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 */
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 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];
67 static u32
virtio_transport_get_local_cid(void)
69 struct virtio_vsock
*vsock
;
73 vsock
= rcu_dereference(the_virtio_vsock
);
79 ret
= vsock
->guest_cid
;
86 virtio_transport_send_pkt_work(struct work_struct
*work
)
88 struct virtio_vsock
*vsock
=
89 container_of(work
, struct virtio_vsock
, send_pkt_work
);
92 bool restart_rx
= false;
94 mutex_lock(&vsock
->tx_lock
);
99 vq
= vsock
->vqs
[VSOCK_VQ_TX
];
102 struct virtio_vsock_pkt
*pkt
;
103 struct scatterlist hdr
, buf
, *sgs
[2];
104 int ret
, in_sg
= 0, out_sg
= 0;
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
);
113 pkt
= list_first_entry(&vsock
->send_pkt_list
,
114 struct virtio_vsock_pkt
, list
);
115 list_del_init(&pkt
->list
);
116 spin_unlock_bh(&vsock
->send_pkt_list_lock
);
118 virtio_transport_deliver_tap_pkt(pkt
);
122 sg_init_one(&hdr
, &pkt
->hdr
, sizeof(pkt
->hdr
));
123 sgs
[out_sg
++] = &hdr
;
125 sg_init_one(&buf
, pkt
->buf
, pkt
->len
);
126 sgs
[out_sg
++] = &buf
;
129 ret
= virtqueue_add_sgs(vq
, sgs
, out_sg
, in_sg
, pkt
, GFP_KERNEL
);
130 /* Usually this means that there is no more space available in
134 spin_lock_bh(&vsock
->send_pkt_list_lock
);
135 list_add(&pkt
->list
, &vsock
->send_pkt_list
);
136 spin_unlock_bh(&vsock
->send_pkt_list_lock
);
141 struct virtqueue
*rx_vq
= vsock
->vqs
[VSOCK_VQ_RX
];
144 val
= atomic_dec_return(&vsock
->queued_replies
);
146 /* Do we now have resources to resume rx processing? */
147 if (val
+ 1 == virtqueue_get_vring_size(rx_vq
))
158 mutex_unlock(&vsock
->tx_lock
);
161 queue_work(virtio_vsock_workqueue
, &vsock
->rx_work
);
165 virtio_transport_send_pkt(struct virtio_vsock_pkt
*pkt
)
167 struct virtio_vsock
*vsock
;
171 vsock
= rcu_dereference(the_virtio_vsock
);
173 virtio_transport_free_pkt(pkt
);
178 if (le64_to_cpu(pkt
->hdr
.dst_cid
) == vsock
->guest_cid
) {
179 virtio_transport_free_pkt(pkt
);
185 atomic_inc(&vsock
->queued_replies
);
187 spin_lock_bh(&vsock
->send_pkt_list_lock
);
188 list_add_tail(&pkt
->list
, &vsock
->send_pkt_list
);
189 spin_unlock_bh(&vsock
->send_pkt_list_lock
);
191 queue_work(virtio_vsock_workqueue
, &vsock
->send_pkt_work
);
199 virtio_transport_cancel_pkt(struct vsock_sock
*vsk
)
201 struct virtio_vsock
*vsock
;
202 struct virtio_vsock_pkt
*pkt
, *n
;
207 vsock
= rcu_dereference(the_virtio_vsock
);
213 spin_lock_bh(&vsock
->send_pkt_list_lock
);
214 list_for_each_entry_safe(pkt
, n
, &vsock
->send_pkt_list
, list
) {
217 list_move(&pkt
->list
, &freeme
);
219 spin_unlock_bh(&vsock
->send_pkt_list_lock
);
221 list_for_each_entry_safe(pkt
, n
, &freeme
, list
) {
224 list_del(&pkt
->list
);
225 virtio_transport_free_pkt(pkt
);
229 struct virtqueue
*rx_vq
= vsock
->vqs
[VSOCK_VQ_RX
];
232 new_cnt
= atomic_sub_return(cnt
, &vsock
->queued_replies
);
233 if (new_cnt
+ cnt
>= virtqueue_get_vring_size(rx_vq
) &&
234 new_cnt
< virtqueue_get_vring_size(rx_vq
))
235 queue_work(virtio_vsock_workqueue
, &vsock
->rx_work
);
245 static void virtio_vsock_rx_fill(struct virtio_vsock
*vsock
)
247 int buf_len
= VIRTIO_VSOCK_DEFAULT_RX_BUF_SIZE
;
248 struct virtio_vsock_pkt
