1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * Hyper-V transport for vsock
4 *
5 * Hyper-V Sockets supplies a byte-stream based communication mechanism
6 * between the host and the VM. This driver implements the necessary
7 * support in the VM by introducing the new vsock transport.
8 *
9 * Copyright (c) 2017, Microsoft Corporation.
10 */
11#include <linux/module.h>
12#include <linux/vmalloc.h>
13#include <linux/hyperv.h>
14#include <net/sock.h>
15#include <net/af_vsock.h>
16#include <hyperv/hvhdk.h>
17
18/* Older (VMBUS version 'VERSION_WIN10' or before) Windows hosts have some
19 * stricter requirements on the hv_sock ring buffer size of six 4K pages.
20 * HV_HYP_PAGE_SIZE is defined as 4K. Newer hosts don't have this limitation;
21 * but, keep the defaults the same for compat.
22 */
23#define RINGBUFFER_HVS_RCV_SIZE (HV_HYP_PAGE_SIZE * 6)
24#define RINGBUFFER_HVS_SND_SIZE (HV_HYP_PAGE_SIZE * 6)
25#define RINGBUFFER_HVS_MAX_SIZE (HV_HYP_PAGE_SIZE * 64)
26
27/* The MTU is 16KB per the host side's design */
28#define HVS_MTU_SIZE (1024 * 16)
29
30/* How long to wait for graceful shutdown of a connection */
31#define HVS_CLOSE_TIMEOUT (8 * HZ)
32
33struct vmpipe_proto_header {
34 u32 pkt_type;
35 u32 data_size;
36};
37
38/* For recv, we use the VMBus in-place packet iterator APIs to directly copy
39 * data from the ringbuffer into the userspace buffer.
40 */
41struct hvs_recv_buf {
42 /* The header before the payload data */
43 struct vmpipe_proto_header hdr;
44
45 /* The payload */
46 u8 data[HVS_MTU_SIZE];
47};
48
49/* We can send up to HVS_MTU_SIZE bytes of payload to the host, but let's use
50 * a smaller size, i.e. HVS_SEND_BUF_SIZE, to maximize concurrency between the
51 * guest and the host processing as one VMBUS packet is the smallest processing
52 * unit.
53 *
54 * Note: the buffer can be eliminated in the future when we add new VMBus
55 * ringbuffer APIs that allow us to directly copy data from userspace buffer
56 * to VMBus ringbuffer.
57 */
58#define HVS_SEND_BUF_SIZE \
59 (HV_HYP_PAGE_SIZE - sizeof(struct vmpipe_proto_header))
60
61struct hvs_send_buf {
62 /* The header before the payload data */
63 struct vmpipe_proto_header hdr;
64
65 /* The payload */
66 u8 data[HVS_SEND_BUF_SIZE];
67};
68
69#define HVS_HEADER_LEN (sizeof(struct vmpacket_descriptor) + \
70 sizeof(struct vmpipe_proto_header))
71
72/* See 'prev_indices' in hv_ringbuffer_read(), hv_ringbuffer_write(), and
73 * __hv_pkt_iter_next().
74 */
75#define VMBUS_PKT_TRAILER_SIZE (sizeof(u64))
76
77#define HVS_PKT_LEN(payload_len) (HVS_HEADER_LEN + \
78 ALIGN((payload_len), 8) + \
79 VMBUS_PKT_TRAILER_SIZE)
80
81/* Upper bound on the size of a VMbus packet for hv_sock */
82#define HVS_MAX_PKT_SIZE HVS_PKT_LEN(HVS_MTU_SIZE)
83
84union hvs_service_id {
85 guid_t srv_id;
86
87 struct {
88 unsigned int svm_port;
89 unsigned char b[sizeof(guid_t) - sizeof(unsigned int)];
90 };
91};
92
93/* Per-socket state (accessed via vsk->trans) */
94struct hvsock {
95 struct vsock_sock *vsk;
96
97 guid_t vm_srv_id;
98 guid_t host_srv_id;
99
100 struct vmbus_channel *chan;
101 struct vmpacket_descriptor *recv_desc;
102
103 /* The length of the payload not delivered to userland yet */
104 u32 recv_data_len;
105 /* The offset of the payload */
106 u32 recv_data_off;
107
108 /* Have we sent the zero-length packet (FIN)? */
109 bool fin_sent;
110};
111
112/* In the VM, we support Hyper-V Sockets with AF_VSOCK, and the endpoint is
113 * <cid, port> (see struct sockaddr_vm). Note: cid is not really used here:
114 * when we write apps to connect to the host, we can only use VMADDR_CID_ANY
115 * or VMADDR_CID_HOST (both are equivalent) as the remote cid, and when we
116 * write apps to bind() & listen() in the VM, we can only use VMADDR_CID_ANY
117 * as the local cid.
