cm.c revision 336667
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 336667 2018-07-24 13:13:30Z 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 78static spinlock_t req_lock; 79static TAILQ_HEAD(c4iw_ep_list, c4iw_ep_common) req_list; 80static struct work_struct c4iw_task; 81static struct workqueue_struct *c4iw_taskq; 82static LIST_HEAD(err_cqe_list); 83static spinlock_t err_cqe_lock; 84static LIST_HEAD(listen_port_list); 85static DEFINE_MUTEX(listen_port_mutex); 86 87static void process_req(struct work_struct *ctx); 88static void start_ep_timer(struct c4iw_ep *ep); 89static int stop_ep_timer(struct c4iw_ep *ep); 90static int set_tcpinfo(struct c4iw_ep *ep); 91static void process_timeout(struct c4iw_ep *ep); 92static void process_err_cqes(void); 93static void *alloc_ep(int size, gfp_t flags); 94static void close_socket(struct socket *so); 95static int send_mpa_req(struct c4iw_ep *ep); 96static int send_mpa_reject(struct c4iw_ep *ep, const void *pdata, u8 plen); 97static int send_mpa_reply(struct c4iw_ep *ep, const void *pdata, u8 plen); 98static void close_complete_upcall(struct c4iw_ep *ep, int status); 99static int send_abort(struct c4iw_ep *ep); 100static void peer_close_upcall(struct c4iw_ep *ep); 101static void peer_abort_upcall(struct c4iw_ep *ep); 102static void connect_reply_upcall(struct c4iw_ep *ep, int status); 103static int connect_request_upcall(struct c4iw_ep *ep); 104static void established_upcall(struct c4iw_ep *ep); 105static int process_mpa_reply(struct c4iw_ep *ep); 106static int process_mpa_request(struct c4iw_ep *ep); 107static void process_peer_close(struct c4iw_ep *ep); 108static void process_conn_error(struct c4iw_ep *ep); 109static void process_close_complete(struct c4iw_ep *ep); 110static void ep_timeout(unsigned long arg); 111static void setiwsockopt(struct socket *so); 112static void init_iwarp_socket(struct socket *so, void *arg); 113static void uninit_iwarp_socket(struct socket *so); 114static void process_data(struct c4iw_ep *ep); 115static void process_connected(struct c4iw_ep *ep); 116static int c4iw_so_upcall(struct socket *so, void *arg, int waitflag); 117static void process_socket_event(struct c4iw_ep *ep); 118static void release_ep_resources(struct c4iw_ep *ep); 119static int process_terminate(struct c4iw_ep *ep); 120static int terminate(struct sge_iq *iq, const struct rss_header *rss, 121 struct mbuf *m); 122static int add_ep_to_req_list(struct c4iw_ep *ep, int ep_events); 123static struct listen_port_info * 124add_ep_to_listenlist(struct c4iw_listen_ep *lep); 125static int rem_ep_from_listenlist(struct c4iw_listen_ep *lep); 126static struct c4iw_listen_ep * 127find_real_listen_ep(struct c4iw_listen_ep *master_lep, struct socket *so); 128static int get_ifnet_from_raddr(struct sockaddr_storage *raddr, 129 struct ifnet **ifp); 130static void process_newconn(struct c4iw_listen_ep *master_lep, 131 struct socket *new_so); 132#define START_EP_TIMER(ep) \ 133 do { \ 134 CTR3(KTR_IW_CXGBE, "start_ep_timer (%s:%d) ep %p", \ 135 __func__, __LINE__, (ep)); \ 136 start_ep_timer(ep); \ 137 } while (0) 138 139#define STOP_EP_TIMER(ep) \ 140 ({ \ 141 CTR3(KTR_IW_CXGBE, "stop_ep_timer (%s:%d) ep %p", \ 142 __func__, __LINE__, (ep)); \ 143 stop_ep_timer(ep); \ 144 }) 145 146#define GET_LOCAL_ADDR(pladdr, so) \ 147 do { \ 148 struct sockaddr_storage *__a = NULL; \ 149 struct inpcb *__inp = sotoinpcb(so); \ 150 KASSERT(__inp != NULL, \ 151 ("GET_LOCAL_ADDR(%s):so:%p, inp = NULL", __func__, so)); \ 152 if (__inp->inp_vflag & INP_IPV4) \ 153 in_getsockaddr(so, (struct sockaddr **)&__a); \ 154 else \ 155 in6_getsockaddr(so, (struct sockaddr **)&__a); \ 156 *(pladdr) = *__a; \ 157 free(__a, M_SONAME); \ 158 } while (0) 159 160#define GET_REMOTE_ADDR(praddr, so) \ 161 do { \ 162 struct sockaddr_storage *__a = NULL; \ 163 struct inpcb *__inp = sotoinpcb(so); \ 164 KASSERT(__inp != NULL, \ 165 ("GET_REMOTE_ADDR(%s):so:%p, inp = NULL", __func__, so)); \ 166 if (__inp->inp_vflag & INP_IPV4) \ 167 in_getpeeraddr(so, (struct sockaddr **)&__a); \ 168 else \ 169 in6_getpeeraddr(so, (struct sockaddr **)&__a); \ 170 *(praddr) = *__a; \ 171 free(__a, M_SONAME); \ 172 } while (0) 173 174#ifdef KTR 175static char *states[] = { 176 "idle", 177 "listen", 178 "connecting", 179 "mpa_wait_req", 180 "mpa_req_sent", 181 "mpa_req_rcvd", 182 "mpa_rep_sent", 183 "fpdu_mode", 184 "aborting", 185 "closing", 186 "moribund", 187 "dead", 188 NULL, 189}; 190#endif 191 192static void deref_cm_id(struct c4iw_ep_common *epc) 193{ 194 epc->cm_id->rem_ref(epc->cm_id); 195 epc->cm_id = NULL; 196 set_bit(CM_ID_DEREFED, &epc->history); 197} 198 199static void ref_cm_id(struct c4iw_ep_common *epc) 200{ 201 set_bit(CM_ID_REFED, &epc->history); 202 epc->cm_id->add_ref(epc->cm_id); 203} 204 205static void deref_qp(struct c4iw_ep *ep) 206{ 207 c4iw_qp_rem_ref(&ep->com.qp->ibqp); 208 clear_bit(QP_REFERENCED, &ep->com.flags); 209 set_bit(QP_DEREFED, &ep->com.history); 210} 211 212static void ref_qp(struct c4iw_ep *ep) 213{ 214 set_bit(QP_REFERENCED, &ep->com.flags); 215 set_bit(QP_REFED, &ep->com.history); 216 c4iw_qp_add_ref(&ep->com.qp->ibqp); 217} 218/* allocated per TCP port while listening */ 219struct listen_port_info { 220 uint16_t port_num; /* TCP port address */ 221 struct list_head list; /* belongs to listen_port_list */ 222 struct list_head lep_list; /* per port lep list */ 223 uint32_t refcnt; /* number of lep's listening */ 224}; 225 226/* 227 * Following two lists are used to manage INADDR_ANY listeners: 228 * 1)listen_port_list 229 * 2)lep_list 230 * 231 * Below is the INADDR_ANY listener lists overview on a system with a two port 232 * adapter: 233 * |------------------| 234 * |listen_port_list | 235 * |------------------| 236 * | 237 * | |-----------| |-----------| 238 * | | port_num:X| | port_num:X| 239 * |--------------|-list------|-------|-list------|-------.... 240 * | lep_list----| | lep_list----| 241 * | refcnt | | | refcnt | | 242 * | | | | | | 243 * | | | | | | 244 * |-----------| | |-----------| | 245 * | | 246 * | | 247 * | | 248 * | | lep1 lep2 249 * | | |----------------| |----------------| 250 * | |----| listen_ep_list |----| listen_ep_list | 251 * | |----------------| |----------------| 252 * | 253 * | 254 * | lep1 lep2 255 * | |----------------| |----------------| 256 * |---| listen_ep_list |----| listen_ep_list | 257 * |----------------| |----------------| 258 * 259 * Because of two port adapter, the number of lep's are two(lep1 & lep2) for 260 * each TCP port number. 261 * 262 * Here 'lep1' is always marked as Master lep, because solisten() is always 263 * called through first lep. 264 * 265 */ 266static struct listen_port_info * 267add_ep_to_listenlist(struct c4iw_listen_ep *lep) 268{ 269 uint16_t port; 270 struct listen_port_info *port_info = NULL; 271 struct sockaddr_storage *laddr = &lep->com.local_addr; 272 273 port = (laddr->ss_family == AF_INET) ? 274 ((struct sockaddr_in *)laddr)->sin_port : 275 ((struct sockaddr_in6 *)laddr)->sin6_port; 276 277 mutex_lock(&listen_port_mutex); 278 279 list_for_each_entry(port_info, &listen_port_list, list) 280 if (port_info->port_num == port) 281 goto found_port; 282 283 port_info = malloc(sizeof(*port_info), M_CXGBE, M_WAITOK); 284 port_info->port_num = port; 285 port_info->refcnt = 0; 286 287 list_add_tail(&port_info->list, &listen_port_list); 288 INIT_LIST_HEAD(&port_info->lep_list); 289 290found_port: 291 port_info->refcnt++; 292 list_add_tail(&lep->listen_ep_list, &port_info->lep_list); 293 mutex_unlock(&listen_port_mutex); 294 return port_info; 295} 296 297static int 298rem_ep_from_listenlist(struct c4iw_listen_ep *lep) 299{ 300 uint16_t port; 301 struct listen_port_info *port_info = NULL; 302 struct sockaddr_storage *laddr = &lep->com.local_addr; 303 int refcnt = 0; 304 305 port = (laddr->ss_family == AF_INET) ? 306 ((struct sockaddr_in *)laddr)->sin_port : 307 ((struct sockaddr_in6 *)laddr)->sin6_port; 308 309 mutex_lock(&listen_port_mutex); 310 311 /* get the port_info structure based on the lep's port address */ 312 list_for_each_entry(port_info, &listen_port_list, list) { 313 if (port_info->port_num == port) { 314 port_info->refcnt--; 315 refcnt = port_info->refcnt; 316 /* remove the current lep from the listen list */ 317 list_del(&lep->listen_ep_list); 318 if (port_info->refcnt == 0) { 319 /* Remove this entry from the list as there 320 * are no more listeners for this port_num. 321 */ 322 list_del(&port_info->list); 323 kfree(port_info); 324 } 325 break; 326 } 327 } 328 mutex_unlock(&listen_port_mutex); 329 return refcnt; 330} 331 332/* 333 * Find the lep that belongs to the ifnet on which the SYN frame was received. 334 */ 335struct c4iw_listen_ep * 336find_real_listen_ep(struct c4iw_listen_ep *master_lep, struct socket *so) 337{ 338 struct adapter *adap = NULL; 339 struct c4iw_listen_ep *lep = NULL; 340 struct sockaddr_storage remote = { 0 }; 341 struct ifnet *new_conn_ifp = NULL; 342 struct listen_port_info *port_info = NULL; 343 int err = 0, i = 0, 344 found_portinfo = 0, found_lep = 0; 345 uint16_t port; 346 347 /* STEP 1: get 'ifnet' based on socket's remote address */ 348 GET_REMOTE_ADDR(&remote, so); 349 350 err = get_ifnet_from_raddr(&remote, &new_conn_ifp); 351 if (err) { 352 CTR4(KTR_IW_CXGBE, "%s: Failed to get ifnet, sock %p, " 353 "master_lep %p err %d", 354 __func__, so, master_lep, err); 355 return (NULL); 356 } 357 358 /* STEP 2: Find 'port_info' with listener local port address. */ 359 port = (master_lep->com.local_addr.ss_family == AF_INET) ? 360 ((struct sockaddr_in *)&master_lep->com.local_addr)->sin_port : 361 ((struct sockaddr_in6 *)&master_lep->com.local_addr)->sin6_port; 362 363 364 mutex_lock(&listen_port_mutex); 365 list_for_each_entry(port_info, &listen_port_list, list) 366 if (port_info->port_num == port) { 367 found_portinfo =1; 368 break; 369 } 370 if (!found_portinfo) 371 goto out; 372 373 /* STEP 3: Traverse through list of lep's that are bound to the current 374 * TCP port address and find the lep that belongs to the ifnet on which 375 * the SYN frame was received. 