1 /*- 2 * SPDX-License-Identifier: BSD-2-Clause-FreeBSD 3 * 4 * Copyright (c) 2009-2013, 2016 Chelsio, Inc. All rights reserved. 5 * 6 * This software is available to you under a choice of one of two 7 * licenses. You may choose to be licensed under the terms of the GNU 8 * General Public License (GPL) Version 2, available from the file 9 * COPYING in the main directory of this source tree, or the 10 * OpenIB.org BSD license below: 11 * 12 * Redistribution and use in source and binary forms, with or 13 * without modification, are permitted provided that the following 14 * conditions are met: 15 * 16 * - Redistributions of source code must retain the above 17 * copyright notice, this list of conditions and the following 18 * disclaimer. 19 * 20 * - Redistributions in binary form must reproduce the above 21 * copyright notice, this list of conditions and the following 22 * disclaimer in the documentation and/or other materials 23 * provided with the distribution. 24 * 25 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, 26 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF 27 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND 28 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS 29 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN 30 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN 31 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE 32 * SOFTWARE. 33 */ 34 #include <sys/cdefs.h> 35 __FBSDID("$FreeBSD$"); 36 37 #include "opt_inet.h" 38 39 #ifdef TCP_OFFLOAD 40 #include <sys/types.h> 41 #include <sys/malloc.h> 42 #include <sys/socket.h> 43 #include <sys/socketvar.h> 44 #include <sys/sockio.h> 45 #include <sys/taskqueue.h> 46 #include <netinet/in.h> 47 #include <net/route.h> 48 49 #include <netinet/in_systm.h> 50 #include <netinet/in_pcb.h> 51 #include <netinet6/in6_pcb.h> 52 #include <netinet/ip.h> 53 #include <netinet/in_fib.h> 54 #include <netinet6/in6_fib.h> 55 #include <netinet6/scope6_var.h> 56 #include <netinet/ip_var.h> 57 #include <netinet/tcp_var.h> 58 #include <netinet/tcp.h> 59 #include <netinet/tcpip.h> 60 61 #include <netinet/toecore.h> 62 63 struct sge_iq; 64 struct rss_header; 65 struct cpl_set_tcb_rpl; 66 #include <linux/types.h> 67 #include "offload.h" 68 #include "tom/t4_tom.h" 69 70 #define TOEPCB(so) ((struct toepcb *)(so_sototcpcb((so))->t_toe)) 71 72 #include "iw_cxgbe.h" 73 #include <linux/module.h> 74 #include <linux/workqueue.h> 75 #include <linux/notifier.h> 76 #include <linux/inetdevice.h> 77 #include <linux/if_vlan.h> 78 #include <net/netevent.h> 79 #include <rdma/rdma_cm.h> 80 81 static spinlock_t req_lock; 82 static TAILQ_HEAD(c4iw_ep_list, c4iw_ep_common) req_list; 83 static struct work_struct c4iw_task; 84 static struct workqueue_struct *c4iw_taskq; 85 static LIST_HEAD(err_cqe_list); 86 static spinlock_t err_cqe_lock; 87 static LIST_HEAD(listen_port_list); 88 static DEFINE_MUTEX(listen_port_mutex); 89 90 static void process_req(struct work_struct *ctx); 91 static void start_ep_timer(struct c4iw_ep *ep); 92 static int stop_ep_timer(struct c4iw_ep *ep); 93 static int set_tcpinfo(struct c4iw_ep *ep); 94 static void process_timeout(struct c4iw_ep *ep); 95 static void process_err_cqes(void); 96 static void *alloc_ep(int size, gfp_t flags); 97 static void close_socket(struct socket *so); 98 static int send_mpa_req(struct c4iw_ep *ep); 99 static int send_mpa_reject(struct c4iw_ep *ep, const void *pdata, u8 plen); 100 static int send_mpa_reply(struct c4iw_ep *ep, const void *pdata, u8 plen); 101 static void close_complete_upcall(struct c4iw_ep *ep, int status); 102 static int send_abort(struct c4iw_ep *ep); 103 static void peer_close_upcall(struct c4iw_ep *ep); 104 static void peer_abort_upcall(struct c4iw_ep *ep); 105 static void connect_reply_upcall(struct c4iw_ep *ep, int status); 106 static int connect_request_upcall(struct c4iw_ep *ep); 107 static void established_upcall(struct c4iw_ep *ep); 108 static int process_mpa_reply(struct c4iw_ep *ep); 109 static int process_mpa_request(struct c4iw_ep *ep); 110 static void process_peer_close(struct c4iw_ep *ep); 111 static void process_conn_error(struct c4iw_ep *ep); 112 static void process_close_complete(struct c4iw_ep *ep); 113 static void ep_timeout(unsigned long arg); 114 static void setiwsockopt(struct socket *so); 115 static void init_iwarp_socket(struct socket *so, void *arg); 116 static void uninit_iwarp_socket(struct socket *so); 117 static void process_data(struct c4iw_ep *ep); 118 static void process_connected(struct c4iw_ep *ep); 119 static int c4iw_so_upcall(struct socket *so, void *arg, int waitflag); 120 static void process_socket_event(struct c4iw_ep *ep); 121 static void release_ep_resources(struct c4iw_ep *ep); 122 static int process_terminate(struct c4iw_ep *ep); 123 static int terminate(struct sge_iq *iq, const struct rss_header *rss, 124 struct mbuf *m); 125 static int add_ep_to_req_list(struct c4iw_ep *ep, int ep_events); 126 static struct listen_port_info * 127 add_ep_to_listenlist(struct c4iw_listen_ep *lep); 128 static int rem_ep_from_listenlist(struct c4iw_listen_ep *lep); 129 static struct c4iw_listen_ep * 130 find_real_listen_ep(struct c4iw_listen_ep *master_lep, struct socket *so); 131 static int get_ifnet_from_raddr(struct sockaddr_storage *raddr, 132 struct ifnet **ifp); 133 static void process_newconn(struct c4iw_listen_ep *master_lep, 134 struct socket *new_so); 135 #define START_EP_TIMER(ep) \ 136 do { \ 137 CTR3(KTR_IW_CXGBE, "start_ep_timer (%s:%d) ep %p", \ 138 __func__, __LINE__, (ep)); \ 139 start_ep_timer(ep); \ 140 } while (0) 141 142 #define STOP_EP_TIMER(ep) \ 143 ({ \ 144 CTR3(KTR_IW_CXGBE, "stop_ep_timer (%s:%d) ep %p", \ 145 __func__, __LINE__, (ep)); \ 146 stop_ep_timer(ep); \ 147 }) 148 149 #define GET_LOCAL_ADDR(pladdr, so) \ 150 do { \ 151 struct sockaddr_storage *__a = NULL; \ 152 struct inpcb *__inp = sotoinpcb(so); \ 153 KASSERT(__inp != NULL, \ 154 ("GET_LOCAL_ADDR(%s):so:%p, inp = NULL", __func__, so)); \ 155 if (__inp->inp_vflag & INP_IPV4) \ 156 in_getsockaddr(so, (struct sockaddr **)&__a); \ 157 else \ 158 in6_getsockaddr(so, (struct sockaddr **)&__a); \ 159 *(pladdr) = *__a; \ 160 free(__a, M_SONAME); \ 161 } while (0) 162 163 #define GET_REMOTE_ADDR(praddr, so) \ 164 do { \ 165 struct sockaddr_storage *__a = NULL; \ 166 struct inpcb *__inp = sotoinpcb(so); \ 167 KASSERT(__inp != NULL, \ 168 ("GET_REMOTE_ADDR(%s):so:%p, inp = NULL", __func__, so)); \ 169 if (__inp->inp_vflag & INP_IPV4) \ 170 in_getpeeraddr(so, (struct sockaddr **)&__a); \ 171 else \ 172 in6_getpeeraddr(so, (struct sockaddr **)&__a); \ 173 *(praddr) = *__a; \ 174 free(__a, M_SONAME); \ 175 } while (0) 176 177 static char *states[] = { 178 "idle", 179 "listen", 180 "connecting", 181 "mpa_wait_req", 182 "mpa_req_sent", 183 "mpa_req_rcvd", 184 "mpa_rep_sent", 185 "fpdu_mode", 186 "aborting", 187 "closing", 188 "moribund", 189 "dead", 190 NULL, 191 }; 192 193 static void deref_cm_id(struct c4iw_ep_common *epc) 194 { 195 epc->cm_id->rem_ref(epc->cm_id); 196 epc->cm_id = NULL; 197 set_bit(CM_ID_DEREFED, &epc->history); 198 } 199 200 static void ref_cm_id(struct c4iw_ep_common *epc) 201 { 202 set_bit(CM_ID_REFED, &epc->history); 203 epc->cm_id->add_ref(epc->cm_id); 204 } 205 206 static void deref_qp(struct c4iw_ep *ep) 207 { 208 c4iw_qp_rem_ref(&ep->com.qp->ibqp); 209 clear_bit(QP_REFERENCED, &ep->com.flags); 210 set_bit(QP_DEREFED, &ep->com.history); 211 } 212 213 static void ref_qp(struct c4iw_ep *ep) 214 { 215 set_bit(QP_REFERENCED, &ep->com.flags); 216 set_bit(QP_REFED, &ep->com.history); 217 c4iw_qp_add_ref(&ep->com.qp->ibqp); 218 } 219 /* allocated per TCP port while listening */ 220 struct listen_port_info { 221 uint16_t port_num; /* TCP port address */ 222 struct list_head list; /* belongs to listen_port_list */ 223 struct list_head lep_list; /* per port lep list */ 224 uint32_t refcnt; /* number of lep's listening */ 225 }; 226 227 /* 228 * Following two lists are used to manage INADDR_ANY listeners: 229 * 1)listen_port_list 230 * 2)lep_list 231 * 232 * Below is the INADDR_ANY listener lists overview on a system with a two port 233 * adapter: 234 * |------------------| 235 * |listen_port_list | 236 * |------------------| 237 * | 238 * | |-----------| |-----------| 239 * | | port_num:X| | port_num:X| 240 * |--------------|-list------|-------|-list------|-------.... 241 * | lep_list----| | lep_list----| 242 * | refcnt | | | refcnt | | 243 * | | | | | | 244 * | | | | | | 245 * |-----------| | |-----------| | 246 * | | 247 * | | 248 * | | 249 * | | lep1 lep2 250 * | | |----------------| |----------------| 251 * | |----| listen_ep_list |----| listen_ep_list | 252 * | |----------------| |----------------| 253 * | 254 * | 255 * | lep1 lep2 256 * | |----------------| |----------------| 257 * |---| listen_ep_list |----| listen_ep_list | 258 * |----------------| |----------------| 259 * 260 * Because of two port adapter, the number of lep's are two(lep1 & lep2) for 261 * each TCP port number. 262 * 263 * Here 'lep1' is always marked as Master lep, because solisten() is always 264 * called through first lep. 265 * 266 */ 267 static struct listen_port_info * 268 add_ep_to_listenlist(struct c4iw_listen_ep *lep) 269 { 270 uint16_t port; 271 struct listen_port_info *port_info = NULL; 272 struct sockaddr_storage *laddr = &lep->com.local_addr; 273 274 port = (laddr->ss_family == AF_INET) ? 275 ((struct sockaddr_in *)laddr)->sin_port : 276 ((struct sockaddr_in6 *)laddr)->sin6_port; 277 278 mutex_lock(&listen_port_mutex); 279 280 list_for_each_entry(port_info, &listen_port_list, list) 281 if (port_info->port_num == port) 282 goto found_port; 283 284 port_info = malloc(sizeof(*port_info), M_CXGBE, M_WAITOK); 285 port_info->port_num = port; 286 port_info->refcnt = 0; 287 288 list_add_tail(&port_info->list, &listen_port_list); 289 INIT_LIST_HEAD(&port_info->lep_list); 290 291 found_port: 292 port_info->refcnt++; 293 list_add_tail(&lep->listen_ep_list, &port_info->lep_list); 294 mutex_unlock(&listen_port_mutex); 295 return port_info; 296 } 297 298 static int 299 rem_ep_from_listenlist(struct c4iw_listen_ep *lep) 300 { 301 uint16_t port; 302 struct listen_port_info *port_info = NULL; 303 struct sockaddr_storage *laddr = &lep->com.local_addr; 304 int refcnt = 0; 305 306 port = (laddr->ss_family == AF_INET) ? 307 ((struct sockaddr_in *)laddr)->sin_port : 308 ((struct sockaddr_in6 *)laddr)->sin6_port; 309 310 mutex_lock(&listen_port_mutex); 311 312 /* get the port_info structure based on the lep's port address */ 313 list_for_each_entry(port_info, &listen_port_list, list) { 314 if (port_info->port_num == port) { 315 port_info->refcnt--; 316 refcnt = port_info->refcnt; 317 /* remove the current lep from the listen list */ 318 list_del(&lep->listen_ep_list); 319 if (port_info->refcnt == 0) { 320 /* Remove this entry from the list as there 321 * are no more listeners for this port_num. 322 */ 323 list_del(&port_info->list); 324 kfree(port_info); 325 } 326 break; 327 } 328 } 329 mutex_unlock(&listen_port_mutex); 330 return refcnt; 331 } 332 333 /* 334 * Find the lep that belongs to the ifnet on which the SYN frame was received. 