xref: /freebsd/sys/dev/cxgbe/iw_cxgbe/cm.c (revision 657729a89dd578d8cfc70d6616f5c65a48a8b33a)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
3  *
4  * Copyright (c) 2009-2013, 2016 Chelsio, Inc. All rights reserved.
5  *
6  * This software is available to you under a choice of one of two
7  * licenses.  You may choose to be licensed under the terms of the GNU
8  * General Public License (GPL) Version 2, available from the file
9  * COPYING in the main directory of this source tree, or the
10  * OpenIB.org BSD license below:
11  *
12  *     Redistribution and use in source and binary forms, with or
13  *     without modification, are permitted provided that the following
14  *     conditions are met:
15  *
16  *      - Redistributions of source code must retain the above
17  *	  copyright notice, this list of conditions and the following
18  *	  disclaimer.
19  *
20  *      - Redistributions in binary form must reproduce the above
21  *	  copyright notice, this list of conditions and the following
22  *	  disclaimer in the documentation and/or other materials
23  *	  provided with the distribution.
24  *
25  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
26  * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
27  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
28  * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
29  * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
30  * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
31  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
32  * SOFTWARE.
33  */
34 #include <sys/cdefs.h>
35 __FBSDID("$FreeBSD$");
36 
37 #include "opt_inet.h"
38 
39 #ifdef TCP_OFFLOAD
40 #include <sys/types.h>
41 #include <sys/malloc.h>
42 #include <sys/socket.h>
43 #include <sys/socketvar.h>
44 #include <sys/sockio.h>
45 #include <sys/taskqueue.h>
46 #include <netinet/in.h>
47 #include <net/route.h>
48 #include <net/route/nhop.h>
49 
50 #include <netinet/in_systm.h>
51 #include <netinet/in_pcb.h>
52 #include <netinet6/in6_pcb.h>
53 #include <netinet/ip.h>
54 #include <netinet/in_fib.h>
55 #include <netinet6/in6_fib.h>
56 #include <netinet6/scope6_var.h>
57 #include <netinet/ip_var.h>
58 #include <netinet/tcp_var.h>
59 #include <netinet/tcp.h>
60 #include <netinet/tcpip.h>
61 
62 #include <netinet/toecore.h>
63 
64 struct sge_iq;
65 struct rss_header;
66 struct cpl_set_tcb_rpl;
67 #include <linux/types.h>
68 #include "offload.h"
69 #include "tom/t4_tom.h"
70 
71 #define TOEPCB(so)  ((struct toepcb *)(sototcpcb((so))->t_toe))
72 
73 #include "iw_cxgbe.h"
74 #include <linux/module.h>
75 #include <linux/workqueue.h>
76 #include <linux/if_vlan.h>
77 #include <net/netevent.h>
78 #include <rdma/rdma_cm.h>
79 
80 static spinlock_t req_lock;
81 static TAILQ_HEAD(c4iw_ep_list, c4iw_ep_common) req_list;
82 static struct work_struct c4iw_task;
83 static struct workqueue_struct *c4iw_taskq;
84 static LIST_HEAD(err_cqe_list);
85 static spinlock_t err_cqe_lock;
86 static LIST_HEAD(listen_port_list);
87 static DEFINE_MUTEX(listen_port_mutex);
88 
89 static void process_req(struct work_struct *ctx);
90 static void start_ep_timer(struct c4iw_ep *ep);
91 static int stop_ep_timer(struct c4iw_ep *ep);
92 static int set_tcpinfo(struct c4iw_ep *ep);
93 static void process_timeout(struct c4iw_ep *ep);
94 static void process_err_cqes(void);
95 static void *alloc_ep(int size, gfp_t flags);
96 static void close_socket(struct socket *so);
97 static int send_mpa_req(struct c4iw_ep *ep);
98 static int send_mpa_reject(struct c4iw_ep *ep, const void *pdata, u8 plen);
99 static int send_mpa_reply(struct c4iw_ep *ep, const void *pdata, u8 plen);
100 static void close_complete_upcall(struct c4iw_ep *ep, int status);
101 static int send_abort(struct c4iw_ep *ep);
102 static void peer_close_upcall(struct c4iw_ep *ep);
103 static void peer_abort_upcall(struct c4iw_ep *ep);
104 static void connect_reply_upcall(struct c4iw_ep *ep, int status);
105 static int connect_request_upcall(struct c4iw_ep *ep);
106 static void established_upcall(struct c4iw_ep *ep);
107 static int process_mpa_reply(struct c4iw_ep *ep);
108 static int process_mpa_request(struct c4iw_ep *ep);
109 static void process_peer_close(struct c4iw_ep *ep);
110 static void process_conn_error(struct c4iw_ep *ep);
111 static void process_close_complete(struct c4iw_ep *ep);
112 static void ep_timeout(unsigned long arg);
113 static void setiwsockopt(struct socket *so);
114 static void init_iwarp_socket(struct socket *so, void *arg);
115 static void uninit_iwarp_socket(struct socket *so);
116 static void process_data(struct c4iw_ep *ep);
