1 // SPDX-License-Identifier: GPL-2.0-only
3 * Hyper-V transport for vsock
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.
9 * Copyright (c) 2017, Microsoft Corporation.
11 #include <linux/module.h>
12 #include <linux/vmalloc.h>
13 #include <linux/hyperv.h>
15 #include <net/af_vsock.h>
16 #include <asm/hyperv-tlfs.h>
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.
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)
27 /* The MTU is 16KB per the host side's design */
28 #define HVS_MTU_SIZE (1024 * 16)
30 /* How long to wait for graceful shutdown of a connection */
31 #define HVS_CLOSE_TIMEOUT (8 * HZ)
33 struct vmpipe_proto_header {
38 /* For recv, we use the VMBus in-place packet iterator APIs to directly copy
39 * data from the ringbuffer into the userspace buffer.
42 /* The header before the payload data */
43 struct vmpipe_proto_header hdr;
46 u8 data[HVS_MTU_SIZE];
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
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.
58 #define HVS_SEND_BUF_SIZE \
59 (HV_HYP_PAGE_SIZE - sizeof(struct vmpipe_proto_header))
62 /* The header before the payload data */
63 struct vmpipe_proto_header hdr;
66 u8 data[HVS_SEND_BUF_SIZE];
69 #define HVS_HEADER_LEN (sizeof(struct vmpacket_descriptor) + \
70 sizeof(struct vmpipe_proto_header))
72 /* See 'prev_indices' in hv_ringbuffer_read(), hv_ringbuffer_write(), and
73 * __hv_pkt_iter_next().
75 #define VMBUS_PKT_TRAILER_SIZE (sizeof(u64))
77 #define HVS_PKT_LEN(payload_len) (HVS_HEADER_LEN + \
78 ALIGN((payload_len), 8) + \
79 VMBUS_PKT_TRAILER_SIZE)
81 union hvs_service_id {
85 unsigned int svm_port;
86 unsigned char b[sizeof(guid_t) - sizeof(unsigned int)];
90 /* Per-socket state (accessed via vsk->trans) */
92 struct vsock_sock *vsk;
97 struct vmbus_channel *chan;
98 struct vmpacket_descriptor *recv_desc;
100 /* The length of the payload not delivered to userland yet */
102 /* The offset of the payload */
105 /* Have we sent the zero-length packet (FIN)? */
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
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:
123 * ADDRESS_FAMILY Family;
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.
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:
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 ****************************************************************************
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.
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);
155 static bool hvs_check_transport(struct vsock_sock *vsk);
157 static bool is_valid_srv_id(const guid_t *id)
159 return !memcmp(&id->b[4], &srv_id_template.b[4], sizeof(guid_t) - 4);
162 static unsigned int get_port_by_srv_id(const guid_t *svr_id)
164 return *((unsigned int *)svr_id);
167 static void hvs_addr_init(struct sockaddr_vm *addr, const guid_t *svr_id)
169 unsigned int port = get_port_by_srv_id(svr_id);
171 vsock_addr_init(addr, VMADDR_CID_ANY, port);
174 static void hvs_set_channel_pending_send_size(struct vmbus_channel *chan)
176 set_channel_pending_send_size(chan,
177 HVS_PKT_LEN(HVS_SEND_BUF_SIZE));
182 static bool hvs_channel_readable(struct vmbus_channel *chan)
184 u32 readable = hv_get_bytes_to_read(&chan->inbound);
186 /* 0-size payload means FIN */
187 return readable >= HVS_PKT_LEN(0);
190 static int hvs_channel_readable_payload(struct vmbus_channel *chan)
192 u32 readable = hv_get_bytes_to_read(&chan->inbound);
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().
202 if (readable == HVS_PKT_LEN(0)) {
203 /* 0-size payload means FIN */
207 /* No payload or FIN */
211 static size_t hvs_channel_writable_bytes(struct vmbus_channel *chan)
213 u32 writeable = hv_get_bytes_to_write(&chan->outbound);
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().
