeable) ? UIO_MAXIOV : 1);
 332                 /* On error, stop handling until the next kick. */
 333                 if (unlikely(headcount < 0))
 334                         break;
 335                 /* OK, now we need to know about added descriptors. */
 336                 if (!headcount) {
 337                         if (unlikely(vhost_enable_notify(&net->dev, vq))) {
 338                                 /* They have slipped one in as we were
 339                                  * doing that: check again. */
 340                                 vhost_disable_notify(&net->dev, vq);
 341                                 continue;
 342                         }
 343                         /* Nothing new?  Wait for eventfd to tell us
 344                          * they refilled. */
 345                         break;
 346                 }
 347                 /* We don't need to be notified again. */
 348                 if (unlikely((vhost_hlen)))
 349                         /* Skip header. TODO: support TSO. */
 350                         move_iovec_hdr(vq->iov, vq->hdr, vhost_hlen, in);
 351                 else
 352                         /* Copy the header for use in VIRTIO_NET_F_MRG_RXBUF:
 353                          * needed because recvmsg can modify msg_iov. */
 354                         copy_iovec_hdr(vq->iov, vq->hdr, sock_hlen, in);
 355                 msg.msg_iovlen = in;
 356                 err = sock->ops->recvmsg(NULL, sock, &msg,
 357                                          sock_len, MSG_DONTWAIT | MSG_TRUNC);
 358                 /* Userspace might have consumed the packet meanwhile:
 359                  * it's not supposed to do this usually, but might be hard
 360                  * to prevent. Discard data we got (if any) and keep going. */
 361                 if (unlikely(err != sock_len)) {
 362                         pr_debug("Discarded rx packet: "
 363                                  " len %d, expected %zd\n", err, sock_len);
 364                         vhost_discard_vq_desc(vq, headcount);
 365                         continue;
 366                 }
 367                 if (unlikely(vhost_hlen) &&
 368                     memcpy_toiovecend(vq->hdr, (unsigned char *)&hdr, 0,
 369                                       vhost_hlen)) {
 370                         vq_err(vq, "Unable to write vnet_hdr at addr %p\n",
 371                                vq->iov->iov_base);
 372                         break;
 373                 }
 374                 /* TODO: Should check and handle checksum. */
 375                 if (likely(mergeable) &&
 376                     memcpy_toiovecend(vq->hdr, (unsigned char *)&headcount,
 377                                       offsetof(typeof(hdr), num_buffers),
 378                                       sizeof hdr.num_buffers)) {
 379                         vq_err(vq, "Failed num_buffers write");
 380                         vhost_discard_vq_desc(vq, headcount);
 381                         break;
 382                 }
 383                 vhost_add_used_and_signal_n(&net->dev, vq, vq->heads,
 384                                             headcount);
 385                 if (unlikely(vq_log))
 386                         vhost_log_write(vq, vq_log, log, vhost_len);
 387                 total_len += vhost_len;
 388                 if (unlikely(total_len >= VHOST_NET_WEIGHT)) {
 389                         vhost_poll_queue(&vq->poll);
 390                         break;
 391                 }
 392         }
 393
 394         mutex_unlock(&vq->mutex);
 395 }
 396
 397 static void handle_tx_kick(struct vhost_work *work)
 398 {
 399         struct vhost_virtqueue *vq = container_of(work, struct vhost_virtqueue,
 400                                                   poll.work);
 401         struct vhost_net *net = container_of(vq->dev, struct vhost_net, dev);
 402
 403         handle_tx(net);
 404 }
 405
 406 static void handle_rx_kick(struct vhost_work *work)
 407 {
 408         struct vhost_virtqueue *vq = container_of(work, struct vhost_virtqueue,
 409                                                   poll.work);
 410         struct vhost_net *net = container_of(vq->dev, struct vhost_net, dev);
 411
 412         handle_rx(net);
 413 }
 414
 415 static void handle_tx_net(struct vhost_work *work)
 416 {
 417         struct vhost_net *net = container_of(work, struct vhost_net,
