xref: /linux/drivers/vhost/vsock.c (revision 9052e9c95d908d6c3d7570aadc8898e1d871c8bb)
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 			       (1ULL << VIRTIO_VSOCK_F_SEQPACKET)
36 };
37 
38 enum {
39 	VHOST_VSOCK_BACKEND_FEATURES = (1ULL << VHOST_BACKEND_F_IOTLB_MSG_V2)
40 };
41 
42 /* Used to track all the vhost_vsock instances on the system. */
43 static DEFINE_MUTEX(vhost_vsock_mutex);
44 static DEFINE_READ_MOSTLY_HASHTABLE(vhost_vsock_hash, 8);
45 
46 struct vhost_vsock {
47 	struct vhost_dev dev;
48 	struct vhost_virtqueue vqs[2];
49 
50 	/* Link to global vhost_vsock_hash, writes use vhost_vsock_mutex */
51 	struct hlist_node hash;
52 
53 	struct vhost_work send_pkt_work;
54 	spinlock_t send_pkt_list_lock;
55 	struct list_head send_pkt_list;	/* host->guest pending packets */
56 
57 	atomic_t queued_replies;
58 
59 	u32 guest_cid;
60 	bool seqpacket_allow;
61 };
62 
63 static u32 vhost_transport_get_local_cid(void)
64 {
65 	return VHOST_VSOCK_DEFAULT_HOST_CID;
66 }
67 
68 /* Callers that dereference the return value must hold vhost_vsock_mutex or the
69  * RCU read lock.
70  */
71 static struct vhost_vsock *vhost_vsock_get(u32 guest_cid)
72 {
73 	struct vhost_vsock *vsock;
74 
75 	hash_for_each_possible_rcu(vhost_vsock_hash, vsock, hash, guest_cid) {
76 		u32 other_cid = vsock->guest_cid;
77 
78 		/* Skip instances that have no CID yet */
79 		if (other_cid == 0)
80 			continue;
81 
82 		if (other_cid == guest_cid)
83 			return vsock;
84 
85 	}
86 
87 	return NULL;
88 }
89 
90 static void
91 vhost_transport_do_send_pkt(struct vhost_vsock *vsock,
92 			    struct vhost_virtqueue *vq)
93 {
94 	struct vhost_virtqueue *tx_vq = &vsock->vqs[VSOCK_VQ_TX];
95 	int pkts = 0, total_len = 0;
96 	bool added = false;
97 	bool restart_tx = false;
98 
99 	mutex_lock(&vq->mutex);
100 
101 	if (!vhost_vq_get_backend(vq))
102 		goto out;
103 
104 	if (!vq_meta_prefetch(vq))
105 		goto out;
106 
107 	/* Avoid further vmexits, we're already processing the virtqueue */
108 	vhost_disable_notify(&vsock->dev, vq);
109 
110 	do {
111 		struct virtio_vsock_pkt *pkt;
112 		struct iov_iter iov_iter;
113 		unsigned out, in;
114 		size_t nbytes;
115 		size_t iov_len, payload_len;
116 		int head;
117 		u32 flags_to_restore = 0;
118 
119 		spin_lock_bh(&vsock->send_pkt_list_lock);
120 		if (list_empty(&vsock->send_pkt_list)) {
121 			spin_unlock_bh(&vsock->send_pkt_list_lock);
122 			vhost_enable_notify(&vsock->dev, vq);
123 			break;
124 		}
125 
126 		pkt = list_first_entry(&vsock->send_pkt_list,
127 				       struct virtio_vsock_pkt, list);
128 		list_del_init(&pkt->list);
129 		spin_unlock_bh(&vsock->send_pkt_list_lock);
130 
131 		head = vhost_get_vq_desc(vq, vq->iov, ARRAY_SIZE(vq->iov),
132 					 &out, &in, NULL, NULL);
133 		if (head < 0) {
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);
137 			break;
138 		}
139 
140 		if (head == vq->num) {
141 			spin_lock_bh(&vsock->send_pkt_list_lock);
142 			list_add(&pkt->list, &vsock->send_pkt_list);
143 			spin_unlock_bh(&vsock->send_pkt_list_lock);
144 
145 			/* We cannot finish yet if more buffers snuck in while
146 			 * re-enabling notify.
