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 spinlock_t send_pkt_list_lock;
46 struct list_head send_pkt_list;
48 atomic_t queued_replies;
50 /* The following fields are protected by rx_lock. vqs[VSOCK_VQ_RX]
51 * must be accessed with rx_lock held.
58 /* The following fields are protected by event_lock.
59 * vqs[VSOCK_VQ_EVENT] must be accessed with event_lock held.
61 struct mutex event_lock;
63 struct virtio_vsock_event event_list[8];
69 static u32 virtio_transport_get_local_cid(void)
71 struct virtio_vsock *vsock;
75 vsock = rcu_dereference(the_virtio_vsock);
81 ret = vsock->guest_cid;
88 virtio_transport_send_pkt_work(struct work_struct *work)
90 struct virtio_vsock *vsock =
91 container_of(work, struct virtio_vsock, send_pkt_work);
94 bool restart_rx = false;
96 mutex_lock(&vsock->tx_lock);
101 vq = vsock->vqs[VSOCK_VQ_TX];
104 struct virtio_vsock_pkt *pkt;
105 struct scatterlist hdr, buf, *sgs[2];
106 int ret, in_sg = 0, out_sg = 0;
109 spin_lock_bh(&vsock->send_pkt_list_lock);
110 if (list_empty(&vsock->send_pkt_list)) {
111 spin_unlock_bh(&vsock->send_pkt_list_lock);
115 pkt = list_first_entry(&vsock->send_pkt_list,
116 struct virtio_vsock_pkt, list);
117 list_del_init(&pkt->list);
118 spin_unlock_bh(&vsock->send_pkt_list_lock);
120 virtio_transport_deliver_tap_pkt(pkt);
124 sg_init_one(&hdr, &pkt->hdr, sizeof(pkt->hdr));
125 sgs[out_sg++] = &hdr;
127 sg_init_one(&buf, pkt->buf, pkt->len);
128 sgs[out_sg++] = &buf;
131 ret = virtqueue_add_sgs(vq, sgs, out_sg, in_sg, pkt, GFP_KERNEL);
132 /* Usually this means that there is no more space available in
136 spin_lock_bh(&vsock->send_pkt_list_lock);
137 list_add(&pkt->list, &vsock->send_pkt_list);
138 spin_unlock_bh(&vsock->send_pkt_list_lock);
143 struct virtqueue *rx_vq = vsock->vqs[VSOCK_VQ_RX];
146 val = atomic_dec_return(&vsock->queued_replies);
148 /* Do we now have resources to resume rx processing? */
149 if (val + 1 == virtqueue_get_vring_size(rx_vq))
160 mutex_unlock(&vsock->tx_lock);
163 queue_work(virtio_vsock_workqueue, &vsock->rx_work);
167 virtio_transport_send_pkt(struct virtio_vsock_pkt *pkt)
169 struct virtio_vsock *vsock;
173 vsock = rcu_dereference(the_virtio_vsock);
175 virtio_transport_free_pkt(pkt);
180 if (le64_to_cpu(pkt->hdr.dst_cid) == vsock->guest_cid) {
181 virtio_transport_free_pkt(pkt);
187 atomic_inc(&vsock->queued_replies);
189 spin_lock_bh(&vsock->send_pkt_list_lock);
190 list_add_tail(&pkt->list, &vsock->send_pkt_list);
191 spin_unlock_bh(&vsock->send_pkt_list_lock);
193 queue_work(virtio_vsock_workqueue, &vsock->send_pkt_work);
201 virtio_transport_cancel_pkt(struct vsock_sock *vsk)
203 struct virtio_vsock *vsock;
204 struct virtio_vsock_pkt *pkt, *n;
209 vsock = rcu_dereference(the_virtio_vsock);
215 spin_lock_bh(&vsock->send_pkt_list_lock);
216 list_for_each_entry_safe(pkt, n, &vsock->send_pkt_list, list) {
219 list_move(&pkt->list, &freeme);
221 spin_unlock_bh(&vsock->send_pkt_list_lock);
223 list_for_each_entry_safe(pkt, n, &freeme, list) {
226 list_del(&pkt->list);
227 virtio_transport_free_pkt(pkt);
231 struct virtqueue *rx_vq = vsock->vqs[VSOCK_VQ_RX];
234 new_cnt = atomic_sub_return(cnt, &vsock->queued_replies);
235 if (new_cnt + cnt >= virtqueue_get_vring_size(rx_vq) &&
236 new_cnt < virtqueue_get_vring_size(rx_vq))
237 queue_work(virtio_vsock_workqueue, &vsock->rx_work);
247 static void virtio_vsock_rx_fill(struct virtio_vsock *vsock)
249 int buf_len = VIRTIO_VSOCK_DEFAULT_RX_BUF_SIZE;