376 */ 377 list_for_each_entry(lep, &port_info->lep_list, listen_ep_list) { 378 adap = lep->com.dev->rdev.adap; 379 for_each_port(adap, i) { 380 if (new_conn_ifp == adap->port[i]->vi[0].ifp) { 381 found_lep =1; 382 goto out; 383 } 384 } 385 } 386out: 387 mutex_unlock(&listen_port_mutex); 388 return found_lep ? lep : (NULL); 389} 390 391static void process_timeout(struct c4iw_ep *ep) 392{ 393 struct c4iw_qp_attributes attrs = {0}; 394 int abort = 1; 395 396 CTR4(KTR_IW_CXGBE, "%s ep :%p, tid:%u, state %d", __func__, 397 ep, ep->hwtid, ep->com.state); 398 set_bit(TIMEDOUT, &ep->com.history); 399 switch (ep->com.state) { 400 case MPA_REQ_SENT: 401 connect_reply_upcall(ep, -ETIMEDOUT); 402 break; 403 case MPA_REQ_WAIT: 404 case MPA_REQ_RCVD: 405 case MPA_REP_SENT: 406 case FPDU_MODE: 407 break; 408 case CLOSING: 409 case MORIBUND: 410 if (ep->com.cm_id && ep->com.qp) { 411 attrs.next_state = C4IW_QP_STATE_ERROR; 412 c4iw_modify_qp(ep->com.dev, ep->com.qp, 413 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1); 414 } 415 close_complete_upcall(ep, -ETIMEDOUT); 416 break; 417 case ABORTING: 418 case DEAD: 419 /* 420 * These states are expected if the ep timed out at the same 421 * time as another thread was calling stop_ep_timer(). 422 * So we silently do nothing for these states. 423 */ 424 abort = 0; 425 break; 426 default: 427 CTR4(KTR_IW_CXGBE, "%s unexpected state ep %p tid %u state %u\n" 428 , __func__, ep, ep->hwtid, ep->com.state); 429 abort = 0; 430 } 431 if (abort) 432 c4iw_ep_disconnect(ep, 1, GFP_KERNEL); 433 c4iw_put_ep(&ep->com); 434 return; 435} 436 437struct cqe_list_entry { 438 struct list_head entry; 439 struct c4iw_dev *rhp; 440 struct t4_cqe err_cqe; 441}; 442 443static void 444process_err_cqes(void) 445{ 446 unsigned long flag; 447 struct cqe_list_entry *cle; 448 449 spin_lock_irqsave(&err_cqe_lock, flag); 450 while (!list_empty(&err_cqe_list)) { 451 struct list_head *tmp; 452 tmp = err_cqe_list.next; 453 list_del(tmp); 454 tmp->next = tmp->prev = NULL; 455 spin_unlock_irqrestore(&err_cqe_lock, flag); 456 cle = list_entry(tmp, struct cqe_list_entry, entry); 457 c4iw_ev_dispatch(cle->rhp, &cle->err_cqe); 458 free(cle, M_CXGBE); 459 spin_lock_irqsave(&err_cqe_lock, flag); 460 } 461 spin_unlock_irqrestore(&err_cqe_lock, flag); 462 463 return; 464} 465 466static void 467process_req(struct work_struct *ctx) 468{ 469 struct c4iw_ep_common *epc; 470 unsigned long flag; 471 int ep_events; 472 473 process_err_cqes(); 474 spin_lock_irqsave(&req_lock, flag); 475 while (!TAILQ_EMPTY(&req_list)) { 476 epc = TAILQ_FIRST(&req_list); 477 TAILQ_REMOVE(&req_list, epc, entry); 478 epc->entry.tqe_prev = NULL; 479 ep_events = epc->ep_events; 480 epc->ep_events = 0; 481 spin_unlock_irqrestore(&req_lock, flag); 482 mutex_lock(&epc->mutex); 483 CTR5(KTR_IW_CXGBE, "%s: so %p, ep %p, ep_state %s events 0x%x", 484 __func__, epc->so, epc, states[epc->state], ep_events); 485 if (ep_events & C4IW_EVENT_TERM) 486 process_terminate((struct c4iw_ep *)epc); 487 if (ep_events & C4IW_EVENT_TIMEOUT) 488 process_timeout((struct c4iw_ep *)epc); 489 if (ep_events & C4IW_EVENT_SOCKET) 490 process_socket_event((struct c4iw_ep *)epc); 491 mutex_unlock(&epc->mutex); 492 c4iw_put_ep(epc); 493 process_err_cqes(); 494 spin_lock_irqsave(&req_lock, flag); 495 } 496 spin_unlock_irqrestore(&req_lock, flag); 497} 498 499/* 500 * XXX: doesn't belong here in the iWARP driver. 501 * XXX: assumes that the connection was offloaded by cxgbe/t4_tom if TF_TOE is 502 * set. Is this a valid assumption for active open? 503 */ 504static int 505set_tcpinfo(struct c4iw_ep *ep) 506{ 507 struct socket *so = ep->com.so; 508 struct inpcb *inp = sotoinpcb(so); 509 struct tcpcb *tp; 510 struct toepcb *toep; 511 int rc = 0; 512 513 INP_WLOCK(inp); 514 tp = intotcpcb(inp); 515 if ((tp->t_flags & TF_TOE) == 0) { 516 rc = EINVAL; 517 log(LOG_ERR, "%s: connection not offloaded (so %p, ep %p)\n", 518 __func__, so, ep); 519 goto done; 520 } 521 toep = TOEPCB(so); 522 523 ep->hwtid = toep->tid; 524 ep->snd_seq = tp->snd_nxt; 525 ep->rcv_seq = tp->rcv_nxt; 526 ep->emss = max(tp->t_maxseg, 128); 527done: 528 INP_WUNLOCK(inp); 529 return (rc); 530 531} 532static int 533get_ifnet_from_raddr(struct sockaddr_storage *raddr, struct ifnet **ifp) 534{ 535 int err = 0; 536 537 if (raddr->ss_family == AF_INET) { 538 struct sockaddr_in *raddr4 = (struct sockaddr_in *)raddr; 539 struct nhop4_extended nh4 = {0}; 540 541 err = fib4_lookup_nh_ext(RT_DEFAULT_FIB, raddr4->sin_addr, 542 NHR_REF, 0, &nh4); 543 *ifp = nh4.nh_ifp; 544 if (err) 545 fib4_free_nh_ext(RT_DEFAULT_FIB, &nh4); 546 } else { 547 struct sockaddr_in6 *raddr6 = (struct sockaddr_in6 *)raddr; 548 struct nhop6_extended nh6 = {0}; 549 struct in6_addr addr6; 550 uint32_t scopeid; 551 552 memset(&addr6, 0, sizeof(addr6)); 553 in6_splitscope((struct in6_addr *)&raddr6->sin6_addr, 554 &addr6, &scopeid); 555 err = fib6_lookup_nh_ext(RT_DEFAULT_FIB, &addr6, scopeid, 556 NHR_REF, 0, &nh6); 557 *ifp = nh6.nh_ifp; 558 if (err) 559 fib6_free_nh_ext(RT_DEFAULT_FIB, &nh6); 560 } 561 562 CTR2(KTR_IW_CXGBE, "%s: return: %d", __func__, err); 563 return err; 564} 565 566static void 567close_socket(struct socket *so) 568{ 569 uninit_iwarp_socket(so); 570 soclose(so); 571} 572 573static void 574process_peer_close(struct c4iw_ep *ep) 575{ 576 struct c4iw_qp_attributes attrs = {0}; 577 int disconnect = 1; 578 int release = 0; 579 580 CTR4(KTR_IW_CXGBE, "%s:ppcB ep %p so %p state %s", __func__, ep, 581 ep->com.so, states[ep->com.state]); 582 583 switch (ep->com.state) { 584 585 case MPA_REQ_WAIT: 586 CTR2(KTR_IW_CXGBE, "%s:ppc1 %p MPA_REQ_WAIT DEAD", 587 __func__, ep); 588 /* Fallthrough */ 589 case MPA_REQ_SENT: 590 CTR2(KTR_IW_CXGBE, "%s:ppc2 %p MPA_REQ_SENT DEAD", 591 __func__, ep); 592 ep->com.state = DEAD; 593 connect_reply_upcall(ep, -ECONNABORTED); 594 595 disconnect = 0; 596 STOP_EP_TIMER(ep); 597 close_socket(ep->com.so); 598 deref_cm_id(&ep->com); 599 release = 1; 600 break; 601 602 case MPA_REQ_RCVD: 603 604 /* 605 * We're gonna mark this puppy DEAD, but keep 606 * the reference on it until the ULP accepts or 607 * rejects the CR. 608 */ 609 CTR2(KTR_IW_CXGBE, "%s:ppc3 %p MPA_REQ_RCVD CLOSING", 610 __func__, ep); 611 ep->com.state = CLOSING; 612 break; 613 614 case MPA_REP_SENT: 615 CTR2(KTR_IW_CXGBE, "%s:ppc4 %p MPA_REP_SENT CLOSING", 616 __func__, ep); 617 ep->com.state = CLOSING; 618 break; 619 620 case FPDU_MODE: 621 CTR2(KTR_IW_CXGBE, "%s:ppc5 %p FPDU_MODE CLOSING", 622 __func__, ep); 623 START_EP_TIMER(ep); 624 ep->com.state = CLOSING; 625 attrs.next_state = C4IW_QP_STATE_CLOSING; 626 c4iw_modify_qp(ep->com.dev, ep->com.qp, 627 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1); 628 peer_close_upcall(ep); 629 break; 630 631 case ABORTING: 632 CTR2(KTR_IW_CXGBE, "%s:ppc6 %p ABORTING (disconn)", 633 __func__, ep); 634 disconnect = 0; 635 break; 636 637 case CLOSING: 638 CTR2(KTR_IW_CXGBE, "%s:ppc7 %p CLOSING MORIBUND", 639 __func__, ep); 640 ep->com.state = MORIBUND; 641 disconnect = 0; 642 break; 643 644 case MORIBUND: 645 CTR2(KTR_IW_CXGBE, "%s:ppc8 %p MORIBUND DEAD", __func__, 646 ep); 647 STOP_EP_TIMER(ep); 648 if (ep->com.cm_id && ep->com.qp) { 649 attrs.next_state = C4IW_QP_STATE_IDLE; 650 c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp, 651 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1); 652 } 653 close_socket(ep->com.so); 654 close_complete_upcall(ep, 0); 655 ep->com.state = DEAD; 656 release = 1; 657 disconnect = 0; 658 break; 659 660 case DEAD: 661 CTR2(KTR_IW_CXGBE, "%s:ppc9 %p DEAD (disconn)", 662 __func__, ep); 663 disconnect = 0; 664 break; 665 666 default: 667 panic("%s: ep %p state %d", __func__, ep, 668 ep->com.state); 669 break; 670 } 671 672 673 if (disconnect) { 674 675 CTR2(KTR_IW_CXGBE, "%s:ppca %p", __func__, ep); 676 c4iw_ep_disconnect(ep, 0, M_NOWAIT); 677 } 678 if (release) { 679 680 CTR2(KTR_IW_CXGBE, "%s:ppcb %p", __func__, ep); 681 c4iw_put_ep(&ep->com); 682 } 683 CTR2(KTR_IW_CXGBE, "%s:ppcE %p", __func__, ep); 684 return; 685} 686 687static void 688process_conn_error(struct c4iw_ep *ep) 689{ 690 struct c4iw_qp_attributes attrs = {0}; 691 int ret; 692 int state; 693 694 state = ep->com.state; 695 CTR5(KTR_IW_CXGBE, "%s:pceB ep %p so %p so->so_error %u state %s", 696 __func__, ep, ep->com.so, ep->com.so->so_error, 697 states[ep->com.state]); 698 699 switch (state) { 700 701 case MPA_REQ_WAIT: 702 STOP_EP_TIMER(ep); 703 c4iw_put_ep(&ep->parent_ep->com); 704 break; 705 706 case MPA_REQ_SENT: 707 STOP_EP_TIMER(ep); 708 connect_reply_upcall(ep, -ECONNRESET); 709 break; 710 711 case MPA_REP_SENT: 712 ep->com.rpl_err = ECONNRESET; 713 CTR1(KTR_IW_CXGBE, "waking up ep %p", ep); 714 break; 715 716 case MPA_REQ_RCVD: 717 break; 718 719 case MORIBUND: 720 case CLOSING: 721 STOP_EP_TIMER(ep); 722 /*FALLTHROUGH*/ 723 case FPDU_MODE: 724 725 if (ep->com.cm_id && ep->com.qp) { 726 727 attrs.next_state = C4IW_QP_STATE_ERROR; 728 ret = c4iw_modify_qp(ep->com.qp->rhp, 729 ep->com.qp, C4IW_QP_ATTR_NEXT_STATE, 730 &attrs, 1); 731 if (ret) 732 log(LOG_ERR, 733 "%s - qp <- error failed!