cm.c revision 331769
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 331769 2018-03-30 18:06:29Z hselasky $"); 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 struct sockaddr_in *remote = NULL; 951 int ret = 0; 952 953 MPASS(new_so != NULL); 954 955 if (c4iw_any_addr((struct sockaddr *)&master_lep->com.local_addr)) { 956 /* Here we need to find the 'real_lep' that belongs to the 957 * incomming socket's network interface, such that the newly 958 * created 'ep' can be attached to the real 'lep'. 959 */ 960 real_lep = find_real_listen_ep(master_lep, new_so); 961 if (real_lep == NULL) { 962 CTR2(KTR_IW_CXGBE, "%s: Could not find the real listen " 963 "ep for sock: %p", __func__, new_so); 964 log(LOG_ERR,"%s: Could not find the real listen ep for " 965 "sock: %p\n", __func__, new_so); 966 /* FIXME: properly free the 'new_so' in failure case. 967 * Use of soabort() and soclose() are not legal 968 * here(before soaccept()). 969 */ 970 return; 971 } 972 } else /* for Non-Wildcard address, master_lep is always the real_lep */ 973 real_lep = master_lep; 974 975 new_ep = alloc_ep(sizeof(*new_ep), GFP_KERNEL); 976 977 CTR6(KTR_IW_CXGBE, "%s: master_lep %p, real_lep: %p, new ep %p, " 978 "listening so %p, new so %p", __func__, master_lep, real_lep, 979 new_ep, master_lep->com.so, new_so); 980 981 new_ep->com.dev = real_lep->com.dev; 982 new_ep->com.so = new_so; 983 new_ep->com.cm_id = NULL; 984 new_ep->com.thread = real_lep->com.thread; 985 new_ep->parent_ep = real_lep; 986 987 GET_LOCAL_ADDR(&new_ep->com.local_addr, new_so); 988 GET_REMOTE_ADDR(&new_ep->com.remote_addr, new_so); 989 c4iw_get_ep(&real_lep->com); 990 init_timer(&new_ep->timer); 991 new_ep->com.state = MPA_REQ_WAIT; 992 START_EP_TIMER(new_ep); 993 994 setiwsockopt(new_so); 995 ret = soaccept(new_so, (struct sockaddr **)&remote); 996 if (ret != 0) { 997 CTR4(KTR_IW_CXGBE, 998 "%s:listen sock:%p, new sock:%p, ret:%d\n", 999 __func__, master_lep->com.so, new_so, ret); 1000 if (remote != NULL) 1001 free(remote, M_SONAME); 1002 uninit_iwarp_socket(new_so); 1003 soclose(new_so); 1004 c4iw_put_ep(&new_ep->com); 1005 c4iw_put_ep(&real_lep->com); 1006 return; 1007 } 1008 free(remote, M_SONAME); 1009 1010 /* MPA request might have been queued up on the socket already, so we 1011 * initialize the socket/upcall_handler under lock to prevent processing 1012 * MPA request on another thread(via process_req()) simultaniously. 1013 */ 1014 c4iw_get_ep(&new_ep->com); /* Dereferenced at the end below, this is to 1015 avoid freeing of ep before ep unlock. */ 1016 mutex_lock(&new_ep->com.mutex); 1017 init_iwarp_socket(new_so, &new_ep->com); 1018 1019 ret = process_mpa_request(new_ep); 1020 if (ret) { 1021 /* ABORT */ 1022 c4iw_ep_disconnect(new_ep, 1, GFP_KERNEL); 1023 c4iw_put_ep(&real_lep->com); 1024 } 1025 mutex_unlock(&new_ep->com.mutex); 1026 c4iw_put_ep(&new_ep->com); 1027 return; 1028} 1029 1030static int 1031add_ep_to_req_list(struct c4iw_ep *ep, int new_ep_event) 1032{ 1033 unsigned long flag; 1034 1035 spin_lock_irqsave(&req_lock, flag); 1036 if (ep && ep->com.so) { 1037 ep->com.ep_events |= new_ep_event; 1038 if (!ep->com.entry.tqe_prev) { 1039 c4iw_get_ep(&ep->com); 1040 TAILQ_INSERT_TAIL(&req_list, &ep->com, entry); 1041 queue_work(c4iw_taskq, &c4iw_task); 1042 } 1043 } 1044 spin_unlock_irqrestore(&req_lock, flag); 1045 1046 return (0); 1047} 1048 1049static int 1050c4iw_so_upcall(struct socket *so, void *arg, int waitflag) 1051{ 1052 struct c4iw_ep *ep = arg; 1053 1054 CTR6(KTR_IW_CXGBE, 1055 "%s: so %p, so_state 0x%x, ep %p, ep_state %s, tqe_prev %p", 1056 __func__, so, so->so_state, ep, states[ep->com.state], 1057 ep->com.entry.tqe_prev); 1058 1059 MPASS(ep->com.so == so); 1060 /* 1061 * Wake up any threads waiting in rdma_init()/rdma_fini(), 1062 * with locks held. 1063 */ 1064 if (so->so_error) 1065 c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET); 1066 add_ep_to_req_list(ep, C4IW_EVENT_SOCKET); 1067 1068 return (SU_OK); 1069} 1070 1071 1072static int 1073terminate(struct sge_iq *iq, const struct rss_header *rss, struct mbuf *m) 1074{ 1075 struct adapter *sc = iq->adapter; 1076 const struct cpl_rdma_terminate *cpl = mtod(m, const void *); 1077 unsigned int tid = GET_TID(cpl); 1078 struct toepcb *toep = lookup_tid(sc, tid); 1079 struct socket *so; 1080 struct c4iw_ep *ep; 1081 1082 INP_WLOCK(toep->inp); 1083 so = inp_inpcbtosocket(toep->inp); 1084 ep = so->so_rcv.sb_upcallarg; 1085 INP_WUNLOCK(toep->inp); 1086 1087 CTR3(KTR_IW_CXGBE, "%s: so %p, ep %p", __func__, so, ep); 1088 add_ep_to_req_list(ep, C4IW_EVENT_TERM); 1089 1090 return 0; 1091} 1092 1093static struct socket * 1094dequeue_socket(struct socket *head) 1095{ 1096 struct socket *so; 1097 struct sockaddr_in *remote; 1098 1099 ACCEPT_LOCK(); 1100 so = TAILQ_FIRST(&head->so_comp); 1101 if (!so) { 1102 ACCEPT_UNLOCK(); 1103 return NULL; 1104 } 1105 1106 SOCK_LOCK(so); 1107 /* 1108 * Before changing the flags on the socket, we have to bump the 1109 * reference count. Otherwise, if the protocol calls sofree(), 1110 * the socket will be released due to a zero refcount. 1111 */ 1112 soref(so); 1113 TAILQ_REMOVE(&head->so_comp, so, so_list); 1114 head->so_qlen--; 1115 so->so_qstate &= ~SQ_COMP; 1116 so->so_head = NULL; 1117 so->so_state |= SS_NBIO; 1118 SOCK_UNLOCK(so); 1119 ACCEPT_UNLOCK(); 1120 remote = NULL; 1121 soaccept(so, (struct sockaddr **)&remote); 1122 1123 free(remote, M_SONAME); 1124 return so; 1125} 1126 1127static void 1128process_socket_event(struct c4iw_ep *ep) 1129{ 1130 int state = ep->com.state; 1131 struct socket *so = ep->com.so; 1132 1133 if (ep->com.state == DEAD) { 1134 CTR3(KTR_IW_CXGBE, "%s: Pending socket event discarded " 1135 "ep %p ep_state %s", __func__, ep, states[state]); 1136 return; 1137 } 1138 1139 CTR6(KTR_IW_CXGBE, "process_socket_event: so %p, so_state 0x%x, " 1140 "so_err %d, sb_state 0x%x, ep %p, ep_state %s", so, so->so_state, 1141 so->so_error, so->so_rcv.sb_state, ep, states[state]); 1142 1143 if (state == CONNECTING) { 1144 process_connected(ep); 1145 return; 1146 } 1147 1148 if (state == LISTEN) { 1149 struct c4iw_listen_ep *lep = (struct c4iw_listen_ep *)ep; 1150 struct socket *new_so; 1151 1152 while ((new_so = dequeue_socket(so)) != NULL) { 1153 process_newconn(lep, new_so); 1154 } 1155 return; 1156 } 1157 1158 /* connection error */ 1159 if (so->so_error) { 1160 process_conn_error(ep); 1161 return; 1162 } 1163 1164 /* peer close */ 1165 if ((so->so_rcv.sb_state & SBS_CANTRCVMORE) && state <= CLOSING) { 1166 process_peer_close(ep); 1167 /* 1168 * check whether socket disconnect event is pending before 1169 * returning. Fallthrough if yes. 1170 */ 1171 if (!