1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * common code for virtio 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/spinlock.h>
10 #include <linux/module.h>
11 #include <linux/sched/signal.h>
12 #include <linux/ctype.h>
13 #include <linux/list.h>
14 #include <linux/virtio_vsock.h>
15 #include <uapi/linux/vsockmon.h>
16 
17 #include <net/sock.h>
18 #include <net/af_vsock.h>
19 
20 #define CREATE_TRACE_POINTS
21 #include <trace/events/vsock_virtio_transport_common.h>
22 
23 /* How long to wait for graceful shutdown of a connection */
24 #define VSOCK_CLOSE_TIMEOUT (8 * HZ)
25 
26 /* Threshold for detecting small packets to copy */
27 #define GOOD_COPY_LEN  128
28 
29 static const struct virtio_transport *
30 virtio_transport_get_ops(struct vsock_sock *vsk)
31 {
32 	const struct vsock_transport *t = vsock_core_get_transport(vsk);
33 
34 	if (WARN_ON(!t))
35 		return NULL;
36 
37 	return container_of(t, struct virtio_transport, transport);
38 }
39 
40 /* Returns a new packet on success, otherwise returns NULL.
41  *
42  * If NULL is returned, errp is set to a negative errno.
43  */
44 static struct sk_buff *
45 virtio_transport_alloc_skb(struct virtio_vsock_pkt_info *info,
46 			   size_t len,
47 			   u32 src_cid,
48 			   u32 src_port,
49 			   u32 dst_cid,
50 			   u32 dst_port)
51 {
52 	const size_t skb_len = VIRTIO_VSOCK_SKB_HEADROOM + len;
53 	struct virtio_vsock_hdr *hdr;
54 	struct sk_buff *skb;
55 	void *payload;
56 	int err;
57 
58 	skb = virtio_vsock_alloc_skb(skb_len, GFP_KERNEL);
59 	if (!skb)
60 		return NULL;
61 
62 	hdr = virtio_vsock_hdr(skb);
63 	hdr->type	= cpu_to_le16(info->type);
64 	hdr->op		= cpu_to_le16(info->op);
65 	hdr->src_cid	= cpu_to_le64(src_cid);
66 	hdr->dst_cid	= cpu_to_le64(dst_cid);
67 	hdr->src_port	= cpu_to_le32(src_port);
68 	hdr->dst_port	= cpu_to_le32(dst_port);
69 	hdr->flags	= cpu_to_le32(info->flags);
70 	hdr->len	= cpu_to_le32(len);
71 	hdr->buf_alloc	= cpu_to_le32(0);
72 	hdr->fwd_cnt	= cpu_to_le32(0);
73 
74 	if (info->msg && len > 0) {
75 		payload = skb_put(skb, len);
76 		err = memcpy_from_msg(payload, info->msg, len);
77 		if (err)
78 			goto out;
79 
80 		if (msg_data_left(info->msg) == 0 &&
81 		    info->type == VIRTIO_VSOCK_TYPE_SEQPACKET) {
82 			hdr->flags |= cpu_to_le32(VIRTIO_VSOCK_SEQ_EOM);
83 
84 			if (info->msg->msg_flags & MSG_EOR)
85 				hdr->flags |= cpu_to_le32(VIRTIO_VSOCK_SEQ_EOR);
86 		}
87 	}
88 
89 	if (info->reply)
90 		virtio_vsock_skb_set_reply(skb);
91 
92 	trace_virtio_transport_alloc_pkt(src_cid, src_port,
93 					 dst_cid, dst_port,
94 					 len,
95 					 info->type,
96 					 info->op,
97 					 info->flags);
98 
99 	if (info->vsk && !skb_set_owner_sk_safe(skb, sk_vsock(info->vsk))) {
100 		WARN_ONCE(1, "failed to allocate skb on vsock socket with sk_refcnt == 0\n");
101 		goto out;
102 	}
103 
104 	return skb;
105 
106 out:
107 	kfree_skb(skb);
108 	return NULL;
109 }
110 
111 /* Packet capture */
112 static struct sk_buff *virtio_transport_build_skb(void *opaque)
113 {
114 	struct virtio_vsock_hdr *pkt_hdr;
115 	struct sk_buff *pkt = opaque;
116 	struct af_vsockmon_hdr *hdr;
117 	struct sk_buff *skb;
118 	size_t payload_len;
119 	void *payload_buf;
120 
121 	/* A packet could be split to fit the RX buffer, so we can retrieve
122 	 * the payload length from the header and the buffer pointer taking
123 	 * care of the offset in the original packet.
124 	 */
125 	pkt_hdr = virtio_vsock_hdr(pkt);
126 	payload_len = pkt->len;
127 	payload_buf = pkt->data;
128 
129 	skb = alloc_skb(sizeof(*hdr) + sizeof(*pkt_hdr) + payload_len,
130 			GFP_ATOMIC);
131 	if (!skb)
132 		return NULL;
133 
134 	hdr = skb_put(skb, sizeof(*hdr));
135 
136 	/* pkt->hdr is little-endian so no need to byteswap here */
137 	hdr->src_cid = pkt_hdr->src_cid;
138 	hdr->src_port = pkt_hdr->src_port;
139 	hdr->dst_cid = pkt_hdr->dst_cid;
140 	hdr->dst_port = pkt_hdr->dst_port;
141 
142 	hdr->transport = cpu_to_le16(AF_VSOCK_TRANSPORT_VIRTIO);
143 	hdr->len = cpu_to_le16(sizeof(*pkt_hdr));
144 	memset(hdr->reserved, 0, sizeof(hdr->reserved));
145 
146 	switch (le16_to_cpu(pkt_hdr->op)) {
147 	case VIRTIO_VSOCK_OP_REQUEST:
148 	case VIRTIO_VSOCK_OP_RESPONSE:
149 		hdr->op = cpu_to_le16(AF_VSOCK_OP_CONNECT);
150 		break;
151 	case VIRTIO_VSOCK_OP_RST:
152 	case VIRTIO_VSOCK_OP_SHUTDOWN:
153 		hdr->op = cpu_to_le16(AF_VSOCK_OP_DISCONNECT);
154 		break;
155 	case VIRTIO_VSOCK_OP_RW:
156 		hdr->op = cpu_to_le16(AF_VSOCK_OP_PAYLOAD);
157 		break;
158 	case VIRTIO_VSOCK_OP_CREDIT_UPDATE:
159 	case VIRTIO_VSOCK_OP_CREDIT_REQUEST:
160 		hdr->op = cpu_to_le16(AF_VSOCK_OP_CONTROL);
161 		break;
162 	default:
163 		hdr->op = cpu_to_le16(AF_VSOCK_OP_UNKNOWN);
164 		break;
165 	}
166 
167 	skb_put_data(skb, pkt_hdr, sizeof(*pkt_hdr));
168 
169 	if (payload_len) {
170 		skb_put_data(skb, payload_buf, payload_len);
171 	}
172 
173 	return skb;
174 }
175 
176 void virtio_transport_deliver_tap_pkt(struct sk_buff *skb)
177 {
178 	if (virtio_vsock_skb_tap_delivered(skb))
179 		return;
180 
181 	vsock_deliver_tap(virtio_transport_build_skb, skb);
182 	virtio_vsock_skb_set_tap_delivered(skb);
183 }
184 EXPORT_SYMBOL_GPL(virtio_transport_deliver_tap_pkt);
185 
186 static u16 virtio_transport_get_type(struct sock *sk)
187 {
188 	if (sk->sk_type == SOCK_STREAM)
189 		return VIRTIO_VSOCK_TYPE_STREAM;
190 	else
191 		return VIRTIO_VSOCK_TYPE_SEQPACKET;
192 }
193 
194 /* This function can only be used on connecting/connected sockets,
195  * since a socket assigned to a transport is required.
