2 * vhost transport for vsock
4 * Copyright (C) 2013-2015 Red Hat, Inc.
8 * This work is licensed under the terms of the GNU GPL, version 2.
10 #include <linux/miscdevice.h>
11 #include <linux/atomic.h>
12 #include <linux/module.h>
13 #include <linux/mutex.h>
14 #include <linux/vmalloc.h>
16 #include <linux/virtio_vsock.h>
17 #include <linux/vhost.h>
19 #include <net/af_vsock.h>
22 #define VHOST_VSOCK_DEFAULT_HOST_CID 2
25 VHOST_VSOCK_FEATURES = VHOST_FEATURES,
28 /* Used to track all the vhost_vsock instances on the system. */
29 static DEFINE_SPINLOCK(vhost_vsock_lock);
30 static LIST_HEAD(vhost_vsock_list);
34 struct vhost_virtqueue vqs[2];
36 /* Link to global vhost_vsock_list, protected by vhost_vsock_lock */
37 struct list_head list;
39 struct vhost_work send_pkt_work;
40 spinlock_t send_pkt_list_lock;
41 struct list_head send_pkt_list; /* host->guest pending packets */
43 atomic_t queued_replies;
48 static u32 vhost_transport_get_local_cid(void)
50 return VHOST_VSOCK_DEFAULT_HOST_CID;
53 static struct vhost_vsock *__vhost_vsock_get(u32 guest_cid)
55 struct vhost_vsock *vsock;
57 list_for_each_entry(vsock, &vhost_vsock_list, list) {
58 u32 other_cid = vsock->guest_cid;
60 /* Skip instances that have no CID yet */
64 if (other_cid == guest_cid) {
72 static struct vhost_vsock *vhost_vsock_get(u32 guest_cid)
74 struct vhost_vsock *vsock;
76 spin_lock_bh(&vhost_vsock_lock);
77 vsock = __vhost_vsock_get(guest_cid);
78 spin_unlock_bh(&vhost_vsock_lock);
84 vhost_transport_do_send_pkt(struct vhost_vsock *vsock,
85 struct vhost_virtqueue *vq)
87 struct vhost_virtqueue *tx_vq = &vsock->vqs[VSOCK_VQ_TX];
89 bool restart_tx = false;
91 mutex_lock(&vq->mutex);
93 if (!vq->private_data)
96 /* Avoid further vmexits, we're already processing the virtqueue */
97 vhost_disable_notify(&vsock->dev, vq);
100 struct virtio_vsock_pkt *pkt;
101 struct iov_iter iov_iter;
107 spin_lock_bh(&vsock->send_pkt_list_lock);
108 if (list_empty(&vsock->send_pkt_list)) {
109 spin_unlock_bh(&vsock->send_pkt_list_lock);
110 vhost_enable_notify(&vsock->dev, vq);
114 pkt = list_first_entry(&vsock->send_pkt_list,
115 struct virtio_vsock_pkt, list);
116 list_del_init(&pkt->list);
117 spin_unlock_bh(&vsock->send_pkt_list_lock);
119 head = vhost_get_vq_desc(vq, vq->iov, ARRAY_SIZE(vq->iov),
120 &out, &in, NULL, NULL);
122 spin_lock_bh(&vsock->send_pkt_list_lock);
123 list_add(&pkt->list, &vsock->send_pkt_list);
124 spin_unlock_bh(&vsock->send_pkt_list_lock);
128 if (head == vq->num) {
129 spin_lock_bh(&vsock->send_pkt_list_lock);
130 list_add(&pkt->list, &vsock->send_pkt_list);
131 spin_unlock_bh(&vsock->send_pkt_list_lock);
133 /* We cannot finish yet if more buffers snuck in while
134 * re-enabling notify.
