Commit | Line | Data |
---|---|---|
ae0078fc DC |
1 | /* |
2 | * Hyper-V transport for vsock | |
3 | * | |
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. | |
7 | * | |
8 | * Copyright (c) 2017, Microsoft Corporation. | |
9 | * | |
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. | |
13 | * | |
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 | |
17 | * more details. | |
18 | * | |
19 | */ | |
20 | #include <linux/module.h> | |
21 | #include <linux/vmalloc.h> | |
22 | #include <linux/hyperv.h> | |
23 | #include <net/sock.h> | |
24 | #include <net/af_vsock.h> | |
25 | ||
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. | |
30 | */ | |
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) | |
34 | ||
35 | /* The MTU is 16KB per the host side's design */ | |
36 | #define HVS_MTU_SIZE (1024 * 16) | |
37 | ||
38 | struct vmpipe_proto_header { | |
39 | u32 pkt_type; | |
40 | u32 data_size; | |
41 | }; | |
42 | ||
43 | /* For recv, we use the VMBus in-place packet iterator APIs to directly copy | |
44 | * data from the ringbuffer into the userspace buffer. | |
45 | */ | |
46 | struct hvs_recv_buf { | |
47 | /* The header before the payload data */ | |
48 | struct vmpipe_proto_header hdr; | |
49 | ||
50 | /* The payload */ | |
51 | u8 data[HVS_MTU_SIZE]; | |
52 | }; | |
53 | ||
54 | /* We can send up to HVS_MTU_SIZE bytes of payload to the host, but let's use | |
55 | * a small size, i.e. HVS_SEND_BUF_SIZE, to minimize the dynamically-allocated | |
56 | * buffer, because tests show there is no significant performance difference. | |
57 | * | |
58 | * Note: the buffer can be eliminated in the future when we add new VMBus | |
59 | * ringbuffer APIs that allow us to directly copy data from userspace buffer | |
60 | * to VMBus ringbuffer. | |
61 | */ | |
62 | #define HVS_SEND_BUF_SIZE (PAGE_SIZE_4K - sizeof(struct vmpipe_proto_header)) | |
63 | ||
64 | struct hvs_send_buf { | |
65 | /* The header before the payload data */ | |
66 | struct vmpipe_proto_header hdr; | |
67 | ||
68 | /* The payload */ | |
69 | u8 data[HVS_SEND_BUF_SIZE]; | |
70 | }; | |
71 | ||
72 | #define HVS_HEADER_LEN (sizeof(struct vmpacket_descriptor) + \ | |
73 | sizeof(struct vmpipe_proto_header)) | |
74 | ||
75 | /* See 'prev_indices' in hv_ringbuffer_read(), hv_ringbuffer_write(), and | |
76 | * __hv_pkt_iter_next(). | |
77 | */ | |
78 | #define VMBUS_PKT_TRAILER_SIZE (sizeof(u64)) | |
79 | ||
80 | #define HVS_PKT_LEN(payload_len) (HVS_HEADER_LEN + \ | |
81 | ALIGN((payload_len), 8) + \ | |
82 | VMBUS_PKT_TRAILER_SIZE) | |
83 | ||
84 | union hvs_service_id { | |
85 | uuid_le srv_id; | |
86 | ||
87 | struct { | |
88 | unsigned int svm_port; | |
89 | unsigned char b[sizeof(uuid_le) - sizeof(unsigned int)]; | |
90 | }; | |
91 | }; | |
92 | ||
93 | /* Per-socket state (accessed via vsk->trans) */ | |
94 | struct hvsock { | |
95 | struct vsock_sock *vsk; | |
96 | ||
97 | uuid_le vm_srv_id; | |
98 | uuid_le 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, and the "port" must be in * | |
145 | * this range [0, 0x7FFFFFFF]. * | |
146 | **************************************************************************** | |
147 | * | |
148 | * When we write apps on the host to connect(), the GUID ServiceID is used. | |
149 | * When we write apps in Linux VM to connect(), we only need to specify the | |
150 | * port and the driver will form the GUID and use that to request the host. | |
151 | * | |
152 | * From the perspective of Linux VM: | |