335 */ 336 struct c4iw_listen_ep * 337 find_real_listen_ep(struct c4iw_listen_ep *master_lep, struct socket *so) 338 { 339 struct adapter *adap = NULL; 340 struct c4iw_listen_ep *lep = NULL; 341 struct ifnet *ifp = NULL, *hw_ifp = NULL; 342 struct listen_port_info *port_info = NULL; 343 int i = 0, found_portinfo = 0, found_lep = 0; 344 uint16_t port; 345 346 /* 347 * STEP 1: Figure out 'ifp' of the physical interface, not pseudo 348 * interfaces like vlan, lagg, etc.. 349 * TBD: lagg support, lagg + vlan support. 350 */ 351 ifp = TOEPCB(so)->l2te->ifp; 352 if (ifp->if_type == IFT_L2VLAN) { 353 hw_ifp = VLAN_TRUNKDEV(ifp); 354 if (hw_ifp == NULL) { 355 CTR4(KTR_IW_CXGBE, "%s: Failed to get parent ifnet of " 356 "vlan ifnet %p, sock %p, master_lep %p", 357 __func__, ifp, so, master_lep); 358 return (NULL); 359 } 360 } else 361 hw_ifp = ifp; 362 363 /* STEP 2: Find 'port_info' with listener local port address. */ 364 port = (master_lep->com.local_addr.ss_family == AF_INET) ? 365 ((struct sockaddr_in *)&master_lep->com.local_addr)->sin_port : 366 ((struct sockaddr_in6 *)&master_lep->com.local_addr)->sin6_port; 367 368 369 mutex_lock(&listen_port_mutex); 370 list_for_each_entry(port_info, &listen_port_list, list) 371 if (port_info->port_num == port) { 372 found_portinfo =1; 373 break; 374 } 375 if (!found_portinfo) 376 goto out; 377 378 /* STEP 3: Traverse through list of lep's that are bound to the current 379 * TCP port address and find the lep that belongs to the ifnet on which 380 * the SYN frame was received. 381 */ 382 list_for_each_entry(lep, &port_info->lep_list, listen_ep_list) { 383 adap = lep->com.dev->rdev.adap; 384 for_each_port(adap, i) { 385 if (hw_ifp == adap->port[i]->vi[0].ifp) { 386 found_lep =1; 387 goto out; 388 } 389 } 390 } 391 out: 392 mutex_unlock(&listen_port_mutex); 393 return found_lep ? lep : (NULL); 394 } 395 396 static void process_timeout(struct c4iw_ep *ep) 397 { 398 struct c4iw_qp_attributes attrs = {0}; 399 int abort = 1; 400 401 CTR4(KTR_IW_CXGBE, "%s ep :%p, tid:%u, state %d", __func__, 402 ep, ep->hwtid, ep->com.state); 403 set_bit(TIMEDOUT, &ep->com.history); 404 switch (ep->com.state) { 405 case MPA_REQ_SENT: 406 connect_reply_upcall(ep, -ETIMEDOUT); 407 break; 408 case MPA_REQ_WAIT: 409 case MPA_REQ_RCVD: 410 case MPA_REP_SENT: 411 case FPDU_MODE: 412 break; 413 case CLOSING: 414 case MORIBUND: 415 if (ep->com.cm_id && ep->com.qp) { 416 attrs.next_state = C4IW_QP_STATE_ERROR; 417 c4iw_modify_qp(ep->com.dev, ep->com.qp, 418 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1); 419 } 420 close_complete_upcall(ep, -ETIMEDOUT); 421 break; 422 case ABORTING: 423 case DEAD: 424 /* 425 * These states are expected if the ep timed out at the same 426 * time as another thread was calling stop_ep_timer(). 427 * So we silently do nothing for these states. 428 */ 429 abort = 0; 430 break; 431 default: 432 CTR4(KTR_IW_CXGBE, "%s unexpected state ep %p tid %u state %u\n" 433 , __func__, ep, ep->hwtid, ep->com.state); 434 abort = 0; 435 } 436 if (abort) 437 c4iw_ep_disconnect(ep, 1, GFP_KERNEL); 438 c4iw_put_ep(&ep->com); 439 return; 440 } 441 442 struct cqe_list_entry { 443 struct list_head entry; 444 struct c4iw_dev *rhp; 445 struct t4_cqe err_cqe; 446 }; 447 448 static void 449 process_err_cqes(void) 450 { 451 unsigned long flag; 452 struct cqe_list_entry *cle; 453 454 spin_lock_irqsave(&err_cqe_lock, flag); 455 while (!list_empty(&err_cqe_list)) { 456 struct list_head *tmp; 457 tmp = err_cqe_list.next; 458 list_del(tmp); 459 tmp->next = tmp->prev = NULL; 460 spin_unlock_irqrestore(&err_cqe_lock, flag); 461 cle = list_entry(tmp, struct cqe_list_entry, entry); 462 c4iw_ev_dispatch(cle->rhp, &cle->err_cqe); 463 free(cle, M_CXGBE); 464 spin_lock_irqsave(&err_cqe_lock, flag); 465 } 466 spin_unlock_irqrestore(&err_cqe_lock, flag); 467 468 return; 469 } 470 471 static void 472 process_req(struct work_struct *ctx) 473 { 474 struct c4iw_ep_common *epc; 475 unsigned long flag; 476 int ep_events; 477 478 process_err_cqes(); 479 spin_lock_irqsave(&req_lock, flag); 480 while (!TAILQ_EMPTY(&req_list)) { 481 epc = TAILQ_FIRST(&req_list); 482 TAILQ_REMOVE(&req_list, epc, entry); 483 epc->entry.tqe_prev = NULL; 484 ep_events = epc->ep_events; 485 epc->ep_events = 0; 486 spin_unlock_irqrestore(&req_lock, flag); 487 mutex_lock(&epc->mutex); 488 CTR5(KTR_IW_CXGBE, "%s: so %p, ep %p, ep_state %s events 0x%x", 489 __func__, epc->so, epc, states[epc->state], ep_events); 490 if (ep_events & C4IW_EVENT_TERM) 491 process_terminate((struct c4iw_ep *)epc); 492 if (ep_events & C4IW_EVENT_TIMEOUT) 493 process_timeout((struct c4iw_ep *)epc); 494 if (ep_events & C4IW_EVENT_SOCKET) 495 process_socket_event((struct c4iw_ep *)epc); 496 mutex_unlock(&epc->mutex); 497 c4iw_put_ep(epc); 498 process_err_cqes(); 499 spin_lock_irqsave(&req_lock, flag); 500 } 501 spin_unlock_irqrestore(&req_lock, flag); 502 } 503 504 /* 505 * XXX: doesn't belong here in the iWARP driver. 506 * XXX: assumes that the connection was offloaded by cxgbe/t4_tom if TF_TOE is 507 * set. Is this a valid assumption for active open? 508 */ 509 static int 510 set_tcpinfo(struct c4iw_ep *ep) 511 { 512 struct socket *so = ep->com.so; 513 struct inpcb *inp = sotoinpcb(so); 514 struct tcpcb *tp; 515 struct toepcb *toep; 516 int rc = 0; 517 518 INP_WLOCK(inp); 519 tp = intotcpcb(inp); 520 if ((tp->t_flags & TF_TOE) == 0) { 521 rc = EINVAL; 522 log(LOG_ERR, "%s: connection not offloaded (so %p, ep %p)\n", 523 __func__, so, ep); 524 goto done; 525 } 526 toep = TOEPCB(so); 527 528 ep->hwtid = toep->tid; 529 ep->snd_seq = tp->snd_nxt; 530 ep->rcv_seq = tp->rcv_nxt; 531 ep->emss = max(tp->t_maxseg, 128); 532 done: 533 INP_WUNLOCK(inp); 534 return (rc); 535 536 } 537 static int 538 get_ifnet_from_raddr(struct sockaddr_storage *raddr, struct ifnet **ifp) 539 { 540 int err = 0; 541 542 if (raddr->ss_family == AF_INET) { 543 struct sockaddr_in *raddr4 = (struct sockaddr_in *)raddr; 544 struct nhop4_extended nh4 = {0}; 545 546 err = fib4_lookup_nh_ext(RT_DEFAULT_FIB, raddr4->sin_addr, 547 NHR_REF, 0, &nh4); 548 *ifp = nh4.nh_ifp; 549 if (err) 550 fib4_free_nh_ext(RT_DEFAULT_FIB, &nh4); 551 } else { 552 struct sockaddr_in6 *raddr6 = (struct sockaddr_in6 *)raddr; 553 struct nhop6_extended nh6 = {0}; 554 struct in6_addr addr6; 555 uint32_t scopeid; 556 557 memset(&addr6, 0, sizeof(addr6)); 558 in6_splitscope((struct in6_addr *)&raddr6->sin6_addr, 559 &addr6, &scopeid); 560 err = fib6_lookup_nh_ext(RT_DEFAULT_FIB, &addr6, scopeid, 561 NHR_REF, 0, &nh6); 562 *ifp = nh6.nh_ifp; 563 if (err) 564 fib6_free_nh_ext(RT_DEFAULT_FIB, &nh6); 565 } 566 567 CTR2(KTR_IW_CXGBE, "%s: return: %d", __func__, err); 568 return err; 569 } 570 571 static void 572 close_socket(struct socket *so) 573 { 574 uninit_iwarp_socket(so); 575 soclose(so); 576 } 577 578 static void 579 process_peer_close(struct c4iw_ep *ep) 580 { 581 struct c4iw_qp_attributes attrs = {0}; 582 int disconnect = 1; 583 int release = 0; 584 585 CTR4(KTR_IW_CXGBE, "%s:ppcB ep %p so %p state %s", __func__, ep, 586 ep->com.so, states[ep->com.state]); 587 588 switch (ep->com.state) { 589 590 case MPA_REQ_WAIT: 591 CTR2(KTR_IW_CXGBE, "%s:ppc1 %p MPA_REQ_WAIT DEAD", 592 __func__, ep); 593 /* Fallthrough */ 594 case MPA_REQ_SENT: 595 CTR2(KTR_IW_CXGBE, "%s:ppc2 %p MPA_REQ_SENT DEAD", 596 __func__, ep); 597 ep->com.state = DEAD; 598 connect_reply_upcall(ep, -ECONNABORTED); 599 600 disconnect = 0; 601 STOP_EP_TIMER(ep); 602 close_socket(ep->com.so); 603 deref_cm_id(&ep->com); 604 release = 1; 605 break; 606 607 case MPA_REQ_RCVD: 608 609 /* 610 * We're gonna mark this puppy DEAD, but keep 611 * the reference on it until the ULP accepts or 612 * rejects the CR. 613 */ 614 CTR2(KTR_IW_CXGBE, "%s:ppc3 %p MPA_REQ_RCVD CLOSING", 615 __func__, ep); 616 ep->com.state = CLOSING; 617 break; 618 619 case MPA_REP_SENT: 620 CTR2(KTR_IW_CXGBE, "%s:ppc4 %p MPA_REP_SENT CLOSING", 621 __func__, ep); 622 ep->com.state = CLOSING; 623 break; 624 625 case FPDU_MODE: 626 CTR2(KTR_IW_CXGBE, "%s:ppc5 %p FPDU_MODE CLOSING", 627 __func__, ep); 628 START_EP_TIMER(ep); 629 ep->com.state = CLOSING; 630 attrs.next_state = C4IW_QP_STATE_CLOSING; 631 c4iw_modify_qp(ep->com.dev, ep->com.qp, 632 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1); 633 peer_close_upcall(ep); 634 break; 635 636 case ABORTING: 637 CTR2(KTR_IW_CXGBE, "%s:ppc6 %p ABORTING (disconn)", 638 __func__, ep); 639 disconnect = 0; 640 break; 641 642 case CLOSING: 643 CTR2(KTR_IW_CXGBE, "%s:ppc7 %p CLOSING MORIBUND", 644 __func__, ep); 645 ep->com.state = MORIBUND; 646 disconnect = 0; 647 break; 648 649 case MORIBUND: 650 CTR2(KTR_IW_CXGBE, "%s:ppc8 %p MORIBUND DEAD", __func__, 651 ep); 652 STOP_EP_TIMER(ep); 653 if (ep->com.cm_id && ep->com.qp) { 654 attrs.next_state = C4IW_QP_STATE_IDLE; 655 c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp, 656 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1); 657 } 658 close_socket(ep->com.so); 659 close_complete_upcall(ep, 0); 660 ep->com.state = DEAD; 661 release = 1; 662 disconnect = 0; 663 break; 664 665 case DEAD: 666 CTR2(KTR_IW_CXGBE, "%s:ppc9 %p DEAD (disconn)", 667 __func__, ep); 668 disconnect = 0; 669 break; 670 671 default: 672 panic("%s: ep %p state %d", __func__, ep, 673 ep->com.state); 674 break; 675 } 676 677 678 if (disconnect) { 679 680 CTR2(KTR_IW_CXGBE, "%s:ppca %p", __func__, ep); 681 c4iw_ep_disconnect(ep, 0, M_NOWAIT); 682 } 683 if (release) { 684 685 CTR2(KTR_IW_CXGBE, "%s:ppcb %p", __func__, ep); 686 c4iw_put_ep(&ep->com); 687 } 688 CTR2(KTR_IW_CXGBE, "%s:ppcE %p", __func__, ep); 689 return; 690 } 691 692 static void 693 process_conn_error(struct c4iw_ep *ep) 694 { 695 struct c4iw_qp_attributes attrs = {0}; 696 int ret; 697 int state; 698 699 state = ep->com.state; 700 CTR5(KTR_IW_CXGBE, "%s:pceB ep %p so %p so->so_error %u state %s", 701 __func__, ep, ep->com.so, ep->com.so->so_error, 702 states[ep->com.state]); 703 704 switch (state) { 705 706 case MPA_REQ_WAIT: 707 STOP_EP_TIMER(ep); 708 c4iw_put_ep(&ep->parent_ep->com); 709 break; 710 711 case MPA_REQ_SENT: 712 STOP_EP_TIMER(ep); 713 connect_reply_upcall(ep, -ECONNRESET); 714 break; 715 716 case MPA_REP_SENT: 717 ep->com.rpl_err = ECONNRESET; 718 CTR1(KTR_IW_CXGBE, "waking up ep %p", ep); 719 break; 720 721 case MPA_REQ_RCVD: 722 break; 723 724 case MORIBUND: 725 case CLOSING: 726 STOP_EP_TIMER(ep); 727 /*FALLTHROUGH*/ 728 case FPDU_MODE: 729 730 if (ep->com.cm_id && ep->com.qp) { 731 732 attrs.next_state = C4IW_QP_STATE_ERROR; 733 ret = c4iw_modify_qp(ep->com.qp->rhp, 734 ep->com.qp, C4IW_QP_ATTR_NEXT_STATE, 735 &attrs, 1); 736 if (ret) 737 log(LOG_ERR, 738 "%s - qp <- error failed!