117 static void process_connected(struct c4iw_ep *ep);
118 static int c4iw_so_upcall(struct socket *so, void *arg, int waitflag);
119 static void process_socket_event(struct c4iw_ep *ep);
120 static void release_ep_resources(struct c4iw_ep *ep);
121 static int process_terminate(struct c4iw_ep *ep);
122 static int terminate(struct sge_iq *iq, const struct rss_header *rss,
123     struct mbuf *m);
124 static int add_ep_to_req_list(struct c4iw_ep *ep, int ep_events);
125 static struct listen_port_info *
126 add_ep_to_listenlist(struct c4iw_listen_ep *lep);
127 static int rem_ep_from_listenlist(struct c4iw_listen_ep *lep);
128 static struct c4iw_listen_ep *
129 find_real_listen_ep(struct c4iw_listen_ep *master_lep, struct socket *so);
130 static int get_ifnet_from_raddr(struct sockaddr_storage *raddr,
131 		struct ifnet **ifp);
132 static void process_newconn(struct c4iw_listen_ep *master_lep,
133 		struct socket *new_so);
134 #define START_EP_TIMER(ep) \
135     do { \
136 	    CTR3(KTR_IW_CXGBE, "start_ep_timer (%s:%d) ep %p", \
137 		__func__, __LINE__, (ep)); \
138 	    start_ep_timer(ep); \
139     } while (0)
140 
141 #define STOP_EP_TIMER(ep) \
142     ({ \
143 	    CTR3(KTR_IW_CXGBE, "stop_ep_timer (%s:%d) ep %p", \
144 		__func__, __LINE__, (ep)); \
145 	    stop_ep_timer(ep); \
146     })
147 
148 #define GET_LOCAL_ADDR(pladdr, so) \
149 	do { \
150 		struct sockaddr_storage *__a = NULL; \
151 		struct  inpcb *__inp = sotoinpcb(so); \
152 		KASSERT(__inp != NULL, \
153 		   ("GET_LOCAL_ADDR(%s):so:%p, inp = NULL", __func__, so)); \
154 		if (__inp->inp_vflag & INP_IPV4) \
155 			in_getsockaddr(so, (struct sockaddr **)&__a); \
156 		else \
157 			in6_getsockaddr(so, (struct sockaddr **)&__a); \
158 		*(pladdr) = *__a; \
159 		free(__a, M_SONAME); \
160 	} while (0)
161 
162 #define GET_REMOTE_ADDR(praddr, so) \
163 	do { \
164 		struct sockaddr_storage *__a = NULL; \
165 		struct  inpcb *__inp = sotoinpcb(so); \
166 		KASSERT(__inp != NULL, \
167 		   ("GET_REMOTE_ADDR(%s):so:%p, inp = NULL", __func__, so)); \
168 		if (__inp->inp_vflag & INP_IPV4) \
169 			in_getpeeraddr(so, (struct sockaddr **)&__a); \
170 		else \
171 			in6_getpeeraddr(so, (struct sockaddr **)&__a); \
172 		*(praddr) = *__a; \
173 		free(__a, M_SONAME); \
174 	} while (0)
175 
176 static char *states[] = {
177 	"idle",
178 	"listen",
179 	"connecting",
180 	"mpa_wait_req",
181 	"mpa_req_sent",
182 	"mpa_req_rcvd",
183 	"mpa_rep_sent",
184 	"fpdu_mode",
185 	"aborting",
186 	"closing",
187 	"moribund",
188 	"dead",
189 	NULL,
190 };
191 
192 static 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 
199 static 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 
205 static 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 
212 static 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 */
219 struct 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  */
266 static struct listen_port_info *
267 add_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 
290 found_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 
297 static int
298 rem_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  */
335 struct c4iw_listen_ep *
336 find_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 ifnet *ifp = NULL, *hw_ifp = NULL;
341 	struct listen_port_info *port_info = NULL;
342 	int i = 0, found_portinfo = 0, found_lep = 0;
343 	uint16_t port;
344 
345 	/*
346 	 * STEP 1: Figure out 'ifp' of the physical interface, not pseudo
347 	 * interfaces like vlan, lagg, etc..
348 	 * TBD: lagg support, lagg + vlan support.
349 	 */
350 	ifp = TOEPCB(so)->l2te->ifp;
351 	if (ifp->if_type == IFT_L2VLAN) {
352 		hw_ifp = VLAN_TRUNKDEV(ifp);
353 		if (hw_ifp == NULL) {
354 			CTR4(KTR_IW_CXGBE, "%s: Failed to get parent ifnet of "
355 				"vlan ifnet %p, sock %p, master_lep %p",
356 				__func__, ifp, so, master_lep);
357 			return (NULL);
358 		}
359 	} else
360 		hw_ifp = ifp;
361 
362 	/* STEP 2: Find 'port_info' with listener local port address. */
363 	port = (master_lep->com.local_addr.ss_family == AF_INET) ?
364 		((struct sockaddr_in *)&master_lep->com.local_addr)->sin_port :
365 		((struct sockaddr_in6 *)&master_lep->com.local_addr)->sin6_port;
366 
367 
368 	mutex_lock(&listen_port_mutex);
369 	list_for_each_entry(port_info, &listen_port_list, list)
370 		if (port_info->port_num == port) {
371 			found_portinfo =1;
372 			break;
373 		}
374 	if (!found_portinfo)
375 		goto out;
376 
377 	/* STEP 3: Traverse through list of lep's that are bound to the current
378 	 * TCP port address and find the lep that belongs to the ifnet on which
379 	 * the SYN frame was received.
