xref: /linux/drivers/vhost/vsock.c (revision 05e352444b2430de4b183b4a988085381e5fd6ad)
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 static const int vhost_vsock_bits[] = {
33 	VHOST_FEATURES,
34 	VIRTIO_F_ACCESS_PLATFORM,
35 	VIRTIO_VSOCK_F_SEQPACKET
36 };
37 
38 #define VHOST_VSOCK_FEATURES VHOST_FEATURES_U64(vhost_vsock_bits, 0)
39 
40 enum {
41 	VHOST_VSOCK_BACKEND_FEATURES = (1ULL << VHOST_BACKEND_F_IOTLB_MSG_V2)
42 };
43 
44 /* Used to track all the vhost_vsock instances on the system. */
45 static DEFINE_MUTEX(vhost_vsock_mutex);
46 static DEFINE_READ_MOSTLY_HASHTABLE(vhost_vsock_hash, 8);
47 
48 struct vhost_vsock {
49 	struct vhost_dev dev;
50 	struct vhost_virtqueue vqs[2];
51 	struct net *net;
52 	netns_tracker ns_tracker;
53 
54 	/* Link to global vhost_vsock_hash, writes use vhost_vsock_mutex */
55 	struct hlist_node hash;
56 
57 	struct vhost_work send_pkt_work;
58 	struct sk_buff_head send_pkt_queue; /* host->guest pending packets */
59 
60 	atomic_t queued_replies;
61 
62 	u32 guest_cid;
63 	bool seqpacket_allow;
64 };
65 
66 static u32 vhost_transport_get_local_cid(void)
67 {
68 	return VHOST_VSOCK_DEFAULT_HOST_CID;
69 }
70 
71 /* Callers must be in an RCU read section or hold the vhost_vsock_mutex.
72  * The return value can only be dereferenced while within the section.
73  */
74 static struct vhost_vsock *vhost_vsock_get(u32 guest_cid, struct net *net)
75 {
76 	struct vhost_vsock *vsock;
77 
78 	hash_for_each_possible_rcu(vhost_vsock_hash, vsock, hash, guest_cid,
79 				   lockdep_is_held(&vhost_vsock_mutex)) {
80 		u32 other_cid = vsock->guest_cid;
81 
82 		/* Skip instances that have no CID yet */
83 		if (other_cid == 0)
84 			continue;
85 
86 		if (other_cid == guest_cid &&
87 		    vsock_net_check_mode(net, vsock->net))
88 			return vsock;
89 	}
90 
91 	return NULL;
92 }
93 
94 static bool vhost_transport_has_remote_cid(struct vsock_sock *vsk, u32 cid)
95 {
96 	struct sock *sk = sk_vsock(vsk);
97 	struct net *net = sock_net(sk);
98 	bool found;
99 
100 	rcu_read_lock();
101 	found = !!vhost_vsock_get(cid, net);
102 	rcu_read_unlock();
103 	return found;
104 }
105 
106 static void
107 vhost_transport_do_send_pkt(struct vhost_vsock *vsock,
108 			    struct vhost_virtqueue *vq)
109 {
110 	struct vhost_virtqueue *tx_vq = &vsock->vqs[VSOCK_VQ_TX];
111 	int pkts = 0, total_len = 0;
112 	bool added = false;
113 	bool restart_tx = false;
114 
115 	mutex_lock(&vq->mutex);
116 
117 	if (!vhost_vq_get_backend(vq))
118 		goto out;
119 
120 	if (!vq_meta_prefetch(vq))
121 		goto out;
122 
123 	/* Avoid further vmexits, we're already processing the virtqueue */
124 	vhost_disable_notify(&vsock->dev, vq);
125 
126 	do {
127 		struct virtio_vsock_hdr *hdr;
128 		size_t iov_len, payload_len;
129 		struct iov_iter iov_iter;
130 		u32 flags_to_restore = 0;
131 		struct sk_buff *skb;
132 		unsigned out, in;
133 		size_t nbytes;
134 		u32 offset;
135 		int head;
136 
137 		skb = virtio_vsock_skb_dequeue(&vsock->send_pkt_queue);
138 
139 		if (!skb) {
140 			vhost_enable_notify(&vsock->dev, vq);
141 			break;
142 		}
143 
144 		head = vhost_get_vq_desc(vq, vq->iov, ARRAY_SIZE(vq->iov),
145 					 &out, &in, NULL, NULL);
146 		if (head < 0) {
147 			virtio_vsock_skb_queue_head(&vsock->send_pkt_queue, skb);
148 			break;
149 		}
150 
151 		if (head == vq->num) {
152 			virtio_vsock_skb_queue_head(&vsock->send_pkt_queue, skb);
153 			/* We cannot finish yet if more buffers snuck in while
154 			 * re-enabling notify.