 418                                              poll[VHOST_NET_VQ_TX].work);
 419         handle_tx(net);
 420 }
 421
 422 static void handle_rx_net(struct vhost_work *work)
 423 {
 424         struct vhost_net *net = container_of(work, struct vhost_net,
 425                                              poll[VHOST_NET_VQ_RX].work);
 426         handle_rx(net);
 427 }
 428
 429 static int vhost_net_open(struct inode *inode, struct file *f)
 430 {
 431         struct vhost_net *n = kmalloc(sizeof *n, GFP_KERNEL);
 432         struct vhost_dev *dev;
 433         int r;
 434
 435         if (!n)
 436                 return -ENOMEM;
 437
 438         dev = &n->dev;
 439         n->vqs[VHOST_NET_VQ_TX].handle_kick = handle_tx_kick;
 440         n->vqs[VHOST_NET_VQ_RX].handle_kick = handle_rx_kick;
 441         r = vhost_dev_init(dev, n->vqs, VHOST_NET_VQ_MAX);
 442         if (r < 0) {
 443                 kfree(n);
 444                 return r;
 445         }
 446
 447         vhost_poll_init(n->poll + VHOST_NET_VQ_TX, handle_tx_net, POLLOUT, dev);
 448         vhost_poll_init(n->poll + VHOST_NET_VQ_RX, handle_rx_net, POLLIN, dev);
 449         n->tx_poll_state = VHOST_NET_POLL_DISABLED;
 450
 451         f->private_data = n;
 452
 453         return 0;
 454 }
 455
 456 static void vhost_net_disable_vq(struct vhost_net *n,
 457                                  struct vhost_virtqueue *vq)
 458 {
 459         if (!vq->private_data)
 460                 return;
 461         if (vq == n->vqs + VHOST_NET_VQ_TX) {
 462                 tx_poll_stop(n);
 463                 n->tx_poll_state = VHOST_NET_POLL_DISABLED;
 464         } else
 465                 vhost_poll_stop(n->poll + VHOST_NET_VQ_RX);
 466 }
 467
 468 static void vhost_net_enable_vq(struct vhost_net *n,
 469                                 struct vhost_virtqueue *vq)
 470 {
 471         struct socket *sock;
 472
 473         sock = rcu_dereference_protected(vq->private_data,
 474                                          lockdep_is_held(&vq->mutex));
 475         if (!sock)
 476                 return;
 477         if (vq == n->vqs + VHOST_NET_VQ_TX) {
 478                 n->tx_poll_state = VHOST_NET_POLL_STOPPED;
 479                 tx_poll_start(n, sock);
 480         } else
 481                 vhost_poll_start(n->poll + VHOST_NET_VQ_RX, sock->file);
 482 }
 483
 484 static struct socket *vhost_net_stop_vq(struct vhost_net *n,
 485                                         struct vhost_virtqueue *vq)
 486 {
 487         struct socket *sock;
 488
 489         mutex_lock(&vq->mutex);
 490         sock = rcu_dereference_protected(vq->private_data,
 491                                          lockdep_is_held(&vq->mutex));
 492         vhost_net_disable_vq(n, vq);
 493         rcu_assign_pointer(vq->private_data, NULL);
 494         mutex_unlock(&vq->mutex);
 495         return sock;
 496 }
 497
 498 static void vhost_net_stop(struct vhost_net *n, struct socket **tx_sock,
 499                            struct socket **rx_sock)
 500 {
 501         *tx_sock = vhost_net_stop_vq(n, n->vqs + VHOST_NET_VQ_TX);
 502         *rx_sock = vhost_net_stop_vq(n, n->vqs + VHOST_NET_VQ_RX);
 503 }
 504
 505 static void vhost_net_flush_vq(struct vhost_net *n, int index)
 506 {
 507         vhost_poll_flush(n->poll + index);
 508         vhost_poll_flush(&n->dev.vqs[index].poll);
 509 }
 510
 511 static void vhost_net_flush(struct vhost_net *n)
 512 {
 513         vhost_net_flush_vq(n, VHOST_NET_VQ_TX);
 514         vhost_net_flush_vq(n, VHOST_NET_VQ_RX);
 515 }
 516
 517 static int vhost_net_release(struct inode *inode, struct file *f)
 518 {
 519         struct vhost_net *n = f->private_data;
 520         struct socket *tx_sock;
 521         struct socket *rx_sock;
 522
 523         vhost_net_stop(n, &tx_sock, &rx_sock);
 524         vhost_net_flush(n);
 525         vhost_dev_cleanup(&n->dev);
 526         if (tx_sock)
 527                 fput(tx_sock->file);
 528         if (rx_sock)
 529                 fput(rx_sock->file);
 530         /* We do an extra flush before freeing memory,
 531          * since jobs can re-queue themselves. */
 532         vhost_net_flush(n);
 533         kfree(n);
 534         return 0;
 535 }
 536
 537 static struct socket *get_raw_socket(int fd)
 538 {
 539         struct {
 540                 struct sockaddr_ll sa;
