1 /*
2  * Hyper-V transport for vsock
3  *
4  * Hyper-V Sockets supplies a byte-stream based communication mechanism
5  * between the host and the VM. This driver implements the necessary
6  * support in the VM by introducing the new vsock transport.
7  *
8  * Copyright (c) 2017, Microsoft Corporation.
9  *
10  * This program is free software; you can redistribute it and/or modify it
11  * under the terms and conditions of the GNU General Public License,
12  * version 2, as published by the Free Software Foundation.
13  *
14  * This program is distributed in the hope it will be useful, but WITHOUT
15  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
16  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
17  * more details.
18  *
19  */
20 #include <linux/module.h>
21 #include <linux/vmalloc.h>
22 #include <linux/hyperv.h>
23 #include <net/sock.h>
24 #include <net/af_vsock.h>
25 
26 /* The host side's design of the feature requires 6 exact 4KB pages for
27  * recv/send rings respectively -- this is suboptimal considering memory
28  * consumption, however unluckily we have to live with it, before the
29  * host comes up with a better design in the future.
30  */
31 #define PAGE_SIZE_4K		4096
32 #define RINGBUFFER_HVS_RCV_SIZE (PAGE_SIZE_4K * 6)
33 #define RINGBUFFER_HVS_SND_SIZE (PAGE_SIZE_4K * 6)
34 
35 /* The MTU is 16KB per the host side's design */
36 #define HVS_MTU_SIZE		(1024 * 16)
37 
38 struct vmpipe_proto_header {
39 	u32 pkt_type;
40 	u32 data_size;
41 };
42 
43 /* For recv, we use the VMBus in-place packet iterator APIs to directly copy
44  * data from the ringbuffer into the userspace buffer.
45  */
46 struct hvs_recv_buf {
47 	/* The header before the payload data */
48 	struct vmpipe_proto_header hdr;
49 
50 	/* The payload */
51 	u8 data[HVS_MTU_SIZE];
52 };
53 
54 /* We can send up to HVS_MTU_SIZE bytes of payload to the host, but let's use
55  * a small size, i.e. HVS_SEND_BUF_SIZE, to minimize the dynamically-allocated
56  * buffer, because tests show there is no significant performance difference.
57  *
58  * Note: the buffer can be eliminated in the future when we add new VMBus
59  * ringbuffer APIs that allow us to directly copy data from userspace buffer
60  * to VMBus ringbuffer.
61  */
62 #define HVS_SEND_BUF_SIZE (PAGE_SIZE_4K - sizeof(struct vmpipe_proto_header))
63 
64 struct hvs_send_buf {
65 	/* The header before the payload data */
66 	struct vmpipe_proto_header hdr;
67 
68 	/* The payload */
69 	u8 data[HVS_SEND_BUF_SIZE];
70 };
71 
72 #define HVS_HEADER_LEN	(sizeof(struct vmpacket_descriptor) + \
73 			 sizeof(struct vmpipe_proto_header))
74 
75 /* See 'prev_indices' in hv_ringbuffer_read(), hv_ringbuffer_write(), and
76  * __hv_pkt_iter_next().
77  */
78 #define VMBUS_PKT_TRAILER_SIZE	(sizeof(u64))
79 
80 #define HVS_PKT_LEN(payload_len)	(HVS_HEADER_LEN + \
81 					 ALIGN((payload_len), 8) + \
82 					 VMBUS_PKT_TRAILER_SIZE)
83 
84 union hvs_service_id {
85 	uuid_le	srv_id;
86 
87 	struct {
88 		unsigned int svm_port;
89 		unsigned char b[sizeof(uuid_le) - sizeof(unsigned int)];
90 	};
91 };
92 
93 /* Per-socket state (accessed via vsk->trans) */
94 struct hvsock {
95 	struct vsock_sock *vsk;
96 
97 	uuid_le vm_srv_id;
98 	uuid_le host_srv_id;
99 
100 	struct vmbus_channel *chan;
101 	struct vmpacket_descriptor *recv_desc;
102 
103 	/* The length of the payload not delivered to userland yet */
104 	u32 recv_data_len;
105 	/* The offset of the payload */
106 	u32 recv_data_off;
107 
108 	/* Have we sent the zero-length packet (FIN)? */
109 	bool fin_sent;
110 };
111 
112 /* In the VM, we support Hyper-V Sockets with AF_VSOCK, and the endpoint is
113  * <cid, port> (see struct sockaddr_vm). Note: cid is not really used here:
114  * when we write apps to connect to the host, we can only use VMADDR_CID_ANY
115  * or VMADDR_CID_HOST (both are equivalent) as the remote cid, and when we
116  * write apps to bind() & listen() in the VM, we can only use VMADDR_CID_ANY
117  * as the local cid.
