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 "t4_clip.h" 66 #include "tom/t4_tom_l2t.h" 67 #include "tom/t4_tom.h" 68 69 /* 70 * Active open succeeded. 71 */ 72 static int 73 do_act_establish(struct sge_iq *iq, const struct rss_header *rss, 74 struct mbuf *m) 75 { 76 struct adapter *sc = iq->adapter; 77 const struct cpl_act_establish *cpl = (const void *)(rss + 1); 78 u_int tid = GET_TID(cpl); 79 u_int atid = G_TID_TID(ntohl(cpl->tos_atid)); 80 struct toepcb *toep = lookup_atid(sc, atid); 81 struct inpcb *inp = toep->inp; 82 83 KASSERT(m == NULL, ("%s: wasn't expecting payload", __func__)); 84 KASSERT(toep->tid == atid, ("%s: toep tid/atid mismatch", __func__)); 85 86 CTR3(KTR_CXGBE, "%s: atid %u, tid %u", __func__, atid, tid); 87 free_atid(sc, atid); 88 89 CURVNET_SET(toep->vnet); 90 INP_WLOCK(inp); 91 toep->tid = tid; 92 insert_tid(sc, tid, toep, inp->inp_vflag & INP_IPV6 ? 2 : 1); 93 if (inp->inp_flags & INP_DROPPED) { 94 95 /* socket closed by the kernel before hw told us it connected */ 96 97 send_flowc_wr(toep, NULL); 98 send_reset(sc, toep, be32toh(cpl->snd_isn)); 99 goto done; 100 } 101 102 make_established(toep, be32toh(cpl->snd_isn) - 1, 103 be32toh(cpl->rcv_isn) - 1, cpl->tcp_opt); 104 105 if (toep->ulp_mode == ULP_MODE_TLS) 106 tls_establish(toep); 107 108 done: 109 INP_WUNLOCK(inp); 110 CURVNET_RESTORE(); 111 return (0); 112 } 113 114 void 115 act_open_failure_cleanup(struct adapter *sc, u_int atid, u_int status) 116 { 117 struct toepcb *toep = lookup_atid(sc, atid); 118 struct inpcb *inp = toep->inp; 119 struct toedev *tod = &toep->td->tod; 120 struct epoch_tracker et; 121 122 free_atid(sc, atid); 123 toep->tid = -1; 124 125 CURVNET_SET(toep->vnet); 126 if (status != EAGAIN) 127 INP_INFO_RLOCK_ET(&V_tcbinfo, et); 128 INP_WLOCK(inp); 129 toe_connect_failed(tod, inp, status); 130 final_cpl_received(toep); /* unlocks inp */ 131 if (status != EAGAIN) 132 INP_INFO_RUNLOCK_ET(&V_tcbinfo, et); 133 CURVNET_RESTORE(); 134 } 135 136 /* 137 * Active open failed. 138 */ 139 static int 140 do_act_open_rpl(struct sge_iq *iq, const struct rss_header *rss, 141 struct mbuf *m) 142 { 143 struct adapter *sc = iq->adapter; 144 const struct cpl_act_open_rpl *cpl = (const void *)(rss + 1); 145 u_int atid = G_TID_TID(G_AOPEN_ATID(be32toh(cpl->atid_status))); 146 u_int status = G_AOPEN_STATUS(be32toh(cpl->atid_status)); 147 struct toepcb *toep = lookup_atid(sc, atid); 148 int rc; 149 150 KASSERT(m == NULL, ("%s: wasn't expecting payload", __func__)); 151 KASSERT(toep->tid == atid, ("%s: toep tid/atid mismatch", __func__)); 152 153 CTR3(KTR_CXGBE, "%s: atid %u, status %u ", __func__, atid, status); 