]>
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 | ||
5a044fee SM |
38 | /* How long to wait for graceful shutdown of a connection */ |
39 | #define HVS_CLOSE_TIMEOUT (8 * HZ) | |
40 | ||
ae0078fc DC |
41 | struct vmpipe_proto_header { |
42 | u32 pkt_type; | |
43 | u32 data_size; | |
44 | }; | |
45 | ||
46 | /* For recv, we use the VMBus in-place packet iterator APIs to directly copy | |
47 | * data from the ringbuffer into the userspace buffer. | |
48 | */ | |
49 | struct hvs_recv_buf { | |
50 | /* The header before the payload data */ | |
51 | struct vmpipe_proto_header hdr; | |
52 | ||
53 | /* The payload */ | |
54 | u8 data[HVS_MTU_SIZE]; | |
55 | }; | |
56 | ||
57 | /* We can send up to HVS_MTU_SIZE bytes of payload to the host, but let's use | |
58 | * a small size, i.e. HVS_SEND_BUF_SIZE, to minimize the dynamically-allocated | |
59 | * buffer, because tests show there is no significant performance difference. | |
60 | * | |
61 | * Note: the buffer can be eliminated in the future when we add new VMBus | |
62 | * ringbuffer APIs that allow us to directly copy data from userspace buffer | |
63 | * to VMBus ringbuffer. | |
64 | */ | |
65 | #define HVS_SEND_BUF_SIZE (PAGE_SIZE_4K - sizeof(struct vmpipe_proto_header)) | |
66 | ||
67 | struct hvs_send_buf { | |
68 | /* The header before the payload data */ | |
69 | struct vmpipe_proto_header hdr; | |
70 | ||
71 | /* The payload */ | |
72 | u8 data[HVS_SEND_BUF_SIZE]; | |
73 | }; | |
74 | ||
75 | #define HVS_HEADER_LEN (sizeof(struct vmpacket_descriptor) + \ | |
76 | sizeof(struct vmpipe_proto_header)) | |
77 | ||
78 | /* See 'prev_indices' in hv_ringbuffer_read(), hv_ringbuffer_write(), and | |
79 | * __hv_pkt_iter_next(). | |
80 | */ | |
81 | #define VMBUS_PKT_TRAILER_SIZE (sizeof(u64)) | |
82 | ||
83 | #define HVS_PKT_LEN(payload_len) (HVS_HEADER_LEN + \ | |
84 | ALIGN((payload_len), 8) + \ | |
85 | VMBUS_PKT_TRAILER_SIZE) | |
86 | ||
87 | union hvs_service_id { | |
88 | uuid_le srv_id; | |
89 | ||
90 | struct { | |
91 | unsigned int svm_port; | |
92 | unsigned char b[sizeof(uuid_le) - sizeof(unsigned int)]; | |
93 | }; | |
94 | }; | |
95 | ||
96 | /* Per-socket state (accessed via vsk->trans) */ | |
97 | struct hvsock { | |
98 | struct vsock_sock *vsk; | |
99 | ||
100 | uuid_le vm_srv_id; | |
101 | uuid_le host_srv_id; | |
102 | ||
103 | struct vmbus_channel *chan; | |
104 | struct vmpacket_descriptor *recv_desc; | |
105 | ||
106 | /* The length of the payload not delivered to userland yet */ | |
107 | u32 recv_data_len; | |
108 | /* The offset of the payload */ | |
109 | u32 recv_data_off; | |
110 | ||
111 | /* Have we sent the zero-length packet (FIN)? */ | |
112 | bool fin_sent; | |
113 | }; | |
114 | ||
115 | /* In the VM, we support Hyper-V Sockets with AF_VSOCK, and the endpoint is | |
116 | * <cid, port> (see struct sockaddr_vm). Note: cid is not really used here: | |
117 | * when we write apps to connect to the host, we can only use VMADDR_CID_ANY | |
118 | * or VMADDR_CID_HOST (both are equivalent) as the remote cid, and when we | |
119 | * write apps to bind() & listen() in the VM, we can only use VMADDR_CID_ANY | |
120 | * as the local cid. | |
121 | * | |
122 | * On the host, Hyper-V Sockets are supported by Winsock AF_HYPERV: | |
123 | * https://docs.microsoft.com/en-us/virtualization/hyper-v-on-windows/user- | |
124 | * guide/make-integration-service, and the endpoint is <VmID, ServiceId> with | |
125 | * the below sockaddr: | |
126 | * | |
127 | * struct SOCKADDR_HV | |
128 | * { | |
129 | * ADDRESS_FAMILY Family; | |
130 | * USHORT Reserved; | |
131 | * GUID VmId; | |
132 | * GUID ServiceId; | |