147 			 */
148 			if (unlikely(vhost_enable_notify(&vsock->dev, vq))) {
149 				vhost_disable_notify(&vsock->dev, vq);
150 				continue;
151 			}
152 			break;
153 		}
154 
155 		if (out) {
156 			virtio_transport_free_pkt(pkt);
157 			vq_err(vq, "Expected 0 output buffers, got %u\n", out);
158 			break;
159 		}
160 
161 		iov_len = iov_length(&vq->iov[out], in);
162 		if (iov_len < sizeof(pkt->hdr)) {
163 			virtio_transport_free_pkt(pkt);
164 			vq_err(vq, "Buffer len [%zu] too small\n", iov_len);
165 			break;
166 		}
167 
168 		iov_iter_init(&iov_iter, READ, &vq->iov[out], in, iov_len);
169 		payload_len = pkt->len - pkt->off;
170 
171 		/* If the packet is greater than the space available in the
172 		 * buffer, we split it using multiple buffers.
173 		 */
174 		if (payload_len > iov_len - sizeof(pkt->hdr)) {
175 			payload_len = iov_len - sizeof(pkt->hdr);
176 
177 			/* As we are copying pieces of large packet's buffer to
178 			 * small rx buffers, headers of packets in rx queue are
179 			 * created dynamically and are initialized with header
180 			 * of current packet(except length). But in case of
181 			 * SOCK_SEQPACKET, we also must clear message delimeter
182 			 * bit (VIRTIO_VSOCK_SEQ_EOM) and MSG_EOR bit
183 			 * (VIRTIO_VSOCK_SEQ_EOR) if set. Otherwise,
184 			 * there will be sequence of packets with these
185 			 * bits set. After initialized header will be copied to
186 			 * rx buffer, these required bits will be restored.
187 			 */
188 			if (le32_to_cpu(pkt->hdr.flags) & VIRTIO_VSOCK_SEQ_EOM) {
189 				pkt->hdr.flags &= ~cpu_to_le32(VIRTIO_VSOCK_SEQ_EOM);
190 				flags_to_restore |= VIRTIO_VSOCK_SEQ_EOM;
191 
192 				if (le32_to_cpu(pkt->hdr.flags) & VIRTIO_VSOCK_SEQ_EOR) {
193 					pkt->hdr.flags &= ~cpu_to_le32(VIRTIO_VSOCK_SEQ_EOR);
194 					flags_to_restore |= VIRTIO_VSOCK_SEQ_EOR;
195 				}
196 			}
197 		}
198 
199 		/* Set the correct length in the header */
200 		pkt->hdr.len = cpu_to_le32(payload_len);
201 
202 		nbytes = copy_to_iter(&pkt->hdr, sizeof(pkt->hdr), &iov_iter);
203 		if (nbytes != sizeof(pkt->hdr)) {
204 			virtio_transport_free_pkt(pkt);
205 			vq_err(vq, "Faulted on copying pkt hdr\n");
206 			break;
207 		}
208 
209 		nbytes = copy_to_iter(pkt->buf + pkt->off, payload_len,
210 				      &iov_iter);
211 		if (nbytes != payload_len) {
212 			virtio_transport_free_pkt(pkt);
213 			vq_err(vq, "Faulted on copying pkt buf\n");
214 			break;
215 		}
216 
217 		/* Deliver to monitoring devices all packets that we
218 		 * will transmit.
219 		 */
220 		virtio_transport_deliver_tap_pkt(pkt);
221 
222 		vhost_add_used(vq, head, sizeof(pkt->hdr) + payload_len);
223 		added = true;
224 
225 		pkt->off += payload_len;
226 		total_len += payload_len;
227 
228 		/* If we didn't send all the payload we can requeue the packet
229 		 * to send it with the next available buffer.
230 		 */
231 		if (pkt->off < pkt->len) {
232 			pkt->hdr.flags |= cpu_to_le32(flags_to_restore);
233 
234 			/* We are queueing the same virtio_vsock_pkt to handle
235 			 * the remaining bytes, and we want to deliver it
236 			 * to monitoring devices in the next iteration.
237 			 */
238 			pkt->tap_delivered = false;
239 
240 			spin_lock_bh(&vsock->send_pkt_list_lock);
241 			list_add(&pkt->list, &vsock->send_pkt_list);
242 			spin_unlock_bh(&vsock->send_pkt_list_lock);
243 		} else {
244 			if (pkt->reply) {
245 				int val;
246 
247 				val = atomic_dec_return(&vsock->queued_replies);
248 
249 				/* Do we have resources to resume tx
250 				 * processing?