153 | * 1. the local ephemeral port (i.e. the local auto-bound port when we call | |
154 | * connect() without explicit bind()) is generated by __vsock_bind_stream(), | |
155 | * and the range is [1024, 0xFFFFFFFF). | |
156 | * 2. the remote ephemeral port (i.e. the auto-generated remote port for | |
157 | * a connect request initiated by the host's connect()) is generated by | |
158 | * hvs_remote_addr_init() and the range is [0x80000000, 0xFFFFFFFF). | |
159 | */ | |
160 | ||
161 | #define MAX_LISTEN_PORT ((u32)0x7FFFFFFF) | |
162 | #define MAX_VM_LISTEN_PORT MAX_LISTEN_PORT | |
163 | #define MAX_HOST_LISTEN_PORT MAX_LISTEN_PORT | |
164 | #define MIN_HOST_EPHEMERAL_PORT (MAX_HOST_LISTEN_PORT + 1) | |
165 | ||
166 | /* 00000000-facb-11e6-bd58-64006a7986d3 */ | |
167 | static const uuid_le srv_id_template = | |
168 | UUID_LE(0x00000000, 0xfacb, 0x11e6, 0xbd, 0x58, | |
169 | 0x64, 0x00, 0x6a, 0x79, 0x86, 0xd3); | |
170 | ||
171 | static bool is_valid_srv_id(const uuid_le *id) | |
172 | { | |
173 | return !memcmp(&id->b[4], &srv_id_template.b[4], sizeof(uuid_le) - 4); | |
174 | } | |
175 | ||
176 | static unsigned int get_port_by_srv_id(const uuid_le *svr_id) | |
177 | { | |
178 | return *((unsigned int *)svr_id); | |
179 | } | |
180 | ||
181 | static void hvs_addr_init(struct sockaddr_vm *addr, const uuid_le *svr_id) | |
182 | { | |
183 | unsigned int port = get_port_by_srv_id(svr_id); | |
184 | ||
185 | vsock_addr_init(addr, VMADDR_CID_ANY, port); | |
186 | } | |
187 | ||
188 | static void hvs_remote_addr_init(struct sockaddr_vm *remote, | |
189 | struct sockaddr_vm *local) | |
190 | { | |
191 | static u32 host_ephemeral_port = MIN_HOST_EPHEMERAL_PORT; | |
192 | struct sock *sk; | |
193 | ||
194 | vsock_addr_init(remote, VMADDR_CID_ANY, VMADDR_PORT_ANY); | |
195 | ||
196 | while (1) { | |
197 | /* Wrap around ? */ | |
198 | if (host_ephemeral_port < MIN_HOST_EPHEMERAL_PORT || | |
199 | host_ephemeral_port == VMADDR_PORT_ANY) | |
200 | host_ephemeral_port = MIN_HOST_EPHEMERAL_PORT; | |
201 | ||
202 | remote->svm_port = host_ephemeral_port++; | |
203 | ||
204 | sk = vsock_find_connected_socket(remote, local); | |
205 | if (!sk) { | |
206 | /* Found an available ephemeral port */ | |
207 | return; | |
208 | } | |
209 | ||
210 | /* Release refcnt got in vsock_find_connected_socket */ | |
211 | sock_put(sk); | |
212 | } | |
213 | } | |
214 | ||
215 | static void hvs_set_channel_pending_send_size(struct vmbus_channel *chan) | |
216 | { | |
217 | set_channel_pending_send_size(chan, | |
218 | HVS_PKT_LEN(HVS_SEND_BUF_SIZE)); | |
219 | ||
220 | /* See hvs_stream_has_space(): we must make sure the host has seen | |
221 | * the new pending send size, before we can re-check the writable | |
222 | * bytes. | |
223 | */ | |
224 | virt_mb(); | |
225 | } | |
226 | ||
227 | static void hvs_clear_channel_pending_send_size(struct vmbus_channel *chan) | |
228 | { | |
229 | set_channel_pending_send_size(chan, 0); | |
230 | ||
231 | /* Ditto */ | |
232 | virt_mb(); | |
233 | } | |
234 | ||
235 | static bool hvs_channel_readable(struct vmbus_channel *chan) | |
236 | { | |
237 | u32 readable = hv_get_bytes_to_read(&chan->inbound); | |
238 | ||
239 | /* 0-size payload means FIN */ | |
240 | return readable >= HVS_PKT_LEN(0); | |
241 | } | |
242 | ||
243 | static int hvs_channel_readable_payload(struct vmbus_channel *chan) | |
244 | { | |
245 | u32 readable = hv_get_bytes_to_read(&chan->inbound); | |