155 			 */
156 			if (unlikely(vhost_enable_notify(&vsock->dev, vq))) {
157 				vhost_disable_notify(&vsock->dev, vq);
158 				continue;
159 			}
160 			break;
161 		}
162 
163 		if (out) {
164 			kfree_skb(skb);
165 			vq_err(vq, "Expected 0 output buffers, got %u\n", out);
166 			break;
167 		}
168 
169 		iov_len = iov_length(&vq->iov[out], in);
170 		if (iov_len < sizeof(*hdr)) {
171 			kfree_skb(skb);
172 			vq_err(vq, "Buffer len [%zu] too small\n", iov_len);
173 			break;
174 		}
175 
176 		iov_iter_init(&iov_iter, ITER_DEST, &vq->iov[out], in, iov_len);
177 		offset = VIRTIO_VSOCK_SKB_CB(skb)->offset;
178 		payload_len = skb->len - offset;
179 		hdr = virtio_vsock_hdr(skb);
180 
181 		/* If the packet is greater than the space available in the
182 		 * buffer, we split it using multiple buffers.
183 		 */
184 		if (payload_len > iov_len - sizeof(*hdr)) {
185 			payload_len = iov_len - sizeof(*hdr);
186 
187 			/* As we are copying pieces of large packet's buffer to
188 			 * small rx buffers, headers of packets in rx queue are
189 			 * created dynamically and are initialized with header
190 			 * of current packet(except length). But in case of
191 			 * SOCK_SEQPACKET, we also must clear message delimeter
192 			 * bit (VIRTIO_VSOCK_SEQ_EOM) and MSG_EOR bit
193 			 * (VIRTIO_VSOCK_SEQ_EOR) if set. Otherwise,
194 			 * there will be sequence of packets with these
195 			 * bits set. After initialized header will be copied to
196 			 * rx buffer, these required bits will be restored.
197 			 */
198 			if (le32_to_cpu(hdr->flags) & VIRTIO_VSOCK_SEQ_EOM) {
199 				hdr->flags &= ~cpu_to_le32(VIRTIO_VSOCK_SEQ_EOM);
200 				flags_to_restore |= VIRTIO_VSOCK_SEQ_EOM;
201 
202 				if (le32_to_cpu(hdr->flags) & VIRTIO_VSOCK_SEQ_EOR) {
203 					hdr->flags &= ~cpu_to_le32(VIRTIO_VSOCK_SEQ_EOR);
204 					flags_to_restore |= VIRTIO_VSOCK_SEQ_EOR;
205 				}
206 			}
207 		}
208 
209 		/* Set the correct length in the header */
210 		hdr->len = cpu_to_le32(payload_len);
211 
212 		nbytes = copy_to_iter(hdr, sizeof(*hdr), &iov_iter);
213 		if (nbytes != sizeof(*hdr)) {
214 			kfree_skb(skb);
215 			vq_err(vq, "Faulted on copying pkt hdr\n");
216 			break;
217 		}
218 
219 		if (skb_copy_datagram_iter(skb,
220 					   offset,
221 					   &iov_iter,
222 					   payload_len)) {
223 			kfree_skb(skb);
224 			vq_err(vq, "Faulted on copying pkt buf\n");
225 			break;
226 		}
227 
228 		/* Deliver to monitoring devices all packets that we
229 		 * will transmit.
230 		 */
231 		virtio_transport_deliver_tap_pkt(skb);
232 
233 		vhost_add_used(vq, head, sizeof(*hdr) + payload_len);
234 		added = true;
235 
236 		VIRTIO_VSOCK_SKB_CB(skb)->offset += payload_len;
237 		total_len += payload_len;
238 
239 		/* If we didn't send all the payload we can requeue the packet
240 		 * to send it with the next available buffer.
241 		 */
242 		if (VIRTIO_VSOCK_SKB_CB(skb)->offset < skb->len) {
243 			hdr->flags |= cpu_to_le32(flags_to_restore);
244 
245 			/* We are queueing the same skb to handle
246 			 * the remaining bytes, and we want to deliver it
247 			 * to monitoring devices in the next iteration.
248 			 */
249 			virtio_vsock_skb_clear_tap_delivered(skb);
250 			virtio_vsock_skb_queue_head(&vsock->send_pkt_queue, skb);
251 		} else {
252 			if (virtio_vsock_skb_reply(skb)) {
253 				int val;
254 
255 				val = atomic_dec_return(&vsock->queued_replies);
256 
257 				/* Do we have resources to resume tx
258 				 * processing?
