xref: /linux/drivers/vhost/vsock.c (revision e5e95a7639ed5f7dc3e404858ad7910de5fa2057)
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 			       (1ULL << VIRTIO_F_ACCESS_PLATFORM)
35 };
36 
37 enum {
38 	VHOST_VSOCK_BACKEND_FEATURES = (1ULL << VHOST_BACKEND_F_IOTLB_MSG_V2)
39 };
40 
41 /* Used to track all the vhost_vsock instances on the system. */
42 static DEFINE_MUTEX(vhost_vsock_mutex);
43 static DEFINE_READ_MOSTLY_HASHTABLE(vhost_vsock_hash, 8);
44 
45 struct vhost_vsock {
46 	struct vhost_dev dev;
47 	struct vhost_virtqueue vqs[2];
48 
49 	/* Link to global vhost_vsock_hash, writes use vhost_vsock_mutex */
50 	struct hlist_node hash;
51 
52 	struct vhost_work send_pkt_work;
53 	spinlock_t send_pkt_list_lock;
54 	struct list_head send_pkt_list;	/* host->guest pending packets */
55 
56 	atomic_t queued_replies;
57 
58 	u32 guest_cid;
59 };
60 
61 static u32 vhost_transport_get_local_cid(void)
62 {
63 	return VHOST_VSOCK_DEFAULT_HOST_CID;
64 }
65 
66 /* Callers that dereference the return value must hold vhost_vsock_mutex or the
67  * RCU read lock.
68  */
69 static struct vhost_vsock *vhost_vsock_get(u32 guest_cid)
70 {
71 	struct vhost_vsock *vsock;
72 
73 	hash_for_each_possible_rcu(vhost_vsock_hash, vsock, hash, guest_cid) {
74 		u32 other_cid = vsock->guest_cid;
75 
76 		/* Skip instances that have no CID yet */
77 		if (other_cid == 0)
78 			continue;
79 
80 		if (other_cid == guest_cid)
81 			return vsock;
82 
83 	}
84 
85 	return NULL;
86 }
87 
88 static void
89 vhost_transport_do_send_pkt(struct vhost_vsock *vsock,
90 			    struct vhost_virtqueue *vq)
91 {
92 	struct vhost_virtqueue *tx_vq = &vsock->vqs[VSOCK_VQ_TX];
93 	int pkts = 0, total_len = 0;
94 	bool added = false;
95 	bool restart_tx = false;
96 
97 	mutex_lock(&vq->mutex);
98 
99 	if (!vhost_vq_get_backend(vq))
100 		goto out;
101 
102 	if (!vq_meta_prefetch(vq))
103 		goto out;
104 
105 	/* Avoid further vmexits, we're already processing the virtqueue */
106 	vhost_disable_notify(&vsock->dev, vq);
107 
108 	do {
109 		struct virtio_vsock_pkt *pkt;
110 		struct iov_iter iov_iter;
111 		unsigned out, in;
112 		size_t nbytes;
113 		size_t iov_len, payload_len;
114 		int head;
115 
116 		spin_lock_bh(&vsock->send_pkt_list_lock);
117 		if (list_empty(&vsock->send_pkt_list)) {
118 			spin_unlock_bh(&vsock->send_pkt_list_lock);
119 			vhost_enable_notify(&vsock->dev, vq);
120 			break;
121 		}
122 
123 		pkt = list_first_entry(&vsock->send_pkt_list,
124 				       struct virtio_vsock_pkt, list);
125 		list_del_init(&pkt->list);
126 		spin_unlock_bh(&vsock->send_pkt_list_lock);
127 
128 		head = vhost_get_vq_desc(vq, vq->iov, ARRAY_SIZE(vq->iov),
129 					 &out, &in, NULL, NULL);
130 		if (head < 0) {
131 			spin_lock_bh(&vsock->send_pkt_list_lock);
132 			list_add(&pkt->list, &vsock->send_pkt_list);
133 			spin_unlock_bh(&vsock->send_pkt_list_lock);
134 			break;
135 		}
136 
137 		if (head == vq->num) {
138 			spin_lock_bh(&vsock->send_pkt_list_lock);
139 			list_add(&pkt->list, &vsock->send_pkt_list);
140 			spin_unlock_bh(&vsock->send_pkt_list_lock);
141 
142 			/* We cannot finish yet if more buffers snuck in while
143 			 * re-enabling notify.