133 | * }; | |
134 | * Note: VmID is not used by Linux VM and actually it isn't transmitted via | |
135 | * VMBus, because here it's obvious the host and the VM can easily identify | |
136 | * each other. Though the VmID is useful on the host, especially in the case | |
137 | * of Windows container, Linux VM doesn't need it at all. | |
138 | * | |
139 | * To make use of the AF_VSOCK infrastructure in Linux VM, we have to limit | |
140 | * the available GUID space of SOCKADDR_HV so that we can create a mapping | |
141 | * between AF_VSOCK port and SOCKADDR_HV Service GUID. The rule of writing | |
142 | * Hyper-V Sockets apps on the host and in Linux VM is: | |
143 | * | |
144 | **************************************************************************** | |
145 | * The only valid Service GUIDs, from the perspectives of both the host and * | |
146 | * Linux VM, that can be connected by the other end, must conform to this * | |
b9331933 | 147 | * format: <port>-facb-11e6-bd58-64006a7986d3. * |
ae0078fc DC |
148 | **************************************************************************** |
149 | * | |
150 | * When we write apps on the host to connect(), the GUID ServiceID is used. | |
151 | * When we write apps in Linux VM to connect(), we only need to specify the | |
152 | * port and the driver will form the GUID and use that to request the host. | |
153 | * | |
ae0078fc DC |
154 | */ |
155 | ||
ae0078fc DC |
156 | /* 00000000-facb-11e6-bd58-64006a7986d3 */ |
157 | static const uuid_le srv_id_template = | |
158 | UUID_LE(0x00000000, 0xfacb, 0x11e6, 0xbd, 0x58, | |
159 | 0x64, 0x00, 0x6a, 0x79, 0x86, 0xd3); | |
160 | ||
161 | static bool is_valid_srv_id(const uuid_le *id) | |
162 | { | |
163 | return !memcmp(&id->b[4], &srv_id_template.b[4], sizeof(uuid_le) - 4); | |
164 | } | |
165 | ||
166 | static unsigned int get_port_by_srv_id(const uuid_le *svr_id) | |
167 | { | |
168 | return *((unsigned int *)svr_id); | |
169 | } | |
170 | ||
171 | static void hvs_addr_init(struct sockaddr_vm *addr, const uuid_le *svr_id) | |
172 | { | |
173 | unsigned int port = get_port_by_srv_id(svr_id); | |
174 | ||
175 | vsock_addr_init(addr, VMADDR_CID_ANY, port); | |
176 | } | |
177 | ||
ae0078fc DC |
178 | static void hvs_set_channel_pending_send_size(struct vmbus_channel *chan) |
179 | { | |
180 | set_channel_pending_send_size(chan, | |
181 | HVS_PKT_LEN(HVS_SEND_BUF_SIZE)); | |
182 | ||
ae0078fc DC |
183 | virt_mb(); |
184 | } | |
185 | ||
186 | static bool hvs_channel_readable(struct vmbus_channel *chan) | |
187 | { | |
188 | u32 readable = hv_get_bytes_to_read(&chan->inbound); | |
189 | ||
190 | /* 0-size payload means FIN */ | |
191 | return readable >= HVS_PKT_LEN(0); | |
192 | } | |
193 | ||
194 | static int hvs_channel_readable_payload(struct vmbus_channel *chan) | |
195 | { | |
196 | u32 readable = hv_get_bytes_to_read(&chan->inbound); | |
197 | ||
198 | if (readable > HVS_PKT_LEN(0)) { | |
199 | /* At least we have 1 byte to read. We don't need to return | |
200 | * the exact readable bytes: see vsock_stream_recvmsg() -> | |
201 | * vsock_stream_has_data(). | |
202 | */ | |
203 | return 1; | |
204 | } | |
205 | ||
206 | if (readable == HVS_PKT_LEN(0)) { | |
207 | /* 0-size payload means FIN */ | |
208 | return 0; | |
209 | } | |
210 | ||
211 | /* No payload or FIN */ | |
212 | return -1; | |
213 | } | |
214 | ||
215 | static size_t hvs_channel_writable_bytes(struct vmbus_channel *chan) | |
216 | { | |
217 | u32 writeable = hv_get_bytes_to_write(&chan->outbound); | |
218 | size_t ret; | |
219 | ||
220 | /* The ringbuffer mustn't be 100% full, and we should reserve a | |