136 if (unlikely(vhost_enable_notify(&vsock->dev, vq))) {
137 vhost_disable_notify(&vsock->dev, vq);
144 virtio_transport_free_pkt(pkt);
145 vq_err(vq, "Expected 0 output buffers, got %u\n", out);
149 len = iov_length(&vq->iov[out], in);
150 iov_iter_init(&iov_iter, READ, &vq->iov[out], in, len);
152 nbytes = copy_to_iter(&pkt->hdr, sizeof(pkt->hdr), &iov_iter);
153 if (nbytes != sizeof(pkt->hdr)) {
154 virtio_transport_free_pkt(pkt);
155 vq_err(vq, "Faulted on copying pkt hdr\n");
159 nbytes = copy_to_iter(pkt->buf, pkt->len, &iov_iter);
160 if (nbytes != pkt->len) {
161 virtio_transport_free_pkt(pkt);
162 vq_err(vq, "Faulted on copying pkt buf\n");
166 vhost_add_used(vq, head, sizeof(pkt->hdr) + pkt->len);
172 val = atomic_dec_return(&vsock->queued_replies);
174 /* Do we have resources to resume tx processing? */
175 if (val + 1 == tx_vq->num)
179 virtio_transport_free_pkt(pkt);
182 vhost_signal(&vsock->dev, vq);
185 mutex_unlock(&vq->mutex);
188 vhost_poll_queue(&tx_vq->poll);
191 static void vhost_transport_send_pkt_work(struct vhost_work *work)
193 struct vhost_virtqueue *vq;
194 struct vhost_vsock *vsock;
196 vsock = container_of(work, struct vhost_vsock, send_pkt_work);
197 vq = &vsock->vqs[VSOCK_VQ_RX];
199 vhost_transport_do_send_pkt(vsock, vq);
203 vhost_transport_send_pkt(struct virtio_vsock_pkt *pkt)
205 struct vhost_vsock *vsock;
208 /* Find the vhost_vsock according to guest context id */
209 vsock = vhost_vsock_get(le64_to_cpu(pkt->hdr.dst_cid));
211 virtio_transport_free_pkt(pkt);
216 atomic_inc(&vsock->queued_replies);
218 spin_lock_bh(&vsock->send_pkt_list_lock);
219 list_add_tail(&pkt->list, &vsock->send_pkt_list);
220 spin_unlock_bh(&vsock->send_pkt_list_lock);
222 vhost_work_queue(&vsock->dev, &vsock->send_pkt_work);
227 vhost_transport_cancel_pkt(struct vsock_sock *vsk)
229 struct vhost_vsock *vsock;
230 struct virtio_vsock_pkt *pkt, *n;
234 /* Find the vhost_vsock according to guest context id */
235 vsock = vhost_vsock_get(vsk->remote_addr.svm_cid);
239 spin_lock_bh(&vsock->send_pkt_list_lock);
240 list_for_each_entry_safe(pkt, n, &vsock->send_pkt_list, list) {
243 list_move(&pkt->list, &freeme);
245 spin_unlock_bh(&vsock->send_pkt_list_lock);
247 list_for_each_entry_safe(pkt, n, &freeme, list) {
250 list_del(&pkt->list);
251 virtio_transport_free_pkt(pkt);
255 struct vhost_virtqueue *tx_vq = &vsock->vqs[VSOCK_VQ_TX];
258 new_cnt = atomic_sub_return(cnt, &vsock->queued_replies);
259 if (new_cnt + cnt >= tx_vq->num && new_cnt < tx_vq->num)
260 vhost_poll_queue(&tx_vq->poll);
266 static struct virtio_vsock_pkt *
267 vhost_vsock_alloc_pkt(struct vhost_virtqueue *vq,
268 unsigned int out, unsigned int in)
270 struct virtio_vsock_pkt *pkt;
271 struct iov_iter iov_iter;
276 vq_err(vq, "Expected 0 input buffers, got %u\n", in);
280 pkt = kzalloc(sizeof(*pkt), GFP_KERNEL);
284 len = iov_length(vq->iov, out);
285 iov_iter_init(&iov_iter, WRITE, vq->iov, out, len);
287 nbytes = copy_from_iter(&pkt->hdr, sizeof(pkt->hdr), &iov_iter);
288 if (nbytes != sizeof(pkt->hdr)) {
289 vq_err(vq, "Expected %zu bytes for pkt->hdr, got %zu bytes\n",
290 sizeof(pkt->hdr), nbytes);
295 if (le16_to_cpu(pkt->hdr.type) == VIRTIO_VSOCK_TYPE_STREAM)
296 pkt->len = le32_to_cpu(pkt->hdr.len);
302 /* The pkt is too big */
303 if (pkt->len > VIRTIO_VSOCK_MAX_PKT_BUF_SIZE) {
308 pkt->buf = kmalloc(pkt->len, GFP_KERNEL);
314 nbytes = copy_from_iter(pkt->buf, pkt->len, &iov_iter);
315 if (nbytes != pkt->len) {
316 vq_err(vq, "Expected %u byte payload, got %zu bytes\n",
318 virtio_transport_free_pkt(pkt);
325 /* Is there space left for replies to rx packets? */
326 static bool vhost_vsock_more_replies(struct vhost_vsock *vsock)
328 struct vhost_virtqueue *vq = &vsock->vqs[VSOCK_VQ_TX];
331 smp_rmb(); /* paired with atomic_inc() and atomic_dec_return() */
332 val = atomic_read(&vsock->queued_replies);
334 return val < vq->num;
337 static void vhost_vsock_handle_tx_kick(struct vhost_work *work)
339 struct vhost_virtqueue *vq = container_of(work, struct vhost_virtqueue,
341 struct vhost_vsock *vsock = container_of(vq->dev, struct vhost_vsock,
343 struct virtio_vsock_pkt *pkt;
345 unsigned int out, in;
348 mutex_lock(&vq->mutex);
350 if (!vq->private_data)
353 vhost_disable_notify(&vsock->dev, vq);
357 if (!vhost_vsock_more_replies(vsock)) {
358 /* Stop tx until the device processes already
359 * pending replies. Leave tx virtqueue
360 * callbacks disabled.
