cm.c revision 346922
1/* 2 * Copyright (c) 2009-2013, 2016 Chelsio, Inc. All rights reserved. 3 * 4 * This software is available to you under a choice of one of two 5 * licenses. You may choose to be licensed under the terms of the GNU 6 * General Public License (GPL) Version 2, available from the file 7 * COPYING in the main directory of this source tree, or the 8 * OpenIB.org BSD license below: 9 * 10 * Redistribution and use in source and binary forms, with or 11 * without modification, are permitted provided that the following 12 * conditions are met: 13 * 14 * - Redistributions of source code must retain the above 15 * copyright notice, this list of conditions and the following 16 * disclaimer. 17 * 18 * - Redistributions in binary form must reproduce the above 19 * copyright notice, this list of conditions and the following 20 * disclaimer in the documentation and/or other materials 21 * provided with the distribution. 22 * 23 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, 24 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF 25 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND 26 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS 27 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN 28 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN 29 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE 30 * SOFTWARE. 31 */ 32#include <sys/cdefs.h> 33__FBSDID("$FreeBSD: stable/11/sys/dev/cxgbe/iw_cxgbe/cm.c 346922 2019-04-29 19:47:21Z np $"); 34 35#include "opt_inet.h" 36 37#ifdef TCP_OFFLOAD 38#include <sys/types.h> 39#include <sys/malloc.h> 40#include <sys/socket.h> 41#include <sys/socketvar.h> 42#include <sys/sockio.h> 43#include <sys/taskqueue.h> 44#include <netinet/in.h> 45#include <net/route.h> 46 47#include <netinet/in_systm.h> 48#include <netinet/in_pcb.h> 49#include <netinet6/in6_pcb.h> 50#include <netinet/ip.h> 51#include <netinet/in_fib.h> 52#include <netinet6/in6_fib.h> 53#include <netinet6/scope6_var.h> 54#include <netinet/ip_var.h> 55#include <netinet/tcp_var.h> 56#include <netinet/tcp.h> 57#include <netinet/tcpip.h> 58 59#include <netinet/toecore.h> 60 61struct sge_iq; 62struct rss_header; 63struct cpl_set_tcb_rpl; 64#include <linux/types.h> 65#include "offload.h" 66#include "tom/t4_tom.h" 67 68#define TOEPCB(so) ((struct toepcb *)(so_sototcpcb((so))->t_toe)) 69 70#include "iw_cxgbe.h" 71#include <linux/module.h> 72#include <linux/workqueue.h> 73#include <linux/notifier.h> 74#include <linux/inetdevice.h> 75#include <linux/if_vlan.h> 76#include <net/netevent.h> 77#include <rdma/rdma_cm.h> 78 79static spinlock_t req_lock; 80static TAILQ_HEAD(c4iw_ep_list, c4iw_ep_common) req_list; 81static struct work_struct c4iw_task; 82static struct workqueue_struct *c4iw_taskq; 83static LIST_HEAD(err_cqe_list); 84static spinlock_t err_cqe_lock; 85static LIST_HEAD(listen_port_list); 86static DEFINE_MUTEX(listen_port_mutex); 87 88static void process_req(struct work_struct *ctx); 89static void start_ep_timer(struct c4iw_ep *ep); 90static int stop_ep_timer(struct c4iw_ep *ep); 91static int set_tcpinfo(struct c4iw_ep *ep); 92static void process_timeout(struct c4iw_ep *ep); 93static void process_err_cqes(void); 94static void *alloc_ep(int size, gfp_t flags); 95static void close_socket(struct socket *so); 96static int send_mpa_req(struct c4iw_ep *ep); 97static int send_mpa_reject(struct c4iw_ep *ep, const void *pdata, u8 plen); 98static int send_mpa_reply(struct c4iw_ep *ep, const void *pdata, u8 plen); 99static void close_complete_upcall(struct c4iw_ep *ep, int status); 100static int send_abort(struct c4iw_ep *ep); 101static void peer_close_upcall(struct c4iw_ep *ep); 102static void peer_abort_upcall(struct c4iw_ep *ep); 103static void connect_reply_upcall(struct c4iw_ep *ep, int status); 104static int connect_request_upcall(struct c4iw_ep *ep); 105static void established_upcall(struct c4iw_ep *ep); 106static int process_mpa_reply(struct c4iw_ep *ep); 107static int process_mpa_request(struct c4iw_ep *ep); 108static void process_peer_close(struct c4iw_ep *ep); 109static void process_conn_error(struct c4iw_ep *ep); 110static void process_close_complete(struct c4iw_ep *ep); 111static void ep_timeout(unsigned long arg); 112static void setiwsockopt(struct socket *so); 113static void init_iwarp_socket(struct socket *so, void *arg); 114static void uninit_iwarp_socket(struct socket *so); 115static void process_data(struct c4iw_ep *ep); 116static void process_connected(struct c4iw_ep *ep); 117static int c4iw_so_upcall(struct socket *so, void *arg, int waitflag); 118static void process_socket_event(struct c4iw_ep *ep); 119static void release_ep_resources(struct c4iw_ep *ep); 120static int process_terminate(struct c4iw_ep *ep); 121static int terminate(struct sge_iq *iq, const struct rss_header *rss, 122 struct mbuf *m); 123static int add_ep_to_req_list(struct c4iw_ep *ep, int ep_events); 124static struct listen_port_info * 125add_ep_to_listenlist(struct c4iw_listen_ep *lep); 126static int rem_ep_from_listenlist(struct c4iw_listen_ep *lep); 127static struct c4iw_listen_ep * 128find_real_listen_ep(struct c4iw_listen_ep *master_lep, struct socket *so); 129static int get_ifnet_from_raddr(struct sockaddr_storage *raddr, 130 struct ifnet **ifp); 131static void process_newconn(struct c4iw_listen_ep *master_lep, 132 struct socket *new_so); 133#define START_EP_TIMER(ep) \ 134 do { \ 135 CTR3(KTR_IW_CXGBE, "start_ep_timer (%s:%d) ep %p", \ 136 __func__, __LINE__, (ep)); \ 137 start_ep_timer(ep); \ 138 } while (0) 139 140#define STOP_EP_TIMER(ep) \ 141 ({ \ 142 CTR3(KTR_IW_CXGBE, "stop_ep_timer (%s:%d) ep %p", \ 143 __func__, __LINE__, (ep)); \ 144 stop_ep_timer(ep); \ 145 }) 146 147#define GET_LOCAL_ADDR(pladdr, so) \ 148 do { \ 149 struct sockaddr_storage *__a = NULL; \ 150 struct inpcb *__inp = sotoinpcb(so); \ 151 KASSERT(__inp != NULL, \ 152 ("GET_LOCAL_ADDR(%s):so:%p, inp = NULL", __func__, so)); \ 153 if (__inp->inp_vflag & INP_IPV4) \ 154 in_getsockaddr(so, (struct sockaddr **)&__a); \ 155 else \ 156 in6_getsockaddr(so, (struct sockaddr **)&__a); \ 157 *(pladdr) = *__a; \ 158 free(__a, M_SONAME); \ 159 } while (0) 160 161#define GET_REMOTE_ADDR(praddr, so) \ 162 do { \ 163 struct sockaddr_storage *__a = NULL; \ 164 struct inpcb *__inp = sotoinpcb(so); \ 165 KASSERT(__inp != NULL, \ 166 ("GET_REMOTE_ADDR(%s):so:%p, inp = NULL", __func__, so)); \ 167 if (__inp->inp_vflag & INP_IPV4) \ 168 in_getpeeraddr(so, (struct sockaddr **)&__a); \ 169 else \ 170 in6_getpeeraddr(so, (struct sockaddr **)&__a); \ 171 *(praddr) = *__a; \ 172 free(__a, M_SONAME); \ 173 } while (0) 174 175#ifdef KTR 176static char *states[] = { 177 "idle", 178 "listen", 179 "connecting", 180 "mpa_wait_req", 181 "mpa_req_sent", 182 "mpa_req_rcvd", 183 "mpa_rep_sent", 184 "fpdu_mode", 185 "aborting", 186 "closing", 187 "moribund", 188 "dead", 189 NULL, 190}; 191#endif 192 193static 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 200static 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 206static 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 213static 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 */ 220struct 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 */ 267static struct listen_port_info * 268add_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 291found_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 298static int 299rem_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 */ 336struct c4iw_listen_ep * 337find_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 } 391out: 392 mutex_unlock(&listen_port_mutex); 393 return found_lep ? lep : (NULL); 394} 395 396static 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 442struct cqe_list_entry { 443 struct list_head entry; 444 struct c4iw_dev *rhp; 445 struct t4_cqe err_cqe; 446}; 447 448static void 449process_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 471static void 472process_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 */ 509static int 510set_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); 532done: 533 INP_WUNLOCK(inp); 534 return (rc); 535 536} 537static int 538get_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 571static void 572close_socket(struct socket *so) 573{ 574 uninit_iwarp_socket(so); 575 soclose(so); 576} 577 578static void 579process_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 692static void 693process_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 766static void 767process_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 831static void 832setiwsockopt(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 851static void 852init_iwarp_socket(struct socket *so, void *arg) 853{ 854 855 SOCKBUF_LOCK(&so->so_rcv); 856 