t4_connect.c revision 346805
1/*- 2 * Copyright (c) 2012 Chelsio Communications, Inc. 3 * All rights reserved. 4 * Written by: Navdeep Parhar <np@FreeBSD.org> 5 * 6 * Redistribution and use in source and binary forms, with or without 7 * modification, are permitted provided that the following conditions 8 * are met: 9 * 1. Redistributions of source code must retain the above copyright 10 * notice, this list of conditions and the following disclaimer. 11 * 2. Redistributions in binary form must reproduce the above copyright 12 * notice, this list of conditions and the following disclaimer in the 13 * documentation and/or other materials provided with the distribution. 14 * 15 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND 16 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 17 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 18 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE 19 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 20 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 21 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 22 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 23 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 24 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 25 * SUCH DAMAGE. 26 */ 27 28#include <sys/cdefs.h> 29__FBSDID("$FreeBSD: stable/11/sys/dev/cxgbe/tom/t4_connect.c 346805 2019-04-28 06:51:59Z np $"); 30 31#include "opt_inet.h" 32#include "opt_inet6.h" 33 34#ifdef TCP_OFFLOAD 35#include <sys/param.h> 36#include <sys/systm.h> 37#include <sys/kernel.h> 38#include <sys/ktr.h> 39#include <sys/module.h> 40#include <sys/protosw.h> 41#include <sys/domain.h> 42#include <sys/socket.h> 43#include <sys/socketvar.h> 44#include <sys/sysctl.h> 45#include <net/ethernet.h> 46#include <net/if.h> 47#include <net/if_types.h> 48#include <net/if_vlan_var.h> 49#include <net/route.h> 50#include <netinet/in.h> 51#include <netinet/in_pcb.h> 52#include <netinet/ip.h> 53#define TCPSTATES 54#include <netinet/tcp_fsm.h> 55#include <netinet/tcp_var.h> 56#include <netinet/toecore.h> 57#include <netinet/cc/cc.h> 58 59#include "common/common.h" 60#include "common/t4_msg.h" 61#include "common/t4_regs.h" 62#include "common/t4_regs_values.h" 63#include "tom/t4_tom_l2t.h" 64#include "tom/t4_tom.h" 65 66/* atid services */ 67static int alloc_atid(struct adapter *, void *); 68static void *lookup_atid(struct adapter *, int); 69static void free_atid(struct adapter *, int); 70 71static int 72alloc_atid(struct adapter *sc, void *ctx) 73{ 74 struct tid_info *t = &sc->tids; 75 int atid = -1; 76 77 mtx_lock(&t->atid_lock); 78 if (t->afree) { 79 union aopen_entry *p = t->afree; 80 81 atid = p - t->atid_tab; 82 t->afree = p->next; 83 p->data = ctx; 84 t->atids_in_use++; 85 } 86 