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 /* Upper bound on the size of a VMbus packet for hv_sock */
82 #define HVS_MAX_PKT_SIZE	HVS_PKT_LEN(HVS_MTU_SIZE)
83 
84 union hvs_service_id {
85 	guid_t	srv_id;
86 
87 	struct {
88 		unsigned int svm_port;
89 		unsigned char b[sizeof(guid_t) - sizeof(unsigned int)];
90 	};
91 };
92 
93 /* Per-socket state (accessed via vsk->trans) */
94 struct hvsock {
95 	struct vsock_sock *vsk;
96 
97 	guid_t vm_srv_id;
98 	guid_t 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.                              *
145  ****************************************************************************
146  *
147  * When we write apps on the host to connect(), the GUID ServiceID is used.
148  * When we write apps in Linux VM to connect(), we only need to specify the
149  * port and the driver will form the GUID and use that to request the host.
150  *
151  */
152 
153 /* 00000000-facb-11e6-bd58-64006a7986d3 */
154 static const guid_t srv_id_template =
155 	GUID_INIT(0x00000000, 0xfacb, 0x11e6, 0xbd, 0x58,
156 		  0x64, 0x00, 0x6a, 0x79, 0x86, 0xd3);
157 
158 static bool hvs_check_transport(struct vsock_sock *vsk);
159 
160 static bool is_valid_srv_id(const guid_t *id)
161 {
162 	return !memcmp(&id->b[4], &srv_id_template.b[4], sizeof(guid_t) - 4);
163 }
164 
165 static unsigned int get_port_by_srv_id(const guid_t *svr_id)
166 {
167 	return *((unsigned int *)svr_id);
168 }
169 
170 static void hvs_addr_init(struct sockaddr_vm *addr, const guid_t *svr_id)
171 {
172 	unsigned int port = get_port_by_srv_id(svr_id);
173 
174 	vsock_addr_init(addr, VMADDR_CID_ANY, port);
175 }
176 
177 static void hvs_set_channel_pending_send_size(struct vmbus_channel *chan)
178 {
179 	set_channel_pending_send_size(chan,
180 				      HVS_PKT_LEN(HVS_SEND_BUF_SIZE));
181 
182 	virt_mb();
183 }
184 
185 static bool hvs_channel_readable(struct vmbus_channel *chan)
186 {
187 	u32 readable = hv_get_bytes_to_read(&chan->inbound);
188 
189 	/* 0-size payload means FIN */
190 	return readable >= HVS_PKT_LEN(0);
191 }
192 
193 static int hvs_channel_readable_payload(struct vmbus_channel *chan)
194 {
195 	u32 readable = hv_get_bytes_to_read(&chan->inbound);
196 
197 	if (readable > HVS_PKT_LEN(0)) {
198 		/* At least we have 1 byte to read. We don't need to return
199 		 * the exact readable bytes: see vsock_stream_recvmsg() ->
200 		 * vsock_stream_has_data().
201 		 */
202 		return 1;
203 	}
204 
205 	if (readable == HVS_PKT_LEN(0)) {
206 		/* 0-size payload means FIN */
207 		return 0;
208 	}
209 
210 	/* No payload or FIN */
211 	return -1;
212 }
213 
214 static size_t hvs_channel_writable_bytes(struct vmbus_channel *chan)
215 {
216 	u32 writeable = hv_get_bytes_to_write(&chan->outbound);
217 	size_t ret;
218 
219 	/* The ringbuffer mustn't be 100% full, and we should reserve a
220 	 * zero-length-payload packet for the FIN: see hv_ringbuffer_write()
221 	 * and hvs_shutdown().
