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 #ifdef TCP_OFFLOAD 37 #include <sys/param.h> 38 #include <sys/systm.h> 39 #include <sys/kernel.h> 40 #include <sys/ktr.h> 41 #include <sys/module.h> 42 #include <sys/protosw.h> 43 #include <sys/domain.h> 44 #include <sys/socket.h> 45 #include <sys/socketvar.h> 46 #include <sys/sysctl.h> 47 #include <net/ethernet.h> 48 #include <net/if.h> 49 #include <net/if_types.h> 50 #include <net/if_vlan_var.h> 51 #include <net/route.h> 52 #include <netinet/in.h> 53 #include <netinet/in_pcb.h> 54 #include <netinet/ip.h> 55 #define TCPSTATES 56 #include <netinet/tcp_fsm.h> 57 #include <netinet/tcp_var.h> 58 #include <netinet/toecore.h> 59 #include <netinet/cc/cc.h> 60 61 #include "common/common.h" 62 #include "common/t4_msg.h" 63 #include "common/t4_regs.h" 64 #include "common/t4_regs_values.h" 65 #include "tom/t4_tom_l2t.h" 66 #include "tom/t4_tom.h" 67 68 /* atid services */ 69 static int alloc_atid(struct adapter *, void *); 70 static void *lookup_atid(struct adapter *, int); 71 static void free_atid(struct adapter *, int); 72 73 static int 74 alloc_atid(struct adapter *sc, void *ctx) 75 { 76 struct tid_info *t = &sc->tids; 77 int atid = -1; 78 79 mtx_lock(&t->atid_lock); 80 if (t->afree) { 81 union aopen_entry *p = t->afree; 82 83 atid = p - t->atid_tab; 84 t->afree = p->next; 85 p->data = ctx; 86 t->atids_in_use++; 87 } 88 mtx_unlock(&t->atid_lock); 89 return (atid); 90 } 91 92 static void * 93 lookup_atid(struct adapter *sc, int atid) 94 { 95 struct tid_info *t = &sc->tids; 96 97 return (t->atid_tab[atid].data); 98 } 99 100 static void 101 free_atid(struct adapter *sc, int atid) 102 { 103 struct tid_info *t = &sc->tids; 104 union aopen_entry *p = &t->atid_tab[atid]; 105 106 mtx_lock(&t->atid_lock); 107 p->next = t->afree; 108 t->afree = p; 109 t->atids_in_use--; 110 mtx_unlock(&t->atid_lock); 111 } 112 113 /* 114 * Active open succeeded. 115 */ 116 static int 117 do_act_establish(struct sge_iq *iq, const struct rss_header *rss, 118 struct mbuf *m) 119 { 120 struct adapter *sc = iq->adapter; 121 const struct cpl_act_establish *cpl = (const void *)(rss + 1); 122 u_int tid = GET_TID(cpl); 123 u_int atid = G_TID_TID(ntohl(cpl->tos_atid)); 124 struct toepcb *toep = lookup_atid(sc, atid); 125 struct inpcb *inp = toep->inp; 126 127 KASSERT(m == NULL, ("%s: wasn't expecting payload", __func__)); 128 KASSERT(toep->tid == atid, ("%s: toep tid/atid mismatch", __func__)); 129 130 CTR3(KTR_CXGBE, "%s: atid %u, tid %u", __func__, atid, tid); 131 free_atid(sc, atid); 132 133 CURVNET_SET(toep->vnet); 134 INP_WLOCK(inp); 135 toep->tid = tid; 136 insert_tid(sc, tid, toep, inp->inp_vflag & INP_IPV6 ? 2 : 1); 137 if (inp->inp_flags & INP_DROPPED) { 138 139 /* socket closed by the kernel before hw told us it connected */ 140 141 send_flowc_wr(toep, NULL); 142 send_reset(sc, toep, be32toh(cpl->snd_isn)); 143 goto done; 144 } 145 146 make_established(toep, cpl->snd_isn, cpl->rcv_isn, cpl->tcp_opt); 147 148 if (toep->ulp_mode == ULP_MODE_TLS) 149 tls_establish(toep); 150 151 done: 152 INP_WUNLOCK(inp); 153 CURVNET_RESTORE(); 154 return (0); 155 } 156 157 /* 158 * Convert an ACT_OPEN_RPL status to an errno. 