251 				 */
252 				if (val + 1 == tx_vq->num)
253 					restart_tx = true;
254 			}
255 
256 			virtio_transport_free_pkt(pkt);
257 		}
258 	} while(likely(!vhost_exceeds_weight(vq, ++pkts, total_len)));
259 	if (added)
260 		vhost_signal(&vsock->dev, vq);
261 
262 out:
263 	mutex_unlock(&vq->mutex);
264 
265 	if (restart_tx)
266 		vhost_poll_queue(&tx_vq->poll);
267 }
268 
269 static void vhost_transport_send_pkt_work(struct vhost_work *work)
270 {
271 	struct vhost_virtqueue *vq;
272 	struct vhost_vsock *vsock;
273 
274 	vsock = container_of(work, struct vhost_vsock, send_pkt_work);
275 	vq = &vsock->vqs[VSOCK_VQ_RX];
276 
277 	vhost_transport_do_send_pkt(vsock, vq);
278 }
279 
280 static int
281 vhost_transport_send_pkt(struct virtio_vsock_pkt *pkt)
282 {
283 	struct vhost_vsock *vsock;
284 	int len = pkt->len;
285 
286 	rcu_read_lock();
287 
288 	/* Find the vhost_vsock according to guest context id  */
289 	vsock = vhost_vsock_get(le64_to_cpu(pkt->hdr.dst_cid));
290 	if (!vsock) {
291 		rcu_read_unlock();
292 		virtio_transport_free_pkt(pkt);
293 		return -ENODEV;
294 	}
295 
296 	if (pkt->reply)
297 		atomic_inc(&vsock->queued_replies);
298 
299 	spin_lock_bh(&vsock->send_pkt_list_lock);
300 	list_add_tail(&pkt->list, &vsock->send_pkt_list);
301 	spin_unlock_bh(&vsock->send_pkt_list_lock);
302 
303 	vhost_work_queue(&vsock->dev, &vsock->send_pkt_work);
304 
305 	rcu_read_unlock();
306 	return len;
307 }
308 
309 static int
310 vhost_transport_cancel_pkt(struct vsock_sock *vsk)
311 {
312 	struct vhost_vsock *vsock;
313 	struct virtio_vsock_pkt *pkt, *n;
314 	int cnt = 0;
315 	int ret = -ENODEV;
316 	LIST_HEAD(freeme);
317 
318 	rcu_read_lock();
319 
320 	/* Find the vhost_vsock according to guest context id  */
321 	vsock = vhost_vsock_get(vsk->remote_addr.svm_cid);
322 	if (!vsock)
323 		goto out;
324 
325 	spin_lock_bh(&vsock->send_pkt_list_lock);
326 	list_for_each_entry_safe(pkt, n, &vsock->send_pkt_list, list) {
327 		if (pkt->vsk != vsk)
328 			continue;
329 		list_move(&pkt->list, &freeme);
330 	}
331 	spin_unlock_bh(&vsock->send_pkt_list_lock);
332 
333 	list_for_each_entry_safe(pkt, n, &freeme, list) {
334 		if (pkt->reply)
335 			cnt++;
336 		list_del(&pkt->list);
337 		virtio_transport_free_pkt(pkt);
338 	}
339 
340 	if (cnt) {
341 		struct vhost_virtqueue *tx_vq = &vsock->vqs[VSOCK_VQ_TX];
342 		int new_cnt;
343 
344 		new_cnt = atomic_sub_return(cnt, &vsock->queued_replies);
345 		if (new_cnt + cnt >= tx_vq->num && new_cnt < tx_vq->num)
346 			vhost_poll_queue(&tx_vq->poll);
347 	}
348 
349 	ret = 0;
350 out:
351 	rcu_read_unlock();
352 	return ret;
353 }
354 
355 static struct virtio_vsock_pkt *
356 vhost_vsock_alloc_pkt(struct vhost_virtqueue *vq,
357 		      unsigned int out, unsigned int in)
358 {
359 	struct virtio_vsock_pkt *pkt;
360 	struct iov_iter iov_iter;
361 	size_t nbytes;
362 	size_t len;
363 
364 	if (in != 0) {
365 		vq_err(vq, "Expected 0 input buffers, got %u\n", in);
366 		return NULL;
367 	}
368 
369 	pkt = kzalloc(sizeof(*pkt), GFP_KERNEL);
370 	if (!pkt)
371 		return NULL;
372 
373 	len = iov_length(vq->iov, out);
374 	iov_iter_init(&iov_iter, WRITE, vq->iov, out, len);
375 
376 	nbytes = copy_from_iter(&pkt->hdr, sizeof(pkt->hdr), &iov_iter);
377 	if (nbytes != sizeof(pkt->hdr)) {
378 		vq_err(vq, "Expected %zu bytes for pkt->hdr, got %zu bytes\n",
379 		       sizeof(pkt->hdr), nbytes);
380 		kfree(pkt);
381 		return NULL;
382 	}
383 
384 	pkt->len = le32_to_cpu(pkt->hdr.len);
385 
386 	/* No payload */
387 	if (!pkt->len)
388 		return pkt;
389 
390 	/* The pkt is too big */
391 	if (pkt->len > VIRTIO_VSOCK_MAX_PKT_BUF_SIZE) {
392 		kfree(pkt);
393 		return NULL;
394 	}
395 
396 	pkt->buf = kmalloc(pkt->len, GFP_KERNEL);
397 	if (!pkt->buf) {
398 		kfree(pkt);
399 		return NULL;
400 	}
401 
402 	pkt->buf_len = pkt->len;
403 
404 	nbytes = copy_from_iter(pkt->buf, pkt->len, &iov_iter);
405 	if (nbytes != pkt->len) {
406 		vq_err(vq, "Expected %u byte payload, got %zu bytes\n",
407 		       pkt->len, nbytes);
408 		virtio_transport_free_pkt(pkt);
409 		return NULL;
410 	}