221 | * zero-length-payload packet for the FIN: see hv_ringbuffer_write() | |
222 | * and hvs_shutdown(). | |
223 | */ | |
224 | if (writeable <= HVS_PKT_LEN(1) + HVS_PKT_LEN(0)) | |
225 | return 0; | |
226 | ||
227 | ret = writeable - HVS_PKT_LEN(1) - HVS_PKT_LEN(0); | |
228 | ||
229 | return round_down(ret, 8); | |
230 | } | |
231 | ||
232 | static int hvs_send_data(struct vmbus_channel *chan, | |
233 | struct hvs_send_buf *send_buf, size_t to_write) | |
234 | { | |
235 | send_buf->hdr.pkt_type = 1; | |
236 | send_buf->hdr.data_size = to_write; | |
237 | return vmbus_sendpacket(chan, &send_buf->hdr, | |
238 | sizeof(send_buf->hdr) + to_write, | |
239 | 0, VM_PKT_DATA_INBAND, 0); | |
240 | } | |
241 | ||
242 | static void hvs_channel_cb(void *ctx) | |
243 | { | |
244 | struct sock *sk = (struct sock *)ctx; | |
245 | struct vsock_sock *vsk = vsock_sk(sk); | |
246 | struct hvsock *hvs = vsk->trans; | |
247 | struct vmbus_channel *chan = hvs->chan; | |
248 | ||
249 | if (hvs_channel_readable(chan)) | |
250 | sk->sk_data_ready(sk); | |
251 | ||
ae0078fc DC |
252 | if (hv_get_bytes_to_write(&chan->outbound) > 0) |
253 | sk->sk_write_space(sk); | |
254 | } | |
255 | ||
5a044fee SM |
256 | static void hvs_do_close_lock_held(struct vsock_sock *vsk, |
257 | bool cancel_timeout) | |
ae0078fc | 258 | { |
5a044fee | 259 | struct sock *sk = sk_vsock(vsk); |
b4562ca7 | 260 | |
ae0078fc | 261 | sock_set_flag(sk, SOCK_DONE); |
5a044fee SM |
262 | vsk->peer_shutdown = SHUTDOWN_MASK; |
263 | if (vsock_stream_has_data(vsk) <= 0) | |
264 | sk->sk_state = TCP_CLOSING; | |
ae0078fc | 265 | sk->sk_state_change(sk); |
5a044fee SM |
266 | if (vsk->close_work_scheduled && |
267 | (!cancel_timeout || cancel_delayed_work(&vsk->close_work))) { | |
268 | vsk->close_work_scheduled = false; | |
269 | vsock_remove_sock(vsk); | |
270 | ||
271 | /* Release the reference taken while scheduling the timeout */ | |
272 | sock_put(sk); | |
273 | } | |
274 | } | |
275 | ||
276 | static void hvs_close_connection(struct vmbus_channel *chan) | |
277 | { | |
278 | struct sock *sk = get_per_channel_state(chan); | |
b4562ca7 | 279 | |
5a044fee SM |
280 | lock_sock(sk); |
281 | hvs_do_close_lock_held(vsock_sk(sk), true); | |
b4562ca7 | 282 | release_sock(sk); |
94836a8a DC |
283 | |
284 | /* Release the refcnt for the channel that's opened in | |
285 | * hvs_open_connection(). | |
286 | */ | |
287 | sock_put(sk); | |
ae0078fc DC |
288 | } |
289 | ||
290 | static void hvs_open_connection(struct vmbus_channel *chan) | |
291 | { | |
292 | uuid_le *if_instance, *if_type; | |
293 | unsigned char conn_from_host; | |
294 | ||
295 | struct sockaddr_vm addr; | |
296 | struct sock *sk, *new = NULL; | |
f48d5590 SM |
297 | struct vsock_sock *vnew = NULL; |
298 | struct hvsock *hvs = NULL; | |
299 | struct hvsock *hvs_new = NULL; | |
ae0078fc DC |
300 | int ret; |
301 | ||
302 | if_type = &chan->offermsg.offer.if_type; | |
303 | if_instance = &chan->offermsg.offer.if_instance; | |
304 | conn_from_host = chan->offermsg.offer.u.pipe.user_def[0]; | |
b9331933 | 305 | if (!is_valid_srv_id(if_type)) |
ae0078fc DC |
306 | return; |
307 | ||
308 | hvs_addr_init(&addr, conn_from_host ? if_type : if_instance); | |
309 | sk = vsock_find_bound_socket(&addr); | |
310 | if (!sk) | |
311 | return; | |
312 | ||
b4562ca7 | 313 | lock_sock(sk); |
3b4477d2 SH |
314 | if ((conn_from_host && sk->sk_state != TCP_LISTEN) || |
315 | (!conn_from_host && sk->sk_state != TCP_SYN_SENT)) | |
ae0078fc DC |
316 | goto out; |
317 | ||
318 | if (conn_from_host) { | |
319 | if (sk->sk_ack_backlog >= sk->sk_max_ack_backlog) | |
320 | goto out; | |
321 | ||
