2 * Hyper-V transport for vsock
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.
8 * Copyright (c) 2017, Microsoft Corporation.
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.
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
20 #include <linux/module.h>
21 #include <linux/vmalloc.h>
22 #include <linux/hyperv.h>
24 #include <net/af_vsock.h>
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.
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)
35 /* The MTU is 16KB per the host side's design */
36 #define HVS_MTU_SIZE (1024 * 16)
38 /* How long to wait for graceful shutdown of a connection */
39 #define HVS_CLOSE_TIMEOUT (8 * HZ)
41 struct vmpipe_proto_header {
46 /* For recv, we use the VMBus in-place packet iterator APIs to directly copy
47 * data from the ringbuffer into the userspace buffer.
50 /* The header before the payload data */
51 struct vmpipe_proto_header hdr;
54 u8 data[HVS_MTU_SIZE];
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.
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.
65 #define HVS_SEND_BUF_SIZE (PAGE_SIZE_4K - sizeof(struct vmpipe_proto_header))
68 /* The header before the payload data */
69 struct vmpipe_proto_header hdr;
72 u8 data[HVS_SEND_BUF_SIZE];
75 #define HVS_HEADER_LEN (sizeof(struct vmpacket_descriptor) + \
76 sizeof(struct vmpipe_proto_header))
78 /* See 'prev_indices' in hv_ringbuffer_read(), hv_ringbuffer_write(), and
79 * __hv_pkt_iter_next().
81 #define VMBUS_PKT_TRAILER_SIZE (sizeof(u64))
83 #define HVS_PKT_LEN(payload_len) (HVS_HEADER_LEN + \
84 ALIGN((payload_len), 8) + \
85 VMBUS_PKT_TRAILER_SIZE)
87 union hvs_service_id {
91 unsigned int svm_port;
92 unsigned char b[sizeof(uuid_le) - sizeof(unsigned int)];
96 /* Per-socket state (accessed via vsk->trans) */
98 struct vsock_sock *vsk;
103 struct vmbus_channel *chan;
104 struct vmpacket_descriptor *recv_desc;
106 /* The length of the payload not delivered to userland yet */
108 /* The offset of the payload */
111 /* Have we sent the zero-length packet (FIN)? */
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
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:
129 * ADDRESS_FAMILY Family;
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.
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:
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 ****************************************************************************
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.
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).
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)
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);
174 static bool is_valid_srv_id(const uuid_le *id)
176 return !memcmp(&id->b[4], &srv_id_template.b[4], sizeof(uuid_le) - 4);
179 static unsigned int get_port_by_srv_id(const uuid_le *svr_id)
181 return *((unsigned int *)svr_id);
184 static void hvs_addr_init(struct sockaddr_vm *addr, const uuid_le *svr_id)
186 unsigned int port = get_port_by_srv_id(svr_id);
188 vsock_addr_init(addr, VMADDR_CID_ANY, port);
191 static void hvs_remote_addr_init(struct sockaddr_vm *remote,
192 struct sockaddr_vm *local)
194 static u32 host_ephemeral_port = MIN_HOST_EPHEMERAL_PORT;
197 vsock_addr_init(remote, VMADDR_CID_ANY, VMADDR_PORT_ANY);
201 if (host_ephemeral_port < MIN_HOST_EPHEMERAL_PORT ||
202 host_ephemeral_port == VMADDR_PORT_ANY)
203 host_ephemeral_port = MIN_HOST_EPHEMERAL_PORT;
205 remote->svm_port = host_ephemeral_port++;
207 sk = vsock_find_connected_socket(remote, local);
209 /* Found an available ephemeral port */
213 /* Release refcnt got in vsock_find_connected_socket */
218 static void hvs_set_channel_pending_send_size(struct vmbus_channel *chan)
220 set_channel_pending_send_size(chan,
221 HVS_PKT_LEN(HVS_SEND_BUF_SIZE));
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
230 static void hvs_clear_channel_pending_send_size(struct vmbus_channel *chan)
232 set_channel_pending_send_size(chan, 0);
238 static bool hvs_channel_readable(struct vmbus_channel *chan)
240 u32 readable = hv_get_bytes_to_read(&chan->inbound);
242 /* 0-size payload means FIN */
243 return readable >= HVS_PKT_LEN(0);
246 static int hvs_channel_readable_payload(struct vmbus_channel *chan)
248 u32 readable = hv_get_bytes_to_read(&chan->inbound);
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().