220 if (writeable <= HVS_PKT_LEN(1) + HVS_PKT_LEN(0))
223 ret = writeable - HVS_PKT_LEN(1) - HVS_PKT_LEN(0);
225 return round_down(ret, 8);
228 static int hvs_send_data(struct vmbus_channel *chan,
229 struct hvs_send_buf *send_buf, size_t to_write)
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);
238 static void hvs_channel_cb(void *ctx)
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;
245 if (hvs_channel_readable(chan))
246 sk->sk_data_ready(sk);
248 if (hv_get_bytes_to_write(&chan->outbound) > 0)
249 sk->sk_write_space(sk);
252 static void hvs_do_close_lock_held(struct vsock_sock *vsk,
255 struct sock *sk = sk_vsock(vsk);
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);
267 /* Release the reference taken while scheduling the timeout */
272 static void hvs_close_connection(struct vmbus_channel *chan)
274 struct sock *sk = get_per_channel_state(chan);
277 hvs_do_close_lock_held(vsock_sk(sk), true);
280 /* Release the refcnt for the channel that's opened in
281 * hvs_open_connection().
286 static void hvs_open_connection(struct vmbus_channel *chan)
288 guid_t *if_instance, *if_type;
289 unsigned char conn_from_host;
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;
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))
306 hvs_addr_init(&addr, conn_from_host ? if_type : if_instance);
307 sk = vsock_find_bound_socket(&addr);
312 if ((conn_from_host && sk->sk_state != TCP_LISTEN) ||
313 (!conn_from_host && sk->sk_state != TCP_SYN_SENT))
316 if (conn_from_host) {
317 if (sk->sk_ack_backlog >= sk->sk_max_ack_backlog)
320 new = vsock_create_connected(sk);
324 new->sk_state = TCP_SYN_SENT;
325 vnew = vsock_sk(new);
327 hvs_addr_init(&vnew->local_addr, if_type);
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.
337 if (ret || !hvs_check_transport(vnew)) {
341 hvs_new = vnew->trans;
342 hvs_new->chan = chan;
344 hvs = vsock_sk(sk)->trans;
348 set_channel_read_mode(chan, HV_CALL_DIRECT);
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
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
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
363 if (vmbus_proto_version < VERSION_WIN10_V5) {
364 sndbuf = RINGBUFFER_HVS_SND_SIZE;
365 rcvbuf = RINGBUFFER_HVS_RCV_SIZE;
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);
375 ret = vmbus_open(chan, sndbuf, rcvbuf, NULL, 0, hvs_channel_cb,
376 conn_from_host ? new : sk);
378 if (conn_from_host) {
379 hvs_new->chan = NULL;
387 set_per_channel_state(chan, conn_from_host ? new : sk);
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);
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.