250 struct virtio_vsock_pkt *pkt;
251 struct scatterlist hdr, buf, *sgs[2];
252 struct virtqueue *vq;
255 vq = vsock->vqs[VSOCK_VQ_RX];
258 pkt = kzalloc(sizeof(*pkt), GFP_KERNEL);
262 pkt->buf = kmalloc(buf_len, GFP_KERNEL);
264 virtio_transport_free_pkt(pkt);
268 pkt->buf_len = buf_len;
271 sg_init_one(&hdr, &pkt->hdr, sizeof(pkt->hdr));
274 sg_init_one(&buf, pkt->buf, buf_len);
276 ret = virtqueue_add_sgs(vq, sgs, 0, 2, pkt, GFP_KERNEL);
278 virtio_transport_free_pkt(pkt);
282 } while (vq->num_free);
283 if (vsock->rx_buf_nr > vsock->rx_buf_max_nr)
284 vsock->rx_buf_max_nr = vsock->rx_buf_nr;
288 static void virtio_transport_tx_work(struct work_struct *work)
290 struct virtio_vsock *vsock =
291 container_of(work, struct virtio_vsock, tx_work);
292 struct virtqueue *vq;
295 vq = vsock->vqs[VSOCK_VQ_TX];
296 mutex_lock(&vsock->tx_lock);
302 struct virtio_vsock_pkt *pkt;
305 virtqueue_disable_cb(vq);
306 while ((pkt = virtqueue_get_buf(vq, &len)) != NULL) {
307 virtio_transport_free_pkt(pkt);
310 } while (!virtqueue_enable_cb(vq));
313 mutex_unlock(&vsock->tx_lock);
316 queue_work(virtio_vsock_workqueue, &vsock->send_pkt_work);
319 /* Is there space left for replies to rx packets? */
320 static bool virtio_transport_more_replies(struct virtio_vsock *vsock)
322 struct virtqueue *vq = vsock->vqs[VSOCK_VQ_RX];
325 smp_rmb(); /* paired with atomic_inc() and atomic_dec_return() */
326 val = atomic_read(&vsock->queued_replies);
328 return val < virtqueue_get_vring_size(vq);
331 /* event_lock must be held */
332 static int virtio_vsock_event_fill_one(struct virtio_vsock *vsock,
333 struct virtio_vsock_event *event)
335 struct scatterlist sg;
336 struct virtqueue *vq;
338 vq = vsock->vqs[VSOCK_VQ_EVENT];
340 sg_init_one(&sg, event, sizeof(*event));
342 return virtqueue_add_inbuf(vq, &sg, 1, event, GFP_KERNEL);
345 /* event_lock must be held */
346 static void virtio_vsock_event_fill(struct virtio_vsock *vsock)
350 for (i = 0; i < ARRAY_SIZE(vsock->event_list); i++) {
351 struct virtio_vsock_event *event = &vsock->event_list[i];
353 virtio_vsock_event_fill_one(vsock, event);
356 virtqueue_kick(vsock->vqs[VSOCK_VQ_EVENT]);
359 static void virtio_vsock_reset_sock(struct sock *sk)
361 /* vmci_transport.c doesn't take sk_lock here either. At least we're
362 * under vsock_table_lock so the sock cannot disappear while we're
366 sk->sk_state = TCP_CLOSE;
367 sk->sk_err = ECONNRESET;
371 static void virtio_vsock_update_guest_cid(struct virtio_vsock *vsock)
373 struct virtio_device *vdev = vsock->vdev;
376 vdev->config->get(vdev, offsetof(struct virtio_vsock_config, guest_cid),
377 &guest_cid, sizeof(guest_cid));
378 vsock->guest_cid = le64_to_cpu(guest_cid);
381 /* event_lock must be held */
382 static void virtio_vsock_event_handle(struct virtio_vsock *vsock,
383 struct virtio_vsock_event *event)
385 switch (le32_to_cpu(event->id)) {
386 case VIRTIO_VSOCK_EVENT_TRANSPORT_RESET:
387 virtio_vsock_update_guest_cid(vsock);
388 vsock_for_each_connected_socket(&virtio_transport.transport,
389 virtio_vsock_reset_sock);
394 static void virtio_transport_event_work(struct work_struct *work)
396 struct virtio_vsock *vsock =
397 container_of(work, struct virtio_vsock, event_work);
398 struct virtqueue *vq;
400 vq = vsock->vqs[VSOCK_VQ_EVENT];
402 mutex_lock(&vsock->event_lock);
404 if (!vsock->event_run)
408 struct virtio_vsock_event *event;
411 virtqueue_disable_cb(vq);