380 	 */
381 	list_for_each_entry(lep, &port_info->lep_list, listen_ep_list) {
382 		adap = lep->com.dev->rdev.adap;
383 		for_each_port(adap, i) {
384 			if (hw_ifp == adap->port[i]->vi[0].ifp) {
385 				found_lep =1;
386 				goto out;
387 			}
388 		}
389 	}
390 out:
391 	mutex_unlock(&listen_port_mutex);
392 	return found_lep ? lep : (NULL);
393 }
394 
395 static void process_timeout(struct c4iw_ep *ep)
396 {
397 	struct c4iw_qp_attributes attrs = {0};
398 	int abort = 1;
399 
400 	CTR4(KTR_IW_CXGBE, "%s ep :%p, tid:%u, state %d", __func__,
401 			ep, ep->hwtid, ep->com.state);
402 	set_bit(TIMEDOUT, &ep->com.history);
403 	switch (ep->com.state) {
404 	case MPA_REQ_SENT:
405 		connect_reply_upcall(ep, -ETIMEDOUT);
406 		break;
407 	case MPA_REQ_WAIT:
408 	case MPA_REQ_RCVD:
409 	case MPA_REP_SENT:
410 	case FPDU_MODE:
411 		break;
412 	case CLOSING:
413 	case MORIBUND:
414 		if (ep->com.cm_id && ep->com.qp) {
415 			attrs.next_state = C4IW_QP_STATE_ERROR;
416 			c4iw_modify_qp(ep->com.dev, ep->com.qp,
417 					C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
418 		}
419 		close_complete_upcall(ep, -ETIMEDOUT);
420 		break;
421 	case ABORTING:
422 	case DEAD:
423 		/*
424 		 * These states are expected if the ep timed out at the same
425 		 * time as another thread was calling stop_ep_timer().
426 		 * So we silently do nothing for these states.
427 		 */
428 		abort = 0;
429 		break;
430 	default:
431 		CTR4(KTR_IW_CXGBE, "%s unexpected state ep %p tid %u state %u"
432 				, __func__, ep, ep->hwtid, ep->com.state);
433 		abort = 0;
434 	}
435 	if (abort)
436 		c4iw_ep_disconnect(ep, 1, GFP_KERNEL);
437 	c4iw_put_ep(&ep->com);
438 	return;
439 }
440 
441 struct cqe_list_entry {
442 	struct list_head entry;
443 	struct c4iw_dev *rhp;
444 	struct t4_cqe err_cqe;
445 };
446 
447 static void
448 process_err_cqes(void)
449 {
450 	unsigned long flag;
451 	struct cqe_list_entry *cle;
452 
453 	spin_lock_irqsave(&err_cqe_lock, flag);
454 	while (!list_empty(&err_cqe_list)) {
455 		struct list_head *tmp;
456 		tmp = err_cqe_list.next;
457 		list_del(tmp);
458 		tmp->next = tmp->prev = NULL;
459 		spin_unlock_irqrestore(&err_cqe_lock, flag);
460 		cle = list_entry(tmp, struct cqe_list_entry, entry);
461 		c4iw_ev_dispatch(cle->rhp, &cle->err_cqe);
462 		free(cle, M_CXGBE);
463 		spin_lock_irqsave(&err_cqe_lock, flag);
464 	}
465 	spin_unlock_irqrestore(&err_cqe_lock, flag);
466 
467 	return;
468 }
469 
470 static void
471 process_req(struct work_struct *ctx)
472 {
473 	struct c4iw_ep_common *epc;
474 	unsigned long flag;
475 	int ep_events;
476 
477 	process_err_cqes();
478 	spin_lock_irqsave(&req_lock, flag);
479 	while (!TAILQ_EMPTY(&req_list)) {
480 		epc = TAILQ_FIRST(&req_list);
481 		TAILQ_REMOVE(&req_list, epc, entry);
482 		epc->entry.tqe_prev = NULL;
483 		ep_events = epc->ep_events;
484 		epc->ep_events = 0;
485 		spin_unlock_irqrestore(&req_lock, flag);
486 		mutex_lock(&epc->mutex);
487 		CTR5(KTR_IW_CXGBE, "%s: so %p, ep %p, ep_state %s events 0x%x",
488 		    __func__, epc->so, epc, states[epc->state], ep_events);
489 		if (ep_events & C4IW_EVENT_TERM)
490 			process_terminate((struct c4iw_ep *)epc);
491 		if (ep_events & C4IW_EVENT_TIMEOUT)
492 			process_timeout((struct c4iw_ep *)epc);
493 		if (ep_events & C4IW_EVENT_SOCKET)
494 			process_socket_event((struct c4iw_ep *)epc);
495 		mutex_unlock(&epc->mutex);
496 		c4iw_put_ep(epc);
497 		process_err_cqes();
498 		spin_lock_irqsave(&req_lock, flag);
499 	}
500 	spin_unlock_irqrestore(&req_lock, flag);
501 }
502 
503 /*
504  * XXX: doesn't belong here in the iWARP driver.
505  * XXX: assumes that the connection was offloaded by cxgbe/t4_tom if TF_TOE is
506  *      set.  Is this a valid assumption for active open?
