12025cf9eSThomas Gleixner // SPDX-License-Identifier: GPL-2.0-only
2ae0078fcSDexuan Cui /*
3ae0078fcSDexuan Cui  * Hyper-V transport for vsock
4ae0078fcSDexuan Cui  *
5ae0078fcSDexuan Cui  * Hyper-V Sockets supplies a byte-stream based communication mechanism
6ae0078fcSDexuan Cui  * between the host and the VM. This driver implements the necessary
7ae0078fcSDexuan Cui  * support in the VM by introducing the new vsock transport.
8ae0078fcSDexuan Cui  *
9ae0078fcSDexuan Cui  * Copyright (c) 2017, Microsoft Corporation.
10ae0078fcSDexuan Cui  */
11ae0078fcSDexuan Cui #include <linux/module.h>
12ae0078fcSDexuan Cui #include <linux/vmalloc.h>
13ae0078fcSDexuan Cui #include <linux/hyperv.h>
14ae0078fcSDexuan Cui #include <net/sock.h>
15ae0078fcSDexuan Cui #include <net/af_vsock.h>
16ae0078fcSDexuan Cui 
17ac383f58SSunil Muthuswamy /* Older (VMBUS version 'VERSION_WIN10' or before) Windows hosts have some
18ac383f58SSunil Muthuswamy  * stricter requirements on the hv_sock ring buffer size of six 4K pages. Newer
19ac383f58SSunil Muthuswamy  * hosts don't have this limitation; but, keep the defaults the same for compat.
20ae0078fcSDexuan Cui  */
21ae0078fcSDexuan Cui #define PAGE_SIZE_4K		4096
22ae0078fcSDexuan Cui #define RINGBUFFER_HVS_RCV_SIZE (PAGE_SIZE_4K * 6)
23ae0078fcSDexuan Cui #define RINGBUFFER_HVS_SND_SIZE (PAGE_SIZE_4K * 6)
24ac383f58SSunil Muthuswamy #define RINGBUFFER_HVS_MAX_SIZE (PAGE_SIZE_4K * 64)
25ae0078fcSDexuan Cui 
26ae0078fcSDexuan Cui /* The MTU is 16KB per the host side's design */
27ae0078fcSDexuan Cui #define HVS_MTU_SIZE		(1024 * 16)
28ae0078fcSDexuan Cui 
29a9eeb998SSunil Muthuswamy /* How long to wait for graceful shutdown of a connection */
30a9eeb998SSunil Muthuswamy #define HVS_CLOSE_TIMEOUT (8 * HZ)
31a9eeb998SSunil Muthuswamy 
32ae0078fcSDexuan Cui struct vmpipe_proto_header {
33ae0078fcSDexuan Cui 	u32 pkt_type;
34ae0078fcSDexuan Cui 	u32 data_size;
35ae0078fcSDexuan Cui };
36ae0078fcSDexuan Cui 
37ae0078fcSDexuan Cui /* For recv, we use the VMBus in-place packet iterator APIs to directly copy
38ae0078fcSDexuan Cui  * data from the ringbuffer into the userspace buffer.
39ae0078fcSDexuan Cui  */
40ae0078fcSDexuan Cui struct hvs_recv_buf {
41ae0078fcSDexuan Cui 	/* The header before the payload data */
42ae0078fcSDexuan Cui 	struct vmpipe_proto_header hdr;
43ae0078fcSDexuan Cui 
44ae0078fcSDexuan Cui 	/* The payload */
45ae0078fcSDexuan Cui 	u8 data[HVS_MTU_SIZE];
46ae0078fcSDexuan Cui };
47ae0078fcSDexuan Cui 
48ae0078fcSDexuan Cui /* We can send up to HVS_MTU_SIZE bytes of payload to the host, but let's use
4914a1eaa8SSunil Muthuswamy  * a smaller size, i.e. HVS_SEND_BUF_SIZE, to maximize concurrency between the
5014a1eaa8SSunil Muthuswamy  * guest and the host processing as one VMBUS packet is the smallest processing
5114a1eaa8SSunil Muthuswamy  * unit.
52ae0078fcSDexuan Cui  *
53ae0078fcSDexuan Cui  * Note: the buffer can be eliminated in the future when we add new VMBus
54ae0078fcSDexuan Cui  * ringbuffer APIs that allow us to directly copy data from userspace buffer
55ae0078fcSDexuan Cui  * to VMBus ringbuffer.
56ae0078fcSDexuan Cui  */
57ae0078fcSDexuan Cui #define HVS_SEND_BUF_SIZE (PAGE_SIZE_4K - sizeof(struct vmpipe_proto_header))
58ae0078fcSDexuan Cui 
59ae0078fcSDexuan Cui struct hvs_send_buf {
60ae0078fcSDexuan Cui 	/* The header before the payload data */
61ae0078fcSDexuan Cui 	struct vmpipe_proto_header hdr;
62ae0078fcSDexuan Cui 
63ae0078fcSDexuan Cui 	/* The payload */
64ae0078fcSDexuan Cui 	u8 data[HVS_SEND_BUF_SIZE];
65ae0078fcSDexuan Cui };
66ae0078fcSDexuan Cui 
67ae0078fcSDexuan Cui #define HVS_HEADER_LEN	(sizeof(struct vmpacket_descriptor) + \
68ae0078fcSDexuan Cui 			 sizeof(struct vmpipe_proto_header))
69ae0078fcSDexuan Cui 
70ae0078fcSDexuan Cui /* See 'prev_indices' in hv_ringbuffer_read(), hv_ringbuffer_write(), and
71ae0078fcSDexuan Cui  * __hv_pkt_iter_next().
72ae0078fcSDexuan Cui  */
73ae0078fcSDexuan Cui #define VMBUS_PKT_TRAILER_SIZE	(sizeof(u64))
74ae0078fcSDexuan Cui 
75ae0078fcSDexuan Cui #define HVS_PKT_LEN(payload_len)	(HVS_HEADER_LEN + \
76ae0078fcSDexuan Cui 					 ALIGN((payload_len), 8) + \
77ae0078fcSDexuan Cui 					 VMBUS_PKT_TRAILER_SIZE)
78ae0078fcSDexuan Cui 
79ae0078fcSDexuan Cui union hvs_service_id {
80ae0078fcSDexuan Cui 	uuid_le	srv_id;
81ae0078fcSDexuan Cui 
82ae0078fcSDexuan Cui 	struct {
83ae0078fcSDexuan Cui 		unsigned int svm_port;
84ae0078fcSDexuan Cui 		unsigned char b[sizeof(uuid_le) - sizeof(unsigned int)];
85ae0078fcSDexuan Cui 	};
86ae0078fcSDexuan Cui };
87ae0078fcSDexuan Cui 
88ae0078fcSDexuan Cui /* Per-socket state (accessed via vsk->trans) */
89ae0078fcSDexuan Cui struct hvsock {
90ae0078fcSDexuan Cui 	struct vsock_sock *vsk;
91ae0078fcSDexuan Cui 
92ae0078fcSDexuan Cui 	uuid_le vm_srv_id;
93ae0078fcSDexuan Cui 	uuid_le host_srv_id;
94ae0078fcSDexuan Cui 
95ae0078fcSDexuan Cui 	struct vmbus_channel *chan;
96ae0078fcSDexuan Cui 	struct vmpacket_descriptor *recv_desc;
97ae0078fcSDexuan Cui 
98ae0078fcSDexuan Cui 	/* The length of the payload not delivered to userland yet */
99ae0078fcSDexuan Cui 	u32 recv_data_len;
100ae0078fcSDexuan Cui 	/* The offset of the payload */
101ae0078fcSDexuan Cui 	u32 recv_data_off;
102ae0078fcSDexuan Cui 
103ae0078fcSDexuan Cui 	/* Have we sent the zero-length packet (FIN)? */
104ae0078fcSDexuan Cui 	bool fin_sent;
105ae0078fcSDexuan Cui };
106ae0078fcSDexuan Cui 
107ae0078fcSDexuan Cui /* In the VM, we support Hyper-V Sockets with AF_VSOCK, and the endpoint is
108ae0078fcSDexuan Cui  * <cid, port> (see struct sockaddr_vm). Note: cid is not really used here:
109ae0078fcSDexuan Cui  * when we write apps to connect to the host, we can only use VMADDR_CID_ANY
110ae0078fcSDexuan Cui  * or VMADDR_CID_HOST (both are equivalent) as the remote cid, and when we
111ae0078fcSDexuan Cui  * write apps to bind() & listen() in the VM, we can only use VMADDR_CID_ANY
112ae0078fcSDexuan Cui  * as the local cid.