259 				 */
260 				if (val + 1 == tx_vq->num)
261 					restart_tx = true;
262 			}
263 
264 			virtio_transport_consume_skb_sent(skb, true);
265 		}
266 	} while(likely(!vhost_exceeds_weight(vq, ++pkts, total_len)));
267 	if (added)
268 		vhost_signal(&vsock->dev, vq);
269 
270 out:
271 	mutex_unlock(&vq->mutex);
272 
273 	if (restart_tx)
274 		vhost_poll_queue(&tx_vq->poll);
275 }
276 
277 static void vhost_transport_send_pkt_work(struct vhost_work *work)
278 {
279 	struct vhost_virtqueue *vq;
280 	struct vhost_vsock *vsock;
281 
282 	vsock = container_of(work, struct vhost_vsock, send_pkt_work);
283 	vq = &vsock->vqs[VSOCK_VQ_RX];
284 
285 	vhost_transport_do_send_pkt(vsock, vq);
286 }
287 
288 static int
289 vhost_transport_send_pkt(struct sk_buff *skb, struct net *net)
290 {
291 	struct virtio_vsock_hdr *hdr = virtio_vsock_hdr(skb);
292 	struct vhost_vsock *vsock;
293 	int len = skb->len;
294 
295 	rcu_read_lock();
296 
297 	/* Find the vhost_vsock according to guest context id  */
298 	vsock = vhost_vsock_get(le64_to_cpu(hdr->dst_cid), net);
299 	if (!vsock) {
300 		rcu_read_unlock();
301 		kfree_skb(skb);
302 		return -ENODEV;
303 	}
304 
305 	if (virtio_vsock_skb_reply(skb))
306 		atomic_inc(&vsock->queued_replies);
307 
308 	virtio_vsock_skb_queue_tail(&vsock->send_pkt_queue, skb);
309 	vhost_vq_work_queue(&vsock->vqs[VSOCK_VQ_RX], &vsock->send_pkt_work);
310 
311 	rcu_read_unlock();
312 	return len;
313 }
314 
315 static int
316 vhost_transport_cancel_pkt(struct vsock_sock *vsk)
317 {
318 	struct vhost_vsock *vsock;
319 	int cnt = 0;
320 	int ret = -ENODEV;
321 
322 	rcu_read_lock();
323 
324 	/* Find the vhost_vsock according to guest context id  */
325 	vsock = vhost_vsock_get(vsk->remote_addr.svm_cid,
326 				sock_net(sk_vsock(vsk)));
327 	if (!vsock)
328 		goto out;
329 
330 	cnt = virtio_transport_purge_skbs(vsk, &vsock->send_pkt_queue);
331 
332 	if (cnt) {
333 		struct vhost_virtqueue *tx_vq = &vsock->vqs[VSOCK_VQ_TX];
334 		int new_cnt;
335 
336 		new_cnt = atomic_sub_return(cnt, &vsock->queued_replies);
337 		if (new_cnt + cnt >= tx_vq->num && new_cnt < tx_vq->num)
338 			vhost_poll_queue(&tx_vq->poll);
339 	}
340 
341 	ret = 0;
342 out:
343 	rcu_read_unlock();
344 	return ret;
345 }
346 
347 static struct sk_buff *
348 vhost_vsock_alloc_skb(struct vhost_virtqueue *vq,
349 		      unsigned int out, unsigned int in)
350 {
351 	struct virtio_vsock_hdr *hdr;
352 	struct iov_iter iov_iter;
353 	struct sk_buff *skb;
354 	size_t payload_len;
355 	size_t nbytes;
356 	size_t len;
357 
358 	if (in != 0) {
359 		vq_err(vq, "Expected 0 input buffers, got %u\n", in);
360 		return NULL;
361 	}
362 
363 	len = iov_length(vq->iov, out);
364 
365 	if (len < VIRTIO_VSOCK_SKB_HEADROOM ||
366 	    len > VIRTIO_VSOCK_MAX_PKT_BUF_SIZE + VIRTIO_VSOCK_SKB_HEADROOM)
367 		return NULL;
368 
369 	/* len contains both payload and hdr */
370 	skb = virtio_vsock_alloc_skb(len, GFP_KERNEL);
371 	if (!skb)
372 		return NULL;
373 
374 	iov_iter_init(&iov_iter, ITER_SOURCE, vq->iov, out, len);
375 
376 	hdr = virtio_vsock_hdr(skb);
377 	nbytes = copy_from_iter(hdr, sizeof(*hdr), &iov_iter);
378 	if (nbytes != sizeof(*hdr)) {
379 		vq_err(vq, "Expected %zu bytes for pkt->hdr, got %zu bytes\n",
380 		       sizeof(*hdr), nbytes);
381 		kfree_skb(skb);
382 		return NULL;
383 	}
384 
385 	payload_len = le32_to_cpu(hdr->len);
386 
387 	/* No payload */
388 	if (!payload_len)
389 		return skb;
390 
391 	/* The pkt is too big or the length in the header is invalid */
392 	if (payload_len + sizeof(*hdr) > len) {
393 		kfree_skb(skb);
394 		return NULL;
395 	}
396 
397 	virtio_vsock_skb_put(skb, payload_len);
398 
399 	if (skb_copy_datagram_from_iter(skb, 0, &iov_iter, payload_len)) {
400 		vq_err(vq, "Failed to copy %zu byte payload\n", payload_len);
401 		kfree_skb(skb);
402 		return NULL;
403 	}
404 
405 	return skb;
406 }
407 
408 /* Is there space left for replies to rx packets? */
409 static bool vhost_vsock_more_replies(struct vhost_vsock *vsock)
410 {
411 	struct vhost_virtqueue *vq = &vsock->vqs[VSOCK_VQ_TX];
412 	int val;
413 
414 	smp_rmb(); /* paired with atomic_inc() and atomic_dec_return() */
415 	val = atomic_read(&vsock->queued_replies);
416 
417 	return val < vq->num;
418 }
419 
420 static bool vhost_transport_msgzerocopy_allow(void)
