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 struct work_struct loopback_work; 48 spinlock_t loopback_list_lock; /* protects loopback_list */ 49 struct list_head loopback_list; 50 51 atomic_t queued_replies; 52 53 /* The following fields are protected by rx_lock. vqs[VSOCK_VQ_RX] 54 * must be accessed with rx_lock held. 55 */ 56 struct mutex rx_lock; 57 int rx_buf_nr; 58 int rx_buf_max_nr; 59 60 /* The following fields are protected by event_lock. 61 * vqs[VSOCK_VQ_EVENT] must be accessed with event_lock held. 62 */ 63 struct mutex event_lock; 64 struct virtio_vsock_event event_list[8]; 65 66 u32 guest_cid; 67 }; 68 69 static struct virtio_vsock *virtio_vsock_get(void) 70 { 71 return the_virtio_vsock; 72 } 73 74 static u32 virtio_transport_get_local_cid(void) 75 { 76 struct virtio_vsock *vsock = virtio_vsock_get(); 77 78 return vsock->guest_cid; 79 } 80 81 static void virtio_transport_loopback_work(struct work_struct *work) 82 { 83 struct virtio_vsock *vsock = 84 container_of(work, struct virtio_vsock, loopback_work); 85 LIST_HEAD(pkts); 86 87 spin_lock_bh(&vsock->loopback_list_lock); 88 list_splice_init(&vsock->loopback_list, &pkts); 89 spin_unlock_bh(&vsock->loopback_list_lock); 90 91 mutex_lock(&vsock->rx_lock); 92 while (!list_empty(&pkts)) { 93 struct virtio_vsock_pkt *pkt; 94 95 pkt = list_first_entry(&pkts, struct virtio_vsock_pkt, list); 96 list_del_init(&pkt->list); 97 98 virtio_transport_recv_pkt(pkt); 99 } 100 mutex_unlock(&vsock->rx_lock); 101 } 102 103 static int virtio_transport_send_pkt_loopback(struct virtio_vsock *vsock, 104 struct virtio_vsock_pkt *pkt) 105 { 106 int len = pkt->len; 107 108 spin_lock_bh(&vsock->loopback_list_lock); 109 list_add_tail(&pkt->list, &vsock->loopback_list); 110 spin_unlock_bh(&vsock->loopback_list_lock); 111 112 queue_work(virtio_vsock_workqueue, &vsock->loopback_work); 113 114 return len; 115 } 116 117 static void 118 virtio_transport_send_pkt_work(struct work_struct *work) 119 { 120 struct virtio_vsock *vsock = 121 container_of(work, struct virtio_vsock, send_pkt_work); 122 struct virtqueue *vq; 123 bool added = false; 124 bool restart_rx = false; 125 126 mutex_lock(&vsock->tx_lock); 127 128 vq = vsock->vqs[VSOCK_VQ_TX]; 129 130 for (;;) { 131 struct virtio_vsock_pkt *pkt; 132 struct scatterlist hdr, buf, *sgs[2]; 133 int ret, in_sg = 0, out_sg = 0; 134 bool reply; 135 136 spin_lock_bh(&vsock->send_pkt_list_lock); 137 if (list_empty(&vsock->send_pkt_list)) { 138 spin_unlock_bh(&vsock->send_pkt_list_lock); 139 break; 140 } 141 142 pkt = list_first_entry(&vsock->send_pkt_list, 143 struct virtio_vsock_pkt, list); 144 list_del_init(&pkt->list); 145 spin_unlock_bh(&vsock->send_pkt_list_lock); 146 147 reply = pkt->reply; 148 149 sg_init_one(&hdr, &pkt->hdr, sizeof(pkt->hdr)); 150 sgs[out_sg++] = &hdr; 151 if (pkt->buf) { 152 sg_init_one(&buf, pkt->buf, pkt->len); 153 sgs[out_sg++] = &buf; 154 } 155 156 ret = virtqueue_add_sgs(vq, sgs, out_sg, in_sg, pkt, GFP_KERNEL); 157 /* Usually this means that there is no more space available in 158 * the vq 159 */ 160 if (ret < 0) { 161 spin_lock_bh(&vsock->send_pkt_list_lock); 162 list_add(&pkt->list, &vsock->send_pkt_list); 163 spin_unlock_bh(&vsock->send_pkt_list_lock); 164 break; 165 } 166 167 if (reply) { 168 struct virtqueue *rx_vq = vsock->vqs[VSOCK_VQ_RX]; 169 int val; 170 171 val = atomic_dec_return(&vsock->queued_replies); 172 173 /* Do we now have resources to resume rx processing? */ 174 if (val + 1 == virtqueue_get_vring_size(rx_vq)) 175 restart_rx = true; 176 } 177 178 added = true; 179 } 180 181 if (added) 182 virtqueue_kick(vq); 183 184 mutex_unlock(&vsock->tx_lock); 185 186 if (restart_rx) 187 queue_work(virtio_vsock_workqueue, &vsock->rx_work); 188 } 189 190 static int 191 virtio_transport_send_pkt(struct virtio_vsock_pkt *pkt) 192 { 193 struct virtio_vsock *vsock; 194 int len = pkt->len; 195 196 vsock = virtio_vsock_get(); 197 if (!vsock) { 198 virtio_transport_free_pkt(pkt); 199 return -ENODEV; 200 } 201 202 if (le32_to_cpu(pkt->hdr.dst_cid) == vsock->guest_cid) 203 return virtio_transport_send_pkt_loopback(vsock, pkt); 204 205 if (pkt->reply) 206 atomic_inc(&vsock->queued_replies); 207 208 spin_lock_bh(&vsock->send_pkt_list_lock); 209 list_add_tail(&pkt->list, &vsock->send_pkt_list); 210 spin_unlock_bh(&vsock->send_pkt_list_lock); 211 212 queue_work(virtio_vsock_workqueue, &vsock->send_pkt_work); 213 return len; 214 } 215 216 static void virtio_vsock_rx_fill(struct virtio_vsock *vsock) 217 { 218 int buf_len = VIRTIO_VSOCK_DEFAULT_RX_BUF_SIZE; 219 struct virtio_vsock_pkt *pkt; 220 struct scatterlist hdr, buf, *sgs[2]; 221 struct virtqueue *vq; 222 int ret; 223 224 vq = vsock->vqs[VSOCK_VQ_RX]; 225 226 do { 227 pkt = kzalloc(sizeof(*pkt), GFP_KERNEL); 228 if (!pkt) 229 break; 230 231 pkt->buf = kmalloc(buf_len, GFP_KERNEL); 232 if (!pkt->buf) { 233 virtio_transport_free_pkt(pkt); 234 break; 235 } 236 237 pkt->len = buf_len; 238 239 sg_init_one(&hdr, &pkt->hdr, sizeof(pkt->hdr)); 240 sgs[0] = &hdr; 241 242 sg_init_one(&buf, pkt->buf, buf_len); 243 sgs[1] = &buf; 244 ret = virtqueue_add_sgs(vq, sgs, 0, 2, pkt, GFP_KERNEL); 245 if (ret) { 246 virtio_transport_free_pkt(pkt); 247 break; 248 } 249 vsock->rx_buf_nr++; 250 } while (vq->num_free); 251 if (vsock->rx_buf_nr > vsock->rx_buf_max_nr) 252 vsock->rx_buf_max_nr = vsock->rx_buf_nr; 253 virtqueue_kick(vq); 254 } 255 256 static void virtio_transport_tx_work(struct work_struct *work) 257 { 258 struct virtio_vsock *vsock = 259 container_of(work, struct virtio_vsock, tx_work); 260 struct virtqueue *vq; 261 bool added = false; 262 263 vq = vsock->vqs[VSOCK_VQ_TX]; 264 mutex_lock(&vsock->tx_lock); 265 do { 266 struct virtio_vsock_pkt *pkt; 267 unsigned int len; 268 269 virtqueue_disable_cb(vq); 270 while ((pkt = virtqueue_get_buf(vq, &len)) != NULL) { 271 virtio_transport_free_pkt(pkt); 