net/vmw_vsock/virtio_transport.c

   1 /*
   2  * virtio transport for vsock
   3  *
   4  * Copyright (C) 2013-2015 Red Hat, Inc.
   5  * Author: Asias He <asias@redhat.com>
   6  *         Stefan Hajnoczi <stefanha@redhat.com>
   7  *
   8  * Some of the code is take from Gerd Hoffmann <kraxel@redhat.com>'s
   9  * early virtio-vsock proof-of-concept bits.
  10  *
  11  * This work is licensed under the terms of the GNU GPL, version 2.
  12  */
  13 #include <linux/spinlock.h>
  14 #include <linux/module.h>
  15 #include <linux/list.h>
  16 #include <linux/atomic.h>
  17 #include <linux/virtio.h>
  18 #include <linux/virtio_ids.h>
  19 #include <linux/virtio_config.h>
  20 #include <linux/virtio_vsock.h>
  21 #include <net/sock.h>
  22 #include <linux/mutex.h>
  23 #include <net/af_vsock.h>
  24
  25 static struct workqueue_struct *virtio_vsock_workqueue;
  26 static struct virtio_vsock *the_virtio_vsock;
  27 static DEFINE_MUTEX(the_virtio_vsock_mutex); /* protects the_virtio_vsock */
  28
  29 struct virtio_vsock {
  30         struct virtio_device *vdev;
  31         struct virtqueue *vqs[VSOCK_VQ_MAX];
  32
  33         /* Virtqueue processing is deferred to a workqueue */
  34         struct work_struct tx_work;
  35         struct work_struct rx_work;
  36         struct work_struct event_work;
  37
  38         /* The following fields are protected by tx_lock.  vqs[VSOCK_VQ_TX]
  39          * must be accessed with tx_lock held.
  40          */
  41         struct mutex tx_lock;
  42
  43         struct work_struct send_pkt_work;
  44         spinlock_t send_pkt_list_lock;
  45         struct list_head send_pkt_list;
  46
  47         atomic_t queued_replies;
  48
  49         /* The following fields are protected by rx_lock.  vqs[VSOCK_VQ_RX]
  50          * must be accessed with rx_lock held.
  51          */
  52         struct mutex rx_lock;
  53         int rx_buf_nr;
  54         int rx_buf_max_nr;
  55
  56         /* The following fields are protected by event_lock.
  57          * vqs[VSOCK_VQ_EVENT] must be accessed with event_lock held.
  58          */
  59         struct mutex event_lock;
  60         struct virtio_vsock_event event_list[8];
  61
  62         u32 guest_cid;
  63 };
  64
  65 static struct virtio_vsock *virtio_vsock_get(void)
  66 {
  67         return the_virtio_vsock;
  68 }
  69
  70 static u32 virtio_transport_get_local_cid(void)
  71 {
  72         struct virtio_vsock *vsock = virtio_vsock_get();
  73
  74         if (!vsock)
  75                 return VMADDR_CID_ANY;
  76
  77         return vsock->guest_cid;
  78 }
  79
  80 static void
  81 virtio_transport_send_pkt_work(struct work_struct *work)
  82 {
  83         struct virtio_vsock *vsock =
  84                 container_of(work, struct virtio_vsock, send_pkt_work);
  85         struct virtqueue *vq;
  86         bool added = false;
  87         bool restart_rx = false;
  88
  89         mutex_lock(&vsock->tx_lock);
  90
  91         vq = vsock->vqs[VSOCK_VQ_TX];
  92
  93         for (;;) {
  94                 struct virtio_vsock_pkt *pkt;
  95                 struct scatterlist hdr, buf, *sgs[2];
  96                 int ret, in_sg = 0, out_sg = 0;
  97                 bool reply;
  98
  99                 spin_lock_bh(&vsock->send_pkt_list_lock);
 100                 if (list_empty(&vsock->send_pkt_list)) {
 101                         spin_unlock_bh(&vsock->send_pkt_list_lock);
 102                         break;
 103                 }
 104
 105                 pkt = list_first_entry(&vsock->send_pkt_list,
 106                                        struct virtio_vsock_pkt, list);
 107                 list_del_init(&pkt->list);
 108                 spin_unlock_bh(&vsock->send_pkt_list_lock);
 109
 110                 reply = pkt->reply;
 111
 112                 sg_init_one(&hdr, &pkt->hdr, sizeof(pkt->hdr));
 113                 sgs[out_sg++] = &hdr;
