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 17 /* Older (VMBUS version 'VERSION_WIN10' or before) Windows hosts have some 18 * stricter requirements on the hv_sock ring buffer size of six 4K pages. Newer 19 * hosts don't have this limitation; but, keep the defaults the same for compat. 20 */ 21 #define PAGE_SIZE_4K 4096 22 #define RINGBUFFER_HVS_RCV_SIZE (PAGE_SIZE_4K * 6) 23 #define RINGBUFFER_HVS_SND_SIZE (PAGE_SIZE_4K * 6) 24 #define RINGBUFFER_HVS_MAX_SIZE (PAGE_SIZE_4K * 64) 25 26 /* The MTU is 16KB per the host side's design */ 27 #define HVS_MTU_SIZE (1024 * 16) 28 29 /* How long to wait for graceful shutdown of a connection */ 30 #define HVS_CLOSE_TIMEOUT (8 * HZ) 31 32 struct vmpipe_proto_header { 33 u32 pkt_type; 34 u32 data_size; 35 }; 36 37 /* For recv, we use the VMBus in-place packet iterator APIs to directly copy 38 * data from the ringbuffer into the userspace buffer. 39 */ 40 struct hvs_recv_buf { 41 /* The header before the payload data */ 42 struct vmpipe_proto_header hdr; 43 44 /* The payload */ 45 u8 data[HVS_MTU_SIZE]; 46 }; 47 48 /* We can send up to HVS_MTU_SIZE bytes of payload to the host, but let's use 49 * a smaller size, i.e. HVS_SEND_BUF_SIZE, to maximize concurrency between the 50 * guest and the host processing as one VMBUS packet is the smallest processing 51 * unit. 52 * 53 * Note: the buffer can be eliminated in the future when we add new VMBus 54 * ringbuffer APIs that allow us to directly copy data from userspace buffer 55 * to VMBus ringbuffer. 56 */ 57 #define HVS_SEND_BUF_SIZE (PAGE_SIZE_4K - sizeof(struct vmpipe_proto_header)) 58 59 struct hvs_send_buf { 60 /* The header before the payload data */ 61 struct vmpipe_proto_header hdr; 62 63 /* The payload */ 64 u8 data[HVS_SEND_BUF_SIZE]; 65 }; 66 67 #define HVS_HEADER_LEN (sizeof(struct vmpacket_descriptor) + \ 68 sizeof(struct vmpipe_proto_header)) 69 70 /* See 'prev_indices' in hv_ringbuffer_read(), hv_ringbuffer_write(), and 71 * __hv_pkt_iter_next(). 72 */ 73 #define VMBUS_PKT_TRAILER_SIZE (sizeof(u64)) 74 75 #define HVS_PKT_LEN(payload_len) (HVS_HEADER_LEN + \ 76 ALIGN((payload_len), 8) + \ 77 VMBUS_PKT_TRAILER_SIZE) 78 79 union hvs_service_id { 80 uuid_le srv_id; 81 82 struct { 83 unsigned int svm_port; 84 unsigned char b[sizeof(uuid_le) - sizeof(unsigned int)]; 85 }; 86 }; 87 88 /* Per-socket state (accessed via vsk->trans) */ 89 struct hvsock { 90 struct vsock_sock *vsk; 91 92 uuid_le vm_srv_id; 93 uuid_le host_srv_id; 94 95 struct vmbus_channel *chan; 96 struct vmpacket_descriptor *recv_desc; 97 98 /* The length of the payload not delivered to userland yet */ 99 u32 recv_data_len; 100 /* The offset of the payload */ 101 u32 recv_data_off; 102 103 /* Have we sent the zero-length packet (FIN)? */ 104 bool fin_sent; 105 }; 106 107 /* In the VM, we support Hyper-V Sockets with AF_VSOCK, and the endpoint is 108 * <cid, port> (see struct sockaddr_vm). Note: cid is not really used here: 109 * when we write apps to connect to the host, we can only use VMADDR_CID_ANY 110 * or VMADDR_CID_HOST (both are equivalent) as the remote cid, and when we 111 * write apps to bind() & listen() in the VM, we can only use VMADDR_CID_ANY 112 * as the local cid. 113 * 114 * On the host, Hyper-V Sockets are supported by Winsock AF_HYPERV: 115 * https://docs.microsoft.com/en-us/virtualization/hyper-v-on-windows/user- 116 * guide/make-integration-service, and