1 /*- 2 * SPDX-License-Identifier: BSD-2-Clause-FreeBSD 3 * 4 * Copyright (c) 2020 Microsoft Corp. 5 * All rights reserved. 6 * 7 * Redistribution and use in source and binary forms, with or without 8 * modification, are permitted provided that the following conditions 9 * are met: 10 * 1. Redistributions of source code must retain the above copyright 11 * notice unmodified, this list of conditions, and the following 12 * disclaimer. 13 * 2. Redistributions in binary form must reproduce the above copyright 14 * notice, this list of conditions and the following disclaimer in the 15 * documentation and/or other materials provided with the distribution. 16 * 17 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR 18 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES 19 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. 20 * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, 21 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT 22 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, 23 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY 24 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT 25 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF 26 * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 27 */ 28 29 #include <sys/cdefs.h> 30 __FBSDID("$FreeBSD$"); 31 32 #include <sys/param.h> 33 #include <sys/bus.h> 34 #include <sys/domain.h> 35 #include <sys/lock.h> 36 #include <sys/kernel.h> 37 #include <sys/types.h> 38 #include <sys/malloc.h> 39 #include <sys/module.h> 40 #include <sys/mutex.h> 41 #include <sys/proc.h> 42 #include <sys/protosw.h> 43 #include <sys/socket.h> 44 #include <sys/sysctl.h> 45 #include <sys/sysproto.h> 46 #include <sys/systm.h> 47 #include <sys/sockbuf.h> 48 #include <sys/sx.h> 49 #include <sys/uio.h> 50 51 #include <net/vnet.h> 52 53 #include <dev/hyperv/vmbus/vmbus_reg.h> 54 55 #include "hv_sock.h" 56 57 #define HVSOCK_DBG_NONE 0x0 58 #define HVSOCK_DBG_INFO 0x1 59 #define HVSOCK_DBG_ERR 0x2 60 #define HVSOCK_DBG_VERBOSE 0x3 61 62 63 SYSCTL_NODE(_net, OID_AUTO, hvsock, CTLFLAG_RD, 0, "HyperV socket"); 64 65 static int hvs_dbg_level; 66 SYSCTL_INT(_net_hvsock, OID_AUTO, hvs_dbg_level, CTLFLAG_RWTUN, &hvs_dbg_level, 67 0, "hyperv socket debug level: 0 = none, 1 = info, 2 = error, 3 = verbose"); 68 69 70 #define HVSOCK_DBG(level, ...) do { \ 71 if (hvs_dbg_level >= (level)) \ 72 printf(__VA_ARGS__); \ 73 } while (0) 74 75 MALLOC_DEFINE(M_HVSOCK, "hyperv_socket", "hyperv socket control structures"); 76 77 static int hvs_dom_probe(void); 78 79 /* The MTU is 16KB per host side's design */ 80 #define HVSOCK_MTU_SIZE (1024 * 16) 81 #define HVSOCK_SEND_BUF_SZ (PAGE_SIZE - sizeof(struct vmpipe_proto_header)) 82 83 #define HVSOCK_HEADER_LEN (sizeof(struct hvs_pkt_header)) 84 85 #define HVSOCK_PKT_LEN(payload_len) (HVSOCK_HEADER_LEN + \ 86 roundup2(payload_len, 8) + \ 87 sizeof(uint64_t)) 88 89 90 static struct domain hv_socket_domain; 91 92 /* 93 * HyperV Transport sockets 94 */ 95 static struct pr_usrreqs hvs_trans_usrreqs = { 96 .pru_attach = hvs_trans_attach, 97 .pru_bind = hvs_trans_bind, 98 .pru_listen = hvs_trans_listen, 99 .pru_accept = hvs_trans_accept, 100 .pru_connect = hvs_trans_connect, 101 .pru_peeraddr = hvs_trans_peeraddr, 102 .pru_sockaddr = hvs_trans_sockaddr, 103 .pru_soreceive = hvs_trans_soreceive, 104 .pru_sosend = hvs_trans_sosend, 105 .pru_disconnect = hvs_trans_disconnect, 106 .pru_close = hvs_trans_close, 107 .pru_detach = hvs_trans_detach, 108 .pru_shutdown = hvs_trans_shutdown, 109 .pru_abort = hvs_trans_abort, 110 }; 111 112 /* 113 * Definitions of protocols supported in HyperV socket domain 114 */ 115 static struct protosw hv_socket_protosw[] = { 116 { 117 .pr_type = SOCK_STREAM, 118 .pr_domain = &hv_socket_domain, 119 .pr_protocol = HYPERV_SOCK_PROTO_TRANS, 120 .pr_flags = PR_CONNREQUIRED, 121 .pr_init = hvs_trans_init, 122 .pr_usrreqs = &hvs_trans_usrreqs, 123 }, 124 }; 125 126 static struct domain hv_socket_domain = { 127 .dom_family = AF_HYPERV, 128 .dom_name = "hyperv", 129 .dom_probe = hvs_dom_probe, 130 .dom_protosw = hv_socket_protosw, 131 .dom_protoswNPROTOSW = &hv_socket_protosw[nitems(hv_socket_protosw)] 132 }; 133 134 VNET_DOMAIN_SET(hv_socket_); 135 136 #define MAX_PORT ((uint32_t)0xFFFFFFFF) 137 #define MIN_PORT ((uint32_t)0x0) 138 139 /* 00000000-facb-11e6-bd58-64006a7986d3 */ 140 static const struct hyperv_guid srv_id_template = { 141 .hv_guid = { 142 0x00, 0x00, 0x00, 0x00, 0xcb, 0xfa, 0xe6, 0x11, 143 0xbd, 0x58, 0x64, 0x00, 0x6a, 0x79, 0x86, 0xd3 } 144 }; 145 146 static int hvsock_br_callback(void *, int, void *); 147 static uint32_t hvsock_canread_check(struct hvs_pcb *); 148 static uint32_t hvsock_canwrite_check(struct hvs_pcb *); 149 static int hvsock_send_data(struct vmbus_channel *chan, 150 struct uio *uio, uint32_t to_write, struct sockbuf *sb); 151 152 153 154 /* Globals */ 155 static struct sx hvs_trans_socks_sx; 156 static struct mtx hvs_trans_socks_mtx; 157 static LIST_HEAD(, hvs_pcb) hvs_trans_bound_socks; 158 static LIST_HEAD(, hvs_pcb) hvs_trans_connected_socks; 159 static uint32_t previous_auto_bound_port; 160 161 static void 162 hvsock_print_guid(struct hyperv_guid *guid) 163 { 164 unsigned char *p = (unsigned char *)guid; 165 166 HVSOCK_DBG(HVSOCK_DBG_INFO, 167 "0x%x-0x%x-0x%x-0x%x-0x%x-0x%x-0x%x-0x%x-0x%x-0x%x-0x%x\n", 168 *(unsigned int *)p, 169 *((unsigned short *) &p[4]), 170 *((unsigned short *) &p[6]), 171 p[8], p[9], p[10], p[11], p[12], p[13], p[14], p[15]); 172 } 173 174 static bool 175 is_valid_srv_id(const struct hyperv_guid *id) 176 { 177 return !memcmp(&id->hv_guid[4], 178 &srv_id_template.hv_guid[4], sizeof(struct hyperv_guid) - 4); 179 } 180 181 static unsigned int 182 get_port_by_srv_id(const struct hyperv_guid *srv_id) 183 { 184 return *((const unsigned int *)srv_id); 185 } 186 187 static void 188 set_port_by_srv_id(struct hyperv_guid *srv_id, unsigned int port) 189 { 190 *((unsigned int *)srv_id) = port; 191 } 192 193 194 static void 195 __hvs_remove_pcb_from_list(struct hvs_pcb *pcb, unsigned char list) 196 { 197 struct hvs_pcb *p = NULL; 198 199 