soupcall_set(so, SO_RCV, c4iw_so_upcall, arg); 857 so->so_state |= SS_NBIO; 858 SOCKBUF_UNLOCK(&so->so_rcv); 859} 860 861static void 862uninit_iwarp_socket(struct socket *so) 863{ 864 865 SOCKBUF_LOCK(&so->so_rcv); 866 soupcall_clear(so, SO_RCV); 867 SOCKBUF_UNLOCK(&so->so_rcv); 868} 869 870static void 871process_data(struct c4iw_ep *ep) 872{ 873 int disconnect = 0; 874 875 CTR5(KTR_IW_CXGBE, "%s: so %p, ep %p, state %s, sbused %d", __func__, 876 ep->com.so, ep, states[ep->com.state], sbused(&ep->com.so->so_rcv)); 877 878 switch (ep->com.state) { 879 case MPA_REQ_SENT: 880 disconnect = process_mpa_reply(ep); 881 break; 882 case MPA_REQ_WAIT: 883 disconnect = process_mpa_request(ep); 884 if (disconnect) 885 /* Refered in process_newconn() */ 886 c4iw_put_ep(&ep->parent_ep->com); 887 break; 888 default: 889 if (sbused(&ep->com.so->so_rcv)) 890 log(LOG_ERR, "%s: Unexpected streaming data. ep %p, " 891 "state %d, so %p, so_state 0x%x, sbused %u\n", 892 __func__, ep, ep->com.state, ep->com.so, 893 ep->com.so->so_state, sbused(&ep->com.so->so_rcv)); 894 break; 895 } 896 if (disconnect) 897 c4iw_ep_disconnect(ep, disconnect == 2, GFP_KERNEL); 898 899} 900 901static void 902process_connected(struct c4iw_ep *ep) 903{ 904 struct socket *so = ep->com.so; 905 906 if ((so->so_state & SS_ISCONNECTED) && !so->so_error) { 907 if (send_mpa_req(ep)) 908 goto err; 909 } else { 910 connect_reply_upcall(ep, -so->so_error); 911 goto err; 912 } 913 return; 914err: 915 close_socket(so); 916 ep->com.state = DEAD; 917 c4iw_put_ep(&ep->com); 918 return; 919} 920 921static inline int c4iw_zero_addr(struct sockaddr *addr) 922{ 923 struct in6_addr *ip6; 924 925 if (addr->sa_family == AF_INET) 926 return IN_ZERONET( 927 ntohl(((struct sockaddr_in *)addr)->sin_addr.s_addr)); 928 else { 929 ip6 = &((struct sockaddr_in6 *) addr)->sin6_addr; 930 return (ip6->s6_addr32[0] | ip6->s6_addr32[1] | 931 ip6->s6_addr32[2] | ip6->s6_addr32[3]) == 0; 932 } 933} 934 935static inline int c4iw_loopback_addr(struct sockaddr *addr) 936{ 937 if (addr->sa_family == AF_INET) 938 return IN_LOOPBACK( 939 ntohl(((struct sockaddr_in *) addr)->sin_addr.s_addr)); 940 else 941 return IN6_IS_ADDR_LOOPBACK( 942 &((struct sockaddr_in6 *) addr)->sin6_addr); 943} 944 945static inline int c4iw_any_addr(struct sockaddr *addr) 946{ 947 return c4iw_zero_addr(addr) || c4iw_loopback_addr(addr); 948} 949 950static void 951process_newconn(struct c4iw_listen_ep *master_lep, struct socket *new_so) 952{ 953 struct c4iw_listen_ep *real_lep = NULL; 954 struct c4iw_ep *new_ep = NULL; 955 int ret = 0; 956 957 MPASS(new_so != NULL); 958 959 if (c4iw_any_addr((struct sockaddr *)&master_lep->com.local_addr)) { 960 /* Here we need to find the 'real_lep' that belongs to the 961 * incomming socket's network interface, such that the newly 962 * created 'ep' can be attached to the real 'lep'. 963 */ 964 real_lep = find_real_listen_ep(master_lep, new_so); 965 if (real_lep == NULL) { 966 CTR2(KTR_IW_CXGBE, "%s: Could not find the real listen " 967 "ep for sock: %p", __func__, new_so); 968 log(LOG_ERR,"%s: Could not find the real listen ep for " 969 "sock: %p\n", __func__, new_so); 970 /* FIXME: properly free the 'new_so' in failure case. 971 * Use of soabort() and soclose() are not legal 972 * here(before soaccept()). 973 */ 974 return; 975 } 976 } else /* for Non-Wildcard address, master_lep is always the real_lep */ 977 real_lep = master_lep; 978 979 new_ep = alloc_ep(sizeof(*new_ep), GFP_KERNEL); 980 981 CTR6(KTR_IW_CXGBE, "%s: master_lep %p, real_lep: %p, new ep %p, " 982 "listening so %p, new so %p", __func__, master_lep, real_lep, 983 new_ep, master_lep->com.so, new_so); 984 985 new_ep->com.dev = real_lep->com.dev; 986 new_ep->com.so = new_so; 987 new_ep->com.cm_id = NULL; 988 new_ep->com.thread = real_lep->com.thread; 989 new_ep->parent_ep = real_lep; 990 991 GET_LOCAL_ADDR(&new_ep->com.local_addr, new_so); 992 GET_REMOTE_ADDR(&new_ep->com.remote_addr, new_so); 993 c4iw_get_ep(&real_lep->com); 994 init_timer(&new_ep->timer); 995 new_ep->com.state = MPA_REQ_WAIT; 996 START_EP_TIMER(new_ep); 997 998 setiwsockopt(new_so); 999 1000 /* MPA request might have been queued up on the socket already, so we 1001 * initialize the socket/upcall_handler under lock to prevent processing 1002 * MPA request on another thread(via process_req()) simultaniously. 1003 */ 1004 c4iw_get_ep(&new_ep->com); /* Dereferenced at the end below, this is to 1005 avoid freeing of ep before ep unlock. */ 1006 mutex_lock(&new_ep->com.mutex); 1007 init_iwarp_socket(new_so, &new_ep->com); 1008 1009 ret = process_mpa_request(new_ep); 1010 if (ret) { 1011 /* ABORT */ 1012 c4iw_ep_disconnect(new_ep, 1, GFP_KERNEL); 1013 c4iw_put_ep(&real_lep->com); 1014 } 1015 mutex_unlock(&new_ep->com.mutex); 1016 c4iw_put_ep(&new_ep->com); 1017 return; 1018} 1019 1020static int 1021add_ep_to_req_list(struct c4iw_ep *ep, int new_ep_event) 1022{ 1023 unsigned long flag; 1024 1025 spin_lock_irqsave(&req_lock, flag); 1026 if (ep && ep->com.so) { 1027 ep->com.ep_events |= new_ep_event; 1028 if (!ep->com.entry.tqe_prev) { 1029 c4iw_get_ep(&ep->com); 1030 TAILQ_INSERT_TAIL(&req_list, &ep->com, entry); 1031 queue_work(c4iw_taskq, &c4iw_task); 1032 } 1033 } 1034 spin_unlock_irqrestore(&req_lock, flag); 1035 1036 return (0); 1037} 1038 1039static int 1040c4iw_so_upcall(struct socket *so, void *arg, int waitflag) 1041{ 1042 struct c4iw_ep *ep = arg; 1043 1044 CTR6(KTR_IW_CXGBE, 1045 "%s: so %p, so_state 0x%x, ep %p, ep_state %s, tqe_prev %p", 1046 __func__, so, so->so_state, ep, states[ep->com.state], 1047 ep->com.entry.tqe_prev); 1048 1049 MPASS(ep->com.so == so); 1050 /* 1051 * Wake up any threads waiting in rdma_init()/rdma_fini(), 1052 * with locks held. 1053 */ 1054 if (so->so_error) 1055 c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET); 1056 add_ep_to_req_list(ep, C4IW_EVENT_SOCKET); 1057 1058 return (SU_OK); 1059} 1060 1061 1062static int 1063terminate(struct sge_iq *iq, const struct rss_header *rss, struct mbuf *m) 1064{ 1065 struct adapter *sc = iq->adapter; 1066 const struct cpl_rdma_terminate *cpl = mtod(m, const void *); 1067 unsigned int tid = GET_TID(cpl); 1068 struct toepcb *toep = lookup_tid(sc, tid); 1069 struct socket *so; 1070 struct c4iw_ep *ep; 1071 1072 INP_WLOCK(toep->inp); 1073 so = inp_inpcbtosocket(toep->inp); 1074 ep = so->so_rcv.sb_upcallarg; 1075 INP_WUNLOCK(toep->inp); 1076 1077 CTR3(KTR_IW_CXGBE, "%s: so %p, ep %p", __func__, so, ep); 1078 add_ep_to_req_list(ep, C4IW_EVENT_TERM); 1079 1080 return 0; 1081} 1082 1083static struct socket * 1084dequeue_socket(struct socket *head) 1085{ 1086 struct socket *so; 1087 struct sockaddr_in *remote; 1088 1089 ACCEPT_LOCK(); 1090 so = TAILQ_FIRST(&head->so_comp); 1091 if (!so) { 1092 ACCEPT_UNLOCK(); 1093 return NULL; 1094 } 1095 1096 SOCK_LOCK(so); 1097 /* 1098 * Before changing the flags on the socket, we have to bump the 1099 * reference count. Otherwise, if the protocol calls sofree(), 1100 * the socket will be released due to a zero refcount. 1101 */ 1102 soref(so); 1103 TAILQ_REMOVE(&head->so_comp, so, so_list); 1104 head->so_qlen--; 1105 so->so_qstate &= ~SQ_COMP; 1106 so->so_head = NULL; 1107 so->so_state |= SS_NBIO; 1108 SOCK_UNLOCK(so); 1109 ACCEPT_UNLOCK(); 1110 remote = NULL; 1111 if (soaccept(so, (struct sockaddr **)&remote) != 0) { 1112 soclose(so); 1113 so = NULL; 1114 } 1115 free(remote, M_SONAME); 1116 return so; 1117} 1118 1119static void 1120process_socket_event(struct c4iw_ep *ep) 1121{ 1122 int state = ep->com.state; 1123 struct socket *so = ep->com.so; 1124 1125 if (ep->com.state == DEAD) { 1126 CTR3(KTR_IW_CXGBE, "%s: Pending socket event discarded " 1127 "ep %p ep_state %s", __func__, ep, states[state]); 1128 return; 1129 } 1130 1131 CTR6(KTR_IW_CXGBE, "process_socket_event: so %p, so_state 0x%x, " 1132 "so_err %d, sb_state 0x%x, ep %p, ep_state %s", so, so->so_state, 1133 so->so_error, so->so_rcv.sb_state, ep, states[state]); 1134 1135 if (state == CONNECTING) { 1136 process_connected(ep); 1137 return; 1138 } 1139 1140 if (state == LISTEN) { 1141 struct c4iw_listen_ep *lep = (struct c4iw_listen_ep *)ep; 1142 struct socket *new_so; 1143 1144 while ((new_so = dequeue_socket(so)) != NULL) { 1145 process_newconn(lep, new_so); 1146 } 1147 return; 1148 } 1149 1150 /* connection error */ 1151 if (so->so_error) { 1152 process_conn_error(ep); 1153 return; 1154 } 1155 1156 /* peer close */ 1157 if ((so->so_rcv.sb_state & SBS_CANTRCVMORE) && state <= CLOSING) { 1158 process_peer_close(ep); 1159 /* 1160 * check whether socket disconnect event is pending before 1161 * returning. Fallthrough if yes. 1162 */ 1163 if (!