xref: /openbmc/linux/drivers/vhost/vsock.c (revision 8ec90bfd)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * vhost 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 #include <linux/miscdevice.h>
10 #include <linux/atomic.h>
11 #include <linux/module.h>
12 #include <linux/mutex.h>
13 #include <linux/vmalloc.h>
14 #include <net/sock.h>
15 #include <linux/virtio_vsock.h>
16 #include <linux/vhost.h>
17 #include <linux/hashtable.h>
18 
19 #include <net/af_vsock.h>
20 #include "vhost.h"
21 
22 #define VHOST_VSOCK_DEFAULT_HOST_CID	2
23 /* Max number of bytes transferred before requeueing the job.
24  * Using this limit prevents one virtqueue from starving others. */
25 #define VHOST_VSOCK_WEIGHT 0x80000
26 /* Max number of packets transferred before requeueing the job.
27  * Using this limit prevents one virtqueue from starving others with
28  * small pkts.
29  */
30 #define VHOST_VSOCK_PKT_WEIGHT 256
31 
32 enum {
33 	VHOST_VSOCK_FEATURES = VHOST_FEATURES,
34 };
35 
36 /* Used to track all the vhost_vsock instances on the system. */
37 static DEFINE_MUTEX(vhost_vsock_mutex);
38 static DEFINE_READ_MOSTLY_HASHTABLE(vhost_vsock_hash, 8);
39 
40 struct vhost_vsock {
41 	struct vhost_dev dev;
42 	struct vhost_virtqueue vqs[2];
43 
44 	/* Link to global vhost_vsock_hash, writes use vhost_vsock_mutex */
45 	struct hlist_node hash;
46 
47 	struct vhost_work send_pkt_work;
48 	spinlock_t send_pkt_list_lock;
49 	struct list_head send_pkt_list;	/* host->guest pending packets */
50 
51 	atomic_t queued_replies;
52 
53 	u32 guest_cid;
54 };
55 
56 static u32 vhost_transport_get_local_cid(void)
57 {
58 	return VHOST_VSOCK_DEFAULT_HOST_CID;
59 }
60 
61 /* Callers that dereference the return value must hold vhost_vsock_mutex or the
62  * RCU read lock.
63  */
64 static struct vhost_vsock *vhost_vsock_get(u32 guest_cid)
65 {
66 	struct vhost_vsock *vsock;
67 
68 	hash_for_each_possible_rcu(vhost_vsock_hash, vsock, hash, guest_cid) {
69 		u32 other_cid = vsock->guest_cid;
70 
71 		/* Skip instances that have no CID yet */
72 		if (other_cid == 0)
73 			continue;
74 
75 		if (other_cid == guest_cid)
76 			return vsock;
77 
78 	}
79 
80 	return NULL;
81 }
82 
83 static void
84 vhost_transport_do_send_pkt(struct vhost_vsock *vsock,
85 			    struct vhost_virtqueue *vq)
86 {
87 	struct vhost_virtqueue *tx_vq = &vsock->vqs[VSOCK_VQ_TX];
88 	int pkts = 0, total_len = 0;
89 	bool added = false;
90 	bool restart_tx = false;
91 
92 	mutex_lock(&vq->mutex);
93 
94 	if (!vhost_vq_get_backend(vq))
95 		goto out;
96 
97 	/* Avoid further vmexits, we're already processing the virtqueue */
98 	vhost_disable_notify(&vsock->dev, vq);
99 
100 	do {
101 		struct virtio_vsock_pkt *pkt;
102 		struct iov_iter iov_iter;
103 		unsigned out, in;
104 		size_t nbytes;
105 		size_t iov_len, payload_len;
106 		int head;
107 
108 		spin_lock_bh(&vsock->send_pkt_list_lock);
109 		if (list_empty(&vsock->send_pkt_list)) {
110 			spin_unlock_bh(&vsock->send_pkt_list_lock);
111 			vhost_enable_notify(&vsock->dev, vq);
112 			break;
113 		}
114 
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);
119 
120 		head = vhost_get_vq_desc(vq, vq->iov, ARRAY_SIZE(vq->iov),
121 					 &out, &in, NULL, NULL);
122 		if (head < 0) {
123 			spin_lock_bh(&vsock->send_pkt_list_lock);
124 			list_add(&pkt->list, &vsock->send_pkt_list);
125 			spin_unlock_bh(&vsock->send_pkt_list_lock);
126 			break;
127 		}
128 
129 		if (head == vq->num) {
130 			spin_lock_bh(&vsock->send_pkt_list_lock);
131 			list_add(&pkt->list, &vsock->send_pkt_list);
132 			spin_unlock_bh(&vsock->send_pkt_list_lock);
133 
134 			/* We cannot finish yet if more buffers snuck in while
135 			 * re-enabling notify.