\n", 739 __func__); 740 } 741 peer_abort_upcall(ep); 742 break; 743 744 case ABORTING: 745 break; 746 747 case DEAD: 748 CTR2(KTR_IW_CXGBE, "%s so_error %d IN DEAD STATE!!!!", 749 __func__, ep->com.so->so_error); 750 return; 751 752 default: 753 panic("%s: ep %p state %d", __func__, ep, state); 754 break; 755 } 756 757 if (state != ABORTING) { 758 close_socket(ep->com.so); 759 ep->com.state = DEAD; 760 c4iw_put_ep(&ep->com); 761 } 762 CTR2(KTR_IW_CXGBE, "%s:pceE %p", __func__, ep); 763 return; 764 } 765 766 static void 767 process_close_complete(struct c4iw_ep *ep) 768 { 769 struct c4iw_qp_attributes attrs = {0}; 770 int release = 0; 771 772 CTR4(KTR_IW_CXGBE, "%s:pccB ep %p so %p state %s", __func__, ep, 773 ep->com.so, states[ep->com.state]); 774 775 /* The cm_id may be null if we failed to connect */ 776 set_bit(CLOSE_CON_RPL, &ep->com.history); 777 778 switch (ep->com.state) { 779 780 case CLOSING: 781 CTR2(KTR_IW_CXGBE, "%s:pcc1 %p CLOSING MORIBUND", 782 __func__, ep); 783 ep->com.state = MORIBUND; 784 break; 785 786 case MORIBUND: 787 CTR2(KTR_IW_CXGBE, "%s:pcc1 %p MORIBUND DEAD", __func__, 788 ep); 789 STOP_EP_TIMER(ep); 790 791 if ((ep->com.cm_id) && (ep->com.qp)) { 792 793 CTR2(KTR_IW_CXGBE, "%s:pcc2 %p QP_STATE_IDLE", 794 __func__, ep); 795 attrs.next_state = C4IW_QP_STATE_IDLE; 796 c4iw_modify_qp(ep->com.dev, 797 ep->com.qp, 798 C4IW_QP_ATTR_NEXT_STATE, 799 &attrs, 1); 800 } 801 802 close_socket(ep->com.so); 803 close_complete_upcall(ep, 0); 804 ep->com.state = DEAD; 805 release = 1; 806 break; 807 808 case ABORTING: 809 CTR2(KTR_IW_CXGBE, "%s:pcc5 %p ABORTING", __func__, ep); 810 break; 811 812 case DEAD: 813 CTR2(KTR_IW_CXGBE, "%s:pcc6 %p DEAD", __func__, ep); 814 break; 815 default: 816 CTR2(KTR_IW_CXGBE, "%s:pcc7 %p unknown ep state", 817 __func__, ep); 818 panic("%s:pcc6 %p unknown ep state", __func__, ep); 819 break; 820 } 821 822 if (release) { 823 824 CTR2(KTR_IW_CXGBE, "%s:pcc8 %p", __func__, ep); 825 release_ep_resources(ep); 826 } 827 CTR2(KTR_IW_CXGBE, "%s:pccE %p", __func__, ep); 828 return; 829 } 830 831 static void 832 setiwsockopt(struct socket *so) 833 { 834 int rc; 835 struct sockopt sopt; 836 int on = 1; 837 838 sopt.sopt_dir = SOPT_SET; 839 sopt.sopt_level = IPPROTO_TCP; 840 sopt.sopt_name = TCP_NODELAY; 841 sopt.sopt_val = (caddr_t)&on; 842 sopt.sopt_valsize = sizeof on; 843 sopt.sopt_td = NULL; 844 rc = sosetopt(so, &sopt); 845 if (rc) { 846 log(LOG_ERR, "%s: can't set TCP_NODELAY on so %p (%d)\n", 847 __func__, so, rc); 848 } 849 } 850 851 static void 852 init_iwarp_socket(struct socket *so, void *arg) 853 { 854 if (SOLISTENING(so)) { 855 SOLISTEN_LOCK(so); 856 solisten_upcall_set(so, c4iw_so_upcall, arg); 857 so->so_state |= SS_NBIO; 858 SOLISTEN_UNLOCK(so); 859 } else { 860 SOCKBUF_LOCK(&so->so_rcv); 861 soupcall_set(so, SO_RCV, c4iw_so_upcall, arg); 862 so->so_state |= SS_NBIO; 863 SOCKBUF_UNLOCK(&so->so_rcv); 864 } 865 } 866 867 static void 868 uninit_iwarp_socket(struct socket *so) 869 { 870 if (SOLISTENING(so)) { 871 SOLISTEN_LOCK(so); 872 solisten_upcall_set(so, NULL, NULL); 873 SOLISTEN_UNLOCK(so); 874 } else { 875 SOCKBUF_LOCK(&so->so_rcv); 876 soupcall_clear(so, SO_RCV); 877 SOCKBUF_UNLOCK(&so->so_rcv); 878 } 879 } 880 881 static void 882 process_data(struct c4iw_ep *ep) 883 { 884 int ret = 0; 885 int disconnect = 0; 886 struct c4iw_qp_attributes attrs = {0}; 887 888 CTR5(KTR_IW_CXGBE, "%s: so %p, ep %p, state %s, sbused %d", __func__, 889 ep->com.so, ep, states[ep->com.state], sbused(&ep->com.so->so_rcv)); 890 891 switch (ep->com.state) { 892 case MPA_REQ_SENT: 893 disconnect = process_mpa_reply(ep); 894 break; 895 case MPA_REQ_WAIT: 896 disconnect = process_mpa_request(ep); 897 if (disconnect) 898 /* Refered in process_newconn() */ 899 c4iw_put_ep(&ep->parent_ep->com); 900 break; 901 case FPDU_MODE: 902 MPASS(ep->com.qp != NULL); 903 attrs.next_state = C4IW_QP_STATE_TERMINATE; 904 ret = c4iw_modify_qp(ep->com.dev, ep->com.qp, 905 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1); 906 if (ret != -EINPROGRESS) 907 disconnect = 1; 908 break; 909 default: 910 log(LOG_ERR, "%s: Unexpected streaming data. ep %p, " 911 "state %d, so %p, so_state 0x%x, sbused %u\n", 912 __func__, ep, ep->com.state, ep->com.so, 913 ep->com.so->so_state, sbused(&ep->com.so->so_rcv)); 914 break; 915 } 916 if (disconnect) 917 c4iw_ep_disconnect(ep, disconnect == 2, GFP_KERNEL); 918 919 } 920 921 static void 922 process_connected(struct c4iw_ep *ep) 923 { 924 struct socket *so = ep->com.so; 925 926 if ((so->so_state & SS_ISCONNECTED) && !so->so_error) { 927 if (send_mpa_req(ep)) 928 goto err; 929 } else { 930 connect_reply_upcall(ep, -so->so_error); 931 goto err; 932 } 933 return; 934 err: 935 close_socket(so); 936 ep->com.state = DEAD; 937 c4iw_put_ep(&ep->com); 938 return; 939 } 940 941 static inline int c4iw_zero_addr(struct sockaddr *addr) 942 { 943 struct in6_addr *ip6; 944 945 if (addr->sa_family == AF_INET) 946 return IN_ZERONET( 947 ntohl(((struct sockaddr_in *)addr)->sin_addr.s_addr)); 948 else { 949 ip6 = &((struct sockaddr_in6 *) addr)->sin6_addr; 950 return (ip6->s6_addr32[0] | ip6->s6_addr32[1] | 951 ip6->s6_addr32[2] | ip6->s6_addr32[3]) == 0; 952 } 953 } 954 955 static inline int c4iw_loopback_addr(struct sockaddr *addr) 956 { 957 if (addr->sa_family == AF_INET) 958 return IN_LOOPBACK( 959 ntohl(((struct sockaddr_in *) addr)->sin_addr.s_addr)); 960 else 961 return IN6_IS_ADDR_LOOPBACK( 962 &((struct sockaddr_in6 *) addr)->sin6_addr); 963 } 964 965 static inline int c4iw_any_addr(struct sockaddr *addr) 966 { 967 return c4iw_zero_addr(addr) || c4iw_loopback_addr(addr); 968 } 969 970 static void 971 process_newconn(struct c4iw_listen_ep *master_lep, struct socket *new_so) 972 { 973 struct c4iw_listen_ep *real_lep = NULL; 974 struct c4iw_ep *new_ep = NULL; 975 struct sockaddr_in *remote = NULL; 976 int ret = 0; 977 978 MPASS(new_so != NULL); 979 980 if (c4iw_any_addr((struct sockaddr *)&master_lep->com.local_addr)) { 981 /* Here we need to find the 'real_lep' that belongs to the 982 * incomming socket's network interface, such that the newly 983 * created 'ep' can be attached to the real 'lep'. 984 */ 985 real_lep = find_real_listen_ep(master_lep, new_so); 986 if (real_lep == NULL) { 987 CTR2(KTR_IW_CXGBE, "%s: Could not find the real listen " 988 "ep for sock: %p", __func__, new_so); 989 log(LOG_ERR,"%s: Could not find the real listen ep for " 990 "sock: %p\n", __func__, new_so); 991 /* FIXME: properly free the 'new_so' in failure case. 992 * Use of soabort() and soclose() are not legal 993 * here(before soaccept()). 994 */ 995 return; 996 } 997 } else /* for Non-Wildcard address, master_lep is always the real_lep */ 998 real_lep = master_lep; 999 1000 new_ep = alloc_ep(sizeof(*new_ep), GFP_KERNEL); 1001 1002 CTR6(KTR_IW_CXGBE, "%s: master_lep %p, real_lep: %p, new ep %p, " 1003 "listening so %p, new so %p", __func__, master_lep, real_lep, 1004 new_ep, master_lep->com.so, new_so); 1005 1006 new_ep->com.dev = real_lep->com.dev; 1007 new_ep->com.so = new_so; 1008 new_ep->com.cm_id = NULL; 1009 new_ep->com.thread = real_lep->com.thread; 1010 new_ep->parent_ep = real_lep; 1011 1012 GET_LOCAL_ADDR(&new_ep->com.local_addr, new_so); 1013 GET_REMOTE_ADDR(&new_ep->com.remote_addr, new_so); 1014 c4iw_get_ep(&real_lep->com); 1015 init_timer(&new_ep->timer); 1016 new_ep->com.state = MPA_REQ_WAIT; 1017 START_EP_TIMER(new_ep); 1018 1019 setiwsockopt(new_so); 1020 ret = soaccept(new_so, (struct sockaddr **)&remote); 1021 if (ret != 0) { 1022 CTR4(KTR_IW_CXGBE, 1023 "%s:listen sock:%p, new sock:%p, ret:%d\n", 1024 __func__, master_lep->com.so, new_so, ret); 1025 if (remote != NULL) 1026 free(remote, M_SONAME); 1027 uninit_iwarp_socket(new_so); 1028 soclose(new_so); 1029 c4iw_put_ep(&new_ep->com); 1030 c4iw_put_ep(&real_lep->com); 1031 return; 1032 } 1033 free(remote, M_SONAME); 1034 1035 /* MPA request might have been queued up on the socket already, so we 1036 * initialize the socket/upcall_handler under lock to prevent processing 1037 * MPA request on another thread(via process_req()) simultaniously. 1038 */ 1039 c4iw_get_ep(&new_ep->com); /* Dereferenced at the end below, this is to 1040 avoid freeing of ep before ep unlock. */ 1041 mutex_lock(&new_ep->com.mutex); 1042 init_iwarp_socket(new_so, &new_ep->com); 1043 1044 ret = process_mpa_request(new_ep); 1045 if (ret) { 1046 /* ABORT */ 1047 c4iw_ep_disconnect(new_ep, 1, GFP_KERNEL); 1048 c4iw_put_ep(&real_lep->com); 1049 } 1050 mutex_unlock(&new_ep->com.mutex); 1051 c4iw_put_ep(&new_ep->com); 1052 return; 1053 } 1054 1055 static int 1056 add_ep_to_req_list(struct c4iw_ep *ep, int new_ep_event) 1057 { 1058 unsigned long flag; 1059 1060 spin_lock_irqsave(&req_lock, flag); 1061 if (ep && ep->com.so) { 1062 ep->com.ep_events |= new_ep_event; 1063 if (!ep->com.entry.tqe_prev) { 1064 c4iw_get_ep(&ep->com); 1065 TAILQ_INSERT_TAIL(&req_list, &ep->com, entry); 1066 queue_work(c4iw_taskq, &c4iw_task); 1067 } 1068 } 1069 spin_unlock_irqrestore(&req_lock, flag); 1070 1071 return (0); 1072 } 1073 1074 static int 1075 c4iw_so_upcall(struct socket *so, void *arg, int waitflag) 1076 { 1077 struct c4iw_ep *ep = arg; 1078 1079 CTR6(KTR_IW_CXGBE, 1080 "%s: so %p, so_state 0x%x, ep %p, ep_state %s, tqe_prev %p", 1081 __func__, so, so->so_state, ep, states[ep->com.state], 1082 ep->com.entry.tqe_prev); 1083 1084 MPASS(ep->com.so == so); 1085 /* 1086 * Wake up any threads waiting in rdma_init()/rdma_fini(), 1087 * with locks held. 1088 */ 1089 if (so->so_error) 1090 c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET); 1091 add_ep_to_req_list(ep, C4IW_EVENT_SOCKET); 1092 1093 return (SU_OK); 1094 } 1095 1096 1097 static int 1098 terminate(struct sge_iq *iq, const struct rss_header *rss, struct mbuf *m) 1099 { 1100 struct adapter *sc = iq->adapter; 1101 const struct cpl_rdma_terminate *cpl = mtod(m, const void *); 1102 unsigned int tid = GET_TID(cpl); 1103 struct toepcb *toep = lookup_tid(sc, tid); 1104 struct socket *so; 1105 struct c4iw_ep *ep; 1106 1107 INP_WLOCK(toep->inp); 1108 so = inp_inpcbtosocket(toep->inp); 1109 ep = so->so_rcv.sb_upcallarg; 1110 INP_WUNLOCK(toep->inp); 1111 1112 CTR3(KTR_IW_CXGBE, "%s: so %p, ep %p", __func__, so, ep); 1113 add_ep_to_req_list(ep, C4IW_EVENT_TERM); 1114 1115 return 0; 1116 } 1117 1118 static void 1119 process_socket_event(struct c4iw_ep *ep) 1120 { 1121 int state = ep->com.state; 1122 struct socket *so = ep->com.so; 1123 1124 if (ep->com.state == DEAD) { 1125 CTR3(KTR_IW_CXGBE, "%s: Pending socket event discarded " 1126 "ep %p ep_state %s", __func__, ep, states[state]); 1127 return; 1128 } 1129 1130 CTR6(KTR_IW_CXGBE, "process_socket_event: so %p, so_state 0x%x, " 1131 "so_err %d, sb_state 0x%x, ep %p, ep_state %s", so, so->so_state, 1132 so->so_error, so->so_rcv.sb_state, ep, states[state]); 1133 1134 if (state == CONNECTING) { 1135 process_connected(ep); 1136 return; 1137 } 1138 1139 if (state == LISTEN) { 1140 struct c4iw_listen_ep *lep = (struct c4iw_listen_ep *)ep; 1141 struct socket *listen_so = so, *new_so = NULL; 1142 int error = 0; 1143 1144 SOLISTEN_LOCK(listen_so); 1145 do { 1146 error = solisten_dequeue(listen_so, &new_so, 1147 SOCK_NONBLOCK); 1148 if (error) { 1149 CTR4(KTR_IW_CXGBE, "%s: lep %p listen_so %p " 1150 "error %d", __func__, lep, listen_so, 1151 error); 1152 return; 1153 } 1154 process_newconn(lep, new_so); 1155 1156 /* solisten_dequeue() unlocks while return, so aquire 1157 * lock again for sol_qlen and also for next iteration. 