113ae0078fcSDexuan Cui  *
114ae0078fcSDexuan Cui  * On the host, Hyper-V Sockets are supported by Winsock AF_HYPERV:
115ae0078fcSDexuan Cui  * https://docs.microsoft.com/en-us/virtualization/hyper-v-on-windows/user-
116ae0078fcSDexuan Cui  * guide/make-integration-service, and the endpoint is <VmID, ServiceId> with
117ae0078fcSDexuan Cui  * the below sockaddr:
118ae0078fcSDexuan Cui  *
119ae0078fcSDexuan Cui  * struct SOCKADDR_HV
120ae0078fcSDexuan Cui  * {
121ae0078fcSDexuan Cui  *    ADDRESS_FAMILY Family;
122ae0078fcSDexuan Cui  *    USHORT Reserved;
123ae0078fcSDexuan Cui  *    GUID VmId;
124ae0078fcSDexuan Cui  *    GUID ServiceId;
125ae0078fcSDexuan Cui  * };
126ae0078fcSDexuan Cui  * Note: VmID is not used by Linux VM and actually it isn't transmitted via
127ae0078fcSDexuan Cui  * VMBus, because here it's obvious the host and the VM can easily identify
128ae0078fcSDexuan Cui  * each other. Though the VmID is useful on the host, especially in the case
129ae0078fcSDexuan Cui  * of Windows container, Linux VM doesn't need it at all.
130ae0078fcSDexuan Cui  *
131ae0078fcSDexuan Cui  * To make use of the AF_VSOCK infrastructure in Linux VM, we have to limit
132ae0078fcSDexuan Cui  * the available GUID space of SOCKADDR_HV so that we can create a mapping
133ae0078fcSDexuan Cui  * between AF_VSOCK port and SOCKADDR_HV Service GUID. The rule of writing
134ae0078fcSDexuan Cui  * Hyper-V Sockets apps on the host and in Linux VM is:
135ae0078fcSDexuan Cui  *
136ae0078fcSDexuan Cui  ****************************************************************************
137ae0078fcSDexuan Cui  * The only valid Service GUIDs, from the perspectives of both the host and *
138ae0078fcSDexuan Cui  * Linux VM, that can be connected by the other end, must conform to this   *
139ae0078fcSDexuan Cui  * format: <port>-facb-11e6-bd58-64006a7986d3, and the "port" must be in    *
140ae0078fcSDexuan Cui  * this range [0, 0x7FFFFFFF].                                              *
141ae0078fcSDexuan Cui  ****************************************************************************
142ae0078fcSDexuan Cui  *
143ae0078fcSDexuan Cui  * When we write apps on the host to connect(), the GUID ServiceID is used.
144ae0078fcSDexuan Cui  * When we write apps in Linux VM to connect(), we only need to specify the
145ae0078fcSDexuan Cui  * port and the driver will form the GUID and use that to request the host.
146ae0078fcSDexuan Cui  *
147ae0078fcSDexuan Cui  * From the perspective of Linux VM:
148ae0078fcSDexuan Cui  * 1. the local ephemeral port (i.e. the local auto-bound port when we call
149ae0078fcSDexuan Cui  * connect() without explicit bind()) is generated by __vsock_bind_stream(),
150ae0078fcSDexuan Cui  * and the range is [1024, 0xFFFFFFFF).
151ae0078fcSDexuan Cui  * 2. the remote ephemeral port (i.e. the auto-generated remote port for
152ae0078fcSDexuan Cui  * a connect request initiated by the host's connect()) is generated by
153ae0078fcSDexuan Cui  * hvs_remote_addr_init() and the range is [0x80000000, 0xFFFFFFFF).
154ae0078fcSDexuan Cui  */
155ae0078fcSDexuan Cui 
156ae0078fcSDexuan Cui #define MAX_LISTEN_PORT			((u32)0x7FFFFFFF)
157ae0078fcSDexuan Cui #define MAX_VM_LISTEN_PORT		MAX_LISTEN_PORT
158ae0078fcSDexuan Cui #define MAX_HOST_LISTEN_PORT		MAX_LISTEN_PORT
159ae0078fcSDexuan Cui #define MIN_HOST_EPHEMERAL_PORT		(MAX_HOST_LISTEN_PORT + 1)
160ae0078fcSDexuan Cui 
161ae0078fcSDexuan Cui /* 00000000-facb-11e6-bd58-64006a7986d3 */
162ae0078fcSDexuan Cui static const uuid_le srv_id_template =
163ae0078fcSDexuan Cui 	UUID_LE(0x00000000, 0xfacb, 0x11e6, 0xbd, 0x58,
164ae0078fcSDexuan Cui 		0x64, 0x00, 0x6a, 0x79, 0x86, 0xd3);
165ae0078fcSDexuan Cui 
166ae0078fcSDexuan Cui static bool is_valid_srv_id(const uuid_le *id)
167ae0078fcSDexuan Cui {
168ae0078fcSDexuan Cui 	return !memcmp(&id->b[4], &srv_id_template.b[4], sizeof(uuid_le) - 4);
169ae0078fcSDexuan Cui }
170ae0078fcSDexuan Cui 
171ae0078fcSDexuan Cui static unsigned int get_port_by_srv_id(const uuid_le *svr_id)
172ae0078fcSDexuan Cui {
173ae0078fcSDexuan Cui 	return *((unsigned int *)svr_id);
174ae0078fcSDexuan Cui }
175ae0078fcSDexuan Cui 
176ae0078fcSDexuan Cui static void hvs_addr_init(struct sockaddr_vm *addr, const uuid_le *svr_id)
177ae0078fcSDexuan Cui {
178ae0078fcSDexuan Cui 	unsigned int port = get_port_by_srv_id(svr_id);
179ae0078fcSDexuan Cui 
180ae0078fcSDexuan Cui 	vsock_addr_init(addr, VMADDR_CID_ANY, port);
181ae0078fcSDexuan Cui }
182ae0078fcSDexuan Cui 
183ae0078fcSDexuan Cui static void hvs_remote_addr_init(struct sockaddr_vm *remote,
184ae0078fcSDexuan Cui 				 struct sockaddr_vm *local)
185ae0078fcSDexuan Cui {
186ae0078fcSDexuan Cui 	static u32 host_ephemeral_port = MIN_HOST_EPHEMERAL_PORT;
187ae0078fcSDexuan Cui 	struct sock *sk;
188ae0078fcSDexuan Cui 
189ae0078fcSDexuan Cui 	vsock_addr_init(remote, VMADDR_CID_ANY, VMADDR_PORT_ANY);
190ae0078fcSDexuan Cui 
191ae0078fcSDexuan Cui 	while (1) {
192ae0078fcSDexuan Cui 		/* Wrap around ? */
193ae0078fcSDexuan Cui 		if (host_ephemeral_port < MIN_HOST_EPHEMERAL_PORT ||
194ae0078fcSDexuan Cui 		    host_ephemeral_port == VMADDR_PORT_ANY)
195ae0078fcSDexuan Cui 			host_ephemeral_port = MIN_HOST_EPHEMERAL_PORT;
196ae0078fcSDexuan Cui 
197ae0078fcSDexuan Cui 		remote->svm_port = host_ephemeral_port++;
198ae0078fcSDexuan Cui 
199ae0078fcSDexuan Cui 		sk = vsock_find_connected_socket(remote, local);
200ae0078fcSDexuan Cui 		if (!sk) {
201ae0078fcSDexuan Cui 			/* Found an available ephemeral port */
202ae0078fcSDexuan Cui 			return;
203ae0078fcSDexuan Cui 		}
204ae0078fcSDexuan Cui 
205ae0078fcSDexuan Cui 		/* Release refcnt got in vsock_find_connected_socket */
206ae0078fcSDexuan Cui 		sock_put(sk);
207ae0078fcSDexuan Cui 	}
208ae0078fcSDexuan Cui }
209ae0078fcSDexuan Cui 
210ae0078fcSDexuan Cui static void hvs_set_channel_pending_send_size(struct vmbus_channel *chan)
211ae0078fcSDexuan Cui {
212ae0078fcSDexuan Cui 	set_channel_pending_send_size(chan,
213ae0078fcSDexuan Cui 				      HVS_PKT_LEN(HVS_SEND_BUF_SIZE));
214ae0078fcSDexuan Cui 
215ae0078fcSDexuan Cui 	virt_mb();
216ae0078fcSDexuan Cui }
217ae0078fcSDexuan Cui 
218ae0078fcSDexuan Cui static bool hvs_channel_readable(struct vmbus_channel *chan)
219ae0078fcSDexuan Cui {
220ae0078fcSDexuan Cui 	u32 readable = hv_get_bytes_to_read(&chan->inbound);
221ae0078fcSDexuan Cui 
222ae0078fcSDexuan Cui 	/* 0-size payload means FIN */
223ae0078fcSDexuan Cui 	return readable >= HVS_PKT_LEN(0);
224ae0078fcSDexuan Cui }
225ae0078fcSDexuan Cui 
226ae0078fcSDexuan Cui static int hvs_channel_readable_payload(struct vmbus_channel *chan)
227ae0078fcSDexuan Cui {
228ae0078fcSDexuan Cui 	u32 readable = hv_get_bytes_to_read(&chan->inbound);
229ae0078fcSDexuan Cui 
230ae0078fcSDexuan Cui 	if (readable > HVS_PKT_LEN(0)) {
231ae0078fcSDexuan Cui 		/* At least we have 1 byte to read. We don't need to return
232ae0078fcSDexuan Cui 		 * the exact readable bytes: see vsock_stream_recvmsg() ->
233ae0078fcSDexuan Cui 		 * vsock_stream_has_data().