 114                 if (pkt->buf) {
 115                         sg_init_one(&buf, pkt->buf, pkt->len);
 116                         sgs[out_sg++] = &buf;
 117                 }
 118
 119                 ret = virtqueue_add_sgs(vq, sgs, out_sg, in_sg, pkt, GFP_KERNEL);
 120                 /* Usually this means that there is no more space available in
 121                  * the vq
 122                  */
 123                 if (ret < 0) {
 124                         spin_lock_bh(&vsock->send_pkt_list_lock);
 125                         list_add(&pkt->list, &vsock->send_pkt_list);
 126                         spin_unlock_bh(&vsock->send_pkt_list_lock);
 127                         break;
 128                 }
 129
 130                 if (reply) {
 131                         struct virtqueue *rx_vq = vsock->vqs[VSOCK_VQ_RX];
 132                         int val;
 133
 134                         val = atomic_dec_return(&vsock->queued_replies);
 135
 136                         /* Do we now have resources to resume rx processing? */
 137                         if (val + 1 == virtqueue_get_vring_size(rx_vq))
 138                                 restart_rx = true;
 139                 }
 140
 141                 added = true;
 142         }
 143
 144         if (added)
 145                 virtqueue_kick(vq);
 146
 147         mutex_unlock(&vsock->tx_lock);
 148
 149         if (restart_rx)
 150                 queue_work(virtio_vsock_workqueue, &vsock->rx_work);
 151 }
 152
 153 static int
 154 virtio_transport_send_pkt(struct virtio_vsock_pkt *pkt)
 155 {
 156         struct virtio_vsock *vsock;
 157         int len = pkt->len;
 158
 159         vsock = virtio_vsock_get();
 160         if (!vsock) {
 161                 virtio_transport_free_pkt(pkt);
 162                 return -ENODEV;
 163         }
 164
 165         if (pkt->reply)
 166                 atomic_inc(&vsock->queued_replies);
 167
 168         spin_lock_bh(&vsock->send_pkt_list_lock);
 169         list_add_tail(&pkt->list, &vsock->send_pkt_list);
 170         spin_unlock_bh(&vsock->send_pkt_list_lock);
 171
 172         queue_work(virtio_vsock_workqueue, &vsock->send_pkt_work);
 173         return len;
 174 }
 175
 176 static void virtio_vsock_rx_fill(struct virtio_vsock *vsock)
 177 {
 178         int buf_len = VIRTIO_VSOCK_DEFAULT_RX_BUF_SIZE;
 179         struct virtio_vsock_pkt *pkt;
 180         struct scatterlist hdr, buf, *sgs[2];
 181         struct virtqueue *vq;
 182         int ret;
 183
 184         vq = vsock->vqs[VSOCK_VQ_RX];
 185
 186         do {
 187                 pkt = kzalloc(sizeof(*pkt), GFP_KERNEL);
 188                 if (!pkt)
 189                         break;
 190
 191                 pkt->buf = kmalloc(buf_len, GFP_KERNEL);
 192                 if (!pkt->buf) {
 193                         virtio_transport_free_pkt(pkt);
 194                         break;
 195                 }
 196
 197                 pkt->len = buf_len;
 198
 199                 sg_init_one(&hdr, &pkt->hdr, sizeof(pkt->hdr));
 200                 sgs[0] = &hdr;
 201
 202                 sg_init_one(&buf, pkt->buf, buf_len);
 203                 sgs[1] = &buf;
 204                 ret = virtqueue_add_sgs(vq, sgs, 0, 2, pkt, GFP_KERNEL);
 205                 if (ret) {
 206                         virtio_transport_free_pkt(pkt);
 207                         break;
 208                 }
 209                 vsock->rx_buf_nr++;
 210         } while (vq->num_free);
 211         if (vsock->rx_buf_nr > vsock->rx_buf_max_nr)
 212                 vsock->rx_buf_max_nr = vsock->rx_buf_nr;
 213         virtqueue_kick(vq);
 214 }
 215
 216 static void virtio_transport_tx_work(struct work_struct *work)
 217 {
 218         struct virtio_vsock *vsock =
 219                 container_of(work, struct virtio_vsock, tx_work);
 220         struct virtqueue *vq;
 221         bool added = false;