\n", 734 __func__); 735 } 736 peer_abort_upcall(ep); 737 break; 738 739 case ABORTING: 740 break; 741 742 case DEAD: 743 CTR2(KTR_IW_CXGBE, "%s so_error %d IN DEAD STATE!!!!", 744 __func__, ep->com.so->so_error); 745 return; 746 747 default: 748 panic("%s: ep %p state %d", __func__, ep, state); 749 break; 750 } 751 752 if (state != ABORTING) { 753 close_socket(ep->com.so); 754 ep->com.state = DEAD; 755 c4iw_put_ep(&ep->com); 756 } 757 CTR2(KTR_IW_CXGBE, "%s:pceE %p", __func__, ep); 758 return; 759} 760 761static void 762process_close_complete(struct c4iw_ep *ep) 763{ 764 struct c4iw_qp_attributes attrs = {0}; 765 int release = 0; 766 767 CTR4(KTR_IW_CXGBE, "%s:pccB ep %p so %p state %s", __func__, ep, 768 ep->com.so, states[ep->com.state]); 769 770 /* The cm_id may be null if we failed to connect */ 771 set_bit(CLOSE_CON_RPL, &ep->com.history); 772 773 switch (ep->com.state) { 774 775 case CLOSING: 776 CTR2(KTR_IW_CXGBE, "%s:pcc1 %p CLOSING MORIBUND", 777 __func__, ep); 778 ep->com.state = MORIBUND; 779 break; 780 781 case MORIBUND: 782 CTR2(KTR_IW_CXGBE, "%s:pcc1 %p MORIBUND DEAD", __func__, 783 ep); 784 STOP_EP_TIMER(ep); 785 786 if ((ep->com.cm_id) && (ep->com.qp)) { 787 788 CTR2(KTR_IW_CXGBE, "%s:pcc2 %p QP_STATE_IDLE", 789 __func__, ep); 790 attrs.next_state = C4IW_QP_STATE_IDLE; 791 c4iw_modify_qp(ep->com.dev, 792 ep->com.qp, 793 C4IW_QP_ATTR_NEXT_STATE, 794 &attrs, 1); 795 } 796 797 close_socket(ep->com.so); 798 close_complete_upcall(ep, 0); 799 ep->com.state = DEAD; 800 release = 1; 801 break; 802 803 case ABORTING: 804 CTR2(KTR_IW_CXGBE, "%s:pcc5 %p ABORTING", __func__, ep); 805 break; 806 807 case DEAD: 808 CTR2(KTR_IW_CXGBE, "%s:pcc6 %p DEAD", __func__, ep); 809 break; 810 default: 811 CTR2(KTR_IW_CXGBE, "%s:pcc7 %p unknown ep state", 812 __func__, ep); 813 panic("%s:pcc6 %p unknown ep state", __func__, ep); 814 break; 815 } 816 817 if (release) { 818 819 CTR2(KTR_IW_CXGBE, "%s:pcc8 %p", __func__, ep); 820 release_ep_resources(ep); 821 } 822 CTR2(KTR_IW_CXGBE, "%s:pccE %p", __func__, ep); 823 return; 824} 825 826static void 827setiwsockopt(struct socket *so) 828{ 829 int rc; 830 struct sockopt sopt; 831 int on = 1; 832 833 sopt.sopt_dir = SOPT_SET; 834 sopt.sopt_level = IPPROTO_TCP; 835 sopt.sopt_name = TCP_NODELAY; 836 sopt.sopt_val = (caddr_t)&on; 837 sopt.sopt_valsize = sizeof on; 838 sopt.sopt_td = NULL; 839 rc = sosetopt(so, &sopt); 840 if (rc) { 841 log(LOG_ERR, "%s: can't set TCP_NODELAY on so %p (%d)\n", 842 __func__, so, rc); 843 } 844} 845 846static void 847init_iwarp_socket(struct socket *so, void *arg) 848{ 849 850 SOCKBUF_LOCK(&so->so_rcv); 851 soupcall_set(so, SO_RCV, c4iw_so_upcall, arg); 852 so->so_state |= SS_NBIO; 853 SOCKBUF_UNLOCK(&so->so_rcv); 854} 855 856static void 857uninit_iwarp_socket(struct socket *so) 858{ 859 860 SOCKBUF_LOCK(&so->so_rcv); 861 soupcall_clear(so, SO_RCV); 862 SOCKBUF_UNLOCK(&so->so_rcv); 863} 864 865static void 866process_data(struct c4iw_ep *ep) 867{ 868 int disconnect = 0; 869 870 CTR5(KTR_IW_CXGBE, "%s: so %p, ep %p, state %s, sbused %d", __func__, 871 ep->com.so, ep, states[ep->com.state], sbused(&ep->com.so->so_rcv)); 872 873 switch (ep->com.state) { 874 case MPA_REQ_SENT: 875 disconnect = process_mpa_reply(ep); 876 break; 877 case MPA_REQ_WAIT: 878 disconnect = process_mpa_request(ep); 879 if (disconnect) 880 /* Refered in process_newconn() */ 881 c4iw_put_ep(&ep->parent_ep->com); 882 break; 883 default: 884 if (sbused(&ep->com.so->so_rcv)) 885 log(LOG_ERR, "%s: Unexpected streaming data. ep %p, " 886 "state %d, so %p, so_state 0x%x, sbused %u\n", 887 __func__, ep, ep->com.state, ep->com.so, 888 ep->com.so->so_state, sbused(&ep->com.so->so_rcv)); 889 break; 890 } 891 if (disconnect) 892 c4iw_ep_disconnect(ep, disconnect == 2, GFP_KERNEL); 893 894} 895 896static void 897process_connected(struct c4iw_ep *ep) 898{ 899 struct socket *so = ep->com.so; 900 901 if ((so->so_state & SS_ISCONNECTED) && !so->so_error) { 902 if (send_mpa_req(ep)) 903 goto err; 904 } else { 905 connect_reply_upcall(ep, -so->so_error); 906 goto err; 907 } 908 return; 909err: 910 close_socket(so); 911 ep->com.state = DEAD; 912 c4iw_put_ep(&ep->com); 913 return; 914} 915 916static inline int c4iw_zero_addr(struct sockaddr *addr) 917{ 918 struct in6_addr *ip6; 919 920 if (addr->sa_family == AF_INET) 921 return IN_ZERONET( 922 ntohl(((struct sockaddr_in *)addr)->sin_addr.s_addr)); 923 else { 924 ip6 = &((struct sockaddr_in6 *) addr)->sin6_addr; 925 return (ip6->s6_addr32[0] | ip6->s6_addr32[1] | 926 ip6->s6_addr32[2] | ip6->s6_addr32[3]) == 0; 927 } 928} 929 930static inline int c4iw_loopback_addr(struct sockaddr *addr) 931{ 932 if (addr->sa_family == AF_INET) 933 return IN_LOOPBACK( 934 ntohl(((struct sockaddr_in *) addr)->sin_addr.s_addr)); 935 else 936 return IN6_IS_ADDR_LOOPBACK( 937 &((struct sockaddr_in6 *) addr)->sin6_addr); 938} 939 940static inline int c4iw_any_addr(struct sockaddr *addr) 941{ 942 return c4iw_zero_addr(addr) || c4iw_loopback_addr(addr); 943} 944 945static void 946process_newconn(struct c4iw_listen_ep *master_lep, struct socket *new_so) 947{ 948 struct c4iw_listen_ep *real_lep = NULL; 949 struct c4iw_ep *new_ep = NULL; 950 int ret = 0; 951 952 MPASS(new_so != NULL); 953 954 if (c4iw_any_addr((struct sockaddr *)&master_lep->com.local_addr)) { 955 /* Here we need to find the 'real_lep' that belongs to the 956 * incomming socket's network interface, such that the newly 957 * created 'ep' can be attached to the real 'lep'. 958 */ 959 real_lep = find_real_listen_ep(master_lep, new_so); 960 if (real_lep == NULL) { 961 CTR2(KTR_IW_CXGBE, "%s: Could not find the real listen " 962 "ep for sock: %p", __func__, new_so); 963 log(LOG_ERR,"%s: Could not find the real listen ep for " 964 "sock: %p\n", __func__, new_so); 965 /* FIXME: properly free the 'new_so' in failure case. 966 * Use of soabort() and soclose() are not legal 967 * here(before soaccept()). 968 */ 969 return; 970 } 971 } else /* for Non-Wildcard address, master_lep is always the real_lep */ 972 real_lep = master_lep; 973 974 new_ep = alloc_ep(sizeof(*new_ep), GFP_KERNEL); 975 976 CTR6(KTR_IW_CXGBE, "%s: master_lep %p, real_lep: %p, new ep %p, " 977 "listening so %p, new so %p", __func__, master_lep, real_lep, 978 new_ep, master_lep->com.so, new_so); 979 980 new_ep->com.dev = real_lep->com.dev; 981 new_ep->com.so = new_so; 982 new_ep->com.cm_id = NULL; 983 new_ep->com.thread = real_lep->com.thread; 984 new_ep->parent_ep = real_lep; 985 986 GET_LOCAL_ADDR(&new_ep->com.local_addr, new_so); 987 GET_REMOTE_ADDR(&new_ep->com.remote_addr, new_so); 988 c4iw_get_ep(&real_lep->com); 989 init_timer(&new_ep->timer); 990 new_ep->com.state = MPA_REQ_WAIT; 991 START_EP_TIMER(new_ep); 992 993 setiwsockopt(new_so); 994 995 /* MPA request might have been queued up on the socket already, so we 996 * initialize the socket/upcall_handler under lock to prevent processing 997 * MPA request on another thread(via process_req()) simultaniously. 998 */ 999 c4iw_get_ep(&new_ep->com); /* Dereferenced at the end below, this is to 1000 avoid freeing of ep before ep unlock. */ 1001 mutex_lock(&new_ep->com.mutex); 1002 init_iwarp_socket(new_so, &new_ep->com); 1003 1004 ret = process_mpa_request(new_ep); 1005 if (ret) { 1006 /* ABORT */ 1007 c4iw_ep_disconnect(new_ep, 1, GFP_KERNEL); 1008 c4iw_put_ep(&real_lep->com); 1009 } 1010 mutex_unlock(&new_ep->com.mutex); 1011 c4iw_put_ep(&new_ep->com); 1012 return; 1013} 1014 1015static int 1016add_ep_to_req_list(struct c4iw_ep *ep, int new_ep_event) 1017{ 1018 unsigned long flag; 1019 1020 spin_lock_irqsave(&req_lock, flag); 1021 if (ep && ep->com.so) { 1022 ep->com.ep_events |= new_ep_event; 1023 if (!ep->com.entry.tqe_prev) { 1024 c4iw_get_ep(&ep->com); 1025 TAILQ_INSERT_TAIL(&req_list, &ep->com, entry); 1026 queue_work(c4iw_taskq, &c4iw_task); 1027 } 1028 } 1029 spin_unlock_irqrestore(&req_lock, flag); 1030 1031 return (0); 1032} 1033 1034static int 1035c4iw_so_upcall(struct socket *so, void *arg, int waitflag) 1036{ 1037 struct c4iw_ep *ep = arg; 1038 1039 CTR6(KTR_IW_CXGBE, 1040 "%s: so %p, so_state 0x%x, ep %p, ep_state %s, tqe_prev %p", 1041 __func__, so, so->so_state, ep, states[ep->com.state], 1042 ep->com.entry.tqe_prev); 1043 1044 MPASS(ep->com.so == so); 1045 /* 1046 * Wake up any threads waiting in rdma_init()/rdma_fini(), 1047 * with locks held. 1048 */ 1049 if (so->so_error) 1050 c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET); 1051 add_ep_to_req_list(ep, C4IW_EVENT_SOCKET); 1052 1053 return (SU_OK); 1054} 1055 1056 1057static int 1058terminate(struct sge_iq *iq, const struct rss_header *rss, struct mbuf *m) 1059{ 1060 struct adapter *sc = iq->adapter; 1061 const struct cpl_rdma_terminate *cpl = mtod(m, const void *); 1062 unsigned int tid = GET_TID(cpl); 1063 struct toepcb *toep = lookup_tid(sc, tid); 1064 struct socket *so; 1065 struct c4iw_ep *ep; 1066 1067 INP_WLOCK(toep->inp); 1068 so = inp_inpcbtosocket(toep->inp); 1069 ep = so->so_rcv.sb_upcallarg; 1070 INP_WUNLOCK(toep->inp); 1071 1072 CTR3(KTR_IW_CXGBE, "%s: so %p, ep %p", __func__, so, ep); 1073 add_ep_to_req_list(ep, C4IW_EVENT_TERM); 1074 1075 return 0; 1076} 1077 1078static struct socket * 1079dequeue_socket(struct socket *head) 1080{ 1081 struct socket *so; 1082 struct sockaddr_in *remote; 1083 1084 ACCEPT_LOCK(); 1085 so = TAILQ_FIRST(&head->so_comp); 1086 if (!so) { 1087 ACCEPT_UNLOCK(); 1088 return NULL; 1089 } 1090 1091 SOCK_LOCK(so); 1092 /* 1093 * Before changing the flags on the socket, we have to bump the 1094 * reference count. Otherwise, if the protocol calls sofree(), 1095 * the socket will be released due to a zero refcount. 1096 */ 1097 soref(so); 1098 TAILQ_REMOVE(&head->so_comp, so, so_list); 1099 head->so_qlen--; 1100 so->so_qstate &= ~SQ_COMP; 1101 so->so_head = NULL; 1102 so->so_state |= SS_NBIO; 1103 SOCK_UNLOCK(so); 1104 ACCEPT_UNLOCK(); 1105 remote = NULL; 1106 if (soaccept(so, (struct sockaddr **)&remote) != 0) { 1107 soclose(so); 1108 so = NULL; 1109 } 1110 free(remote, M_SONAME); 1111 return so; 1112} 1113 1114static void 1115process_socket_event(struct c4iw_ep *ep) 1116{ 1117 int state = ep->com.state; 1118 struct socket *so = ep->com.so; 1119 1120 if (ep->com.state == DEAD) { 1121 CTR3(KTR_IW_CXGBE, "%s: Pending socket event discarded " 1122 "ep %p ep_state %s", __func__, ep, states[state]); 1123 return; 1124 } 1125 1126 CTR6(KTR_IW_CXGBE, "process_socket_event: so %p, so_state 0x%x, " 1127 "so_err %d, sb_state 0x%x, ep %p, ep_state %s", so, so->so_state, 1128 so->so_error, so->so_rcv.sb_state, ep, states[state]); 1129 1130 if (state == CONNECTING) { 1131 process_connected(ep); 1132 return; 1133 } 1134 1135 if (state == LISTEN) { 1136 struct c4iw_listen_ep *lep = (struct c4iw_listen_ep *)ep; 1137 struct socket *new_so; 1138 1139 while ((new_so = dequeue_socket(so)) != NULL) { 1140 process_newconn(lep, new_so); 1141 } 1142 return; 1143 } 1144 1145 /* connection error */ 1146 if (so->so_error) { 1147 process_conn_error(ep); 1148 return; 1149 } 1150 1151 /* peer close */ 1152 if ((so->so_rcv.sb_state & SBS_CANTRCVMORE) && state <= CLOSING) { 1153 process_peer_close(ep); 1154 /* 1155 * check whether socket disconnect event is pending before 1156 * returning. Fallthrough if yes. 1157 */ 1158 if (!