272 added = true; 273 } 274 } while (!virtqueue_enable_cb(vq)); 275 mutex_unlock(&vsock->tx_lock); 276 277 if (added) 278 queue_work(virtio_vsock_workqueue, &vsock->send_pkt_work); 279 } 280 281 /* Is there space left for replies to rx packets? */ 282 static bool virtio_transport_more_replies(struct virtio_vsock *vsock) 283 { 284 struct virtqueue *vq = vsock->vqs[VSOCK_VQ_RX]; 285 int val; 286 287 smp_rmb(); /* paired with atomic_inc() and atomic_dec_return() */ 288 val = atomic_read(&vsock->queued_replies); 289 290 return val < virtqueue_get_vring_size(vq); 291 } 292 293 static void virtio_transport_rx_work(struct work_struct *work) 294 { 295 struct virtio_vsock *vsock = 296 container_of(work, struct virtio_vsock, rx_work); 297 struct virtqueue *vq; 298 299 vq = vsock->vqs[VSOCK_VQ_RX]; 300 301 mutex_lock(&vsock->rx_lock); 302 303 do { 304 virtqueue_disable_cb(vq); 305 for (;;) { 306 struct virtio_vsock_pkt *pkt; 307 unsigned int len; 308 309 if (!virtio_transport_more_replies(vsock)) { 310 /* Stop rx until the device processes already 311 * pending replies. Leave rx virtqueue 312 * callbacks disabled. 313 */ 314 goto out; 315 } 316 317 pkt = virtqueue_get_buf(vq, &len); 318 if (!pkt) { 319 break; 320 } 321 322 vsock->rx_buf_nr--; 323 324 /* Drop short/long packets */ 325 if (unlikely(len < sizeof(pkt->hdr) || 326 len > sizeof(pkt->hdr) + pkt->len)) { 327 virtio_transport_free_pkt(pkt); 328 continue; 329 } 330 331 pkt->len = len - sizeof(pkt->hdr); 332 virtio_transport_recv_pkt(pkt); 333 } 334 } while (!virtqueue_enable_cb(vq)); 335 336 out: 337 if (vsock->rx_buf_nr < vsock->rx_buf_max_nr / 2) 338 virtio_vsock_rx_fill(vsock); 339 mutex_unlock(&vsock->rx_lock); 340 } 341 342 /* event_lock must be held */ 343 static int virtio_vsock_event_fill_one(struct virtio_vsock *vsock, 344 struct virtio_vsock_event *event) 345 { 346 struct scatterlist sg; 347 struct virtqueue *vq; 348 349 vq = vsock->vqs[VSOCK_VQ_EVENT]; 350 351 sg_init_one(&sg, event, sizeof(*event)); 352 353 return virtqueue_add_inbuf(vq, &sg, 1, event, GFP_KERNEL); 354 } 355 356 /* event_lock must be held */ 357 static void virtio_vsock_event_fill(struct virtio_vsock *vsock) 358 { 359 size_t i; 360 361 for (i = 0; i < ARRAY_SIZE(vsock->event_list); i++) { 362 struct virtio_vsock_event *event = &vsock->event_list[i]; 363 364 virtio_vsock_event_fill_one(vsock, event); 365 } 366 367 virtqueue_kick(vsock->vqs[VSOCK_VQ_EVENT]); 368 } 369 370 static void virtio_vsock_reset_sock(struct sock *sk) 371 { 372 lock_sock(sk); 373 sk->sk_state = SS_UNCONNECTED; 374 sk->sk_err = ECONNRESET; 375 sk->sk_error_report(sk); 376 release_sock(sk); 377 } 378 379 static void virtio_vsock_update_guest_cid(struct virtio_vsock *vsock) 380 { 381 struct virtio_device *vdev = vsock->vdev; 382 __le64 guest_cid; 383 384 vdev->config->get(vdev, offsetof(struct virtio_vsock_config, guest_cid), 385 &guest_cid, sizeof(guest_cid)); 