HVSOCK_DBG(HVSOCK_DBG_VERBOSE, "%s: pcb is %p\n", __func__, pcb); 200 201 if (!pcb) 202 return; 203 204 if (list & HVS_LIST_BOUND) { 205 LIST_FOREACH(p, &hvs_trans_bound_socks, bound_next) 206 if (p == pcb) 207 LIST_REMOVE(p, bound_next); 208 } 209 210 if (list & HVS_LIST_CONNECTED) { 211 LIST_FOREACH(p, &hvs_trans_connected_socks, connected_next) 212 if (p == pcb) 213 LIST_REMOVE(pcb, connected_next); 214 } 215 } 216 217 static void 218 __hvs_remove_socket_from_list(struct socket *so, unsigned char list) 219 { 220 struct hvs_pcb *pcb = so2hvspcb(so); 221 222 HVSOCK_DBG(HVSOCK_DBG_VERBOSE, "%s: pcb is %p\n", __func__, pcb); 223 224 __hvs_remove_pcb_from_list(pcb, list); 225 } 226 227 static void 228 __hvs_insert_socket_on_list(struct socket *so, unsigned char list) 229 { 230 struct hvs_pcb *pcb = so2hvspcb(so); 231 232 if (list & HVS_LIST_BOUND) 233 LIST_INSERT_HEAD(&hvs_trans_bound_socks, 234 pcb, bound_next); 235 236 if (list & HVS_LIST_CONNECTED) 237 LIST_INSERT_HEAD(&hvs_trans_connected_socks, 238 pcb, connected_next); 239 } 240 241 void 242 hvs_remove_socket_from_list(struct socket *so, unsigned char list) 243 { 244 if (!so || !so->so_pcb) { 245 HVSOCK_DBG(HVSOCK_DBG_VERBOSE, 246 "%s: socket or so_pcb is null\n", __func__); 247 return; 248 } 249 250 mtx_lock(&hvs_trans_socks_mtx); 251 __hvs_remove_socket_from_list(so, list); 252 mtx_unlock(&hvs_trans_socks_mtx); 253 } 254 255 static void 256 hvs_insert_socket_on_list(struct socket *so, unsigned char list) 257 { 258 if (!so || !so->so_pcb) { 259 HVSOCK_DBG(HVSOCK_DBG_VERBOSE, 260 "%s: socket or so_pcb is null\n", __func__); 261 return; 262 } 263 264 mtx_lock(&hvs_trans_socks_mtx); 265 __hvs_insert_socket_on_list(so, list); 266 mtx_unlock(&hvs_trans_socks_mtx); 267 } 268 269 static struct socket * 270 __hvs_find_socket_on_list(struct sockaddr_hvs *addr, unsigned char list) 271 { 272 struct hvs_pcb *p = NULL; 273 274 if (list & HVS_LIST_BOUND) 275 LIST_FOREACH(p, &hvs_trans_bound_socks, bound_next) 276 if (p->so != NULL && 277 addr->hvs_port == p->local_addr.hvs_port) 278 return p->so; 279 280 if (list & HVS_LIST_CONNECTED) 281 LIST_FOREACH(p, &hvs_trans_connected_socks, connected_next) 282 if (p->so != NULL && 283 addr->hvs_port == p->local_addr.hvs_port) 284 return p->so; 285 286 return NULL; 287 } 288 289 static struct socket * 290 hvs_find_socket_on_list(struct sockaddr_hvs *addr, unsigned char list) 291 { 292 struct socket *s = NULL; 293 294 mtx_lock(&hvs_trans_socks_mtx); 295 s = __hvs_find_socket_on_list(addr, list); 296 mtx_unlock(&hvs_trans_socks_mtx); 297 298 return s; 299 } 300 301 static inline void 302 hvs_addr_set(struct sockaddr_hvs *addr, unsigned int port) 303 { 304 memset(addr, 0, sizeof(*addr)); 305 addr->sa_family = AF_HYPERV; 306 addr->sa_len = sizeof(*addr); 307 addr->hvs_port = port; 308 } 309 310 void 311 hvs_addr_init(struct sockaddr_hvs *addr, const struct hyperv_guid *svr_id) 312 { 313 hvs_addr_set(addr, get_port_by_srv_id(svr_id)); 314 } 315 316 int 317 hvs_trans_lock(void) 318 { 319 sx_xlock(&hvs_trans_socks_sx); 320 return (0); 321 } 322 323 void 324 hvs_trans_unlock(void) 325 { 326 sx_xunlock(&hvs_trans_socks_sx); 327 } 328 329 static int 330 hvs_dom_probe(void) 331 { 332 333 /* Don't even give us a chance to attach on non-HyperV. */ 334 if (vm_guest != VM_GUEST_HV) 335 return (ENXIO); 336 return (0); 337 } 338 339 void 340 hvs_trans_init(void) 341 { 342 /* Skip initialization of globals for non-default instances. */ 343 if (!IS_DEFAULT_VNET(curvnet)) 344 return; 345 346 HVSOCK_DBG(HVSOCK_DBG_VERBOSE, 347 "%s: HyperV Socket hvs_trans_init called\n", __func__); 348 349 /* Initialize Globals */ 350 previous_auto_bound_port = MAX_PORT; 351 sx_init(&hvs_trans_socks_sx, "hvs_trans_sock_sx"); 352 mtx_init(&hvs_trans_socks_mtx, 353 "hvs_trans_socks_mtx", NULL, MTX_DEF); 354 LIST_INIT(&hvs_trans_bound_socks); 355 LIST_INIT(&hvs_trans_connected_socks); 356 } 357 358 /* 359 * Called in two cases: 360 * 1) When user calls socket(); 361 * 2) When we accept new incoming conneciton and call sonewconn(). 362 */ 363 int 364 hvs_trans_attach(struct socket *so, int proto, struct thread *td) 365 { 366 struct hvs_pcb *pcb = so2hvspcb(so); 367 368 HVSOCK_DBG(HVSOCK_DBG_VERBOSE, 369 "%s: HyperV Socket hvs_trans_attach called\n", __func__); 370 371 if (so->so_type != SOCK_STREAM) 372 return (ESOCKTNOSUPPORT); 373 374 if (proto != 0 && proto != HYPERV_SOCK_PROTO_TRANS) 375 return (EPROTONOSUPPORT); 376 377 if (pcb != NULL) 378 return (EISCONN); 379 pcb = malloc(sizeof(struct hvs_pcb), M_HVSOCK, M_NOWAIT | M_ZERO); 380 if (pcb == NULL) 381 return (ENOMEM); 382 383 pcb->so = so; 384 so->so_pcb = (void *)pcb; 385 386 return (0); 387 } 388 389 void 390 hvs_trans_detach(struct socket *so) 391 { 392 struct hvs_pcb *pcb; 393 394 HVSOCK_DBG(HVSOCK_DBG_VERBOSE, 395 "%s: HyperV Socket hvs_trans_detach called\n", __func__); 396 397 (void) hvs_trans_lock(); 398 pcb = so2hvspcb(so); 399 if (pcb == NULL) { 400 hvs_trans_unlock(); 401 return; 402 } 403 404 if (SOLISTENING(so)) { 405 bzero(pcb, sizeof(*pcb)); 406 free(pcb, M_HVSOCK); 407 } 408 409 so->so_pcb = NULL; 410 411 hvs_trans_unlock(); 412 } 413 414 int 415 hvs_trans_bind(struct socket *so, struct sockaddr *addr, struct thread *td) 416 { 417 struct hvs_pcb *pcb = so2hvspcb(so); 418 struct sockaddr_hvs *sa = (struct sockaddr_hvs *) addr; 419 int error = 0; 420 421 HVSOCK_DBG(HVSOCK_DBG_VERBOSE, 422 "%s: HyperV Socket hvs_trans_bind called\n", __func__); 423 424 if (sa == NULL) { 425 return (EINVAL); 426 } 427 428 if (pcb == NULL) { 429 return (EINVAL); 430 } 431 432 if (sa->sa_family != AF_HYPERV) { 433 HVSOCK_DBG(HVSOCK_DBG_ERR, 434 "%s: Not supported, sa_family is %u\n", 435 __func__, sa->sa_family); 436 return (EAFNOSUPPORT); 437 } 438 if (sa->sa_len != sizeof(*sa)) { 439 HVSOCK_DBG(HVSOCK_DBG_ERR, 440 "%s: Not supported, sa_len is %u\n", 441 __func__, sa->sa_len); 442 return (EINVAL); 443 } 444 445 HVSOCK_DBG(HVSOCK_DBG_VERBOSE, 446 "%s: binding port = 0x%x\n", __func__, sa->hvs_port); 447 448 mtx_lock(&hvs_trans_socks_mtx); 449 if (__hvs_find_socket_on_list(sa, 450 