118  *
119  * On the host, Hyper-V Sockets are supported by Winsock AF_HYPERV:
120  * https://docs.microsoft.com/en-us/virtualization/hyper-v-on-windows/user-
121  * guide/make-integration-service, and the endpoint is <VmID, ServiceId> with
122  * the below sockaddr:
123  *
124  * struct SOCKADDR_HV
125  * {
126  *    ADDRESS_FAMILY Family;
127  *    USHORT Reserved;
128  *    GUID VmId;
129  *    GUID ServiceId;
130  * };
131  * Note: VmID is not used by Linux VM and actually it isn't transmitted via
132  * VMBus, because here it's obvious the host and the VM can easily identify
133  * each other. Though the VmID is useful on the host, especially in the case
134  * of Windows container, Linux VM doesn't need it at all.
135  *
136  * To make use of the AF_VSOCK infrastructure in Linux VM, we have to limit
137  * the available GUID space of SOCKADDR_HV so that we can create a mapping
138  * between AF_VSOCK port and SOCKADDR_HV Service GUID. The rule of writing
139  * Hyper-V Sockets apps on the host and in Linux VM is:
140  *
141  ****************************************************************************
142  * The only valid Service GUIDs, from the perspectives of both the host and *
143  * Linux VM, that can be connected by the other end, must conform to this   *
144  * format: <port>-facb-11e6-bd58-64006a7986d3, and the "port" must be in    *
145  * this range [0, 0x7FFFFFFF].                                              *
146  ****************************************************************************
147  *
148  * When we write apps on the host to connect(), the GUID ServiceID is used.
149  * When we write apps in Linux VM to connect(), we only need to specify the
150  * port and the driver will form the GUID and use that to request the host.
151  *
152  * From the perspective of Linux VM:
153  * 1. the local ephemeral port (i.e. the local auto-bound port when we call
154  * connect() without explicit bind()) is generated by __vsock_bind_stream(),
155  * and the range is [1024, 0xFFFFFFFF).
156  * 2. the remote ephemeral port (i.e. the auto-generated remote port for
157  * a connect request initiated by the host's connect()) is generated by
158  * hvs_remote_addr_init() and the range is [0x80000000, 0xFFFFFFFF).
159  */
160 
161 #define MAX_LISTEN_PORT			((u32)0x7FFFFFFF)
162 #define MAX_VM_LISTEN_PORT		MAX_LISTEN_PORT
163 #define MAX_HOST_LISTEN_PORT		MAX_LISTEN_PORT
164 #define MIN_HOST_EPHEMERAL_PORT		(MAX_HOST_LISTEN_PORT + 1)
165 
166 /* 00000000-facb-11e6-bd58-64006a7986d3 */
167 static const uuid_le srv_id_template =
168 	UUID_LE(0x00000000, 0xfacb, 0x11e6, 0xbd, 0x58,
169 		0x64, 0x00, 0x6a, 0x79, 0x86, 0xd3);
170 
171 static bool is_valid_srv_id(const uuid_le *id)
172 {
173 	return !memcmp(&id->b[4], &srv_id_template.b[4], sizeof(uuid_le) - 4);
174 }
175 
176 static unsigned int get_port_by_srv_id(const uuid_le *svr_id)
177 {
178 	return *((unsigned int *)svr_id);
179 }
180 
181 static void hvs_addr_init(struct sockaddr_vm *addr, const uuid_le *svr_id)
182 {
183 	unsigned int port = get_port_by_srv_id(svr_id);
184 
185 	vsock_addr_init(addr, VMADDR_CID_ANY, port);
186 }
187 
188 static void hvs_remote_addr_init(struct sockaddr_vm *remote,
189 				 struct sockaddr_vm *local)
190 {
191 	static u32 host_ephemeral_port = MIN_HOST_EPHEMERAL_PORT;
192 	struct sock *sk;
193 
194 	vsock_addr_init(remote, VMADDR_CID_ANY, VMADDR_PORT_ANY);
195 
196 	while (1) {
197 		/* Wrap around ? */
198 		if (host_ephemeral_port < MIN_HOST_EPHEMERAL_PORT ||
199 		    host_ephemeral_port == VMADDR_PORT_ANY)
200 			host_ephemeral_port = MIN_HOST_EPHEMERAL_PORT;
201 
202 		remote->svm_port = host_ephemeral_port++;
203 
204 		sk = vsock_find_connected_socket(remote, local);
205 		if (!sk) {
206 			/* Found an available ephemeral port */
207 			return;
208 		}
209 
210 		/* Release refcnt got in vsock_find_connected_socket */
211 		sock_put(sk);
212 	}
213 }
214 
215 static void hvs_set_channel_pending_send_size(struct vmbus_channel *chan)
216 {
217 	set_channel_pending_send_size(chan,
218 				      HVS_PKT_LEN(HVS_SEND_BUF_SIZE));
219 
220 	/* See hvs_stream_has_space(): we must make sure the host has seen
221 	 * the new pending send size, before we can re-check the writable
222 	 * bytes.