118 *
119 * On the host, Hyper-V Sockets are supported by Winsock AF_HYPERV:
120 * https://docs.microsoft.com/en-us/virtualization/hyper-v-on-windows/user-
121 * guide/make-integration-service, and the endpoint is <VmID, ServiceId> with
122 * the below sockaddr:
123 *
124 * struct SOCKADDR_HV
125 * {
126 * ADDRESS_FAMILY Family;
127 * USHORT Reserved;
128 * GUID VmId;
129 * GUID ServiceId;
130 * };
131 * Note: VmID is not used by Linux VM and actually it isn't transmitted via
132 * VMBus, because here it's obvious the host and the VM can easily identify
133 * each other. Though the VmID is useful on the host, especially in the case
134 * of Windows container, Linux VM doesn't need it at all.
135 *
136 * To make use of the AF_VSOCK infrastructure in Linux VM, we have to limit
137 * the available GUID space of SOCKADDR_HV so that we can create a mapping
138 * between AF_VSOCK port and SOCKADDR_HV Service GUID. The rule of writing
139 * Hyper-V Sockets apps on the host and in Linux VM is:
140 *
141 ****************************************************************************
142 * The only valid Service GUIDs, from the perspectives of both the host and *
143 * Linux VM, that can be connected by the other end, must conform to this *
144 * format: <port>-facb-11e6-bd58-64006a7986d3. *
145 ****************************************************************************
146 *
147 * When we write apps on the host to connect(), the GUID ServiceID is used.
148 * When we write apps in Linux VM to connect(), we only need to specify the
149 * port and the driver will form the GUID and use that to request the host.
150 *
151 */
152
153/* 00000000-facb-11e6-bd58-64006a7986d3 */
154static const guid_t srv_id_template =
155 GUID_INIT(0x00000000, 0xfacb, 0x11e6, 0xbd, 0x58,
156 0x64, 0x00, 0x6a, 0x79, 0x86, 0xd3);
157
158static bool hvs_check_transport(struct vsock_sock *vsk);
159
160static bool is_valid_srv_id(const guid_t *id)
161{
162 return !memcmp(p: &id->b[4], q: &srv_id_template.b[4], size: sizeof(guid_t) - 4);
163}
164
165static unsigned int get_port_by_srv_id(const guid_t *svr_id)
166{
167 return *((unsigned int *)svr_id);
168}
169
170static void hvs_addr_init(struct sockaddr_vm *addr, const guid_t *svr_id)
171{
172 unsigned int port = get_port_by_srv_id(svr_id);
173
174 vsock_addr_init(addr, VMADDR_CID_ANY, port);
175}
176
177static void hvs_set_channel_pending_send_size(struct vmbus_channel *chan)
178{
179 set_channel_pending_send_size(c: chan,
180 HVS_PKT_LEN(HVS_SEND_BUF_SIZE));
181
182 virt_mb();
183}
184
185static bool hvs_channel_readable(struct vmbus_channel *chan)
186{
187 u32 readable = hv_get_bytes_to_read(rbi: &chan->inbound);
188
189 /* 0-size payload means FIN */
190 return readable >= HVS_PKT_LEN(0);
191}
192
193static int hvs_channel_readable_payload(struct vmbus_channel *chan)
194{
195 u32 readable = hv_get_bytes_to_read(rbi: &chan->inbound);
196
197 if (readable > HVS_PKT_LEN(0)) {
198 /* At least we have 1 byte to read. We don't need to return
199 * the exact readable bytes: see vsock_stream_recvmsg() ->
200 * vsock_stream_has_data().
201 */
202 return 1;
203 }
204
205 if (readable == HVS_PKT_LEN(0)) {
206 /* 0-size payload means FIN */
207 return 0;
208 }
209
210 /* No payload or FIN */
211 return -1;
212}
213
214static size_t hvs_channel_writable_bytes(struct vmbus_channel *chan)
215{
216 u32 writeable = hv_get_bytes_to_write(rbi: &chan->outbound);
217 size_t ret;
218
219 /* The ringbuffer mustn't be 100% full, and we should reserve a
220 * zero-length-payload packet for the FIN: see hv_ringbuffer_write()
221 * and hvs_shutdown().