HVS_LIST_BOUND | HVS_LIST_CONNECTED)) { 451 error = EADDRINUSE; 452 } else { 453 /* 454 * The address is available for us to bind. 455 * Add socket to the bound list. 456 */ 457 hvs_addr_set(&pcb->local_addr, sa->hvs_port); 458 hvs_addr_set(&pcb->remote_addr, HVADDR_PORT_ANY); 459 __hvs_insert_socket_on_list(so, HVS_LIST_BOUND); 460 } 461 mtx_unlock(&hvs_trans_socks_mtx); 462 463 return (error); 464 } 465 466 int 467 hvs_trans_listen(struct socket *so, int backlog, struct thread *td) 468 { 469 struct hvs_pcb *pcb = so2hvspcb(so); 470 struct socket *bound_so; 471 int error; 472 473 HVSOCK_DBG(HVSOCK_DBG_VERBOSE, 474 "%s: HyperV Socket hvs_trans_listen called\n", __func__); 475 476 if (pcb == NULL) 477 return (EINVAL); 478 479 /* Check if the address is already bound and it was by us. */ 480 bound_so = hvs_find_socket_on_list(&pcb->local_addr, HVS_LIST_BOUND); 481 if (bound_so == NULL || bound_so != so) { 482 HVSOCK_DBG(HVSOCK_DBG_ERR, 483 "%s: Address not bound or not by us.\n", __func__); 484 return (EADDRNOTAVAIL); 485 } 486 487 SOCK_LOCK(so); 488 error = solisten_proto_check(so); 489 if (error == 0) 490 solisten_proto(so, backlog); 491 SOCK_UNLOCK(so); 492 493 HVSOCK_DBG(HVSOCK_DBG_VERBOSE, 494 "%s: HyperV Socket listen error = %d\n", __func__, error); 495 return (error); 496 } 497 498 int 499 hvs_trans_accept(struct socket *so, struct sockaddr **nam) 500 { 501 struct hvs_pcb *pcb = so2hvspcb(so); 502 503 HVSOCK_DBG(HVSOCK_DBG_VERBOSE, 504 "%s: HyperV Socket hvs_trans_accept called\n", __func__); 505 506 if (pcb == NULL) 507 return (EINVAL); 508 509 *nam = sodupsockaddr((struct sockaddr *) &pcb->remote_addr, 510 M_NOWAIT); 511 512 return ((*nam == NULL) ? ENOMEM : 0); 513 } 514 515 int 516 hvs_trans_connect(struct socket *so, struct sockaddr *nam, struct thread *td) 517 { 518 struct hvs_pcb *pcb = so2hvspcb(so); 519 struct sockaddr_hvs *raddr = (struct sockaddr_hvs *)nam; 520 bool found_auto_bound_port = false; 521 int i, error = 0; 522 523 HVSOCK_DBG(HVSOCK_DBG_VERBOSE, 524 "%s: HyperV Socket hvs_trans_connect called, remote port is %x\n", 525 __func__, raddr->hvs_port); 526 527 if (pcb == NULL) 528 return (EINVAL); 529 530 /* Verify the remote address */ 531 if (raddr == NULL) 532 return (EINVAL); 533 if (raddr->sa_family != AF_HYPERV) 534 return (EAFNOSUPPORT); 535 if (raddr->sa_len != sizeof(*raddr)) 536 return (EINVAL); 537 538 mtx_lock(&hvs_trans_socks_mtx); 539 if (so->so_state & 540 (SS_ISCONNECTED|SS_ISDISCONNECTING|SS_ISCONNECTING)) { 541 HVSOCK_DBG(HVSOCK_DBG_ERR, 542 "%s: socket connect in progress\n", 543 __func__); 544 error = EINPROGRESS; 545 goto out; 546 } 547 548 /* 549 * Find an available port for us to auto bind the local 550 * address. 551 */ 552 hvs_addr_set(&pcb->local_addr, 0); 553 554 for (i = previous_auto_bound_port - 1; 555 i != previous_auto_bound_port; i --) { 556 if (i == MIN_PORT) 557 i = MAX_PORT; 558 559 pcb->local_addr.hvs_port = i; 560 561 if (__hvs_find_socket_on_list(&pcb->local_addr, 562 HVS_LIST_BOUND | HVS_LIST_CONNECTED) == NULL) { 563 found_auto_bound_port = true; 564 previous_auto_bound_port = i; 565 HVSOCK_DBG(HVSOCK_DBG_VERBOSE, 566 "%s: found local bound port is %x\n", 567 __func__, pcb->local_addr.hvs_port); 568 break; 569 } 570 } 571 572 if (found_auto_bound_port == true) { 573 /* Found available port for auto bound, put on list */ 574 __hvs_insert_socket_on_list(so, HVS_LIST_BOUND); 575 /* Set VM service ID */ 576 pcb->vm_srv_id = srv_id_template; 577 set_port_by_srv_id(&pcb->vm_srv_id, pcb->local_addr.hvs_port); 578 /* Set host service ID and remote port */ 579 pcb->host_srv_id = srv_id_template; 580 set_port_by_srv_id(&pcb->host_srv_id, raddr->hvs_port); 581 hvs_addr_set(&pcb->remote_addr, raddr->hvs_port); 582 583 /* Change the socket state to SS_ISCONNECTING */ 584 soisconnecting(so); 585 } else { 586 HVSOCK_DBG(HVSOCK_DBG_ERR, 587 "%s: No local port available for auto bound\n", 588 __func__); 589 error = EADDRINUSE; 590 } 591 592 HVSOCK_DBG(HVSOCK_DBG_INFO, "Connect vm_srv_id is "); 593 hvsock_print_guid(&pcb->vm_srv_id); 594 HVSOCK_DBG(HVSOCK_DBG_INFO, "Connect host_srv_id is "); 595 hvsock_print_guid(&pcb->host_srv_id); 596 597 out: 598 mtx_unlock(&hvs_trans_socks_mtx); 599 600 if (found_auto_bound_port == true) 601 vmbus_req_tl_connect(&pcb->vm_srv_id, &pcb->host_srv_id); 602 603 return (error); 604 } 605 606 int 607 hvs_trans_disconnect(struct socket *so) 608 { 609 struct hvs_pcb *pcb; 610 611 HVSOCK_DBG(HVSOCK_DBG_VERBOSE, 612 "%s: HyperV Socket hvs_trans_disconnect called\n", __func__); 613 614 (void) hvs_trans_lock(); 615 pcb = so2hvspcb(so); 616 if (pcb == NULL) { 617 hvs_trans_unlock(); 618 return (EINVAL); 619 } 620 621 /* If socket is already disconnected, skip this */ 622 if ((so->so_state & SS_ISDISCONNECTED) == 0) 623 soisdisconnecting(so); 624 625 hvs_trans_unlock(); 626 627 return (0); 628 } 629 630 #define SBLOCKWAIT(f) (((f) & MSG_DONTWAIT) ? 0 : SBL_WAIT) 631 struct hvs_callback_arg { 632 struct uio *uio; 633 struct sockbuf *sb; 634 }; 635 636 int 637 hvs_trans_soreceive(struct socket *so, struct sockaddr **paddr, 638 struct uio *uio, struct mbuf **mp0, struct mbuf **controlp, int *flagsp) 639 { 640 struct hvs_pcb *pcb = so2hvspcb(so); 641 struct sockbuf *sb; 642 ssize_t orig_resid; 643 uint32_t canread, to_read; 644 int flags, error = 0; 645 struct hvs_callback_arg cbarg; 646 647 HVSOCK_DBG(HVSOCK_DBG_VERBOSE, 648 "%s: HyperV Socket hvs_trans_soreceive called\n", __func__); 649 650 if (so->so_type != SOCK_STREAM) 651 return (EINVAL); 652 if (pcb == NULL) 653 return (EINVAL); 654 655 if (flagsp != NULL) 656 flags = *flagsp &~ MSG_EOR; 657 else 658 flags = 0; 659 660 if (flags & MSG_PEEK) 661 return (EOPNOTSUPP); 662 663 /* If no space to copy out anything */ 664 if (uio->uio_resid == 0 || uio->uio_rw != UIO_READ) 665 return (EINVAL); 666 667 sb = &so->so_rcv; 668 669 orig_resid = uio->uio_resid; 670 671 /* Prevent other readers from entering the socket. */ 672 error = sblock(sb, SBLOCKWAIT(flags)); 673 if (error) { 674 HVSOCK_DBG(HVSOCK_DBG_ERR, 675 "%s: sblock returned error = %d\n", __func__, error); 