223 	 */
224 	virt_mb();
225 }
226 
227 static void hvs_clear_channel_pending_send_size(struct vmbus_channel *chan)
228 {
229 	set_channel_pending_send_size(chan, 0);
230 
231 	/* Ditto */
232 	virt_mb();
233 }
234 
235 static bool hvs_channel_readable(struct vmbus_channel *chan)
236 {
237 	u32 readable = hv_get_bytes_to_read(&chan->inbound);
238 
239 	/* 0-size payload means FIN */
240 	return readable >= HVS_PKT_LEN(0);
241 }
242 
243 static int hvs_channel_readable_payload(struct vmbus_channel *chan)
244 {
245 	u32 readable = hv_get_bytes_to_read(&chan->inbound);
246 
247 	if (readable > HVS_PKT_LEN(0)) {
248 		/* At least we have 1 byte to read. We don't need to return
249 		 * the exact readable bytes: see vsock_stream_recvmsg() ->
250 		 * vsock_stream_has_data().
251 		 */
252 		return 1;
253 	}
254 
255 	if (readable == HVS_PKT_LEN(0)) {
256 		/* 0-size payload means FIN */
257 		return 0;
258 	}
259 
260 	/* No payload or FIN */
261 	return -1;
262 }
263 
264 static size_t hvs_channel_writable_bytes(struct vmbus_channel *chan)
265 {
266 	u32 writeable = hv_get_bytes_to_write(&chan->outbound);
267 	size_t ret;
268 
269 	/* The ringbuffer mustn't be 100% full, and we should reserve a
270 	 * zero-length-payload packet for the FIN: see hv_ringbuffer_write()
271 	 * and hvs_shutdown().
272 	 */
273 	if (writeable <= HVS_PKT_LEN(1) + HVS_PKT_LEN(0))
274 		return 0;
275 
276 	ret = writeable - HVS_PKT_LEN(1) - HVS_PKT_LEN(0);
277 
278 	return round_down(ret, 8);
279 }
280 
281 static int hvs_send_data(struct vmbus_channel *chan,
282 			 struct hvs_send_buf *send_buf, size_t to_write)
283 {
284 	send_buf->hdr.pkt_type = 1;
285 	send_buf->hdr.data_size = to_write;
286 	return vmbus_sendpacket(chan, &send_buf->hdr,
287 				sizeof(send_buf->hdr) + to_write,
288 				0, VM_PKT_DATA_INBAND, 0);
289 }
290 
291 static void hvs_channel_cb(void *ctx)
292 {
293 	struct sock *sk = (struct sock *)ctx;
294 	struct vsock_sock *vsk = vsock_sk(sk);
295 	struct hvsock *hvs = vsk->trans;
296 	struct vmbus_channel *chan = hvs->chan;
297 
298 	if (hvs_channel_readable(chan))
299 		sk->sk_data_ready(sk);
300 
301 	/* See hvs_stream_has_space(): when we reach here, the writable bytes
302 	 * may be already less than HVS_PKT_LEN(HVS_SEND_BUF_SIZE).