222 	 */
223 	if (writeable <= HVS_PKT_LEN(1) + HVS_PKT_LEN(0))
224 		return 0;
225 
226 	ret = writeable - HVS_PKT_LEN(1) - HVS_PKT_LEN(0);
227 
228 	return round_down(ret, 8);
229 }
230 
231 static int __hvs_send_data(struct vmbus_channel *chan,
232 			   struct vmpipe_proto_header *hdr,
233 			   size_t to_write)
234 {
235 	hdr->pkt_type = 1;
236 	hdr->data_size = to_write;
237 	return vmbus_sendpacket(chan, hdr, sizeof(*hdr) + to_write,
238 				0, VM_PKT_DATA_INBAND, 0);
239 }
240 
241 static int hvs_send_data(struct vmbus_channel *chan,
242 			 struct hvs_send_buf *send_buf, size_t to_write)
243 {
244 	return __hvs_send_data(chan, &send_buf->hdr, to_write);
245 }
246 
247 static void hvs_channel_cb(void *ctx)
248 {
249 	struct sock *sk = (struct sock *)ctx;
250 	struct vsock_sock *vsk = vsock_sk(sk);
251 	struct hvsock *hvs = vsk->trans;
252 	struct vmbus_channel *chan = hvs->chan;
253 
254 	if (hvs_channel_readable(chan))
255 		sk->sk_data_ready(sk);
256 
257 	if (hv_get_bytes_to_write(&chan->outbound) > 0)
258 		sk->sk_write_space(sk);
259 }
260 
261 static void hvs_do_close_lock_held(struct vsock_sock *vsk,
262 				   bool cancel_timeout)
263 {
264 	struct sock *sk = sk_vsock(vsk);
265 
266 	sock_set_flag(sk, SOCK_DONE);
267 	vsk->peer_shutdown = SHUTDOWN_MASK;
268 	if (vsock_stream_has_data(vsk) <= 0)
269 		sk->sk_state = TCP_CLOSING;
270 	sk->sk_state_change(sk);
271 	if (vsk->close_work_scheduled &&
272 	    (!cancel_timeout || cancel_delayed_work(&vsk->close_work))) {
273 		vsk->close_work_scheduled = false;
274 		vsock_remove_sock(vsk);
275 
276 		/* Release the reference taken while scheduling the timeout */
277 		sock_put(sk);
278 	}
279 }
280 
281 static void hvs_close_connection(struct vmbus_channel *chan)
282 {
283 	struct sock *sk = get_per_channel_state(chan);
284 
285 	lock_sock(sk);
286 	hvs_do_close_lock_held(vsock_sk(sk), true);
287 	release_sock(sk);
288 
289 	/* Release the refcnt for the channel that's opened in
290 	 * hvs_open_connection().
291 	 */
292 	sock_put(sk);
293 }
294 
295 static void hvs_open_connection(struct vmbus_channel *chan)
296 {
297 	guid_t *if_instance, *if_type;
298 	unsigned char conn_from_host;
299 
300 	struct sockaddr_vm addr;
301 	struct sock *sk, *new = NULL;
302 	struct vsock_sock *vnew = NULL;
303 	struct hvsock *hvs = NULL;
304 	struct hvsock *hvs_new = NULL;
305 	int rcvbuf;
306 	int ret;
307 	int sndbuf;
308 
309 	if_type = &chan->offermsg.offer.if_type;
310 	if_instance = &chan->offermsg.offer.if_instance;
311 	conn_from_host = chan->offermsg.offer.u.pipe.user_def[0];
312 	if (!is_valid_srv_id(if_type))
313 		return;
314 
315 	hvs_addr_init(&addr, conn_from_host ? if_type : if_instance);
316 	sk = vsock_find_bound_socket(&addr);
317 	if (!sk)
318 		return;
319 
320 	lock_sock(sk);
321 	if ((conn_from_host && sk->sk_state != TCP_LISTEN) ||
322 	    (!conn_from_host && sk->sk_state != TCP_SYN_SENT))
323 		goto out;
324 
325 	if (conn_from_host) {
326 		if (sk->sk_ack_backlog >= sk->sk_max_ack_backlog)
327 			goto out;
328 
329 		new = vsock_create_connected(sk);
330 		if (!new)
331 			goto out;
332 
333 		new->sk_state = TCP_SYN_SENT;
334 		vnew = vsock_sk(new);
335 
336 		hvs_addr_init(&vnew->local_addr, if_type);
337 
338 		/* Remote peer is always the host */
339 		vsock_addr_init(&vnew->remote_addr,
340 				VMADDR_CID_HOST, VMADDR_PORT_ANY);
341 		vnew->remote_addr.svm_port = get_port_by_srv_id(if_instance);
342 		ret = vsock_assign_transport(vnew, vsock_sk(sk));
343 		/* Transport assigned (looking at remote_addr) must be the
344 		 * same where we received the request.
