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 /* How long to wait for graceful shutdown of a connection */
39 #define HVS_CLOSE_TIMEOUT (8 * HZ)
40 
41 struct vmpipe_proto_header {
42 	u32 pkt_type;
43 	u32 data_size;
44 };
45 
46 /* For recv, we use the VMBus in-place packet iterator APIs to directly copy
47  * data from the ringbuffer into the userspace buffer.
48  */
49 struct hvs_recv_buf {
50 	/* The header before the payload data */
51 	struct vmpipe_proto_header hdr;
52 
53 	/* The payload */
54 	u8 data[HVS_MTU_SIZE];
55 };
56 
57 /* We can send up to HVS_MTU_SIZE bytes of payload to the host, but let's use
58  * a small size, i.e. HVS_SEND_BUF_SIZE, to minimize the dynamically-allocated
59  * buffer, because tests show there is no significant performance difference.
60  *
61  * Note: the buffer can be eliminated in the future when we add new VMBus
62  * ringbuffer APIs that allow us to directly copy data from userspace buffer
63  * to VMBus ringbuffer.
64  */
65 #define HVS_SEND_BUF_SIZE (PAGE_SIZE_4K - sizeof(struct vmpipe_proto_header))
66 
67 struct hvs_send_buf {
68 	/* The header before the payload data */
69 	struct vmpipe_proto_header hdr;
70 
71 	/* The payload */
72 	u8 data[HVS_SEND_BUF_SIZE];
73 };
74 
75 #define HVS_HEADER_LEN	(sizeof(struct vmpacket_descriptor) + \
76 			 sizeof(struct vmpipe_proto_header))
77 
78 /* See 'prev_indices' in hv_ringbuffer_read(), hv_ringbuffer_write(), and
79  * __hv_pkt_iter_next().
80  */
81 #define VMBUS_PKT_TRAILER_SIZE	(sizeof(u64))
82 
83 #define HVS_PKT_LEN(payload_len)	(HVS_HEADER_LEN + \
84 					 ALIGN((payload_len), 8) + \
85 					 VMBUS_PKT_TRAILER_SIZE)
86 
87 union hvs_service_id {
88 	uuid_le	srv_id;
89 
90 	struct {
91 		unsigned int svm_port;
92 		unsigned char b[sizeof(uuid_le) - sizeof(unsigned int)];
93 	};
94 };
95 
96 /* Per-socket state (accessed via vsk->trans) */
97 struct hvsock {
98 	struct vsock_sock *vsk;
99 
100 	uuid_le vm_srv_id;
101 	uuid_le host_srv_id;
102 
103 	struct vmbus_channel *chan;
104 	struct vmpacket_descriptor *recv_desc;
105 
106 	/* The length of the payload not delivered to userland yet */
107 	u32 recv_data_len;
108 	/* The offset of the payload */
109 	u32 recv_data_off;
110 
111 	/* Have we sent the zero-length packet (FIN)? */
112 	bool fin_sent;
113 };
114 
115 /* In the VM, we support Hyper-V Sockets with AF_VSOCK, and the endpoint is
116  * <cid, port> (see struct sockaddr_vm). Note: cid is not really used here:
117  * when we write apps to connect to the host, we can only use VMADDR_CID_ANY
118  * or VMADDR_CID_HOST (both are equivalent) as the remote cid, and when we
119  * write apps to bind() & listen() in the VM, we can only use VMADDR_CID_ANY
120  * as the local cid.
121  *
122  * On the host, Hyper-V Sockets are supported by Winsock AF_HYPERV:
123  * https://docs.microsoft.com/en-us/virtualization/hyper-v-on-windows/user-
124  * guide/make-integration-service, and the endpoint is <VmID, ServiceId> with
125  * the below sockaddr:
126  *
127  * struct SOCKADDR_HV
128  * {
129  *    ADDRESS_FAMILY Family;
130  *    USHORT Reserved;
131  *    GUID VmId;
132  *    GUID ServiceId;
133  * };
134  * Note: VmID is not used by Linux VM and actually it isn't transmitted via
135  * VMBus, because here it's obvious the host and the VM can easily identify
136  * each other. Though the VmID is useful on the host, especially in the case
137  * of Windows container, Linux VM doesn't need it at all.