303 	 */
304 	if (hv_get_bytes_to_write(&chan->outbound) > 0)
305 		sk->sk_write_space(sk);
306 }
307 
308 static void hvs_close_connection(struct vmbus_channel *chan)
309 {
310 	struct sock *sk = get_per_channel_state(chan);
311 	struct vsock_sock *vsk = vsock_sk(sk);
312 
313 	lock_sock(sk);
314 
315 	sk->sk_state = SS_UNCONNECTED;
316 	sock_set_flag(sk, SOCK_DONE);
317 	vsk->peer_shutdown |= SEND_SHUTDOWN | RCV_SHUTDOWN;
318 
319 	sk->sk_state_change(sk);
320 
321 	release_sock(sk);
322 }
323 
324 static void hvs_open_connection(struct vmbus_channel *chan)
325 {
326 	uuid_le *if_instance, *if_type;
327 	unsigned char conn_from_host;
328 
329 	struct sockaddr_vm addr;
330 	struct sock *sk, *new = NULL;
331 	struct vsock_sock *vnew;
332 	struct hvsock *hvs, *hvs_new;
333 	int ret;
334 
335 	if_type = &chan->offermsg.offer.if_type;
336 	if_instance = &chan->offermsg.offer.if_instance;
337 	conn_from_host = chan->offermsg.offer.u.pipe.user_def[0];
338 
339 	/* The host or the VM should only listen on a port in
340 	 * [0, MAX_LISTEN_PORT]
341 	 */
342 	if (!is_valid_srv_id(if_type) ||
343 	    get_port_by_srv_id(if_type) > MAX_LISTEN_PORT)
344 		return;
345 
346 	hvs_addr_init(&addr, conn_from_host ? if_type : if_instance);
347 	sk = vsock_find_bound_socket(&addr);
348 	if (!sk)
349 		return;
350 
351 	lock_sock(sk);
352 
353 	if ((conn_from_host && sk->sk_state != VSOCK_SS_LISTEN) ||
354 	    (!conn_from_host && sk->sk_state != SS_CONNECTING))
355 		goto out;
356 
357 	if (conn_from_host) {
358 		if (sk->sk_ack_backlog >= sk->sk_max_ack_backlog)
359 			goto out;
360 
361 		new = __vsock_create(sock_net(sk), NULL, sk, GFP_KERNEL,
362 				     sk->sk_type, 0);
363 		if (!new)
364 			goto out;
365 
366 		new->sk_state = SS_CONNECTING;
367 		vnew = vsock_sk(new);
368 		hvs_new = vnew->trans;
369 		hvs_new->chan = chan;
370 	} else {
371 		hvs = vsock_sk(sk)->trans;
372 		hvs->chan = chan;
373 	}
374 
375 	set_channel_read_mode(chan, HV_CALL_DIRECT);
376 	ret = vmbus_open(chan, RINGBUFFER_HVS_SND_SIZE,
377 			 RINGBUFFER_HVS_RCV_SIZE, NULL, 0,
378 			 hvs_channel_cb, conn_from_host ? new : sk);
379 	if (ret != 0) {
380 		if (conn_from_host) {
381 			hvs_new->chan = NULL;
382 			sock_put(new);
383 		} else {
384 			hvs->chan = NULL;
385 		}
386 		goto out;
387 	}
388 
389 	set_per_channel_state(chan, conn_from_host ? new : sk);
390 	vmbus_set_chn_rescind_callback(chan, hvs_close_connection);
391 
392 	if (conn_from_host) {
393 		new->sk_state = SS_CONNECTED;
394 		sk->sk_ack_backlog++;
395 
396 		hvs_addr_init(&vnew->local_addr, if_type);
397 		hvs_remote_addr_init(&vnew->remote_addr, &vnew->local_addr);
398 
399 		hvs_new->vm_srv_id = *if_type;
400 		hvs_new->host_srv_id = *if_instance;
401 
402 		vsock_insert_connected(vnew);
403 
404 		vsock_enqueue_accept(sk, new);
405 	} else {
406 		sk->sk_state = SS_CONNECTED;
407 		sk->sk_socket->state = SS_CONNECTED;
408 
409 		vsock_insert_connected(vsock_sk(sk));
410 	}
411 
412 	sk->sk_state_change(sk);
413 
414 out:
415 	/* Release refcnt obtained when we called vsock_find_bound_socket() */
416 	sock_put(sk);
417 
418 	release_sock(sk);
419 }
420 
421 static u32 hvs_get_local_cid(void)
422 {
423 	return VMADDR_CID_ANY;
424 }
425 