144 			 */
145 			if (unlikely(vhost_enable_notify(&vsock->dev, vq))) {
146 				vhost_disable_notify(&vsock->dev, vq);
147 				continue;
148 			}
149 			break;
150 		}
151 
152 		if (out) {
153 			virtio_transport_free_pkt(pkt);
154 			vq_err(vq, "Expected 0 output buffers, got %u\n", out);
155 			break;
156 		}
157 
158 		iov_len = iov_length(&vq->iov[out], in);
159 		if (iov_len < sizeof(pkt->hdr)) {
160 			virtio_transport_free_pkt(pkt);
161 			vq_err(vq, "Buffer len [%zu] too small\n", iov_len);
162 			break;
163 		}
164 
165 		iov_iter_init(&iov_iter, READ, &vq->iov[out], in, iov_len);
166 		payload_len = pkt->len - pkt->off;
167 
168 		/* If the packet is greater than the space available in the
169 		 * buffer, we split it using multiple buffers.
170 		 */
171 		if (payload_len > iov_len - sizeof(pkt->hdr))
172 			payload_len = iov_len - sizeof(pkt->hdr);
173 
174 		/* Set the correct length in the header */
175 		pkt->hdr.len = cpu_to_le32(payload_len);
176 
177 		nbytes = copy_to_iter(&pkt->hdr, sizeof(pkt->hdr), &iov_iter);
178 		if (nbytes != sizeof(pkt->hdr)) {
179 			virtio_transport_free_pkt(pkt);
180 			vq_err(vq, "Faulted on copying pkt hdr\n");
181 			break;
182 		}
183 
184 		nbytes = copy_to_iter(pkt->buf + pkt->off, payload_len,
185 				      &iov_iter);
186 		if (nbytes != payload_len) {
187 			virtio_transport_free_pkt(pkt);
188 			vq_err(vq, "Faulted on copying pkt buf\n");
189 			break;
190 		}
191 
192 		/* Deliver to monitoring devices all packets that we
193 		 * will transmit.
194 		 */
195 		virtio_transport_deliver_tap_pkt(pkt);
196 
197 		vhost_add_used(vq, head, sizeof(pkt->hdr) + payload_len);
198 		added = true;
199 
200 		pkt->off += payload_len;
201 		total_len += payload_len;
202 
203 		/* If we didn't send all the payload we can requeue the packet
204 		 * to send it with the next available buffer.
205 		 */
206 		if (pkt->off < pkt->len) {
207 			/* We are queueing the same virtio_vsock_pkt to handle
208 			 * the remaining bytes, and we want to deliver it
209 			 * to monitoring devices in the next iteration.
210 			 */
211 			pkt->tap_delivered = false;
212 
213 			spin_lock_bh(&vsock->send_pkt_list_lock);
214 			list_add(&pkt->list, &vsock->send_pkt_list);
215 			spin_unlock_bh(&vsock->send_pkt_list_lock);
216 		} else {
217 			if (pkt->reply) {
218 				int val;
219 
220 				val = atomic_dec_return(&vsock->queued_replies);
221 
222 				/* Do we have resources to resume tx
223 				 * processing?
224 				 */
225 				if (val + 1 == tx_vq->num)
226 					restart_tx = true;
227 			}
228 
229 			virtio_transport_free_pkt(pkt);
230 		}
231 	} while(likely(!vhost_exceeds_weight(vq, ++pkts, total_len)));
232 	if (added)
233 		vhost_signal(&vsock->dev, vq);
234 
235 out:
236 	mutex_unlock(&vq->mutex);
237 
238 	if (restart_tx)
239 		vhost_poll_queue(&tx_vq->poll);
240 }
241 
242 static void vhost_transport_send_pkt_work(struct vhost_work *work)
243 {
244 	struct vhost_virtqueue *vq;
245 	struct vhost_vsock *vsock;
246 
247 	vsock = container_of(work, struct vhost_vsock, send_pkt_work);
248 	vq = &vsock->vqs[VSOCK_VQ_RX];
249 
250 	vhost_transport_do_send_pkt(vsock, vq);
251 }
252 
253 static int
254 vhost_transport_send_pkt(struct virtio_vsock_pkt *pkt)
255 {
256 	struct vhost_vsock *vsock;
257 	int len = pkt->len;
258 
259 	rcu_read_lock();
260 