*pkt
;
249 struct scatterlist hdr
, buf
, *sgs
[2];
250 struct virtqueue
*vq
;
253 vq
= vsock
->vqs
[VSOCK_VQ_RX
];
256 pkt
= kzalloc(sizeof(*pkt
), GFP_KERNEL
);
260 pkt
->buf
= kmalloc(buf_len
, GFP_KERNEL
);
262 virtio_transport_free_pkt(pkt
);
266 pkt
->buf_len
= buf_len
;
269 sg_init_one(&hdr
, &pkt
->hdr
, sizeof(pkt
->hdr
));
272 sg_init_one(&buf
, pkt
->buf
, buf_len
);
274 ret
= virtqueue_add_sgs(vq
, sgs
, 0, 2, pkt
, GFP_KERNEL
);
276 virtio_transport_free_pkt(pkt
);
280 } while (vq
->num_free
);
281 if (vsock
->rx_buf_nr
> vsock
->rx_buf_max_nr
)
282 vsock
->rx_buf_max_nr
= vsock
->rx_buf_nr
;
286 static void virtio_transport_tx_work(struct work_struct
*work
)
288 struct virtio_vsock
*vsock
=
289 container_of(work
, struct virtio_vsock
, tx_work
);
290 struct virtqueue
*vq
;
293 vq
= vsock
->vqs
[VSOCK_VQ_TX
];
294 mutex_lock(&vsock
->tx_lock
);
300 struct virtio_vsock_pkt
*pkt
;
303 virtqueue_disable_cb(vq
);
304 while ((pkt
= virtqueue_get_buf(vq
, &len
)) != NULL
) {
305 virtio_transport_free_pkt(pkt
);
308 } while (!virtqueue_enable_cb(vq
));
311 mutex_unlock(&vsock
->tx_lock
);
314 queue_work(virtio_vsock_workqueue
, &vsock
->send_pkt_work
);
317 /* Is there space left for replies to rx packets? */
318 static bool virtio_transport_more_replies(struct virtio_vsock
*vsock
)
320 struct virtqueue
*vq
= vsock
->vqs
[VSOCK_VQ_RX
];
323 smp_rmb(); /* paired with atomic_inc() and atomic_dec_return() */
324 val
= atomic_read(&vsock
->queued_replies
);
326 return val
< virtqueue_get_vring_size(vq
);
329 /* event_lock must be held */
330 static int virtio_vsock_event_fill_one(struct virtio_vsock
*vsock
,
331 struct virtio_vsock_event
*event
)
333 struct scatterlist sg
;
334 struct virtqueue
*vq
;
336 vq
= vsock
->vqs
[VSOCK_VQ_EVENT
];
338 sg_init_one(&sg
, event
, sizeof(*event
));
340 return virtqueue_add_inbuf(vq
, &sg
, 1, event
, GFP_KERNEL
);
343 /* event_lock must be held */
344 static void virtio_vsock_event_fill(struct virtio_vsock
*vsock
)
348 for (i
= 0; i
< ARRAY_SIZE(vsock
->event_list
); i
++) {
349 struct virtio_vsock_event
*event
= &vsock
->event_list
[i
];
351 virtio_vsock_event_fill_one(vsock
, event
);
354 virtqueue_kick(vsock
->vqs
[VSOCK_VQ_EVENT
]);
357 static void virtio_vsock_reset_sock(struct sock
*sk
)
360 sk
->sk_state
= TCP_CLOSE
;
361 sk
->sk_err
= ECONNRESET
;
362 sk
->sk_error_report(sk
);
366 static void virtio_vsock_update_guest_cid(struct virtio_vsock
*vsock
)
368 struct virtio_device
*vdev
= vsock
->vdev
;
371 vdev
->config
->get(vdev
, offsetof(struct virtio_vsock_config
, guest_cid
),
372 &guest_cid
, sizeof(guest_cid
));
373 vsock
->guest_cid
= le64_to_cpu(guest_cid
);
376 /* event_lock must be held */
377 static void virtio_vsock_event_handle(struct virtio_vsock
*vsock
,
378 struct virtio_vsock_event
*event
)
380 switch (le32_to_cpu(event
->id
)) {
381 case VIRTIO_VSOCK_EVENT_TRANSPORT_RESET
:
382 virtio_vsock_update_guest_cid(vsock
);
383 vsock_for_each_connected_socket(virtio_vsock_reset_sock
);
388 static void virtio_transport_event_work(struct work_struct
*work
)
390 struct virtio_vsock
*vsock
=
391 container_of(work
, struct virtio_vsock
, event_work
);
392 struct virtqueue
*vq
;
394 vq
= vsock
->vqs
[VSOCK_VQ_EVENT
];
396 mutex_lock(&vsock
