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 /* See hvs_stream_has_space(): we must make sure the host has seen 216 * the new pending send size, before we can re-check the writable 217 * bytes. 218 */ 219 virt_mb(); 220 } 221 222 static void hvs_clear_channel_pending_send_size(struct vmbus_channel *chan) 223 { 224 set_channel_pending_send_size(chan, 0); 225 226 /* Ditto */ 227 virt_mb(); 228 } 229 230 static bool hvs_channel_readable(struct vmbus_channel *chan) 231 { 232 u32 readable = hv_get_bytes_to_read(&chan->inbound); 233 234 /* 0-size payload means FIN */ 235 return readable >= HVS_PKT_LEN(0); 236 } 237 238 static int hvs_channel_readable_payload(struct vmbus_channel *chan) 239 { 240 u32 readable = hv_get_bytes_to_read(&chan->inbound); 241 242 if (readable > HVS_PKT_LEN(0)) { 243 /* At least we have 1 byte to read. We don't need to return 244 * the exact readable bytes: see vsock_stream_recvmsg() -> 245 * vsock_stream_has_data(). 246 */ 247 return 1; 248 } 249 250 if (readable == HVS_PKT_LEN(0)) { 251 /* 0-size payload means FIN */ 252 return 0; 253 } 254 255 /* No payload or FIN */ 256 return -1; 257 } 258 259 static size_t hvs_channel_writable_bytes(struct vmbus_channel *chan) 260 { 261 u32 writeable = hv_get_bytes_to_write(&chan->outbound); 262 size_t ret; 263 264 /* The ringbuffer mustn't be 100% full, and we should reserve a 265 * zero-length-payload packet for the FIN: see hv_ringbuffer_write() 266 * and hvs_shutdown(). 267 */ 268 if (writeable <= HVS_PKT_LEN(1) + HVS_PKT_LEN(0)) 269 return 0; 270 271 ret = writeable - HVS_PKT_LEN(1) - HVS_PKT_LEN(0); 272 273 return round_down(ret, 8); 274 } 275 276 static int hvs_send_data(struct vmbus_channel *chan, 277 struct hvs_send_buf *send_buf, size_t to_write) 278 { 279 send_buf->hdr.pkt_type = 1; 280 send_buf->hdr.data_size = to_write; 281 return vmbus_sendpacket(chan, &send_buf->hdr, 282 sizeof(send_buf->hdr) + to_write, 283 0, VM_PKT_DATA_INBAND, 0); 284 } 285 286 static void hvs_channel_cb(void *ctx) 287 { 288 struct sock *sk = (struct sock *)ctx; 289 struct vsock_sock *vsk = vsock_sk(sk); 290 struct hvsock *hvs = vsk->trans; 291 struct vmbus_channel *chan = hvs->chan; 292 293 if (hvs_channel_readable(chan)) 294 sk->sk_data_ready(sk); 295 296 /* See hvs_stream_has_space(): when we reach here, the writable bytes 297 * may be already less than HVS_PKT_LEN(HVS_SEND_BUF_SIZE). 298 */ 299 if (hv_get_bytes_to_write(&chan->outbound) > 0) 300 sk->sk_write_space(sk); 301 } 302 303 static void hvs_do_close_lock_held(struct vsock_sock *vsk, 304 bool cancel_timeout) 305 { 306 struct sock *sk = sk_vsock(vsk); 307 308 sock_set_flag(sk, SOCK_DONE); 309 vsk->peer_shutdown = SHUTDOWN_MASK; 310 if (vsock_stream_has_data(vsk) <= 0) 311 sk->sk_state = TCP_CLOSING; 312 sk->sk_state_change(sk); 313 if (vsk->close_work_scheduled && 314 (!cancel_timeout || cancel_delayed_work(&vsk->close_work))) { 315 vsk->close_work_scheduled = false; 316 vsock_remove_sock(vsk); 