196  *
197  * Do not use on listener sockets!
198  */
199 static int virtio_transport_send_pkt_info(struct vsock_sock *vsk,
200 					  struct virtio_vsock_pkt_info *info)
201 {
202 	u32 src_cid, src_port, dst_cid, dst_port;
203 	const struct virtio_transport *t_ops;
204 	struct virtio_vsock_sock *vvs;
205 	u32 pkt_len = info->pkt_len;
206 	u32 rest_len;
207 	int ret;
208 
209 	info->type = virtio_transport_get_type(sk_vsock(vsk));
210 
211 	t_ops = virtio_transport_get_ops(vsk);
212 	if (unlikely(!t_ops))
213 		return -EFAULT;
214 
215 	src_cid = t_ops->transport.get_local_cid();
216 	src_port = vsk->local_addr.svm_port;
217 	if (!info->remote_cid) {
218 		dst_cid	= vsk->remote_addr.svm_cid;
219 		dst_port = vsk->remote_addr.svm_port;
220 	} else {
221 		dst_cid = info->remote_cid;
222 		dst_port = info->remote_port;
223 	}
224 
225 	vvs = vsk->trans;
226 
227 	/* virtio_transport_get_credit might return less than pkt_len credit */
228 	pkt_len = virtio_transport_get_credit(vvs, pkt_len);
229 
230 	/* Do not send zero length OP_RW pkt */
231 	if (pkt_len == 0 && info->op == VIRTIO_VSOCK_OP_RW)
232 		return pkt_len;
233 
234 	rest_len = pkt_len;
235 
236 	do {
237 		struct sk_buff *skb;
238 		size_t skb_len;
239 
240 		skb_len = min_t(u32, VIRTIO_VSOCK_MAX_PKT_BUF_SIZE, rest_len);
241 
242 		skb = virtio_transport_alloc_skb(info, skb_len,
243 						 src_cid, src_port,
244 						 dst_cid, dst_port);
245 		if (!skb) {
246 			ret = -ENOMEM;
247 			break;
248 		}
249 
250 		virtio_transport_inc_tx_pkt(vvs, skb);
251 
252 		ret = t_ops->send_pkt(skb);
253 		if (ret < 0)
254 			break;
255 
256 		/* Both virtio and vhost 'send_pkt()' returns 'skb_len',
257 		 * but for reliability use 'ret' instead of 'skb_len'.
258 		 * Also if partial send happens (e.g. 'ret' != 'skb_len')
259 		 * somehow, we break this loop, but account such returned
260 		 * value in 'virtio_transport_put_credit()'.
261 		 */
262 		rest_len -= ret;
263 
264 		if (WARN_ONCE(ret != skb_len,
265 			      "'send_pkt()' returns %i, but %zu expected\n",
266 			      ret, skb_len))
267 			break;
268 	} while (rest_len);
269 
270 	virtio_transport_put_credit(vvs, rest_len);
271 
272 	/* Return number of bytes, if any data has been sent. */
273 	if (rest_len != pkt_len)
274 		ret = pkt_len - rest_len;
275 
276 	return ret;
277 }
278 
279 static bool virtio_transport_inc_rx_pkt(struct virtio_vsock_sock *vvs,
280 					u32 len)
281 {
282 	if (vvs->rx_bytes + len > vvs->buf_alloc)
283 		return false;
284 
285 	vvs->rx_bytes += len;
286 	return true;
287 }
288 
289 static void virtio_transport_dec_rx_pkt(struct virtio_vsock_sock *vvs,
290 					u32 len)
291 {
292 	vvs->rx_bytes -= len;
293 	vvs->fwd_cnt += len;
294 }
295 
296 void virtio_transport_inc_tx_pkt(struct virtio_vsock_sock *vvs, struct sk_buff *skb)
297 {
298 	struct virtio_vsock_hdr *hdr = virtio_vsock_hdr(skb);
299 
300 	spin_lock_bh(&vvs->rx_lock);
301 	vvs->last_fwd_cnt = vvs->fwd_cnt;
302 	hdr->fwd_cnt = cpu_to_le32(vvs->fwd_cnt);
303 	hdr->buf_alloc = cpu_to_le32(vvs->buf_alloc);
304 	spin_unlock_bh(&vvs->rx_lock);
305 }
306 EXPORT_SYMBOL_GPL(virtio_transport_inc_tx_pkt);
307 
308 u32 virtio_transport_get_credit(struct virtio_vsock_sock *vvs, u32 credit)
309 {
310 	u32 ret;
311 
312 	if (!credit)
313 		return 0;
314 
315 	spin_lock_bh(&vvs->tx_lock);
316 	ret = vvs->peer_buf_alloc - (vvs->tx_cnt - vvs->peer_fwd_cnt);
317 	if (ret > credit)
318 		ret = credit;
319 	vvs->tx_cnt += ret;
320 	spin_unlock_bh(&vvs->tx_lock);
321 
322 	return ret;
323 }
324 EXPORT_SYMBOL_GPL(virtio_transport_get_credit);
325 
326 void virtio_transport_put_credit(struct virtio_vsock_sock *vvs, u32 credit)
327 {
328 	if (!credit)
329 		return;
330 
331 	spin_lock_bh(&vvs->tx_lock);
332 	vvs->tx_cnt -= credit;
333 	spin_unlock_bh(&vvs->tx_lock);
334 }
335 EXPORT_SYMBOL_GPL(virtio_transport_put_credit);
336 
337 static int virtio_transport_send_credit_update(struct vsock_sock *vsk)
338 {
339 	struct virtio_vsock_pkt_info info = {
340 		.op = VIRTIO_VSOCK_OP_CREDIT_UPDATE,
341 		.vsk = vsk,
342 	};
343 
344 	return virtio_transport_send_pkt_info(vsk, &info);
345 }
346 
347 static ssize_t
348 virtio_transport_stream_do_peek(struct vsock_sock *vsk,
349 				struct msghdr *msg,
350 				size_t len)
351 {
352 	struct virtio_vsock_sock *vvs = vsk->trans;
353 	struct sk_buff *skb;
354 	size_t total = 0;
355 	int err;
356 
357 	spin_lock_bh(&vvs->rx_lock);
358 
359 	skb_queue_walk(&vvs->rx_queue, skb) {
360 		size_t bytes;
361 
362 		bytes = len - total;
363 		if (bytes > skb->len)
364 			bytes = skb->len;
365 
366 		spin_unlock_bh(&vvs->rx_lock);
367 
368 		/* sk_lock is held by caller so no one else can dequeue.
369 		 * Unlock rx_lock since memcpy_to_msg() may sleep.
370 		 */
371 		err = memcpy_to_msg(msg, skb->data, bytes);
372 		if (err)
373 			goto out;
374 
375 		total += bytes;
376 
377 		spin_lock_bh(&vvs->rx_lock);
378 
379 		if (total == len)
380 			break;
381 	}
382 
383 	spin_unlock_bh(&vvs->rx_lock);
384 
385 	return total;
386 
387 out:
388 	if (total)
389 		err = total;
390 	return err;
391 }
392 
393 static ssize_t
394 virtio_transport_stream_do_dequeue(struct vsock_sock *vsk,
395 				   struct msghdr *msg,
396 				   size_t len)
397 {
398 	struct virtio_vsock_sock *vvs = vsk->trans;
399 	size_t bytes, total = 0;
400 	struct sk_buff *skb;
401 	u32 fwd_cnt_delta;
402 	bool low_rx_bytes;
403 	int err = -EFAULT;
404 	u32 free_space;
405 
406 	spin_lock_bh(&vvs->rx_lock);
407 
408 	if (WARN_ONCE(skb_queue_empty(&vvs->rx_queue) && vvs->rx_bytes,
409 		      "rx_queue is empty, but rx_bytes is non-zero\n")) {
410 		spin_unlock_bh(&vvs->rx_lock);
411 		return err;
412 	}
413 
414 	while (total < len && !skb_queue_empty(&vvs->rx_queue)) {
415 		skb = skb_peek(&vvs->rx_queue);
416 
417 		bytes = len - total;
418 		if (bytes > skb->len)
419 			bytes = skb->len;
420 
421 		/* sk_lock is held by caller so no one else can dequeue.