412 while ((event = virtqueue_get_buf(vq, &len)) != NULL) {
413 if (len == sizeof(*event))
414 virtio_vsock_event_handle(vsock, event);
416 virtio_vsock_event_fill_one(vsock, event);
418 } while (!virtqueue_enable_cb(vq));
420 virtqueue_kick(vsock->vqs[VSOCK_VQ_EVENT]);
422 mutex_unlock(&vsock->event_lock);
425 static void virtio_vsock_event_done(struct virtqueue *vq)
427 struct virtio_vsock *vsock = vq->vdev->priv;
431 queue_work(virtio_vsock_workqueue, &vsock->event_work);
434 static void virtio_vsock_tx_done(struct virtqueue *vq)
436 struct virtio_vsock *vsock = vq->vdev->priv;
440 queue_work(virtio_vsock_workqueue, &vsock->tx_work);
443 static void virtio_vsock_rx_done(struct virtqueue *vq)
445 struct virtio_vsock *vsock = vq->vdev->priv;
449 queue_work(virtio_vsock_workqueue, &vsock->rx_work);
452 static bool virtio_transport_seqpacket_allow(u32 remote_cid);
454 static struct virtio_transport virtio_transport = {
456 .module = THIS_MODULE,
458 .get_local_cid = virtio_transport_get_local_cid,
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 .cancel_pkt = virtio_transport_cancel_pkt,
467 .dgram_bind = virtio_transport_dgram_bind,
468 .dgram_dequeue = virtio_transport_dgram_dequeue,
469 .dgram_enqueue = virtio_transport_dgram_enqueue,
470 .dgram_allow = virtio_transport_dgram_allow,
472 .stream_dequeue = virtio_transport_stream_dequeue,
473 .stream_enqueue = virtio_transport_stream_enqueue,
474 .stream_has_data = virtio_transport_stream_has_data,
475 .stream_has_space = virtio_transport_stream_has_space,
476 .stream_rcvhiwat = virtio_transport_stream_rcvhiwat,
477 .stream_is_active = virtio_transport_stream_is_active,
478 .stream_allow = virtio_transport_stream_allow,
480 .seqpacket_dequeue = virtio_transport_seqpacket_dequeue,
481 .seqpacket_enqueue = virtio_transport_seqpacket_enqueue,
482 .seqpacket_allow = virtio_transport_seqpacket_allow,
483 .seqpacket_has_data = virtio_transport_seqpacket_has_data,
485 .notify_poll_in = virtio_transport_notify_poll_in,
486 .notify_poll_out = virtio_transport_notify_poll_out,
487 .notify_recv_init = virtio_transport_notify_recv_init,
488 .notify_recv_pre_block = virtio_transport_notify_recv_pre_block,
489 .notify_recv_pre_dequeue = virtio_transport_notify_recv_pre_dequeue,
490 .notify_recv_post_dequeue = virtio_transport_notify_recv_post_dequeue,
491 .notify_send_init = virtio_transport_notify_send_init,
492 .notify_send_pre_block = virtio_transport_notify_send_pre_block,
493 .notify_send_pre_enqueue = virtio_transport_notify_send_pre_enqueue,
494 .notify_send_post_enqueue = virtio_transport_notify_send_post_enqueue,
495 .notify_buffer_size = virtio_transport_notify_buffer_size,
498 .send_pkt = virtio_transport_send_pkt,
501 static bool virtio_transport_seqpacket_allow(u32 remote_cid)
503 struct virtio_vsock *vsock;
504 bool seqpacket_allow;
506 seqpacket_allow = false;
508 vsock = rcu_dereference(the_virtio_vsock);
510 seqpacket_allow = vsock->seqpacket_allow;
513 return seqpacket_allow;
516 static void virtio_transport_rx_work(struct work_struct *work)
518 struct virtio_vsock *vsock =
519 container_of(work, struct virtio_vsock, rx_work);
520 struct virtqueue *vq;
522 vq = vsock->vqs[VSOCK_VQ_RX];
524 mutex_lock(&vsock->rx_lock);
530 virtqueue_disable_cb(vq);
532 struct virtio_vsock_pkt *pkt;
535 if (!virtio_transport_more_replies(vsock)) {
536 /* Stop rx until the device processes already
537 * pending replies. Leave rx virtqueue
538 * callbacks disabled.