136 			 */
137 			if (unlikely(vhost_enable_notify(&vsock->dev, vq))) {
138 				vhost_disable_notify(&vsock->dev, vq);
139 				continue;
140 			}
141 			break;
142 		}
143 
144 		if (out) {
145 			virtio_transport_free_pkt(pkt);
146 			vq_err(vq, "Expected 0 output buffers, got %u\n", out);
147 			break;
148 		}
149 
150 		iov_len = iov_length(&vq->iov[out], in);
151 		if (iov_len < sizeof(pkt->hdr)) {
152 			virtio_transport_free_pkt(pkt);
153 			vq_err(vq, "Buffer len [%zu] too small\n", iov_len);
154 			break;
155 		}
156 
157 		iov_iter_init(&iov_iter, READ, &vq->iov[out], in, iov_len);
158 		payload_len = pkt->len - pkt->off;
159 
160 		/* If the packet is greater than the space available in the
161 		 * buffer, we split it using multiple buffers.
162 		 */
163 		if (payload_len > iov_len - sizeof(pkt->hdr))
164 			payload_len = iov_len - sizeof(pkt->hdr);
165 
166 		/* Set the correct length in the header */
167 		pkt->hdr.len = cpu_to_le32(payload_len);
168 
169 		nbytes = copy_to_iter(&pkt->hdr, sizeof(pkt->hdr), &iov_iter);
170 		if (nbytes != sizeof(pkt->hdr)) {
171 			virtio_transport_free_pkt(pkt);
172 			vq_err(vq, "Faulted on copying pkt hdr\n");
173 			break;
174 		}
175 
176 		nbytes = copy_to_iter(pkt->buf + pkt->off, payload_len,
177 				      &iov_iter);
178 		if (nbytes != payload_len) {
179 			virtio_transport_free_pkt(pkt);
180 			vq_err(vq, "Faulted on copying pkt buf\n");
181 			break;
182 		}
183 
184 		/* Deliver to monitoring devices all packets that we
185 		 * will transmit.
186 		 */
187 		virtio_transport_deliver_tap_pkt(pkt);
188 
189 		vhost_add_used(vq, head, sizeof(pkt->hdr) + payload_len);
190 		added = true;
191 
192 		pkt->off += payload_len;
193 		total_len += payload_len;
194 
195 		/* If we didn't send all the payload we can requeue the packet
196 		 * to send it with the next available buffer.
197 		 */
198 		if (pkt->off < pkt->len) {
199 			/* We are queueing the same virtio_vsock_pkt to handle
200 			 * the remaining bytes, and we want to deliver it
201 			 * to monitoring devices in the next iteration.
202 			 */
203 			pkt->tap_delivered = false;
204 
205 			spin_lock_bh(&vsock->send_pkt_list_lock);
206 			list_add(&pkt->list, &vsock->send_pkt_list);
207 			spin_unlock_bh(&vsock->send_pkt_list_lock);
208 		} else {
209 			if (pkt->reply) {
210 				int val;
211 
212 				val = atomic_dec_return(&vsock->queued_replies);
213 
214 				/* Do we have resources to resume tx
215 				 * processing?