 222
 223         vq = vsock->vqs[VSOCK_VQ_TX];
 224         mutex_lock(&vsock->tx_lock);
 225         do {
 226                 struct virtio_vsock_pkt *pkt;
 227                 unsigned int len;
 228
 229                 virtqueue_disable_cb(vq);
 230                 while ((pkt = virtqueue_get_buf(vq, &len)) != NULL) {
 231                         virtio_transport_free_pkt(pkt);
 232                         added = true;
 233                 }
 234         } while (!virtqueue_enable_cb(vq));
 235         mutex_unlock(&vsock->tx_lock);
 236
 237         if (added)
 238                 queue_work(virtio_vsock_workqueue, &vsock->send_pkt_work);
 239 }
 240
 241 /* Is there space left for replies to rx packets? */
 242 static bool virtio_transport_more_replies(struct virtio_vsock *vsock)
 243 {
 244         struct virtqueue *vq = vsock->vqs[VSOCK_VQ_RX];
 245         int val;
 246
 247         smp_rmb(); /* paired with atomic_inc() and atomic_dec_return() */
 248         val = atomic_read(&vsock->queued_replies);
 249
 250         return val < virtqueue_get_vring_size(vq);
 251 }
 252
 253 static void virtio_transport_rx_work(struct work_struct *work)
 254 {
 255         struct virtio_vsock *vsock =
 256                 container_of(work, struct virtio_vsock, rx_work);
 257         struct virtqueue *vq;
 258
 259         vq = vsock->vqs[VSOCK_VQ_RX];
 260
 261         mutex_lock(&vsock->rx_lock);
 262
 263         do {
 264                 virtqueue_disable_cb(vq);
 265                 for (;;) {
 266                         struct virtio_vsock_pkt *pkt;
 267                         unsigned int len;
 268
 269                         if (!virtio_transport_more_replies(vsock)) {
 270                                 /* Stop rx until the device processes already
 271                                  * pending replies.  Leave rx virtqueue
 272                                  * callbacks disabled.
 273                                  */
 274                                 goto out;
 275                         }
 276
 277                         pkt = virtqueue_get_buf(vq, &len);
 278                         if (!pkt) {
 279                                 break;
 280                         }
 281
 282                         vsock->rx_buf_nr--;
 283
 284                         /* Drop short/long packets */
 285                         if (unlikely(len < sizeof(pkt->hdr) ||
 286                                      len > sizeof(pkt->hdr) + pkt->len)) {
 287                                 virtio_transport_free_pkt(pkt);
 288                                 continue;
 289                         }
 290
 291                         pkt->len = len - sizeof(pkt->hdr);
 292                         virtio_transport_recv_pkt(pkt);
 293                 }
 294         } while (!virtqueue_enable_cb(vq));
 295
 296 out:
 297         if (vsock->rx_buf_nr < vsock->rx_buf_max_nr / 2)
 298                 virtio_vsock_rx_fill(vsock);
 299         mutex_unlock(&vsock->rx_lock);
 300 }
 301
 302 /* event_lock must be held */
 303 static int virtio_vsock_event_fill_one(struct virtio_vsock *vsock,
 304                                        struct virtio_vsock_event *event)
 305 {
 306         struct scatterlist sg;
 307         struct virtqueue *vq;
 308
 309         vq = vsock->vqs[VSOCK_VQ_EVENT];
 310
 311         sg_init_one(&sg, event, sizeof(*event));
 312
 313         return virtqueue_add_inbuf(vq, &sg, 1, event, GFP_KERNEL);
 314 }
 315
 316 /* event_lock must be held */
 317 static void virtio_vsock_event_fill(struct virtio_vsock *vsock)
 318 {
 319         size_t i;
 320
 321         for (i = 0; i < ARRAY_SIZE(vsock->event_list); i++) {
 322                 struct virtio_vsock_event *event = &vsock->event_list[i];
 323
 324                 virtio_vsock_event_fill_one(vsock, event);
 325         }
 326
 327         virtqueue_kick(vsock->vqs[VSOCK_VQ_EVENT]);
