1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * virtio transport for vsock
4  *
5  * Copyright (C) 2013-2015 Red Hat, Inc.
6  * Author: Asias He <asias@redhat.com>
7  *         Stefan Hajnoczi <stefanha@redhat.com>
8  *
9  * Some of the code is take from Gerd Hoffmann <kraxel@redhat.com>'s
10  * early virtio-vsock proof-of-concept bits.
11  */
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>
20 #include <net/sock.h>
21 #include <linux/mutex.h>
22 #include <net/af_vsock.h>
23 
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 */
28 
29 struct virtio_vsock {
30 	struct virtio_device *vdev;
31 	struct virtqueue *vqs[VSOCK_VQ_MAX];
32 
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;
37 
38 	/* The following fields are protected by tx_lock.  vqs[VSOCK_VQ_TX]
39 	 * must be accessed with tx_lock held.
40 	 */
41 	struct mutex tx_lock;
42 	bool tx_run;
43 
44 	struct work_struct send_pkt_work;
45 	struct sk_buff_head send_pkt_queue;
46 
47 	atomic_t queued_replies;
48 
49 	/* The following fields are protected by rx_lock.  vqs[VSOCK_VQ_RX]
50 	 * must be accessed with rx_lock held.
51 	 */
52 	struct mutex rx_lock;
53 	bool rx_run;
54 	int rx_buf_nr;
55 	int rx_buf_max_nr;
56 
57 	/* The following fields are protected by event_lock.
58 	 * vqs[VSOCK_VQ_EVENT] must be accessed with event_lock held.
59 	 */
60 	struct mutex event_lock;
61 	bool event_run;
62 	struct virtio_vsock_event event_list[8];
63 
64 	u32 guest_cid;
65 	bool seqpacket_allow;
66 };
67 
68 static u32 virtio_transport_get_local_cid(void)
69 {
70 	struct virtio_vsock *vsock;
71 	u32 ret;
72 
73 	rcu_read_lock();
74 	vsock = rcu_dereference(the_virtio_vsock);
75 	if (!vsock) {
76 		ret = VMADDR_CID_ANY;
77 		goto out_rcu;
78 	}
79 
80 	ret = vsock->guest_cid;
81 out_rcu:
82 	rcu_read_unlock();
83 	return ret;
84 }
85 
86 static void
87 virtio_transport_send_pkt_work(struct work_struct *work)
88 {
89 	struct virtio_vsock *vsock =
90 		container_of(work, struct virtio_vsock, send_pkt_work);
91 	struct virtqueue *vq;
92 	bool added = false;
93 	bool restart_rx = false;
94 
95 	mutex_lock(&vsock->tx_lock);
96 
97 	if (!vsock->tx_run)
98 		goto out;
99 
100 	vq = vsock->vqs[VSOCK_VQ_TX];
101 
102 	for (;;) {
103 		struct scatterlist hdr, buf, *sgs[2];
104 		int ret, in_sg = 0, out_sg = 0;
105 		struct sk_buff *skb;
106 		bool reply;
107 
108 		skb = virtio_vsock_skb_dequeue(&vsock->send_pkt_queue);
109 		if (!skb)
110 			break;
111 
112 		virtio_transport_deliver_tap_pkt(skb);
113 		reply = virtio_vsock_skb_reply(skb);
114 
115 		sg_init_one(&hdr, virtio_vsock_hdr(skb), sizeof(*virtio_vsock_hdr(skb)));
116 		sgs[out_sg++] = &hdr;
117 		if (skb->len > 0) {
118 			sg_init_one(&buf, skb->data, skb->len);
119 			sgs[out_sg++] = &buf;
120 		}
121 
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
124 		 * the vq
125 		 */
126 		if (ret < 0) {
127 			virtio_vsock_skb_queue_head(&vsock->send_pkt_queue, skb);
128 			break;
129 		}
130 
131 		if (reply) {
132 			struct virtqueue *rx_vq = vsock->vqs[VSOCK_VQ_RX];
133 			int val;
134 
135 			val = atomic_dec_return(&vsock->queued_replies);
136 
137 			/* Do we now have resources to resume rx processing? */
138 			if (val + 1 == virtqueue_get_vring_size(rx_vq))
139 				restart_rx = true;
140 		}
141 
142 		added = true;
143 	}
144 
145 	if (added)
146 		virtqueue_kick(vq);
147 
148 out:
149 	mutex_unlock(&vsock->tx_lock);
150 
151 	if (restart_rx)
152 		queue_work(virtio_vsock_workqueue, &vsock->rx_work);
153 }
154 
155 static int
156 virtio_transport_send_pkt(struct sk_buff *skb)
157 {
158 	struct virtio_vsock_hdr *hdr;
159 	struct virtio_vsock *vsock;
160 	int len = skb->len;
161 
162 	hdr = virtio_vsock_hdr(skb);
163 
164 	rcu_read_lock();
165 	vsock = rcu_dereference(the_virtio_vsock);
166 	if (!vsock) {