258 if (readable == HVS_PKT_LEN(0)) {
259 /* 0-size payload means FIN */
263 /* No payload or FIN */
267 static size_t hvs_channel_writable_bytes(struct vmbus_channel *chan)
269 u32 writeable = hv_get_bytes_to_write(&chan->outbound);
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().
276 if (writeable <= HVS_PKT_LEN(1) + HVS_PKT_LEN(0))
279 ret = writeable - HVS_PKT_LEN(1) - HVS_PKT_LEN(0);
281 return round_down(ret, 8);
284 static int hvs_send_data(struct vmbus_channel *chan,
285 struct hvs_send_buf *send_buf, size_t to_write)
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);
294 static void hvs_channel_cb(void *ctx)
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;
301 if (hvs_channel_readable(chan))
302 sk->sk_data_ready(sk);
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).
307 if (hv_get_bytes_to_write(&chan->outbound) > 0)
308 sk->sk_write_space(sk);
311 static void hvs_do_close_lock_held(struct vsock_sock *vsk,
314 struct sock *sk = sk_vsock(vsk);
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);
326 /* Release the reference taken while scheduling the timeout */
331 static void hvs_close_connection(struct vmbus_channel *chan)
333 struct sock *sk = get_per_channel_state(chan);
336 hvs_do_close_lock_held(vsock_sk(sk), true);
340 static void hvs_open_connection(struct vmbus_channel *chan)
342 uuid_le *if_instance, *if_type;
343 unsigned char conn_from_host;
345 struct sockaddr_vm addr;
346 struct sock *sk, *new = NULL;
347 struct vsock_sock *vnew;
348 struct hvsock *hvs, *hvs_new;
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];
355 /* The host or the VM should only listen on a port in
356 * [0, MAX_LISTEN_PORT]
358 if (!is_valid_srv_id(if_type) ||
359 get_port_by_srv_id(if_type) > MAX_LISTEN_PORT)
362 hvs_addr_init(&addr, conn_from_host ? if_type : if_instance);
363 sk = vsock_find_bound_socket(&addr);
368 if ((conn_from_host && sk->sk_state != TCP_LISTEN) ||
369 (!conn_from_host && sk->sk_state != TCP_SYN_SENT))
372 if (conn_from_host) {
373 if (sk->sk_ack_backlog >= sk->sk_max_ack_backlog)
376 new = __vsock_create(sock_net(sk), NULL, sk, GFP_KERNEL,
381 new->sk_state = TCP_SYN_SENT;
382 vnew = vsock_sk(new);
383 hvs_new = vnew->trans;
384 hvs_new->chan = chan;
386 hvs = vsock_sk(sk)->trans;
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);
395 if (conn_from_host) {
396 hvs_new->chan = NULL;
404 set_per_channel_state(chan, conn_from_host ? new : sk);
405 vmbus_set_chn_rescind_callback(chan, hvs_close_connection);
407 if (conn_from_host) {
408 new->sk_state = TCP_ESTABLISHED;
409 sk->sk_ack_backlog++;
411 hvs_addr_init(&vnew->local_addr, if_type);
412 hvs_remote_addr_init(&vnew->remote_addr, &vnew->local_addr);
414 hvs_new->vm_srv_id = *if_type;
415 hvs_new->host_srv_id = *if_instance;
417 vsock_insert_connected(vnew);
419 vsock_enqueue_accept(sk, new);
421 sk->sk_state = TCP_ESTABLISHED;
422 sk->sk_socket->state = SS_CONNECTED;
424 vsock_insert_connected(vsock_sk(sk));
427 sk->sk_state_change(sk);
430 /* Release refcnt obtained when we called vsock_find_bound_socket() */
436 static u32 hvs_get_local_cid(void)
438 return VMADDR_CID_ANY;
441 static int hvs_sock_init(struct vsock_sock *vsk, struct vsock_sock *psk)
445 hvs = kzalloc(sizeof(*hvs), GFP_KERNEL);
455 static int hvs_connect(struct vsock_sock *vsk)
457 union hvs_service_id vm, host;
458 struct hvsock *h = vsk->trans;
460 vm.srv_id = srv_id_template;
461 vm.svm_port = vsk->local_addr.svm_port;
462 h->vm_srv_id = vm.srv_id;
464 host.srv_id = srv_id_template;
465 host.svm_port = vsk->remote_addr.svm_port;
466 h->host_srv_id = host.srv_id;
468 return vmbus_send_tl_connect_request(&h->vm_srv_id, &h->host_srv_id);
471 static void hvs_shutdown_lock_held(struct hvsock *hvs, int mode)
473 struct vmpipe_proto_header hdr;
475 if (hvs->fin_sent || !hvs->chan)
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;
483 static int hvs_shutdown(struct vsock_sock *vsk, int mode)