507  */
508 static int
509 set_tcpinfo(struct c4iw_ep *ep)
510 {
511 	struct socket *so = ep->com.so;
512 	struct inpcb *inp = sotoinpcb(so);
513 	struct tcpcb *tp;
514 	struct toepcb *toep;
515 	int rc = 0;
516 
517 	INP_WLOCK(inp);
518 	tp = intotcpcb(inp);
519 	if ((tp->t_flags & TF_TOE) == 0) {
520 		rc = EINVAL;
521 		log(LOG_ERR, "%s: connection not offloaded (so %p, ep %p)\n",
522 		    __func__, so, ep);
523 		goto done;
524 	}
525 	toep = TOEPCB(so);
526 
527 	ep->hwtid = toep->tid;
528 	ep->snd_seq = tp->snd_nxt;
529 	ep->rcv_seq = tp->rcv_nxt;
530 done:
531 	INP_WUNLOCK(inp);
532 	return (rc);
533 
534 }
535 static int
536 get_ifnet_from_raddr(struct sockaddr_storage *raddr, struct ifnet **ifp)
537 {
538 	int err = 0;
539 	struct nhop_object *nh;
540 
541 	if (raddr->ss_family == AF_INET) {
542 		struct sockaddr_in *raddr4 = (struct sockaddr_in *)raddr;
543 
544 		nh = fib4_lookup(RT_DEFAULT_FIB, raddr4->sin_addr, 0,
545 				NHR_NONE, 0);
546 	} else {
547 		struct sockaddr_in6 *raddr6 = (struct sockaddr_in6 *)raddr;
548 		struct in6_addr addr6;
549 		uint32_t scopeid;
550 
551 		memset(&addr6, 0, sizeof(addr6));
552 		in6_splitscope((struct in6_addr *)&raddr6->sin6_addr,
553 					&addr6, &scopeid);
554 		nh = fib6_lookup(RT_DEFAULT_FIB, &addr6, scopeid,
555 				NHR_NONE, 0);
556 	}
557 
558 	if (nh == NULL)
559 		err = EHOSTUNREACH;
560 	else
561 		*ifp = nh->nh_ifp;
562 	CTR2(KTR_IW_CXGBE, "%s: return: %d", __func__, err);
563 	return err;
564 }
565 
566 static void
567 close_socket(struct socket *so)
568 {
569 	uninit_iwarp_socket(so);
570 	soclose(so);
571 }
572 
573 static void
574 process_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 
687 static void
688 process_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 
761 static void
762 process_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 
826 static void
827 setiwsockopt(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 
846 static void
847 init_iwarp_socket(struct socket *so, void *arg)
848 {
849 	if (SOLISTENING(so)) {
850 		SOLISTEN_LOCK(so);
851 		solisten_upcall_set(so, c4iw_so_upcall, arg);
852 		so->so_state |= SS_NBIO;
853 		SOLISTEN_UNLOCK(so);
854 	} else {
855 		SOCKBUF_LOCK(&so->so_rcv);
856 		soupcall_set(so, SO_RCV, c4iw_so_upcall, arg);
857 		so->so_state |= SS_NBIO;
858 		SOCKBUF_UNLOCK(&so->so_rcv);
859 	}
860 }
861 
862 static void
863 uninit_iwarp_socket(struct socket *so)
864 {
865 	if (SOLISTENING(so)) {
866 		SOLISTEN_LOCK(so);
867 		solisten_upcall_set(so, NULL, NULL);
868 		SOLISTEN_UNLOCK(so);
869 	} else {
870 		SOCKBUF_LOCK(&so->so_rcv);
871 		soupcall_clear(so, SO_RCV);
872 		SOCKBUF_UNLOCK(&so->so_rcv);
873 	}
874 }
875 
876 static void
877 process_data(struct c4iw_ep *ep)
878 {
879 	int ret = 0;
880 	int disconnect = 0;
881 	struct c4iw_qp_attributes attrs = {0};
882 
883 	CTR5(KTR_IW_CXGBE, "%s: so %p, ep %p, state %s, sbused %d", __func__,
884 	    ep->com.so, ep, states[ep->com.state], sbused(&ep->com.so->so_rcv));
885 
886 	switch (ep->com.state) {
887 	case MPA_REQ_SENT:
888 		disconnect = process_mpa_reply(ep);
889 		break;
890 	case MPA_REQ_WAIT:
891 		disconnect = process_mpa_request(ep);
892 		if (disconnect)
893 			/* Refered in process_newconn() */
894 			c4iw_put_ep(&ep->parent_ep->com);
895 		break;
896 	case FPDU_MODE:
897 		MPASS(ep->com.qp != NULL);
898 		attrs.next_state = C4IW_QP_STATE_TERMINATE;
899 		ret = c4iw_modify_qp(ep->com.dev, ep->com.qp,
900 					C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
901 		if (ret != -EINPROGRESS)
902 			disconnect = 1;
903 		break;
904 	default:
905 		log(LOG_ERR, "%s: Unexpected streaming data. ep %p, "
906 			    "state %d, so %p, so_state 0x%x, sbused %u\n",
907 			    __func__, ep, ep->com.state, ep->com.so,
908 			    ep->com.so->so_state, sbused(&ep->com.so->so_rcv));
909 		break;
910 	}
911 	if (disconnect)
912 		c4iw_ep_disconnect(ep, disconnect == 2, GFP_KERNEL);
913 
914 }
915 
916 static void
917 process_connected(struct c4iw_ep *ep)
918 {
919 	struct socket *so = ep->com.so;
920 
921 	if ((so->so_state & SS_ISCONNECTED) && !so->so_error) {
922 		if (send_mpa_req(ep))
923 			goto err;
924 	} else {
925 		connect_reply_upcall(ep, -so->so_error);
926 		goto err;
927 	}
928 	return;
929 err:
930 	close_socket(so);
931 	ep->com.state = DEAD;
932 	c4iw_put_ep(&ep->com);
933 	return;
934 }
935 
936 static inline int c4iw_zero_addr(struct sockaddr *addr)
937 {
938 	struct in6_addr *ip6;
939 
940 	if (addr->sa_family == AF_INET)
941 		return (((struct sockaddr_in *)addr)->sin_addr.s_addr == 0);
942 	else {
943 		ip6 = &((struct sockaddr_in6 *) addr)->sin6_addr;
944 		return (ip6->s6_addr32[0] | ip6->s6_addr32[1] |
945 				ip6->s6_addr32[2] | ip6->s6_addr32[3]) == 0;
946 	}
947 }
948 
949 static inline int c4iw_loopback_addr(struct sockaddr *addr)
950 {
951 	if (addr->sa_family == AF_INET)
952 		return IN_LOOPBACK(
953 			ntohl(((struct sockaddr_in *) addr)->sin_addr.s_addr));
954 	else
955 		return IN6_IS_ADDR_LOOPBACK(
956 				&((struct sockaddr_in6 *) addr)->sin6_addr);
957 }
958 
959 static inline int c4iw_any_addr(struct sockaddr *addr)
960 {
961 	return c4iw_zero_addr(addr) || c4iw_loopback_addr(addr);
962 }
963 
964 static void
965 process_newconn(struct c4iw_listen_ep *master_lep, struct socket *new_so)
966 {
967 	struct c4iw_listen_ep *real_lep = NULL;
968 	struct c4iw_ep *new_ep = NULL;
969 	struct sockaddr_in *remote = NULL;
970 	int ret = 0;
971 
972 	MPASS(new_so != NULL);
973 
974 	if (c4iw_any_addr((struct sockaddr *)&master_lep->com.local_addr)) {
975 		/* Here we need to find the 'real_lep' that belongs to the
976 		 * incomming socket's network interface, such that the newly
977 		 * created 'ep' can be attached to the real 'lep'.