167 		kfree_skb(skb);
168 		len = -ENODEV;
169 		goto out_rcu;
170 	}
171 
172 	if (le64_to_cpu(hdr->dst_cid) == vsock->guest_cid) {
173 		kfree_skb(skb);
174 		len = -ENODEV;
175 		goto out_rcu;
176 	}
177 
178 	if (virtio_vsock_skb_reply(skb))
179 		atomic_inc(&vsock->queued_replies);
180 
181 	virtio_vsock_skb_queue_tail(&vsock->send_pkt_queue, skb);
182 	queue_work(virtio_vsock_workqueue, &vsock->send_pkt_work);
183 
184 out_rcu:
185 	rcu_read_unlock();
186 	return len;
187 }
188 
189 static int
190 virtio_transport_cancel_pkt(struct vsock_sock *vsk)
191 {
192 	struct virtio_vsock *vsock;
193 	int cnt = 0, ret;
194 
195 	rcu_read_lock();
196 	vsock = rcu_dereference(the_virtio_vsock);
197 	if (!vsock) {
198 		ret = -ENODEV;
199 		goto out_rcu;
200 	}
201 
202 	cnt = virtio_transport_purge_skbs(vsk, &vsock->send_pkt_queue);
203 
204 	if (cnt) {
205 		struct virtqueue *rx_vq = vsock->vqs[VSOCK_VQ_RX];
206 		int new_cnt;
207 
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);
212 	}
213 
214 	ret = 0;
215 
216 out_rcu:
217 	rcu_read_unlock();
218 	return ret;
219 }
220 
221 static void virtio_vsock_rx_fill(struct virtio_vsock *vsock)
222 {
223 	int total_len = VIRTIO_VSOCK_DEFAULT_RX_BUF_SIZE + VIRTIO_VSOCK_SKB_HEADROOM;
224 	struct scatterlist pkt, *p;
225 	struct virtqueue *vq;
226 	struct sk_buff *skb;
227 	int ret;
228 
229 	vq = vsock->vqs[VSOCK_VQ_RX];
230 
231 	do {
232 		skb = virtio_vsock_alloc_skb(total_len, GFP_KERNEL);
233 		if (!skb)
234 			break;
235 
236 		memset(skb->head, 0, VIRTIO_VSOCK_SKB_HEADROOM);
237 		sg_init_one(&pkt, virtio_vsock_hdr(skb), total_len);
238 		p = &pkt;
239 		ret = virtqueue_add_sgs(vq, &p, 0, 1, skb, GFP_KERNEL);
240 		if (ret < 0) {
241 			kfree_skb(skb);
242 			break;
243 		}
244 
245 		vsock->rx_buf_nr++;
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;
249 	virtqueue_kick(vq);
250 }
251 
252 static void virtio_transport_tx_work(struct work_struct *work)
253 {
254 	struct virtio_vsock *vsock =
255 		container_of(work, struct virtio_vsock, tx_work);
256 	struct virtqueue *vq;
257 	bool added = false;
258 
259 	vq = vsock->vqs[VSOCK_VQ_TX];
260 	mutex_lock(&vsock->tx_lock);
261 
262 	if (!vsock->tx_run)
263 		goto out;
264 
265 	do {
266 		struct sk_buff *skb;
267 		unsigned int len;
268 
269 		virtqueue_disable_cb(vq);
270 		while ((skb = virtqueue_get_buf(vq, &len)) != NULL) {
271 			consume_skb(skb);
272 			added = true;
273 		}
274 	} while (!virtqueue_enable_cb(vq));
275 
276 out:
277 	mutex_unlock(&vsock->tx_lock);
278 
279 	if (added)
280 		queue_work(virtio_vsock_workqueue, &vsock->send_pkt_work);
281 }
282 
283 /* Is there space left for replies to rx packets? */
284 static bool virtio_transport_more_replies(struct virtio_vsock *vsock)
285 {
286 	struct virtqueue *vq = vsock->vqs[VSOCK_VQ_RX];
287 	int val;
288 
289 	smp_rmb(); /* paired with atomic_inc() and atomic_dec_return() */
290 	val = atomic_read(&vsock->queued_replies);
291 
292 	return val < virtqueue_get_vring_size(vq);
293 }
294 
295 /* event_lock must be held */
296 static int virtio_vsock_event_fill_one(struct virtio_vsock *vsock,
297 				       struct virtio_vsock_event *event)
298 {
299 	struct scatterlist sg;
300 	struct virtqueue *vq;
301 
302 	vq = vsock->vqs[VSOCK_VQ_EVENT];
303 
304 	sg_init_one(&sg, event, sizeof(*event));
305 
306 	return virtqueue_add_inbuf(vq, &sg, 1, event, GFP_KERNEL);
307 }
308 
309 /* event_lock must be held */
310 static void virtio_vsock_event_fill(struct virtio_vsock *vsock)
311 {
312 	size_t i;
313 
314 	for (i = 0; i < ARRAY_SIZE(vsock->event_list); i++) {
315 		struct virtio_vsock_event *event = &vsock->event_list[i];
316 
317 		virtio_vsock_event_fill_one(vsock, event);
318 	}
319 
320 	virtqueue_kick(vsock->vqs[VSOCK_VQ_EVENT]);
321 }
322 
