1 /*- 2 * SPDX-License-Identifier: BSD-2-Clause-FreeBSD 3 * 4 * Copyright (c) 2012 Chelsio Communications, Inc. 5 * All rights reserved. 6 * Written by: Navdeep Parhar <np@FreeBSD.org> 7 * 8 * Redistribution and use in source and binary forms, with or without 9 * modification, are permitted provided that the following conditions 10 * are met: 11 * 1. Redistributions of source code must retain the above copyright 12 * notice, this list of conditions and the following 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 AND CONTRIBUTORS ``AS IS'' AND 18 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 19 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 20 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE 21 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 22 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 23 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 24 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 25 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 26 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 27 * SUCH DAMAGE. 28 */ 29 30 #include <sys/cdefs.h> 31 __FBSDID("$FreeBSD$"); 32 33 #include "opt_inet.h" 34 #include "opt_inet6.h" 35 36 #include <sys/param.h> 37 #include <sys/eventhandler.h> 38 #include <sys/kernel.h> 39 #include <sys/systm.h> 40 #include <sys/malloc.h> 41 #include <sys/mbuf.h> 42 #include <sys/module.h> 43 #include <sys/types.h> 44 #include <sys/sockopt.h> 45 #include <sys/sysctl.h> 46 #include <sys/socket.h> 47 48 #include <net/ethernet.h> 49 #include <net/if.h> 50 #include <net/if_var.h> 51 #include <net/if_types.h> 52 #include <net/if_vlan_var.h> 53 #include <net/if_llatbl.h> 54 #include <net/route.h> 55 56 #include <netinet/if_ether.h> 57 #include <netinet/in.h> 58 #include <netinet/in_pcb.h> 59 #include <netinet/in_var.h> 60 #include <netinet6/in6_var.h> 61 #include <netinet6/in6_pcb.h> 62 #include <netinet6/nd6.h> 63 #define TCPSTATES 64 #include <netinet/tcp.h> 65 #include <netinet/tcp_fsm.h> 66 #include <netinet/tcp_timer.h> 67 #include <netinet/tcp_var.h> 68 #include <netinet/tcp_syncache.h> 69 #include <netinet/tcp_offload.h> 70 #include <netinet/toecore.h> 71 72 static struct mtx toedev_lock; 73 static TAILQ_HEAD(, toedev) toedev_list; 74 static eventhandler_tag listen_start_eh; 75 static eventhandler_tag listen_stop_eh; 76 static eventhandler_tag lle_event_eh; 77 78 static int 79 toedev_connect(struct toedev *tod __unused, struct socket *so __unused, 80 struct rtentry *rt __unused, struct sockaddr *nam __unused) 81 { 82 83 return (ENOTSUP); 84 } 85 86 static int 87 toedev_listen_start(struct toedev *tod __unused, struct tcpcb *tp __unused) 88 { 89 90 return (ENOTSUP); 91 } 92 93 static int 94 toedev_listen_stop(struct toedev *tod __unused, struct tcpcb *tp __unused) 95 { 96 97 return (ENOTSUP); 98 } 99 100 static void 101 toedev_input(struct toedev *tod __unused, struct tcpcb *tp __unused, 102 struct mbuf *m) 103 { 104 105 m_freem(m); 106 return; 107 } 108 109 static void 110 toedev_rcvd(struct