 541                 char  buf[MAX_ADDR_LEN];
 542         } uaddr;
 543         int uaddr_len = sizeof uaddr, r;
 544         struct socket *sock = sockfd_lookup(fd, &r);
 545
 546         if (!sock)
 547                 return ERR_PTR(-ENOTSOCK);
 548
 549         /* Parameter checking */
 550         if (sock->sk->sk_type != SOCK_RAW) {
 551                 r = -ESOCKTNOSUPPORT;
 552                 goto err;
 553         }
 554
 555         r = sock->ops->getname(sock, (struct sockaddr *)&uaddr.sa,
 556                                &uaddr_len, 0);
 557         if (r)
 558                 goto err;
 559
 560         if (uaddr.sa.sll_family != AF_PACKET) {
 561                 r = -EPFNOSUPPORT;
 562                 goto err;
 563         }
 564         return sock;
 565 err:
 566         fput(sock->file);
 567         return ERR_PTR(r);
 568 }
 569
 570 static struct socket *get_tap_socket(int fd)
 571 {
 572         struct file *file = fget(fd);
 573         struct socket *sock;
 574
 575         if (!file)
 576                 return ERR_PTR(-EBADF);
 577         sock = tun_get_socket(file);
 578         if (!IS_ERR(sock))
 579                 return sock;
 580         sock = macvtap_get_socket(file);
 581         if (IS_ERR(sock))
 582                 fput(file);
 583         return sock;
 584 }
 585
 586 static struct socket *get_socket(int fd)
 587 {
 588         struct socket *sock;
 589
 590         /* special case to disable backend */
 591         if (fd == -1)
 592                 return NULL;
 593         sock = get_raw_socket(fd);
 594         if (!IS_ERR(sock))
 595                 return sock;
 596         sock = get_tap_socket(fd);
 597         if (!IS_ERR(sock))
 598                 return sock;
 599         return ERR_PTR(-ENOTSOCK);
 600 }
 601
 602 static long vhost_net_set_backend(struct vhost_net *n, unsigned index, int fd)
 603 {
 604         struct socket *sock, *oldsock;
 605         struct vhost_virtqueue *vq;
 606         int r;
 607
 608         mutex_lock(&n->dev.mutex);
 609         r = vhost_dev_check_owner(&n->dev);
 610         if (r)
 611                 goto err;
 612
 613         if (index >= VHOST_NET_VQ_MAX) {
 614                 r = -ENOBUFS;
 615                 goto err;
 616         }
 617         vq = n->vqs + index;
 618         mutex_lock(&vq->mutex);
 619
 620         /* Verify that ring has been setup correctly. */
 621         if (!vhost_vq_access_ok(vq)) {
 622                 r = -EFAULT;
 623                 goto err_vq;
 624         }
 625         sock = get_socket(fd);
 626         if (IS_ERR(sock)) {
 627                 r = PTR_ERR(sock);
 628                 goto err_vq;
 629         }
 630
 631         /* start polling new socket */
 632         oldsock = rcu_dereference_protected(vq->private_data,
 633                                             lockdep_is_held(&vq->mutex));
 634         if (sock != oldsock) {
 635                 vhost_net_disable_vq(n, vq);
 636                 rcu_assign_pointer(vq->private_data, sock);
 637                 vhost_net_enable_vq(n, vq);
 638         }
 639
 640         mutex_unlock(&vq->mutex);
 641
 642         if (oldsock) {
 643                 vhost_net_flush_vq(n, index);
 644                 fput(oldsock->file);
 645         }
 646
 647         mutex_unlock(&n->dev.mutex);
 648         return 0;
 649
 650 err_vq:
 651         mutex_unlock(&vq->mutex);
 652 err:
 653         mutex_unlock(&n->dev.mutex);
 654         return r;
 655 }
 656
 657 static long vhost_net_reset_owner(struct vhost_net *n)
 658 {
 659         struct socket *tx_sock = NULL;
 660         struct socket *rx_sock = NULL;
 661         long err;
 662
 663         mutex_lock(&n->dev.mutex);
 664         err = vhost_dev_check_owner(&n->dev);
 665         if (err)
 666                 goto done;
 667         vhost_net_stop(n, &tx_sock, &rx_sock);
 668         vhost_net_flush(n);
 669         err = vhost_dev_reset_owner(&n->dev);
 670 done:
 671         mutex_unlock(&n->dev.mutex);
 672         if (tx_sock)
 673                 fput(tx_sock->file);
 674         if (rx_sock)
 675                 fput(rx_sock->file);
 676         return err;
 677 }
 678
 679 static int vhost_net_set_features(struct vhost_net *n, u64 features)
 680 {
 681         size_t vhost_hlen, sock_hlen, hdr_len;
 682         int i;
 683
 684         hdr_len = (features & (1 << VIRTIO_NET_F_MRG_RXBUF)) ?