362 goto no_more_replies;
365 head = vhost_get_vq_desc(vq, vq->iov, ARRAY_SIZE(vq->iov),
366 &out, &in, NULL, NULL);
370 if (head == vq->num) {
371 if (unlikely(vhost_enable_notify(&vsock->dev, vq))) {
372 vhost_disable_notify(&vsock->dev, vq);
378 pkt = vhost_vsock_alloc_pkt(vq, out, in);
380 vq_err(vq, "Faulted on pkt\n");
386 /* Only accept correctly addressed packets */
387 if (le64_to_cpu(pkt->hdr.src_cid) == vsock->guest_cid)
388 virtio_transport_recv_pkt(pkt);
390 virtio_transport_free_pkt(pkt);
392 vhost_add_used(vq, head, sizeof(pkt->hdr) + len);
398 vhost_signal(&vsock->dev, vq);
401 mutex_unlock(&vq->mutex);
404 static void vhost_vsock_handle_rx_kick(struct vhost_work *work)
406 struct vhost_virtqueue *vq = container_of(work, struct vhost_virtqueue,
408 struct vhost_vsock *vsock = container_of(vq->dev, struct vhost_vsock,
411 vhost_transport_do_send_pkt(vsock, vq);
414 static int vhost_vsock_start(struct vhost_vsock *vsock)
416 struct vhost_virtqueue *vq;
420 mutex_lock(&vsock->dev.mutex);
422 ret = vhost_dev_check_owner(&vsock->dev);
426 for (i = 0; i < ARRAY_SIZE(vsock->vqs); i++) {
429 mutex_lock(&vq->mutex);
431 if (!vhost_vq_access_ok(vq)) {
436 if (!vq->private_data) {
437 vq->private_data = vsock;
438 ret = vhost_vq_init_access(vq);
443 mutex_unlock(&vq->mutex);
446 mutex_unlock(&vsock->dev.mutex);
450 vq->private_data = NULL;
451 mutex_unlock(&vq->mutex);
453 for (i = 0; i < ARRAY_SIZE(vsock->vqs); i++) {
456 mutex_lock(&vq->mutex);
457 vq->private_data = NULL;
458 mutex_unlock(&vq->mutex);
461 mutex_unlock(&vsock->dev.mutex);
465 static int vhost_vsock_stop(struct vhost_vsock *vsock)
470 mutex_lock(&vsock->dev.mutex);
472 ret = vhost_dev_check_owner(&vsock->dev);
476 for (i = 0; i < ARRAY_SIZE(vsock->vqs); i++) {
477 struct vhost_virtqueue *vq = &vsock->vqs[i];
479 mutex_lock(&vq->mutex);
480 vq->private_data = NULL;
481 mutex_unlock(&vq->mutex);
485 mutex_unlock(&vsock->dev.mutex);
489 static void vhost_vsock_free(struct vhost_vsock *vsock)
494 static int vhost_vsock_dev_open(struct inode *inode, struct file *file)
496 struct vhost_virtqueue **vqs;
497 struct vhost_vsock *vsock;
500 /* This struct is large and allocation could fail, fall back to vmalloc
501 * if there is no other way.
