1 // SPDX-License-Identifier: GPL-2.0-only 2 /* 3 * Hyper-V transport for vsock 4 * 5 * Hyper-V Sockets supplies a byte-stream based communication mechanism 6 * between the host and the VM. This driver implements the necessary 7 * support in the VM by introducing the new vsock transport. 8 * 9 * Copyright (c) 2017, Microsoft Corporation. 10 */ 11 #include <linux/module.h> 12 #include <linux/vmalloc.h> 13 #include <linux/hyperv.h> 14 #include <net/sock.h> 15 #include <net/af_vsock.h> 16 #include <hyperv/hvhdk.h> 17 18 /* Older (VMBUS version 'VERSION_WIN10' or before) Windows hosts have some 19 * stricter requirements on the hv_sock ring buffer size of six 4K pages. 20 * HV_HYP_PAGE_SIZE is defined as 4K. Newer hosts don't have this limitation; 21 * but, keep the defaults the same for compat. 22 */ 23 #define RINGBUFFER_HVS_RCV_SIZE (HV_HYP_PAGE_SIZE * 6) 24 #define RINGBUFFER_HVS_SND_SIZE (HV_HYP_PAGE_SIZE * 6) 25 #define RINGBUFFER_HVS_MAX_SIZE (HV_HYP_PAGE_SIZE * 64) 26 27 /* The MTU is 16KB per the host side's design */ 28 #define HVS_MTU_SIZE (1024 * 16) 29 30 /* How long to wait for graceful shutdown of a connection */ 31 #define HVS_CLOSE_TIMEOUT (8 * HZ) 32 33 struct vmpipe_proto_header { 34 u32 pkt_type; 35 u32 data_size; 36 }; 37 38 /* For recv, we use the VMBus in-place packet iterator APIs to directly copy 39 * data from the ringbuffer into the userspace buffer. 40 */ 41 struct hvs_recv_buf { 42 /* The header before the payload data */ 43 struct vmpipe_proto_header hdr; 44 45 /* The payload */ 46 u8 data[HVS_MTU_SIZE]; 47 }; 48 49 /* We can send up to HVS_MTU_SIZE bytes of payload to the host, but let's use 50 * a smaller size, i.e. HVS_SEND_BUF_SIZE, to maximize concurrency between the 51 * guest and the host processing as one VMBUS packet is the smallest processing 52 * unit. 53 * 54 * Note: the buffer can be eliminated in the future when we add new VMBus 55 * ringbuffer APIs that allow us to directly copy data from userspace buffer 56 * to VMBus ringbuffer. 57 */ 58 #define HVS_SEND_BUF_SIZE \ 59 (HV_HYP_PAGE_SIZE - sizeof(struct vmpipe_proto_header)) 60 61 struct hvs_send_buf { 62 /* The header before the payload data */ 63 struct vmpipe_proto_header hdr; 64 65 /* The payload */ 66 u8 data[HVS_SEND_BUF_SIZE]; 67 }; 68 69 #define HVS_HEADER_LEN (sizeof(struct vmpacket_descriptor) + \ 70 sizeof(struct vmpipe_proto_header)) 71 72 /* See 'prev_indices' in hv_ringbuffer_read(), hv_ringbuffer_write(), and 73 * __hv_pkt_iter_next(). 74 */ 75 #define VMBUS_PKT_TRAILER_SIZE (sizeof(u64)) 76 77 #define HVS_PKT_LEN(payload_len) (HVS_HEADER_LEN + \ 78 ALIGN((payload_len), 8) + \ 79 VMBUS_PKT_TRAILER_SIZE) 80 81 /* Upper bound on the size of a VMbus packet for hv_sock */ 82 #define HVS_MAX_PKT_SIZE HVS_PKT_LEN(HVS_MTU_SIZE) 83 84 union hvs_service_id { 85 guid_t srv_id; 86 87 struct { 88 unsigned int svm_port; 89 unsigned char b[sizeof(guid_t) - sizeof(unsigned int)]; 90 }; 91 }; 92 93 /* Per-socket state (accessed via vsk->trans) */ 94 struct hvsock { 95 struct vsock_sock *vsk; 96 97 guid_t vm_srv_id; 98 guid_t host_srv_id; 99 100 struct vmbus_channel *chan; 101 struct vmpacket_descriptor *recv_desc; 102 103 /* The length of the payload not delivered to userland yet */ 104 u32 recv_data_len; 105 /* The offset of the payload */ 106 u32 recv_data_off; 107 108 /* Have we sent the zero-length packet (FIN)? */ 109 bool fin_sent; 110 }; 111 112 /* In the VM, we support Hyper-V Sockets with AF_VSOCK, and the endpoint is 113 * <cid, port> (see struct