1158 */ 1159 SOLISTEN_LOCK(listen_so); 1160 } while (listen_so->sol_qlen); 1161 SOLISTEN_UNLOCK(listen_so); 1162 1163 return; 1164 } 1165 1166 /* connection error */ 1167 if (so->so_error) { 1168 process_conn_error(ep); 1169 return; 1170 } 1171 1172 /* peer close */ 1173 if ((so->so_rcv.sb_state & SBS_CANTRCVMORE) && state <= CLOSING) { 1174 process_peer_close(ep); 1175 /* 1176 * check whether socket disconnect event is pending before 1177 * returning. Fallthrough if yes. 1178 */ 1179 if (!(so->so_state & SS_ISDISCONNECTED)) 1180 return; 1181 } 1182 1183 /* close complete */ 1184 if (so->so_state & SS_ISDISCONNECTED) { 1185 process_close_complete(ep); 1186 return; 1187 } 1188 1189 /* rx data */ 1190 if (sbused(&ep->com.so->so_rcv)) { 1191 process_data(ep); 1192 return; 1193 } 1194 1195 /* Socket events for 'MPA Request Received' and 'Close Complete' 1196 * were already processed earlier in their previous events handlers. 1197 * Hence, these socket events are skipped. 1198 * And any other socket events must have handled above. 1199 */ 1200 MPASS((ep->com.state == MPA_REQ_RCVD) || (ep->com.state == MORIBUND)); 1201 1202 if ((ep->com.state != MPA_REQ_RCVD) && (ep->com.state != MORIBUND)) 1203 log(LOG_ERR, "%s: Unprocessed socket event so %p, " 1204 "so_state 0x%x, so_err %d, sb_state 0x%x, ep %p, ep_state %s\n", 1205 __func__, so, so->so_state, so->so_error, so->so_rcv.sb_state, 1206 ep, states[state]); 1207 1208 } 1209 1210 SYSCTL_NODE(_hw, OID_AUTO, iw_cxgbe, CTLFLAG_RD, 0, "iw_cxgbe driver parameters"); 1211 1212 static int dack_mode = 0; 1213 SYSCTL_INT(_hw_iw_cxgbe, OID_AUTO, dack_mode, CTLFLAG_RWTUN, &dack_mode, 0, 1214 "Delayed ack mode (default = 0)"); 1215 1216 int c4iw_max_read_depth = 8; 1217 SYSCTL_INT(_hw_iw_cxgbe, OID_AUTO, c4iw_max_read_depth, CTLFLAG_RWTUN, &c4iw_max_read_depth, 0, 1218 "Per-connection max ORD/IRD (default = 8)"); 1219 1220 static int enable_tcp_timestamps; 1221 SYSCTL_INT(_hw_iw_cxgbe, OID_AUTO, enable_tcp_timestamps, CTLFLAG_RWTUN, &enable_tcp_timestamps, 0, 1222 "Enable tcp timestamps (default = 0)"); 1223 1224 static int enable_tcp_sack; 1225 SYSCTL_INT(_hw_iw_cxgbe, OID_AUTO, enable_tcp_sack, CTLFLAG_RWTUN, &enable_tcp_sack, 0, 1226 "Enable tcp SACK (default = 0)"); 1227 1228 static int enable_tcp_window_scaling = 1; 1229 SYSCTL_INT(_hw_iw_cxgbe, OID_AUTO, enable_tcp_window_scaling, CTLFLAG_RWTUN, &enable_tcp_window_scaling, 0, 1230 "Enable tcp window scaling (default = 1)"); 1231 1232 int c4iw_debug = 0; 1233 SYSCTL_INT(_hw_iw_cxgbe, OID_AUTO, c4iw_debug, CTLFLAG_RWTUN, &c4iw_debug, 0, 1234 "Enable debug logging (default = 0)"); 1235 1236 static int peer2peer = 1; 1237 SYSCTL_INT(_hw_iw_cxgbe, OID_AUTO, peer2peer, CTLFLAG_RWTUN, &peer2peer, 0, 1238 "Support peer2peer ULPs (default = 1)"); 1239 1240 static int p2p_type = FW_RI_INIT_P2PTYPE_READ_REQ; 1241 SYSCTL_INT(_hw_iw_cxgbe, OID_AUTO, p2p_type, CTLFLAG_RWTUN, &p2p_type, 0, 1242 "RDMAP opcode to use for the RTR message: 1 = RDMA_READ 0 = RDMA_WRITE (default 1)"); 1243 1244 static int ep_timeout_secs = 60; 1245 SYSCTL_INT(_hw_iw_cxgbe, OID_AUTO, ep_timeout_secs, CTLFLAG_RWTUN, &ep_timeout_secs, 0, 1246 "CM Endpoint operation timeout in seconds (default = 60)"); 1247 1248 static int mpa_rev = 1; 1249 SYSCTL_INT(_hw_iw_cxgbe, OID_AUTO, mpa_rev, CTLFLAG_RWTUN, &mpa_rev, 0, 1250 "MPA Revision, 0 supports amso1100, 1 is RFC5044 spec compliant, 2 is IETF MPA Peer Connect Draft compliant (default = 1)"); 1251 1252 static int markers_enabled; 1253 SYSCTL_INT(_hw_iw_cxgbe, OID_AUTO, markers_enabled, CTLFLAG_RWTUN, &markers_enabled, 0, 1254 "Enable MPA MARKERS (default(0) = disabled)"); 1255 1256 static int crc_enabled = 1; 1257 SYSCTL_INT(_hw_iw_cxgbe, OID_AUTO, crc_enabled, CTLFLAG_RWTUN, &crc_enabled, 0, 1258 "Enable MPA CRC (default(1) = enabled)"); 1259 1260 static int rcv_win = 256 * 1024; 1261 SYSCTL_INT(_hw_iw_cxgbe, OID_AUTO, rcv_win, CTLFLAG_RWTUN, &rcv_win, 0, 1262 "TCP receive window in bytes (default = 256KB)"); 1263 1264 static int snd_win = 128 * 1024; 1265 SYSCTL_INT(_hw_iw_cxgbe, OID_AUTO, snd_win, CTLFLAG_RWTUN, &snd_win, 0, 1266 "TCP send window in bytes (default = 128KB)"); 1267 1268 static void 1269 start_ep_timer(struct c4iw_ep *ep) 1270 { 1271 1272 if (timer_pending(&ep->timer)) { 1273 CTR2(KTR_IW_CXGBE, "%s: ep %p, already started", __func__, ep); 1274 printk(KERN_ERR "%s timer already started! ep %p\n", __func__, 1275 ep); 1276 return; 1277 } 1278 clear_bit(TIMEOUT, &ep->com.flags); 1279 c4iw_get_ep(&ep->com); 1280 ep->timer.expires = jiffies + ep_timeout_secs * HZ; 1281 ep->timer.data = (unsigned long)ep; 1282 ep->timer.function = ep_timeout; 1283 add_timer(&ep->timer); 1284 } 1285 1286 static int 1287 stop_ep_timer(struct c4iw_ep *ep) 1288 { 1289 1290 del_timer_sync(&ep->timer); 1291 if (!test_and_set_bit(TIMEOUT, &ep->com.flags)) { 1292 c4iw_put_ep(&ep->com); 1293 return 0; 1294 } 1295 return 1; 1296 } 1297 1298 static void * 1299 alloc_ep(int size, gfp_t gfp) 1300 { 1301 struct c4iw_ep_common *epc; 1302 1303 epc = kzalloc(size, gfp); 1304 if (epc == NULL) 1305 return (NULL); 1306 1307 kref_init(&epc->kref); 1308 mutex_init(&epc->mutex); 1309 c4iw_init_wr_wait(&epc->wr_wait); 1310 1311 return (epc); 1312 } 1313 1314 void _c4iw_free_ep(struct kref *kref) 1315 { 1316 struct c4iw_ep *ep; 1317 struct c4iw_ep_common *epc; 1318 1319 ep = container_of(kref, struct c4iw_ep, com.kref); 1320 epc = &ep->com; 1321 KASSERT(!epc->entry.tqe_prev, ("%s epc %p still on req list", 1322 __func__, epc)); 1323 if (test_bit(QP_REFERENCED, &ep->com.flags)) 1324 deref_qp(ep); 1325 CTR4(KTR_IW_CXGBE, "%s: ep %p, history 0x%lx, flags 0x%lx", 1326 __func__, ep, epc->history, epc->flags); 1327 kfree(ep); 1328 } 1329 1330 static void release_ep_resources(struct c4iw_ep *ep) 1331 { 1332 CTR2(KTR_IW_CXGBE, "%s:rerB %p", __func__, ep); 1333 set_bit(RELEASE_RESOURCES, &ep->com.flags); 1334 c4iw_put_ep(&ep->com); 1335 CTR2(KTR_IW_CXGBE, "%s:rerE %p", __func__, ep); 1336 } 1337 1338 static int 1339 send_mpa_req(struct c4iw_ep *ep) 1340 { 1341 int mpalen; 1342 struct mpa_message *mpa; 1343 struct mpa_v2_conn_params mpa_v2_params; 1344 struct mbuf *m; 1345 char mpa_rev_to_use = mpa_rev; 1346 int err = 0; 1347 1348 if (ep->retry_with_mpa_v1) 1349 mpa_rev_to_use = 1; 1350 mpalen = sizeof(*mpa) + ep->plen; 1351 if (mpa_rev_to_use == 2) 1352 mpalen += sizeof(struct mpa_v2_conn_params); 1353 1354 mpa = malloc(mpalen, M_CXGBE, M_NOWAIT); 1355 if (mpa == NULL) { 1356 err = -ENOMEM; 1357 CTR3(KTR_IW_CXGBE, "%s:smr1 ep: %p , error: %d", 1358 __func__, ep, err); 1359 goto err; 1360 } 1361 1362 memset(mpa, 0, mpalen); 1363 memcpy(mpa->key, MPA_KEY_REQ, sizeof(mpa->key)); 1364 mpa->flags = (crc_enabled ? MPA_CRC : 0) | 1365 (markers_enabled ? MPA_MARKERS : 0) | 1366 (mpa_rev_to_use == 2 ? MPA_ENHANCED_RDMA_CONN : 0); 1367 mpa->private_data_size = htons(ep->plen); 1368 mpa->revision = mpa_rev_to_use; 1369 1370 if (mpa_rev_to_use == 1) { 1371 ep->tried_with_mpa_v1 = 1; 1372 ep->retry_with_mpa_v1 = 0; 1373 } 1374 1375 if (mpa_rev_to_use == 2) { 1376 mpa->private_data_size = htons(ntohs(mpa->private_data_size) + 1377 sizeof(struct mpa_v2_conn_params)); 1378 mpa_v2_params.ird = htons((u16)ep->ird); 1379 mpa_v2_params.ord = htons((u16)ep->ord); 1380 1381 if (peer2peer) { 1382 mpa_v2_params.ird |= htons(MPA_V2_PEER2PEER_MODEL); 1383 1384 if (p2p_type == FW_RI_INIT_P2PTYPE_RDMA_WRITE) { 1385 mpa_v2_params.ord |= 1386 htons(MPA_V2_RDMA_WRITE_RTR); 1387 } else if (p2p_type == FW_RI_INIT_P2PTYPE_READ_REQ) { 1388 mpa_v2_params.ord |= 1389 htons(MPA_V2_RDMA_READ_RTR); 1390 } 1391 } 1392 memcpy(mpa->private_data, &mpa_v2_params, 1393 sizeof(struct mpa_v2_conn_params)); 1394 1395 if (ep->plen) { 1396 1397 memcpy(mpa->private_data + 1398 sizeof(struct mpa_v2_conn_params), 1399 ep->mpa_pkt + sizeof(*mpa), ep->plen); 1400 } 1401 } else { 1402 1403 if (ep->plen) 1404 memcpy(mpa->private_data, 1405 ep->mpa_pkt + sizeof(*mpa), ep->plen); 1406 CTR2(KTR_IW_CXGBE, "%s:smr7 %p", __func__, ep); 1407 } 1408 1409 m = m_getm(NULL, mpalen, M_NOWAIT, MT_DATA); 1410 if (m == NULL) { 1411 err = -ENOMEM; 1412 CTR3(KTR_IW_CXGBE, "%s:smr2 ep: %p , error: %d", 1413 __func__, ep, err); 1414 free(mpa, M_CXGBE); 1415 goto err; 1416 } 1417 m_copyback(m, 0, mpalen, (void *)mpa); 1418 free(mpa, M_CXGBE); 1419 1420 err = -sosend(ep->com.so, NULL, NULL, m, NULL, MSG_DONTWAIT, 1421 ep->com.thread); 1422 if (err) { 1423 CTR3(KTR_IW_CXGBE, "%s:smr3 ep: %p , error: %d", 1424 __func__, ep, err); 1425 goto err; 1426 } 1427 1428 START_EP_TIMER(ep); 1429 ep->com.state = MPA_REQ_SENT; 1430 ep->mpa_attr.initiator = 1; 1431 CTR3(KTR_IW_CXGBE, "%s:smrE %p, error: %d", __func__, ep, err); 1432 return 0; 1433 err: 1434 connect_reply_upcall(ep, err); 1435 CTR3(KTR_IW_CXGBE, "%s:smrE %p, error: %d", __func__, ep, err); 1436 return err; 1437 } 1438 1439 static int send_mpa_reject(struct c4iw_ep *ep, const void *pdata, u8 plen) 1440 { 1441 int mpalen ; 1442 struct mpa_message *mpa; 1443 struct mpa_v2_conn_params mpa_v2_params; 1444 struct mbuf *m; 1445 int err; 1446 1447 CTR4(KTR_IW_CXGBE, "%s:smrejB %p %u %d", __func__, ep, ep->hwtid, 1448 ep->plen); 1449 1450 mpalen = sizeof(*mpa) + plen; 1451 1452 if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) { 1453 1454 mpalen += sizeof(struct mpa_v2_conn_params); 1455 CTR4(KTR_IW_CXGBE, "%s:smrej1 %p %u %d", __func__, ep, 1456 ep->mpa_attr.version, mpalen); 1457 } 1458 1459 mpa = malloc(mpalen, M_CXGBE, M_NOWAIT); 1460 if (mpa == NULL) 1461 return (-ENOMEM); 1462 1463 memset(mpa, 0, mpalen); 1464 memcpy(mpa->key, MPA_KEY_REP, sizeof(mpa->key)); 1465 mpa->flags = MPA_REJECT; 1466 mpa->revision = mpa_rev; 1467 mpa->private_data_size = htons(plen); 1468 1469 if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) { 1470 1471 mpa->flags |= MPA_ENHANCED_RDMA_CONN; 1472 mpa->private_data_size = htons(ntohs(mpa->private_data_size) + 1473 sizeof(struct mpa_v2_conn_params)); 1474 mpa_v2_params.ird = htons(((u16)ep->ird) | 1475 (peer2peer ? MPA_V2_PEER2PEER_MODEL : 1476 0)); 1477 mpa_v2_params.ord = htons(((u16)ep->ord) | (peer2peer ? 1478 (p2p_type == 1479 FW_RI_INIT_P2PTYPE_RDMA_WRITE ? 1480 MPA_V2_RDMA_WRITE_RTR : p2p_type == 1481 FW_RI_INIT_P2PTYPE_READ_REQ ? 