411 
412 	return pkt;
413 }
414 
415 /* Is there space left for replies to rx packets? */
416 static bool vhost_vsock_more_replies(struct vhost_vsock *vsock)
417 {
418 	struct vhost_virtqueue *vq = &vsock->vqs[VSOCK_VQ_TX];
419 	int val;
420 
421 	smp_rmb(); /* paired with atomic_inc() and atomic_dec_return() */
422 	val = atomic_read(&vsock->queued_replies);
423 
424 	return val < vq->num;
425 }
426 
427 static bool vhost_transport_seqpacket_allow(u32 remote_cid);
428 
429 static struct virtio_transport vhost_transport = {
430 	.transport = {
431 		.module                   = THIS_MODULE,
432 
433 		.get_local_cid            = vhost_transport_get_local_cid,
434 
435 		.init                     = virtio_transport_do_socket_init,
436 		.destruct                 = virtio_transport_destruct,
437 		.release                  = virtio_transport_release,
438 		.connect                  = virtio_transport_connect,
439 		.shutdown                 = virtio_transport_shutdown,
440 		.cancel_pkt               = vhost_transport_cancel_pkt,
441 
442 		.dgram_enqueue            = virtio_transport_dgram_enqueue,
443 		.dgram_dequeue            = virtio_transport_dgram_dequeue,
444 		.dgram_bind               = virtio_transport_dgram_bind,
445 		.dgram_allow              = virtio_transport_dgram_allow,
446 
447 		.stream_enqueue           = virtio_transport_stream_enqueue,
448 		.stream_dequeue           = virtio_transport_stream_dequeue,
449 		.stream_has_data          = virtio_transport_stream_has_data,
450 		.stream_has_space         = virtio_transport_stream_has_space,
451 		.stream_rcvhiwat          = virtio_transport_stream_rcvhiwat,
452 		.stream_is_active         = virtio_transport_stream_is_active,
453 		.stream_allow             = virtio_transport_stream_allow,
454 
455 		.seqpacket_dequeue        = virtio_transport_seqpacket_dequeue,
456 		.seqpacket_enqueue        = virtio_transport_seqpacket_enqueue,
457 		.seqpacket_allow          = vhost_transport_seqpacket_allow,
458 		.seqpacket_has_data       = virtio_transport_seqpacket_has_data,
459 
460 		.notify_poll_in           = virtio_transport_notify_poll_in,
461 		.notify_poll_out          = virtio_transport_notify_poll_out,
462 		.notify_recv_init         = virtio_transport_notify_recv_init,
463 		.notify_recv_pre_block    = virtio_transport_notify_recv_pre_block,
464 		.notify_recv_pre_dequeue  = virtio_transport_notify_recv_pre_dequeue,
465 		.notify_recv_post_dequeue = virtio_transport_notify_recv_post_dequeue,
466 		.notify_send_init         = virtio_transport_notify_send_init,
467 		.notify_send_pre_block    = virtio_transport_notify_send_pre_block,
468 		.notify_send_pre_enqueue  = virtio_transport_notify_send_pre_enqueue,
469 		.notify_send_post_enqueue = virtio_transport_notify_send_post_enqueue,
470 		.notify_buffer_size       = virtio_transport_notify_buffer_size,
471 
472 	},
473 
474 	.send_pkt = vhost_transport_send_pkt,
475 };
476 
477 static bool vhost_transport_seqpacket_allow(u32 remote_cid)
478 {
479 	struct vhost_vsock *vsock;
480 	bool seqpacket_allow = false;
481 
482 	rcu_read_lock();
483 	vsock = vhost_vsock_get(remote_cid);
484 
485 	if (vsock)
486 		seqpacket_allow = vsock->seqpacket_allow;
487 
488 	rcu_read_unlock();
489 
490 	return seqpacket_allow;
491 }
492 
493 static void vhost_vsock_handle_tx_kick(struct vhost_work *work)
494 {
495 	struct vhost_virtqueue *vq = container_of(work, struct vhost_virtqueue,
496 						  poll.work);
497 	struct vhost_vsock *vsock = container_of(vq->dev, struct vhost_vsock,
498 						 dev);
499 	struct virtio_vsock_pkt *pkt;
500 	int head, pkts = 0, total_len = 0;
501 	unsigned int out, in;
502 	bool added = false;
503 
504 	mutex_lock(&vq->mutex);
505 
506 	if (!vhost_vq_get_backend(vq))
507 		goto out;
508 
509 	if (!vq_meta_prefetch(vq))
510 		goto out;
511 
512 	vhost_disable_notify(&vsock->dev, vq);
513 	do {
514 		u32 len;
515 
516 		if (!vhost_vsock_more_replies(vsock)) {
517 			/* Stop tx until the device processes already
518 			 * pending replies.  Leave tx virtqueue
519 			 * callbacks disabled.