(so->so_state & SS_ISDISCONNECTED)) 1159 return; 1160 } 1161 1162 /* close complete */ 1163 if (so->so_state & SS_ISDISCONNECTED) { 1164 process_close_complete(ep); 1165 return; 1166 } 1167 1168 /* rx data */ 1169 process_data(ep); 1170} 1171 1172SYSCTL_NODE(_hw, OID_AUTO, iw_cxgbe, CTLFLAG_RD, 0, "iw_cxgbe driver parameters"); 1173 1174static int dack_mode = 0; 1175SYSCTL_INT(_hw_iw_cxgbe, OID_AUTO, dack_mode, CTLFLAG_RWTUN, &dack_mode, 0, 1176 "Delayed ack mode (default = 0)"); 1177 1178int c4iw_max_read_depth = 8; 1179SYSCTL_INT(_hw_iw_cxgbe, OID_AUTO, c4iw_max_read_depth, CTLFLAG_RWTUN, &c4iw_max_read_depth, 0, 1180 "Per-connection max ORD/IRD (default = 8)"); 1181 1182static int enable_tcp_timestamps; 1183SYSCTL_INT(_hw_iw_cxgbe, OID_AUTO, enable_tcp_timestamps, CTLFLAG_RWTUN, &enable_tcp_timestamps, 0, 1184 "Enable tcp timestamps (default = 0)"); 1185 1186static int enable_tcp_sack; 1187SYSCTL_INT(_hw_iw_cxgbe, OID_AUTO, enable_tcp_sack, CTLFLAG_RWTUN, &enable_tcp_sack, 0, 1188 "Enable tcp SACK (default = 0)"); 1189 1190static int enable_tcp_window_scaling = 1; 1191SYSCTL_INT(_hw_iw_cxgbe, OID_AUTO, enable_tcp_window_scaling, CTLFLAG_RWTUN, &enable_tcp_window_scaling, 0, 1192 "Enable tcp window scaling (default = 1)"); 1193 1194int c4iw_debug = 1; 1195SYSCTL_INT(_hw_iw_cxgbe, OID_AUTO, c4iw_debug, CTLFLAG_RWTUN, &c4iw_debug, 0, 1196 "Enable debug logging (default = 0)"); 1197 1198static int peer2peer = 1; 1199SYSCTL_INT(_hw_iw_cxgbe, OID_AUTO, peer2peer, CTLFLAG_RWTUN, &peer2peer, 0, 1200 "Support peer2peer ULPs (default = 1)"); 1201 1202static int p2p_type = FW_RI_INIT_P2PTYPE_READ_REQ; 1203SYSCTL_INT(_hw_iw_cxgbe, OID_AUTO, p2p_type, CTLFLAG_RWTUN, &p2p_type, 0, 1204 "RDMAP opcode to use for the RTR message: 1 = RDMA_READ 0 = RDMA_WRITE (default 1)"); 1205 1206static int ep_timeout_secs = 60; 1207SYSCTL_INT(_hw_iw_cxgbe, OID_AUTO, ep_timeout_secs, CTLFLAG_RWTUN, &ep_timeout_secs, 0, 1208 "CM Endpoint operation timeout in seconds (default = 60)"); 1209 1210static int mpa_rev = 1; 1211SYSCTL_INT(_hw_iw_cxgbe, OID_AUTO, mpa_rev, CTLFLAG_RWTUN, &mpa_rev, 0, 1212 "MPA Revision, 0 supports amso1100, 1 is RFC5044 spec compliant, 2 is IETF MPA Peer Connect Draft compliant (default = 1)"); 1213 1214static int markers_enabled; 1215SYSCTL_INT(_hw_iw_cxgbe, OID_AUTO, markers_enabled, CTLFLAG_RWTUN, &markers_enabled, 0, 1216 "Enable MPA MARKERS (default(0) = disabled)"); 1217 1218static int crc_enabled = 1; 1219SYSCTL_INT(_hw_iw_cxgbe, OID_AUTO, crc_enabled, CTLFLAG_RWTUN, &crc_enabled, 0, 1220 "Enable MPA CRC (default(1) = enabled)"); 1221 1222static int rcv_win = 256 * 1024; 1223SYSCTL_INT(_hw_iw_cxgbe, OID_AUTO, rcv_win, CTLFLAG_RWTUN, &rcv_win, 0, 1224 "TCP receive window in bytes (default = 256KB)"); 1225 1226static int snd_win = 128 * 1024; 1227SYSCTL_INT(_hw_iw_cxgbe, OID_AUTO, snd_win, CTLFLAG_RWTUN, &snd_win, 0, 1228 "TCP send window in bytes (default = 128KB)"); 1229 1230static void 1231start_ep_timer(struct c4iw_ep *ep) 1232{ 1233 1234 if (timer_pending(&ep->timer)) { 1235 CTR2(KTR_IW_CXGBE, "%s: ep %p, already started", __func__, ep); 1236 printk(KERN_ERR "%s timer already started! ep %p\n", __func__, 1237 ep); 1238 return; 1239 } 1240 clear_bit(TIMEOUT, &ep->com.flags); 1241 c4iw_get_ep(&ep->com); 1242 ep->timer.expires = jiffies + ep_timeout_secs * HZ; 1243 ep->timer.data = (unsigned long)ep; 1244 ep->timer.function = ep_timeout; 1245 add_timer(&ep->timer); 1246} 1247 1248static int 1249stop_ep_timer(struct c4iw_ep *ep) 1250{ 1251 1252 del_timer_sync(&ep->timer); 1253 if (!test_and_set_bit(TIMEOUT, &ep->com.flags)) { 1254 c4iw_put_ep(&ep->com); 1255 return 0; 1256 } 1257 return 1; 1258} 1259 1260static void * 1261alloc_ep(int size, gfp_t gfp) 1262{ 1263 struct c4iw_ep_common *epc; 1264 1265 epc = kzalloc(size, gfp); 1266 if (epc == NULL) 1267 return (NULL); 1268 1269 kref_init(&epc->kref); 1270 mutex_init(&epc->mutex); 1271 c4iw_init_wr_wait(&epc->wr_wait); 1272 1273 return (epc); 1274} 1275 1276void _c4iw_free_ep(struct kref *kref) 1277{ 1278 struct c4iw_ep *ep; 1279 struct c4iw_ep_common *epc; 1280 1281 ep = container_of(kref, struct c4iw_ep, com.kref); 1282 epc = &ep->com; 1283 KASSERT(!epc->entry.tqe_prev, ("%s epc %p still on req list", 1284 __func__, epc)); 1285 if (test_bit(QP_REFERENCED, &ep->com.flags)) 1286 deref_qp(ep); 1287 CTR4(KTR_IW_CXGBE, "%s: ep %p, history 0x%lx, flags 0x%lx", 1288 __func__, ep, epc->history, epc->flags); 1289 kfree(ep); 1290} 1291 1292static void release_ep_resources(struct c4iw_ep *ep) 1293{ 1294 CTR2(KTR_IW_CXGBE, "%s:rerB %p", __func__, ep); 1295 set_bit(RELEASE_RESOURCES, &ep->com.flags); 1296 c4iw_put_ep(&ep->com); 1297 CTR2(KTR_IW_CXGBE, "%s:rerE %p", __func__, ep); 1298} 1299 1300static int 1301send_mpa_req(struct c4iw_ep *ep) 1302{ 1303 int mpalen; 1304 struct mpa_message *mpa; 1305 struct mpa_v2_conn_params mpa_v2_params; 1306 struct mbuf *m; 1307 char mpa_rev_to_use = mpa_rev; 1308 int err = 0; 1309 1310 if (ep->retry_with_mpa_v1) 1311 mpa_rev_to_use = 1; 1312 mpalen = sizeof(*mpa) + ep->plen; 1313 if (mpa_rev_to_use == 2) 1314 mpalen += sizeof(struct mpa_v2_conn_params); 1315 1316 mpa = malloc(mpalen, M_CXGBE, M_NOWAIT); 1317 if (mpa == NULL) { 1318 err = -ENOMEM; 1319 CTR3(KTR_IW_CXGBE, "%s:smr1 ep: %p , error: %d", 1320 __func__, ep, err); 1321 goto err; 1322 } 1323 1324 memset(mpa, 0, mpalen); 1325 memcpy(mpa->key, MPA_KEY_REQ, sizeof(mpa->key)); 1326 mpa->flags = (crc_enabled ? MPA_CRC : 0) | 1327 (markers_enabled ? MPA_MARKERS : 0) | 1328 (mpa_rev_to_use == 2 ? MPA_ENHANCED_RDMA_CONN : 0); 1329 mpa->private_data_size = htons(ep->plen); 1330 mpa->revision = mpa_rev_to_use; 1331 1332 if (mpa_rev_to_use == 1) { 1333 ep->tried_with_mpa_v1 = 1; 1334 ep->retry_with_mpa_v1 = 0; 1335 } 1336 1337 if (mpa_rev_to_use == 2) { 1338 mpa->private_data_size = htons(ntohs(mpa->private_data_size) + 1339 sizeof(struct mpa_v2_conn_params)); 1340 mpa_v2_params.ird = htons((u16)ep->ird); 1341 mpa_v2_params.ord = htons((u16)ep->ord); 1342 1343 if (peer2peer) { 1344 mpa_v2_params.ird |= htons(MPA_V2_PEER2PEER_MODEL); 1345 1346 if (p2p_type == FW_RI_INIT_P2PTYPE_RDMA_WRITE) { 1347 mpa_v2_params.ord |= 1348 htons(MPA_V2_RDMA_WRITE_RTR); 1349 } else if (p2p_type == FW_RI_INIT_P2PTYPE_READ_REQ) { 1350 mpa_v2_params.ord |= 1351 htons(MPA_V2_RDMA_READ_RTR); 1352 } 1353 } 1354 memcpy(mpa->private_data, &mpa_v2_params, 1355 sizeof(struct mpa_v2_conn_params)); 1356 1357 if (ep->plen) { 1358 1359 memcpy(mpa->private_data + 1360 sizeof(struct mpa_v2_conn_params), 1361 ep->mpa_pkt + sizeof(*mpa), ep->plen); 1362 } 1363 } else { 1364 1365 if (ep->plen) 1366 memcpy(mpa->private_data, 1367 ep->mpa_pkt + sizeof(*mpa), ep->plen); 1368 CTR2(KTR_IW_CXGBE, "%s:smr7 %p", __func__, ep); 1369 } 1370 1371 m = m_getm(NULL, mpalen, M_NOWAIT, MT_DATA); 1372 if (m == NULL) { 1373 err = -ENOMEM; 1374 CTR3(KTR_IW_CXGBE, "%s:smr2 ep: %p , error: %d", 1375 __func__, ep, err); 1376 free(mpa, M_CXGBE); 1377 goto err; 1378 } 1379 m_copyback(m, 0, mpalen, (void *)mpa); 1380 free(mpa, M_CXGBE); 1381 1382 err = -sosend(ep->com.so, NULL, NULL, m, NULL, MSG_DONTWAIT, 1383 ep->com.thread); 1384 if (err) { 1385 CTR3(KTR_IW_CXGBE, "%s:smr3 ep: %p , error: %d", 1386 __func__, ep, err); 1387 goto err; 1388 } 1389 1390 START_EP_TIMER(ep); 1391 ep->com.state = MPA_REQ_SENT; 1392 ep->mpa_attr.initiator = 1; 1393 CTR3(KTR_IW_CXGBE, "%s:smrE %p, error: %d", __func__, ep, err); 1394 return 0; 1395err: 1396 connect_reply_upcall(ep, err); 1397 CTR3(KTR_IW_CXGBE, "%s:smrE %p, error: %d", __func__, ep, err); 1398 return err; 1399} 1400 1401static int send_mpa_reject(struct c4iw_ep *ep, const void *pdata, u8 plen) 1402{ 1403 int mpalen ; 1404 struct mpa_message *mpa; 1405 struct mpa_v2_conn_params mpa_v2_params; 1406 struct mbuf *m; 1407 int err; 1408 1409 CTR4(KTR_IW_CXGBE, "%s:smrejB %p %u %d", __func__, ep, ep->hwtid, 1410 ep->plen); 1411 1412 mpalen = sizeof(*mpa) + plen; 1413 1414 if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) { 1415 1416 mpalen += sizeof(struct mpa_v2_conn_params); 1417 CTR4(KTR_IW_CXGBE, "%s:smrej1 %p %u %d", __func__, ep, 1418 ep->mpa_attr.version, mpalen); 1419 } 1420 1421 mpa = malloc(mpalen, M_CXGBE, M_NOWAIT); 1422 if (mpa == NULL) 1423 return (-ENOMEM); 1424 1425 memset(mpa, 0, mpalen); 1426 memcpy(mpa->key, MPA_KEY_REP, sizeof(mpa->key)); 1427 mpa->flags = MPA_REJECT; 1428 mpa->revision = mpa_rev; 1429 mpa->private_data_size = htons(plen); 1430 1431 if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) { 1432 1433 mpa->flags |= MPA_ENHANCED_RDMA_CONN; 1434 mpa->private_data_size = htons(ntohs(mpa->private_data_size) + 1435 sizeof(struct mpa_v2_conn_params)); 1436 mpa_v2_params.ird = htons(((u16)ep->ird) | 1437 (peer2peer ? MPA_V2_PEER2PEER_MODEL : 1438 0)); 1439 mpa_v2_params.ord = htons(((u16)ep->ord) | (peer2peer ? 1440 (p2p_type == 1441 FW_RI_INIT_P2PTYPE_RDMA_WRITE ? 1442 MPA_V2_RDMA_WRITE_RTR : p2p_type == 1443 FW_RI_INIT_P2PTYPE_READ_REQ ? 1444 MPA_V2_RDMA_READ_RTR : 0) : 0)); 1445 memcpy(mpa->private_data, &mpa_v2_params, 1446 sizeof(struct mpa_v2_conn_params)); 1447 1448 if (ep->plen) 1449 memcpy(mpa->private_data + 1450 sizeof(struct mpa_v2_conn_params), pdata, plen); 1451 CTR5(KTR_IW_CXGBE, "%s:smrej3 %p %d %d %d", __func__, ep, 1452 mpa_v2_params.ird, mpa_v2_params.ord, ep->plen); 1453 } else 1454 if (plen) 1455 memcpy(mpa->private_data, pdata, plen); 1456 1457 m = m_getm(NULL, mpalen, M_NOWAIT, MT_DATA); 1458 if (m == NULL) { 1459 free(mpa, M_CXGBE); 1460 return (-ENOMEM); 1461 } 1462 m_copyback(m, 0, mpalen, (void *)mpa); 1463 free(mpa, M_CXGBE); 1464 1465 err = -sosend(ep->com.so, NULL, NULL, m, NULL, MSG_DONTWAIT, ep->com.thread); 1466 if (!err) 1467 ep->snd_seq += mpalen; 1468 CTR4(KTR_IW_CXGBE, "%s:smrejE %p %u %d", __func__, ep, ep->hwtid, err); 1469 return err; 1470} 1471 1472static int send_mpa_reply(struct c4iw_ep *ep, const void *pdata, u8 plen) 1473{ 1474 int mpalen; 1475 struct mpa_message *mpa; 1476 struct mbuf *m; 1477 struct mpa_v2_conn_params mpa_v2_params; 1478 int err; 1479 1480 CTR2(KTR_IW_CXGBE, "%s:smrepB %p", __func__, ep); 1481 1482 mpalen = sizeof(*mpa) + plen; 1483 1484 if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) { 1485 1486 CTR3(KTR_IW_CXGBE, "%s:smrep1 %p %d", __func__, ep, 1487 ep->mpa_attr.version); 1488 mpalen += sizeof(struct mpa_v2_conn_params); 1489 } 1490 1491 mpa = malloc(mpalen, M_CXGBE, M_NOWAIT); 1492 if (mpa == NULL) 1493 return (-ENOMEM); 1494 1495 memset(mpa, 0, sizeof(*mpa)); 1496 memcpy(mpa->key, MPA_KEY_REP, sizeof(mpa->key)); 1497 mpa->flags = (ep->mpa_attr.crc_enabled ? MPA_CRC : 0) | 1498 (markers_enabled ? MPA_MARKERS : 0); 1499 mpa->revision = ep->mpa_attr.version; 1500 mpa->private_data_size = htons(plen); 1501 1502 if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) { 1503 1504 mpa->flags |= MPA_ENHANCED_RDMA_CONN; 1505 mpa->private_data_size += 1506 htons(sizeof(struct