1482 MPA_V2_RDMA_READ_RTR : 0) : 0)); 1483 memcpy(mpa->private_data, &mpa_v2_params, 1484 sizeof(struct mpa_v2_conn_params)); 1485 1486 if (ep->plen) 1487 memcpy(mpa->private_data + 1488 sizeof(struct mpa_v2_conn_params), pdata, plen); 1489 CTR5(KTR_IW_CXGBE, "%s:smrej3 %p %d %d %d", __func__, ep, 1490 mpa_v2_params.ird, mpa_v2_params.ord, ep->plen); 1491 } else 1492 if (plen) 1493 memcpy(mpa->private_data, pdata, plen); 1494 1495 m = m_getm(NULL, mpalen, M_NOWAIT, MT_DATA); 1496 if (m == NULL) { 1497 free(mpa, M_CXGBE); 1498 return (-ENOMEM); 1499 } 1500 m_copyback(m, 0, mpalen, (void *)mpa); 1501 free(mpa, M_CXGBE); 1502 1503 err = -sosend(ep->com.so, NULL, NULL, m, NULL, MSG_DONTWAIT, ep->com.thread); 1504 if (!err) 1505 ep->snd_seq += mpalen; 1506 CTR4(KTR_IW_CXGBE, "%s:smrejE %p %u %d", __func__, ep, ep->hwtid, err); 1507 return err; 1508 } 1509 1510 static int send_mpa_reply(struct c4iw_ep *ep, const void *pdata, u8 plen) 1511 { 1512 int mpalen; 1513 struct mpa_message *mpa; 1514 struct mbuf *m; 1515 struct mpa_v2_conn_params mpa_v2_params; 1516 int err; 1517 1518 CTR2(KTR_IW_CXGBE, "%s:smrepB %p", __func__, ep); 1519 1520 mpalen = sizeof(*mpa) + plen; 1521 1522 if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) { 1523 1524 CTR3(KTR_IW_CXGBE, "%s:smrep1 %p %d", __func__, ep, 1525 ep->mpa_attr.version); 1526 mpalen += sizeof(struct mpa_v2_conn_params); 1527 } 1528 1529 mpa = malloc(mpalen, M_CXGBE, M_NOWAIT); 1530 if (mpa == NULL) 1531 return (-ENOMEM); 1532 1533 memset(mpa, 0, sizeof(*mpa)); 1534 memcpy(mpa->key, MPA_KEY_REP, sizeof(mpa->key)); 1535 mpa->flags = (ep->mpa_attr.crc_enabled ? MPA_CRC : 0) | 1536 (markers_enabled ? MPA_MARKERS : 0); 1537 mpa->revision = ep->mpa_attr.version; 1538 mpa->private_data_size = htons(plen); 1539 1540 if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) { 1541 1542 mpa->flags |= MPA_ENHANCED_RDMA_CONN; 1543 mpa->private_data_size += 1544 htons(sizeof(struct mpa_v2_conn_params)); 1545 mpa_v2_params.ird = htons((u16)ep->ird); 1546 mpa_v2_params.ord = htons((u16)ep->ord); 1547 CTR5(KTR_IW_CXGBE, "%s:smrep3 %p %d %d %d", __func__, ep, 1548 ep->mpa_attr.version, mpa_v2_params.ird, mpa_v2_params.ord); 1549 1550 if (peer2peer && (ep->mpa_attr.p2p_type != 1551 FW_RI_INIT_P2PTYPE_DISABLED)) { 1552 1553 mpa_v2_params.ird |= htons(MPA_V2_PEER2PEER_MODEL); 1554 1555 if (p2p_type == FW_RI_INIT_P2PTYPE_RDMA_WRITE) { 1556 1557 mpa_v2_params.ord |= 1558 htons(MPA_V2_RDMA_WRITE_RTR); 1559 CTR5(KTR_IW_CXGBE, "%s:smrep4 %p %d %d %d", 1560 __func__, ep, p2p_type, mpa_v2_params.ird, 1561 mpa_v2_params.ord); 1562 } 1563 else if (p2p_type == FW_RI_INIT_P2PTYPE_READ_REQ) { 1564 1565 mpa_v2_params.ord |= 1566 htons(MPA_V2_RDMA_READ_RTR); 1567 CTR5(KTR_IW_CXGBE, "%s:smrep5 %p %d %d %d", 1568 __func__, ep, p2p_type, mpa_v2_params.ird, 1569 mpa_v2_params.ord); 1570 } 1571 } 1572 1573 memcpy(mpa->private_data, &mpa_v2_params, 1574 sizeof(struct mpa_v2_conn_params)); 1575 1576 if (ep->plen) 1577 memcpy(mpa->private_data + 1578 sizeof(struct mpa_v2_conn_params), pdata, plen); 1579 } else 1580 if (plen) 1581 memcpy(mpa->private_data, pdata, plen); 1582 1583 m = m_getm(NULL, mpalen, M_NOWAIT, MT_DATA); 1584 if (m == NULL) { 1585 free(mpa, M_CXGBE); 1586 return (-ENOMEM); 1587 } 1588 m_copyback(m, 0, mpalen, (void *)mpa); 1589 free(mpa, M_CXGBE); 1590 1591 1592 ep->com.state = MPA_REP_SENT; 1593 ep->snd_seq += mpalen; 1594 err = -sosend(ep->com.so, NULL, NULL, m, NULL, MSG_DONTWAIT, 1595 ep->com.thread); 1596 CTR3(KTR_IW_CXGBE, "%s:smrepE %p %d", __func__, ep, err); 1597 return err; 1598 } 1599 1600 1601 1602 static void close_complete_upcall(struct c4iw_ep *ep, int status) 1603 { 1604 struct iw_cm_event event; 1605 1606 CTR2(KTR_IW_CXGBE, "%s:ccuB %p", __func__, ep); 1607 memset(&event, 0, sizeof(event)); 1608 event.event = IW_CM_EVENT_CLOSE; 1609 event.status = status; 1610 1611 if (ep->com.cm_id) { 1612 1613 CTR2(KTR_IW_CXGBE, "%s:ccu1 %1", __func__, ep); 1614 ep->com.cm_id->event_handler(ep->com.cm_id, &event); 1615 deref_cm_id(&ep->com); 1616 set_bit(CLOSE_UPCALL, &ep->com.history); 1617 } 1618 CTR2(KTR_IW_CXGBE, "%s:ccuE %p", __func__, ep); 1619 } 1620 1621 static int 1622 send_abort(struct c4iw_ep *ep) 1623 { 1624 struct socket *so = ep->com.so; 1625 struct sockopt sopt; 1626 int rc; 1627 struct linger l; 1628 1629 CTR5(KTR_IW_CXGBE, "%s ep %p so %p state %s tid %d", __func__, ep, so, 1630 states[ep->com.state], ep->hwtid); 1631 1632 l.l_onoff = 1; 1633 l.l_linger = 0; 1634 1635 /* linger_time of 0 forces RST to be sent */ 1636 sopt.sopt_dir = SOPT_SET; 1637 sopt.sopt_level = SOL_SOCKET; 1638 sopt.sopt_name = SO_LINGER; 1639 sopt.sopt_val = (caddr_t)&l; 1640 sopt.sopt_valsize = sizeof l; 1641 sopt.sopt_td = NULL; 1642 rc = sosetopt(so, &sopt); 1643 if (rc != 0) { 1644 log(LOG_ERR, "%s: sosetopt(%p, linger = 0) failed with %d.\n", 1645 __func__, so, rc); 1646 } 1647 1648 uninit_iwarp_socket(so); 1649 soclose(so); 1650 set_bit(ABORT_CONN, &ep->com.history); 1651 1652 /* 1653 * TBD: iw_cxgbe driver should receive ABORT reply for every ABORT 1654 * request it has sent. But the current TOE driver is not propagating 1655 * this ABORT reply event (via do_abort_rpl) to iw_cxgbe. So as a work- 1656 * around de-refererece 'ep' here instead of doing it in abort_rpl() 1657 * handler(not yet implemented) of iw_cxgbe driver. 1658 */ 1659 release_ep_resources(ep); 1660 ep->com.state = DEAD; 1661 1662 return (0); 1663 } 1664 1665 static void peer_close_upcall(struct c4iw_ep *ep) 1666 { 1667 struct iw_cm_event event; 1668 1669 CTR2(KTR_IW_CXGBE, "%s:pcuB %p", __func__, ep); 1670 memset(&event, 0, sizeof(event)); 1671 event.event = IW_CM_EVENT_DISCONNECT; 1672 1673 if (ep->com.cm_id) { 1674 1675 CTR2(KTR_IW_CXGBE, "%s:pcu1 %p", __func__, ep); 1676 ep->com.cm_id->event_handler(ep->com.cm_id, &event); 1677 set_bit(DISCONN_UPCALL, &ep->com.history); 1678 } 1679 CTR2(KTR_IW_CXGBE, "%s:pcuE %p", __func__, ep); 1680 } 1681 1682 static void peer_abort_upcall(struct c4iw_ep *ep) 1683 { 1684 struct iw_cm_event event; 1685 1686 CTR2(KTR_IW_CXGBE, "%s:pauB %p", __func__, ep); 1687 memset(&event, 0, sizeof(event)); 1688 event.event = IW_CM_EVENT_CLOSE; 1689 event.status = -ECONNRESET; 1690 1691 if (ep->com.cm_id) { 1692 1693 CTR2(KTR_IW_CXGBE, "%s:pau1 %p", __func__, ep); 1694 ep->com.cm_id->event_handler(ep->com.cm_id, &event); 1695 deref_cm_id(&ep->com); 1696 set_bit(ABORT_UPCALL, &ep->com.history); 1697 } 1698 CTR2(KTR_IW_CXGBE, "%s:pauE %p", __func__, ep); 1699 } 1700 1701 static void connect_reply_upcall(struct c4iw_ep *ep, int status) 1702 { 1703 struct iw_cm_event event; 1704 1705 CTR3(KTR_IW_CXGBE, "%s:cruB %p, status: %d", __func__, ep, status); 1706 memset(&event, 0, sizeof(event)); 1707 event.event = IW_CM_EVENT_CONNECT_REPLY; 1708 event.status = ((status == -ECONNABORTED) || (status == -EPIPE)) ? 1709 -ECONNRESET : status; 1710 event.local_addr = ep->com.local_addr; 1711 event.remote_addr = ep->com.remote_addr; 1712 1713 if ((status == 0) || (status == -ECONNREFUSED)) { 1714 1715 if (!ep->tried_with_mpa_v1) { 1716 1717 CTR2(KTR_IW_CXGBE, "%s:cru1 %p", __func__, ep); 1718 /* this means MPA_v2 is used */ 1719 event.ord = ep->ird; 1720 event.ird = ep->ord; 1721 event.private_data_len = ep->plen - 1722 sizeof(struct mpa_v2_conn_params); 1723 event.private_data = ep->mpa_pkt + 1724 sizeof(struct mpa_message) + 1725 sizeof(struct mpa_v2_conn_params); 1726 } else { 1727 1728 CTR2(KTR_IW_CXGBE, "%s:cru2 %p", __func__, ep); 1729 /* this means MPA_v1 is used */ 1730 event.ord = c4iw_max_read_depth; 1731 event.ird = c4iw_max_read_depth; 1732 event.private_data_len = ep->plen; 1733 event.private_data = ep->mpa_pkt + 1734 sizeof(struct mpa_message); 1735 } 1736 } 1737 1738 if (ep->com.cm_id) { 1739 1740 CTR2(KTR_IW_CXGBE, "%s:cru3 %p", __func__, ep); 1741 set_bit(CONN_RPL_UPCALL, &ep->com.history); 1742 ep->com.cm_id->event_handler(ep->com.cm_id, &event); 1743 } 1744 1745 if(status == -ECONNABORTED) { 1746 1747 CTR3(KTR_IW_CXGBE, "%s:cruE %p %d", __func__, ep, status); 1748 return; 1749 } 1750 1751 if (status < 0) { 1752 1753 CTR3(KTR_IW_CXGBE, "%s:cru4 %p %d", __func__, ep, status); 1754 deref_cm_id(&ep->com); 1755 } 1756 1757 CTR2(KTR_IW_CXGBE, "%s:cruE %p", __func__, ep); 1758 } 1759 1760 static int connect_request_upcall(struct c4iw_ep *ep) 1761 { 1762 struct iw_cm_event event; 1763 int ret; 1764 1765 CTR3(KTR_IW_CXGBE, "%s: ep %p, mpa_v1 %d", __func__, ep, 1766 ep->tried_with_mpa_v1); 1767 1768 memset(&event, 0, sizeof(event)); 1769 event.event = IW_CM_EVENT_CONNECT_REQUEST; 1770 event.local_addr = ep->com.local_addr; 1771 event.remote_addr = ep->com.remote_addr; 1772 event.provider_data = ep; 1773 1774 if (!ep->tried_with_mpa_v1) { 1775 /* this means MPA_v2 is used */ 1776 event.ord = ep->ord; 1777 event.ird = ep->ird; 1778 event.private_data_len = ep->plen - 1779 sizeof(struct mpa_v2_conn_params); 1780 event.private_data = ep->mpa_pkt + sizeof(struct mpa_message) + 1781 sizeof(struct mpa_v2_conn_params); 1782 } else { 1783 1784 /* this means MPA_v1 is used. Send max supported */ 1785 event.ord = c4iw_max_read_depth; 1786 event.ird = c4iw_max_read_depth; 1787 event.private_data_len = ep->plen; 1788 event.private_data = ep->mpa_pkt + sizeof(struct mpa_message); 1789 } 1790 1791 c4iw_get_ep(&ep->com); 1792 ret = ep->parent_ep->com.cm_id->event_handler(ep->parent_ep->com.cm_id, 1793 &event); 1794 if(ret) { 1795 CTR3(KTR_IW_CXGBE, "%s: ep %p, Failure while notifying event to" 1796 " IWCM, err:%d", __func__, ep, ret); 1797 c4iw_put_ep(&ep->com); 1798 } else 1799 /* Dereference parent_ep only in success case. 1800 * In case of failure, parent_ep is dereferenced by the caller 1801 * of process_mpa_request(). 1802 */ 1803 c4iw_put_ep(&ep->parent_ep->com); 1804 1805 set_bit(CONNREQ_UPCALL, &ep->com.history); 1806 return ret; 1807 } 1808 1809 static void established_upcall(struct c4iw_ep *ep) 1810 { 1811 struct iw_cm_event event; 1812 1813 CTR2(KTR_IW_CXGBE, "%s:euB %p", __func__, ep); 1814 memset(&event, 0, sizeof(event)); 1815 event.event = IW_CM_EVENT_ESTABLISHED; 1816 event.ird = ep->ord; 1817 event.ord = ep->ird; 1818 1819 if (ep->com.cm_id) { 1820 1821 CTR2(KTR_IW_CXGBE, "%s:eu1 %p", __func__, ep); 1822 ep->com.cm_id->event_handler(ep->com.cm_id, &event); 1823 set_bit(ESTAB_UPCALL, &ep->com.history); 1824 } 1825 CTR2(KTR_IW_CXGBE, "%s:euE %p", __func__, ep); 1826 } 1827 1828 1829 #define RELAXED_IRD_NEGOTIATION 1 1830 1831 /* 1832 * process_mpa_reply - process streaming mode MPA reply 1833 * 1834 * Returns: 1835 * 1836 * 0 upon success indicating a connect request was delivered to the ULP 1837 * or the mpa request is incomplete but valid so far. 1838 * 1839 * 1 if a failure requires the caller to close the connection. 1840 * 1841 * 2 if a failure requires the caller to abort the connection. 