(so->so_state & SS_ISDISCONNECTED)) 1164 return; 1165 } 1166 1167 /* close complete */ 1168 if (so->so_state & SS_ISDISCONNECTED) { 1169 process_close_complete(ep); 1170 return; 1171 } 1172 1173 /* rx data */ 1174 process_data(ep); 1175} 1176 1177SYSCTL_NODE(_hw, OID_AUTO, iw_cxgbe, CTLFLAG_RD, 0, "iw_cxgbe driver parameters"); 1178 1179static int dack_mode = 0; 1180SYSCTL_INT(_hw_iw_cxgbe, OID_AUTO, dack_mode, CTLFLAG_RWTUN, &dack_mode, 0, 1181 "Delayed ack mode (default = 0)"); 1182 1183int c4iw_max_read_depth = 8; 1184SYSCTL_INT(_hw_iw_cxgbe, OID_AUTO, c4iw_max_read_depth, CTLFLAG_RWTUN, &c4iw_max_read_depth, 0, 1185 "Per-connection max ORD/IRD (default = 8)"); 1186 1187static int enable_tcp_timestamps; 1188SYSCTL_INT(_hw_iw_cxgbe, OID_AUTO, enable_tcp_timestamps, CTLFLAG_RWTUN, &enable_tcp_timestamps, 0, 1189 "Enable tcp timestamps (default = 0)"); 1190 1191static int enable_tcp_sack; 1192SYSCTL_INT(_hw_iw_cxgbe, OID_AUTO, enable_tcp_sack, CTLFLAG_RWTUN, &enable_tcp_sack, 0, 1193 "Enable tcp SACK (default = 0)"); 1194 1195static int enable_tcp_window_scaling = 1; 1196SYSCTL_INT(_hw_iw_cxgbe, OID_AUTO, enable_tcp_window_scaling, CTLFLAG_RWTUN, &enable_tcp_window_scaling, 0, 1197 "Enable tcp window scaling (default = 1)"); 1198 1199int c4iw_debug = 0; 1200SYSCTL_INT(_hw_iw_cxgbe, OID_AUTO, c4iw_debug, CTLFLAG_RWTUN, &c4iw_debug, 0, 1201 "Enable debug logging (default = 0)"); 1202 1203static int peer2peer = 1; 1204SYSCTL_INT(_hw_iw_cxgbe, OID_AUTO, peer2peer, CTLFLAG_RWTUN, &peer2peer, 0, 1205 "Support peer2peer ULPs (default = 1)"); 1206 1207static int p2p_type = FW_RI_INIT_P2PTYPE_READ_REQ; 1208SYSCTL_INT(_hw_iw_cxgbe, OID_AUTO, p2p_type, CTLFLAG_RWTUN, &p2p_type, 0, 1209 "RDMAP opcode to use for the RTR message: 1 = RDMA_READ 0 = RDMA_WRITE (default 1)"); 1210 1211static int ep_timeout_secs = 60; 1212SYSCTL_INT(_hw_iw_cxgbe, OID_AUTO, ep_timeout_secs, CTLFLAG_RWTUN, &ep_timeout_secs, 0, 1213 "CM Endpoint operation timeout in seconds (default = 60)"); 1214 1215static int mpa_rev = 1; 1216SYSCTL_INT(_hw_iw_cxgbe, OID_AUTO, mpa_rev, CTLFLAG_RWTUN, &mpa_rev, 0, 1217 "MPA Revision, 0 supports amso1100, 1 is RFC5044 spec compliant, 2 is IETF MPA Peer Connect Draft compliant (default = 1)"); 1218 1219static int markers_enabled; 1220SYSCTL_INT(_hw_iw_cxgbe, OID_AUTO, markers_enabled, CTLFLAG_RWTUN, &markers_enabled, 0, 1221 "Enable MPA MARKERS (default(0) = disabled)"); 1222 1223static int crc_enabled = 1; 1224SYSCTL_INT(_hw_iw_cxgbe, OID_AUTO, crc_enabled, CTLFLAG_RWTUN, &crc_enabled, 0, 1225 "Enable MPA CRC (default(1) = enabled)"); 1226 1227static int rcv_win = 256 * 1024; 1228SYSCTL_INT(_hw_iw_cxgbe, OID_AUTO, rcv_win, CTLFLAG_RWTUN, &rcv_win, 0, 1229 "TCP receive window in bytes (default = 256KB)"); 1230 1231static int snd_win = 128 * 1024; 1232SYSCTL_INT(_hw_iw_cxgbe, OID_AUTO, snd_win, CTLFLAG_RWTUN, &snd_win, 0, 1233 "TCP send window in bytes (default = 128KB)"); 1234 1235static void 1236start_ep_timer(struct c4iw_ep *ep) 1237{ 1238 1239 if (timer_pending(&ep->timer)) { 1240 CTR2(KTR_IW_CXGBE, "%s: ep %p, already started", __func__, ep); 1241 printk(KERN_ERR "%s timer already started! ep %p\n", __func__, 1242 ep); 1243 return; 1244 } 1245 clear_bit(TIMEOUT, &ep->com.flags); 1246 c4iw_get_ep(&ep->com); 1247 ep->timer.expires = jiffies + ep_timeout_secs * HZ; 1248 ep->timer.data = (unsigned long)ep; 1249 ep->timer.function = ep_timeout; 1250 add_timer(&ep->timer); 1251} 1252 1253static int 1254stop_ep_timer(struct c4iw_ep *ep) 1255{ 1256 1257 del_timer_sync(&ep->timer); 1258 if (!test_and_set_bit(TIMEOUT, &ep->com.flags)) { 1259 c4iw_put_ep(&ep->com); 1260 return 0; 1261 } 1262 return 1; 1263} 1264 1265static void * 1266alloc_ep(int size, gfp_t gfp) 1267{ 1268 struct c4iw_ep_common *epc; 1269 1270 epc = kzalloc(size, gfp); 1271 if (epc == NULL) 1272 return (NULL); 1273 1274 kref_init(&epc->kref); 1275 mutex_init(&epc->mutex); 1276 c4iw_init_wr_wait(&epc->wr_wait); 1277 1278 return (epc); 1279} 1280 1281void _c4iw_free_ep(struct kref *kref) 1282{ 1283 struct c4iw_ep *ep; 1284 struct c4iw_ep_common *epc; 1285 1286 ep = container_of(kref, struct c4iw_ep, com.kref); 1287 epc = &ep->com; 1288 KASSERT(!epc->entry.tqe_prev, ("%s epc %p still on req list", 1289 __func__, epc)); 1290 if (test_bit(QP_REFERENCED, &ep->com.flags)) 1291 deref_qp(ep); 1292 CTR4(KTR_IW_CXGBE, "%s: ep %p, history 0x%lx, flags 0x%lx", 1293 __func__, ep, epc->history, epc->flags); 1294 kfree(ep); 1295} 1296 1297static void release_ep_resources(struct c4iw_ep *ep) 1298{ 1299 CTR2(KTR_IW_CXGBE, "%s:rerB %p", __func__, ep); 1300 set_bit(RELEASE_RESOURCES, &ep->com.flags); 1301 c4iw_put_ep(&ep->com); 1302 CTR2(KTR_IW_CXGBE, "%s:rerE %p", __func__, ep); 1303} 1304 1305static int 1306send_mpa_req(struct c4iw_ep *ep) 1307{ 1308 int mpalen; 1309 struct mpa_message *mpa; 1310 struct mpa_v2_conn_params mpa_v2_params; 1311 struct mbuf *m; 1312 char mpa_rev_to_use = mpa_rev; 1313 int err = 0; 1314 1315 if (ep->retry_with_mpa_v1) 1316 mpa_rev_to_use = 1; 1317 mpalen = sizeof(*mpa) + ep->plen; 1318 if (mpa_rev_to_use == 2) 1319 mpalen += sizeof(struct mpa_v2_conn_params); 1320 1321 mpa = malloc(mpalen, M_CXGBE, M_NOWAIT); 1322 if (mpa == NULL) { 1323 err = -ENOMEM; 1324 CTR3(KTR_IW_CXGBE, "%s:smr1 ep: %p , error: %d", 1325 __func__, ep, err); 1326 goto err; 1327 } 1328 1329 memset(mpa, 0, mpalen); 1330 memcpy(mpa->key, MPA_KEY_REQ, sizeof(mpa->key)); 1331 mpa->flags = (crc_enabled ? MPA_CRC : 0) | 1332 (markers_enabled ? MPA_MARKERS : 0) | 1333 (mpa_rev_to_use == 2 ? MPA_ENHANCED_RDMA_CONN : 0); 1334 mpa->private_data_size = htons(ep->plen); 1335 mpa->revision = mpa_rev_to_use; 1336 1337 if (mpa_rev_to_use == 1) { 1338 ep->tried_with_mpa_v1 = 1; 1339 ep->retry_with_mpa_v1 = 0; 1340 } 1341 1342 if (mpa_rev_to_use == 2) { 1343 mpa->private_data_size = htons(ntohs(mpa->private_data_size) + 1344 sizeof(struct mpa_v2_conn_params)); 1345 mpa_v2_params.ird = htons((u16)ep->ird); 1346 mpa_v2_params.ord = htons((u16)ep->ord); 1347 1348 if (peer2peer) { 1349 mpa_v2_params.ird |= htons(MPA_V2_PEER2PEER_MODEL); 1350 1351 if (p2p_type == FW_RI_INIT_P2PTYPE_RDMA_WRITE) { 1352 mpa_v2_params.ord |= 1353 htons(MPA_V2_RDMA_WRITE_RTR); 1354 } else if (p2p_type == FW_RI_INIT_P2PTYPE_READ_REQ) { 1355 mpa_v2_params.ord |= 1356 htons(MPA_V2_RDMA_READ_RTR); 1357 } 1358 } 1359 memcpy(mpa->private_data, &mpa_v2_params, 1360 sizeof(struct mpa_v2_conn_params)); 1361 1362 if (ep->plen) { 1363 1364 memcpy(mpa->private_data + 1365 sizeof(struct mpa_v2_conn_params), 1366 ep->mpa_pkt + sizeof(*mpa), ep->plen); 1367 } 1368 } else { 1369 1370 if (ep->plen) 1371 memcpy(mpa->private_data, 1372 ep->mpa_pkt + sizeof(*mpa), ep->plen); 1373 CTR2(KTR_IW_CXGBE, "%s:smr7 %p", __func__, ep); 1374 } 1375 1376 m = m_getm(NULL, mpalen, M_NOWAIT, MT_DATA); 1377 if (m == NULL) { 1378 err = -ENOMEM; 1379 CTR3(KTR_IW_CXGBE, "%s:smr2 ep: %p , error: %d", 1380 __func__, ep, err); 1381 free(mpa, M_CXGBE); 1382 goto err; 1383 } 1384 m_copyback(m, 0, mpalen, (void *)mpa); 1385 free(mpa, M_CXGBE); 1386 1387 err = -sosend(ep->com.so, NULL, NULL, m, NULL, MSG_DONTWAIT, 1388 ep->com.thread); 1389 if (err) { 1390 CTR3(KTR_IW_CXGBE, "%s:smr3 ep: %p , error: %d", 1391 __func__, ep, err); 1392 goto err; 1393 } 1394 1395 START_EP_TIMER(ep); 1396 ep->com.state = MPA_REQ_SENT; 1397 ep->mpa_attr.initiator = 1; 1398 CTR3(KTR_IW_CXGBE, "%s:smrE %p, error: %d", __func__, ep, err); 1399 return 0; 1400err: 1401 connect_reply_upcall(ep, err); 1402 CTR3(KTR_IW_CXGBE, "%s:smrE %p, error: %d", __func__, ep, err); 1403 return err; 1404} 1405 1406static int send_mpa_reject(struct c4iw_ep *ep, const void *pdata, u8 plen) 1407{ 1408 int mpalen ; 1409 struct mpa_message *mpa; 1410 struct mpa_v2_conn_params mpa_v2_params; 1411 struct mbuf *m; 1412 int err; 1413 1414 CTR4(KTR_IW_CXGBE, "%s:smrejB %p %u %d", __func__, ep, ep->hwtid, 1415 ep->plen); 1416 1417 mpalen = sizeof(*mpa) + plen; 1418 1419 if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) { 1420 1421 mpalen += sizeof(struct mpa_v2_conn_params); 1422 CTR4(KTR_IW_CXGBE, "%s:smrej1 %p %u %d", __func__, ep, 1423 ep->mpa_attr.version, mpalen); 1424 } 1425 1426 mpa = malloc(mpalen, M_CXGBE, M_NOWAIT); 1427 if (mpa == NULL) 1428 return (-ENOMEM); 1429 1430 memset(mpa, 0, mpalen); 1431 memcpy(mpa->key, MPA_KEY_REP, sizeof(mpa->key)); 1432 mpa->flags = MPA_REJECT; 1433 mpa->revision = mpa_rev; 1434 mpa->private_data_size = htons(plen); 1435 1436 if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) { 1437 1438 mpa->flags |= MPA_ENHANCED_RDMA_CONN; 1439 mpa->private_data_size = htons(ntohs(mpa->private_data_size) + 1440 sizeof(struct mpa_v2_conn_params)); 1441 mpa_v2_params.ird = htons(((u16)ep->ird) | 1442 (peer2peer ? MPA_V2_PEER2PEER_MODEL : 1443 0)); 1444 mpa_v2_params.ord = htons(((u16)ep->ord) | (peer2peer ? 