216 				 */
217 				if (val + 1 == tx_vq->num)
218 					restart_tx = true;
219 			}
220 
221 			virtio_transport_free_pkt(pkt);
222 		}
223 	} while(likely(!vhost_exceeds_weight(vq, ++pkts, total_len)));
224 	if (added)
225 		vhost_signal(&vsock->dev, vq);
226 
227 out:
228 	mutex_unlock(&vq->mutex);
229 
230 	if (restart_tx)
231 		vhost_poll_queue(&tx_vq->poll);
232 }
233 
234 static void vhost_transport_send_pkt_work(struct vhost_work *work)
235 {
236 	struct vhost_virtqueue *vq;
237 	struct vhost_vsock *vsock;
238 
239 	vsock = container_of(work, struct vhost_vsock, send_pkt_work);
240 	vq = &vsock->vqs[VSOCK_VQ_RX];
241 
242 	vhost_transport_do_send_pkt(vsock, vq);
243 }
244 
245 static int
246 vhost_transport_send_pkt(struct virtio_vsock_pkt *pkt)
247 {
248 	struct vhost_vsock *vsock;
249 	int len = pkt->len;
250 
251 	rcu_read_lock();
252 
253 	/* Find the vhost_vsock according to guest context id  */
254 	vsock = vhost_vsock_get(le64_to_cpu(pkt->hdr.dst_cid));
255 	if (!vsock) {
256 		rcu_read_unlock();
257 		virtio_transport_free_pkt(pkt);
258 		return -ENODEV;
259 	}
260 
261 	if (pkt->reply)
262 		atomic_inc(&vsock->queued_replies);
263 
264 	spin_lock_bh(&vsock->send_pkt_list_lock);
265 	list_add_tail(&pkt->list, &vsock->send_pkt_list);
266 	spin_unlock_bh(&vsock->send_pkt_list_lock);
267 
268 	vhost_work_queue(&vsock->dev, &vsock->send_pkt_work);
269 
270 	rcu_read_unlock();
271 	return len;
272 }
273 
274 static int
275 vhost_transport_cancel_pkt(struct vsock_sock *vsk)
276 {
277 	struct vhost_vsock *vsock;
278 	struct virtio_vsock_pkt *pkt, *n;
279 	int cnt = 0;
280 	int ret = -ENODEV;
281 	LIST_HEAD(freeme);
282 
283 	rcu_read_lock();
284 
285 	/* Find the vhost_vsock according to guest context id  */
286 	vsock = vhost_vsock_get(vsk->remote_addr.svm_cid);
287 	if (!vsock)
288 		goto out;
289 
290 	spin_lock_bh(&vsock->send_pkt_list_lock);
291 	list_for_each_entry_safe(pkt, n, &vsock->send_pkt_list, list) {
292 		if (pkt->vsk != vsk)
293 			continue;
294 		list_move(&pkt->list, &freeme);
295 	}
296 	spin_unlock_bh(&vsock->send_pkt_list_lock);
297 
298 	list_for_each_entry_safe(pkt, n, &freeme, list) {
299 		if (pkt->reply)
300 			cnt++;
301 		list_del(&pkt->list);
302 		virtio_transport_free_pkt(pkt);
303 	}
304 
305 	if (cnt) {
306 		struct vhost_virtqueue *tx_vq = &vsock->vqs[VSOCK_VQ_TX];
307 		int new_cnt;
308 
309 		new_cnt = atomic_sub_return(cnt, &vsock->queued_replies);
310 		if (new_cnt + cnt >= tx_vq->num && new_cnt < tx_vq->num)
311 			vhost_poll_queue(&tx_vq->poll);
312 	}
313 
314 	ret = 0;
315 out:
316 	rcu_read_unlock();
317 	return ret;
318 }
319 
320 static struct virtio_vsock_pkt *
321 vhost_vsock_alloc_pkt(struct vhost_virtqueue *vq,
322 		      unsigned int out, unsigned int in)
323 {
324 	struct virtio_vsock_pkt *pkt;
325 	struct iov_iter iov_iter;
326 	size_t nbytes;
327 	size_t len;
328 
329 	if (in != 0) {
330 		vq_err(vq, "Expected 0 input buffers, got %u\n", in);
331 		return NULL;
332 	}
333 
334 	pkt = kzalloc(sizeof(*pkt), GFP_KERNEL);
335 	if (!pkt)
336 		return NULL;
337 
338 	len = iov_length(vq->iov, out);
339 	iov_iter_init(&iov_iter, WRITE, vq->iov, out, len);
340 
341 	nbytes = copy_from_iter(&pkt->hdr, sizeof(pkt->hdr), &iov_iter);
342 	if (nbytes != sizeof(pkt->hdr)) {
343 		vq_err(vq, "Expected %zu bytes for pkt->hdr, got %zu bytes\n",
344 		       sizeof(pkt->hdr), nbytes);
345 		kfree(pkt);
346 		return NULL;
347 	}
348 
349 	if (le16_to_cpu(pkt->hdr.type) == VIRTIO_VSOCK_TYPE_STREAM)
350 		pkt->len = le32_to_cpu(pkt->hdr.len);
351 
352 	/* No payload */
353 	if (!pkt->len)
354 		return pkt;
355 
356 	/* The pkt is too big */
357 	if (pkt->len > VIRTIO_VSOCK_MAX_PKT_BUF_SIZE) {
358 		kfree(pkt);
359 		return NULL;
360 	}