426 static int hvs_sock_init(struct vsock_sock *vsk, struct vsock_sock *psk)
427 {
428 	struct hvsock *hvs;
429 
430 	hvs = kzalloc(sizeof(*hvs), GFP_KERNEL);
431 	if (!hvs)
432 		return -ENOMEM;
433 
434 	vsk->trans = hvs;
435 	hvs->vsk = vsk;
436 
437 	return 0;
438 }
439 
440 static int hvs_connect(struct vsock_sock *vsk)
441 {
442 	union hvs_service_id vm, host;
443 	struct hvsock *h = vsk->trans;
444 
445 	vm.srv_id = srv_id_template;
446 	vm.svm_port = vsk->local_addr.svm_port;
447 	h->vm_srv_id = vm.srv_id;
448 
449 	host.srv_id = srv_id_template;
450 	host.svm_port = vsk->remote_addr.svm_port;
451 	h->host_srv_id = host.srv_id;
452 
453 	return vmbus_send_tl_connect_request(&h->vm_srv_id, &h->host_srv_id);
454 }
455 
456 static int hvs_shutdown(struct vsock_sock *vsk, int mode)
457 {
458 	struct sock *sk = sk_vsock(vsk);
459 	struct vmpipe_proto_header hdr;
460 	struct hvs_send_buf *send_buf;
461 	struct hvsock *hvs;
462 
463 	if (!(mode & SEND_SHUTDOWN))
464 		return 0;
465 
466 	lock_sock(sk);
467 
468 	hvs = vsk->trans;
469 	if (hvs->fin_sent)
470 		goto out;
471 
472 	send_buf = (struct hvs_send_buf *)&hdr;
473 
474 	/* It can't fail: see hvs_channel_writable_bytes(). */
475 	(void)hvs_send_data(hvs->chan, send_buf, 0);
476 
477 	hvs->fin_sent = true;
478 out:
479 	release_sock(sk);
480 	return 0;
481 }
482 
483 static void hvs_release(struct vsock_sock *vsk)
484 {
485 	struct sock *sk = sk_vsock(vsk);
486 	struct hvsock *hvs = vsk->trans;
487 	struct vmbus_channel *chan;
488 
489 	lock_sock(sk);
490 
491 	sk->sk_state = SS_DISCONNECTING;
492 	vsock_remove_sock(vsk);
493 
494 	release_sock(sk);
495 
496 	chan = hvs->chan;
497 	if (chan)
498 		hvs_shutdown(vsk, RCV_SHUTDOWN | SEND_SHUTDOWN);
499 
500 }
501 
502 static void hvs_destruct(struct vsock_sock *vsk)
503 {
504 	struct hvsock *hvs = vsk->trans;
505 	struct vmbus_channel *chan = hvs->chan;
506 
507 	if (chan)
508 		vmbus_hvsock_device_unregister(chan);
509 
510 	kfree(hvs);
511 }
512 
513 static int hvs_dgram_bind(struct vsock_sock *vsk, struct sockaddr_vm *addr)
514 {
515 	return -EOPNOTSUPP;
516 }
517 
518 static int hvs_dgram_dequeue(struct vsock_sock *vsk, struct msghdr *msg,
519 			     size_t len, int flags)
520 {
521 	return -EOPNOTSUPP;
522 }
523 
524 static int hvs_dgram_enqueue(struct vsock_sock *vsk,
525 			     struct sockaddr_vm *remote, struct msghdr *msg,
526 			     size_t dgram_len)
527 {
528 	return -EOPNOTSUPP;
529 }
530 
531 static bool hvs_dgram_allow(u32 cid, u32 port)
532 {
533 	return false;
534 }
535 
536 static int hvs_update_recv_data(struct hvsock *hvs)
537 {
538 	struct hvs_recv_buf *recv_buf;
539 	u32 payload_len;
540 
541 	recv_buf = (struct hvs_recv_buf *)(hvs->recv_desc + 1);
542 	payload_len = recv_buf->hdr.data_size;
543 
544 	if (payload_len > HVS_MTU_SIZE)
545 		return -EIO;
546 
547 	if (payload_len == 0)
548 		hvs->vsk->peer_shutdown |= SEND_SHUTDOWN;
549 
550 	hvs->recv_data_len = payload_len;
551 	hvs->recv_data_off = 0;
552 
553 	return 0;
554 }
555 
556 static ssize_t hvs_stream_dequeue(struct vsock_sock *vsk, struct msghdr *msg,
557 				  size_t len, int flags)
558 {
559 	struct hvsock *hvs = vsk->trans;
560 	bool need_refill = !hvs->recv_desc;
561 	struct hvs_recv_buf *recv_buf;
562 	u32 to_read;
