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 __rcu *the_virtio_vsock;
26 static DEFINE_MUTEX(the_virtio_vsock_mutex); /* protects the_virtio_vsock */
27 static struct virtio_transport virtio_transport; /* forward declaration */
30 struct virtio_device *vdev;
31 struct virtqueue *vqs[VSOCK_VQ_MAX];
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;
38 /* The following fields are protected by tx_lock. vqs[VSOCK_VQ_TX]
39 * must be accessed with tx_lock held.
44 struct work_struct send_pkt_work;
45 struct sk_buff_head send_pkt_queue;
47 atomic_t queued_replies;
49 /* The following fields are protected by rx_lock. vqs[VSOCK_VQ_RX]
50 * must be accessed with rx_lock held.
57 /* The following fields are protected by event_lock.
58 * vqs[VSOCK_VQ_EVENT] must be accessed with event_lock held.
60 struct mutex event_lock;
62 struct virtio_vsock_event event_list[8];
68 static u32 virtio_transport_get_local_cid(void)
70 struct virtio_vsock *vsock;
74 vsock = rcu_dereference(the_virtio_vsock);
80 ret = vsock->guest_cid;
87 virtio_transport_send_pkt_work(struct work_struct *work)
89 struct virtio_vsock *vsock =
90 container_of(work, struct virtio_vsock, send_pkt_work);
93 bool restart_rx = false;
95 mutex_lock(&vsock->tx_lock);
100 vq = vsock->vqs[VSOCK_VQ_TX];
103 struct scatterlist hdr, buf, *sgs[2];
104 int ret, in_sg = 0, out_sg = 0;
108 skb = virtio_vsock_skb_dequeue(&vsock->send_pkt_queue);
112 virtio_transport_deliver_tap_pkt(skb);
113 reply = virtio_vsock_skb_reply(skb);
115 sg_init_one(&hdr, virtio_vsock_hdr(skb), sizeof(*virtio_vsock_hdr(skb)));
116 sgs[out_sg++] = &hdr;
118 sg_init_one(&buf, skb->data, skb->len);
119 sgs[out_sg++] = &buf;
122 ret = virtqueue_add_sgs(vq, sgs, out_sg, in_sg, skb, GFP_KERNEL);
123 /* Usually this means that there is no more space available in
127 virtio_vsock_skb_queue_head(&vsock->send_pkt_queue, skb);
132 struct virtqueue *rx_vq = vsock->vqs[VSOCK_VQ_RX];
135 val = atomic_dec_return(&vsock->queued_replies);
137 /* Do we now have resources to resume rx processing? */
138 if (val + 1 == virtqueue_get_vring_size(rx_vq))
149 mutex_unlock(&vsock->tx_lock);
152 queue_work(virtio_vsock_workqueue, &vsock->rx_work);
156 virtio_transport_send_pkt(struct sk_buff *skb)
158 struct virtio_vsock_hdr *hdr;
159 struct virtio_vsock *vsock;
162 hdr = virtio_vsock_hdr(skb);
165 vsock = rcu_dereference(the_virtio_vsock);
172 if (le64_to_cpu(hdr->dst_cid) == vsock->guest_cid) {
178 if (virtio_vsock_skb_reply(skb))
179 atomic_inc(&vsock->queued_replies);
181 virtio_vsock_skb_queue_tail(&vsock->send_pkt_queue, skb);
182 queue_work(virtio_vsock_workqueue, &vsock->send_pkt_work);
190 virtio_transport_cancel_pkt(struct vsock_sock *vsk)
192 struct virtio_vsock *vsock;
196 vsock = rcu_dereference(the_virtio_vsock);
202 cnt = virtio_transport_purge_skbs(vsk, &vsock->send_pkt_queue);
205 struct virtqueue *rx_vq = vsock->vqs[VSOCK_VQ_RX];
208 new_cnt = atomic_sub_return(cnt, &vsock->queued_replies);
209 if (new_cnt + cnt >= virtqueue_get_vring_size(rx_vq) &&
210 new_cnt < virtqueue_get_vring_size(rx_vq))
211 queue_work(virtio_vsock_workqueue, &vsock->rx_work);
221 static void virtio_vsock_rx_fill(struct virtio_vsock *vsock)
223 int total_len = VIRTIO_VSOCK_DEFAULT_RX_BUF_SIZE + VIRTIO_VSOCK_SKB_HEADROOM;
224 struct scatterlist pkt, *p;
225 struct virtqueue *vq;