676 return (error); 677 } 678 679 SOCKBUF_LOCK(sb); 680 681 cbarg.uio = uio; 682 cbarg.sb = sb; 683 /* 684 * If the socket is closing, there might still be some data 685 * in rx br to read. However we need to make sure 686 * the channel is still open. 687 */ 688 if ((sb->sb_state & SBS_CANTRCVMORE) && 689 (so->so_state & SS_ISDISCONNECTED)) { 690 /* Other thread already closed the channel */ 691 error = EPIPE; 692 goto out; 693 } 694 695 while (true) { 696 while (uio->uio_resid > 0 && 697 (canread = hvsock_canread_check(pcb)) > 0) { 698 to_read = MIN(canread, uio->uio_resid); 699 HVSOCK_DBG(HVSOCK_DBG_VERBOSE, 700 "%s: to_read = %u, skip = %u\n", __func__, to_read, 701 (unsigned int)(sizeof(struct hvs_pkt_header) + 702 pcb->recv_data_off)); 703 704 error = vmbus_chan_recv_peek_call(pcb->chan, to_read, 705 sizeof(struct hvs_pkt_header) + pcb->recv_data_off, 706 hvsock_br_callback, (void *)&cbarg); 707 /* 708 * It is possible socket is disconnected becasue 709 * we released lock in hvsock_br_callback. So we 710 * need to check the state to make sure it is not 711 * disconnected. 712 */ 713 if (error || so->so_state & SS_ISDISCONNECTED) { 714 break; 715 } 716 717 pcb->recv_data_len -= to_read; 718 pcb->recv_data_off += to_read; 719 } 720 721 if (error) 722 break; 723 724 /* Abort if socket has reported problems. */ 725 if (so->so_error) { 726 if (so->so_error == ESHUTDOWN && 727 orig_resid > uio->uio_resid) { 728 /* 729 * Although we got a FIN, we also received 730 * some data in this round. Delivery it 731 * to user. 732 */ 733 error = 0; 734 } else { 735 if (so->so_error != ESHUTDOWN) 736 error = so->so_error; 737 } 738 739 break; 740 } 741 742 /* Cannot received more. */ 743 if (sb->sb_state & SBS_CANTRCVMORE) 744 break; 745 746 /* We are done if buffer has been filled */ 747 if (uio->uio_resid == 0) 748 break; 749 750 if (!(flags & MSG_WAITALL) && orig_resid > uio->uio_resid) 751 break; 752 753 /* Buffer ring is empty and we shall not block */ 754 if ((so->so_state & SS_NBIO) || 755 (flags & (MSG_DONTWAIT|MSG_NBIO))) { 756 if (orig_resid == uio->uio_resid) { 757 /* We have not read anything */ 758 error = EAGAIN; 759 } 760 HVSOCK_DBG(HVSOCK_DBG_VERBOSE, 761 "%s: non blocked read return, error %d.\n", 762 __func__, error); 763 break; 764 } 765 766 /* 767 * Wait and block until (more) data comes in. 768 * Note: Drops the sockbuf lock during wait. 769 */ 770 error = sbwait(sb); 771 772 if (error) 773 break; 774 775 HVSOCK_DBG(HVSOCK_DBG_VERBOSE, 776 "%s: wake up from sbwait, read available is %u\n", 777 __func__, vmbus_chan_read_available(pcb->chan)); 778 } 779 780 out: 781 SOCKBUF_UNLOCK(sb); 782 783 sbunlock(sb); 784 785 /* We recieved a FIN in this call */ 786 if (so->so_error == ESHUTDOWN) { 787 if (so->so_snd.sb_state & SBS_CANTSENDMORE) { 788 /* Send has already closed */ 789 soisdisconnecting(so); 790 } else { 791 /* Just close the receive side */ 792 socantrcvmore(so); 793 } 794 } 795 796 HVSOCK_DBG(HVSOCK_DBG_VERBOSE, 797 "%s: returning error = %d, so_error = %d\n", 798 __func__, error, so->so_error); 799 800 return (error); 801 } 802 803 int 804 hvs_trans_sosend(struct socket *so, struct sockaddr *addr, struct uio *uio, 805 struct mbuf *top, struct mbuf *controlp, int flags, struct thread *td) 806 { 807 struct hvs_pcb *pcb = so2hvspcb(so); 808 struct sockbuf *sb; 809 ssize_t orig_resid; 810 uint32_t canwrite, to_write; 811 int error = 0; 812 813 HVSOCK_DBG(HVSOCK_DBG_VERBOSE, 814 "%s: HyperV Socket hvs_trans_sosend called, uio_resid = %zd\n", 815 __func__, uio->uio_resid); 816 817 if (so->so_type != SOCK_STREAM) 818 return (EINVAL); 819 if (pcb == NULL) 820 return (EINVAL); 821 822 /* If nothing to send */ 823 if (uio->uio_resid == 0 || uio->uio_rw != UIO_WRITE) 824 return (EINVAL); 825 826 sb = &so->so_snd; 827 828 orig_resid = uio->uio_resid; 829 830 /* Prevent other writers from entering the socket. */ 831 error = sblock(sb, SBLOCKWAIT(flags)); 832 if (error) { 833 HVSOCK_DBG(HVSOCK_DBG_ERR, 834 "%s: sblock returned error = %d\n", __func__, error); 835 return (error); 836 } 837 838 SOCKBUF_LOCK(sb); 839 840 if ((sb->sb_state & SBS_CANTSENDMORE) || 841 so->so_error == ESHUTDOWN) { 842 error = EPIPE; 843 goto out; 844 } 845 846 while (uio->uio_resid > 0) { 847 canwrite = hvsock_canwrite_check(pcb); 848 if (canwrite == 0) { 849 /* We have sent some data */ 850 if (orig_resid > uio->uio_resid) 851 break; 852 /* 853 * We have not sent any data and it is 854 * non-blocked io 855 */ 856 if (so->so_state & SS_NBIO || 857 (flags & (MSG_NBIO | MSG_DONTWAIT)) != 0) { 858 error = EWOULDBLOCK; 859 break; 860 } else { 861 /* 862 * We are here because there is no space on 863 * send buffer ring. Signal the other side 864 * to read and free more space. 865 * Sleep wait until space avaiable to send 866 * Note: Drops the sockbuf lock during wait. 867 */ 868 error = sbwait(sb); 869 870 if (error) 871 break; 872 873 HVSOCK_DBG(HVSOCK_DBG_VERBOSE, 874 "%s: wake up from sbwait, space avail on " 875 "tx ring is %u\n", 876 __func__, 877 vmbus_chan_write_available(pcb->chan)); 878 879 continue; 880 } 881 } 882 to_write = MIN(canwrite, uio->uio_resid); 883 to_write = MIN(to_write, HVSOCK_SEND_BUF_SZ); 884 885 HVSOCK_DBG(HVSOCK_DBG_VERBOSE, 886 "%s: canwrite is %u, to_write = %u\n", __func__, 887 canwrite, to_write); 888 error = hvsock_send_data(pcb->chan, uio, to_write, sb); 889 890 if (error) 891 break; 892 } 893 894 out: 895 SOCKBUF_UNLOCK(sb); 896 sbunlock(sb); 897 898 return (error); 899 } 900 901 int 902 hvs_trans_peeraddr(struct socket *so, struct sockaddr **nam) 903 { 904 struct hvs_pcb *pcb = so2hvspcb(so); 905 906 HVSOCK_DBG(HVSOCK_DBG_VERBOSE, 907 "%s: HyperV Socket hvs_trans_peeraddr called\n", __func__); 908 909 if (pcb == NULL) 910 return (EINVAL); 911 912 *nam = sodupsockaddr((struct sockaddr *) &pcb->remote_addr, M_NOWAIT); 913 914 return ((*nam == NULL)? ENOMEM : 0); 915 } 916 917 int 918 hvs_trans_sockaddr(struct socket *so, struct sockaddr **nam) 919 { 920 struct hvs_pcb *pcb = so2hvspcb(so); 921 922 HVSOCK_DBG(HVSOCK_DBG_VERBOSE, 923 "%s: HyperV Socket