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