(so->so_state & SS_ISDISCONNECTED)) 1172 return; 1173 } 1174 1175 /* close complete */ 1176 if (so->so_state & SS_ISDISCONNECTED) { 1177 process_close_complete(ep); 1178 return; 1179 } 1180 1181 /* rx data */ 1182 process_data(ep); 1183} 1184 1185SYSCTL_NODE(_hw, OID_AUTO, iw_cxgbe, CTLFLAG_RD, 0, "iw_cxgbe driver parameters"); 1186 1187static int dack_mode = 0; 1188SYSCTL_INT(_hw_iw_cxgbe, OID_AUTO, dack_mode, CTLFLAG_RWTUN, &dack_mode, 0, 1189 "Delayed ack mode (default = 0)"); 1190 1191int c4iw_max_read_depth = 8; 1192SYSCTL_INT(_hw_iw_cxgbe, OID_AUTO, c4iw_max_read_depth, CTLFLAG_RWTUN, &c4iw_max_read_depth, 0, 1193 "Per-connection max ORD/IRD (default = 8)"); 1194 1195static int enable_tcp_timestamps; 1196SYSCTL_INT(_hw_iw_cxgbe, OID_AUTO, enable_tcp_timestamps, CTLFLAG_RWTUN, &enable_tcp_timestamps, 0, 1197 "Enable tcp timestamps (default = 0)"); 1198 1199static int enable_tcp_sack; 1200SYSCTL_INT(_hw_iw_cxgbe, OID_AUTO, enable_tcp_sack, CTLFLAG_RWTUN, &enable_tcp_sack, 0, 1201 "Enable tcp SACK (default = 0)"); 1202 1203static int enable_tcp_window_scaling = 1; 1204SYSCTL_INT(_hw_iw_cxgbe, OID_AUTO, enable_tcp_window_scaling, CTLFLAG_RWTUN, &enable_tcp_window_scaling, 0, 1205 "Enable tcp window scaling (default = 1)"); 1206 1207int c4iw_debug = 1; 1208SYSCTL_INT(_hw_iw_cxgbe, OID_AUTO, c4iw_debug, CTLFLAG_RWTUN, &c4iw_debug, 0, 1209 "Enable debug logging (default = 0)"); 1210 1211static int peer2peer = 1; 1212SYSCTL_INT(_hw_iw_cxgbe, OID_AUTO, peer2peer, CTLFLAG_RWTUN, &peer2peer, 0, 1213 "Support peer2peer ULPs (default = 1)"); 1214 1215static int p2p_type = FW_RI_INIT_P2PTYPE_READ_REQ; 1216SYSCTL_INT(_hw_iw_cxgbe, OID_AUTO, p2p_type, CTLFLAG_RWTUN, &p2p_type, 0, 1217 "RDMAP opcode to use for the RTR message: 1 = RDMA_READ 0 = RDMA_WRITE (default 1)"); 1218 1219static int ep_timeout_secs = 60; 1220SYSCTL_INT(_hw_iw_cxgbe, OID_AUTO, ep_timeout_secs, CTLFLAG_RWTUN, &ep_timeout_secs, 0, 1221 "CM Endpoint operation timeout in seconds (default = 60)"); 1222 1223static int mpa_rev = 1; 1224SYSCTL_INT(_hw_iw_cxgbe, OID_AUTO, mpa_rev, CTLFLAG_RWTUN, &mpa_rev, 0, 1225 "MPA Revision, 0 supports amso1100, 1 is RFC5044 spec compliant, 2 is IETF MPA Peer Connect Draft compliant (default = 1)"); 1226 1227static int markers_enabled; 1228SYSCTL_INT(_hw_iw_cxgbe, OID_AUTO, markers_enabled, CTLFLAG_RWTUN, &markers_enabled, 0, 1229 "Enable MPA MARKERS (default(0) = disabled)"); 1230 1231static int crc_enabled = 1; 1232SYSCTL_INT(_hw_iw_cxgbe, OID_AUTO, crc_enabled, CTLFLAG_RWTUN, &crc_enabled, 0, 1233 "Enable MPA CRC (default(1) = enabled)"); 1234 1235static int rcv_win = 256 * 1024; 1236SYSCTL_INT(_hw_iw_cxgbe, OID_AUTO, rcv_win, CTLFLAG_RWTUN, &rcv_win, 0, 1237 "TCP receive window in bytes (default = 256KB)"); 1238 1239static int snd_win = 128 * 1024; 1240SYSCTL_INT(_hw_iw_cxgbe, OID_AUTO, snd_win, CTLFLAG_RWTUN, &snd_win, 0, 1241 "TCP send window in bytes (default = 128KB)"); 1242 1243static void 1244start_ep_timer(struct c4iw_ep *ep) 1245{ 1246 1247 if (timer_pending(&ep->timer)) { 1248 CTR2(KTR_IW_CXGBE, "%s: ep %p, already started", __func__, ep); 1249 printk(KERN_ERR "%s timer already started! ep %p\n", __func__, 1250 ep); 1251 return; 1252 } 1253 clear_bit(TIMEOUT, &ep->com.flags); 1254 c4iw_get_ep(&ep->com); 1255 ep->timer.expires = jiffies + ep_timeout_secs * HZ; 1256 ep->timer.data = (unsigned long)ep; 1257 ep->timer.function = ep_timeout; 1258 add_timer(&ep->timer); 1259} 1260 1261static int 1262stop_ep_timer(struct c4iw_ep *ep) 1263{ 1264 1265 del_timer_sync(&ep->timer); 1266 if (!test_and_set_bit(TIMEOUT, &ep->com.flags)) { 1267 c4iw_put_ep(&ep->com); 1268 return 0; 1269 } 1270 return 1; 1271} 1272 1273static void * 1274alloc_ep(int size, gfp_t gfp) 1275{ 1276 struct c4iw_ep_common *epc; 1277 1278 epc = kzalloc(size, gfp); 1279 if (epc == NULL) 1280 return (NULL); 1281 1282 kref_init(&epc->kref); 1283 mutex_init(&epc->mutex); 1284 c4iw_init_wr_wait(&epc->wr_wait); 1285 1286 return (epc); 1287} 1288 1289void _c4iw_free_ep(struct kref *kref) 1290{ 1291 struct c4iw_ep *ep; 1292 struct c4iw_ep_common *epc; 1293 1294 ep = container_of(kref, struct c4iw_ep, com.kref); 1295 epc = &ep->com; 1296 KASSERT(!epc->entry.tqe_prev, ("%s epc %p still on req list", 1297 __func__, epc)); 1298 if (test_bit(QP_REFERENCED, &ep->com.flags)) 1299 deref_qp(ep); 1300 CTR4(KTR_IW_CXGBE, "%s: ep %p, history 0x%lx, flags 0x%lx", 1301 __func__, ep, epc->history, epc->flags); 1302 kfree(ep); 1303} 1304 1305static void release_ep_resources(struct c4iw_ep *ep) 1306{ 1307 CTR2(KTR_IW_CXGBE, "%s:rerB %p", __func__, ep); 1308 set_bit(RELEASE_RESOURCES, &ep->com.flags); 1309 c4iw_put_ep(&ep->com); 1310 CTR2(KTR_IW_CXGBE, "%s:rerE %p", __func__, ep); 1311} 1312 1313static int 1314send_mpa_req(struct c4iw_ep *ep) 1315{ 1316 int mpalen; 1317 struct mpa_message *mpa; 1318 struct mpa_v2_conn_params mpa_v2_params; 1319 struct mbuf *m; 1320 char mpa_rev_to_use = mpa_rev; 1321 int err = 0; 1322 1323 if (ep->retry_with_mpa_v1) 1324 mpa_rev_to_use = 1; 1325 mpalen = sizeof(*mpa) + ep->plen; 1326 if (mpa_rev_to_use == 2) 1327 mpalen += sizeof(struct mpa_v2_conn_params); 1328 1329 mpa = malloc(mpalen, M_CXGBE, M_NOWAIT); 1330 if (mpa == NULL) { 1331 err = -ENOMEM; 1332 CTR3(KTR_IW_CXGBE, "%s:smr1 ep: %p , error: %d", 1333 __func__, ep, err); 1334 goto err; 1335 } 1336 1337 memset(mpa, 0, mpalen); 1338 memcpy(mpa->key, MPA_KEY_REQ, sizeof(mpa->key)); 1339 mpa->flags = (crc_enabled ? MPA_CRC : 0) | 1340 (markers_enabled ? MPA_MARKERS : 0) | 1341 (mpa_rev_to_use == 2 ? MPA_ENHANCED_RDMA_CONN : 0); 1342 mpa->private_data_size = htons(ep->plen); 1343 mpa->revision = mpa_rev_to_use; 1344 1345 if (mpa_rev_to_use == 1) { 1346 ep->tried_with_mpa_v1 = 1; 1347 ep->retry_with_mpa_v1 = 0; 1348 } 1349 1350 if (mpa_rev_to_use == 2) { 1351 mpa->private_data_size = htons(ntohs(mpa->private_data_size) + 1352 sizeof(struct mpa_v2_conn_params)); 1353 mpa_v2_params.ird = htons((u16)ep->ird); 1354 mpa_v2_params.ord = htons((u16)ep->ord); 1355 1356 if (peer2peer) { 1357 mpa_v2_params.ird |= htons(MPA_V2_PEER2PEER_MODEL); 1358 1359 if (p2p_type == FW_RI_INIT_P2PTYPE_RDMA_WRITE) { 1360 mpa_v2_params.ord |= 1361 htons(MPA_V2_RDMA_WRITE_RTR); 1362 } else if (p2p_type == FW_RI_INIT_P2PTYPE_READ_REQ) { 1363 mpa_v2_params.ord |= 1364 htons(MPA_V2_RDMA_READ_RTR); 1365 } 1366 } 1367 memcpy(mpa->private_data, &mpa_v2_params, 1368 sizeof(struct