hvs_trans_sockaddr called\n", __func__); 924 925 if (pcb == NULL) 926 return (EINVAL); 927 928 *nam = sodupsockaddr((struct sockaddr *) &pcb->local_addr, M_NOWAIT); 929 930 return ((*nam == NULL)? ENOMEM : 0); 931 } 932 933 void 934 hvs_trans_close(struct socket *so) 935 { 936 struct hvs_pcb *pcb; 937 938 HVSOCK_DBG(HVSOCK_DBG_VERBOSE, 939 "%s: HyperV Socket hvs_trans_close called\n", __func__); 940 941 (void) hvs_trans_lock(); 942 pcb = so2hvspcb(so); 943 if (!pcb) { 944 hvs_trans_unlock(); 945 return; 946 } 947 948 if (so->so_state & SS_ISCONNECTED) { 949 /* Send a FIN to peer */ 950 HVSOCK_DBG(HVSOCK_DBG_VERBOSE, 951 "%s: hvs_trans_close sending a FIN to host\n", __func__); 952 (void) hvsock_send_data(pcb->chan, NULL, 0, NULL); 953 } 954 955 if (so->so_state & 956 (SS_ISCONNECTED|SS_ISCONNECTING|SS_ISDISCONNECTING)) 957 soisdisconnected(so); 958 959 pcb->chan = NULL; 960 pcb->so = NULL; 961 962 if (SOLISTENING(so)) { 963 mtx_lock(&hvs_trans_socks_mtx); 964 /* Remove from bound list */ 965 __hvs_remove_socket_from_list(so, HVS_LIST_BOUND); 966 mtx_unlock(&hvs_trans_socks_mtx); 967 } 968 969 hvs_trans_unlock(); 970 971 return; 972 } 973 974 void 975 hvs_trans_abort(struct socket *so) 976 { 977 struct hvs_pcb *pcb = so2hvspcb(so); 978 979 HVSOCK_DBG(HVSOCK_DBG_VERBOSE, 980 "%s: HyperV Socket hvs_trans_abort called\n", __func__); 981 982 (void) hvs_trans_lock(); 983 if (pcb == NULL) { 984 hvs_trans_unlock(); 985 return; 986 } 987 988 if (SOLISTENING(so)) { 989 mtx_lock(&hvs_trans_socks_mtx); 990 /* Remove from bound list */ 991 __hvs_remove_socket_from_list(so, HVS_LIST_BOUND); 992 mtx_unlock(&hvs_trans_socks_mtx); 993 } 994 995 if (so->so_state & SS_ISCONNECTED) { 996 (void) sodisconnect(so); 997 } 998 hvs_trans_unlock(); 999 1000 return; 1001 } 1002 1003 int 1004 hvs_trans_shutdown(struct socket *so) 1005 { 1006 struct hvs_pcb *pcb = so2hvspcb(so); 1007 struct sockbuf *sb; 1008 1009 HVSOCK_DBG(HVSOCK_DBG_VERBOSE, 1010 "%s: HyperV Socket hvs_trans_shutdown called\n", __func__); 1011 1012 if (pcb == NULL) 1013 return (EINVAL); 1014 1015 /* 1016 * Only get called with the shutdown method is SHUT_WR or 1017 * SHUT_RDWR. 1018 * When the method is SHUT_RD or SHUT_RDWR, the caller 1019 * already set the SBS_CANTRCVMORE on receive side socket 1020 * buffer. 1021 */ 1022 if ((so->so_rcv.sb_state & SBS_CANTRCVMORE) == 0) { 1023 /* 1024 * SHUT_WR only case. 1025 * Receive side is still open. Just close 1026 * the send side. 1027 */ 1028 socantsendmore(so); 1029 } else { 1030 /* SHUT_RDWR case */ 1031 if (so->so_state & SS_ISCONNECTED) { 1032 /* Send a FIN to peer */ 1033 sb = &so->so_snd; 1034 SOCKBUF_LOCK(sb); 1035 (void) hvsock_send_data(pcb->chan, NULL, 0, sb); 1036 SOCKBUF_UNLOCK(sb); 1037 1038 soisdisconnecting(so); 1039 } 1040 } 1041 1042 return (0); 1043 } 1044 1045 /* In the VM, we support Hyper-V Sockets with AF_HYPERV, and the endpoint is 1046 * <port> (see struct sockaddr_hvs). 1047 * 1048 * On the host, Hyper-V Sockets are supported by Winsock AF_HYPERV: 1049 * https://docs.microsoft.com/en-us/virtualization/hyper-v-on-windows/user- 1050 * guide/make-integration-service, and the endpoint is <VmID, ServiceId> with 1051 * the below sockaddr: 1052 * 1053 * struct SOCKADDR_HV 1054 * { 1055 * ADDRESS_FAMILY Family; 1056 * USHORT Reserved; 1057 * GUID VmId; 1058 * GUID ServiceId; 1059 * }; 1060 * Note: VmID is not used by FreeBSD VM and actually it isn't transmitted via 1061 * VMBus, because here it's obvious the host and the VM can easily identify 1062 * each other. Though the VmID is useful on the host, especially in the case 1063 * of Windows container, FreeBSD VM doesn't need it at all. 1064 * 1065 * To be compatible with similar infrastructure in Linux VMs, we have 1066 * to limit the available GUID space of SOCKADDR_HV so that we can create 1067 * a mapping between FreeBSD AF_HYPERV port and SOCKADDR_HV Service GUID. 1068 * The rule of writing Hyper-V Sockets apps on the host and in FreeBSD VM is: 1069 * 1070 **************************************************************************** 1071 * The only valid Service GUIDs, from the perspectives of both the host and * 1072 * FreeBSD VM, that can be connected by the other end, must conform to this * 1073 * format: <port>-facb-11e6-bd58-64006a7986d3. * 1074 **************************************************************************** 1075 * 1076 * When we write apps on the host to connect(), the GUID ServiceID is used. 1077 * When we write apps in FreeBSD VM to connect(), we only need to specify the 1078 * port and the driver will form the GUID and use that to request the host. 1079 * 1080 * From the perspective of FreeBSD VM, the remote ephemeral port (i.e. the 1081 * auto-generated remote port for a connect request initiated by the host's 1082 * connect()) is set to HVADDR_PORT_UNKNOWN, which is not realy used on the 1083 * FreeBSD guest. 1084 */ 1085 1086 /* 1087 * Older HyperV hosts (vmbus version 'VMBUS_VERSION_WIN10' or before) 1088 * restricts HyperV socket ring buffer size to six 4K pages. Newer 1089 * HyperV hosts doen't have this limit. 1090 */ 1091 #define HVS_RINGBUF_RCV_SIZE (PAGE_SIZE * 6) 1092 #define HVS_RINGBUF_SND_SIZE (PAGE_SIZE * 6) 1093 #define HVS_RINGBUF_MAX_SIZE (PAGE_SIZE * 64) 1094 1095 struct hvsock_sc { 1096 device_t dev; 1097 struct hvs_pcb *pcb; 1098 struct vmbus_channel *channel; 1099 }; 1100 1101 static bool 1102 hvsock_chan_readable(struct vmbus_channel *chan) 1103 { 1104 uint32_t readable = vmbus_chan_read_available(chan); 1105 1106 return (readable >= HVSOCK_PKT_LEN(0)); 1107 } 1108 1109 static void 1110 hvsock_chan_cb(struct vmbus_channel *chan, void *context) 1111 { 1112 struct hvs_pcb *pcb = (struct hvs_pcb *) context; 1113 struct socket *so; 1114 uint32_t canwrite; 1115 1116 HVSOCK_DBG(HVSOCK_DBG_VERBOSE, 1117 "%s: host send us a wakeup on rb data, pcb = %p\n", 1118 __func__, pcb); 1119 1120 /* 1121 * Check if the socket is still attached and valid. 1122 * Here we know channel is still open. Need to make 1123 * sure the socket has not been closed or freed. 