520 			 */
521 			goto no_more_replies;
522 		}
523 
524 		head = vhost_get_vq_desc(vq, vq->iov, ARRAY_SIZE(vq->iov),
525 					 &out, &in, NULL, NULL);
526 		if (head < 0)
527 			break;
528 
529 		if (head == vq->num) {
530 			if (unlikely(vhost_enable_notify(&vsock->dev, vq))) {
531 				vhost_disable_notify(&vsock->dev, vq);
532 				continue;
533 			}
534 			break;
535 		}
536 
537 		pkt = vhost_vsock_alloc_pkt(vq, out, in);
538 		if (!pkt) {
539 			vq_err(vq, "Faulted on pkt\n");
540 			continue;
541 		}
542 
543 		len = pkt->len;
544 
545 		/* Deliver to monitoring devices all received packets */
546 		virtio_transport_deliver_tap_pkt(pkt);
547 
548 		/* Only accept correctly addressed packets */
549 		if (le64_to_cpu(pkt->hdr.src_cid) == vsock->guest_cid &&
550 		    le64_to_cpu(pkt->hdr.dst_cid) ==
551 		    vhost_transport_get_local_cid())
552 			virtio_transport_recv_pkt(&vhost_transport, pkt);
553 		else
554 			virtio_transport_free_pkt(pkt);
555 
556 		len += sizeof(pkt->hdr);
557 		vhost_add_used(vq, head, len);
558 		total_len += len;
559 		added = true;
560 	} while(likely(!vhost_exceeds_weight(vq, ++pkts, total_len)));
561 
562 no_more_replies:
563 	if (added)
564 		vhost_signal(&vsock->dev, vq);
565 
566 out:
567 	mutex_unlock(&vq->mutex);
568 }
569 
570 static void vhost_vsock_handle_rx_kick(struct vhost_work *work)
571 {
572 	struct vhost_virtqueue *vq = container_of(work, struct vhost_virtqueue,
573 						poll.work);
574 	struct vhost_vsock *vsock = container_of(vq->dev, struct vhost_vsock,
575 						 dev);
576 
577 	vhost_transport_do_send_pkt(vsock, vq);
578 }
579 
580 static int vhost_vsock_start(struct vhost_vsock *vsock)
581 {
582 	struct vhost_virtqueue *vq;
583 	size_t i;
584 	int ret;
585 
586 	mutex_lock(&vsock->dev.mutex);
587 
588 	ret = vhost_dev_check_owner(&vsock->dev);
589 	if (ret)
590 		goto err;
591 
592 	for (i = 0; i < ARRAY_SIZE(vsock->vqs); i++) {
593 		vq = &vsock->vqs[i];
594 
595 		mutex_lock(&vq->mutex);
596 
597 		if (!vhost_vq_access_ok(vq)) {
598 			ret = -EFAULT;
599 			goto err_vq;
600 		}
601 
602 		if (!vhost_vq_get_backend(vq)) {
603 			vhost_vq_set_backend(vq, vsock);
604 			ret = vhost_vq_init_access(vq);
605 			if (ret)
606 				goto err_vq;
607 		}
608 
609 		mutex_unlock(&vq->mutex);
610 	}
611 
612 	/* Some packets may have been queued before the device was started,
613 	 * let's kick the send worker to send them.