246 | ||
247 | if (readable > HVS_PKT_LEN(0)) { | |
248 | /* At least we have 1 byte to read. We don't need to return | |
249 | * the exact readable bytes: see vsock_stream_recvmsg() -> | |
250 | * vsock_stream_has_data(). | |
251 | */ | |
252 | return 1; | |
253 | } | |
254 | ||
255 | if (readable == HVS_PKT_LEN(0)) { | |
256 | /* 0-size payload means FIN */ | |
257 | return 0; | |
258 | } | |
259 | ||
260 | /* No payload or FIN */ | |
261 | return -1; | |
262 | } | |
263 | ||
264 | static size_t hvs_channel_writable_bytes(struct vmbus_channel *chan) | |
265 | { | |
266 | u32 writeable = hv_get_bytes_to_write(&chan->outbound); | |
267 | size_t ret; | |
268 | ||
269 | /* The ringbuffer mustn't be 100% full, and we should reserve a | |
270 | * zero-length-payload packet for the FIN: see hv_ringbuffer_write() | |
271 | * and hvs_shutdown(). | |
272 | */ | |
273 | if (writeable <= HVS_PKT_LEN(1) + HVS_PKT_LEN(0)) | |
274 | return 0; | |
275 | ||
276 | ret = writeable - HVS_PKT_LEN(1) - HVS_PKT_LEN(0); | |
277 | ||
278 | return round_down(ret, 8); | |
279 | } | |
280 | ||
281 | static int hvs_send_data(struct vmbus_channel *chan, | |
282 | struct hvs_send_buf *send_buf, size_t to_write) | |
283 | { | |
284 | send_buf->hdr.pkt_type = 1; | |
285 | send_buf->hdr.data_size = to_write; | |
286 | return vmbus_sendpacket(chan, &send_buf->hdr, | |
287 | sizeof(send_buf->hdr) + to_write, | |
288 | 0, VM_PKT_DATA_INBAND, 0); | |
289 | } | |
290 | ||
291 | static void hvs_channel_cb(void *ctx) | |
292 | { | |
293 | struct sock *sk = (struct sock *)ctx; | |
294 | struct vsock_sock *vsk = vsock_sk(sk); | |
295 | struct hvsock *hvs = vsk->trans; | |
296 | struct vmbus_channel *chan = hvs->chan; | |
297 | ||
298 | if (hvs_channel_readable(chan)) | |
299 | sk->sk_data_ready(sk); | |
300 | ||
301 | /* See hvs_stream_has_space(): when we reach here, the writable bytes | |
302 | * may be already less than HVS_PKT_LEN(HVS_SEND_BUF_SIZE). | |
303 | */ | |
304 | if (hv_get_bytes_to_write(&chan->outbound) > 0) | |
305 | sk->sk_write_space(sk); | |
306 | } | |
307 | ||
308 | static void hvs_close_connection(struct vmbus_channel *chan) | |
309 | { | |
310 | struct sock *sk = get_per_channel_state(chan); | |
311 | struct vsock_sock *vsk = vsock_sk(sk); | |
312 | ||
b4562ca7 DC |
313 | lock_sock(sk); |
314 | ||
3b4477d2 | 315 | sk->sk_state = TCP_CLOSE; |
ae0078fc DC |
316 | sock_set_flag(sk, SOCK_DONE); |
317 | vsk->peer_shutdown |= SEND_SHUTDOWN | RCV_SHUTDOWN; | |
318 | ||
319 | sk->sk_state_change(sk); | |
b4562ca7 DC |
320 | |
321 | release_sock(sk); | |
ae0078fc DC |
322 | } |
323 | ||
324 | static void hvs_open_connection(struct vmbus_channel *chan) | |
325 | { | |
326 | uuid_le *if_instance, *if_type; | |
327 | unsigned char conn_from_host; | |
328 | ||
329 | struct sockaddr_vm addr; | |
330 | struct sock *sk, *new = NULL; | |
331 | struct vsock_sock *vnew; | |
332 | struct hvsock *hvs, *hvs_new; | |
333 | int ret; | |
334 | ||
335 | if_type = &chan->offermsg.offer.if_type; | |
336 | if_instance = &chan->offermsg.offer.if_instance; | |
337 | conn_from_host = chan->offermsg.offer.u.pipe.user_def[0]; | |
338 | ||
339 | /* The host or the VM should only listen on a port in | |
340 | * [0, MAX_LISTEN_PORT] | |
341 | */ | |
342 | if (!is_valid_srv_id(if_type) || | |
343 | get_port_by_srv_id(if_type) > MAX_LISTEN_PORT) | |
344 | return; | |
345 | ||
346 | hvs_addr_init(&addr, conn_from_host ? if_type : if_instance); | |