421 {
422 	return true;
423 }
424 
425 static bool vhost_transport_seqpacket_allow(struct vsock_sock *vsk,
426 					    u32 remote_cid);
427 
428 static bool
429 vhost_transport_stream_allow(struct vsock_sock *vsk, u32 cid, u32 port)
430 {
431 	return true;
432 }
433 
434 static struct virtio_transport vhost_transport = {
435 	.transport = {
436 		.module                   = THIS_MODULE,
437 
438 		.get_local_cid            = vhost_transport_get_local_cid,
439 		.has_remote_cid           = vhost_transport_has_remote_cid,
440 
441 		.init                     = virtio_transport_do_socket_init,
442 		.destruct                 = virtio_transport_destruct,
443 		.release                  = virtio_transport_release,
444 		.connect                  = virtio_transport_connect,
445 		.shutdown                 = virtio_transport_shutdown,
446 		.cancel_pkt               = vhost_transport_cancel_pkt,
447 
448 		.dgram_enqueue            = virtio_transport_dgram_enqueue,
449 		.dgram_dequeue            = virtio_transport_dgram_dequeue,
450 		.dgram_bind               = virtio_transport_dgram_bind,
451 		.dgram_allow              = virtio_transport_dgram_allow,
452 
453 		.stream_enqueue           = virtio_transport_stream_enqueue,
454 		.stream_dequeue           = virtio_transport_stream_dequeue,
455 		.stream_has_data          = virtio_transport_stream_has_data,
456 		.stream_has_space         = virtio_transport_stream_has_space,
457 		.stream_rcvhiwat          = virtio_transport_stream_rcvhiwat,
458 		.stream_is_active         = virtio_transport_stream_is_active,
459 		.stream_allow             = vhost_transport_stream_allow,
460 
461 		.seqpacket_dequeue        = virtio_transport_seqpacket_dequeue,
462 		.seqpacket_enqueue        = virtio_transport_seqpacket_enqueue,
463 		.seqpacket_allow          = vhost_transport_seqpacket_allow,
464 		.seqpacket_has_data       = virtio_transport_seqpacket_has_data,
465 
466 		.msgzerocopy_allow        = vhost_transport_msgzerocopy_allow,
467 
468 		.notify_poll_in           = virtio_transport_notify_poll_in,
469 		.notify_poll_out          = virtio_transport_notify_poll_out,
470 		.notify_recv_init         = virtio_transport_notify_recv_init,
471 		.notify_recv_pre_block    = virtio_transport_notify_recv_pre_block,
472 		.notify_recv_pre_dequeue  = virtio_transport_notify_recv_pre_dequeue,
473 		.notify_recv_post_dequeue = virtio_transport_notify_recv_post_dequeue,
474 		.notify_send_init         = virtio_transport_notify_send_init,
475 		.notify_send_pre_block    = virtio_transport_notify_send_pre_block,
476 		.notify_send_pre_enqueue  = virtio_transport_notify_send_pre_enqueue,
477 		.notify_send_post_enqueue = virtio_transport_notify_send_post_enqueue,
478 		.notify_buffer_size       = virtio_transport_notify_buffer_size,
479 		.notify_set_rcvlowat      = virtio_transport_notify_set_rcvlowat,
480 
481 		.unsent_bytes             = virtio_transport_unsent_bytes,
482 
483 		.read_skb = virtio_transport_read_skb,
484 	},
485 
486 	.send_pkt = vhost_transport_send_pkt,
487 };
488 
489 static bool vhost_transport_seqpacket_allow(struct vsock_sock *vsk,
490 					    u32 remote_cid)
491 {
492 	struct net *net = sock_net(sk_vsock(vsk));
493 	struct vhost_vsock *vsock;
494 	bool seqpacket_allow = false;
495 
496 	rcu_read_lock();
497 	vsock = vhost_vsock_get(remote_cid, net);
498 
499 	if (vsock)
500 		seqpacket_allow = vsock->seqpacket_allow;
501 
502 	rcu_read_unlock();
503 
504 	return seqpacket_allow;
505 }
506 
507 static void vhost_vsock_handle_tx_kick(struct vhost_work *work)
508 {
509 	struct vhost_virtqueue *vq = container_of(work, struct vhost_virtqueue,
510 						  poll.work);
511 	struct vhost_vsock *vsock = container_of(vq->dev, struct vhost_vsock,
512 						 dev);
513 	int head, pkts = 0, total_len = 0;
514 	unsigned int out, in;
515 	struct sk_buff *skb;
516 	bool added = false;
517 
518 	mutex_lock(&vq->mutex);
519 
520 	if (!vhost_vq_get_backend(vq))
521 		goto out;
522 
523 	if (!vq_meta_prefetch(vq))
524 		goto out;
525 
526 	vhost_disable_notify(&vsock->dev, vq);
527 	do {
528 		struct virtio_vsock_hdr *hdr;
529 
530 		if (!vhost_vsock_more_replies(vsock)) {
531 			/* Stop tx until the device processes already
532 			 * pending replies.  Leave tx virtqueue
533 			 * callbacks disabled.