422 		 * Unlock rx_lock since memcpy_to_msg() may sleep.
423 		 */
424 		spin_unlock_bh(&vvs->rx_lock);
425 
426 		err = memcpy_to_msg(msg, skb->data, bytes);
427 		if (err)
428 			goto out;
429 
430 		spin_lock_bh(&vvs->rx_lock);
431 
432 		total += bytes;
433 		skb_pull(skb, bytes);
434 
435 		if (skb->len == 0) {
436 			u32 pkt_len = le32_to_cpu(virtio_vsock_hdr(skb)->len);
437 
438 			virtio_transport_dec_rx_pkt(vvs, pkt_len);
439 			__skb_unlink(skb, &vvs->rx_queue);
440 			consume_skb(skb);
441 		}
442 	}
443 
444 	fwd_cnt_delta = vvs->fwd_cnt - vvs->last_fwd_cnt;
445 	free_space = vvs->buf_alloc - fwd_cnt_delta;
446 	low_rx_bytes = (vvs->rx_bytes <
447 			sock_rcvlowat(sk_vsock(vsk), 0, INT_MAX));
448 
449 	spin_unlock_bh(&vvs->rx_lock);
450 
451 	/* To reduce the number of credit update messages,
452 	 * don't update credits as long as lots of space is available.
453 	 * Note: the limit chosen here is arbitrary. Setting the limit
454 	 * too high causes extra messages. Too low causes transmitter
455 	 * stalls. As stalls are in theory more expensive than extra
456 	 * messages, we set the limit to a high value. TODO: experiment
457 	 * with different values. Also send credit update message when
458 	 * number of bytes in rx queue is not enough to wake up reader.
459 	 */
460 	if (fwd_cnt_delta &&
461 	    (free_space < VIRTIO_VSOCK_MAX_PKT_BUF_SIZE || low_rx_bytes))
462 		virtio_transport_send_credit_update(vsk);
463 
464 	return total;
465 
466 out:
467 	if (total)
468 		err = total;
469 	return err;
470 }
471 
472 static ssize_t
473 virtio_transport_seqpacket_do_peek(struct vsock_sock *vsk,
474 				   struct msghdr *msg)
475 {
476 	struct virtio_vsock_sock *vvs = vsk->trans;
477 	struct sk_buff *skb;
478 	size_t total, len;
479 
480 	spin_lock_bh(&vvs->rx_lock);
481 
482 	if (!vvs->msg_count) {
483 		spin_unlock_bh(&vvs->rx_lock);
484 		return 0;
485 	}
486 
487 	total = 0;
488 	len = msg_data_left(msg);
489 
490 	skb_queue_walk(&vvs->rx_queue, skb) {
491 		struct virtio_vsock_hdr *hdr;
492 
493 		if (total < len) {
494 			size_t bytes;
495 			int err;
496 
497 			bytes = len - total;
498 			if (bytes > skb->len)
499 				bytes = skb->len;
500 
501 			spin_unlock_bh(&vvs->rx_lock);
502 
503 			/* sk_lock is held by caller so no one else can dequeue.
504 			 * Unlock rx_lock since memcpy_to_msg() may sleep.
505 			 */
506 			err = memcpy_to_msg(msg, skb->data, bytes);
507 			if (err)
508 				return err;
509 
510 			spin_lock_bh(&vvs->rx_lock);
511 		}
512 
513 		total += skb->len;
514 		hdr = virtio_vsock_hdr(skb);
515 
516 		if (le32_to_cpu(hdr->flags) & VIRTIO_VSOCK_SEQ_EOM) {
517 			if (le32_to_cpu(hdr->flags) & VIRTIO_VSOCK_SEQ_EOR)
518 				msg->msg_flags |= MSG_EOR;
519 
520 			break;
521 		}
522 	}
523 
524 	spin_unlock_bh(&vvs->rx_lock);
525 
526 	return total;
527 }
528 
529 static int virtio_transport_seqpacket_do_dequeue(struct vsock_sock *vsk,
530 						 struct msghdr *msg,
531 						 int flags)
532 {
533 	struct virtio_vsock_sock *vvs = vsk->trans;
534 	int dequeued_len = 0;
535 	size_t user_buf_len = msg_data_left(msg);
536 	bool msg_ready = false;
537 	struct sk_buff *skb;
538 
539 	spin_lock_bh(&vvs->rx_lock);
540 
541 	if (vvs->msg_count == 0) {
542 		spin_unlock_bh(&vvs->rx_lock);
543 		return 0;
544 	}
545 
546 	while (!msg_ready) {
547 		struct virtio_vsock_hdr *hdr;
548 		size_t pkt_len;
549 
550 		skb = __skb_dequeue(&vvs->rx_queue);
551 		if (!skb)
552 			break;
553 		hdr = virtio_vsock_hdr(skb);
554 		pkt_len = (size_t)le32_to_cpu(hdr->len);
555 
556 		if (dequeued_len >= 0) {
557 			size_t bytes_to_copy;
558 
559 			bytes_to_copy = min(user_buf_len, pkt_len);
560 
561 			if (bytes_to_copy) {
562 				int err;
563 
564 				/* sk_lock is held by caller so no one else can dequeue.
565 				 * Unlock rx_lock since memcpy_to_msg() may sleep.
566 				 */
567 				spin_unlock_bh(&vvs->rx_lock);
568 
569 				err = memcpy_to_msg(msg, skb->data, bytes_to_copy);
570 				if (err) {
571 					/* Copy of message failed. Rest of
572 					 * fragments will be freed without copy.