mpa_v2_conn_params)); 1369 1370 if (ep->plen) { 1371 1372 memcpy(mpa->private_data + 1373 sizeof(struct mpa_v2_conn_params), 1374 ep->mpa_pkt + sizeof(*mpa), ep->plen); 1375 } 1376 } else { 1377 1378 if (ep->plen) 1379 memcpy(mpa->private_data, 1380 ep->mpa_pkt + sizeof(*mpa), ep->plen); 1381 CTR2(KTR_IW_CXGBE, "%s:smr7 %p", __func__, ep); 1382 } 1383 1384 m = m_getm(NULL, mpalen, M_NOWAIT, MT_DATA); 1385 if (m == NULL) { 1386 err = -ENOMEM; 1387 CTR3(KTR_IW_CXGBE, "%s:smr2 ep: %p , error: %d", 1388 __func__, ep, err); 1389 free(mpa, M_CXGBE); 1390 goto err; 1391 } 1392 m_copyback(m, 0, mpalen, (void *)mpa); 1393 free(mpa, M_CXGBE); 1394 1395 err = -sosend(ep->com.so, NULL, NULL, m, NULL, MSG_DONTWAIT, 1396 ep->com.thread); 1397 if (err) { 1398 CTR3(KTR_IW_CXGBE, "%s:smr3 ep: %p , error: %d", 1399 __func__, ep, err); 1400 goto err; 1401 } 1402 1403 START_EP_TIMER(ep); 1404 ep->com.state = MPA_REQ_SENT; 1405 ep->mpa_attr.initiator = 1; 1406 CTR3(KTR_IW_CXGBE, "%s:smrE %p, error: %d", __func__, ep, err); 1407 return 0; 1408err: 1409 connect_reply_upcall(ep, err); 1410 CTR3(KTR_IW_CXGBE, "%s:smrE %p, error: %d", __func__, ep, err); 1411 return err; 1412} 1413 1414static int send_mpa_reject(struct c4iw_ep *ep, const void *pdata, u8 plen) 1415{ 1416 int mpalen ; 1417 struct mpa_message *mpa; 1418 struct mpa_v2_conn_params mpa_v2_params; 1419 struct mbuf *m; 1420 int err; 1421 1422 CTR4(KTR_IW_CXGBE, "%s:smrejB %p %u %d", __func__, ep, ep->hwtid, 1423 ep->plen); 1424 1425 mpalen = sizeof(*mpa) + plen; 1426 1427 if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) { 1428 1429 mpalen += sizeof(struct mpa_v2_conn_params); 1430 CTR4(KTR_IW_CXGBE, "%s:smrej1 %p %u %d", __func__, ep, 1431 ep->mpa_attr.version, mpalen); 1432 } 1433 1434 mpa = malloc(mpalen, M_CXGBE, M_NOWAIT); 1435 if (mpa == NULL) 1436 return (-ENOMEM); 1437 1438 memset(mpa, 0, mpalen); 1439 memcpy(mpa->key, MPA_KEY_REP, sizeof(mpa->key)); 1440 mpa->flags = MPA_REJECT; 1441 mpa->revision = mpa_rev; 1442 mpa->private_data_size = htons(plen); 1443 1444 if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) { 1445 1446 mpa->flags |= MPA_ENHANCED_RDMA_CONN; 1447 mpa->private_data_size = htons(ntohs(mpa->private_data_size) + 1448 sizeof(struct mpa_v2_conn_params)); 1449 mpa_v2_params.ird = htons(((u16)ep->ird) | 1450 (peer2peer ? MPA_V2_PEER2PEER_MODEL : 1451 0)); 1452 mpa_v2_params.ord = htons(((u16)ep->ord) | (peer2peer ? 1453 (p2p_type == 1454 FW_RI_INIT_P2PTYPE_RDMA_WRITE ? 1455 MPA_V2_RDMA_WRITE_RTR : p2p_type == 1456 FW_RI_INIT_P2PTYPE_READ_REQ ? 1457 MPA_V2_RDMA_READ_RTR : 0) : 0)); 1458 memcpy(mpa->private_data, &mpa_v2_params, 1459 sizeof(struct mpa_v2_conn_params)); 1460 1461 if (ep->plen) 1462 memcpy(mpa->private_data + 1463 sizeof(struct mpa_v2_conn_params), pdata, plen); 1464 CTR5(KTR_IW_CXGBE, "%s:smrej3 %p %d %d %d", __func__, ep, 1465 mpa_v2_params.ird, mpa_v2_params.ord, ep->plen); 1466 } else 1467 if (plen) 1468 memcpy(mpa->private_data, pdata, plen); 1469 1470 m = m_getm(NULL, mpalen, M_NOWAIT, MT_DATA); 1471 if (m == NULL) { 1472 free(mpa, M_CXGBE); 1473 return (-ENOMEM); 1474 } 1475 m_copyback(m, 0, mpalen, (void *)mpa); 1476 free(mpa, M_CXGBE); 1477 1478 err = -sosend(ep->com.so, NULL, NULL, m, NULL, MSG_DONTWAIT, ep->com.thread); 1479 if (!err) 1480 ep->snd_seq += mpalen; 1481 CTR4(KTR_IW_CXGBE, "%s:smrejE %p %u %d", __func__, ep, ep->hwtid, err); 1482 return err; 1483} 1484 1485static int send_mpa_reply(struct c4iw_ep *ep, const void *pdata, u8 plen) 1486{ 1487 int mpalen; 1488 struct mpa_message *mpa; 1489 struct mbuf *m; 1490 struct mpa_v2_conn_params mpa_v2_params; 1491 int err; 1492 1493 CTR2(KTR_IW_CXGBE, "%s:smrepB %p", __func__, ep); 1494 1495 mpalen = sizeof(*mpa) + plen; 1496 1497 if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) { 1498 1499 CTR3(KTR_IW_CXGBE, "%s:smrep1 %p %d", __func__, ep, 1500 ep->mpa_attr.version); 1501 mpalen += sizeof(struct mpa_v2_conn_params); 1502 } 1503 1504 mpa = malloc(mpalen, M_CXGBE, M_NOWAIT); 1505 if (mpa == NULL) 1506 return (-ENOMEM); 1507 1508 memset(mpa, 0, sizeof(*mpa)); 1509 memcpy(mpa->key, MPA_KEY_REP, sizeof(mpa->key)); 1510 mpa->flags = (ep->mpa_attr.crc_enabled ? MPA_CRC : 0) | 1511 (markers_enabled ? MPA_MARKERS : 0); 1512 mpa->revision = ep->mpa_attr.version; 1513 mpa->private_data_size = htons(plen); 1514 1515 if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) { 1516 1517 mpa->flags |= MPA_ENHANCED_RDMA_CONN; 1518 mpa->private_data_size += 1519 htons(sizeof(struct mpa_v2_conn_params)); 1520 mpa_v2_params.ird = htons((u16)ep->ird); 1521 mpa_v2_params.ord = htons((u16)ep->ord); 1522 CTR5(KTR_IW_CXGBE, "%s:smrep3 %p %d %d %d", __func__, ep, 1523 ep->mpa_attr.version, mpa_v2_params.ird, mpa_v2_params.ord); 1524 1525 if (peer2peer && (ep->mpa_attr.p2p_type != 1526 FW_RI_INIT_P2PTYPE_DISABLED)) { 1527 1528 mpa_v2_params.ird |= htons(MPA_V2_PEER2PEER_MODEL); 1529 1530 if (p2p_type == FW_RI_INIT_P2PTYPE_RDMA_WRITE) { 1531 1532 mpa_v2_params.ord |= 1533 htons(MPA_V2_RDMA_WRITE_RTR); 1534 CTR5(KTR_IW_CXGBE, "%s:smrep4 %p %d %d %d", 1535 __func__, ep, p2p_type, mpa_v2_params.ird, 1536 mpa_v2_params.ord); 1537 } 1538 else if (p2p_type == FW_RI_INIT_P2PTYPE_READ_REQ) { 1539 1540 mpa_v2_params.ord |= 1541 htons(MPA_V2_RDMA_READ_RTR); 1542 CTR5(KTR_IW_CXGBE, "%s:smrep5 %p %d %d %d", 1543 __func__, ep, p2p_type, mpa_v2_params.ird, 1544 mpa_v2_params.ord); 1545 } 1546 } 1547 1548 memcpy(mpa->private_data, &mpa_v2_params, 1549 sizeof(struct mpa_v2_conn_params)); 1550 1551 if (ep->plen) 1552 memcpy(mpa->private_data + 1553 sizeof(struct mpa_v2_conn_params), pdata, plen); 1554 } else 1555 if (plen) 1556 memcpy(mpa->private_data, pdata, plen); 1557 1558 m = m_getm(NULL, mpalen, M_NOWAIT, MT_DATA); 1559 if (m == NULL) { 1560 free(mpa, M_CXGBE); 1561 return (-ENOMEM); 1562 } 1563 m_copyback(m, 0, mpalen, (void *)mpa); 1564 free(mpa, M_CXGBE); 1565 1566 1567 ep->com.state = MPA_REP_SENT; 1568 ep->snd_seq += mpalen; 1569 err = -sosend(ep->com.so, NULL, NULL, m, NULL, MSG_DONTWAIT, 1570 ep->com.thread); 1571 CTR3(KTR_IW_CXGBE, "%s:smrepE %p %d", __func__, ep, err); 1572 return err; 1573} 1574 1575 1576 1577static void close_complete_upcall(struct c4iw_ep *ep, int status) 1578{ 1579 struct iw_cm_event event; 1580 1581 CTR2(KTR_IW_CXGBE, "%s:ccuB %p", __func__, ep); 1582 memset(&event, 0, sizeof(event)); 1583 event.event = IW_CM_EVENT_CLOSE; 1584 event.status = status; 1585 1586 if (ep->com.cm_id) { 1587 1588 CTR2(KTR_IW_CXGBE, "%s:ccu1 %1", __func__, ep); 1589 ep->com.cm_id->event_handler(ep->com.cm_id, &event); 1590 deref_cm_id(&ep->com); 1591 set_bit(CLOSE_UPCALL, &ep->com.history); 1592 } 1593 CTR2(KTR_IW_CXGBE, "%s:ccuE %p", __func__, ep); 1594} 1595 1596static int 1597send_abort(struct c4iw_ep *ep) 1598{ 1599 struct socket *so = ep->com.so; 1600 struct sockopt sopt; 1601 int rc; 1602 struct linger l; 1603 1604 CTR5(KTR_IW_CXGBE, "%s ep %p so %p state %s tid %d", __func__, ep, so, 1605 states[ep->com.state], ep->hwtid); 1606 1607 l.l_onoff = 1; 1608 l.l_linger = 0; 1609 1610 /* linger_time of 0 forces RST to be sent */ 1611 sopt.sopt_dir = SOPT_SET; 1612 sopt.sopt_level = SOL_SOCKET; 1613 sopt.sopt_name = SO_LINGER; 1614 sopt.sopt_val = (caddr_t)&l; 1615 sopt.sopt_valsize = sizeof l; 1616 sopt.sopt_td = NULL; 1617 rc = sosetopt(so, &sopt); 1618 if (rc != 0) { 1619 log(LOG_ERR, "%s: sosetopt(%p, linger = 0) failed with %d.