317 318 /* Release the reference taken while scheduling the timeout */ 319 sock_put(sk); 320 } 321 } 322 323 static void hvs_close_connection(struct vmbus_channel *chan) 324 { 325 struct sock *sk = get_per_channel_state(chan); 326 327 lock_sock(sk); 328 hvs_do_close_lock_held(vsock_sk(sk), true); 329 release_sock(sk); 330 } 331 332 static void hvs_open_connection(struct vmbus_channel *chan) 333 { 334 uuid_le *if_instance, *if_type; 335 unsigned char conn_from_host; 336 337 struct sockaddr_vm addr; 338 struct sock *sk, *new = NULL; 339 struct vsock_sock *vnew = NULL; 340 struct hvsock *hvs = NULL; 341 struct hvsock *hvs_new = NULL; 342 int rcvbuf; 343 int ret; 344 int sndbuf; 345 346 if_type = &chan->offermsg.offer.if_type; 347 if_instance = &chan->offermsg.offer.if_instance; 348 conn_from_host = chan->offermsg.offer.u.pipe.user_def[0]; 349 350 /* The host or the VM should only listen on a port in 351 * [0, MAX_LISTEN_PORT] 352 */ 353 if (!is_valid_srv_id(if_type) || 354 get_port_by_srv_id(if_type) > MAX_LISTEN_PORT) 355 return; 356 357 hvs_addr_init(&addr, conn_from_host ? if_type : if_instance); 358 sk = vsock_find_bound_socket(&addr); 359 if (!sk) 360 return; 361 362 lock_sock(sk); 363 if ((conn_from_host && sk->sk_state != TCP_LISTEN) || 364 (!conn_from_host && sk->sk_state != TCP_SYN_SENT)) 365 goto out; 366 367 if (conn_from_host) { 368 if (sk->sk_ack_backlog >= sk->sk_max_ack_backlog) 369 goto out; 370 371 new = __vsock_create(sock_net(sk), NULL, sk, GFP_KERNEL, 372 sk->sk_type, 0); 373 if (!new) 374 goto out; 375 376 new->sk_state = TCP_SYN_SENT; 377 vnew = vsock_sk(new); 378 hvs_new = vnew->trans; 379 hvs_new->chan = chan; 380 } else { 381 hvs = vsock_sk(sk)->trans; 382 hvs->chan = chan; 383 } 384 385 set_channel_read_mode(chan, HV_CALL_DIRECT); 386 387 /* Use the socket buffer sizes as hints for the VMBUS ring size. For 388 * server side sockets, 'sk' is the parent socket and thus, this will 389 * allow the child sockets to inherit the size from the parent. Keep 390 * the mins to the default value and align to page size as per VMBUS 391 * requirements. 392 * For the max, the socket core library will limit the socket buffer 393 * size that can be set by the user, but, since currently, the hv_sock 394 * VMBUS ring buffer is physically contiguous allocation, restrict it 395 * further. 396 * Older versions of hv_sock host side code cannot handle bigger VMBUS 397 * ring buffer size. Use the version number to limit the change to newer 398 * versions. 399 */ 400 if (vmbus_proto_version < VERSION_WIN10_V5) { 401 sndbuf = RINGBUFFER_HVS_SND_SIZE; 402 rcvbuf = RINGBUFFER_HVS_RCV_SIZE; 403 } else { 404 sndbuf = max_t(int, sk->sk_sndbuf, RINGBUFFER_HVS_SND_SIZE); 405 sndbuf = min_t(int, sndbuf, RINGBUFFER_HVS_MAX_SIZE); 406 sndbuf = ALIGN(sndbuf, PAGE_SIZE); 407 rcvbuf = max_t(int, sk->sk_rcvbuf, RINGBUFFER_HVS_RCV_SIZE); 408 rcvbuf = min_t(int, rcvbuf, RINGBUFFER_HVS_MAX_SIZE); 409 rcvbuf = ALIGN(rcvbuf, PAGE_SIZE); 410 } 411 412 ret = vmbus_open(chan, sndbuf, rcvbuf, NULL, 0, hvs_channel_cb, 413 conn_from_host ? new : sk); 414 if (ret != 0) { 415 if (conn_from_host) { 416 hvs_new->chan = NULL; 417 sock_put(new); 418 } else { 419 hvs->chan = NULL; 420 } 421 goto out; 422 } 423 424 set_per_channel_state(chan, conn_from_host ? new : sk); 425 vmbus_set_chn_rescind_callback(chan, hvs_close_connection); 426 427 if (conn_from_host) { 428 new->sk_state = TCP_ESTABLISHED; 429 sk->sk_ack_backlog++; 430 431 hvs_addr_init(&vnew->local_addr, if_type); 432 hvs_remote_addr_init(&vnew->remote_addr, &vnew->local_addr); 433 434 hvs_new->vm_srv_id = *if_type; 435 hvs_new->host_srv_id = *if_instance; 436 437 vsock_insert_connected(vnew); 438 439 vsock_enqueue_accept(sk, new); 440 } else { 441 sk->sk_state = TCP_ESTABLISHED; 442 sk->sk_socket->state = SS_CONNECTED; 443 444 vsock_insert_connected(vsock_sk(sk)); 445 } 446 447 sk->sk_state_change(sk); 448 449 out: 450 /* Release refcnt obtained when we called vsock_find_bound_socket() */ 451 sock_put(sk); 452 453 release_sock(sk); 454 } 455 456 static u32 hvs_get_local_cid(void) 457 { 458 return VMADDR_CID_ANY; 459 } 460 461 static int hvs_sock_init(struct vsock_sock *vsk, struct vsock_sock *psk) 462 { 463 struct hvsock *hvs; 464 struct sock *sk = sk_vsock(vsk); 465 466 hvs = kzalloc(sizeof(*hvs), GFP_KERNEL); 467 if (!hvs) 468 return -ENOMEM; 469 470 vsk->trans = hvs; 471 hvs->vsk = vsk; 472 sk->sk_sndbuf = RINGBUFFER_HVS_SND_SIZE; 473 sk->sk_rcvbuf = RINGBUFFER_HVS_RCV_SIZE; 474 return 0; 475 } 476 477 static int hvs_connect(struct vsock_sock *vsk) 478 { 479 union hvs_service_id vm, host; 480 struct hvsock *h = vsk->trans; 481 482 vm.srv_id = srv_id_template; 483 vm.svm_port = vsk->local_addr.svm_port; 484 h->vm_srv_id = vm.srv_id; 485 486 host.srv_id = srv_id_template; 487 host.svm_port = vsk->remote_addr.svm_port; 488 h->host_srv_id = host.srv_id; 489 490 return vmbus_send_tl_connect_request(&h->vm_srv_id, &h->host_srv_id); 491 } 492 493 static void hvs_shutdown_lock_held(struct hvsock *hvs, int mode) 494 { 495 struct vmpipe_proto_header hdr; 496 497 if (hvs->fin_sent || !hvs->chan) 498 return; 499 500 /* It can't fail: see hvs_channel_writable_bytes(). */ 501 (void)hvs_send_data(hvs->chan, (struct hvs_send_buf *)&hdr, 0); 502 hvs->fin_sent = true; 503 } 504 505 static int hvs_shutdown(struct vsock_sock *vsk, int mode) 506 { 507 struct sock *sk = sk_vsock(vsk); 508 509 if (!(mode & SEND_SHUTDOWN)) 510 return 0; 511 512 lock_sock(sk); 513 hvs_shutdown_lock_held(vsk->trans, mode); 514 release_sock(sk); 515 return 0; 516 } 517 518 static void hvs_close_timeout(struct work_struct *work) 519 { 520 struct vsock_sock *vsk = 521 container_of(work, struct vsock_sock, close_work.work); 522 struct sock *sk = sk_vsock(vsk); 523 524 sock_hold(sk); 525 lock_sock(sk); 526 if (!sock_flag(sk, SOCK_DONE)) 527 hvs_do_close_lock_held(vsk, false); 528 529 vsk->close_work_scheduled = false; 530 release_sock(sk); 531 sock_put(sk); 532 } 533 534 /* Returns true, if it is safe to remove socket; false otherwise */ 535 static bool hvs_close_lock_held(struct vsock_sock *vsk) 536 { 537 struct sock *sk = sk_vsock(vsk); 538 539 if (!