->event_lock
);
398 if (!vsock
->event_run
)
402 struct virtio_vsock_event
*event
;
405 virtqueue_disable_cb(vq
);
406 while ((event
= virtqueue_get_buf(vq
, &len
)) != NULL
) {
407 if (len
== sizeof(*event
))
408 virtio_vsock_event_handle(vsock
, event
);
410 virtio_vsock_event_fill_one(vsock
, event
);
412 } while (!virtqueue_enable_cb(vq
));
414 virtqueue_kick(vsock
->vqs
[VSOCK_VQ_EVENT
]);
416 mutex_unlock(&vsock
->event_lock
);
419 static void virtio_vsock_event_done(struct virtqueue
*vq
)
421 struct virtio_vsock
*vsock
= vq
->vdev
->priv
;
425 queue_work(virtio_vsock_workqueue
, &vsock
->event_work
);
428 static void virtio_vsock_tx_done(struct virtqueue
*vq
)
430 struct virtio_vsock
*vsock
= vq
->vdev
->priv
;
434 queue_work(virtio_vsock_workqueue
, &vsock
->tx_work
);
437 static void virtio_vsock_rx_done(struct virtqueue
*vq
)
439 struct virtio_vsock
*vsock
= vq
->vdev
->priv
;
443 queue_work(virtio_vsock_workqueue
, &vsock
->rx_work
);
446 static struct virtio_transport virtio_transport
= {
448 .module
= THIS_MODULE
,
450 .get_local_cid
= virtio_transport_get_local_cid
,
452 .init
= virtio_transport_do_socket_init
,
453 .destruct
= virtio_transport_destruct
,
454 .release
= virtio_transport_release
,
455 .connect
= virtio_transport_connect
,
456 .shutdown
= virtio_transport_shutdown
,
457 .cancel_pkt
= virtio_transport_cancel_pkt
,
459 .dgram_bind
= virtio_transport_dgram_bind
,
460 .dgram_dequeue
= virtio_transport_dgram_dequeue
,
461 .dgram_enqueue
= virtio_transport_dgram_enqueue
,
462 .dgram_allow
= virtio_transport_dgram_allow
,
464 .stream_dequeue
= virtio_transport_stream_dequeue
,
465 .stream_enqueue
= virtio_transport_stream_enqueue
,
466 .stream_has_data
= virtio_transport_stream_has_data
,
467 .stream_has_space
= virtio_transport_stream_has_space
,
468 .stream_rcvhiwat
= virtio_transport_stream_rcvhiwat
,
469 .stream_is_active
= virtio_transport_stream_is_active
,
470 .stream_allow
= virtio_transport_stream_allow
,
472 .notify_poll_in
= virtio_transport_notify_poll_in
,
473 .notify_poll_out
= virtio_transport_notify_poll_out
,
474 .notify_recv_init
= virtio_transport_notify_recv_init
,
475 .notify_recv_pre_block
= virtio_transport_notify_recv_pre_block
,
476 .notify_recv_pre_dequeue
= virtio_transport_notify_recv_pre_dequeue
,
477 .notify_recv_post_dequeue
= virtio_transport_notify_recv_post_dequeue
,
478 .notify_send_init
= virtio_transport_notify_send_init
,
479 .notify_send_pre_block
= virtio_transport_notify_send_pre_block
,
480 .notify_send_pre_enqueue
= virtio_transport_notify_send_pre_enqueue
,
481 .notify_send_post_enqueue
= virtio_transport_notify_send_post_enqueue
,
482 .notify_buffer_size
= virtio_transport_notify_buffer_size
,
485 .send_pkt
= virtio_transport_send_pkt
,
488 static void virtio_transport_rx_work(struct work_struct
*work
)
490 struct virtio_vsock
*vsock
=
491 container_of(work
, struct virtio_vsock
, rx_work
);
492 struct virtqueue
*vq
;
494 vq
= vsock
->vqs
[VSOCK_VQ_RX
];
496 mutex_lock(&vsock
->rx_lock
);
502 virtqueue_disable_cb(vq
);
504 struct virtio_vsock_pkt
*pkt
;
507 if (!virtio_transport_more_replies(vsock
)) {
508 /* Stop rx until the device processes already
509 * pending replies. Leave rx virtqueue
510 * callbacks disabled.