1842 */ 1843 static int process_mpa_reply(struct c4iw_ep *ep) 1844 { 1845 struct mpa_message *mpa; 1846 struct mpa_v2_conn_params *mpa_v2_params; 1847 u16 plen; 1848 u16 resp_ird, resp_ord; 1849 u8 rtr_mismatch = 0, insuff_ird = 0; 1850 struct c4iw_qp_attributes attrs = {0}; 1851 enum c4iw_qp_attr_mask mask; 1852 int err; 1853 struct mbuf *top, *m; 1854 int flags = MSG_DONTWAIT; 1855 struct uio uio; 1856 int disconnect = 0; 1857 1858 CTR2(KTR_IW_CXGBE, "%s:pmrB %p", __func__, ep); 1859 1860 /* 1861 * Stop mpa timer. If it expired, then 1862 * we ignore the MPA reply. process_timeout() 1863 * will abort the connection. 1864 */ 1865 if (STOP_EP_TIMER(ep)) 1866 return 0; 1867 1868 uio.uio_resid = 1000000; 1869 uio.uio_td = ep->com.thread; 1870 err = soreceive(ep->com.so, NULL, &uio, &top, NULL, &flags); 1871 1872 if (err) { 1873 1874 if (err == EWOULDBLOCK) { 1875 1876 CTR2(KTR_IW_CXGBE, "%s:pmr1 %p", __func__, ep); 1877 START_EP_TIMER(ep); 1878 return 0; 1879 } 1880 err = -err; 1881 CTR2(KTR_IW_CXGBE, "%s:pmr2 %p", __func__, ep); 1882 goto err; 1883 } 1884 1885 if (ep->com.so->so_rcv.sb_mb) { 1886 1887 CTR2(KTR_IW_CXGBE, "%s:pmr3 %p", __func__, ep); 1888 printf("%s data after soreceive called! so %p sb_mb %p top %p\n", 1889 __func__, ep->com.so, ep->com.so->so_rcv.sb_mb, top); 1890 } 1891 1892 m = top; 1893 1894 do { 1895 1896 CTR2(KTR_IW_CXGBE, "%s:pmr4 %p", __func__, ep); 1897 /* 1898 * If we get more than the supported amount of private data 1899 * then we must fail this connection. 1900 */ 1901 if (ep->mpa_pkt_len + m->m_len > sizeof(ep->mpa_pkt)) { 1902 1903 CTR3(KTR_IW_CXGBE, "%s:pmr5 %p %d", __func__, ep, 1904 ep->mpa_pkt_len + m->m_len); 1905 err = (-EINVAL); 1906 goto err_stop_timer; 1907 } 1908 1909 /* 1910 * copy the new data into our accumulation buffer. 1911 */ 1912 m_copydata(m, 0, m->m_len, &(ep->mpa_pkt[ep->mpa_pkt_len])); 1913 ep->mpa_pkt_len += m->m_len; 1914 if (!m->m_next) 1915 m = m->m_nextpkt; 1916 else 1917 m = m->m_next; 1918 } while (m); 1919 1920 m_freem(top); 1921 /* 1922 * if we don't even have the mpa message, then bail. 1923 */ 1924 if (ep->mpa_pkt_len < sizeof(*mpa)) { 1925 return 0; 1926 } 1927 mpa = (struct mpa_message *) ep->mpa_pkt; 1928 1929 /* Validate MPA header. */ 1930 if (mpa->revision > mpa_rev) { 1931 1932 CTR4(KTR_IW_CXGBE, "%s:pmr6 %p %d %d", __func__, ep, 1933 mpa->revision, mpa_rev); 1934 printk(KERN_ERR MOD "%s MPA version mismatch. Local = %d, " 1935 " Received = %d\n", __func__, mpa_rev, mpa->revision); 1936 err = -EPROTO; 1937 goto err_stop_timer; 1938 } 1939 1940 if (memcmp(mpa->key, MPA_KEY_REP, sizeof(mpa->key))) { 1941 1942 CTR2(KTR_IW_CXGBE, "%s:pmr7 %p", __func__, ep); 1943 err = -EPROTO; 1944 goto err_stop_timer; 1945 } 1946 1947 plen = ntohs(mpa->private_data_size); 1948 1949 /* 1950 * Fail if there's too much private data. 1951 */ 1952 if (plen > MPA_MAX_PRIVATE_DATA) { 1953 1954 CTR2(KTR_IW_CXGBE, "%s:pmr8 %p", __func__, ep); 1955 err = -EPROTO; 1956 goto err_stop_timer; 1957 } 1958 1959 /* 1960 * If plen does not account for pkt size 1961 */ 1962 if (ep->mpa_pkt_len > (sizeof(*mpa) + plen)) { 1963 1964 CTR2(KTR_IW_CXGBE, "%s:pmr9 %p", __func__, ep); 1965 STOP_EP_TIMER(ep); 1966 err = -EPROTO; 1967 goto err_stop_timer; 1968 } 1969 1970 ep->plen = (u8) plen; 1971 1972 /* 1973 * If we don't have all the pdata yet, then bail. 1974 * We'll continue process when more data arrives. 1975 */ 1976 if (ep->mpa_pkt_len < (sizeof(*mpa) + plen)) { 1977 1978 CTR2(KTR_IW_CXGBE, "%s:pmra %p", __func__, ep); 1979 return 0; 1980 } 1981 1982 if (mpa->flags & MPA_REJECT) { 1983 1984 CTR2(KTR_IW_CXGBE, "%s:pmrb %p", __func__, ep); 1985 err = -ECONNREFUSED; 1986 goto err_stop_timer; 1987 } 1988 1989 /* 1990 * If we get here we have accumulated the entire mpa 1991 * start reply message including private data. And 1992 * the MPA header is valid. 1993 */ 1994 ep->com.state = FPDU_MODE; 1995 ep->mpa_attr.crc_enabled = (mpa->flags & MPA_CRC) | crc_enabled ? 1 : 0; 1996 ep->mpa_attr.recv_marker_enabled = markers_enabled; 1997 ep->mpa_attr.xmit_marker_enabled = mpa->flags & MPA_MARKERS ? 1 : 0; 1998 ep->mpa_attr.version = mpa->revision; 1999 ep->mpa_attr.p2p_type = FW_RI_INIT_P2PTYPE_DISABLED; 2000 2001 if (mpa->revision == 2) { 2002 2003 CTR2(KTR_IW_CXGBE, "%s:pmrc %p", __func__, ep); 2004 ep->mpa_attr.enhanced_rdma_conn = 2005 mpa->flags & MPA_ENHANCED_RDMA_CONN ? 1 : 0; 2006 2007 if (ep->mpa_attr.enhanced_rdma_conn) { 2008 2009 CTR2(KTR_IW_CXGBE, "%s:pmrd %p", __func__, ep); 2010 mpa_v2_params = (struct mpa_v2_conn_params *) 2011 (ep->mpa_pkt + sizeof(*mpa)); 2012 resp_ird = ntohs(mpa_v2_params->ird) & 2013 MPA_V2_IRD_ORD_MASK; 2014 resp_ord = ntohs(mpa_v2_params->ord) & 2015 MPA_V2_IRD_ORD_MASK; 2016 2017 /* 2018 * This is a double-check. Ideally, below checks are 2019 * not required since ird/ord stuff has been taken 2020 * care of in c4iw_accept_cr 2021 */ 2022 if (ep->ird < resp_ord) { 2023 if (RELAXED_IRD_NEGOTIATION && resp_ord <= 2024 ep->com.dev->rdev.adap->params.max_ordird_qp) 2025 ep->ird = resp_ord; 2026 else 2027 insuff_ird = 1; 2028 } else if (ep->ird > resp_ord) { 2029 ep->ird = resp_ord; 2030 } 2031 if (ep->ord > resp_ird) { 2032 if (RELAXED_IRD_NEGOTIATION) 2033 ep->ord = resp_ird; 2034 else 2035 insuff_ird = 1; 2036 } 2037 if (insuff_ird) { 2038 err = -ENOMEM; 2039 ep->ird = resp_ord; 2040 ep->ord = resp_ird; 2041 } 2042 2043 if (ntohs(mpa_v2_params->ird) & 2044 MPA_V2_PEER2PEER_MODEL) { 2045 2046 CTR2(KTR_IW_CXGBE, "%s:pmrf %p", __func__, ep); 2047 if (ntohs(mpa_v2_params->ord) & 2048 MPA_V2_RDMA_WRITE_RTR) { 2049 2050 CTR2(KTR_IW_CXGBE, "%s:pmrg %p", __func__, ep); 2051 ep->mpa_attr.p2p_type = 2052 FW_RI_INIT_P2PTYPE_RDMA_WRITE; 2053 } 2054 else if (ntohs(mpa_v2_params->ord) & 2055 MPA_V2_RDMA_READ_RTR) { 2056 2057 CTR2(KTR_IW_CXGBE, "%s:pmrh %p", __func__, ep); 2058 ep->mpa_attr.p2p_type = 2059 FW_RI_INIT_P2PTYPE_READ_REQ; 2060 } 2061 } 2062 } 2063 } else { 2064 2065 CTR2(KTR_IW_CXGBE, "%s:pmri %p", __func__, ep); 2066 2067 if (mpa->revision == 1) { 2068 2069 CTR2(KTR_IW_CXGBE, "%s:pmrj %p", __func__, ep); 2070 2071 if (peer2peer) { 2072 2073 CTR2(KTR_IW_CXGBE, "%s:pmrk %p", __func__, ep); 2074 ep->mpa_attr.p2p_type = p2p_type; 2075 } 2076 } 2077 } 2078 2079 if (set_tcpinfo(ep)) { 2080 2081 CTR2(KTR_IW_CXGBE, "%s:pmrl %p", __func__, ep); 2082 printf("%s set_tcpinfo error\n", __func__); 2083 err = -ECONNRESET; 2084 goto err; 2085 } 2086 2087 CTR6(KTR_IW_CXGBE, "%s - crc_enabled = %d, recv_marker_enabled = %d, " 2088 "xmit_marker_enabled = %d, version = %d p2p_type = %d", __func__, 2089 ep->mpa_attr.crc_enabled, ep->mpa_attr.recv_marker_enabled, 2090 ep->mpa_attr.xmit_marker_enabled, ep->mpa_attr.version, 2091 ep->mpa_attr.p2p_type); 2092 2093 /* 2094 * If responder's RTR does not match with that of initiator, assign 2095 * FW_RI_INIT_P2PTYPE_DISABLED in mpa attributes so that RTR is not 2096 * generated when moving QP to RTS state. 2097 * A TERM message will be sent after QP has moved to RTS state 2098 */ 2099 if ((ep->mpa_attr.version == 2) && peer2peer && 2100 (ep->mpa_attr.p2p_type != p2p_type)) { 2101 2102 CTR2(KTR_IW_CXGBE, "%s:pmrm %p", __func__, ep); 2103 ep->mpa_attr.p2p_type = FW_RI_INIT_P2PTYPE_DISABLED; 2104 rtr_mismatch = 1; 2105 } 2106 2107 2108 //ep->ofld_txq = TOEPCB(ep->com.so)->ofld_txq; 2109 attrs.mpa_attr = ep->mpa_attr; 2110 attrs.max_ird = ep->ird; 2111 attrs.max_ord = ep->ord; 2112 attrs.llp_stream_handle = ep; 2113 attrs.next_state = C4IW_QP_STATE_RTS; 2114 2115 mask = C4IW_QP_ATTR_NEXT_STATE | 2116 C4IW_QP_ATTR_LLP_STREAM_HANDLE | C4IW_QP_ATTR_MPA_ATTR | 2117 C4IW_QP_ATTR_MAX_IRD | C4IW_QP_ATTR_MAX_ORD; 2118 2119 /* bind QP and TID with INIT_WR */ 2120 err = c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp, mask, &attrs, 1); 2121 2122 if (err) { 2123 2124 CTR2(KTR_IW_CXGBE, "%s:pmrn %p", __func__, ep); 2125 goto err; 2126 } 2127 2128 /* 2129 * If responder's RTR requirement did not match with what initiator 2130 * supports, generate TERM message 2131 */ 2132 if (rtr_mismatch) { 2133 2134 CTR2(KTR_IW_CXGBE, "%s:pmro %p", __func__, ep); 2135 printk(KERN_ERR "%s: RTR mismatch, sending TERM\n", __func__); 2136 attrs.layer_etype = LAYER_MPA | DDP_LLP; 2137 attrs.ecode = MPA_NOMATCH_RTR; 2138 attrs.next_state = C4IW_QP_STATE_TERMINATE; 2139 attrs.send_term = 1; 2140 err = c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp, 2141 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1); 2142 err = -ENOMEM; 2143 disconnect = 1; 2144 goto out; 2145 } 2146 2147 /* 2148 * Generate TERM if initiator IRD is not sufficient for responder 2149 * provided ORD. Currently, we do the same behaviour even when 2150 * responder provided IRD is also not sufficient as regards to 2151 * initiator ORD. 2152 */ 2153 if (insuff_ird) { 2154 2155 CTR2(KTR_IW_CXGBE, "%s:pmrp %p", __func__, ep); 2156 printk(KERN_ERR "%s: Insufficient IRD, sending TERM\n", 2157 __func__); 2158 attrs.layer_etype = LAYER_MPA | DDP_LLP; 2159 attrs.ecode = MPA_INSUFF_IRD; 2160 attrs.next_state = C4IW_QP_STATE_TERMINATE; 2161 attrs.send_term = 1; 2162 err = c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp, 2163 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1); 2164 err = -ENOMEM; 2165 disconnect = 1; 2166 goto out; 2167 } 2168 goto out; 2169 err_stop_timer: 2170 STOP_EP_TIMER(ep); 2171 err: 2172 disconnect = 2; 2173 out: 2174 connect_reply_upcall(ep, err); 2175 CTR2(KTR_IW_CXGBE, "%s:pmrE %p", __func__, ep); 2176 return disconnect; 2177 } 2178 2179 /* 2180 * process_mpa_request - process streaming mode MPA request 2181 * 2182 * Returns: 2183 * 2184 * 0 upon success indicating a connect request was delivered to the ULP 2185 * or the mpa request is incomplete but valid so far. 2186 * 2187 * 1 if a failure requires the caller to close the connection. 2188 * 2189 * 2 if a failure requires the caller to abort the connection. 2190 */ 2191 static int 2192 process_mpa_request(struct c4iw_ep *ep) 2193 { 2194 struct mpa_message *mpa; 2195 struct mpa_v2_conn_params *mpa_v2_params; 2196 u16 plen; 2197 int flags = MSG_DONTWAIT; 2198 int rc; 2199 struct iovec iov; 2200 struct uio uio; 2201 enum c4iw_ep_state state = ep->com.state; 2202 2203 CTR3(KTR_IW_CXGBE, "%s: ep %p, state %s", __func__, ep, states[state]); 2204 2205 if (state != MPA_REQ_WAIT) 2206 return 0; 2207 2208 iov.iov_base = &ep->mpa_pkt[ep->mpa_pkt_len]; 2209 iov.iov_len = sizeof(ep->mpa_pkt) - ep->mpa_pkt_len; 2210 uio.uio_iov = &iov; 2211 uio.uio_iovcnt = 1; 2212 uio.uio_offset = 0; 2213 uio.uio_resid = sizeof(ep->mpa_pkt) - ep->mpa_pkt_len; 2214 uio.uio_segflg = UIO_SYSSPACE; 2215 uio.uio_rw = UIO_READ; 2216 uio.uio_td = NULL; /* uio.uio_td = ep->com.thread; */ 2217 2218 rc = soreceive(ep->com.so, NULL, &uio, NULL, NULL, &flags); 2219 if (rc == EAGAIN) 2220 return 0; 2221 else if (rc) 2222 goto err_stop_timer; 2223 2224 KASSERT(uio.uio_offset > 0, ("%s: sorecieve on so %p read no data", 2225 __func__, ep->com.so)); 2226 ep->mpa_pkt_len += uio.uio_offset; 2227 2228 /* 2229 * If we get more than the supported amount of private data then we must 2230 * fail this connection. XXX: check so_rcv->sb_cc, or peek with another 2231 * soreceive, or increase the size of mpa_pkt by 1 and abort if the last 2232 * byte is filled by the soreceive above. 