1 // SPDX-License-Identifier: GPL-2.0-only
3 * virtio transport for vsock
5 * Copyright (C) 2013-2015 Red Hat, Inc.
6 * Author: Asias He <asias@redhat.com>
7 * Stefan Hajnoczi <stefanha@redhat.com>
9 * Some of the code is take from Gerd Hoffmann <kraxel@redhat.com>'s
10 * early virtio-vsock proof-of-concept bits.
12 #include <linux/spinlock.h>
13 #include <linux/module.h>
14 #include <linux/list.h>
15 #include <linux/atomic.h>
16 #include <linux/virtio.h>
17 #include <linux/virtio_ids.h>
18 #include <linux/virtio_config.h>
19 #include <linux/virtio_vsock.h>
21 #include <linux/mutex.h>
22 #include <net/af_vsock.h>
24 static struct workqueue_struct *virtio_vsock_workqueue;
25 static struct virtio_vsock *the_virtio_vsock;
26 static DEFINE_MUTEX(the_virtio_vsock_mutex); /* protects the_virtio_vsock */
29 struct virtio_device *vdev;
30 struct virtqueue *vqs[VSOCK_VQ_MAX];
32 /* Virtqueue processing is deferred to a workqueue */
33 struct work_struct tx_work;
34 struct work_struct rx_work;
35 struct work_struct event_work;
37 /* The following fields are protected by tx_lock. vqs[VSOCK_VQ_TX]
38 * must be accessed with tx_lock held.
43 struct work_struct send_pkt_work;
44 spinlock_t send_pkt_list_lock;
45 struct list_head send_pkt_list;
47 struct work_struct loopback_work;
48 spinlock_t loopback_list_lock; /* protects loopback_list */
49 struct list_head loopback_list;
51 atomic_t queued_replies;
53 /* The following fields are protected by rx_lock. vqs[VSOCK_VQ_RX]
54 * must be accessed with rx_lock held.
61 /* The following fields are protected by event_lock.
62 * vqs[VSOCK_VQ_EVENT] must be accessed with event_lock held.
64 struct mutex event_lock;
66 struct virtio_vsock_event event_list[8];
71 static u32 virtio_transport_get_local_cid(void)
73 struct virtio_vsock *vsock;
77 vsock = rcu_dereference(the_virtio_vsock);
83 ret = vsock->guest_cid;
89 static int virtio_transport_send_pkt_loopback(struct virtio_vsock *vsock,
90 struct virtio_vsock_pkt *pkt)
94 spin_lock_bh(&vsock->loopback_list_lock);
95 list_add_tail(&pkt->list, &vsock->loopback_list);
96 spin_unlock_bh(&vsock->loopback_list_lock);
98 queue_work(virtio_vsock_workqueue, &vsock->loopback_work);
104 virtio_transport_send_pkt_work(struct work_struct *work)
106 struct virtio_vsock *vsock =
107 container_of(work, struct virtio_vsock, send_pkt_work);
108 struct virtqueue *vq;
110 bool restart_rx = false;
112 mutex_lock(&vsock->tx_lock);
117 vq = vsock->vqs[VSOCK_VQ_TX];
120 struct virtio_vsock_pkt *pkt;
121 struct scatterlist hdr, buf, *sgs[2];
122 int ret, in_sg = 0, out_sg = 0;
125 spin_lock_bh(&vsock->send_pkt_list_lock);
126 if (list_empty(&vsock->send_pkt_list)) {
127 spin_unlock_bh(&vsock->send_pkt_list_lock);
131 pkt = list_first_entry(&vsock->send_pkt_list,
132 struct virtio_vsock_pkt, list);
133 list_del_init(&pkt->list);
134 spin_unlock_bh(&vsock->send_pkt_list_lock);
136 virtio_transport_deliver_tap_pkt(pkt);
140 sg_init_one(&hdr, &pkt->hdr, sizeof(pkt->hdr));
141 sgs[out_sg++] = &hdr;
