1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * Hyper-V transport for vsock
4  *
5  * Hyper-V Sockets supplies a byte-stream based communication mechanism
6  * between the host and the VM. This driver implements the necessary
7  * support in the VM by introducing the new vsock transport.
8  *
9  * Copyright (c) 2017, Microsoft Corporation.
10  */
11 #include <linux/module.h>
12 #include <linux/vmalloc.h>
13 #include <linux/hyperv.h>
14 #include <net/sock.h>
15 #include <net/af_vsock.h>
16 #include <asm/hyperv-tlfs.h>
17 
18 /* Older (VMBUS version 'VERSION_WIN10' or before) Windows hosts have some
19  * stricter requirements on the hv_sock ring buffer size of six 4K pages.
20  * hyperv-tlfs defines HV_HYP_PAGE_SIZE as 4K. Newer hosts don't have this
21  * limitation; but, keep the defaults the same for compat.
22  */
23 #define RINGBUFFER_HVS_RCV_SIZE (HV_HYP_PAGE_SIZE * 6)
24 #define RINGBUFFER_HVS_SND_SIZE (HV_HYP_PAGE_SIZE * 6)
25 #define RINGBUFFER_HVS_MAX_SIZE (HV_HYP_PAGE_SIZE * 64)
26 
27 /* The MTU is 16KB per the host side's design */
28 #define HVS_MTU_SIZE		(1024 * 16)
29 
30 /* How long to wait for graceful shutdown of a connection */
31 #define HVS_CLOSE_TIMEOUT (8 * HZ)
32 
33 struct vmpipe_proto_header {
34 	u32 pkt_type;
35 	u32 data_size;
36 };
37 
38 /* For recv, we use the VMBus in-place packet iterator APIs to directly copy
39  * data from the ringbuffer into the userspace buffer.
40  */
41 struct hvs_recv_buf {
42 	/* The header before the payload data */
43 	struct vmpipe_proto_header hdr;
44 
45 	/* The payload */
46 	u8 data[HVS_MTU_SIZE];
47 };
48 
49 /* We can send up to HVS_MTU_SIZE bytes of payload to the host, but let's use
50  * a smaller size, i.e. HVS_SEND_BUF_SIZE, to maximize concurrency between the
51  * guest and the host processing as one VMBUS packet is the smallest processing
52  * unit.
53  *
54  * Note: the buffer can be eliminated in the future when we add new VMBus
55  * ringbuffer APIs that allow us to directly copy data from userspace buffer
56  * to VMBus ringbuffer.
57  */
58 #define HVS_SEND_BUF_SIZE \
59 		(HV_HYP_PAGE_SIZE - sizeof(struct vmpipe_proto_header))
60 
61 struct hvs_send_buf {
62 	/* The header before the payload data */
63 	struct vmpipe_proto_header hdr;
64 
65 	/* The payload */
66 	u8 data[HVS_SEND_BUF_SIZE];
67 };
68 
69 #define HVS_HEADER_LEN	(sizeof(struct vmpacket_descriptor) + \
70 			 sizeof(struct vmpipe_proto_header))
71 
72 /* See 'prev_indices' in hv_ringbuffer_read(), hv_ringbuffer_write(), and
73  * __hv_pkt_iter_next().
74  */
75 #define VMBUS_PKT_TRAILER_SIZE	(sizeof(u64))
76 
77 #define HVS_PKT_LEN(payload_len)	(HVS_HEADER_LEN + \
78 					 ALIGN((payload_len), 8) + \
79 					 VMBUS_PKT_TRAILER_SIZE)
80 
81 union hvs_service_id {
82 	guid_t	srv_id;
83 
84 	struct {
85 		unsigned int svm_port;
86 		unsigned char b[sizeof(guid_t) - sizeof(unsigned int)];
87 	};
88 };
89 
90 /* Per-socket state (accessed via vsk->trans) */
91 struct hvsock {
92 	struct vsock_sock *vsk;
93 
94 	guid_t vm_srv_id;
95 	guid_t host_srv_id;
96 
97 	struct vmbus_channel *chan;
98 	struct vmpacket_descriptor *recv_desc;
99 
100 	/* The length of the payload not delivered to userland yet */
101 	u32 recv_data_len;
102 	/* The offset of the payload */
103 	u32 recv_data_off;
104 
105 	/* Have we sent the zero-length packet (FIN)? */
106 	bool fin_sent;
107 };
108 
109 /* In the VM, we support Hyper-V Sockets with AF_VSOCK, and the endpoint is
110  * <cid, port> (see struct sockaddr_vm). Note: cid is not really used here:
111  * when we write apps to connect to the host, we can only use VMADDR_CID_ANY
112  * or VMADDR_CID_HOST (both are equivalent) as the remote cid, and when we
113  * write apps to bind() & listen() in the VM, we can only use VMADDR_CID_ANY
114  * as the local cid.