563 	int ret;
564 
565 	if (flags & MSG_PEEK)
566 		return -EOPNOTSUPP;
567 
568 	if (need_refill) {
569 		hvs->recv_desc = hv_pkt_iter_first(hvs->chan);
570 		ret = hvs_update_recv_data(hvs);
571 		if (ret)
572 			return ret;
573 	}
574 
575 	recv_buf = (struct hvs_recv_buf *)(hvs->recv_desc + 1);
576 	to_read = min_t(u32, len, hvs->recv_data_len);
577 	ret = memcpy_to_msg(msg, recv_buf->data + hvs->recv_data_off, to_read);
578 	if (ret != 0)
579 		return ret;
580 
581 	hvs->recv_data_len -= to_read;
582 	if (hvs->recv_data_len == 0) {
583 		hvs->recv_desc = hv_pkt_iter_next(hvs->chan, hvs->recv_desc);
584 		if (hvs->recv_desc) {
585 			ret = hvs_update_recv_data(hvs);
586 			if (ret)
587 				return ret;
588 		}
589 	} else {
590 		hvs->recv_data_off += to_read;
591 	}
592 
593 	return to_read;
594 }
595 
596 static ssize_t hvs_stream_enqueue(struct vsock_sock *vsk, struct msghdr *msg,
597 				  size_t len)
598 {
599 	struct hvsock *hvs = vsk->trans;
600 	struct vmbus_channel *chan = hvs->chan;
601 	struct hvs_send_buf *send_buf;
602 	ssize_t to_write, max_writable, ret;
603 
604 	BUILD_BUG_ON(sizeof(*send_buf) != PAGE_SIZE_4K);
605 
606 	send_buf = kmalloc(sizeof(*send_buf), GFP_KERNEL);
607 	if (!send_buf)
608 		return -ENOMEM;
609 
610 	max_writable = hvs_channel_writable_bytes(chan);
611 	to_write = min_t(ssize_t, len, max_writable);
612 	to_write = min_t(ssize_t, to_write, HVS_SEND_BUF_SIZE);
613 
614 	ret = memcpy_from_msg(send_buf->data, msg, to_write);
615 	if (ret < 0)
616 		goto out;
617 
618 	ret = hvs_send_data(hvs->chan, send_buf, to_write);
619 	if (ret < 0)
620 		goto out;
621 
622 	ret = to_write;
623 out:
624 	kfree(send_buf);
625 	return ret;
626 }
627 
628 static s64 hvs_stream_has_data(struct vsock_sock *vsk)
629 {
630 	struct hvsock *hvs = vsk->trans;
631 	s64 ret;
632 
633 	if (hvs->recv_data_len > 0)
634 		return 1;
635 
636 	switch (hvs_channel_readable_payload(hvs->chan)) {
637 	case 1:
638 		ret = 1;
639 		break;
640 	case 0:
641 		vsk->peer_shutdown |= SEND_SHUTDOWN;
642 		ret = 0;
643 		break;
644 	default: /* -1 */
645 		ret = 0;
646 		break;
647 	}
648 
649 	return ret;
650 }
651 
652 static s64 hvs_stream_has_space(struct vsock_sock *vsk)
653 {
654 	struct hvsock *hvs = vsk->trans;
655 	struct vmbus_channel *chan = hvs->chan;
656 	s64 ret;
657 
658 	ret = hvs_channel_writable_bytes(chan);
659 	if (ret > 0)  {
660 		hvs_clear_channel_pending_send_size(chan);
661 	} else {
662 		/* See hvs_channel_cb() */
663 		hvs_set_channel_pending_send_size(chan);
664 
665 		/* Re-check the writable bytes to avoid race */
666 		ret = hvs_channel_writable_bytes(chan);
667 		if (ret > 0)
668 			hvs_clear_channel_pending_send_size(chan);
669 	}
670 
671 	return ret;
672 }
673 
674 static u64 hvs_stream_rcvhiwat(struct vsock_sock *vsk)
675 {
676 	return HVS_MTU_SIZE + 1;
677 }
678 
679 static bool hvs_stream_is_active(struct vsock_sock *vsk)
680 {
681 	struct hvsock *hvs = vsk->trans;
682 
683 	return hvs->chan != NULL;
684 }
685 
686 static bool hvs_stream_allow(u32 cid, u32 port)
687 {
688 	/* The host's port range [MIN_HOST_EPHEMERAL_PORT, 0xFFFFFFFF) is
689 	 * reserved as ephemeral ports, which are used as the host's ports
690 	 * when the host initiates connections.