159 */ 160 static inline int 161 act_open_rpl_status_to_errno(int status) 162 { 163 164 switch (status) { 165 case CPL_ERR_CONN_RESET: 166 return (ECONNREFUSED); 167 case CPL_ERR_ARP_MISS: 168 return (EHOSTUNREACH); 169 case CPL_ERR_CONN_TIMEDOUT: 170 return (ETIMEDOUT); 171 case CPL_ERR_TCAM_FULL: 172 return (EAGAIN); 173 case CPL_ERR_CONN_EXIST: 174 log(LOG_ERR, "ACTIVE_OPEN_RPL: 4-tuple in use\n"); 175 return (EAGAIN); 176 default: 177 return (EIO); 178 } 179 } 180 181 void 182 act_open_failure_cleanup(struct adapter *sc, u_int atid, u_int status) 183 { 184 struct toepcb *toep = lookup_atid(sc, atid); 185 struct inpcb *inp = toep->inp; 186 struct toedev *tod = &toep->td->tod; 187 188 free_atid(sc, atid); 189 toep->tid = -1; 190 191 CURVNET_SET(toep->vnet); 192 if (status != EAGAIN) 193 INP_INFO_RLOCK(&V_tcbinfo); 194 INP_WLOCK(inp); 195 toe_connect_failed(tod, inp, status); 196 final_cpl_received(toep); /* unlocks inp */ 197 if (status != EAGAIN) 198 INP_INFO_RUNLOCK(&V_tcbinfo); 199 CURVNET_RESTORE(); 200 } 201 202 /* 203 * Active open failed. 204 */ 205 static int 206 do_act_open_rpl(struct sge_iq *iq, const struct rss_header *rss, 207 struct mbuf *m) 208 { 209 struct adapter *sc = iq->adapter; 210 const struct cpl_act_open_rpl *cpl = (const void *)(rss + 1); 211 u_int atid = G_TID_TID(G_AOPEN_ATID(be32toh(cpl->atid_status))); 212 u_int status = G_AOPEN_STATUS(be32toh(cpl->atid_status)); 213 struct toepcb *toep = lookup_atid(sc, atid); 214 int rc; 215 216 KASSERT(m == NULL, ("%s: wasn't expecting payload", __func__)); 217 KASSERT(toep->tid == atid, ("%s: toep tid/atid mismatch", __func__)); 218 219 CTR3(KTR_CXGBE, "%s: atid %u, status %u ", __func__, atid, status); 220 221 /* Ignore negative advice */ 222 if (negative_advice(status)) 223 return (0); 224 225 if (status && act_open_has_tid(status)) 226 release_tid(sc, GET_TID(cpl), toep->ctrlq); 227 228 rc = act_open_rpl_status_to_errno(status); 229 act_open_failure_cleanup(sc, atid, rc); 230 231 return (0); 232 } 233 234 /* 235 * Options2 for active open. 236 */ 237 static uint32_t 238 calc_opt2a(struct socket *so, struct toepcb *toep, 239 const struct offload_settings *s) 240 { 241 struct tcpcb *tp = so_sototcpcb(so); 242 struct port_info *pi = toep->vi->pi; 243 struct adapter *sc = pi->adapter; 244 uint32_t opt2 = 0; 245 246 /* 247 * rx flow control, rx coalesce, congestion control, and tx pace are all 248 * explicitly set by the driver. On T5+ the ISS is also set by the 249 * driver to the value picked by the kernel. 