222 */
223 if (writeable <= HVS_PKT_LEN(1) + HVS_PKT_LEN(0))
224 return 0;
225
226 ret = writeable - HVS_PKT_LEN(1) - HVS_PKT_LEN(0);
227
228 return round_down(ret, 8);
229}
230
231static int __hvs_send_data(struct vmbus_channel *chan,
232 struct vmpipe_proto_header *hdr,
233 size_t to_write)
234{
235 hdr->pkt_type = 1;
236 hdr->data_size = to_write;
237 return vmbus_sendpacket(channel: chan, buffer: hdr, bufferLen: sizeof(*hdr) + to_write,
238 requestid: 0, type: VM_PKT_DATA_INBAND, flags: 0);
239}
240
241static int hvs_send_data(struct vmbus_channel *chan,
242 struct hvs_send_buf *send_buf, size_t to_write)
243{
244 return __hvs_send_data(chan, hdr: &send_buf->hdr, to_write);
245}
246
247static void hvs_channel_cb(void *ctx)
248{
249 struct sock *sk = (struct sock *)ctx;
250 struct vsock_sock *vsk = vsock_sk(sk);
251 struct hvsock *hvs = vsk->trans;
252 struct vmbus_channel *chan = hvs->chan;
253
254 if (hvs_channel_readable(chan))
255 sk->sk_data_ready(sk);
256
257 if (hv_get_bytes_to_write(rbi: &chan->outbound) > 0)
258 sk->sk_write_space(sk);
259}
260
261static void hvs_do_close_lock_held(struct vsock_sock *vsk,
262 bool cancel_timeout)
263{
264 struct sock *sk = sk_vsock(vsk);
265
266 sock_set_flag(sk, flag: SOCK_DONE);
267 vsk->peer_shutdown = SHUTDOWN_MASK;
268 if (vsock_stream_has_data(vsk) <= 0)
269 sk->sk_state = TCP_CLOSING;
270 sk->sk_state_change(sk);
271 if (vsk->close_work_scheduled &&
272 (!cancel_timeout || cancel_delayed_work(dwork: &vsk->close_work))) {
273 vsk->close_work_scheduled = false;
274 vsock_remove_sock(vsk);
275
276 /* Release the reference taken while scheduling the timeout */
277 sock_put(sk);
278 }
279}
280
281static void hvs_close_connection(struct vmbus_channel *chan)
282{
283 struct sock *sk = get_per_channel_state(c: chan);
284
285 lock_sock(sk);
286 hvs_do_close_lock_held(vsock_sk(sk), cancel_timeout: true);
287 release_sock(sk);
288
289 /* Release the refcnt for the channel that's opened in
290 * hvs_open_connection().
291 */
292 sock_put(sk);
293}
294
295static void hvs_open_connection(struct vmbus_channel *chan)
296{
297 guid_t *if_instance, *if_type;
298 unsigned char conn_from_host;
299
300 struct sockaddr_vm addr;
301 struct sock *sk, *new = NULL;
302 struct vsock_sock *vnew = NULL;
303 struct hvsock *hvs = NULL;
304 struct hvsock *hvs_new = NULL;
305 int rcvbuf;
306 int ret;
307 int sndbuf;
308
309 if_type = &chan->offermsg.offer.if_type;
310 if_instance = &chan->offermsg.offer.if_instance;
311 conn_from_host = chan->offermsg.offer.u.pipe.user_def[0];
312 if (!is_valid_srv_id(id: if_type))
313 return;
314
315 hvs_addr_init(addr: &addr, svr_id: conn_from_host ? if_type : if_instance);
316 sk = vsock_find_bound_socket(addr: &addr);
317 if (!sk)
318 return;
319
320 lock_sock(sk);
321 if ((conn_from_host && sk->sk_state != TCP_LISTEN) ||
322 (!conn_from_host && sk->sk_state != TCP_SYN_SENT))
323 goto out;
324
325 if (conn_from_host) {
326 if (sk->sk_ack_backlog >= sk->sk_max_ack_backlog)
327 goto out;
328
329 new = vsock_create_connected(parent: sk);
330 if (!new)
331 goto out;
332
333 new->sk_state = TCP_SYN_SENT;
334 vnew = vsock_sk(new);
335
336 hvs_addr_init(addr: &vnew->local_addr, svr_id: if_type);
337
338 /* Remote peer is always the host */
339 vsock_addr_init(addr: &vnew->remote_addr,
340 VMADDR_CID_HOST, VMADDR_PORT_ANY);
341 vnew->remote_addr.svm_port = get_port_by_srv_id(svr_id: if_instance);
342 ret = vsock_assign_transport(vsk: vnew, vsock_sk(sk));
343 /* Transport assigned (looking at remote_addr) must be the
344 * same where we received the request.