361 
362 	pkt->buf = kmalloc(pkt->len, GFP_KERNEL);
363 	if (!pkt->buf) {
364 		kfree(pkt);
365 		return NULL;
366 	}
367 
368 	pkt->buf_len = pkt->len;
369 
370 	nbytes = copy_from_iter(pkt->buf, pkt->len, &iov_iter);
371 	if (nbytes != pkt->len) {
372 		vq_err(vq, "Expected %u byte payload, got %zu bytes\n",
373 		       pkt->len, nbytes);
374 		virtio_transport_free_pkt(pkt);
375 		return NULL;
376 	}
377 
378 	return pkt;
379 }
380 
381 /* Is there space left for replies to rx packets? */
382 static bool vhost_vsock_more_replies(struct vhost_vsock *vsock)
383 {
384 	struct vhost_virtqueue *vq = &vsock->vqs[VSOCK_VQ_TX];
385 	int val;
386 
387 	smp_rmb(); /* paired with atomic_inc() and atomic_dec_return() */
388 	val = atomic_read(&vsock->queued_replies);
389 
390 	return val < vq->num;
391 }
392 
393 static struct virtio_transport vhost_transport = {
394 	.transport = {
395 		.module                   = THIS_MODULE,
396 
397 		.get_local_cid            = vhost_transport_get_local_cid,
398 
399 		.init                     = virtio_transport_do_socket_init,
400 		.destruct                 = virtio_transport_destruct,
401 		.release                  = virtio_transport_release,
402 		.connect                  = virtio_transport_connect,
403 		.shutdown                 = virtio_transport_shutdown,
404 		.cancel_pkt               = vhost_transport_cancel_pkt,
405 
406 		.dgram_enqueue            = virtio_transport_dgram_enqueue,
407 		.dgram_dequeue            = virtio_transport_dgram_dequeue,
408 		.dgram_bind               = virtio_transport_dgram_bind,
409 		.dgram_allow              = virtio_transport_dgram_allow,
410 
411 		.stream_enqueue           = virtio_transport_stream_enqueue,
412 		.stream_dequeue           = virtio_transport_stream_dequeue,
413 		.stream_has_data          = virtio_transport_stream_has_data,
414 		.stream_has_space         = virtio_transport_stream_has_space,
415 		.stream_rcvhiwat          = virtio_transport_stream_rcvhiwat,
416 		.stream_is_active         = virtio_transport_stream_is_active,
417 		.stream_allow             = virtio_transport_stream_allow,
418 
419 		.notify_poll_in           = virtio_transport_notify_poll_in,
420 		.notify_poll_out          = virtio_transport_notify_poll_out,
421 		.notify_recv_init         = virtio_transport_notify_recv_init,
422 		.notify_recv_pre_block    = virtio_transport_notify_recv_pre_block,
423 		.notify_recv_pre_dequeue  = virtio_transport_notify_recv_pre_dequeue,
424 		.notify_recv_post_dequeue = virtio_transport_notify_recv_post_dequeue,
425 		.notify_send_init         = virtio_transport_notify_send_init,
426 		.notify_send_pre_block    = virtio_transport_notify_send_pre_block,
427 		.notify_send_pre_enqueue  = virtio_transport_notify_send_pre_enqueue,
428 		.notify_send_post_enqueue = virtio_transport_notify_send_post_enqueue,
429 		.notify_buffer_size       = virtio_transport_notify_buffer_size,
430 
431 	},
432 
433 	.send_pkt = vhost_transport_send_pkt,
434 };
435 
436 static void vhost_vsock_handle_tx_kick(struct vhost_work *work)
437 {
438 	struct vhost_virtqueue *vq = container_of(work, struct vhost_virtqueue,
439 						  poll.work);
440 	struct vhost_vsock *vsock = container_of(vq->dev, struct vhost_vsock,
441 						 dev);
442 	struct virtio_vsock_pkt *pkt;
443 	int head, pkts = 0, total_len = 0;
444 	unsigned int out, in;
445 	bool added = false;
446 
447 	mutex_lock(&vq->mutex);
448 
449 	if (!vhost_vq_get_backend(vq))
450 		goto out;
451 
452 	vhost_disable_notify(&vsock->dev, vq);
453 	do {
454 		u32 len;
455 
456 		if (!vhost_vsock_more_replies(vsock)) {
457 			/* Stop tx until the device processes already
458 			 * pending replies.  Leave tx virtqueue
459 			 * callbacks disabled.