115  *
116  * On the host, Hyper-V Sockets are supported by Winsock AF_HYPERV:
117  * https://docs.microsoft.com/en-us/virtualization/hyper-v-on-windows/user-
118  * guide/make-integration-service, and the endpoint is <VmID, ServiceId> with
119  * the below sockaddr:
120  *
121  * struct SOCKADDR_HV
122  * {
123  *    ADDRESS_FAMILY Family;
124  *    USHORT Reserved;
125  *    GUID VmId;
126  *    GUID ServiceId;
127  * };
128  * Note: VmID is not used by Linux VM and actually it isn't transmitted via
129  * VMBus, because here it's obvious the host and the VM can easily identify
130  * each other. Though the VmID is useful on the host, especially in the case
131  * of Windows container, Linux VM doesn't need it at all.
132  *
133  * To make use of the AF_VSOCK infrastructure in Linux VM, we have to limit
134  * the available GUID space of SOCKADDR_HV so that we can create a mapping
135  * between AF_VSOCK port and SOCKADDR_HV Service GUID. The rule of writing
136  * Hyper-V Sockets apps on the host and in Linux VM is:
137  *
138  ****************************************************************************
139  * The only valid Service GUIDs, from the perspectives of both the host and *
140  * Linux VM, that can be connected by the other end, must conform to this   *
141  * format: <port>-facb-11e6-bd58-64006a7986d3.                              *
142  ****************************************************************************
143  *
144  * When we write apps on the host to connect(), the GUID ServiceID is used.
145  * When we write apps in Linux VM to connect(), we only need to specify the
146  * port and the driver will form the GUID and use that to request the host.
147  *
148  */
149 
150 /* 00000000-facb-11e6-bd58-64006a7986d3 */
151 static const guid_t srv_id_template =
152 	GUID_INIT(0x00000000, 0xfacb, 0x11e6, 0xbd, 0x58,
153 		  0x64, 0x00, 0x6a, 0x79, 0x86, 0xd3);
154 
155 static bool hvs_check_transport(struct vsock_sock *vsk);
156 
157 static bool is_valid_srv_id(const guid_t *id)
158 {
159 	return !memcmp(&id->b[4], &srv_id_template.b[4], sizeof(guid_t) - 4);
160 }
161 
162 static unsigned int get_port_by_srv_id(const guid_t *svr_id)
163 {
164 	return *((unsigned int *)svr_id);
165 }
166 
167 static void hvs_addr_init(struct sockaddr_vm *addr, const guid_t *svr_id)
168 {
169 	unsigned int port = get_port_by_srv_id(svr_id);
170 
171 	vsock_addr_init(addr, VMADDR_CID_ANY, port);
172 }
173 
174 static void hvs_set_channel_pending_send_size(struct vmbus_channel *chan)
175 {
176 	set_channel_pending_send_size(chan,
177 				      HVS_PKT_LEN(HVS_SEND_BUF_SIZE));
178 
179 	virt_mb();
180 }
181 
182 static bool hvs_channel_readable(struct vmbus_channel *chan)
183 {
184 	u32 readable = hv_get_bytes_to_read(&chan->inbound);
185 
186 	/* 0-size payload means FIN */
187 	return readable >= HVS_PKT_LEN(0);
188 }
189 
190 static int hvs_channel_readable_payload(struct vmbus_channel *chan)
191 {
192 	u32 readable = hv_get_bytes_to_read(&chan->inbound);
193 
194 	if (readable > HVS_PKT_LEN(0)) {
195 		/* At least we have 1 byte to read. We don't need to return
196 		 * the exact readable bytes: see vsock_stream_recvmsg() ->
197 		 * vsock_stream_has_data().
198 		 */
199 		return 1;
200 	}
201 
202 	if (readable == HVS_PKT_LEN(0)) {
203 		/* 0-size payload means FIN */
204 		return 0;
205 	}
206 
207 	/* No payload or FIN */
208 	return -1;
209 }
210 
211 static size_t hvs_channel_writable_bytes(struct vmbus_channel *chan)
212 {
213 	u32 writeable = hv_get_bytes_to_write(&chan->outbound);
214 	size_t ret;
215 
216 	/* The ringbuffer mustn't be 100% full, and we should reserve a
217 	 * zero-length-payload packet for the FIN: see hv_ringbuffer_write()
218 	 * and hvs_shutdown().