573 					 */
574 					dequeued_len = err;
575 				} else {
576 					user_buf_len -= bytes_to_copy;
577 				}
578 
579 				spin_lock_bh(&vvs->rx_lock);
580 			}
581 
582 			if (dequeued_len >= 0)
583 				dequeued_len += pkt_len;
584 		}
585 
586 		if (le32_to_cpu(hdr->flags) & VIRTIO_VSOCK_SEQ_EOM) {
587 			msg_ready = true;
588 			vvs->msg_count--;
589 
590 			if (le32_to_cpu(hdr->flags) & VIRTIO_VSOCK_SEQ_EOR)
591 				msg->msg_flags |= MSG_EOR;
592 		}
593 
594 		virtio_transport_dec_rx_pkt(vvs, pkt_len);
595 		kfree_skb(skb);
596 	}
597 
598 	spin_unlock_bh(&vvs->rx_lock);
599 
600 	virtio_transport_send_credit_update(vsk);
601 
602 	return dequeued_len;
603 }
604 
605 ssize_t
606 virtio_transport_stream_dequeue(struct vsock_sock *vsk,
607 				struct msghdr *msg,
608 				size_t len, int flags)
609 {
610 	if (flags & MSG_PEEK)
611 		return virtio_transport_stream_do_peek(vsk, msg, len);
612 	else
613 		return virtio_transport_stream_do_dequeue(vsk, msg, len);
614 }
615 EXPORT_SYMBOL_GPL(virtio_transport_stream_dequeue);
616 
617 ssize_t
618 virtio_transport_seqpacket_dequeue(struct vsock_sock *vsk,
619 				   struct msghdr *msg,
620 				   int flags)
621 {
622 	if (flags & MSG_PEEK)
623 		return virtio_transport_seqpacket_do_peek(vsk, msg);
624 	else
625 		return virtio_transport_seqpacket_do_dequeue(vsk, msg, flags);
626 }
627 EXPORT_SYMBOL_GPL(virtio_transport_seqpacket_dequeue);
628 
629 int
630 virtio_transport_seqpacket_enqueue(struct vsock_sock *vsk,
631 				   struct msghdr *msg,
632 				   size_t len)
633 {
634 	struct virtio_vsock_sock *vvs = vsk->trans;
635 
636 	spin_lock_bh(&vvs->tx_lock);
637 
638 	if (len > vvs->peer_buf_alloc) {
639 		spin_unlock_bh(&vvs->tx_lock);
640 		return -EMSGSIZE;
641 	}
642 
643 	spin_unlock_bh(&vvs->tx_lock);
644 
645 	return virtio_transport_stream_enqueue(vsk, msg, len);
646 }
647 EXPORT_SYMBOL_GPL(virtio_transport_seqpacket_enqueue);
648 
649 int
650 virtio_transport_dgram_dequeue(struct vsock_sock *vsk,
651 			       struct msghdr *msg,
652 			       size_t len, int flags)
653 {
654 	return -EOPNOTSUPP;
655 }
656 EXPORT_SYMBOL_GPL(virtio_transport_dgram_dequeue);
657 
658 s64 virtio_transport_stream_has_data(struct vsock_sock *vsk)
659 {
660 	struct virtio_vsock_sock *vvs = vsk->trans;
661 	s64 bytes;
662 
663 	spin_lock_bh(&vvs->rx_lock);
664 	bytes = vvs->rx_bytes;
665 	spin_unlock_bh(&vvs->rx_lock);
666 
667 	return bytes;
668 }
669 EXPORT_SYMBOL_GPL(virtio_transport_stream_has_data);
670 
671 u32 virtio_transport_seqpacket_has_data(struct vsock_sock *vsk)
672 {
673 	struct virtio_vsock_sock *vvs = vsk->trans;
674 	u32 msg_count;
675 
676 	spin_lock_bh(&vvs->rx_lock);
677 	msg_count = vvs->msg_count;
678 	spin_unlock_bh(&vvs->rx_lock);
679 
680 	return msg_count;
681 }
682 EXPORT_SYMBOL_GPL(virtio_transport_seqpacket_has_data);
683 
684 static s64 virtio_transport_has_space(struct vsock_sock *vsk)
685 {
686 	struct virtio_vsock_sock *vvs = vsk->trans;
687 	s64 bytes;
688 
689 	bytes = (s64)vvs->peer_buf_alloc - (vvs->tx_cnt - vvs->peer_fwd_cnt);
690 	if (bytes < 0)
691 		bytes = 0;
692 
693 	return bytes;
694 }
695 
696 s64 virtio_transport_stream_has_space(struct vsock_sock *vsk)
697 {
698 	struct virtio_vsock_sock *vvs = vsk->trans;
699 	s64 bytes;
700 
701 	spin_lock_bh(&vvs->tx_lock);
702 	bytes = virtio_transport_has_space(vsk);
703 	spin_unlock_bh(&vvs->tx_lock);
704 
705 	return bytes;
706 }
707 EXPORT_SYMBOL_GPL(virtio_transport_stream_has_space);
708 
709 int virtio_transport_do_socket_init(struct vsock_sock *vsk,
710 				    struct vsock_sock *psk)
711 {
712 	struct virtio_vsock_sock *vvs;
713 
714 	vvs = kzalloc(sizeof(*vvs), GFP_KERNEL);
715 	if (!vvs)
716 		return -ENOMEM;
717 
718 	vsk->trans = vvs;
719 	vvs->vsk = vsk;
720 	if (psk && psk->trans) {
721 		struct virtio_vsock_sock *ptrans = psk->trans;
722 
723 		vvs->peer_buf_alloc = ptrans->peer_buf_alloc;
724 	}
725 
726 	if (vsk->buffer_size > VIRTIO_VSOCK_MAX_BUF_SIZE)
727 		vsk->buffer_size = VIRTIO_VSOCK_MAX_BUF_SIZE;
728 
729 	vvs->buf_alloc = vsk->buffer_size;
730 
731 	spin_lock_init(&vvs->rx_lock);
732 	spin_lock_init(&vvs->tx_lock);
733 	skb_queue_head_init(&vvs->rx_queue);
734 
735 	return 0;
736 }
737 EXPORT_SYMBOL_GPL(virtio_transport_do_socket_init);
738 
739 /* sk_lock held by the caller */
740 void virtio_transport_notify_buffer_size(struct vsock_sock *vsk, u64 *val)
741 {
742 	struct virtio_vsock_sock *vvs = vsk->trans;
743 
744 	if (*val > VIRTIO_VSOCK_MAX_BUF_SIZE)
745 		*val = VIRTIO_VSOCK_MAX_BUF_SIZE;
746 
747 	vvs->buf_alloc = *val;
748 
749 	virtio_transport_send_credit_update(vsk);
750 }
751 EXPORT_SYMBOL_GPL(virtio_transport_notify_buffer_size);
752 
753 int
754 virtio_transport_notify_poll_in(struct vsock_sock *vsk,
755 				size_t target,
756 				bool *data_ready_now)
757 {
758 	*data_ready_now = vsock_stream_has_data(vsk) >= target;
759 
760 	return 0;
761 }
762 EXPORT_SYMBOL_GPL(virtio_transport_notify_poll_in);
763 
764 int
765 virtio_transport_notify_poll_out(struct vsock_sock *vsk,
766 				 size_t target,
767 				 bool *space_avail_now)
768 {
769 	s64 free_space;
770 
771 	free_space = vsock_stream_has_space(vsk);
772 	if (free_space > 0)
773 		*space_avail_now = true;
774 	else if (free_space == 0)
775 		*space_avail_now = false;
776 
777 	return 0;
778 }
779 EXPORT_SYMBOL_GPL(virtio_transport_notify_poll_out);
780 
781 int virtio_transport_notify_recv_init(struct vsock_sock *vsk,
782 	size_t target, struct vsock_transport_recv_notify_data *data)
783 {
784 	return 0;
785 }
786 EXPORT_SYMBOL_GPL(virtio_transport_notify_recv_init);
787 
788 int virtio_transport_notify_recv_pre_block(struct vsock_sock *vsk,
789 	size_t target, struct vsock_transport_recv_notify_data *data)
790 {
791 	return 0;
792 }
793 EXPORT_SYMBOL_GPL(virtio_transport_notify_recv_pre_block);
794 
795 int virtio_transport_notify_recv_pre_dequeue(struct vsock_sock *vsk,
796 	size_t target, struct vsock_transport_recv_notify_data *data)
797 {
798 	return 0;
799 }
800 EXPORT_SYMBOL_GPL(virtio_transport_notify_recv_pre_dequeue);
801 
802 int virtio_transport_notify_recv_post_dequeue(struct vsock_sock *vsk,
803 	size_t target, ssize_t copied, bool data_read,
804 	struct vsock_transport_recv_notify_data *data)
805 {
806 	return 0;
807 }
808 EXPORT_SYMBOL_GPL(virtio_transport_notify_recv_post_dequeue);
809 
810 int virtio_transport_notify_send_init(struct vsock_sock *vsk,
811 	struct vsock_transport_send_notify_data *data)
812 {
813 	return 0;
814 }
815 EXPORT_SYMBOL_GPL(virtio_transport_notify_send_init);
816 
817 int virtio_transport_notify_send_pre_block(struct vsock_sock *vsk,
818 	struct vsock_transport_send_notify_data *data)
819 {
820 	return 0;
821 }
822 EXPORT_SYMBOL_GPL(virtio_transport_notify_send_pre_block);
823 
824 int virtio_transport_notify_send_pre_enqueue(struct vsock_sock *vsk,
825 	struct vsock_transport_send_notify_data *data)
826 {
827 	return 0;
828 }
829 EXPORT_SYMBOL_GPL(virtio_transport_notify_send_pre_enqueue);
830 
831 int virtio_transport_notify_send_post_enqueue(struct vsock_sock *vsk,
832 	ssize_t written, struct vsock_transport_send_notify_data *data)
833 {
834 	return 0;
835 }
836 EXPORT_SYMBOL_GPL(virtio_transport_notify_send_post_enqueue);
837 
838 u64 virtio_transport_stream_rcvhiwat(struct vsock_sock *vsk)
839 {
840 	return vsk->buffer_size;
841 }
842 EXPORT_SYMBOL_GPL(virtio_transport_stream_rcvhiwat);
843 
844 bool virtio_transport_stream_is_active(struct vsock_sock *vsk)
845 {
846 	return true;
847 }
848 EXPORT_SYMBOL_GPL(virtio_transport_stream_is_active);
849 
850 bool virtio_transport_stream_allow(u32 cid, u32 port)
851 {
852 	return true;
853 }
854 EXPORT_SYMBOL_GPL(virtio_transport_stream_allow);
855 
856 int virtio_transport_dgram_bind(struct vsock_sock *vsk,
857 				struct sockaddr_vm *addr)
858 {
859 	return -EOPNOTSUPP;
860 }
861 EXPORT_SYMBOL_GPL(virtio_transport_dgram_bind);
862 
863 bool virtio_transport_dgram_allow(u32 cid, u32 port)
864 {
865 	return false;
866 }
867 EXPORT_SYMBOL_GPL(virtio_transport_dgram_allow);
868 
869 int virtio_transport_connect(struct vsock_sock *vsk)
870 {
871 	struct virtio_vsock_pkt_info info = {
872 		.op = VIRTIO_VSOCK_OP_REQUEST,
873 		.vsk = vsk,
874 	};
875 
876 	return virtio_transport_send_pkt_info(vsk, &info);
877 }
878 EXPORT_SYMBOL_GPL(virtio_transport_connect);
879 
880 int virtio_transport_shutdown(struct vsock_sock *vsk, int mode)
881 {
882 	struct virtio_vsock_pkt_info info = {
883 		.op = VIRTIO_VSOCK_OP_SHUTDOWN,
884 		.flags = (mode & RCV_SHUTDOWN ?