322 | new = __vsock_create(sock_net(sk), NULL, sk, GFP_KERNEL, | |
323 | sk->sk_type, 0); | |
324 | if (!new) | |
325 | goto out; | |
326 | ||
3b4477d2 | 327 | new->sk_state = TCP_SYN_SENT; |
ae0078fc | 328 | vnew = vsock_sk(new); |
b9331933 SM |
329 | |
330 | hvs_addr_init(&vnew->local_addr, if_type); | |
331 | ||
332 | /* Remote peer is always the host */ | |
333 | vsock_addr_init(&vnew->remote_addr, | |
334 | VMADDR_CID_HOST, VMADDR_PORT_ANY); | |
335 | vnew->remote_addr.svm_port = get_port_by_srv_id(if_instance); | |
ae0078fc DC |
336 | hvs_new = vnew->trans; |
337 | hvs_new->chan = chan; | |
338 | } else { | |
339 | hvs = vsock_sk(sk)->trans; | |
340 | hvs->chan = chan; | |
341 | } | |
342 | ||
343 | set_channel_read_mode(chan, HV_CALL_DIRECT); | |
344 | ret = vmbus_open(chan, RINGBUFFER_HVS_SND_SIZE, | |
345 | RINGBUFFER_HVS_RCV_SIZE, NULL, 0, | |
346 | hvs_channel_cb, conn_from_host ? new : sk); | |
347 | if (ret != 0) { | |
348 | if (conn_from_host) { | |
349 | hvs_new->chan = NULL; | |
350 | sock_put(new); | |
351 | } else { | |
352 | hvs->chan = NULL; | |
353 | } | |
354 | goto out; | |
355 | } | |
356 | ||
357 | set_per_channel_state(chan, conn_from_host ? new : sk); | |
94836a8a DC |
358 | |
359 | /* This reference will be dropped by hvs_close_connection(). */ | |
360 | sock_hold(conn_from_host ? new : sk); | |
ae0078fc DC |
361 | vmbus_set_chn_rescind_callback(chan, hvs_close_connection); |
362 | ||
f48d5590 SM |
363 | /* Set the pending send size to max packet size to always get |
364 | * notifications from the host when there is enough writable space. | |
365 | * The host is optimized to send notifications only when the pending | |
366 | * size boundary is crossed, and not always. | |
367 | */ | |
368 | hvs_set_channel_pending_send_size(chan); | |
369 | ||
ae0078fc | 370 | if (conn_from_host) { |
3b4477d2 | 371 | new->sk_state = TCP_ESTABLISHED; |
ae0078fc DC |
372 | sk->sk_ack_backlog++; |
373 | ||
374 | hvs_addr_init(&vnew->local_addr, if_type); | |
ae0078fc DC |
375 | hvs_new->vm_srv_id = *if_type; |
376 | hvs_new->host_srv_id = *if_instance; | |
377 | ||
378 | vsock_insert_connected(vnew); | |
379 | ||
ae0078fc | 380 | vsock_enqueue_accept(sk, new); |
ae0078fc | 381 | } else { |
3b4477d2 | 382 | sk->sk_state = TCP_ESTABLISHED; |
ae0078fc DC |
383 | sk->sk_socket->state = SS_CONNECTED; |
384 | ||
385 | vsock_insert_connected(vsock_sk(sk)); | |
386 | } | |
387 | ||
388 | sk->sk_state_change(sk); | |
389 | ||
390 | out: | |
391 | /* Release refcnt obtained when we called vsock_find_bound_socket() */ | |
392 | sock_put(sk); | |
b4562ca7 DC |
393 | |
394 | release_sock(sk); | |
ae0078fc DC |
395 | } |
396 | ||
397 | static u32 hvs_get_local_cid(void) | |
398 | { | |
399 | return VMADDR_CID_ANY; | |
400 | } | |
401 | ||
402 | static int hvs_sock_init(struct vsock_sock *vsk, struct vsock_sock *psk) | |
403 | { | |
404 | struct hvsock *hvs; | |
405 | ||
406 | hvs = kzalloc(sizeof(*hvs), GFP_KERNEL); | |
407 | if (!hvs) | |
408 | return -ENOMEM; | |
409 | ||
410 | vsk->trans = hvs; | |
411 | hvs->vsk = vsk; | |
412 | ||
413 | return 0; | |
414 | } | |
415 | ||
416 | static int hvs_connect(struct vsock_sock *vsk) | |
417 | { | |
418 | union hvs_service_id vm, host; | |
419 | struct hvsock *h = vsk->trans; | |
420 | ||
421 | vm.srv_id = srv_id_template; | |
422 | vm.svm_port = vsk->local_addr.svm_port; | |
423 | h->vm_srv_id = vm.srv_id; | |
424 | ||
425 | host.srv_id = srv_id_template; | |
426 | host.svm_port = vsk->remote_addr.svm_port; | |
427 | h->host_srv_id = host.srv_id; | |
428 | ||
429 | return vmbus_send_tl_connect_request(&h->vm_srv_id, &h->host_srv_id); | |
430 | } | |
431 | ||
5a044fee SM |