614 	 */
615 	vhost_work_queue(&vsock->dev, &vsock->send_pkt_work);
616 
617 	mutex_unlock(&vsock->dev.mutex);
618 	return 0;
619 
620 err_vq:
621 	vhost_vq_set_backend(vq, NULL);
622 	mutex_unlock(&vq->mutex);
623 
624 	for (i = 0; i < ARRAY_SIZE(vsock->vqs); i++) {
625 		vq = &vsock->vqs[i];
626 
627 		mutex_lock(&vq->mutex);
628 		vhost_vq_set_backend(vq, NULL);
629 		mutex_unlock(&vq->mutex);
630 	}
631 err:
632 	mutex_unlock(&vsock->dev.mutex);
633 	return ret;
634 }
635 
636 static int vhost_vsock_stop(struct vhost_vsock *vsock)
637 {
638 	size_t i;
639 	int ret;
640 
641 	mutex_lock(&vsock->dev.mutex);
642 
643 	ret = vhost_dev_check_owner(&vsock->dev);
644 	if (ret)
645 		goto err;
646 
647 	for (i = 0; i < ARRAY_SIZE(vsock->vqs); i++) {
648 		struct vhost_virtqueue *vq = &vsock->vqs[i];
649 
650 		mutex_lock(&vq->mutex);
651 		vhost_vq_set_backend(vq, NULL);
652 		mutex_unlock(&vq->mutex);
653 	}
654 
655 err:
656 	mutex_unlock(&vsock->dev.mutex);
657 	return ret;
658 }
659 
660 static void vhost_vsock_free(struct vhost_vsock *vsock)
661 {
662 	kvfree(vsock);
663 }
664 
665 static int vhost_vsock_dev_open(struct inode *inode, struct file *file)
666 {
667 	struct vhost_virtqueue **vqs;
668 	struct vhost_vsock *vsock;
669 	int ret;
670 
671 	/* This struct is large and allocation could fail, fall back to vmalloc
672 	 * if there is no other way.
673 	 */
674 	vsock = kvmalloc(sizeof(*vsock), GFP_KERNEL | __GFP_RETRY_MAYFAIL);
675 	if (!vsock)
676 		return -ENOMEM;
677 
678 	vqs = kmalloc_array(ARRAY_SIZE(vsock->vqs), sizeof(*vqs), GFP_KERNEL);
679 	if (!vqs) {
680 		ret = -ENOMEM;
681 		goto out;
682 	}
683 
684 	vsock->guest_cid = 0; /* no CID assigned yet */
685 
686 	atomic_set(&vsock->queued_replies, 0);
687 
688 	vqs[VSOCK_VQ_TX] = &vsock->vqs[VSOCK_VQ_TX];
689 	vqs[VSOCK_VQ_RX] = &vsock->vqs[VSOCK_VQ_RX];
690 	vsock->vqs[VSOCK_VQ_TX].handle_kick = vhost_vsock_handle_tx_kick;
691 	vsock->vqs[VSOCK_VQ_RX].handle_kick = vhost_vsock_handle_rx_kick;
692 
693 	vhost_dev_init(&vsock->dev, vqs, ARRAY_SIZE(vsock->vqs),
694 		       UIO_MAXIOV, VHOST_VSOCK_PKT_WEIGHT,
695 		       VHOST_VSOCK_WEIGHT, true, NULL);
696 
697 	file->private_data = vsock;
698 	spin_lock_init(&vsock->send_pkt_list_lock);
699 	INIT_LIST_HEAD(&vsock->send_pkt_list);
700 	vhost_work_init(&vsock->send_pkt_work, vhost_transport_send_pkt_work);
701 	return 0;
702 
703 out:
704 	vhost_vsock_free(vsock);
705 	return ret;
706 }
707 
708 static void vhost_vsock_flush(struct vhost_vsock *vsock)
709 {
710 	int i;
711 
712 	for (i = 0; i < ARRAY_SIZE(vsock->vqs); i++)
713 		if (vsock->vqs[i].handle_kick)
714 			vhost_poll_flush(&vsock->vqs[i].poll);
715 	vhost_work_dev_flush(&vsock->dev);
716 }
717 
718 static void vhost_vsock_reset_orphans(struct sock *sk)
719 {
720 	struct vsock_sock *vsk = vsock_sk(sk);
721 
722 	/* vmci_transport.c doesn't take sk_lock here either.  At least we're
723 	 * under vsock_table_lock so the sock cannot disappear while we're
724 	 * executing.
725 	 */
726 
727 	/* If the peer is still valid, no need to reset connection */
728 	if (vhost_vsock_get(vsk->remote_addr.svm_cid))
729 		return;
730 
731 	/* If the close timeout is pending, let it expire.  This avoids races
732 	 * with the timeout callback.