the endpoint is <VmID, ServiceId> with 117 * the below sockaddr: 118 * 119 * struct SOCKADDR_HV 120 * { 121 * ADDRESS_FAMILY Family; 122 * USHORT Reserved; 123 * GUID VmId; 124 * GUID ServiceId; 125 * }; 126 * Note: VmID is not used by Linux VM and actually it isn't transmitted via 127 * VMBus, because here it's obvious the host and the VM can easily identify 128 * each other. Though the VmID is useful on the host, especially in the case 129 * of Windows container, Linux VM doesn't need it at all. 130 * 131 * To make use of the AF_VSOCK infrastructure in Linux VM, we have to limit 132 * the available GUID space of SOCKADDR_HV so that we can create a mapping 133 * between AF_VSOCK port and SOCKADDR_HV Service GUID. The rule of writing 134 * Hyper-V Sockets apps on the host and in Linux VM is: 135 * 136 **************************************************************************** 137 * The only valid Service GUIDs, from the perspectives of both the host and * 138 * Linux VM, that can be connected by the other end, must conform to this * 139 * format: <port>-facb-11e6-bd58-64006a7986d3, and the "port" must be in * 140 * this range [0, 0x7FFFFFFF]. * 141 **************************************************************************** 142 * 143 * When we write apps on the host to connect(), the GUID ServiceID is used. 144 * When we write apps in Linux VM to connect(), we only need to specify the 145 * port and the driver will form the GUID and use that to request the host. 146 * 147 * From the perspective of Linux VM: 148 * 1. the local ephemeral port (i.e. the local auto-bound port when we call 149 * connect() without explicit bind()) is generated by __vsock_bind_stream(), 150 * and the range is [1024, 0xFFFFFFFF). 151 * 2. the remote ephemeral port (i.e. the auto-generated remote port for 152 * a connect request initiated by the host's connect()) is generated by 153 * hvs_remote_addr_init() and the range is [0x80000000, 0xFFFFFFFF). 154 */ 155 156 #define MAX_LISTEN_PORT ((u32)0x7FFFFFFF) 157 #define MAX_VM_LISTEN_PORT MAX_LISTEN_PORT 158 #define MAX_HOST_LISTEN_PORT MAX_LISTEN_PORT 159 #define MIN_HOST_EPHEMERAL_PORT (MAX_HOST_LISTEN_PORT + 1) 160 161 /* 00000000-facb-11e6-bd58-64006a7986d3 */ 162 static const uuid_le srv_id_template = 163 UUID_LE(0x00000000, 0xfacb, 0x11e6, 0xbd, 0x58, 164 0x64, 0x00, 0x6a, 0x79, 0x86, 0xd3); 165 166 static bool is_valid_srv_id(const uuid_le *id) 167 { 168 return !memcmp(&id->b[4], &srv_id_template.b[4], sizeof(uuid_le) - 4); 169 } 170 171 static unsigned int get_port_by_srv_id(const uuid_le *svr_id) 172 { 173 return *((unsigned int *)svr_id); 174 } 175 176 static void hvs_addr_init(struct sockaddr_vm *addr, const uuid_le *svr_id) 177 { 178 unsigned int port = get_port_by_srv_id(svr_id); 179 180 vsock_addr_init(addr, VMADDR_CID_ANY, port); 181 } 182 183 static void hvs_remote_addr_init(struct sockaddr_vm *remote, 184 struct sockaddr_vm *local) 185 { 186 static u32 host_ephemeral_port = MIN_HOST_EPHEMERAL_PORT; 187 struct sock *sk; 188 189 vsock_addr_init(remote, VMADDR_CID_ANY, VMADDR_PORT_ANY); 190 191 while (1) { 192 /* Wrap around ? */ 193 if (host_ephemeral_port < MIN_HOST_EPHEMERAL_PORT || 194 host_ephemeral_port == VMADDR_PORT_ANY) 195 host_ephemeral_port = MIN_HOST_EPHEMERAL_PORT; 196 197 remote->svm_port = host_ephemeral_port++; 198 199 sk = vsock_find_connected_socket(remote, local); 200 if (!sk) { 201 /* Found an available ephemeral port */ 202 return; 203 } 204 205 /* Release refcnt got in vsock_find_connected_socket */ 206 sock_put(sk); 207 } 208 } 209 210 static void hvs_set_channel_pending_send_size(struct vmbus_channel *chan) 211 { 212 set_channel_pending_send_size(chan, 213 HVS_PKT_LEN(HVS_SEND_BUF_SIZE)); 214 215 virt_mb(); 216 } 217 218 static bool hvs_channel_readable(struct vmbus_channel *chan) 219 { 220 u32 readable = hv_get_bytes_to_read(&chan->inbound); 221 222 /* 0-size payload means FIN */ 223 return readable >= HVS_PKT_LEN(0); 224 } 225 226 static int hvs_channel_readable_payload(struct vmbus_channel *chan) 227 { 228 u32 readable = hv_get_bytes_to_read(&chan->inbound); 229 230 if (readable > HVS_PKT_LEN(0)) { 231 /* At least we have 1 byte to read. We don't need to return 232 * the exact readable bytes: see vsock_stream_recvmsg() -> 233 * vsock_stream_has_data(). 234 */ 235 return 1; 236 } 237 238 if (readable == HVS_PKT_LEN(0)) { 239 /* 0-size payload means FIN */ 240 return 0; 241 } 242 243 /* No payload or FIN */ 244 return -1; 245 } 246 247 static size_t hvs_channel_writable_bytes(struct vmbus_channel *chan) 248 { 249 u32 writeable = hv_get_bytes_to_write(&chan->outbound); 250 size_t ret; 251 252 /* The ringbuffer mustn't be 100% full, and we should reserve a 253 * zero-length-payload packet for the FIN: see hv_ringbuffer_write() 254 * and hvs_shutdown(). 255 */ 256 if (writeable <= HVS_PKT_LEN(1) + HVS_PKT_LEN(0)) 257 return 0; 258 259 ret = writeable - HVS_PKT_LEN(1) - HVS_PKT_LEN(0); 260 261 return round_down(ret, 8); 262 } 263 264 static int hvs_send_data(struct vmbus_channel *chan, 265 struct hvs_send_buf *send_buf, size_t to_write) 266 { 267 send_buf->hdr.pkt_type = 1; 268 send_buf->hdr.data_size = to_write; 269 return vmbus_sendpacket(chan, &send_buf->hdr, 270 sizeof(send_buf->hdr) + to_write, 271 0, VM_PKT_DATA_INBAND, 0); 272 } 273 274 static void hvs_channel_cb(void *ctx) 275 { 276 struct sock *sk = (struct sock *)ctx; 277 struct vsock_sock *vsk = vsock_sk(sk); 278 struct hvsock *hvs = vsk->trans; 279 struct vmbus_channel *chan = hvs->chan; 280 281 if (hvs_channel_readable(chan)) 282 sk->sk_data_ready(sk); 283 284 if (hv_get_bytes_to_write(&chan->outbound) > 0) 285 sk->sk_write_space(sk); 286 } 287 288 static void hvs_do_close_lock_held(struct vsock_sock *vsk, 289 bool cancel_timeout) 290 { 291 struct sock *sk = sk_vsock(vsk); 292 293 sock_set_flag(sk, SOCK_DONE); 294 vsk->peer_shutdown = SHUTDOWN_MASK; 295 if (vsock_stream_has_data(vsk) <= 0) 296 sk->sk_state = TCP_CLOSING; 297 sk->sk_state_change(sk); 298 if (vsk->close_work_scheduled && 299 (!cancel_timeout || cancel_delayed_work(&vsk->close_work))) { 300 vsk->close_work_scheduled = false; 301 vsock_remove_sock(vsk); 302 303 /* Release the reference taken while scheduling the timeout */ 304 sock_put(sk); 305 } 306 } 307 308 static void hvs_close_connection(struct vmbus_channel *chan) 309 { 310 struct sock *sk = get_per_channel_state(chan); 311 312 lock_sock(sk); 313 hvs_do_close_lock_held(vsock_sk(sk), true); 314 release_sock(sk); 315 } 316 317 static void hvs_open_connection(struct vmbus_channel *chan) 318 { 319 uuid_le *if_instance, *if_type; 320 unsigned char conn_from_host; 321 322 struct sockaddr_vm addr; 323 struct sock *sk, *new = NULL; 324 struct vsock_sock *vnew = NULL; 325 struct hvsock *hvs = NULL; 326 struct hvsock *hvs_new = NULL; 327 int rcvbuf; 328 int ret; 329 int sndbuf; 330 331 if_type = &chan->offermsg.offer.if_type; 332 if_instance = &chan->offermsg.offer.if_instance; 333 conn_from_host = chan->offermsg.offer.u.pipe.user_def[0]; 334 335 /* The host or the VM should only listen on a port in 336 * [0, MAX_LISTEN_PORT] 337 */ 338 if (!is_valid_srv_id(if_type) || 339 get_port_by_srv_id(if_type) > MAX_LISTEN_PORT) 340 return; 341 342 hvs_addr_init(&addr, conn_from_host ? if_type : if_instance); 343 sk = vsock_find_bound_socket(&addr); 344 if (!sk) 345 return; 346 347 lock_sock(sk); 348 if ((conn_from_host && sk->sk_state != TCP_LISTEN) || 349 (!conn_from_host && sk->sk_state != TCP_SYN_SENT)) 350 goto out; 351 352 if (conn_from_host) { 353 if (sk->sk_ack_backlog >= sk->sk_max_ack_backlog) 354 goto out; 355 356 new = __vsock_create(sock_net(sk), NULL, sk, GFP_KERNEL, 357 sk->sk_type, 0); 358 if (!new) 359 goto out; 360 361 new->sk_state = TCP_SYN_SENT; 362 vnew = vsock_sk(new); 363 hvs_new = vnew->trans; 364 hvs_new->chan = chan; 365 } else { 366 hvs = vsock_sk(sk)->trans; 367 hvs->chan = chan; 368 } 369 370 set_channel_read_mode(chan, HV_CALL_DIRECT); 371 372 /* Use the socket buffer sizes as hints for the VMBUS ring size. For 373 * server side sockets, 'sk' is the parent socket and thus, this will 374 * allow the child sockets to inherit the size from the parent. Keep 375 * the mins to the default value and align to page size as per VMBUS 376 * requirements. 377 * For the max, the socket core library will limit the socket buffer 378 * size that can be set by the user, but, since currently, the hv_sock 379 * VMBUS ring buffer is physically contiguous allocation, restrict it 380 * further. 381 * Older versions of hv_sock host side code cannot handle bigger VMBUS 382 * ring buffer size. Use the version number to limit the change to newer 383 * versions. 384 */ 385 if (vmbus_proto_version < VERSION_WIN10_V5) { 386 sndbuf = RINGBUFFER_HVS_SND_SIZE; 387 rcvbuf = RINGBUFFER_HVS_RCV_SIZE; 388 } else { 389 sndbuf = max_t(int, sk->sk_sndbuf, RINGBUFFER_HVS_SND_SIZE); 390 sndbuf = min_t(int, sndbuf, RINGBUFFER_HVS_MAX_SIZE); 391 sndbuf = ALIGN(sndbuf, PAGE_SIZE); 392 rcvbuf = max_t(int, sk->sk_rcvbuf, RINGBUFFER_HVS_RCV_SIZE); 393 rcvbuf = min_t(int, rcvbuf, RINGBUFFER_HVS_MAX_SIZE); 394 rcvbuf = ALIGN(rcvbuf, PAGE_SIZE); 395 } 396 397 ret = vmbus_open(chan, sndbuf, rcvbuf, NULL, 0, hvs_channel_cb, 398 conn_from_host ? new : sk); 399 if (ret != 0) { 400 if (conn_from_host) { 401 hvs_new->chan = NULL; 402 sock_put(new); 403 } else { 404 hvs->chan = NULL; 405 } 406 goto out; 407 } 408 409 set_per_channel_state(chan, conn_from_host ? new : sk); 410 vmbus_set_chn_rescind_callback(chan, hvs_close_connection); 411 412 /* Set the pending send size to max packet size to always get 413 * notifications from the host when there is enough writable space. 414 * The host is optimized to send notifications only when the pending 415 * size boundary is crossed, and not always. 