\n", 1620 __func__, so, rc); 1621 } 1622 1623 uninit_iwarp_socket(so); 1624 soclose(so); 1625 set_bit(ABORT_CONN, &ep->com.history); 1626 1627 /* 1628 * TBD: iw_cxgbe driver should receive ABORT reply for every ABORT 1629 * request it has sent. But the current TOE driver is not propagating 1630 * this ABORT reply event (via do_abort_rpl) to iw_cxgbe. So as a work- 1631 * around de-refererece 'ep' here instead of doing it in abort_rpl() 1632 * handler(not yet implemented) of iw_cxgbe driver. 1633 */ 1634 release_ep_resources(ep); 1635 1636 return (0); 1637} 1638 1639static void peer_close_upcall(struct c4iw_ep *ep) 1640{ 1641 struct iw_cm_event event; 1642 1643 CTR2(KTR_IW_CXGBE, "%s:pcuB %p", __func__, ep); 1644 memset(&event, 0, sizeof(event)); 1645 event.event = IW_CM_EVENT_DISCONNECT; 1646 1647 if (ep->com.cm_id) { 1648 1649 CTR2(KTR_IW_CXGBE, "%s:pcu1 %p", __func__, ep); 1650 ep->com.cm_id->event_handler(ep->com.cm_id, &event); 1651 set_bit(DISCONN_UPCALL, &ep->com.history); 1652 } 1653 CTR2(KTR_IW_CXGBE, "%s:pcuE %p", __func__, ep); 1654} 1655 1656static void peer_abort_upcall(struct c4iw_ep *ep) 1657{ 1658 struct iw_cm_event event; 1659 1660 CTR2(KTR_IW_CXGBE, "%s:pauB %p", __func__, ep); 1661 memset(&event, 0, sizeof(event)); 1662 event.event = IW_CM_EVENT_CLOSE; 1663 event.status = -ECONNRESET; 1664 1665 if (ep->com.cm_id) { 1666 1667 CTR2(KTR_IW_CXGBE, "%s:pau1 %p", __func__, ep); 1668 ep->com.cm_id->event_handler(ep->com.cm_id, &event); 1669 deref_cm_id(&ep->com); 1670 set_bit(ABORT_UPCALL, &ep->com.history); 1671 } 1672 CTR2(KTR_IW_CXGBE, "%s:pauE %p", __func__, ep); 1673} 1674 1675static void connect_reply_upcall(struct c4iw_ep *ep, int status) 1676{ 1677 struct iw_cm_event event; 1678 1679 CTR3(KTR_IW_CXGBE, "%s:cruB %p, status: %d", __func__, ep, status); 1680 memset(&event, 0, sizeof(event)); 1681 event.event = IW_CM_EVENT_CONNECT_REPLY; 1682 event.status = ((status == -ECONNABORTED) || (status == -EPIPE)) ? 1683 -ECONNRESET : status; 1684 event.local_addr = ep->com.local_addr; 1685 event.remote_addr = ep->com.remote_addr; 1686 1687 if ((status == 0) || (status == -ECONNREFUSED)) { 1688 1689 if (!ep->tried_with_mpa_v1) { 1690 1691 CTR2(KTR_IW_CXGBE, "%s:cru1 %p", __func__, ep); 1692 /* this means MPA_v2 is used */ 1693 event.ord = ep->ird; 1694 event.ird = ep->ord; 1695 event.private_data_len = ep->plen - 1696 sizeof(struct mpa_v2_conn_params); 1697 event.private_data = ep->mpa_pkt + 1698 sizeof(struct mpa_message) + 1699 sizeof(struct mpa_v2_conn_params); 1700 } else { 1701 1702 CTR2(KTR_IW_CXGBE, "%s:cru2 %p", __func__, ep); 1703 /* this means MPA_v1 is used */ 1704 event.ord = c4iw_max_read_depth; 1705 event.ird = c4iw_max_read_depth; 1706 event.private_data_len = ep->plen; 1707 event.private_data = ep->mpa_pkt + 1708 sizeof(struct mpa_message); 1709 } 1710 } 1711 1712 if (ep->com.cm_id) { 1713 1714 CTR2(KTR_IW_CXGBE, "%s:cru3 %p", __func__, ep); 1715 set_bit(CONN_RPL_UPCALL, &ep->com.history); 1716 ep->com.cm_id->event_handler(ep->com.cm_id, &event); 1717 } 1718 1719 if(status == -ECONNABORTED) { 1720 1721 CTR3(KTR_IW_CXGBE, "%s:cruE %p %d", __func__, ep, status); 1722 return; 1723 } 1724 1725 if (status < 0) { 1726 1727 CTR3(KTR_IW_CXGBE, "%s:cru4 %p %d", __func__, ep, status); 1728 deref_cm_id(&ep->com); 1729 } 1730 1731 CTR2(KTR_IW_CXGBE, "%s:cruE %p", __func__, ep); 1732} 1733 1734static int connect_request_upcall(struct c4iw_ep *ep) 1735{ 1736 struct iw_cm_event event; 1737 int ret; 1738 1739 CTR3(KTR_IW_CXGBE, "%s: ep %p, mpa_v1 %d", __func__, ep, 1740 ep->tried_with_mpa_v1); 1741 1742 memset(&event, 0, sizeof(event)); 1743 event.event = IW_CM_EVENT_CONNECT_REQUEST; 1744 event.local_addr = ep->com.local_addr; 1745 event.remote_addr = ep->com.remote_addr; 1746 event.provider_data = ep; 1747 1748 if (!ep->tried_with_mpa_v1) { 1749 /* this means MPA_v2 is used */ 1750 event.ord = ep->ord; 1751 event.ird = ep->ird; 1752 event.private_data_len = ep->plen - 1753 sizeof(struct mpa_v2_conn_params); 1754 event.private_data = ep->mpa_pkt + sizeof(struct mpa_message) + 1755 sizeof(struct mpa_v2_conn_params); 1756 } else { 1757 1758 /* this means MPA_v1 is used. Send max supported */ 1759 event.ord = c4iw_max_read_depth; 1760 event.ird = c4iw_max_read_depth; 1761 event.private_data_len = ep->plen; 1762 event.private_data = ep->mpa_pkt + sizeof(struct mpa_message); 1763 } 1764 1765 c4iw_get_ep(&ep->com); 1766 ret = ep->parent_ep->com.cm_id->event_handler(ep->parent_ep->com.cm_id, 1767 &event); 1768 if(ret) { 1769 CTR3(KTR_IW_CXGBE, "%s: ep %p, Failure while notifying event to" 1770 " IWCM, err:%d", __func__, ep, ret); 1771 c4iw_put_ep(&ep->com); 1772 } else 1773 /* Dereference parent_ep only in success case. 1774 * In case of failure, parent_ep is dereferenced by the caller 1775 * of process_mpa_request(). 1776 */ 1777 c4iw_put_ep(&ep->parent_ep->com); 1778 1779 set_bit(CONNREQ_UPCALL, &ep->com.history); 1780 return ret; 1781} 1782 1783static void established_upcall(struct c4iw_ep *ep) 1784{ 1785 struct iw_cm_event event; 1786 1787 CTR2(KTR_IW_CXGBE, "%s:euB %p", __func__, ep); 1788 memset(&event, 0, sizeof(event)); 1789 event.event = IW_CM_EVENT_ESTABLISHED; 1790 event.ird = ep->ord; 1791 event.ord = ep->ird; 1792 1793 if (ep->com.cm_id) { 1794 1795 CTR2(KTR_IW_CXGBE, "%s:eu1 %p", __func__, ep); 1796 ep->com.cm_id->event_handler(ep->com.cm_id, &event); 1797 set_bit(ESTAB_UPCALL, &ep->com.history); 1798 } 1799 CTR2(KTR_IW_CXGBE, "%s:euE %p", __func__, ep); 1800} 1801 1802 1803#define RELAXED_IRD_NEGOTIATION 1 1804 1805/* 1806 * process_mpa_reply - process streaming mode MPA reply 1807 * 1808 * Returns: 1809 * 1810 * 0 upon success indicating a connect request was delivered to the ULP 1811 * or the mpa request is incomplete but valid so far. 1812 * 1813 * 1 if a failure requires the caller to close the connection. 