885 			  VIRTIO_VSOCK_SHUTDOWN_RCV : 0) |
886 			 (mode & SEND_SHUTDOWN ?
887 			  VIRTIO_VSOCK_SHUTDOWN_SEND : 0),
888 		.vsk = vsk,
889 	};
890 
891 	return virtio_transport_send_pkt_info(vsk, &info);
892 }
893 EXPORT_SYMBOL_GPL(virtio_transport_shutdown);
894 
895 int
896 virtio_transport_dgram_enqueue(struct vsock_sock *vsk,
897 			       struct sockaddr_vm *remote_addr,
898 			       struct msghdr *msg,
899 			       size_t dgram_len)
900 {
901 	return -EOPNOTSUPP;
902 }
903 EXPORT_SYMBOL_GPL(virtio_transport_dgram_enqueue);
904 
905 ssize_t
906 virtio_transport_stream_enqueue(struct vsock_sock *vsk,
907 				struct msghdr *msg,
908 				size_t len)
909 {
910 	struct virtio_vsock_pkt_info info = {
911 		.op = VIRTIO_VSOCK_OP_RW,
912 		.msg = msg,
913 		.pkt_len = len,
914 		.vsk = vsk,
915 	};
916 
917 	return virtio_transport_send_pkt_info(vsk, &info);
918 }
919 EXPORT_SYMBOL_GPL(virtio_transport_stream_enqueue);
920 
921 void virtio_transport_destruct(struct vsock_sock *vsk)
922 {
923 	struct virtio_vsock_sock *vvs = vsk->trans;
924 
925 	kfree(vvs);
926 	vsk->trans = NULL;
927 }
928 EXPORT_SYMBOL_GPL(virtio_transport_destruct);
929 
930 static int virtio_transport_reset(struct vsock_sock *vsk,
931 				  struct sk_buff *skb)
932 {
933 	struct virtio_vsock_pkt_info info = {
934 		.op = VIRTIO_VSOCK_OP_RST,
935 		.reply = !!skb,
936 		.vsk = vsk,
937 	};
938 
939 	/* Send RST only if the original pkt is not a RST pkt */
940 	if (skb && le16_to_cpu(virtio_vsock_hdr(skb)->op) == VIRTIO_VSOCK_OP_RST)
941 		return 0;
942 
943 	return virtio_transport_send_pkt_info(vsk, &info);
944 }
945 
946 /* Normally packets are associated with a socket.  There may be no socket if an
947  * attempt was made to connect to a socket that does not exist.
948  */
949 static int virtio_transport_reset_no_sock(const struct virtio_transport *t,
950 					  struct sk_buff *skb)
951 {
952 	struct virtio_vsock_hdr *hdr = virtio_vsock_hdr(skb);
953 	struct virtio_vsock_pkt_info info = {
954 		.op = VIRTIO_VSOCK_OP_RST,
955 		.type = le16_to_cpu(hdr->type),
956 		.reply = true,
957 	};
958 	struct sk_buff *reply;
959 
960 	/* Send RST only if the original pkt is not a RST pkt */
961 	if (le16_to_cpu(hdr->op) == VIRTIO_VSOCK_OP_RST)
962 		return 0;
963 
964 	if (!t)
965 		return -ENOTCONN;
966 
967 	reply = virtio_transport_alloc_skb(&info, 0,
968 					   le64_to_cpu(hdr->dst_cid),
969 					   le32_to_cpu(hdr->dst_port),
970 					   le64_to_cpu(hdr->src_cid),
971 					   le32_to_cpu(hdr->src_port));
972 	if (!reply)
973 		return -ENOMEM;
974 
975 	return t->send_pkt(reply);
976 }
977 
978 /* This function should be called with sk_lock held and SOCK_DONE set */
979 static void virtio_transport_remove_sock(struct vsock_sock *vsk)
980 {
981 	struct virtio_vsock_sock *vvs = vsk->trans;
982 
983 	/* We don't need to take rx_lock, as the socket is closing and we are
984 	 * removing it.