503 vsock = kzalloc(sizeof(*vsock), GFP_KERNEL | __GFP_NOWARN | __GFP_REPEAT);
505 vsock = vmalloc(sizeof(*vsock));
510 vqs = kmalloc_array(ARRAY_SIZE(vsock->vqs), sizeof(*vqs), GFP_KERNEL);
516 atomic_set(&vsock->queued_replies, 0);
518 vqs[VSOCK_VQ_TX] = &vsock->vqs[VSOCK_VQ_TX];
519 vqs[VSOCK_VQ_RX] = &vsock->vqs[VSOCK_VQ_RX];
520 vsock->vqs[VSOCK_VQ_TX].handle_kick = vhost_vsock_handle_tx_kick;
521 vsock->vqs[VSOCK_VQ_RX].handle_kick = vhost_vsock_handle_rx_kick;
523 vhost_dev_init(&vsock->dev, vqs, ARRAY_SIZE(vsock->vqs));
525 file->private_data = vsock;
526 spin_lock_init(&vsock->send_pkt_list_lock);
527 INIT_LIST_HEAD(&vsock->send_pkt_list);
528 vhost_work_init(&vsock->send_pkt_work, vhost_transport_send_pkt_work);
530 spin_lock_bh(&vhost_vsock_lock);
531 list_add_tail(&vsock->list, &vhost_vsock_list);
532 spin_unlock_bh(&vhost_vsock_lock);
536 vhost_vsock_free(vsock);
540 static void vhost_vsock_flush(struct vhost_vsock *vsock)
544 for (i = 0; i < ARRAY_SIZE(vsock->vqs); i++)
545 if (vsock->vqs[i].handle_kick)
546 vhost_poll_flush(&vsock->vqs[i].poll);
547 vhost_work_flush(&vsock->dev, &vsock->send_pkt_work);
550 static void vhost_vsock_reset_orphans(struct sock *sk)
552 struct vsock_sock *vsk = vsock_sk(sk);
554 /* vmci_transport.c doesn't take sk_lock here either. At least we're
555 * under vsock_table_lock so the sock cannot disappear while we're
559 if (!vhost_vsock_get(vsk->remote_addr.svm_cid)) {
560 sock_set_flag(sk, SOCK_DONE);
561 vsk->peer_shutdown = SHUTDOWN_MASK;
562 sk->sk_state = SS_UNCONNECTED;
563 sk->sk_err = ECONNRESET;
564 sk->sk_error_report(sk);
568 static int vhost_vsock_dev_release(struct inode *inode, struct file *file)
570 struct vhost_vsock *vsock = file->private_data;
572 spin_lock_bh(&vhost_vsock_lock);
573 list_del(&vsock->list);
574 spin_unlock_bh(&vhost_vsock_lock);
576 /* Iterating over all connections for all CIDs to find orphans is
577 * inefficient. Room for improvement here. */
578 vsock_for_each_connected_socket(vhost_vsock_reset_orphans);
580 vhost_vsock_stop(vsock);
581 vhost_vsock_flush(vsock);
582 vhost_dev_stop(&vsock->dev);
584 spin_lock_bh(&vsock->send_pkt_list_lock);
585 while (!list_empty(&vsock->send_pkt_list)) {
586 struct virtio_vsock_pkt *pkt;
588 pkt = list_first_entry(&vsock->send_pkt_list,
589 struct virtio_vsock_pkt, list);
590 list_del_init(&pkt->list);
591 virtio_transport_free_pkt(pkt);
593 spin_unlock_bh(&vsock->send_pkt_list_lock);
595 vhost_dev_cleanup(&vsock->dev, false);
596 kfree(vsock->dev.vqs);
597 vhost_vsock_free(vsock);
601 static int vhost_vsock_set_cid(struct vhost_vsock *vsock, u64 guest_cid)
603 struct vhost_vsock *other;
605 /* Refuse reserved CIDs */
606 if (guest_cid <= VMADDR_CID_HOST ||
607 guest_cid == U32_MAX)
610 /* 64-bit CIDs are not yet supported */
611 if (guest_cid > U32_MAX)
614 /* Refuse if CID is already in use */
615 spin_lock_bh(&vhost_vsock_lock);
616 other = __vhost_vsock_get(guest_cid);
617 if (other && other != vsock) {
618 spin_unlock_bh(&vhost_vsock_lock);
621 vsock->guest_cid = guest_cid;
622 spin_unlock_bh(&vhost_vsock_lock);
627 static int vhost_vsock_set_features(struct vhost_vsock *vsock, u64 features)
629 struct vhost_virtqueue *vq;
632 if (features & ~VHOST_VSOCK_FEATURES)
635 mutex_lock(&vsock->dev.mutex);
636 if ((features & (1 << VHOST_F_LOG_ALL)) &&
637 !vhost_log_access_ok(&vsock->dev)) {
638 mutex_unlock(&vsock->dev.mutex);
642 for (i = 0; i < ARRAY_SIZE(vsock->vqs); i++) {
644 mutex_lock(&vq->mutex);
645 vq->acked_features = features;
646 mutex_unlock(&vq->mutex);
648 mutex_unlock(&vsock->dev.mutex);
652 static long vhost_vsock_dev_ioctl(struct file *f, unsigned int ioctl,
655 struct vhost_vsock *vsock = f->private_data;