143 sg_init_one(&buf, pkt->buf, pkt->len);
144 sgs[out_sg++] = &buf;
147 ret = virtqueue_add_sgs(vq, sgs, out_sg, in_sg, pkt, GFP_KERNEL);
148 /* Usually this means that there is no more space available in
152 spin_lock_bh(&vsock->send_pkt_list_lock);
153 list_add(&pkt->list, &vsock->send_pkt_list);
154 spin_unlock_bh(&vsock->send_pkt_list_lock);
159 struct virtqueue *rx_vq = vsock->vqs[VSOCK_VQ_RX];
162 val = atomic_dec_return(&vsock->queued_replies);
164 /* Do we now have resources to resume rx processing? */
165 if (val + 1 == virtqueue_get_vring_size(rx_vq))
176 mutex_unlock(&vsock->tx_lock);
179 queue_work(virtio_vsock_workqueue, &vsock->rx_work);
183 virtio_transport_send_pkt(struct virtio_vsock_pkt *pkt)
185 struct virtio_vsock *vsock;
189 vsock = rcu_dereference(the_virtio_vsock);
191 virtio_transport_free_pkt(pkt);
196 if (le64_to_cpu(pkt->hdr.dst_cid) == vsock->guest_cid) {
197 len = virtio_transport_send_pkt_loopback(vsock, pkt);
202 atomic_inc(&vsock->queued_replies);
204 spin_lock_bh(&vsock->send_pkt_list_lock);
205 list_add_tail(&pkt->list, &vsock->send_pkt_list);
206 spin_unlock_bh(&vsock->send_pkt_list_lock);
208 queue_work(virtio_vsock_workqueue, &vsock->send_pkt_work);
216 virtio_transport_cancel_pkt(struct vsock_sock *vsk)
218 struct virtio_vsock *vsock;
219 struct virtio_vsock_pkt *pkt, *n;
224 vsock = rcu_dereference(the_virtio_vsock);
230 spin_lock_bh(&vsock->send_pkt_list_lock);
231 list_for_each_entry_safe(pkt, n, &vsock->send_pkt_list, list) {
234 list_move(&pkt->list, &freeme);
236 spin_unlock_bh(&vsock->send_pkt_list_lock);
238 list_for_each_entry_safe(pkt, n, &freeme, list) {
241 list_del(&pkt->list);
242 virtio_transport_free_pkt(pkt);
246 struct virtqueue *rx_vq = vsock->vqs[VSOCK_VQ_RX];
249 new_cnt = atomic_sub_return(cnt, &vsock->queued_replies);
250 if (new_cnt + cnt >= virtqueue_get_vring_size(rx_vq) &&
251 new_cnt < virtqueue_get_vring_size(rx_vq))
252 queue_work(virtio_vsock_workqueue, &vsock->rx_work);
262 static void virtio_vsock_rx_fill(struct virtio_vsock *vsock)
264 int buf_len = VIRTIO_VSOCK_DEFAULT_RX_BUF_SIZE;
265 struct virtio_vsock_pkt *pkt;
266 struct scatterlist hdr, buf, *sgs[2];
267 struct virtqueue *vq;
270 vq = vsock->vqs[VSOCK_VQ_RX];
273 pkt = kzalloc(sizeof(*pkt), GFP_KERNEL);
277 pkt->buf = kmalloc(buf_len, GFP_KERNEL);
279 virtio_transport_free_pkt(pkt);
283 pkt->buf_len = buf_len;
286 sg_init_one(&hdr, &pkt->hdr, sizeof(pkt->hdr));
289 sg_init_one(&buf, pkt->buf, buf_len);
291 ret = virtqueue_add_sgs(vq, sgs, 0, 2, pkt, GFP_KERNEL);
293 virtio_transport_free_pkt(pkt);
297 } while (vq->num_free);
298 if (vsock->rx_buf_nr > vsock->rx_buf_max_nr)
299 vsock->rx_buf_max_nr = vsock->rx_buf_nr;
303 static void virtio_transport_tx_work(struct work_struct *work)
305 struct virtio_vsock *vsock =
306 container_of(work, struct virtio_vsock, tx_work);
307 struct virtqueue *vq;
310 vq = vsock->vqs[VSOCK_VQ_TX];
311 mutex_lock(&vsock->tx_lock);
317 struct virtio_vsock_pkt *pkt;
320 virtqueue_disable_cb(vq);