460 			 */
461 			goto no_more_replies;
462 		}
463 
464 		head = vhost_get_vq_desc(vq, vq->iov, ARRAY_SIZE(vq->iov),
465 					 &out, &in, NULL, NULL);
466 		if (head < 0)
467 			break;
468 
469 		if (head == vq->num) {
470 			if (unlikely(vhost_enable_notify(&vsock->dev, vq))) {
471 				vhost_disable_notify(&vsock->dev, vq);
472 				continue;
473 			}
474 			break;
475 		}
476 
477 		pkt = vhost_vsock_alloc_pkt(vq, out, in);
478 		if (!pkt) {
479 			vq_err(vq, "Faulted on pkt\n");
480 			continue;
481 		}
482 
483 		len = pkt->len;
484 
485 		/* Deliver to monitoring devices all received packets */
486 		virtio_transport_deliver_tap_pkt(pkt);
487 
488 		/* Only accept correctly addressed packets */
489 		if (le64_to_cpu(pkt->hdr.src_cid) == vsock->guest_cid &&
490 		    le64_to_cpu(pkt->hdr.dst_cid) ==
491 		    vhost_transport_get_local_cid())
492 			virtio_transport_recv_pkt(&vhost_transport, pkt);
493 		else
494 			virtio_transport_free_pkt(pkt);
495 
496 		len += sizeof(pkt->hdr);
497 		vhost_add_used(vq, head, len);
498 		total_len += len;
499 		added = true;
500 	} while(likely(!vhost_exceeds_weight(vq, ++pkts, total_len)));
501 
502 no_more_replies:
503 	if (added)
504 		vhost_signal(&vsock->dev, vq);
505 
506 out:
507 	mutex_unlock(&vq->mutex);
508 }
509 
510 static void vhost_vsock_handle_rx_kick(struct vhost_work *work)
511 {
512 	struct vhost_virtqueue *vq = container_of(work, struct vhost_virtqueue,
513 						poll.work);
514 	struct vhost_vsock *vsock = container_of(vq->dev, struct vhost_vsock,
515 						 dev);
516 
517 	vhost_transport_do_send_pkt(vsock, vq);
518 }
519 
520 static int vhost_vsock_start(struct vhost_vsock *vsock)
521 {
522 	struct vhost_virtqueue *vq;
523 	size_t i;
524 	int ret;
525 
526 	mutex_lock(&vsock->dev.mutex);
527 
528 	ret = vhost_dev_check_owner(&vsock->dev);
529 	if (ret)
530 		goto err;
531 
532 	for (i = 0; i < ARRAY_SIZE(vsock->vqs); i++) {
533 		vq = &vsock->vqs[i];
534 
535 		mutex_lock(&vq->mutex);
536 
537 		if (!vhost_vq_access_ok(vq)) {
538 			ret = -EFAULT;
539 			goto err_vq;
540 		}
541 
542 		if (!vhost_vq_get_backend(vq)) {
543 			vhost_vq_set_backend(vq, vsock);
544 			ret = vhost_vq_init_access(vq);
545 			if (ret)
546 				goto err_vq;
547 		}
548 
549 		mutex_unlock(&vq->mutex);
550 	}
551 
552 	/* Some packets may have been queued before the device was started,
553 	 * let's kick the send worker to send them.