sockaddr_vm). Note: cid is not really used here: 114 * when we write apps to connect to the host, we can only use VMADDR_CID_ANY 115 * or VMADDR_CID_HOST (both are equivalent) as the remote cid, and when we 116 * write apps to bind() & listen() in the VM, we can only use VMADDR_CID_ANY 117 * as the local cid. 118 * 119 * On the host, Hyper-V Sockets are supported by Winsock AF_HYPERV: 120 * https://docs.microsoft.com/en-us/virtualization/hyper-v-on-windows/user- 121 * guide/make-integration-service, and the endpoint is <VmID, ServiceId> with 122 * the below sockaddr: 123 * 124 * struct SOCKADDR_HV 125 * { 126 * ADDRESS_FAMILY Family; 127 * USHORT Reserved; 128 * GUID VmId; 129 * GUID ServiceId; 130 * }; 131 * Note: VmID is not used by Linux VM and actually it isn't transmitted via 132 * VMBus, because here it's obvious the host and the VM can easily identify 133 * each other. Though the VmID is useful on the host, especially in the case 134 * of Windows container, Linux VM doesn't need it at all. 135 * 136 * To make use of the AF_VSOCK infrastructure in Linux VM, we have to limit 137 * the available GUID space of SOCKADDR_HV so that we can create a mapping 138 * between AF_VSOCK port and SOCKADDR_HV Service GUID. The rule of writing 139 * Hyper-V Sockets apps on the host and in Linux VM is: 140 * 141 **************************************************************************** 142 * The only valid Service GUIDs, from the perspectives of both the host and * 143 * Linux VM, that can be connected by the other end, must conform to this * 144 * format: <port>-facb-11e6-bd58-64006a7986d3. * 145 **************************************************************************** 146 * 147 * When we write apps on the host to connect(), the GUID ServiceID is used. 148 * When we write apps in Linux VM to connect(), we only need to specify the 149 * port and the driver will form the GUID and use that to request the host. 150 * 151 */ 152 153 /* 00000000-facb-11e6-bd58-64006a7986d3 */ 154 static const guid_t srv_id_template = 155 GUID_INIT(0x00000000, 0xfacb, 0x11e6, 0xbd, 0x58, 156 0x64, 0x00, 0x6a, 0x79, 0x86, 0xd3); 157 158 static bool hvs_check_transport(struct vsock_sock *vsk); 159 160 static bool is_valid_srv_id(const guid_t *id) 161 { 162 return !memcmp(&id->b[4], &srv_id_template.b[4], sizeof(guid_t) - 4); 163 } 164 165 static unsigned int get_port_by_srv_id(const guid_t *svr_id) 166 { 167 return *((unsigned int *)svr_id); 168 } 169 170 static void hvs_addr_init(struct sockaddr_vm *addr, const guid_t *svr_id) 171 { 172 unsigned int port = get_port_by_srv_id(svr_id); 173 174 vsock_addr_init(addr, VMADDR_CID_ANY, port); 175 } 176 177 static void hvs_set_channel_pending_send_size(struct vmbus_channel *chan) 178 { 179 set_channel_pending_send_size(chan, 180 HVS_PKT_LEN(HVS_SEND_BUF_SIZE)); 181 182 virt_mb(); 183 } 184 185 static bool hvs_channel_readable(struct vmbus_channel *chan) 186 { 187 u32 readable = hv_get_bytes_to_read(&chan->inbound); 188 189 /* 0-size payload means FIN */ 190 return readable >= HVS_PKT_LEN(0); 191 } 192 193 static int hvs_channel_readable_payload(struct vmbus_channel *chan) 194 { 195 u32 readable = hv_get_bytes_to_read(&chan->inbound); 196 197 if (readable > HVS_PKT_LEN(0)) { 198 /* At least we have 1 byte to read. We don't need to return 199 * the exact readable bytes: see vsock_connectible_recvmsg() -> 200 * vsock_stream_has_data(). 