323 static void virtio_vsock_reset_sock(struct sock *sk)
324 {
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
327 	 * executing.
328 	 */
329 
330 	sk->sk_state = TCP_CLOSE;
331 	sk->sk_err = ECONNRESET;
332 	sk_error_report(sk);
333 }
334 
335 static void virtio_vsock_update_guest_cid(struct virtio_vsock *vsock)
336 {
337 	struct virtio_device *vdev = vsock->vdev;
338 	__le64 guest_cid;
339 
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);
343 }
344 
345 /* event_lock must be held */
346 static void virtio_vsock_event_handle(struct virtio_vsock *vsock,
347 				      struct virtio_vsock_event *event)
348 {
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);
354 		break;
355 	}
356 }
357 
358 static void virtio_transport_event_work(struct work_struct *work)
359 {
360 	struct virtio_vsock *vsock =
361 		container_of(work, struct virtio_vsock, event_work);
362 	struct virtqueue *vq;
363 
364 	vq = vsock->vqs[VSOCK_VQ_EVENT];
365 
366 	mutex_lock(&vsock->event_lock);
367 
368 	if (!vsock->event_run)
369 		goto out;
370 
371 	do {
372 		struct virtio_vsock_event *event;
373 		unsigned int len;
374 
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);
379 
380 			virtio_vsock_event_fill_one(vsock, event);
381 		}
382 	} while (!virtqueue_enable_cb(vq));
383 
384 	virtqueue_kick(vsock->vqs[VSOCK_VQ_EVENT]);
385 out:
386 	mutex_unlock(&vsock->event_lock);
387 }
388 
389 static void virtio_vsock_event_done(struct virtqueue *vq)
390 {
391 	struct virtio_vsock *vsock = vq->vdev->priv;
392 
393 	if (!vsock)
394 		return;
395 	queue_work(virtio_vsock_workqueue, &vsock->event_work);
396 }
397 
398 static void virtio_vsock_tx_done(struct virtqueue *vq)
399 {
400 	struct virtio_vsock *vsock = vq->vdev->priv;
401 
402 	if (!vsock)
403 		return;
404 	queue_work(virtio_vsock_workqueue, &vsock->tx_work);
405 }
406 
407 static void virtio_vsock_rx_done(struct virtqueue *vq)
408 {
409 	struct virtio_vsock *vsock = vq->vdev->priv;
410 
411 	if (!vsock)
412 		return;
413 	queue_work(virtio_vsock_workqueue, &vsock->rx_work);
414 }
415 
416 static bool virtio_transport_seqpacket_allow(u32 remote_cid);
417 
418 static struct virtio_transport virtio_transport = {
419 	.transport = {
420 		.module                   = THIS_MODULE,
421 
422 		.get_local_cid            = virtio_transport_get_local_cid,
423 
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,
430 
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,
435 
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,
443 
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,
448 
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,
460 		.notify_set_rcvlowat      = virtio_transport_notify_set_rcvlowat,
461 
462 		.read_skb = virtio_transport_read_skb,
463 	},
464 
465 	.send_pkt = virtio_transport_send_pkt,
466 };
467 
468 static bool virtio_transport_seqpacket_allow(u32 remote_cid)
469 {
470 	struct virtio_vsock *vsock;
471 	bool seqpacket_allow;
472 
473 	seqpacket_allow = false;
474 	rcu_read_lock();
475 	vsock = rcu_dereference(the_virtio_vsock);
476 	if (vsock)
477 		seqpacket_allow = vsock->seqpacket_allow;
478 	rcu_read_unlock();
479 
480 	return seqpacket_allow;
481 }
482 
483 static void virtio_transport_rx_work(struct work_struct *work)
484 {
485 	struct virtio_vsock *vsock =
486 		container_of(work, struct virtio_vsock, rx_work);
487 	struct virtqueue *vq;
488 
489 	vq = vsock->vqs[VSOCK_VQ_RX];
490 
491 	mutex_lock(&vsock->rx_lock);
492 
493 	if (!vsock->rx_run)
494 		goto out;
495 
496 	do {
497 		virtqueue_disable_cb(vq);
498 		for (;;) {
499 			struct sk_buff *skb;
500 			unsigned int len;
501 
502 			if (!virtio_transport_more_replies(vsock)) {
503 				/* Stop rx until the device processes already
504 				 * pending replies.  Leave rx virtqueue
505 				 * callbacks disabled.