229 vq = vsock->vqs[VSOCK_VQ_RX];
232 skb = virtio_vsock_alloc_skb(total_len, GFP_KERNEL);
236 memset(skb->head, 0, VIRTIO_VSOCK_SKB_HEADROOM);
237 sg_init_one(&pkt, virtio_vsock_hdr(skb), total_len);
239 ret = virtqueue_add_sgs(vq, &p, 0, 1, skb, GFP_KERNEL);
246 } while (vq->num_free);
247 if (vsock->rx_buf_nr > vsock->rx_buf_max_nr)
248 vsock->rx_buf_max_nr = vsock->rx_buf_nr;
252 static void virtio_transport_tx_work(struct work_struct *work)
254 struct virtio_vsock *vsock =
255 container_of(work, struct virtio_vsock, tx_work);
256 struct virtqueue *vq;
259 vq = vsock->vqs[VSOCK_VQ_TX];
260 mutex_lock(&vsock->tx_lock);
269 virtqueue_disable_cb(vq);
270 while ((skb = virtqueue_get_buf(vq, &len)) != NULL) {
274 } while (!virtqueue_enable_cb(vq));
277 mutex_unlock(&vsock->tx_lock);
280 queue_work(virtio_vsock_workqueue, &vsock->send_pkt_work);
283 /* Is there space left for replies to rx packets? */
284 static bool virtio_transport_more_replies(struct virtio_vsock *vsock)
286 struct virtqueue *vq = vsock->vqs[VSOCK_VQ_RX];
289 smp_rmb(); /* paired with atomic_inc() and atomic_dec_return() */
290 val = atomic_read(&vsock->queued_replies);
292 return val < virtqueue_get_vring_size(vq);
295 /* event_lock must be held */
296 static int virtio_vsock_event_fill_one(struct virtio_vsock *vsock,
297 struct virtio_vsock_event *event)
299 struct scatterlist sg;
300 struct virtqueue *vq;
302 vq = vsock->vqs[VSOCK_VQ_EVENT];
304 sg_init_one(&sg, event, sizeof(*event));
306 return virtqueue_add_inbuf(vq, &sg, 1, event, GFP_KERNEL);
309 /* event_lock must be held */
310 static void virtio_vsock_event_fill(struct virtio_vsock *vsock)
314 for (i = 0; i < ARRAY_SIZE(vsock->event_list); i++) {
315 struct virtio_vsock_event *event = &vsock->event_list[i];
317 virtio_vsock_event_fill_one(vsock, event);
320 virtqueue_kick(vsock->vqs[VSOCK_VQ_EVENT]);
323 static void virtio_vsock_reset_sock(struct sock *sk)
325 /* vmci_transport.c doesn't take sk_lock here either. At least we're
326 * under vsock_table_lock so the sock cannot disappear while we're
330 sk->sk_state = TCP_CLOSE;
331 sk->sk_err = ECONNRESET;
335 static void virtio_vsock_update_guest_cid(struct virtio_vsock *vsock)
337 struct virtio_device *vdev = vsock->vdev;
340 vdev->config->get(vdev, offsetof(struct virtio_vsock_config, guest_cid),
341 &guest_cid, sizeof(guest_cid));
342 vsock->guest_cid = le64_to_cpu(guest_cid);
345 /* event_lock must be held */
346 static void virtio_vsock_event_handle(struct virtio_vsock *vsock,
347 struct virtio_vsock_event *event)
349 switch (le32_to_cpu(event->id)) {
350 case VIRTIO_VSOCK_EVENT_TRANSPORT_RESET:
351 virtio_vsock_update_guest_cid(vsock);
352 vsock_for_each_connected_socket(&virtio_transport.transport,
353 virtio_vsock_reset_sock);
358 static void virtio_transport_event_work(struct work_struct *work)
360 struct virtio_vsock *vsock =
361 container_of(work, struct virtio_vsock, event_work);
362 struct virtqueue *vq;
364 vq = vsock->vqs[VSOCK_VQ_EVENT];
366 mutex_lock(&vsock->event_lock);
368 if (!vsock->event_run)
372 struct virtio_vsock_event *event;
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);
380 virtio_vsock_event_fill_one(vsock, event);
382 } while (!virtqueue_enable_cb(vq));
384 virtqueue_kick(vsock->vqs[VSOCK_VQ_EVENT]);
386 mutex_unlock(&vsock->event_lock);