543 pkt = virtqueue_get_buf(vq, &len);
550 /* Drop short/long packets */
551 if (unlikely(len < sizeof(pkt->hdr) ||
552 len > sizeof(pkt->hdr) + pkt->len)) {
553 virtio_transport_free_pkt(pkt);
557 pkt->len = len - sizeof(pkt->hdr);
558 virtio_transport_deliver_tap_pkt(pkt);
559 virtio_transport_recv_pkt(&virtio_transport, pkt);
561 } while (!virtqueue_enable_cb(vq));
564 if (vsock->rx_buf_nr < vsock->rx_buf_max_nr / 2)
565 virtio_vsock_rx_fill(vsock);
566 mutex_unlock(&vsock->rx_lock);
569 static int virtio_vsock_vqs_init(struct virtio_vsock *vsock)
571 struct virtio_device *vdev = vsock->vdev;
572 static const char * const names[] = {
577 vq_callback_t *callbacks[] = {
578 virtio_vsock_rx_done,
579 virtio_vsock_tx_done,
580 virtio_vsock_event_done,
584 ret = virtio_find_vqs(vdev, VSOCK_VQ_MAX, vsock->vqs, callbacks, names,
589 virtio_vsock_update_guest_cid(vsock);
591 virtio_device_ready(vdev);
593 mutex_lock(&vsock->tx_lock);
594 vsock->tx_run = true;
595 mutex_unlock(&vsock->tx_lock);
597 mutex_lock(&vsock->rx_lock);
598 virtio_vsock_rx_fill(vsock);
599 vsock->rx_run = true;
600 mutex_unlock(&vsock->rx_lock);
602 mutex_lock(&vsock->event_lock);
603 virtio_vsock_event_fill(vsock);
604 vsock->event_run = true;
605 mutex_unlock(&vsock->event_lock);
610 static void virtio_vsock_vqs_del(struct virtio_vsock *vsock)
612 struct virtio_device *vdev = vsock->vdev;
613 struct virtio_vsock_pkt *pkt;
615 /* Reset all connected sockets when the VQs disappear */
616 vsock_for_each_connected_socket(&virtio_transport.transport,
617 virtio_vsock_reset_sock);
619 /* Stop all work handlers to make sure no one is accessing the device,
620 * so we can safely call virtio_reset_device().
622 mutex_lock(&vsock->rx_lock);
623 vsock->rx_run = false;
624 mutex_unlock(&vsock->rx_lock);
626 mutex_lock(&vsock->tx_lock);
627 vsock->tx_run = false;
628 mutex_unlock(&vsock->tx_lock);
630 mutex_lock(&vsock->event_lock);
631 vsock->event_run = false;
632 mutex_unlock(&vsock->event_lock);
634 /* Flush all device writes and interrupts, device will not use any
637 virtio_reset_device(vdev);
639 mutex_lock(&vsock->rx_lock);
640 while ((pkt = virtqueue_detach_unused_buf(vsock->vqs[VSOCK_VQ_RX])))
641 virtio_transport_free_pkt(pkt);
642 mutex_unlock(&vsock->rx_lock);
644 mutex_lock(&vsock->tx_lock);
645 while ((pkt = virtqueue_detach_unused_buf(vsock->vqs[VSOCK_VQ_TX])))
646 virtio_transport_free_pkt(pkt);
647 mutex_unlock(&vsock->tx_lock);
649 spin_lock_bh(&vsock->send_pkt_list_lock);
650 while (!list_empty(&vsock->send_pkt_list)) {
651 pkt = list_first_entry(&vsock->send_pkt_list,
652 struct virtio_vsock_pkt, list);
653 list_del(&pkt->list);
654 virtio_transport_free_pkt(pkt);
656 spin_unlock_bh(&vsock->send_pkt_list_lock);
658 /* Delete virtqueues and flush outstanding callbacks if any */
659 vdev->config->del_vqs(vdev);
662 static int virtio_vsock_probe(struct virtio_device *vdev)
664 struct virtio_vsock *vsock = NULL;
667 ret = mutex_lock_interruptible(&the_virtio_vsock_mutex);
671 /* Only one virtio-vsock device per guest is supported */
672 if (rcu_dereference_protected(the_virtio_vsock,
673 lockdep_is_held(&the_virtio_vsock_mutex))) {
678 vsock = kzalloc(sizeof(*vsock), GFP_KERNEL);
686 vsock->rx_buf_nr = 0;
687 vsock->rx_buf_max_nr = 0;
688 atomic_set(&vsock->queued_replies, 0);
690 mutex_init(&vsock->tx_lock);
691 mutex_init(&vsock->rx_lock);