(sk->sk_state == TCP_ESTABLISHED || 540 sk->sk_state == TCP_CLOSING)) 541 return true; 542 543 if ((sk->sk_shutdown & SHUTDOWN_MASK) != SHUTDOWN_MASK) 544 hvs_shutdown_lock_held(vsk->trans, SHUTDOWN_MASK); 545 546 if (sock_flag(sk, SOCK_DONE)) 547 return true; 548 549 /* This reference will be dropped by the delayed close routine */ 550 sock_hold(sk); 551 INIT_DELAYED_WORK(&vsk->close_work, hvs_close_timeout); 552 vsk->close_work_scheduled = true; 553 schedule_delayed_work(&vsk->close_work, HVS_CLOSE_TIMEOUT); 554 return false; 555 } 556 557 static void hvs_release(struct vsock_sock *vsk) 558 { 559 struct sock *sk = sk_vsock(vsk); 560 bool remove_sock; 561 562 lock_sock(sk); 563 remove_sock = hvs_close_lock_held(vsk); 564 release_sock(sk); 565 if (remove_sock) 566 vsock_remove_sock(vsk); 567 } 568 569 static void hvs_destruct(struct vsock_sock *vsk) 570 { 571 struct hvsock *hvs = vsk->trans; 572 struct vmbus_channel *chan = hvs->chan; 573 574 if (chan) 575 vmbus_hvsock_device_unregister(chan); 576 577 kfree(hvs); 578 } 579 580 static int hvs_dgram_bind(struct vsock_sock *vsk, struct sockaddr_vm *addr) 581 { 582 return -EOPNOTSUPP; 583 } 584 585 static int hvs_dgram_dequeue(struct vsock_sock *vsk, struct msghdr *msg, 586 size_t len, int flags) 587 { 588 return -EOPNOTSUPP; 589 } 590 591 static int hvs_dgram_enqueue(struct vsock_sock *vsk, 592 struct sockaddr_vm *remote, struct msghdr *msg, 593 size_t dgram_len) 594 { 595 return -EOPNOTSUPP; 596 } 597 598 static bool hvs_dgram_allow(u32 cid, u32 port) 599 { 600 return false; 601 } 602 603 static int hvs_update_recv_data(struct hvsock *hvs) 604 { 605 struct hvs_recv_buf *recv_buf; 606 u32 payload_len; 607 608 recv_buf = (struct hvs_recv_buf *)(hvs->recv_desc + 1); 609 payload_len = recv_buf->hdr.data_size; 610 611 if (payload_len > HVS_MTU_SIZE) 612 return -EIO; 613 614 if (payload_len == 0) 615 hvs->vsk->peer_shutdown |= SEND_SHUTDOWN; 616 617 hvs->recv_data_len = payload_len; 618 hvs->recv_data_off = 0; 619 620 return 0; 621 } 622 623 static ssize_t hvs_stream_dequeue(struct vsock_sock *vsk, struct msghdr *msg, 624 size_t len, int flags) 625 { 626 struct hvsock *hvs = vsk->trans; 627 bool need_refill = !hvs->recv_desc; 628 struct hvs_recv_buf *recv_buf; 629 u32 to_read; 630 int ret; 631 632 if (flags & MSG_PEEK) 633 return -EOPNOTSUPP; 634 635 if (need_refill) { 636 hvs->recv_desc = hv_pkt_iter_first(hvs->chan); 637 ret = hvs_update_recv_data(hvs); 638 if (ret) 639 return ret; 640 } 641 642 recv_buf = (struct hvs_recv_buf *)(hvs->recv_desc + 1); 643 to_read = min_t(u32, len, hvs->recv_data_len); 644 ret = memcpy_to_msg(msg, recv_buf->data + hvs->recv_data_off, to_read); 645 if (ret != 0) 646 return