345 		 */
346 		if (ret || !hvs_check_transport(vnew)) {
347 			sock_put(new);
348 			goto out;
349 		}
350 		hvs_new = vnew->trans;
351 		hvs_new->chan = chan;
352 	} else {
353 		hvs = vsock_sk(sk)->trans;
354 		hvs->chan = chan;
355 	}
356 
357 	set_channel_read_mode(chan, HV_CALL_DIRECT);
358 
359 	/* Use the socket buffer sizes as hints for the VMBUS ring size. For
360 	 * server side sockets, 'sk' is the parent socket and thus, this will
361 	 * allow the child sockets to inherit the size from the parent. Keep
362 	 * the mins to the default value and align to page size as per VMBUS
363 	 * requirements.
364 	 * For the max, the socket core library will limit the socket buffer
365 	 * size that can be set by the user, but, since currently, the hv_sock
366 	 * VMBUS ring buffer is physically contiguous allocation, restrict it
367 	 * further.
368 	 * Older versions of hv_sock host side code cannot handle bigger VMBUS
369 	 * ring buffer size. Use the version number to limit the change to newer
370 	 * versions.
371 	 */
372 	if (vmbus_proto_version < VERSION_WIN10_V5) {
373 		sndbuf = RINGBUFFER_HVS_SND_SIZE;
374 		rcvbuf = RINGBUFFER_HVS_RCV_SIZE;
375 	} else {
376 		sndbuf = max_t(int, sk->sk_sndbuf, RINGBUFFER_HVS_SND_SIZE);
377 		sndbuf = min_t(int, sndbuf, RINGBUFFER_HVS_MAX_SIZE);
378 		sndbuf = ALIGN(sndbuf, HV_HYP_PAGE_SIZE);
379 		rcvbuf = max_t(int, sk->sk_rcvbuf, RINGBUFFER_HVS_RCV_SIZE);
380 		rcvbuf = min_t(int, rcvbuf, RINGBUFFER_HVS_MAX_SIZE);
381 		rcvbuf = ALIGN(rcvbuf, HV_HYP_PAGE_SIZE);
382 	}
383 
384 	chan->max_pkt_size = HVS_MAX_PKT_SIZE;
385 
386 	ret = vmbus_open(chan, sndbuf, rcvbuf, NULL, 0, hvs_channel_cb,
387 			 conn_from_host ? new : sk);
388 	if (ret != 0) {
389 		if (conn_from_host) {
390 			hvs_new->chan = NULL;
391 			sock_put(new);
392 		} else {
393 			hvs->chan = NULL;
394 		}
395 		goto out;
396 	}
397 
398 	set_per_channel_state(chan, conn_from_host ? new : sk);
399 
400 	/* This reference will be dropped by hvs_close_connection(). */
401 	sock_hold(conn_from_host ? new : sk);
402 	vmbus_set_chn_rescind_callback(chan, hvs_close_connection);
403 
404 	/* Set the pending send size to max packet size to always get
405 	 * notifications from the host when there is enough writable space.
406 	 * The host is optimized to send notifications only when the pending
407 	 * size boundary is crossed, and not always.