398 hvs_set_channel_pending_send_size(chan);
400 if (conn_from_host) {
401 new->sk_state = TCP_ESTABLISHED;
402 sk_acceptq_added(sk);
404 hvs_new->vm_srv_id = *if_type;
405 hvs_new->host_srv_id = *if_instance;
407 vsock_insert_connected(vnew);
409 vsock_enqueue_accept(sk, new);
411 sk->sk_state = TCP_ESTABLISHED;
412 sk->sk_socket->state = SS_CONNECTED;
414 vsock_insert_connected(vsock_sk(sk));
417 sk->sk_state_change(sk);
420 /* Release refcnt obtained when we called vsock_find_bound_socket() */
426 static u32 hvs_get_local_cid(void)
428 return VMADDR_CID_ANY;
431 static int hvs_sock_init(struct vsock_sock *vsk, struct vsock_sock *psk)
434 struct sock *sk = sk_vsock(vsk);
436 hvs = kzalloc(sizeof(*hvs), GFP_KERNEL);
442 sk->sk_sndbuf = RINGBUFFER_HVS_SND_SIZE;
443 sk->sk_rcvbuf = RINGBUFFER_HVS_RCV_SIZE;
447 static int hvs_connect(struct vsock_sock *vsk)
449 union hvs_service_id vm, host;
450 struct hvsock *h = vsk->trans;
452 vm.srv_id = srv_id_template;
453 vm.svm_port = vsk->local_addr.svm_port;
454 h->vm_srv_id = vm.srv_id;
456 host.srv_id = srv_id_template;
457 host.svm_port = vsk->remote_addr.svm_port;
458 h->host_srv_id = host.srv_id;
460 return vmbus_send_tl_connect_request(&h->vm_srv_id, &h->host_srv_id);
463 static void hvs_shutdown_lock_held(struct hvsock *hvs, int mode)
465 struct vmpipe_proto_header hdr;
467 if (hvs->fin_sent || !hvs->chan)
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;
475 static int hvs_shutdown(struct vsock_sock *vsk, int mode)
477 struct sock *sk = sk_vsock(vsk);
479 if (!(mode & SEND_SHUTDOWN))
483 hvs_shutdown_lock_held(vsk->trans, mode);
488 static void hvs_close_timeout(struct work_struct *work)
490 struct vsock_sock *vsk =
491 container_of(work, struct vsock_sock, close_work.work);
492 struct sock *sk = sk_vsock(vsk);
496 if (!sock_flag(sk, SOCK_DONE))
497 hvs_do_close_lock_held(vsk, false);
499 vsk->close_work_scheduled = false;
504 /* Returns true, if it is safe to remove socket; false otherwise */
505 static bool hvs_close_lock_held(struct vsock_sock *vsk)
507 struct sock *sk = sk_vsock(vsk);
509 if (!(sk->sk_state == TCP_ESTABLISHED ||
510 sk->sk_state == TCP_CLOSING))
513 if ((sk->sk_shutdown & SHUTDOWN_MASK) != SHUTDOWN_MASK)
514 hvs_shutdown_lock_held(vsk->trans, SHUTDOWN_MASK);
516 if (sock_flag(sk, SOCK_DONE))
519 /* This reference will be dropped by the delayed close routine */
521 INIT_DELAYED_WORK(&vsk->close_work, hvs_close_timeout);
522 vsk->close_work_scheduled = true;
523 schedule_delayed_work(&vsk->close_work, HVS_CLOSE_TIMEOUT);
527 static void hvs_release(struct vsock_sock *vsk)
529 struct sock *sk = sk_vsock(vsk);
532 lock_sock_nested(sk, SINGLE_DEPTH_NESTING);
533 remove_sock = hvs_close_lock_held(vsk);
536 vsock_remove_sock(vsk);
539 static void hvs_destruct(struct vsock_sock *vsk)
541 struct hvsock *hvs = vsk->trans;
542 struct vmbus_channel *chan = hvs->chan;
545 vmbus_hvsock_device_unregister(chan);
550 static int hvs_dgram_bind(struct vsock_sock *vsk, struct sockaddr_vm *addr)
555 static int hvs_dgram_dequeue(struct vsock_sock *vsk, struct msghdr *msg,
556 size_t len, int flags)
561 static int hvs_dgram_enqueue(struct vsock_sock *vsk,
562 struct sockaddr_vm *remote, struct msghdr *msg,