mtx_unlock(&t->atid_lock); 87 return (atid); 88} 89 90static void * 91lookup_atid(struct adapter *sc, int atid) 92{ 93 struct tid_info *t = &sc->tids; 94 95 return (t->atid_tab[atid].data); 96} 97 98static void 99free_atid(struct adapter *sc, int atid) 100{ 101 struct tid_info *t = &sc->tids; 102 union aopen_entry *p = &t->atid_tab[atid]; 103 104 mtx_lock(&t->atid_lock); 105 p->next = t->afree; 106 t->afree = p; 107 t->atids_in_use--; 108 mtx_unlock(&t->atid_lock); 109} 110 111/* 112 * Active open succeeded. 113 */ 114static int 115do_act_establish(struct sge_iq *iq, const struct rss_header *rss, 116 struct mbuf *m) 117{ 118 struct adapter *sc = iq->adapter; 119 const struct cpl_act_establish *cpl = (const void *)(rss + 1); 120 u_int tid = GET_TID(cpl); 121 u_int atid = G_TID_TID(ntohl(cpl->tos_atid)); 122 struct toepcb *toep = lookup_atid(sc, atid); 123 struct inpcb *inp = toep->inp; 124 125 KASSERT(m == NULL, ("%s: wasn't expecting payload", __func__)); 126 KASSERT(toep->tid == atid, ("%s: toep tid/atid mismatch", __func__)); 127 128 CTR3(KTR_CXGBE, "%s: atid %u, tid %u", __func__, atid, tid); 129 free_atid(sc, atid); 130 131 CURVNET_SET(toep->vnet); 132 INP_WLOCK(inp); 133 toep->tid = tid; 134 insert_tid(sc, tid, toep, inp->inp_vflag & INP_IPV6 ? 2 : 1); 135 if (inp->inp_flags & INP_DROPPED) { 136 137 /* socket closed by the kernel before hw told us it connected */ 138 139 send_flowc_wr(toep, NULL); 140 send_reset(sc, toep, be32toh(cpl->snd_isn)); 141 goto done; 142 } 143 144 make_established(toep, cpl->snd_isn, cpl->rcv_isn, cpl->tcp_opt); 145 146 if (toep->ulp_mode == ULP_MODE_TLS) 147 tls_establish(toep); 148 149done: 150 INP_WUNLOCK(inp); 151 CURVNET_RESTORE(); 152 return (0); 153} 154 155/* 156 * Convert an ACT_OPEN_RPL status to an errno. 157 */ 158static inline int 159act_open_rpl_status_to_errno(int status) 160{ 161 162 switch (status) { 163 case CPL_ERR_CONN_RESET: 164 return (ECONNREFUSED); 165 case CPL_ERR_ARP_MISS: 166 return (EHOSTUNREACH); 167 case CPL_ERR_CONN_TIMEDOUT: 168 return (ETIMEDOUT); 169 case CPL_ERR_TCAM_FULL: 170 return (EAGAIN); 171 case CPL_ERR_CONN_EXIST: 172 log(LOG_ERR, "ACTIVE_OPEN_RPL: 4-tuple in use\n"); 173 return (EAGAIN); 174 default: 175 return (EIO); 176 } 177} 178 179void 180act_open_failure_cleanup(struct adapter *sc, u_int atid, u_int status) 181{ 182 struct toepcb *toep = lookup_atid(sc, atid); 183 struct inpcb *inp = toep->inp; 184 struct toedev *tod = &toep->td->tod; 185 186 free_atid(sc, atid); 187 toep->tid = -1; 188 189 CURVNET_SET(toep->vnet); 190 if (status != EAGAIN) 191 INP_INFO_RLOCK(&V_tcbinfo); 192 INP_WLOCK(inp); 193 toe_connect_failed(tod, inp, status); 194 final_cpl_received(toep); /* unlocks inp */ 195 if (status != EAGAIN) 196 INP_INFO_RUNLOCK(&V_tcbinfo); 197 CURVNET_RESTORE(); 198} 199 200/* 201 * Active open failed. 