432 | static void hvs_shutdown_lock_held(struct hvsock *hvs, int mode) |
433 | { | |
434 | struct vmpipe_proto_header hdr; | |
435 | ||
436 | if (hvs->fin_sent || !hvs->chan) | |
437 | return; | |
438 | ||
439 | /* It can't fail: see hvs_channel_writable_bytes(). */ | |
440 | (void)hvs_send_data(hvs->chan, (struct hvs_send_buf *)&hdr, 0); | |
441 | hvs->fin_sent = true; | |
442 | } | |
443 | ||
ae0078fc DC |
444 | static int hvs_shutdown(struct vsock_sock *vsk, int mode) |
445 | { | |
446 | struct sock *sk = sk_vsock(vsk); | |
ae0078fc DC |
447 | |
448 | if (!(mode & SEND_SHUTDOWN)) | |
449 | return 0; | |
450 | ||
451 | lock_sock(sk); | |
5a044fee SM |
452 | hvs_shutdown_lock_held(vsk->trans, mode); |
453 | release_sock(sk); | |
454 | return 0; | |
455 | } | |
ae0078fc | 456 | |
5a044fee SM |
457 | static void hvs_close_timeout(struct work_struct *work) |
458 | { | |
459 | struct vsock_sock *vsk = | |
460 | container_of(work, struct vsock_sock, close_work.work); | |
461 | struct sock *sk = sk_vsock(vsk); | |
ae0078fc | 462 | |
5a044fee SM |
463 | sock_hold(sk); |
464 | lock_sock(sk); | |
465 | if (!sock_flag(sk, SOCK_DONE)) | |
466 | hvs_do_close_lock_held(vsk, false); | |
ae0078fc | 467 | |
5a044fee | 468 | vsk->close_work_scheduled = false; |
ae0078fc | 469 | release_sock(sk); |
5a044fee | 470 | sock_put(sk); |
ae0078fc DC |
471 | } |
472 | ||
5a044fee SM |
473 | /* Returns true, if it is safe to remove socket; false otherwise */ |
474 | static bool hvs_close_lock_held(struct vsock_sock *vsk) | |
ae0078fc | 475 | { |
b4562ca7 | 476 | struct sock *sk = sk_vsock(vsk); |
ae0078fc | 477 | |
5a044fee SM |
478 | if (!(sk->sk_state == TCP_ESTABLISHED || |
479 | sk->sk_state == TCP_CLOSING)) | |
480 | return true; | |
b4562ca7 | 481 | |
5a044fee SM |
482 | if ((sk->sk_shutdown & SHUTDOWN_MASK) != SHUTDOWN_MASK) |
483 | hvs_shutdown_lock_held(vsk->trans, SHUTDOWN_MASK); | |
b4562ca7 | 484 | |
5a044fee SM |
485 | if (sock_flag(sk, SOCK_DONE)) |
486 | return true; | |
ae0078fc | 487 | |
5a044fee SM |
488 | /* This reference will be dropped by the delayed close routine */ |
489 | sock_hold(sk); | |
490 | INIT_DELAYED_WORK(&vsk->close_work, hvs_close_timeout); | |
491 | vsk->close_work_scheduled = true; | |
492 | schedule_delayed_work(&vsk->close_work, HVS_CLOSE_TIMEOUT); | |
493 | return false; | |
494 | } | |
ae0078fc | 495 | |
5a044fee SM |
496 | static void hvs_release(struct vsock_sock *vsk) |
497 | { | |
498 | struct sock *sk = sk_vsock(vsk); | |
499 | bool remove_sock; | |
500 | ||
e28078fc | 501 | lock_sock_nested(sk, SINGLE_DEPTH_NESTING); |
5a044fee SM |
502 | remove_sock = hvs_close_lock_held(vsk); |
503 | release_sock(sk); | |
504 | if (remove_sock) | |
505 | vsock_remove_sock(vsk); | |
ae0078fc DC |
506 | } |
507 | ||
508 | static void hvs_destruct(struct vsock_sock *vsk) | |
509 | { | |
510 | struct hvsock *hvs = vsk->trans; | |
511 | struct vmbus_channel *chan = hvs->chan; | |
512 | ||
513 | if (chan) | |
514 | vmbus_hvsock_device_unregister(chan); | |
515 | ||
516 | kfree(hvs); | |
517 | } | |
518 | ||
519 | static int hvs_dgram_bind(struct vsock_sock *vsk, struct sockaddr_vm *addr) | |
520 | { | |
521 | return -EOPNOTSUPP; | |
522 | } | |
523 | ||
524 | static int hvs_dgram_dequeue(struct vsock_sock *vsk, struct msghdr *msg, | |
525 | size_t len, int flags) | |
526 | { | |
527 | return -EOPNOTSUPP; | |
528 | } | |
529 | ||
530 | static int hvs_dgram_enqueue(struct vsock_sock *vsk, | |
531 | struct sockaddr_vm *remote, struct msghdr *msg, | |
532 | size_t dgram_len) | |
533 | { | |
534 | return -EOPNOTSUPP; | |
535 | } | |
536 | ||
537 | static bool hvs_dgram_allow(u32 cid, u32 port) | |