138  *
139  * To make use of the AF_VSOCK infrastructure in Linux VM, we have to limit
140  * the available GUID space of SOCKADDR_HV so that we can create a mapping
141  * between AF_VSOCK port and SOCKADDR_HV Service GUID. The rule of writing
142  * Hyper-V Sockets apps on the host and in Linux VM is:
143  *
144  ****************************************************************************
145  * The only valid Service GUIDs, from the perspectives of both the host and *
146  * Linux VM, that can be connected by the other end, must conform to this   *
147  * format: <port>-facb-11e6-bd58-64006a7986d3, and the "port" must be in    *
148  * this range [0, 0x7FFFFFFF].                                              *
149  ****************************************************************************
150  *
151  * When we write apps on the host to connect(), the GUID ServiceID is used.
152  * When we write apps in Linux VM to connect(), we only need to specify the
153  * port and the driver will form the GUID and use that to request the host.
154  *
155  * From the perspective of Linux VM:
156  * 1. the local ephemeral port (i.e. the local auto-bound port when we call
157  * connect() without explicit bind()) is generated by __vsock_bind_stream(),
158  * and the range is [1024, 0xFFFFFFFF).
159  * 2. the remote ephemeral port (i.e. the auto-generated remote port for
160  * a connect request initiated by the host's connect()) is generated by
161  * hvs_remote_addr_init() and the range is [0x80000000, 0xFFFFFFFF).
162  */
163 
164 #define MAX_LISTEN_PORT			((u32)0x7FFFFFFF)
165 #define MAX_VM_LISTEN_PORT		MAX_LISTEN_PORT
166 #define MAX_HOST_LISTEN_PORT		MAX_LISTEN_PORT
167 #define MIN_HOST_EPHEMERAL_PORT		(MAX_HOST_LISTEN_PORT + 1)
168 
169 /* 00000000-facb-11e6-bd58-64006a7986d3 */
170 static const uuid_le srv_id_template =
171 	UUID_LE(0x00000000, 0xfacb, 0x11e6, 0xbd, 0x58,
172 		0x64, 0x00, 0x6a, 0x79, 0x86, 0xd3);
173 
174 static bool is_valid_srv_id(const uuid_le *id)
175 {
176 	return !memcmp(&id->b[4], &srv_id_template.b[4], sizeof(uuid_le) - 4);
177 }
178 
179 static unsigned int get_port_by_srv_id(const uuid_le *svr_id)
180 {
181 	return *((unsigned int *)svr_id);
182 }
183 
184 static void hvs_addr_init(struct sockaddr_vm *addr, const uuid_le *svr_id)
185 {
186 	unsigned int port = get_port_by_srv_id(svr_id);
187 
188 	vsock_addr_init(addr, VMADDR_CID_ANY, port);
189 }
190 
191 static void hvs_remote_addr_init(struct sockaddr_vm *remote,
192 				 struct sockaddr_vm *local)
193 {
194 	static u32 host_ephemeral_port = MIN_HOST_EPHEMERAL_PORT;
195 	struct sock *sk;
196 
197 	vsock_addr_init(remote, VMADDR_CID_ANY, VMADDR_PORT_ANY);
198 
199 	while (1) {
200 		/* Wrap around ? */
201 		if (host_ephemeral_port < MIN_HOST_EPHEMERAL_PORT ||
202 		    host_ephemeral_port == VMADDR_PORT_ANY)
203 			host_ephemeral_port = MIN_HOST_EPHEMERAL_PORT;
204 
205 		remote->svm_port = host_ephemeral_port++;
206 
207 		sk = vsock_find_connected_socket(remote, local);
208 		if (!sk) {
209 			/* Found an available ephemeral port */
210 			return;
211 		}
212 
213 		/* Release refcnt got in vsock_find_connected_socket */
214 		sock_put(sk);
215 	}
216 }
217 
218 static void hvs_set_channel_pending_send_size(struct vmbus_channel *chan)
219 {
220 	set_channel_pending_send_size(chan,
221 				      HVS_PKT_LEN(HVS_SEND_BUF_SIZE));
222 
223 	/* See hvs_stream_has_space(): we must make sure the host has seen
224 	 * the new pending send size, before we can re-check the writable
225 	 * bytes.