978 		 */
979 		real_lep = find_real_listen_ep(master_lep, new_so);
980 		if (real_lep == NULL) {
981 			CTR2(KTR_IW_CXGBE, "%s: Could not find the real listen "
982 					"ep for sock: %p", __func__, new_so);
983 			log(LOG_ERR,"%s: Could not find the real listen ep for "
984 					"sock: %p\n", __func__, new_so);
985 			/* FIXME: properly free the 'new_so' in failure case.
986 			 * Use of soabort() and  soclose() are not legal
987 			 * here(before soaccept()).
988 			 */
989 			return;
990 		}
991 	} else /* for Non-Wildcard address, master_lep is always the real_lep */
992 		real_lep = master_lep;
993 
994 	new_ep = alloc_ep(sizeof(*new_ep), GFP_KERNEL);
995 
996 	CTR6(KTR_IW_CXGBE, "%s: master_lep %p, real_lep: %p, new ep %p, "
997 	    "listening so %p, new so %p", __func__, master_lep, real_lep,
998 	    new_ep, master_lep->com.so, new_so);
999 
1000 	new_ep->com.dev = real_lep->com.dev;
1001 	new_ep->com.so = new_so;
1002 	new_ep->com.cm_id = NULL;
1003 	new_ep->com.thread = real_lep->com.thread;
1004 	new_ep->parent_ep = real_lep;
1005 
1006 	GET_LOCAL_ADDR(&new_ep->com.local_addr, new_so);
1007 	GET_REMOTE_ADDR(&new_ep->com.remote_addr, new_so);
1008 	c4iw_get_ep(&real_lep->com);
1009 	init_timer(&new_ep->timer);
1010 	new_ep->com.state = MPA_REQ_WAIT;
1011 
1012 	setiwsockopt(new_so);
1013 	ret = soaccept(new_so, (struct sockaddr **)&remote);
1014 	if (ret != 0) {
1015 		CTR4(KTR_IW_CXGBE,
1016 				"%s:listen sock:%p, new sock:%p, ret:%d",
1017 				__func__, master_lep->com.so, new_so, ret);
1018 		if (remote != NULL)
1019 			free(remote, M_SONAME);
1020 		soclose(new_so);
1021 		c4iw_put_ep(&new_ep->com);
1022 		c4iw_put_ep(&real_lep->com);
1023 		return;
1024 	}
1025 	free(remote, M_SONAME);
1026 
1027 	START_EP_TIMER(new_ep);
1028 
1029 	/* MPA request might have been queued up on the socket already, so we
1030 	 * initialize the socket/upcall_handler under lock to prevent processing
1031 	 * MPA request on another thread(via process_req()) simultaneously.
1032 	 */
1033 	c4iw_get_ep(&new_ep->com); /* Dereferenced at the end below, this is to
1034 				      avoid freeing of ep before ep unlock. */
1035 	mutex_lock(&new_ep->com.mutex);
1036 	init_iwarp_socket(new_so, &new_ep->com);
1037 
1038 	ret = process_mpa_request(new_ep);
1039 	if (ret) {
1040 		/* ABORT */
1041 		c4iw_ep_disconnect(new_ep, 1, GFP_KERNEL);
1042 		c4iw_put_ep(&real_lep->com);
1043 	}
1044 	mutex_unlock(&new_ep->com.mutex);
1045 	c4iw_put_ep(&new_ep->com);
1046 	return;
1047 }
1048 
1049 static int
1050 add_ep_to_req_list(struct c4iw_ep *ep, int new_ep_event)
1051 {
1052 	unsigned long flag;
1053 
1054 	spin_lock_irqsave(&req_lock, flag);
1055 	if (ep && ep->com.so) {
1056 		ep->com.ep_events |= new_ep_event;
1057 		if (!ep->com.entry.tqe_prev) {
1058 			c4iw_get_ep(&ep->com);
1059 			TAILQ_INSERT_TAIL(&req_list, &ep->com, entry);
1060 			queue_work(c4iw_taskq, &c4iw_task);
1061 		}
1062 	}
1063 	spin_unlock_irqrestore(&req_lock, flag);
1064 
1065 	return (0);
1066 }
1067 
1068 static int
1069 c4iw_so_upcall(struct socket *so, void *arg, int waitflag)
1070 {
1071 	struct c4iw_ep *ep = arg;
1072 
1073 	CTR6(KTR_IW_CXGBE,
1074 	    "%s: so %p, so_state 0x%x, ep %p, ep_state %s, tqe_prev %p",
1075 	    __func__, so, so->so_state, ep, states[ep->com.state],
1076 	    ep->com.entry.tqe_prev);
1077 
1078 	MPASS(ep->com.so == so);
1079 	/*
1080 	 * Wake up any threads waiting in rdma_init()/rdma_fini(),
1081 	 * with locks held.