538 | { | |
539 | return false; | |
540 | } | |
541 | ||
542 | static int hvs_update_recv_data(struct hvsock *hvs) | |
543 | { | |
544 | struct hvs_recv_buf *recv_buf; | |
545 | u32 payload_len; | |
546 | ||
547 | recv_buf = (struct hvs_recv_buf *)(hvs->recv_desc + 1); | |
548 | payload_len = recv_buf->hdr.data_size; | |
549 | ||
550 | if (payload_len > HVS_MTU_SIZE) | |
551 | return -EIO; | |
552 | ||
553 | if (payload_len == 0) | |
554 | hvs->vsk->peer_shutdown |= SEND_SHUTDOWN; | |
555 | ||
556 | hvs->recv_data_len = payload_len; | |
557 | hvs->recv_data_off = 0; | |
558 | ||
559 | return 0; | |
560 | } | |
561 | ||
562 | static ssize_t hvs_stream_dequeue(struct vsock_sock *vsk, struct msghdr *msg, | |
563 | size_t len, int flags) | |
564 | { | |
565 | struct hvsock *hvs = vsk->trans; | |
566 | bool need_refill = !hvs->recv_desc; | |
567 | struct hvs_recv_buf *recv_buf; | |
568 | u32 to_read; | |
569 | int ret; | |
570 | ||
571 | if (flags & MSG_PEEK) | |
572 | return -EOPNOTSUPP; | |
573 | ||
574 | if (need_refill) { | |
575 | hvs->recv_desc = hv_pkt_iter_first(hvs->chan); | |
576 | ret = hvs_update_recv_data(hvs); | |
577 | if (ret) | |
578 | return ret; | |
579 | } | |
580 | ||
581 | recv_buf = (struct hvs_recv_buf *)(hvs->recv_desc + 1); | |
582 | to_read = min_t(u32, len, hvs->recv_data_len); | |
583 | ret = memcpy_to_msg(msg, recv_buf->data + hvs->recv_data_off, to_read); | |
584 | if (ret != 0) | |
585 | return ret; | |
586 | ||
587 | hvs->recv_data_len -= to_read; | |
588 | if (hvs->recv_data_len == 0) { | |
589 | hvs->recv_desc = hv_pkt_iter_next(hvs->chan, hvs->recv_desc); | |
590 | if (hvs->recv_desc) { | |
591 | ret = hvs_update_recv_data(hvs); | |
592 | if (ret) | |
593 | return ret; | |
594 | } | |
595 | } else { | |
596 | hvs->recv_data_off += to_read; | |
597 | } | |
598 | ||
599 | return to_read; | |
600 | } | |
601 | ||
602 | static ssize_t hvs_stream_enqueue(struct vsock_sock *vsk, struct msghdr *msg, | |
603 | size_t len) | |
604 | { | |
605 | struct hvsock *hvs = vsk->trans; | |
606 | struct vmbus_channel *chan = hvs->chan; | |
607 | struct hvs_send_buf *send_buf; | |
608 | ssize_t to_write, max_writable, ret; | |
609 | ||
610 | BUILD_BUG_ON(sizeof(*send_buf) != PAGE_SIZE_4K); | |
611 | ||
612 | send_buf = kmalloc(sizeof(*send_buf), GFP_KERNEL); | |
613 | if (!send_buf) | |
614 | return -ENOMEM; | |
615 | ||
616 | max_writable = hvs_channel_writable_bytes(chan); | |
617 | to_write = min_t(ssize_t, len, max_writable); | |
618 | to_write = min_t(ssize_t, to_write, HVS_SEND_BUF_SIZE); | |
619 | ||
620 | ret = memcpy_from_msg(send_buf->data, msg, to_write); | |
621 | if (ret < 0) | |
622 | goto out; | |
623 | ||
624 | ret = hvs_send_data(hvs->chan, send_buf, to_write); | |
625 | if (ret < 0) | |
626 | goto out; | |
627 | ||
628 | ret = to_write; | |
629 | out: | |
630 | kfree(send_buf); | |
631 | return ret; | |
632 | } | |
633 | ||
634 | static s64 hvs_stream_has_data(struct vsock_sock *vsk) | |
635 | { | |
636 | struct hvsock *hvs = vsk->trans; | |
637 | s64 ret; | |
638 | ||
639 | if (hvs->recv_data_len > 0) | |
640 | return 1; | |
641 | ||
642 | switch (hvs_channel_readable_payload(hvs->chan)) { | |
643 | case 1: | |
644 | ret = 1; | |
645 | break; | |
646 | case 0: | |
647 | vsk->peer_shutdown |= SEND_SHUTDOWN; | |
648 | ret = 0; | |
649 | break; | |
650 | default: /* -1 */ | |
651 | ret = 0; | |
652 | break; | |
653 | } | |
654 | ||
655 | return ret; | |
656 | } | |
657 | ||
658 | static s64 hvs_stream_has_space(struct vsock_sock *vsk) | |
659 | { | |
660 | struct hvsock *hvs = vsk->trans; | |
ae0078fc | 661 | |