347 | sk = vsock_find_bound_socket(&addr); | |
348 | if (!sk) | |
349 | return; | |
350 | ||
b4562ca7 | 351 | lock_sock(sk); |
3b4477d2 SH |
352 | if ((conn_from_host && sk->sk_state != TCP_LISTEN) || |
353 | (!conn_from_host && sk->sk_state != TCP_SYN_SENT)) | |
ae0078fc DC |
354 | goto out; |
355 | ||
356 | if (conn_from_host) { | |
357 | if (sk->sk_ack_backlog >= sk->sk_max_ack_backlog) | |
358 | goto out; | |
359 | ||
360 | new = __vsock_create(sock_net(sk), NULL, sk, GFP_KERNEL, | |
361 | sk->sk_type, 0); | |
362 | if (!new) | |
363 | goto out; | |
364 | ||
3b4477d2 | 365 | new->sk_state = TCP_SYN_SENT; |
ae0078fc DC |
366 | vnew = vsock_sk(new); |
367 | hvs_new = vnew->trans; | |
368 | hvs_new->chan = chan; | |
369 | } else { | |
370 | hvs = vsock_sk(sk)->trans; | |
371 | hvs->chan = chan; | |
372 | } | |
373 | ||
374 | set_channel_read_mode(chan, HV_CALL_DIRECT); | |
375 | ret = vmbus_open(chan, RINGBUFFER_HVS_SND_SIZE, | |
376 | RINGBUFFER_HVS_RCV_SIZE, NULL, 0, | |
377 | hvs_channel_cb, conn_from_host ? new : sk); | |
378 | if (ret != 0) { | |
379 | if (conn_from_host) { | |
380 | hvs_new->chan = NULL; | |
381 | sock_put(new); | |
382 | } else { | |
383 | hvs->chan = NULL; | |
384 | } | |
385 | goto out; | |
386 | } | |
387 | ||
388 | set_per_channel_state(chan, conn_from_host ? new : sk); | |
389 | vmbus_set_chn_rescind_callback(chan, hvs_close_connection); | |
390 | ||
391 | if (conn_from_host) { | |
3b4477d2 | 392 | new->sk_state = TCP_ESTABLISHED; |
ae0078fc DC |
393 | sk->sk_ack_backlog++; |
394 | ||
395 | hvs_addr_init(&vnew->local_addr, if_type); | |
396 | hvs_remote_addr_init(&vnew->remote_addr, &vnew->local_addr); | |
397 | ||
398 | hvs_new->vm_srv_id = *if_type; | |
399 | hvs_new->host_srv_id = *if_instance; | |
400 | ||
401 | vsock_insert_connected(vnew); | |
402 | ||
ae0078fc | 403 | vsock_enqueue_accept(sk, new); |
ae0078fc | 404 | } else { |
3b4477d2 | 405 | sk->sk_state = TCP_ESTABLISHED; |
ae0078fc DC |
406 | sk->sk_socket->state = SS_CONNECTED; |
407 | ||
408 | vsock_insert_connected(vsock_sk(sk)); | |
409 | } | |
410 | ||
411 | sk->sk_state_change(sk); | |
412 | ||
413 | out: | |
414 | /* Release refcnt obtained when we called vsock_find_bound_socket() */ | |
415 | sock_put(sk); | |
b4562ca7 DC |
416 | |
417 | release_sock(sk); | |
ae0078fc DC |
418 | } |
419 | ||
420 | static u32 hvs_get_local_cid(void) | |
421 | { | |
422 | return VMADDR_CID_ANY; | |
423 | } | |
424 | ||
425 | static int hvs_sock_init(struct vsock_sock *vsk, struct vsock_sock *psk) | |
426 | { | |
427 | struct hvsock *hvs; | |
428 | ||
429 | hvs = kzalloc(sizeof(*hvs), GFP_KERNEL); | |
430 | if (!hvs) | |
431 | return -ENOMEM; | |
432 | ||
433 | vsk->trans = hvs; | |
434 | hvs->vsk = vsk; | |
435 | ||
436 | return 0; | |
437 | } | |
438 | ||
439 | static int hvs_connect(struct vsock_sock *vsk) | |
440 | { | |
441 | union hvs_service_id vm, host; | |
442 | struct hvsock *h = vsk->trans; | |
443 | ||
444 | vm.srv_id = srv_id_template; | |
445 | vm.svm_port = vsk->local_addr.svm_port; | |
446 | h->vm_srv_id = vm.srv_id; | |
447 | ||
448 | host.srv_id = srv_id_template; | |
449 | host.svm_port = vsk->remote_addr.svm_port; | |
450 | h->host_srv_id = host.srv_id; | |
451 | ||
452 | return vmbus_send_tl_connect_request(&h->vm_srv_id, &h->host_srv_id); | |
453 | } | |
454 | ||
455 | static int hvs_shutdown(struct vsock_sock *vsk, int mode) | |
456 | { | |
457 | struct sock *sk = sk_vsock(vsk); | |