250 */ 251 if (is_t4(sc)) { 252 opt2 |= F_RX_FC_VALID | F_RX_COALESCE_VALID; 253 opt2 |= F_CONG_CNTRL_VALID | F_PACE_VALID; 254 } else { 255 opt2 |= F_T5_OPT_2_VALID; /* all 4 valid */ 256 opt2 |= F_T5_ISS; /* ISS provided in CPL */ 257 } 258 259 if (s->sack > 0 || (s->sack < 0 && (tp->t_flags & TF_SACK_PERMIT))) 260 opt2 |= F_SACK_EN; 261 262 if (s->tstamp > 0 || (s->tstamp < 0 && (tp->t_flags & TF_REQ_TSTMP))) 263 opt2 |= F_TSTAMPS_EN; 264 265 if (tp->t_flags & TF_REQ_SCALE) 266 opt2 |= F_WND_SCALE_EN; 267 268 if (s->ecn > 0 || (s->ecn < 0 && V_tcp_do_ecn == 1)) 269 opt2 |= F_CCTRL_ECN; 270 271 /* XXX: F_RX_CHANNEL for multiple rx c-chan support goes here. */ 272 273 opt2 |= V_TX_QUEUE(sc->params.tp.tx_modq[pi->tx_chan]); 274 275 /* These defaults are subject to ULP specific fixups later. */ 276 opt2 |= V_RX_FC_DDP(0) | V_RX_FC_DISABLE(0); 277 278 opt2 |= V_PACE(0); 279 280 if (s->cong_algo >= 0) 281 opt2 |= V_CONG_CNTRL(s->cong_algo); 282 else if (sc->tt.cong_algorithm >= 0) 283 opt2 |= V_CONG_CNTRL(sc->tt.cong_algorithm & M_CONG_CNTRL); 284 else { 285 struct cc_algo *cc = CC_ALGO(tp); 286 287 if (strcasecmp(cc->name, "reno") == 0) 288 opt2 |= V_CONG_CNTRL(CONG_ALG_RENO); 289 else if (strcasecmp(cc->name, "tahoe") == 0) 290 opt2 |= V_CONG_CNTRL(CONG_ALG_TAHOE); 291 if (strcasecmp(cc->name, "newreno") == 0) 292 opt2 |= V_CONG_CNTRL(CONG_ALG_NEWRENO); 293 if (strcasecmp(cc->name, "highspeed") == 0) 294 opt2 |= V_CONG_CNTRL(CONG_ALG_HIGHSPEED); 295 else { 296 /* 297 * Use newreno in case the algorithm selected by the 298 * host stack is not supported by the hardware. 299 */ 300 opt2 |= V_CONG_CNTRL(CONG_ALG_NEWRENO); 301 } 302 } 303 304 if (s->rx_coalesce > 0 || (s->rx_coalesce < 0 && sc->tt.rx_coalesce)) 305 opt2 |= V_RX_COALESCE(M_RX_COALESCE); 306 307 /* Note that ofld_rxq is already set according to s->rxq. */ 308 opt2 |= F_RSS_QUEUE_VALID; 309 opt2 |= V_RSS_QUEUE(toep->ofld_rxq->iq.abs_id); 310 311 #ifdef USE_DDP_RX_FLOW_CONTROL 312 if (toep->ulp_mode == ULP_MODE_TCPDDP) 313 opt2 |= F_RX_FC_DDP; 314 #endif 315 316 if (toep->ulp_mode == ULP_MODE_TLS) { 317 opt2 &= ~V_RX_COALESCE(M_RX_COALESCE); 318 opt2 |= F_RX_FC_DISABLE; 319 } 320 321 return (htobe32(opt2)); 322 } 323 324 void 325 t4_init_connect_cpl_handlers(void) 326 { 327 328 t4_register_cpl_handler(CPL_ACT_ESTABLISH, do_act_establish); 329 t4_register_cpl_handler(CPL_ACT_OPEN_RPL, do_act_open_rpl); 330 } 331 332 void 333 t4_uninit_connect_cpl_handlers(void) 334 { 335 336 t4_register_cpl_handler(CPL_ACT_ESTABLISH, NULL); 337 t4_register_cpl_handler(CPL_ACT_OPEN_RPL, NULL); 338 } 339 340 #define DONT_OFFLOAD_ACTIVE_OPEN(x) do { \ 341 reason = __LINE__; \ 342 rc = (x); \ 343 goto failed; \ 344 } while (0) 345 346 static inline int 347 act_open_cpl_size(struct adapter *sc, int isipv6) 348 { 349 int idx; 350 static const int sz_table[3][2] = { 351 { 352 sizeof (struct cpl_act_open_req), 353 sizeof (struct cpl_act_open_req6) 354 }, 355 { 356 sizeof (struct cpl_t5_act_open_req), 357 sizeof (struct cpl_t5_act_open_req6) 358 }, 359 { 360 sizeof (struct cpl_t6_act_open_req), 361 sizeof (struct cpl_t6_act_open_req6) 362 }, 363 }; 364 365 MPASS(chip_id(sc) >= CHELSIO_T4); 366 idx = min(chip_id(sc) - CHELSIO_T4, 2); 367 368 return (sz_table[idx][!!isipv6]); 369 } 370 371 /* 372 * active open (soconnect). 