408 	 */
409 	hvs_set_channel_pending_send_size(chan);
410 
411 	if (conn_from_host) {
412 		new->sk_state = TCP_ESTABLISHED;
413 		sk_acceptq_added(sk);
414 
415 		hvs_new->vm_srv_id = *if_type;
416 		hvs_new->host_srv_id = *if_instance;
417 
418 		vsock_insert_connected(vnew);
419 
420 		vsock_enqueue_accept(sk, new);
421 	} else {
422 		sk->sk_state = TCP_ESTABLISHED;
423 		sk->sk_socket->state = SS_CONNECTED;
424 
425 		vsock_insert_connected(vsock_sk(sk));
426 	}
427 
428 	sk->sk_state_change(sk);
429 
430 out:
431 	/* Release refcnt obtained when we called vsock_find_bound_socket() */
432 	sock_put(sk);
433 
434 	release_sock(sk);
435 }
436 
437 static u32 hvs_get_local_cid(void)
438 {
439 	return VMADDR_CID_ANY;
440 }
441 
442 static int hvs_sock_init(struct vsock_sock *vsk, struct vsock_sock *psk)
443 {
444 	struct hvsock *hvs;
445 	struct sock *sk = sk_vsock(vsk);
446 
447 	hvs = kzalloc(sizeof(*hvs), GFP_KERNEL);
448 	if (!hvs)
449 		return -ENOMEM;
450 
451 	vsk->trans = hvs;
452 	hvs->vsk = vsk;
453 	sk->sk_sndbuf = RINGBUFFER_HVS_SND_SIZE;
454 	sk->sk_rcvbuf = RINGBUFFER_HVS_RCV_SIZE;
455 	return 0;
456 }
457 
458 static int hvs_connect(struct vsock_sock *vsk)
459 {
460 	union hvs_service_id vm, host;
461 	struct hvsock *h = vsk->trans;
462 
463 	vm.srv_id = srv_id_template;
464 	vm.svm_port = vsk->local_addr.svm_port;
465 	h->vm_srv_id = vm.srv_id;
466 
467 	host.srv_id = srv_id_template;
468 	host.svm_port = vsk->remote_addr.svm_port;
469 	h->host_srv_id = host.srv_id;
470 
471 	return vmbus_send_tl_connect_request(&h->vm_srv_id, &h->host_srv_id);
472 }
473 
474 static void hvs_shutdown_lock_held(struct hvsock *hvs, int mode)
475 {
476 	struct vmpipe_proto_header hdr;
477 
478 	if (hvs->fin_sent || !hvs->chan)
479 		return;
480 
481 	/* It can't fail: see hvs_channel_writable_bytes(). */
482 	(void)__hvs_send_data(hvs->chan, &hdr, 0);
483 	hvs->fin_sent = true;
484 }
485 
486 static int hvs_shutdown(struct vsock_sock *vsk, int mode)
487 {
488 	if (!(mode & SEND_SHUTDOWN))
489 		return 0;
490 
491 	hvs_shutdown_lock_held(vsk->trans, mode);
492 	return 0;
493 }
494 
495 static void hvs_close_timeout(struct work_struct *work)
496 {
497 	struct vsock_sock *vsk =
498 		container_of(work, struct vsock_sock, close_work.work);
499 	struct sock *sk = sk_vsock(vsk);
500 
501 	sock_hold(sk);
502 	lock_sock(sk);
503 	if (!sock_flag(sk, SOCK_DONE))
504 		hvs_do_close_lock_held(vsk, false);
505 
506 	vsk->close_work_scheduled = false;
507 	release_sock(sk);
508 	sock_put(sk);
509 }
510 
511 /* Returns true, if it is safe to remove socket; false otherwise */
512 static bool hvs_close_lock_held(struct vsock_sock *vsk)
513 {
514 	struct sock *sk = sk_vsock(vsk);
515 
516 	if (!(sk->sk_state == TCP_ESTABLISHED ||
517 	      sk->sk_state == TCP_CLOSING))
518 		return true;
519 
520 	if ((sk->sk_shutdown & SHUTDOWN_MASK) != SHUTDOWN_MASK)