985 	 */
986 	__skb_queue_purge(&vvs->rx_queue);
987 	vsock_remove_sock(vsk);
988 }
989 
990 static void virtio_transport_wait_close(struct sock *sk, long timeout)
991 {
992 	if (timeout) {
993 		DEFINE_WAIT_FUNC(wait, woken_wake_function);
994 
995 		add_wait_queue(sk_sleep(sk), &wait);
996 
997 		do {
998 			if (sk_wait_event(sk, &timeout,
999 					  sock_flag(sk, SOCK_DONE), &wait))
1000 				break;
1001 		} while (!signal_pending(current) && timeout);
1002 
1003 		remove_wait_queue(sk_sleep(sk), &wait);
1004 	}
1005 }
1006 
1007 static void virtio_transport_do_close(struct vsock_sock *vsk,
1008 				      bool cancel_timeout)
1009 {
1010 	struct sock *sk = sk_vsock(vsk);
1011 
1012 	sock_set_flag(sk, SOCK_DONE);
1013 	vsk->peer_shutdown = SHUTDOWN_MASK;
1014 	if (vsock_stream_has_data(vsk) <= 0)
1015 		sk->sk_state = TCP_CLOSING;
1016 	sk->sk_state_change(sk);
1017 
1018 	if (vsk->close_work_scheduled &&
1019 	    (!cancel_timeout || cancel_delayed_work(&vsk->close_work))) {
1020 		vsk->close_work_scheduled = false;
1021 
1022 		virtio_transport_remove_sock(vsk);
1023 
1024 		/* Release refcnt obtained when we scheduled the timeout */
1025 		sock_put(sk);
1026 	}
1027 }
1028 
1029 static void virtio_transport_close_timeout(struct work_struct *work)
1030 {
1031 	struct vsock_sock *vsk =
1032 		container_of(work, struct vsock_sock, close_work.work);
1033 	struct sock *sk = sk_vsock(vsk);
1034 
1035 	sock_hold(sk);
1036 	lock_sock(sk);
1037 
1038 	if (!sock_flag(sk, SOCK_DONE)) {
1039 		(void)virtio_transport_reset(vsk, NULL);
1040 
1041 		virtio_transport_do_close(vsk, false);
1042 	}
1043 
1044 	vsk->close_work_scheduled = false;
1045 
1046 	release_sock(sk);
1047 	sock_put(sk);
1048 }
1049 
1050 /* User context, vsk->sk is locked */
1051 static bool virtio_transport_close(struct vsock_sock *vsk)
1052 {
1053 	struct sock *sk = &vsk->sk;
1054 
1055 	if (!(sk->sk_state == TCP_ESTABLISHED ||
1056 	      sk->sk_state == TCP_CLOSING))
1057 		return true;
1058 
1059 	/* Already received SHUTDOWN from peer, reply with RST */
1060 	if ((vsk->peer_shutdown & SHUTDOWN_MASK) == SHUTDOWN_MASK) {
1061 		(void)virtio_transport_reset(vsk, NULL);
1062 		return true;
1063 	}
1064 
1065 	if ((sk->sk_shutdown & SHUTDOWN_MASK) != SHUTDOWN_MASK)
1066 		(void)virtio_transport_shutdown(vsk, SHUTDOWN_MASK);
1067 
1068 	if (sock_flag(sk, SOCK_LINGER) && !(current->flags & PF_EXITING))
1069 		virtio_transport_wait_close(sk, sk->sk_lingertime);
1070 
1071 	if (sock_flag(sk, SOCK_DONE)) {
1072 		return true;
1073 	}
1074 
1075 	sock_hold(sk);
1076 	INIT_DELAYED_WORK(&vsk->close_work,
1077 			  virtio_transport_close_timeout);
1078 	vsk->close_work_scheduled = true;
1079 	schedule_delayed_work(&vsk->close_work, VSOCK_CLOSE_TIMEOUT);
1080 	return false;
1081 }
1082 
1083 void virtio_transport_release(struct vsock_sock *vsk)
1084 {
1085 	struct sock *sk = &vsk->sk;
1086 	bool remove_sock = true;
1087 
1088 	if (sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET)
1089 		remove_sock = virtio_transport_close(vsk);
1090 
1091 	if (remove_sock) {
1092 		sock_set_flag(sk, SOCK_DONE);
1093 		virtio_transport_remove_sock(vsk);
1094 	}
1095 }
1096 EXPORT_SYMBOL_GPL(virtio_transport_release);
1097 
1098 static int
1099 virtio_transport_recv_connecting(struct sock *sk,
1100 				 struct sk_buff *skb)
1101 {
1102 	struct virtio_vsock_hdr *hdr = virtio_vsock_hdr(skb);
1103 	struct vsock_sock *vsk = vsock_sk(sk);
1104 	int skerr;
1105 	int err;
1106 
1107 	switch (le16_to_cpu(hdr->op)) {
1108 	case VIRTIO_VSOCK_OP_RESPONSE:
1109 		sk->sk_state = TCP_ESTABLISHED;
1110 		sk->sk_socket->state = SS_CONNECTED;
1111 		vsock_insert_connected(vsk);
1112 		sk->sk_state_change(sk);
1113 		break;
1114 	case VIRTIO_VSOCK_OP_INVALID:
1115 		break;
1116 	case VIRTIO_VSOCK_OP_RST:
1117 		skerr = ECONNRESET;
1118 		err = 0;
1119 		goto destroy;
1120 	default:
1121 		skerr = EPROTO;
1122 		err = -EINVAL;
1123 		goto destroy;
1124 	}
1125 	return 0;
1126 
1127 destroy:
1128 	virtio_transport_reset(vsk, skb);
1129 	sk->sk_state = TCP_CLOSE;
1130 	sk->sk_err = skerr;
1131 	sk_error_report(sk);
1132 	return err;
1133 }
1134 
1135 static void
1136 virtio_transport_recv_enqueue(struct vsock_sock *vsk,
1137 			      struct sk_buff *skb)
1138 {
1139 	struct virtio_vsock_sock *vvs = vsk->trans;
1140 	bool can_enqueue, free_pkt = false;
1141 	struct virtio_vsock_hdr *hdr;
1142 	u32 len;
1143 
1144 	hdr = virtio_vsock_hdr(skb);
1145 	len = le32_to_cpu(hdr->len);
1146 
1147 	spin_lock_bh(&vvs->rx_lock);
1148 
1149 	can_enqueue = virtio_transport_inc_rx_pkt(vvs, len);
1150 	if (!can_enqueue) {
1151 		free_pkt = true;
1152 		goto out;
1153 	}
1154 
1155 	if (le32_to_cpu(hdr->flags) & VIRTIO_VSOCK_SEQ_EOM)
1156 		vvs->msg_count++;
1157 
1158 	/* Try to copy small packets into the buffer of last packet queued,
1159 	 * to avoid wasting memory queueing the entire buffer with a small
1160 	 * payload.
1161 	 */
1162 	if (len <= GOOD_COPY_LEN && !skb_queue_empty(&vvs->rx_queue)) {
1163 		struct virtio_vsock_hdr *last_hdr;
1164 		struct sk_buff *last_skb;
1165 
1166 		last_skb = skb_peek_tail(&vvs->rx_queue);
1167 		last_hdr = virtio_vsock_hdr(last_skb);
1168 
1169 		/* If there is space in the last packet queued, we copy the
1170 		 * new packet in its buffer. We avoid this if the last packet
1171 		 * queued has VIRTIO_VSOCK_SEQ_EOM set, because this is
1172 		 * delimiter of SEQPACKET message, so 'pkt' is the first packet
1173 		 * of a new message.
1174 		 */
1175 		if (skb->len < skb_tailroom(last_skb) &&
1176 		    !(le32_to_cpu(last_hdr->flags) & VIRTIO_VSOCK_SEQ_EOM)) {
1177 			memcpy(skb_put(last_skb, skb->len), skb->data, skb->len);
1178 			free_pkt = true;
1179 			last_hdr->flags |= hdr->flags;
1180 			le32_add_cpu(&last_hdr->len, len);
1181 			goto out;
1182 		}
1183 	}
1184 
1185 	__skb_queue_tail(&vvs->rx_queue, skb);
1186 
1187 out:
1188 	spin_unlock_bh(&vvs->rx_lock);
1189 	if (free_pkt)
1190 		kfree_skb(skb);
1191 }
1192 
1193 static int
1194 virtio_transport_recv_connected(struct sock *sk,
1195 				struct sk_buff *skb)
1196 {
1197 	struct virtio_vsock_hdr *hdr = virtio_vsock_hdr(skb);
1198 	struct vsock_sock *vsk = vsock_sk(sk);
1199 	int err = 0;
1200 
1201 	switch (le16_to_cpu(hdr->op)) {
1202 	case VIRTIO_VSOCK_OP_RW:
1203 		virtio_transport_recv_enqueue(vsk, skb);
1204 		vsock_data_ready(sk);
1205 		return err;
1206 	case VIRTIO_VSOCK_OP_CREDIT_REQUEST:
1207 		virtio_transport_send_credit_update(vsk);
1208 		break;
1209 	case VIRTIO_VSOCK_OP_CREDIT_UPDATE:
1210 		sk->sk_write_space(sk);
1211 		break;
1212 	case VIRTIO_VSOCK_OP_SHUTDOWN:
1213 		if (le32_to_cpu(hdr->flags) & VIRTIO_VSOCK_SHUTDOWN_RCV)
1214 			vsk->peer_shutdown |= RCV_SHUTDOWN;
1215 		if (le32_to_cpu(hdr->flags) & VIRTIO_VSOCK_SHUTDOWN_SEND)
1216 			vsk->peer_shutdown |= SEND_SHUTDOWN;
1217 		if (vsk->peer_shutdown == SHUTDOWN_MASK) {
1218 			if (vsock_stream_has_data(vsk) <= 0 && !sock_flag(sk, SOCK_DONE)) {
1219 				(void)virtio_transport_reset(vsk, NULL);
1220 				virtio_transport_do_close(vsk, true);
1221 			}
1222 			/* Remove this socket anyway because the remote peer sent
1223 			 * the shutdown. This way a new connection will succeed
1224 			 * if the remote peer uses the same source port,
1225 			 * even if the old socket is still unreleased, but now disconnected.