416 */ 417 hvs_set_channel_pending_send_size(chan); 418 419 if (conn_from_host) { 420 new->sk_state = TCP_ESTABLISHED; 421 sk->sk_ack_backlog++; 422 423 hvs_addr_init(&vnew->local_addr, if_type); 424 hvs_remote_addr_init(&vnew->remote_addr, &vnew->local_addr); 425 426 hvs_new->vm_srv_id = *if_type; 427 hvs_new->host_srv_id = *if_instance; 428 429 vsock_insert_connected(vnew); 430 431 vsock_enqueue_accept(sk, new); 432 } else { 433 sk->sk_state = TCP_ESTABLISHED; 434 sk->sk_socket->state = SS_CONNECTED; 435 436 vsock_insert_connected(vsock_sk(sk)); 437 } 438 439 sk->sk_state_change(sk); 440 441 out: 442 /* Release refcnt obtained when we called vsock_find_bound_socket() */ 443 sock_put(sk); 444 445 release_sock(sk); 446 } 447 448 static u32 hvs_get_local_cid(void) 449 { 450 return VMADDR_CID_ANY; 451 } 452 453 static int hvs_sock_init(struct vsock_sock *vsk, struct vsock_sock *psk) 454 { 455 struct hvsock *hvs; 456 struct sock *sk = sk_vsock(vsk); 457 458 hvs = kzalloc(sizeof(*hvs), GFP_KERNEL); 459 if (!hvs) 460 return -ENOMEM; 461 462 vsk->trans = hvs; 463 hvs->vsk = vsk; 464 sk->sk_sndbuf = RINGBUFFER_HVS_SND_SIZE; 465 sk->sk_rcvbuf = RINGBUFFER_HVS_RCV_SIZE; 466 return 0; 467 } 468 469 static int hvs_connect(struct vsock_sock *vsk) 470 { 471 union hvs_service_id vm, host; 472 struct hvsock *h = vsk->trans; 473 474 vm.srv_id = srv_id_template; 475 vm.svm_port = vsk->local_addr.svm_port; 476 h->vm_srv_id = vm.srv_id; 477 478 host.srv_id = srv_id_template; 479 host.svm_port = vsk->remote_addr.svm_port; 480 h->host_srv_id = host.srv_id; 481 482 return vmbus_send_tl_connect_request(&h->vm_srv_id, &h->host_srv_id); 483 } 484 485 static void hvs_shutdown_lock_held(struct hvsock *hvs, int mode) 486 { 487 struct vmpipe_proto_header hdr; 488 489 if (hvs->fin_sent || !hvs->chan) 490 return; 491 492 /* It can't fail: see hvs_channel_writable_bytes(). */ 493 (void)hvs_send_data(hvs->chan, (struct hvs_send_buf *)&hdr, 0); 494 hvs->fin_sent = true; 495 } 496 497 static int hvs_shutdown(struct vsock_sock *vsk, int mode) 498 { 499 struct sock *sk = sk_vsock(vsk); 500 501 if (!(mode & SEND_SHUTDOWN)) 502 return 0; 503 504 lock_sock(sk); 505 hvs_shutdown_lock_held(vsk->trans, mode); 506 release_sock(sk); 507 return 0; 508 } 509 510 static void hvs_close_timeout(struct work_struct *work) 511 { 512 struct vsock_sock *vsk = 513 container_of(work, struct vsock_sock, close_work.work); 514 struct sock *sk = sk_vsock(vsk); 515 516 sock_hold(sk); 517 lock_sock(sk); 518 if (!sock_flag(sk, SOCK_DONE)) 519 hvs_do_close_lock_held(vsk, false); 520 521 vsk->close_work_scheduled = false; 522 release_sock(sk); 523 sock_put(sk); 524 } 525 526 /* Returns true, if it is safe to remove socket; false otherwise */ 527 static bool hvs_close_lock_held(struct vsock_sock *vsk) 528 { 529 struct sock *sk = sk_vsock(vsk); 530 531 if (!(sk->sk_state == TCP_ESTABLISHED || 532 sk->sk_state == TCP_CLOSING)) 533 return true; 534 535 if ((sk->sk_shutdown & SHUTDOWN_MASK) != SHUTDOWN_MASK) 536 hvs_shutdown_lock_held(vsk->trans, SHUTDOWN_MASK); 537 538 if (sock_flag(sk, SOCK_DONE)) 539 return true; 540 541 /* This reference will be dropped by the delayed close routine */ 542 sock_hold(sk); 543 INIT_DELAYED_WORK(&vsk->close_work, hvs_close_timeout); 544 vsk->close_work_scheduled = true; 545 schedule_delayed_work(&vsk->close_work, HVS_CLOSE_TIMEOUT); 546 return false; 547 } 548 549 static void hvs_release(struct vsock_sock *vsk) 550 { 551 struct sock *sk = sk_vsock(vsk); 552 bool remove_sock; 553 554 lock_sock(sk); 555 remove_sock = hvs_close_lock_held(vsk); 556 release_sock(sk); 557 if (remove_sock) 558 vsock_remove_sock(vsk); 559 } 560 561 static void hvs_destruct(struct vsock_sock *vsk) 562 { 563 struct hvsock *hvs = vsk->trans; 564 struct vmbus_channel *chan = hvs->chan; 565 566 if (chan) 567 vmbus_hvsock_device_unregister(chan); 568 569 kfree(hvs); 570 } 571 572 static int hvs_dgram_bind(struct vsock_sock *vsk, struct sockaddr_vm *addr) 573 { 574 return -EOPNOTSUPP; 575 } 576 577 static int hvs_dgram_dequeue(struct vsock_sock *vsk, struct msghdr *msg, 578 size_t len, int flags) 579 { 580 return -EOPNOTSUPP; 581 } 582 583 static int hvs_dgram_enqueue(struct vsock_sock *vsk, 584 struct sockaddr_vm *remote, struct msghdr *msg, 585 size_t dgram_len) 586 { 587 return -EOPNOTSUPP; 588 } 589 590 static bool hvs_dgram_allow(u32 cid, u32 port) 591 { 592 return false; 593 } 594 595 static int hvs_update_recv_data(struct hvsock *hvs) 596 { 597 struct hvs_recv_buf *recv_buf; 598 u32 payload_len; 599 600 recv_buf = (struct hvs_recv_buf *)(hvs->recv_desc + 1); 601 payload_len = recv_buf->hdr.data_size; 602 603 if (payload_len > HVS_MTU_SIZE) 604 return -EIO; 605 606 if (payload_len == 0) 607 hvs->vsk->peer_shutdown |= SEND_SHUTDOWN; 608 609 hvs->recv_data_len = payload_len; 610 hvs->recv_data_off = 0; 611 612 return 0; 613 } 614 615 static ssize_t hvs_stream_dequeue(struct vsock_sock *vsk, struct msghdr *msg, 616 size_t len, int flags) 617 { 618 struct hvsock *hvs = vsk->trans; 619 bool need_refill = !hvs->recv_desc; 620 struct hvs_recv_buf *recv_buf; 621 u32 to_read; 622 int ret; 623 624 if (flags & MSG_PEEK) 625 return -EOPNOTSUPP; 626 627 if (need_refill) { 628 hvs->recv_desc = hv_pkt_iter_first(hvs->chan); 629 ret = hvs_update_recv_data(hvs); 630 if (ret) 631 return ret; 632 } 633 634 recv_buf = (struct hvs_recv_buf *)(hvs->recv_desc + 1); 635 to_read = min_t(u32, len, hvs->recv_data_len); 636 ret = memcpy_to_msg(msg, recv_buf->data + hvs->recv_data_off, to_read); 637 if (ret != 0) 638 return ret; 639 640 hvs->recv_data_len -= to_read; 641 if (hvs->recv_data_len == 0) { 642 hvs->recv_desc = hv_pkt_iter_next(hvs->chan, hvs->recv_desc); 643 if (hvs->recv_desc) { 644 ret = hvs_update_recv_data(hvs); 645 if (ret) 646 return ret; 647 } 648 } else { 649 hvs->recv_data_off += to_read; 650 } 651 652 return to_read; 653 } 654 655 static ssize_t hvs_stream_enqueue(struct vsock_sock *vsk, struct msghdr *msg, 656 size_t len) 657 { 658 struct hvsock *hvs = vsk->trans; 659 struct vmbus_channel *chan = hvs->chan; 660 struct hvs_send_buf *send_buf; 661 ssize_t to_write, max_writable; 662 ssize_t ret = 0; 663 ssize_t bytes_written = 0; 664 665 BUILD_BUG_ON(sizeof(*send_buf) != PAGE_SIZE_4K); 666 667 send_buf = kmalloc(sizeof(*send_buf), GFP_KERNEL); 668 if (!send_buf) 669 return -ENOMEM; 670 671 /* Reader(s) could be draining data from the channel as we write. 672 * Maximize bandwidth, by iterating until the channel is found to be 673 * full. 674 */ 675 while (len) { 676 max_writable = hvs_channel_writable_bytes(chan); 677 if (!max_writable) 678 break; 679 to_write = min_t(ssize_t, len, max_writable); 680 to_write = min_t(ssize_t, to_write, HVS_SEND_BUF_SIZE); 681 /* memcpy_from_msg is safe for loop as it advances the offsets 682 * within the message iterator. 