321 while ((pkt = virtqueue_get_buf(vq, &len)) != NULL) {
322 virtio_transport_free_pkt(pkt);
325 } while (!virtqueue_enable_cb(vq));
328 mutex_unlock(&vsock->tx_lock);
331 queue_work(virtio_vsock_workqueue, &vsock->send_pkt_work);
334 /* Is there space left for replies to rx packets? */
335 static bool virtio_transport_more_replies(struct virtio_vsock *vsock)
337 struct virtqueue *vq = vsock->vqs[VSOCK_VQ_RX];
340 smp_rmb(); /* paired with atomic_inc() and atomic_dec_return() */
341 val = atomic_read(&vsock->queued_replies);
343 return val < virtqueue_get_vring_size(vq);
346 /* event_lock must be held */
347 static int virtio_vsock_event_fill_one(struct virtio_vsock *vsock,
348 struct virtio_vsock_event *event)
350 struct scatterlist sg;
351 struct virtqueue *vq;
353 vq = vsock->vqs[VSOCK_VQ_EVENT];
355 sg_init_one(&sg, event, sizeof(*event));
357 return virtqueue_add_inbuf(vq, &sg, 1, event, GFP_KERNEL);
360 /* event_lock must be held */
361 static void virtio_vsock_event_fill(struct virtio_vsock *vsock)
365 for (i = 0; i < ARRAY_SIZE(vsock->event_list); i++) {
366 struct virtio_vsock_event *event = &vsock->event_list[i];
368 virtio_vsock_event_fill_one(vsock, event);
371 virtqueue_kick(vsock->vqs[VSOCK_VQ_EVENT]);
374 static void virtio_vsock_reset_sock(struct sock *sk)
377 sk->sk_state = TCP_CLOSE;
378 sk->sk_err = ECONNRESET;
379 sk->sk_error_report(sk);
383 static void virtio_vsock_update_guest_cid(struct virtio_vsock *vsock)
385 struct virtio_device *vdev = vsock->vdev;
388 vdev->config->get(vdev, offsetof(struct virtio_vsock_config, guest_cid),
389 &guest_cid, sizeof(guest_cid));
390 vsock->guest_cid = le64_to_cpu(guest_cid);
393 /* event_lock must be held */
394 static void virtio_vsock_event_handle(struct virtio_vsock *vsock,
395 struct virtio_vsock_event *event)
397 switch (le32_to_cpu(event->id)) {
398 case VIRTIO_VSOCK_EVENT_TRANSPORT_RESET:
399 virtio_vsock_update_guest_cid(vsock);
400 vsock_for_each_connected_socket(virtio_vsock_reset_sock);
405 static void virtio_transport_event_work(struct work_struct *work)
407 struct virtio_vsock *vsock =
408 container_of(work, struct virtio_vsock, event_work);
409 struct virtqueue *vq;
411 vq = vsock->vqs[VSOCK_VQ_EVENT];
413 mutex_lock(&vsock->event_lock);
415 if (!vsock->event_run)
419 struct virtio_vsock_event *event;
422 virtqueue_disable_cb(vq);
423 while ((event = virtqueue_get_buf(vq, &len)) != NULL) {
424 if (len == sizeof(*event))
425 virtio_vsock_event_handle(vsock, event);
427 virtio_vsock_event_fill_one(vsock, event);
429 } while (!virtqueue_enable_cb(vq));
431 virtqueue_kick(vsock->vqs[VSOCK_VQ_EVENT]);
433 mutex_unlock(&vsock->event_lock);
436 static void virtio_vsock_event_done(struct virtqueue *vq)
438 struct virtio_vsock *vsock = vq->vdev->priv;
442 queue_work(virtio_vsock_workqueue, &vsock->event_work);
445 static void virtio_vsock_tx_done(struct virtqueue *vq)
447 struct virtio_vsock *vsock = vq->vdev->priv;
451 queue_work(virtio_vsock_workqueue, &vsock->tx_work);