f48d5590 | 662 | return hvs_channel_writable_bytes(hvs->chan); |
ae0078fc DC |
663 | } |
664 | ||
665 | static u64 hvs_stream_rcvhiwat(struct vsock_sock *vsk) | |
666 | { | |
667 | return HVS_MTU_SIZE + 1; | |
668 | } | |
669 | ||
670 | static bool hvs_stream_is_active(struct vsock_sock *vsk) | |
671 | { | |
672 | struct hvsock *hvs = vsk->trans; | |
673 | ||
674 | return hvs->chan != NULL; | |
675 | } | |
676 | ||
677 | static bool hvs_stream_allow(u32 cid, u32 port) | |
678 | { | |
ae0078fc DC |
679 | if (cid == VMADDR_CID_HOST) |
680 | return true; | |
681 | ||
682 | return false; | |
683 | } | |
684 | ||
685 | static | |
686 | int hvs_notify_poll_in(struct vsock_sock *vsk, size_t target, bool *readable) | |
687 | { | |
688 | struct hvsock *hvs = vsk->trans; | |
689 | ||
690 | *readable = hvs_channel_readable(hvs->chan); | |
691 | return 0; | |
692 | } | |
693 | ||
694 | static | |
695 | int hvs_notify_poll_out(struct vsock_sock *vsk, size_t target, bool *writable) | |
696 | { | |
697 | *writable = hvs_stream_has_space(vsk) > 0; | |
698 | ||
699 | return 0; | |
700 | } | |
701 | ||
702 | static | |
703 | int hvs_notify_recv_init(struct vsock_sock *vsk, size_t target, | |
704 | struct vsock_transport_recv_notify_data *d) | |
705 | { | |
706 | return 0; | |
707 | } | |
708 | ||
709 | static | |
710 | int hvs_notify_recv_pre_block(struct vsock_sock *vsk, size_t target, | |
711 | struct vsock_transport_recv_notify_data *d) | |
712 | { | |
713 | return 0; | |
714 | } | |
715 | ||
716 | static | |
717 | int hvs_notify_recv_pre_dequeue(struct vsock_sock *vsk, size_t target, | |
718 | struct vsock_transport_recv_notify_data *d) | |
719 | { | |
720 | return 0; | |
721 | } | |
722 | ||
723 | static | |
724 | int hvs_notify_recv_post_dequeue(struct vsock_sock *vsk, size_t target, | |
725 | ssize_t copied, bool data_read, | |
726 | struct vsock_transport_recv_notify_data *d) | |
727 | { | |
728 | return 0; | |
729 | } | |
730 | ||
731 | static | |
732 | int hvs_notify_send_init(struct vsock_sock *vsk, | |
733 | struct vsock_transport_send_notify_data *d) | |
734 | { | |
735 | return 0; | |
736 | } | |
737 | ||
738 | static | |
739 | int hvs_notify_send_pre_block(struct vsock_sock *vsk, | |
740 | struct vsock_transport_send_notify_data *d) | |
741 | { | |
742 | return 0; | |
743 | } | |
744 | ||
745 | static | |
746 | int hvs_notify_send_pre_enqueue(struct vsock_sock *vsk, | |
747 | struct vsock_transport_send_notify_data *d) | |
748 | { | |
749 | return 0; | |
750 | } | |
751 | ||
752 | static | |
753 | int hvs_notify_send_post_enqueue(struct vsock_sock *vsk, ssize_t written, | |
754 | struct vsock_transport_send_notify_data *d) | |
755 | { | |
756 | return 0; | |
757 | } | |
758 | ||
759 | static void hvs_set_buffer_size(struct vsock_sock *vsk, u64 val) | |
760 | { | |
761 | /* Ignored. */ | |
762 | } | |
763 | ||
764 | static void hvs_set_min_buffer_size(struct vsock_sock *vsk, u64 val) | |
765 | { | |
766 | /* Ignored. */ | |
767 | } | |
768 | ||
769 | static void hvs_set_max_buffer_size(struct vsock_sock *vsk, u64 val) | |
770 | { | |
771 | /* Ignored. */ | |
772 | } | |
773 | ||
774 | static u64 hvs_get_buffer_size(struct vsock_sock *vsk) | |
775 | { | |
776 | return -ENOPROTOOPT; | |
777 | } | |
778 | ||
779 | static u64 hvs_get_min_buffer_size(struct vsock_sock *vsk) | |
780 | { | |
781 | return -ENOPROTOOPT; | |
782 | } | |
783 | ||
784 | static u64 hvs_get_max_buffer_size(struct vsock_sock *vsk) | |
785 | { | |
786 | return -ENOPROTOOPT; | |
787 | } | |
788 | ||
789 | static struct vsock_transport hvs_transport = { | |
790 | .get_local_cid = hvs_get_local_cid, | |
791 | ||
792 | .init = hvs_sock_init, | |