mpa_v2_conn_params)); 1507 mpa_v2_params.ird = htons((u16)ep->ird); 1508 mpa_v2_params.ord = htons((u16)ep->ord); 1509 CTR5(KTR_IW_CXGBE, "%s:smrep3 %p %d %d %d", __func__, ep, 1510 ep->mpa_attr.version, mpa_v2_params.ird, mpa_v2_params.ord); 1511 1512 if (peer2peer && (ep->mpa_attr.p2p_type != 1513 FW_RI_INIT_P2PTYPE_DISABLED)) { 1514 1515 mpa_v2_params.ird |= htons(MPA_V2_PEER2PEER_MODEL); 1516 1517 if (p2p_type == FW_RI_INIT_P2PTYPE_RDMA_WRITE) { 1518 1519 mpa_v2_params.ord |= 1520 htons(MPA_V2_RDMA_WRITE_RTR); 1521 CTR5(KTR_IW_CXGBE, "%s:smrep4 %p %d %d %d", 1522 __func__, ep, p2p_type, mpa_v2_params.ird, 1523 mpa_v2_params.ord); 1524 } 1525 else if (p2p_type == FW_RI_INIT_P2PTYPE_READ_REQ) { 1526 1527 mpa_v2_params.ord |= 1528 htons(MPA_V2_RDMA_READ_RTR); 1529 CTR5(KTR_IW_CXGBE, "%s:smrep5 %p %d %d %d", 1530 __func__, ep, p2p_type, mpa_v2_params.ird, 1531 mpa_v2_params.ord); 1532 } 1533 } 1534 1535 memcpy(mpa->private_data, &mpa_v2_params, 1536 sizeof(struct mpa_v2_conn_params)); 1537 1538 if (ep->plen) 1539 memcpy(mpa->private_data + 1540 sizeof(struct mpa_v2_conn_params), pdata, plen); 1541 } else 1542 if (plen) 1543 memcpy(mpa->private_data, pdata, plen); 1544 1545 m = m_getm(NULL, mpalen, M_NOWAIT, MT_DATA); 1546 if (m == NULL) { 1547 free(mpa, M_CXGBE); 1548 return (-ENOMEM); 1549 } 1550 m_copyback(m, 0, mpalen, (void *)mpa); 1551 free(mpa, M_CXGBE); 1552 1553 1554 ep->com.state = MPA_REP_SENT; 1555 ep->snd_seq += mpalen; 1556 err = -sosend(ep->com.so, NULL, NULL, m, NULL, MSG_DONTWAIT, 1557 ep->com.thread); 1558 CTR3(KTR_IW_CXGBE, "%s:smrepE %p %d", __func__, ep, err); 1559 return err; 1560} 1561 1562 1563 1564static void close_complete_upcall(struct c4iw_ep *ep, int status) 1565{ 1566 struct iw_cm_event event; 1567 1568 CTR2(KTR_IW_CXGBE, "%s:ccuB %p", __func__, ep); 1569 memset(&event, 0, sizeof(event)); 1570 event.event = IW_CM_EVENT_CLOSE; 1571 event.status = status; 1572 1573 if (ep->com.cm_id) { 1574 1575 CTR2(KTR_IW_CXGBE, "%s:ccu1 %1", __func__, ep); 1576 ep->com.cm_id->event_handler(ep->com.cm_id, &event); 1577 deref_cm_id(&ep->com); 1578 set_bit(CLOSE_UPCALL, &ep->com.history); 1579 } 1580 CTR2(KTR_IW_CXGBE, "%s:ccuE %p", __func__, ep); 1581} 1582 1583static int 1584send_abort(struct c4iw_ep *ep) 1585{ 1586 struct socket *so = ep->com.so; 1587 struct sockopt sopt; 1588 int rc; 1589 struct linger l; 1590 1591 CTR5(KTR_IW_CXGBE, "%s ep %p so %p state %s tid %d", __func__, ep, so, 1592 states[ep->com.state], ep->hwtid); 1593 1594 l.l_onoff = 1; 1595 l.l_linger = 0; 1596 1597 /* linger_time of 0 forces RST to be sent */ 1598 sopt.sopt_dir = SOPT_SET; 1599 sopt.sopt_level = SOL_SOCKET; 1600 sopt.sopt_name = SO_LINGER; 1601 sopt.sopt_val = (caddr_t)&l; 1602 sopt.sopt_valsize = sizeof l; 1603 sopt.sopt_td = NULL; 1604 rc = sosetopt(so, &sopt); 1605 if (rc != 0) { 1606 log(LOG_ERR, "%s: sosetopt(%p, linger = 0) failed with %d.\n", 1607 __func__, so, rc); 1608 } 1609 1610 uninit_iwarp_socket(so); 1611 soclose(so); 1612 set_bit(ABORT_CONN, &ep->com.history); 1613 1614 /* 1615 * TBD: iw_cxgbe driver should receive ABORT reply for every ABORT 1616 * request it has sent. But the current TOE driver is not propagating 1617 * this ABORT reply event (via do_abort_rpl) to iw_cxgbe. So as a work- 1618 * around de-refererece 'ep' here instead of doing it in abort_rpl() 1619 * handler(not yet implemented) of iw_cxgbe driver. 1620 */ 1621 release_ep_resources(ep); 1622 1623 return (0); 1624} 1625 1626static void peer_close_upcall(struct c4iw_ep *ep) 1627{ 1628 struct iw_cm_event event; 1629 1630 CTR2(KTR_IW_CXGBE, "%s:pcuB %p", __func__, ep); 1631 memset(&event, 0, sizeof(event)); 1632 event.event = IW_CM_EVENT_DISCONNECT; 1633 1634 if (ep->com.cm_id) { 1635 1636 CTR2(KTR_IW_CXGBE, "%s:pcu1 %p", __func__, ep); 1637 ep->com.cm_id->event_handler(ep->com.cm_id, &event); 1638 set_bit(DISCONN_UPCALL, &ep->com.history); 1639 } 1640 CTR2(KTR_IW_CXGBE, "%s:pcuE %p", __func__, ep); 1641} 1642 1643static void peer_abort_upcall(struct c4iw_ep *ep) 1644{ 1645 struct iw_cm_event event; 1646 1647 CTR2(KTR_IW_CXGBE, "%s:pauB %p", __func__, ep); 1648 memset(&event, 0, sizeof(event)); 1649 event.event = IW_CM_EVENT_CLOSE; 1650 event.status = -ECONNRESET; 1651 1652 if (ep->com.cm_id) { 1653 1654 CTR2(KTR_IW_CXGBE, "%s:pau1 %p", __func__, ep); 1655 ep->com.cm_id->event_handler(ep->com.cm_id, &event); 1656 deref_cm_id(&ep->com); 1657 set_bit(ABORT_UPCALL, &ep->com.history); 1658 } 1659 CTR2(KTR_IW_CXGBE, "%s:pauE %p", __func__, ep); 1660} 1661 1662static void connect_reply_upcall(struct c4iw_ep *ep, int status) 1663{ 1664 struct iw_cm_event event; 1665 1666 CTR3(KTR_IW_CXGBE, "%s:cruB %p, status: %d", __func__, ep, status); 1667 memset(&event, 0, sizeof(event)); 1668 event.event = IW_CM_EVENT_CONNECT_REPLY; 1669 event.status = ((status == -ECONNABORTED) || (status == -EPIPE)) ? 1670 -ECONNRESET : status; 1671 event.local_addr = ep->com.local_addr; 1672 event.remote_addr = ep->com.remote_addr; 1673 1674 if ((status == 0) || (status == -ECONNREFUSED)) { 1675 1676 if (!ep->tried_with_mpa_v1) { 1677 1678 CTR2(KTR_IW_CXGBE, "%s:cru1 %p", __func__, ep); 1679 /* this means MPA_v2 is used */ 1680 event.ord = ep->ird; 1681 event.ird = ep->ord; 1682 event.private_data_len = ep->plen - 1683 sizeof(struct mpa_v2_conn_params); 1684 event.private_data = ep->mpa_pkt + 1685 sizeof(struct mpa_message) + 1686 sizeof(struct mpa_v2_conn_params); 1687 } else { 1688 1689 CTR2(KTR_IW_CXGBE, "%s:cru2 %p", __func__, ep); 1690 /* this means MPA_v1 is used */ 1691 event.ord = c4iw_max_read_depth; 1692 event.ird = c4iw_max_read_depth; 1693 event.private_data_len = ep->plen; 1694 event.private_data = ep->mpa_pkt + 1695 sizeof(struct mpa_message); 1696 } 1697 } 1698 1699 if (ep->com.cm_id) { 1700 1701 CTR2(KTR_IW_CXGBE, "%s:cru3 %p", __func__, ep); 1702 set_bit(CONN_RPL_UPCALL, &ep->com.history); 1703 ep->com.cm_id->event_handler(ep->com.cm_id, &event); 1704 } 1705 1706 if(status == -ECONNABORTED) { 1707 1708 CTR3(KTR_IW_CXGBE, "%s:cruE %p %d", __func__, ep, status); 1709 return; 1710 } 1711 1712 if (status < 0) { 1713 1714 CTR3(KTR_IW_CXGBE, "%s:cru4 %p %d", __func__, ep, status); 1715 deref_cm_id(&ep->com); 1716 } 1717 1718 CTR2(KTR_IW_CXGBE, "%s:cruE %p", __func__, ep); 1719} 1720 1721static int connect_request_upcall(struct c4iw_ep *ep) 1722{ 1723 struct iw_cm_event event; 1724 int ret; 1725 1726 CTR3(KTR_IW_CXGBE, "%s: ep %p, mpa_v1 %d", __func__, ep, 1727 ep->tried_with_mpa_v1); 1728 1729 memset(&event, 0, sizeof(event)); 1730 event.event = IW_CM_EVENT_CONNECT_REQUEST; 1731 event.local_addr = ep->com.local_addr; 1732 event.remote_addr = ep->com.remote_addr; 1733 event.provider_data = ep; 1734 1735 if (!ep->tried_with_mpa_v1) { 1736 /* this means MPA_v2 is used */ 1737 event.ord = ep->ord; 1738 event.ird = ep->ird; 1739 event.private_data_len = ep->plen - 1740 sizeof(struct mpa_v2_conn_params); 1741 event.private_data = ep->mpa_pkt + sizeof(struct mpa_message) + 1742 sizeof(struct mpa_v2_conn_params); 1743 } else { 1744 1745 /* this means MPA_v1 is used. Send max supported */ 1746 event.ord = c4iw_max_read_depth; 1747 event.ird = c4iw_max_read_depth; 1748 event.private_data_len = ep->plen; 1749 event.private_data = ep->mpa_pkt + sizeof(struct mpa_message); 1750 } 1751 1752 c4iw_get_ep(&ep->com); 1753 ret = ep->parent_ep->com.cm_id->event_handler(ep->parent_ep->com.cm_id, 1754 &event); 1755 if(ret) { 1756 CTR3(KTR_IW_CXGBE, "%s: ep %p, Failure while notifying event to" 1757 " IWCM, err:%d", __func__, ep, ret); 1758 c4iw_put_ep(&ep->com); 1759 } else 1760 /* Dereference parent_ep only in success case. 1761 * In case of failure, parent_ep is dereferenced by the caller 1762 * of process_mpa_request(). 1763 */ 1764 c4iw_put_ep(&ep->parent_ep->com); 1765 1766 set_bit(CONNREQ_UPCALL, &ep->com.history); 1767 return ret; 1768} 1769 1770static void established_upcall(struct c4iw_ep *ep) 1771{ 1772 struct iw_cm_event event; 1773 1774 CTR2(KTR_IW_CXGBE, "%s:euB %p", __func__, ep); 1775 memset(&event, 0, sizeof(event)); 1776 event.event = IW_CM_EVENT_ESTABLISHED; 1777 event.ird = ep->ord; 1778 event.ord = ep->ird; 1779 1780 if (ep->com.cm_id) { 1781 1782 CTR2(KTR_IW_CXGBE, "%s:eu1 %p", __func__, ep); 1783 ep->com.cm_id->event_handler(ep->com.cm_id, &event); 1784 set_bit(ESTAB_UPCALL, &ep->com.history); 1785 } 1786 CTR2(KTR_IW_CXGBE, "%s:euE %p", __func__, ep); 1787} 1788 1789 1790#define RELAXED_IRD_NEGOTIATION 1 1791 1792/* 1793 * process_mpa_reply - process streaming mode MPA reply 1794 * 1795 * Returns: 1796 * 1797 * 0 upon success indicating a connect request was delivered to the ULP 1798 * or the mpa request is incomplete but valid so far. 1799 * 1800 * 1 if a failure requires the caller to close the connection. 1801 * 1802 * 2 if a failure requires the caller to abort the connection. 1803 */ 1804static int process_mpa_reply(struct c4iw_ep *ep) 1805{ 1806 struct mpa_message *mpa; 1807 struct mpa_v2_conn_params *mpa_v2_params; 1808 u16 plen; 1809 u16 resp_ird, resp_ord; 1810 u8 rtr_mismatch = 0, insuff_ird = 0; 1811 struct c4iw_qp_attributes attrs = {0}; 1812 enum c4iw_qp_attr_mask mask; 1813 int err; 1814 struct mbuf *top, *m; 1815 int flags = MSG_DONTWAIT; 1816 struct uio uio; 1817 int disconnect = 0; 1818 1819 CTR2(KTR_IW_CXGBE, "%s:pmrB %p", __func__, ep); 1820 1821 /* 1822 * Stop mpa timer. If it expired, then 1823 * we ignore the MPA reply. process_timeout() 1824 * will abort the connection. 1825 */ 1826 if (STOP_EP_TIMER(ep)) 1827 return 0; 1828 1829 uio.uio_resid = 1000000; 1830 uio.uio_td = ep->com.thread; 1831 err = soreceive(ep->com.so, NULL, &uio, &top, NULL, &flags); 1832 1833 if (err) { 1834 1835 if (err == EWOULDBLOCK) { 1836 1837 CTR2(KTR_IW_CXGBE, "%s:pmr1 %p", __func__, ep); 1838 START_EP_TIMER(ep); 1839 return 0; 1840 } 1841 err = -err; 1842 CTR2(KTR_IW_CXGBE, "%s:pmr2 %p", __func__, ep); 1843 goto err; 1844 } 1845 1846 if (ep->com.so->so_rcv.sb_mb) { 1847 1848 CTR2(KTR_IW_CXGBE, "%s:pmr3 %p", __func__, ep); 1849 printf("%s data after soreceive called! so %p sb_mb %p top %p\n", 1850 __func__, ep->com.so, ep->com.so->so_rcv.sb_mb, top); 1851 } 1852 1853 m = top; 1854 1855 do { 1856 1857 CTR2(KTR_IW_CXGBE, "%s:pmr4 %p", __func__, ep); 1858 /* 1859 * If we get more than the supported amount of private data 1860 * then we must fail this connection. 