373 * 374 * State of affairs on entry: 375 * soisconnecting (so_state |= SS_ISCONNECTING) 376 * tcbinfo not locked (This has changed - used to be WLOCKed) 377 * inp WLOCKed 378 * tp->t_state = TCPS_SYN_SENT 379 * rtalloc1, RT_UNLOCK on rt. 380 */ 381 int 382 t4_connect(struct toedev *tod, struct socket *so, struct rtentry *rt, 383 struct sockaddr *nam) 384 { 385 struct adapter *sc = tod->tod_softc; 386 struct tom_data *td = tod_td(tod); 387 struct toepcb *toep = NULL; 388 struct wrqe *wr = NULL; 389 struct ifnet *rt_ifp = rt->rt_ifp; 390 struct vi_info *vi; 391 int mtu_idx, rscale, qid_atid, rc, isipv6, txqid, rxqid; 392 struct inpcb *inp = sotoinpcb(so); 393 struct tcpcb *tp = intotcpcb(inp); 394 int reason; 395 struct offload_settings settings; 396 uint16_t vid = 0xffff; 397 398 INP_WLOCK_ASSERT(inp); 399 KASSERT(nam->sa_family == AF_INET || nam->sa_family == AF_INET6, 400 ("%s: dest addr %p has family %u", __func__, nam, nam->sa_family)); 401 402 if (rt_ifp->if_type == IFT_ETHER) 403 vi = rt_ifp->if_softc; 404 else if (rt_ifp->if_type == IFT_L2VLAN) { 405 struct ifnet *ifp = VLAN_COOKIE(rt_ifp); 406 407 vi = ifp->if_softc; 408 VLAN_TAG(ifp, &vid); 409 } else if (rt_ifp->if_type == IFT_IEEE8023ADLAG) 410 DONT_OFFLOAD_ACTIVE_OPEN(ENOSYS); /* XXX: implement lagg+TOE */ 411 else 412 DONT_OFFLOAD_ACTIVE_OPEN(ENOTSUP); 413 414 rw_rlock(&sc->policy_lock); 415 settings = *lookup_offload_policy(sc, OPEN_TYPE_ACTIVE, NULL, vid, inp); 416 rw_runlock(&sc->policy_lock); 417 if (!settings.offload) 418 DONT_OFFLOAD_ACTIVE_OPEN(EPERM); 419 420 if (settings.txq >= 0 && settings.txq < vi->nofldtxq) 421 txqid = settings.txq; 422 else 423 txqid = arc4random() % vi->nofldtxq; 424 txqid += vi->first_ofld_txq; 425 if (settings.rxq >= 0 && settings.rxq < vi->nofldrxq) 426 rxqid = settings.rxq; 427 else 428 rxqid = arc4random() % vi->nofldrxq; 429 rxqid += vi->first_ofld_rxq; 430 431 toep = alloc_toepcb(vi, txqid, rxqid, M_NOWAIT | M_ZERO); 432 if (toep == NULL) 433 DONT_OFFLOAD_ACTIVE_OPEN(ENOMEM); 434 435 toep->tid = alloc_atid(sc, toep); 436 if (toep->tid < 0) 437 DONT_OFFLOAD_ACTIVE_OPEN(ENOMEM); 438 439 toep->l2te = t4_l2t_get(vi->pi, rt_ifp, 440 rt->rt_flags & RTF_GATEWAY ? rt->rt_gateway : nam); 441 if (toep->l2te == NULL) 442 DONT_OFFLOAD_ACTIVE_OPEN(ENOMEM); 443 444 isipv6 = nam->sa_family == AF_INET6; 445 wr = alloc_wrqe(act_open_cpl_size(sc, isipv6), toep->ctrlq); 446 if (wr == NULL) 447 DONT_OFFLOAD_ACTIVE_OPEN(ENOMEM); 448 449 toep->vnet = so->so_vnet; 450 set_ulp_mode(toep, select_ulp_mode(so, sc, &settings)); 451 SOCKBUF_LOCK(&so->so_rcv); 452 /* opt0 rcv_bufsiz initially, assumes its normal meaning later */ 453 toep->rx_credits = min(select_rcv_wnd(so) >> 10, M_RCV_BUFSIZ); 454 SOCKBUF_UNLOCK(&so->so_rcv); 455 456 /* 457 * The kernel sets request_r_scale based on sb_max whereas we need to 458 * take hardware's MAX_RCV_WND into account too. This is normally a 459 * no-op as MAX_RCV_WND is much larger than the default sb_max. 460 */ 461 if (tp->t_flags & TF_REQ_SCALE) 462 rscale = tp->request_r_scale = select_rcv_wscale(); 463 else 464 rscale = 0; 465 mtu_idx = find_best_mtu_idx(sc, &inp->inp_inc, &settings); 466 qid_atid = (toep->ofld_rxq->iq.abs_id << 14) | toep->tid; 467 468 if (isipv6) { 469 struct cpl_act_open_req6 *cpl = wrtod(wr); 470 struct cpl_t5_act_open_req6 *cpl5 = (void *)cpl; 471 struct cpl_t6_act_open_req6 *cpl6 = (void *)cpl; 472 473 if ((inp->inp_vflag & INP_IPV6) == 0) 474 DONT_OFFLOAD_ACTIVE_OPEN(ENOTSUP); 475 476 toep->ce = hold_lip(td, &inp->in6p_laddr, NULL); 477 if (toep->ce == NULL) 478 DONT_OFFLOAD_ACTIVE_OPEN(ENOENT); 479 480 switch (chip_id(sc)) { 481 case CHELSIO_T4: 482 INIT_TP_WR(cpl, 0); 483 cpl->params = select_ntuple(vi, toep->l2te); 484 break; 485 case CHELSIO_T5: 486 INIT_TP_WR(cpl5, 0); 487 cpl5->iss = htobe32(tp->iss); 488 cpl5->params = select_ntuple(vi, toep->l2te); 489 break; 490 case CHELSIO_T6: 491 default: 492 INIT_TP_WR(cpl6, 0); 493 cpl6->iss = htobe32(tp->iss); 494 cpl6->params = select_ntuple(vi, toep->l2te); 495 break; 496 } 497 OPCODE_TID(cpl) = htobe32(MK_OPCODE_TID(CPL_ACT_OPEN_REQ6, 498 qid_atid)); 499 cpl->local_port = inp->inp_lport; 500 cpl->local_ip_hi = *(uint64_t *)&inp->in6p_laddr.s6_addr[0]; 501 cpl->local_ip_lo = *(uint64_t *)&inp->in6p_laddr.s6_addr[8]; 502 cpl->peer_port = inp->inp_fport; 503 cpl->peer_ip_hi = *(uint64_t *)&inp->in6p_faddr.s6_addr[0]; 504 cpl->peer_ip_lo = *(uint64_t *)&inp->in6p_faddr.s6_addr[8]; 505 cpl->opt0 = calc_opt0(so, vi, toep->l2te, mtu_idx, rscale, 506 toep->rx_credits, toep->ulp_mode, &settings); 507 cpl->opt2 = calc_opt2a(so, toep, &settings); 508 } else { 509 struct cpl_act_open_req *cpl = wrtod(wr); 510 struct cpl_t5_act_open_req *cpl5 = (void *)cpl; 511 struct cpl_t6_act_open_req *cpl6 = (void *)cpl; 512 513 switch (chip_id(sc)) { 514 case CHELSIO_T4: 515 INIT_TP_WR(cpl, 0); 516 cpl->params = select_ntuple(vi, toep->l2te); 517 break; 518 case CHELSIO_T5: 519 INIT_TP_WR(cpl5, 0); 520 cpl5->iss = htobe32(tp->iss); 521 cpl5->params = select_ntuple(vi, toep->l2te); 522 break; 523 case CHELSIO_T6: 524 default: 525 INIT_TP_WR(cpl6, 0); 526 cpl6->iss = htobe32(tp->iss); 527 cpl6->params = select_ntuple(vi, toep->l2te); 528 break; 529 } 530 OPCODE_TID(cpl) = htobe32(MK_OPCODE_TID(CPL_ACT_OPEN_REQ, 531 qid_atid)); 532 inp_4tuple_get(inp, &cpl->local_ip, &cpl->local_port, 533 &cpl->peer_ip, &cpl->peer_port); 534 cpl->opt0 = calc_opt0(so, vi, toep->l2te, mtu_idx, rscale, 535 toep->rx_credits, toep->ulp_mode, &settings); 536 cpl->opt2 = calc_opt2a(so, toep, &settings); 537 } 538 539 CTR5(KTR_CXGBE, "%s: atid %u (%s), toep %p, inp %p", __func__, 540 toep->tid, tcpstates[tp->t_state], toep, inp); 541 542 offload_socket(so, toep); 543 rc = t4_l2t_send(sc, wr, toep->l2te); 544 if (rc == 0) { 545 toep->flags |= TPF_CPL_PENDING; 546 return (0); 547 } 548 549 undo_offload_socket(so); 550 reason = __LINE__; 551 failed: 552 CTR3(KTR_CXGBE, "%s: not offloading (%d), rc %d", __func__, reason, rc); 553 554 if (wr) 555 free_wrqe(wr); 556 557 if (toep) { 558 if (toep->tid >= 0) 559 free_atid(sc, toep->tid); 560 if (toep->l2te) 561 t4_l2t_release(toep->l2te); 562 if (toep->ce) 563 release_lip(td, toep->ce); 564 free_toepcb(toep); 565 } 566 567 return (rc); 568 } 569 #endif 570