521 		hvs_shutdown_lock_held(vsk->trans, SHUTDOWN_MASK);
522 
523 	if (sock_flag(sk, SOCK_DONE))
524 		return true;
525 
526 	/* This reference will be dropped by the delayed close routine */
527 	sock_hold(sk);
528 	INIT_DELAYED_WORK(&vsk->close_work, hvs_close_timeout);
529 	vsk->close_work_scheduled = true;
530 	schedule_delayed_work(&vsk->close_work, HVS_CLOSE_TIMEOUT);
531 	return false;
532 }
533 
534 static void hvs_release(struct vsock_sock *vsk)
535 {
536 	bool remove_sock;
537 
538 	remove_sock = hvs_close_lock_held(vsk);
539 	if (remove_sock)
540 		vsock_remove_sock(vsk);
541 }
542 
543 static void hvs_destruct(struct vsock_sock *vsk)
544 {
545 	struct hvsock *hvs = vsk->trans;
546 	struct vmbus_channel *chan = hvs->chan;
547 
548 	if (chan)
549 		vmbus_hvsock_device_unregister(chan);
550 
551 	kfree(hvs);
552 	vsk->trans = NULL;
553 }
554 
555 static int hvs_dgram_bind(struct vsock_sock *vsk, struct sockaddr_vm *addr)
556 {
557 	return -EOPNOTSUPP;
558 }
559 
560 static int hvs_dgram_dequeue(struct vsock_sock *vsk, struct msghdr *msg,
561 			     size_t len, int flags)
562 {
563 	return -EOPNOTSUPP;
564 }
565 
566 static int hvs_dgram_enqueue(struct vsock_sock *vsk,
567 			     struct sockaddr_vm *remote, struct msghdr *msg,
568 			     size_t dgram_len)
569 {
570 	return -EOPNOTSUPP;
571 }
572 
573 static bool hvs_dgram_allow(u32 cid, u32 port)
574 {
575 	return false;
576 }
577 
578 static int hvs_update_recv_data(struct hvsock *hvs)
579 {
580 	struct hvs_recv_buf *recv_buf;
581 	u32 pkt_len, payload_len;
582 
583 	pkt_len = hv_pkt_len(hvs->recv_desc);
584 
585 	if (pkt_len < HVS_HEADER_LEN)
586 		return -EIO;
587 
588 	recv_buf = (struct hvs_recv_buf *)(hvs->recv_desc + 1);
589 	payload_len = recv_buf->hdr.data_size;
590 
591 	if (payload_len > pkt_len - HVS_HEADER_LEN ||
592 	    payload_len > HVS_MTU_SIZE)
593 		return -EIO;
594 
595 	if (payload_len == 0)
596 		hvs->vsk->peer_shutdown |= SEND_SHUTDOWN;
597 
598 	hvs->recv_data_len = payload_len;
599 	hvs->recv_data_off = 0;
600 
601 	return 0;
602 }
603 
604 static ssize_t hvs_stream_dequeue(struct vsock_sock *vsk, struct msghdr *msg,
605 				  size_t len, int flags)
606 {
607 	struct hvsock *hvs = vsk->trans;
608 	bool need_refill = !hvs->recv_desc;
609 	struct hvs_recv_buf *recv_buf;
610 	u32 to_read;
611 	int ret;
612 
613 	if (flags & MSG_PEEK)
614 		return -EOPNOTSUPP;
615 
616 	if (need_refill) {
617 		hvs->recv_desc = hv_pkt_iter_first(hvs->chan);
618 		if (!hvs->recv_desc)
619 			return -ENOBUFS;
620 		ret = hvs_update_recv_data(hvs);
621 		if (ret)
622 			return ret;
623 	}
624 
625 	recv_buf = (struct hvs_recv_buf *)(hvs->recv_desc + 1);
626 	to_read = min_t(u32, len, hvs->recv_data_len);
627 	ret = memcpy_to_msg(msg, recv_buf->data + hvs->recv_data_off, to_read);
628 	if (ret != 0)
629 		return ret;
630 
631 	hvs->recv_data_len -= to_read;
632 	if (hvs->recv_data_len == 0) {