234ae0078fcSDexuan Cui 		 */
235ae0078fcSDexuan Cui 		return 1;
236ae0078fcSDexuan Cui 	}
237ae0078fcSDexuan Cui 
238ae0078fcSDexuan Cui 	if (readable == HVS_PKT_LEN(0)) {
239ae0078fcSDexuan Cui 		/* 0-size payload means FIN */
240ae0078fcSDexuan Cui 		return 0;
241ae0078fcSDexuan Cui 	}
242ae0078fcSDexuan Cui 
243ae0078fcSDexuan Cui 	/* No payload or FIN */
244ae0078fcSDexuan Cui 	return -1;
245ae0078fcSDexuan Cui }
246ae0078fcSDexuan Cui 
247ae0078fcSDexuan Cui static size_t hvs_channel_writable_bytes(struct vmbus_channel *chan)
248ae0078fcSDexuan Cui {
249ae0078fcSDexuan Cui 	u32 writeable = hv_get_bytes_to_write(&chan->outbound);
250ae0078fcSDexuan Cui 	size_t ret;
251ae0078fcSDexuan Cui 
252ae0078fcSDexuan Cui 	/* The ringbuffer mustn't be 100% full, and we should reserve a
253ae0078fcSDexuan Cui 	 * zero-length-payload packet for the FIN: see hv_ringbuffer_write()
254ae0078fcSDexuan Cui 	 * and hvs_shutdown().
255ae0078fcSDexuan Cui 	 */
256ae0078fcSDexuan Cui 	if (writeable <= HVS_PKT_LEN(1) + HVS_PKT_LEN(0))
257ae0078fcSDexuan Cui 		return 0;
258ae0078fcSDexuan Cui 
259ae0078fcSDexuan Cui 	ret = writeable - HVS_PKT_LEN(1) - HVS_PKT_LEN(0);
260ae0078fcSDexuan Cui 
261ae0078fcSDexuan Cui 	return round_down(ret, 8);
262ae0078fcSDexuan Cui }
263ae0078fcSDexuan Cui 
264ae0078fcSDexuan Cui static int hvs_send_data(struct vmbus_channel *chan,
265ae0078fcSDexuan Cui 			 struct hvs_send_buf *send_buf, size_t to_write)
266ae0078fcSDexuan Cui {
267ae0078fcSDexuan Cui 	send_buf->hdr.pkt_type = 1;
268ae0078fcSDexuan Cui 	send_buf->hdr.data_size = to_write;
269ae0078fcSDexuan Cui 	return vmbus_sendpacket(chan, &send_buf->hdr,
270ae0078fcSDexuan Cui 				sizeof(send_buf->hdr) + to_write,
271ae0078fcSDexuan Cui 				0, VM_PKT_DATA_INBAND, 0);
272ae0078fcSDexuan Cui }
273ae0078fcSDexuan Cui 
274ae0078fcSDexuan Cui static void hvs_channel_cb(void *ctx)
275ae0078fcSDexuan Cui {
276ae0078fcSDexuan Cui 	struct sock *sk = (struct sock *)ctx;
277ae0078fcSDexuan Cui 	struct vsock_sock *vsk = vsock_sk(sk);
278ae0078fcSDexuan Cui 	struct hvsock *hvs = vsk->trans;
279ae0078fcSDexuan Cui 	struct vmbus_channel *chan = hvs->chan;
280ae0078fcSDexuan Cui 
281ae0078fcSDexuan Cui 	if (hvs_channel_readable(chan))
282ae0078fcSDexuan Cui 		sk->sk_data_ready(sk);
283ae0078fcSDexuan Cui 
284ae0078fcSDexuan Cui 	if (hv_get_bytes_to_write(&chan->outbound) > 0)
285ae0078fcSDexuan Cui 		sk->sk_write_space(sk);
286ae0078fcSDexuan Cui }
287ae0078fcSDexuan Cui 
288a9eeb998SSunil Muthuswamy static void hvs_do_close_lock_held(struct vsock_sock *vsk,
289a9eeb998SSunil Muthuswamy 				   bool cancel_timeout)
290a9eeb998SSunil Muthuswamy {
291a9eeb998SSunil Muthuswamy 	struct sock *sk = sk_vsock(vsk);
292a9eeb998SSunil Muthuswamy 
293a9eeb998SSunil Muthuswamy 	sock_set_flag(sk, SOCK_DONE);
294a9eeb998SSunil Muthuswamy 	vsk->peer_shutdown = SHUTDOWN_MASK;
295a9eeb998SSunil Muthuswamy 	if (vsock_stream_has_data(vsk) <= 0)
296a9eeb998SSunil Muthuswamy 		sk->sk_state = TCP_CLOSING;
297a9eeb998SSunil Muthuswamy 	sk->sk_state_change(sk);
298a9eeb998SSunil Muthuswamy 	if (vsk->close_work_scheduled &&
299a9eeb998SSunil Muthuswamy 	    (!cancel_timeout || cancel_delayed_work(&vsk->close_work))) {
300a9eeb998SSunil Muthuswamy 		vsk->close_work_scheduled = false;
301a9eeb998SSunil Muthuswamy 		vsock_remove_sock(vsk);
302a9eeb998SSunil Muthuswamy 
303a9eeb998SSunil Muthuswamy 		/* Release the reference taken while scheduling the timeout */
304a9eeb998SSunil Muthuswamy 		sock_put(sk);
305a9eeb998SSunil Muthuswamy 	}
306a9eeb998SSunil Muthuswamy }
307a9eeb998SSunil Muthuswamy 
308ae0078fcSDexuan Cui static void hvs_close_connection(struct vmbus_channel *chan)
309ae0078fcSDexuan Cui {
310ae0078fcSDexuan Cui 	struct sock *sk = get_per_channel_state(chan);
311ae0078fcSDexuan Cui 
312b4562ca7SDexuan Cui 	lock_sock(sk);
313a9eeb998SSunil Muthuswamy 	hvs_do_close_lock_held(vsock_sk(sk), true);
314b4562ca7SDexuan Cui 	release_sock(sk);
315685703b4SDexuan Cui 
316685703b4SDexuan Cui 	/* Release the refcnt for the channel that's opened in
317685703b4SDexuan Cui 	 * hvs_open_connection().
318685703b4SDexuan Cui 	 */
319685703b4SDexuan Cui 	sock_put(sk);
320ae0078fcSDexuan Cui }
321ae0078fcSDexuan Cui 
322ae0078fcSDexuan Cui static void hvs_open_connection(struct vmbus_channel *chan)
323ae0078fcSDexuan Cui {
324ae0078fcSDexuan Cui 	uuid_le *if_instance, *if_type;
325ae0078fcSDexuan Cui 	unsigned char conn_from_host;
326ae0078fcSDexuan Cui 
327ae0078fcSDexuan Cui 	struct sockaddr_vm addr;
328ae0078fcSDexuan Cui 	struct sock *sk, *new = NULL;
329ac383f58SSunil Muthuswamy 	struct vsock_sock *vnew = NULL;
330ac383f58SSunil Muthuswamy 	struct hvsock *hvs = NULL;
331ac383f58SSunil Muthuswamy 	struct hvsock *hvs_new = NULL;
332ac383f58SSunil Muthuswamy 	int rcvbuf;
333ae0078fcSDexuan Cui 	int ret;
334ac383f58SSunil Muthuswamy 	int sndbuf;
335ae0078fcSDexuan Cui 
336ae0078fcSDexuan Cui 	if_type = &chan->offermsg.offer.if_type;
337ae0078fcSDexuan Cui 	if_instance = &chan->offermsg.offer.if_instance;
338ae0078fcSDexuan Cui 	conn_from_host = chan->offermsg.offer.u.pipe.user_def[0];
339ae0078fcSDexuan Cui 
340ae0078fcSDexuan Cui 	/* The host or the VM should only listen on a port in
341ae0078fcSDexuan Cui 	 * [0, MAX_LISTEN_PORT]
342ae0078fcSDexuan Cui 	 */
343ae0078fcSDexuan Cui 	if (!is_valid_srv_id(if_type) ||
344ae0078fcSDexuan Cui 	    get_port_by_srv_id(if_type) > MAX_LISTEN_PORT)
345ae0078fcSDexuan Cui 		return;
346ae0078fcSDexuan Cui 
347ae0078fcSDexuan Cui 	hvs_addr_init(&addr, conn_from_host ? if_type : if_instance);
348ae0078fcSDexuan Cui 	sk = vsock_find_bound_socket(&addr);
349ae0078fcSDexuan Cui 	if (!sk)
350ae0078fcSDexuan Cui 		return;
351ae0078fcSDexuan Cui 
352b4562ca7SDexuan Cui 	lock_sock(sk);
3533b4477d2SStefan Hajnoczi 	if ((conn_from_host && sk->sk_state != TCP_LISTEN) ||
3543b4477d2SStefan Hajnoczi 	    (!conn_from_host && sk->sk_state != TCP_SYN_SENT))
355ae0078fcSDexuan Cui 		goto out;
356ae0078fcSDexuan Cui 
357ae0078fcSDexuan Cui 	if (conn_from_host) {
358ae0078fcSDexuan Cui 		if (sk->sk_ack_backlog >= sk->sk_max_ack_backlog)
359ae0078fcSDexuan Cui 			goto out;
360ae0078fcSDexuan Cui 
361ae0078fcSDexuan Cui 		new = __vsock_create(sock_net(sk), NULL, sk, GFP_KERNEL,
362ae0078fcSDexuan Cui 				     sk->sk_type, 0);
363ae0078fcSDexuan Cui 		if (!new)
364ae0078fcSDexuan Cui 			goto out;
365ae0078fcSDexuan Cui 
3663b4477d2SStefan Hajnoczi 		new->sk_state = TCP_SYN_SENT;
367ae0078fcSDexuan Cui 		vnew = vsock_sk(new);
368ae0078fcSDexuan Cui 		hvs_new = vnew->trans;
369ae0078fcSDexuan Cui 		hvs_new->chan = chan;
370ae0078fcSDexuan Cui 	} else {
371ae0078fcSDexuan Cui 		hvs = vsock_sk(sk)->trans;
372ae0078fcSDexuan Cui 		hvs->chan = chan;
373ae0078fcSDexuan Cui 	}
374ae0078fcSDexuan Cui 
375ae0078fcSDexuan Cui 	set_channel_read_mode(chan, HV_CALL_DIRECT);
376ac383f58SSunil Muthuswamy 
377ac383f58SSunil Muthuswamy 	/* Use the socket buffer sizes as hints for the VMBUS ring size. For
378ac383f58SSunil Muthuswamy 	 * server side sockets, 'sk' is the parent socket and thus, this will
379ac383f58SSunil Muthuswamy 	 * allow the child sockets to inherit the size from the parent. Keep
380ac383f58SSunil Muthuswamy 	 * the mins to the default value and align to page size as per VMBUS
381ac383f58SSunil Muthuswamy 	 * requirements.