683 */ 684 ret = memcpy_from_msg(send_buf->data, msg, to_write); 685 if (ret < 0) 686 goto out; 687 688 ret = hvs_send_data(hvs->chan, send_buf, to_write); 689 if (ret < 0) 690 goto out; 691 692 bytes_written += to_write; 693 len -= to_write; 694 } 695 out: 696 /* If any data has been sent, return that */ 697 if (bytes_written) 698 ret = bytes_written; 699 kfree(send_buf); 700 return ret; 701 } 702 703 static s64 hvs_stream_has_data(struct vsock_sock *vsk) 704 { 705 struct hvsock *hvs = vsk->trans; 706 s64 ret; 707 708 if (hvs->recv_data_len > 0) 709 return 1; 710 711 switch (hvs_channel_readable_payload(hvs->chan)) { 712 case 1: 713 ret = 1; 714 break; 715 case 0: 716 vsk->peer_shutdown |= SEND_SHUTDOWN; 717 ret = 0; 718 break; 719 default: /* -1 */ 720 ret = 0; 721 break; 722 } 723 724 return ret; 725 } 726 727 static s64 hvs_stream_has_space(struct vsock_sock *vsk) 728 { 729 struct hvsock *hvs = vsk->trans; 730 731 return hvs_channel_writable_bytes(hvs->chan); 732 } 733 734 static u64 hvs_stream_rcvhiwat(struct vsock_sock *vsk) 735 { 736 return HVS_MTU_SIZE + 1; 737 } 738 739 static bool hvs_stream_is_active(struct vsock_sock *vsk) 740 { 741 struct hvsock *hvs = vsk->trans; 742 743 return hvs->chan != NULL; 744 } 745 746 static bool hvs_stream_allow(u32 cid, u32 port) 747 { 748 /* The host's port range [MIN_HOST_EPHEMERAL_PORT, 0xFFFFFFFF) is 749 * reserved as ephemeral ports, which are used as the host's ports 750 * when the host initiates connections. 751 * 752 * Perform this check in the guest so an immediate error is produced 753 * instead of a timeout. 754 */ 755 if (port > MAX_HOST_LISTEN_PORT) 756 return false; 757 758 if (cid == VMADDR_CID_HOST) 759 return true; 760 761 return false; 762 } 763 764 static 765 int hvs_notify_poll_in(struct vsock_sock *vsk, size_t target, bool *readable) 766 { 767 struct hvsock *hvs = vsk->trans; 768 769 *readable = hvs_channel_readable(hvs->chan); 770 return 0; 771 } 772 773 static 774 int hvs_notify_poll_out(struct vsock_sock *vsk, size_t target, bool *writable) 775 { 776 *writable = hvs_stream_has_space(vsk) > 0; 777 778 return 0; 779 } 780 781 static 782 int hvs_notify_recv_init(struct vsock_sock *vsk, size_t target, 783 struct vsock_transport_recv_notify_data *d) 784 { 785 return 0; 786 } 787 788 static 789 int hvs_notify_recv_pre_block(struct vsock_sock *vsk, size_t target, 790 struct vsock_transport_recv_notify_data *d) 791 { 792 return 0; 793 } 794 795 static 796 int hvs_notify_recv_pre_dequeue(struct vsock_sock *vsk, size_t target, 797 struct vsock_transport_recv_notify_data *d) 798 { 799 return 0; 800 } 801 802 static 803 int hvs_notify_recv_post_dequeue(struct vsock_sock *vsk, size_t target, 804 ssize_t copied, bool data_read, 805 struct vsock_transport_recv_notify_data *d) 806 { 807 return 0; 808 } 809 810 static 811 int hvs_notify_send_init(struct vsock_sock *vsk, 812 struct vsock_transport_send_notify_data *d) 813 { 814 return 0; 815 } 816 817 static 818 int hvs_notify_send_pre_block(struct vsock_sock *vsk, 819 struct vsock_transport_send_notify_data *d) 820 { 821 return 0; 822 } 823 824 static 825 int hvs_notify_send_pre_enqueue(struct vsock_sock *vsk, 826 struct vsock_transport_send_notify_data *d) 827 { 828 return 0; 829 } 830 831 static 832 int hvs_notify_send_post_enqueue(struct vsock_sock *vsk, ssize_t written, 833 struct vsock_transport_send_notify_data *d) 834 { 835 return 0; 836 } 837 838 static void hvs_set_buffer_size(struct vsock_sock *vsk, u64 val) 839 { 840 /* Ignored. */ 841 } 842 843 static void hvs_set_min_buffer_size(struct vsock_sock *vsk, u64 val) 844 { 845 /* Ignored. */ 846 } 847 848 static void hvs_set_max_buffer_size(struct vsock_sock *vsk, u64 val) 849 { 850 /* Ignored. */ 851 } 852 853 static u64 hvs_get_buffer_size(struct vsock_sock *vsk) 854 { 855 return -ENOPROTOOPT; 856 } 857 858 static u64 hvs_get_min_buffer_size(struct vsock_sock *vsk) 859 { 860 return -ENOPROTOOPT; 861 } 862 863 static u64 hvs_get_max_buffer_size(struct vsock_sock *vsk) 864 { 865 return -ENOPROTOOPT; 866 } 867 868 static struct vsock_transport hvs_transport = { 869 .get_local_cid = hvs_get_local_cid, 870 871 .init = hvs_sock_init, 872 .destruct = hvs_destruct, 873 .release = hvs_release, 874 .connect = hvs_connect, 875 .shutdown = hvs_shutdown, 876 877 .dgram_bind = hvs_dgram_bind, 878 .dgram_dequeue = hvs_dgram_dequeue, 879 .dgram_enqueue = hvs_dgram_enqueue, 880 .dgram_allow = hvs_dgram_allow, 881 882 .stream_dequeue = hvs_stream_dequeue, 883 .stream_enqueue = hvs_stream_enqueue, 884 .stream_has_data = hvs_stream_has_data, 885 .stream_has_space = hvs_stream_has_space, 886 .stream_rcvhiwat = hvs_stream_rcvhiwat, 887 .stream_is_active = hvs_stream_is_active, 888 .stream_allow = hvs_stream_allow, 889 890 .notify_poll_in = hvs_notify_poll_in, 891 .notify_poll_out = hvs_notify_poll_out, 892 .notify_recv_init = hvs_notify_recv_init, 893 .notify_recv_pre_block = hvs_notify_recv_pre_block, 894 .notify_recv_pre_dequeue = hvs_notify_recv_pre_dequeue, 895 .notify_recv_post_dequeue = hvs_notify_recv_post_dequeue, 896 .notify_send_init = hvs_notify_send_init, 897 .notify_send_pre_block = hvs_notify_send_pre_block, 898 .notify_send_pre_enqueue = hvs_notify_send_pre_enqueue, 899 .notify_send_post_enqueue = hvs_notify_send_post_enqueue, 900 901 .set_buffer_size = hvs_set_buffer_size, 902 .set_min_buffer_size = hvs_set_min_buffer_size, 903 .set_max_buffer_size = hvs_set_max_buffer_size, 904 .get_buffer_size = hvs_get_buffer_size, 905 .get_min_buffer_size = hvs_get_min_buffer_size, 906 .get_max_buffer_size = hvs_get_max_buffer_size, 907 }; 908 909 static int hvs_probe(struct hv_device *hdev, 910 const struct hv_vmbus_device_id *dev_id) 911 { 912 struct vmbus_channel *chan = hdev->channel; 913 914 hvs_open_connection(chan); 915 916 /* Always return success to suppress the unnecessary error message 917 * in vmbus_probe(): on error the host will rescind the device in 918 * 30 seconds and we can do cleanup at that time in 919 * vmbus_onoffer_rescind(). 920 */ 921 return 0; 922 } 923 924 static int hvs_remove(struct hv_device *hdev) 925 { 926 struct vmbus_channel *chan = hdev->channel; 927 928 vmbus_close(chan); 929 930 return 0; 931 } 932 933 /* This isn't really used. See vmbus_match() and vmbus_probe() */ 934 static const struct hv_vmbus_device_id id_table[] = { 935 {}, 936 }; 937 938 static struct hv_driver hvs_drv = { 939 .name = "hv_sock", 940 .hvsock = true, 941 .id_table = id_table, 942 .probe = hvs_probe, 943 .remove = hvs_remove, 944 }; 945 946 static int __init hvs_init(void) 947 { 948 int ret; 949 950 if (vmbus_proto_version < VERSION_WIN10) 951 return -ENODEV; 952 953 ret = vmbus_driver_register(&hvs_drv); 954 if (ret != 0) 955 return ret; 956 957 ret = vsock_core_init(&hvs_transport); 958 if (ret) { 959 vmbus_driver_unregister(&hvs_drv); 960 return ret; 961 } 962 963 return 0; 964 } 965 966 static void __exit hvs_exit(void) 967 { 968 vsock_core_exit(); 969 vmbus_driver_unregister(&hvs_drv); 970 } 971 972 module_init(hvs_init); 973 module_exit(hvs_exit); 974 975 MODULE_DESCRIPTION("Hyper-V Sockets"); 976 MODULE_VERSION("1.0.0"); 977 MODULE_LICENSE("GPL"); 978 MODULE_ALIAS_NETPROTO(PF_VSOCK); 979