1861 */ 1862 if (ep->mpa_pkt_len + m->m_len > sizeof(ep->mpa_pkt)) { 1863 1864 CTR3(KTR_IW_CXGBE, "%s:pmr5 %p %d", __func__, ep, 1865 ep->mpa_pkt_len + m->m_len); 1866 err = (-EINVAL); 1867 goto err_stop_timer; 1868 } 1869 1870 /* 1871 * copy the new data into our accumulation buffer. 1872 */ 1873 m_copydata(m, 0, m->m_len, &(ep->mpa_pkt[ep->mpa_pkt_len])); 1874 ep->mpa_pkt_len += m->m_len; 1875 if (!m->m_next) 1876 m = m->m_nextpkt; 1877 else 1878 m = m->m_next; 1879 } while (m); 1880 1881 m_freem(top); 1882 /* 1883 * if we don't even have the mpa message, then bail. 1884 */ 1885 if (ep->mpa_pkt_len < sizeof(*mpa)) { 1886 return 0; 1887 } 1888 mpa = (struct mpa_message *) ep->mpa_pkt; 1889 1890 /* Validate MPA header. */ 1891 if (mpa->revision > mpa_rev) { 1892 1893 CTR4(KTR_IW_CXGBE, "%s:pmr6 %p %d %d", __func__, ep, 1894 mpa->revision, mpa_rev); 1895 printk(KERN_ERR MOD "%s MPA version mismatch. Local = %d, " 1896 " Received = %d\n", __func__, mpa_rev, mpa->revision); 1897 err = -EPROTO; 1898 goto err_stop_timer; 1899 } 1900 1901 if (memcmp(mpa->key, MPA_KEY_REP, sizeof(mpa->key))) { 1902 1903 CTR2(KTR_IW_CXGBE, "%s:pmr7 %p", __func__, ep); 1904 err = -EPROTO; 1905 goto err_stop_timer; 1906 } 1907 1908 plen = ntohs(mpa->private_data_size); 1909 1910 /* 1911 * Fail if there's too much private data. 1912 */ 1913 if (plen > MPA_MAX_PRIVATE_DATA) { 1914 1915 CTR2(KTR_IW_CXGBE, "%s:pmr8 %p", __func__, ep); 1916 err = -EPROTO; 1917 goto err_stop_timer; 1918 } 1919 1920 /* 1921 * If plen does not account for pkt size 1922 */ 1923 if (ep->mpa_pkt_len > (sizeof(*mpa) + plen)) { 1924 1925 CTR2(KTR_IW_CXGBE, "%s:pmr9 %p", __func__, ep); 1926 STOP_EP_TIMER(ep); 1927 err = -EPROTO; 1928 goto err_stop_timer; 1929 } 1930 1931 ep->plen = (u8) plen; 1932 1933 /* 1934 * If we don't have all the pdata yet, then bail. 1935 * We'll continue process when more data arrives. 1936 */ 1937 if (ep->mpa_pkt_len < (sizeof(*mpa) + plen)) { 1938 1939 CTR2(KTR_IW_CXGBE, "%s:pmra %p", __func__, ep); 1940 return 0; 1941 } 1942 1943 if (mpa->flags & MPA_REJECT) { 1944 1945 CTR2(KTR_IW_CXGBE, "%s:pmrb %p", __func__, ep); 1946 err = -ECONNREFUSED; 1947 goto err_stop_timer; 1948 } 1949 1950 /* 1951 * If we get here we have accumulated the entire mpa 1952 * start reply message including private data. And 1953 * the MPA header is valid. 1954 */ 1955 ep->com.state = FPDU_MODE; 1956 ep->mpa_attr.crc_enabled = (mpa->flags & MPA_CRC) | crc_enabled ? 1 : 0; 1957 ep->mpa_attr.recv_marker_enabled = markers_enabled; 1958 ep->mpa_attr.xmit_marker_enabled = mpa->flags & MPA_MARKERS ? 1 : 0; 1959 ep->mpa_attr.version = mpa->revision; 1960 ep->mpa_attr.p2p_type = FW_RI_INIT_P2PTYPE_DISABLED; 1961 1962 if (mpa->revision == 2) { 1963 1964 CTR2(KTR_IW_CXGBE, "%s:pmrc %p", __func__, ep); 1965 ep->mpa_attr.enhanced_rdma_conn = 1966 mpa->flags & MPA_ENHANCED_RDMA_CONN ? 1 : 0; 1967 1968 if (ep->mpa_attr.enhanced_rdma_conn) { 1969 1970 CTR2(KTR_IW_CXGBE, "%s:pmrd %p", __func__, ep); 1971 mpa_v2_params = (struct mpa_v2_conn_params *) 1972 (ep->mpa_pkt + sizeof(*mpa)); 1973 resp_ird = ntohs(mpa_v2_params->ird) & 1974 MPA_V2_IRD_ORD_MASK; 1975 resp_ord = ntohs(mpa_v2_params->ord) & 1976 MPA_V2_IRD_ORD_MASK; 1977 1978 /* 1979 * This is a double-check. Ideally, below checks are 1980 * not required since ird/ord stuff has been taken 1981 * care of in c4iw_accept_cr 1982 */ 1983 if (ep->ird < resp_ord) { 1984 if (RELAXED_IRD_NEGOTIATION && resp_ord <= 1985 ep->com.dev->rdev.adap->params.max_ordird_qp) 1986 ep->ird = resp_ord; 1987 else 1988 insuff_ird = 1; 1989 } else if (ep->ird > resp_ord) { 1990 ep->ird = resp_ord; 1991 } 1992 if (ep->ord > resp_ird) { 1993 if (RELAXED_IRD_NEGOTIATION) 1994 ep->ord = resp_ird; 1995 else 1996 insuff_ird = 1; 1997 } 1998 if (insuff_ird) { 1999 err = -ENOMEM; 2000 ep->ird = resp_ord; 2001 ep->ord = resp_ird; 2002 } 2003 2004 if (ntohs(mpa_v2_params->ird) & 2005 MPA_V2_PEER2PEER_MODEL) { 2006 2007 CTR2(KTR_IW_CXGBE, "%s:pmrf %p", __func__, ep); 2008 if (ntohs(mpa_v2_params->ord) & 2009 MPA_V2_RDMA_WRITE_RTR) { 2010 2011 CTR2(KTR_IW_CXGBE, "%s:pmrg %p", __func__, ep); 2012 ep->mpa_attr.p2p_type = 2013 FW_RI_INIT_P2PTYPE_RDMA_WRITE; 2014 } 2015 else if (ntohs(mpa_v2_params->ord) & 2016 MPA_V2_RDMA_READ_RTR) { 2017 2018 CTR2(KTR_IW_CXGBE, "%s:pmrh %p", __func__, ep); 2019 ep->mpa_attr.p2p_type = 2020 FW_RI_INIT_P2PTYPE_READ_REQ; 2021 } 2022 } 2023 } 2024 } else { 2025 2026 CTR2(KTR_IW_CXGBE, "%s:pmri %p", __func__, ep); 2027 2028 if (mpa->revision == 1) { 2029 2030 CTR2(KTR_IW_CXGBE, "%s:pmrj %p", __func__, ep); 2031 2032 if (peer2peer) { 2033 2034 CTR2(KTR_IW_CXGBE, "%s:pmrk %p", __func__, ep); 2035 ep->mpa_attr.p2p_type = p2p_type; 2036 } 2037 } 2038 } 2039 2040 if (set_tcpinfo(ep)) { 2041 2042 CTR2(KTR_IW_CXGBE, "%s:pmrl %p", __func__, ep); 2043 printf("%s set_tcpinfo error\n", __func__); 2044 err = -ECONNRESET; 2045 goto err; 2046 } 2047 2048 CTR6(KTR_IW_CXGBE, "%s - crc_enabled = %d, recv_marker_enabled = %d, " 2049 "xmit_marker_enabled = %d, version = %d p2p_type = %d", __func__, 2050 ep->mpa_attr.crc_enabled, ep->mpa_attr.recv_marker_enabled, 2051 ep->mpa_attr.xmit_marker_enabled, ep->mpa_attr.version, 2052 ep->mpa_attr.p2p_type); 2053 2054 /* 2055 * If responder's RTR does not match with that of initiator, assign 2056 * FW_RI_INIT_P2PTYPE_DISABLED in mpa attributes so that RTR is not 2057 * generated when moving QP to RTS state. 2058 * A TERM message will be sent after QP has moved to RTS state 2059 */ 2060 if ((ep->mpa_attr.version == 2) && peer2peer && 2061 (ep->mpa_attr.p2p_type != p2p_type)) { 2062 2063 CTR2(KTR_IW_CXGBE, "%s:pmrm %p", __func__, ep); 2064 ep->mpa_attr.p2p_type = FW_RI_INIT_P2PTYPE_DISABLED; 2065 rtr_mismatch = 1; 2066 } 2067 2068 2069 //ep->ofld_txq = TOEPCB(ep->com.so)->ofld_txq; 2070 attrs.mpa_attr = ep->mpa_attr; 2071 attrs.max_ird = ep->ird; 2072 attrs.max_ord = ep->ord; 2073 attrs.llp_stream_handle = ep; 2074 attrs.next_state = C4IW_QP_STATE_RTS; 2075 2076 mask = C4IW_QP_ATTR_NEXT_STATE | 2077 C4IW_QP_ATTR_LLP_STREAM_HANDLE | C4IW_QP_ATTR_MPA_ATTR | 2078 C4IW_QP_ATTR_MAX_IRD | C4IW_QP_ATTR_MAX_ORD; 2079 2080 /* bind QP and TID with INIT_WR */ 2081 err = c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp, mask, &attrs, 1); 2082 2083 if (err) { 2084 2085 CTR2(KTR_IW_CXGBE, "%s:pmrn %p", __func__, ep); 2086 goto err; 2087 } 2088 2089 /* 2090 * If responder's RTR requirement did not match with what initiator 2091 * supports, generate TERM message 2092 */ 2093 if (rtr_mismatch) { 2094 2095 CTR2(KTR_IW_CXGBE, "%s:pmro %p", __func__, ep); 2096 printk(KERN_ERR "%s: RTR mismatch, sending TERM\n", __func__); 2097 attrs.layer_etype = LAYER_MPA | DDP_LLP; 2098 attrs.ecode = MPA_NOMATCH_RTR; 2099 attrs.next_state = C4IW_QP_STATE_TERMINATE; 2100 attrs.send_term = 1; 2101 err = c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp, 2102 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1); 2103 err = -ENOMEM; 2104 disconnect = 1; 2105 goto out; 2106 } 2107 2108 /* 2109 * Generate TERM if initiator IRD is not sufficient for responder 2110 * provided ORD. Currently, we do the same behaviour even when 2111 * responder provided IRD is also not sufficient as regards to 2112 * initiator ORD. 2113 */ 2114 if (insuff_ird) { 2115 2116 CTR2(KTR_IW_CXGBE, "%s:pmrp %p", __func__, ep); 2117 printk(KERN_ERR "%s: Insufficient IRD, sending TERM\n", 2118 __func__); 2119 attrs.layer_etype = LAYER_MPA | DDP_LLP; 2120 attrs.ecode = MPA_INSUFF_IRD; 2121 attrs.next_state = C4IW_QP_STATE_TERMINATE; 2122 attrs.send_term = 1; 2123 err = c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp, 2124 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1); 2125 err = -ENOMEM; 2126 disconnect = 1; 2127 goto out; 2128 } 2129 goto out; 2130err_stop_timer: 2131 STOP_EP_TIMER(ep); 2132err: 2133 disconnect = 2; 2134out: 2135 connect_reply_upcall(ep, err); 2136 CTR2(KTR_IW_CXGBE, "%s:pmrE %p", __func__, ep); 2137 return disconnect; 2138} 2139 2140/* 2141 * process_mpa_request - process streaming mode MPA request 2142 * 2143 * Returns: 2144 * 2145 * 0 upon success indicating a connect request was delivered to the ULP 2146 * or the mpa request is incomplete but valid so far. 2147 * 2148 * 1 if a failure requires the caller to close the connection. 2149 * 2150 * 2 if a failure requires the caller to abort the connection. 2151 */ 2152static int 2153process_mpa_request(struct c4iw_ep *ep) 2154{ 2155 struct mpa_message *mpa; 2156 struct mpa_v2_conn_params *mpa_v2_params; 2157 u16 plen; 2158 int flags = MSG_DONTWAIT; 2159 int rc; 2160 struct iovec iov; 2161 struct uio uio; 2162 enum c4iw_ep_state state = ep->com.state; 2163 2164 CTR3(KTR_IW_CXGBE, "%s: ep %p, state %s", __func__, ep, states[state]); 2165 2166 if (state != MPA_REQ_WAIT) 2167 return 0; 2168 2169 iov.iov_base = &ep->mpa_pkt[ep->mpa_pkt_len]; 2170 iov.iov_len = sizeof(ep->mpa_pkt) - ep->mpa_pkt_len; 2171 uio.uio_iov = &iov; 2172 uio.uio_iovcnt = 1; 2173 uio.uio_offset = 0; 2174 uio.uio_resid = sizeof(ep->mpa_pkt) - ep->mpa_pkt_len; 2175 uio.uio_segflg = UIO_SYSSPACE; 2176 uio.uio_rw = UIO_READ; 2177 uio.uio_td = NULL; /* uio.uio_td = ep->com.thread; */ 2178 2179 rc = soreceive(ep->com.so, NULL, &uio, NULL, NULL, &flags); 2180 if (rc == EAGAIN) 2181 return 0; 2182 else if (rc) 2183 goto err_stop_timer; 2184 2185 KASSERT(uio.uio_offset > 0, ("%s: sorecieve on so %p read no data", 2186 __func__, ep->com.so)); 2187 ep->mpa_pkt_len += uio.uio_offset; 2188 2189 /* 2190 * If we get more than the supported amount of private data then we must 2191 * fail this connection. XXX: check so_rcv->sb_cc, or peek with another 2192 * soreceive, or increase the size of mpa_pkt by 1 and abort if the last 2193 * byte is filled by the soreceive above. 2194 */ 2195 2196 /* Don't even have the MPA message. Wait for more data to arrive. */ 2197 if (ep->mpa_pkt_len < sizeof(*mpa)) 2198 return 0; 2199 mpa = (struct mpa_message *) ep->mpa_pkt; 2200 2201 /* 2202 * Validate MPA Header. 