202 */ 203static int 204do_act_open_rpl(struct sge_iq *iq, const struct rss_header *rss, 205 struct mbuf *m) 206{ 207 struct adapter *sc = iq->adapter; 208 const struct cpl_act_open_rpl *cpl = (const void *)(rss + 1); 209 u_int atid = G_TID_TID(G_AOPEN_ATID(be32toh(cpl->atid_status))); 210 u_int status = G_AOPEN_STATUS(be32toh(cpl->atid_status)); 211 struct toepcb *toep = lookup_atid(sc, atid); 212 int rc; 213 214 KASSERT(m == NULL, ("%s: wasn't expecting payload", __func__)); 215 KASSERT(toep->tid == atid, ("%s: toep tid/atid mismatch", __func__)); 216 217 CTR3(KTR_CXGBE, "%s: atid %u, status %u ", __func__, atid, status); 218 219 /* Ignore negative advice */ 220 if (negative_advice(status)) 221 return (0); 222 223 if (status && act_open_has_tid(status)) 224 release_tid(sc, GET_TID(cpl), toep->ctrlq); 225 226 rc = act_open_rpl_status_to_errno(status); 227 act_open_failure_cleanup(sc, atid, rc); 228 229 return (0); 230} 231 232/* 233 * Options2 for active open. 234 */ 235static uint32_t 236calc_opt2a(struct socket *so, struct toepcb *toep, 237 const struct offload_settings *s) 238{ 239 struct tcpcb *tp = so_sototcpcb(so); 240 struct port_info *pi = toep->vi->pi; 241 struct adapter *sc = pi->adapter; 242 uint32_t opt2 = 0; 243 244 /* 245 * rx flow control, rx coalesce, congestion control, and tx pace are all 246 * explicitly set by the driver. On T5+ the ISS is also set by the 247 * driver to the value picked by the kernel. 248 */ 249 if (is_t4(sc)) { 250 opt2 |= F_RX_FC_VALID | F_RX_COALESCE_VALID; 251 opt2 |= F_CONG_CNTRL_VALID | F_PACE_VALID; 252 } else { 253 opt2 |= F_T5_OPT_2_VALID; /* all 4 valid */ 254 opt2 |= F_T5_ISS; /* ISS provided in CPL */ 255 } 256 257 if (s->sack > 0 || (s->sack < 0 && (tp->t_flags & TF_SACK_PERMIT))) 258 opt2 |= F_SACK_EN; 259 260 if (s->tstamp > 0 || (s->tstamp < 0 && (tp->t_flags & TF_REQ_TSTMP))) 261 opt2 |= F_TSTAMPS_EN; 262 263 if (tp->t_flags & TF_REQ_SCALE) 264 opt2 |= F_WND_SCALE_EN; 265 266 if (s->ecn > 0 || (s->ecn < 0 && V_tcp_do_ecn == 1)) 267 opt2 |= F_CCTRL_ECN; 268 269 /* XXX: F_RX_CHANNEL for multiple rx c-chan support goes here. */ 270 271 opt2 |= V_TX_QUEUE(sc->params.tp.tx_modq[pi->tx_chan]); 272 273 /* These defaults are subject to ULP specific fixups later. */ 274 opt2 |= V_RX_FC_DDP(0) | V_RX_FC_DISABLE(0); 275 276 opt2 |= V_PACE(0); 277 278 if (s->cong_algo >= 0) 279 opt2 |= V_CONG_CNTRL(s->cong_algo); 280 else if (sc->tt.cong_algorithm >= 0) 281 opt2 |= V_CONG_CNTRL(sc->tt.cong_algorithm & M_CONG_CNTRL); 282 else { 283 struct cc_algo *cc = CC_ALGO(tp); 284 285 if (strcasecmp(cc->name, "reno") == 0) 286 opt2 |= V_CONG_CNTRL(CONG_ALG_RENO); 287 else if (strcasecmp(cc->name, "tahoe") == 0) 288 opt2 |= V_CONG_CNTRL(CONG_ALG_TAHOE); 289 if (strcasecmp(cc->name, "newreno") == 0) 290 opt2 |= V_CONG_CNTRL(CONG_ALG_NEWRENO); 291 if (strcasecmp(cc->name, "highspeed") == 0) 292 opt2 |= V_CONG_CNTRL(CONG_ALG_HIGHSPEED); 293 else { 294 /* 295 * Use newreno in case the algorithm selected by the 296 * host stack is not supported by the hardware. 