 328 }
 329
 330 static void virtio_vsock_reset_sock(struct sock *sk)
 331 {
 332         lock_sock(sk);
 333         sk->sk_state = SS_UNCONNECTED;
 334         sk->sk_err = ECONNRESET;
 335         sk->sk_error_report(sk);
 336         release_sock(sk);
 337 }
 338
 339 static void virtio_vsock_update_guest_cid(struct virtio_vsock *vsock)
 340 {
 341         struct virtio_device *vdev = vsock->vdev;
 342         u64 guest_cid;
 343
 344         vdev->config->get(vdev, offsetof(struct virtio_vsock_config, guest_cid),
 345                           &guest_cid, sizeof(guest_cid));
 346         vsock->guest_cid = le64_to_cpu(guest_cid);
 347 }
 348
 349 /* event_lock must be held */
 350 static void virtio_vsock_event_handle(struct virtio_vsock *vsock,
 351                                       struct virtio_vsock_event *event)
 352 {
 353         switch (le32_to_cpu(event->id)) {
 354         case VIRTIO_VSOCK_EVENT_TRANSPORT_RESET:
 355                 virtio_vsock_update_guest_cid(vsock);
 356                 vsock_for_each_connected_socket(virtio_vsock_reset_sock);
 357                 break;
 358         }
 359 }
 360
 361 static void virtio_transport_event_work(struct work_struct *work)
 362 {
 363         struct virtio_vsock *vsock =
 364                 container_of(work, struct virtio_vsock, event_work);
 365         struct virtqueue *vq;
 366
 367         vq = vsock->vqs[VSOCK_VQ_EVENT];
 368
 369         mutex_lock(&vsock->event_lock);
 370
 371         do {
 372                 struct virtio_vsock_event *event;
 373                 unsigned int len;
 374
 375                 virtqueue_disable_cb(vq);
 376                 while ((event = virtqueue_get_buf(vq, &len)) != NULL) {
 377                         if (len == sizeof(*event))
 378                                 virtio_vsock_event_handle(vsock, event);
 379
 380                         virtio_vsock_event_fill_one(vsock, event);
 381                 }
 382         } while (!virtqueue_enable_cb(vq));
 383
 384         virtqueue_kick(vsock->vqs[VSOCK_VQ_EVENT]);
 385
 386         mutex_unlock(&vsock->event_lock);
 387 }
 388
 389 static void virtio_vsock_event_done(struct virtqueue *vq)
 390 {
 391         struct virtio_vsock *vsock = vq->vdev->priv;
 392
 393         if (!vsock)
 394                 return;
 395         queue_work(virtio_vsock_workqueue, &vsock->event_work);
 396 }
 397
 398 static void virtio_vsock_tx_done(struct virtqueue *vq)
 399 {
 400         struct virtio_vsock *vsock = vq->vdev->priv;
 401
 402         if (!vsock)
 403                 return;
 404         queue_work(virtio_vsock_workqueue, &vsock->tx_work);
 405 }
 406
 407 static void virtio_vsock_rx_done(struct virtqueue *vq)
 408 {
 409         struct virtio_vsock *vsock = vq->vdev->priv;
 410
 411         if (!vsock)
 412                 return;
 413         queue_work(virtio_vsock_workqueue, &vsock->rx_work);
 414 }
 415
 416 static struct virtio_transport virtio_transport = {
 417         .transport = {
 418                 .get_local_cid            = virtio_transport_get_local_cid,
 419
 420                 .init                     = virtio_transport_do_socket_init,
 421                 .destruct                 = virtio_transport_destruct,
 422                 .release                  = virtio_transport_release,
 423                 .connect                  = virtio_transport_connect,
 424                 .shutdown                 = virtio_transport_shutdown,
 425
 426                 .dgram_bind               = virtio_transport_dgram_bind,
 427                 .dgram_dequeue            = virtio_transport_dgram_dequeue,
 428                 .dgram_enqueue            = virtio_transport_dgram_enqueue,
 429                 .dgram_allow              = virtio_transport_dgram_allow,
 430
 431                 .stream_dequeue           = virtio_transport_stream_dequeue,