345 */
346 if (ret || !hvs_check_transport(vsk: vnew)) {
347 sock_put(sk: new);
348 goto out;
349 }
350 hvs_new = vnew->trans;
351 hvs_new->chan = chan;
352 } else {
353 hvs = vsock_sk(sk)->trans;
354 hvs->chan = chan;
355 }
356
357 set_channel_read_mode(c: chan, mode: HV_CALL_DIRECT);
358
359 /* Use the socket buffer sizes as hints for the VMBUS ring size. For
360 * server side sockets, 'sk' is the parent socket and thus, this will
361 * allow the child sockets to inherit the size from the parent. Keep
362 * the mins to the default value and align to page size as per VMBUS
363 * requirements.
364 * For the max, the socket core library will limit the socket buffer
365 * size that can be set by the user, but, since currently, the hv_sock
366 * VMBUS ring buffer is physically contiguous allocation, restrict it
367 * further.
368 * Older versions of hv_sock host side code cannot handle bigger VMBUS
369 * ring buffer size. Use the version number to limit the change to newer
370 * versions.
371 */
372 if (vmbus_proto_version < VERSION_WIN10_V5) {
373 sndbuf = RINGBUFFER_HVS_SND_SIZE;
374 rcvbuf = RINGBUFFER_HVS_RCV_SIZE;
375 } else {
376 sndbuf = max_t(int, sk->sk_sndbuf, RINGBUFFER_HVS_SND_SIZE);
377 sndbuf = min_t(int, sndbuf, RINGBUFFER_HVS_MAX_SIZE);
378 sndbuf = ALIGN(sndbuf, HV_HYP_PAGE_SIZE);
379 rcvbuf = max_t(int, sk->sk_rcvbuf, RINGBUFFER_HVS_RCV_SIZE);
380 rcvbuf = min_t(int, rcvbuf, RINGBUFFER_HVS_MAX_SIZE);
381 rcvbuf = ALIGN(rcvbuf, HV_HYP_PAGE_SIZE);
382 }
383
384 chan->max_pkt_size = HVS_MAX_PKT_SIZE;
385
386 ret = vmbus_open(channel: chan, send_ringbuffersize: sndbuf, recv_ringbuffersize: rcvbuf, NULL, userdatalen: 0, onchannel_callback: hvs_channel_cb,
387 context: conn_from_host ? new : sk);
388 if (ret != 0) {
389 if (conn_from_host) {
390 hvs_new->chan = NULL;
391 sock_put(sk: new);
392 } else {
393 hvs->chan = NULL;
394 }
395 goto out;
396 }
397
398 set_per_channel_state(c: chan, s: conn_from_host ? new : sk);
399
400 /* This reference will be dropped by hvs_close_connection(). */
401 sock_hold(sk: conn_from_host ? new : sk);
402 vmbus_set_chn_rescind_callback(channel: chan, chn_rescind_cb: hvs_close_connection);
403
404 /* Set the pending send size to max packet size to always get
405 * notifications from the host when there is enough writable space.
406 * The host is optimized to send notifications only when the pending
407 * size boundary is crossed, and not always.