534 			 */
535 			goto no_more_replies;
536 		}
537 
538 		head = vhost_get_vq_desc(vq, vq->iov, ARRAY_SIZE(vq->iov),
539 					 &out, &in, NULL, NULL);
540 		if (head < 0)
541 			break;
542 
543 		if (head == vq->num) {
544 			if (unlikely(vhost_enable_notify(&vsock->dev, vq))) {
545 				vhost_disable_notify(&vsock->dev, vq);
546 				continue;
547 			}
548 			break;
549 		}
550 
551 		skb = vhost_vsock_alloc_skb(vq, out, in);
552 		if (!skb) {
553 			vq_err(vq, "Faulted on pkt\n");
554 			continue;
555 		}
556 
557 		total_len += sizeof(*hdr) + skb->len;
558 
559 		/* Deliver to monitoring devices all received packets */
560 		virtio_transport_deliver_tap_pkt(skb);
561 
562 		hdr = virtio_vsock_hdr(skb);
563 
564 		/* Only accept correctly addressed packets */
565 		if (le64_to_cpu(hdr->src_cid) == vsock->guest_cid &&
566 		    le64_to_cpu(hdr->dst_cid) ==
567 		    vhost_transport_get_local_cid())
568 			virtio_transport_recv_pkt(&vhost_transport, skb,
569 						  vsock->net);
570 		else
571 			kfree_skb(skb);
572 
573 		vhost_add_used(vq, head, 0);
574 		added = true;
575 	} while(likely(!vhost_exceeds_weight(vq, ++pkts, total_len)));
576 
577 no_more_replies:
578 	if (added)
579 		vhost_signal(&vsock->dev, vq);
580 
581 out:
582 	mutex_unlock(&vq->mutex);
583 }
584 
585 static void vhost_vsock_handle_rx_kick(struct vhost_work *work)
586 {
587 	struct vhost_virtqueue *vq = container_of(work, struct vhost_virtqueue,
588 						poll.work);
589 	struct vhost_vsock *vsock = container_of(vq->dev, struct vhost_vsock,
590 						 dev);
591 
592 	vhost_transport_do_send_pkt(vsock, vq);
593 }
594 
595 static int vhost_vsock_start(struct vhost_vsock *vsock)
596 {
597 	struct vhost_virtqueue *vq;
598 	size_t i;
599 	int ret;
600 
601 	mutex_lock(&vsock->dev.mutex);
602 
603 	ret = vhost_dev_check_owner(&vsock->dev);
604 	if (ret)
605 		goto err;
606 
607 	for (i = 0; i < ARRAY_SIZE(vsock->vqs); i++) {
608 		vq = &vsock->vqs[i];
609 
610 		mutex_lock(&vq->mutex);
611 
612 		if (!vhost_vq_access_ok(vq)) {
613 			ret = -EFAULT;
614 			goto err_vq;
615 		}
616 
617 		if (!vhost_vq_get_backend(vq)) {
618 			vhost_vq_set_backend(vq, vsock);
619 			ret = vhost_vq_init_access(vq);
620 			if (ret)
621 				goto err_vq;
622 		}
623 
624 		mutex_unlock(&vq->mutex);
625 	}
626 
627 	/* Some packets may have been queued before the device was started,
628 	 * let's kick the send worker to send them.