458 | struct vmpipe_proto_header hdr; | |
459 | struct hvs_send_buf *send_buf; | |
460 | struct hvsock *hvs; | |
461 | ||
462 | if (!(mode & SEND_SHUTDOWN)) | |
463 | return 0; | |
464 | ||
465 | lock_sock(sk); | |
466 | ||
467 | hvs = vsk->trans; | |
468 | if (hvs->fin_sent) | |
469 | goto out; | |
470 | ||
471 | send_buf = (struct hvs_send_buf *)&hdr; | |
472 | ||
473 | /* It can't fail: see hvs_channel_writable_bytes(). */ | |
474 | (void)hvs_send_data(hvs->chan, send_buf, 0); | |
475 | ||
476 | hvs->fin_sent = true; | |
477 | out: | |
478 | release_sock(sk); | |
479 | return 0; | |
480 | } | |
481 | ||
482 | static void hvs_release(struct vsock_sock *vsk) | |
483 | { | |
b4562ca7 | 484 | struct sock *sk = sk_vsock(vsk); |
ae0078fc | 485 | struct hvsock *hvs = vsk->trans; |
b4562ca7 | 486 | struct vmbus_channel *chan; |
ae0078fc | 487 | |
b4562ca7 DC |
488 | lock_sock(sk); |
489 | ||
490 | sk->sk_state = SS_DISCONNECTING; | |
491 | vsock_remove_sock(vsk); | |
492 | ||
493 | release_sock(sk); | |
ae0078fc | 494 | |
b4562ca7 | 495 | chan = hvs->chan; |
ae0078fc DC |
496 | if (chan) |
497 | hvs_shutdown(vsk, RCV_SHUTDOWN | SEND_SHUTDOWN); | |
498 | ||
ae0078fc DC |
499 | } |
500 | ||
501 | static void hvs_destruct(struct vsock_sock *vsk) | |
502 | { | |
503 | struct hvsock *hvs = vsk->trans; | |
504 | struct vmbus_channel *chan = hvs->chan; | |
505 | ||
506 | if (chan) | |
507 | vmbus_hvsock_device_unregister(chan); | |
508 | ||
509 | kfree(hvs); | |
510 | } | |
511 | ||
512 | static int hvs_dgram_bind(struct vsock_sock *vsk, struct sockaddr_vm *addr) | |
513 | { | |
514 | return -EOPNOTSUPP; | |
515 | } | |
516 | ||
517 | static int hvs_dgram_dequeue(struct vsock_sock *vsk, struct msghdr *msg, | |
518 | size_t len, int flags) | |
519 | { | |
520 | return -EOPNOTSUPP; | |
521 | } | |
522 | ||
523 | static int hvs_dgram_enqueue(struct vsock_sock *vsk, | |
524 | struct sockaddr_vm *remote, struct msghdr *msg, | |
525 | size_t dgram_len) | |
526 | { | |
527 | return -EOPNOTSUPP; | |
528 | } | |
529 | ||
530 | static bool hvs_dgram_allow(u32 cid, u32 port) | |
531 | { | |
532 | return false; | |
533 | } | |
534 | ||
535 | static int hvs_update_recv_data(struct hvsock *hvs) | |
536 | { | |
537 | struct hvs_recv_buf *recv_buf; | |
538 | u32 payload_len; | |
539 | ||
540 | recv_buf = (struct hvs_recv_buf *)(hvs->recv_desc + 1); | |
541 | payload_len = recv_buf->hdr.data_size; | |
542 | ||
543 | if (payload_len > HVS_MTU_SIZE) | |
544 | return -EIO; | |
545 | ||
546 | if (payload_len == 0) | |
547 | hvs->vsk->peer_shutdown |= SEND_SHUTDOWN; | |
548 | ||
549 | hvs->recv_data_len = payload_len; | |
550 | hvs->recv_data_off = 0; | |
551 | ||
552 | return 0; | |
553 | } | |
554 | ||
555 | static ssize_t hvs_stream_dequeue(struct vsock_sock *vsk, struct msghdr *msg, | |
556 | size_t len, int flags) | |
557 | { | |
558 | struct hvsock *hvs = vsk->trans; | |
559 | bool need_refill = !hvs->recv_desc; | |
560 | struct hvs_recv_buf *recv_buf; | |
561 | u32 to_read; | |
562 | int ret; | |
563 | ||
564 | if (flags & MSG_PEEK) | |
565 | return -EOPNOTSUPP; | |
566 | ||
567 | if (need_refill) { | |
568 | hvs->recv_desc = hv_pkt_iter_first(hvs->chan); | |
569 | ret = hvs_update_recv_data(hvs); | |
570 | if (ret) | |
571 | return ret; | |
572 | } | |
573 | ||
574 | recv_buf = (struct hvs_recv_buf *)(hvs->recv_desc + 1); | |
575 | to_read = min_t(u32, len, hvs->recv_data_len); | |
576 | ret = memcpy_to_msg(msg, recv_buf->data + hvs->recv_data_off, to_read); | |