485 struct sock *sk = sk_vsock(vsk);
487 if (!(mode & SEND_SHUTDOWN))
491 hvs_shutdown_lock_held(vsk->trans, mode);
496 static void hvs_close_timeout(struct work_struct *work)
498 struct vsock_sock *vsk =
499 container_of(work, struct vsock_sock, close_work.work);
500 struct sock *sk = sk_vsock(vsk);
504 if (!sock_flag(sk, SOCK_DONE))
505 hvs_do_close_lock_held(vsk, false);
507 vsk->close_work_scheduled = false;
512 /* Returns true, if it is safe to remove socket; false otherwise */
513 static bool hvs_close_lock_held(struct vsock_sock *vsk)
515 struct sock *sk = sk_vsock(vsk);
517 if (!(sk->sk_state == TCP_ESTABLISHED ||
518 sk->sk_state == TCP_CLOSING))
521 if ((sk->sk_shutdown & SHUTDOWN_MASK) != SHUTDOWN_MASK)
522 hvs_shutdown_lock_held(vsk->trans, SHUTDOWN_MASK);
524 if (sock_flag(sk, SOCK_DONE))
527 /* This reference will be dropped by the delayed close routine */
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);
535 static void hvs_release(struct vsock_sock *vsk)
537 struct sock *sk = sk_vsock(vsk);
541 remove_sock = hvs_close_lock_held(vsk);
544 vsock_remove_sock(vsk);
547 static void hvs_destruct(struct vsock_sock *vsk)
549 struct hvsock *hvs = vsk->trans;
550 struct vmbus_channel *chan = hvs->chan;
553 vmbus_hvsock_device_unregister(chan);
558 static int hvs_dgram_bind(struct vsock_sock *vsk, struct sockaddr_vm *addr)
563 static int hvs_dgram_dequeue(struct vsock_sock *vsk, struct msghdr *msg,
564 size_t len, int flags)
569 static int hvs_dgram_enqueue(struct vsock_sock *vsk,
570 struct sockaddr_vm *remote, struct msghdr *msg,
576 static bool hvs_dgram_allow(u32 cid, u32 port)
581 static int hvs_update_recv_data(struct hvsock *hvs)
583 struct hvs_recv_buf *recv_buf;
586 recv_buf = (struct hvs_recv_buf *)(hvs->recv_desc + 1);
587 payload_len = recv_buf->hdr.data_size;
589 if (payload_len > HVS_MTU_SIZE)
592 if (payload_len == 0)
593 hvs->vsk->peer_shutdown |= SEND_SHUTDOWN;
595 hvs->recv_data_len = payload_len;
596 hvs->recv_data_off = 0;
601 static ssize_t hvs_stream_dequeue(struct vsock_sock *vsk, struct msghdr *msg,
602 size_t len, int flags)
604 struct hvsock *hvs = vsk->trans;
605 bool need_refill = !hvs->recv_desc;
606 struct hvs_recv_buf *recv_buf;
610 if (flags & MSG_PEEK)
614 hvs->recv_desc = hv_pkt_iter_first(hvs->chan);
615 ret = hvs_update_recv_data(hvs);
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);
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);
635 hvs->recv_data_off += to_read;
641 static ssize_t hvs_stream_enqueue(struct vsock_sock *vsk, struct msghdr *msg,
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;
649 BUILD_BUG_ON(sizeof(*send_buf) != PAGE_SIZE_4K);
651 send_buf = kmalloc(sizeof(*send_buf), GFP_KERNEL);
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);
659 ret = memcpy_from_msg(send_buf->data, msg, to_write);
663 ret = hvs_send_data(hvs->chan, send_buf, to_write);
673 static s64 hvs_stream_has_data(struct vsock_sock *vsk)
675 struct hvsock *hvs = vsk->trans;
678 if (hvs->recv_data_len > 0)
681 switch (hvs_channel_readable_payload(hvs->chan)) {
686 vsk->peer_shutdown |= SEND_SHUTDOWN;
697 static s64 hvs_stream_has_space(struct vsock_sock *vsk)
699 struct hvsock *hvs = vsk->trans;
700 struct vmbus_channel *chan = hvs->chan;
703 ret = hvs_channel_writable_bytes(chan);
705 hvs_clear_channel_pending_send_size(chan);
707 /* See hvs_channel_cb() */
708 hvs_set_channel_pending_send_size(chan);
710 /* Re-check the writable bytes to avoid race */
711 ret = hvs_channel_writable_bytes(chan);
713 hvs_clear_channel_pending_send_size(chan);
719 static u64 hvs_stream_rcvhiwat(struct vsock_sock *vsk)
721 return HVS_MTU_SIZE + 1;
724 static bool hvs_stream_is_active(struct vsock_sock *vsk)
726 struct hvsock *hvs = vsk->trans;
728 return hvs->chan != NULL;
731 static bool hvs_stream_allow(u32 cid, u32 port)
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.