219 	 */
220 	if (writeable <= HVS_PKT_LEN(1) + HVS_PKT_LEN(0))
221 		return 0;
222 
223 	ret = writeable - HVS_PKT_LEN(1) - HVS_PKT_LEN(0);
224 
225 	return round_down(ret, 8);
226 }
227 
228 static int hvs_send_data(struct vmbus_channel *chan,
229 			 struct hvs_send_buf *send_buf, size_t to_write)
230 {
231 	send_buf->hdr.pkt_type = 1;
232 	send_buf->hdr.data_size = to_write;
233 	return vmbus_sendpacket(chan, &send_buf->hdr,
234 				sizeof(send_buf->hdr) + to_write,
235 				0, VM_PKT_DATA_INBAND, 0);
236 }
237 
238 static void hvs_channel_cb(void *ctx)
239 {
240 	struct sock *sk = (struct sock *)ctx;
241 	struct vsock_sock *vsk = vsock_sk(sk);
242 	struct hvsock *hvs = vsk->trans;
243 	struct vmbus_channel *chan = hvs->chan;
244 
245 	if (hvs_channel_readable(chan))
246 		sk->sk_data_ready(sk);
247 
248 	if (hv_get_bytes_to_write(&chan->outbound) > 0)
249 		sk->sk_write_space(sk);
250 }
251 
252 static void hvs_do_close_lock_held(struct vsock_sock *vsk,
253 				   bool cancel_timeout)
254 {
255 	struct sock *sk = sk_vsock(vsk);
256 
257 	sock_set_flag(sk, SOCK_DONE);
258 	vsk->peer_shutdown = SHUTDOWN_MASK;
259 	if (vsock_stream_has_data(vsk) <= 0)
260 		sk->sk_state = TCP_CLOSING;
261 	sk->sk_state_change(sk);
262 	if (vsk->close_work_scheduled &&
263 	    (!cancel_timeout || cancel_delayed_work(&vsk->close_work))) {
264 		vsk->close_work_scheduled = false;
265 		vsock_remove_sock(vsk);
266 
267 		/* Release the reference taken while scheduling the timeout */
268 		sock_put(sk);
269 	}
270 }
271 
272 static void hvs_close_connection(struct vmbus_channel *chan)
273 {
274 	struct sock *sk = get_per_channel_state(chan);
275 
276 	lock_sock(sk);
277 	hvs_do_close_lock_held(vsock_sk(sk), true);
278 	release_sock(sk);
279 
280 	/* Release the refcnt for the channel that's opened in
281 	 * hvs_open_connection().
282 	 */
283 	sock_put(sk);
284 }
285 
286 static void hvs_open_connection(struct vmbus_channel *chan)
287 {
288 	guid_t *if_instance, *if_type;
289 	unsigned char conn_from_host;
290 
291 	struct sockaddr_vm addr;
292 	struct sock *sk, *new = NULL;
293 	struct vsock_sock *vnew = NULL;
294 	struct hvsock *hvs = NULL;
295 	struct hvsock *hvs_new = NULL;
296 	int rcvbuf;
297 	int ret;
298 	int sndbuf;
299 
300 	if_type = &chan->offermsg.offer.if_type;
301 	if_instance = &chan->offermsg.offer.if_instance;
302 	conn_from_host = chan->offermsg.offer.u.pipe.user_def[0];
303 	if (!is_valid_srv_id(if_type))
304 		return;
305 
306 	hvs_addr_init(&addr, conn_from_host ? if_type : if_instance);
307 	sk = vsock_find_bound_socket(&addr);
308 	if (!sk)
309 		return;
310 
311 	lock_sock(sk);
312 	if ((conn_from_host && sk->sk_state != TCP_LISTEN) ||
313 	    (!conn_from_host && sk->sk_state != TCP_SYN_SENT))
314 		goto out;
315 
316 	if (conn_from_host) {
317 		if (sk->sk_ack_backlog >= sk->sk_max_ack_backlog)
318 			goto out;
319 
320 		new = vsock_create_connected(sk);
321 		if (!new)
322 			goto out;
323 
324 		new->sk_state = TCP_SYN_SENT;
325 		vnew = vsock_sk(new);
326 
327 		hvs_addr_init(&vnew->local_addr, if_type);
328 
329 		/* Remote peer is always the host */
330 		vsock_addr_init(&vnew->remote_addr,
331 				VMADDR_CID_HOST, VMADDR_PORT_ANY);
332 		vnew->remote_addr.svm_port = get_port_by_srv_id(if_instance);
333 		ret = vsock_assign_transport(vnew, vsock_sk(sk));
334 		/* Transport assigned (looking at remote_addr) must be the
335 		 * same where we received the request.