1445 (p2p_type == 1446 FW_RI_INIT_P2PTYPE_RDMA_WRITE ? 1447 MPA_V2_RDMA_WRITE_RTR : p2p_type == 1448 FW_RI_INIT_P2PTYPE_READ_REQ ? 1449 MPA_V2_RDMA_READ_RTR : 0) : 0)); 1450 memcpy(mpa->private_data, &mpa_v2_params, 1451 sizeof(struct mpa_v2_conn_params)); 1452 1453 if (ep->plen) 1454 memcpy(mpa->private_data + 1455 sizeof(struct mpa_v2_conn_params), pdata, plen); 1456 CTR5(KTR_IW_CXGBE, "%s:smrej3 %p %d %d %d", __func__, ep, 1457 mpa_v2_params.ird, mpa_v2_params.ord, ep->plen); 1458 } else 1459 if (plen) 1460 memcpy(mpa->private_data, pdata, plen); 1461 1462 m = m_getm(NULL, mpalen, M_NOWAIT, MT_DATA); 1463 if (m == NULL) { 1464 free(mpa, M_CXGBE); 1465 return (-ENOMEM); 1466 } 1467 m_copyback(m, 0, mpalen, (void *)mpa); 1468 free(mpa, M_CXGBE); 1469 1470 err = -sosend(ep->com.so, NULL, NULL, m, NULL, MSG_DONTWAIT, ep->com.thread); 1471 if (!err) 1472 ep->snd_seq += mpalen; 1473 CTR4(KTR_IW_CXGBE, "%s:smrejE %p %u %d", __func__, ep, ep->hwtid, err); 1474 return err; 1475} 1476 1477static int send_mpa_reply(struct c4iw_ep *ep, const void *pdata, u8 plen) 1478{ 1479 int mpalen; 1480 struct mpa_message *mpa; 1481 struct mbuf *m; 1482 struct mpa_v2_conn_params mpa_v2_params; 1483 int err; 1484 1485 CTR2(KTR_IW_CXGBE, "%s:smrepB %p", __func__, ep); 1486 1487 mpalen = sizeof(*mpa) + plen; 1488 1489 if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) { 1490 1491 CTR3(KTR_IW_CXGBE, "%s:smrep1 %p %d", __func__, ep, 1492 ep->mpa_attr.version); 1493 mpalen += sizeof(struct mpa_v2_conn_params); 1494 } 1495 1496 mpa = malloc(mpalen, M_CXGBE, M_NOWAIT); 1497 if (mpa == NULL) 1498 return (-ENOMEM); 1499 1500 memset(mpa, 0, sizeof(*mpa)); 1501 memcpy(mpa->key, MPA_KEY_REP, sizeof(mpa->key)); 1502 mpa->flags = (ep->mpa_attr.crc_enabled ? MPA_CRC : 0) | 1503 (markers_enabled ? MPA_MARKERS : 0); 1504 mpa->revision = ep->mpa_attr.version; 1505 mpa->private_data_size = htons(plen); 1506 1507 if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) { 1508 1509 mpa->flags |= MPA_ENHANCED_RDMA_CONN; 1510 mpa->private_data_size += 1511 htons(sizeof(struct mpa_v2_conn_params)); 1512 mpa_v2_params.ird = htons((u16)ep->ird); 1513 mpa_v2_params.ord = htons((u16)ep->ord); 1514 CTR5(KTR_IW_CXGBE, "%s:smrep3 %p %d %d %d", __func__, ep, 1515 ep->mpa_attr.version, mpa_v2_params.ird, mpa_v2_params.ord); 1516 1517 if (peer2peer && (ep->mpa_attr.p2p_type != 1518 FW_RI_INIT_P2PTYPE_DISABLED)) { 1519 1520 mpa_v2_params.ird |= htons(MPA_V2_PEER2PEER_MODEL); 1521 1522 if (p2p_type == FW_RI_INIT_P2PTYPE_RDMA_WRITE) { 1523 1524 mpa_v2_params.ord |= 1525 htons(MPA_V2_RDMA_WRITE_RTR); 1526 CTR5(KTR_IW_CXGBE, "%s:smrep4 %p %d %d %d", 1527 __func__, ep, p2p_type, mpa_v2_params.ird, 1528 mpa_v2_params.ord); 1529 } 1530 else if (p2p_type == FW_RI_INIT_P2PTYPE_READ_REQ) { 1531 1532 mpa_v2_params.ord |= 1533 htons(MPA_V2_RDMA_READ_RTR); 1534 CTR5(KTR_IW_CXGBE, "%s:smrep5 %p %d %d %d", 1535 __func__, ep, p2p_type, mpa_v2_params.ird, 1536 mpa_v2_params.ord); 1537 } 1538 } 1539 1540 memcpy(mpa->private_data, &mpa_v2_params, 1541 sizeof(struct mpa_v2_conn_params)); 1542 1543 if (ep->plen) 1544 memcpy(mpa->private_data + 1545 sizeof(struct mpa_v2_conn_params), pdata, plen); 1546 } else 1547 if (plen) 1548 memcpy(mpa->private_data, pdata, plen); 1549 1550 m = m_getm(NULL, mpalen, M_NOWAIT, MT_DATA); 1551 if (m == NULL) { 1552 free(mpa, M_CXGBE); 1553 return (-ENOMEM); 1554 } 1555 m_copyback(m, 0, mpalen, (void *)mpa); 1556 free(mpa, M_CXGBE); 1557 1558 1559 ep->com.state = MPA_REP_SENT; 1560 ep->snd_seq += mpalen; 1561 err = -sosend(ep->com.so, NULL, NULL, m, NULL, MSG_DONTWAIT, 1562 ep->com.thread); 1563 CTR3(KTR_IW_CXGBE, "%s:smrepE %p %d", __func__, ep, err); 1564 return err; 1565} 1566 1567 1568 1569static void close_complete_upcall(struct c4iw_ep *ep, int status) 1570{ 1571 struct iw_cm_event event; 1572 1573 CTR2(KTR_IW_CXGBE, "%s:ccuB %p", __func__, ep); 1574 memset(&event, 0, sizeof(event)); 1575 event.event = IW_CM_EVENT_CLOSE; 1576 event.status = status; 1577 1578 if (ep->com.cm_id) { 1579 1580 CTR2(KTR_IW_CXGBE, "%s:ccu1 %1", __func__, ep); 1581 ep->com.cm_id->event_handler(ep->com.cm_id, &event); 1582 deref_cm_id(&ep->com); 1583 set_bit(CLOSE_UPCALL, &ep->com.history); 1584 } 1585 CTR2(KTR_IW_CXGBE, "%s:ccuE %p", __func__, ep); 1586} 1587 1588static int 1589send_abort(struct c4iw_ep *ep) 1590{ 1591 struct socket *so = ep->com.so; 1592 struct sockopt sopt; 1593 int rc; 1594 struct linger l; 1595 1596 CTR5(KTR_IW_CXGBE, "%s ep %p so %p state %s tid %d", __func__, ep, so, 1597 states[ep->com.state], ep->hwtid); 1598 1599 l.l_onoff = 1; 1600 l.l_linger = 0; 1601 1602 /* linger_time of 0 forces RST to be sent */ 1603 sopt.sopt_dir = SOPT_SET; 1604 sopt.sopt_level = SOL_SOCKET; 1605 sopt.sopt_name = SO_LINGER; 1606 sopt.sopt_val = (caddr_t)&l; 1607 sopt.sopt_valsize = sizeof l; 1608 sopt.sopt_td = NULL; 1609 rc = sosetopt(so, &sopt); 1610 if (rc != 0) { 1611 log(LOG_ERR, "%s: sosetopt(%p, linger = 0) failed with %d.\n", 1612 __func__, so, rc); 1613 } 1614 1615 uninit_iwarp_socket(so); 1616 soclose(so); 1617 set_bit(ABORT_CONN, &ep->com.history); 1618 1619 /* 1620 * TBD: iw_cxgbe driver should receive ABORT reply for every ABORT 1621 * request it has sent. But the current TOE driver is not propagating 1622 * this ABORT reply event (via do_abort_rpl) to iw_cxgbe. So as a work- 1623 * around de-refererece 'ep' here instead of doing it in abort_rpl() 1624 * handler(not yet implemented) of iw_cxgbe driver. 1625 */ 1626 release_ep_resources(ep); 1627 1628 return (0); 1629} 1630 1631static void peer_close_upcall(struct c4iw_ep *ep) 1632{ 1633 struct iw_cm_event event; 1634 1635 CTR2(KTR_IW_CXGBE, "%s:pcuB %p", __func__, ep); 1636 memset(&event, 0, sizeof(event)); 1637 event.event = IW_CM_EVENT_DISCONNECT; 1638 1639 if (ep->com.cm_id) { 1640 1641 CTR2(KTR_IW_CXGBE, "%s:pcu1 %p", __func__, ep); 1642 ep->com.cm_id->event_handler(ep->com.cm_id, &event); 1643 set_bit(DISCONN_UPCALL, &ep->com.history); 1644 } 1645 CTR2(KTR_IW_CXGBE, "%s:pcuE %p", __func__, ep); 1646} 1647 1648static void peer_abort_upcall(struct c4iw_ep *ep) 1649{ 1650 struct iw_cm_event event; 1651 1652 CTR2(KTR_IW_CXGBE, "%s:pauB %p", __func__, ep); 1653 memset(&event, 0, sizeof(event)); 1654 event.event = IW_CM_EVENT_CLOSE; 1655 event.status = -ECONNRESET; 1656 1657 if (ep->com.cm_id) { 1658 1659 CTR2(KTR_IW_CXGBE, "%s:pau1 %p", __func__, ep); 1660 ep->com.cm_id->event_handler(ep->com.cm_id, &event); 1661 deref_cm_id(&ep->com); 1662 set_bit(ABORT_UPCALL, &ep->com.history); 1663 } 1664 CTR2(KTR_IW_CXGBE, "%s:pauE %p", __func__, ep); 1665} 1666 1667static void connect_reply_upcall(struct c4iw_ep *ep, int status) 1668{ 1669 struct iw_cm_event event; 1670 1671 CTR3(KTR_IW_CXGBE, "%s:cruB %p, status: %d", __func__, ep, status); 1672 memset(&event, 0, sizeof(event)); 1673 event.event = IW_CM_EVENT_CONNECT_REPLY; 1674 event.status = ((status == -ECONNABORTED) || (status == -EPIPE)) ? 1675 -ECONNRESET : status; 1676 event.local_addr = ep->com.local_addr; 1677 event.remote_addr = ep->com.remote_addr; 1678 1679 if ((status == 0) || (status == -ECONNREFUSED)) { 1680 1681 if (!ep->tried_with_mpa_v1) { 1682 1683 CTR2(KTR_IW_CXGBE, "%s:cru1 %p", __func__, ep); 1684 /* this means MPA_v2 is used */ 1685 event.ord = ep->ird; 1686 event.ird = ep->ord; 1687 event.private_data_len = ep->plen - 1688 sizeof(struct mpa_v2_conn_params); 1689 event.private_data = ep->mpa_pkt + 1690 sizeof(struct mpa_message) + 1691 sizeof(struct mpa_v2_conn_params); 1692 } else { 1693 1694 CTR2(KTR_IW_CXGBE, "%s:cru2 %p", __func__, ep); 1695 /* this means MPA_v1 is used */ 1696 event.ord = c4iw_max_read_depth; 1697 event.ird = c4iw_max_read_depth; 1698 event.private_data_len = ep->plen; 1699 event.private_data = ep->mpa_pkt + 1700 sizeof(struct mpa_message); 1701 } 1702 } 1703 1704 if (ep->com.cm_id) { 1705 1706 CTR2(KTR_IW_CXGBE, "%s:cru3 %p", __func__, ep); 1707 set_bit(CONN_RPL_UPCALL, &ep->com.history); 1708 ep->com.cm_id->event_handler(ep->com.cm_id, &event); 1709 } 1710 1711 if(status == -ECONNABORTED) { 1712 1713 CTR3(KTR_IW_CXGBE, "%s:cruE %p %d", __func__, ep, status); 1714 return; 1715 } 1716 1717 if (status < 0) { 1718 1719 CTR3(KTR_IW_CXGBE, "%s:cru4 %p %d", __func__, ep, status); 1720 deref_cm_id(&ep->com); 1721 } 1722 1723 CTR2(KTR_IW_CXGBE, "%s:cruE %p", __func__, ep); 1724} 1725 1726static int connect_request_upcall(struct c4iw_ep *ep) 1727{ 1728 struct iw_cm_event event; 1729 int ret; 1730 1731 CTR3(KTR_IW_CXGBE, "%s: ep %p, mpa_v1 %d", __func__, ep, 1732 ep->tried_with_mpa_v1); 1733 1734 memset(&event, 0, sizeof(event)); 1735 event.event = IW_CM_EVENT_CONNECT_REQUEST; 1736 event.local_addr = ep->com.local_addr; 1737 event.remote_addr = ep->com.remote_addr; 1738 event.provider_data = ep; 1739 1740 if (!ep->tried_with_mpa_v1) { 1741 /* this means MPA_v2 is used */ 1742 event.ord = ep->ord; 1743 event.ird = ep->ird; 1744 event.private_data_len = ep->plen - 1745 sizeof(struct mpa_v2_conn_params); 1746 event.private_data = ep->mpa_pkt + sizeof(struct mpa_message) + 1747 sizeof(struct mpa_v2_conn_params); 1748 } else { 1749 1750 /* this means MPA_v1 is used. Send max supported */ 1751 event.ord = c4iw_max_read_depth; 1752 event.ird = c4iw_max_read_depth; 1753 event.private_data_len = ep->plen; 1754 event.private_data = ep->mpa_pkt + sizeof(struct mpa_message); 1755 } 1756 1757 c4iw_get_ep(&ep->com); 1758 ret = ep->parent_ep->com.cm_id->event_handler(ep->parent_ep->com.cm_id, 1759 &event); 1760 if(ret) { 1761 CTR3(KTR_IW_CXGBE, "%s: ep %p, Failure while notifying event to" 1762 " IWCM, err:%d", __func__, ep, ret); 1763 c4iw_put_ep(&ep->com); 1764 } else 1765 /* Dereference parent_ep only in success case. 1766 * In case of failure, parent_ep is dereferenced by the caller 1767 * of process_mpa_request(). 