577 | if (ret != 0) | |
578 | return ret; | |
579 | ||
580 | hvs->recv_data_len -= to_read; | |
581 | if (hvs->recv_data_len == 0) { | |
582 | hvs->recv_desc = hv_pkt_iter_next(hvs->chan, hvs->recv_desc); | |
583 | if (hvs->recv_desc) { | |
584 | ret = hvs_update_recv_data(hvs); | |
585 | if (ret) | |
586 | return ret; | |
587 | } | |
588 | } else { | |
589 | hvs->recv_data_off += to_read; | |
590 | } | |
591 | ||
592 | return to_read; | |
593 | } | |
594 | ||
595 | static ssize_t hvs_stream_enqueue(struct vsock_sock *vsk, struct msghdr *msg, | |
596 | size_t len) | |
597 | { | |
598 | struct hvsock *hvs = vsk->trans; | |
599 | struct vmbus_channel *chan = hvs->chan; | |
600 | struct hvs_send_buf *send_buf; | |
601 | ssize_t to_write, max_writable, ret; | |
602 | ||
603 | BUILD_BUG_ON(sizeof(*send_buf) != PAGE_SIZE_4K); | |
604 | ||
605 | send_buf = kmalloc(sizeof(*send_buf), GFP_KERNEL); | |
606 | if (!send_buf) | |
607 | return -ENOMEM; | |
608 | ||
609 | max_writable = hvs_channel_writable_bytes(chan); | |
610 | to_write = min_t(ssize_t, len, max_writable); | |
611 | to_write = min_t(ssize_t, to_write, HVS_SEND_BUF_SIZE); | |
612 | ||
613 | ret = memcpy_from_msg(send_buf->data, msg, to_write); | |
614 | if (ret < 0) | |
615 | goto out; | |
616 | ||
617 | ret = hvs_send_data(hvs->chan, send_buf, to_write); | |
618 | if (ret < 0) | |
619 | goto out; | |
620 | ||
621 | ret = to_write; | |
622 | out: | |
623 | kfree(send_buf); | |
624 | return ret; | |
625 | } | |
626 | ||
627 | static s64 hvs_stream_has_data(struct vsock_sock *vsk) | |
628 | { | |
629 | struct hvsock *hvs = vsk->trans; | |
630 | s64 ret; | |
631 | ||
632 | if (hvs->recv_data_len > 0) | |
633 | return 1; | |
634 | ||
635 | switch (hvs_channel_readable_payload(hvs->chan)) { | |
636 | case 1: | |
637 | ret = 1; | |
638 | break; | |
639 | case 0: | |
640 | vsk->peer_shutdown |= SEND_SHUTDOWN; | |
641 | ret = 0; | |
642 | break; | |
643 | default: /* -1 */ | |
644 | ret = 0; | |
645 | break; | |
646 | } | |
647 | ||
648 | return ret; | |
649 | } | |
650 | ||
651 | static s64 hvs_stream_has_space(struct vsock_sock *vsk) | |
652 | { | |
653 | struct hvsock *hvs = vsk->trans; | |
654 | struct vmbus_channel *chan = hvs->chan; | |
655 | s64 ret; | |
656 | ||
657 | ret = hvs_channel_writable_bytes(chan); | |
658 | if (ret > 0) { | |
659 | hvs_clear_channel_pending_send_size(chan); | |
660 | } else { | |
661 | /* See hvs_channel_cb() */ | |
662 | hvs_set_channel_pending_send_size(chan); | |
663 | ||
664 | /* Re-check the writable bytes to avoid race */ | |
665 | ret = hvs_channel_writable_bytes(chan); | |
666 | if (ret > 0) | |
667 | hvs_clear_channel_pending_send_size(chan); | |
668 | } | |
669 | ||
670 | return ret; | |
671 | } | |
672 | ||
673 | static u64 hvs_stream_rcvhiwat(struct vsock_sock *vsk) | |
674 | { | |
675 | return HVS_MTU_SIZE + 1; | |
676 | } | |
677 | ||
678 | static bool hvs_stream_is_active(struct vsock_sock *vsk) | |
679 | { | |
680 | struct hvsock *hvs = vsk->trans; | |
681 | ||
682 | return hvs->chan != NULL; | |
683 | } | |
684 | ||
685 | static bool hvs_stream_allow(u32 cid, u32 port) | |
686 | { | |
687 | /* The host's port range [MIN_HOST_EPHEMERAL_PORT, 0xFFFFFFFF) is | |
688 | * reserved as ephemeral ports, which are used as the host's ports | |
689 | * when the host initiates connections. | |
690 | * | |
691 | * Perform this check in the guest so an immediate error is produced | |