226 	 */
227 	virt_mb();
228 }
229 
230 static void hvs_clear_channel_pending_send_size(struct vmbus_channel *chan)
231 {
232 	set_channel_pending_send_size(chan, 0);
233 
234 	/* Ditto */
235 	virt_mb();
236 }
237 
238 static bool hvs_channel_readable(struct vmbus_channel *chan)
239 {
240 	u32 readable = hv_get_bytes_to_read(&chan->inbound);
241 
242 	/* 0-size payload means FIN */
243 	return readable >= HVS_PKT_LEN(0);
244 }
245 
246 static int hvs_channel_readable_payload(struct vmbus_channel *chan)
247 {
248 	u32 readable = hv_get_bytes_to_read(&chan->inbound);
249 
250 	if (readable > HVS_PKT_LEN(0)) {
251 		/* At least we have 1 byte to read. We don't need to return
252 		 * the exact readable bytes: see vsock_stream_recvmsg() ->
253 		 * vsock_stream_has_data().
254 		 */
255 		return 1;
256 	}
257 
258 	if (readable == HVS_PKT_LEN(0)) {
259 		/* 0-size payload means FIN */
260 		return 0;
261 	}
262 
263 	/* No payload or FIN */
264 	return -1;
265 }
266 
267 static size_t hvs_channel_writable_bytes(struct vmbus_channel *chan)
268 {
269 	u32 writeable = hv_get_bytes_to_write(&chan->outbound);
270 	size_t ret;
271 
272 	/* The ringbuffer mustn't be 100% full, and we should reserve a
273 	 * zero-length-payload packet for the FIN: see hv_ringbuffer_write()
274 	 * and hvs_shutdown().
275 	 */
276 	if (writeable <= HVS_PKT_LEN(1) + HVS_PKT_LEN(0))
277 		return 0;
278 
279 	ret = writeable - HVS_PKT_LEN(1) - HVS_PKT_LEN(0);
280 
281 	return round_down(ret, 8);
282 }
283 
284 static int hvs_send_data(struct vmbus_channel *chan,
285 			 struct hvs_send_buf *send_buf, size_t to_write)
286 {
287 	send_buf->hdr.pkt_type = 1;
288 	send_buf->hdr.data_size = to_write;
289 	return vmbus_sendpacket(chan, &send_buf->hdr,
290 				sizeof(send_buf->hdr) + to_write,
291 				0, VM_PKT_DATA_INBAND, 0);
292 }
293 
294 static void hvs_channel_cb(void *ctx)
295 {
296 	struct sock *sk = (struct sock *)ctx;
297 	struct vsock_sock *vsk = vsock_sk(sk);
298 	struct hvsock *hvs = vsk->trans;
299 	struct vmbus_channel *chan = hvs->chan;
300 
301 	if (hvs_channel_readable(chan))
302 		sk->sk_data_ready(sk);
303 
304 	/* See hvs_stream_has_space(): when we reach here, the writable bytes
305 	 * may be already less than HVS_PKT_LEN(HVS_SEND_BUF_SIZE).
306 	 */
307 	if (hv_get_bytes_to_write(&chan->outbound) > 0)
308 		sk->sk_write_space(sk);
309 }
310 
311 static void hvs_do_close_lock_held(struct vsock_sock *vsk,
312 				   bool cancel_timeout)
313 {
314 	struct sock *sk = sk_vsock(vsk);
315 
316 	sock_set_flag(sk, SOCK_DONE);
317 	vsk->peer_shutdown = SHUTDOWN_MASK;
318 	if (vsock_stream_has_data(vsk) <= 0)
319 		sk->sk_state = TCP_CLOSING;
320 	sk->sk_state_change(sk);
321 	if (vsk->close_work_scheduled &&
322 	    (!cancel_timeout || cancel_delayed_work(&vsk->close_work))) {
323 		vsk->close_work_scheduled = false;
324 		vsock_remove_sock(vsk);
325 
326 		/* Release the reference taken while scheduling the timeout */
327 		sock_put(sk);
328 	}
329 }
330 
331 static void hvs_close_connection(struct vmbus_channel *chan)
332 {
333 	struct sock *sk = get_per_channel_state(chan);
334 
335 	lock_sock(sk);
336 	hvs_do_close_lock_held(vsock_sk(sk), true);
337 	release_sock(sk);
338 }
339 
340 static void hvs_open_connection(struct vmbus_channel *chan)
341 {
342 	uuid_le *if_instance, *if_type;
343 	unsigned char conn_from_host;
344 