506 				 */
507 				goto out;
508 			}
509 
510 			skb = virtqueue_get_buf(vq, &len);
511 			if (!skb)
512 				break;
513 
514 			vsock->rx_buf_nr--;
515 
516 			/* Drop short/long packets */
517 			if (unlikely(len < sizeof(struct virtio_vsock_hdr) ||
518 				     len > virtio_vsock_skb_len(skb))) {
519 				kfree_skb(skb);
520 				continue;
521 			}
522 
523 			virtio_vsock_skb_rx_put(skb);
524 			virtio_transport_deliver_tap_pkt(skb);
525 			virtio_transport_recv_pkt(&virtio_transport, skb);
526 		}
527 	} while (!virtqueue_enable_cb(vq));
528 
529 out:
530 	if (vsock->rx_buf_nr < vsock->rx_buf_max_nr / 2)
531 		virtio_vsock_rx_fill(vsock);
532 	mutex_unlock(&vsock->rx_lock);
533 }
534 
535 static int virtio_vsock_vqs_init(struct virtio_vsock *vsock)
536 {
537 	struct virtio_device *vdev = vsock->vdev;
538 	static const char * const names[] = {
539 		"rx",
540 		"tx",
541 		"event",
542 	};
543 	vq_callback_t *callbacks[] = {
544 		virtio_vsock_rx_done,
545 		virtio_vsock_tx_done,
546 		virtio_vsock_event_done,
547 	};
548 	int ret;
549 
550 	ret = virtio_find_vqs(vdev, VSOCK_VQ_MAX, vsock->vqs, callbacks, names,
551 			      NULL);
552 	if (ret < 0)
553 		return ret;
554 
555 	virtio_vsock_update_guest_cid(vsock);
556 
557 	virtio_device_ready(vdev);
558 
559 	return 0;
560 }
561 
562 static void virtio_vsock_vqs_start(struct virtio_vsock *vsock)
563 {
564 	mutex_lock(&vsock->tx_lock);
565 	vsock->tx_run = true;
566 	mutex_unlock(&vsock->tx_lock);
567 
568 	mutex_lock(&vsock->rx_lock);
569 	virtio_vsock_rx_fill(vsock);
570 	vsock->rx_run = true;
571 	mutex_unlock(&vsock->rx_lock);
572 
573 	mutex_lock(&vsock->event_lock);
574 	virtio_vsock_event_fill(vsock);
575 	vsock->event_run = true;
576 	mutex_unlock(&vsock->event_lock);
577 
578 	/* virtio_transport_send_pkt() can queue packets once
579 	 * the_virtio_vsock is set, but they won't be processed until
580 	 * vsock->tx_run is set to true. We queue vsock->send_pkt_work
581 	 * when initialization finishes to send those packets queued
582 	 * earlier.
583 	 * We don't need to queue the other workers (rx, event) because
584 	 * as long as we don't fill the queues with empty buffers, the
585 	 * host can't send us any notification.