629 	 */
630 	vhost_vq_work_queue(&vsock->vqs[VSOCK_VQ_RX], &vsock->send_pkt_work);
631 
632 	mutex_unlock(&vsock->dev.mutex);
633 	return 0;
634 
635 err_vq:
636 	vhost_vq_set_backend(vq, NULL);
637 	mutex_unlock(&vq->mutex);
638 
639 	for (i = 0; i < ARRAY_SIZE(vsock->vqs); i++) {
640 		vq = &vsock->vqs[i];
641 
642 		mutex_lock(&vq->mutex);
643 		vhost_vq_set_backend(vq, NULL);
644 		mutex_unlock(&vq->mutex);
645 	}
646 err:
647 	mutex_unlock(&vsock->dev.mutex);
648 	return ret;
649 }
650 
651 static int vhost_vsock_stop(struct vhost_vsock *vsock, bool check_owner)
652 {
653 	size_t i;
654 	int ret = 0;
655 
656 	mutex_lock(&vsock->dev.mutex);
657 
658 	if (check_owner) {
659 		ret = vhost_dev_check_owner(&vsock->dev);
660 		if (ret)
661 			goto err;
662 	}
663 
664 	for (i = 0; i < ARRAY_SIZE(vsock->vqs); i++) {
665 		struct vhost_virtqueue *vq = &vsock->vqs[i];
666 
667 		mutex_lock(&vq->mutex);
668 		vhost_vq_set_backend(vq, NULL);
669 		mutex_unlock(&vq->mutex);
670 	}
671 
672 err:
673 	mutex_unlock(&vsock->dev.mutex);
674 	return ret;
675 }
676 
677 static void vhost_vsock_free(struct vhost_vsock *vsock)
678 {
679 	kvfree(vsock);
680 }
681 
682 static int vhost_vsock_dev_open(struct inode *inode, struct file *file)
683 {
684 	struct vhost_virtqueue **vqs;
685 	struct vhost_vsock *vsock;
686 	struct net *net;
687 	int ret;
688 
689 	/* This struct is large and allocation could fail, fall back to vmalloc
690 	 * if there is no other way.
691 	 */
692 	vsock = kvmalloc_obj(*vsock, GFP_KERNEL | __GFP_RETRY_MAYFAIL);
693 	if (!vsock)
694 		return -ENOMEM;
695 
696 	vqs = kmalloc_objs(*vqs, ARRAY_SIZE(vsock->vqs));
697 	if (!vqs) {
698 		ret = -ENOMEM;
699 		goto out;
700 	}
701 
702 	net = current->nsproxy->net_ns;
703 	vsock->net = get_net_track(net, &vsock->ns_tracker, GFP_KERNEL);
704 
705 	vsock->guest_cid = 0; /* no CID assigned yet */
706 	vsock->seqpacket_allow = false;
707 
708 	atomic_set(&vsock->queued_replies, 0);
709 
710 	vqs[VSOCK_VQ_TX] = &vsock->vqs[VSOCK_VQ_TX];
711 	vqs[VSOCK_VQ_RX] = &vsock->vqs[VSOCK_VQ_RX];
712 	vsock->vqs[VSOCK_VQ_TX].handle_kick = vhost_vsock_handle_tx_kick;
713 	vsock->vqs[VSOCK_VQ_RX].handle_kick = vhost_vsock_handle_rx_kick;
714 
715 	vhost_dev_init(&vsock->dev, vqs, ARRAY_SIZE(vsock->vqs),
716 		       UIO_MAXIOV, VHOST_VSOCK_PKT_WEIGHT,
717 		       VHOST_VSOCK_WEIGHT, true, NULL);
718 
719 	file->private_data = vsock;
720 	skb_queue_head_init(&vsock->send_pkt_queue);
721 	vhost_work_init(&vsock->send_pkt_work, vhost_transport_send_pkt_work);
722 	return 0;
723 
724 out:
725 	vhost_vsock_free(vsock);
726 	return ret;
727 }
728 
729 static void vhost_vsock_flush(struct vhost_vsock *vsock)
730 {
731 	vhost_dev_flush(&vsock->dev);
732 }
733 
734 static void vhost_vsock_reset_orphans(struct sock *sk)
735 {
736 	struct vsock_sock *vsk = vsock_sk(sk);
737 
738 	/* vmci_transport.c doesn't take sk_lock here either.  At least we're
739 	 * under vsock_table_lock so the sock cannot disappear while we're
740 	 * executing.
741 	 */
742 
743 	rcu_read_lock();
744 
745 	/* If the peer is still valid, no need to reset connection */
746 	if (vhost_vsock_get(vsk->remote_addr.svm_cid, sock_net(sk))) {
747 		rcu_read_unlock();
748 		return;
749 	}
750 
751 	rcu_read_unlock();
752 
753 	/* If the close timeout is pending, let it expire.  This avoids races
754 	 * with the timeout callback.