1768 */ 1769 c4iw_put_ep(&ep->parent_ep->com); 1770 1771 set_bit(CONNREQ_UPCALL, &ep->com.history); 1772 return ret; 1773} 1774 1775static void established_upcall(struct c4iw_ep *ep) 1776{ 1777 struct iw_cm_event event; 1778 1779 CTR2(KTR_IW_CXGBE, "%s:euB %p", __func__, ep); 1780 memset(&event, 0, sizeof(event)); 1781 event.event = IW_CM_EVENT_ESTABLISHED; 1782 event.ird = ep->ord; 1783 event.ord = ep->ird; 1784 1785 if (ep->com.cm_id) { 1786 1787 CTR2(KTR_IW_CXGBE, "%s:eu1 %p", __func__, ep); 1788 ep->com.cm_id->event_handler(ep->com.cm_id, &event); 1789 set_bit(ESTAB_UPCALL, &ep->com.history); 1790 } 1791 CTR2(KTR_IW_CXGBE, "%s:euE %p", __func__, ep); 1792} 1793 1794 1795#define RELAXED_IRD_NEGOTIATION 1 1796 1797/* 1798 * process_mpa_reply - process streaming mode MPA reply 1799 * 1800 * Returns: 1801 * 1802 * 0 upon success indicating a connect request was delivered to the ULP 1803 * or the mpa request is incomplete but valid so far. 1804 * 1805 * 1 if a failure requires the caller to close the connection. 1806 * 1807 * 2 if a failure requires the caller to abort the connection. 1808 */ 1809static int process_mpa_reply(struct c4iw_ep *ep) 1810{ 1811 struct mpa_message *mpa; 1812 struct mpa_v2_conn_params *mpa_v2_params; 1813 u16 plen; 1814 u16 resp_ird, resp_ord; 1815 u8 rtr_mismatch = 0, insuff_ird = 0; 1816 struct c4iw_qp_attributes attrs = {0}; 1817 enum c4iw_qp_attr_mask mask; 1818 int err; 1819 struct mbuf *top, *m; 1820 int flags = MSG_DONTWAIT; 1821 struct uio uio; 1822 int disconnect = 0; 1823 1824 CTR2(KTR_IW_CXGBE, "%s:pmrB %p", __func__, ep); 1825 1826 /* 1827 * Stop mpa timer. If it expired, then 1828 * we ignore the MPA reply. process_timeout() 1829 * will abort the connection. 1830 */ 1831 if (STOP_EP_TIMER(ep)) 1832 return 0; 1833 1834 uio.uio_resid = 1000000; 1835 uio.uio_td = ep->com.thread; 1836 err = soreceive(ep->com.so, NULL, &uio, &top, NULL, &flags); 1837 1838 if (err) { 1839 1840 if (err == EWOULDBLOCK) { 1841 1842 CTR2(KTR_IW_CXGBE, "%s:pmr1 %p", __func__, ep); 1843 START_EP_TIMER(ep); 1844 return 0; 1845 } 1846 err = -err; 1847 CTR2(KTR_IW_CXGBE, "%s:pmr2 %p", __func__, ep); 1848 goto err; 1849 } 1850 1851 if (ep->com.so->so_rcv.sb_mb) { 1852 1853 CTR2(KTR_IW_CXGBE, "%s:pmr3 %p", __func__, ep); 1854 printf("%s data after soreceive called! so %p sb_mb %p top %p\n", 1855 __func__, ep->com.so, ep->com.so->so_rcv.sb_mb, top); 1856 } 1857 1858 m = top; 1859 1860 do { 1861 1862 CTR2(KTR_IW_CXGBE, "%s:pmr4 %p", __func__, ep); 1863 /* 1864 * If we get more than the supported amount of private data 1865 * then we must fail this connection. 1866 */ 1867 if (ep->mpa_pkt_len + m->m_len > sizeof(ep->mpa_pkt)) { 1868 1869 CTR3(KTR_IW_CXGBE, "%s:pmr5 %p %d", __func__, ep, 1870 ep->mpa_pkt_len + m->m_len); 1871 err = (-EINVAL); 1872 goto err_stop_timer; 1873 } 1874 1875 /* 1876 * copy the new data into our accumulation buffer. 1877 */ 1878 m_copydata(m, 0, m->m_len, &(ep->mpa_pkt[ep->mpa_pkt_len])); 1879 ep->mpa_pkt_len += m->m_len; 1880 if (!m->m_next) 1881 m = m->m_nextpkt; 1882 else 1883 m = m->m_next; 1884 } while (m); 1885 1886 m_freem(top); 1887 /* 1888 * if we don't even have the mpa message, then bail. 1889 */ 1890 if (ep->mpa_pkt_len < sizeof(*mpa)) { 1891 return 0; 1892 } 1893 mpa = (struct mpa_message *) ep->mpa_pkt; 1894 1895 /* Validate MPA header. */ 1896 if (mpa->revision > mpa_rev) { 1897 1898 CTR4(KTR_IW_CXGBE, "%s:pmr6 %p %d %d", __func__, ep, 1899 mpa->revision, mpa_rev); 1900 printk(KERN_ERR MOD "%s MPA version mismatch. Local = %d, " 1901 " Received = %d\n", __func__, mpa_rev, mpa->revision); 1902 err = -EPROTO; 1903 goto err_stop_timer; 1904 } 1905 1906 if (memcmp(mpa->key, MPA_KEY_REP, sizeof(mpa->key))) { 1907 1908 CTR2(KTR_IW_CXGBE, "%s:pmr7 %p", __func__, ep); 1909 err = -EPROTO; 1910 goto err_stop_timer; 1911 } 1912 1913 plen = ntohs(mpa->private_data_size); 1914 1915 /* 1916 * Fail if there's too much private data. 1917 */ 1918 if (plen > MPA_MAX_PRIVATE_DATA) { 1919 1920 CTR2(KTR_IW_CXGBE, "%s:pmr8 %p", __func__, ep); 1921 err = -EPROTO; 1922 goto err_stop_timer; 1923 } 1924 1925 /* 1926 * If plen does not account for pkt size 1927 */ 1928 if (ep->mpa_pkt_len > (sizeof(*mpa) + plen)) { 1929 1930 CTR2(KTR_IW_CXGBE, "%s:pmr9 %p", __func__, ep); 1931 STOP_EP_TIMER(ep); 1932 err = -EPROTO; 1933 goto err_stop_timer; 1934 } 1935 1936 ep->plen = (u8) plen; 1937 1938 /* 1939 * If we don't have all the pdata yet, then bail. 1940 * We'll continue process when more data arrives. 1941 */ 1942 if (ep->mpa_pkt_len < (sizeof(*mpa) + plen)) { 1943 1944 CTR2(KTR_IW_CXGBE, "%s:pmra %p", __func__, ep); 1945 return 0; 1946 } 1947 1948 if (mpa->flags & MPA_REJECT) { 1949 1950 CTR2(KTR_IW_CXGBE, "%s:pmrb %p", __func__, ep); 1951 err = -ECONNREFUSED; 1952 goto err_stop_timer; 1953 } 1954 1955 /* 1956 * If we get here we have accumulated the entire mpa 1957 * start reply message including private data. And 1958 * the MPA header is valid. 1959 */ 1960 ep->com.state = FPDU_MODE; 1961 ep->mpa_attr.crc_enabled = (mpa->flags & MPA_CRC) | crc_enabled ? 1 : 0; 1962 ep->mpa_attr.recv_marker_enabled = markers_enabled; 1963 ep->mpa_attr.xmit_marker_enabled = mpa->flags & MPA_MARKERS ? 1 : 0; 1964 ep->mpa_attr.version = mpa->revision; 1965 ep->mpa_attr.p2p_type = FW_RI_INIT_P2PTYPE_DISABLED; 1966 1967 if (mpa->revision == 2) { 1968 1969 CTR2(KTR_IW_CXGBE, "%s:pmrc %p", __func__, ep); 1970 ep->mpa_attr.enhanced_rdma_conn = 1971 mpa->flags & MPA_ENHANCED_RDMA_CONN ? 1 : 0; 1972 1973 if (ep->mpa_attr.enhanced_rdma_conn) { 1974 1975 CTR2(KTR_IW_CXGBE, "%s:pmrd %p", __func__, ep); 1976 mpa_v2_params = (struct mpa_v2_conn_params *) 1977 (ep->mpa_pkt + sizeof(*mpa)); 1978 resp_ird = ntohs(mpa_v2_params->ird) & 1979 MPA_V2_IRD_ORD_MASK; 1980 resp_ord = ntohs(mpa_v2_params->ord) & 1981 MPA_V2_IRD_ORD_MASK; 1982 1983 /* 1984 * This is a double-check. Ideally, below checks are 1985 * not required since ird/ord stuff has been taken 1986 * care of in c4iw_accept_cr 1987 */ 1988 if (ep->ird < resp_ord) { 1989 if (RELAXED_IRD_NEGOTIATION && resp_ord <= 1990 ep->com.dev->rdev.adap->params.max_ordird_qp) 1991 ep->ird = resp_ord; 1992 else 1993 insuff_ird = 1; 1994 } else if (ep->ird > resp_ord) { 1995 ep->ird = resp_ord; 1996 } 1997 if (ep->ord > resp_ird) { 1998 if (RELAXED_IRD_NEGOTIATION) 1999 ep->ord = resp_ird; 2000 else 2001 insuff_ird = 1; 2002 } 2003 if (insuff_ird) { 2004 err = -ENOMEM; 2005 ep->ird = resp_ord; 2006 ep->ord = resp_ird; 2007 } 2008 2009 if (ntohs(mpa_v2_params->ird) & 2010 MPA_V2_PEER2PEER_MODEL) { 2011 2012 CTR2(KTR_IW_CXGBE, "%s:pmrf %p", __func__, ep); 2013 if (ntohs(mpa_v2_params->ord) & 2014 MPA_V2_RDMA_WRITE_RTR) { 2015 2016 CTR2(KTR_IW_CXGBE, "%s:pmrg %p", __func__, ep); 2017 ep->mpa_attr.p2p_type = 2018 FW_RI_INIT_P2PTYPE_RDMA_WRITE; 2019 } 2020 else if (ntohs(mpa_v2_params->ord) & 2021 MPA_V2_RDMA_READ_RTR) { 2022 2023 CTR2(KTR_IW_CXGBE, "%s:pmrh %p", __func__, ep); 2024 ep->mpa_attr.p2p_type = 2025 FW_RI_INIT_P2PTYPE_READ_REQ; 2026 } 2027 } 2028 } 2029 } else { 2030 2031 CTR2(KTR_IW_CXGBE, "%s:pmri %p", __func__, ep); 2032 2033 if (mpa->revision == 1) { 2034 2035 CTR2(KTR_IW_CXGBE, "%s:pmrj %p", __func__, ep); 2036 2037 if (peer2peer) { 2038 2039 CTR2(KTR_IW_CXGBE, "%s:pmrk %p", __func__, ep); 2040 ep->mpa_attr.p2p_type = p2p_type; 2041 } 2042 } 2043 } 2044 2045 if (set_tcpinfo(ep)) { 2046 2047 CTR2(KTR_IW_CXGBE, "%s:pmrl %p", __func__, ep); 2048 printf("%s set_tcpinfo error\n", __func__); 2049 err = -ECONNRESET; 2050 goto err; 2051 } 2052 2053 CTR6(KTR_IW_CXGBE, "%s - crc_enabled = %d, recv_marker_enabled = %d, " 2054 "xmit_marker_enabled = %d, version = %d p2p_type = %d", __func__, 2055 ep->mpa_attr.crc_enabled, ep->mpa_attr.recv_marker_enabled, 2056 ep->mpa_attr.xmit_marker_enabled, ep->mpa_attr.version, 2057 ep->mpa_attr.p2p_type); 2058 2059 /* 2060 * If responder's RTR does not match with that of initiator, assign 2061 * FW_RI_INIT_P2PTYPE_DISABLED in mpa attributes so that RTR is not 2062 * generated when moving QP to RTS state. 