297 */ 298 opt2 |= V_CONG_CNTRL(CONG_ALG_NEWRENO); 299 } 300 } 301 302 if (s->rx_coalesce > 0 || (s->rx_coalesce < 0 && sc->tt.rx_coalesce)) 303 opt2 |= V_RX_COALESCE(M_RX_COALESCE); 304 305 /* Note that ofld_rxq is already set according to s->rxq. */ 306 opt2 |= F_RSS_QUEUE_VALID; 307 opt2 |= V_RSS_QUEUE(toep->ofld_rxq->iq.abs_id); 308 309#ifdef USE_DDP_RX_FLOW_CONTROL 310 if (toep->ulp_mode == ULP_MODE_TCPDDP) 311 opt2 |= F_RX_FC_DDP; 312#endif 313 314 if (toep->ulp_mode == ULP_MODE_TLS) { 315 opt2 &= ~V_RX_COALESCE(M_RX_COALESCE); 316 opt2 |= F_RX_FC_DISABLE; 317 } 318 319 return (htobe32(opt2)); 320} 321 322void 323t4_init_connect_cpl_handlers(void) 324{ 325 326 t4_register_cpl_handler(CPL_ACT_ESTABLISH, do_act_establish); 327 t4_register_cpl_handler(CPL_ACT_OPEN_RPL, do_act_open_rpl); 328} 329 330void 331t4_uninit_connect_cpl_handlers(void) 332{ 333 334 t4_register_cpl_handler(CPL_ACT_ESTABLISH, NULL); 335 t4_register_cpl_handler(CPL_ACT_OPEN_RPL, NULL); 336} 337 338#define DONT_OFFLOAD_ACTIVE_OPEN(x) do { \ 339 reason = __LINE__; \ 340 rc = (x); \ 341 goto failed; \ 342} while (0) 343 344static inline int 345act_open_cpl_size(struct adapter *sc, int isipv6) 346{ 347 int idx; 348 static const int sz_table[3][2] = { 349 { 350 sizeof (struct cpl_act_open_req), 351 sizeof (struct cpl_act_open_req6) 352 }, 353 { 354 sizeof (struct cpl_t5_act_open_req), 355 sizeof (struct cpl_t5_act_open_req6) 356 }, 357 { 358 sizeof (struct cpl_t6_act_open_req), 359 sizeof (struct cpl_t6_act_open_req6) 360 }, 361 }; 362 363 MPASS(chip_id(sc) >= CHELSIO_T4); 364 idx = min(chip_id(sc) - CHELSIO_T4, 2); 365 366 return (sz_table[idx][!!isipv6]); 367} 368 369/* 370 * active open (soconnect). 371 * 372 * State of affairs on entry: 373 * soisconnecting (so_state |= SS_ISCONNECTING) 374 * tcbinfo not locked (This has changed - used to be WLOCKed) 375 * inp WLOCKed 376 * tp->t_state = TCPS_SYN_SENT 377 * rtalloc1, RT_UNLOCK on rt. 378 */ 379int 380t4_connect(struct toedev *tod, struct socket *so, struct rtentry *rt, 381 struct sockaddr *nam) 382{ 383 struct adapter *sc = tod->tod_softc; 384 struct tom_data *td = tod_td(tod); 385 struct toepcb *toep = NULL; 386 struct wrqe *wr = NULL; 387 struct ifnet *rt_ifp = rt->rt_ifp; 388 struct vi_info *vi; 389 int mtu_idx, rscale, qid_atid, rc, isipv6, txqid, rxqid; 390 struct inpcb *inp = sotoinpcb(so); 391 struct tcpcb *tp = intotcpcb(inp); 392 int reason; 393 