568 static bool hvs_dgram_allow(u32 cid, u32 port)
573 static int hvs_update_recv_data(struct hvsock *hvs)
575 struct hvs_recv_buf *recv_buf;
578 recv_buf = (struct hvs_recv_buf *)(hvs->recv_desc + 1);
579 payload_len = recv_buf->hdr.data_size;
581 if (payload_len > HVS_MTU_SIZE)
584 if (payload_len == 0)
585 hvs->vsk->peer_shutdown |= SEND_SHUTDOWN;
587 hvs->recv_data_len = payload_len;
588 hvs->recv_data_off = 0;
593 static ssize_t hvs_stream_dequeue(struct vsock_sock *vsk, struct msghdr *msg,
594 size_t len, int flags)
596 struct hvsock *hvs = vsk->trans;
597 bool need_refill = !hvs->recv_desc;
598 struct hvs_recv_buf *recv_buf;
602 if (flags & MSG_PEEK)
606 hvs->recv_desc = hv_pkt_iter_first(hvs->chan);
607 ret = hvs_update_recv_data(hvs);
612 recv_buf = (struct hvs_recv_buf *)(hvs->recv_desc + 1);
613 to_read = min_t(u32, len, hvs->recv_data_len);
614 ret = memcpy_to_msg(msg, recv_buf->data + hvs->recv_data_off, to_read);
618 hvs->recv_data_len -= to_read;
619 if (hvs->recv_data_len == 0) {
620 hvs->recv_desc = hv_pkt_iter_next(hvs->chan, hvs->recv_desc);
621 if (hvs->recv_desc) {
622 ret = hvs_update_recv_data(hvs);
627 hvs->recv_data_off += to_read;
633 static ssize_t hvs_stream_enqueue(struct vsock_sock *vsk, struct msghdr *msg,
636 struct hvsock *hvs = vsk->trans;
637 struct vmbus_channel *chan = hvs->chan;
638 struct hvs_send_buf *send_buf;
639 ssize_t to_write, max_writable;
641 ssize_t bytes_written = 0;
643 BUILD_BUG_ON(sizeof(*send_buf) != HV_HYP_PAGE_SIZE);
645 send_buf = kmalloc(sizeof(*send_buf), GFP_KERNEL);
649 /* Reader(s) could be draining data from the channel as we write.
650 * Maximize bandwidth, by iterating until the channel is found to be
654 max_writable = hvs_channel_writable_bytes(chan);
657 to_write = min_t(ssize_t, len, max_writable);
658 to_write = min_t(ssize_t, to_write, HVS_SEND_BUF_SIZE);
659 /* memcpy_from_msg is safe for loop as it advances the offsets
660 * within the message iterator.
662 ret = memcpy_from_msg(send_buf->data, msg, to_write);
666 ret = hvs_send_data(hvs->chan, send_buf, to_write);
670 bytes_written += to_write;
674 /* If any data has been sent, return that */
681 static s64 hvs_stream_has_data(struct vsock_sock *vsk)
683 struct hvsock *hvs = vsk->trans;
686 if (hvs->recv_data_len > 0)
689 switch (hvs_channel_readable_payload(hvs->chan)) {
694 vsk->peer_shutdown |= SEND_SHUTDOWN;
705 static s64 hvs_stream_has_space(struct vsock_sock *vsk)
707 struct hvsock *hvs = vsk->trans;
709 return hvs_channel_writable_bytes(hvs->chan);
712 static u64 hvs_stream_rcvhiwat(struct vsock_sock *vsk)
714 return HVS_MTU_SIZE + 1;
717 static bool hvs_stream_is_active(struct vsock_sock *vsk)
719 struct hvsock *hvs = vsk->trans;
721 return hvs->chan != NULL;
724 static bool hvs_stream_allow(u32 cid, u32 port)
726 if (cid == VMADDR_CID_HOST)
733 int hvs_notify_poll_in(struct vsock_sock *vsk, size_t target, bool *readable)
735 struct hvsock *hvs = vsk->trans;
737 *readable = hvs_channel_readable(hvs->chan);
742 int hvs_notify_poll_out(struct vsock_sock *vsk, size_t target, bool *writable)
744 *writable = hvs_stream_has_space(vsk) > 0;
750 int hvs_notify_recv_init(struct vsock_sock *vsk, size_t target,
751 struct vsock_transport_recv_notify_data *d)