2233 */ 2234 2235 /* Don't even have the MPA message. Wait for more data to arrive. */ 2236 if (ep->mpa_pkt_len < sizeof(*mpa)) 2237 return 0; 2238 mpa = (struct mpa_message *) ep->mpa_pkt; 2239 2240 /* 2241 * Validate MPA Header. 2242 */ 2243 if (mpa->revision > mpa_rev) { 2244 log(LOG_ERR, "%s: MPA version mismatch. Local = %d," 2245 " Received = %d\n", __func__, mpa_rev, mpa->revision); 2246 goto err_stop_timer; 2247 } 2248 2249 if (memcmp(mpa->key, MPA_KEY_REQ, sizeof(mpa->key))) 2250 goto err_stop_timer; 2251 2252 /* 2253 * Fail if there's too much private data. 2254 */ 2255 plen = ntohs(mpa->private_data_size); 2256 if (plen > MPA_MAX_PRIVATE_DATA) 2257 goto err_stop_timer; 2258 2259 /* 2260 * If plen does not account for pkt size 2261 */ 2262 if (ep->mpa_pkt_len > (sizeof(*mpa) + plen)) 2263 goto err_stop_timer; 2264 2265 ep->plen = (u8) plen; 2266 2267 /* 2268 * If we don't have all the pdata yet, then bail. 2269 */ 2270 if (ep->mpa_pkt_len < (sizeof(*mpa) + plen)) 2271 return 0; 2272 2273 /* 2274 * If we get here we have accumulated the entire mpa 2275 * start reply message including private data. 2276 */ 2277 ep->mpa_attr.initiator = 0; 2278 ep->mpa_attr.crc_enabled = (mpa->flags & MPA_CRC) | crc_enabled ? 1 : 0; 2279 ep->mpa_attr.recv_marker_enabled = markers_enabled; 2280 ep->mpa_attr.xmit_marker_enabled = mpa->flags & MPA_MARKERS ? 1 : 0; 2281 ep->mpa_attr.version = mpa->revision; 2282 if (mpa->revision == 1) 2283 ep->tried_with_mpa_v1 = 1; 2284 ep->mpa_attr.p2p_type = FW_RI_INIT_P2PTYPE_DISABLED; 2285 2286 if (mpa->revision == 2) { 2287 ep->mpa_attr.enhanced_rdma_conn = 2288 mpa->flags & MPA_ENHANCED_RDMA_CONN ? 1 : 0; 2289 if (ep->mpa_attr.enhanced_rdma_conn) { 2290 mpa_v2_params = (struct mpa_v2_conn_params *) 2291 (ep->mpa_pkt + sizeof(*mpa)); 2292 ep->ird = ntohs(mpa_v2_params->ird) & 2293 MPA_V2_IRD_ORD_MASK; 2294 ep->ird = min_t(u32, ep->ird, 2295 cur_max_read_depth(ep->com.dev)); 2296 ep->ord = ntohs(mpa_v2_params->ord) & 2297 MPA_V2_IRD_ORD_MASK; 2298 ep->ord = min_t(u32, ep->ord, 2299 cur_max_read_depth(ep->com.dev)); 2300 CTR3(KTR_IW_CXGBE, "%s initiator ird %u ord %u\n", 2301 __func__, ep->ird, ep->ord); 2302 if (ntohs(mpa_v2_params->ird) & MPA_V2_PEER2PEER_MODEL) 2303 if (peer2peer) { 2304 if (ntohs(mpa_v2_params->ord) & 2305 MPA_V2_RDMA_WRITE_RTR) 2306 ep->mpa_attr.p2p_type = 2307 FW_RI_INIT_P2PTYPE_RDMA_WRITE; 2308 else if (ntohs(mpa_v2_params->ord) & 2309 MPA_V2_RDMA_READ_RTR) 2310 ep->mpa_attr.p2p_type = 2311 FW_RI_INIT_P2PTYPE_READ_REQ; 2312 } 2313 } 2314 } else if (mpa->revision == 1 && peer2peer) 2315 ep->mpa_attr.p2p_type = p2p_type; 2316 2317 if (set_tcpinfo(ep)) 2318 goto err_stop_timer; 2319 2320 CTR5(KTR_IW_CXGBE, "%s: crc_enabled = %d, recv_marker_enabled = %d, " 2321 "xmit_marker_enabled = %d, version = %d", __func__, 2322 ep->mpa_attr.crc_enabled, ep->mpa_attr.recv_marker_enabled, 2323 ep->mpa_attr.xmit_marker_enabled, ep->mpa_attr.version); 2324 2325 ep->com.state = MPA_REQ_RCVD; 2326 STOP_EP_TIMER(ep); 2327 2328 /* drive upcall */ 2329 if (ep->parent_ep->com.state != DEAD) 2330 if (connect_request_upcall(ep)) 2331 goto err_out; 2332 return 0; 2333 2334 err_stop_timer: 2335 STOP_EP_TIMER(ep); 2336 err_out: 2337 return 2; 2338 } 2339 2340 /* 2341 * Upcall from the adapter indicating data has been transmitted. 2342 * For us its just the single MPA request or reply. We can now free 2343 * the skb holding the mpa message. 2344 */ 2345 int c4iw_reject_cr(struct iw_cm_id *cm_id, const void *pdata, u8 pdata_len) 2346 { 2347 int err; 2348 struct c4iw_ep *ep = to_ep(cm_id); 2349 int abort = 0; 2350 2351 mutex_lock(&ep->com.mutex); 2352 CTR2(KTR_IW_CXGBE, "%s:crcB %p", __func__, ep); 2353 2354 if ((ep->com.state == DEAD) || 2355 (ep->com.state != MPA_REQ_RCVD)) { 2356 2357 CTR2(KTR_IW_CXGBE, "%s:crc1 %p", __func__, ep); 2358 mutex_unlock(&ep->com.mutex); 2359 c4iw_put_ep(&ep->com); 2360 return -ECONNRESET; 2361 } 2362 set_bit(ULP_REJECT, &ep->com.history); 2363 2364 if (mpa_rev == 0) { 2365 2366 CTR2(KTR_IW_CXGBE, "%s:crc2 %p", __func__, ep); 2367 abort = 1; 2368 } 2369 else { 2370 2371 CTR2(KTR_IW_CXGBE, "%s:crc3 %p", __func__, ep); 2372 abort = send_mpa_reject(ep, pdata, pdata_len); 2373 } 2374 STOP_EP_TIMER(ep); 2375 err = c4iw_ep_disconnect(ep, abort != 0, GFP_KERNEL); 2376 mutex_unlock(&ep->com.mutex); 2377 c4iw_put_ep(&ep->com); 2378 CTR3(KTR_IW_CXGBE, "%s:crc4 %p, err: %d", __func__, ep, err); 2379 return 0; 2380 } 2381 2382 int c4iw_accept_cr(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param) 2383 { 2384 int err; 2385 struct c4iw_qp_attributes attrs = {0}; 2386 enum c4iw_qp_attr_mask mask; 2387 struct c4iw_ep *ep = to_ep(cm_id); 2388 struct c4iw_dev *h = to_c4iw_dev(cm_id->device); 2389 struct c4iw_qp *qp = get_qhp(h, conn_param->qpn); 2390 int abort = 0; 2391 2392 mutex_lock(&ep->com.mutex); 2393 CTR2(KTR_IW_CXGBE, "%s:cacB %p", __func__, ep); 2394 2395 if ((ep->com.state == DEAD) || 2396 (ep->com.state != MPA_REQ_RCVD)) { 2397 2398 CTR2(KTR_IW_CXGBE, "%s:cac1 %p", __func__, ep); 2399 err = -ECONNRESET; 2400 goto err_out; 2401 } 2402 2403 BUG_ON(!qp); 2404 2405 set_bit(ULP_ACCEPT, &ep->com.history); 2406 2407 if ((conn_param->ord > c4iw_max_read_depth) || 2408 (conn_param->ird > c4iw_max_read_depth)) { 2409 2410 CTR2(KTR_IW_CXGBE, "%s:cac2 %p", __func__, ep); 2411 err = -EINVAL; 2412 goto err_abort; 2413 } 2414 2415 if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) { 2416 2417 CTR2(KTR_IW_CXGBE, "%s:cac3 %p", __func__, ep); 2418 2419 if (conn_param->ord > ep->ird) { 2420 if (RELAXED_IRD_NEGOTIATION) { 2421 conn_param->ord = ep->ird; 2422 } else { 2423 ep->ird = conn_param->ird; 2424 ep->ord = conn_param->ord; 2425 send_mpa_reject(ep, conn_param->private_data, 2426 conn_param->private_data_len); 2427 err = -ENOMEM; 2428 goto err_abort; 2429 } 2430 } 2431 if (conn_param->ird < ep->ord) { 2432 if (RELAXED_IRD_NEGOTIATION && 2433 ep->ord <= h->rdev.adap->params.max_ordird_qp) { 2434 conn_param->ird = ep->ord; 2435 } else { 2436 err = -ENOMEM; 2437 goto err_abort; 2438 } 2439 } 2440 } 2441 ep->ird = conn_param->ird; 2442 ep->ord = conn_param->ord; 2443 2444 if (ep->mpa_attr.version == 1) { 2445 if (peer2peer && ep->ird == 0) 2446 ep->ird = 1; 2447 } else { 2448 if (peer2peer && 2449 (ep->mpa_attr.p2p_type != FW_RI_INIT_P2PTYPE_DISABLED) && 2450 (p2p_type == FW_RI_INIT_P2PTYPE_READ_REQ) && ep->ird == 0) 2451 ep->ird = 1; 2452 } 2453 2454 CTR4(KTR_IW_CXGBE, "%s %d ird %d ord %d\n", __func__, __LINE__, 2455 ep->ird, ep->ord); 2456 2457 ep->com.cm_id = cm_id; 2458 ref_cm_id(&ep->com); 2459 ep->com.qp = qp; 2460 ref_qp(ep); 2461 //ep->ofld_txq = TOEPCB(ep->com.so)->ofld_txq; 2462 2463 /* bind QP to EP and move to RTS */ 2464 attrs.mpa_attr = ep->mpa_attr; 2465 attrs.max_ird = ep->ird; 2466 attrs.max_ord = ep->ord; 2467 attrs.llp_stream_handle = ep; 2468 attrs.next_state = C4IW_QP_STATE_RTS; 2469 2470 /* bind QP and TID with INIT_WR */ 2471 mask = C4IW_QP_ATTR_NEXT_STATE | 2472 C4IW_QP_ATTR_LLP_STREAM_HANDLE | 2473 C4IW_QP_ATTR_MPA_ATTR | 2474 C4IW_QP_ATTR_MAX_IRD | 2475 C4IW_QP_ATTR_MAX_ORD; 2476 2477 err = c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp, mask, &attrs, 1); 2478 if (err) { 2479 CTR3(KTR_IW_CXGBE, "%s:caca %p, err: %d", __func__, ep, err); 2480 goto err_defef_cm_id; 2481 } 2482 2483 err = send_mpa_reply(ep, conn_param->private_data, 2484 conn_param->private_data_len); 2485 if (err) { 2486 CTR3(KTR_IW_CXGBE, "%s:cacb %p, err: %d", __func__, ep, err); 2487 goto err_defef_cm_id; 2488 } 2489 2490 ep->com.state = FPDU_MODE; 2491 established_upcall(ep); 2492 mutex_unlock(&ep->com.mutex); 2493 c4iw_put_ep(&ep->com); 2494 CTR2(KTR_IW_CXGBE, "%s:cacE %p", __func__, ep); 2495 return 0; 2496 err_defef_cm_id: 2497 deref_cm_id(&ep->com); 2498 err_abort: 2499 abort = 1; 2500 err_out: 2501 if (abort) 2502 c4iw_ep_disconnect(ep, 1, GFP_KERNEL); 2503 mutex_unlock(&ep->com.mutex); 2504 c4iw_put_ep(&ep->com); 2505 CTR2(KTR_IW_CXGBE, "%s:cacE err %p", __func__, ep); 2506 return err; 2507 } 2508 2509 static int 2510 c4iw_sock_create(struct sockaddr_storage *laddr, struct socket **so) 2511 { 2512 int ret; 2513 int size; 2514 struct socket *sock = NULL; 2515 2516 ret = sock_create_kern(laddr->ss_family, 2517 SOCK_STREAM, IPPROTO_TCP, &sock); 2518 if (ret) { 2519 CTR2(KTR_IW_CXGBE, "%s:Failed to create TCP socket. err %d", 2520 __func__, ret); 2521 return ret; 2522 } 2523 2524 ret = sobind(sock, (struct sockaddr *)laddr, curthread); 2525 if (ret) { 2526 CTR2(KTR_IW_CXGBE, "%s:Failed to bind socket. err %p", 2527 __func__, ret); 2528 sock_release(sock); 2529 return ret; 2530 } 2531 2532 size = laddr->ss_family == AF_INET6 ? 2533 sizeof(struct sockaddr_in6) : sizeof(struct sockaddr_in); 2534 ret = sock_getname(sock, (struct sockaddr *)laddr, &size, 0); 2535 if (ret) { 2536 CTR2(KTR_IW_CXGBE, "%s:sock_getname failed. err %p", 2537 __func__, ret); 2538 sock_release(sock); 2539 return ret; 2540 } 2541 2542 *so = sock; 2543 return 0; 2544 } 2545 2546 int c4iw_connect(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param) 2547 { 2548 int err = 0; 2549 struct c4iw_dev *dev = to_c4iw_dev(cm_id->device); 2550 struct c4iw_ep *ep = NULL; 2551 struct ifnet *nh_ifp; /* Logical egress interface */ 2552 #ifdef VIMAGE 2553 struct rdma_cm_id *rdma_id = (struct rdma_cm_id*)cm_id->context; 2554 struct vnet *vnet = rdma_id->route.addr.dev_addr.net; 2555 #endif 2556 2557 CTR2(KTR_IW_CXGBE, "%s:ccB %p", __func__, cm_id); 2558 2559 2560 if ((conn_param->ord > c4iw_max_read_depth) || 2561 (conn_param->ird > c4iw_max_read_depth)) { 2562 2563 CTR2(KTR_IW_CXGBE, "%s:cc1 %p", __func__, cm_id); 2564 err = -EINVAL; 2565 goto out; 2566 } 2567 ep = alloc_ep(sizeof(*ep), GFP_KERNEL); 2568 cm_id->provider_data = ep; 2569 2570 init_timer(&ep->timer); 2571 ep->plen = conn_param->private_data_len; 2572 2573 if (ep->plen) { 2574 2575 CTR2(KTR_IW_CXGBE, "%s:cc3 %p", __func__, ep); 2576 memcpy(ep->mpa_pkt + sizeof(struct mpa_message), 2577 conn_param->private_data, ep->plen); 2578 } 2579 ep->ird = conn_param->ird; 2580 ep->ord = conn_param->ord; 2581 2582 if (peer2peer && ep->ord == 0) { 2583 2584 CTR2(KTR_IW_CXGBE, "%s:cc4 %p", __func__, ep); 2585 ep->ord = 1; 2586 } 2587 2588 ep->com.dev = dev; 2589 ep->com.cm_id = cm_id; 2590 ref_cm_id(&ep->com); 2591 ep->com.qp = get_qhp(dev, conn_param->qpn); 2592 2593 if (!ep->com.qp) { 2594 2595 CTR2(KTR_IW_CXGBE, "%s:cc5 %p", __func__, ep); 2596 err = -EINVAL; 2597 goto fail; 2598 } 2599 ref_qp(ep); 2600 ep->com.thread = curthread; 2601 2602 CURVNET_SET(vnet); 2603 err = get_ifnet_from_raddr(&cm_id->remote_addr, &nh_ifp); 2604 CURVNET_RESTORE(); 2605 2606 if (err) { 2607 2608 CTR2(KTR_IW_CXGBE, "%s:cc7 %p", __func__, ep); 2609 printk(KERN_ERR MOD "%s - cannot find route.\n", __func__); 2610 err = EHOSTUNREACH; 2611 return err; 2612 } 2613 2614 if (!