554 	 */
555 	vhost_work_queue(&vsock->dev, &vsock->send_pkt_work);
556 
557 	mutex_unlock(&vsock->dev.mutex);
558 	return 0;
559 
560 err_vq:
561 	vhost_vq_set_backend(vq, NULL);
562 	mutex_unlock(&vq->mutex);
563 
564 	for (i = 0; i < ARRAY_SIZE(vsock->vqs); i++) {
565 		vq = &vsock->vqs[i];
566 
567 		mutex_lock(&vq->mutex);
568 		vhost_vq_set_backend(vq, NULL);
569 		mutex_unlock(&vq->mutex);
570 	}
571 err:
572 	mutex_unlock(&vsock->dev.mutex);
573 	return ret;
574 }
575 
576 static int vhost_vsock_stop(struct vhost_vsock *vsock)
577 {
578 	size_t i;
579 	int ret;
580 
581 	mutex_lock(&vsock->dev.mutex);
582 
583 	ret = vhost_dev_check_owner(&vsock->dev);
584 	if (ret)
585 		goto err;
586 
587 	for (i = 0; i < ARRAY_SIZE(vsock->vqs); i++) {
588 		struct vhost_virtqueue *vq = &vsock->vqs[i];
589 
590 		mutex_lock(&vq->mutex);
591 		vhost_vq_set_backend(vq, NULL);
592 		mutex_unlock(&vq->mutex);
593 	}
594 
595 err:
596 	mutex_unlock(&vsock->dev.mutex);
597 	return ret;
598 }
599 
600 static void vhost_vsock_free(struct vhost_vsock *vsock)
601 {
602 	kvfree(vsock);
603 }
604 
605 static int vhost_vsock_dev_open(struct inode *inode, struct file *file)
606 {
607 	struct vhost_virtqueue **vqs;
608 	struct vhost_vsock *vsock;
609 	int ret;
610 
611 	/* This struct is large and allocation could fail, fall back to vmalloc
612 	 * if there is no other way.
613 	 */
614 	vsock = kvmalloc(sizeof(*vsock), GFP_KERNEL | __GFP_RETRY_MAYFAIL);
615 	if (!vsock)
616 		return -ENOMEM;
617 
618 	vqs = kmalloc_array(ARRAY_SIZE(vsock->vqs), sizeof(*vqs), GFP_KERNEL);
619 	if (!vqs) {
620 		ret = -ENOMEM;
621 		goto out;
622 	}
623 
624 	vsock->guest_cid = 0; /* no CID assigned yet */
625 
626 	atomic_set(&vsock->queued_replies, 0);
627 
628 	vqs[VSOCK_VQ_TX] = &vsock->vqs[VSOCK_VQ_TX];
629 	vqs[VSOCK_VQ_RX] = &vsock->vqs[VSOCK_VQ_RX];
630 	vsock->vqs[VSOCK_VQ_TX].handle_kick = vhost_vsock_handle_tx_kick;
631 	vsock->vqs[VSOCK_VQ_RX].handle_kick = vhost_vsock_handle_rx_kick;
632 
633 	vhost_dev_init(&vsock->dev, vqs, ARRAY_SIZE(vsock->vqs),
634 		       UIO_MAXIOV, VHOST_VSOCK_PKT_WEIGHT,
635 		       VHOST_VSOCK_WEIGHT, true, NULL);
636 
637 	file->private_data = vsock;
638 	spin_lock_init(&vsock->send_pkt_list_lock);
639 	INIT_LIST_HEAD(&vsock->send_pkt_list);
640 	vhost_work_init(&vsock->send_pkt_work, vhost_transport_send_pkt_work);
641 	return 0;
642 
643 out:
644 	vhost_vsock_free(vsock);
645 	return ret;
646 }
647 
648 static void vhost_vsock_flush(struct vhost_vsock *vsock)
649 {
650 	int i;
651 
652 	for (i = 0; i < ARRAY_SIZE(vsock->vqs); i++)
653 		if (vsock->vqs[i].handle_kick)
654 			vhost_poll_flush(&vsock->vqs[i].poll);
655 	vhost_work_flush(&vsock->dev, &vsock->send_pkt_work);
656 }
657 
658 static void vhost_vsock_reset_orphans(struct sock *sk)
659 {
660 	struct vsock_sock *vsk = vsock_sk(sk);
661 
662 	/* vmci_transport.c doesn't take sk_lock here either.  At least we're
663 	 * under vsock_table_lock so the sock cannot disappear while we're
664 	 * executing.