382ac383f58SSunil Muthuswamy 	 * For the max, the socket core library will limit the socket buffer
383ac383f58SSunil Muthuswamy 	 * size that can be set by the user, but, since currently, the hv_sock
384ac383f58SSunil Muthuswamy 	 * VMBUS ring buffer is physically contiguous allocation, restrict it
385ac383f58SSunil Muthuswamy 	 * further.
386ac383f58SSunil Muthuswamy 	 * Older versions of hv_sock host side code cannot handle bigger VMBUS
387ac383f58SSunil Muthuswamy 	 * ring buffer size. Use the version number to limit the change to newer
388ac383f58SSunil Muthuswamy 	 * versions.
389ac383f58SSunil Muthuswamy 	 */
390ac383f58SSunil Muthuswamy 	if (vmbus_proto_version < VERSION_WIN10_V5) {
391ac383f58SSunil Muthuswamy 		sndbuf = RINGBUFFER_HVS_SND_SIZE;
392ac383f58SSunil Muthuswamy 		rcvbuf = RINGBUFFER_HVS_RCV_SIZE;
393ac383f58SSunil Muthuswamy 	} else {
394ac383f58SSunil Muthuswamy 		sndbuf = max_t(int, sk->sk_sndbuf, RINGBUFFER_HVS_SND_SIZE);
395ac383f58SSunil Muthuswamy 		sndbuf = min_t(int, sndbuf, RINGBUFFER_HVS_MAX_SIZE);
396ac383f58SSunil Muthuswamy 		sndbuf = ALIGN(sndbuf, PAGE_SIZE);
397ac383f58SSunil Muthuswamy 		rcvbuf = max_t(int, sk->sk_rcvbuf, RINGBUFFER_HVS_RCV_SIZE);
398ac383f58SSunil Muthuswamy 		rcvbuf = min_t(int, rcvbuf, RINGBUFFER_HVS_MAX_SIZE);
399ac383f58SSunil Muthuswamy 		rcvbuf = ALIGN(rcvbuf, PAGE_SIZE);
400ac383f58SSunil Muthuswamy 	}
401ac383f58SSunil Muthuswamy 
402ac383f58SSunil Muthuswamy 	ret = vmbus_open(chan, sndbuf, rcvbuf, NULL, 0, hvs_channel_cb,
403ac383f58SSunil Muthuswamy 			 conn_from_host ? new : sk);
404ae0078fcSDexuan Cui 	if (ret != 0) {
405ae0078fcSDexuan Cui 		if (conn_from_host) {
406ae0078fcSDexuan Cui 			hvs_new->chan = NULL;
407ae0078fcSDexuan Cui 			sock_put(new);
408ae0078fcSDexuan Cui 		} else {
409ae0078fcSDexuan Cui 			hvs->chan = NULL;
410ae0078fcSDexuan Cui 		}
411ae0078fcSDexuan Cui 		goto out;
412ae0078fcSDexuan Cui 	}
413ae0078fcSDexuan Cui 
414ae0078fcSDexuan Cui 	set_per_channel_state(chan, conn_from_host ? new : sk);
415685703b4SDexuan Cui 
416685703b4SDexuan Cui 	/* This reference will be dropped by hvs_close_connection(). */
417685703b4SDexuan Cui 	sock_hold(conn_from_host ? new : sk);
418ae0078fcSDexuan Cui 	vmbus_set_chn_rescind_callback(chan, hvs_close_connection);
419ae0078fcSDexuan Cui 
420cb359b60SSunil Muthuswamy 	/* Set the pending send size to max packet size to always get
421cb359b60SSunil Muthuswamy 	 * notifications from the host when there is enough writable space.
422cb359b60SSunil Muthuswamy 	 * The host is optimized to send notifications only when the pending
423cb359b60SSunil Muthuswamy 	 * size boundary is crossed, and not always.
424cb359b60SSunil Muthuswamy 	 */
425cb359b60SSunil Muthuswamy 	hvs_set_channel_pending_send_size(chan);
426cb359b60SSunil Muthuswamy 
427ae0078fcSDexuan Cui 	if (conn_from_host) {
4283b4477d2SStefan Hajnoczi 		new->sk_state = TCP_ESTABLISHED;
429ae0078fcSDexuan Cui 		sk->sk_ack_backlog++;
430ae0078fcSDexuan Cui 
431ae0078fcSDexuan Cui 		hvs_addr_init(&vnew->local_addr, if_type);
432ae0078fcSDexuan Cui 		hvs_remote_addr_init(&vnew->remote_addr, &vnew->local_addr);
433ae0078fcSDexuan Cui 
434ae0078fcSDexuan Cui 		hvs_new->vm_srv_id = *if_type;
435ae0078fcSDexuan Cui 		hvs_new->host_srv_id = *if_instance;
436ae0078fcSDexuan Cui 
437ae0078fcSDexuan Cui 		vsock_insert_connected(vnew);
438ae0078fcSDexuan Cui 
439ae0078fcSDexuan Cui 		vsock_enqueue_accept(sk, new);
440ae0078fcSDexuan Cui 	} else {
4413b4477d2SStefan Hajnoczi 		sk->sk_state = TCP_ESTABLISHED;
442ae0078fcSDexuan Cui 		sk->sk_socket->state = SS_CONNECTED;
443ae0078fcSDexuan Cui 
444ae0078fcSDexuan Cui 		vsock_insert_connected(vsock_sk(sk));
445ae0078fcSDexuan Cui 	}
446ae0078fcSDexuan Cui 
447ae0078fcSDexuan Cui 	sk->sk_state_change(sk);
448ae0078fcSDexuan Cui 
449ae0078fcSDexuan Cui out:
450ae0078fcSDexuan Cui 	/* Release refcnt obtained when we called vsock_find_bound_socket() */
451ae0078fcSDexuan Cui 	sock_put(sk);
452b4562ca7SDexuan Cui 
453b4562ca7SDexuan Cui 	release_sock(sk);
454ae0078fcSDexuan Cui }
455ae0078fcSDexuan Cui 
456ae0078fcSDexuan Cui static u32 hvs_get_local_cid(void)
457ae0078fcSDexuan Cui {
458ae0078fcSDexuan Cui 	return VMADDR_CID_ANY;
459ae0078fcSDexuan Cui }
460ae0078fcSDexuan Cui 
461ae0078fcSDexuan Cui static int hvs_sock_init(struct vsock_sock *vsk, struct vsock_sock *psk)
462ae0078fcSDexuan Cui {
463ae0078fcSDexuan Cui 	struct hvsock *hvs;
464ac383f58SSunil Muthuswamy 	struct sock *sk = sk_vsock(vsk);
465ae0078fcSDexuan Cui 
466ae0078fcSDexuan Cui 	hvs = kzalloc(sizeof(*hvs), GFP_KERNEL);
467ae0078fcSDexuan Cui 	if (!hvs)
468ae0078fcSDexuan Cui 		return -ENOMEM;
469ae0078fcSDexuan Cui 
470ae0078fcSDexuan Cui 	vsk->trans = hvs;
471ae0078fcSDexuan Cui 	hvs->vsk = vsk;
472ac383f58SSunil Muthuswamy 	sk->sk_sndbuf = RINGBUFFER_HVS_SND_SIZE;
473ac383f58SSunil Muthuswamy 	sk->sk_rcvbuf = RINGBUFFER_HVS_RCV_SIZE;
474ae0078fcSDexuan Cui 	return 0;
475ae0078fcSDexuan Cui }
476ae0078fcSDexuan Cui 
477ae0078fcSDexuan Cui static int hvs_connect(struct vsock_sock *vsk)
478ae0078fcSDexuan Cui {
479ae0078fcSDexuan Cui 	union hvs_service_id vm, host;