656 void __user *argp = (void __user *)arg;
663 case VHOST_VSOCK_SET_GUEST_CID:
664 if (copy_from_user(&guest_cid, argp, sizeof(guest_cid)))
666 return vhost_vsock_set_cid(vsock, guest_cid);
667 case VHOST_VSOCK_SET_RUNNING:
668 if (copy_from_user(&start, argp, sizeof(start)))
671 return vhost_vsock_start(vsock);
673 return vhost_vsock_stop(vsock);
674 case VHOST_GET_FEATURES:
675 features = VHOST_VSOCK_FEATURES;
676 if (copy_to_user(argp, &features, sizeof(features)))
679 case VHOST_SET_FEATURES:
680 if (copy_from_user(&features, argp, sizeof(features)))
682 return vhost_vsock_set_features(vsock, features);
684 mutex_lock(&vsock->dev.mutex);
685 r = vhost_dev_ioctl(&vsock->dev, ioctl, argp);
686 if (r == -ENOIOCTLCMD)
687 r = vhost_vring_ioctl(&vsock->dev, ioctl, argp);
689 vhost_vsock_flush(vsock);
690 mutex_unlock(&vsock->dev.mutex);
695 static const struct file_operations vhost_vsock_fops = {
696 .owner = THIS_MODULE,
697 .open = vhost_vsock_dev_open,
698 .release = vhost_vsock_dev_release,
699 .llseek = noop_llseek,
700 .unlocked_ioctl = vhost_vsock_dev_ioctl,
703 static struct miscdevice vhost_vsock_misc = {
704 .minor = MISC_DYNAMIC_MINOR,
705 .name = "vhost-vsock",
706 .fops = &vhost_vsock_fops,
709 static struct virtio_transport vhost_transport = {
711 .get_local_cid = vhost_transport_get_local_cid,
713 .init = virtio_transport_do_socket_init,
714 .destruct = virtio_transport_destruct,
715 .release = virtio_transport_release,
716 .connect = virtio_transport_connect,
717 .shutdown = virtio_transport_shutdown,
718 .cancel_pkt = vhost_transport_cancel_pkt,
720 .dgram_enqueue = virtio_transport_dgram_enqueue,
721 .dgram_dequeue = virtio_transport_dgram_dequeue,
722 .dgram_bind = virtio_transport_dgram_bind,
723 .dgram_allow = virtio_transport_dgram_allow,
725 .stream_enqueue = virtio_transport_stream_enqueue,
726 .stream_dequeue = virtio_transport_stream_dequeue,
727 .stream_has_data = virtio_transport_stream_has_data,
728 .stream_has_space = virtio_transport_stream_has_space,
729 .stream_rcvhiwat = virtio_transport_stream_rcvhiwat,
730 .stream_is_active = virtio_transport_stream_is_active,
731 .stream_allow = virtio_transport_stream_allow,
733 .notify_poll_in = virtio_transport_notify_poll_in,
734 .notify_poll_out = virtio_transport_notify_poll_out,
735 .notify_recv_init = virtio_transport_notify_recv_init,
736 .notify_recv_pre_block = virtio_transport_notify_recv_pre_block,
737 .notify_recv_pre_dequeue = virtio_transport_notify_recv_pre_dequeue,
738 .notify_recv_post_dequeue = virtio_transport_notify_recv_post_dequeue,
739 .notify_send_init = virtio_transport_notify_send_init,
740 .notify_send_pre_block = virtio_transport_notify_send_pre_block,
741 .notify_send_pre_enqueue = virtio_transport_notify_send_pre_enqueue,
742 .notify_send_post_enqueue = virtio_transport_notify_send_post_enqueue,
744 .set_buffer_size = virtio_transport_set_buffer_size,
745 .set_min_buffer_size = virtio_transport_set_min_buffer_size,
746 .set_max_buffer_size = virtio_transport_set_max_buffer_size,
747 .get_buffer_size = virtio_transport_get_buffer_size,
748 .get_min_buffer_size = virtio_transport_get_min_buffer_size,
749 .get_max_buffer_size = virtio_transport_get_max_buffer_size,
752 .send_pkt = vhost_transport_send_pkt,
755 static int __init vhost_vsock_init(void)
759 ret = vsock_core_init(&vhost_transport.transport);
762 return misc_register(&vhost_vsock_misc);
765 static void __exit vhost_vsock_exit(void)
767 misc_deregister(&vhost_vsock_misc);
771 module_init(vhost_vsock_init);
772 module_exit(vhost_vsock_exit);
773 MODULE_LICENSE("GPL v2");
774 MODULE_AUTHOR("Asias He");
775 MODULE_DESCRIPTION("vhost transport for vsock ");