586 	 */
587 	queue_work(virtio_vsock_workqueue, &vsock->send_pkt_work);
588 }
589 
590 static void virtio_vsock_vqs_del(struct virtio_vsock *vsock)
591 {
592 	struct virtio_device *vdev = vsock->vdev;
593 	struct sk_buff *skb;
594 
595 	/* Reset all connected sockets when the VQs disappear */
596 	vsock_for_each_connected_socket(&virtio_transport.transport,
597 					virtio_vsock_reset_sock);
598 
599 	/* Stop all work handlers to make sure no one is accessing the device,
600 	 * so we can safely call virtio_reset_device().
601 	 */
602 	mutex_lock(&vsock->rx_lock);
603 	vsock->rx_run = false;
604 	mutex_unlock(&vsock->rx_lock);
605 
606 	mutex_lock(&vsock->tx_lock);
607 	vsock->tx_run = false;
608 	mutex_unlock(&vsock->tx_lock);
609 
610 	mutex_lock(&vsock->event_lock);
611 	vsock->event_run = false;
612 	mutex_unlock(&vsock->event_lock);
613 
614 	/* Flush all device writes and interrupts, device will not use any
615 	 * more buffers.
616 	 */
617 	virtio_reset_device(vdev);
618 
619 	mutex_lock(&vsock->rx_lock);
620 	while ((skb = virtqueue_detach_unused_buf(vsock->vqs[VSOCK_VQ_RX])))
621 		kfree_skb(skb);
622 	mutex_unlock(&vsock->rx_lock);
623 
624 	mutex_lock(&vsock->tx_lock);
625 	while ((skb = virtqueue_detach_unused_buf(vsock->vqs[VSOCK_VQ_TX])))
626 		kfree_skb(skb);
627 	mutex_unlock(&vsock->tx_lock);
628 
629 	virtio_vsock_skb_queue_purge(&vsock->send_pkt_queue);
630 
631 	/* Delete virtqueues and flush outstanding callbacks if any */
632 	vdev->config->del_vqs(vdev);
633 }
634 
635 static int virtio_vsock_probe(struct virtio_device *vdev)
636 {
637 	struct virtio_vsock *vsock = NULL;
638 	int ret;
639 
640 	ret = mutex_lock_interruptible(&the_virtio_vsock_mutex);
641 	if (ret)
642 		return ret;
643 
644 	/* Only one virtio-vsock device per guest is supported */
645 	if (rcu_dereference_protected(the_virtio_vsock,
646 				lockdep_is_held(&the_virtio_vsock_mutex))) {
647 		ret = -EBUSY;
648 		goto out;
649 	}
650 
651 	vsock = kzalloc(sizeof(*vsock), GFP_KERNEL);
652 	if (!vsock) {
653 		ret = -ENOMEM;
654 		goto out;
655 	}
656 
657 	vsock->vdev = vdev;
658 
659 	vsock->rx_buf_nr = 0;
660 	vsock->rx_buf_max_nr = 0;
661 	atomic_set(&vsock->queued_replies, 0);
662 
663 	mutex_init(&vsock->tx_lock);
664 	mutex_init(&vsock->rx_lock);
665 	mutex_init(&vsock->event_lock);
666 	skb_queue_head_init(&vsock->send_pkt_queue);
667 	INIT_WORK(&vsock->rx_work, virtio_transport_rx_work);
668 	INIT_WORK(&vsock->tx_work, virtio_transport_tx_work);
669 	INIT_WORK(&vsock->event_work, virtio_transport_event_work);
670 	INIT_WORK(&vsock->send_pkt_work, virtio_transport_send_pkt_work);
671 
672 	if (virtio_has_feature(vdev, VIRTIO_VSOCK_F_SEQPACKET))
673 		vsock->seqpacket_allow = true;
674 
675 	vdev->priv = vsock;
676 
677 	ret = virtio_vsock_vqs_init(vsock);
678 	if (ret < 0)
679 		goto out;
680 
681 	rcu_assign_pointer(the_virtio_vsock, vsock);
682 	virtio_vsock_vqs_start(vsock);
683 
684 	mutex_unlock(&the_virtio_vsock_mutex);
685 
686 	return 0;
687 
688 out:
689 	kfree(vsock);
690 	mutex_unlock(&the_virtio_vsock_mutex);
691 	return ret;
692 }
693 
694 static void virtio_vsock_remove(struct virtio_device *vdev)
695 {
696 	struct virtio_vsock *vsock = vdev->priv;
697 
698 	mutex_lock(&the_virtio_vsock_mutex);
699 
700 	vdev->priv = NULL;
701 	rcu_assign_pointer(the_virtio_vsock, NULL);
702 	synchronize_rcu();
703 
704 	virtio_vsock_vqs_del(vsock);
705 
706 	/* Other works can be queued before 'config->del_vqs()', so we flush
707 	 * all works before to free the vsock object to avoid use after free.