454 static void virtio_vsock_rx_done(struct virtqueue *vq)
456 struct virtio_vsock *vsock = vq->vdev->priv;
460 queue_work(virtio_vsock_workqueue, &vsock->rx_work);
463 static struct virtio_transport virtio_transport = {
465 .module = THIS_MODULE,
467 .get_local_cid = virtio_transport_get_local_cid,
469 .init = virtio_transport_do_socket_init,
470 .destruct = virtio_transport_destruct,
471 .release = virtio_transport_release,
472 .connect = virtio_transport_connect,
473 .shutdown = virtio_transport_shutdown,
474 .cancel_pkt = virtio_transport_cancel_pkt,
476 .dgram_bind = virtio_transport_dgram_bind,
477 .dgram_dequeue = virtio_transport_dgram_dequeue,
478 .dgram_enqueue = virtio_transport_dgram_enqueue,
479 .dgram_allow = virtio_transport_dgram_allow,
481 .stream_dequeue = virtio_transport_stream_dequeue,
482 .stream_enqueue = virtio_transport_stream_enqueue,
483 .stream_has_data = virtio_transport_stream_has_data,
484 .stream_has_space = virtio_transport_stream_has_space,
485 .stream_rcvhiwat = virtio_transport_stream_rcvhiwat,
486 .stream_is_active = virtio_transport_stream_is_active,
487 .stream_allow = virtio_transport_stream_allow,
489 .notify_poll_in = virtio_transport_notify_poll_in,
490 .notify_poll_out = virtio_transport_notify_poll_out,
491 .notify_recv_init = virtio_transport_notify_recv_init,
492 .notify_recv_pre_block = virtio_transport_notify_recv_pre_block,
493 .notify_recv_pre_dequeue = virtio_transport_notify_recv_pre_dequeue,
494 .notify_recv_post_dequeue = virtio_transport_notify_recv_post_dequeue,
495 .notify_send_init = virtio_transport_notify_send_init,
496 .notify_send_pre_block = virtio_transport_notify_send_pre_block,
497 .notify_send_pre_enqueue = virtio_transport_notify_send_pre_enqueue,
498 .notify_send_post_enqueue = virtio_transport_notify_send_post_enqueue,
499 .notify_buffer_size = virtio_transport_notify_buffer_size,
502 .send_pkt = virtio_transport_send_pkt,
505 static void virtio_transport_loopback_work(struct work_struct *work)
507 struct virtio_vsock *vsock =
508 container_of(work, struct virtio_vsock, loopback_work);
511 spin_lock_bh(&vsock->loopback_list_lock);
512 list_splice_init(&vsock->loopback_list, &pkts);
513 spin_unlock_bh(&vsock->loopback_list_lock);
515 mutex_lock(&vsock->rx_lock);
520 while (!list_empty(&pkts)) {
521 struct virtio_vsock_pkt *pkt;
523 pkt = list_first_entry(&pkts, struct virtio_vsock_pkt, list);
524 list_del_init(&pkt->list);
526 virtio_transport_recv_pkt(&virtio_transport, pkt);
529 mutex_unlock(&vsock->rx_lock);
532 static void virtio_transport_rx_work(struct work_struct *work)
534 struct virtio_vsock *vsock =
535 container_of(work, struct virtio_vsock, rx_work);
536 struct virtqueue *vq;
538 vq = vsock->vqs[VSOCK_VQ_RX];
540 mutex_lock(&vsock->rx_lock);
546 virtqueue_disable_cb(vq);
548 struct virtio_vsock_pkt *pkt;
551 if (!virtio_transport_more_replies(vsock)) {
552 /* Stop rx until the device processes already
553 * pending replies. Leave rx virtqueue
554 * callbacks disabled.