2203 */ 2204 if (mpa->revision > mpa_rev) { 2205 log(LOG_ERR, "%s: MPA version mismatch. Local = %d," 2206 " Received = %d\n", __func__, mpa_rev, mpa->revision); 2207 goto err_stop_timer; 2208 } 2209 2210 if (memcmp(mpa->key, MPA_KEY_REQ, sizeof(mpa->key))) 2211 goto err_stop_timer; 2212 2213 /* 2214 * Fail if there's too much private data. 2215 */ 2216 plen = ntohs(mpa->private_data_size); 2217 if (plen > MPA_MAX_PRIVATE_DATA) 2218 goto err_stop_timer; 2219 2220 /* 2221 * If plen does not account for pkt size 2222 */ 2223 if (ep->mpa_pkt_len > (sizeof(*mpa) + plen)) 2224 goto err_stop_timer; 2225 2226 ep->plen = (u8) plen; 2227 2228 /* 2229 * If we don't have all the pdata yet, then bail. 2230 */ 2231 if (ep->mpa_pkt_len < (sizeof(*mpa) + plen)) 2232 return 0; 2233 2234 /* 2235 * If we get here we have accumulated the entire mpa 2236 * start reply message including private data. 2237 */ 2238 ep->mpa_attr.initiator = 0; 2239 ep->mpa_attr.crc_enabled = (mpa->flags & MPA_CRC) | crc_enabled ? 1 : 0; 2240 ep->mpa_attr.recv_marker_enabled = markers_enabled; 2241 ep->mpa_attr.xmit_marker_enabled = mpa->flags & MPA_MARKERS ? 1 : 0; 2242 ep->mpa_attr.version = mpa->revision; 2243 if (mpa->revision == 1) 2244 ep->tried_with_mpa_v1 = 1; 2245 ep->mpa_attr.p2p_type = FW_RI_INIT_P2PTYPE_DISABLED; 2246 2247 if (mpa->revision == 2) { 2248 ep->mpa_attr.enhanced_rdma_conn = 2249 mpa->flags & MPA_ENHANCED_RDMA_CONN ? 1 : 0; 2250 if (ep->mpa_attr.enhanced_rdma_conn) { 2251 mpa_v2_params = (struct mpa_v2_conn_params *) 2252 (ep->mpa_pkt + sizeof(*mpa)); 2253 ep->ird = ntohs(mpa_v2_params->ird) & 2254 MPA_V2_IRD_ORD_MASK; 2255 ep->ird = min_t(u32, ep->ird, 2256 cur_max_read_depth(ep->com.dev)); 2257 ep->ord = ntohs(mpa_v2_params->ord) & 2258 MPA_V2_IRD_ORD_MASK; 2259 ep->ord = min_t(u32, ep->ord, 2260 cur_max_read_depth(ep->com.dev)); 2261 CTR3(KTR_IW_CXGBE, "%s initiator ird %u ord %u\n", 2262 __func__, ep->ird, ep->ord); 2263 if (ntohs(mpa_v2_params->ird) & MPA_V2_PEER2PEER_MODEL) 2264 if (peer2peer) { 2265 if (ntohs(mpa_v2_params->ord) & 2266 MPA_V2_RDMA_WRITE_RTR) 2267 ep->mpa_attr.p2p_type = 2268 FW_RI_INIT_P2PTYPE_RDMA_WRITE; 2269 else if (ntohs(mpa_v2_params->ord) & 2270 MPA_V2_RDMA_READ_RTR) 2271 ep->mpa_attr.p2p_type = 2272 FW_RI_INIT_P2PTYPE_READ_REQ; 2273 } 2274 } 2275 } else if (mpa->revision == 1 && peer2peer) 2276 ep->mpa_attr.p2p_type = p2p_type; 2277 2278 if (set_tcpinfo(ep)) 2279 goto err_stop_timer; 2280 2281 CTR5(KTR_IW_CXGBE, "%s: crc_enabled = %d, recv_marker_enabled = %d, " 2282 "xmit_marker_enabled = %d, version = %d", __func__, 2283 ep->mpa_attr.crc_enabled, ep->mpa_attr.recv_marker_enabled, 2284 ep->mpa_attr.xmit_marker_enabled, ep->mpa_attr.version); 2285 2286 ep->com.state = MPA_REQ_RCVD; 2287 STOP_EP_TIMER(ep); 2288 2289 /* drive upcall */ 2290 if (ep->parent_ep->com.state != DEAD) 2291 if (connect_request_upcall(ep)) 2292 goto err_out; 2293 return 0; 2294 2295err_stop_timer: 2296 STOP_EP_TIMER(ep); 2297err_out: 2298 return 2; 2299} 2300 2301/* 2302 * Upcall from the adapter indicating data has been transmitted. 2303 * For us its just the single MPA request or reply. We can now free 2304 * the skb holding the mpa message. 2305 */ 2306int c4iw_reject_cr(struct iw_cm_id *cm_id, const void *pdata, u8 pdata_len) 2307{ 2308 int err; 2309 struct c4iw_ep *ep = to_ep(cm_id); 2310 int abort = 0; 2311 2312 mutex_lock(&ep->com.mutex); 2313 CTR2(KTR_IW_CXGBE, "%s:crcB %p", __func__, ep); 2314 2315 if ((ep->com.state == DEAD) || 2316 (ep->com.state != MPA_REQ_RCVD)) { 2317 2318 CTR2(KTR_IW_CXGBE, "%s:crc1 %p", __func__, ep); 2319 mutex_unlock(&ep->com.mutex); 2320 c4iw_put_ep(&ep->com); 2321 return -ECONNRESET; 2322 } 2323 set_bit(ULP_REJECT, &ep->com.history); 2324 2325 if (mpa_rev == 0) { 2326 2327 CTR2(KTR_IW_CXGBE, "%s:crc2 %p", __func__, ep); 2328 abort = 1; 2329 } 2330 else { 2331 2332 CTR2(KTR_IW_CXGBE, "%s:crc3 %p", __func__, ep); 2333 abort = send_mpa_reject(ep, pdata, pdata_len); 2334 } 2335 STOP_EP_TIMER(ep); 2336 err = c4iw_ep_disconnect(ep, abort != 0, GFP_KERNEL); 2337 mutex_unlock(&ep->com.mutex); 2338 c4iw_put_ep(&ep->com); 2339 CTR3(KTR_IW_CXGBE, "%s:crc4 %p, err: %d", __func__, ep, err); 2340 return 0; 2341} 2342 2343int c4iw_accept_cr(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param) 2344{ 2345 int err; 2346 struct c4iw_qp_attributes attrs = {0}; 2347 enum c4iw_qp_attr_mask mask; 2348 struct c4iw_ep *ep = to_ep(cm_id); 2349 struct c4iw_dev *h = to_c4iw_dev(cm_id->device); 2350 struct c4iw_qp *qp = get_qhp(h, conn_param->qpn); 2351 int abort = 0; 2352 2353 mutex_lock(&ep->com.mutex); 2354 CTR2(KTR_IW_CXGBE, "%s:cacB %p", __func__, ep); 2355 2356 if ((ep->com.state == DEAD) || 2357 (ep->com.state != MPA_REQ_RCVD)) { 2358 2359 CTR2(KTR_IW_CXGBE, "%s:cac1 %p", __func__, ep); 2360 err = -ECONNRESET; 2361 goto err_out; 2362 } 2363 2364 BUG_ON(!qp); 2365 2366 set_bit(ULP_ACCEPT, &ep->com.history); 2367 2368 if ((conn_param->ord > c4iw_max_read_depth) || 2369 (conn_param->ird > c4iw_max_read_depth)) { 2370 2371 CTR2(KTR_IW_CXGBE, "%s:cac2 %p", __func__, ep); 2372 err = -EINVAL; 2373 goto err_abort; 2374 } 2375 2376 if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) { 2377 2378 CTR2(KTR_IW_CXGBE, "%s:cac3 %p", __func__, ep); 2379 2380 if (conn_param->ord > ep->ird) { 2381 if (RELAXED_IRD_NEGOTIATION) { 2382 conn_param->ord = ep->ird; 2383 } else { 2384 ep->ird = conn_param->ird; 2385 ep->ord = conn_param->ord; 2386 send_mpa_reject(ep, conn_param->private_data, 2387 conn_param->private_data_len); 2388 err = -ENOMEM; 2389 goto err_abort; 2390 } 2391 } 2392 if (conn_param->ird < ep->ord) { 2393 if (RELAXED_IRD_NEGOTIATION && 2394 ep->ord <= h->rdev.adap->params.max_ordird_qp) { 2395 conn_param->ird = ep->ord; 2396 } else { 2397 err = -ENOMEM; 2398 goto err_abort; 2399 } 2400 } 2401 } 2402 ep->ird = conn_param->ird; 2403 ep->ord = conn_param->ord; 2404 2405 if (ep->mpa_attr.version == 1) { 2406 if (peer2peer && ep->ird == 0) 2407 ep->ird = 1; 2408 } else { 2409 if (peer2peer && 2410 (ep->mpa_attr.p2p_type != FW_RI_INIT_P2PTYPE_DISABLED) && 2411 (p2p_type == FW_RI_INIT_P2PTYPE_READ_REQ) && ep->ird == 0) 2412 ep->ird = 1; 2413 } 2414 2415 CTR4(KTR_IW_CXGBE, "%s %d ird %d ord %d\n", __func__, __LINE__, 2416 ep->ird, ep->ord); 2417 2418 ep->com.cm_id = cm_id; 2419 ref_cm_id(&ep->com); 2420 ep->com.qp = qp; 2421 ref_qp(ep); 2422 //ep->ofld_txq = TOEPCB(ep->com.so)->ofld_txq; 2423 2424 /* bind QP to EP and move to RTS */ 2425 attrs.mpa_attr = ep->mpa_attr; 2426 attrs.max_ird = ep->ird; 2427 attrs.max_ord = ep->ord; 2428 attrs.llp_stream_handle = ep; 2429 attrs.next_state = C4IW_QP_STATE_RTS; 2430 2431 /* bind QP and TID with INIT_WR */ 2432 mask = C4IW_QP_ATTR_NEXT_STATE | 2433 C4IW_QP_ATTR_LLP_STREAM_HANDLE | 2434 C4IW_QP_ATTR_MPA_ATTR | 2435 C4IW_QP_ATTR_MAX_IRD | 2436 C4IW_QP_ATTR_MAX_ORD; 2437 2438 err = c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp, mask, &attrs, 1); 2439 if (err) { 2440 CTR3(KTR_IW_CXGBE, "%s:caca %p, err: %d", __func__, ep, err); 2441 goto err_defef_cm_id; 2442 } 2443 2444 err = send_mpa_reply(ep, conn_param->private_data, 2445 conn_param->private_data_len); 2446 if (err) { 2447 CTR3(KTR_IW_CXGBE, "%s:cacb %p, err: %d", __func__, ep, err); 2448 goto err_defef_cm_id; 2449 } 2450 2451 ep->com.state = FPDU_MODE; 2452 established_upcall(ep); 2453 mutex_unlock(&ep->com.mutex); 2454 c4iw_put_ep(&ep->com); 2455 CTR2(KTR_IW_CXGBE, "%s:cacE %p", __func__, ep); 2456 return 0; 2457err_defef_cm_id: 2458 deref_cm_id(&ep->com); 2459err_abort: 2460 abort = 1; 2461err_out: 2462 if (abort) 2463 c4iw_ep_disconnect(ep, 1, GFP_KERNEL); 2464 mutex_unlock(&ep->com.mutex); 2465 c4iw_put_ep(&ep->com); 2466 CTR2(KTR_IW_CXGBE, "%s:cacE err %p", __func__, ep); 2467 return err; 2468} 2469 2470static int 2471c4iw_sock_create(struct sockaddr_storage *laddr, struct socket **so) 2472{ 2473 int ret; 2474 int size; 2475 struct socket *sock = NULL; 2476 2477 ret = sock_create_kern(laddr->ss_family, 2478 SOCK_STREAM, IPPROTO_TCP, &sock); 2479 if (ret) { 2480 CTR2(KTR_IW_CXGBE, "%s:Failed to create TCP socket. err %d", 2481 __func__, ret); 2482 return ret; 2483 } 2484 2485 ret = sobind(sock, (struct sockaddr *)laddr, curthread); 2486 if (ret) { 2487 CTR2(KTR_IW_CXGBE, "%s:Failed to bind socket. err %p", 2488 __func__, ret); 2489 sock_release(sock); 2490 return ret; 2491 } 2492 2493 size = laddr->ss_family == AF_INET6 ? 2494 sizeof(struct sockaddr_in6) : sizeof(struct sockaddr_in); 2495 ret = sock_getname(sock, (struct sockaddr *)laddr, &size, 0); 2496 if (ret) { 2497 CTR2(KTR_IW_CXGBE, "%s:sock_getname failed. err %p", 2498 __func__, ret); 2499 sock_release(sock); 2500 return ret; 2501 } 2502 2503 *so = sock; 2504 return 0; 2505} 2506 2507int c4iw_connect(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param) 2508{ 2509 int err = 0; 2510 struct c4iw_dev *dev = to_c4iw_dev(cm_id->device); 2511 struct c4iw_ep *ep = NULL; 2512 struct ifnet *nh_ifp; /* Logical egress interface */ 2513 2514 CTR2(KTR_IW_CXGBE, "%s:ccB %p", __func__, cm_id); 2515 2516 2517 if ((conn_param->ord > c4iw_max_read_depth) || 2518 (conn_param->ird > c4iw_max_read_depth)) { 2519 2520 CTR2(KTR_IW_CXGBE, "%s:cc1 %p", __func__, cm_id); 2521 err = -EINVAL; 2522 goto out; 2523 } 2524 ep = alloc_ep(sizeof(*ep), GFP_KERNEL); 2525 2526 init_timer(&ep->timer); 2527 ep->plen = conn_param->private_data_len; 2528 2529 if (ep->plen) { 2530 2531 CTR2(KTR_IW_CXGBE, "%s:cc3 %p", __func__, ep); 2532 memcpy(ep->mpa_pkt + sizeof(struct mpa_message), 2533 conn_param->private_data, ep->plen); 2534 } 2535 ep->ird = conn_param->ird; 2536 ep->ord = conn_param->ord; 2537 2538 if (peer2peer && ep->ord == 0) { 2539 2540 CTR2(KTR_IW_CXGBE, "%s:cc4 %p", __func__, ep); 2541 ep->ord = 1; 2542 } 2543 2544 ep->com.dev = dev; 2545 ep->com.cm_id = cm_id; 2546 ref_cm_id(&ep->com); 2547 ep->com.qp = get_qhp(dev, conn_param->qpn); 2548 2549 if (!ep->com.qp) { 2550 2551 CTR2(KTR_IW_CXGBE, "%s:cc5 %p", __func__, ep); 2552 err = -EINVAL; 2553 goto fail; 2554 } 2555 ref_qp(ep); 2556 ep->com.thread = curthread; 2557 2558 err = get_ifnet_from_raddr(&cm_id->remote_addr, &nh_ifp); 2559 if (err) { 2560 2561 CTR2(KTR_IW_CXGBE, "%s:cc7 %p", __func__, ep); 2562 printk(KERN_ERR MOD "%s - cannot find route.\n", __func__); 2563 err = EHOSTUNREACH; 2564 return err; 2565 } 2566 2567 if (!