480ae0078fcSDexuan Cui 	struct hvsock *h = vsk->trans;
481ae0078fcSDexuan Cui 
482ae0078fcSDexuan Cui 	vm.srv_id = srv_id_template;
483ae0078fcSDexuan Cui 	vm.svm_port = vsk->local_addr.svm_port;
484ae0078fcSDexuan Cui 	h->vm_srv_id = vm.srv_id;
485ae0078fcSDexuan Cui 
486ae0078fcSDexuan Cui 	host.srv_id = srv_id_template;
487ae0078fcSDexuan Cui 	host.svm_port = vsk->remote_addr.svm_port;
488ae0078fcSDexuan Cui 	h->host_srv_id = host.srv_id;
489ae0078fcSDexuan Cui 
490ae0078fcSDexuan Cui 	return vmbus_send_tl_connect_request(&h->vm_srv_id, &h->host_srv_id);
491ae0078fcSDexuan Cui }
492ae0078fcSDexuan Cui 
493a9eeb998SSunil Muthuswamy static void hvs_shutdown_lock_held(struct hvsock *hvs, int mode)
494a9eeb998SSunil Muthuswamy {
495a9eeb998SSunil Muthuswamy 	struct vmpipe_proto_header hdr;
496a9eeb998SSunil Muthuswamy 
497a9eeb998SSunil Muthuswamy 	if (hvs->fin_sent || !hvs->chan)
498a9eeb998SSunil Muthuswamy 		return;
499a9eeb998SSunil Muthuswamy 
500a9eeb998SSunil Muthuswamy 	/* It can't fail: see hvs_channel_writable_bytes(). */
501a9eeb998SSunil Muthuswamy 	(void)hvs_send_data(hvs->chan, (struct hvs_send_buf *)&hdr, 0);
502a9eeb998SSunil Muthuswamy 	hvs->fin_sent = true;
503a9eeb998SSunil Muthuswamy }
504a9eeb998SSunil Muthuswamy 
505ae0078fcSDexuan Cui static int hvs_shutdown(struct vsock_sock *vsk, int mode)
506ae0078fcSDexuan Cui {
507ae0078fcSDexuan Cui 	struct sock *sk = sk_vsock(vsk);
508ae0078fcSDexuan Cui 
509ae0078fcSDexuan Cui 	if (!(mode & SEND_SHUTDOWN))
510ae0078fcSDexuan Cui 		return 0;
511ae0078fcSDexuan Cui 
512ae0078fcSDexuan Cui 	lock_sock(sk);
513a9eeb998SSunil Muthuswamy 	hvs_shutdown_lock_held(vsk->trans, mode);
514ae0078fcSDexuan Cui 	release_sock(sk);
515ae0078fcSDexuan Cui 	return 0;
516ae0078fcSDexuan Cui }
517ae0078fcSDexuan Cui 
518a9eeb998SSunil Muthuswamy static void hvs_close_timeout(struct work_struct *work)
519a9eeb998SSunil Muthuswamy {
520a9eeb998SSunil Muthuswamy 	struct vsock_sock *vsk =
521a9eeb998SSunil Muthuswamy 		container_of(work, struct vsock_sock, close_work.work);
522a9eeb998SSunil Muthuswamy 	struct sock *sk = sk_vsock(vsk);
523a9eeb998SSunil Muthuswamy 
524a9eeb998SSunil Muthuswamy 	sock_hold(sk);
525a9eeb998SSunil Muthuswamy 	lock_sock(sk);
526a9eeb998SSunil Muthuswamy 	if (!sock_flag(sk, SOCK_DONE))
527a9eeb998SSunil Muthuswamy 		hvs_do_close_lock_held(vsk, false);
528a9eeb998SSunil Muthuswamy 
529a9eeb998SSunil Muthuswamy 	vsk->close_work_scheduled = false;
530a9eeb998SSunil Muthuswamy 	release_sock(sk);
531a9eeb998SSunil Muthuswamy 	sock_put(sk);
532a9eeb998SSunil Muthuswamy }
533a9eeb998SSunil Muthuswamy 
534a9eeb998SSunil Muthuswamy /* Returns true, if it is safe to remove socket; false otherwise */
535a9eeb998SSunil Muthuswamy static bool hvs_close_lock_held(struct vsock_sock *vsk)
536a9eeb998SSunil Muthuswamy {
537a9eeb998SSunil Muthuswamy 	struct sock *sk = sk_vsock(vsk);
538a9eeb998SSunil Muthuswamy 
539a9eeb998SSunil Muthuswamy 	if (!(sk->sk_state == TCP_ESTABLISHED ||
540a9eeb998SSunil Muthuswamy 	      sk->sk_state == TCP_CLOSING))
541a9eeb998SSunil Muthuswamy 		return true;
542a9eeb998SSunil Muthuswamy 
543a9eeb998SSunil Muthuswamy 	if ((sk->sk_shutdown & SHUTDOWN_MASK) != SHUTDOWN_MASK)
544a9eeb998SSunil Muthuswamy 		hvs_shutdown_lock_held(vsk->trans, SHUTDOWN_MASK);
545a9eeb998SSunil Muthuswamy 
546a9eeb998SSunil Muthuswamy 	if (sock_flag(sk, SOCK_DONE))
547a9eeb998SSunil Muthuswamy 		return true;
548a9eeb998SSunil Muthuswamy 
549a9eeb998SSunil Muthuswamy 	/* This reference will be dropped by the delayed close routine */
550a9eeb998SSunil Muthuswamy 	sock_hold(sk);
551a9eeb998SSunil Muthuswamy 	INIT_DELAYED_WORK(&vsk->close_work, hvs_close_timeout);
552a9eeb998SSunil Muthuswamy 	vsk->close_work_scheduled = true;
553a9eeb998SSunil Muthuswamy 	schedule_delayed_work(&vsk->close_work, HVS_CLOSE_TIMEOUT);
554a9eeb998SSunil Muthuswamy 	return false;
555a9eeb998SSunil Muthuswamy }
556a9eeb998SSunil Muthuswamy 
557ae0078fcSDexuan Cui static void hvs_release(struct vsock_sock *vsk)
558ae0078fcSDexuan Cui {
559b4562ca7SDexuan Cui 	struct sock *sk = sk_vsock(vsk);
560a9eeb998SSunil Muthuswamy 	bool remove_sock;
561ae0078fcSDexuan Cui 
562b4562ca7SDexuan Cui 	lock_sock(sk);
563a9eeb998SSunil Muthuswamy 	remove_sock = hvs_close_lock_held(vsk);
564b4562ca7SDexuan Cui 	release_sock(sk);
565a9eeb998SSunil Muthuswamy 	if (remove_sock)
566a9eeb998SSunil Muthuswamy 		vsock_remove_sock(vsk);
567ae0078fcSDexuan Cui }
568ae0078fcSDexuan Cui 
569ae0078fcSDexuan Cui static void hvs_destruct(struct vsock_sock *vsk)
570ae0078fcSDexuan Cui {
571ae0078fcSDexuan Cui 	struct hvsock *hvs = vsk->trans;
572ae0078fcSDexuan Cui 	struct vmbus_channel *chan = hvs->chan;
573ae0078fcSDexuan Cui 
574ae0078fcSDexuan Cui 	if (chan)
575ae0078fcSDexuan Cui 		vmbus_hvsock_device_unregister(chan);
576ae0078fcSDexuan Cui 
577ae0078fcSDexuan Cui 	kfree(hvs);
578ae0078fcSDexuan Cui }
579ae0078fcSDexuan Cui 
580ae0078fcSDexuan Cui static int hvs_dgram_bind(struct vsock_sock *vsk, struct sockaddr_vm *addr)
581ae0078fcSDexuan Cui {
582ae0078fcSDexuan Cui 	return -EOPNOTSUPP;