389 static void virtio_vsock_event_done(struct virtqueue *vq)
391 struct virtio_vsock *vsock = vq->vdev->priv;
395 queue_work(virtio_vsock_workqueue, &vsock->event_work);
398 static void virtio_vsock_tx_done(struct virtqueue *vq)
400 struct virtio_vsock *vsock = vq->vdev->priv;
404 queue_work(virtio_vsock_workqueue, &vsock->tx_work);
407 static void virtio_vsock_rx_done(struct virtqueue *vq)
409 struct virtio_vsock *vsock = vq->vdev->priv;
413 queue_work(virtio_vsock_workqueue, &vsock->rx_work);
416 static bool virtio_transport_seqpacket_allow(u32 remote_cid);
418 static struct virtio_transport virtio_transport = {
420 .module = THIS_MODULE,
422 .get_local_cid = virtio_transport_get_local_cid,
424 .init = virtio_transport_do_socket_init,
425 .destruct = virtio_transport_destruct,
426 .release = virtio_transport_release,
427 .connect = virtio_transport_connect,
428 .shutdown = virtio_transport_shutdown,
429 .cancel_pkt = virtio_transport_cancel_pkt,
431 .dgram_bind = virtio_transport_dgram_bind,
432 .dgram_dequeue = virtio_transport_dgram_dequeue,
433 .dgram_enqueue = virtio_transport_dgram_enqueue,
434 .dgram_allow = virtio_transport_dgram_allow,
436 .stream_dequeue = virtio_transport_stream_dequeue,
437 .stream_enqueue = virtio_transport_stream_enqueue,
438 .stream_has_data = virtio_transport_stream_has_data,
439 .stream_has_space = virtio_transport_stream_has_space,
440 .stream_rcvhiwat = virtio_transport_stream_rcvhiwat,
441 .stream_is_active = virtio_transport_stream_is_active,
442 .stream_allow = virtio_transport_stream_allow,
444 .seqpacket_dequeue = virtio_transport_seqpacket_dequeue,
445 .seqpacket_enqueue = virtio_transport_seqpacket_enqueue,
446 .seqpacket_allow = virtio_transport_seqpacket_allow,
447 .seqpacket_has_data = virtio_transport_seqpacket_has_data,
449 .notify_poll_in = virtio_transport_notify_poll_in,
450 .notify_poll_out = virtio_transport_notify_poll_out,
451 .notify_recv_init = virtio_transport_notify_recv_init,
452 .notify_recv_pre_block = virtio_transport_notify_recv_pre_block,
453 .notify_recv_pre_dequeue = virtio_transport_notify_recv_pre_dequeue,
454 .notify_recv_post_dequeue = virtio_transport_notify_recv_post_dequeue,
455 .notify_send_init = virtio_transport_notify_send_init,
456 .notify_send_pre_block = virtio_transport_notify_send_pre_block,
457 .notify_send_pre_enqueue = virtio_transport_notify_send_pre_enqueue,
458 .notify_send_post_enqueue = virtio_transport_notify_send_post_enqueue,
459 .notify_buffer_size = virtio_transport_notify_buffer_size,
461 .read_skb = virtio_transport_read_skb,
464 .send_pkt = virtio_transport_send_pkt,
467 static bool virtio_transport_seqpacket_allow(u32 remote_cid)
469 struct virtio_vsock *vsock;
470 bool seqpacket_allow;
472 seqpacket_allow = false;
474 vsock = rcu_dereference(the_virtio_vsock);
476 seqpacket_allow = vsock->seqpacket_allow;
479 return seqpacket_allow;
482 static void virtio_transport_rx_work(struct work_struct *work)
484 struct virtio_vsock *vsock =
485 container_of(work, struct virtio_vsock, rx_work);
486 struct virtqueue *vq;
488 vq = vsock->vqs[VSOCK_VQ_RX];
490 mutex_lock(&vsock->rx_lock);
496 virtqueue_disable_cb(vq);
501 if (!virtio_transport_more_replies(vsock)) {
502 /* Stop rx until the device processes already
503 * pending replies. Leave rx virtqueue
504 * callbacks disabled.