201 */ 202 return 1; 203 } 204 205 if (readable == HVS_PKT_LEN(0)) { 206 /* 0-size payload means FIN */ 207 return 0; 208 } 209 210 /* No payload or FIN */ 211 return -1; 212 } 213 214 static size_t hvs_channel_writable_bytes(struct vmbus_channel *chan) 215 { 216 u32 writeable = hv_get_bytes_to_write(&chan->outbound); 217 size_t ret; 218 219 /* The ringbuffer mustn't be 100% full, and we should reserve a 220 * zero-length-payload packet for the FIN: see hv_ringbuffer_write() 221 * and hvs_shutdown(). 222 */ 223 if (writeable <= HVS_PKT_LEN(1) + HVS_PKT_LEN(0)) 224 return 0; 225 226 ret = writeable - HVS_PKT_LEN(1) - HVS_PKT_LEN(0); 227 228 return round_down(ret, 8); 229 } 230 231 static int __hvs_send_data(struct vmbus_channel *chan, 232 struct vmpipe_proto_header *hdr, 233 size_t to_write) 234 { 235 hdr->pkt_type = 1; 236 hdr->data_size = to_write; 237 return vmbus_sendpacket(chan, hdr, sizeof(*hdr) + to_write, 238 0, VM_PKT_DATA_INBAND, 0); 239 } 240 241 static int hvs_send_data(struct vmbus_channel *chan, 242 struct hvs_send_buf *send_buf, size_t to_write) 243 { 244 return __hvs_send_data(chan, &send_buf->hdr, to_write); 245 } 246 247 static void hvs_channel_cb(void *ctx) 248 { 249 struct sock *sk = (struct sock *)ctx; 250 struct vsock_sock *vsk = vsock_sk(sk); 251 struct hvsock *hvs = vsk->trans; 252 struct vmbus_channel *chan = hvs->chan; 253 254 if (hvs_channel_readable(chan)) 255 sk->sk_data_ready(sk); 256 257 if (hv_get_bytes_to_write(&chan->outbound) > 0) 258 sk->sk_write_space(sk); 259 } 260 261 static void hvs_do_close_lock_held(struct vsock_sock *vsk, 262 bool cancel_timeout) 263 { 264 struct sock *sk = sk_vsock(vsk); 265 266 sock_set_flag(sk, SOCK_DONE); 267 WRITE_ONCE(vsk->peer_shutdown, SHUTDOWN_MASK); 268 if (vsock_stream_has_data(vsk) <= 0) 269 sk->sk_state = TCP_CLOSING; 270 sk->sk_state_change(sk); 271 if (vsk->close_work_scheduled && 272 (!cancel_timeout || cancel_delayed_work(&vsk->close_work))) { 273 vsk->close_work_scheduled = false; 274 vsock_remove_sock(vsk); 275 276 /* Release the reference taken while scheduling the timeout */ 277 sock_put(sk); 278 } 279 } 280 281 static void hvs_close_connection(struct vmbus_channel *chan) 282 { 283 struct sock *sk = get_per_channel_state(chan); 284 285 lock_sock(sk); 286 hvs_do_close_lock_held(vsock_sk(sk), true); 287 release_sock(sk); 288 289 /* Release the refcnt for the channel that's opened in 290 * hvs_open_connection(). 291 */ 292 sock_put(sk); 293 } 294 295 static void hvs_open_connection(struct vmbus_channel *chan) 296 { 297 guid_t *if_instance, *if_type; 298 unsigned char conn_from_host; 299 300 struct sockaddr_vm addr; 301 struct sock *sk, *new = NULL; 302 struct vsock_sock *vnew = NULL; 303 struct hvsock *hvs = NULL; 304 struct hvsock *hvs_new = NULL; 305 int rcvbuf; 306 int ret; 307 int sndbuf; 308 309 if_type = &chan->offermsg.offer.if_type; 310 if_instance = &chan->offermsg.offer.if_instance; 311 conn_from_host = chan->offermsg.offer.u.pipe.user_def[0]; 312 if (!is_valid_srv_id(if_type)) 313 return; 314 315 hvs_addr_init(&addr, conn_from_host ? if_type : if_instance); 316 sk = vsock_find_bound_socket(&addr); 317 if (!sk) 318 return; 319 320 lock_sock(sk); 321 if ((conn_from_host && sk->sk_state != TCP_LISTEN) || 322 (!conn_from_host && sk->sk_state != TCP_SYN_SENT)) 323 goto out; 324 325 if (conn_from_host) { 326 if (sk->sk_ack_backlog >= sk->sk_max_ack_backlog) 327 goto out; 328 329 new = vsock_create_connected(sk); 330 if (!new) 331 goto out; 332 333 new->sk_state = TCP_SYN_SENT; 334 vnew = vsock_sk(new); 335 336 hvs_addr_init(&vnew->local_addr, if_type); 337 338 /* Remote peer is always the host */ 339 vsock_addr_init(&vnew->remote_addr, 340 VMADDR_CID_HOST, VMADDR_PORT_ANY); 341 vnew->remote_addr.svm_port = get_port_by_srv_id(if_instance); 342 ret = vsock_assign_transport(vnew, vsock_sk(sk)); 343 /* Transport assigned (looking at remote_addr) must be the 344 * same where we received the request. 