1124 */ 1125 (void) hvs_trans_lock(); 1126 so = hsvpcb2so(pcb); 1127 1128 if (pcb->chan != NULL && so != NULL) { 1129 /* 1130 * Wake up reader if there are data to read. 1131 */ 1132 SOCKBUF_LOCK(&(so)->so_rcv); 1133 1134 HVSOCK_DBG(HVSOCK_DBG_VERBOSE, 1135 "%s: read available = %u\n", __func__, 1136 vmbus_chan_read_available(pcb->chan)); 1137 1138 if (hvsock_chan_readable(pcb->chan)) 1139 sorwakeup_locked(so); 1140 else 1141 SOCKBUF_UNLOCK(&(so)->so_rcv); 1142 1143 /* 1144 * Wake up sender if space becomes available to write. 1145 */ 1146 SOCKBUF_LOCK(&(so)->so_snd); 1147 canwrite = hvsock_canwrite_check(pcb); 1148 1149 HVSOCK_DBG(HVSOCK_DBG_VERBOSE, 1150 "%s: canwrite = %u\n", __func__, canwrite); 1151 1152 if (canwrite > 0) { 1153 sowwakeup_locked(so); 1154 } else { 1155 SOCKBUF_UNLOCK(&(so)->so_snd); 1156 } 1157 } 1158 1159 hvs_trans_unlock(); 1160 1161 return; 1162 } 1163 1164 static int 1165 hvsock_br_callback(void *datap, int cplen, void *cbarg) 1166 { 1167 struct hvs_callback_arg *arg = (struct hvs_callback_arg *)cbarg; 1168 struct uio *uio = arg->uio; 1169 struct sockbuf *sb = arg->sb; 1170 int error = 0; 1171 1172 if (cbarg == NULL || datap == NULL) 1173 return (EINVAL); 1174 1175 HVSOCK_DBG(HVSOCK_DBG_VERBOSE, 1176 "%s: called, uio_rw = %s, uio_resid = %zd, cplen = %u, " 1177 "datap = %p\n", 1178 __func__, (uio->uio_rw == UIO_READ) ? "read from br":"write to br", 1179 uio->uio_resid, cplen, datap); 1180 1181 if (sb) 1182 SOCKBUF_UNLOCK(sb); 1183 1184 error = uiomove(datap, cplen, uio); 1185 1186 if (sb) 1187 SOCKBUF_LOCK(sb); 1188 1189 HVSOCK_DBG(HVSOCK_DBG_VERBOSE, 1190 "%s: after uiomove, uio_resid = %zd, error = %d\n", 1191 __func__, uio->uio_resid, error); 1192 1193 return (error); 1194 } 1195 1196 static int 1197 hvsock_send_data(struct vmbus_channel *chan, struct uio *uio, 1198 uint32_t to_write, struct sockbuf *sb) 1199 { 1200 struct hvs_pkt_header hvs_pkt; 1201 int hvs_pkthlen, hvs_pktlen, pad_pktlen, hlen, error = 0; 1202 uint64_t pad = 0; 1203 struct iovec iov[3]; 1204 struct hvs_callback_arg cbarg; 1205 1206 if (chan == NULL) 1207 return (ENOTCONN); 1208 1209 hlen = sizeof(struct vmbus_chanpkt_hdr); 1210 hvs_pkthlen = sizeof(struct hvs_pkt_header); 1211 hvs_pktlen = hvs_pkthlen + to_write; 1212 pad_pktlen = VMBUS_CHANPKT_TOTLEN(hvs_pktlen); 1213 1214 HVSOCK_DBG(HVSOCK_DBG_VERBOSE, 1215 "%s: hlen = %u, hvs_pkthlen = %u, hvs_pktlen = %u, " 1216 "pad_pktlen = %u, data_len = %u\n", 1217 __func__, hlen, hvs_pkthlen, hvs_pktlen, pad_pktlen, to_write); 1218 1219 hvs_pkt.chan_pkt_hdr.cph_type = VMBUS_CHANPKT_TYPE_INBAND; 1220 hvs_pkt.chan_pkt_hdr.cph_flags = 0; 1221 VMBUS_CHANPKT_SETLEN(hvs_pkt.chan_pkt_hdr.cph_hlen, hlen); 1222 VMBUS_CHANPKT_SETLEN(hvs_pkt.chan_pkt_hdr.cph_tlen, pad_pktlen); 1223 hvs_pkt.chan_pkt_hdr.cph_xactid = 0; 1224 1225 hvs_pkt.vmpipe_pkt_hdr.vmpipe_pkt_type = 1; 1226 hvs_pkt.vmpipe_pkt_hdr.vmpipe_data_size = to_write; 1227 1228 cbarg.uio = uio; 1229 cbarg.sb = sb; 1230 1231 if (uio && to_write > 0) { 1232 iov[0].iov_base = &hvs_pkt; 1233 iov[0].iov_len = hvs_pkthlen; 1234 iov[1].iov_base = NULL; 1235 iov[1].iov_len = to_write; 1236 iov[2].iov_base = &pad; 1237 iov[2].iov_len = pad_pktlen - hvs_pktlen; 1238 1239 error = vmbus_chan_iov_send(chan, iov, 3, 1240 hvsock_br_callback, &cbarg); 1241 } else { 1242 if (to_write == 0) { 1243 iov[0].iov_base = &hvs_pkt; 1244 iov[0].iov_len = hvs_pkthlen; 1245 iov[1].iov_base = &pad; 1246 iov[1].iov_len = pad_pktlen - hvs_pktlen; 1247 error = vmbus_chan_iov_send(chan, iov, 2, NULL, NULL); 1248 } 1249 } 1250 1251 if (error) { 1252 HVSOCK_DBG(HVSOCK_DBG_VERBOSE, 1253 "%s: error = %d\n", __func__, error); 1254 } 1255 1256 return (error); 1257 } 1258 1259 /* 1260 * Check if we have data on current ring buffer to read 1261 * or not. If not, advance the ring buffer read index to 1262 * next packet. Update the recev_data_len and recev_data_off 1263 * to new value. 1264 * Return the number of bytes can read. 1265 */ 1266 static uint32_t 1267 hvsock_canread_check(struct hvs_pcb *pcb) 1268 { 1269 uint32_t advance; 1270 uint32_t tlen, hlen, dlen; 1271 uint32_t bytes_canread = 0; 1272 int error; 1273 1274 if (pcb == NULL || pcb->chan == NULL) { 1275 pcb->so->so_error = EIO; 1276 return (0); 1277 } 1278 1279 /* Still have data not read yet on current packet */ 1280 if (pcb->recv_data_len > 0) 1281 return (pcb->recv_data_len); 1282 1283 if (pcb->rb_init) 1284 advance = 1285 VMBUS_CHANPKT_GETLEN(pcb->hvs_pkt.chan_pkt_hdr.cph_tlen); 1286 else 1287 advance = 0; 1288 1289 bytes_canread = vmbus_chan_read_available(pcb->chan); 1290 1291 HVSOCK_DBG(HVSOCK_DBG_VERBOSE, 1292 "%s: bytes_canread on br = %u, advance = %u\n", 1293 __func__, bytes_canread, advance); 1294 1295 if (pcb->rb_init && bytes_canread == (advance + sizeof(uint64_t))) { 1296 /* 1297 * Nothing to read. Need to advance the rindex before 1298 * calling sbwait, so host knows to wake us up when data 1299 * is available to read on rb. 1300 */ 1301 error = vmbus_chan_recv_idxadv(pcb->chan, advance); 1302 if (error) { 1303 HVSOCK_DBG(HVSOCK_DBG_ERR, 1304 "%s: after calling vmbus_chan_recv_idxadv, " 1305 "got error = %d\n", __func__, error); 1306 return (0); 1307 } else { 1308 pcb->rb_init = false; 1309 pcb->recv_data_len = 0; 1310 pcb->recv_data_off = 0; 1311 bytes_canread = vmbus_chan_read_available(pcb->chan); 1312 1313 HVSOCK_DBG(HVSOCK_DBG_VERBOSE, 1314 "%s: advanced %u bytes, " 1315 " bytes_canread on br now = %u\n", 1316 __func__, advance, bytes_canread); 1317 1318 if (bytes_canread == 0) 1319 return (0); 1320 else 1321 advance = 0; 1322 } 1323 } 1324 1325 if (bytes_canread < 1326 advance + (sizeof(struct hvs_pkt_header) + sizeof(uint64_t))) 1327 return (0); 1328 1329 error = vmbus_chan_recv_peek(pcb->chan, &pcb->hvs_pkt, 1330 sizeof(struct hvs_pkt_header), advance); 1331 1332 /* Don't have anything to read */ 1333 if (error) { 1334 HVSOCK_DBG(HVSOCK_DBG_VERBOSE, 1335 "%s: after calling vmbus_chan_recv_peek, got error = %d\n", 1336 __func__, error); 1337 return (0); 1338 } 1339 1340 /* 1341 * We just read in a new packet header. Do some sanity checks. 