336 		 */
337 		if (ret || !hvs_check_transport(vnew)) {
338 			sock_put(new);
339 			goto out;
340 		}
341 		hvs_new = vnew->trans;
342 		hvs_new->chan = chan;
343 	} else {
344 		hvs = vsock_sk(sk)->trans;
345 		hvs->chan = chan;
346 	}
347 
348 	set_channel_read_mode(chan, HV_CALL_DIRECT);
349 
350 	/* Use the socket buffer sizes as hints for the VMBUS ring size. For
351 	 * server side sockets, 'sk' is the parent socket and thus, this will
352 	 * allow the child sockets to inherit the size from the parent. Keep
353 	 * the mins to the default value and align to page size as per VMBUS
354 	 * requirements.
355 	 * For the max, the socket core library will limit the socket buffer
356 	 * size that can be set by the user, but, since currently, the hv_sock
357 	 * VMBUS ring buffer is physically contiguous allocation, restrict it
358 	 * further.
359 	 * Older versions of hv_sock host side code cannot handle bigger VMBUS
360 	 * ring buffer size. Use the version number to limit the change to newer
361 	 * versions.
362 	 */
363 	if (vmbus_proto_version < VERSION_WIN10_V5) {
364 		sndbuf = RINGBUFFER_HVS_SND_SIZE;
365 		rcvbuf = RINGBUFFER_HVS_RCV_SIZE;
366 	} else {
367 		sndbuf = max_t(int, sk->sk_sndbuf, RINGBUFFER_HVS_SND_SIZE);
368 		sndbuf = min_t(int, sndbuf, RINGBUFFER_HVS_MAX_SIZE);
369 		sndbuf = ALIGN(sndbuf, HV_HYP_PAGE_SIZE);
370 		rcvbuf = max_t(int, sk->sk_rcvbuf, RINGBUFFER_HVS_RCV_SIZE);
371 		rcvbuf = min_t(int, rcvbuf, RINGBUFFER_HVS_MAX_SIZE);
372 		rcvbuf = ALIGN(rcvbuf, HV_HYP_PAGE_SIZE);
373 	}
374 
375 	ret = vmbus_open(chan, sndbuf, rcvbuf, NULL, 0, hvs_channel_cb,
376 			 conn_from_host ? new : sk);
377 	if (ret != 0) {
378 		if (conn_from_host) {
379 			hvs_new->chan = NULL;
380 			sock_put(new);
381 		} else {
382 			hvs->chan = NULL;
383 		}
384 		goto out;
385 	}
386 
387 	set_per_channel_state(chan, conn_from_host ? new : sk);
388 
389 	/* This reference will be dropped by hvs_close_connection(). */
390 	sock_hold(conn_from_host ? new : sk);
391 	vmbus_set_chn_rescind_callback(chan, hvs_close_connection);
392 
393 	/* Set the pending send size to max packet size to always get
394 	 * notifications from the host when there is enough writable space.
395 	 * The host is optimized to send notifications only when the pending
396 	 * size boundary is crossed, and not always.