345 */ 346 if (ret || !hvs_check_transport(vnew)) { 347 sock_put(new); 348 goto out; 349 } 350 hvs_new = vnew->trans; 351 hvs_new->chan = chan; 352 } else { 353 hvs = vsock_sk(sk)->trans; 354 hvs->chan = chan; 355 } 356 357 set_channel_read_mode(chan, HV_CALL_DIRECT); 358 359 /* Use the socket buffer sizes as hints for the VMBUS ring size. For 360 * server side sockets, 'sk' is the parent socket and thus, this will 361 * allow the child sockets to inherit the size from the parent. Keep 362 * the mins to the default value and align to page size as per VMBUS 363 * requirements. 364 * For the max, the socket core library will limit the socket buffer 365 * size that can be set by the user, but, since currently, the hv_sock 366 * VMBUS ring buffer is physically contiguous allocation, restrict it 367 * further. 368 * Older versions of hv_sock host side code cannot handle bigger VMBUS 369 * ring buffer size. Use the version number to limit the change to newer 370 * versions. 371 */ 372 if (vmbus_proto_version < VERSION_WIN10_V5) { 373 sndbuf = RINGBUFFER_HVS_SND_SIZE; 374 rcvbuf = RINGBUFFER_HVS_RCV_SIZE; 375 } else { 376 sndbuf = max_t(int, sk->sk_sndbuf, RINGBUFFER_HVS_SND_SIZE); 377 sndbuf = min_t(int, sndbuf, RINGBUFFER_HVS_MAX_SIZE); 378 sndbuf = VMBUS_RING_SIZE(sndbuf); 379 rcvbuf = max_t(int, sk->sk_rcvbuf, RINGBUFFER_HVS_RCV_SIZE); 380 rcvbuf = min_t(int, rcvbuf, RINGBUFFER_HVS_MAX_SIZE); 381 rcvbuf = VMBUS_RING_SIZE(rcvbuf); 382 } 383 384 chan->max_pkt_size = HVS_MAX_PKT_SIZE; 385 386 ret = vmbus_open(chan, sndbuf, rcvbuf, NULL, 0, hvs_channel_cb, 387 conn_from_host ? new : sk); 388 if (ret != 0) { 389 if (conn_from_host) { 390 hvs_new->chan = NULL; 391 sock_put(new); 392 } else { 393 hvs->chan = NULL; 394 } 395 goto out; 396 } 397 398 set_per_channel_state(chan, conn_from_host ? new : sk); 399 400 /* This reference will be dropped by hvs_close_connection(). */ 401 sock_hold(conn_from_host ? new : sk); 402 vmbus_set_chn_rescind_callback(chan, hvs_close_connection); 403 404 /* Set the pending send size to max packet size to always get 405 * notifications from the host when there is enough writable space. 406 * The host is optimized to send notifications only when the pending 407 * size boundary is crossed, and not always. 408 */ 409 hvs_set_channel_pending_send_size(chan); 410 411 if (conn_from_host) { 412 new->sk_state = TCP_ESTABLISHED; 413 sk_acceptq_added(sk); 414 415 hvs_new->vm_srv_id = *if_type; 416 hvs_new->host_srv_id = *if_instance; 417 418 vsock_insert_connected(vnew); 419 420 vsock_enqueue_accept(sk, new); 421 } else { 422 sk->sk_state = TCP_ESTABLISHED; 423 sk->sk_socket->state = SS_CONNECTED; 424 425 vsock_insert_connected(vsock_sk(sk)); 426 } 427 428 sk->sk_state_change(sk); 429 430 out: 431 /* Release refcnt obtained when we called vsock_find_bound_socket() */ 432 sock_put(sk); 433 434 release_sock(sk); 435 } 436 437 static u32 hvs_get_local_cid(void) 438 { 439 return VMADDR_CID_ANY; 440 } 441 442 static int hvs_sock_init(struct vsock_sock *vsk, struct vsock_sock *psk) 443 { 444 struct hvsock *hvs; 445 struct sock *sk = sk_vsock(vsk); 446 447 hvs = kzalloc_obj(*hvs); 448 if (!hvs) 449 return -ENOMEM; 