755 	 */
756 	if (vsk->close_work_scheduled)
757 		return;
758 
759 	sock_set_flag(sk, SOCK_DONE);
760 	vsk->peer_shutdown = SHUTDOWN_MASK;
761 	sk->sk_state = SS_UNCONNECTED;
762 	sk->sk_err = ECONNRESET;
763 	sk_error_report(sk);
764 }
765 
766 static int vhost_vsock_dev_release(struct inode *inode, struct file *file)
767 {
768 	struct vhost_vsock *vsock = file->private_data;
769 
770 	mutex_lock(&vhost_vsock_mutex);
771 	if (vsock->guest_cid)
772 		hash_del_rcu(&vsock->hash);
773 	mutex_unlock(&vhost_vsock_mutex);
774 
775 	/* Wait for other CPUs to finish using vsock */
776 	synchronize_rcu();
777 
778 	/* Iterating over all connections for all CIDs to find orphans is
779 	 * inefficient.  Room for improvement here. */
780 	vsock_for_each_connected_socket(&vhost_transport.transport,
781 					vhost_vsock_reset_orphans);
782 
783 	/* Don't check the owner, because we are in the release path, so we
784 	 * need to stop the vsock device in any case.
785 	 * vhost_vsock_stop() can not fail in this case, so we don't need to
786 	 * check the return code.
787 	 */
788 	vhost_vsock_stop(vsock, false);
789 	vhost_vsock_flush(vsock);
790 	vhost_dev_stop(&vsock->dev);
791 
792 	virtio_vsock_skb_queue_purge(&vsock->send_pkt_queue);
793 
794 	vhost_dev_cleanup(&vsock->dev);
795 	put_net_track(vsock->net, &vsock->ns_tracker);
796 	kfree(vsock->dev.vqs);
797 	vhost_vsock_free(vsock);
798 	return 0;
799 }
800 
801 static int vhost_vsock_set_cid(struct vhost_vsock *vsock, u64 guest_cid)
802 {
803 	struct vhost_vsock *other;
804 
805 	/* Refuse reserved CIDs */
806 	if (guest_cid <= VMADDR_CID_HOST ||
807 	    guest_cid == U32_MAX)
808 		return -EINVAL;
809 
810 	/* 64-bit CIDs are not yet supported */
811 	if (guest_cid > U32_MAX)
812 		return -EINVAL;
813 
814 	/* Refuse if CID is assigned to the guest->host transport (i.e. nested
815 	 * VM), to make the loopback work.
816 	 */
817 	if (vsock_find_cid(guest_cid))
818 		return -EADDRINUSE;
819 
820 	/* Refuse if CID is already in use */
821 	mutex_lock(&vhost_vsock_mutex);
822 	other = vhost_vsock_get(guest_cid, vsock->net);
823 	if (other && other != vsock) {
824 		mutex_unlock(&vhost_vsock_mutex);
825 		return -EADDRINUSE;
826 	}
827 
828 	if (vsock->guest_cid)
829 		hash_del_rcu(&vsock->hash);
830 
831 	vsock->guest_cid = guest_cid;
832 	hash_add_rcu(vhost_vsock_hash, &vsock->hash, vsock->guest_cid);
833 	mutex_unlock(&vhost_vsock_mutex);
834 
835 	return 0;
836 }
837 
838 static int vhost_vsock_set_features(struct vhost_vsock *vsock, u64 features)
839 {
840 	struct vhost_virtqueue *vq;
841 	int i;
842 
843 	if (features & ~VHOST_VSOCK_FEATURES)
844 		return -EOPNOTSUPP;
845 
846 	mutex_lock(&vsock->dev.mutex);
847 	if ((features & (1 << VHOST_F_LOG_ALL)) &&
848 	    !vhost_log_access_ok(&vsock->dev)) {
849 		goto err;
850 	}
851 
852 	if ((features & (1ULL << VIRTIO_F_ACCESS_PLATFORM))) {
853 		if (vhost_init_device_iotlb(&vsock->dev))
854 			goto err;
855 	}
856 
857 	vsock->seqpacket_allow = features & (1ULL << VIRTIO_VSOCK_F_SEQPACKET);
858 
859 	for (i = 0; i < ARRAY_SIZE(vsock->vqs); i++) {
860 		vq = &vsock->vqs[i];
861 		mutex_lock(&vq->mutex);
862 		vq->acked_features = features;
863 		mutex_unlock(&vq->mutex);
864 	}
865 	mutex_unlock(&vsock->dev.mutex);
866 	return 0;
867 
868 err:
869 	mutex_unlock(&vsock->dev.mutex);
870 	return -EFAULT;
871 }
872 
873 static long vhost_vsock_dev_ioctl(struct file *f, unsigned int ioctl,
874 				  unsigned long arg)
875 {
876 	struct vhost_vsock *vsock = f->private_data;
877 	void __user *argp = (void __user *)arg;
878 	u64 guest_cid;
879 	u64 features;
880 	int start;
881 	int r;
882 
883 	switch (ioctl) {
884 	case VHOST_VSOCK_SET_GUEST_CID:
885 		if (copy_from_user(&guest_cid, argp, sizeof(guest_cid)))