692 | * instead of a timeout. | |
693 | */ | |
694 | if (port > MAX_HOST_LISTEN_PORT) | |
695 | return false; | |
696 | ||
697 | if (cid == VMADDR_CID_HOST) | |
698 | return true; | |
699 | ||
700 | return false; | |
701 | } | |
702 | ||
703 | static | |
704 | int hvs_notify_poll_in(struct vsock_sock *vsk, size_t target, bool *readable) | |
705 | { | |
706 | struct hvsock *hvs = vsk->trans; | |
707 | ||
708 | *readable = hvs_channel_readable(hvs->chan); | |
709 | return 0; | |
710 | } | |
711 | ||
712 | static | |
713 | int hvs_notify_poll_out(struct vsock_sock *vsk, size_t target, bool *writable) | |
714 | { | |
715 | *writable = hvs_stream_has_space(vsk) > 0; | |
716 | ||
717 | return 0; | |
718 | } | |
719 | ||
720 | static | |
721 | int hvs_notify_recv_init(struct vsock_sock *vsk, size_t target, | |
722 | struct vsock_transport_recv_notify_data *d) | |
723 | { | |
724 | return 0; | |
725 | } | |
726 | ||
727 | static | |
728 | int hvs_notify_recv_pre_block(struct vsock_sock *vsk, size_t target, | |
729 | struct vsock_transport_recv_notify_data *d) | |
730 | { | |
731 | return 0; | |
732 | } | |
733 | ||
734 | static | |
735 | int hvs_notify_recv_pre_dequeue(struct vsock_sock *vsk, size_t target, | |
736 | struct vsock_transport_recv_notify_data *d) | |
737 | { | |
738 | return 0; | |
739 | } | |
740 | ||
741 | static | |
742 | int hvs_notify_recv_post_dequeue(struct vsock_sock *vsk, size_t target, | |
743 | ssize_t copied, bool data_read, | |
744 | struct vsock_transport_recv_notify_data *d) | |
745 | { | |
746 | return 0; | |
747 | } | |
748 | ||
749 | static | |
750 | int hvs_notify_send_init(struct vsock_sock *vsk, | |
751 | struct vsock_transport_send_notify_data *d) | |
752 | { | |
753 | return 0; | |
754 | } | |
755 | ||
756 | static | |
757 | int hvs_notify_send_pre_block(struct vsock_sock *vsk, | |
758 | struct vsock_transport_send_notify_data *d) | |
759 | { | |
760 | return 0; | |
761 | } | |
762 | ||
763 | static | |
764 | int hvs_notify_send_pre_enqueue(struct vsock_sock *vsk, | |
765 | struct vsock_transport_send_notify_data *d) | |
766 | { | |
767 | return 0; | |
768 | } | |
769 | ||
770 | static | |
771 | int hvs_notify_send_post_enqueue(struct vsock_sock *vsk, ssize_t written, | |
772 | struct vsock_transport_send_notify_data *d) | |
773 | { | |
774 | return 0; | |
775 | } | |
776 | ||
777 | static void hvs_set_buffer_size(struct vsock_sock *vsk, u64 val) | |
778 | { | |
779 | /* Ignored. */ | |
780 | } | |
781 | ||
782 | static void hvs_set_min_buffer_size(struct vsock_sock *vsk, u64 val) | |
783 | { | |
784 | /* Ignored. */ | |
785 | } | |
786 | ||
787 | static void hvs_set_max_buffer_size(struct vsock_sock *vsk, u64 val) | |
788 | { | |
789 | /* Ignored. */ | |
790 | } | |
791 | ||
792 | static u64 hvs_get_buffer_size(struct vsock_sock *vsk) | |
793 | { | |
794 | return -ENOPROTOOPT; | |
795 | } | |
796 | ||
797 | static u64 hvs_get_min_buffer_size(struct vsock_sock *vsk) | |
798 | { | |
799 | return -ENOPROTOOPT; | |
800 | } | |
801 | ||
802 | static u64 hvs_get_max_buffer_size(struct vsock_sock *vsk) | |
803 | { | |
804 | return -ENOPROTOOPT; | |
805 | } | |
806 | ||
807 | static struct vsock_transport hvs_transport = { | |
808 | .get_local_cid = hvs_get_local_cid, | |
809 | ||
810 | .init = hvs_sock_init, | |
811 | .destruct = hvs_destruct, | |
812 | .release = hvs_release, | |
813 | .connect = hvs_connect, | |
814 | .shutdown = hvs_shutdown, | |
815 | ||
816 | .dgram_bind = hvs_dgram_bind, | |