154 155 /* Ignore negative advice */ 156 if (negative_advice(status)) 157 return (0); 158 159 if (status && act_open_has_tid(status)) 160 release_tid(sc, GET_TID(cpl), toep->ctrlq); 161 162 rc = act_open_rpl_status_to_errno(status); 163 act_open_failure_cleanup(sc, atid, rc); 164 165 return (0); 166 } 167 168 /* 169 * Options2 for active open. 170 */ 171 static uint32_t 172 calc_opt2a(struct socket *so, struct toepcb *toep, 173 const struct offload_settings *s) 174 { 175 struct tcpcb *tp = so_sototcpcb(so); 176 struct port_info *pi = toep->vi->pi; 177 struct adapter *sc = pi->adapter; 178 uint32_t opt2 = 0; 179 180 /* 181 * rx flow control, rx coalesce, congestion control, and tx pace are all 182 * explicitly set by the driver. On T5+ the ISS is also set by the 183 * driver to the value picked by the kernel. 184 */ 185 if (is_t4(sc)) { 186 opt2 |= F_RX_FC_VALID | F_RX_COALESCE_VALID; 187 opt2 |= F_CONG_CNTRL_VALID | F_PACE_VALID; 188 } else { 189 opt2 |= F_T5_OPT_2_VALID; /* all 4 valid */ 190 opt2 |= F_T5_ISS; /* ISS provided in CPL */ 191 } 192 193 if (s->sack > 0 || (s->sack < 0 && (tp->t_flags & TF_SACK_PERMIT))) 194 opt2 |= F_SACK_EN; 195 196 if (s->tstamp > 0 || (s->tstamp < 0 && (tp->t_flags & TF_REQ_TSTMP))) 197 opt2 |= F_TSTAMPS_EN; 198 199 if (tp->t_flags & TF_REQ_SCALE) 200 opt2 |= F_WND_SCALE_EN; 201 202 if (s->ecn > 0 || (s->ecn < 0 && V_tcp_do_ecn == 1)) 203 opt2 |= F_CCTRL_ECN; 204 205 /* XXX: F_RX_CHANNEL for multiple rx c-chan support goes here. */ 206 207 opt2 |= V_TX_QUEUE(sc->params.tp.tx_modq[pi->tx_chan]); 208 209 /* These defaults are subject to ULP specific fixups later. */ 210 opt2 |= V_RX_FC_DDP(0) | V_RX_FC_DISABLE(0); 211 212 opt2 |= V_PACE(0); 213 214 if (s->cong_algo >= 0) 215 opt2 |= V_CONG_CNTRL(s->cong_algo); 216 else if (sc->tt.cong_algorithm >= 0) 217 opt2 |= V_CONG_CNTRL(sc->tt.cong_algorithm & M_CONG_CNTRL); 218 else { 219 struct cc_algo *cc = CC_ALGO(tp); 220 221 if (strcasecmp(cc->name, "reno") == 0) 222 opt2 |= V_CONG_CNTRL(CONG_ALG_RENO); 223 else if (strcasecmp(cc->name, "tahoe") == 0) 224 opt2 |= V_CONG_CNTRL(CONG_ALG_TAHOE); 225 if (strcasecmp(cc->name, "newreno") == 0) 226 opt2 |= V_CONG_CNTRL(CONG_ALG_NEWRENO); 227 if (strcasecmp(cc->name, "highspeed") == 0) 228 opt2 |= V_CONG_CNTRL(CONG_ALG_HIGHSPEED); 229 else { 230 /* 231 * Use newreno in case the algorithm selected by the 232 * host stack is not supported by the hardware. 