886 			return -EFAULT;
887 		return vhost_vsock_set_cid(vsock, guest_cid);
888 	case VHOST_VSOCK_SET_RUNNING:
889 		if (copy_from_user(&start, argp, sizeof(start)))
890 			return -EFAULT;
891 		if (start)
892 			return vhost_vsock_start(vsock);
893 		else
894 			return vhost_vsock_stop(vsock, true);
895 	case VHOST_GET_FEATURES:
896 		features = VHOST_VSOCK_FEATURES;
897 		if (copy_to_user(argp, &features, sizeof(features)))
898 			return -EFAULT;
899 		return 0;
900 	case VHOST_SET_FEATURES:
901 		if (copy_from_user(&features, argp, sizeof(features)))
902 			return -EFAULT;
903 		return vhost_vsock_set_features(vsock, features);
904 	case VHOST_GET_BACKEND_FEATURES:
905 		features = VHOST_VSOCK_BACKEND_FEATURES;
906 		if (copy_to_user(argp, &features, sizeof(features)))
907 			return -EFAULT;
908 		return 0;
909 	case VHOST_SET_BACKEND_FEATURES:
910 		if (copy_from_user(&features, argp, sizeof(features)))
911 			return -EFAULT;
912 		if (features & ~VHOST_VSOCK_BACKEND_FEATURES)
913 			return -EOPNOTSUPP;
914 		vhost_set_backend_features(&vsock->dev, features);
915 		return 0;
916 	default:
917 		mutex_lock(&vsock->dev.mutex);
918 		r = vhost_dev_ioctl(&vsock->dev, ioctl, argp);
919 		if (r == -ENOIOCTLCMD)
920 			r = vhost_vring_ioctl(&vsock->dev, ioctl, argp);
921 		else
922 			vhost_vsock_flush(vsock);
923 		mutex_unlock(&vsock->dev.mutex);
924 		return r;
925 	}
926 }
927 
928 static ssize_t vhost_vsock_chr_read_iter(struct kiocb *iocb, struct iov_iter *to)
929 {
930 	struct file *file = iocb->ki_filp;
931 	struct vhost_vsock *vsock = file->private_data;
932 	struct vhost_dev *dev = &vsock->dev;
933 	int noblock = file->f_flags & O_NONBLOCK;
934 
935 	return vhost_chr_read_iter(dev, to, noblock);
936 }
937 
938 static ssize_t vhost_vsock_chr_write_iter(struct kiocb *iocb,
939 					struct iov_iter *from)
940 {
941 	struct file *file = iocb->ki_filp;
942 	struct vhost_vsock *vsock = file->private_data;
943 	struct vhost_dev *dev = &vsock->dev;
944 
945 	return vhost_chr_write_iter(dev, from);
946 }
947 
948 static __poll_t vhost_vsock_chr_poll(struct file *file, poll_table *wait)
949 {
950 	struct vhost_vsock *vsock = file->private_data;
951 	struct vhost_dev *dev = &vsock->dev;
952 
953 	return vhost_chr_poll(file, dev, wait);
954 }
955 
956 static const struct file_operations vhost_vsock_fops = {
957 	.owner          = THIS_MODULE,
958 	.open           = vhost_vsock_dev_open,
959 	.release        = vhost_vsock_dev_release,
960 	.llseek		= noop_llseek,
961 	.unlocked_ioctl = vhost_vsock_dev_ioctl,
962 	.compat_ioctl   = compat_ptr_ioctl,
963 	.read_iter      = vhost_vsock_chr_read_iter,
964 	.write_iter     = vhost_vsock_chr_write_iter,
965 	.poll           = vhost_vsock_chr_poll,
966 };
967 
968 static struct miscdevice vhost_vsock_misc = {
969 	.minor = VHOST_VSOCK_MINOR,
970 	.name = "vhost-vsock",
971 	.fops = &vhost_vsock_fops,
972 };
973 
974 static int __init vhost_vsock_init(void)
975 {
976 	int ret;
977 
978 	ret = vsock_core_register(&vhost_transport.transport,
979 				  VSOCK_TRANSPORT_F_H2G);
980 	if (ret < 0)
981 		return ret;
982 
983 	ret = misc_register(&vhost_vsock_misc);
984 	if (ret) {
985 		vsock_core_unregister(&vhost_transport.transport);
986 		return ret;
987 	}
988 
989 	return 0;
990 };
991 
992 static void __exit vhost_vsock_exit(void)
993 {
994 	misc_deregister(&vhost_vsock_misc);
995 	vsock_core_unregister(&vhost_transport.transport);
996 };
997 
998 module_init(vhost_vsock_init);
999 module_exit(vhost_vsock_exit);
1000 MODULE_LICENSE("GPL v2");
1001 MODULE_AUTHOR("Asias He");
1002 MODULE_DESCRIPTION("vhost transport for vsock ");
1003 MODULE_ALIAS_MISCDEV(VHOST_VSOCK_MINOR);
1004 MODULE_ALIAS("devname:vhost-vsock");
1005