397 	 */
398 	hvs_set_channel_pending_send_size(chan);
399 
400 	if (conn_from_host) {
401 		new->sk_state = TCP_ESTABLISHED;
402 		sk_acceptq_added(sk);
403 
404 		hvs_new->vm_srv_id = *if_type;
405 		hvs_new->host_srv_id = *if_instance;
406 
407 		vsock_insert_connected(vnew);
408 
409 		vsock_enqueue_accept(sk, new);
410 	} else {
411 		sk->sk_state = TCP_ESTABLISHED;
412 		sk->sk_socket->state = SS_CONNECTED;
413 
414 		vsock_insert_connected(vsock_sk(sk));
415 	}
416 
417 	sk->sk_state_change(sk);
418 
419 out:
420 	/* Release refcnt obtained when we called vsock_find_bound_socket() */
421 	sock_put(sk);
422 
423 	release_sock(sk);
424 }
425 
426 static u32 hvs_get_local_cid(void)
427 {
428 	return VMADDR_CID_ANY;
429 }
430 
431 static int hvs_sock_init(struct vsock_sock *vsk, struct vsock_sock *psk)
432 {
433 	struct hvsock *hvs;
434 	struct sock *sk = sk_vsock(vsk);
435 
436 	hvs = kzalloc(sizeof(*hvs), GFP_KERNEL);
437 	if (!hvs)
438 		return -ENOMEM;
439 
440 	vsk->trans = hvs;
441 	hvs->vsk = vsk;
442 	sk->sk_sndbuf = RINGBUFFER_HVS_SND_SIZE;
443 	sk->sk_rcvbuf = RINGBUFFER_HVS_RCV_SIZE;
444 	return 0;
445 }
446 
447 static int hvs_connect(struct vsock_sock *vsk)
448 {
449 	union hvs_service_id vm, host;
450 	struct hvsock *h = vsk->trans;
451 
452 	vm.srv_id = srv_id_template;
453 	vm.svm_port = vsk->local_addr.svm_port;
454 	h->vm_srv_id = vm.srv_id;
455 
456 	host.srv_id = srv_id_template;
457 	host.svm_port = vsk->remote_addr.svm_port;
458 	h->host_srv_id = host.srv_id;
459 
460 	return vmbus_send_tl_connect_request(&h->vm_srv_id, &h->host_srv_id);
461 }
462 
463 static void hvs_shutdown_lock_held(struct hvsock *hvs, int mode)
464 {
465 	struct vmpipe_proto_header hdr;
466 
467 	if (hvs->fin_sent || !hvs->chan)
468 		return;
469 
470 	/* It can't fail: see hvs_channel_writable_bytes(). */
471 	(void)hvs_send_data(hvs->chan, (struct hvs_send_buf *)&hdr, 0);
472 	hvs->fin_sent = true;
473 }
474 
475 static int hvs_shutdown(struct vsock_sock *vsk, int mode)
476 {
477 	if (!(mode & SEND_SHUTDOWN))
478 		return 0;
479 
480 	hvs_shutdown_lock_held(vsk->trans, mode);
481 	return 0;
482 }
483 
484 static void hvs_close_timeout(struct work_struct *work)
485 {
486 	struct vsock_sock *vsk =
487 		container_of(work, struct vsock_sock, close_work.work);
488 	struct sock *sk = sk_vsock(vsk);
489 
490 	sock_hold(sk);
491 	lock_sock(sk);
492 	if (!sock_flag(sk, SOCK_DONE))
493 		hvs_do_close_lock_held(vsk, false);
494 
495 	vsk->close_work_scheduled = false;
496 	release_sock(sk);
497 	sock_put(sk);
498 }
499 
500 /* Returns true, if it is safe to remove socket; false otherwise */
501 static bool hvs_close_lock_held(struct vsock_sock *vsk)
502 {
503 	struct sock *sk = sk_vsock(vsk);
504 
505 	if (!(sk->sk_state == TCP_ESTABLISHED ||
506 	      sk->sk_state == TCP_CLOSING))
507 		return true;
508 
509 	if ((sk->sk_shutdown & SHUTDOWN_MASK) != SHUTDOWN_MASK)
510 		hvs_shutdown_lock_held(vsk->trans, SHUTDOWN_MASK);
511 
512 	if (sock_flag(sk, SOCK_DONE))
513 		return true;
514 
515 	/* This reference will be dropped by the delayed close routine */
516 	sock_hold(sk);
517 	INIT_DELAYED_WORK(&vsk->close_work, hvs_close_timeout);
518 	vsk->close_work_scheduled = true;
519 	schedule_delayed_work(&vsk->close_work, HVS_CLOSE_TIMEOUT);
520 	return false;
521 }
522 
523 static void hvs_release(struct vsock_sock *vsk)
524 {
525 	bool remove_sock;
526 
527 	remove_sock = hvs_close_lock_held(vsk);