261 	/* Find the vhost_vsock according to guest context id  */
262 	vsock = vhost_vsock_get(le64_to_cpu(pkt->hdr.dst_cid));
263 	if (!vsock) {
264 		rcu_read_unlock();
265 		virtio_transport_free_pkt(pkt);
266 		return -ENODEV;
267 	}
268 
269 	if (pkt->reply)
270 		atomic_inc(&vsock->queued_replies);
271 
272 	spin_lock_bh(&vsock->send_pkt_list_lock);
273 	list_add_tail(&pkt->list, &vsock->send_pkt_list);
274 	spin_unlock_bh(&vsock->send_pkt_list_lock);
275 
276 	vhost_work_queue(&vsock->dev, &vsock->send_pkt_work);
277 
278 	rcu_read_unlock();
279 	return len;
280 }
281 
282 static int
283 vhost_transport_cancel_pkt(struct vsock_sock *vsk)
284 {
285 	struct vhost_vsock *vsock;
286 	struct virtio_vsock_pkt *pkt, *n;
287 	int cnt = 0;
288 	int ret = -ENODEV;
289 	LIST_HEAD(freeme);
290 
291 	rcu_read_lock();
292 
293 	/* Find the vhost_vsock according to guest context id  */
294 	vsock = vhost_vsock_get(vsk->remote_addr.svm_cid);
295 	if (!vsock)
296 		goto out;
297 
298 	spin_lock_bh(&vsock->send_pkt_list_lock);
299 	list_for_each_entry_safe(pkt, n, &vsock->send_pkt_list, list) {
300 		if (pkt->vsk != vsk)
301 			continue;
302 		list_move(&pkt->list, &freeme);
303 	}
304 	spin_unlock_bh(&vsock->send_pkt_list_lock);
305 
306 	list_for_each_entry_safe(pkt, n, &freeme, list) {
307 		if (pkt->reply)
308 			cnt++;
309 		list_del(&pkt->list);
310 		virtio_transport_free_pkt(pkt);
311 	}
312 
313 	if (cnt) {
314 		struct vhost_virtqueue *tx_vq = &vsock->vqs[VSOCK_VQ_TX];
315 		int new_cnt;
316 
317 		new_cnt = atomic_sub_return(cnt, &vsock->queued_replies);
318 		if (new_cnt + cnt >= tx_vq->num && new_cnt < tx_vq->num)
319 			vhost_poll_queue(&tx_vq->poll);
320 	}
321 
322 	ret = 0;
323 out:
324 	rcu_read_unlock();
325 	return ret;
326 }
327 
328 static struct virtio_vsock_pkt *
329 vhost_vsock_alloc_pkt(struct vhost_virtqueue *vq,
330 		      unsigned int out, unsigned int in)
331 {
332 	struct virtio_vsock_pkt *pkt;
333 	struct iov_iter iov_iter;
334 	size_t nbytes;
335 	size_t len;
336 
337 	if (in != 0) {
338 		vq_err(vq, "Expected 0 input buffers, got %u\n", in);
339 		return NULL;
340 	}
341 
342 	pkt = kzalloc(sizeof(*pkt), GFP_KERNEL);
343 	if (!pkt)
344 		return NULL;
345 
346 	len = iov_length(vq->iov, out);
347 	iov_iter_init(&iov_iter, WRITE, vq->iov, out, len);
348 
349 	nbytes = copy_from_iter(&pkt->hdr, sizeof(pkt->hdr), &iov_iter);
350 	if (nbytes != sizeof(pkt->hdr)) {
351 		vq_err(vq, "Expected %zu bytes for pkt->hdr, got %zu bytes\n",
352 		       sizeof(pkt->hdr), nbytes);
353 		kfree(pkt);
354 		return NULL;
355 	}
356 
357 	if (le16_to_cpu(pkt->hdr.type) == VIRTIO_VSOCK_TYPE_STREAM)
358 		pkt->len = le32_to_cpu(pkt->hdr.len);
359 
360 	/* No payload */
361 	if (!pkt->len)
362 		return pkt;
363 
364 	/* The pkt is too big */
365 	if (pkt->len > VIRTIO_VSOCK_MAX_PKT_BUF_SIZE) {
366 		kfree(pkt);
367 		return NULL;
368 	}
369 
370 	pkt->buf = kmalloc(pkt->len, GFP_KERNEL);
371 	if (!pkt->buf) {
372 		kfree(pkt);
373 		return NULL;
374 	}
375 
376 	pkt->buf_len = pkt->len;
377 
378 	nbytes = copy_from_iter(pkt->buf, pkt->len, &iov_iter);
379 	if (nbytes != pkt->len) {
380 		vq_err(vq, "Expected %u byte payload, got %zu bytes\n",
381 		       pkt->len, nbytes);
382 		virtio_transport_free_pkt(pkt);
383 		return NULL;
384 	}
385 
386 	return pkt;
387 }
388 