408 */
409 hvs_set_channel_pending_send_size(chan);
410
411 if (conn_from_host) {
412 new->sk_state = TCP_ESTABLISHED;
413 sk_acceptq_added(sk);
414
415 hvs_new->vm_srv_id = *if_type;
416 hvs_new->host_srv_id = *if_instance;
417
418 vsock_insert_connected(vsk: vnew);
419
420 vsock_enqueue_accept(listener: sk, connected: new);
421 } else {
422 sk->sk_state = TCP_ESTABLISHED;
423 sk->sk_socket->state = SS_CONNECTED;
424
425 vsock_insert_connected(vsock_sk(sk));
426 }
427
428 sk->sk_state_change(sk);
429
430out:
431 /* Release refcnt obtained when we called vsock_find_bound_socket() */
432 sock_put(sk);
433
434 release_sock(sk);
435}
436
437static u32 hvs_get_local_cid(void)
438{
439 return VMADDR_CID_ANY;
440}
441
442static int hvs_sock_init(struct vsock_sock *vsk, struct vsock_sock *psk)
443{
444 struct hvsock *hvs;
445 struct sock *sk = sk_vsock(vsk);
446
447 hvs = kzalloc(sizeof(*hvs), GFP_KERNEL);
448 if (!hvs)
449 return -ENOMEM;
450
451 vsk->trans = hvs;
452 hvs->vsk = vsk;
453 sk->sk_sndbuf = RINGBUFFER_HVS_SND_SIZE;
454 sk->sk_rcvbuf = RINGBUFFER_HVS_RCV_SIZE;
455 return 0;
456}
457
458static int hvs_connect(struct vsock_sock *vsk)
459{
460 union hvs_service_id vm, host;
461 struct hvsock *h = vsk->trans;
462
463 vm.srv_id = srv_id_template;
464 vm.svm_port = vsk->local_addr.svm_port;
465 h->vm_srv_id = vm.srv_id;
466
467 host.srv_id = srv_id_template;
468 host.svm_port = vsk->remote_addr.svm_port;
469 h->host_srv_id = host.srv_id;
470
471 return vmbus_send_tl_connect_request(shv_guest_servie_id: &h->vm_srv_id, shv_host_servie_id: &h->host_srv_id);
472}
473
474static void hvs_shutdown_lock_held(struct hvsock *hvs, int mode)
475{
476 struct vmpipe_proto_header hdr;
477
478 if (hvs->fin_sent || !hvs->chan)
479 return;
480
481 /* It can't fail: see hvs_channel_writable_bytes(). */
482 (void)__hvs_send_data(chan: hvs->chan, hdr: &hdr, to_write: 0);
483 hvs->fin_sent = true;
484}
485
486static int hvs_shutdown(struct vsock_sock *vsk, int mode)
487{
488 if (!(mode & SEND_SHUTDOWN))
489 return 0;
490
491 hvs_shutdown_lock_held(hvs: vsk->trans, mode);
492 return 0;
493}
494
495static void hvs_close_timeout(struct work_struct *work)
496{
497 struct vsock_sock *vsk =
498 container_of(work, struct vsock_sock, close_work.work);
499 struct sock *sk = sk_vsock(vsk);
500
501 sock_hold(sk);
502 lock_sock(sk);
503 if (!sock_flag(sk, flag: SOCK_DONE))
504 hvs_do_close_lock_held(vsk, cancel_timeout: false);
505
506 vsk->close_work_scheduled = false;
507 release_sock(sk);
508 sock_put(sk);
509}
510
511/* Returns true, if it is safe to remove socket; false otherwise */
512static bool hvs_close_lock_held(struct vsock_sock *vsk)
513{
514 struct sock *sk = sk_vsock(vsk);
515
516 if (!(sk->sk_state == TCP_ESTABLISHED ||
517 sk->sk_state == TCP_CLOSING))
518 return true;
519
520 if ((sk->sk_shutdown & SHUTDOWN_MASK) != SHUTDOWN_MASK)
521 hvs_shutdown_lock_held(hvs: vsk->trans, SHUTDOWN_MASK);
522
523 if (sock_flag(sk, flag: SOCK_DONE))
524 return true;
525
526 /* This reference will be dropped by the delayed close routine */
527 sock_hold(sk);
528 INIT_DELAYED_WORK(&vsk->close_work, hvs_close_timeout);
529 vsk->close_work_scheduled = true;
530 schedule_delayed_work(dwork: &vsk->close_work, HVS_CLOSE_TIMEOUT);
531 return false;
532}
533
534static void hvs_release(struct vsock_sock *vsk)
535{
536 bool remove_sock;
537
538 remove_sock = hvs_close_lock_held(vsk);
539 if (remove_sock)
540 vsock_remove_sock(vsk);