528 	if (remove_sock)
529 		vsock_remove_sock(vsk);
530 }
531 
532 static void hvs_destruct(struct vsock_sock *vsk)
533 {
534 	struct hvsock *hvs = vsk->trans;
535 	struct vmbus_channel *chan = hvs->chan;
536 
537 	if (chan)
538 		vmbus_hvsock_device_unregister(chan);
539 
540 	kfree(hvs);
541 }
542 
543 static int hvs_dgram_bind(struct vsock_sock *vsk, struct sockaddr_vm *addr)
544 {
545 	return -EOPNOTSUPP;
546 }
547 
548 static int hvs_dgram_dequeue(struct vsock_sock *vsk, struct msghdr *msg,
549 			     size_t len, int flags)
550 {
551 	return -EOPNOTSUPP;
552 }
553 
554 static int hvs_dgram_enqueue(struct vsock_sock *vsk,
555 			     struct sockaddr_vm *remote, struct msghdr *msg,
556 			     size_t dgram_len)
557 {
558 	return -EOPNOTSUPP;
559 }
560 
561 static bool hvs_dgram_allow(u32 cid, u32 port)
562 {
563 	return false;
564 }
565 
566 static int hvs_update_recv_data(struct hvsock *hvs)
567 {
568 	struct hvs_recv_buf *recv_buf;
569 	u32 payload_len;
570 
571 	recv_buf = (struct hvs_recv_buf *)(hvs->recv_desc + 1);
572 	payload_len = recv_buf->hdr.data_size;
573 
574 	if (payload_len > HVS_MTU_SIZE)
575 		return -EIO;
576 
577 	if (payload_len == 0)
578 		hvs->vsk->peer_shutdown |= SEND_SHUTDOWN;
579 
580 	hvs->recv_data_len = payload_len;
581 	hvs->recv_data_off = 0;
582 
583 	return 0;
584 }
585 
586 static ssize_t hvs_stream_dequeue(struct vsock_sock *vsk, struct msghdr *msg,
587 				  size_t len, int flags)
588 {
589 	struct hvsock *hvs = vsk->trans;
590 	bool need_refill = !hvs->recv_desc;
591 	struct hvs_recv_buf *recv_buf;
592 	u32 to_read;
593 	int ret;
594 
595 	if (flags & MSG_PEEK)
596 		return -EOPNOTSUPP;
597 
598 	if (need_refill) {
599 		hvs->recv_desc = hv_pkt_iter_first_raw(hvs->chan);
600 		ret = hvs_update_recv_data(hvs);
601 		if (ret)
602 			return ret;
603 	}
604 
605 	recv_buf = (struct hvs_recv_buf *)(hvs->recv_desc + 1);
606 	to_read = min_t(u32, len, hvs->recv_data_len);
607 	ret = memcpy_to_msg(msg, recv_buf->data + hvs->recv_data_off, to_read);
608 	if (ret != 0)
609 		return ret;
610 
611 	hvs->recv_data_len -= to_read;
612 	if (hvs->recv_data_len == 0) {
613 		hvs->recv_desc = hv_pkt_iter_next_raw(hvs->chan, hvs->recv_desc);
614 		if (hvs->recv_desc) {
615 			ret = hvs_update_recv_data(hvs);
616 			if (ret)
617 				return ret;
618 		}
619 	} else {
620 		hvs->recv_data_off += to_read;
621 	}
622 
623 	return to_read;
624 }
625 
626 static ssize_t hvs_stream_enqueue(struct vsock_sock *vsk, struct msghdr *msg,
627 				  size_t len)
628 {
629 	struct hvsock *hvs = vsk->trans;
630 	struct vmbus_channel *chan = hvs->chan;
631 	struct hvs_send_buf *send_buf;
632 	ssize_t to_write, max_writable;
633 	ssize_t ret = 0;
634 	ssize_t bytes_written = 0;
635 
636 	BUILD_BUG_ON(sizeof(*send_buf) != HV_HYP_PAGE_SIZE);
637 
638 	send_buf = kmalloc(sizeof(*send_buf), GFP_KERNEL);
639 	if (!send_buf)
640 		return -ENOMEM;
641 
642 	/* Reader(s) could be draining data from the channel as we write.
643 	 * Maximize bandwidth, by iterating until the channel is found to be
644 	 * full.
645 	 */
646 	while (len) {
647 		max_writable = hvs_channel_writable_bytes(chan);
648 		if (!max_writable)
649 			break;
650 		to_write = min_t(ssize_t, len, max_writable);
651 		to_write = min_t(ssize_t, to_write, HVS_SEND_BUF_SIZE);
652 		/* memcpy_from_msg is safe for loop as it advances the offsets
653 		 * within the message iterator.