389 /* Is there space left for replies to rx packets? */
390 static bool vhost_vsock_more_replies(struct vhost_vsock *vsock)
391 {
392 	struct vhost_virtqueue *vq = &vsock->vqs[VSOCK_VQ_TX];
393 	int val;
394 
395 	smp_rmb(); /* paired with atomic_inc() and atomic_dec_return() */
396 	val = atomic_read(&vsock->queued_replies);
397 
398 	return val < vq->num;
399 }
400 
401 static struct virtio_transport vhost_transport = {
402 	.transport = {
403 		.module                   = THIS_MODULE,
404 
405 		.get_local_cid            = vhost_transport_get_local_cid,
406 
407 		.init                     = virtio_transport_do_socket_init,
408 		.destruct                 = virtio_transport_destruct,
409 		.release                  = virtio_transport_release,
410 		.connect                  = virtio_transport_connect,
411 		.shutdown                 = virtio_transport_shutdown,
412 		.cancel_pkt               = vhost_transport_cancel_pkt,
413 
414 		.dgram_enqueue            = virtio_transport_dgram_enqueue,
415 		.dgram_dequeue            = virtio_transport_dgram_dequeue,
416 		.dgram_bind               = virtio_transport_dgram_bind,
417 		.dgram_allow              = virtio_transport_dgram_allow,
418 
419 		.stream_enqueue           = virtio_transport_stream_enqueue,
420 		.stream_dequeue           = virtio_transport_stream_dequeue,
421 		.stream_has_data          = virtio_transport_stream_has_data,
422 		.stream_has_space         = virtio_transport_stream_has_space,
423 		.stream_rcvhiwat          = virtio_transport_stream_rcvhiwat,
424 		.stream_is_active         = virtio_transport_stream_is_active,
425 		.stream_allow             = virtio_transport_stream_allow,
426 
427 		.notify_poll_in           = virtio_transport_notify_poll_in,
428 		.notify_poll_out          = virtio_transport_notify_poll_out,
429 		.notify_recv_init         = virtio_transport_notify_recv_init,
430 		.notify_recv_pre_block    = virtio_transport_notify_recv_pre_block,
431 		.notify_recv_pre_dequeue  = virtio_transport_notify_recv_pre_dequeue,
432 		.notify_recv_post_dequeue = virtio_transport_notify_recv_post_dequeue,
433 		.notify_send_init         = virtio_transport_notify_send_init,
434 		.notify_send_pre_block    = virtio_transport_notify_send_pre_block,
435 		.notify_send_pre_enqueue  = virtio_transport_notify_send_pre_enqueue,
436 		.notify_send_post_enqueue = virtio_transport_notify_send_post_enqueue,
437 		.notify_buffer_size       = virtio_transport_notify_buffer_size,
438 
439 	},
440 
441 	.send_pkt = vhost_transport_send_pkt,
442 };
443 
444 static void vhost_vsock_handle_tx_kick(struct vhost_work *work)
445 {
446 	struct vhost_virtqueue *vq = container_of(work, struct vhost_virtqueue,
447 						  poll.work);
448 	struct vhost_vsock *vsock = container_of(vq->dev, struct vhost_vsock,
449 						 dev);
450 	struct virtio_vsock_pkt *pkt;
451 	int head, pkts = 0, total_len = 0;
452 	unsigned int out, in;
453 	bool added = false;
454 
455 	mutex_lock(&vq->mutex);
456 
457 	if (!vhost_vq_get_backend(vq))
458 		goto out;
459 
460 	if (!vq_meta_prefetch(vq))
461 		goto out;
462 
463 	vhost_disable_notify(&vsock->dev, vq);
464 	do {
465 		u32 len;
466 
467 		if (!vhost_vsock_more_replies(vsock)) {
468 			/* Stop tx until the device processes already
469 			 * pending replies.  Leave tx virtqueue
470 			 * callbacks disabled.