345 	struct sockaddr_vm addr;
346 	struct sock *sk, *new = NULL;
347 	struct vsock_sock *vnew;
348 	struct hvsock *hvs, *hvs_new;
349 	int ret;
350 
351 	if_type = &chan->offermsg.offer.if_type;
352 	if_instance = &chan->offermsg.offer.if_instance;
353 	conn_from_host = chan->offermsg.offer.u.pipe.user_def[0];
354 
355 	/* The host or the VM should only listen on a port in
356 	 * [0, MAX_LISTEN_PORT]
357 	 */
358 	if (!is_valid_srv_id(if_type) ||
359 	    get_port_by_srv_id(if_type) > MAX_LISTEN_PORT)
360 		return;
361 
362 	hvs_addr_init(&addr, conn_from_host ? if_type : if_instance);
363 	sk = vsock_find_bound_socket(&addr);
364 	if (!sk)
365 		return;
366 
367 	lock_sock(sk);
368 	if ((conn_from_host && sk->sk_state != TCP_LISTEN) ||
369 	    (!conn_from_host && sk->sk_state != TCP_SYN_SENT))
370 		goto out;
371 
372 	if (conn_from_host) {
373 		if (sk->sk_ack_backlog >= sk->sk_max_ack_backlog)
374 			goto out;
375 
376 		new = __vsock_create(sock_net(sk), NULL, sk, GFP_KERNEL,
377 				     sk->sk_type, 0);
378 		if (!new)
379 			goto out;
380 
381 		new->sk_state = TCP_SYN_SENT;
382 		vnew = vsock_sk(new);
383 		hvs_new = vnew->trans;
384 		hvs_new->chan = chan;
385 	} else {
386 		hvs = vsock_sk(sk)->trans;
387 		hvs->chan = chan;
388 	}
389 
390 	set_channel_read_mode(chan, HV_CALL_DIRECT);
391 	ret = vmbus_open(chan, RINGBUFFER_HVS_SND_SIZE,
392 			 RINGBUFFER_HVS_RCV_SIZE, NULL, 0,
393 			 hvs_channel_cb, conn_from_host ? new : sk);
394 	if (ret != 0) {
395 		if (conn_from_host) {
396 			hvs_new->chan = NULL;
397 			sock_put(new);
398 		} else {
399 			hvs->chan = NULL;
400 		}
401 		goto out;
402 	}
403 
404 	set_per_channel_state(chan, conn_from_host ? new : sk);
405 	vmbus_set_chn_rescind_callback(chan, hvs_close_connection);
406 
407 	if (conn_from_host) {
408 		new->sk_state = TCP_ESTABLISHED;
409 		sk->sk_ack_backlog++;
410 
411 		hvs_addr_init(&vnew->local_addr, if_type);
412 		hvs_remote_addr_init(&vnew->remote_addr, &vnew->local_addr);
413 
414 		hvs_new->vm_srv_id = *if_type;
415 		hvs_new->host_srv_id = *if_instance;
416 
417 		vsock_insert_connected(vnew);
418 
419 		vsock_enqueue_accept(sk, new);
420 	} else {
421 		sk->sk_state = TCP_ESTABLISHED;
422 		sk->sk_socket->state = SS_CONNECTED;
423 
424 		vsock_insert_connected(vsock_sk(sk));
425 	}
426 
427 	sk->sk_state_change(sk);
428 
429 out:
430 	/* Release refcnt obtained when we called vsock_find_bound_socket() */
431 	sock_put(sk);
432 
433 	release_sock(sk);
434 }
435 
436 static u32 hvs_get_local_cid(void)
437 {
438 	return VMADDR_CID_ANY;
439 }
440 
441 static int hvs_sock_init(struct vsock_sock *vsk, struct vsock_sock *psk)
442 {
443 	struct hvsock *hvs;
444 
445 	hvs = kzalloc(sizeof(*hvs), GFP_KERNEL);
446 	if (!hvs)
447 		return -ENOMEM;
448 
449 	vsk->trans = hvs;
450 	hvs->vsk = vsk;
451 
452 	return 0;
453 }
454 
455 static int hvs_connect(struct vsock_sock *vsk)
456 {
457 	union hvs_service_id vm, host;
458 	struct hvsock *h = vsk->trans;
459 
460 	vm.srv_id = srv_id_template;
461 	vm.svm_port = vsk->local_addr.svm_port;
462 	h->vm_srv_id = vm.srv_id;
463 
464 	host.srv_id = srv_id_template;
465 	host.svm_port = vsk->remote_addr.svm_port;
466 	h->host_srv_id = host.srv_id;
467 
468 	return vmbus_send_tl_connect_request(&h->vm_srv_id, &h->host_srv_id);
469 }
470 
471 static void hvs_shutdown_lock_held(struct hvsock *hvs, int mode)
472 {