583ae0078fcSDexuan Cui }
584ae0078fcSDexuan Cui 
585ae0078fcSDexuan Cui static int hvs_dgram_dequeue(struct vsock_sock *vsk, struct msghdr *msg,
586ae0078fcSDexuan Cui 			     size_t len, int flags)
587ae0078fcSDexuan Cui {
588ae0078fcSDexuan Cui 	return -EOPNOTSUPP;
589ae0078fcSDexuan Cui }
590ae0078fcSDexuan Cui 
591ae0078fcSDexuan Cui static int hvs_dgram_enqueue(struct vsock_sock *vsk,
592ae0078fcSDexuan Cui 			     struct sockaddr_vm *remote, struct msghdr *msg,
593ae0078fcSDexuan Cui 			     size_t dgram_len)
594ae0078fcSDexuan Cui {
595ae0078fcSDexuan Cui 	return -EOPNOTSUPP;
596ae0078fcSDexuan Cui }
597ae0078fcSDexuan Cui 
598ae0078fcSDexuan Cui static bool hvs_dgram_allow(u32 cid, u32 port)
599ae0078fcSDexuan Cui {
600ae0078fcSDexuan Cui 	return false;
601ae0078fcSDexuan Cui }
602ae0078fcSDexuan Cui 
603ae0078fcSDexuan Cui static int hvs_update_recv_data(struct hvsock *hvs)
604ae0078fcSDexuan Cui {
605ae0078fcSDexuan Cui 	struct hvs_recv_buf *recv_buf;
606ae0078fcSDexuan Cui 	u32 payload_len;
607ae0078fcSDexuan Cui 
608ae0078fcSDexuan Cui 	recv_buf = (struct hvs_recv_buf *)(hvs->recv_desc + 1);
609ae0078fcSDexuan Cui 	payload_len = recv_buf->hdr.data_size;
610ae0078fcSDexuan Cui 
611ae0078fcSDexuan Cui 	if (payload_len > HVS_MTU_SIZE)
612ae0078fcSDexuan Cui 		return -EIO;
613ae0078fcSDexuan Cui 
614ae0078fcSDexuan Cui 	if (payload_len == 0)
615ae0078fcSDexuan Cui 		hvs->vsk->peer_shutdown |= SEND_SHUTDOWN;
616ae0078fcSDexuan Cui 
617ae0078fcSDexuan Cui 	hvs->recv_data_len = payload_len;
618ae0078fcSDexuan Cui 	hvs->recv_data_off = 0;
619ae0078fcSDexuan Cui 
620ae0078fcSDexuan Cui 	return 0;
621ae0078fcSDexuan Cui }
622ae0078fcSDexuan Cui 
623ae0078fcSDexuan Cui static ssize_t hvs_stream_dequeue(struct vsock_sock *vsk, struct msghdr *msg,
624ae0078fcSDexuan Cui 				  size_t len, int flags)
625ae0078fcSDexuan Cui {
626ae0078fcSDexuan Cui 	struct hvsock *hvs = vsk->trans;
627ae0078fcSDexuan Cui 	bool need_refill = !hvs->recv_desc;
628ae0078fcSDexuan Cui 	struct hvs_recv_buf *recv_buf;
629ae0078fcSDexuan Cui 	u32 to_read;
630ae0078fcSDexuan Cui 	int ret;
631ae0078fcSDexuan Cui 
632ae0078fcSDexuan Cui 	if (flags & MSG_PEEK)
633ae0078fcSDexuan Cui 		return -EOPNOTSUPP;
634ae0078fcSDexuan Cui 
635ae0078fcSDexuan Cui 	if (need_refill) {
636ae0078fcSDexuan Cui 		hvs->recv_desc = hv_pkt_iter_first(hvs->chan);
637ae0078fcSDexuan Cui 		ret = hvs_update_recv_data(hvs);
638ae0078fcSDexuan Cui 		if (ret)
639ae0078fcSDexuan Cui 			return ret;
640ae0078fcSDexuan Cui 	}
641ae0078fcSDexuan Cui 
642ae0078fcSDexuan Cui 	recv_buf = (struct hvs_recv_buf *)(hvs->recv_desc + 1);
643ae0078fcSDexuan Cui 	to_read = min_t(u32, len, hvs->recv_data_len);
644ae0078fcSDexuan Cui 	ret = memcpy_to_msg(msg, recv_buf->data + hvs->recv_data_off, to_read);
645ae0078fcSDexuan Cui 	if (ret != 0)
646ae0078fcSDexuan Cui 		return ret;
647ae0078fcSDexuan Cui 
648ae0078fcSDexuan Cui 	hvs->recv_data_len -= to_read;
649ae0078fcSDexuan Cui 	if (hvs->recv_data_len == 0) {
650ae0078fcSDexuan Cui 		hvs->recv_desc = hv_pkt_iter_next(hvs->chan, hvs->recv_desc);
651ae0078fcSDexuan Cui 		if (hvs->recv_desc) {
652ae0078fcSDexuan Cui 			ret = hvs_update_recv_data(hvs);
653ae0078fcSDexuan Cui 			if (ret)
654ae0078fcSDexuan Cui 				return ret;
655ae0078fcSDexuan Cui 		}
656ae0078fcSDexuan Cui 	} else {
657ae0078fcSDexuan Cui 		hvs->recv_data_off += to_read;
658ae0078fcSDexuan Cui 	}
659ae0078fcSDexuan Cui 
660ae0078fcSDexuan Cui 	return to_read;
661ae0078fcSDexuan Cui }
662ae0078fcSDexuan Cui 
663ae0078fcSDexuan Cui static ssize_t hvs_stream_enqueue(struct vsock_sock *vsk, struct msghdr *msg,
664ae0078fcSDexuan Cui 				  size_t len)
665ae0078fcSDexuan Cui {
666ae0078fcSDexuan Cui 	struct hvsock *hvs = vsk->trans;
667ae0078fcSDexuan Cui 	struct vmbus_channel *chan = hvs->chan;
668ae0078fcSDexuan Cui 	struct hvs_send_buf *send_buf;
66914a1eaa8SSunil Muthuswamy 	ssize_t to_write, max_writable;
67014a1eaa8SSunil Muthuswamy 	ssize_t ret = 0;
67114a1eaa8SSunil Muthuswamy 	ssize_t bytes_written = 0;
672ae0078fcSDexuan Cui 
673ae0078fcSDexuan Cui 	BUILD_BUG_ON(sizeof(*send_buf) != PAGE_SIZE_4K);
674ae0078fcSDexuan Cui 
675ae0078fcSDexuan Cui 	send_buf = kmalloc(sizeof(*send_buf), GFP_KERNEL);
676ae0078fcSDexuan Cui 	if (!send_buf)
677ae0078fcSDexuan Cui 		return -ENOMEM;
678ae0078fcSDexuan Cui 
67914a1eaa8SSunil Muthuswamy 	/* Reader(s) could be draining data from the channel as we write.
68014a1eaa8SSunil Muthuswamy 	 * Maximize bandwidth, by iterating until the channel is found to be
68114a1eaa8SSunil Muthuswamy 	 * full.
68214a1eaa8SSunil Muthuswamy 	 */
68314a1eaa8SSunil Muthuswamy 	while (len) {
684ae0078fcSDexuan Cui 		max_writable = hvs_channel_writable_bytes(chan);
68514a1eaa8SSunil Muthuswamy 		if (!max_writable)
68614a1eaa8SSunil Muthuswamy 			break;
687ae0078fcSDexuan Cui 		to_write = min_t(ssize_t, len, max_writable);
688ae0078fcSDexuan Cui 		to_write = min_t(ssize_t, to_write, HVS_SEND_BUF_SIZE);
68914a1eaa8SSunil Muthuswamy 		/* memcpy_from_msg is safe for loop as it advances the offsets
69014a1eaa8SSunil Muthuswamy 		 * within the message iterator.