737 * Perform this check in the guest so an immediate error is produced
738 * instead of a timeout.
740 if (port > MAX_HOST_LISTEN_PORT)
743 if (cid == VMADDR_CID_HOST)
750 int hvs_notify_poll_in(struct vsock_sock *vsk, size_t target, bool *readable)
752 struct hvsock *hvs = vsk->trans;
754 *readable = hvs_channel_readable(hvs->chan);
759 int hvs_notify_poll_out(struct vsock_sock *vsk, size_t target, bool *writable)
761 *writable = hvs_stream_has_space(vsk) > 0;
767 int hvs_notify_recv_init(struct vsock_sock *vsk, size_t target,
768 struct vsock_transport_recv_notify_data *d)
774 int hvs_notify_recv_pre_block(struct vsock_sock *vsk, size_t target,
775 struct vsock_transport_recv_notify_data *d)
781 int hvs_notify_recv_pre_dequeue(struct vsock_sock *vsk, size_t target,
782 struct vsock_transport_recv_notify_data *d)
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)
796 int hvs_notify_send_init(struct vsock_sock *vsk,
797 struct vsock_transport_send_notify_data *d)
803 int hvs_notify_send_pre_block(struct vsock_sock *vsk,
804 struct vsock_transport_send_notify_data *d)
810 int hvs_notify_send_pre_enqueue(struct vsock_sock *vsk,
811 struct vsock_transport_send_notify_data *d)
817 int hvs_notify_send_post_enqueue(struct vsock_sock *vsk, ssize_t written,
818 struct vsock_transport_send_notify_data *d)
823 static void hvs_set_buffer_size(struct vsock_sock *vsk, u64 val)
828 static void hvs_set_min_buffer_size(struct vsock_sock *vsk, u64 val)
833 static void hvs_set_max_buffer_size(struct vsock_sock *vsk, u64 val)
838 static u64 hvs_get_buffer_size(struct vsock_sock *vsk)
843 static u64 hvs_get_min_buffer_size(struct vsock_sock *vsk)
848 static u64 hvs_get_max_buffer_size(struct vsock_sock *vsk)
853 static struct vsock_transport hvs_transport = {
854 .get_local_cid = hvs_get_local_cid,
856 .init = hvs_sock_init,
857 .destruct = hvs_destruct,
858 .release = hvs_release,
859 .connect = hvs_connect,
860 .shutdown = hvs_shutdown,
862 .dgram_bind = hvs_dgram_bind,
863 .dgram_dequeue = hvs_dgram_dequeue,
864 .dgram_enqueue = hvs_dgram_enqueue,
865 .dgram_allow = hvs_dgram_allow,
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,
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,
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,
894 static int hvs_probe(struct hv_device *hdev,
895 const struct hv_vmbus_device_id *dev_id)
897 struct vmbus_channel *chan = hdev->channel;
899 hvs_open_connection(chan);
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().
909 static int hvs_remove(struct hv_device *hdev)
911 struct vmbus_channel *chan = hdev->channel;
918 /* This isn't really used. See vmbus_match() and vmbus_probe() */
919 static const struct hv_vmbus_device_id id_table[] = {
923 static struct hv_driver hvs_drv = {
926 .id_table = id_table,
928 .remove = hvs_remove,
931 static int __init hvs_init(void)
935 if (vmbus_proto_version < VERSION_WIN10)
938 ret = vmbus_driver_register(&hvs_drv);
942 ret = vsock_core_init(&hvs_transport);
944 vmbus_driver_unregister(&hvs_drv);
951 static void __exit hvs_exit(void)
954 vmbus_driver_unregister(&hvs_drv);
957 module_init(hvs_init);
958 module_exit(hvs_exit);
960 MODULE_DESCRIPTION("Hyper-V Sockets");
961 MODULE_VERSION("1.0.0");
962 MODULE_LICENSE("GPL");
963 MODULE_ALIAS_NETPROTO(PF_VSOCK);