473 	struct vmpipe_proto_header hdr;
474 
475 	if (hvs->fin_sent || !hvs->chan)
476 		return;
477 
478 	/* It can't fail: see hvs_channel_writable_bytes(). */
479 	(void)hvs_send_data(hvs->chan, (struct hvs_send_buf *)&hdr, 0);
480 	hvs->fin_sent = true;
481 }
482 
483 static int hvs_shutdown(struct vsock_sock *vsk, int mode)
484 {
485 	struct sock *sk = sk_vsock(vsk);
486 
487 	if (!(mode & SEND_SHUTDOWN))
488 		return 0;
489 
490 	lock_sock(sk);
491 	hvs_shutdown_lock_held(vsk->trans, mode);
492 	release_sock(sk);
493 	return 0;
494 }
495 
496 static void hvs_close_timeout(struct work_struct *work)
497 {
498 	struct vsock_sock *vsk =
499 		container_of(work, struct vsock_sock, close_work.work);
500 	struct sock *sk = sk_vsock(vsk);
501 
502 	sock_hold(sk);
503 	lock_sock(sk);
504 	if (!sock_flag(sk, SOCK_DONE))
505 		hvs_do_close_lock_held(vsk, false);
506 
507 	vsk->close_work_scheduled = false;
508 	release_sock(sk);
509 	sock_put(sk);
510 }
511 
512 /* Returns true, if it is safe to remove socket; false otherwise */
513 static bool hvs_close_lock_held(struct vsock_sock *vsk)
514 {
515 	struct sock *sk = sk_vsock(vsk);
516 
517 	if (!(sk->sk_state == TCP_ESTABLISHED ||
518 	      sk->sk_state == TCP_CLOSING))
519 		return true;
520 
521 	if ((sk->sk_shutdown & SHUTDOWN_MASK) != SHUTDOWN_MASK)
522 		hvs_shutdown_lock_held(vsk->trans, SHUTDOWN_MASK);
523 
524 	if (sock_flag(sk, SOCK_DONE))
525 		return true;
526 
527 	/* This reference will be dropped by the delayed close routine */
528 	sock_hold(sk);
529 	INIT_DELAYED_WORK(&vsk->close_work, hvs_close_timeout);
530 	vsk->close_work_scheduled = true;
531 	schedule_delayed_work(&vsk->close_work, HVS_CLOSE_TIMEOUT);
532 	return false;
533 }
534 
535 static void hvs_release(struct vsock_sock *vsk)
536 {
537 	struct sock *sk = sk_vsock(vsk);
538 	bool remove_sock;
539 
540 	lock_sock(sk);
541 	remove_sock = hvs_close_lock_held(vsk);
542 	release_sock(sk);
543 	if (remove_sock)
544 		vsock_remove_sock(vsk);
545 }
546 
547 static void hvs_destruct(struct vsock_sock *vsk)
548 {
549 	struct hvsock *hvs = vsk->trans;
550 	struct vmbus_channel *chan = hvs->chan;
551 
552 	if (chan)
553 		vmbus_hvsock_device_unregister(chan);
554 
555 	kfree(hvs);
556 }
557 
558 static int hvs_dgram_bind(struct vsock_sock *vsk, struct sockaddr_vm *addr)
559 {
560 	return -EOPNOTSUPP;
561 }
562 
563 static int hvs_dgram_dequeue(struct vsock_sock *vsk, struct msghdr *msg,
564 			     size_t len, int flags)
565 {
566 	return -EOPNOTSUPP;
567 }
568 
569 static int hvs_dgram_enqueue(struct vsock_sock *vsk,
570 			     struct sockaddr_vm *remote, struct msghdr *msg,
571 			     size_t dgram_len)
572 {
573 	return -EOPNOTSUPP;
574 }
575 
576 static bool hvs_dgram_allow(u32 cid, u32 port)
577 {
578 	return false;
579 }
580 
581 static int hvs_update_recv_data(struct hvsock *hvs)
582 {
583 	struct hvs_recv_buf *recv_buf;
584 	u32 payload_len;
585 
586 	recv_buf = (struct hvs_recv_buf *)(hvs->recv_desc + 1);
587 	payload_len = recv_buf->hdr.data_size;
588 
589 	if (payload_len > HVS_MTU_SIZE)
590 		return -EIO;
591 
592 	if (payload_len == 0)
593 		hvs->vsk->peer_shutdown |= SEND_SHUTDOWN;
594 
595 	hvs->recv_data_len = payload_len;
596 	hvs->recv_data_off = 0;
597 
598 	return 0;
599 }
600 
601 static ssize_t hvs_stream_dequeue(struct vsock_sock *vsk, struct msghdr *msg,
602 				  size_t len, int flags)
603 {