541}
542
543static void hvs_destruct(struct vsock_sock *vsk)
544{
545 struct hvsock *hvs = vsk->trans;
546 struct vmbus_channel *chan = hvs->chan;
547
548 if (chan)
549 vmbus_hvsock_device_unregister(channel: chan);
550
551 kfree(objp: hvs);
552 vsk->trans = NULL;
553}
554
555static int hvs_dgram_bind(struct vsock_sock *vsk, struct sockaddr_vm *addr)
556{
557 return -EOPNOTSUPP;
558}
559
560static int hvs_dgram_dequeue(struct vsock_sock *vsk, struct msghdr *msg,
561 size_t len, int flags)
562{
563 return -EOPNOTSUPP;
564}
565
566static int hvs_dgram_enqueue(struct vsock_sock *vsk,
567 struct sockaddr_vm *remote, struct msghdr *msg,
568 size_t dgram_len)
569{
570 return -EOPNOTSUPP;
571}
572
573static bool hvs_dgram_allow(u32 cid, u32 port)
574{
575 return false;
576}
577
578static int hvs_update_recv_data(struct hvsock *hvs)
579{
580 struct hvs_recv_buf *recv_buf;
581 u32 pkt_len, payload_len;
582
583 pkt_len = hv_pkt_len(desc: hvs->recv_desc);
584
585 if (pkt_len < HVS_HEADER_LEN)
586 return -EIO;
587
588 recv_buf = (struct hvs_recv_buf *)(hvs->recv_desc + 1);
589 payload_len = recv_buf->hdr.data_size;
590
591 if (payload_len > pkt_len - HVS_HEADER_LEN ||
592 payload_len > HVS_MTU_SIZE)
593 return -EIO;
594
595 if (payload_len == 0)
596 hvs->vsk->peer_shutdown |= SEND_SHUTDOWN;
597
598 hvs->recv_data_len = payload_len;
599 hvs->recv_data_off = 0;
600
601 return 0;
602}
603
604static ssize_t hvs_stream_dequeue(struct vsock_sock *vsk, struct msghdr *msg,
605 size_t len, int flags)
606{
607 struct hvsock *hvs = vsk->trans;
608 bool need_refill = !hvs->recv_desc;
609 struct hvs_recv_buf *recv_buf;
610 u32 to_read;
611 int ret;
612
613 if (flags & MSG_PEEK)
614 return -EOPNOTSUPP;
615
616 if (need_refill) {
617 hvs->recv_desc = hv_pkt_iter_first(channel: hvs->chan);
618 if (!hvs->recv_desc)
619 return -ENOBUFS;
620 ret = hvs_update_recv_data(hvs);
621 if (ret)
622 return ret;
623 }
624
625 recv_buf = (struct hvs_recv_buf *)(hvs->recv_desc + 1);
626 to_read = min_t(u32, len, hvs->recv_data_len);
627 ret = memcpy_to_msg(msg, data: recv_buf->data + hvs->recv_data_off, len: to_read);
628 if (ret != 0)
629 return ret;
630
631 hvs->recv_data_len -= to_read;
632 if (hvs->recv_data_len == 0) {
633 hvs->recv_desc = hv_pkt_iter_next(channel: hvs->chan, pkt: hvs->recv_desc);
634 if (hvs->recv_desc) {
635 ret = hvs_update_recv_data(hvs);
636 if (ret)
637 return ret;
638 }
639 } else {
640 hvs->recv_data_off += to_read;
641 }
642
643 return to_read;
644}
645
646static ssize_t hvs_stream_enqueue(struct vsock_sock *vsk, struct msghdr *msg,
647 size_t len)
648{
649 struct hvsock *hvs = vsk->trans;
650 struct vmbus_channel *chan = hvs->chan;
651 struct hvs_send_buf *send_buf;
652 ssize_t to_write, max_writable;
653 ssize_t ret = 0;
654 ssize_t bytes_written = 0;
655
656 BUILD_BUG_ON(sizeof(*send_buf) != HV_HYP_PAGE_SIZE);
657
658 send_buf = kmalloc(sizeof(*send_buf), GFP_KERNEL);
659 if (!send_buf)
660 return -ENOMEM;
661
662 /* Reader(s) could be draining data from the channel as we write.
663 * Maximize bandwidth, by iterating until the channel is found to be
664 * full.
665 */
666 while (len) {
667 max_writable = hvs_channel_writable_bytes(chan);
668 if (!max_writable)
669 break;
670 to_write = min_t(ssize_t, len, max_writable);
671 to_write = min_t(ssize_t, to_write, HVS_SEND_BUF_SIZE);
672 /* memcpy_from_msg is safe for loop as it advances the offsets
673 * within the message iterator.