633 		hvs->recv_desc = hv_pkt_iter_next(hvs->chan, hvs->recv_desc);
634 		if (hvs->recv_desc) {
635 			ret = hvs_update_recv_data(hvs);
636 			if (ret)
637 				return ret;
638 		}
639 	} else {
640 		hvs->recv_data_off += to_read;
641 	}
642 
643 	return to_read;
644 }
645 
646 static ssize_t hvs_stream_enqueue(struct vsock_sock *vsk, struct msghdr *msg,
647 				  size_t len)
648 {
649 	struct hvsock *hvs = vsk->trans;
650 	struct vmbus_channel *chan = hvs->chan;
651 	struct hvs_send_buf *send_buf;
652 	ssize_t to_write, max_writable;
653 	ssize_t ret = 0;
654 	ssize_t bytes_written = 0;
655 
656 	BUILD_BUG_ON(sizeof(*send_buf) != HV_HYP_PAGE_SIZE);
657 
658 	send_buf = kmalloc(sizeof(*send_buf), GFP_KERNEL);
659 	if (!send_buf)
660 		return -ENOMEM;
661 
662 	/* Reader(s) could be draining data from the channel as we write.
663 	 * Maximize bandwidth, by iterating until the channel is found to be
664 	 * full.
665 	 */
666 	while (len) {
667 		max_writable = hvs_channel_writable_bytes(chan);
668 		if (!max_writable)
669 			break;
670 		to_write = min_t(ssize_t, len, max_writable);
671 		to_write = min_t(ssize_t, to_write, HVS_SEND_BUF_SIZE);
672 		/* memcpy_from_msg is safe for loop as it advances the offsets
673 		 * within the message iterator.
674 		 */
675 		ret = memcpy_from_msg(send_buf->data, msg, to_write);
676 		if (ret < 0)
677 			goto out;
678 
679 		ret = hvs_send_data(hvs->chan, send_buf, to_write);
680 		if (ret < 0)
681 			goto out;
682 
683 		bytes_written += to_write;
684 		len -= to_write;
685 	}
686 out:
687 	/* If any data has been sent, return that */
688 	if (bytes_written)
689 		ret = bytes_written;
690 	kfree(send_buf);
691 	return ret;
692 }
693 
694 static s64 hvs_stream_has_data(struct vsock_sock *vsk)
695 {
696 	struct hvsock *hvs = vsk->trans;
697 	s64 ret;
698 
699 	if (hvs->recv_data_len > 0)
700 		return 1;
701 
702 	switch (hvs_channel_readable_payload(hvs->chan)) {
703 	case 1:
704 		ret = 1;
705 		break;
706 	case 0:
707 		vsk->peer_shutdown |= SEND_SHUTDOWN;
708 		ret = 0;
709 		break;
710 	default: /* -1 */
711 		ret = 0;
712 		break;
713 	}
714 
715 	return ret;
716 }
717 
718 static s64 hvs_stream_has_space(struct vsock_sock *vsk)
719 {
720 	struct hvsock *hvs = vsk->trans;
721 
722 	return hvs_channel_writable_bytes(hvs->chan);
723 }
724 
725 static u64 hvs_stream_rcvhiwat(struct vsock_sock *vsk)
726 {
727 	return HVS_MTU_SIZE + 1;
728 }
729 
730 static bool hvs_stream_is_active(struct vsock_sock *vsk)
731 {
732 	struct hvsock *hvs = vsk->trans;
733 
734 	return hvs->chan != NULL;
735 }
736 
737 static bool hvs_stream_allow(u32 cid, u32 port)
738 {
739 	if (cid == VMADDR_CID_HOST)
740 		return true;
741 
742 	return false;
743 }
744 
745 static
746 int hvs_notify_poll_in(struct vsock_sock *vsk, size_t target, bool *readable)
747 {
748 	struct hvsock *hvs = vsk->trans;
749 
750 	*readable = hvs_channel_readable(hvs->chan);