559 pkt = virtqueue_get_buf(vq, &len);
566 /* Drop short/long packets */
567 if (unlikely(len < sizeof(pkt->hdr) ||
568 len > sizeof(pkt->hdr) + pkt->len)) {
569 virtio_transport_free_pkt(pkt);
573 pkt->len = len - sizeof(pkt->hdr);
574 virtio_transport_deliver_tap_pkt(pkt);
575 virtio_transport_recv_pkt(&virtio_transport, pkt);
577 } while (!virtqueue_enable_cb(vq));
580 if (vsock->rx_buf_nr < vsock->rx_buf_max_nr / 2)
581 virtio_vsock_rx_fill(vsock);
582 mutex_unlock(&vsock->rx_lock);
585 static int virtio_vsock_probe(struct virtio_device *vdev)
587 vq_callback_t *callbacks[] = {
588 virtio_vsock_rx_done,
589 virtio_vsock_tx_done,
590 virtio_vsock_event_done,
592 static const char * const names[] = {
597 struct virtio_vsock *vsock = NULL;
600 ret = mutex_lock_interruptible(&the_virtio_vsock_mutex);
604 /* Only one virtio-vsock device per guest is supported */
605 if (rcu_dereference_protected(the_virtio_vsock,
606 lockdep_is_held(&the_virtio_vsock_mutex))) {
611 vsock = kzalloc(sizeof(*vsock), GFP_KERNEL);
619 ret = virtio_find_vqs(vsock->vdev, VSOCK_VQ_MAX,
620 vsock->vqs, callbacks, names,
625 virtio_vsock_update_guest_cid(vsock);
627 vsock->rx_buf_nr = 0;
628 vsock->rx_buf_max_nr = 0;
629 atomic_set(&vsock->queued_replies, 0);
631 mutex_init(&vsock->tx_lock);
632 mutex_init(&vsock->rx_lock);
633 mutex_init(&vsock->event_lock);
634 spin_lock_init(&vsock->send_pkt_list_lock);
635 INIT_LIST_HEAD(&vsock->send_pkt_list);
636 spin_lock_init(&vsock->loopback_list_lock);
637 INIT_LIST_HEAD(&vsock->loopback_list);
638 INIT_WORK(&vsock->rx_work, virtio_transport_rx_work);
639 INIT_WORK(&vsock->tx_work, virtio_transport_tx_work);
640 INIT_WORK(&vsock->event_work, virtio_transport_event_work);
641 INIT_WORK(&vsock->send_pkt_work, virtio_transport_send_pkt_work);
642 INIT_WORK(&vsock->loopback_work, virtio_transport_loopback_work);
644 mutex_lock(&vsock->tx_lock);
645 vsock->tx_run = true;
646 mutex_unlock(&vsock->tx_lock);
648 mutex_lock(&vsock->rx_lock);
649 virtio_vsock_rx_fill(vsock);
650 vsock->rx_run = true;
651 mutex_unlock(&vsock->rx_lock);
653 mutex_lock(&vsock->event_lock);
654 virtio_vsock_event_fill(vsock);
655 vsock->event_run = true;
656 mutex_unlock(&vsock->event_lock);
659 rcu_assign_pointer(the_virtio_vsock, vsock);
661 mutex_unlock(&the_virtio_vsock_mutex);
666 mutex_unlock(&the_virtio_vsock_mutex);
670 static void virtio_vsock_remove(struct virtio_device *vdev)
672 struct virtio_vsock *vsock = vdev->priv;
673 struct virtio_vsock_pkt *pkt;
675 mutex_lock(&the_virtio_vsock_mutex);
678 rcu_assign_pointer(the_virtio_vsock, NULL);
681 /* Reset all connected sockets when the device disappear */
682 vsock_for_each_connected_socket(virtio_vsock_reset_sock);
684 /* Stop all work handlers to make sure no one is accessing the device,
685 * so we can safely call vdev->config->reset().