692 mutex_init(&vsock->event_lock);
693 spin_lock_init(&vsock->send_pkt_list_lock);
694 INIT_LIST_HEAD(&vsock->send_pkt_list);
695 INIT_WORK(&vsock->rx_work, virtio_transport_rx_work);
696 INIT_WORK(&vsock->tx_work, virtio_transport_tx_work);
697 INIT_WORK(&vsock->event_work, virtio_transport_event_work);
698 INIT_WORK(&vsock->send_pkt_work, virtio_transport_send_pkt_work);
700 if (virtio_has_feature(vdev, VIRTIO_VSOCK_F_SEQPACKET))
701 vsock->seqpacket_allow = true;
705 ret = virtio_vsock_vqs_init(vsock);
709 rcu_assign_pointer(the_virtio_vsock, vsock);
711 mutex_unlock(&the_virtio_vsock_mutex);
717 mutex_unlock(&the_virtio_vsock_mutex);
721 static void virtio_vsock_remove(struct virtio_device *vdev)
723 struct virtio_vsock *vsock = vdev->priv;
725 mutex_lock(&the_virtio_vsock_mutex);
728 rcu_assign_pointer(the_virtio_vsock, NULL);
731 virtio_vsock_vqs_del(vsock);
733 /* Other works can be queued before 'config->del_vqs()', so we flush
734 * all works before to free the vsock object to avoid use after free.
736 flush_work(&vsock->rx_work);
737 flush_work(&vsock->tx_work);
738 flush_work(&vsock->event_work);
739 flush_work(&vsock->send_pkt_work);
741 mutex_unlock(&the_virtio_vsock_mutex);
746 #ifdef CONFIG_PM_SLEEP
747 static int virtio_vsock_freeze(struct virtio_device *vdev)
749 struct virtio_vsock *vsock = vdev->priv;
751 mutex_lock(&the_virtio_vsock_mutex);
753 rcu_assign_pointer(the_virtio_vsock, NULL);
756 virtio_vsock_vqs_del(vsock);
758 mutex_unlock(&the_virtio_vsock_mutex);
763 static int virtio_vsock_restore(struct virtio_device *vdev)
765 struct virtio_vsock *vsock = vdev->priv;
768 mutex_lock(&the_virtio_vsock_mutex);
770 /* Only one virtio-vsock device per guest is supported */
771 if (rcu_dereference_protected(the_virtio_vsock,
772 lockdep_is_held(&the_virtio_vsock_mutex))) {
777 ret = virtio_vsock_vqs_init(vsock);
781 rcu_assign_pointer(the_virtio_vsock, vsock);
784 mutex_unlock(&the_virtio_vsock_mutex);
787 #endif /* CONFIG_PM_SLEEP */
789 static struct virtio_device_id id_table[] = {
790 { VIRTIO_ID_VSOCK, VIRTIO_DEV_ANY_ID },
794 static unsigned int features[] = {
795 VIRTIO_VSOCK_F_SEQPACKET
798 static struct virtio_driver virtio_vsock_driver = {
799 .feature_table = features,
800 .feature_table_size = ARRAY_SIZE(features),
801 .driver.name = KBUILD_MODNAME,
802 .driver.owner = THIS_MODULE,
803 .id_table = id_table,
804 .probe = virtio_vsock_probe,
805 .remove = virtio_vsock_remove,
806 #ifdef CONFIG_PM_SLEEP
807 .freeze = virtio_vsock_freeze,
808 .restore = virtio_vsock_restore,
812 static int __init virtio_vsock_init(void)
816 virtio_vsock_workqueue = alloc_workqueue("virtio_vsock", 0, 0);
817 if (!virtio_vsock_workqueue)
820 ret = vsock_core_register(&virtio_transport.transport,
821 VSOCK_TRANSPORT_F_G2H);
825 ret = register_virtio_driver(&virtio_vsock_driver);
832 vsock_core_unregister(&virtio_transport.transport);
834 destroy_workqueue(virtio_vsock_workqueue);
838 static void __exit virtio_vsock_exit(void)
840 unregister_virtio_driver(&virtio_vsock_driver);
841 vsock_core_unregister(&virtio_transport.transport);
842 destroy_workqueue(virtio_vsock_workqueue);
845 module_init(virtio_vsock_init);
846 module_exit(virtio_vsock_exit);
847 MODULE_LICENSE("GPL v2");
848 MODULE_AUTHOR("Asias He");
849 MODULE_DESCRIPTION("virtio transport for vsock");
850 MODULE_DEVICE_TABLE(virtio, id_table);