1226 			 */
1227 			vsock_remove_sock(vsk);
1228 		}
1229 		if (le32_to_cpu(virtio_vsock_hdr(skb)->flags))
1230 			sk->sk_state_change(sk);
1231 		break;
1232 	case VIRTIO_VSOCK_OP_RST:
1233 		virtio_transport_do_close(vsk, true);
1234 		break;
1235 	default:
1236 		err = -EINVAL;
1237 		break;
1238 	}
1239 
1240 	kfree_skb(skb);
1241 	return err;
1242 }
1243 
1244 static void
1245 virtio_transport_recv_disconnecting(struct sock *sk,
1246 				    struct sk_buff *skb)
1247 {
1248 	struct virtio_vsock_hdr *hdr = virtio_vsock_hdr(skb);
1249 	struct vsock_sock *vsk = vsock_sk(sk);
1250 
1251 	if (le16_to_cpu(hdr->op) == VIRTIO_VSOCK_OP_RST)
1252 		virtio_transport_do_close(vsk, true);
1253 }
1254 
1255 static int
1256 virtio_transport_send_response(struct vsock_sock *vsk,
1257 			       struct sk_buff *skb)
1258 {
1259 	struct virtio_vsock_hdr *hdr = virtio_vsock_hdr(skb);
1260 	struct virtio_vsock_pkt_info info = {
1261 		.op = VIRTIO_VSOCK_OP_RESPONSE,
1262 		.remote_cid = le64_to_cpu(hdr->src_cid),
1263 		.remote_port = le32_to_cpu(hdr->src_port),
1264 		.reply = true,
1265 		.vsk = vsk,
1266 	};
1267 
1268 	return virtio_transport_send_pkt_info(vsk, &info);
1269 }
1270 
1271 static bool virtio_transport_space_update(struct sock *sk,
1272 					  struct sk_buff *skb)
1273 {
1274 	struct virtio_vsock_hdr *hdr = virtio_vsock_hdr(skb);
1275 	struct vsock_sock *vsk = vsock_sk(sk);
1276 	struct virtio_vsock_sock *vvs = vsk->trans;
1277 	bool space_available;
1278 
1279 	/* Listener sockets are not associated with any transport, so we are
1280 	 * not able to take the state to see if there is space available in the
1281 	 * remote peer, but since they are only used to receive requests, we
1282 	 * can assume that there is always space available in the other peer.
1283 	 */
1284 	if (!vvs)
1285 		return true;
1286 
1287 	/* buf_alloc and fwd_cnt is always included in the hdr */
1288 	spin_lock_bh(&vvs->tx_lock);
1289 	vvs->peer_buf_alloc = le32_to_cpu(hdr->buf_alloc);
1290 	vvs->peer_fwd_cnt = le32_to_cpu(hdr->fwd_cnt);
1291 	space_available = virtio_transport_has_space(vsk);
1292 	spin_unlock_bh(&vvs->tx_lock);
1293 	return space_available;
1294 }
1295 
1296 /* Handle server socket */
1297 static int
1298 virtio_transport_recv_listen(struct sock *sk, struct sk_buff *skb,
1299 			     struct virtio_transport *t)
1300 {
1301 	struct virtio_vsock_hdr *hdr = virtio_vsock_hdr(skb);
1302 	struct vsock_sock *vsk = vsock_sk(sk);
1303 	struct vsock_sock *vchild;
1304 	struct sock *child;
1305 	int ret;
1306 
1307 	if (le16_to_cpu(hdr->op) != VIRTIO_VSOCK_OP_REQUEST) {
1308 		virtio_transport_reset_no_sock(t, skb);
1309 		return -EINVAL;
1310 	}
1311 
1312 	if (sk_acceptq_is_full(sk)) {
1313 		virtio_transport_reset_no_sock(t, skb);
1314 		return -ENOMEM;
1315 	}
1316 
1317 	/* __vsock_release() might have already flushed accept_queue.
1318 	 * Subsequent enqueues would lead to a memory leak.
1319 	 */
1320 	if (sk->sk_shutdown == SHUTDOWN_MASK) {
1321 		virtio_transport_reset_no_sock(t, skb);
1322 		return -ESHUTDOWN;
1323 	}
1324 
1325 	child = vsock_create_connected(sk);
1326 	if (!child) {
1327 		virtio_transport_reset_no_sock(t, skb);
1328 		return -ENOMEM;
1329 	}
1330 
1331 	sk_acceptq_added(sk);
1332 
1333 	lock_sock_nested(child, SINGLE_DEPTH_NESTING);
1334 
1335 	child->sk_state = TCP_ESTABLISHED;
1336 
1337 	vchild = vsock_sk(child);
1338 	vsock_addr_init(&vchild->local_addr, le64_to_cpu(hdr->dst_cid),
1339 			le32_to_cpu(hdr->dst_port));
1340 	vsock_addr_init(&vchild->remote_addr, le64_to_cpu(hdr->src_cid),
1341 			le32_to_cpu(hdr->src_port));
1342 
1343 	ret = vsock_assign_transport(vchild, vsk);
1344 	/* Transport assigned (looking at remote_addr) must be the same
1345 	 * where we received the request.
1346 	 */
1347 	if (ret || vchild->transport != &t->transport) {
1348 		release_sock(child);
1349 		virtio_transport_reset_no_sock(t, skb);
1350 		sock_put(child);
1351 		return ret;
1352 	}
1353 
1354 	if (virtio_transport_space_update(child, skb))
1355 		child->sk_write_space(child);
1356 
1357 	vsock_insert_connected(vchild);
1358 	vsock_enqueue_accept(sk, child);
1359 	virtio_transport_send_response(vchild, skb);
1360 
1361 	release_sock(child);
1362 
1363 	sk->sk_data_ready(sk);
1364 	return 0;
1365 }
1366 
1367 static bool virtio_transport_valid_type(u16 type)
1368 {
1369 	return (type == VIRTIO_VSOCK_TYPE_STREAM) ||
1370 	       (type == VIRTIO_VSOCK_TYPE_SEQPACKET);
1371 }
1372 
1373 /* We are under the virtio-vsock's vsock->rx_lock or vhost-vsock's vq->mutex
1374  * lock.
