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 #include <netinet/tcp_var.h> 53 #define TCPSTATES 54 #include <netinet/tcp_fsm.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 failed. 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 INP_WLOCK(inp); 130 toep->tid = tid; 131 insert_tid(sc, tid, toep); 132 if (inp->inp_flags & INP_DROPPED) { 133 134 /* socket closed by the kernel before hw told us it connected */ 135 136 send_flowc_wr(toep, NULL); 137 send_reset(sc, toep, be32toh(cpl->snd_isn)); 138 goto done; 139 } 140 141 make_established(toep, cpl->snd_isn, cpl->rcv_isn, cpl->tcp_opt); 142 done: 143 INP_WUNLOCK(inp); 144 return (0); 145 } 146 147 static inline int 148 act_open_has_tid(unsigned int status) 149 { 150 151 return (status != CPL_ERR_TCAM_FULL && 152 status != CPL_ERR_TCAM_PARITY && 153 status != CPL_ERR_CONN_EXIST && 154 status != CPL_ERR_ARP_MISS); 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 if (status != EAGAIN) 192 INP_INFO_RLOCK(&V_tcbinfo); 193 INP_WLOCK(inp); 194 toe_connect_failed(tod, inp, status); 195 final_cpl_received(toep); /* unlocks inp */ 196 if (status != EAGAIN) 197 INP_INFO_RUNLOCK(&V_tcbinfo); 198 } 199 200 static int 201 do_act_open_rpl(struct sge_iq *iq, const struct rss_header *rss, 202 struct mbuf *m) 203 { 204 struct adapter *sc = iq->adapter; 205 const struct cpl_act_open_rpl *cpl = (const void *)(rss + 1); 206 u_int atid = G_TID_TID(G_AOPEN_ATID(be32toh(cpl->atid_status))); 207 u_int status = G_AOPEN_STATUS(be32toh(cpl->atid_status)); 208 struct toepcb *toep = lookup_atid(sc, atid); 209 int rc; 210 211 KASSERT(m == NULL, ("%s: wasn't expecting payload", __func__)); 212 KASSERT(toep->tid == atid, ("%s: toep tid/atid mismatch", __func__)); 213 214 CTR3(KTR_CXGBE, "%s: atid %u, status %u ", __func__, atid, status); 215 216 /* Ignore negative advice */ 217 if (negative_advice(status)) 218 return (0); 219 220 if (status && act_open_has_tid(status)) 221 release_tid(sc, GET_TID(cpl), toep->ctrlq); 222 223 rc = act_open_rpl_status_to_errno(status); 224 act_open_failure_cleanup(sc, atid, rc); 225 226 return (0); 227 } 228 229 /* 230 * Options2 for active open. 231 */ 232 static uint32_t 233 calc_opt2a(struct socket *so, struct toepcb *toep) 234 { 235 struct tcpcb *tp = so_sototcpcb(so); 236 struct port_info *pi = toep->port; 237 struct adapter *sc = pi->adapter; 238 uint32_t opt2; 239 240 opt2 = V_TX_QUEUE(sc->params.tp.tx_modq[pi->tx_chan]) | 241 F_RSS_QUEUE_VALID | V_RSS_QUEUE(toep->ofld_rxq->iq.abs_id); 242 243 if (tp->t_flags & TF_SACK_PERMIT) 244 opt2 |= F_SACK_EN; 245 246 if (tp->t_flags & TF_REQ_TSTMP) 247 opt2 |= F_TSTAMPS_EN; 248 249 if (tp->t_flags & TF_REQ_SCALE) 250 opt2 |= F_WND_SCALE_EN; 251 252 if (V_tcp_do_ecn) 253 opt2 |= F_CCTRL_ECN; 254 255 /* RX_COALESCE is always a valid value (M_RX_COALESCE). */ 256 if (is_t4(sc)) 257 opt2 |= F_RX_COALESCE_VALID; 258 else { 259 opt2 |= F_T5_OPT_2_VALID; 260 opt2 |= F_CONG_CNTRL_VALID; /* OPT_2_ISS really, for T5 */ 261 } 262 if (sc->tt.rx_coalesce) 263 opt2 |= V_RX_COALESCE(M_RX_COALESCE); 264 265 #ifdef USE_DDP_RX_FLOW_CONTROL 266 if (toep->ulp_mode == ULP_MODE_TCPDDP) 267 opt2 |= F_RX_FC_VALID | F_RX_FC_DDP; 268 #endif 269 270 return (htobe32(opt2)); 271 } 272 273 void 274 t4_init_connect_cpl_handlers(struct adapter *sc) 275 { 276 277 t4_register_cpl_handler(sc, CPL_ACT_ESTABLISH, do_act_establish); 278 t4_register_cpl_handler(sc, CPL_ACT_OPEN_RPL, do_act_open_rpl); 279 } 280 281 #define DONT_OFFLOAD_ACTIVE_OPEN(x) do { \ 282 reason = __LINE__; \ 283 rc = (x); \ 284 goto failed; \ 285 } while (0) 286 287 static inline int 288 act_open_cpl_size(struct adapter *sc, int isipv6) 289 { 290 static const int sz_t4[] = { 291 sizeof (struct cpl_act_open_req), 292 sizeof (struct cpl_act_open_req6) 293 }; 294 static const int sz_t5[] = { 295 sizeof (struct cpl_t5_act_open_req), 296 sizeof (struct cpl_t5_act_open_req6) 297 }; 298 299 if (is_t4(sc)) 300 return (sz_t4[!!isipv6]); 301 else 302 return (sz_t5[!!isipv6]); 303 } 304 305 /* 306 * active open (soconnect). 307 * 308 * State of affairs on entry: 309 * soisconnecting (so_state |= SS_ISCONNECTING) 310 * tcbinfo not locked (This has changed - used to be WLOCKed) 311 * inp WLOCKed 312 * tp->t_state = TCPS_SYN_SENT 313 * rtalloc1, RT_UNLOCK on rt. 314 */ 315 int 316 t4_connect(struct toedev *tod, struct socket *so, struct rtentry *rt, 317 struct sockaddr *nam) 318 { 319 struct adapter *sc = tod->tod_softc; 320 struct tom_data *td = tod_td(tod); 321 struct toepcb *toep = NULL; 322 struct wrqe *wr = NULL; 323 struct ifnet *rt_ifp = rt->rt_ifp; 324 struct port_info *pi; 325 int mtu_idx, rscale, qid_atid, rc, isipv6; 326 struct inpcb *inp = sotoinpcb(so); 327 struct tcpcb *tp = intotcpcb(inp); 328 int reason; 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 pi = rt_ifp->if_softc; 336 else if (rt_ifp->if_type == IFT_L2VLAN) { 337 struct ifnet *ifp = VLAN_COOKIE(rt_ifp); 338 339 pi = ifp->if_softc; 340 } else if (rt_ifp->if_type == IFT_IEEE8023ADLAG) 341 DONT_OFFLOAD_ACTIVE_OPEN(ENOSYS); /* XXX: implement lagg+TOE */ 342 else 343 DONT_OFFLOAD_ACTIVE_OPEN(ENOTSUP); 344 345 toep = alloc_toepcb(pi, -1, -1, M_NOWAIT); 346 if (toep == NULL) 347 DONT_OFFLOAD_ACTIVE_OPEN(ENOMEM); 348 349 toep->tid = alloc_atid(sc, toep); 350 if (toep->tid < 0) 351 DONT_OFFLOAD_ACTIVE_OPEN(ENOMEM); 352 353 toep->l2te = t4_l2t_get(pi, rt_ifp, 354 rt->rt_flags & RTF_GATEWAY ? rt->rt_gateway : nam); 355 if (toep->l2te == NULL) 356 DONT_OFFLOAD_ACTIVE_OPEN(ENOMEM); 357 358 isipv6 = nam->sa_family == AF_INET6; 359 wr = alloc_wrqe(act_open_cpl_size(sc, isipv6), toep->ctrlq); 360 if (wr == NULL) 361 DONT_OFFLOAD_ACTIVE_OPEN(ENOMEM); 362 363 if (sc->tt.ddp && (so->so_options & SO_NO_DDP) == 0) 364 set_tcpddp_ulp_mode(toep); 365 else 366 toep->ulp_mode = ULP_MODE_NONE; 367 SOCKBUF_LOCK(&so->so_rcv); 368 /* opt0 rcv_bufsiz initially, assumes its normal meaning later */ 369 toep->rx_credits = min(select_rcv_wnd(so) >> 10, M_RCV_BUFSIZ); 370 SOCKBUF_UNLOCK(&so->so_rcv); 371 372 /* 373 * The kernel sets request_r_scale