struct offload_settings settings; 394 uint16_t vid = 0xffff; 395 396 INP_WLOCK_ASSERT(inp); 397 KASSERT(nam->sa_family == AF_INET || nam->sa_family == AF_INET6, 398 ("%s: dest addr %p has family %u", __func__, nam, nam->sa_family)); 399 400 if (rt_ifp->if_type == IFT_ETHER) 401 vi = rt_ifp->if_softc; 402 else if (rt_ifp->if_type == IFT_L2VLAN) { 403 struct ifnet *ifp = VLAN_COOKIE(rt_ifp); 404 405 vi = ifp->if_softc; 406 VLAN_TAG(rt_ifp, &vid); 407 } else if (rt_ifp->if_type == IFT_IEEE8023ADLAG) 408 DONT_OFFLOAD_ACTIVE_OPEN(ENOSYS); /* XXX: implement lagg+TOE */ 409 else 410 DONT_OFFLOAD_ACTIVE_OPEN(ENOTSUP); 411 412 rw_rlock(&sc->policy_lock); 413 settings = *lookup_offload_policy(sc, OPEN_TYPE_ACTIVE, NULL, vid, inp); 414 rw_runlock(&sc->policy_lock); 415 if (!settings.offload) 416 DONT_OFFLOAD_ACTIVE_OPEN(EPERM); 417 418 if (settings.txq >= 0 && settings.txq < vi->nofldtxq) 419 txqid = settings.txq; 420 else 421 txqid = arc4random() % vi->nofldtxq; 422 txqid += vi->first_ofld_txq; 423 if (settings.rxq >= 0 && settings.rxq < vi->nofldrxq) 424 rxqid = settings.rxq; 425 else 426 rxqid = arc4random() % vi->nofldrxq; 427 rxqid += vi->first_ofld_rxq; 428 429 toep = alloc_toepcb(vi, txqid, rxqid, M_NOWAIT | M_ZERO); 430 if (toep == NULL) 431 DONT_OFFLOAD_ACTIVE_OPEN(ENOMEM); 432 433 toep->tid = alloc_atid(sc, toep); 434 if (toep->tid < 0) 435 DONT_OFFLOAD_ACTIVE_OPEN(ENOMEM); 436 437 toep->l2te = t4_l2t_get(vi->pi, rt_ifp, 438 rt->rt_flags & RTF_GATEWAY ? rt->rt_gateway : nam); 439 if (toep->l2te == NULL) 440 DONT_OFFLOAD_ACTIVE_OPEN(ENOMEM); 441 442 isipv6 = nam->sa_family == AF_INET6; 443 wr = alloc_wrqe(act_open_cpl_size(sc, isipv6), toep->ctrlq); 444 if (wr == NULL) 445 DONT_OFFLOAD_ACTIVE_OPEN(ENOMEM); 446 447 toep->vnet = so->so_vnet; 448 set_ulp_mode(toep, select_ulp_mode(so, sc, &settings)); 449 SOCKBUF_LOCK(&so->so_rcv); 450 /* opt0 rcv_bufsiz initially, assumes its normal meaning later */ 451 toep->rx_credits = min(select_rcv_wnd(so) >> 10, M_RCV_BUFSIZ); 452 SOCKBUF_UNLOCK(&so->so_rcv); 453 454 /* 455 * The kernel sets request_r_scale based on sb_max whereas we need to 456 * take hardware's MAX_RCV_WND into account too. This is normally a 457 * no-op as MAX_RCV_WND is much larger than the default sb_max. 458 */ 459 if (tp->t_flags & TF_REQ_SCALE) 460 rscale = tp->request_r_scale = select_rcv_wscale(); 461 else 462 rscale = 0; 463 mtu_idx = find_best_mtu_idx(sc, &inp->inp_inc, &settings); 464 qid_atid = (toep->ofld_rxq->iq.abs_id << 14) | toep->tid; 465 466 if (isipv6) { 467 struct cpl_act_open_req6 *cpl = wrtod(wr); 468 struct cpl_t5_act_open_req6 *cpl5 = (void *)cpl; 469 struct cpl_t6_act_open_req6 *cpl6 = (void *)cpl; 470 471 if ((inp->inp_vflag & INP_IPV6) == 0) 472 DONT_OFFLOAD_ACTIVE_OPEN(ENOTSUP); 473 