674 */
675 ret = memcpy_from_msg(data: send_buf->data, msg, len: to_write);
676 if (ret < 0)
677 goto out;
678
679 ret = hvs_send_data(chan: hvs->chan, send_buf, to_write);
680 if (ret < 0)
681 goto out;
682
683 bytes_written += to_write;
684 len -= to_write;
685 }
686out:
687 /* If any data has been sent, return that */
688 if (bytes_written)
689 ret = bytes_written;
690 kfree(objp: send_buf);
691 return ret;
692}
693
694static s64 hvs_stream_has_data(struct vsock_sock *vsk)
695{
696 struct hvsock *hvs = vsk->trans;
697 bool need_refill;
698 s64 ret;
699
700 if (hvs->recv_data_len > 0)
701 return hvs->recv_data_len;
702
703 switch (hvs_channel_readable_payload(chan: hvs->chan)) {
704 case 1:
705 need_refill = !hvs->recv_desc;
706 if (!need_refill)
707 return -EIO;
708
709 hvs->recv_desc = hv_pkt_iter_first(channel: hvs->chan);
710 if (!hvs->recv_desc)
711 return -ENOBUFS;
712
713 ret = hvs_update_recv_data(hvs);
714 if (ret)
715 return ret;
716 return hvs->recv_data_len;
717 case 0:
718 vsk->peer_shutdown |= SEND_SHUTDOWN;
719 ret = 0;
720 break;
721 default: /* -1 */
722 ret = 0;
723 break;
724 }
725
726 return ret;
727}
728
729static s64 hvs_stream_has_space(struct vsock_sock *vsk)
730{
731 struct hvsock *hvs = vsk->trans;
732
733 return hvs_channel_writable_bytes(chan: hvs->chan);
734}
735
736static u64 hvs_stream_rcvhiwat(struct vsock_sock *vsk)
737{
738 return HVS_MTU_SIZE + 1;
739}
740
741static bool hvs_stream_is_active(struct vsock_sock *vsk)
742{
743 struct hvsock *hvs = vsk->trans;
744
745 return hvs->chan != NULL;
746}
747
748static bool hvs_stream_allow(u32 cid, u32 port)
749{
750 if (cid == VMADDR_CID_HOST)
751 return true;
752
753 return false;
754}
755
756static
757int hvs_notify_poll_in(struct vsock_sock *vsk, size_t target, bool *readable)
758{
759 struct hvsock *hvs = vsk->trans;
760
761 *readable = hvs_channel_readable(chan: hvs->chan);
762 return 0;
763}
764
765static
766int hvs_notify_poll_out(struct vsock_sock *vsk, size_t target, bool *writable)
767{
768 *writable = hvs_stream_has_space(vsk) > 0;
769
770 return 0;
771}
772
773static
774int hvs_notify_recv_init(struct vsock_sock *vsk, size_t target,
775 struct vsock_transport_recv_notify_data *d)
776{
777 return 0;
778}
779
780static
781int hvs_notify_recv_pre_block(struct vsock_sock *vsk, size_t target,
782 struct vsock_transport_recv_notify_data *d)
783{
784 return 0;
785}
786
787static
788int hvs_notify_recv_pre_dequeue(struct vsock_sock *vsk, size_t target,
789 struct vsock_transport_recv_notify_data *d)
790{
791 return 0;
792}
793
794static
795int hvs_notify_recv_post_dequeue(struct vsock_sock *vsk, size_t target,
796 ssize_t copied, bool data_read,
797 struct vsock_transport_recv_notify_data *d)
798{
799 return 0;
800}
801
802static
803int hvs_notify_send_init(struct vsock_sock *vsk,
804 struct vsock_transport_send_notify_data *d)
805{
806 return 0;
807}
808
809static
810int hvs_notify_send_pre_block(struct vsock_sock *vsk,
811 struct vsock_transport_send_notify_data *d)
812{
813 return 0;
814}
815
816static
817int hvs_notify_send_pre_enqueue(struct vsock_sock *vsk,
818 struct vsock_transport_send_notify_data *d)
819{
820 return 0;
821}
822
823static
824int hvs_notify_send_post_enqueue(struct vsock_sock *vsk, ssize_t written,
825 struct vsock_transport_send_notify_data *d)
826{
827 return 0;
828}
829
830static
831int hvs_notify_set_rcvlowat(struct vsock_sock *vsk, int val)
832{
833 return -EOPNOTSUPP;
834}
835
836static struct vsock_transport hvs_transport = {
837 .module = THIS_MODULE,
838
839 .get_local_cid = hvs_get_local_cid,
840
841 .init = hvs_sock_init,