515 pkt
= virtqueue_get_buf(vq
, &len
);
522 /* Drop short/long packets */
523 if (unlikely(len
< sizeof(pkt
->hdr
) ||
524 len
> sizeof(pkt
->hdr
) + pkt
->len
)) {
525 virtio_transport_free_pkt(pkt
);
529 pkt
->len
= len
- sizeof(pkt
->hdr
);
530 virtio_transport_deliver_tap_pkt(pkt
);
531 virtio_transport_recv_pkt(&virtio_transport
, pkt
);
533 } while (!virtqueue_enable_cb(vq
));
536 if (vsock
->rx_buf_nr
< vsock
->rx_buf_max_nr
/ 2)
537 virtio_vsock_rx_fill(vsock
);
538 mutex_unlock(&vsock
->rx_lock
);
541 static int virtio_vsock_probe(struct virtio_device
*vdev
)
543 vq_callback_t
*callbacks
[] = {
544 virtio_vsock_rx_done
,
545 virtio_vsock_tx_done
,
546 virtio_vsock_event_done
,
548 static const char * const names
[] = {
553 struct virtio_vsock
*vsock
= NULL
;
556 ret
= mutex_lock_interruptible(&the_virtio_vsock_mutex
);
560 /* Only one virtio-vsock device per guest is supported */
561 if (rcu_dereference_protected(the_virtio_vsock
,
562 lockdep_is_held(&the_virtio_vsock_mutex
))) {
567 vsock
= kzalloc(sizeof(*vsock
), GFP_KERNEL
);
575 ret
= virtio_find_vqs(vsock
->vdev
, VSOCK_VQ_MAX
,
576 vsock
->vqs
, callbacks
, names
,
581 virtio_vsock_update_guest_cid(vsock
);
583 vsock
->rx_buf_nr
= 0;
584 vsock
->rx_buf_max_nr
= 0;
585 atomic_set(&vsock
->queued_replies
, 0);
587 mutex_init(&vsock
->tx_lock
);
588 mutex_init(&vsock
->rx_lock
);
589 mutex_init(&vsock
->event_lock
);
590 spin_lock_init(&vsock
->send_pkt_list_lock
);
591 INIT_LIST_HEAD(&vsock
->send_pkt_list
);
592 INIT_WORK(&vsock
->rx_work
, virtio_transport_rx_work
);
593 INIT_WORK(&vsock
->tx_work
, virtio_transport_tx_work
);
594 INIT_WORK(&vsock
->event_work
, virtio_transport_event_work
);
595 INIT_WORK(&vsock
->send_pkt_work
, virtio_transport_send_pkt_work
);
597 mutex_lock(&vsock
->tx_lock
);
598 vsock
->tx_run
= true;
599 mutex_unlock(&vsock
->tx_lock
);
601 mutex_lock(&vsock
->rx_lock
);
602 virtio_vsock_rx_fill(vsock
);
603 vsock
->rx_run
= true;
604 mutex_unlock(&vsock
->rx_lock
);
606 mutex_lock(&vsock
->event_lock
);
607 virtio_vsock_event_fill(vsock
);
608 vsock
->event_run
= true;
609 mutex_unlock(&vsock
->event_lock
);
612 rcu_assign_pointer(the_virtio_vsock
, vsock
);
614 mutex_unlock(&the_virtio_vsock_mutex
);
619 mutex_unlock(&the_virtio_vsock_mutex
);
623 static void virtio_vsock_remove(struct virtio_device
*vdev
)
625 struct virtio_vsock
*vsock
= vdev
->priv
;
626 struct virtio_vsock_pkt
*pkt
;
628 mutex_lock(&the_virtio_vsock_mutex
);
631 rcu_assign_pointer(the_virtio_vsock
, NULL
);
634 /* Reset all connected sockets when the device disappear */
635 vsock_for_each_connected_socket(virtio_vsock_reset_sock
);
637 /* Stop all work handlers to make sure no one is accessing the device,
638 * so we can safely call vdev->config->reset().