604 	struct hvsock *hvs = vsk->trans;
605 	bool need_refill = !hvs->recv_desc;
606 	struct hvs_recv_buf *recv_buf;
607 	u32 to_read;
608 	int ret;
609 
610 	if (flags & MSG_PEEK)
611 		return -EOPNOTSUPP;
612 
613 	if (need_refill) {
614 		hvs->recv_desc = hv_pkt_iter_first(hvs->chan);
615 		ret = hvs_update_recv_data(hvs);
616 		if (ret)
617 			return ret;
618 	}
619 
620 	recv_buf = (struct hvs_recv_buf *)(hvs->recv_desc + 1);
621 	to_read = min_t(u32, len, hvs->recv_data_len);
622 	ret = memcpy_to_msg(msg, recv_buf->data + hvs->recv_data_off, to_read);
623 	if (ret != 0)
624 		return ret;
625 
626 	hvs->recv_data_len -= to_read;
627 	if (hvs->recv_data_len == 0) {
628 		hvs->recv_desc = hv_pkt_iter_next(hvs->chan, hvs->recv_desc);
629 		if (hvs->recv_desc) {
630 			ret = hvs_update_recv_data(hvs);
631 			if (ret)
632 				return ret;
633 		}
634 	} else {
635 		hvs->recv_data_off += to_read;
636 	}
637 
638 	return to_read;
639 }
640 
641 static ssize_t hvs_stream_enqueue(struct vsock_sock *vsk, struct msghdr *msg,
642 				  size_t len)
643 {
644 	struct hvsock *hvs = vsk->trans;
645 	struct vmbus_channel *chan = hvs->chan;
646 	struct hvs_send_buf *send_buf;
647 	ssize_t to_write, max_writable, ret;
648 
649 	BUILD_BUG_ON(sizeof(*send_buf) != PAGE_SIZE_4K);
650 
651 	send_buf = kmalloc(sizeof(*send_buf), GFP_KERNEL);
652 	if (!send_buf)
653 		return -ENOMEM;
654 
655 	max_writable = hvs_channel_writable_bytes(chan);
656 	to_write = min_t(ssize_t, len, max_writable);
657 	to_write = min_t(ssize_t, to_write, HVS_SEND_BUF_SIZE);
658 
659 	ret = memcpy_from_msg(send_buf->data, msg, to_write);
660 	if (ret < 0)
661 		goto out;
662 
663 	ret = hvs_send_data(hvs->chan, send_buf, to_write);
664 	if (ret < 0)
665 		goto out;
666 
667 	ret = to_write;
668 out:
669 	kfree(send_buf);
670 	return ret;
671 }
672 
673 static s64 hvs_stream_has_data(struct vsock_sock *vsk)
674 {
675 	struct hvsock *hvs = vsk->trans;
676 	s64 ret;
677 
678 	if (hvs->recv_data_len > 0)
679 		return 1;
680 
681 	switch (hvs_channel_readable_payload(hvs->chan)) {
682 	case 1:
683 		ret = 1;
684 		break;
685 	case 0:
686 		vsk->peer_shutdown |= SEND_SHUTDOWN;
687 		ret = 0;
688 		break;
689 	default: /* -1 */
690 		ret = 0;
691 		break;
692 	}
693 
694 	return ret;
695 }
696 
697 static s64 hvs_stream_has_space(struct vsock_sock *vsk)
698 {
699 	struct hvsock *hvs = vsk->trans;
700 	struct vmbus_channel *chan = hvs->chan;
701 	s64 ret;
702 
703 	ret = hvs_channel_writable_bytes(chan);
704 	if (ret > 0)  {
705 		hvs_clear_channel_pending_send_size(chan);
706 	} else {
707 		/* See hvs_channel_cb() */
708 		hvs_set_channel_pending_send_size(chan);
709 
710 		/* Re-check the writable bytes to avoid race */
711 		ret = hvs_channel_writable_bytes(chan);
712 		if (ret > 0)
713 			hvs_clear_channel_pending_send_size(chan);
714 	}
715 
716 	return ret;
717 }
718 
719 static u64 hvs_stream_rcvhiwat(struct vsock_sock *vsk)
720 {
721 	return HVS_MTU_SIZE + 1;
722 }
723 
724 static bool hvs_stream_is_active(struct vsock_sock *vsk)
725 {
726 	struct hvsock *hvs = vsk->trans;
727 
728 	return hvs->chan != NULL;
729 }
730 
731 static bool hvs_stream_allow(u32 cid, u32 port)
732 {
733 	/* The host's port range [MIN_HOST_EPHEMERAL_PORT, 0xFFFFFFFF) is
734 	 * reserved as ephemeral ports, which are used as the host's ports
735 	 * when the host initiates connections.