654 		 */
655 		ret = memcpy_from_msg(send_buf->data, msg, to_write);
656 		if (ret < 0)
657 			goto out;
658 
659 		ret = hvs_send_data(hvs->chan, send_buf, to_write);
660 		if (ret < 0)
661 			goto out;
662 
663 		bytes_written += to_write;
664 		len -= to_write;
665 	}
666 out:
667 	/* If any data has been sent, return that */
668 	if (bytes_written)
669 		ret = bytes_written;
670 	kfree(send_buf);
671 	return ret;
672 }
673 
674 static s64 hvs_stream_has_data(struct vsock_sock *vsk)
675 {
676 	struct hvsock *hvs = vsk->trans;
677 	s64 ret;
678 
679 	if (hvs->recv_data_len > 0)
680 		return 1;
681 
682 	switch (hvs_channel_readable_payload(hvs->chan)) {
683 	case 1:
684 		ret = 1;
685 		break;
686 	case 0:
687 		vsk->peer_shutdown |= SEND_SHUTDOWN;
688 		ret = 0;
689 		break;
690 	default: /* -1 */
691 		ret = 0;
692 		break;
693 	}
694 
695 	return ret;
696 }
697 
698 static s64 hvs_stream_has_space(struct vsock_sock *vsk)
699 {
700 	struct hvsock *hvs = vsk->trans;
701 
702 	return hvs_channel_writable_bytes(hvs->chan);
703 }
704 
705 static u64 hvs_stream_rcvhiwat(struct vsock_sock *vsk)
706 {
707 	return HVS_MTU_SIZE + 1;
708 }
709 
710 static bool hvs_stream_is_active(struct vsock_sock *vsk)
711 {
712 	struct hvsock *hvs = vsk->trans;
713 
714 	return hvs->chan != NULL;
715 }
716 
717 static bool hvs_stream_allow(u32 cid, u32 port)
718 {
719 	if (cid == VMADDR_CID_HOST)
720 		return true;
721 
722 	return false;
723 }
724 
725 static
726 int hvs_notify_poll_in(struct vsock_sock *vsk, size_t target, bool *readable)
727 {
728 	struct hvsock *hvs = vsk->trans;
729 
730 	*readable = hvs_channel_readable(hvs->chan);
731 	return 0;
732 }
733 
734 static
735 int hvs_notify_poll_out(struct vsock_sock *vsk, size_t target, bool *writable)
736 {
737 	*writable = hvs_stream_has_space(vsk) > 0;
738 
739 	return 0;
740 }
741 
742 static
743 int hvs_notify_recv_init(struct vsock_sock *vsk, size_t target,
744 			 struct vsock_transport_recv_notify_data *d)
745 {
746 	return 0;
747 }
748 
749 static
750 int hvs_notify_recv_pre_block(struct vsock_sock *vsk, size_t target,
751 			      struct vsock_transport_recv_notify_data *d)
752 {
753 	return 0;
754 }
755 
756 static
757 int hvs_notify_recv_pre_dequeue(struct vsock_sock *vsk, size_t target,
758 				struct vsock_transport_recv_notify_data *d)
759 {
760 	return 0;
761 }
762 
763 static
764 int hvs_notify_recv_post_dequeue(struct vsock_sock *vsk, size_t target,
765 				 ssize_t copied, bool data_read,
766 				 struct vsock_transport_recv_notify_data *d)
767 {
768 	return 0;
769 }
770 
771 static
772 int hvs_notify_send_init(struct vsock_sock *vsk,
773 			 struct vsock_transport_send_notify_data *d)
774 {
775 	return 0;
776 }
777 
778 static
779 int hvs_notify_send_pre_block(struct vsock_sock *vsk,
780 			      struct vsock_transport_send_notify_data *d)
781 {
782 	return 0;
783 }
784 
785 static
786 int hvs_notify_send_pre_enqueue(struct vsock_sock *vsk,
787 				struct vsock_transport_send_notify_data *d)
788 {
789 	return 0;
790 }
791 
792 static
793 int hvs_notify_send_post_enqueue(struct vsock_sock *vsk, ssize_t written,
794 				 struct vsock_transport_send_notify_data *d)
795 {
796 	return 0;
797 }
798 
799 static struct vsock_transport hvs_transport = {
800 	.module                   = THIS_MODULE,
801 
802 	.get_local_cid            = hvs_get_local_cid,
803 
804 	.init                     = hvs_sock_init,
805 	.destruct                 = hvs_destruct,
806 	.release                  = hvs_release,
807 	.connect                  = hvs_connect,
808 	.shutdown                 = hvs_shutdown,
809 
810 	.dgram_bind               = hvs_dgram_bind,