(nh_ifp->if_capenable & IFCAP_TOE) || 2615 TOEDEV(nh_ifp) == NULL) { 2616 err = -ENOPROTOOPT; 2617 goto fail; 2618 } 2619 ep->com.state = CONNECTING; 2620 ep->tos = 0; 2621 ep->com.local_addr = cm_id->local_addr; 2622 ep->com.remote_addr = cm_id->remote_addr; 2623 2624 err = c4iw_sock_create(&cm_id->local_addr, &ep->com.so); 2625 if (err) 2626 goto fail; 2627 2628 setiwsockopt(ep->com.so); 2629 init_iwarp_socket(ep->com.so, &ep->com); 2630 err = -soconnect(ep->com.so, (struct sockaddr *)&ep->com.remote_addr, 2631 ep->com.thread); 2632 if (err) 2633 goto fail_free_so; 2634 CTR2(KTR_IW_CXGBE, "%s:ccE, ep %p", __func__, ep); 2635 return 0; 2636 2637 fail_free_so: 2638 uninit_iwarp_socket(ep->com.so); 2639 ep->com.state = DEAD; 2640 sock_release(ep->com.so); 2641 fail: 2642 deref_cm_id(&ep->com); 2643 c4iw_put_ep(&ep->com); 2644 ep = NULL; 2645 out: 2646 CTR2(KTR_IW_CXGBE, "%s:ccE Error %d", __func__, err); 2647 return err; 2648 } 2649 2650 /* 2651 * iwcm->create_listen. Returns -errno on failure. 2652 */ 2653 int 2654 c4iw_create_listen(struct iw_cm_id *cm_id, int backlog) 2655 { 2656 struct c4iw_dev *dev = to_c4iw_dev(cm_id->device); 2657 struct c4iw_listen_ep *lep = NULL; 2658 struct listen_port_info *port_info = NULL; 2659 int rc = 0; 2660 2661 CTR3(KTR_IW_CXGBE, "%s: cm_id %p, backlog %s", __func__, cm_id, 2662 backlog); 2663 lep = alloc_ep(sizeof(*lep), GFP_KERNEL); 2664 lep->com.cm_id = cm_id; 2665 ref_cm_id(&lep->com); 2666 lep->com.dev = dev; 2667 lep->backlog = backlog; 2668 lep->com.local_addr = cm_id->local_addr; 2669 lep->com.thread = curthread; 2670 cm_id->provider_data = lep; 2671 lep->com.state = LISTEN; 2672 2673 /* In case of INDADDR_ANY, ibcore creates cmid for each device and 2674 * invokes iw_cxgbe listener callbacks assuming that iw_cxgbe creates 2675 * HW listeners for each device seperately. But toecore expects single 2676 * solisten() call with INADDR_ANY address to create HW listeners on 2677 * all devices for a given port number. So iw_cxgbe driver calls 2678 * solisten() only once for INADDR_ANY(usually done at first time 2679 * listener callback from ibcore). And all the subsequent INADDR_ANY 2680 * listener callbacks from ibcore(for the same port address) do not 2681 * invoke solisten() as first listener callback has already created 2682 * listeners for all other devices(via solisten). 2683 */ 2684 if (c4iw_any_addr((struct sockaddr *)&lep->com.local_addr)) { 2685 port_info = add_ep_to_listenlist(lep); 2686 /* skip solisten() if refcnt > 1, as the listeners were 2687 * alredy created by 'Master lep' 2688 */ 2689 if (port_info->refcnt > 1) { 2690 /* As there will be only one listener socket for a TCP 2691 * port, copy Master lep's socket pointer to other lep's 2692 * that are belonging to same TCP port. 2693 */ 2694 struct c4iw_listen_ep *head_lep = 2695 container_of(port_info->lep_list.next, 2696 struct c4iw_listen_ep, listen_ep_list); 2697 lep->com.so = head_lep->com.so; 2698 goto out; 2699 } 2700 } 2701 rc = c4iw_sock_create(&cm_id->local_addr, &lep->com.so); 2702 if (rc) { 2703 CTR2(KTR_IW_CXGBE, "%s:Failed to create socket. err %d", 2704 __func__, rc); 2705 goto fail; 2706 } 2707 2708 rc = solisten(lep->com.so, backlog, curthread); 2709 if (rc) { 2710 CTR3(KTR_IW_CXGBE, "%s:Failed to listen on sock:%p. err %d", 2711 __func__, lep->com.so, rc); 2712 goto fail_free_so; 2713 } 2714 init_iwarp_socket(lep->com.so, &lep->com); 2715 out: 2716 return 0; 2717 2718 fail_free_so: 2719 sock_release(lep->com.so); 2720 fail: 2721 if (port_info) 2722 rem_ep_from_listenlist(lep); 2723 deref_cm_id(&lep->com); 2724 c4iw_put_ep(&lep->com); 2725 return rc; 2726 } 2727 2728 int 2729 c4iw_destroy_listen(struct iw_cm_id *cm_id) 2730 { 2731 struct c4iw_listen_ep *lep = to_listen_ep(cm_id); 2732 2733 mutex_lock(&lep->com.mutex); 2734 CTR3(KTR_IW_CXGBE, "%s: cm_id %p, state %s", __func__, cm_id, 2735 states[lep->com.state]); 2736 2737 lep->com.state = DEAD; 2738 if (c4iw_any_addr((struct sockaddr *)&lep->com.local_addr)) { 2739 /* if no refcount then close listen socket */ 2740 if (!rem_ep_from_listenlist(lep)) 2741 close_socket(lep->com.so); 2742 } else 2743 close_socket(lep->com.so); 2744 deref_cm_id(&lep->com); 2745 mutex_unlock(&lep->com.mutex); 2746 c4iw_put_ep(&lep->com); 2747 return 0; 2748 } 2749 2750 int __c4iw_ep_disconnect(struct c4iw_ep *ep, int abrupt, gfp_t gfp) 2751 { 2752 int ret; 2753 mutex_lock(&ep->com.mutex); 2754 ret = c4iw_ep_disconnect(ep, abrupt, gfp); 2755 mutex_unlock(&ep->com.mutex); 2756 return ret; 2757 } 2758 2759 int c4iw_ep_disconnect(struct c4iw_ep *ep, int abrupt, gfp_t gfp) 2760 { 2761 int ret = 0; 2762 int close = 0; 2763 int fatal = 0; 2764 struct c4iw_rdev *rdev; 2765 2766 2767 CTR2(KTR_IW_CXGBE, "%s:cedB %p", __func__, ep); 2768 2769 rdev = &ep->com.dev->rdev; 2770 2771 if (c4iw_fatal_error(rdev)) { 2772 2773 CTR2(KTR_IW_CXGBE, "%s:ced1 %p", __func__, ep); 2774 fatal = 1; 2775 close_complete_upcall(ep, -ECONNRESET); 2776 send_abort(ep); 2777 ep->com.state = DEAD; 2778 } 2779 CTR3(KTR_IW_CXGBE, "%s:ced2 %p %s", __func__, ep, 2780 states[ep->com.state]); 2781 2782 /* 2783 * Ref the ep here in case we have fatal errors causing the 2784 * ep to be released and freed. 2785 */ 2786 c4iw_get_ep(&ep->com); 2787 switch (ep->com.state) { 2788 2789 case MPA_REQ_WAIT: 2790 case MPA_REQ_SENT: 2791 case MPA_REQ_RCVD: 2792 case MPA_REP_SENT: 2793 case FPDU_MODE: 2794 close = 1; 2795 if (abrupt) 2796 ep->com.state = ABORTING; 2797 else { 2798 ep->com.state = CLOSING; 2799 START_EP_TIMER(ep); 2800 } 2801 set_bit(CLOSE_SENT, &ep->com.flags); 2802 break; 2803 2804 case CLOSING: 2805 2806 if (!test_and_set_bit(CLOSE_SENT, &ep->com.flags)) { 2807 2808 close = 1; 2809 if (abrupt) { 2810 STOP_EP_TIMER(ep); 2811 ep->com.state = ABORTING; 2812 } else 2813 ep->com.state = MORIBUND; 2814 } 2815 break; 2816 2817 case MORIBUND: 2818 case ABORTING: 2819 case DEAD: 2820 CTR3(KTR_IW_CXGBE, 2821 "%s ignoring disconnect ep %p state %u", __func__, 2822 ep, ep->com.state); 2823 break; 2824 2825 default: 2826 BUG(); 2827 break; 2828 } 2829 2830 2831 if (close) { 2832 2833 CTR2(KTR_IW_CXGBE, "%s:ced3 %p", __func__, ep); 2834 2835 if (abrupt) { 2836 2837 CTR2(KTR_IW_CXGBE, "%s:ced4 %p", __func__, ep); 2838 set_bit(EP_DISC_ABORT, &ep->com.history); 2839 close_complete_upcall(ep, -ECONNRESET); 2840 ret = send_abort(ep); 2841 if (ret) 2842 fatal = 1; 2843 } else { 2844 2845 CTR2(KTR_IW_CXGBE, "%s:ced5 %p", __func__, ep); 2846 set_bit(EP_DISC_CLOSE, &ep->com.history); 2847 2848 if (!ep->parent_ep) 2849 ep->com.state = MORIBUND; 2850 2851 CURVNET_SET(ep->com.so->so_vnet); 2852 sodisconnect(ep->com.so); 2853 CURVNET_RESTORE(); 2854 } 2855 2856 } 2857 2858 if (fatal) { 2859 set_bit(EP_DISC_FAIL, &ep->com.history); 2860 if (!abrupt) { 2861 STOP_EP_TIMER(ep); 2862 close_complete_upcall(ep, -EIO); 2863 } 2864 if (ep->com.qp) { 2865 struct c4iw_qp_attributes attrs = {0}; 2866 2867 attrs.next_state = C4IW_QP_STATE_ERROR; 2868 ret = c4iw_modify_qp(ep->com.dev, ep->com.qp, 2869 C4IW_QP_ATTR_NEXT_STATE, 2870 &attrs, 1); 2871 if (ret) { 2872 CTR2(KTR_IW_CXGBE, "%s:ced7 %p", __func__, ep); 2873 printf("%s - qp <- error failed!\n", __func__); 2874 } 2875 } 2876 release_ep_resources(ep); 2877 ep->com.state = DEAD; 2878 CTR2(KTR_IW_CXGBE, "%s:ced6 %p", __func__, ep); 2879 } 2880 c4iw_put_ep(&ep->com); 2881 CTR2(KTR_IW_CXGBE, "%s:cedE %p", __func__, ep); 2882 return ret; 2883 } 2884 2885 #ifdef C4IW_EP_REDIRECT 2886 int c4iw_ep_redirect(void *ctx, struct dst_entry *old, struct dst_entry *new, 2887 struct l2t_entry *l2t) 2888 { 2889 struct c4iw_ep *ep = ctx; 2890 2891 if (ep->dst != old) 2892 return 0; 2893 2894 PDBG("%s ep %p redirect to dst %p l2t %p\n", __func__, ep, new, 2895 l2t); 2896 dst_hold(new); 2897 cxgb4_l2t_release(ep->l2t); 2898 ep->l2t = l2t; 2899 dst_release(old); 2900 ep->dst = new; 2901 return 1; 2902 } 2903 #endif 2904 2905 2906 2907 static void ep_timeout(unsigned long arg) 2908 { 2909 struct c4iw_ep *ep = (struct c4iw_ep *)arg; 2910 2911 if (!test_and_set_bit(TIMEOUT, &ep->com.flags)) { 2912 2913 /* 2914 * Only insert if it is not already on the list. 2915 */ 2916 if (!(ep->com.ep_events & C4IW_EVENT_TIMEOUT)) { 2917 CTR2(KTR_IW_CXGBE, "%s:et1 %p", __func__, ep); 2918 add_ep_to_req_list(ep, C4IW_EVENT_TIMEOUT); 2919 } 2920 } 2921 } 2922 2923 static int fw6_wr_rpl(struct adapter *sc, const __be64 *rpl) 2924 { 2925 uint64_t val = be64toh(*rpl); 2926 int ret; 2927 struct c4iw_wr_wait *wr_waitp; 2928 2929 ret = (int)((val >> 8) & 0xff); 2930 wr_waitp = (struct c4iw_wr_wait *)rpl[1]; 2931 CTR3(KTR_IW_CXGBE, "%s wr_waitp %p ret %u", __func__, wr_waitp, ret); 2932 if (wr_waitp) 2933 c4iw_wake_up(wr_waitp, ret ? -ret : 0); 2934 2935 return (0); 2936 } 2937 2938 static int fw6_cqe_handler(struct adapter *sc, const __be64 *rpl) 2939 { 2940 struct cqe_list_entry *cle; 2941 unsigned long flag; 2942 2943 cle = malloc(sizeof(*cle), M_CXGBE, M_NOWAIT); 2944 cle->rhp = sc->iwarp_softc; 2945 cle->err_cqe = *(const struct t4_cqe *)(&rpl[0]); 2946 2947 spin_lock_irqsave(&err_cqe_lock, flag); 2948 list_add_tail(&cle->entry, &err_cqe_list); 2949 queue_work(c4iw_taskq, &c4iw_task); 2950 spin_unlock_irqrestore(&err_cqe_lock, flag); 2951 2952 return (0); 2953 } 2954 2955 static int 2956 process_terminate(struct c4iw_ep *ep) 2957 { 2958 struct c4iw_qp_attributes attrs = {0}; 2959 2960 CTR2(KTR_IW_CXGBE, "%s:tB %p %d", __func__, ep); 2961 2962 if (ep && ep->com.qp) { 2963 2964 printk(KERN_WARNING MOD "TERM received tid %u qpid %u\n", 2965 ep->hwtid, ep->com.qp->wq.sq.qid); 2966 attrs.next_state = C4IW_QP_STATE_TERMINATE; 2967 c4iw_modify_qp(ep->com.dev, ep->com.qp, C4IW_QP_ATTR_NEXT_STATE, &attrs, 2968 1); 2969 } else 2970 printk(KERN_WARNING MOD "TERM received tid %u no ep/qp\n", 2971 ep->hwtid); 2972 CTR2(KTR_IW_CXGBE, "%s:tE %p %d", __func__, ep); 2973 2974 return 0; 2975 } 2976 2977 int __init c4iw_cm_init(void) 2978 { 2979 2980 t4_register_cpl_handler(CPL_RDMA_TERMINATE, terminate); 2981 t4_register_fw_msg_handler(FW6_TYPE_WR_RPL, fw6_wr_rpl); 2982 t4_register_fw_msg_handler(FW6_TYPE_CQE, fw6_cqe_handler); 2983 t4_register_an_handler(c4iw_ev_handler); 2984 2985 TAILQ_INIT(&req_list); 2986 spin_lock_init(&req_lock); 2987 INIT_LIST_HEAD(&err_cqe_list); 2988 spin_lock_init(&err_cqe_lock); 2989 2990 INIT_WORK(&c4iw_task, process_req); 2991 2992 c4iw_taskq = create_singlethread_workqueue("iw_cxgbe"); 2993 if (!c4iw_taskq) 2994 return -ENOMEM; 2995 2996 return 0; 2997 } 2998 2999 void __exit c4iw_cm_term(void) 3000 { 3001 WARN_ON(!TAILQ_EMPTY(&req_list)); 3002 WARN_ON(!list_empty(&err_cqe_list)); 3003 flush_workqueue(c4iw_taskq); 3004 destroy_workqueue(c4iw_taskq); 3005 3006 t4_register_cpl_handler(CPL_RDMA_TERMINATE, NULL); 3007 t4_register_fw_msg_handler(FW6_TYPE_WR_RPL, NULL); 3008 t4_register_fw_msg_handler(FW6_TYPE_CQE, NULL); 3009 t4_register_an_handler(NULL); 3010 } 3011 #endif 3012