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