1375  */
1376 void virtio_transport_recv_pkt(struct virtio_transport *t,
1377 			       struct sk_buff *skb)
1378 {
1379 	struct virtio_vsock_hdr *hdr = virtio_vsock_hdr(skb);
1380 	struct sockaddr_vm src, dst;
1381 	struct vsock_sock *vsk;
1382 	struct sock *sk;
1383 	bool space_available;
1384 
1385 	vsock_addr_init(&src, le64_to_cpu(hdr->src_cid),
1386 			le32_to_cpu(hdr->src_port));
1387 	vsock_addr_init(&dst, le64_to_cpu(hdr->dst_cid),
1388 			le32_to_cpu(hdr->dst_port));
1389 
1390 	trace_virtio_transport_recv_pkt(src.svm_cid, src.svm_port,
1391 					dst.svm_cid, dst.svm_port,
1392 					le32_to_cpu(hdr->len),
1393 					le16_to_cpu(hdr->type),
1394 					le16_to_cpu(hdr->op),
1395 					le32_to_cpu(hdr->flags),
1396 					le32_to_cpu(hdr->buf_alloc),
1397 					le32_to_cpu(hdr->fwd_cnt));
1398 
1399 	if (!virtio_transport_valid_type(le16_to_cpu(hdr->type))) {
1400 		(void)virtio_transport_reset_no_sock(t, skb);
1401 		goto free_pkt;
1402 	}
1403 
1404 	/* The socket must be in connected or bound table
1405 	 * otherwise send reset back
1406 	 */
1407 	sk = vsock_find_connected_socket(&src, &dst);
1408 	if (!sk) {
1409 		sk = vsock_find_bound_socket(&dst);
1410 		if (!sk) {
1411 			(void)virtio_transport_reset_no_sock(t, skb);
1412 			goto free_pkt;
1413 		}
1414 	}
1415 
1416 	if (virtio_transport_get_type(sk) != le16_to_cpu(hdr->type)) {
1417 		(void)virtio_transport_reset_no_sock(t, skb);
1418 		sock_put(sk);
1419 		goto free_pkt;
1420 	}
1421 
1422 	if (!skb_set_owner_sk_safe(skb, sk)) {
1423 		WARN_ONCE(1, "receiving vsock socket has sk_refcnt == 0\n");
1424 		goto free_pkt;
1425 	}
1426 
1427 	vsk = vsock_sk(sk);
1428 
1429 	lock_sock(sk);
1430 
1431 	/* Check if sk has been closed before lock_sock */
1432 	if (sock_flag(sk, SOCK_DONE)) {
1433 		(void)virtio_transport_reset_no_sock(t, skb);
1434 		release_sock(sk);
1435 		sock_put(sk);
1436 		goto free_pkt;
1437 	}
1438 
1439 	space_available = virtio_transport_space_update(sk, skb);
1440 
1441 	/* Update CID in case it has changed after a transport reset event */
1442 	if (vsk->local_addr.svm_cid != VMADDR_CID_ANY)
1443 		vsk->local_addr.svm_cid = dst.svm_cid;
1444 
1445 	if (space_available)
1446 		sk->sk_write_space(sk);
1447 
1448 	switch (sk->sk_state) {
1449 	case TCP_LISTEN:
1450 		virtio_transport_recv_listen(sk, skb, t);
1451 		kfree_skb(skb);
1452 		break;
1453 	case TCP_SYN_SENT:
1454 		virtio_transport_recv_connecting(sk, skb);
1455 		kfree_skb(skb);
1456 		break;
1457 	case TCP_ESTABLISHED:
1458 		virtio_transport_recv_connected(sk, skb);
1459 		break;
1460 	case TCP_CLOSING:
1461 		virtio_transport_recv_disconnecting(sk, skb);
1462 		kfree_skb(skb);
1463 		break;
1464 	default:
1465 		(void)virtio_transport_reset_no_sock(t, skb);
1466 		kfree_skb(skb);
1467 		break;
1468 	}
1469 
1470 	release_sock(sk);
1471 
1472 	/* Release refcnt obtained when we fetched this socket out of the
1473 	 * bound or connected list.
1474 	 */
1475 	sock_put(sk);
1476 	return;
1477 
1478 free_pkt:
1479 	kfree_skb(skb);
1480 }
1481 EXPORT_SYMBOL_GPL(virtio_transport_recv_pkt);
1482 
1483 /* Remove skbs found in a queue that have a vsk that matches.
1484  *
1485  * Each skb is freed.
1486  *
1487  * Returns the count of skbs that were reply packets.
1488  */
1489 int virtio_transport_purge_skbs(void *vsk, struct sk_buff_head *queue)
1490 {
1491 	struct sk_buff_head freeme;
1492 	struct sk_buff *skb, *tmp;
1493 	int cnt = 0;
1494 
1495 	skb_queue_head_init(&freeme);
1496 
1497 	spin_lock_bh(&queue->lock);
1498 	skb_queue_walk_safe(queue, skb, tmp) {
1499 		if (vsock_sk(skb->sk) != vsk)
1500 			continue;
1501 
1502 		__skb_unlink(skb, queue);
1503 		__skb_queue_tail(&freeme, skb);
1504 
1505 		if (virtio_vsock_skb_reply(skb))
1506 			cnt++;
1507 	}
1508 	spin_unlock_bh(&queue->lock);
1509 
1510 	__skb_queue_purge(&freeme);
1511 
1512 	return cnt;
1513 }
1514 EXPORT_SYMBOL_GPL(virtio_transport_purge_skbs);
1515 
1516 int virtio_transport_read_skb(struct vsock_sock *vsk, skb_read_actor_t recv_actor)
1517 {
1518 	struct virtio_vsock_sock *vvs = vsk->trans;
1519 	struct sock *sk = sk_vsock(vsk);
1520 	struct virtio_vsock_hdr *hdr;
1521 	struct sk_buff *skb;
1522 	int off = 0;
1523 	int err;
1524 
1525 	spin_lock_bh(&vvs->rx_lock);
1526 	/* Use __skb_recv_datagram() for race-free handling of the receive. It
1527 	 * works for types other than dgrams.
1528 	 */
1529 	skb = __skb_recv_datagram(sk, &vvs->rx_queue, MSG_DONTWAIT, &off, &err);
1530 	if (!skb) {
1531 		spin_unlock_bh(&vvs->rx_lock);
1532 		return err;
1533 	}
1534 
1535 	hdr = virtio_vsock_hdr(skb);
1536 	if (le32_to_cpu(hdr->flags) & VIRTIO_VSOCK_SEQ_EOM)
1537 		vvs->msg_count--;
1538 
1539 	virtio_transport_dec_rx_pkt(vvs, le32_to_cpu(hdr->len));
1540 	spin_unlock_bh(&vvs->rx_lock);
1541 
1542 	virtio_transport_send_credit_update(vsk);
1543 
1544 	return recv_actor(sk, skb);
1545 }
1546 EXPORT_SYMBOL_GPL(virtio_transport_read_skb);
1547 
1548 int virtio_transport_notify_set_rcvlowat(struct vsock_sock *vsk, int val)
1549 {
1550 	struct virtio_vsock_sock *vvs = vsk->trans;
1551 	bool send_update;
1552 
1553 	spin_lock_bh(&vvs->rx_lock);
1554 
1555 	/* If number of available bytes is less than new SO_RCVLOWAT value,
1556 	 * kick sender to send more data, because sender may sleep in its
1557 	 * 'send()' syscall waiting for enough space at our side. Also
1558 	 * don't send credit update when peer already knows actual value -
1559 	 * such transmission will be useless.
1560 	 */
1561 	send_update = (vvs->rx_bytes < val) &&
1562 		      (vvs->fwd_cnt != vvs->last_fwd_cnt);
1563 
1564 	spin_unlock_bh(&vvs->rx_lock);
1565 
1566 	if (send_update) {
1567 		int err;
1568 
1569 		err = virtio_transport_send_credit_update(vsk);
1570 		if (err < 0)
1571 			return err;
1572 	}
1573 
1574 	return 0;
1575 }
1576 EXPORT_SYMBOL_GPL(virtio_transport_notify_set_rcvlowat);
1577 
1578 MODULE_LICENSE("GPL v2");
1579 MODULE_AUTHOR("Asias He");
1580 MODULE_DESCRIPTION("common code for virtio vsock");
1581