450 451 vsk->trans = hvs; 452 hvs->vsk = vsk; 453 sk->sk_sndbuf = RINGBUFFER_HVS_SND_SIZE; 454 sk->sk_rcvbuf = RINGBUFFER_HVS_RCV_SIZE; 455 return 0; 456 } 457 458 static int hvs_connect(struct vsock_sock *vsk) 459 { 460 union hvs_service_id vm, host; 461 struct hvsock *h = vsk->trans; 462 463 vm.srv_id = srv_id_template; 464 vm.svm_port = vsk->local_addr.svm_port; 465 h->vm_srv_id = vm.srv_id; 466 467 host.srv_id = srv_id_template; 468 host.svm_port = vsk->remote_addr.svm_port; 469 h->host_srv_id = host.srv_id; 470 471 return vmbus_send_tl_connect_request(&h->vm_srv_id, &h->host_srv_id); 472 } 473 474 static void hvs_shutdown_lock_held(struct hvsock *hvs, int mode) 475 { 476 struct vmpipe_proto_header hdr; 477 478 if (hvs->fin_sent || !hvs->chan) 479 return; 480 481 /* It can't fail: see hvs_channel_writable_bytes(). */ 482 (void)__hvs_send_data(hvs->chan, &hdr, 0); 483 hvs->fin_sent = true; 484 } 485 486 static int hvs_shutdown(struct vsock_sock *vsk, int mode) 487 { 488 if (!(mode & SEND_SHUTDOWN)) 489 return 0; 490 491 hvs_shutdown_lock_held(vsk->trans, mode); 492 return 0; 493 } 494 495 static void hvs_close_timeout(struct work_struct *work) 496 { 497 struct vsock_sock *vsk = 498 container_of(work, struct vsock_sock, close_work.work); 499 struct sock *sk = sk_vsock(vsk); 500 501 sock_hold(sk); 502 lock_sock(sk); 503 if (!sock_flag(sk, SOCK_DONE)) 504 hvs_do_close_lock_held(vsk, false); 505 506 vsk->close_work_scheduled = false; 507 release_sock(sk); 508 sock_put(sk); 509 } 510 511 /* Returns true, if it is safe to remove socket; false otherwise */ 512 static bool hvs_close_lock_held(struct vsock_sock *vsk) 513 { 514 struct sock *sk = sk_vsock(vsk); 515 516 if (!(sk->sk_state == TCP_ESTABLISHED || 517 sk->sk_state == TCP_CLOSING)) 518 return true; 519 520 if ((sk->sk_shutdown & SHUTDOWN_MASK) != SHUTDOWN_MASK) 521 hvs_shutdown_lock_held(vsk->trans, SHUTDOWN_MASK); 522 523 if (sock_flag(sk, SOCK_DONE)) 524 return true; 525 526 /* This reference will be dropped by the delayed close routine */ 527 sock_hold(sk); 528 INIT_DELAYED_WORK(&vsk->close_work, hvs_close_timeout); 529 vsk->close_work_scheduled = true; 530 schedule_delayed_work(&vsk->close_work, HVS_CLOSE_TIMEOUT); 531 return false; 532 } 533 534 static void hvs_release(struct vsock_sock *vsk) 535 { 536 bool remove_sock; 537 538 remove_sock = hvs_close_lock_held(vsk); 539 if (remove_sock) 540 vsock_remove_sock(vsk); 541 } 542 543 static void hvs_destruct(struct vsock_sock *vsk) 544 { 545 struct hvsock *hvs = vsk->trans; 546 struct vmbus_channel *chan = hvs->chan; 547 548 if (chan) 549 vmbus_hvsock_device_unregister(chan); 550 551 kfree(hvs); 552 vsk->trans = NULL; 553 } 554 555 static int hvs_dgram_bind(struct vsock_sock *vsk, struct sockaddr_vm *addr) 556 { 557 return -EOPNOTSUPP; 558 } 559 560 static int hvs_dgram_dequeue(struct vsock_sock *vsk, struct msghdr *msg, 561 size_t len, int flags) 562 { 563 return -EOPNOTSUPP; 564 } 565 566 static int hvs_dgram_enqueue(struct vsock_sock *vsk, 567 struct sockaddr_vm *remote, struct msghdr *msg, 568 size_t dgram_len) 569 { 570 return -EOPNOTSUPP; 571 } 572 573 static bool hvs_dgram_allow(struct vsock_sock *vsk, u32 cid, u32 port) 574 { 575 return false; 576 } 577 578 static int hvs_update_recv_data(struct hvsock *hvs) 579 { 580 struct hvs_recv_buf *recv_buf; 581 u32 pkt_len, payload_len; 582 583 pkt_len = hv_pkt_len(hvs->recv_desc); 584 585 if (pkt_len < HVS_HEADER_LEN) 586 return -EIO; 587 588 recv_buf = (struct hvs_recv_buf *)(hvs->recv_desc + 1); 589 payload_len = recv_buf->hdr.data_size; 590 591 if (payload_len > pkt_len - HVS_HEADER_LEN || 592 payload_len > HVS_MTU_SIZE) 593 return -EIO; 594 595 if (payload_len == 0) 596 WRITE_ONCE(hvs->vsk->peer_shutdown, 597 READ_ONCE(hvs->vsk->peer_shutdown) | 598 SEND_SHUTDOWN); 599 600 hvs->recv_data_len = payload_len; 601 hvs->recv_data_off = 0; 602 603 return 0; 604 } 605 606 static ssize_t hvs_stream_dequeue(struct vsock_sock *vsk, struct msghdr *msg, 607 size_t len, int flags) 608 { 609 struct hvsock *hvs = vsk->trans; 610 bool need_refill = !hvs->recv_desc; 611 struct hvs_recv_buf *recv_buf; 612 u32 to_read; 613 int ret; 614 615 if (flags & MSG_PEEK) 616 return -EOPNOTSUPP; 617 618 if (need_refill) { 619 hvs->recv_desc = hv_pkt_iter_first(hvs->chan); 620 if (!hvs->recv_desc) 621 return -ENOBUFS; 622 ret = hvs_update_recv_data(hvs); 623 if (ret) 624 return ret; 625 } 626 627 recv_buf = (struct hvs_recv_buf *)(hvs->recv_desc + 1); 628 to_read = min_t(u32, len, hvs->recv_data_len); 629 ret = memcpy_to_msg(msg, recv_buf->data + hvs->recv_data_off, to_read); 630 if (ret != 0) 631 return ret; 632 633 hvs->recv_data_len -= to_read; 634 if (hvs->recv_data_len == 0) { 635 hvs->recv_desc = hv_pkt_iter_next(hvs->chan, hvs->recv_desc); 636 if (hvs->recv_desc) { 637 ret = hvs_update_recv_data(hvs); 638 if (ret) 639 return ret; 640 } 641 } else { 642 hvs->recv_data_off += to_read; 643 } 644 645 return to_read; 646 } 647 648 static ssize_t hvs_stream_enqueue(struct vsock_sock *vsk, struct msghdr *msg, 649 size_t len) 650 { 651 struct hvsock *hvs = vsk->trans; 652 struct vmbus_channel *chan = hvs->chan; 653 struct hvs_send_buf *send_buf; 654 ssize_t to_write, max_writable; 655 ssize_t ret = 0; 656 ssize_t bytes_written = 0; 657 658 BUILD_BUG_ON(sizeof(*send_buf) != HV_HYP_PAGE_SIZE); 659 660 send_buf = kmalloc_obj(*send_buf); 661 if (!send_buf) 662 return -ENOMEM; 663 664 /* Reader(s) could be draining data from the channel as we write. 665 * Maximize bandwidth, by iterating until the channel is found to be 666 * full. 667 */ 668 while (len) { 669 max_writable = hvs_channel_writable_bytes(chan); 670 if (!max_writable) 671 break; 672 to_write = min_t(ssize_t, len, max_writable); 673 to_write = min_t(ssize_t, to_write, HVS_SEND_BUF_SIZE); 674 /* memcpy_from_msg is safe for loop as it advances the offsets 675 * within the message iterator. 676 */ 677 ret = memcpy_from_msg(send_buf->data, msg, to_write); 678 if (ret < 0) 679 goto out; 680 681 ret = hvs_send_data(hvs->chan, send_buf, to_write); 682 if (ret < 0) 683 goto out; 684 685 bytes_written += to_write; 686 len -= to_write; 687 } 688 out: 689 /* If any data has been sent, return that */ 690 if (bytes_written) 691 ret = bytes_written; 692 kfree(send_buf); 693 return ret; 694 } 695 696 static s64 hvs_stream_has_data(struct vsock_sock *vsk) 697 { 698 struct hvsock *hvs = vsk->trans; 699 s64 ret; 700 701 if (hvs->recv_data_len > 0) 702 return hvs->recv_data_len; 703 704 switch (hvs_channel_readable_payload(hvs->chan)) { 705 case 1: 706 if (hvs->recv_desc) { 707 /* Here hvs->recv_data_len is 0, so hvs->recv_desc must 708 * be NULL unless it points to the 0-byte-payload FIN 709 * packet or a malformed/short packet: see 710 * hvs_update_recv_data(). 