ret; 647 648 hvs->recv_data_len -= to_read; 649 if (hvs->recv_data_len == 0) { 650 hvs->recv_desc = hv_pkt_iter_next(hvs->chan, hvs->recv_desc); 651 if (hvs->recv_desc) { 652 ret = hvs_update_recv_data(hvs); 653 if (ret) 654 return ret; 655 } 656 } else { 657 hvs->recv_data_off += to_read; 658 } 659 660 return to_read; 661 } 662 663 static ssize_t hvs_stream_enqueue(struct vsock_sock *vsk, struct msghdr *msg, 664 size_t len) 665 { 666 struct hvsock *hvs = vsk->trans; 667 struct vmbus_channel *chan = hvs->chan; 668 struct hvs_send_buf *send_buf; 669 ssize_t to_write, max_writable; 670 ssize_t ret = 0; 671 ssize_t bytes_written = 0; 672 673 BUILD_BUG_ON(sizeof(*send_buf) != PAGE_SIZE_4K); 674 675 send_buf = kmalloc(sizeof(*send_buf), GFP_KERNEL); 676 if (!send_buf) 677 return -ENOMEM; 678 679 /* Reader(s) could be draining data from the channel as we write. 680 * Maximize bandwidth, by iterating until the channel is found to be 681 * full. 682 */ 683 while (len) { 684 max_writable = hvs_channel_writable_bytes(chan); 685 if (!max_writable) 686 break; 687 to_write = min_t(ssize_t, len, max_writable); 688 to_write = min_t(ssize_t, to_write, HVS_SEND_BUF_SIZE); 689 /* memcpy_from_msg is safe for loop as it advances the offsets 690 * within the message iterator. 691 */ 692 ret = memcpy_from_msg(send_buf->data, msg, to_write); 693 if (ret < 0) 694 goto out; 695 696 ret = hvs_send_data(hvs->chan, send_buf, to_write); 697 if (ret < 0) 698 goto out; 699 700 bytes_written += to_write; 701 len -= to_write; 702 } 703 out: 704 /* If any data has been sent, return that */ 705 if (bytes_written) 706 ret = bytes_written; 707 kfree(send_buf); 708 return ret; 709 } 710 711 static s64 hvs_stream_has_data(struct vsock_sock *vsk) 712 { 713 struct hvsock *hvs = vsk->trans; 714 s64 ret; 715 716 if (hvs->recv_data_len > 0) 717 return 1; 718 719 switch (hvs_channel_readable_payload(hvs->chan)) { 720 case 1: 721 ret = 1; 722 break; 723 case 0: 724 vsk->peer_shutdown |= SEND_SHUTDOWN; 725 ret = 0; 726 break; 727 default: /* -1 */ 728 ret = 0; 729 break; 730 } 731 732 return ret; 733 } 734 735 static s64 hvs_stream_has_space(struct vsock_sock *vsk) 736 { 737 struct hvsock *hvs = vsk->trans; 738 struct vmbus_channel *chan = hvs->chan; 739 s64 ret; 740 741 ret = hvs_channel_writable_bytes(chan); 742 if (ret > 0) { 743 hvs_clear_channel_pending_send_size(chan); 744 } else { 745 /* See hvs_channel_cb() */ 746 hvs_set_channel_pending_send_size(chan); 747 748 /* Re-check the writable bytes to avoid race */ 749 ret = hvs_channel_writable_bytes(chan); 750 if (ret > 0) 751 hvs_clear_channel_pending_send_size(chan); 752 } 753 754 return ret; 755 } 756 757 static u64 hvs_stream_rcvhiwat(struct vsock_sock *vsk) 758 { 759 return HVS_MTU_SIZE + 1; 760 } 761 762 static bool hvs_stream_is_active(struct vsock_sock *vsk) 763 { 764 struct hvsock *hvs = vsk->trans; 765 766 return hvs->chan != NULL; 767 } 768 769 static bool hvs_stream_allow(u32 cid, u32 port) 770 { 771 /* The host's port range [MIN_HOST_EPHEMERAL_PORT, 0xFFFFFFFF) is 772 * reserved as ephemeral ports, which are used as the host's ports 773 * when the host initiates connections. 