665 	 */
666 
667 	/* If the peer is still valid, no need to reset connection */
668 	if (vhost_vsock_get(vsk->remote_addr.svm_cid))
669 		return;
670 
671 	/* If the close timeout is pending, let it expire.  This avoids races
672 	 * with the timeout callback.
673 	 */
674 	if (vsk->close_work_scheduled)
675 		return;
676 
677 	sock_set_flag(sk, SOCK_DONE);
678 	vsk->peer_shutdown = SHUTDOWN_MASK;
679 	sk->sk_state = SS_UNCONNECTED;
680 	sk->sk_err = ECONNRESET;
681 	sk->sk_error_report(sk);
682 }
683 
684 static int vhost_vsock_dev_release(struct inode *inode, struct file *file)
685 {
686 	struct vhost_vsock *vsock = file->private_data;
687 
688 	mutex_lock(&vhost_vsock_mutex);
689 	if (vsock->guest_cid)
690 		hash_del_rcu(&vsock->hash);
691 	mutex_unlock(&vhost_vsock_mutex);
692 
693 	/* Wait for other CPUs to finish using vsock */
694 	synchronize_rcu();
695 
696 	/* Iterating over all connections for all CIDs to find orphans is
697 	 * inefficient.  Room for improvement here. */
698 	vsock_for_each_connected_socket(vhost_vsock_reset_orphans);
699 
700 	vhost_vsock_stop(vsock);
701 	vhost_vsock_flush(vsock);
702 	vhost_dev_stop(&vsock->dev);
703 
704 	spin_lock_bh(&vsock->send_pkt_list_lock);
705 	while (!list_empty(&vsock->send_pkt_list)) {
706 		struct virtio_vsock_pkt *pkt;
707 
708 		pkt = list_first_entry(&vsock->send_pkt_list,
709 				struct virtio_vsock_pkt, list);
710 		list_del_init(&pkt->list);
711 		virtio_transport_free_pkt(pkt);
712 	}
713 	spin_unlock_bh(&vsock->send_pkt_list_lock);
714 
715 	vhost_dev_cleanup(&vsock->dev);
716 	kfree(vsock->dev.vqs);
717 	vhost_vsock_free(vsock);
718 	return 0;
719 }
720 
721 static int vhost_vsock_set_cid(struct vhost_vsock *vsock, u64 guest_cid)
722 {
723 	struct vhost_vsock *other;
724 
725 	/* Refuse reserved CIDs */
726 	if (guest_cid <= VMADDR_CID_HOST ||
727 	    guest_cid == U32_MAX)
728 		return -EINVAL;
729 
730 	/* 64-bit CIDs are not yet supported */
731 	if (guest_cid > U32_MAX)
732 		return -EINVAL;
733 
734 	/* Refuse if CID is assigned to the guest->host transport (i.e. nested
735 	 * VM), to make the loopback work.