1814 * 1815 * 2 if a failure requires the caller to abort the connection. 1816 */ 1817static int process_mpa_reply(struct c4iw_ep *ep) 1818{ 1819 struct mpa_message *mpa; 1820 struct mpa_v2_conn_params *mpa_v2_params; 1821 u16 plen; 1822 u16 resp_ird, resp_ord; 1823 u8 rtr_mismatch = 0, insuff_ird = 0; 1824 struct c4iw_qp_attributes attrs = {0}; 1825 enum c4iw_qp_attr_mask mask; 1826 int err; 1827 struct mbuf *top, *m; 1828 int flags = MSG_DONTWAIT; 1829 struct uio uio; 1830 int disconnect = 0; 1831 1832 CTR2(KTR_IW_CXGBE, "%s:pmrB %p", __func__, ep); 1833 1834 /* 1835 * Stop mpa timer. If it expired, then 1836 * we ignore the MPA reply. process_timeout() 1837 * will abort the connection. 1838 */ 1839 if (STOP_EP_TIMER(ep)) 1840 return 0; 1841 1842 uio.uio_resid = 1000000; 1843 uio.uio_td = ep->com.thread; 1844 err = soreceive(ep->com.so, NULL, &uio, &top, NULL, &flags); 1845 1846 if (err) { 1847 1848 if (err == EWOULDBLOCK) { 1849 1850 CTR2(KTR_IW_CXGBE, "%s:pmr1 %p", __func__, ep); 1851 START_EP_TIMER(ep); 1852 return 0; 1853 } 1854 err = -err; 1855 CTR2(KTR_IW_CXGBE, "%s:pmr2 %p", __func__, ep); 1856 goto err; 1857 } 1858 1859 if (ep->com.so->so_rcv.sb_mb) { 1860 1861 CTR2(KTR_IW_CXGBE, "%s:pmr3 %p", __func__, ep); 1862 printf("%s data after soreceive called! so %p sb_mb %p top %p\n", 1863 __func__, ep->com.so, ep->com.so->so_rcv.sb_mb, top); 1864 } 1865 1866 m = top; 1867 1868 do { 1869 1870 CTR2(KTR_IW_CXGBE, "%s:pmr4 %p", __func__, ep); 1871 /* 1872 * If we get more than the supported amount of private data 1873 * then we must fail this connection. 1874 */ 1875 if (ep->mpa_pkt_len + m->m_len > sizeof(ep->mpa_pkt)) { 1876 1877 CTR3(KTR_IW_CXGBE, "%s:pmr5 %p %d", __func__, ep, 1878 ep->mpa_pkt_len + m->m_len); 1879 err = (-EINVAL); 1880 goto err_stop_timer; 1881 } 1882 1883 /* 1884 * copy the new data into our accumulation buffer. 1885 */ 1886 m_copydata(m, 0, m->m_len, &(ep->mpa_pkt[ep->mpa_pkt_len])); 1887 ep->mpa_pkt_len += m->m_len; 1888 if (!m->m_next) 1889 m = m->m_nextpkt; 1890 else 1891 m = m->m_next; 1892 } while (m); 1893 1894 m_freem(top); 1895 /* 1896 * if we don't even have the mpa message, then bail. 1897 */ 1898 if (ep->mpa_pkt_len < sizeof(*mpa)) { 1899 return 0; 1900 } 1901 mpa = (struct mpa_message *) ep->mpa_pkt; 1902 1903 /* Validate MPA header. */ 1904 if (mpa->revision > mpa_rev) { 1905 1906 CTR4(KTR_IW_CXGBE, "%s:pmr6 %p %d %d", __func__, ep, 1907 mpa->revision, mpa_rev); 1908 printk(KERN_ERR MOD "%s MPA version mismatch. Local = %d, " 1909 " Received = %d\n", __func__, mpa_rev, mpa->revision); 1910 err = -EPROTO; 1911 goto err_stop_timer; 1912 } 1913 1914 if (memcmp(mpa->key, MPA_KEY_REP, sizeof(mpa->key))) { 1915 1916 CTR2(KTR_IW_CXGBE, "%s:pmr7 %p", __func__, ep); 1917 err = -EPROTO; 1918 goto err_stop_timer; 1919 } 1920 1921 plen = ntohs(mpa->private_data_size); 1922 1923 /* 1924 * Fail if there's too much private data. 1925 */ 1926 if (plen > MPA_MAX_PRIVATE_DATA) { 1927 1928 CTR2(KTR_IW_CXGBE, "%s:pmr8 %p", __func__, ep); 1929 err = -EPROTO; 1930 goto err_stop_timer; 1931 } 1932 1933 /* 1934 * If plen does not account for pkt size 1935 */ 1936 if (ep->mpa_pkt_len > (sizeof(*mpa) + plen)) { 1937 1938 CTR2(KTR_IW_CXGBE, "%s:pmr9 %p", __func__, ep); 1939 STOP_EP_TIMER(ep); 1940 err = -EPROTO; 1941 goto err_stop_timer; 1942 } 1943 1944 ep->plen = (u8) plen; 1945 1946 /* 1947 * If we don't have all the pdata yet, then bail. 1948 * We'll continue process when more data arrives. 1949 */ 1950 if (ep->mpa_pkt_len < (sizeof(*mpa) + plen)) { 1951 1952 CTR2(KTR_IW_CXGBE, "%s:pmra %p", __func__, ep); 1953 return 0; 1954 } 1955 1956 if (mpa->flags & MPA_REJECT) { 1957 1958 CTR2(KTR_IW_CXGBE, "%s:pmrb %p", __func__, ep); 1959 err = -ECONNREFUSED; 1960 goto err_stop_timer; 1961 } 1962 1963 /* 1964 * If we get here we have accumulated the entire mpa 1965 * start reply message including private data. And 1966 * the MPA header is valid. 1967 */ 1968 ep->com.state = FPDU_MODE; 1969 ep->mpa_attr.crc_enabled = (mpa->flags & MPA_CRC) | crc_enabled ? 1 : 0; 1970 ep->mpa_attr.recv_marker_enabled = markers_enabled; 1971 ep->mpa_attr.xmit_marker_enabled = mpa->flags & MPA_MARKERS ? 1 : 0; 1972 ep->mpa_attr.version = mpa->revision; 1973 ep->mpa_attr.p2p_type = FW_RI_INIT_P2PTYPE_DISABLED; 1974 1975 if (mpa->revision == 2) { 1976 1977 CTR2(KTR_IW_CXGBE, "%s:pmrc %p", __func__, ep); 1978 ep->mpa_attr.enhanced_rdma_conn = 1979 mpa->flags & MPA_ENHANCED_RDMA_CONN ? 1 : 0; 1980 1981 if (ep->mpa_attr.enhanced_rdma_conn) { 1982 1983 CTR2(KTR_IW_CXGBE, "%s:pmrd %p", __func__, ep); 1984 mpa_v2_params = (struct mpa_v2_conn_params *) 1985 (ep->mpa_pkt + sizeof(*mpa)); 1986 resp_ird = ntohs(mpa_v2_params->ird) & 1987 MPA_V2_IRD_ORD_MASK; 1988 resp_ord = ntohs(mpa_v2_params->ord) & 1989 MPA_V2_IRD_ORD_MASK; 1990 1991 /* 1992 * This is a double-check. Ideally, below checks are 1993 * not required since ird/ord stuff has been taken 1994 * care of in c4iw_accept_cr 1995 */ 1996 if (ep->ird < resp_ord) { 1997 if (RELAXED_IRD_NEGOTIATION && resp_ord <= 1998 ep->com.dev->rdev.adap->params.max_ordird_qp) 1999 ep->ird = resp_ord; 2000 else 2001 insuff_ird = 1; 2002 } else if (ep->ird > resp_ord) { 2003 ep->ird = resp_ord; 2004 } 2005 if (ep->ord > resp_ird) { 2006 if (RELAXED_IRD_NEGOTIATION) 2007 ep->ord = resp_ird; 2008 else 2009 insuff_ird = 1; 2010 } 2011 if (insuff_ird) { 2012 err = -ENOMEM; 2013 ep->ird = resp_ord; 2014 ep->ord = resp_ird; 2015 } 2016 2017 if (ntohs(mpa_v2_params->ird) & 2018 MPA_V2_PEER2PEER_MODEL) { 2019 2020 CTR2(KTR_IW_CXGBE, "%s:pmrf %p", __func__, ep); 2021 if (ntohs(mpa_v2_params->ord) & 2022 MPA_V2_RDMA_WRITE_RTR) { 2023 2024 CTR2(KTR_IW_CXGBE, "%s:pmrg %p", __func__, ep); 2025 ep->mpa_attr.p2p_type = 2026 FW_RI_INIT_P2PTYPE_RDMA_WRITE; 2027 } 2028 else if (ntohs(mpa_v2_params->ord) & 2029 MPA_V2_RDMA_READ_RTR) { 2030 2031 CTR2(KTR_IW_CXGBE, "%s:pmrh %p", __func__, ep); 2032 ep->mpa_attr.p2p_type = 2033 FW_RI_INIT_P2PTYPE_READ_REQ; 2034 } 2035 } 2036 } 2037 } else { 2038 2039 CTR2(KTR_IW_CXGBE, "%s:pmri %p", __func__, ep); 2040 2041 if (mpa->revision == 1) { 2042 2043 CTR2(KTR_IW_CXGBE, "%s:pmrj %p", __func__, ep); 2044 2045 if (peer2peer) { 2046 2047 CTR2(KTR_IW_CXGBE, "%s:pmrk %p", __func__, ep); 2048 ep->mpa_attr.p2p_type = p2p_type; 2049 } 2050 } 2051 } 2052 2053 if (set_tcpinfo(ep)) { 2054 2055 CTR2(KTR_IW_CXGBE, "%s:pmrl %p", __func__, ep); 2056 printf("%s set_tcpinfo error\n", __func__); 2057 err = -ECONNRESET; 2058 goto err; 2059 } 2060 2061 CTR6(KTR_IW_CXGBE, "%s - crc_enabled = %d, recv_marker_enabled = %d, " 2062 "xmit_marker_enabled = %d, version = %d p2p_type = %d", __func__, 2063 ep->mpa_attr.crc_enabled, ep->mpa_attr.recv_marker_enabled, 2064 ep->mpa_attr.xmit_marker_enabled, ep->mpa_attr.version, 2065 ep->mpa_attr.p2p_type); 2066 2067 /* 2068 * If responder's RTR does not match with that of initiator, assign 2069 * FW_RI_INIT_P2PTYPE_DISABLED in mpa attributes so that RTR is not 2070 * generated when moving QP to RTS state. 