toedev *tod __unused, struct tcpcb *tp __unused) 111 { 112 113 return; 114 } 115 116 static int 117 toedev_output(struct toedev *tod __unused, struct tcpcb *tp __unused) 118 { 119 120 return (ENOTSUP); 121 } 122 123 static void 124 toedev_pcb_detach(struct toedev *tod __unused, struct tcpcb *tp __unused) 125 { 126 127 return; 128 } 129 130 static void 131 toedev_l2_update(struct toedev *tod __unused, struct ifnet *ifp __unused, 132 struct sockaddr *sa __unused, uint8_t *lladdr __unused, 133 uint16_t vtag __unused) 134 { 135 136 return; 137 } 138 139 static void 140 toedev_route_redirect(struct toedev *tod __unused, struct ifnet *ifp __unused, 141 struct rtentry *rt0 __unused, struct rtentry *rt1 __unused) 142 { 143 144 return; 145 } 146 147 static void 148 toedev_syncache_added(struct toedev *tod __unused, void *ctx __unused) 149 { 150 151 return; 152 } 153 154 static void 155 toedev_syncache_removed(struct toedev *tod __unused, void *ctx __unused) 156 { 157 158 return; 159 } 160 161 static int 162 toedev_syncache_respond(struct toedev *tod __unused, void *ctx __unused, 163 struct mbuf *m) 164 { 165 166 m_freem(m); 167 return (0); 168 } 169 170 static void 171 toedev_offload_socket(struct toedev *tod __unused, void *ctx __unused, 172 struct socket *so __unused) 173 { 174 175 return; 176 } 177 178 static void 179 toedev_ctloutput(struct toedev *tod __unused, struct tcpcb *tp __unused, 180 int sopt_dir __unused, int sopt_name __unused) 181 { 182 183 return; 184 } 185 186 static void 187 toedev_tcp_info(struct toedev *tod __unused, struct tcpcb *tp __unused, 188 struct tcp_info *ti __unused) 189 { 190 191 return; 192 } 193 194 static int 195 toedev_alloc_tls_session(struct toedev *tod __unused, struct tcpcb *tp __unused, 196 struct ktls_session *tls __unused) 197 { 198 199 return (EINVAL); 200 } 201 202 /* 203 * Inform one or more TOE devices about a listening socket. 204 */ 205 static void 206 toe_listen_start(struct inpcb *inp, void *arg) 207 { 208 struct toedev *t, *tod; 209 struct tcpcb *tp; 210 211 INP_WLOCK_ASSERT(inp); 212 KASSERT(inp->inp_pcbinfo == &V_tcbinfo, 213 ("%s: inp is not a TCP inp", __func__)); 214 215 if (inp->inp_flags & (INP_DROPPED | INP_TIMEWAIT)) 216 return; 217 218 tp = intotcpcb(inp); 219 if (tp->t_state != TCPS_LISTEN) 220 return; 221 222 t = arg; 223 mtx_lock(&toedev_lock); 224 TAILQ_FOREACH(tod, &toedev_list, link) { 225 if (t == NULL || t == tod) 226 tod->tod_listen_start(tod, tp); 227 } 228 mtx_unlock(&toedev_lock); 229 } 230 231 static void 232 toe_listen_start_event(void *arg __unused, struct tcpcb *tp) 233 { 234 struct inpcb *inp = tp->t_inpcb; 235 236 INP_WLOCK_ASSERT(inp); 237 KASSERT(tp->t_state == TCPS_LISTEN, 238 ("%s: t_state %s", __func__, tcpstates[tp->t_state])); 239 240 toe_listen_start(inp, NULL); 241 } 242 243 static void 244 toe_listen_stop_event(void *arg __unused, struct tcpcb *tp) 245 { 246 struct toedev *tod; 247 #ifdef INVARIANTS 248 struct inpcb *inp = tp->t_inpcb; 249 #endif 250 251 INP_WLOCK_ASSERT(inp); 252 KASSERT(tp->t_state == TCPS_LISTEN, 253 ("%s: t_state %s", __func__, tcpstates[tp->t_state])); 254 255 mtx_lock(&toedev_lock); 256 TAILQ_FOREACH(tod, &toedev_list, link) 257 tod->tod_listen_stop(tod, tp); 258 mtx_unlock(&toedev_lock); 259 } 260 261 /* 262 * Fill up a freshly allocated toedev struct with reasonable defaults. 