751 	return 0;
752 }
753 
754 static
755 int hvs_notify_poll_out(struct vsock_sock *vsk, size_t target, bool *writable)
756 {
757 	*writable = hvs_stream_has_space(vsk) > 0;
758 
759 	return 0;
760 }
761 
762 static
763 int hvs_notify_recv_init(struct vsock_sock *vsk, size_t target,
764 			 struct vsock_transport_recv_notify_data *d)
765 {
766 	return 0;
767 }
768 
769 static
770 int hvs_notify_recv_pre_block(struct vsock_sock *vsk, size_t target,
771 			      struct vsock_transport_recv_notify_data *d)
772 {
773 	return 0;
774 }
775 
776 static
777 int hvs_notify_recv_pre_dequeue(struct vsock_sock *vsk, size_t target,
778 				struct vsock_transport_recv_notify_data *d)
779 {
780 	return 0;
781 }
782 
783 static
784 int hvs_notify_recv_post_dequeue(struct vsock_sock *vsk, size_t target,
785 				 ssize_t copied, bool data_read,
786 				 struct vsock_transport_recv_notify_data *d)
787 {
788 	return 0;
789 }
790 
791 static
792 int hvs_notify_send_init(struct vsock_sock *vsk,
793 			 struct vsock_transport_send_notify_data *d)
794 {
795 	return 0;
796 }
797 
798 static
799 int hvs_notify_send_pre_block(struct vsock_sock *vsk,
800 			      struct vsock_transport_send_notify_data *d)
801 {
802 	return 0;
803 }
804 
805 static
806 int hvs_notify_send_pre_enqueue(struct vsock_sock *vsk,
807 				struct vsock_transport_send_notify_data *d)
808 {
809 	return 0;
810 }
811 
812 static
813 int hvs_notify_send_post_enqueue(struct vsock_sock *vsk, ssize_t written,
814 				 struct vsock_transport_send_notify_data *d)
815 {
816 	return 0;
817 }
818 
819 static
820 int hvs_notify_set_rcvlowat(struct vsock_sock *vsk, int val)
821 {
822 	return -EOPNOTSUPP;
823 }
824 
825 static struct vsock_transport hvs_transport = {
826 	.module                   = THIS_MODULE,
827 
828 	.get_local_cid            = hvs_get_local_cid,
829 
830 	.init                     = hvs_sock_init,
831 	.destruct                 = hvs_destruct,
832 	.release                  = hvs_release,
833 	.connect                  = hvs_connect,
834 	.shutdown                 = hvs_shutdown,
835 
836 	.dgram_bind               = hvs_dgram_bind,
837 	.dgram_dequeue            = hvs_dgram_dequeue,
838 	.dgram_enqueue            = hvs_dgram_enqueue,
839 	.dgram_allow              = hvs_dgram_allow,
840 
841 	.stream_dequeue           = hvs_stream_dequeue,
842 	.stream_enqueue           = hvs_stream_enqueue,
843 	.stream_has_data          = hvs_stream_has_data,
844 	.stream_has_space         = hvs_stream_has_space,
845 	.stream_rcvhiwat          = hvs_stream_rcvhiwat,
846 	.stream_is_active         = hvs_stream_is_active,
847 	.stream_allow             = hvs_stream_allow,
848 
849 	.notify_poll_in           = hvs_notify_poll_in,
850 	.notify_poll_out          = hvs_notify_poll_out,
851 	.notify_recv_init         = hvs_notify_recv_init,
852 	.notify_recv_pre_block    = hvs_notify_recv_pre_block,