386 vsock->guest_cid = le64_to_cpu(guest_cid); 387 } 388 389 /* event_lock must be held */ 390 static void virtio_vsock_event_handle(struct virtio_vsock *vsock, 391 struct virtio_vsock_event *event) 392 { 393 switch (le32_to_cpu(event->id)) { 394 case VIRTIO_VSOCK_EVENT_TRANSPORT_RESET: 395 virtio_vsock_update_guest_cid(vsock); 396 vsock_for_each_connected_socket(virtio_vsock_reset_sock); 397 break; 398 } 399 } 400 401 static void virtio_transport_event_work(struct work_struct *work) 402 { 403 struct virtio_vsock *vsock = 404 container_of(work, struct virtio_vsock, event_work); 405 struct virtqueue *vq; 406 407 vq = vsock->vqs[VSOCK_VQ_EVENT]; 408 409 mutex_lock(&vsock->event_lock); 410 411 do { 412 struct virtio_vsock_event *event; 413 unsigned int len; 414 415 virtqueue_disable_cb(vq); 416 while ((event = virtqueue_get_buf(vq, &len)) != NULL) { 417 if (len == sizeof(*event)) 418 virtio_vsock_event_handle(vsock, event); 419 420 virtio_vsock_event_fill_one(vsock, event); 421 } 422 } while (!virtqueue_enable_cb(vq)); 423 424 virtqueue_kick(vsock->vqs[VSOCK_VQ_EVENT]); 425 426 mutex_unlock(&vsock->event_lock); 427 } 428 429 static void virtio_vsock_event_done(struct virtqueue *vq) 430 { 431 struct virtio_vsock *vsock = vq->vdev->priv; 432 433 if (!vsock) 434 return; 435 queue_work(virtio_vsock_workqueue, &vsock->event_work); 436 } 437 438 static void virtio_vsock_tx_done(struct virtqueue *vq) 439 { 440 struct virtio_vsock *vsock = vq->vdev->priv; 441 442 if (!vsock) 443 return; 444 queue_work(virtio_vsock_workqueue, &vsock->tx_work); 445 } 446 447 static void virtio_vsock_rx_done(struct virtqueue *vq) 448 { 449 struct virtio_vsock *vsock = vq->vdev->priv; 450 451 if (!vsock) 452 return; 453 queue_work(virtio_vsock_workqueue, &vsock->rx_work); 454 } 455 456 static struct virtio_transport virtio_transport = { 457 .transport = { 458 .get_local_cid = virtio_transport_get_local_cid, 459 460 .init = virtio_transport_do_socket_init, 461 .destruct = virtio_transport_destruct, 462 .release = virtio_transport_release, 463 .connect = virtio_transport_connect, 464 .shutdown = virtio_transport_shutdown, 465 466 .dgram_bind = virtio_transport_dgram_bind, 467 .dgram_dequeue = virtio_transport_dgram_dequeue, 468 .dgram_enqueue = virtio_transport_dgram_enqueue, 469 .dgram_allow = virtio_transport_dgram_allow, 470 471 .stream_dequeue = virtio_transport_stream_dequeue, 472 .stream_enqueue = virtio_transport_stream_enqueue, 473 .stream_has_data = virtio_transport_stream_has_data, 474 .stream_has_space = virtio_transport_stream_has_space, 475 .stream_rcvhiwat = virtio_transport_stream_rcvhiwat, 476 .stream_is_active = virtio_transport_stream_is_active, 477 .stream_allow = virtio_transport_stream_allow, 478 479 .notify_poll_in = virtio_transport_notify_poll_in, 480 .notify_poll_out = virtio_transport_notify_poll_out, 481 .notify_recv_init = virtio_transport_notify_recv_init, 482 .notify_recv_pre_block = virtio_transport_notify_recv_pre_block, 483 .notify_recv_pre_dequeue = virtio_transport_notify_recv_pre_dequeue, 484 .notify_recv_post_dequeue = virtio_transport_notify_recv_post_dequeue, 485 .notify_send_init = virtio_transport_notify_send_init, 486 .notify_send_pre_block = virtio_transport_notify_send_pre_block, 487 .notify_send_pre_enqueue = virtio_transport_notify_send_pre_enqueue, 488 .notify_send_post_enqueue = virtio_transport_notify_send_post_enqueue, 489 490 .set_buffer_size = virtio_transport_set_buffer_size, 491 .set_min_buffer_size = virtio_transport_set_min_buffer_size, 492 .set_max_buffer_size = virtio_transport_set_max_buffer_size, 493 .get_buffer_size = virtio_transport_get_buffer_size, 494 .get_min_buffer_size = virtio_transport_get_min_buffer_size, 495 .get_max_buffer_size = virtio_transport_get_max_buffer_size, 496 }, 497 498 .send_pkt = virtio_transport_send_pkt, 499 }; 500 501 static int virtio_vsock_probe(struct virtio_device *vdev) 502 { 503 vq_callback_t *callbacks[] = { 504 virtio_vsock_rx_done, 505 virtio_vsock_tx_done, 506 virtio_vsock_event_done, 507 }; 508 static const char * const names[] = { 509 "rx", 510 "tx", 511 "event", 512 }; 513 struct virtio_vsock *vsock = NULL; 514 int ret; 515 516 ret = mutex_lock_interruptible(&the_virtio_vsock_mutex); 517 if (ret) 518 return ret; 519 520 /* Only one virtio-vsock device per guest is supported */ 521 if (the_virtio_vsock) { 522 ret = -EBUSY; 523 goto out; 524 } 525 526 vsock = kzalloc(sizeof(*vsock), GFP_KERNEL); 527 if (!vsock) { 528 ret = -ENOMEM; 529 goto out; 530 } 531 532 vsock->vdev = vdev; 533 534 ret = vsock->vdev->config->find_vqs(vsock->vdev, VSOCK_VQ_MAX, 535 vsock->vqs, callbacks, names); 536 if (ret < 0) 537 goto out; 538 539 virtio_vsock_update_guest_cid(vsock); 540 541 ret = vsock_core_init(&virtio_transport.transport); 542 if (ret < 0) 543 goto out_vqs; 544 545 vsock->rx_buf_nr = 0; 546 vsock->rx_buf_max_nr = 0; 547 atomic_set(&vsock->queued_replies, 0); 548 549 vdev->priv = vsock; 550 the_virtio_vsock = vsock; 551 mutex_init(&vsock->tx_lock); 552 mutex_init(&vsock->rx_lock); 553 mutex_init(&vsock->event_lock); 554 spin_lock_init(&vsock->send_pkt_list_lock); 555 INIT_LIST_HEAD(&vsock->send_pkt_list); 556 spin_lock_init(&vsock->loopback_list_lock); 557 INIT_LIST_HEAD(&vsock->loopback_list); 558 INIT_WORK(&vsock->rx_work, virtio_transport_rx_work); 559 INIT_WORK(&vsock->tx_work, virtio_transport_tx_work); 560 INIT_WORK(&vsock->event_work, virtio_transport_event_work); 561 INIT_WORK(&vsock->send_pkt_work, virtio_transport_send_pkt_work); 562 INIT_WORK(&vsock->loopback_work, virtio_transport_loopback_work); 563 564 mutex_lock(&vsock->rx_lock); 565 virtio_vsock_rx_fill(vsock); 566 mutex_unlock(&vsock->rx_lock); 567 568 mutex_lock(&vsock->event_lock); 569 virtio_vsock_event_fill(vsock); 570 mutex_unlock(&vsock->event_lock); 571 572 mutex_unlock(&the_virtio_vsock_mutex); 