 432                 .stream_enqueue           = virtio_transport_stream_enqueue,
 433                 .stream_has_data          = virtio_transport_stream_has_data,
 434                 .stream_has_space         = virtio_transport_stream_has_space,
 435                 .stream_rcvhiwat          = virtio_transport_stream_rcvhiwat,
 436                 .stream_is_active         = virtio_transport_stream_is_active,
 437                 .stream_allow             = virtio_transport_stream_allow,
 438
 439                 .notify_poll_in           = virtio_transport_notify_poll_in,
 440                 .notify_poll_out          = virtio_transport_notify_poll_out,
 441                 .notify_recv_init         = virtio_transport_notify_recv_init,
 442                 .notify_recv_pre_block    = virtio_transport_notify_recv_pre_block,
 443                 .notify_recv_pre_dequeue  = virtio_transport_notify_recv_pre_dequeue,
 444                 .notify_recv_post_dequeue = virtio_transport_notify_recv_post_dequeue,
 445                 .notify_send_init         = virtio_transport_notify_send_init,
 446                 .notify_send_pre_block    = virtio_transport_notify_send_pre_block,
 447                 .notify_send_pre_enqueue  = virtio_transport_notify_send_pre_enqueue,
 448                 .notify_send_post_enqueue = virtio_transport_notify_send_post_enqueue,
 449
 450                 .set_buffer_size          = virtio_transport_set_buffer_size,
 451                 .set_min_buffer_size      = virtio_transport_set_min_buffer_size,
 452                 .set_max_buffer_size      = virtio_transport_set_max_buffer_size,
 453                 .get_buffer_size          = virtio_transport_get_buffer_size,
 454                 .get_min_buffer_size      = virtio_transport_get_min_buffer_size,
 455                 .get_max_buffer_size      = virtio_transport_get_max_buffer_size,
 456         },
 457
 458         .send_pkt = virtio_transport_send_pkt,
 459 };
 460
 461 static int virtio_vsock_probe(struct virtio_device *vdev)
 462 {
 463         vq_callback_t *callbacks[] = {
 464                 virtio_vsock_rx_done,
 465                 virtio_vsock_tx_done,
 466                 virtio_vsock_event_done,
 467         };
 468         static const char * const names[] = {
 469                 "rx",
 470                 "tx",
 471                 "event",
 472         };
 473         struct virtio_vsock *vsock = NULL;
 474         int ret;
 475
 476         ret = mutex_lock_interruptible(&the_virtio_vsock_mutex);
 477         if (ret)
 478                 return ret;
 479
 480         /* Only one virtio-vsock device per guest is supported */
 481         if (the_virtio_vsock) {
 482                 ret = -EBUSY;
 483                 goto out;
 484         }
 485
 486         vsock = kzalloc(sizeof(*vsock), GFP_KERNEL);
 487         if (!vsock) {
 488                 ret = -ENOMEM;
 489                 goto out;
 490         }
 491
 492         vsock->vdev = vdev;
 493
 494         ret = vsock->vdev->config->find_vqs(vsock->vdev, VSOCK_VQ_MAX,
 495                                             vsock->vqs, callbacks, names);
 496         if (ret < 0)
 497                 goto out;
 498
 499         virtio_vsock_update_guest_cid(vsock);
 500
 501         vsock->rx_buf_nr = 0;
 502         vsock->rx_buf_max_nr = 0;
 503         atomic_set(&vsock->queued_replies, 0);
 504
 505         vdev->priv = vsock;
 506         the_virtio_vsock = vsock;
 507         mutex_init(&vsock->tx_lock);
 508         mutex_init(&vsock->rx_lock);
 509         mutex_init(&vsock->event_lock);
 510         spin_lock_init(&vsock->send_pkt_list_lock);
 511         INIT_LIST_HEAD(&vsock->send_pkt_list);
 512         INIT_WORK(&vsock->rx_work, virtio_transport_rx_work);
 513         INIT_WORK(&vsock->tx_work, virtio_transport_tx_work);
 514         INIT_WORK(&vsock->event_work, virtio_transport_event_work);