708 	 */
709 	flush_work(&vsock->rx_work);
710 	flush_work(&vsock->tx_work);
711 	flush_work(&vsock->event_work);
712 	flush_work(&vsock->send_pkt_work);
713 
714 	mutex_unlock(&the_virtio_vsock_mutex);
715 
716 	kfree(vsock);
717 }
718 
719 #ifdef CONFIG_PM_SLEEP
720 static int virtio_vsock_freeze(struct virtio_device *vdev)
721 {
722 	struct virtio_vsock *vsock = vdev->priv;
723 
724 	mutex_lock(&the_virtio_vsock_mutex);
725 
726 	rcu_assign_pointer(the_virtio_vsock, NULL);
727 	synchronize_rcu();
728 
729 	virtio_vsock_vqs_del(vsock);
730 
731 	mutex_unlock(&the_virtio_vsock_mutex);
732 
733 	return 0;
734 }
735 
736 static int virtio_vsock_restore(struct virtio_device *vdev)
737 {
738 	struct virtio_vsock *vsock = vdev->priv;
739 	int ret;
740 
741 	mutex_lock(&the_virtio_vsock_mutex);
742 
743 	/* Only one virtio-vsock device per guest is supported */
744 	if (rcu_dereference_protected(the_virtio_vsock,
745 				lockdep_is_held(&the_virtio_vsock_mutex))) {
746 		ret = -EBUSY;
747 		goto out;
748 	}
749 
750 	ret = virtio_vsock_vqs_init(vsock);
751 	if (ret < 0)
752 		goto out;
753 
754 	rcu_assign_pointer(the_virtio_vsock, vsock);
755 	virtio_vsock_vqs_start(vsock);
756 
757 out:
758 	mutex_unlock(&the_virtio_vsock_mutex);
759 	return ret;
760 }
761 #endif /* CONFIG_PM_SLEEP */
762 
763 static struct virtio_device_id id_table[] = {
764 	{ VIRTIO_ID_VSOCK, VIRTIO_DEV_ANY_ID },
765 	{ 0 },
766 };
767 
768 static unsigned int features[] = {
769 	VIRTIO_VSOCK_F_SEQPACKET
770 };
771 
772 static struct virtio_driver virtio_vsock_driver = {
773 	.feature_table = features,
774 	.feature_table_size = ARRAY_SIZE(features),
775 	.driver.name = KBUILD_MODNAME,
776 	.driver.owner = THIS_MODULE,
777 	.id_table = id_table,
778 	.probe = virtio_vsock_probe,
779 	.remove = virtio_vsock_remove,
780 #ifdef CONFIG_PM_SLEEP
781 	.freeze = virtio_vsock_freeze,
782 	.restore = virtio_vsock_restore,
783 #endif
784 };
785 
786 static int __init virtio_vsock_init(void)
787 {
788 	int ret;
789 
790 	virtio_vsock_workqueue = alloc_workqueue("virtio_vsock", 0, 0);
791 	if (!virtio_vsock_workqueue)
792 		return -ENOMEM;
793 
794 	ret = vsock_core_register(&virtio_transport.transport,
795 				  VSOCK_TRANSPORT_F_G2H);
796 	if (ret)
797 		goto out_wq;
798 
799 	ret = register_virtio_driver(&virtio_vsock_driver);
800 	if (ret)
801 		goto out_vci;
802 
803 	return 0;
804 
805 out_vci:
806 	vsock_core_unregister(&virtio_transport.transport);
807 out_wq:
808 	destroy_workqueue(virtio_vsock_workqueue);
809 	return ret;
810 }
811 
812 static void __exit virtio_vsock_exit(void)
813 {
814 	unregister_virtio_driver(&virtio_vsock_driver);
815 	vsock_core_unregister(&virtio_transport.transport);
816 	destroy_workqueue(virtio_vsock_workqueue);
817 }
818 
819 module_init(virtio_vsock_init);
820 module_exit(virtio_vsock_exit);
821 MODULE_LICENSE("GPL v2");
822 MODULE_AUTHOR("Asias He");
823 MODULE_DESCRIPTION("virtio transport for vsock");
824 MODULE_DEVICE_TABLE(virtio, id_table);
825