263 */ 264 void 265 init_toedev(struct toedev *tod) 266 { 267 268 tod->tod_softc = NULL; 269 270 /* 271 * Provide no-op defaults so that the kernel can call any toedev 272 * function without having to check whether the TOE driver supplied one 273 * or not. 274 */ 275 tod->tod_connect = toedev_connect; 276 tod->tod_listen_start = toedev_listen_start; 277 tod->tod_listen_stop = toedev_listen_stop; 278 tod->tod_input = toedev_input; 279 tod->tod_rcvd = toedev_rcvd; 280 tod->tod_output = toedev_output; 281 tod->tod_send_rst = toedev_output; 282 tod->tod_send_fin = toedev_output; 283 tod->tod_pcb_detach = toedev_pcb_detach; 284 tod->tod_l2_update = toedev_l2_update; 285 tod->tod_route_redirect = toedev_route_redirect; 286 tod->tod_syncache_added = toedev_syncache_added; 287 tod->tod_syncache_removed = toedev_syncache_removed; 288 tod->tod_syncache_respond = toedev_syncache_respond; 289 tod->tod_offload_socket = toedev_offload_socket; 290 tod->tod_ctloutput = toedev_ctloutput; 291 tod->tod_tcp_info = toedev_tcp_info; 292 tod->tod_alloc_tls_session = toedev_alloc_tls_session; 293 } 294 295 /* 296 * Register an active TOE device with the system. This allows it to receive 297 * notifications from the kernel. 298 */ 299 int 300 register_toedev(struct toedev *tod) 301 { 302 struct toedev *t; 303 304 mtx_lock(&toedev_lock); 305 TAILQ_FOREACH(t, &toedev_list, link) { 306 if (t == tod) { 307 mtx_unlock(&toedev_lock); 308 return (EEXIST); 309 } 310 } 311 312 TAILQ_INSERT_TAIL(&toedev_list, tod, link); 313 registered_toedevs++; 314 mtx_unlock(&toedev_lock); 315 316 inp_apply_all(toe_listen_start, tod); 317 318 return (0); 319 } 320 321 /* 322 * Remove the TOE device from the global list of active TOE devices. It is the 323 * caller's responsibility to ensure that the TOE device is quiesced prior to 324 * this call. 325 */ 326 int 327 unregister_toedev(struct toedev *tod) 328 { 329 struct toedev *t, *t2; 330 int rc = ENODEV; 331 332 mtx_lock(&toedev_lock); 333 TAILQ_FOREACH_SAFE(t, &toedev_list, link, t2) { 334 if (t == tod) { 335 TAILQ_REMOVE(&toedev_list, tod, link); 336 registered_toedevs--; 337 rc = 0; 338 break; 339 } 340 } 341 KASSERT(registered_toedevs >= 0, 342 ("%s: registered_toedevs (%d) < 0", __func__, registered_toedevs)); 343 mtx_unlock(&toedev_lock); 344 return (rc); 345 } 346 347 void 348 toe_syncache_add(struct in_conninfo *inc, struct tcpopt *to, struct tcphdr *th, 349 struct inpcb *inp, void *tod, void *todctx) 350 { 351 struct socket *lso = inp->inp_socket; 352 353 INP_WLOCK_ASSERT(inp); 354 355 syncache_add(inc, to, th, inp, &lso, NULL, tod, todctx); 356 } 357 358 int 359 toe_syncache_expand(struct in_conninfo *inc, struct tcpopt *to, 360 struct tcphdr *th, struct