471 			 */
472 			goto no_more_replies;
473 		}
474 
475 		head = vhost_get_vq_desc(vq, vq->iov, ARRAY_SIZE(vq->iov),
476 					 &out, &in, NULL, NULL);
477 		if (head < 0)
478 			break;
479 
480 		if (head == vq->num) {
481 			if (unlikely(vhost_enable_notify(&vsock->dev, vq))) {
482 				vhost_disable_notify(&vsock->dev, vq);
483 				continue;
484 			}
485 			break;
486 		}
487 
488 		pkt = vhost_vsock_alloc_pkt(vq, out, in);
489 		if (!pkt) {
490 			vq_err(vq, "Faulted on pkt\n");
491 			continue;
492 		}
493 
494 		len = pkt->len;
495 
496 		/* Deliver to monitoring devices all received packets */
497 		virtio_transport_deliver_tap_pkt(pkt);
498 
499 		/* Only accept correctly addressed packets */
500 		if (le64_to_cpu(pkt->hdr.src_cid) == vsock->guest_cid &&
501 		    le64_to_cpu(pkt->hdr.dst_cid) ==
502 		    vhost_transport_get_local_cid())
503 			virtio_transport_recv_pkt(&vhost_transport, pkt);
504 		else
505 			virtio_transport_free_pkt(pkt);
506 
507 		len += sizeof(pkt->hdr);
508 		vhost_add_used(vq, head, len);
509 		total_len += len;
510 		added = true;
511 	} while(likely(!vhost_exceeds_weight(vq, ++pkts, total_len)));
512 
513 no_more_replies:
514 	if (added)
515 		vhost_signal(&vsock->dev, vq);
516 
517 out:
518 	mutex_unlock(&vq->mutex);
519 }
520 
521 static void vhost_vsock_handle_rx_kick(struct vhost_work *work)
522 {
523 	struct vhost_virtqueue *vq = container_of(work, struct vhost_virtqueue,
524 						poll.work);
525 	struct vhost_vsock *vsock = container_of(vq->dev, struct vhost_vsock,
526 						 dev);
527 
528 	vhost_transport_do_send_pkt(vsock, vq);
529 }
530 
531 static int vhost_vsock_start(struct vhost_vsock *vsock)
532 {
533 	struct vhost_virtqueue *vq;
534 	size_t i;
535 	int ret;
536 
537 	mutex_lock(&vsock->dev.mutex);
538 
539 	ret = vhost_dev_check_owner(&vsock->dev);
540 	if (ret)
541 		goto err;
542 
543 	for (i = 0; i < ARRAY_SIZE(vsock->vqs); i++) {
544 		vq = &vsock->vqs[i];
545 
546 		mutex_lock(&vq->mutex);
547 
548 		if (!vhost_vq_access_ok(vq)) {
549 			ret = -EFAULT;
550 			goto err_vq;
551 		}
552 
553 		if (!vhost_vq_get_backend(vq)) {
554 			vhost_vq_set_backend(vq, vsock);
555 			ret = vhost_vq_init_access(vq);
556 			if (ret)
557 				goto err_vq;
558 		}
559 
560 		mutex_unlock(&vq->mutex);
561 	}
562 
563 	/* Some packets may have been queued before the device was started,
564 	 * let's kick the send worker to send them.
565 	 */
566 	vhost_work_queue(&vsock->dev, &vsock->send_pkt_work);
567 
568 	mutex_unlock(&vsock->dev.mutex);
569 	return 0;
570 
571 err_vq:
572 	vhost_vq_set_backend(vq, NULL);
573 	mutex_unlock(&vq->mutex);
574 
575 	for (i = 0; i < ARRAY_SIZE(vsock->vqs); i++) {
576 		vq = &vsock->vqs[i];
577 
578 		mutex_lock(&vq->mutex);
579 		vhost_vq_set_backend(vq, NULL);
580 		mutex_unlock(&vq->mutex);
581 	}
582 err:
583 	mutex_unlock(&vsock->dev.mutex);
584 	return ret;
585 }
586 
587 static int vhost_vsock_stop(struct vhost_vsock *vsock)
588 {
589 	size_t i;
590 	int ret;
591 
592 	mutex_lock(&vsock->dev.mutex);
593 
594 	ret = vhost_dev_check_owner(&vsock->dev);
595 	if (ret)
596 		goto err;
597 
598 	for (i = 0; i < ARRAY_SIZE(vsock->vqs); i++) {
599 		struct vhost_virtqueue *vq = &vsock->vqs[i];
600 
601 		mutex_lock(&vq->mutex);
602 		vhost_vq_set_backend(vq, NULL);
603 		mutex_unlock(&vq->mutex);
604 	}
605 
606 err:
607 	mutex_unlock(&vsock->dev.mutex);
608 	return ret;
609 }
610 
611 static void vhost_vsock_free(struct vhost_vsock *vsock)
612 {
613 	kvfree(vsock);
614 }
615 
616 static int vhost_vsock_dev_open(struct inode *inode, struct file *file)
617 {
618 	struct vhost_virtqueue **vqs;
619 	struct vhost_vsock *vsock;
620 	int ret;
621 
622 	/* This struct is large and allocation could fail, fall back to vmalloc
623 	 * if there is no other way.