711 * 712 * If hvs->recv_desc points to the FIN packet, here all 713 * the payload has been dequeued and the peer_shutdown 714 * flag is set, but hvs_channel_readable_payload() still 715 * returns 1, because the VMBus ringbuffer's read_index 716 * is not updated for the FIN packet: 717 * hvs_stream_dequeue() -> hv_pkt_iter_next() updates 718 * the cached priv_read_index but has no opportunity to 719 * update the read_index in hv_pkt_iter_close() as 720 * hvs_stream_has_data() returns 0 for the FIN packet, 721 * so it won't get dequeued. 722 * 723 * In case hvs->recv_desc points to a malformed/short 724 * packet, return -EIO. 725 */ 726 if (!(vsk->peer_shutdown & SEND_SHUTDOWN)) 727 return -EIO; 728 729 return 0; 730 } 731 732 hvs->recv_desc = hv_pkt_iter_first(hvs->chan); 733 if (!hvs->recv_desc) 734 return -ENOBUFS; 735 736 ret = hvs_update_recv_data(hvs); 737 if (ret) 738 return ret; 739 return hvs->recv_data_len; 740 case 0: 741 WRITE_ONCE(vsk->peer_shutdown, 742 READ_ONCE(vsk->peer_shutdown) | SEND_SHUTDOWN); 743 ret = 0; 744 break; 745 default: /* -1 */ 746 ret = 0; 747 break; 748 } 749 750 return ret; 751 } 752 753 static s64 hvs_stream_has_space(struct vsock_sock *vsk) 754 { 755 struct hvsock *hvs = vsk->trans; 756 757 return hvs_channel_writable_bytes(hvs->chan); 758 } 759 760 static u64 hvs_stream_rcvhiwat(struct vsock_sock *vsk) 761 { 762 return HVS_MTU_SIZE + 1; 763 } 764 765 static bool hvs_stream_is_active(struct vsock_sock *vsk) 766 { 767 struct hvsock *hvs = vsk->trans; 768 769 return hvs->chan != NULL; 770 } 771 772 static bool hvs_stream_allow(struct vsock_sock *vsk, u32 cid, u32 port) 773 { 774 if (!vsock_net_mode_global(vsk)) 775 return false; 776 777 if (cid == VMADDR_CID_HOST) 778 return true; 779 780 return false; 781 } 782 783 static 784 int hvs_notify_poll_in(struct vsock_sock *vsk, size_t target, bool *readable) 785 { 786 struct hvsock *hvs = vsk->trans; 787 788 *readable = hvs_channel_readable(hvs->chan); 789 return 0; 790 } 791 792 static 793 int hvs_notify_poll_out(struct vsock_sock *vsk, size_t target, bool *writable) 794 { 795 *writable = hvs_stream_has_space(vsk) > 0; 796 797 return 0; 798 } 799 800 static 801 int hvs_notify_recv_init(struct vsock_sock *vsk, size_t target, 802 struct vsock_transport_recv_notify_data *d) 803 { 804 return 0; 805 } 806 807 static 808 int hvs_notify_recv_pre_block(struct vsock_sock *vsk, size_t target, 809 struct vsock_transport_recv_notify_data *d) 810 { 811 return 0; 812 } 813 814 static 815 int hvs_notify_recv_pre_dequeue(struct vsock_sock *vsk, size_t target, 816 struct vsock_transport_recv_notify_data *d) 817 { 818 return 0; 819 } 820 821 static 822 int hvs_notify_recv_post_dequeue(struct vsock_sock *vsk, size_t target, 823 ssize_t copied, bool data_read, 824 struct vsock_transport_recv_notify_data *d) 825 { 826 return 0; 827 } 828 829 static 830 int hvs_notify_send_init(struct vsock_sock *vsk, 831 struct vsock_transport_send_notify_data *d) 832 { 833 return 0; 834 } 835 836 static 837 int hvs_notify_send_pre_block(struct vsock_sock *vsk, 838 struct vsock_transport_send_notify_data *d) 839 { 840 return 0; 841 } 842 843 static 844 int hvs_notify_send_pre_enqueue(struct vsock_sock *vsk, 845 struct vsock_transport_send_notify_data *d) 846 { 847 return 0; 848 } 849 850 static 851 int hvs_notify_send_post_enqueue(struct vsock_sock *vsk, ssize_t written, 852 struct vsock_transport_send_notify_data *d) 853 { 854 return 0; 855 } 856 857 static 858 int hvs_notify_set_rcvlowat(struct vsock_sock *vsk, int val) 859 { 860 return -EOPNOTSUPP; 861 } 862 863 static struct vsock_transport hvs_transport = { 864 .module = THIS_MODULE, 865 866 .get_local_cid = hvs_get_local_cid, 867 868 .init = hvs_sock_init, 869 .destruct = hvs_destruct, 870 .release = hvs_release, 871 .connect = hvs_connect, 872 .shutdown = hvs_shutdown, 873 874 .dgram_bind = hvs_dgram_bind, 875 .dgram_dequeue = hvs_dgram_dequeue, 876 .dgram_enqueue = hvs_dgram_enqueue, 877 .dgram_allow = hvs_dgram_allow, 878 879 .stream_dequeue = hvs_stream_dequeue, 880 .stream_enqueue = hvs_stream_enqueue, 881 .stream_has_data = hvs_stream_has_data, 882 .stream_has_space = hvs_stream_has_space, 883 .stream_rcvhiwat = hvs_stream_rcvhiwat, 884 .stream_is_active = hvs_stream_is_active, 885 .stream_allow = hvs_stream_allow, 886 887 .notify_poll_in = hvs_notify_poll_in, 888 .notify_poll_out = hvs_notify_poll_out, 889 .notify_recv_init = hvs_notify_recv_init, 890 .notify_recv_pre_block = hvs_notify_recv_pre_block, 891 .notify_recv_pre_dequeue = hvs_notify_recv_pre_dequeue, 892 .notify_recv_post_dequeue = hvs_notify_recv_post_dequeue, 893 .notify_send_init = hvs_notify_send_init, 894 .notify_send_pre_block = hvs_notify_send_pre_block, 895 .notify_send_pre_enqueue = hvs_notify_send_pre_enqueue, 896 .notify_send_post_enqueue = hvs_notify_send_post_enqueue, 897 898 .notify_set_rcvlowat = hvs_notify_set_rcvlowat 899 }; 900 901 static bool hvs_check_transport(struct vsock_sock *vsk) 902 { 903 return vsk->transport == &hvs_transport; 904 } 905 906 static int hvs_probe(struct hv_device *hdev, 907 const struct hv_vmbus_device_id *dev_id) 908 { 909 struct vmbus_channel *chan = hdev->channel; 910 911 hvs_open_connection(chan); 912 913 /* Always return success to suppress the unnecessary error message 914 * in vmbus_probe(): on error the host will rescind the device in 915 * 30 seconds and we can do cleanup at that time in 916 * vmbus_onoffer_rescind(). 917 */ 918 return 0; 919 } 920 921 static void hvs_remove(struct hv_device *hdev) 922 { 923 struct vmbus_channel *chan = hdev->channel; 924 925 vmbus_close(chan); 926 } 927 928 /* hv_sock connections can not persist across hibernation, and all the hv_sock 929 * channels are forced to be rescinded before hibernation: see 930 * vmbus_bus_suspend(). Here the dummy hvs_suspend() and hvs_resume() 931 * are only needed because hibernation requires that every vmbus device's 932 * driver should have a .suspend and .resume callback: see vmbus_suspend(). 933 */ 934 static int hvs_suspend(struct hv_device *hv_dev) 935 { 936 /* Dummy */ 937 return 0; 938 } 939 940 static int hvs_resume(struct hv_device *dev) 941 { 942 /* Dummy */ 943 return 0; 944 } 945 946 /* This isn't really used. See vmbus_match() and vmbus_probe() */ 947 static const struct hv_vmbus_device_id id_table[] = { 948 {}, 949 }; 950 951 static struct hv_driver hvs_drv = { 952 .name = "hv_sock", 953 .hvsock = true, 954 .id_table = id_table, 955 .probe = hvs_probe, 956 .remove = hvs_remove, 957 .suspend = hvs_suspend, 958 .resume = hvs_resume, 959 }; 960 961 static int __init hvs_init(void) 962 { 963 int ret; 964 965 if (vmbus_proto_version < VERSION_WIN10) 966 return -ENODEV; 967 968 ret = vmbus_driver_register(&hvs_drv); 969 if (ret != 0) 970 return ret; 971 972 ret = vsock_core_register(&hvs_transport, VSOCK_TRANSPORT_F_G2H); 973 if (ret) { 974 vmbus_driver_unregister(&hvs_drv); 975 return ret; 976 } 977 978 return 0; 979 } 980 981 static void __exit hvs_exit(void) 982 { 983 vsock_core_unregister(&hvs_transport); 984 vmbus_driver_unregister(&hvs_drv); 985 } 986 987 module_init(hvs_init); 988 module_exit(hvs_exit); 989 990 MODULE_DESCRIPTION("Hyper-V Sockets"); 991 MODULE_VERSION("1.0.0"); 992 MODULE_LICENSE("GPL"); 993 MODULE_ALIAS_NETPROTO(PF_VSOCK); 994