2071 * A TERM message will be sent after QP has moved to RTS state 2072 */ 2073 if ((ep->mpa_attr.version == 2) && peer2peer && 2074 (ep->mpa_attr.p2p_type != p2p_type)) { 2075 2076 CTR2(KTR_IW_CXGBE, "%s:pmrm %p", __func__, ep); 2077 ep->mpa_attr.p2p_type = FW_RI_INIT_P2PTYPE_DISABLED; 2078 rtr_mismatch = 1; 2079 } 2080 2081 2082 //ep->ofld_txq = TOEPCB(ep->com.so)->ofld_txq; 2083 attrs.mpa_attr = ep->mpa_attr; 2084 attrs.max_ird = ep->ird; 2085 attrs.max_ord = ep->ord; 2086 attrs.llp_stream_handle = ep; 2087 attrs.next_state = C4IW_QP_STATE_RTS; 2088 2089 mask = C4IW_QP_ATTR_NEXT_STATE | 2090 C4IW_QP_ATTR_LLP_STREAM_HANDLE | C4IW_QP_ATTR_MPA_ATTR | 2091 C4IW_QP_ATTR_MAX_IRD | C4IW_QP_ATTR_MAX_ORD; 2092 2093 /* bind QP and TID with INIT_WR */ 2094 err = c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp, mask, &attrs, 1); 2095 2096 if (err) { 2097 2098 CTR2(KTR_IW_CXGBE, "%s:pmrn %p", __func__, ep); 2099 goto err; 2100 } 2101 2102 /* 2103 * If responder's RTR requirement did not match with what initiator 2104 * supports, generate TERM message 2105 */ 2106 if (rtr_mismatch) { 2107 2108 CTR2(KTR_IW_CXGBE, "%s:pmro %p", __func__, ep); 2109 printk(KERN_ERR "%s: RTR mismatch, sending TERM\n", __func__); 2110 attrs.layer_etype = LAYER_MPA | DDP_LLP; 2111 attrs.ecode = MPA_NOMATCH_RTR; 2112 attrs.next_state = C4IW_QP_STATE_TERMINATE; 2113 attrs.send_term = 1; 2114 err = c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp, 2115 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1); 2116 err = -ENOMEM; 2117 disconnect = 1; 2118 goto out; 2119 } 2120 2121 /* 2122 * Generate TERM if initiator IRD is not sufficient for responder 2123 * provided ORD. Currently, we do the same behaviour even when 2124 * responder provided IRD is also not sufficient as regards to 2125 * initiator ORD. 2126 */ 2127 if (insuff_ird) { 2128 2129 CTR2(KTR_IW_CXGBE, "%s:pmrp %p", __func__, ep); 2130 printk(KERN_ERR "%s: Insufficient IRD, sending TERM\n", 2131 __func__); 2132 attrs.layer_etype = LAYER_MPA | DDP_LLP; 2133 attrs.ecode = MPA_INSUFF_IRD; 2134 attrs.next_state = C4IW_QP_STATE_TERMINATE; 2135 attrs.send_term = 1; 2136 err = c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp, 2137 C4IW_QP_ATTR_NEXT_STATE, &attrs, 1); 2138 err = -ENOMEM; 2139 disconnect = 1; 2140 goto out; 2141 } 2142 goto out; 2143err_stop_timer: 2144 STOP_EP_TIMER(ep); 2145err: 2146 disconnect = 2; 2147out: 2148 connect_reply_upcall(ep, err); 2149 CTR2(KTR_IW_CXGBE, "%s:pmrE %p", __func__, ep); 2150 return disconnect; 2151} 2152 2153/* 2154 * process_mpa_request - process streaming mode MPA request 2155 * 2156 * Returns: 2157 * 2158 * 0 upon success indicating a connect request was delivered to the ULP 2159 * or the mpa request is incomplete but valid so far. 2160 * 2161 * 1 if a failure requires the caller to close the connection. 2162 * 2163 * 2 if a failure requires the caller to abort the connection. 2164 */ 2165static int 2166process_mpa_request(struct c4iw_ep *ep) 2167{ 2168 struct mpa_message *mpa; 2169 struct mpa_v2_conn_params *mpa_v2_params; 2170 u16 plen; 2171 int flags = MSG_DONTWAIT; 2172 int rc; 2173 struct iovec iov; 2174 struct uio uio; 2175 enum c4iw_ep_state state = ep->com.state; 2176 2177 CTR3(KTR_IW_CXGBE, "%s: ep %p, state %s", __func__, ep, states[state]); 2178 2179 if (state != MPA_REQ_WAIT) 2180 return 0; 2181 2182 iov.iov_base = &ep->mpa_pkt[ep->mpa_pkt_len]; 2183 iov.iov_len = sizeof(ep->mpa_pkt) - ep->mpa_pkt_len; 2184 uio.uio_iov = &iov; 2185 uio.uio_iovcnt = 1; 2186 uio.uio_offset = 0; 2187 uio.uio_resid = sizeof(ep->mpa_pkt) - ep->mpa_pkt_len; 2188 uio.uio_segflg = UIO_SYSSPACE; 2189 uio.uio_rw = UIO_READ; 2190 uio.uio_td = NULL; /* uio.uio_td = ep->com.thread; */ 2191 2192 rc = soreceive(ep->com.so, NULL, &uio, NULL, NULL, &flags); 2193 if (rc == EAGAIN) 2194 return 0; 2195 else if (rc) 2196 goto err_stop_timer; 2197 2198 KASSERT(uio.uio_offset > 0, ("%s: sorecieve on so %p read no data", 2199 __func__, ep->com.so)); 2200 ep->mpa_pkt_len += uio.uio_offset; 2201 2202 /* 2203 * If we get more than the supported amount of private data then we must 2204 * fail this connection. XXX: check so_rcv->sb_cc, or peek with another 2205 * soreceive, or increase the size of mpa_pkt by 1 and abort if the last 2206 * byte is filled by the soreceive above. 2207 */ 2208 2209 /* Don't even have the MPA message. Wait for more data to arrive. */ 2210 if (ep->mpa_pkt_len < sizeof(*mpa)) 2211 return 0; 2212 mpa = (struct mpa_message *) ep->mpa_pkt; 2213 2214 /* 2215 * Validate MPA Header. 2216 */ 2217 if (mpa->revision > mpa_rev) { 2218 log(LOG_ERR, "%s: MPA version mismatch. Local = %d," 2219 " Received = %d\n", __func__, mpa_rev, mpa->revision); 2220 goto err_stop_timer; 2221 } 2222 2223 if (memcmp(mpa->key, MPA_KEY_REQ, sizeof(mpa->key))) 2224 goto err_stop_timer; 2225 2226 /* 2227 * Fail if there's too much private data. 2228 */ 2229 plen = ntohs(mpa->private_data_size); 2230 if (plen > MPA_MAX_PRIVATE_DATA) 2231 goto err_stop_timer; 2232 2233 /* 2234 * If plen does not account for pkt size 2235 */ 2236 if (ep->mpa_pkt_len > (sizeof(*mpa) + plen)) 2237 goto err_stop_timer; 2238 2239 ep->plen = (u8) plen; 2240 2241 /* 2242 * If we don't have all the pdata yet, then bail. 2243 */ 2244 if (ep->mpa_pkt_len < (sizeof(*mpa) + plen)) 2245 return 0; 2246 2247 /* 2248 * If we get here we have accumulated the entire mpa 2249 * start reply message including private data. 2250 */ 2251 ep->mpa_attr.initiator = 0; 2252 ep->mpa_attr.crc_enabled = (mpa->flags & MPA_CRC) | crc_enabled ? 1 : 0; 2253 ep->mpa_attr.recv_marker_enabled = markers_enabled; 2254 ep->mpa_attr.xmit_marker_enabled = mpa->flags & MPA_MARKERS ? 1 : 0; 2255 ep->mpa_attr.version = mpa->revision; 2256 if (mpa->revision == 1) 2257 ep->tried_with_mpa_v1 = 1; 2258 ep->mpa_attr.p2p_type = FW_RI_INIT_P2PTYPE_DISABLED; 2259 2260 if (mpa->revision == 2) { 2261 ep->mpa_attr.enhanced_rdma_conn = 2262 mpa->flags & MPA_ENHANCED_RDMA_CONN ? 