624 	 */
625 	vsock = kvmalloc(sizeof(*vsock), GFP_KERNEL | __GFP_RETRY_MAYFAIL);
626 	if (!vsock)
627 		return -ENOMEM;
628 
629 	vqs = kmalloc_array(ARRAY_SIZE(vsock->vqs), sizeof(*vqs), GFP_KERNEL);
630 	if (!vqs) {
631 		ret = -ENOMEM;
632 		goto out;
633 	}
634 
635 	vsock->guest_cid = 0; /* no CID assigned yet */
636 
637 	atomic_set(&vsock->queued_replies, 0);
638 
639 	vqs[VSOCK_VQ_TX] = &vsock->vqs[VSOCK_VQ_TX];
640 	vqs[VSOCK_VQ_RX] = &vsock->vqs[VSOCK_VQ_RX];
641 	vsock->vqs[VSOCK_VQ_TX].handle_kick = vhost_vsock_handle_tx_kick;
642 	vsock->vqs[VSOCK_VQ_RX].handle_kick = vhost_vsock_handle_rx_kick;
643 
644 	vhost_dev_init(&vsock->dev, vqs, ARRAY_SIZE(vsock->vqs),
645 		       UIO_MAXIOV, VHOST_VSOCK_PKT_WEIGHT,
646 		       VHOST_VSOCK_WEIGHT, true, NULL);
647 
648 	file->private_data = vsock;
649 	spin_lock_init(&vsock->send_pkt_list_lock);
650 	INIT_LIST_HEAD(&vsock->send_pkt_list);
651 	vhost_work_init(&vsock->send_pkt_work, vhost_transport_send_pkt_work);
652 	return 0;
653 
654 out:
655 	vhost_vsock_free(vsock);
656 	return ret;
657 }
658 
659 static void vhost_vsock_flush(struct vhost_vsock *vsock)
660 {
661 	int i;
662 
663 	for (i = 0; i < ARRAY_SIZE(vsock->vqs); i++)
664 		if (vsock->vqs[i].handle_kick)
665 			vhost_poll_flush(&vsock->vqs[i].poll);
666 	vhost_work_flush(&vsock->dev, &vsock->send_pkt_work);
667 }
668 
669 static void vhost_vsock_reset_orphans(struct sock *sk)
670 {
671 	struct vsock_sock *vsk = vsock_sk(sk);
672 
673 	/* vmci_transport.c doesn't take sk_lock here either.  At least we're
674 	 * under vsock_table_lock so the sock cannot disappear while we're
675 	 * executing.
676 	 */
677 
678 	/* If the peer is still valid, no need to reset connection */
679 	if (vhost_vsock_get(vsk->remote_addr.svm_cid))
680 		return;
681 
682 	/* If the close timeout is pending, let it expire.  This avoids races
683 	 * with the timeout callback.