69114a1eaa8SSunil Muthuswamy 		 */
692ae0078fcSDexuan Cui 		ret = memcpy_from_msg(send_buf->data, msg, to_write);
693ae0078fcSDexuan Cui 		if (ret < 0)
694ae0078fcSDexuan Cui 			goto out;
695ae0078fcSDexuan Cui 
696ae0078fcSDexuan Cui 		ret = hvs_send_data(hvs->chan, send_buf, to_write);
697ae0078fcSDexuan Cui 		if (ret < 0)
698ae0078fcSDexuan Cui 			goto out;
699ae0078fcSDexuan Cui 
70014a1eaa8SSunil Muthuswamy 		bytes_written += to_write;
70114a1eaa8SSunil Muthuswamy 		len -= to_write;
70214a1eaa8SSunil Muthuswamy 	}
703ae0078fcSDexuan Cui out:
70414a1eaa8SSunil Muthuswamy 	/* If any data has been sent, return that */
70514a1eaa8SSunil Muthuswamy 	if (bytes_written)
70614a1eaa8SSunil Muthuswamy 		ret = bytes_written;
707ae0078fcSDexuan Cui 	kfree(send_buf);
708ae0078fcSDexuan Cui 	return ret;
709ae0078fcSDexuan Cui }
710ae0078fcSDexuan Cui 
711ae0078fcSDexuan Cui static s64 hvs_stream_has_data(struct vsock_sock *vsk)
712ae0078fcSDexuan Cui {
713ae0078fcSDexuan Cui 	struct hvsock *hvs = vsk->trans;
714ae0078fcSDexuan Cui 	s64 ret;
715ae0078fcSDexuan Cui 
716ae0078fcSDexuan Cui 	if (hvs->recv_data_len > 0)
717ae0078fcSDexuan Cui 		return 1;
718ae0078fcSDexuan Cui 
719ae0078fcSDexuan Cui 	switch (hvs_channel_readable_payload(hvs->chan)) {
720ae0078fcSDexuan Cui 	case 1:
721ae0078fcSDexuan Cui 		ret = 1;
722ae0078fcSDexuan Cui 		break;
723ae0078fcSDexuan Cui 	case 0:
724ae0078fcSDexuan Cui 		vsk->peer_shutdown |= SEND_SHUTDOWN;
725ae0078fcSDexuan Cui 		ret = 0;
726ae0078fcSDexuan Cui 		break;
727ae0078fcSDexuan Cui 	default: /* -1 */
728ae0078fcSDexuan Cui 		ret = 0;
729ae0078fcSDexuan Cui 		break;
730ae0078fcSDexuan Cui 	}
731ae0078fcSDexuan Cui 
732ae0078fcSDexuan Cui 	return ret;
733ae0078fcSDexuan Cui }
734ae0078fcSDexuan Cui 
735ae0078fcSDexuan Cui static s64 hvs_stream_has_space(struct vsock_sock *vsk)
736ae0078fcSDexuan Cui {
737ae0078fcSDexuan Cui 	struct hvsock *hvs = vsk->trans;
738ae0078fcSDexuan Cui 
739cb359b60SSunil Muthuswamy 	return hvs_channel_writable_bytes(hvs->chan);
740ae0078fcSDexuan Cui }
741ae0078fcSDexuan Cui 
742ae0078fcSDexuan Cui static u64 hvs_stream_rcvhiwat(struct vsock_sock *vsk)
743ae0078fcSDexuan Cui {
744ae0078fcSDexuan Cui 	return HVS_MTU_SIZE + 1;
745ae0078fcSDexuan Cui }
746ae0078fcSDexuan Cui 
747ae0078fcSDexuan Cui static bool hvs_stream_is_active(struct vsock_sock *vsk)
748ae0078fcSDexuan Cui {
749ae0078fcSDexuan Cui 	struct hvsock *hvs = vsk->trans;
750ae0078fcSDexuan Cui 
751ae0078fcSDexuan Cui 	return hvs->chan != NULL;
752ae0078fcSDexuan Cui }
753ae0078fcSDexuan Cui 
754ae0078fcSDexuan Cui static bool hvs_stream_allow(u32 cid, u32 port)
755ae0078fcSDexuan Cui {
756ae0078fcSDexuan Cui 	/* The host's port range [MIN_HOST_EPHEMERAL_PORT, 0xFFFFFFFF) is
757ae0078fcSDexuan Cui 	 * reserved as ephemeral ports, which are used as the host's ports
758ae0078fcSDexuan Cui 	 * when the host initiates connections.
759ae0078fcSDexuan Cui 	 *
760ae0078fcSDexuan Cui 	 * Perform this check in the guest so an immediate error is produced
761ae0078fcSDexuan Cui 	 * instead of a timeout.
762ae0078fcSDexuan Cui 	 */
763ae0078fcSDexuan Cui 	if (port > MAX_HOST_LISTEN_PORT)
764ae0078fcSDexuan Cui 		return false;
765ae0078fcSDexuan Cui 
766ae0078fcSDexuan Cui 	if (cid == VMADDR_CID_HOST)
767ae0078fcSDexuan Cui 		return true;
768ae0078fcSDexuan Cui 
769ae0078fcSDexuan Cui 	return false;
770ae0078fcSDexuan Cui }
771ae0078fcSDexuan Cui 
772ae0078fcSDexuan Cui static
773ae0078fcSDexuan Cui int hvs_notify_poll_in(struct vsock_sock *vsk, size_t target, bool *readable)
774ae0078fcSDexuan Cui {
775ae0078fcSDexuan Cui 	struct hvsock *hvs = vsk->trans;
776ae0078fcSDexuan Cui 
777ae0078fcSDexuan Cui 	*readable = hvs_channel_readable(hvs->chan);
778ae0078fcSDexuan Cui 	return 0;
779ae0078fcSDexuan Cui }
780ae0078fcSDexuan Cui 
781ae0078fcSDexuan Cui static
782ae0078fcSDexuan Cui int hvs_notify_poll_out(struct vsock_sock *vsk, size_t target, bool *writable)
783ae0078fcSDexuan Cui {
784ae0078fcSDexuan Cui 	*writable = hvs_stream_has_space(vsk) > 0;
785ae0078fcSDexuan Cui 
786ae0078fcSDexuan Cui 	return 0;
787ae0078fcSDexuan Cui }
788ae0078fcSDexuan Cui 
789ae0078fcSDexuan Cui static
790ae0078fcSDexuan Cui int hvs_notify_recv_init(struct vsock_sock *vsk, size_t target,
791ae0078fcSDexuan Cui 			 struct vsock_transport_recv_notify_data *d)
792ae0078fcSDexuan Cui {
793ae0078fcSDexuan Cui 	return 0;
794ae0078fcSDexuan Cui }
795ae0078fcSDexuan Cui 
796ae0078fcSDexuan Cui static
797ae0078fcSDexuan Cui int hvs_notify_recv_pre_block(struct vsock_sock *vsk, size_t target,
798ae0078fcSDexuan Cui 			      struct vsock_transport_recv_notify_data *d)
799ae0078fcSDexuan Cui {
800ae0078fcSDexuan Cui 	return 0;
801ae0078fcSDexuan Cui }
802ae0078fcSDexuan Cui 
803ae0078fcSDexuan Cui static
804ae0078fcSDexuan Cui int hvs_notify_recv_pre_dequeue(struct vsock_sock *vsk, size_t target,
805ae0078fcSDexuan Cui 				struct vsock_transport_recv_notify_data *d)
806ae0078fcSDexuan Cui {
807ae0078fcSDexuan Cui 	return 0;
808ae0078fcSDexuan Cui }
809ae0078fcSDexuan Cui 
810ae0078fcSDexuan Cui static
811ae0078fcSDexuan Cui int hvs_notify_recv_post_dequeue(struct vsock_sock *vsk, size_t target,
812ae0078fcSDexuan Cui 				 ssize_t copied, bool data_read,
813ae0078fcSDexuan Cui 				 struct vsock_transport_recv_notify_data *d)
814ae0078fcSDexuan Cui {
815ae0078fcSDexuan Cui 	return 0;
816ae0078fcSDexuan Cui }
817ae0078fcSDexuan Cui 
818ae0078fcSDexuan Cui static
819ae0078fcSDexuan Cui int hvs_notify_send_init(struct vsock_sock *vsk,
820ae0078fcSDexuan Cui 			 struct vsock_transport_send_notify_data *d)
821ae0078fcSDexuan Cui {
822ae0078fcSDexuan Cui 	return 0;
823ae0078fcSDexuan Cui }
824ae0078fcSDexuan Cui 
825ae0078fcSDexuan Cui static
826ae0078fcSDexuan Cui int hvs_notify_send_pre_block(struct vsock_sock *vsk,
827ae0078fcSDexuan Cui 			      struct vsock_transport_send_notify_data *d)
828ae0078fcSDexuan Cui {
829ae0078fcSDexuan Cui 	return 0;
830ae0078fcSDexuan Cui }
831ae0078fcSDexuan Cui 
832ae0078fcSDexuan Cui static
833ae0078fcSDexuan Cui int hvs_notify_send_pre_enqueue(struct vsock_sock *vsk,
834ae0078fcSDexuan Cui 				struct vsock_transport_send_notify_data *d)
835ae0078fcSDexuan Cui {
836ae0078fcSDexuan Cui 	return 0;
837ae0078fcSDexuan Cui }
838ae0078fcSDexuan Cui 
839ae0078fcSDexuan Cui static
840ae0078fcSDexuan Cui int hvs_notify_send_post_enqueue(struct vsock_sock *vsk, ssize_t written,
841ae0078fcSDexuan Cui 				 struct vsock_transport_send_notify_data *d)
842ae0078fcSDexuan Cui {
843ae0078fcSDexuan Cui 	return 0;
844ae0078fcSDexuan Cui }
845ae0078fcSDexuan Cui 
846ae0078fcSDexuan Cui static void hvs_set_buffer_size(struct vsock_sock *vsk, u64 val)
847ae0078fcSDexuan Cui {
848ae0078fcSDexuan Cui 	/* Ignored. */