socket **lsop) 361 { 362 363 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 364 365 return (syncache_expand(inc, to, th, lsop, NULL)); 366 } 367 368 /* 369 * General purpose check to see if a 4-tuple is in use by the kernel. If a TCP 370 * header (presumably for an incoming SYN) is also provided, an existing 4-tuple 371 * in TIME_WAIT may be assassinated freeing it up for re-use. 372 * 373 * Note that the TCP header must have been run through tcp_fields_to_host() or 374 * equivalent. 375 */ 376 int 377 toe_4tuple_check(struct in_conninfo *inc, struct tcphdr *th, struct ifnet *ifp) 378 { 379 struct inpcb *inp; 380 381 if (inc->inc_flags & INC_ISIPV6) { 382 inp = in6_pcblookup(&V_tcbinfo, &inc->inc6_faddr, 383 inc->inc_fport, &inc->inc6_laddr, inc->inc_lport, 384 INPLOOKUP_WLOCKPCB, ifp); 385 } else { 386 inp = in_pcblookup(&V_tcbinfo, inc->inc_faddr, inc->inc_fport, 387 inc->inc_laddr, inc->inc_lport, INPLOOKUP_WLOCKPCB, ifp); 388 } 389 if (inp != NULL) { 390 INP_WLOCK_ASSERT(inp); 391 392 if ((inp->inp_flags & INP_TIMEWAIT) && th != NULL) { 393 394 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); /* for twcheck */ 395 if (!tcp_twcheck(inp, NULL, th, NULL, 0)) 396 return (EADDRINUSE); 397 } else { 398 INP_WUNLOCK(inp); 399 return (EADDRINUSE); 400 } 401 } 402 403 return (0); 404 } 405 406 static void 407 toe_lle_event(void *arg __unused, struct llentry *lle, int evt) 408 { 409 struct toedev *tod; 410 struct ifnet *ifp; 411 struct sockaddr *sa; 412 uint8_t *lladdr; 413 uint16_t vid, pcp; 414 int family; 415 struct sockaddr_in6 sin6; 416 417 LLE_WLOCK_ASSERT(lle); 418 419 ifp = lltable_get_ifp(lle->lle_tbl); 420 family = lltable_get_af(lle->lle_tbl); 421 422 if (family != AF_INET && family != AF_INET6) 423 return; 424 /* 425 * Not interested if the interface's TOE capability is not enabled. 426 */ 427 if ((family == AF_INET && !(ifp->if_capenable & IFCAP_TOE4)) || 428 (family == AF_INET6 && !(ifp->if_capenable & IFCAP_TOE6))) 429 return; 430 431 tod = TOEDEV(ifp); 432 if (tod == NULL) 433 return; 434 435 sa = (struct sockaddr *)&sin6; 436 lltable_fill_sa_entry(lle, sa); 437 438 vid = 0xfff; 439 pcp = 0; 440 if (evt != LLENTRY_RESOLVED) { 441 442 /* 443 * LLENTRY_TIMEDOUT, LLENTRY_DELETED, LLENTRY_EXPIRED all mean 444 * this entry is going to be deleted. 445 */ 446 447 lladdr = NULL; 448 } else { 449 450 KASSERT(lle->la_flags & LLE_VALID, 451 ("%s: %p resolved but not valid?", __func__, lle)); 452 453 lladdr = (uint8_t *)lle->ll_addr; 454 VLAN_TAG(ifp, &vid); 455 VLAN_PCP(ifp, &pcp); 456 } 457 458 tod->tod_l2_update(tod, ifp, sa, lladdr, EVL_MAKETAG(vid, pcp, 0)); 459 } 460 461 /* 462 * Returns 0 or EWOULDBLOCK on success (any other value is an error). 