733 	 */
734 	if (vsk->close_work_scheduled)
735 		return;
736 
737 	sock_set_flag(sk, SOCK_DONE);
738 	vsk->peer_shutdown = SHUTDOWN_MASK;
739 	sk->sk_state = SS_UNCONNECTED;
740 	sk->sk_err = ECONNRESET;
741 	sk_error_report(sk);
742 }
743 
744 static int vhost_vsock_dev_release(struct inode *inode, struct file *file)
745 {
746 	struct vhost_vsock *vsock = file->private_data;
747 
748 	mutex_lock(&vhost_vsock_mutex);
749 	if (vsock->guest_cid)
750 		hash_del_rcu(&vsock->hash);
751 	mutex_unlock(&vhost_vsock_mutex);
752 
753 	/* Wait for other CPUs to finish using vsock */
754 	synchronize_rcu();
755 
756 	/* Iterating over all connections for all CIDs to find orphans is
757 	 * inefficient.  Room for improvement here. */
758 	vsock_for_each_connected_socket(vhost_vsock_reset_orphans);
759 
760 	vhost_vsock_stop(vsock);
761 	vhost_vsock_flush(vsock);
762 	vhost_dev_stop(&vsock->dev);
763 
764 	spin_lock_bh(&vsock->send_pkt_list_lock);
765 	while (!list_empty(&vsock->send_pkt_list)) {
766 		struct virtio_vsock_pkt *pkt;
767 
768 		pkt = list_first_entry(&vsock->send_pkt_list,
769 				struct virtio_vsock_pkt, list);
770 		list_del_init(&pkt->list);
771 		virtio_transport_free_pkt(pkt);
772 	}
773 	spin_unlock_bh(&vsock->send_pkt_list_lock);
774 
775 	vhost_dev_cleanup(&vsock->dev);
776 	kfree(vsock->dev.vqs);
777 	vhost_vsock_free(vsock);
778 	return 0;
779 }
780 
781 static int vhost_vsock_set_cid(struct vhost_vsock *vsock, u64 guest_cid)
782 {
783 	struct vhost_vsock *other;
784 
785 	/* Refuse reserved CIDs */
786 	if (guest_cid <= VMADDR_CID_HOST ||
787 	    guest_cid == U32_MAX)
788 		return -EINVAL;
789 
790 	/* 64-bit CIDs are not yet supported */
791 	if (guest_cid > U32_MAX)
792 		return -EINVAL;
793 
794 	/* Refuse if CID is assigned to the guest->host transport (i.e. nested
795 	 * VM), to make the loopback work.
796 	 */
797 	if (vsock_find_cid(guest_cid))
798 		return -EADDRINUSE;
799 
800 	/* Refuse if CID is already in use */
801 	mutex_lock(&vhost_vsock_mutex);
802 	other = vhost_vsock_get(guest_cid);
803 	if (other && other != vsock) {
804 		mutex_unlock(&vhost_vsock_mutex);
805 		return -EADDRINUSE;
806 	}
807 
808 	if (vsock->guest_cid)
809 		hash_del_rcu(&vsock->hash);
810 
811 	vsock->guest_cid = guest_cid;
812 	hash_add_rcu(vhost_vsock_hash, &vsock->hash, vsock->guest_cid);
813 	mutex_unlock(&vhost_vsock_mutex);
814 
815 	return 0;
816 }
817 
818 static int vhost_vsock_set_features(struct vhost_vsock *vsock, u64 features)
819 {
820 	struct vhost_virtqueue *vq;
821 	int i;
822 
823 	if (features & ~VHOST_VSOCK_FEATURES)
824 		return -EOPNOTSUPP;
825 
826 	mutex_lock(&vsock->dev.mutex);
827 	if ((features & (1 << VHOST_F_LOG_ALL)) &&
828 	    !vhost_log_access_ok(&vsock->dev)) {
829 		goto err;
830 	}
831 
832 	if ((features & (1ULL << VIRTIO_F_ACCESS_PLATFORM))) {
833 		if (vhost_init_device_iotlb(&vsock->dev, true))
834 			goto err;
835 	}
836 
837 	if (features & (1ULL << VIRTIO_VSOCK_F_SEQPACKET))
838 		vsock->seqpacket_allow = true;
839 
840 	for (i = 0; i < ARRAY_SIZE(vsock->vqs); i++) {
841 		vq = &vsock->vqs[i];
842 		mutex_lock(&vq->mutex);
843 		vq->acked_features = features;
844 		mutex_unlock(&vq->mutex);
845 	}
846 	mutex_unlock(&vsock->dev.mutex);
847 	return 0;
848 
849 err:
850 	mutex_unlock(&vsock->dev.mutex);
851 	return -EFAULT;
852 }
853 
854 static long vhost_vsock_dev_ioctl(struct file *f, unsigned int ioctl,
855 				  unsigned long arg)
856 {
857 	struct vhost_vsock *vsock = f->private_data;
858 	void __user *argp = (void __user *)arg;
859 	u64 guest_cid;
860 	u64 features;
861 	int start;
862 	int r;
863 
864 	switch (ioctl) {
865 	case VHOST_VSOCK_SET_GUEST_CID:
866 		if (copy_from_user(&guest_cid, argp, sizeof(guest_cid)))
867 			return -EFAULT;
868 		return vhost_vsock_set_cid(vsock, guest_cid);
869 	case VHOST_VSOCK_SET_RUNNING:
870 		if (copy_from_user(&start, argp, sizeof(start)))
871 			return -EFAULT;
872 		if (start)
873 			return vhost_vsock_start(vsock);
874 		else
875 			return vhost_vsock_stop(vsock);
876 	case VHOST_GET_FEATURES:
877 		features = VHOST_VSOCK_FEATURES;
878 		if (copy_to_user(argp, &features, sizeof(features)))
879 			return -EFAULT;
880 		return 0;
881 	case VHOST_SET_FEATURES:
882 		if (copy_from_user(&features, argp, sizeof(features)))
883 			return -EFAULT;
884 		return vhost_vsock_set_features(vsock, features);
885 	case VHOST_GET_BACKEND_FEATURES:
886 		features = VHOST_VSOCK_BACKEND_FEATURES;
887 		if (copy_to_user(argp, &features, sizeof(features)))
888 			return -EFAULT;
889 		return 0;
890 	case VHOST_SET_BACKEND_FEATURES:
891 		if (copy_from_user(&features, argp, sizeof(features)))
892 			return -EFAULT;
893 		if (features & ~VHOST_VSOCK_BACKEND_FEATURES)
894 			return -EOPNOTSUPP;
895 		vhost_set_backend_features(&vsock->dev, features);
896 		return 0;
897 	default:
898 		mutex_lock(&vsock->dev.mutex);
899 		r = vhost_dev_ioctl(&vsock->dev, ioctl, argp);
900 		if (r == -ENOIOCTLCMD)
901 			r = vhost_vring_ioctl(&vsock->dev, ioctl, argp);
902 		else
903 			vhost_vsock_flush(vsock);
904 		mutex_unlock(&vsock->dev.mutex);
905 		return r;
906 	}
907 }
908 
909 static ssize_t vhost_vsock_chr_read_iter(struct kiocb *iocb, struct iov_iter *to)
910 {
911 	struct file *file = iocb->ki_filp;
912 	struct vhost_vsock *vsock = file->private_data;
913 	struct vhost_dev *dev = &vsock->dev;
914 	int noblock = file->f_flags & O_NONBLOCK;
915 
916 	return vhost_chr_read_iter(dev, to, noblock);
917 }
918 
919 static ssize_t vhost_vsock_chr_write_iter(struct kiocb *iocb,
920 					struct iov_iter *from)
921 {
922 	struct file *file = iocb->ki_filp;
923 	struct vhost_vsock *vsock = file->private_data;
924 	struct vhost_dev *dev = &vsock->dev;
925 
926 	return vhost_chr_write_iter(dev, from);
927 }
928 
929 static __poll_t vhost_vsock_chr_poll(struct file *file, poll_table *wait)
930 {
931 	struct vhost_vsock *vsock = file->private_data;
932 	struct vhost_dev *dev = &vsock->dev;
933 
934 	return vhost_chr_poll(file, dev, wait);
935 }
936 
937 static const struct file_operations vhost_vsock_fops = {
938 	.owner          = THIS_MODULE,
939 	.open           = vhost_vsock_dev_open,
940 	.release        = vhost_vsock_dev_release,
941 	.llseek		= noop_llseek,
942 	.unlocked_ioctl = vhost_vsock_dev_ioctl,
943 	.compat_ioctl   = compat_ptr_ioctl,
944 	.read_iter      = vhost_vsock_chr_read_iter,
945 	.write_iter     = vhost_vsock_chr_write_iter,
946 	.poll           = vhost_vsock_chr_poll,
947 };
948 
949 static struct miscdevice vhost_vsock_misc = {
950 	.minor = VHOST_VSOCK_MINOR,
951 	.name = "vhost-vsock",
952 	.fops = &vhost_vsock_fops,
953 };
954 
955 static int __init vhost_vsock_init(void)
956 {
957 	int ret;
958 
959 	ret = vsock_core_register(&vhost_transport.transport,
960 				  VSOCK_TRANSPORT_F_H2G);
961 	if (ret < 0)
962 		return ret;
963 	return misc_register(&vhost_vsock_misc);
964 };
965 
966 static void __exit vhost_vsock_exit(void)
967 {
968 	misc_deregister(&vhost_vsock_misc);
969 	vsock_core_unregister(&vhost_transport.transport);
970 };
971 
972 module_init(vhost_vsock_init);
973 module_exit(vhost_vsock_exit);
974 MODULE_LICENSE("GPL v2");
975 MODULE_AUTHOR("Asias He");
976 MODULE_DESCRIPTION("vhost transport for vsock ");
977 MODULE_ALIAS_MISCDEV(VHOST_VSOCK_MINOR);
978 MODULE_ALIAS("devname:vhost-vsock");
979