2063 * A TERM message will be sent after QP has moved to RTS state 2064 */ 2065 if ((ep->mpa_attr.version == 2) && peer2peer && 2066 (ep->mpa_attr.p2p_type != p2p_type)) { 2067 2068 CTR2(KTR_IW_CXGBE, "%s:pmrm %p", __func__, ep); 2069 ep->mpa_attr.p2p_type = FW_RI_INIT_P2PTYPE_DISABLED; 2070 rtr_mismatch = 1; 2071 } 2072 2073 2074 //ep->ofld_txq = TOEPCB(ep->com.so)->ofld_txq; 2075 attrs.mpa_attr = ep->mpa_attr; 2076 attrs.max_ird = ep->ird; 2077 attrs.max_ord = ep->ord; 2078 attrs.llp_stream_handle = ep; 2079 attrs.next_state = C4IW_QP_STATE_RTS; 2080 2081 mask = C4IW_QP_ATTR_NEXT_STATE | 2082 C4IW_QP_ATTR_LLP_STREAM_HANDLE | C4IW_QP_ATTR_MPA_ATTR | 2083 C4IW_QP_ATTR_MAX_IRD | C4IW_QP_ATTR_MAX_ORD; 2084 2085 /* bind QP and TID with INIT_WR */ 2086 err = c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp, mask, &attrs, 1); 2087 2088 if (err) { 2089 2090 CTR2(KTR_IW_CXGBE, "%s:pmrn %p", __func__, ep); 2091 goto err; 2092 } 2093 2094 /* 2095 * If responder's RTR requirement did not match with what initiator 2096 * supports, generate TERM message 2097 */ 2098 if (rtr_mismatch) { 2099 2100 CTR2(KTR_IW_CXGBE, "%s:pmro %p", __func__, ep); 2101 printk(KERN_ERR "%s: RTR mismatch, sending TERM\n", __func__); 2102 attrs.layer_etype = LAYER_MPA | DDP_LLP; 2103 attrs.ecode = MPA_NOMATCH_RTR; 2104 attrs.next_state = C4IW_QP_STATE_TERMINATE; 2105 attrs.send_term = 1; 2106 err = c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp, 2107 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1); 2108 err = -ENOMEM; 2109 disconnect = 1; 2110 goto out; 2111 } 2112 2113 /* 2114 * Generate TERM if initiator IRD is not sufficient for responder 2115 * provided ORD. Currently, we do the same behaviour even when 2116 * responder provided IRD is also not sufficient as regards to 2117 * initiator ORD. 2118 */ 2119 if (insuff_ird) { 2120 2121 CTR2(KTR_IW_CXGBE, "%s:pmrp %p", __func__, ep); 2122 printk(KERN_ERR "%s: Insufficient IRD, sending TERM\n", 2123 __func__); 2124 attrs.layer_etype = LAYER_MPA | DDP_LLP; 2125 attrs.ecode = MPA_INSUFF_IRD; 2126 attrs.next_state = C4IW_QP_STATE_TERMINATE; 2127 attrs.send_term = 1; 2128 err = c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp, 2129 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1); 2130 err = -ENOMEM; 2131 disconnect = 1; 2132 goto out; 2133 } 2134 goto out; 2135err_stop_timer: 2136 STOP_EP_TIMER(ep); 2137err: 2138 disconnect = 2; 2139out: 2140 connect_reply_upcall(ep, err); 2141 CTR2(KTR_IW_CXGBE, "%s:pmrE %p", __func__, ep); 2142 return disconnect; 2143} 2144 2145/* 2146 * process_mpa_request - process streaming mode MPA request 2147 * 2148 * Returns: 2149 * 2150 * 0 upon success indicating a connect request was delivered to the ULP 2151 * or the mpa request is incomplete but valid so far. 2152 * 2153 * 1 if a failure requires the caller to close the connection. 2154 * 2155 * 2 if a failure requires the caller to abort the connection. 2156 */ 2157static int 2158process_mpa_request(struct c4iw_ep *ep) 2159{ 2160 struct mpa_message *mpa; 2161 struct mpa_v2_conn_params *mpa_v2_params; 2162 u16 plen; 2163 int flags = MSG_DONTWAIT; 2164 int rc; 2165 struct iovec iov; 2166 struct uio uio; 2167 enum c4iw_ep_state state = ep->com.state; 2168 2169 CTR3(KTR_IW_CXGBE, "%s: ep %p, state %s", __func__, ep, states[state]); 2170 2171 if (state != MPA_REQ_WAIT) 2172 return 0; 2173 2174 iov.iov_base = &ep->mpa_pkt[ep->mpa_pkt_len]; 2175 iov.iov_len = sizeof(ep->mpa_pkt) - ep->mpa_pkt_len; 2176 uio.uio_iov = &iov; 2177 uio.uio_iovcnt = 1; 2178 uio.uio_offset = 0; 2179 uio.uio_resid = sizeof(ep->mpa_pkt) - ep->mpa_pkt_len; 2180 uio.uio_segflg = UIO_SYSSPACE; 2181 uio.uio_rw = UIO_READ; 2182 uio.uio_td = NULL; /* uio.uio_td = ep->com.thread; */ 2183 2184 rc = soreceive(ep->com.so, NULL, &uio, NULL, NULL, &flags); 2185 if (rc == EAGAIN) 2186 return 0; 2187 else if (rc) 2188 goto err_stop_timer; 2189 2190 KASSERT(uio.uio_offset > 0, ("%s: sorecieve on so %p read no data", 2191 __func__, ep->com.so)); 2192 ep->mpa_pkt_len += uio.uio_offset; 2193 2194 /* 2195 * If we get more than the supported amount of private data then we must 2196 * fail this connection. XXX: check so_rcv->sb_cc, or peek with another 2197 * soreceive, or increase the size of mpa_pkt by 1 and abort if the last 2198 * byte is filled by the soreceive above. 2199 */ 2200 2201 /* Don't even have the MPA message. Wait for more data to arrive. */ 2202 if (ep->mpa_pkt_len < sizeof(*mpa)) 2203 return 0; 2204 mpa = (struct mpa_message *) ep->mpa_pkt; 2205 2206 /* 2207 * Validate MPA Header. 2208 */ 2209 if (mpa->revision > mpa_rev) { 2210 log(LOG_ERR, "%s: MPA version mismatch. Local = %d," 2211 " Received = %d\n", __func__, mpa_rev, mpa->revision); 2212 goto err_stop_timer; 2213 } 2214 2215 if (memcmp(mpa->key, MPA_KEY_REQ, sizeof(mpa->key))) 2216 goto err_stop_timer; 2217 2218 /* 2219 * Fail if there's too much private data. 2220 */ 2221 plen = ntohs(mpa->private_data_size); 2222 if (plen > MPA_MAX_PRIVATE_DATA) 2223 goto err_stop_timer; 2224 2225 /* 2226 * If plen does not account for pkt size 2227 */ 2228 if (ep->mpa_pkt_len > (sizeof(*mpa) + plen)) 2229 goto err_stop_timer; 2230 2231 ep->plen = (u8) plen; 2232 2233 /* 2234 * If we don't have all the pdata yet, then bail. 2235 */ 2236 if (ep->mpa_pkt_len < (sizeof(*mpa) + plen)) 2237 return 0; 2238 2239 /* 2240 * If we get here we have accumulated the entire mpa 2241 * start reply message including private data. 2242 */ 2243 ep->mpa_attr.initiator = 0; 2244 ep->mpa_attr.crc_enabled = (mpa->flags & MPA_CRC) | crc_enabled ? 1 : 0; 2245 ep->mpa_attr.recv_marker_enabled = markers_enabled; 2246 ep->mpa_attr.xmit_marker_enabled = mpa->flags & MPA_MARKERS ? 1 : 0; 2247 ep->mpa_attr.version = mpa->revision; 2248 if (mpa->revision == 1) 2249 ep->tried_with_mpa_v1 = 1; 2250 ep->mpa_attr.p2p_type = FW_RI_INIT_P2PTYPE_DISABLED; 2251 2252 if (mpa->revision == 2) { 2253 ep->mpa_attr.enhanced_rdma_conn = 2254 mpa->flags & MPA_ENHANCED_RDMA_CONN ? 1 : 0; 2255 if (ep->mpa_attr.enhanced_rdma_conn) { 2256 mpa_v2_params = (struct mpa_v2_conn_params *) 2257 (ep->mpa_pkt + sizeof(*mpa)); 2258 ep->ird = ntohs(mpa_v2_params->ird) & 2259 MPA_V2_IRD_ORD_MASK; 2260 ep->ird = min_t(u32, ep->ird, 2261 cur_max_read_depth(ep->com.dev)); 2262 ep->ord = ntohs(mpa_v2_params->ord) & 2263 MPA_V2_IRD_ORD_MASK; 2264 ep->ord = min_t(u32, ep->ord, 2265 cur_max_read_depth(ep->com.dev)); 2266 CTR3(KTR_IW_CXGBE, "%s initiator ird %u ord %u\n", 2267 __func__, ep->ird, ep->ord); 2268 if (ntohs(mpa_v2_params->ird) & MPA_V2_PEER2PEER_MODEL) 2269 if (peer2peer) { 2270 if (ntohs(mpa_v2_params->ord) & 2271 MPA_V2_RDMA_WRITE_RTR) 2272 ep->mpa_attr.p2p_type = 2273 FW_RI_INIT_P2PTYPE_RDMA_WRITE; 2274 else if (ntohs(mpa_v2_params->ord) & 2275 MPA_V2_RDMA_READ_RTR) 2276 ep->mpa_attr.p2p_type = 2277 FW_RI_INIT_P2PTYPE_READ_REQ; 2278 } 2279 } 2280 } else if (mpa->revision == 1 && peer2peer) 2281 ep->mpa_attr.p2p_type = p2p_type; 2282 2283 if (set_tcpinfo(ep)) 2284 goto err_stop_timer; 2285 2286 CTR5(KTR_IW_CXGBE, "%s: crc_enabled = %d, recv_marker_enabled = %d, " 2287 "xmit_marker_enabled = %d, version = %d", __func__, 2288 ep->mpa_attr.crc_enabled, ep->mpa_attr.recv_marker_enabled, 2289 ep->mpa_attr.xmit_marker_enabled, ep->mpa_attr.version); 2290 2291 ep->com.state = MPA_REQ_RCVD; 2292 STOP_EP_TIMER(ep); 2293 2294 /* drive upcall */ 2295 if (ep->parent_ep->com.state != DEAD) 2296 if (connect_request_upcall(ep)) 2297 goto err_out; 2298 return 0; 2299 2300err_stop_timer: 2301 STOP_EP_TIMER(ep); 2302err_out: 2303 return 2; 2304} 2305 2306/* 2307 * Upcall from the adapter indicating data has been transmitted. 2308 * For us its just the single MPA request or reply. We can now free 2309 * the skb holding the mpa message. 