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 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 /* 148 * Convert an ACT_OPEN_RPL status to an errno. 149 */ 150 static inline int 151 act_open_rpl_status_to_errno(int status) 152 { 153 154 switch (status) { 155 case CPL_ERR_CONN_RESET: 156 return (ECONNREFUSED); 157 case CPL_ERR_ARP_MISS: 158 return (EHOSTUNREACH); 159 case CPL_ERR_CONN_TIMEDOUT: 160 return (ETIMEDOUT); 161 case CPL_ERR_TCAM_FULL: 162 return (EAGAIN); 163 case CPL_ERR_CONN_EXIST: 164 log(LOG_ERR, "ACTIVE_OPEN_RPL: 4-tuple in use\n"); 165 return (EAGAIN); 166 default: 167 return (EIO); 168 } 169 } 170 171 void 172 act_open_failure_cleanup(struct adapter *sc, u_int atid, u_int status) 173 { 174 struct toepcb *toep = lookup_atid(sc, atid); 175 struct inpcb *inp = toep->inp; 176 struct toedev *tod = &toep->td->tod; 177 178 free_atid(sc, atid); 179 toep->tid = -1; 180 181 if (status != EAGAIN) 182 INP_INFO_RLOCK(&V_tcbinfo); 183 INP_WLOCK(inp); 184 toe_connect_failed(tod, inp, status); 185 final_cpl_received(toep); /* unlocks inp */ 186 if (status != EAGAIN) 187 INP_INFO_RUNLOCK(&V_tcbinfo); 188 } 189 190 static int 191 do_act_open_rpl(struct sge_iq *iq, const struct rss_header *rss, 192 struct mbuf *m) 193 { 194 struct adapter *sc = iq->adapter; 195 const struct cpl_act_open_rpl *cpl = (const void *)(rss + 1); 196 u_int atid = G_TID_TID(G_AOPEN_ATID(be32toh(cpl->atid_status))); 197 u_int status = G_AOPEN_STATUS(be32toh(cpl->atid_status)); 198 struct toepcb *toep = lookup_atid(sc, atid); 199 int rc; 200 201 KASSERT(m == NULL, ("%s: wasn't expecting payload", __func__)); 202 KASSERT(toep->tid == atid, ("%s: toep tid/atid mismatch", __func__)); 203 204 CTR3(KTR_CXGBE, "%s: atid %u, status %u ", __func__, atid, status); 205 206 /* Ignore negative advice */ 207 if (negative_advice(status)) 208 return (0); 209 210 if (status && act_open_has_tid(status)) 211 release_tid(sc, GET_TID(cpl), toep->ctrlq); 212 213 rc = act_open_rpl_status_to_errno(status); 214 act_open_failure_cleanup(sc, atid, rc); 215 216 return (0); 217 } 218 219 /* 220 * Options2 for active open. 221 */ 222 static uint32_t 223 calc_opt2a(struct socket *so, struct toepcb *toep) 224 { 225 struct tcpcb *tp = so_sototcpcb(so); 226 struct port_info *pi = toep->vi->pi; 227 struct adapter *sc = pi->adapter; 228 uint32_t opt2; 229 230 opt2 = V_TX_QUEUE(sc->params.tp.tx_modq[pi->tx_chan]) | 231 F_RSS_QUEUE_VALID | V_RSS_QUEUE(toep->ofld_rxq->iq.abs_id); 232 233 if (tp->t_flags & TF_SACK_PERMIT) 234 opt2 |= F_SACK_EN; 235 236 if (tp->t_flags & TF_REQ_TSTMP) 237 opt2 |= F_TSTAMPS_EN; 238 239 if (tp->t_flags & TF_REQ_SCALE) 240 opt2 |= F_WND_SCALE_EN; 241 242 if (V_tcp_do_ecn) 243 opt2 |= F_CCTRL_ECN; 244 245 /* RX_COALESCE is always a valid value (M_RX_COALESCE). */ 246 if (is_t4(sc)) 247 opt2 |= F_RX_COALESCE_VALID; 248 else { 249 opt2 |= F_T5_OPT_2_VALID; 250 opt2 |= F_T5_ISS; 251 } 252 if (sc->tt.rx_coalesce) 253 opt2 |= V_RX_COALESCE(M_RX_COALESCE); 254 255 #ifdef USE_DDP_RX_FLOW_CONTROL 256 if (toep->ulp_mode == ULP_MODE_TCPDDP) 