691 	 *
692 	 * Perform this check in the guest so an immediate error is produced
693 	 * instead of a timeout.
694 	 */
695 	if (port > MAX_HOST_LISTEN_PORT)
696 		return false;
697 
698 	if (cid == VMADDR_CID_HOST)
699 		return true;
700 
701 	return false;
702 }
703 
704 static
705 int hvs_notify_poll_in(struct vsock_sock *vsk, size_t target, bool *readable)
706 {
707 	struct hvsock *hvs = vsk->trans;
708 
709 	*readable = hvs_channel_readable(hvs->chan);
710 	return 0;
711 }
712 
713 static
714 int hvs_notify_poll_out(struct vsock_sock *vsk, size_t target, bool *writable)
715 {
716 	*writable = hvs_stream_has_space(vsk) > 0;
717 
718 	return 0;
719 }
720 
721 static
722 int hvs_notify_recv_init(struct vsock_sock *vsk, size_t target,
723 			 struct vsock_transport_recv_notify_data *d)
724 {
725 	return 0;
726 }
727 
728 static
729 int hvs_notify_recv_pre_block(struct vsock_sock *vsk, size_t target,
730 			      struct vsock_transport_recv_notify_data *d)
731 {
732 	return 0;
733 }
734 
735 static
736 int hvs_notify_recv_pre_dequeue(struct vsock_sock *vsk, size_t target,
737 				struct vsock_transport_recv_notify_data *d)
738 {
739 	return 0;
740 }
741 
742 static
743 int hvs_notify_recv_post_dequeue(struct vsock_sock *vsk, size_t target,
744 				 ssize_t copied, bool data_read,
745 				 struct vsock_transport_recv_notify_data *d)
746 {
747 	return 0;
748 }
749 
750 static
751 int hvs_notify_send_init(struct vsock_sock *vsk,
752 			 struct vsock_transport_send_notify_data *d)
753 {
754 	return 0;
755 }
756 
757 static
758 int hvs_notify_send_pre_block(struct vsock_sock *vsk,
759 			      struct vsock_transport_send_notify_data *d)
760 {
761 	return 0;
762 }
763 
764 static
765 int hvs_notify_send_pre_enqueue(struct vsock_sock *vsk,
766 				struct vsock_transport_send_notify_data *d)
767 {
768 	return 0;
769 }
770 
771 static
772 int hvs_notify_send_post_enqueue(struct vsock_sock *vsk, ssize_t written,
773 				 struct vsock_transport_send_notify_data *d)
774 {
775 	return 0;
776 }
777 
778 static void hvs_set_buffer_size(struct vsock_sock *vsk, u64 val)
779 {
780 	/* Ignored. */
781 }
782 
783 static void hvs_set_min_buffer_size(struct vsock_sock *vsk, u64 val)
784 {
785 	/* Ignored. */
786 }
787 
788 static void hvs_set_max_buffer_size(struct vsock_sock *vsk, u64 val)
789 {
790 	/* Ignored. */
791 }
792 
793 static u64 hvs_get_buffer_size(struct vsock_sock *vsk)
794 {
795 	return -ENOPROTOOPT;
796 }
797 
798 static u64 hvs_get_min_buffer_size(struct vsock_sock *vsk)
799 {
800 	return -ENOPROTOOPT;
801 }
802 
803 static u64 hvs_get_max_buffer_size(struct vsock_sock *vsk)
804 {
805 	return -ENOPROTOOPT;
806 }
807 
808 static struct vsock_transport hvs_transport = {
809 	.get_local_cid            = hvs_get_local_cid,
810 
811 	.init                     = hvs_sock_init,
812 	.destruct                 = hvs_destruct,
813 	.release                  = hvs_release,
814 	.connect                  = hvs_connect,
815 	.shutdown                 = hvs_shutdown,
816 
817 	.dgram_bind               = hvs_dgram_bind,
818 	.dgram_dequeue            = hvs_dgram_dequeue,