(nh_ifp->if_capenable & IFCAP_TOE) || 2568 TOEDEV(nh_ifp) == NULL) { 2569 err = -ENOPROTOOPT; 2570 goto fail; 2571 } 2572 ep->com.state = CONNECTING; 2573 ep->tos = 0; 2574 ep->com.local_addr = cm_id->local_addr; 2575 ep->com.remote_addr = cm_id->remote_addr; 2576 2577 err = c4iw_sock_create(&cm_id->local_addr, &ep->com.so); 2578 if (err) 2579 goto fail; 2580 2581 setiwsockopt(ep->com.so); 2582 err = -soconnect(ep->com.so, (struct sockaddr *)&ep->com.remote_addr, 2583 ep->com.thread); 2584 if (!err) { 2585 init_iwarp_socket(ep->com.so, &ep->com); 2586 goto out; 2587 } else 2588 goto fail_free_so; 2589 2590fail_free_so: 2591 sock_release(ep->com.so); 2592fail: 2593 deref_cm_id(&ep->com); 2594 c4iw_put_ep(&ep->com); 2595 ep = NULL; 2596out: 2597 CTR2(KTR_IW_CXGBE, "%s:ccE ret:%d", __func__, err); 2598 return err; 2599} 2600 2601/* 2602 * iwcm->create_listen. Returns -errno on failure. 2603 */ 2604int 2605c4iw_create_listen(struct iw_cm_id *cm_id, int backlog) 2606{ 2607 struct c4iw_dev *dev = to_c4iw_dev(cm_id->device); 2608 struct c4iw_listen_ep *lep = NULL; 2609 struct listen_port_info *port_info = NULL; 2610 int rc = 0; 2611 2612 CTR3(KTR_IW_CXGBE, "%s: cm_id %p, backlog %s", __func__, cm_id, 2613 backlog); 2614 lep = alloc_ep(sizeof(*lep), GFP_KERNEL); 2615 lep->com.cm_id = cm_id; 2616 ref_cm_id(&lep->com); 2617 lep->com.dev = dev; 2618 lep->backlog = backlog; 2619 lep->com.local_addr = cm_id->local_addr; 2620 lep->com.thread = curthread; 2621 cm_id->provider_data = lep; 2622 lep->com.state = LISTEN; 2623 2624 /* In case of INDADDR_ANY, ibcore creates cmid for each device and 2625 * invokes iw_cxgbe listener callbacks assuming that iw_cxgbe creates 2626 * HW listeners for each device seperately. But toecore expects single 2627 * solisten() call with INADDR_ANY address to create HW listeners on 2628 * all devices for a given port number. So iw_cxgbe driver calls 2629 * solisten() only once for INADDR_ANY(usually done at first time 2630 * listener callback from ibcore). And all the subsequent INADDR_ANY 2631 * listener callbacks from ibcore(for the same port address) do not 2632 * invoke solisten() as first listener callback has already created 2633 * listeners for all other devices(via solisten). 2634 */ 2635 if (c4iw_any_addr((struct sockaddr *)&lep->com.local_addr)) { 2636 port_info = add_ep_to_listenlist(lep); 2637 /* skip solisten() if refcnt > 1, as the listeners were 2638 * alredy created by 'Master lep' 2639 */ 2640 if (port_info->refcnt > 1) { 2641 /* As there will be only one listener socket for a TCP 2642 * port, copy Master lep's socket pointer to other lep's 2643 * that are belonging to same TCP port. 2644 */ 2645 struct c4iw_listen_ep *head_lep = 2646 container_of(port_info->lep_list.next, 2647 struct c4iw_listen_ep, listen_ep_list); 2648 lep->com.so = head_lep->com.so; 2649 goto out; 2650 } 2651 } 2652 rc = c4iw_sock_create(&cm_id->local_addr, &lep->com.so); 2653 if (rc) { 2654 CTR2(KTR_IW_CXGBE, "%s:Failed to create socket. err %d", 2655 __func__, rc); 2656 goto fail; 2657 } 2658 2659 rc = solisten(lep->com.so, backlog, curthread); 2660 if (rc) { 2661 CTR3(KTR_IW_CXGBE, "%s:Failed to listen on sock:%p. err %d", 2662 __func__, lep->com.so, rc); 2663 goto fail_free_so; 2664 } 2665 init_iwarp_socket(lep->com.so, &lep->com); 2666out: 2667 return 0; 2668 2669fail_free_so: 2670 sock_release(lep->com.so); 2671fail: 2672 if (port_info) 2673 rem_ep_from_listenlist(lep); 2674 deref_cm_id(&lep->com); 2675 c4iw_put_ep(&lep->com); 2676 return rc; 2677} 2678 2679int 2680c4iw_destroy_listen(struct iw_cm_id *cm_id) 2681{ 2682 struct c4iw_listen_ep *lep = to_listen_ep(cm_id); 2683 2684 mutex_lock(&lep->com.mutex); 2685 CTR3(KTR_IW_CXGBE, "%s: cm_id %p, state %s", __func__, cm_id, 2686 states[lep->com.state]); 2687 2688 lep->com.state = DEAD; 2689 if (c4iw_any_addr((struct sockaddr *)&lep->com.local_addr)) { 2690 /* if no refcount then close listen socket */ 2691 if (!rem_ep_from_listenlist(lep)) 2692 close_socket(lep->com.so); 2693 } else 2694 close_socket(lep->com.so); 2695 deref_cm_id(&lep->com); 2696 mutex_unlock(&lep->com.mutex); 2697 c4iw_put_ep(&lep->com); 2698 return 0; 2699} 2700 2701int __c4iw_ep_disconnect(struct c4iw_ep *ep, int abrupt, gfp_t gfp) 2702{ 2703 int ret; 2704 mutex_lock(&ep->com.mutex); 2705 ret = c4iw_ep_disconnect(ep, abrupt, gfp); 2706 mutex_unlock(&ep->com.mutex); 2707 return ret; 2708} 2709 2710int c4iw_ep_disconnect(struct c4iw_ep *ep, int abrupt, gfp_t gfp) 2711{ 2712 int ret = 0; 2713 int close = 0; 2714 int fatal = 0; 2715 struct c4iw_rdev *rdev; 2716 2717 2718 CTR2(KTR_IW_CXGBE, "%s:cedB %p", __func__, ep); 2719 2720 rdev = &ep->com.dev->rdev; 2721 2722 if (c4iw_fatal_error(rdev)) { 2723 2724 CTR2(KTR_IW_CXGBE, "%s:ced1 %p", __func__, ep); 2725 fatal = 1; 2726 close_complete_upcall(ep, -ECONNRESET); 2727 send_abort(ep); 2728 ep->com.state = DEAD; 2729 } 2730 CTR3(KTR_IW_CXGBE, "%s:ced2 %p %s", __func__, ep, 2731 states[ep->com.state]); 2732 2733 /* 2734 * Ref the ep here in case we have fatal errors causing the 2735 * ep to be released and freed. 2736 */ 2737 c4iw_get_ep(&ep->com); 2738 switch (ep->com.state) { 2739 2740 case MPA_REQ_WAIT: 2741 case MPA_REQ_SENT: 2742 case MPA_REQ_RCVD: 2743 case MPA_REP_SENT: 2744 case FPDU_MODE: 2745 close = 1; 2746 if (abrupt) 2747 ep->com.state = ABORTING; 2748 else { 2749 ep->com.state = CLOSING; 2750 START_EP_TIMER(ep); 2751 } 2752 set_bit(CLOSE_SENT, &ep->com.flags); 2753 break; 2754 2755 case CLOSING: 2756 2757 if (!test_and_set_bit(CLOSE_SENT, &ep->com.flags)) { 2758 2759 close = 1; 2760 if (abrupt) { 2761 STOP_EP_TIMER(ep); 2762 ep->com.state = ABORTING; 2763 } else 2764 ep->com.state = MORIBUND; 2765 } 2766 break; 2767 2768 case MORIBUND: 2769 case ABORTING: 2770 case DEAD: 2771 CTR3(KTR_IW_CXGBE, 2772 "%s ignoring disconnect ep %p state %u", __func__, 2773 ep, ep->com.state); 2774 break; 2775 2776 default: 2777 BUG(); 2778 break; 2779 } 2780 2781 2782 if (close) { 2783 2784 CTR2(KTR_IW_CXGBE, "%s:ced3 %p", __func__, ep); 2785 2786 if (abrupt) { 2787 2788 CTR2(KTR_IW_CXGBE, "%s:ced4 %p", __func__, ep); 2789 set_bit(EP_DISC_ABORT, &ep->com.history); 2790 close_complete_upcall(ep, -ECONNRESET); 2791 ret = send_abort(ep); 2792 if (ret) 2793 fatal = 1; 2794 } else { 2795 2796 CTR2(KTR_IW_CXGBE, "%s:ced5 %p", __func__, ep); 2797 set_bit(EP_DISC_CLOSE, &ep->com.history); 2798 2799 if (!ep->parent_ep) 2800 ep->com.state = MORIBUND; 2801 sodisconnect(ep->com.so); 2802 } 2803 2804 } 2805 2806 if (fatal) { 2807 set_bit(EP_DISC_FAIL, &ep->com.history); 2808 if (!abrupt) { 2809 STOP_EP_TIMER(ep); 2810 close_complete_upcall(ep, -EIO); 2811 } 2812 if (ep->com.qp) { 2813 struct c4iw_qp_attributes attrs = {0}; 2814 2815 attrs.next_state = C4IW_QP_STATE_ERROR; 2816 ret = c4iw_modify_qp(ep->com.dev, ep->com.qp, 2817 C4IW_QP_ATTR_NEXT_STATE, 2818 &attrs, 1); 2819 if (ret) { 2820 CTR2(KTR_IW_CXGBE, "%s:ced7 %p", __func__, ep); 2821 printf("%s - qp <- error failed!\n", __func__); 2822 } 2823 } 2824 release_ep_resources(ep); 2825 ep->com.state = DEAD; 2826 CTR2(KTR_IW_CXGBE, "%s:ced6 %p", __func__, ep); 2827 } 2828 c4iw_put_ep(&ep->com); 2829 CTR2(KTR_IW_CXGBE, "%s:cedE %p", __func__, ep); 2830 return ret; 2831} 2832 2833#ifdef C4IW_EP_REDIRECT 2834int c4iw_ep_redirect(void *ctx, struct dst_entry *old, struct dst_entry *new, 2835 struct l2t_entry *l2t) 2836{ 2837 struct c4iw_ep *ep = ctx; 2838 2839 if (ep->dst != old) 2840 return 0; 2841 2842 PDBG("%s ep %p redirect to dst %p l2t %p\n", __func__, ep, new, 2843 l2t); 2844 dst_hold(new); 2845 cxgb4_l2t_release(ep->l2t); 2846 ep->l2t = l2t; 2847 dst_release(old); 2848 ep->dst = new; 2849 return 1; 2850} 2851#endif 2852 2853 2854 2855static void ep_timeout(unsigned long arg) 2856{ 2857 struct c4iw_ep *ep = (struct c4iw_ep *)arg; 2858 2859 if (!test_and_set_bit(TIMEOUT, &ep->com.flags)) { 2860 2861 /* 2862 * Only insert if it is not already on the list. 2863 */ 2864 if (!(ep->com.ep_events & C4IW_EVENT_TIMEOUT)) { 2865 CTR2(KTR_IW_CXGBE, "%s:et1 %p", __func__, ep); 2866 add_ep_to_req_list(ep, C4IW_EVENT_TIMEOUT); 2867 } 2868 } 2869} 2870 2871static int fw6_wr_rpl(struct adapter *sc, const __be64 *rpl) 2872{ 2873 uint64_t val = be64toh(*rpl); 2874 int ret; 2875 struct c4iw_wr_wait *wr_waitp; 2876 2877 ret = (int)((val >> 8) & 0xff); 2878 wr_waitp = (struct c4iw_wr_wait *)rpl[1]; 2879 CTR3(KTR_IW_CXGBE, "%s wr_waitp %p ret %u", __func__, wr_waitp, ret); 2880 if (wr_waitp) 2881 c4iw_wake_up(wr_waitp, ret ? -ret : 0); 2882 2883 return (0); 2884} 2885 2886static int fw6_cqe_handler(struct adapter *sc, const __be64 *rpl) 2887{ 2888 struct cqe_list_entry *cle; 2889 unsigned long flag; 2890 2891 cle = malloc(sizeof(*cle), M_CXGBE, M_NOWAIT); 2892 cle->rhp = sc->iwarp_softc; 2893 cle->err_cqe = *(const struct t4_cqe *)(&rpl[0]); 2894 2895 spin_lock_irqsave(&err_cqe_lock, flag); 2896 list_add_tail(&cle->entry, &err_cqe_list); 2897 queue_work(c4iw_taskq, &c4iw_task); 2898 spin_unlock_irqrestore(&err_cqe_lock, flag); 2899 2900 return (0); 2901} 2902 2903static int 2904process_terminate(struct c4iw_ep *ep) 2905{ 2906 struct c4iw_qp_attributes attrs = {0}; 2907 2908 CTR2(KTR_IW_CXGBE, "%s:tB %p %d", __func__, ep); 2909 2910 if (ep && ep->com.qp) { 2911 2912 printk(KERN_WARNING MOD "TERM received tid %u qpid %u\n", 2913 ep->hwtid, ep->com.qp->wq.sq.qid); 2914 attrs.next_state = C4IW_QP_STATE_TERMINATE; 2915 c4iw_modify_qp(ep->com.dev, ep->com.qp, C4IW_QP_ATTR_NEXT_STATE, &attrs, 2916 1); 2917 } else 2918 printk(KERN_WARNING MOD "TERM received tid %u no ep/qp\n", 2919 ep->hwtid); 2920 CTR2(KTR_IW_CXGBE, "%s:tE %p %d", __func__, ep); 2921 2922 return 0; 2923} 2924 2925int __init c4iw_cm_init(void) 2926{ 2927 2928 t4_register_cpl_handler(CPL_RDMA_TERMINATE, terminate); 2929 t4_register_fw_msg_handler(FW6_TYPE_WR_RPL, fw6_wr_rpl); 2930 t4_register_fw_msg_handler(FW6_TYPE_CQE, fw6_cqe_handler); 2931 t4_register_an_handler(c4iw_ev_handler); 2932 2933 TAILQ_INIT(&req_list); 2934 spin_lock_init(&req_lock); 2935 INIT_LIST_HEAD(&err_cqe_list); 2936 spin_lock_init(&err_cqe_lock); 2937 2938 INIT_WORK(&c4iw_task, process_req); 2939 2940 c4iw_taskq = create_singlethread_workqueue("iw_cxgbe"); 2941 if (!c4iw_taskq) 2942 return -ENOMEM; 2943 2944 return 0; 2945} 2946 2947void __exit c4iw_cm_term(void) 2948{ 2949 WARN_ON(!TAILQ_EMPTY(&req_list)); 2950 WARN_ON(!list_empty(&err_cqe_list)); 2951 flush_workqueue(c4iw_taskq); 2952 destroy_workqueue(c4iw_taskq); 2953 2954 t4_register_cpl_handler(CPL_RDMA_TERMINATE, NULL); 2955 t4_register_fw_msg_handler(FW6_TYPE_WR_RPL, NULL); 2956 t4_register_fw_msg_handler(FW6_TYPE_CQE, NULL); 2957 t4_register_an_handler(NULL); 2958} 2959#endif 2960