509 skb = virtqueue_get_buf(vq, &len);
515 /* Drop short/long packets */
516 if (unlikely(len < sizeof(struct virtio_vsock_hdr) ||
517 len > virtio_vsock_skb_len(skb))) {
522 virtio_vsock_skb_rx_put(skb);
523 virtio_transport_deliver_tap_pkt(skb);
524 virtio_transport_recv_pkt(&virtio_transport, skb);
526 } while (!virtqueue_enable_cb(vq));
529 if (vsock->rx_buf_nr < vsock->rx_buf_max_nr / 2)
530 virtio_vsock_rx_fill(vsock);
531 mutex_unlock(&vsock->rx_lock);
534 static int virtio_vsock_vqs_init(struct virtio_vsock *vsock)
536 struct virtio_device *vdev = vsock->vdev;
537 static const char * const names[] = {
542 vq_callback_t *callbacks[] = {
543 virtio_vsock_rx_done,
544 virtio_vsock_tx_done,
545 virtio_vsock_event_done,
549 ret = virtio_find_vqs(vdev, VSOCK_VQ_MAX, vsock->vqs, callbacks, names,
554 virtio_vsock_update_guest_cid(vsock);
556 virtio_device_ready(vdev);
558 mutex_lock(&vsock->tx_lock);
559 vsock->tx_run = true;
560 mutex_unlock(&vsock->tx_lock);
562 mutex_lock(&vsock->rx_lock);
563 virtio_vsock_rx_fill(vsock);
564 vsock->rx_run = true;
565 mutex_unlock(&vsock->rx_lock);
567 mutex_lock(&vsock->event_lock);
568 virtio_vsock_event_fill(vsock);
569 vsock->event_run = true;
570 mutex_unlock(&vsock->event_lock);
575 static void virtio_vsock_vqs_del(struct virtio_vsock *vsock)
577 struct virtio_device *vdev = vsock->vdev;
580 /* Reset all connected sockets when the VQs disappear */
581 vsock_for_each_connected_socket(&virtio_transport.transport,
582 virtio_vsock_reset_sock);
584 /* Stop all work handlers to make sure no one is accessing the device,
585 * so we can safely call virtio_reset_device().
587 mutex_lock(&vsock->rx_lock);
588 vsock->rx_run = false;
589 mutex_unlock(&vsock->rx_lock);
591 mutex_lock(&vsock->tx_lock);
592 vsock->tx_run = false;
593 mutex_unlock(&vsock->tx_lock);
595 mutex_lock(&vsock->event_lock);
596 vsock->event_run = false;
597 mutex_unlock(&vsock->event_lock);
599 /* Flush all device writes and interrupts, device will not use any
602 virtio_reset_device(vdev);
604 mutex_lock(&vsock->rx_lock);
605 while ((skb = virtqueue_detach_unused_buf(vsock->vqs[VSOCK_VQ_RX])))
607 mutex_unlock(&vsock->rx_lock);
609 mutex_lock(&vsock->tx_lock);
610 while ((skb = virtqueue_detach_unused_buf(vsock->vqs[VSOCK_VQ_TX])))
612 mutex_unlock(&vsock->tx_lock);
614 virtio_vsock_skb_queue_purge(&vsock->send_pkt_queue);
616 /* Delete virtqueues and flush outstanding callbacks if any */
617 vdev->config->del_vqs(vdev);
620 static int virtio_vsock_probe(struct virtio_device *vdev)
622 struct virtio_vsock *vsock = NULL;
625 ret = mutex_lock_interruptible(&the_virtio_vsock_mutex);
629 /* Only one virtio-vsock device per guest is supported */
630 if (rcu_dereference_protected(the_virtio_vsock,
631 lockdep_is_held(&the_virtio_vsock_mutex))) {
636 vsock = kzalloc(sizeof(*vsock), GFP_KERNEL);
644 vsock->rx_buf_nr = 0;
645 vsock->rx_buf_max_nr = 0;
646 atomic_set(&vsock->queued_replies, 0);
648 mutex_init(&vsock->tx_lock);
649 mutex_init(&vsock->rx_lock);
650 mutex_init(&vsock->event_lock);
651 skb_queue_head_init(&vsock->send_pkt_queue);
652 INIT_WORK(&vsock->rx_work, virtio_transport_rx_work);
653 INIT_WORK(&vsock->tx_work, virtio_transport_tx_work);
654 INIT_WORK(&vsock->event_work, virtio_transport_event_work);
655 INIT_WORK(&vsock->send_pkt_work, virtio_transport_send_pkt_work);
657 if (virtio_has_feature(vdev, VIRTIO_VSOCK_F_SEQPACKET))
658 vsock->seqpacket_allow = true;
662 ret = virtio_vsock_vqs_init(vsock);
666 rcu_assign_pointer(the_virtio_vsock, vsock);
668 mutex_unlock(&the_virtio_vsock_mutex);
674 mutex_unlock(&the_virtio_vsock_mutex);
678 static void virtio_vsock_remove(struct virtio_device *vdev)
680 struct virtio_vsock *vsock = vdev->priv;
682 mutex_lock(&the_virtio_vsock_mutex);
685 rcu_assign_pointer(the_virtio_vsock, NULL);
688 virtio_vsock_vqs_del(vsock);
690 /* Other works can be queued before 'config->del_vqs()', so we flush
691 * all works before to free the vsock object to avoid use after free.