842 .destruct = hvs_destruct,
843 .release = hvs_release,
844 .connect = hvs_connect,
845 .shutdown = hvs_shutdown,
846
847 .dgram_bind = hvs_dgram_bind,
848 .dgram_dequeue = hvs_dgram_dequeue,
849 .dgram_enqueue = hvs_dgram_enqueue,
850 .dgram_allow = hvs_dgram_allow,
851
852 .stream_dequeue = hvs_stream_dequeue,
853 .stream_enqueue = hvs_stream_enqueue,
854 .stream_has_data = hvs_stream_has_data,
855 .stream_has_space = hvs_stream_has_space,
856 .stream_rcvhiwat = hvs_stream_rcvhiwat,
857 .stream_is_active = hvs_stream_is_active,
858 .stream_allow = hvs_stream_allow,
859
860 .notify_poll_in = hvs_notify_poll_in,
861 .notify_poll_out = hvs_notify_poll_out,
862 .notify_recv_init = hvs_notify_recv_init,
863 .notify_recv_pre_block = hvs_notify_recv_pre_block,
864 .notify_recv_pre_dequeue = hvs_notify_recv_pre_dequeue,
865 .notify_recv_post_dequeue = hvs_notify_recv_post_dequeue,
866 .notify_send_init = hvs_notify_send_init,
867 .notify_send_pre_block = hvs_notify_send_pre_block,
868 .notify_send_pre_enqueue = hvs_notify_send_pre_enqueue,
869 .notify_send_post_enqueue = hvs_notify_send_post_enqueue,
870
871 .notify_set_rcvlowat = hvs_notify_set_rcvlowat
872};
873
874static bool hvs_check_transport(struct vsock_sock *vsk)
875{
876 return vsk->transport == &hvs_transport;
877}
878
879static int hvs_probe(struct hv_device *hdev,
880 const struct hv_vmbus_device_id *dev_id)
881{
882 struct vmbus_channel *chan = hdev->channel;
883
884 hvs_open_connection(chan);
885
886 /* Always return success to suppress the unnecessary error message
887 * in vmbus_probe(): on error the host will rescind the device in
888 * 30 seconds and we can do cleanup at that time in
889 * vmbus_onoffer_rescind().
890 */
891 return 0;
892}
893
894static void hvs_remove(struct hv_device *hdev)
895{
896 struct vmbus_channel *chan = hdev->channel;
897
898 vmbus_close(channel: chan);
899}
900
901/* hv_sock connections can not persist across hibernation, and all the hv_sock
902 * channels are forced to be rescinded before hibernation: see
903 * vmbus_bus_suspend(). Here the dummy hvs_suspend() and hvs_resume()
904 * are only needed because hibernation requires that every vmbus device's
905 * driver should have a .suspend and .resume callback: see vmbus_suspend().
906 */
907static int hvs_suspend(struct hv_device *hv_dev)
908{
909 /* Dummy */
910 return 0;
911}
912
913static int hvs_resume(struct hv_device *dev)
914{
915 /* Dummy */
916 return 0;
917}
918
919/* This isn't really used. See vmbus_match() and vmbus_probe() */
920static const struct hv_vmbus_device_id id_table[] = {
921 {},
922};
923
924static struct hv_driver hvs_drv = {
925 .name = "hv_sock",
926 .hvsock = true,
927 .id_table = id_table,
928 .probe = hvs_probe,
929 .remove = hvs_remove,
930 .suspend = hvs_suspend,
931 .resume = hvs_resume,
932};
933
934static int __init hvs_init(void)
935{
936 int ret;
937
938 if (vmbus_proto_version < VERSION_WIN10)
939 return -ENODEV;
940
941 ret = vmbus_driver_register(&hvs_drv);
942 if (ret != 0)
943 return ret;
944
945 ret = vsock_core_register(t: &hvs_transport, VSOCK_TRANSPORT_F_G2H);
946 if (ret) {
947 vmbus_driver_unregister(hv_driver: &hvs_drv);
948 return ret;
949 }
950
951 return 0;
952}
953
954static void __exit hvs_exit(void)
955{
956 vsock_core_unregister(t: &hvs_transport);
957 vmbus_driver_unregister(hv_driver: &hvs_drv);
958}
959
960module_init(hvs_init);
961module_exit(hvs_exit);
962
963MODULE_DESCRIPTION("Hyper-V Sockets");
964MODULE_VERSION("1.0.0");
965MODULE_LICENSE("GPL");
966MODULE_ALIAS_NETPROTO(PF_VSOCK);
967

source code of linux/net/vmw_vsock/hyperv_transport.c