233 */ 234 opt2 |= V_CONG_CNTRL(CONG_ALG_NEWRENO); 235 } 236 } 237 238 if (s->rx_coalesce > 0 || (s->rx_coalesce < 0 && sc->tt.rx_coalesce)) 239 opt2 |= V_RX_COALESCE(M_RX_COALESCE); 240 241 /* Note that ofld_rxq is already set according to s->rxq. */ 242 opt2 |= F_RSS_QUEUE_VALID; 243 opt2 |= V_RSS_QUEUE(toep->ofld_rxq->iq.abs_id); 244 245 #ifdef USE_DDP_RX_FLOW_CONTROL 246 if (toep->ulp_mode == ULP_MODE_TCPDDP) 247 opt2 |= F_RX_FC_DDP; 248 #endif 249 250 if (toep->ulp_mode == ULP_MODE_TLS) { 251 opt2 &= ~V_RX_COALESCE(M_RX_COALESCE); 252 opt2 |= F_RX_FC_DISABLE; 253 } 254 255 return (htobe32(opt2)); 256 } 257 258 void 259 t4_init_connect_cpl_handlers(void) 260 { 261 262 t4_register_cpl_handler(CPL_ACT_ESTABLISH, do_act_establish); 263 t4_register_shared_cpl_handler(CPL_ACT_OPEN_RPL, do_act_open_rpl, 264 CPL_COOKIE_TOM); 265 } 266 267 void 268 t4_uninit_connect_cpl_handlers(void) 269 { 270 271 t4_register_cpl_handler(CPL_ACT_ESTABLISH, NULL); 272 t4_register_shared_cpl_handler(CPL_ACT_OPEN_RPL, NULL, CPL_COOKIE_TOM); 273 } 274 275 #define DONT_OFFLOAD_ACTIVE_OPEN(x) do { \ 276 reason = __LINE__; \ 277 rc = (x); \ 278 goto failed; \ 279 } while (0) 280 281 static inline int 282 act_open_cpl_size(struct adapter *sc, int isipv6) 283 { 284 int idx; 285 static const int sz_table[3][2] = { 286 { 287 sizeof (struct cpl_act_open_req), 288 sizeof (struct cpl_act_open_req6) 289 }, 290 { 291 sizeof (struct cpl_t5_act_open_req), 292 sizeof (struct cpl_t5_act_open_req6) 293 }, 294 { 295 sizeof (struct cpl_t6_act_open_req), 296 sizeof (struct cpl_t6_act_open_req6) 297 }, 298 }; 299 300 MPASS(chip_id(sc) >= CHELSIO_T4); 301 idx = min(chip_id(sc) - CHELSIO_T4, 2); 302 303 return (sz_table[idx][!!isipv6]); 304 } 305 306 /* 307 * active open (soconnect). 308 * 309 * State of affairs on entry: 310 * soisconnecting (so_state |= SS_ISCONNECTING) 311 * tcbinfo not locked (This has changed - used to be WLOCKed) 312 * inp WLOCKed 313 * tp->t_state = TCPS_SYN_SENT 314 * rtalloc1, RT_UNLOCK on rt. 315 */ 316 int 317 t4_connect(struct toedev *tod, struct socket *so, struct rtentry *rt, 318 struct sockaddr *nam) 319 { 320 struct adapter *sc = tod->tod_softc; 321 struct toepcb *toep = NULL; 322 struct wrqe *wr = NULL; 323 struct ifnet *rt_ifp = rt->rt_ifp; 324 struct vi_info *vi; 325 int mtu_idx, rscale, qid_atid, rc, isipv6, txqid, rxqid; 326 struct inpcb *inp = sotoinpcb(so); 327 struct tcpcb *tp = intotcpcb(inp); 328 int reason; 329 struct offload_settings settings; 330 uint16_t vid = 0xfff, pcp = 0; 331 332 INP_WLOCK_ASSERT(inp); 333 KASSERT(nam->sa_family == AF_INET || nam->sa_family == AF_INET6, 334 ("%s: dest addr %p has family %u", __func__, nam, nam->sa_family)); 335 336 if (rt_ifp->if_type == IFT_ETHER) 337 vi = rt_ifp->if_softc; 338 else if (rt_ifp->if_type == IFT_L2VLAN) { 339 struct ifnet *ifp = VLAN_TRUNKDEV(rt_ifp); 340 341 vi = ifp->if_softc; 342 VLAN_TAG(rt_ifp, &vid); 343 VLAN_PCP(rt_ifp, &pcp); 344 } else if (rt_ifp->if_type == IFT_IEEE8023ADLAG) 345 DONT_OFFLOAD_ACTIVE_OPEN(ENOSYS); /* XXX: implement lagg+TOE */ 346 else 347 DONT_OFFLOAD_ACTIVE_OPEN(ENOTSUP); 348 349 rw_rlock(&sc->policy_lock); 350 settings = *lookup_offload_policy(sc, OPEN_TYPE_ACTIVE, NULL, 351 EVL_MAKETAG(vid, pcp, 0), inp); 352 rw_runlock(&sc->policy_lock); 353 if (!settings.offload) 354 DONT_OFFLOAD_ACTIVE_OPEN(EPERM); 355 356 if (settings.txq >= 0 && settings.txq < vi->nofldtxq) 357 txqid = settings.txq; 358 else 359 txqid = arc4random() % vi->nofldtxq; 360 txqid += vi->first_ofld_txq; 361 if (settings.rxq >= 0 && settings.rxq < vi->nofldrxq) 362 rxqid = settings.rxq; 363 else 364 rxqid = arc4random() % vi->nofldrxq; 365 rxqid += vi->first_ofld_rxq; 366 367 toep = alloc_toepcb(vi, txqid, rxqid, M_NOWAIT | M_ZERO); 368 if (toep == NULL) 369 DONT_OFFLOAD_ACTIVE_OPEN(ENOMEM); 370 371 toep->tid = alloc_atid(sc, toep); 372 if (toep->tid < 0) 373 DONT_OFFLOAD_ACTIVE_OPEN(ENOMEM); 374 375 toep->l2te = t4_l2t_get(vi->pi, rt_ifp, 376 rt->rt_flags & RTF_GATEWAY ? rt->rt_gateway : nam); 377 if (toep->l2te == NULL) 378 DONT_OFFLOAD_ACTIVE_OPEN(ENOMEM); 379 380 isipv6 = nam->sa_family == AF_INET6; 381 wr = alloc_wrqe(act_open_cpl_size(sc, isipv6), toep->ctrlq); 382 if (wr == NULL) 383 DONT_OFFLOAD_ACTIVE_OPEN(ENOMEM); 384 385 toep->vnet = so->so_vnet; 386 set_ulp_mode(toep, select_ulp_mode(so, sc, &settings)); 387 SOCKBUF_LOCK(&so->so_rcv); 388 /* opt0 rcv_bufsiz initially, assumes its normal meaning later */ 389 toep->rx_credits = min(select_rcv_wnd(so) >> 10, M_RCV_BUFSIZ); 390 SOCKBUF_UNLOCK(&so->so_rcv); 391 392 /* 393 * The kernel sets request_r_scale based on sb_max whereas we need to 394 * take hardware's MAX_RCV_WND into account too. This is normally a 395 * no-op as MAX_RCV_WND is much larger than the default sb_max. 396 */ 397 if (tp->t_flags & TF_REQ_SCALE) 398 rscale = tp->request_r_scale = select_rcv_wscale(); 399 else 400 rscale = 0; 401 mtu_idx = find_best_mtu_idx(sc, &inp->inp_inc, &settings); 402 qid_atid = V_TID_QID(toep->ofld_rxq->iq.abs_id) | V_TID_TID(toep->tid) | 403 V_TID_COOKIE(CPL_COOKIE_TOM); 404 405 if (isipv6) { 406 struct cpl_act_open_req6 *cpl = wrtod(wr); 407 struct cpl_t5_act_open_req6 *cpl5 = (void *)cpl; 408 struct cpl_t6_act_open_req6 *cpl6 = (void *)cpl; 409 410 if ((inp->inp_vflag & INP_IPV6) == 0) 411 DONT_OFFLOAD_ACTIVE_OPEN(ENOTSUP); 412 413 toep->ce = t4_hold_lip(sc, &inp->in6p_laddr, NULL); 414 if (toep->ce == NULL) 415 DONT_OFFLOAD_ACTIVE_OPEN(ENOENT); 416 417 switch (chip_id(sc)) { 418 case CHELSIO_T4: 419 INIT_TP_WR(cpl, 0); 420 cpl->params = select_ntuple(vi, toep->l2te); 421 break; 422 case CHELSIO_T5: 423 INIT_TP_WR(cpl5, 0); 424 cpl5->iss = htobe32(tp->iss); 425 cpl5->params = select_ntuple(vi, toep->l2te); 426 break; 427 case CHELSIO_T6: 428 default: 429 INIT_TP_WR(cpl6, 0); 430 cpl6->iss = htobe32(tp->iss); 431 cpl6->params = select_ntuple(vi, toep->l2te); 432 break; 433 } 434 OPCODE_TID(cpl) = htobe32(MK_OPCODE_TID(CPL_ACT_OPEN_REQ6, 435 qid_atid)); 436 cpl->local_port = inp->inp_lport; 437 cpl->local_ip_hi = *(uint64_t *)&inp->in6p_laddr.s6_addr[0]; 438 cpl->local_ip_lo = *(uint64_t *)&inp->in6p_laddr.s6_addr[8]; 439 cpl->peer_port = inp->inp_fport; 440 cpl->peer_ip_hi = *(uint64_t *)&inp->in6p_faddr.s6_addr[0]; 441 cpl->peer_ip_lo = *(uint64_t *)&inp->in6p_faddr.s6_addr[8]; 442 cpl->opt0 = calc_opt0(so, vi, toep->l2te, mtu_idx, rscale, 443 toep->rx_credits, toep->ulp_mode, &settings); 444 cpl->opt2 = calc_opt2a(so, toep, &settings); 445 } else { 446 struct cpl_act_open_req *cpl = wrtod(wr); 447 struct cpl_t5_act_open_req *cpl5 = (void *)cpl; 448 struct cpl_t6_act_open_req *cpl6 = (void *)cpl; 449 450 switch (chip_id(sc)) { 451 case CHELSIO_T4: 452 INIT_TP_WR(cpl, 0); 453 cpl->params = select_ntuple(vi, toep->l2te); 454 break; 455 case CHELSIO_T5: 456 INIT_TP_WR(cpl5, 0); 457 cpl5->iss = htobe32(tp->iss); 458 cpl5->params = select_ntuple(vi, toep->l2te); 459 break; 460 case CHELSIO_T6: 461 default: 462 INIT_TP_WR(cpl6, 0); 463 cpl6->iss = htobe32(tp->iss); 464 cpl6->params = select_ntuple(vi, toep->l2te); 465 break; 466 } 467 OPCODE_TID(cpl) = htobe32(MK_OPCODE_TID(CPL_ACT_OPEN_REQ, 468 qid_atid)); 469 inp_4tuple_get(inp, &cpl->local_ip, &cpl->local_port, 470 &cpl->peer_ip, &cpl->peer_port); 471 cpl->opt0 = calc_opt0(so, vi, toep->l2te, mtu_idx, rscale, 472 toep->rx_credits, toep->ulp_mode, &settings); 473 cpl->opt2 = calc_opt2a(so, toep, &settings); 474 } 475 476 CTR5(KTR_CXGBE, "%s: atid %u (%s), toep %p, inp %p", __func__, 477 toep->tid, tcpstates[tp->t_state], toep, inp); 478 479 offload_socket(so, toep); 480 rc = t4_l2t_send(sc, wr, toep->l2te); 481 if (rc == 0) { 482 toep->flags |= TPF_CPL_PENDING; 483 return (0); 484 } 485 486 undo_offload_socket(so); 487 reason = __LINE__; 488 failed: 489 CTR3(KTR_CXGBE, "%s: not offloading (%d), rc %d", __func__, reason, rc); 490 491 if (wr) 492 free_wrqe(wr); 493 494 if (toep) { 495 if (toep->tid >= 0) 496 free_atid(sc, toep->tid); 497 if (toep->l2te) 498 t4_l2t_release(toep->l2te); 499 if (toep->ce) 500 t4_release_lip(sc, toep->ce); 501 free_toepcb(toep); 502 } 503 504 return (rc); 505 } 506 #endif 507