573 return 0; 574 575 out_vqs: 576 vsock->vdev->config->del_vqs(vsock->vdev); 577 out: 578 kfree(vsock); 579 mutex_unlock(&the_virtio_vsock_mutex); 580 return ret; 581 } 582 583 static void virtio_vsock_remove(struct virtio_device *vdev) 584 { 585 struct virtio_vsock *vsock = vdev->priv; 586 struct virtio_vsock_pkt *pkt; 587 588 flush_work(&vsock->loopback_work); 589 flush_work(&vsock->rx_work); 590 flush_work(&vsock->tx_work); 591 flush_work(&vsock->event_work); 592 flush_work(&vsock->send_pkt_work); 593 594 vdev->config->reset(vdev); 595 596 mutex_lock(&vsock->rx_lock); 597 while ((pkt = virtqueue_detach_unused_buf(vsock->vqs[VSOCK_VQ_RX]))) 598 virtio_transport_free_pkt(pkt); 599 mutex_unlock(&vsock->rx_lock); 600 601 mutex_lock(&vsock->tx_lock); 602 while ((pkt = virtqueue_detach_unused_buf(vsock->vqs[VSOCK_VQ_TX]))) 603 virtio_transport_free_pkt(pkt); 604 mutex_unlock(&vsock->tx_lock); 605 606 spin_lock_bh(&vsock->send_pkt_list_lock); 607 while (!list_empty(&vsock->send_pkt_list)) { 608 pkt = list_first_entry(&vsock->send_pkt_list, 609 struct virtio_vsock_pkt, list); 610 list_del(&pkt->list); 611 virtio_transport_free_pkt(pkt); 612 } 613 spin_unlock_bh(&vsock->send_pkt_list_lock); 614 615 spin_lock_bh(&vsock->loopback_list_lock); 616 while (!list_empty(&vsock->loopback_list)) { 617 pkt = list_first_entry(&vsock->loopback_list, 618 struct virtio_vsock_pkt, list); 619 list_del(&pkt->list); 620 virtio_transport_free_pkt(pkt); 621 } 622 spin_unlock_bh(&vsock->loopback_list_lock); 623 624 mutex_lock(&the_virtio_vsock_mutex); 625 the_virtio_vsock = NULL; 626 vsock_core_exit(); 627 mutex_unlock(&the_virtio_vsock_mutex); 628 629 vdev->config->del_vqs(vdev); 630 631 kfree(vsock); 632 } 633 634 static struct virtio_device_id id_table[] = { 635 { VIRTIO_ID_VSOCK, VIRTIO_DEV_ANY_ID }, 636 { 0 }, 637 }; 638 639 static unsigned int features[] = { 640 }; 641 642 static struct virtio_driver virtio_vsock_driver = { 643 .feature_table = features, 644 .feature_table_size = ARRAY_SIZE(features), 645 .driver.name = KBUILD_MODNAME, 646 .driver.owner = THIS_MODULE, 647 .id_table = id_table, 648 .probe = virtio_vsock_probe, 649 .remove = virtio_vsock_remove, 650 }; 651 652 static int __init virtio_vsock_init(void) 653 { 654 int ret; 655 656 virtio_vsock_workqueue = alloc_workqueue("virtio_vsock", 0, 0); 657 if (!virtio_vsock_workqueue) 658 return -ENOMEM; 659 ret = register_virtio_driver(&virtio_vsock_driver); 660 if (ret) 661 destroy_workqueue(virtio_vsock_workqueue); 662 return ret; 663 } 664 665 static void __exit virtio_vsock_exit(void) 666 { 667 unregister_virtio_driver(&virtio_vsock_driver); 668 destroy_workqueue(virtio_vsock_workqueue); 669 } 670 671 module_init(virtio_vsock_init); 672 module_exit(virtio_vsock_exit); 673 MODULE_LICENSE("GPL v2"); 674 MODULE_AUTHOR("Asias He"); 675 MODULE_DESCRIPTION("virtio transport for vsock"); 676 MODULE_DEVICE_TABLE(virtio, id_table); 677