2310 */ 2311int c4iw_reject_cr(struct iw_cm_id *cm_id, const void *pdata, u8 pdata_len) 2312{ 2313 int err; 2314 struct c4iw_ep *ep = to_ep(cm_id); 2315 int abort = 0; 2316 2317 mutex_lock(&ep->com.mutex); 2318 CTR2(KTR_IW_CXGBE, "%s:crcB %p", __func__, ep); 2319 2320 if ((ep->com.state == DEAD) || 2321 (ep->com.state != MPA_REQ_RCVD)) { 2322 2323 CTR2(KTR_IW_CXGBE, "%s:crc1 %p", __func__, ep); 2324 mutex_unlock(&ep->com.mutex); 2325 c4iw_put_ep(&ep->com); 2326 return -ECONNRESET; 2327 } 2328 set_bit(ULP_REJECT, &ep->com.history); 2329 2330 if (mpa_rev == 0) { 2331 2332 CTR2(KTR_IW_CXGBE, "%s:crc2 %p", __func__, ep); 2333 abort = 1; 2334 } 2335 else { 2336 2337 CTR2(KTR_IW_CXGBE, "%s:crc3 %p", __func__, ep); 2338 abort = send_mpa_reject(ep, pdata, pdata_len); 2339 } 2340 STOP_EP_TIMER(ep); 2341 err = c4iw_ep_disconnect(ep, abort != 0, GFP_KERNEL); 2342 mutex_unlock(&ep->com.mutex); 2343 c4iw_put_ep(&ep->com); 2344 CTR3(KTR_IW_CXGBE, "%s:crc4 %p, err: %d", __func__, ep, err); 2345 return 0; 2346} 2347 2348int c4iw_accept_cr(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param) 2349{ 2350 int err; 2351 struct c4iw_qp_attributes attrs = {0}; 2352 enum c4iw_qp_attr_mask mask; 2353 struct c4iw_ep *ep = to_ep(cm_id); 2354 struct c4iw_dev *h = to_c4iw_dev(cm_id->device); 2355 struct c4iw_qp *qp = get_qhp(h, conn_param->qpn); 2356 int abort = 0; 2357 2358 mutex_lock(&ep->com.mutex); 2359 CTR2(KTR_IW_CXGBE, "%s:cacB %p", __func__, ep); 2360 2361 if ((ep->com.state == DEAD) || 2362 (ep->com.state != MPA_REQ_RCVD)) { 2363 2364 CTR2(KTR_IW_CXGBE, "%s:cac1 %p", __func__, ep); 2365 err = -ECONNRESET; 2366 goto err_out; 2367 } 2368 2369 BUG_ON(!qp); 2370 2371 set_bit(ULP_ACCEPT, &ep->com.history); 2372 2373 if ((conn_param->ord > c4iw_max_read_depth) || 2374 (conn_param->ird > c4iw_max_read_depth)) { 2375 2376 CTR2(KTR_IW_CXGBE, "%s:cac2 %p", __func__, ep); 2377 err = -EINVAL; 2378 goto err_abort; 2379 } 2380 2381 if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) { 2382 2383 CTR2(KTR_IW_CXGBE, "%s:cac3 %p", __func__, ep); 2384 2385 if (conn_param->ord > ep->ird) { 2386 if (RELAXED_IRD_NEGOTIATION) { 2387 conn_param->ord = ep->ird; 2388 } else { 2389 ep->ird = conn_param->ird; 2390 ep->ord = conn_param->ord; 2391 send_mpa_reject(ep, conn_param->private_data, 2392 conn_param->private_data_len); 2393 err = -ENOMEM; 2394 goto err_abort; 2395 } 2396 } 2397 if (conn_param->ird < ep->ord) { 2398 if (RELAXED_IRD_NEGOTIATION && 2399 ep->ord <= h->rdev.adap->params.max_ordird_qp) { 2400 conn_param->ird = ep->ord; 2401 } else { 2402 err = -ENOMEM; 2403 goto err_abort; 2404 } 2405 } 2406 } 2407 ep->ird = conn_param->ird; 2408 ep->ord = conn_param->ord; 2409 2410 if (ep->mpa_attr.version == 1) { 2411 if (peer2peer && ep->ird == 0) 2412 ep->ird = 1; 2413 } else { 2414 if (peer2peer && 2415 (ep->mpa_attr.p2p_type != FW_RI_INIT_P2PTYPE_DISABLED) && 2416 (p2p_type == FW_RI_INIT_P2PTYPE_READ_REQ) && ep->ird == 0) 2417 ep->ird = 1; 2418 } 2419 2420 CTR4(KTR_IW_CXGBE, "%s %d ird %d ord %d\n", __func__, __LINE__, 2421 ep->ird, ep->ord); 2422 2423 ep->com.cm_id = cm_id; 2424 ref_cm_id(&ep->com); 2425 ep->com.qp = qp; 2426 ref_qp(ep); 2427 //ep->ofld_txq = TOEPCB(ep->com.so)->ofld_txq; 2428 2429 /* bind QP to EP and move to RTS */ 2430 attrs.mpa_attr = ep->mpa_attr; 2431 attrs.max_ird = ep->ird; 2432 attrs.max_ord = ep->ord; 2433 attrs.llp_stream_handle = ep; 2434 attrs.next_state = C4IW_QP_STATE_RTS; 2435 2436 /* bind QP and TID with INIT_WR */ 2437 mask = C4IW_QP_ATTR_NEXT_STATE | 2438 C4IW_QP_ATTR_LLP_STREAM_HANDLE | 2439 C4IW_QP_ATTR_MPA_ATTR | 2440 C4IW_QP_ATTR_MAX_IRD | 2441 C4IW_QP_ATTR_MAX_ORD; 2442 2443 err = c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp, mask, &attrs, 1); 2444 if (err) { 2445 CTR3(KTR_IW_CXGBE, "%s:caca %p, err: %d", __func__, ep, err); 2446 goto err_defef_cm_id; 2447 } 2448 2449 err = send_mpa_reply(ep, conn_param->private_data, 2450 conn_param->private_data_len); 2451 if (err) { 2452 CTR3(KTR_IW_CXGBE, "%s:cacb %p, err: %d", __func__, ep, err); 2453 goto err_defef_cm_id; 2454 } 2455 2456 ep->com.state = FPDU_MODE; 2457 established_upcall(ep); 2458 mutex_unlock(&ep->com.mutex); 2459 c4iw_put_ep(&ep->com); 2460 CTR2(KTR_IW_CXGBE, "%s:cacE %p", __func__, ep); 2461 return 0; 2462err_defef_cm_id: 2463 deref_cm_id(&ep->com); 2464err_abort: 2465 abort = 1; 2466err_out: 2467 if (abort) 2468 c4iw_ep_disconnect(ep, 1, GFP_KERNEL); 2469 mutex_unlock(&ep->com.mutex); 2470 c4iw_put_ep(&ep->com); 2471 CTR2(KTR_IW_CXGBE, "%s:cacE err %p", __func__, ep); 2472 return err; 2473} 2474 2475static int 2476c4iw_sock_create(struct sockaddr_storage *laddr, struct socket **so) 2477{ 2478 int ret; 2479 int size; 2480 struct socket *sock = NULL; 2481 2482 ret = sock_create_kern(laddr->ss_family, 2483 SOCK_STREAM, IPPROTO_TCP, &sock); 2484 if (ret) { 2485 CTR2(KTR_IW_CXGBE, "%s:Failed to create TCP socket. err %d", 2486 __func__, ret); 2487 return ret; 2488 } 2489 2490 ret = sobind(sock, (struct sockaddr *)laddr, curthread); 2491 if (ret) { 2492 CTR2(KTR_IW_CXGBE, "%s:Failed to bind socket. err %p", 2493 __func__, ret); 2494 sock_release(sock); 2495 return ret; 2496 } 2497 2498 size = laddr->ss_family == AF_INET6 ? 2499 sizeof(struct sockaddr_in6) : sizeof(struct sockaddr_in); 2500 ret = sock_getname(sock, (struct sockaddr *)laddr, &size, 0); 2501 if (ret) { 2502 CTR2(KTR_IW_CXGBE, "%s:sock_getname failed. err %p", 2503 __func__, ret); 2504 sock_release(sock); 2505 return ret; 2506 } 2507 2508 *so = sock; 2509 return 0; 2510} 2511 2512int c4iw_connect(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param) 2513{ 2514 int err = 0; 2515 struct c4iw_dev *dev = to_c4iw_dev(cm_id->device); 2516 struct c4iw_ep *ep = NULL; 2517 struct ifnet *nh_ifp; /* Logical egress interface */ 2518#ifdef VIMAGE 2519 struct rdma_cm_id *rdma_id = (struct rdma_cm_id*)cm_id->context; 2520 struct vnet *vnet = rdma_id->route.addr.dev_addr.net; 2521#endif 2522 2523 CTR2(KTR_IW_CXGBE, "%s:ccB %p", __func__, cm_id); 2524 2525 2526 if ((conn_param->ord > c4iw_max_read_depth) || 2527 (conn_param->ird > c4iw_max_read_depth)) { 2528 2529 CTR2(KTR_IW_CXGBE, "%s:cc1 %p", __func__, cm_id); 2530 err = -EINVAL; 2531 goto out; 2532 } 2533 ep = alloc_ep(sizeof(*ep), GFP_KERNEL); 2534 cm_id->provider_data = ep; 2535 2536 init_timer(&ep->timer); 2537 ep->plen = conn_param->private_data_len; 2538 2539 if (ep->plen) { 2540 2541 CTR2(KTR_IW_CXGBE, "%s:cc3 %p", __func__, ep); 2542 memcpy(ep->mpa_pkt + sizeof(struct mpa_message), 2543 conn_param->private_data, ep->plen); 2544 } 2545 ep->ird = conn_param->ird; 2546 ep->ord = conn_param->ord; 2547 2548 if (peer2peer && ep->ord == 0) { 2549 2550 CTR2(KTR_IW_CXGBE, "%s:cc4 %p", __func__, ep); 2551 ep->ord = 1; 2552 } 2553 2554 ep->com.dev = dev; 2555 ep->com.cm_id = cm_id; 2556 ref_cm_id(&ep->com); 2557 ep->com.qp = get_qhp(dev, conn_param->qpn); 2558 2559 if (!ep->com.qp) { 2560 2561 CTR2(KTR_IW_CXGBE, "%s:cc5 %p", __func__, ep); 2562 err = -EINVAL; 2563 goto fail; 2564 } 2565 ref_qp(ep); 2566 ep->com.thread = curthread; 2567 2568 CURVNET_SET(vnet); 2569 err = get_ifnet_from_raddr(&cm_id->remote_addr, &nh_ifp); 2570 CURVNET_RESTORE(); 2571 2572 if (err) { 2573 2574 CTR2(KTR_IW_CXGBE, "%s:cc7 %p", __func__, ep); 2575 printk(KERN_ERR MOD "%s - cannot find route.\n", __func__); 2576 err = EHOSTUNREACH; 2577 return err; 2578 } 2579 2580 if (!(nh_ifp->if_capenable & IFCAP_TOE) || 2581 TOEDEV(nh_ifp) == NULL) { 2582 err = -ENOPROTOOPT; 2583 goto fail; 2584 } 2585 ep->com.state = CONNECTING; 2586 ep->tos = 0; 2587 ep->com.local_addr = cm_id->local_addr; 2588 ep->com.remote_addr = cm_id->remote_addr; 2589 2590 err = c4iw_sock_create(&cm_id->local_addr, &ep->com.so); 2591 if (err) 2592 goto fail; 2593 2594 setiwsockopt(ep->com.so); 2595 err = -soconnect(ep->com.so, (struct sockaddr *)&ep->com.remote_addr, 2596 ep->com.thread); 2597 if (!err) { 2598 init_iwarp_socket(ep->com.so, &ep->com); 2599 goto out; 2600 } else 2601 goto fail_free_so; 2602 2603fail_free_so: 2604 sock_release(ep->com.so); 2605fail: 2606 deref_cm_id(&ep->com); 2607 c4iw_put_ep(&ep->com); 2608 ep = NULL; 2609out: 2610 CTR2(KTR_IW_CXGBE, "%s:ccE ret:%d", __func__, err); 2611 return err; 2612} 2613 2614/* 2615 * iwcm->create_listen. Returns -errno on failure. 2616 */ 2617int 2618c4iw_create_listen(struct iw_cm_id *cm_id, int backlog) 2619{ 2620 struct c4iw_dev *dev = to_c4iw_dev(cm_id->device); 2621 struct c4iw_listen_ep *lep = NULL; 2622 struct listen_port_info *port_info = NULL; 2623 int rc = 0; 2624 2625 CTR3(KTR_IW_CXGBE, "%s: cm_id %p, backlog %s", __func__, cm_id, 2626 backlog); 2627 lep = alloc_ep(sizeof(*lep), GFP_KERNEL); 2628 lep->com.cm_id = cm_id; 2629 ref_cm_id(&lep->com); 2630 lep->com.dev = dev; 2631 lep->backlog = backlog; 2632 lep->com.local_addr = cm_id->local_addr; 2633 lep->com.thread = curthread; 2634 cm_id->provider_data = lep; 2635 lep->com.state = LISTEN; 2636 2637 /* In case of INDADDR_ANY, ibcore creates cmid for each device and 2638 * invokes iw_cxgbe listener callbacks assuming that iw_cxgbe creates 2639 * HW listeners for each device seperately. But toecore expects single 2640 * solisten() call with INADDR_ANY address to create HW listeners on 2641 * all devices for a given port number. So iw_cxgbe driver calls 2642 * solisten() only once for INADDR_ANY(usually done at first time 2643 * listener callback from ibcore). And all the subsequent INADDR_ANY 2644 * listener callbacks from ibcore(for the same port address) do not 2645 * invoke solisten() as first listener callback has already created 2646 * listeners for all other devices(via solisten). 