1342 */ 1343 tlen = VMBUS_CHANPKT_GETLEN(pcb->hvs_pkt.chan_pkt_hdr.cph_tlen); 1344 hlen = VMBUS_CHANPKT_GETLEN(pcb->hvs_pkt.chan_pkt_hdr.cph_hlen); 1345 dlen = pcb->hvs_pkt.vmpipe_pkt_hdr.vmpipe_data_size; 1346 if (__predict_false(hlen < sizeof(struct vmbus_chanpkt_hdr)) || 1347 __predict_false(hlen > tlen) || 1348 __predict_false(tlen < dlen + sizeof(struct hvs_pkt_header))) { 1349 HVSOCK_DBG(HVSOCK_DBG_VERBOSE, 1350 "invalid tlen(%u), hlen(%u) or dlen(%u)\n", 1351 tlen, hlen, dlen); 1352 pcb->so->so_error = EIO; 1353 return (0); 1354 } 1355 if (pcb->rb_init == false) 1356 pcb->rb_init = true; 1357 1358 HVSOCK_DBG(HVSOCK_DBG_VERBOSE, 1359 "Got new pkt tlen(%u), hlen(%u) or dlen(%u)\n", 1360 tlen, hlen, dlen); 1361 1362 /* The other side has sent a close FIN */ 1363 if (dlen == 0) { 1364 HVSOCK_DBG(HVSOCK_DBG_VERBOSE, 1365 "%s: Received FIN from other side\n", __func__); 1366 /* inform the caller by seting so_error to ESHUTDOWN */ 1367 pcb->so->so_error = ESHUTDOWN; 1368 } 1369 1370 HVSOCK_DBG(HVSOCK_DBG_VERBOSE, 1371 "%s: canread on receive ring is %u \n", __func__, dlen); 1372 1373 pcb->recv_data_len = dlen; 1374 pcb->recv_data_off = 0; 1375 1376 return (pcb->recv_data_len); 1377 } 1378 1379 static uint32_t 1380 hvsock_canwrite_check(struct hvs_pcb *pcb) 1381 { 1382 uint32_t writeable; 1383 uint32_t ret; 1384 1385 if (pcb == NULL || pcb->chan == NULL) 1386 return (0); 1387 1388 writeable = vmbus_chan_write_available(pcb->chan); 1389 1390 /* 1391 * We must always reserve a 0-length-payload packet for the FIN. 1392 */ 1393 HVSOCK_DBG(HVSOCK_DBG_VERBOSE, 1394 "%s: writeable is %u, should be greater than %ju\n", 1395 __func__, writeable, 1396 (uintmax_t)(HVSOCK_PKT_LEN(1) + HVSOCK_PKT_LEN(0))); 1397 1398 if (writeable < HVSOCK_PKT_LEN(1) + HVSOCK_PKT_LEN(0)) { 1399 /* 1400 * The Tx ring seems full. 1401 */ 1402 return (0); 1403 } 1404 1405 ret = writeable - HVSOCK_PKT_LEN(0) - HVSOCK_PKT_LEN(0); 1406 1407 HVSOCK_DBG(HVSOCK_DBG_VERBOSE, 1408 "%s: available size is %u\n", __func__, rounddown2(ret, 8)); 1409 1410 return (rounddown2(ret, 8)); 1411 } 1412 1413 static void 1414 hvsock_set_chan_pending_send_size(struct vmbus_channel *chan) 1415 { 1416 vmbus_chan_set_pending_send_size(chan, 1417 HVSOCK_PKT_LEN(HVSOCK_SEND_BUF_SZ)); 1418 } 1419 1420 static int 1421 hvsock_open_channel(struct vmbus_channel *chan, struct socket *so) 1422 { 1423 unsigned int rcvbuf, sndbuf; 1424 struct hvs_pcb *pcb = so2hvspcb(so); 1425 int ret; 1426 1427 if (vmbus_current_version < VMBUS_VERSION_WIN10_V5) { 1428 sndbuf = HVS_RINGBUF_SND_SIZE; 1429 rcvbuf = HVS_RINGBUF_RCV_SIZE; 1430 } else { 1431 sndbuf = MAX(so->so_snd.sb_hiwat, HVS_RINGBUF_SND_SIZE); 1432 sndbuf = MIN(sndbuf, HVS_RINGBUF_MAX_SIZE); 1433 sndbuf = rounddown2(sndbuf, PAGE_SIZE); 1434 rcvbuf = MAX(so->so_rcv.sb_hiwat, HVS_RINGBUF_RCV_SIZE); 1435 rcvbuf = MIN(rcvbuf, HVS_RINGBUF_MAX_SIZE); 1436 rcvbuf = rounddown2(rcvbuf, PAGE_SIZE); 1437 } 1438 1439 /* 1440 * Can only read whatever user provided size of data 1441 * from ring buffer. Turn off batched reading. 1442 */ 1443 vmbus_chan_set_readbatch(chan, false); 1444 1445 ret = vmbus_chan_open(chan, sndbuf, rcvbuf, NULL, 0, 1446 hvsock_chan_cb, pcb); 1447 1448 if (ret != 0) { 1449 HVSOCK_DBG(HVSOCK_DBG_ERR, 1450 "%s: failed to open hvsock channel, sndbuf = %u, " 1451 "rcvbuf = %u\n", __func__, sndbuf, rcvbuf); 1452 } else { 1453 HVSOCK_DBG(HVSOCK_DBG_INFO, 1454 "%s: hvsock channel opened, sndbuf = %u, i" 1455 "rcvbuf = %u\n", __func__, sndbuf, rcvbuf); 1456 /* 1457 * Se the pending send size so to receive wakeup 1458 * signals from host when there is enough space on 1459 * rx buffer ring to write. 1460 */ 1461 hvsock_set_chan_pending_send_size(chan); 1462 } 1463 1464 return ret; 1465 } 1466 1467 /* 1468 * Guest is listening passively on the socket. Open channel and 1469 * create a new socket for the conneciton. 1470 */ 1471 static void 1472 hvsock_open_conn_passive(struct vmbus_channel *chan, struct socket *so, 1473 struct hvsock_sc *sc) 1474 { 1475 struct socket *new_so; 1476 struct hvs_pcb *new_pcb, *pcb; 1477 int error; 1478 1479 /* Do nothing if socket is not listening */ 1480 if (!SOLISTENING(so)) { 1481 HVSOCK_DBG(HVSOCK_DBG_ERR, 1482 "%s: socket is not a listening one\n", __func__); 1483 return; 1484 } 1485 1486 /* 1487 * Create a new socket. This will call pru_attach to complete 1488 * the socket initialization and put the new socket onto 1489 * listening socket's sol_incomp list, waiting to be promoted 1490 * to sol_comp list. 1491 * The new socket created has ref count 0. There is no other 1492 * thread that changes the state of this new one at the 1493 * moment, so we don't need to hold its lock while opening 1494 * channel and filling out its pcb information. 1495 */ 1496 new_so = sonewconn(so, 0); 1497 if (!new_so) 1498 HVSOCK_DBG(HVSOCK_DBG_ERR, 1499 "%s: creating new socket failed\n", __func__); 1500 1501 /* 1502 * Now open the vmbus channel. If it fails, the socket will be 1503 * on the listening socket's sol_incomp queue until it is 1504 * replaced and aborted. 1505 */ 1506 error = hvsock_open_channel(chan, new_so); 1507 if (error) { 1508 new_so->so_error = error; 1509 return; 1510 } 1511 1512 pcb = so->so_pcb; 1513 new_pcb = new_so->so_pcb; 1514 1515 hvs_addr_set(&(new_pcb->local_addr), pcb->local_addr.hvs_port); 1516 /* Remote port is unknown to guest in this type of conneciton */ 1517 hvs_addr_set(&(new_pcb->remote_addr), HVADDR_PORT_UNKNOWN); 1518 new_pcb->chan = chan; 1519 new_pcb->recv_data_len = 0; 1520 new_pcb->recv_data_off = 0; 1521 new_pcb->rb_init = false; 1522 1523 new_pcb->vm_srv_id = *vmbus_chan_guid_type(chan); 1524 new_pcb->host_srv_id = *vmbus_chan_guid_inst(chan); 1525 1526 hvs_insert_socket_on_list(new_so, HVS_LIST_CONNECTED); 1527 1528 sc->pcb = new_pcb; 1529 1530 /* 1531 * Change the socket state to SS_ISCONNECTED. This will promote 1532 * the socket to sol_comp queue and wake up the thread which 1533 * is accepting connection. 