684 	 */
685 	if (vsk->close_work_scheduled)
686 		return;
687 
688 	sock_set_flag(sk, SOCK_DONE);
689 	vsk->peer_shutdown = SHUTDOWN_MASK;
690 	sk->sk_state = SS_UNCONNECTED;
691 	sk->sk_err = ECONNRESET;
692 	sk->sk_error_report(sk);
693 }
694 
695 static int vhost_vsock_dev_release(struct inode *inode, struct file *file)
696 {
697 	struct vhost_vsock *vsock = file->private_data;
698 
699 	mutex_lock(&vhost_vsock_mutex);
700 	if (vsock->guest_cid)
701 		hash_del_rcu(&vsock->hash);
702 	mutex_unlock(&vhost_vsock_mutex);
703 
704 	/* Wait for other CPUs to finish using vsock */
705 	synchronize_rcu();
706 
707 	/* Iterating over all connections for all CIDs to find orphans is
708 	 * inefficient.  Room for improvement here. */
709 	vsock_for_each_connected_socket(vhost_vsock_reset_orphans);
710 
711 	vhost_vsock_stop(vsock);
712 	vhost_vsock_flush(vsock);
713 	vhost_dev_stop(&vsock->dev);
714 
715 	spin_lock_bh(&vsock->send_pkt_list_lock);
716 	while (!list_empty(&vsock->send_pkt_list)) {
717 		struct virtio_vsock_pkt *pkt;
718 
719 		pkt = list_first_entry(&vsock->send_pkt_list,
720 				struct virtio_vsock_pkt, list);
721 		list_del_init(&pkt->list);
722 		virtio_transport_free_pkt(pkt);
723 	}
724 	spin_unlock_bh(&vsock->send_pkt_list_lock);
725 
726 	vhost_dev_cleanup(&vsock->dev);
727 	kfree(vsock->dev.vqs);
728 	vhost_vsock_free(vsock);
729 	return 0;
730 }
731 
732 static int vhost_vsock_set_cid(struct vhost_vsock *vsock, u64 guest_cid)
733 {
734 	struct vhost_vsock *other;
735 
736 	/* Refuse reserved CIDs */
737 	if (guest_cid <= VMADDR_CID_HOST ||
738 	    guest_cid == U32_MAX)
739 		return -EINVAL;
740 
741 	/* 64-bit CIDs are not yet supported */
742 	if (guest_cid > U32_MAX)
743 		return -EINVAL;
744 
745 	/* Refuse if CID is assigned to the guest->host transport (i.e. nested
746 	 * VM), to make the loopback work.
747 	 */
748 	if (vsock_find_cid(guest_cid))
749 		return -EADDRINUSE;
750 
751 	/* Refuse if CID is already in use */
752 	mutex_lock(&vhost_vsock_mutex);
753 	other = vhost_vsock_get(guest_cid);
754 	if (other && other != vsock) {
755 		mutex_unlock(&vhost_vsock_mutex);
756 		return -EADDRINUSE;
757 	}
758 
759 	if (vsock->guest_cid)
760 		hash_del_rcu(&vsock->hash);
761 
762 	vsock->guest_cid = guest_cid;
763 	hash_add_rcu(vhost_vsock_hash, &vsock->hash, vsock->guest_cid);
764 	mutex_unlock(&vhost_vsock_mutex);
765 
766 	return 0;
767 }
768 
769 static int vhost_vsock_set_features(struct vhost_vsock *vsock, u64 features)
770 {
771 	struct vhost_virtqueue *vq;
772 	int i;
773 
774 	if (features & ~VHOST_VSOCK_FEATURES)
775 		return -EOPNOTSUPP;
776 
777 	mutex_lock(&vsock->dev.mutex);
778 	if ((features & (1 << VHOST_F_LOG_ALL)) &&
779 	    !vhost_log_access_ok(&vsock->dev)) {
780 		goto err;
781 	}
782 
783 	if ((features & (1ULL << VIRTIO_F_ACCESS_PLATFORM))) {
784 		if (vhost_init_device_iotlb(&vsock->dev, true))
785 			goto err;
786 	}
787 
788 	for (i = 0; i < ARRAY_SIZE(vsock->vqs); i++) {
789 		vq = &vsock->vqs[i];
790 		mutex_lock(&vq->mutex);
791 		vq->acked_features = features;
792 		mutex_unlock(&vq->mutex);
793 	}
794 	mutex_unlock(&vsock->dev.mutex);
795 	return 0;
796 
797 err:
798 	mutex_unlock(&vsock->dev.mutex);
799 	return -EFAULT;
800 }
801 
802 static long vhost_vsock_dev_ioctl(struct file *f, unsigned int ioctl,
803 				  unsigned long arg)
804 {
805 	struct vhost_vsock *vsock = f->private_data;
806 	void __user *argp = (void __user *)arg;
807 	u64 guest_cid;
808 	u64 features;
809 	int start;
810 	int r;
811 
812 	switch (ioctl) {
813 	case VHOST_VSOCK_SET_GUEST_CID:
814 		if (copy_from_user(&guest_cid, argp, sizeof(guest_cid)))