817 | .dgram_dequeue = hvs_dgram_dequeue, | |
818 | .dgram_enqueue = hvs_dgram_enqueue, | |
819 | .dgram_allow = hvs_dgram_allow, | |
820 | ||
821 | .stream_dequeue = hvs_stream_dequeue, | |
822 | .stream_enqueue = hvs_stream_enqueue, | |
823 | .stream_has_data = hvs_stream_has_data, | |
824 | .stream_has_space = hvs_stream_has_space, | |
825 | .stream_rcvhiwat = hvs_stream_rcvhiwat, | |
826 | .stream_is_active = hvs_stream_is_active, | |
827 | .stream_allow = hvs_stream_allow, | |
828 | ||
829 | .notify_poll_in = hvs_notify_poll_in, | |
830 | .notify_poll_out = hvs_notify_poll_out, | |
831 | .notify_recv_init = hvs_notify_recv_init, | |
832 | .notify_recv_pre_block = hvs_notify_recv_pre_block, | |
833 | .notify_recv_pre_dequeue = hvs_notify_recv_pre_dequeue, | |
834 | .notify_recv_post_dequeue = hvs_notify_recv_post_dequeue, | |
835 | .notify_send_init = hvs_notify_send_init, | |
836 | .notify_send_pre_block = hvs_notify_send_pre_block, | |
837 | .notify_send_pre_enqueue = hvs_notify_send_pre_enqueue, | |
838 | .notify_send_post_enqueue = hvs_notify_send_post_enqueue, | |
839 | ||
840 | .set_buffer_size = hvs_set_buffer_size, | |
841 | .set_min_buffer_size = hvs_set_min_buffer_size, | |
842 | .set_max_buffer_size = hvs_set_max_buffer_size, | |
843 | .get_buffer_size = hvs_get_buffer_size, | |
844 | .get_min_buffer_size = hvs_get_min_buffer_size, | |
845 | .get_max_buffer_size = hvs_get_max_buffer_size, | |
846 | }; | |
847 | ||
848 | static int hvs_probe(struct hv_device *hdev, | |
849 | const struct hv_vmbus_device_id *dev_id) | |
850 | { | |
851 | struct vmbus_channel *chan = hdev->channel; | |
852 | ||
853 | hvs_open_connection(chan); | |
854 | ||
855 | /* Always return success to suppress the unnecessary error message | |
856 | * in vmbus_probe(): on error the host will rescind the device in | |
857 | * 30 seconds and we can do cleanup at that time in | |
858 | * vmbus_onoffer_rescind(). | |
859 | */ | |
860 | return 0; | |
861 | } | |
862 | ||
863 | static int hvs_remove(struct hv_device *hdev) | |
864 | { | |
865 | struct vmbus_channel *chan = hdev->channel; | |
866 | ||
867 | vmbus_close(chan); | |
868 | ||
869 | return 0; | |
870 | } | |
871 | ||
872 | /* This isn't really used. See vmbus_match() and vmbus_probe() */ | |
873 | static const struct hv_vmbus_device_id id_table[] = { | |
874 | {}, | |
875 | }; | |
876 | ||
877 | static struct hv_driver hvs_drv = { | |
878 | .name = "hv_sock", | |
879 | .hvsock = true, | |
880 | .id_table = id_table, | |
881 | .probe = hvs_probe, | |
882 | .remove = hvs_remove, | |
883 | }; | |
884 | ||
885 | static int __init hvs_init(void) | |
886 | { | |
887 | int ret; | |
888 | ||
889 | if (vmbus_proto_version < VERSION_WIN10) | |
890 | return -ENODEV; | |
891 | ||
892 | ret = vmbus_driver_register(&hvs_drv); | |
893 | if (ret != 0) | |
894 | return ret; | |
895 | ||
896 | ret = vsock_core_init(&hvs_transport); | |
897 | if (ret) { | |
898 | vmbus_driver_unregister(&hvs_drv); | |
899 | return ret; | |
900 | } | |
901 | ||
902 | return 0; | |
903 | } | |
904 | ||
905 | static void __exit hvs_exit(void) | |
906 | { | |
907 | vsock_core_exit(); | |
908 | vmbus_driver_unregister(&hvs_drv); | |
909 | } | |
910 | ||
911 | module_init(hvs_init); | |
912 | module_exit(hvs_exit); | |
913 | ||
914 | MODULE_DESCRIPTION("Hyper-V Sockets"); | |
915 | MODULE_VERSION("1.0.0"); | |
916 | MODULE_LICENSE("GPL"); | |
917 | MODULE_ALIAS_NETPROTO(PF_VSOCK); |