1082 	 */
1083 	if (so->so_error || (ep->com.dev->rdev.flags & T4_FATAL_ERROR))
1084 		c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET);
1085 	add_ep_to_req_list(ep, C4IW_EVENT_SOCKET);
1086 
1087 	return (SU_OK);
1088 }
1089 
1090 
1091 static int
1092 terminate(struct sge_iq *iq, const struct rss_header *rss, struct mbuf *m)
1093 {
1094 	struct adapter *sc = iq->adapter;
1095 	const struct cpl_rdma_terminate *cpl = mtod(m, const void *);
1096 	unsigned int tid = GET_TID(cpl);
1097 	struct toepcb *toep = lookup_tid(sc, tid);
1098 	struct socket *so;
1099 	struct c4iw_ep *ep;
1100 
1101 	INP_WLOCK(toep->inp);
1102 	so = inp_inpcbtosocket(toep->inp);
1103 	ep = so->so_rcv.sb_upcallarg;
1104 	INP_WUNLOCK(toep->inp);
1105 
1106 	CTR3(KTR_IW_CXGBE, "%s: so %p, ep %p", __func__, so, ep);
1107 	add_ep_to_req_list(ep, C4IW_EVENT_TERM);
1108 
1109 	return 0;
1110 }
1111 
1112 static void
1113 process_socket_event(struct c4iw_ep *ep)
1114 {
1115 	int state = ep->com.state;
1116 	struct socket *so = ep->com.so;
1117 
1118 	if (ep->com.state == DEAD) {
1119 		CTR3(KTR_IW_CXGBE, "%s: Pending socket event discarded "
1120 		    "ep %p ep_state %s", __func__, ep, states[state]);
1121 		return;
1122 	}
1123 
1124 	CTR6(KTR_IW_CXGBE, "process_socket_event: so %p, so_state 0x%x, "
1125 	    "so_err %d, sb_state 0x%x, ep %p, ep_state %s", so, so->so_state,
1126 	    so->so_error, so->so_rcv.sb_state, ep, states[state]);
1127 
1128 	if (state == CONNECTING) {
1129 		process_connected(ep);
1130 		return;
1131 	}
1132 
1133 	if (state == LISTEN) {
1134 		struct c4iw_listen_ep *lep = (struct c4iw_listen_ep *)ep;
1135 		struct socket *listen_so = so, *new_so = NULL;
1136 		int error = 0;
1137 
1138 		SOLISTEN_LOCK(listen_so);
1139 		do {
1140 			error = solisten_dequeue(listen_so, &new_so,
1141 						SOCK_NONBLOCK);
1142 			if (error) {
1143 				CTR4(KTR_IW_CXGBE, "%s: lep %p listen_so %p "
1144 					"error %d", __func__, lep, listen_so,
1145 					error);
1146 				return;
1147 			}
1148 			process_newconn(lep, new_so);
1149 
1150 			/* solisten_dequeue() unlocks while return, so aquire
1151 			 * lock again for sol_qlen and also for next iteration.
1152 			 */
1153 			SOLISTEN_LOCK(listen_so);
1154 		} while (listen_so->sol_qlen);
1155 		SOLISTEN_UNLOCK(listen_so);
1156 
1157 		return;
1158 	}
1159 
1160 	/* connection error */
1161 	if (so->so_error) {
1162 		process_conn_error(ep);
1163 		return;
1164 	}
1165 
1166 	/* peer close */
1167 	if ((so->so_rcv.sb_state & SBS_CANTRCVMORE) && state <= CLOSING) {
1168 		process_peer_close(ep);
1169 		/*
1170 		 * check whether socket disconnect event is pending before
1171 		 * returning. Fallthrough if yes.
1172 		 */
1173 		if (!(so->so_state & SS_ISDISCONNECTED))
1174 			return;
1175 	}
1176 
1177 	/* close complete */
1178 	if (so->so_state & SS_ISDISCONNECTED) {
1179 		process_close_complete(ep);
1180 		return;
1181 	}
1182 
1183 	/* rx data */
1184 	if (sbused(&ep->com.so->so_rcv)) {
1185 		process_data(ep);
1186 		return;
1187 	}
1188 
1189 	/* Socket events for 'MPA Request Received' and 'Close Complete'
1190 	 * were already processed earlier in their previous events handlers.
1191 	 * Hence, these socket events are skipped.
1192 	 * And any other socket events must have handled above.