 515         INIT_WORK(&vsock->send_pkt_work, virtio_transport_send_pkt_work);
 516
 517         mutex_lock(&vsock->rx_lock);
 518         virtio_vsock_rx_fill(vsock);
 519         mutex_unlock(&vsock->rx_lock);
 520
 521         mutex_lock(&vsock->event_lock);
 522         virtio_vsock_event_fill(vsock);
 523         mutex_unlock(&vsock->event_lock);
 524
 525         mutex_unlock(&the_virtio_vsock_mutex);
 526         return 0;
 527
 528 out:
 529         kfree(vsock);
 530         mutex_unlock(&the_virtio_vsock_mutex);
 531         return ret;
 532 }
 533
 534 static void virtio_vsock_remove(struct virtio_device *vdev)
 535 {
 536         struct virtio_vsock *vsock = vdev->priv;
 537         struct virtio_vsock_pkt *pkt;
 538
 539         flush_work(&vsock->rx_work);
 540         flush_work(&vsock->tx_work);
 541         flush_work(&vsock->event_work);
 542         flush_work(&vsock->send_pkt_work);
 543
 544         /* Reset all connected sockets when the device disappear */
 545         vsock_for_each_connected_socket(virtio_vsock_reset_sock);
 546
 547         vdev->config->reset(vdev);
 548
 549         mutex_lock(&vsock->rx_lock);
 550         while ((pkt = virtqueue_detach_unused_buf(vsock->vqs[VSOCK_VQ_RX])))
 551                 virtio_transport_free_pkt(pkt);
 552         mutex_unlock(&vsock->rx_lock);
 553
 554         mutex_lock(&vsock->tx_lock);
 555         while ((pkt = virtqueue_detach_unused_buf(vsock->vqs[VSOCK_VQ_TX])))
 556                 virtio_transport_free_pkt(pkt);
 557         mutex_unlock(&vsock->tx_lock);
 558
 559         spin_lock_bh(&vsock->send_pkt_list_lock);
 560         while (!list_empty(&vsock->send_pkt_list)) {
 561                 pkt = list_first_entry(&vsock->send_pkt_list,
 562                                        struct virtio_vsock_pkt, list);
 563                 list_del(&pkt->list);
 564                 virtio_transport_free_pkt(pkt);
 565         }
 566         spin_unlock_bh(&vsock->send_pkt_list_lock);
 567
 568         mutex_lock(&the_virtio_vsock_mutex);
 569         the_virtio_vsock = NULL;
 570         mutex_unlock(&the_virtio_vsock_mutex);
 571
 572         vdev->config->del_vqs(vdev);
 573
 574         kfree(vsock);
 575 }
 576
 577 static struct virtio_device_id id_table[] = {
 578         { VIRTIO_ID_VSOCK, VIRTIO_DEV_ANY_ID },
 579         { 0 },
 580 };
 581
 582 static unsigned int features[] = {
 583 };
 584
 585 static struct virtio_driver virtio_vsock_driver = {
 586         .feature_table = features,
 587         .feature_table_size = ARRAY_SIZE(features),
 588         .driver.name = KBUILD_MODNAME,
 589         .driver.owner = THIS_MODULE,
 590         .id_table = id_table,
 591         .probe = virtio_vsock_probe,
 592         .remove = virtio_vsock_remove,
 593 };
 594
 595 static int __init virtio_vsock_init(void)
 596 {
 597         int ret;
 598
 599         virtio_vsock_workqueue = alloc_workqueue("virtio_vsock", 0, 0);
 600         if (!virtio_vsock_workqueue)
 601                 return -ENOMEM;
 602
 603         ret = register_virtio_driver(&virtio_vsock_driver);
 604         if (ret)
 605                 goto out_wq;
 606
 607         ret = vsock_core_init(&virtio_transport.transport);
 608         if (ret)
 609                 goto out_vdr;
 610
 611         return 0;
 612
 613 out_vdr:
 614         unregister_virtio_driver(&virtio_vsock_driver);
 615 out_wq:
 616         destroy_workqueue(virtio_vsock_workqueue);
 617         return ret;
 618
 619 }
 620
 621 static void __exit virtio_vsock_exit(void)
 622 {
 623         vsock_core_exit();
 624         unregister_virtio_driver(&virtio_vsock_driver);
 625         destroy_workqueue(virtio_vsock_workqueue);
 626 }
 627
 628 module_init(virtio_vsock_init);
 629 module_exit(virtio_vsock_exit);
 630 MODULE_LICENSE("GPL v2");
 631 MODULE_AUTHOR("Asias He");
 632 MODULE_DESCRIPTION("virtio transport for vsock");
 633 MODULE_DEVICE_TABLE(virtio, id_table);