 685                         sizeof(struct virtio_net_hdr_mrg_rxbuf) :
 686                         sizeof(struct virtio_net_hdr);
 687         if (features & (1 << VHOST_NET_F_VIRTIO_NET_HDR)) {
 688                 /* vhost provides vnet_hdr */
 689                 vhost_hlen = hdr_len;
 690                 sock_hlen = 0;
 691         } else {
 692                 /* socket provides vnet_hdr */
 693                 vhost_hlen = 0;
 694                 sock_hlen = hdr_len;
 695         }
 696         mutex_lock(&n->dev.mutex);
 697         if ((features & (1 << VHOST_F_LOG_ALL)) &&
 698             !vhost_log_access_ok(&n->dev)) {
 699                 mutex_unlock(&n->dev.mutex);
 700                 return -EFAULT;
 701         }
 702         n->dev.acked_features = features;
 703         smp_wmb();
 704         for (i = 0; i < VHOST_NET_VQ_MAX; ++i) {
 705                 mutex_lock(&n->vqs[i].mutex);
 706                 n->vqs[i].vhost_hlen = vhost_hlen;
 707                 n->vqs[i].sock_hlen = sock_hlen;
 708                 mutex_unlock(&n->vqs[i].mutex);
 709         }
 710         vhost_net_flush(n);
 711         mutex_unlock(&n->dev.mutex);
 712         return 0;
 713 }
 714
 715 static long vhost_net_ioctl(struct file *f, unsigned int ioctl,
 716                             unsigned long arg)
 717 {
 718         struct vhost_net *n = f->private_data;
 719         void __user *argp = (void __user *)arg;
 720         u64 __user *featurep = argp;
 721         struct vhost_vring_file backend;
 722         u64 features;
 723         int r;
 724
 725         switch (ioctl) {
 726         case VHOST_NET_SET_BACKEND:
 727                 if (copy_from_user(&backend, argp, sizeof backend))
 728                         return -EFAULT;
 729                 return vhost_net_set_backend(n, backend.index, backend.fd);
 730         case VHOST_GET_FEATURES:
 731                 features = VHOST_FEATURES;
 732                 if (copy_to_user(featurep, &features, sizeof features))
 733                         return -EFAULT;
 734                 return 0;
 735         case VHOST_SET_FEATURES:
 736                 if (copy_from_user(&features, featurep, sizeof features))
 737                         return -EFAULT;
 738                 if (features & ~VHOST_FEATURES)
 739                         return -EOPNOTSUPP;
 740                 return vhost_net_set_features(n, features);
 741         case VHOST_RESET_OWNER:
 742                 return vhost_net_reset_owner(n);
 743         default:
 744                 mutex_lock(&n->dev.mutex);
 745                 r = vhost_dev_ioctl(&n->dev, ioctl, arg);
 746                 vhost_net_flush(n);
 747                 mutex_unlock(&n->dev.mutex);
 748                 return r;
 749         }
 750 }
 751
 752 #ifdef CONFIG_COMPAT
 753 static long vhost_net_compat_ioctl(struct file *f, unsigned int ioctl,
 754                                    unsigned long arg)
 755 {
 756         return vhost_net_ioctl(f, ioctl, (unsigned long)compat_ptr(arg));
 757 }
 758 #endif
 759
 760 static const struct file_operations vhost_net_fops = {
 761         .owner          = THIS_MODULE,
 762         .release        = vhost_net_release,
 763         .unlocked_ioctl = vhost_net_ioctl,
 764 #ifdef CONFIG_COMPAT
 765         .compat_ioctl   = vhost_net_compat_ioctl,
 766 #endif
 767         .open           = vhost_net_open,
 768         .llseek         = noop_llseek,
 769 };
 770
 771 static struct miscdevice vhost_net_misc = {
 772         MISC_DYNAMIC_MINOR,
 773         "vhost-net",
 774         &vhost_net_fops,
 775 };
 776
 777 static int vhost_net_init(void)
 778 {
 779         return misc_register(&vhost_net_misc);
 780 }
 781 module_init(vhost_net_init);
 782
 783 static void vhost_net_exit(void)
 784 {
 785         misc_deregister(&vhost_net_misc);
 786 }
 787 module_exit(vhost_net_exit);
 788
 789 MODULE_VERSION("0.0.1");
 790 MODULE_LICENSE("GPL v2");
 791 MODULE_AUTHOR("Michael S. Tsirkin");
 792 MODULE_DESCRIPTION("Host kernel accelerator for virtio net");