1193 	 */
1194 	MPASS((ep->com.state == MPA_REQ_RCVD) || (ep->com.state == MORIBUND));
1195 
1196 	if ((ep->com.state != MPA_REQ_RCVD) && (ep->com.state != MORIBUND))
1197 		log(LOG_ERR, "%s: Unprocessed socket event so %p, "
1198 		"so_state 0x%x, so_err %d, sb_state 0x%x, ep %p, ep_state %s\n",
1199 		__func__, so, so->so_state, so->so_error, so->so_rcv.sb_state,
1200 			ep, states[state]);
1201 
1202 }
1203 
1204 SYSCTL_NODE(_hw, OID_AUTO, iw_cxgbe, CTLFLAG_RD | CTLFLAG_MPSAFE, 0,
1205     "iw_cxgbe driver parameters");
1206 
1207 static int dack_mode = 0;
1208 SYSCTL_INT(_hw_iw_cxgbe, OID_AUTO, dack_mode, CTLFLAG_RWTUN, &dack_mode, 0,
1209 		"Delayed ack mode (default = 0)");
1210 
1211 int c4iw_max_read_depth = 8;
1212 SYSCTL_INT(_hw_iw_cxgbe, OID_AUTO, c4iw_max_read_depth, CTLFLAG_RWTUN, &c4iw_max_read_depth, 0,
1213 		"Per-connection max ORD/IRD (default = 8)");
1214 
1215 static int enable_tcp_timestamps;
1216 SYSCTL_INT(_hw_iw_cxgbe, OID_AUTO, enable_tcp_timestamps, CTLFLAG_RWTUN, &enable_tcp_timestamps, 0,
1217 		"Enable tcp timestamps (default = 0)");
1218 
1219 static int enable_tcp_sack;
1220 SYSCTL_INT(_hw_iw_cxgbe, OID_AUTO, enable_tcp_sack, CTLFLAG_RWTUN, &enable_tcp_sack, 0,
1221 		"Enable tcp SACK (default = 0)");
1222 
1223 static int enable_tcp_window_scaling = 1;
1224 SYSCTL_INT(_hw_iw_cxgbe, OID_AUTO, enable_tcp_window_scaling, CTLFLAG_RWTUN, &enable_tcp_window_scaling, 0,
1225 		"Enable tcp window scaling (default = 1)");
1226 
1227 int c4iw_debug = 0;
1228 SYSCTL_INT(_hw_iw_cxgbe, OID_AUTO, c4iw_debug, CTLFLAG_RWTUN, &c4iw_debug, 0,
1229 		"Enable debug logging (default = 0)");
1230 
1231 static int peer2peer = 1;
1232 SYSCTL_INT(_hw_iw_cxgbe, OID_AUTO, peer2peer, CTLFLAG_RWTUN, &peer2peer, 0,
1233 		"Support peer2peer ULPs (default = 1)");
1234 
1235 static int p2p_type = FW_RI_INIT_P2PTYPE_READ_REQ;
1236 SYSCTL_INT(_hw_iw_cxgbe, OID_AUTO, p2p_type, CTLFLAG_RWTUN, &p2p_type, 0,
1237 		"RDMAP opcode to use for the RTR message: 1 = RDMA_READ 0 = RDMA_WRITE (default 1)");
1238 
1239 static int ep_timeout_secs = 60;
1240 SYSCTL_INT(_hw_iw_cxgbe, OID_AUTO, ep_timeout_secs, CTLFLAG_RWTUN, &ep_timeout_secs, 0,
1241 		"CM Endpoint operation timeout in seconds (default = 60)");
1242 
1243 static int mpa_rev = 1;
1244 SYSCTL_INT(_hw_iw_cxgbe, OID_AUTO, mpa_rev, CTLFLAG_RWTUN, &mpa_rev, 0,
1245 		"MPA Revision, 0 supports amso1100, 1 is RFC5044 spec compliant, 2 is IETF MPA Peer Connect Draft compliant (default = 1)");
1246 
1247 static int markers_enabled;
1248 SYSCTL_INT(_hw_iw_cxgbe, OID_AUTO, markers_enabled, CTLFLAG_RWTUN, &markers_enabled, 0,
1249 		"Enable MPA MARKERS (default(0) = disabled)");
1250 
1251 static int crc_enabled = 1;
1252 SYSCTL_INT(_hw_iw_cxgbe, OID_AUTO, crc_enabled, CTLFLAG_RWTUN, &crc_enabled, 0,
1253 		"Enable MPA CRC (default(1) = enabled)");
1254 
1255 static int rcv_win = 256 * 1024;
1256 SYSCTL_INT(_hw_iw_cxgbe, OID_AUTO, rcv_win, CTLFLAG_RWTUN, &rcv_win, 0,
1257 		"TCP receive window in bytes (default = 256KB)");
1258 
1259 static int snd_win = 128 * 1024;
1260 SYSCTL_INT(_hw_iw_cxgbe, OID_AUTO, snd_win, CTLFLAG_RWTUN, &snd_win, 0,
1261 		"TCP send window in bytes (default = 128KB)");
1262 
1263 int use_dsgl = 1;
1264 SYSCTL_INT(_hw_iw_cxgbe, OID_AUTO, use_dsgl, CTLFLAG_RWTUN, &use_dsgl, 0,
1265 		"Use DSGL for PBL/FastReg (default=1)");
1266 
1267 int inline_threshold = 128;
1268 SYSCTL_INT(_hw_iw_cxgbe, OID_AUTO, inline_threshold, CTLFLAG_RWTUN, &inline_threshold, 0,
1269 		"inline vs dsgl threshold (default=128)");
1270 
1271 static int reuseaddr = 0;
1272 SYSCTL_INT(_hw_iw_cxgbe, OID_AUTO, reuseaddr, CTLFLAG_RWTUN, &reuseaddr, 0,
1273 		"Enable SO_REUSEADDR & SO_REUSEPORT socket options on all iWARP client connections(default = 0)");
1274 
1275 static void
1276 start_ep_timer(struct c4iw_ep *ep)
1277 {
1278 
1279 	if (timer_pending(&ep->timer)) {
1280 		CTR2(KTR_IW_CXGBE, "%s: ep %p, already started", __func__, ep);
1281 		printk(KERN_ERR "%s timer already started! ep %p\n", __func__,
1282 		    ep);
1283 		return;
1284 	}
1285 	clear_bit(TIMEOUT, &ep->com.flags);