811 	.dgram_dequeue            = hvs_dgram_dequeue,
812 	.dgram_enqueue            = hvs_dgram_enqueue,
813 	.dgram_allow              = hvs_dgram_allow,
814 
815 	.stream_dequeue           = hvs_stream_dequeue,
816 	.stream_enqueue           = hvs_stream_enqueue,
817 	.stream_has_data          = hvs_stream_has_data,
818 	.stream_has_space         = hvs_stream_has_space,
819 	.stream_rcvhiwat          = hvs_stream_rcvhiwat,
820 	.stream_is_active         = hvs_stream_is_active,
821 	.stream_allow             = hvs_stream_allow,
822 
823 	.notify_poll_in           = hvs_notify_poll_in,
824 	.notify_poll_out          = hvs_notify_poll_out,
825 	.notify_recv_init         = hvs_notify_recv_init,
826 	.notify_recv_pre_block    = hvs_notify_recv_pre_block,
827 	.notify_recv_pre_dequeue  = hvs_notify_recv_pre_dequeue,
828 	.notify_recv_post_dequeue = hvs_notify_recv_post_dequeue,
829 	.notify_send_init         = hvs_notify_send_init,
830 	.notify_send_pre_block    = hvs_notify_send_pre_block,
831 	.notify_send_pre_enqueue  = hvs_notify_send_pre_enqueue,
832 	.notify_send_post_enqueue = hvs_notify_send_post_enqueue,
833 
834 };
835 
836 static bool hvs_check_transport(struct vsock_sock *vsk)
837 {
838 	return vsk->transport == &hvs_transport;
839 }
840 
841 static int hvs_probe(struct hv_device *hdev,
842 		     const struct hv_vmbus_device_id *dev_id)
843 {
844 	struct vmbus_channel *chan = hdev->channel;
845 
846 	hvs_open_connection(chan);
847 
848 	/* Always return success to suppress the unnecessary error message
849 	 * in vmbus_probe(): on error the host will rescind the device in
850 	 * 30 seconds and we can do cleanup at that time in
851 	 * vmbus_onoffer_rescind().
852 	 */
853 	return 0;
854 }
855 
856 static int hvs_remove(struct hv_device *hdev)
857 {
858 	struct vmbus_channel *chan = hdev->channel;
859 
860 	vmbus_close(chan);
861 
862 	return 0;
863 }
864 
865 /* hv_sock connections can not persist across hibernation, and all the hv_sock
866  * channels are forced to be rescinded before hibernation: see
867  * vmbus_bus_suspend(). Here the dummy hvs_suspend() and hvs_resume()
868  * are only needed because hibernation requires that every vmbus device's
869  * driver should have a .suspend and .resume callback: see vmbus_suspend().
870  */
871 static int hvs_suspend(struct hv_device *hv_dev)
872 {
873 	/* Dummy */
874 	return 0;
875 }
876 
877 static int hvs_resume(struct hv_device *dev)
878 {
879 	/* Dummy */
880 	return 0;
881 }
882 
883 /* This isn't really used. See vmbus_match() and vmbus_probe() */
884 static const struct hv_vmbus_device_id id_table[] = {
885 	{},
886 };
887 
888 static struct hv_driver hvs_drv = {
889 	.name		= "hv_sock",
890 	.hvsock		= true,
891 	.id_table	= id_table,
892 	.probe		= hvs_probe,
893 	.remove		= hvs_remove,
894 	.suspend	= hvs_suspend,
895 	.resume		= hvs_resume,
896 };
897 
898 static int __init hvs_init(void)
899 {
900 	int ret;
901 
902 	if (vmbus_proto_version < VERSION_WIN10)
903 		return -ENODEV;
904 
905 	ret = vmbus_driver_register(&hvs_drv);
906 	if (ret != 0)
907 		return ret;
908 
909 	ret = vsock_core_register(&hvs_transport, VSOCK_TRANSPORT_F_G2H);
910 	if (ret) {
911 		vmbus_driver_unregister(&hvs_drv);
912 		return ret;
913 	}
914 
915 	return 0;
916 }
917 
918 static void __exit hvs_exit(void)
919 {
920 	vsock_core_unregister(&hvs_transport);
921 	vmbus_driver_unregister(&hvs_drv);
922 }
923 
924 module_init(hvs_init);
925 module_exit(hvs_exit);
926 
927 MODULE_DESCRIPTION("Hyper-V Sockets");
928 MODULE_VERSION("1.0.0");
929 MODULE_LICENSE("GPL");
930 MODULE_ALIAS_NETPROTO(PF_VSOCK);
931