849ae0078fcSDexuan Cui }
850ae0078fcSDexuan Cui 
851ae0078fcSDexuan Cui static void hvs_set_min_buffer_size(struct vsock_sock *vsk, u64 val)
852ae0078fcSDexuan Cui {
853ae0078fcSDexuan Cui 	/* Ignored. */
854ae0078fcSDexuan Cui }
855ae0078fcSDexuan Cui 
856ae0078fcSDexuan Cui static void hvs_set_max_buffer_size(struct vsock_sock *vsk, u64 val)
857ae0078fcSDexuan Cui {
858ae0078fcSDexuan Cui 	/* Ignored. */
859ae0078fcSDexuan Cui }
860ae0078fcSDexuan Cui 
861ae0078fcSDexuan Cui static u64 hvs_get_buffer_size(struct vsock_sock *vsk)
862ae0078fcSDexuan Cui {
863ae0078fcSDexuan Cui 	return -ENOPROTOOPT;
864ae0078fcSDexuan Cui }
865ae0078fcSDexuan Cui 
866ae0078fcSDexuan Cui static u64 hvs_get_min_buffer_size(struct vsock_sock *vsk)
867ae0078fcSDexuan Cui {
868ae0078fcSDexuan Cui 	return -ENOPROTOOPT;
869ae0078fcSDexuan Cui }
870ae0078fcSDexuan Cui 
871ae0078fcSDexuan Cui static u64 hvs_get_max_buffer_size(struct vsock_sock *vsk)
872ae0078fcSDexuan Cui {
873ae0078fcSDexuan Cui 	return -ENOPROTOOPT;
874ae0078fcSDexuan Cui }
875ae0078fcSDexuan Cui 
876ae0078fcSDexuan Cui static struct vsock_transport hvs_transport = {
877ae0078fcSDexuan Cui 	.get_local_cid            = hvs_get_local_cid,
878ae0078fcSDexuan Cui 
879ae0078fcSDexuan Cui 	.init                     = hvs_sock_init,
880ae0078fcSDexuan Cui 	.destruct                 = hvs_destruct,
881ae0078fcSDexuan Cui 	.release                  = hvs_release,
882ae0078fcSDexuan Cui 	.connect                  = hvs_connect,
883ae0078fcSDexuan Cui 	.shutdown                 = hvs_shutdown,
884ae0078fcSDexuan Cui 
885ae0078fcSDexuan Cui 	.dgram_bind               = hvs_dgram_bind,
886ae0078fcSDexuan Cui 	.dgram_dequeue            = hvs_dgram_dequeue,
887ae0078fcSDexuan Cui 	.dgram_enqueue            = hvs_dgram_enqueue,
888ae0078fcSDexuan Cui 	.dgram_allow              = hvs_dgram_allow,
889ae0078fcSDexuan Cui 
890ae0078fcSDexuan Cui 	.stream_dequeue           = hvs_stream_dequeue,
891ae0078fcSDexuan Cui 	.stream_enqueue           = hvs_stream_enqueue,
892ae0078fcSDexuan Cui 	.stream_has_data          = hvs_stream_has_data,
893ae0078fcSDexuan Cui 	.stream_has_space         = hvs_stream_has_space,
894ae0078fcSDexuan Cui 	.stream_rcvhiwat          = hvs_stream_rcvhiwat,
895ae0078fcSDexuan Cui 	.stream_is_active         = hvs_stream_is_active,
896ae0078fcSDexuan Cui 	.stream_allow             = hvs_stream_allow,
897ae0078fcSDexuan Cui 
898ae0078fcSDexuan Cui 	.notify_poll_in           = hvs_notify_poll_in,
899ae0078fcSDexuan Cui 	.notify_poll_out          = hvs_notify_poll_out,
900ae0078fcSDexuan Cui 	.notify_recv_init         = hvs_notify_recv_init,
901ae0078fcSDexuan Cui 	.notify_recv_pre_block    = hvs_notify_recv_pre_block,
902ae0078fcSDexuan Cui 	.notify_recv_pre_dequeue  = hvs_notify_recv_pre_dequeue,
903ae0078fcSDexuan Cui 	.notify_recv_post_dequeue = hvs_notify_recv_post_dequeue,
904ae0078fcSDexuan Cui 	.notify_send_init         = hvs_notify_send_init,
905ae0078fcSDexuan Cui 	.notify_send_pre_block    = hvs_notify_send_pre_block,
906ae0078fcSDexuan Cui 	.notify_send_pre_enqueue  = hvs_notify_send_pre_enqueue,
907ae0078fcSDexuan Cui 	.notify_send_post_enqueue = hvs_notify_send_post_enqueue,
908ae0078fcSDexuan Cui 
909ae0078fcSDexuan Cui 	.set_buffer_size          = hvs_set_buffer_size,
910ae0078fcSDexuan Cui 	.set_min_buffer_size      = hvs_set_min_buffer_size,
911ae0078fcSDexuan Cui 	.set_max_buffer_size      = hvs_set_max_buffer_size,
912ae0078fcSDexuan Cui 	.get_buffer_size          = hvs_get_buffer_size,
913ae0078fcSDexuan Cui 	.get_min_buffer_size      = hvs_get_min_buffer_size,
914ae0078fcSDexuan Cui 	.get_max_buffer_size      = hvs_get_max_buffer_size,
915ae0078fcSDexuan Cui };
916ae0078fcSDexuan Cui 
917ae0078fcSDexuan Cui static int hvs_probe(struct hv_device *hdev,
918ae0078fcSDexuan Cui 		     const struct hv_vmbus_device_id *dev_id)
919ae0078fcSDexuan Cui {
920ae0078fcSDexuan Cui 	struct vmbus_channel *chan = hdev->channel;
921ae0078fcSDexuan Cui 
922ae0078fcSDexuan Cui 	hvs_open_connection(chan);
923ae0078fcSDexuan Cui 
924ae0078fcSDexuan Cui 	/* Always return success to suppress the unnecessary error message
925ae0078fcSDexuan Cui 	 * in vmbus_probe(): on error the host will rescind the device in
926ae0078fcSDexuan Cui 	 * 30 seconds and we can do cleanup at that time in
927ae0078fcSDexuan Cui 	 * vmbus_onoffer_rescind().
928ae0078fcSDexuan Cui 	 */
929ae0078fcSDexuan Cui 	return 0;
930ae0078fcSDexuan Cui }
931ae0078fcSDexuan Cui 
932ae0078fcSDexuan Cui static int hvs_remove(struct hv_device *hdev)
933ae0078fcSDexuan Cui {
934ae0078fcSDexuan Cui 	struct vmbus_channel *chan = hdev->channel;
935ae0078fcSDexuan Cui 
936ae0078fcSDexuan Cui 	vmbus_close(chan);
937ae0078fcSDexuan Cui 
938ae0078fcSDexuan Cui 	return 0;
939ae0078fcSDexuan Cui }
940ae0078fcSDexuan Cui 
941ae0078fcSDexuan Cui /* This isn't really used. See vmbus_match() and vmbus_probe() */
942ae0078fcSDexuan Cui static const struct hv_vmbus_device_id id_table[] = {
943ae0078fcSDexuan Cui 	{},
944ae0078fcSDexuan Cui };
945ae0078fcSDexuan Cui 
946ae0078fcSDexuan Cui static struct hv_driver hvs_drv = {
947ae0078fcSDexuan Cui 	.name		= "hv_sock",
948ae0078fcSDexuan Cui 	.hvsock		= true,
949ae0078fcSDexuan Cui 	.id_table	= id_table,
950ae0078fcSDexuan Cui 	.probe		= hvs_probe,
951ae0078fcSDexuan Cui 	.remove		= hvs_remove,
952ae0078fcSDexuan Cui };
953ae0078fcSDexuan Cui 
954ae0078fcSDexuan Cui static int __init hvs_init(void)
955ae0078fcSDexuan Cui {
956ae0078fcSDexuan Cui 	int ret;
957ae0078fcSDexuan Cui 
958ae0078fcSDexuan Cui 	if (vmbus_proto_version < VERSION_WIN10)
959ae0078fcSDexuan Cui 		return -ENODEV;
960ae0078fcSDexuan Cui 
961ae0078fcSDexuan Cui 	ret = vmbus_driver_register(&hvs_drv);
962ae0078fcSDexuan Cui 	if (ret != 0)
963ae0078fcSDexuan Cui 		return ret;
964ae0078fcSDexuan Cui 
965ae0078fcSDexuan Cui 	ret = vsock_core_init(&hvs_transport);
966ae0078fcSDexuan Cui 	if (ret) {
967ae0078fcSDexuan Cui 		vmbus_driver_unregister(&hvs_drv);
968ae0078fcSDexuan Cui 		return ret;
969ae0078fcSDexuan Cui 	}
970ae0078fcSDexuan Cui 
971ae0078fcSDexuan Cui 	return 0;
972ae0078fcSDexuan Cui }
973ae0078fcSDexuan Cui 
974ae0078fcSDexuan Cui static void __exit hvs_exit(void)
975ae0078fcSDexuan Cui {
976ae0078fcSDexuan Cui 	vsock_core_exit();
977ae0078fcSDexuan Cui 	vmbus_driver_unregister(&hvs_drv);
978ae0078fcSDexuan Cui }
979ae0078fcSDexuan Cui 
980ae0078fcSDexuan Cui module_init(hvs_init);
981ae0078fcSDexuan Cui module_exit(hvs_exit);
982ae0078fcSDexuan Cui 
983ae0078fcSDexuan Cui MODULE_DESCRIPTION("Hyper-V Sockets");
984ae0078fcSDexuan Cui MODULE_VERSION("1.0.0");
985ae0078fcSDexuan Cui MODULE_LICENSE("GPL");
986ae0078fcSDexuan Cui MODULE_ALIAS_NETPROTO(PF_VSOCK);
987