474 toep->ce = hold_lip(td, &inp->in6p_laddr, NULL); 475 if (toep->ce == NULL) 476 DONT_OFFLOAD_ACTIVE_OPEN(ENOENT); 477 478 switch (chip_id(sc)) { 479 case CHELSIO_T4: 480 INIT_TP_WR(cpl, 0); 481 cpl->params = select_ntuple(vi, toep->l2te); 482 break; 483 case CHELSIO_T5: 484 INIT_TP_WR(cpl5, 0); 485 cpl5->iss = htobe32(tp->iss); 486 cpl5->params = select_ntuple(vi, toep->l2te); 487 break; 488 case CHELSIO_T6: 489 default: 490 INIT_TP_WR(cpl6, 0); 491 cpl6->iss = htobe32(tp->iss); 492 cpl6->params = select_ntuple(vi, toep->l2te); 493 break; 494 } 495 OPCODE_TID(cpl) = htobe32(MK_OPCODE_TID(CPL_ACT_OPEN_REQ6, 496 qid_atid)); 497 cpl->local_port = inp->inp_lport; 498 cpl->local_ip_hi = *(uint64_t *)&inp->in6p_laddr.s6_addr[0]; 499 cpl->local_ip_lo = *(uint64_t *)&inp->in6p_laddr.s6_addr[8]; 500 cpl->peer_port = inp->inp_fport; 501 cpl->peer_ip_hi = *(uint64_t *)&inp->in6p_faddr.s6_addr[0]; 502 cpl->peer_ip_lo = *(uint64_t *)&inp->in6p_faddr.s6_addr[8]; 503 cpl->opt0 = calc_opt0(so, vi, toep->l2te, mtu_idx, rscale, 504 toep->rx_credits, toep->ulp_mode, &settings); 505 cpl->opt2 = calc_opt2a(so, toep, &settings); 506 } else { 507 struct cpl_act_open_req *cpl = wrtod(wr); 508 struct cpl_t5_act_open_req *cpl5 = (void *)cpl; 509 struct cpl_t6_act_open_req *cpl6 = (void *)cpl; 510 511 switch (chip_id(sc)) { 512 case CHELSIO_T4: 513 INIT_TP_WR(cpl, 0); 514 cpl->params = select_ntuple(vi, toep->l2te); 515 break; 516 case CHELSIO_T5: 517 INIT_TP_WR(cpl5, 0); 518 cpl5->iss = htobe32(tp->iss); 519 cpl5->params = select_ntuple(vi, toep->l2te); 520 break; 521 case CHELSIO_T6: 522 default: 523 INIT_TP_WR(cpl6, 0); 524 cpl6->iss = htobe32(tp->iss); 525 cpl6->params = select_ntuple(vi, toep->l2te); 526 break; 527 } 528 OPCODE_TID(cpl) = htobe32(MK_OPCODE_TID(CPL_ACT_OPEN_REQ, 529 qid_atid)); 530 inp_4tuple_get(inp, &cpl->local_ip, &cpl->local_port, 531 &cpl->peer_ip, &cpl->peer_port); 532 cpl->opt0 = calc_opt0(so, vi, toep->l2te, mtu_idx, rscale, 533 toep->rx_credits, toep->ulp_mode, &settings); 534 cpl->opt2 = calc_opt2a(so, toep, &settings); 535 } 536 537 CTR5(KTR_CXGBE, "%s: atid %u (%s), toep %p, inp %p", __func__, 538 toep->tid, tcpstates[tp->t_state], toep, inp); 539 540 offload_socket(so, toep); 541 rc = t4_l2t_send(sc, wr, toep->l2te); 542 if (rc == 0) { 543 toep->flags |= TPF_CPL_PENDING; 544 return (0); 545 } 546 547 undo_offload_socket(so); 548 reason = __LINE__; 549failed: 550 CTR3(KTR_CXGBE, "%s: not offloading (%d), rc %d", __func__, reason, rc); 551 552 if (wr) 553 free_wrqe(wr); 554 555 if (toep) { 556 if (toep->tid >= 0) 557 free_atid(sc, toep->tid); 558 if (toep->l2te) 559 t4_l2t_release(toep->l2te); 560 if (toep->ce) 561 release_lip(td, toep->ce); 562 free_toepcb(toep); 563 } 564 565 return (rc); 566} 567#endif 568