736 	 */
737 	if (vsock_find_cid(guest_cid))
738 		return -EADDRINUSE;
739 
740 	/* Refuse if CID is already in use */
741 	mutex_lock(&vhost_vsock_mutex);
742 	other = vhost_vsock_get(guest_cid);
743 	if (other && other != vsock) {
744 		mutex_unlock(&vhost_vsock_mutex);
745 		return -EADDRINUSE;
746 	}
747 
748 	if (vsock->guest_cid)
749 		hash_del_rcu(&vsock->hash);
750 
751 	vsock->guest_cid = guest_cid;
752 	hash_add_rcu(vhost_vsock_hash, &vsock->hash, vsock->guest_cid);
753 	mutex_unlock(&vhost_vsock_mutex);
754 
755 	return 0;
756 }
757 
758 static int vhost_vsock_set_features(struct vhost_vsock *vsock, u64 features)
759 {
760 	struct vhost_virtqueue *vq;
761 	int i;
762 
763 	if (features & ~VHOST_VSOCK_FEATURES)
764 		return -EOPNOTSUPP;
765 
766 	mutex_lock(&vsock->dev.mutex);
767 	if ((features & (1 << VHOST_F_LOG_ALL)) &&
768 	    !vhost_log_access_ok(&vsock->dev)) {
769 		mutex_unlock(&vsock->dev.mutex);
770 		return -EFAULT;
771 	}
772 
773 	for (i = 0; i < ARRAY_SIZE(vsock->vqs); i++) {
774 		vq = &vsock->vqs[i];
775 		mutex_lock(&vq->mutex);
776 		vq->acked_features = features;
777 		mutex_unlock(&vq->mutex);
778 	}
779 	mutex_unlock(&vsock->dev.mutex);
780 	return 0;
781 }
782 
783 static long vhost_vsock_dev_ioctl(struct file *f, unsigned int ioctl,
784 				  unsigned long arg)
785 {
786 	struct vhost_vsock *vsock = f->private_data;
787 	void __user *argp = (void __user *)arg;
788 	u64 guest_cid;
789 	u64 features;
790 	int start;
791 	int r;
792 
793 	switch (ioctl) {
794 	case VHOST_VSOCK_SET_GUEST_CID:
795 		if (copy_from_user(&guest_cid, argp, sizeof(guest_cid)))
796 			return -EFAULT;
797 		return vhost_vsock_set_cid(vsock, guest_cid);
798 	case VHOST_VSOCK_SET_RUNNING:
799 		if (copy_from_user(&start, argp, sizeof(start)))
800 			return -EFAULT;
801 		if (start)
802 			return vhost_vsock_start(vsock);
803 		else
804 			return vhost_vsock_stop(vsock);
805 	case VHOST_GET_FEATURES:
806 		features = VHOST_VSOCK_FEATURES;
807 		if (copy_to_user(argp, &features, sizeof(features)))
808 			return -EFAULT;
809 		return 0;
810 	case VHOST_SET_FEATURES:
811 		if (copy_from_user(&features, argp, sizeof(features)))
812 			return -EFAULT;
813 		return vhost_vsock_set_features(vsock, features);
814 	default:
815 		mutex_lock(&vsock->dev.mutex);
816 		r = vhost_dev_ioctl(&vsock->dev, ioctl, argp);
817 		if (r == -ENOIOCTLCMD)
818 			r = vhost_vring_ioctl(&vsock->dev, ioctl, argp);
819 		else
820 			vhost_vsock_flush(vsock);
821 		mutex_unlock(&vsock->dev.mutex);
822 		return r;
823 	}
824 }
825 
826 static const struct file_operations vhost_vsock_fops = {
827 	.owner          = THIS_MODULE,
828 	.open           = vhost_vsock_dev_open,
829 	.release        = vhost_vsock_dev_release,
830 	.llseek		= noop_llseek,
831 	.unlocked_ioctl = vhost_vsock_dev_ioctl,
832 	.compat_ioctl   = compat_ptr_ioctl,
833 };
834 
835 static struct miscdevice vhost_vsock_misc = {
836 	.minor = VHOST_VSOCK_MINOR,
837 	.name = "vhost-vsock",
838 	.fops = &vhost_vsock_fops,
839 };
840 
841 static int __init vhost_vsock_init(void)
842 {
843 	int ret;
844 
845 	ret = vsock_core_register(&vhost_transport.transport,
846 				  VSOCK_TRANSPORT_F_H2G);
847 	if (ret < 0)
848 		return ret;
849 	return misc_register(&vhost_vsock_misc);
850 };
851 
852 static void __exit vhost_vsock_exit(void)
853 {
854 	misc_deregister(&vhost_vsock_misc);
855 	vsock_core_unregister(&vhost_transport.transport);
856 };
857 
858 module_init(vhost_vsock_init);
859 module_exit(vhost_vsock_exit);
860 MODULE_LICENSE("GPL v2");
861 MODULE_AUTHOR("Asias He");
862 MODULE_DESCRIPTION("vhost transport for vsock ");
863 MODULE_ALIAS_MISCDEV(VHOST_VSOCK_MINOR);
864 MODULE_ALIAS("devname:vhost-vsock");
865