640 mutex_lock(&vsock
->rx_lock
);
641 vsock
->rx_run
= false;
642 mutex_unlock(&vsock
->rx_lock
);
644 mutex_lock(&vsock
->tx_lock
);
645 vsock
->tx_run
= false;
646 mutex_unlock(&vsock
->tx_lock
);
648 mutex_lock(&vsock
->event_lock
);
649 vsock
->event_run
= false;
650 mutex_unlock(&vsock
->event_lock
);
652 /* Flush all device writes and interrupts, device will not use any
655 vdev
->config
->reset(vdev
);
657 mutex_lock(&vsock
->rx_lock
);
658 while ((pkt
= virtqueue_detach_unused_buf(vsock
->vqs
[VSOCK_VQ_RX
])))
659 virtio_transport_free_pkt(pkt
);
660 mutex_unlock(&vsock
->rx_lock
);
662 mutex_lock(&vsock
->tx_lock
);
663 while ((pkt
= virtqueue_detach_unused_buf(vsock
->vqs
[VSOCK_VQ_TX
])))
664 virtio_transport_free_pkt(pkt
);
665 mutex_unlock(&vsock
->tx_lock
);
667 spin_lock_bh(&vsock
->send_pkt_list_lock
);
668 while (!list_empty(&vsock
->send_pkt_list
)) {
669 pkt
= list_first_entry(&vsock
->send_pkt_list
,
670 struct virtio_vsock_pkt
, list
);
671 list_del(&pkt
->list
);
672 virtio_transport_free_pkt(pkt
);
674 spin_unlock_bh(&vsock
->send_pkt_list_lock
);
676 /* Delete virtqueues and flush outstanding callbacks if any */
677 vdev
->config
->del_vqs(vdev
);
679 /* Other works can be queued before 'config->del_vqs()', so we flush
680 * all works before to free the vsock object to avoid use after free.
682 flush_work(&vsock
->rx_work
);
683 flush_work(&vsock
->tx_work
);
684 flush_work(&vsock
->event_work
);
685 flush_work(&vsock
->send_pkt_work
);
687 mutex_unlock(&the_virtio_vsock_mutex
);
692 static struct virtio_device_id id_table
[] = {
693 { VIRTIO_ID_VSOCK
, VIRTIO_DEV_ANY_ID
},
697 static unsigned int features
[] = {
700 static struct virtio_driver virtio_vsock_driver
= {
701 .feature_table
= features
,
702 .feature_table_size
= ARRAY_SIZE(features
),
703 .driver
.name
= KBUILD_MODNAME
,
704 .driver
.owner
= THIS_MODULE
,
705 .id_table
= id_table
,
706 .probe
= virtio_vsock_probe
,
707 .remove
= virtio_vsock_remove
,
710 static int __init
virtio_vsock_init(void)
714 virtio_vsock_workqueue
= alloc_workqueue("virtio_vsock", 0, 0);
715 if (!virtio_vsock_workqueue
)
718 ret
= vsock_core_register(&virtio_transport
.transport
,
719 VSOCK_TRANSPORT_F_G2H
);
723 ret
= register_virtio_driver(&virtio_vsock_driver
);
730 vsock_core_unregister(&virtio_transport
.transport
);
732 destroy_workqueue(virtio_vsock_workqueue
);
736 static void __exit
virtio_vsock_exit(void)
738 unregister_virtio_driver(&virtio_vsock_driver
);
739 vsock_core_unregister(&virtio_transport
.transport
);
740 destroy_workqueue(virtio_vsock_workqueue
);
743 module_init(virtio_vsock_init
);
744 module_exit(virtio_vsock_exit
);
745 MODULE_LICENSE("GPL v2");
746 MODULE_AUTHOR("Asias He");
747 MODULE_DESCRIPTION("virtio transport for vsock");
748 MODULE_DEVICE_TABLE(virtio
, id_table
);