736 	 *
737 	 * Perform this check in the guest so an immediate error is produced
738 	 * instead of a timeout.
739 	 */
740 	if (port > MAX_HOST_LISTEN_PORT)
741 		return false;
742 
743 	if (cid == VMADDR_CID_HOST)
744 		return true;
745 
746 	return false;
747 }
748 
749 static
750 int hvs_notify_poll_in(struct vsock_sock *vsk, size_t target, bool *readable)
751 {
752 	struct hvsock *hvs = vsk->trans;
753 
754 	*readable = hvs_channel_readable(hvs->chan);
755 	return 0;
756 }
757 
758 static
759 int hvs_notify_poll_out(struct vsock_sock *vsk, size_t target, bool *writable)
760 {
761 	*writable = hvs_stream_has_space(vsk) > 0;
762 
763 	return 0;
764 }
765 
766 static
767 int hvs_notify_recv_init(struct vsock_sock *vsk, size_t target,
768 			 struct vsock_transport_recv_notify_data *d)
769 {
770 	return 0;
771 }
772 
773 static
774 int hvs_notify_recv_pre_block(struct vsock_sock *vsk, size_t target,
775 			      struct vsock_transport_recv_notify_data *d)
776 {
777 	return 0;
778 }
779 
780 static
781 int hvs_notify_recv_pre_dequeue(struct vsock_sock *vsk, size_t target,
782 				struct vsock_transport_recv_notify_data *d)
783 {
784 	return 0;
785 }
786 
787 static
788 int hvs_notify_recv_post_dequeue(struct vsock_sock *vsk, size_t target,
789 				 ssize_t copied, bool data_read,
790 				 struct vsock_transport_recv_notify_data *d)
791 {
792 	return 0;
793 }
794 
795 static
796 int hvs_notify_send_init(struct vsock_sock *vsk,
797 			 struct vsock_transport_send_notify_data *d)
798 {
799 	return 0;
800 }
801 
802 static
803 int hvs_notify_send_pre_block(struct vsock_sock *vsk,
804 			      struct vsock_transport_send_notify_data *d)
805 {
806 	return 0;
807 }
808 
809 static
810 int hvs_notify_send_pre_enqueue(struct vsock_sock *vsk,
811 				struct vsock_transport_send_notify_data *d)
812 {
813 	return 0;
814 }
815 
816 static
817 int hvs_notify_send_post_enqueue(struct vsock_sock *vsk, ssize_t written,
818 				 struct vsock_transport_send_notify_data *d)
819 {
820 	return 0;
821 }
822 
823 static void hvs_set_buffer_size(struct vsock_sock *vsk, u64 val)
824 {
825 	/* Ignored. */
826 }
827 
828 static void hvs_set_min_buffer_size(struct vsock_sock *vsk, u64 val)
829 {
830 	/* Ignored. */
831 }
832 
833 static void hvs_set_max_buffer_size(struct vsock_sock *vsk, u64 val)
834 {
835 	/* Ignored. */
836 }
837 
838 static u64 hvs_get_buffer_size(struct vsock_sock *vsk)
839 {
840 	return -ENOPROTOOPT;
841 }
842 
843 static u64 hvs_get_min_buffer_size(struct vsock_sock *vsk)
844 {
845 	return -ENOPROTOOPT;
846 }
847 
848 static u64 hvs_get_max_buffer_size(struct vsock_sock *vsk)
849 {
850 	return -ENOPROTOOPT;
851 }
852 
853 static struct vsock_transport hvs_transport = {
854 	.get_local_cid            = hvs_get_local_cid,
855 
856 	.init                     = hvs_sock_init,
857 	.destruct                 = hvs_destruct,
858 	.release                  = hvs_release,
859 	.connect                  = hvs_connect,
860 	.shutdown                 = hvs_shutdown,
861 
862 	.dgram_bind               = hvs_dgram_bind,
863 	.dgram_dequeue            = hvs_dgram_dequeue,