1534 */ 1535 soisconnected(new_so); 1536 } 1537 1538 1539 /* 1540 * Guest is actively connecting to host. 1541 */ 1542 static void 1543 hvsock_open_conn_active(struct vmbus_channel *chan, struct socket *so) 1544 { 1545 struct hvs_pcb *pcb; 1546 int error; 1547 1548 error = hvsock_open_channel(chan, so); 1549 if (error) { 1550 so->so_error = error; 1551 return; 1552 } 1553 1554 pcb = so->so_pcb; 1555 pcb->chan = chan; 1556 pcb->recv_data_len = 0; 1557 pcb->recv_data_off = 0; 1558 pcb->rb_init = false; 1559 1560 mtx_lock(&hvs_trans_socks_mtx); 1561 __hvs_remove_socket_from_list(so, HVS_LIST_BOUND); 1562 __hvs_insert_socket_on_list(so, HVS_LIST_CONNECTED); 1563 mtx_unlock(&hvs_trans_socks_mtx); 1564 1565 /* 1566 * Change the socket state to SS_ISCONNECTED. This will wake up 1567 * the thread sleeping in connect call. 1568 */ 1569 soisconnected(so); 1570 } 1571 1572 static void 1573 hvsock_open_connection(struct vmbus_channel *chan, struct hvsock_sc *sc) 1574 { 1575 struct hyperv_guid *inst_guid, *type_guid; 1576 bool conn_from_host; 1577 struct sockaddr_hvs addr; 1578 struct socket *so; 1579 struct hvs_pcb *pcb; 1580 1581 type_guid = (struct hyperv_guid *) vmbus_chan_guid_type(chan); 1582 inst_guid = (struct hyperv_guid *) vmbus_chan_guid_inst(chan); 1583 conn_from_host = vmbus_chan_is_hvs_conn_from_host(chan); 1584 1585 HVSOCK_DBG(HVSOCK_DBG_INFO, "type_guid is "); 1586 hvsock_print_guid(type_guid); 1587 HVSOCK_DBG(HVSOCK_DBG_INFO, "inst_guid is "); 1588 hvsock_print_guid(inst_guid); 1589 HVSOCK_DBG(HVSOCK_DBG_INFO, "connection %s host\n", 1590 (conn_from_host == true ) ? "from" : "to"); 1591 1592 /* 1593 * The listening port should be in [0, MAX_LISTEN_PORT] 1594 */ 1595 if (!is_valid_srv_id(type_guid)) 1596 return; 1597 1598 /* 1599 * There should be a bound socket already created no matter 1600 * it is a passive or active connection. 1601 * For host initiated connection (passive on guest side), 1602 * the type_guid contains the port which guest is bound and 1603 * listening. 1604 * For the guest initiated connection (active on guest side), 1605 * the inst_guid contains the port that guest has auto bound 1606 * to. 1607 */ 1608 hvs_addr_init(&addr, conn_from_host ? type_guid : inst_guid); 1609 so = hvs_find_socket_on_list(&addr, HVS_LIST_BOUND); 1610 if (!so) { 1611 HVSOCK_DBG(HVSOCK_DBG_ERR, 1612 "%s: no bound socket found for port %u\n", 1613 __func__, addr.hvs_port); 1614 return; 1615 } 1616 1617 if (conn_from_host) { 1618 hvsock_open_conn_passive(chan, so, sc); 1619 } else { 1620 (void) hvs_trans_lock(); 1621 pcb = so->so_pcb; 1622 if (pcb && pcb->so) { 1623 sc->pcb = so2hvspcb(so); 1624 hvsock_open_conn_active(chan, so); 1625 } else { 1626 HVSOCK_DBG(HVSOCK_DBG_VERBOSE, 1627 "%s: channel detached before open\n", __func__); 1628 } 1629 hvs_trans_unlock(); 1630 } 1631 1632 } 1633 1634 static int 1635 hvsock_probe(device_t dev) 1636 { 1637 struct vmbus_channel *channel = vmbus_get_channel(dev); 1638 1639 if (!channel || !vmbus_chan_is_hvs(channel)) { 1640 HVSOCK_DBG(HVSOCK_DBG_VERBOSE, 1641 "hvsock_probe called but not a hvsock channel id %u\n", 1642 vmbus_chan_id(channel)); 1643 1644 return ENXIO; 1645 } else { 1646 HVSOCK_DBG(HVSOCK_DBG_VERBOSE, 1647 "hvsock_probe got a hvsock channel id %u\n", 1648 vmbus_chan_id(channel)); 1649 1650 return BUS_PROBE_DEFAULT; 1651 } 1652 } 1653 1654 static int 1655 hvsock_attach(device_t dev) 1656 { 1657 struct vmbus_channel *channel = vmbus_get_channel(dev); 1658 struct hvsock_sc *sc = (struct hvsock_sc *)device_get_softc(dev); 1659 1660 HVSOCK_DBG(HVSOCK_DBG_VERBOSE, "hvsock_attach called.\n"); 1661 1662 hvsock_open_connection(channel, sc); 1663 1664 /* 1665 * Always return success. On error the host will rescind the device 1666 * in 30 seconds and we can do cleanup at that time in 1667 * vmbus_chan_msgproc_chrescind(). 1668 */ 1669 return (0); 1670 } 1671 1672 static int 1673 hvsock_detach(device_t dev) 1674 { 1675 struct hvsock_sc *sc = (struct hvsock_sc *)device_get_softc(dev); 1676 struct socket *so; 1677 int error, retry; 1678 1679 if (bootverbose) 1680 device_printf(dev, "hvsock_detach called.\n"); 1681 1682 HVSOCK_DBG(HVSOCK_DBG_VERBOSE, "hvsock_detach called.\n"); 1683 1684 if (sc->pcb != NULL) { 1685 (void) hvs_trans_lock(); 1686 1687 so = hsvpcb2so(sc->pcb); 1688 if (so) { 1689 /* Close the connection */ 1690 if (so->so_state & 1691 (SS_ISCONNECTED|SS_ISCONNECTING|SS_ISDISCONNECTING)) 1692 soisdisconnected(so); 1693 } 1694 1695 mtx_lock(&hvs_trans_socks_mtx); 1696 __hvs_remove_pcb_from_list(sc->pcb, 1697 HVS_LIST_BOUND | HVS_LIST_CONNECTED); 1698 mtx_unlock(&hvs_trans_socks_mtx); 1699 1700 /* 1701 * Close channel while no reader and sender are working 1702 * on the buffer rings. 1703 */ 1704 if (so) { 1705 retry = 0; 1706 while ((error = sblock(&so->so_rcv, 0)) == 1707 EWOULDBLOCK) { 1708 /* 1709 * Someone is reading, rx br is busy 1710 */ 1711 soisdisconnected(so); 1712 DELAY(500); 1713 HVSOCK_DBG(HVSOCK_DBG_VERBOSE, 1714 "waiting for rx reader to exit, " 1715 "retry = %d\n", retry++); 1716 } 1717 retry = 0; 1718 while ((error = sblock(&so->so_snd, 0)) == 1719 EWOULDBLOCK) { 1720 /* 1721 * Someone is sending, tx br is busy 1722 */ 1723 soisdisconnected(so); 1724 DELAY(500); 1725 HVSOCK_DBG(HVSOCK_DBG_VERBOSE, 1726 "waiting for tx sender to exit, " 1727 "retry = %d\n", retry++); 1728 } 1729 } 1730 1731 1732 bzero(sc->pcb, sizeof(struct hvs_pcb)); 1733 free(sc->pcb, M_HVSOCK); 1734 sc->pcb = NULL; 1735 1736 if (so) { 1737 sbunlock(&so->so_rcv); 1738 sbunlock(&so->so_snd); 1739 so->so_pcb = NULL; 1740 } 1741 1742 hvs_trans_unlock(); 1743 } 1744 1745 vmbus_chan_close(vmbus_get_channel(dev)); 1746 1747 return (0); 1748 } 1749 1750 static device_method_t hvsock_methods[] = { 1751 /* Device interface */ 1752 DEVMETHOD(device_probe, hvsock_probe), 1753 DEVMETHOD(device_attach, hvsock_attach), 1754 DEVMETHOD(device_detach, hvsock_detach), 1755 DEVMETHOD_END 1756 }; 1757 1758 static driver_t hvsock_driver = { 1759 "hv_sock", 1760 hvsock_methods, 1761 sizeof(struct hvsock_sc) 1762 }; 1763 1764 static devclass_t hvsock_devclass; 1765 1766 DRIVER_MODULE(hvsock, vmbus, hvsock_driver, hvsock_devclass, NULL, NULL); 1767 MODULE_VERSION(hvsock, 1); 1768 MODULE_DEPEND(hvsock, vmbus, 1, 1, 1); 1769