757 int hvs_notify_recv_pre_block(struct vsock_sock *vsk, size_t target,
758 struct vsock_transport_recv_notify_data *d)
764 int hvs_notify_recv_pre_dequeue(struct vsock_sock *vsk, size_t target,
765 struct vsock_transport_recv_notify_data *d)
771 int hvs_notify_recv_post_dequeue(struct vsock_sock *vsk, size_t target,
772 ssize_t copied, bool data_read,
773 struct vsock_transport_recv_notify_data *d)
779 int hvs_notify_send_init(struct vsock_sock *vsk,
780 struct vsock_transport_send_notify_data *d)
786 int hvs_notify_send_pre_block(struct vsock_sock *vsk,
787 struct vsock_transport_send_notify_data *d)
793 int hvs_notify_send_pre_enqueue(struct vsock_sock *vsk,
794 struct vsock_transport_send_notify_data *d)
800 int hvs_notify_send_post_enqueue(struct vsock_sock *vsk, ssize_t written,
801 struct vsock_transport_send_notify_data *d)
806 static struct vsock_transport hvs_transport = {
807 .module = THIS_MODULE,
809 .get_local_cid = hvs_get_local_cid,
811 .init = hvs_sock_init,
812 .destruct = hvs_destruct,
813 .release = hvs_release,
814 .connect = hvs_connect,
815 .shutdown = hvs_shutdown,
817 .dgram_bind = hvs_dgram_bind,
818 .dgram_dequeue = hvs_dgram_dequeue,
819 .dgram_enqueue = hvs_dgram_enqueue,
820 .dgram_allow = hvs_dgram_allow,
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,
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,
843 static bool hvs_check_transport(struct vsock_sock *vsk)
845 return vsk->transport == &hvs_transport;
848 static int hvs_probe(struct hv_device *hdev,
849 const struct hv_vmbus_device_id *dev_id)
851 struct vmbus_channel *chan = hdev->channel;
853 hvs_open_connection(chan);
855 /* Always return success to suppress the unnecessary error message
856 * in vmbus_probe(): on error the host will rescind the device in
857 * 30 seconds and we can do cleanup at that time in
858 * vmbus_onoffer_rescind().
863 static int hvs_remove(struct hv_device *hdev)
865 struct vmbus_channel *chan = hdev->channel;
872 /* hv_sock connections can not persist across hibernation, and all the hv_sock
873 * channels are forced to be rescinded before hibernation: see
874 * vmbus_bus_suspend(). Here the dummy hvs_suspend() and hvs_resume()
875 * are only needed because hibernation requires that every vmbus device's
876 * driver should have a .suspend and .resume callback: see vmbus_suspend().
878 static int hvs_suspend(struct hv_device *hv_dev)
884 static int hvs_resume(struct hv_device *dev)
890 /* This isn't really used. See vmbus_match() and vmbus_probe() */
891 static const struct hv_vmbus_device_id id_table[] = {
895 static struct hv_driver hvs_drv = {
898 .id_table = id_table,
900 .remove = hvs_remove,
901 .suspend = hvs_suspend,
902 .resume = hvs_resume,
905 static int __init hvs_init(void)
909 if (vmbus_proto_version < VERSION_WIN10)
912 ret = vmbus_driver_register(&hvs_drv);
916 ret = vsock_core_register(&hvs_transport, VSOCK_TRANSPORT_F_G2H);
918 vmbus_driver_unregister(&hvs_drv);
925 static void __exit hvs_exit(void)
927 vsock_core_unregister(&hvs_transport);
928 vmbus_driver_unregister(&hvs_drv);
931 module_init(hvs_init);
932 module_exit(hvs_exit);
934 MODULE_DESCRIPTION("Hyper-V Sockets");
935 MODULE_VERSION("1.0.0");
936 MODULE_LICENSE("GPL");
937 MODULE_ALIAS_NETPROTO(PF_VSOCK);