687 mutex_lock(&vsock->rx_lock);
688 vsock->rx_run = false;
689 mutex_unlock(&vsock->rx_lock);
691 mutex_lock(&vsock->tx_lock);
692 vsock->tx_run = false;
693 mutex_unlock(&vsock->tx_lock);
695 mutex_lock(&vsock->event_lock);
696 vsock->event_run = false;
697 mutex_unlock(&vsock->event_lock);
699 /* Flush all device writes and interrupts, device will not use any
702 vdev->config->reset(vdev);
704 mutex_lock(&vsock->rx_lock);
705 while ((pkt = virtqueue_detach_unused_buf(vsock->vqs[VSOCK_VQ_RX])))
706 virtio_transport_free_pkt(pkt);
707 mutex_unlock(&vsock->rx_lock);
709 mutex_lock(&vsock->tx_lock);
710 while ((pkt = virtqueue_detach_unused_buf(vsock->vqs[VSOCK_VQ_TX])))
711 virtio_transport_free_pkt(pkt);
712 mutex_unlock(&vsock->tx_lock);
714 spin_lock_bh(&vsock->send_pkt_list_lock);
715 while (!list_empty(&vsock->send_pkt_list)) {
716 pkt = list_first_entry(&vsock->send_pkt_list,
717 struct virtio_vsock_pkt, list);
718 list_del(&pkt->list);
719 virtio_transport_free_pkt(pkt);
721 spin_unlock_bh(&vsock->send_pkt_list_lock);
723 spin_lock_bh(&vsock->loopback_list_lock);
724 while (!list_empty(&vsock->loopback_list)) {
725 pkt = list_first_entry(&vsock->loopback_list,
726 struct virtio_vsock_pkt, list);
727 list_del(&pkt->list);
728 virtio_transport_free_pkt(pkt);
730 spin_unlock_bh(&vsock->loopback_list_lock);
732 /* Delete virtqueues and flush outstanding callbacks if any */
733 vdev->config->del_vqs(vdev);
735 /* Other works can be queued before 'config->del_vqs()', so we flush
736 * all works before to free the vsock object to avoid use after free.
738 flush_work(&vsock->loopback_work);
739 flush_work(&vsock->rx_work);
740 flush_work(&vsock->tx_work);
741 flush_work(&vsock->event_work);
742 flush_work(&vsock->send_pkt_work);
744 mutex_unlock(&the_virtio_vsock_mutex);
749 static struct virtio_device_id id_table[] = {
750 { VIRTIO_ID_VSOCK, VIRTIO_DEV_ANY_ID },
754 static unsigned int features[] = {
757 static struct virtio_driver virtio_vsock_driver = {
758 .feature_table = features,
759 .feature_table_size = ARRAY_SIZE(features),
760 .driver.name = KBUILD_MODNAME,
761 .driver.owner = THIS_MODULE,
762 .id_table = id_table,
763 .probe = virtio_vsock_probe,
764 .remove = virtio_vsock_remove,
767 static int __init virtio_vsock_init(void)
771 virtio_vsock_workqueue = alloc_workqueue("virtio_vsock", 0, 0);
772 if (!virtio_vsock_workqueue)
775 ret = vsock_core_register(&virtio_transport.transport,
776 VSOCK_TRANSPORT_F_G2H);
780 ret = register_virtio_driver(&virtio_vsock_driver);
787 vsock_core_unregister(&virtio_transport.transport);
789 destroy_workqueue(virtio_vsock_workqueue);
793 static void __exit virtio_vsock_exit(void)
795 unregister_virtio_driver(&virtio_vsock_driver);
796 vsock_core_unregister(&virtio_transport.transport);
797 destroy_workqueue(virtio_vsock_workqueue);
800 module_init(virtio_vsock_init);
801 module_exit(virtio_vsock_exit);
802 MODULE_LICENSE("GPL v2");
803 MODULE_AUTHOR("Asias He");
804 MODULE_DESCRIPTION("virtio transport for vsock");
805 MODULE_DEVICE_TABLE(virtio, id_table);