2647 */ 2648 if (c4iw_any_addr((struct sockaddr *)&lep->com.local_addr)) { 2649 port_info = add_ep_to_listenlist(lep); 2650 /* skip solisten() if refcnt > 1, as the listeners were 2651 * alredy created by 'Master lep' 2652 */ 2653 if (port_info->refcnt > 1) { 2654 /* As there will be only one listener socket for a TCP 2655 * port, copy Master lep's socket pointer to other lep's 2656 * that are belonging to same TCP port. 2657 */ 2658 struct c4iw_listen_ep *head_lep = 2659 container_of(port_info->lep_list.next, 2660 struct c4iw_listen_ep, listen_ep_list); 2661 lep->com.so = head_lep->com.so; 2662 goto out; 2663 } 2664 } 2665 rc = c4iw_sock_create(&cm_id->local_addr, &lep->com.so); 2666 if (rc) { 2667 CTR2(KTR_IW_CXGBE, "%s:Failed to create socket. err %d", 2668 __func__, rc); 2669 goto fail; 2670 } 2671 2672 rc = solisten(lep->com.so, backlog, curthread); 2673 if (rc) { 2674 CTR3(KTR_IW_CXGBE, "%s:Failed to listen on sock:%p. err %d", 2675 __func__, lep->com.so, rc); 2676 goto fail_free_so; 2677 } 2678 init_iwarp_socket(lep->com.so, &lep->com); 2679out: 2680 return 0; 2681 2682fail_free_so: 2683 sock_release(lep->com.so); 2684fail: 2685 if (port_info) 2686 rem_ep_from_listenlist(lep); 2687 deref_cm_id(&lep->com); 2688 c4iw_put_ep(&lep->com); 2689 return rc; 2690} 2691 2692int 2693c4iw_destroy_listen(struct iw_cm_id *cm_id) 2694{ 2695 struct c4iw_listen_ep *lep = to_listen_ep(cm_id); 2696 2697 mutex_lock(&lep->com.mutex); 2698 CTR3(KTR_IW_CXGBE, "%s: cm_id %p, state %s", __func__, cm_id, 2699 states[lep->com.state]); 2700 2701 lep->com.state = DEAD; 2702 if (c4iw_any_addr((struct sockaddr *)&lep->com.local_addr)) { 2703 /* if no refcount then close listen socket */ 2704 if (!rem_ep_from_listenlist(lep)) 2705 close_socket(lep->com.so); 2706 } else 2707 close_socket(lep->com.so); 2708 deref_cm_id(&lep->com); 2709 mutex_unlock(&lep->com.mutex); 2710 c4iw_put_ep(&lep->com); 2711 return 0; 2712} 2713 2714int __c4iw_ep_disconnect(struct c4iw_ep *ep, int abrupt, gfp_t gfp) 2715{ 2716 int ret; 2717 mutex_lock(&ep->com.mutex); 2718 ret = c4iw_ep_disconnect(ep, abrupt, gfp); 2719 mutex_unlock(&ep->com.mutex); 2720 return ret; 2721} 2722 2723int c4iw_ep_disconnect(struct c4iw_ep *ep, int abrupt, gfp_t gfp) 2724{ 2725 int ret = 0; 2726 int close = 0; 2727 int fatal = 0; 2728 struct c4iw_rdev *rdev; 2729 2730 2731 CTR2(KTR_IW_CXGBE, "%s:cedB %p", __func__, ep); 2732 2733 rdev = &ep->com.dev->rdev; 2734 2735 if (c4iw_fatal_error(rdev)) { 2736 2737 CTR2(KTR_IW_CXGBE, "%s:ced1 %p", __func__, ep); 2738 fatal = 1; 2739 close_complete_upcall(ep, -ECONNRESET); 2740 send_abort(ep); 2741 ep->com.state = DEAD; 2742 } 2743 CTR3(KTR_IW_CXGBE, "%s:ced2 %p %s", __func__, ep, 2744 states[ep->com.state]); 2745 2746 /* 2747 * Ref the ep here in case we have fatal errors causing the 2748 * ep to be released and freed. 2749 */ 2750 c4iw_get_ep(&ep->com); 2751 switch (ep->com.state) { 2752 2753 case MPA_REQ_WAIT: 2754 case MPA_REQ_SENT: 2755 case MPA_REQ_RCVD: 2756 case MPA_REP_SENT: 2757 case FPDU_MODE: 2758 close = 1; 2759 if (abrupt) 2760 ep->com.state = ABORTING; 2761 else { 2762 ep->com.state = CLOSING; 2763 START_EP_TIMER(ep); 2764 } 2765 set_bit(CLOSE_SENT, &ep->com.flags); 2766 break; 2767 2768 case CLOSING: 2769 2770 if (!test_and_set_bit(CLOSE_SENT, &ep->com.flags)) { 2771 2772 close = 1; 2773 if (abrupt) { 2774 STOP_EP_TIMER(ep); 2775 ep->com.state = ABORTING; 2776 } else 2777 ep->com.state = MORIBUND; 2778 } 2779 break; 2780 2781 case MORIBUND: 2782 case ABORTING: 2783 case DEAD: 2784 CTR3(KTR_IW_CXGBE, 2785 "%s ignoring disconnect ep %p state %u", __func__, 2786 ep, ep->com.state); 2787 break; 2788 2789 default: 2790 BUG(); 2791 break; 2792 } 2793 2794 2795 if (close) { 2796 2797 CTR2(KTR_IW_CXGBE, "%s:ced3 %p", __func__, ep); 2798 2799 if (abrupt) { 2800 2801 CTR2(KTR_IW_CXGBE, "%s:ced4 %p", __func__, ep); 2802 set_bit(EP_DISC_ABORT, &ep->com.history); 2803 close_complete_upcall(ep, -ECONNRESET); 2804 ret = send_abort(ep); 2805 if (ret) 2806 fatal = 1; 2807 } else { 2808 2809 CTR2(KTR_IW_CXGBE, "%s:ced5 %p", __func__, ep); 2810 set_bit(EP_DISC_CLOSE, &ep->com.history); 2811 2812 if (!ep->parent_ep) 2813 ep->com.state = MORIBUND; 2814 2815 CURVNET_SET(ep->com.so->so_vnet); 2816 sodisconnect(ep->com.so); 2817 CURVNET_RESTORE(); 2818 } 2819 2820 } 2821 2822 if (fatal) { 2823 set_bit(EP_DISC_FAIL, &ep->com.history); 2824 if (!abrupt) { 2825 STOP_EP_TIMER(ep); 2826 close_complete_upcall(ep, -EIO); 2827 } 2828 if (ep->com.qp) { 2829 struct c4iw_qp_attributes attrs = {0}; 2830 2831 attrs.next_state = C4IW_QP_STATE_ERROR; 2832 ret = c4iw_modify_qp(ep->com.dev, ep->com.qp, 2833 C4IW_QP_ATTR_NEXT_STATE, 2834 &attrs, 1); 2835 if (ret) { 2836 CTR2(KTR_IW_CXGBE, "%s:ced7 %p", __func__, ep); 2837 printf("%s - qp <- error failed!\n", __func__); 2838 } 2839 } 2840 release_ep_resources(ep); 2841 ep->com.state = DEAD; 2842 CTR2(KTR_IW_CXGBE, "%s:ced6 %p", __func__, ep); 2843 } 2844 c4iw_put_ep(&ep->com); 2845 CTR2(KTR_IW_CXGBE, "%s:cedE %p", __func__, ep); 2846 return ret; 2847} 2848 2849#ifdef C4IW_EP_REDIRECT 2850int c4iw_ep_redirect(void *ctx, struct dst_entry *old, struct dst_entry *new, 2851 struct l2t_entry *l2t) 2852{ 2853 struct c4iw_ep *ep = ctx; 2854 2855 if (ep->dst != old) 2856 return 0; 2857 2858 PDBG("%s ep %p redirect to dst %p l2t %p\n", __func__, ep, new, 2859 l2t); 2860 dst_hold(new); 2861 cxgb4_l2t_release(ep->l2t); 2862 ep->l2t = l2t; 2863 dst_release(old); 2864 ep->dst = new; 2865 return 1; 2866} 2867#endif 2868 2869 2870 2871static void ep_timeout(unsigned long arg) 2872{ 2873 struct c4iw_ep *ep = (struct c4iw_ep *)arg; 2874 2875 if (!test_and_set_bit(TIMEOUT, &ep->com.flags)) { 2876 2877 /* 2878 * Only insert if it is not already on the list. 2879 */ 2880 if (!(ep->com.ep_events & C4IW_EVENT_TIMEOUT)) { 2881 CTR2(KTR_IW_CXGBE, "%s:et1 %p", __func__, ep); 2882 add_ep_to_req_list(ep, C4IW_EVENT_TIMEOUT); 2883 } 2884 } 2885} 2886 2887static int fw6_wr_rpl(struct adapter *sc, const __be64 *rpl) 2888{ 2889 uint64_t val = be64toh(*rpl); 2890 int ret; 2891 struct c4iw_wr_wait *wr_waitp; 2892 2893 ret = (int)((val >> 8) & 0xff); 2894 wr_waitp = (struct c4iw_wr_wait *)rpl[1]; 2895 CTR3(KTR_IW_CXGBE, "%s wr_waitp %p ret %u", __func__, wr_waitp, ret); 2896 if (wr_waitp) 2897 c4iw_wake_up(wr_waitp, ret ? -ret : 0); 2898 2899 return (0); 2900} 2901 2902static int fw6_cqe_handler(struct adapter *sc, const __be64 *rpl) 2903{ 2904 struct cqe_list_entry *cle; 2905 unsigned long flag; 2906 2907 cle = malloc(sizeof(*cle), M_CXGBE, M_NOWAIT); 2908 cle->rhp = sc->iwarp_softc; 2909 cle->err_cqe = *(const struct t4_cqe *)(&rpl[0]); 2910 2911 spin_lock_irqsave(&err_cqe_lock, flag); 2912 list_add_tail(&cle->entry, &err_cqe_list); 2913 queue_work(c4iw_taskq, &c4iw_task); 2914 spin_unlock_irqrestore(&err_cqe_lock, flag); 2915 2916 return (0); 2917} 2918 2919static int 2920process_terminate(struct c4iw_ep *ep) 2921{ 2922 struct c4iw_qp_attributes attrs = {0}; 2923 2924 CTR2(KTR_IW_CXGBE, "%s:tB %p %d", __func__, ep); 2925 2926 if (ep && ep->com.qp) { 2927 2928 printk(KERN_WARNING MOD "TERM received tid %u qpid %u\n", 2929 ep->hwtid, ep->com.qp->wq.sq.qid); 2930 attrs.next_state = C4IW_QP_STATE_TERMINATE; 2931 c4iw_modify_qp(ep->com.dev, ep->com.qp, C4IW_QP_ATTR_NEXT_STATE, &attrs, 2932 1); 2933 } else 2934 printk(KERN_WARNING MOD "TERM received tid %u no ep/qp\n", 2935 ep->hwtid); 2936 CTR2(KTR_IW_CXGBE, "%s:tE %p %d", __func__, ep); 2937 2938 return 0; 2939} 2940 2941int __init c4iw_cm_init(void) 2942{ 2943 2944 t4_register_cpl_handler(CPL_RDMA_TERMINATE, terminate); 2945 t4_register_fw_msg_handler(FW6_TYPE_WR_RPL, fw6_wr_rpl); 2946 t4_register_fw_msg_handler(FW6_TYPE_CQE, fw6_cqe_handler); 2947 t4_register_an_handler(c4iw_ev_handler); 2948 2949 TAILQ_INIT(&req_list); 2950 spin_lock_init(&req_lock); 2951 INIT_LIST_HEAD(&err_cqe_list); 2952 spin_lock_init(&err_cqe_lock); 2953 2954 INIT_WORK(&c4iw_task, process_req); 2955 2956 c4iw_taskq = create_singlethread_workqueue("iw_cxgbe"); 2957 if (!c4iw_taskq) 2958 return -ENOMEM; 2959 2960 return 0; 2961} 2962 2963void __exit c4iw_cm_term(void) 2964{ 2965 WARN_ON(!TAILQ_EMPTY(&req_list)); 2966 WARN_ON(!list_empty(&err_cqe_list)); 2967 flush_workqueue(c4iw_taskq); 2968 destroy_workqueue(c4iw_taskq); 2969 2970 t4_register_cpl_handler(CPL_RDMA_TERMINATE, NULL); 2971 t4_register_fw_msg_handler(FW6_TYPE_WR_RPL, NULL); 2972 t4_register_fw_msg_handler(FW6_TYPE_CQE, NULL); 2973 t4_register_an_handler(NULL); 2974} 2975#endif 2976