0 means 463 * lladdr and vtag are valid on return, EWOULDBLOCK means the TOE driver's 464 * tod_l2_update will be called later, when the entry is resolved or times out. 465 */ 466 int 467 toe_l2_resolve(struct toedev *tod, struct ifnet *ifp, struct sockaddr *sa, 468 uint8_t *lladdr, uint16_t *vtag) 469 { 470 int rc; 471 uint16_t vid, pcp; 472 473 switch (sa->sa_family) { 474 #ifdef INET 475 case AF_INET: 476 rc = arpresolve(ifp, 0, NULL, sa, lladdr, NULL, NULL); 477 break; 478 #endif 479 #ifdef INET6 480 case AF_INET6: 481 rc = nd6_resolve(ifp, 0, NULL, sa, lladdr, NULL, NULL); 482 break; 483 #endif 484 default: 485 return (EPROTONOSUPPORT); 486 } 487 488 if (rc == 0) { 489 vid = 0xfff; 490 pcp = 0; 491 if (ifp->if_type == IFT_L2VLAN) { 492 VLAN_TAG(ifp, &vid); 493 VLAN_PCP(ifp, &pcp); 494 } else if (ifp->if_pcp != IFNET_PCP_NONE) { 495 vid = 0; 496 pcp = ifp->if_pcp; 497 } 498 *vtag = EVL_MAKETAG(vid, pcp, 0); 499 } 500 501 return (rc); 502 } 503 504 void 505 toe_connect_failed(struct toedev *tod, struct inpcb *inp, int err) 506 { 507 508 INP_WLOCK_ASSERT(inp); 509 510 if (!(inp->inp_flags & INP_DROPPED)) { 511 struct tcpcb *tp = intotcpcb(inp); 512 513 KASSERT(tp->t_flags & TF_TOE, 514 ("%s: tp %p not offloaded.", __func__, tp)); 515 516 if (err == EAGAIN) { 517 518 /* 519 * Temporary failure during offload, take this PCB back. 520 * Detach from the TOE driver and do the rest of what 521 * TCP's pru_connect would have done if the connection 522 * wasn't offloaded. 523 */ 524 525 tod->tod_pcb_detach(tod, tp); 526 KASSERT(!(tp->t_flags & TF_TOE), 527 ("%s: tp %p still offloaded.", __func__, tp)); 528 tcp_timer_activate(tp, TT_KEEP, TP_KEEPINIT(tp)); 529 (void) tp->t_fb->tfb_tcp_output(tp); 530 } else { 531 532 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 533 tp = tcp_drop(tp, err); 534 if (tp == NULL) 535 INP_WLOCK(inp); /* re-acquire */ 536 } 537 } 538 INP_WLOCK_ASSERT(inp); 539 } 540 541 static int 542 toecore_load(void) 543 { 544 545 mtx_init(&toedev_lock, "toedev lock", NULL, MTX_DEF); 546 TAILQ_INIT(&toedev_list); 547 548 listen_start_eh = EVENTHANDLER_REGISTER(tcp_offload_listen_start, 549 toe_listen_start_event, NULL, EVENTHANDLER_PRI_ANY); 550 listen_stop_eh = EVENTHANDLER_REGISTER(tcp_offload_listen_stop, 551 toe_listen_stop_event, NULL, EVENTHANDLER_PRI_ANY); 552 lle_event_eh = EVENTHANDLER_REGISTER(lle_event, toe_lle_event, NULL, 553 EVENTHANDLER_PRI_ANY); 554 555 return (0); 556 } 557 558 static int 559 toecore_unload(void) 560 { 561 562 mtx_lock(&toedev_lock); 563 if (!TAILQ_EMPTY(&toedev_list)) { 564 mtx_unlock(&toedev_lock); 565 return (EBUSY); 566 } 567 568 EVENTHANDLER_DEREGISTER(tcp_offload_listen_start, listen_start_eh); 569 EVENTHANDLER_DEREGISTER(tcp_offload_listen_stop, listen_stop_eh); 570 EVENTHANDLER_DEREGISTER(lle_event, lle_event_eh); 571 572 mtx_unlock(&toedev_lock); 573 mtx_destroy(&toedev_lock); 574 575 return (0); 576 } 577 578 static int 579 toecore_mod_handler(module_t mod, int cmd, void *arg) 580 { 581 582 if (cmd == MOD_LOAD) 583 return (toecore_load()); 584 585 if (cmd == MOD_UNLOAD) 586 return (toecore_unload()); 587 588 return (EOPNOTSUPP); 589 } 590 591 static moduledata_t mod_data= { 592 "toecore", 593 toecore_mod_handler, 594 0 595 }; 596 597 MODULE_VERSION(toecore, 1); 598 DECLARE_MODULE(toecore, mod_data, SI_SUB_EXEC, SI_ORDER_ANY); 599