1 : 0; 2263 if (ep->mpa_attr.enhanced_rdma_conn) { 2264 mpa_v2_params = (struct mpa_v2_conn_params *) 2265 (ep->mpa_pkt + sizeof(*mpa)); 2266 ep->ird = ntohs(mpa_v2_params->ird) & 2267 MPA_V2_IRD_ORD_MASK; 2268 ep->ird = min_t(u32, ep->ird, 2269 cur_max_read_depth(ep->com.dev)); 2270 ep->ord = ntohs(mpa_v2_params->ord) & 2271 MPA_V2_IRD_ORD_MASK; 2272 ep->ord = min_t(u32, ep->ord, 2273 cur_max_read_depth(ep->com.dev)); 2274 CTR3(KTR_IW_CXGBE, "%s initiator ird %u ord %u\n", 2275 __func__, ep->ird, ep->ord); 2276 if (ntohs(mpa_v2_params->ird) & MPA_V2_PEER2PEER_MODEL) 2277 if (peer2peer) { 2278 if (ntohs(mpa_v2_params->ord) & 2279 MPA_V2_RDMA_WRITE_RTR) 2280 ep->mpa_attr.p2p_type = 2281 FW_RI_INIT_P2PTYPE_RDMA_WRITE; 2282 else if (ntohs(mpa_v2_params->ord) & 2283 MPA_V2_RDMA_READ_RTR) 2284 ep->mpa_attr.p2p_type = 2285 FW_RI_INIT_P2PTYPE_READ_REQ; 2286 } 2287 } 2288 } else if (mpa->revision == 1 && peer2peer) 2289 ep->mpa_attr.p2p_type = p2p_type; 2290 2291 if (set_tcpinfo(ep)) 2292 goto err_stop_timer; 2293 2294 CTR5(KTR_IW_CXGBE, "%s: crc_enabled = %d, recv_marker_enabled = %d, " 2295 "xmit_marker_enabled = %d, version = %d", __func__, 2296 ep->mpa_attr.crc_enabled, ep->mpa_attr.recv_marker_enabled, 2297 ep->mpa_attr.xmit_marker_enabled, ep->mpa_attr.version); 2298 2299 ep->com.state = MPA_REQ_RCVD; 2300 STOP_EP_TIMER(ep); 2301 2302 /* drive upcall */ 2303 if (ep->parent_ep->com.state != DEAD) 2304 if (connect_request_upcall(ep)) 2305 goto err_out; 2306 return 0; 2307 2308err_stop_timer: 2309 STOP_EP_TIMER(ep); 2310err_out: 2311 return 2; 2312} 2313 2314/* 2315 * Upcall from the adapter indicating data has been transmitted. 2316 * For us its just the single MPA request or reply. We can now free 2317 * the skb holding the mpa message. 2318 */ 2319int c4iw_reject_cr(struct iw_cm_id *cm_id, const void *pdata, u8 pdata_len) 2320{ 2321 int err; 2322 struct c4iw_ep *ep = to_ep(cm_id); 2323 int abort = 0; 2324 2325 mutex_lock(&ep->com.mutex); 2326 CTR2(KTR_IW_CXGBE, "%s:crcB %p", __func__, ep); 2327 2328 if ((ep->com.state == DEAD) || 2329 (ep->com.state != MPA_REQ_RCVD)) { 2330 2331 CTR2(KTR_IW_CXGBE, "%s:crc1 %p", __func__, ep); 2332 mutex_unlock(&ep->com.mutex); 2333 c4iw_put_ep(&ep->com); 2334 return -ECONNRESET; 2335 } 2336 set_bit(ULP_REJECT, &ep->com.history); 2337 2338 if (mpa_rev == 0) { 2339 2340 CTR2(KTR_IW_CXGBE, "%s:crc2 %p", __func__, ep); 2341 abort = 1; 2342 } 2343 else { 2344 2345 CTR2(KTR_IW_CXGBE, "%s:crc3 %p", __func__, ep); 2346 abort = send_mpa_reject(ep, pdata, pdata_len); 2347 } 2348 STOP_EP_TIMER(ep); 2349 err = c4iw_ep_disconnect(ep, abort != 0, GFP_KERNEL); 2350 mutex_unlock(&ep->com.mutex); 2351 c4iw_put_ep(&ep->com); 2352 CTR3(KTR_IW_CXGBE, "%s:crc4 %p, err: %d", __func__, ep, err); 2353 return 0; 2354} 2355 2356int c4iw_accept_cr(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param) 2357{ 2358 int err; 2359 struct c4iw_qp_attributes attrs = {0}; 2360 enum c4iw_qp_attr_mask mask; 2361 struct c4iw_ep *ep = to_ep(cm_id); 2362 struct c4iw_dev *h = to_c4iw_dev(cm_id->device); 2363 struct c4iw_qp *qp = get_qhp(h, conn_param->qpn); 2364 int abort = 0; 2365 2366 mutex_lock(&ep->com.mutex); 2367 CTR2(KTR_IW_CXGBE, "%s:cacB %p", __func__, ep); 2368 2369 if ((ep->com.state == DEAD) || 2370 (ep->com.state != MPA_REQ_RCVD)) { 2371 2372 CTR2(KTR_IW_CXGBE, "%s:cac1 %p", __func__, ep); 2373 err = -ECONNRESET; 2374 goto err_out; 2375 } 2376 2377 BUG_ON(!qp); 2378 2379 set_bit(ULP_ACCEPT, &ep->com.history); 2380 2381 if ((conn_param->ord > c4iw_max_read_depth) || 2382 (conn_param->ird > c4iw_max_read_depth)) { 2383 2384 CTR2(KTR_IW_CXGBE, "%s:cac2 %p", __func__, ep); 2385 err = -EINVAL; 2386 goto err_abort; 2387 } 2388 2389 if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) { 2390 2391 CTR2(KTR_IW_CXGBE, "%s:cac3 %p", __func__, ep); 2392 2393 if (conn_param->ord > ep->ird) { 2394 if (RELAXED_IRD_NEGOTIATION) { 2395 conn_param->ord = ep->ird; 2396 } else { 2397 ep->ird = conn_param->ird; 2398 ep->ord = conn_param->ord; 2399 send_mpa_reject(ep, conn_param->private_data, 2400 conn_param->private_data_len); 2401 err = -ENOMEM; 2402 goto err_abort; 2403 } 2404 } 2405 if (conn_param->ird < ep->ord) { 2406 if (RELAXED_IRD_NEGOTIATION && 2407 ep->ord <= h->rdev.adap->params.max_ordird_qp) { 2408 conn_param->ird = ep->ord; 2409 } else { 2410 err = -ENOMEM; 2411 goto err_abort; 2412 } 2413 } 2414 } 2415 ep->ird = conn_param->ird; 2416 ep->ord = conn_param->ord; 2417 2418 if (ep->mpa_attr.version == 1) { 2419 if (peer2peer && ep->ird == 0) 2420 ep->ird = 1; 2421 } else { 2422 if (peer2peer && 2423 (ep->mpa_attr.p2p_type != FW_RI_INIT_P2PTYPE_DISABLED) && 2424 (p2p_type == FW_RI_INIT_P2PTYPE_READ_REQ) && ep->ird == 0) 2425 ep->ird = 1; 2426 } 2427 2428 CTR4(KTR_IW_CXGBE, "%s %d ird %d ord %d\n", __func__, __LINE__, 2429 ep->ird, ep->ord); 2430 2431 ep->com.cm_id = cm_id; 2432 ref_cm_id(&ep->com); 2433 ep->com.qp = qp; 2434 ref_qp(ep); 2435 //ep->ofld_txq = TOEPCB(ep->com.so)->ofld_txq; 2436 2437 /* bind QP to EP and move to RTS */ 2438 attrs.mpa_attr = ep->mpa_attr; 2439 attrs.max_ird = ep->ird; 2440 attrs.max_ord = ep->ord; 2441 attrs.llp_stream_handle = ep; 2442 attrs.next_state = C4IW_QP_STATE_RTS; 2443 2444 /* bind QP and TID with INIT_WR */ 2445 mask = C4IW_QP_ATTR_NEXT_STATE | 2446 C4IW_QP_ATTR_LLP_STREAM_HANDLE | 2447 C4IW_QP_ATTR_MPA_ATTR | 2448 C4IW_QP_ATTR_MAX_IRD | 2449 C4IW_QP_ATTR_MAX_ORD; 2450 2451 err = c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp, mask, &attrs, 1); 2452 if (err) { 2453 CTR3(KTR_IW_CXGBE, "%s:caca %p, err: %d", __func__, ep, err); 2454 goto err_defef_cm_id; 2455 } 2456 2457 err = send_mpa_reply(ep, conn_param->private_data, 2458 conn_param->private_data_len); 2459 if (err) { 2460 CTR3(KTR_IW_CXGBE, "%s:cacb %p, err: %d", __func__, ep, err); 2461 goto err_defef_cm_id; 2462 } 2463 2464 ep->com.state = FPDU_MODE; 2465 established_upcall(ep); 2466 mutex_unlock(&ep->com.mutex); 2467 c4iw_put_ep(&ep->com); 2468 CTR2(KTR_IW_CXGBE, "%s:cacE %p", __func__, ep); 2469 return 0; 2470err_defef_cm_id: 2471 deref_cm_id(&ep->com); 2472err_abort: 2473 abort = 1; 2474err_out: 2475 if (abort) 2476 c4iw_ep_disconnect(ep, 1, GFP_KERNEL); 2477 mutex_unlock(&ep->com.mutex); 2478 c4iw_put_ep(&ep->com); 2479 CTR2(KTR_IW_CXGBE, "%s:cacE err %p", __func__, ep); 2480 return err; 2481} 2482 2483static int 2484c4iw_sock_create(struct sockaddr_storage *laddr, struct socket **so) 2485{ 2486 int ret; 2487 int size; 2488 struct socket *sock = NULL; 2489 2490 ret = sock_create_kern(laddr->ss_family, 2491 SOCK_STREAM, IPPROTO_TCP, &sock); 2492 if (ret) { 2493 CTR2(KTR_IW_CXGBE, "%s:Failed to create TCP socket. err %d", 2494 __func__, ret); 2495 return ret; 2496 } 2497 2498 ret = sobind(sock, (struct sockaddr *)laddr, curthread); 2499 if (ret) { 2500 CTR2(KTR_IW_CXGBE, "%s:Failed to bind socket. err %p", 2501 __func__, ret); 2502 sock_release(sock); 2503 return ret; 2504 } 2505 2506 size = laddr->ss_family == AF_INET6 ? 