853 	.notify_recv_pre_dequeue  = hvs_notify_recv_pre_dequeue,
854 	.notify_recv_post_dequeue = hvs_notify_recv_post_dequeue,
855 	.notify_send_init         = hvs_notify_send_init,
856 	.notify_send_pre_block    = hvs_notify_send_pre_block,
857 	.notify_send_pre_enqueue  = hvs_notify_send_pre_enqueue,
858 	.notify_send_post_enqueue = hvs_notify_send_post_enqueue,
859 
860 	.notify_set_rcvlowat      = hvs_notify_set_rcvlowat
861 };
862 
863 static bool hvs_check_transport(struct vsock_sock *vsk)
864 {
865 	return vsk->transport == &hvs_transport;
866 }
867 
868 static int hvs_probe(struct hv_device *hdev,
869 		     const struct hv_vmbus_device_id *dev_id)
870 {
871 	struct vmbus_channel *chan = hdev->channel;
872 
873 	hvs_open_connection(chan);
874 
875 	/* Always return success to suppress the unnecessary error message
876 	 * in vmbus_probe(): on error the host will rescind the device in
877 	 * 30 seconds and we can do cleanup at that time in
878 	 * vmbus_onoffer_rescind().
879 	 */
880 	return 0;
881 }
882 
883 static void hvs_remove(struct hv_device *hdev)
884 {
885 	struct vmbus_channel *chan = hdev->channel;
886 
887 	vmbus_close(chan);
888 }
889 
890 /* hv_sock connections can not persist across hibernation, and all the hv_sock
891  * channels are forced to be rescinded before hibernation: see
892  * vmbus_bus_suspend(). Here the dummy hvs_suspend() and hvs_resume()
893  * are only needed because hibernation requires that every vmbus device's
894  * driver should have a .suspend and .resume callback: see vmbus_suspend().
895  */
896 static int hvs_suspend(struct hv_device *hv_dev)
897 {
898 	/* Dummy */
899 	return 0;
900 }
901 
902 static int hvs_resume(struct hv_device *dev)
903 {
904 	/* Dummy */
905 	return 0;
906 }
907 
908 /* This isn't really used. See vmbus_match() and vmbus_probe() */
909 static const struct hv_vmbus_device_id id_table[] = {
910 	{},
911 };
912 
913 static struct hv_driver hvs_drv = {
914 	.name		= "hv_sock",
915 	.hvsock		= true,
916 	.id_table	= id_table,
917 	.probe		= hvs_probe,
918 	.remove		= hvs_remove,
919 	.suspend	= hvs_suspend,
920 	.resume		= hvs_resume,
921 };
922 
923 static int __init hvs_init(void)
924 {
925 	int ret;
926 
927 	if (vmbus_proto_version < VERSION_WIN10)
928 		return -ENODEV;
929 
930 	ret = vmbus_driver_register(&hvs_drv);
931 	if (ret != 0)
932 		return ret;
933 
934 	ret = vsock_core_register(&hvs_transport, VSOCK_TRANSPORT_F_G2H);
935 	if (ret) {
936 		vmbus_driver_unregister(&hvs_drv);
937 		return ret;
938 	}
939 
940 	return 0;
941 }
942 
943 static void __exit hvs_exit(void)
944 {
945 	vsock_core_unregister(&hvs_transport);
946 	vmbus_driver_unregister(&hvs_drv);
947 }
948 
949 module_init(hvs_init);
950 module_exit(hvs_exit);
951 
952 MODULE_DESCRIPTION("Hyper-V Sockets");
953 MODULE_VERSION("1.0.0");
954 MODULE_LICENSE("GPL");
955 MODULE_ALIAS_NETPROTO(PF_VSOCK);
956