based on sb_max whereas we need to 374 * take hardware's MAX_RCV_WND into account too. This is normally a 375 * no-op as MAX_RCV_WND is much larger than the default sb_max. 376 */ 377 if (tp->t_flags & TF_REQ_SCALE) 378 rscale = tp->request_r_scale = select_rcv_wscale(); 379 else 380 rscale = 0; 381 mtu_idx = find_best_mtu_idx(sc, &inp->inp_inc, 0); 382 qid_atid = (toep->ofld_rxq->iq.abs_id << 14) | toep->tid; 383 384 if (isipv6) { 385 struct cpl_act_open_req6 *cpl = wrtod(wr); 386 387 if ((inp->inp_vflag & INP_IPV6) == 0) { 388 /* XXX think about this a bit more */ 389 log(LOG_ERR, 390 "%s: time to think about AF_INET6 + vflag 0x%x.\n", 391 __func__, inp->inp_vflag); 392 DONT_OFFLOAD_ACTIVE_OPEN(ENOTSUP); 393 } 394 395 toep->ce = hold_lip(td, &inp->in6p_laddr); 396 if (toep->ce == NULL) 397 DONT_OFFLOAD_ACTIVE_OPEN(ENOENT); 398 399 if (is_t4(sc)) { 400 INIT_TP_WR(cpl, 0); 401 cpl->params = select_ntuple(pi, toep->l2te); 402 } else { 403 struct cpl_t5_act_open_req6 *c5 = (void *)cpl; 404 405 INIT_TP_WR(c5, 0); 406 c5->iss = htobe32(tp->iss); 407 c5->params = select_ntuple(pi, toep->l2te); 408 } 409 OPCODE_TID(cpl) = htobe32(MK_OPCODE_TID(CPL_ACT_OPEN_REQ6, 410 qid_atid)); 411 cpl->local_port = inp->inp_lport; 412 cpl->local_ip_hi = *(uint64_t *)&inp->in6p_laddr.s6_addr[0]; 413 cpl->local_ip_lo = *(uint64_t *)&inp->in6p_laddr.s6_addr[8]; 414 cpl->peer_port = inp->inp_fport; 415 cpl->peer_ip_hi = *(uint64_t *)&inp->in6p_faddr.s6_addr[0]; 416 cpl->peer_ip_lo = *(uint64_t *)&inp->in6p_faddr.s6_addr[8]; 417 cpl->opt0 = calc_opt0(so, pi, toep->l2te, mtu_idx, rscale, 418 toep->rx_credits, toep->ulp_mode); 419 cpl->opt2 = calc_opt2a(so, toep); 420 } else { 421 struct cpl_act_open_req *cpl = wrtod(wr); 422 423 if (is_t4(sc)) { 424 INIT_TP_WR(cpl, 0); 425 cpl->params = select_ntuple(pi, toep->l2te); 426 } else { 427 struct cpl_t5_act_open_req *c5 = (void *)cpl; 428 429 INIT_TP_WR(c5, 0); 430 c5->iss = htobe32(tp->iss); 431 c5->params = select_ntuple(pi, toep->l2te); 432 } 433 OPCODE_TID(cpl) = htobe32(MK_OPCODE_TID(CPL_ACT_OPEN_REQ, 434 qid_atid)); 435 inp_4tuple_get(inp, &cpl->local_ip, &cpl->local_port, 436 &cpl->peer_ip, &cpl->peer_port); 437 cpl->opt0 = calc_opt0(so, pi, toep->l2te, mtu_idx, rscale, 438 toep->rx_credits, toep->ulp_mode); 439 cpl->opt2 = calc_opt2a(so, toep); 440 } 441 442 CTR5(KTR_CXGBE, "%s: atid %u (%s), toep %p, inp %p", __func__, 443 toep->tid, tcpstates[tp->t_state], toep, inp); 444 445 offload_socket(so, toep); 446 rc = t4_l2t_send(sc, wr, toep->l2te); 447 if (rc == 0) { 448 toep->flags |= TPF_CPL_PENDING; 449 return (0); 450 } 451 452 undo_offload_socket(so); 453 reason = __LINE__; 454 failed: 455 CTR3(KTR_CXGBE, "%s: not offloading (%d), rc %d", __func__, reason, rc); 456 457 if (wr) 458 free_wrqe(wr); 459 460 if (toep) { 461 if (toep->tid >= 0) 462 free_atid(sc, toep->tid); 463 if (toep->l2te) 464 t4_l2t_release(toep->l2te); 465 if (toep->ce) 466 release_lip(td, toep->ce); 467 free_toepcb(toep); 468 } 469 470 return (rc); 471 } 472 #endif 473