793 | .destruct = hvs_destruct, | |
794 | .release = hvs_release, | |
795 | .connect = hvs_connect, | |
796 | .shutdown = hvs_shutdown, | |
797 | ||
798 | .dgram_bind = hvs_dgram_bind, | |
799 | .dgram_dequeue = hvs_dgram_dequeue, | |
800 | .dgram_enqueue = hvs_dgram_enqueue, | |
801 | .dgram_allow = hvs_dgram_allow, | |
802 | ||
803 | .stream_dequeue = hvs_stream_dequeue, | |
804 | .stream_enqueue = hvs_stream_enqueue, | |
805 | .stream_has_data = hvs_stream_has_data, | |
806 | .stream_has_space = hvs_stream_has_space, | |
807 | .stream_rcvhiwat = hvs_stream_rcvhiwat, | |
808 | .stream_is_active = hvs_stream_is_active, | |
809 | .stream_allow = hvs_stream_allow, | |
810 | ||
811 | .notify_poll_in = hvs_notify_poll_in, | |
812 | .notify_poll_out = hvs_notify_poll_out, | |
813 | .notify_recv_init = hvs_notify_recv_init, | |
814 | .notify_recv_pre_block = hvs_notify_recv_pre_block, | |
815 | .notify_recv_pre_dequeue = hvs_notify_recv_pre_dequeue, | |
816 | .notify_recv_post_dequeue = hvs_notify_recv_post_dequeue, | |
817 | .notify_send_init = hvs_notify_send_init, | |
818 | .notify_send_pre_block = hvs_notify_send_pre_block, | |
819 | .notify_send_pre_enqueue = hvs_notify_send_pre_enqueue, | |
820 | .notify_send_post_enqueue = hvs_notify_send_post_enqueue, | |
821 | ||
822 | .set_buffer_size = hvs_set_buffer_size, | |
823 | .set_min_buffer_size = hvs_set_min_buffer_size, | |
824 | .set_max_buffer_size = hvs_set_max_buffer_size, | |
825 | .get_buffer_size = hvs_get_buffer_size, | |
826 | .get_min_buffer_size = hvs_get_min_buffer_size, | |
827 | .get_max_buffer_size = hvs_get_max_buffer_size, | |
828 | }; | |
829 | ||
830 | static int hvs_probe(struct hv_device *hdev, | |
831 | const struct hv_vmbus_device_id *dev_id) | |
832 | { | |
833 | struct vmbus_channel *chan = hdev->channel; | |
834 | ||
835 | hvs_open_connection(chan); | |
836 | ||
837 | /* Always return success to suppress the unnecessary error message | |
838 | * in vmbus_probe(): on error the host will rescind the device in | |
839 | * 30 seconds and we can do cleanup at that time in | |
840 | * vmbus_onoffer_rescind(). | |
841 | */ | |
842 | return 0; | |
843 | } | |
844 | ||
845 | static int hvs_remove(struct hv_device *hdev) | |
846 | { | |
847 | struct vmbus_channel *chan = hdev->channel; | |
848 | ||
849 | vmbus_close(chan); | |
850 | ||
851 | return 0; | |
852 | } | |
853 | ||
854 | /* This isn't really used. See vmbus_match() and vmbus_probe() */ | |
855 | static const struct hv_vmbus_device_id id_table[] = { | |
856 | {}, | |
857 | }; | |
858 | ||
859 | static struct hv_driver hvs_drv = { | |
860 | .name = "hv_sock", | |
861 | .hvsock = true, | |
862 | .id_table = id_table, | |
863 | .probe = hvs_probe, | |
864 | .remove = hvs_remove, | |
865 | }; | |
866 | ||
867 | static int __init hvs_init(void) | |
868 | { | |
869 | int ret; | |
870 | ||
871 | if (vmbus_proto_version < VERSION_WIN10) | |
872 | return -ENODEV; | |
873 | ||
874 | ret = vmbus_driver_register(&hvs_drv); | |
875 | if (ret != 0) | |
876 | return ret; | |
877 | ||
878 | ret = vsock_core_init(&hvs_transport); | |
879 | if (ret) { | |
880 | vmbus_driver_unregister(&hvs_drv); | |
881 | return ret; | |
882 | } | |
883 | ||
884 | return 0; | |
885 | } | |
886 | ||
887 | static void __exit hvs_exit(void) | |
888 | { | |
889 | vsock_core_exit(); | |
890 | vmbus_driver_unregister(&hvs_drv); | |
891 | } | |
892 | ||
893 | module_init(hvs_init); | |
894 | module_exit(hvs_exit); | |
895 | ||
896 | MODULE_DESCRIPTION("Hyper-V Sockets"); | |
897 | MODULE_VERSION("1.0.0"); | |
898 | MODULE_LICENSE("GPL"); | |
899 | MODULE_ALIAS_NETPROTO(PF_VSOCK); |