815 			return -EFAULT;
816 		return vhost_vsock_set_cid(vsock, guest_cid);
817 	case VHOST_VSOCK_SET_RUNNING:
818 		if (copy_from_user(&start, argp, sizeof(start)))
819 			return -EFAULT;
820 		if (start)
821 			return vhost_vsock_start(vsock);
822 		else
823 			return vhost_vsock_stop(vsock);
824 	case VHOST_GET_FEATURES:
825 		features = VHOST_VSOCK_FEATURES;
826 		if (copy_to_user(argp, &features, sizeof(features)))
827 			return -EFAULT;
828 		return 0;
829 	case VHOST_SET_FEATURES:
830 		if (copy_from_user(&features, argp, sizeof(features)))
831 			return -EFAULT;
832 		return vhost_vsock_set_features(vsock, features);
833 	case VHOST_GET_BACKEND_FEATURES:
834 		features = VHOST_VSOCK_BACKEND_FEATURES;
835 		if (copy_to_user(argp, &features, sizeof(features)))
836 			return -EFAULT;
837 		return 0;
838 	case VHOST_SET_BACKEND_FEATURES:
839 		if (copy_from_user(&features, argp, sizeof(features)))
840 			return -EFAULT;
841 		if (features & ~VHOST_VSOCK_BACKEND_FEATURES)
842 			return -EOPNOTSUPP;
843 		vhost_set_backend_features(&vsock->dev, features);
844 		return 0;
845 	default:
846 		mutex_lock(&vsock->dev.mutex);
847 		r = vhost_dev_ioctl(&vsock->dev, ioctl, argp);
848 		if (r == -ENOIOCTLCMD)
849 			r = vhost_vring_ioctl(&vsock->dev, ioctl, argp);
850 		else
851 			vhost_vsock_flush(vsock);
852 		mutex_unlock(&vsock->dev.mutex);
853 		return r;
854 	}
855 }
856 
857 static ssize_t vhost_vsock_chr_read_iter(struct kiocb *iocb, struct iov_iter *to)
858 {
859 	struct file *file = iocb->ki_filp;
860 	struct vhost_vsock *vsock = file->private_data;
861 	struct vhost_dev *dev = &vsock->dev;
862 	int noblock = file->f_flags & O_NONBLOCK;
863 
864 	return vhost_chr_read_iter(dev, to, noblock);
865 }
866 
867 static ssize_t vhost_vsock_chr_write_iter(struct kiocb *iocb,
868 					struct iov_iter *from)
869 {
870 	struct file *file = iocb->ki_filp;
871 	struct vhost_vsock *vsock = file->private_data;
872 	struct vhost_dev *dev = &vsock->dev;
873 
874 	return vhost_chr_write_iter(dev, from);
875 }
876 
877 static __poll_t vhost_vsock_chr_poll(struct file *file, poll_table *wait)
878 {
879 	struct vhost_vsock *vsock = file->private_data;
880 	struct vhost_dev *dev = &vsock->dev;
881 
882 	return vhost_chr_poll(file, dev, wait);
883 }
884 
885 static const struct file_operations vhost_vsock_fops = {
886 	.owner          = THIS_MODULE,
887 	.open           = vhost_vsock_dev_open,
888 	.release        = vhost_vsock_dev_release,
889 	.llseek		= noop_llseek,
890 	.unlocked_ioctl = vhost_vsock_dev_ioctl,
891 	.compat_ioctl   = compat_ptr_ioctl,
892 	.read_iter      = vhost_vsock_chr_read_iter,
893 	.write_iter     = vhost_vsock_chr_write_iter,
894 	.poll           = vhost_vsock_chr_poll,
895 };
896 
897 static struct miscdevice vhost_vsock_misc = {
898 	.minor = VHOST_VSOCK_MINOR,
899 	.name = "vhost-vsock",
900 	.fops = &vhost_vsock_fops,
901 };
902 
903 static int __init vhost_vsock_init(void)
904 {
905 	int ret;
906 
907 	ret = vsock_core_register(&vhost_transport.transport,
908 				  VSOCK_TRANSPORT_F_H2G);
909 	if (ret < 0)
910 		return ret;
911 	return misc_register(&vhost_vsock_misc);
912 };
913 
914 static void __exit vhost_vsock_exit(void)
915 {
916 	misc_deregister(&vhost_vsock_misc);
917 	vsock_core_unregister(&vhost_transport.transport);
918 };
919 
920 module_init(vhost_vsock_init);
921 module_exit(vhost_vsock_exit);
922 MODULE_LICENSE("GPL v2");
923 MODULE_AUTHOR("Asias He");
924 MODULE_DESCRIPTION("vhost transport for vsock ");
925 MODULE_ALIAS_MISCDEV(VHOST_VSOCK_MINOR);
926 MODULE_ALIAS("devname:vhost-vsock");
927