774 * 775 * Perform this check in the guest so an immediate error is produced 776 * instead of a timeout. 777 */ 778 if (port > MAX_HOST_LISTEN_PORT) 779 return false; 780 781 if (cid == VMADDR_CID_HOST) 782 return true; 783 784 return false; 785 } 786 787 static 788 int hvs_notify_poll_in(struct vsock_sock *vsk, size_t target, bool *readable) 789 { 790 struct hvsock *hvs = vsk->trans; 791 792 *readable = hvs_channel_readable(hvs->chan); 793 return 0; 794 } 795 796 static 797 int hvs_notify_poll_out(struct vsock_sock *vsk, size_t target, bool *writable) 798 { 799 *writable = hvs_stream_has_space(vsk) > 0; 800 801 return 0; 802 } 803 804 static 805 int hvs_notify_recv_init(struct vsock_sock *vsk, size_t target, 806 struct vsock_transport_recv_notify_data *d) 807 { 808 return 0; 809 } 810 811 static 812 int hvs_notify_recv_pre_block(struct vsock_sock *vsk, size_t target, 813 struct vsock_transport_recv_notify_data *d) 814 { 815 return 0; 816 } 817 818 static 819 int hvs_notify_recv_pre_dequeue(struct vsock_sock *vsk, size_t target, 820 struct vsock_transport_recv_notify_data *d) 821 { 822 return 0; 823 } 824 825 static 826 int hvs_notify_recv_post_dequeue(struct vsock_sock *vsk, size_t target, 827 ssize_t copied, bool data_read, 828 struct vsock_transport_recv_notify_data *d) 829 { 830 return 0; 831 } 832 833 static 834 int hvs_notify_send_init(struct vsock_sock *vsk, 835 struct vsock_transport_send_notify_data *d) 836 { 837 return 0; 838 } 839 840 static 841 int hvs_notify_send_pre_block(struct vsock_sock *vsk, 842 struct vsock_transport_send_notify_data *d) 843 { 844 return 0; 845 } 846 847 static 848 int hvs_notify_send_pre_enqueue(struct vsock_sock *vsk, 849 struct vsock_transport_send_notify_data *d) 850 { 851 return 0; 852 } 853 854 static 855 int hvs_notify_send_post_enqueue(struct vsock_sock *vsk, ssize_t written, 856 struct vsock_transport_send_notify_data *d) 857 { 858 return 0; 859 } 860 861 static void hvs_set_buffer_size(struct vsock_sock *vsk, u64 val) 862 { 863 /* Ignored. */ 864 } 865 866 static void hvs_set_min_buffer_size(struct vsock_sock *vsk, u64 val) 867 { 868 /* Ignored. */ 869 } 870 871 static void hvs_set_max_buffer_size(struct vsock_sock *vsk, u64 val) 872 { 873 /* Ignored. */ 874 } 875 876 static u64 hvs_get_buffer_size(struct vsock_sock *vsk) 877 { 878 return -ENOPROTOOPT; 879 } 880 881 static u64 hvs_get_min_buffer_size(struct vsock_sock *vsk) 882 { 883 return -ENOPROTOOPT; 884 } 885 886 static u64 hvs_get_max_buffer_size(struct vsock_sock *vsk) 887 { 888 return -ENOPROTOOPT; 889 } 890 891 static struct vsock_transport hvs_transport = { 892 .get_local_cid = hvs_get_local_cid, 893 894 .init = hvs_sock_init, 895 .destruct = hvs_destruct, 896 .release = hvs_release, 897 .connect = hvs_connect, 898 .shutdown = hvs_shutdown, 899 900 .dgram_bind = hvs_dgram_bind, 901 .dgram_dequeue = hvs_dgram_dequeue, 902 .dgram_enqueue = hvs_dgram_enqueue, 903 .dgram_allow = hvs_dgram_allow, 904 905 .stream_dequeue = hvs_stream_dequeue, 906 .stream_enqueue = hvs_stream_enqueue, 907 .stream_has_data = hvs_stream_has_data, 908 .stream_has_space = hvs_stream_has_space, 909 .stream_rcvhiwat = hvs_stream_rcvhiwat, 910 .stream_is_active = hvs_stream_is_active, 911 .stream_allow = hvs_stream_allow, 912 913 .notify_poll_in = hvs_notify_poll_in, 914 .notify_poll_out = hvs_notify_poll_out, 915 .notify_recv_init = hvs_notify_recv_init, 916 .notify_recv_pre_block = hvs_notify_recv_pre_block, 917 .notify_recv_pre_dequeue = hvs_notify_recv_pre_dequeue, 918 .notify_recv_post_dequeue = hvs_notify_recv_post_dequeue, 919 .notify_send_init = hvs_notify_send_init, 920 .notify_send_pre_block = hvs_notify_send_pre_block, 921 .notify_send_pre_enqueue = hvs_notify_send_pre_enqueue, 922 .notify_send_post_enqueue = hvs_notify_send_post_enqueue, 923 924 .set_buffer_size = hvs_set_buffer_size, 925 .set_min_buffer_size = hvs_set_min_buffer_size, 926 .set_max_buffer_size = hvs_set_max_buffer_size, 927 .get_buffer_size = hvs_get_buffer_size, 928 .get_min_buffer_size = hvs_get_min_buffer_size, 929 .get_max_buffer_size = hvs_get_max_buffer_size, 930 }; 931 932 static int hvs_probe(struct hv_device *hdev, 933 const struct hv_vmbus_device_id *dev_id) 934 { 935 struct vmbus_channel *chan = hdev->channel; 936 937 hvs_open_connection(chan); 938 939 /* Always return success to suppress the unnecessary error message 940 * in vmbus_probe(): on error the host will rescind the device in 941 * 30 seconds and we can do cleanup at that time in 942 * vmbus_onoffer_rescind(). 943 */ 944 return 0; 945 } 946 947 static int hvs_remove(struct hv_device *hdev) 948 { 949 struct vmbus_channel *chan = hdev->channel; 950 951 vmbus_close(chan); 952 953 return 0; 954 } 955 956 /* This isn't really used. See vmbus_match() and vmbus_probe() */ 957 static const struct hv_vmbus_device_id id_table[] = { 958 {}, 959 }; 960 961 static struct hv_driver hvs_drv = { 962 .name = "hv_sock", 963 .hvsock = true, 964 .id_table = id_table, 965 .probe = hvs_probe, 966 .remove = hvs_remove, 967 }; 968 969 static int __init hvs_init(void) 970 { 971 int ret; 972 973 if (vmbus_proto_version < VERSION_WIN10) 974 return -ENODEV; 975 976 ret = vmbus_driver_register(&hvs_drv); 977 if (ret != 0) 978 return ret; 979 980 ret = vsock_core_init(&hvs_transport); 981 if (ret) { 982 vmbus_driver_unregister(&hvs_drv); 983 return ret; 984 } 985 986 return 0; 987 } 988 989 static void __exit hvs_exit(void) 990 { 991 vsock_core_exit(); 992 vmbus_driver_unregister(&hvs_drv); 993 } 994 995 module_init(hvs_init); 996 module_exit(hvs_exit); 997 998 MODULE_DESCRIPTION("Hyper-V Sockets"); 999 MODULE_VERSION("1.0.0"); 1000 MODULE_LICENSE("GPL"); 1001 MODULE_ALIAS_NETPROTO(PF_VSOCK); 1002