864 	.dgram_enqueue            = hvs_dgram_enqueue,
865 	.dgram_allow              = hvs_dgram_allow,
866 
867 	.stream_dequeue           = hvs_stream_dequeue,
868 	.stream_enqueue           = hvs_stream_enqueue,
869 	.stream_has_data          = hvs_stream_has_data,
870 	.stream_has_space         = hvs_stream_has_space,
871 	.stream_rcvhiwat          = hvs_stream_rcvhiwat,
872 	.stream_is_active         = hvs_stream_is_active,
873 	.stream_allow             = hvs_stream_allow,
874 
875 	.notify_poll_in           = hvs_notify_poll_in,
876 	.notify_poll_out          = hvs_notify_poll_out,
877 	.notify_recv_init         = hvs_notify_recv_init,
878 	.notify_recv_pre_block    = hvs_notify_recv_pre_block,
879 	.notify_recv_pre_dequeue  = hvs_notify_recv_pre_dequeue,
880 	.notify_recv_post_dequeue = hvs_notify_recv_post_dequeue,
881 	.notify_send_init         = hvs_notify_send_init,
882 	.notify_send_pre_block    = hvs_notify_send_pre_block,
883 	.notify_send_pre_enqueue  = hvs_notify_send_pre_enqueue,
884 	.notify_send_post_enqueue = hvs_notify_send_post_enqueue,
885 
886 	.set_buffer_size          = hvs_set_buffer_size,
887 	.set_min_buffer_size      = hvs_set_min_buffer_size,
888 	.set_max_buffer_size      = hvs_set_max_buffer_size,
889 	.get_buffer_size          = hvs_get_buffer_size,
890 	.get_min_buffer_size      = hvs_get_min_buffer_size,
891 	.get_max_buffer_size      = hvs_get_max_buffer_size,
892 };
893 
894 static int hvs_probe(struct hv_device *hdev,
895 		     const struct hv_vmbus_device_id *dev_id)
896 {
897 	struct vmbus_channel *chan = hdev->channel;
898 
899 	hvs_open_connection(chan);
900 
901 	/* Always return success to suppress the unnecessary error message
902 	 * in vmbus_probe(): on error the host will rescind the device in
903 	 * 30 seconds and we can do cleanup at that time in
904 	 * vmbus_onoffer_rescind().
905 	 */
906 	return 0;
907 }
908 
909 static int hvs_remove(struct hv_device *hdev)
910 {
911 	struct vmbus_channel *chan = hdev->channel;
912 
913 	vmbus_close(chan);
914 
915 	return 0;
916 }
917 
918 /* This isn't really used. See vmbus_match() and vmbus_probe() */
919 static const struct hv_vmbus_device_id id_table[] = {
920 	{},
921 };
922 
923 static struct hv_driver hvs_drv = {
924 	.name		= "hv_sock",
925 	.hvsock		= true,
926 	.id_table	= id_table,
927 	.probe		= hvs_probe,
928 	.remove		= hvs_remove,
929 };
930 
931 static int __init hvs_init(void)
932 {
933 	int ret;
934 
935 	if (vmbus_proto_version < VERSION_WIN10)
936 		return -ENODEV;
937 
938 	ret = vmbus_driver_register(&hvs_drv);
939 	if (ret != 0)
940 		return ret;
941 
942 	ret = vsock_core_init(&hvs_transport);
943 	if (ret) {
944 		vmbus_driver_unregister(&hvs_drv);
945 		return ret;
946 	}
947 
948 	return 0;
949 }
950 
951 static void __exit hvs_exit(void)
952 {
953 	vsock_core_exit();
954 	vmbus_driver_unregister(&hvs_drv);
955 }
956 
957 module_init(hvs_init);
958 module_exit(hvs_exit);
959 
960 MODULE_DESCRIPTION("Hyper-V Sockets");
961 MODULE_VERSION("1.0.0");
962 MODULE_LICENSE("GPL");
963 MODULE_ALIAS_NETPROTO(PF_VSOCK);
964