819 	.dgram_enqueue            = hvs_dgram_enqueue,
820 	.dgram_allow              = hvs_dgram_allow,
821 
822 	.stream_dequeue           = hvs_stream_dequeue,
823 	.stream_enqueue           = hvs_stream_enqueue,
824 	.stream_has_data          = hvs_stream_has_data,
825 	.stream_has_space         = hvs_stream_has_space,
826 	.stream_rcvhiwat          = hvs_stream_rcvhiwat,
827 	.stream_is_active         = hvs_stream_is_active,
828 	.stream_allow             = hvs_stream_allow,
829 
830 	.notify_poll_in           = hvs_notify_poll_in,
831 	.notify_poll_out          = hvs_notify_poll_out,
832 	.notify_recv_init         = hvs_notify_recv_init,
833 	.notify_recv_pre_block    = hvs_notify_recv_pre_block,
834 	.notify_recv_pre_dequeue  = hvs_notify_recv_pre_dequeue,
835 	.notify_recv_post_dequeue = hvs_notify_recv_post_dequeue,
836 	.notify_send_init         = hvs_notify_send_init,
837 	.notify_send_pre_block    = hvs_notify_send_pre_block,
838 	.notify_send_pre_enqueue  = hvs_notify_send_pre_enqueue,
839 	.notify_send_post_enqueue = hvs_notify_send_post_enqueue,
840 
841 	.set_buffer_size          = hvs_set_buffer_size,
842 	.set_min_buffer_size      = hvs_set_min_buffer_size,
843 	.set_max_buffer_size      = hvs_set_max_buffer_size,
844 	.get_buffer_size          = hvs_get_buffer_size,
845 	.get_min_buffer_size      = hvs_get_min_buffer_size,
846 	.get_max_buffer_size      = hvs_get_max_buffer_size,
847 };
848 
849 static int hvs_probe(struct hv_device *hdev,
850 		     const struct hv_vmbus_device_id *dev_id)
851 {
852 	struct vmbus_channel *chan = hdev->channel;
853 
854 	hvs_open_connection(chan);
855 
856 	/* Always return success to suppress the unnecessary error message
857 	 * in vmbus_probe(): on error the host will rescind the device in
858 	 * 30 seconds and we can do cleanup at that time in
859 	 * vmbus_onoffer_rescind().
860 	 */
861 	return 0;
862 }
863 
864 static int hvs_remove(struct hv_device *hdev)
865 {
866 	struct vmbus_channel *chan = hdev->channel;
867 
868 	vmbus_close(chan);
869 
870 	return 0;
871 }
872 
873 /* This isn't really used. See vmbus_match() and vmbus_probe() */
874 static const struct hv_vmbus_device_id id_table[] = {
875 	{},
876 };
877 
878 static struct hv_driver hvs_drv = {
879 	.name		= "hv_sock",
880 	.hvsock		= true,
881 	.id_table	= id_table,
882 	.probe		= hvs_probe,
883 	.remove		= hvs_remove,
884 };
885 
886 static int __init hvs_init(void)
887 {
888 	int ret;
889 
890 	if (vmbus_proto_version < VERSION_WIN10)
891 		return -ENODEV;
892 
893 	ret = vmbus_driver_register(&hvs_drv);
894 	if (ret != 0)
895 		return ret;
896 
897 	ret = vsock_core_init(&hvs_transport);
898 	if (ret) {
899 		vmbus_driver_unregister(&hvs_drv);
900 		return ret;
901 	}
902 
903 	return 0;
904 }
905 
906 static void __exit hvs_exit(void)
907 {
908 	vsock_core_exit();
909 	vmbus_driver_unregister(&hvs_drv);
910 }
911 
912 module_init(hvs_init);
913 module_exit(hvs_exit);
914 
915 MODULE_DESCRIPTION("Hyper-V Sockets");
916 MODULE_VERSION("1.0.0");
917 MODULE_LICENSE("GPL");
918 MODULE_ALIAS_NETPROTO(PF_VSOCK);
919