257 opt2 |= F_RX_FC_VALID | F_RX_FC_DDP; 258 #endif 259 260 return (htobe32(opt2)); 261 } 262 263 void 264 t4_init_connect_cpl_handlers(struct adapter *sc) 265 { 266 267 t4_register_cpl_handler(sc, CPL_ACT_ESTABLISH, do_act_establish); 268 t4_register_cpl_handler(sc, CPL_ACT_OPEN_RPL, do_act_open_rpl); 269 } 270 271 #define DONT_OFFLOAD_ACTIVE_OPEN(x) do { \ 272 reason = __LINE__; \ 273 rc = (x); \ 274 goto failed; \ 275 } while (0) 276 277 static inline int 278 act_open_cpl_size(struct adapter *sc, int isipv6) 279 { 280 static const int sz_t4[] = { 281 sizeof (struct cpl_act_open_req), 282 sizeof (struct cpl_act_open_req6) 283 }; 284 static const int sz_t5[] = { 285 sizeof (struct cpl_t5_act_open_req), 286 sizeof (struct cpl_t5_act_open_req6) 287 }; 288 289 if (is_t4(sc)) 290 return (sz_t4[!!isipv6]); 291 else 292 return (sz_t5[!!isipv6]); 293 } 294 295 /* 296 * active open (soconnect). 297 * 298 * State of affairs on entry: 299 * soisconnecting (so_state |= SS_ISCONNECTING) 300 * tcbinfo not locked (This has changed - used to be WLOCKed) 301 * inp WLOCKed 302 * tp->t_state = TCPS_SYN_SENT 303 * rtalloc1, RT_UNLOCK on rt. 304 */ 305 int 306 t4_connect(struct toedev *tod, struct socket *so, struct rtentry *rt, 307 struct sockaddr *nam) 308 { 309 struct adapter *sc = tod->tod_softc; 310 struct tom_data *td = tod_td(tod); 311 struct toepcb *toep = NULL; 312 struct wrqe *wr = NULL; 313 struct ifnet *rt_ifp = rt->rt_ifp; 314 struct vi_info *vi; 315 int mtu_idx, rscale, qid_atid, rc, isipv6; 316 struct inpcb *inp = sotoinpcb(so); 317 struct tcpcb *tp = intotcpcb(inp); 318 int reason; 319 320 INP_WLOCK_ASSERT(inp); 321 KASSERT(nam->sa_family == AF_INET || nam->sa_family == AF_INET6, 322 ("%s: dest addr %p has family %u", __func__, nam, nam->sa_family)); 323 324 if (rt_ifp->if_type == IFT_ETHER) 325 vi = rt_ifp->if_softc; 326 else if (rt_ifp->if_type == IFT_L2VLAN) { 327 struct ifnet *ifp = VLAN_COOKIE(rt_ifp); 328 329 vi = ifp->if_softc; 330 } else if (rt_ifp->if_type == IFT_IEEE8023ADLAG) 331 DONT_OFFLOAD_ACTIVE_OPEN(ENOSYS); /* XXX: implement lagg+TOE */ 332 else 333 DONT_OFFLOAD_ACTIVE_OPEN(ENOTSUP); 334 335 toep = alloc_toepcb(vi, -1, -1, M_NOWAIT); 336 if (toep == NULL) 337 DONT_OFFLOAD_ACTIVE_OPEN(ENOMEM); 338 339 toep->tid = alloc_atid(sc, toep); 340 if (toep->tid < 0) 341 DONT_OFFLOAD_ACTIVE_OPEN(ENOMEM); 342 343 toep->l2te = t4_l2t_get(vi->pi, rt_ifp, 344 rt->rt_flags & RTF_GATEWAY ? rt->rt_gateway : nam); 345 if (toep->l2te == NULL) 346 DONT_OFFLOAD_ACTIVE_OPEN(ENOMEM); 347 348 isipv6 = nam->sa_family == AF_INET6; 349 wr = alloc_wrqe(act_open_cpl_size(sc, isipv6), toep->ctrlq); 350 if (wr == NULL) 351 DONT_OFFLOAD_ACTIVE_OPEN(ENOMEM); 352 353 if (sc->tt.ddp && (so->so_options & SO_NO_DDP) == 0) 354 set_tcpddp_ulp_mode(toep); 355 else 356 toep->ulp_mode = ULP_MODE_NONE; 357 SOCKBUF_LOCK(&so->so_rcv); 358 /* opt0 rcv_bufsiz initially, assumes its normal meaning later */ 359 toep->rx_credits = min(select_rcv_wnd(so) >> 10, M_RCV_BUFSIZ); 360 SOCKBUF_UNLOCK(&so->so_rcv); 361 362 /* 363 * The kernel sets request_r_scale based on sb_max whereas we need to 364 * take hardware's MAX_RCV_WND into account too. This is normally a 365 * no-op as MAX_RCV_WND is much larger than the default sb_max. 366 */ 367 if (tp->t_flags & TF_REQ_SCALE) 368 rscale = tp->request_r_scale = select_rcv_wscale(); 369 else 370 rscale = 0; 371 mtu_idx = find_best_mtu_idx(sc, &inp->inp_inc, 0); 372 qid_atid = (toep->ofld_rxq->iq.abs_id << 14) | toep->tid; 373 374 if (isipv6) { 375 struct cpl_act_open_req6 *cpl = wrtod(wr); 376 377 if ((inp->inp_vflag & INP_IPV6) == 0) { 378 /* XXX think about this a bit more */ 379 log(LOG_ERR, 380 "%s: time to think about AF_INET6 + vflag 0x%x.\n", 381 __func__, inp->inp_vflag); 382 DONT_OFFLOAD_ACTIVE_OPEN(ENOTSUP); 383 } 384 385 toep->ce = hold_lip(td, &inp->in6p_laddr); 386 if (toep->ce == NULL) 387 DONT_OFFLOAD_ACTIVE_OPEN(ENOENT); 388 389 if (is_t4(sc)) { 390 INIT_TP_WR(cpl, 0); 391 cpl->params = select_ntuple(vi, toep->l2te); 392 } else { 393 struct cpl_t5_act_open_req6 *c5 = (void *)cpl; 394 395 INIT_TP_WR(c5, 0); 396 c5->iss = htobe32(tp->iss); 397 c5->params = select_ntuple(vi, toep->l2te); 398 } 399 OPCODE_TID(cpl) = htobe32(MK_OPCODE_TID(CPL_ACT_OPEN_REQ6, 400 qid_atid)); 401 cpl->local_port = inp->inp_lport; 402 cpl->local_ip_hi = *(uint64_t *)&inp->in6p_laddr.s6_addr[0]; 403 cpl->local_ip_lo = *(uint64_t *)&inp->in6p_laddr.s6_addr[8]; 404 cpl->peer_port = inp->inp_fport; 405 cpl->peer_ip_hi = *(uint64_t *)&inp->in6p_faddr.s6_addr[0]; 406 cpl->peer_ip_lo = *(uint64_t *)&inp->in6p_faddr.s6_addr[8]; 407 cpl->opt0 = calc_opt0(so, vi, toep->l2te, mtu_idx, rscale, 408 toep->rx_credits, toep->ulp_mode); 409 cpl->opt2 = calc_opt2a(so, toep); 410 } else { 411 struct cpl_act_open_req *cpl = wrtod(wr); 412 413 if (is_t4(sc)) { 414 INIT_TP_WR(cpl, 0); 415 cpl->params = select_ntuple(vi, toep->l2te); 416 } else { 417 struct cpl_t5_act_open_req *c5 = (void *)cpl; 418 419 INIT_TP_WR(c5, 0); 420 c5->iss = htobe32(tp->iss); 421 c5->params = select_ntuple(vi, toep->l2te); 422 } 423 OPCODE_TID(cpl) = htobe32(MK_OPCODE_TID(CPL_ACT_OPEN_REQ, 424 qid_atid)); 425 inp_4tuple_get(inp, &cpl->local_ip, &cpl->local_port, 426 &cpl->peer_ip, &cpl->peer_port); 427 cpl->opt0 = calc_opt0(so, vi, toep->l2te, mtu_idx, rscale, 428 toep->rx_credits, toep->ulp_mode); 429 cpl->opt2 = calc_opt2a(so, toep); 430 } 431 432 CTR5(KTR_CXGBE, "%s: atid %u (%s), toep %p, inp %p", __func__, 433 toep->tid, tcpstates[tp->t_state], toep, inp); 434 435 offload_socket(so, toep); 436 rc = t4_l2t_send(sc, wr, toep->l2te); 437 if (rc == 0) { 438 toep->flags |= TPF_CPL_PENDING; 439 return (0); 440 } 441 442 undo_offload_socket(so); 443 reason = __LINE__; 444 failed: 445 CTR3(KTR_CXGBE, "%s: not offloading (%d), rc %d", __func__, reason, rc); 446 447 if (wr) 448 free_wrqe(wr); 449 450 if (toep) { 451 if (toep->tid >= 0) 452 free_atid(sc, toep->tid); 453 if (toep->l2te) 454 t4_l2t_release(toep->l2te); 455 if (toep->ce) 456 release_lip(td, toep->ce); 457 free_toepcb(toep); 458 } 459 460 return (rc); 461 } 462 #endif 463