693 flush_work(&vsock->rx_work);
694 flush_work(&vsock->tx_work);
695 flush_work(&vsock->event_work);
696 flush_work(&vsock->send_pkt_work);
698 mutex_unlock(&the_virtio_vsock_mutex);
703 #ifdef CONFIG_PM_SLEEP
704 static int virtio_vsock_freeze(struct virtio_device *vdev)
706 struct virtio_vsock *vsock = vdev->priv;
708 mutex_lock(&the_virtio_vsock_mutex);
710 rcu_assign_pointer(the_virtio_vsock, NULL);
713 virtio_vsock_vqs_del(vsock);
715 mutex_unlock(&the_virtio_vsock_mutex);
720 static int virtio_vsock_restore(struct virtio_device *vdev)
722 struct virtio_vsock *vsock = vdev->priv;
725 mutex_lock(&the_virtio_vsock_mutex);
727 /* Only one virtio-vsock device per guest is supported */
728 if (rcu_dereference_protected(the_virtio_vsock,
729 lockdep_is_held(&the_virtio_vsock_mutex))) {
734 ret = virtio_vsock_vqs_init(vsock);
738 rcu_assign_pointer(the_virtio_vsock, vsock);
741 mutex_unlock(&the_virtio_vsock_mutex);
744 #endif /* CONFIG_PM_SLEEP */
746 static struct virtio_device_id id_table[] = {
747 { VIRTIO_ID_VSOCK, VIRTIO_DEV_ANY_ID },
751 static unsigned int features[] = {
752 VIRTIO_VSOCK_F_SEQPACKET
755 static struct virtio_driver virtio_vsock_driver = {
756 .feature_table = features,
757 .feature_table_size = ARRAY_SIZE(features),
758 .driver.name = KBUILD_MODNAME,
759 .driver.owner = THIS_MODULE,
760 .id_table = id_table,
761 .probe = virtio_vsock_probe,
762 .remove = virtio_vsock_remove,
763 #ifdef CONFIG_PM_SLEEP
764 .freeze = virtio_vsock_freeze,
765 .restore = virtio_vsock_restore,
769 static int __init virtio_vsock_init(void)
773 virtio_vsock_workqueue = alloc_workqueue("virtio_vsock", 0, 0);
774 if (!virtio_vsock_workqueue)
777 ret = vsock_core_register(&virtio_transport.transport,
778 VSOCK_TRANSPORT_F_G2H);
782 ret = register_virtio_driver(&virtio_vsock_driver);
789 vsock_core_unregister(&virtio_transport.transport);
791 destroy_workqueue(virtio_vsock_workqueue);
795 static void __exit virtio_vsock_exit(void)
797 unregister_virtio_driver(&virtio_vsock_driver);
798 vsock_core_unregister(&virtio_transport.transport);
799 destroy_workqueue(virtio_vsock_workqueue);
802 module_init(virtio_vsock_init);
803 module_exit(virtio_vsock_exit);
804 MODULE_LICENSE("GPL v2");
805 MODULE_AUTHOR("Asias He");
806 MODULE_DESCRIPTION("virtio transport for vsock");
807 MODULE_DEVICE_TABLE(virtio, id_table);