2507 sizeof(struct sockaddr_in6) : sizeof(struct sockaddr_in); 2508 ret = sock_getname(sock, (struct sockaddr *)laddr, &size, 0); 2509 if (ret) { 2510 CTR2(KTR_IW_CXGBE, "%s:sock_getname failed. err %p", 2511 __func__, ret); 2512 sock_release(sock); 2513 return ret; 2514 } 2515 2516 *so = sock; 2517 return 0; 2518} 2519 2520int c4iw_connect(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param) 2521{ 2522 int err = 0; 2523 struct c4iw_dev *dev = to_c4iw_dev(cm_id->device); 2524 struct c4iw_ep *ep = NULL; 2525 struct ifnet *nh_ifp; /* Logical egress interface */ 2526 2527 CTR2(KTR_IW_CXGBE, "%s:ccB %p", __func__, cm_id); 2528 2529 2530 if ((conn_param->ord > c4iw_max_read_depth) || 2531 (conn_param->ird > c4iw_max_read_depth)) { 2532 2533 CTR2(KTR_IW_CXGBE, "%s:cc1 %p", __func__, cm_id); 2534 err = -EINVAL; 2535 goto out; 2536 } 2537 ep = alloc_ep(sizeof(*ep), GFP_KERNEL); 2538 2539 init_timer(&ep->timer); 2540 ep->plen = conn_param->private_data_len; 2541 2542 if (ep->plen) { 2543 2544 CTR2(KTR_IW_CXGBE, "%s:cc3 %p", __func__, ep); 2545 memcpy(ep->mpa_pkt + sizeof(struct mpa_message), 2546 conn_param->private_data, ep->plen); 2547 } 2548 ep->ird = conn_param->ird; 2549 ep->ord = conn_param->ord; 2550 2551 if (peer2peer && ep->ord == 0) { 2552 2553 CTR2(KTR_IW_CXGBE, "%s:cc4 %p", __func__, ep); 2554 ep->ord = 1; 2555 } 2556 2557 ep->com.dev = dev; 2558 ep->com.cm_id = cm_id; 2559 ref_cm_id(&ep->com); 2560 ep->com.qp = get_qhp(dev, conn_param->qpn); 2561 2562 if (!ep->com.qp) { 2563 2564 CTR2(KTR_IW_CXGBE, "%s:cc5 %p", __func__, ep); 2565 err = -EINVAL; 2566 goto fail; 2567 } 2568 ref_qp(ep); 2569 ep->com.thread = curthread; 2570 2571 err = get_ifnet_from_raddr(&cm_id->remote_addr, &nh_ifp); 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 sodisconnect(ep->com.so); 2815 } 2816 2817 } 2818 2819 if (fatal) { 2820 set_bit(EP_DISC_FAIL, &ep->com.history); 2821 if (!abrupt) { 2822 STOP_EP_TIMER(ep); 2823 close_complete_upcall(ep, -EIO); 2824 } 2825 if (ep->com.qp) { 2826 struct c4iw_qp_attributes attrs = {0}; 2827 2828 attrs.next_state = C4IW_QP_STATE_ERROR; 2829 ret = c4iw_modify_qp(ep->com.dev, ep->com.qp, 2830 C4IW_QP_ATTR_NEXT_STATE, 2831 &attrs, 1); 2832 if (ret) { 2833 CTR2(KTR_IW_CXGBE, "%s:ced7 %p", __func__, ep); 2834 printf("%s - qp <- error failed!\n", __func__); 2835 } 2836 } 2837 release_ep_resources(ep); 2838 ep->com.state = DEAD; 2839 CTR2(KTR_IW_CXGBE, "%s:ced6 %p", __func__, ep); 2840 } 2841 c4iw_put_ep(&ep->com); 2842 CTR2(KTR_IW_CXGBE, "%s:cedE %p", __func__, ep); 2843 return ret; 2844} 2845 2846#ifdef C4IW_EP_REDIRECT 2847int c4iw_ep_redirect(void *ctx, struct dst_entry *old, struct dst_entry *new, 2848 struct l2t_entry *l2t) 2849{ 2850 struct c4iw_ep *ep = ctx; 2851 2852 if (ep->dst != old) 2853 return 0; 2854 2855 PDBG("%s ep %p redirect to dst %p l2t %p\n", __func__, ep, new, 2856 l2t); 2857 dst_hold(new); 2858 cxgb4_l2t_release(ep->l2t); 2859 ep->l2t = l2t; 2860 dst_release(old); 2861 ep->dst = new; 2862 return 1; 2863} 2864#endif 2865 2866 2867 2868static void ep_timeout(unsigned long arg) 2869{ 2870 struct c4iw_ep *ep = (struct c4iw_ep *)arg; 2871 2872 if (!test_and_set_bit(TIMEOUT, &ep->com.flags)) { 2873 2874 /* 2875 * Only insert if it is not already on the list. 2876 */ 2877 if (!(ep->com.ep_events & C4IW_EVENT_TIMEOUT)) { 2878 CTR2(KTR_IW_CXGBE, "%s:et1 %p", __func__, ep); 2879 add_ep_to_req_list(ep, C4IW_EVENT_TIMEOUT); 2880 } 2881 } 2882} 2883 2884static int fw6_wr_rpl(struct adapter *sc, const __be64 *rpl) 2885{ 2886 uint64_t val = be64toh(*rpl); 2887 int ret; 2888 struct c4iw_wr_wait *wr_waitp; 2889 2890 ret = (int)((val >> 8) & 0xff); 2891 wr_waitp = (struct c4iw_wr_wait *)rpl[1]; 2892 CTR3(KTR_IW_CXGBE, "%s wr_waitp %p ret %u", __func__, wr_waitp, ret); 2893 if (wr_waitp) 2894 c4iw_wake_up(wr_waitp, ret ? -ret : 0); 2895 2896 return (0); 2897} 2898 2899static int fw6_cqe_handler(struct adapter *sc, const __be64 *rpl) 2900{ 2901 struct cqe_list_entry *cle; 2902 unsigned long flag; 2903 2904 cle = malloc(sizeof(*cle), M_CXGBE, M_NOWAIT); 2905 cle->rhp = sc->iwarp_softc; 2906 cle->err_cqe = *(const struct t4_cqe *)(&rpl[0]); 2907 2908 spin_lock_irqsave(&err_cqe_lock, flag); 2909 list_add_tail(&cle->entry, &err_cqe_list); 2910 queue_work(c4iw_taskq, &c4iw_task); 2911 spin_unlock_irqrestore(&err_cqe_lock, flag); 2912 2913 return (0); 2914} 2915 2916static int 2917process_terminate(struct c4iw_ep *ep) 2918{ 2919 struct c4iw_qp_attributes attrs = {0}; 2920 2921 CTR2(KTR_IW_CXGBE, "%s:tB %p %d", __func__, ep); 2922 2923 if (ep && ep->com.qp) { 2924 2925 printk(KERN_WARNING MOD "TERM received tid %u qpid %u\n", 2926 ep->hwtid, ep->com.qp->wq.sq.qid); 2927 attrs.next_state = C4IW_QP_STATE_TERMINATE; 2928 c4iw_modify_qp(ep->com.dev, ep->com.qp, C4IW_QP_ATTR_NEXT_STATE, &attrs, 2929 1); 2930 } else 2931 printk(KERN_WARNING MOD "TERM received tid %u no ep/qp\n", 2932 ep->hwtid); 2933 CTR2(KTR_IW_CXGBE, "%s:tE %p %d", __func__, ep); 2934 2935 return 0; 2936} 2937 2938int __init c4iw_cm_init(void) 2939{ 2940 2941 t4_register_cpl_handler(CPL_RDMA_TERMINATE, terminate); 2942 t4_register_fw_msg_handler(FW6_TYPE_WR_RPL, fw6_wr_rpl); 2943 t4_register_fw_msg_handler(FW6_TYPE_CQE, fw6_cqe_handler); 2944 t4_register_an_handler(c4iw_ev_handler); 2945 2946 TAILQ_INIT(&req_list); 2947 spin_lock_init(&req_lock); 2948 INIT_LIST_HEAD(&err_cqe_list); 2949 spin_lock_init(&err_cqe_lock); 2950 2951 INIT_WORK(&c4iw_task, process_req); 2952 2953 c4iw_taskq = create_singlethread_workqueue("iw_cxgbe"); 2954 if (!c4iw_taskq) 2955 return -ENOMEM; 2956 2957 return 0; 2958} 2959 2960void __exit c4iw_cm_term(void) 2961{ 2962 WARN_ON(!TAILQ_EMPTY(&req_list)); 2963 WARN_ON(!list_empty(&err_cqe_list)); 2964 flush_workqueue(c4iw_taskq); 2965 destroy_workqueue(c4iw_taskq); 2966 2967 t4_register_cpl_handler(CPL_RDMA_TERMINATE, NULL); 2968 t4_register_fw_msg_handler(FW6_TYPE_WR_RPL, NULL); 2969 t4_register_fw_msg_handler(FW6_TYPE_CQE, NULL); 2970 t4_register_an_handler(NULL); 2971} 2972#endif 2973