1286 	c4iw_get_ep(&ep->com);
1287 	ep->timer.expires = jiffies + ep_timeout_secs * HZ;
1288 	ep->timer.data = (unsigned long)ep;
1289 	ep->timer.function = ep_timeout;
1290 	add_timer(&ep->timer);
1291 }
1292 
1293 static int
1294 stop_ep_timer(struct c4iw_ep *ep)
1295 {
1296 
1297 	del_timer_sync(&ep->timer);
1298 	if (!test_and_set_bit(TIMEOUT, &ep->com.flags)) {
1299 		c4iw_put_ep(&ep->com);
1300 		return 0;
1301 	}
1302 	return 1;
1303 }
1304 
1305 static void *
1306 alloc_ep(int size, gfp_t gfp)
1307 {
1308 	struct c4iw_ep_common *epc;
1309 
1310 	epc = kzalloc(size, gfp);
1311 	if (epc == NULL)
1312 		return (NULL);
1313 
1314 	kref_init(&epc->kref);
1315 	mutex_init(&epc->mutex);
1316 	c4iw_init_wr_wait(&epc->wr_wait);
1317 
1318 	return (epc);
1319 }
1320 
1321 void _c4iw_free_ep(struct kref *kref)
1322 {
1323 	struct c4iw_ep *ep;
1324 #if defined(KTR) || defined(INVARIANTS)
1325 	struct c4iw_ep_common *epc;
1326 #endif
1327 
1328 	ep = container_of(kref, struct c4iw_ep, com.kref);
1329 #if defined(KTR) || defined(INVARIANTS)
1330 	epc = &ep->com;
1331 #endif
1332 	KASSERT(!epc->entry.tqe_prev, ("%s epc %p still on req list",
1333 	    __func__, epc));
1334 	if (test_bit(QP_REFERENCED, &ep->com.flags))
1335 		deref_qp(ep);
1336 	CTR4(KTR_IW_CXGBE, "%s: ep %p, history 0x%lx, flags 0x%lx",
1337 	    __func__, ep, epc->history, epc->flags);
1338 	kfree(ep);
1339 }
1340 
1341 static void release_ep_resources(struct c4iw_ep *ep)
1342 {
1343 	CTR2(KTR_IW_CXGBE, "%s:rerB %p", __func__, ep);
1344 	set_bit(RELEASE_RESOURCES, &ep->com.flags);
1345 	c4iw_put_ep(&ep->com);
1346 	CTR2(KTR_IW_CXGBE, "%s:rerE %p", __func__, ep);
1347 }
1348 
1349 static int
1350 send_mpa_req(struct c4iw_ep *ep)
1351 {
1352 	int mpalen;
1353 	struct mpa_message *mpa;
1354 	struct mpa_v2_conn_params mpa_v2_params;
1355 	struct mbuf *m;
1356 	char mpa_rev_to_use = mpa_rev;
1357 	int err = 0;
1358 
1359 	if (ep->retry_with_mpa_v1)
1360 		mpa_rev_to_use = 1;
1361 	mpalen = sizeof(*mpa) + ep->plen;
1362 	if (mpa_rev_to_use == 2)
1363 		mpalen += sizeof(struct mpa_v2_conn_params);
1364 
1365 	mpa = malloc(mpalen, M_CXGBE, M_NOWAIT);
1366 	if (mpa == NULL) {
1367 		err = -ENOMEM;
1368 		CTR3(KTR_IW_CXGBE, "%s:smr1 ep: %p , error: %d",
1369 				__func__, ep, err);
1370 		goto err;
1371 	}
1372 
1373 	memset(mpa, 0, mpalen);
1374 	memcpy(mpa->key, MPA_KEY_REQ, sizeof(mpa->key));
1375 	mpa->flags = (crc_enabled ? MPA_CRC : 0) |
1376 		(markers_enabled ? MPA_MARKERS : 0) |
1377 		(mpa_rev_to_use == 2 ? MPA_ENHANCED_RDMA_CONN : 0);
1378 	mpa->private_data_size = htons(ep->plen);
1379 	mpa->revision = mpa_rev_to_use;
1380 
1381 	if (mpa_rev_to_use == 1) {
1382 		ep->tried_with_mpa_v1 = 1;
1383 		ep->retry_with_mpa_v1 = 0;
1384 	}
1385 
1386 	if (mpa_rev_to_use == 2) {
1387 		mpa->private_data_size = htons(ntohs(mpa->private_data_size) +
1388 					    sizeof(struct mpa_v2_conn_params));
1389 		mpa_v2_params.ird = htons((u16)ep->ird);
1390 		mpa_v2_params.ord = htons((u16)ep->ord);
1391 
1392 		if (peer2peer) {
1393 			mpa_v2_params.ird |= htons(MPA_V2_PEER2PEER_MODEL);
1394 
1395 			if (p2p_type == FW_RI_INIT_P2PTYPE_RDMA_WRITE) {
1396 				mpa_v2_params.ord |=
1397 				    htons(MPA_V2_RDMA_WRITE_RTR);
1398 			} else if (p2p_type == FW_RI_INIT_P2PTYPE_READ_REQ) {
1399 				mpa_v2_params.ord |=
1400 					htons(MPA_V2_RDMA_READ_RTR);
1401 			}
1402 		}
1403 		memcpy(mpa->private_data, &mpa_v2_params,
1404 			sizeof(struct mpa_v2_conn_params));
1405 
1406 		if (ep->plen) {
1407 
1408 			memcpy(mpa->private_data +
1409 				sizeof(struct mpa_v2_conn_params),
1410 				ep->mpa_pkt + sizeof(*mpa), ep->plen);
1411 		}
1412 	} else {
1413 
1414 		if (ep->plen)
1415 			memcpy(mpa->private_data,
1416 					ep->mpa_pkt + sizeof(*mpa), ep->plen);
1417 		CTR2(KTR_IW_CXGBE, "%s:smr7 %p", __func__, ep);
1418 	}
1419 
1420 	m = m_getm(NULL, mpalen, M_NOWAIT, MT_DATA);
1421 	if (m == NULL) {
1422 		err = -ENOMEM;
1423 		CTR3(KTR_IW_CXGBE, "%s:smr2 ep: %p , error: %d",
1424 				__func__, ep, err);
1425 		free(mpa, M_CXGBE);
1426 		goto err;
1427 	}
1428 	m_copyback(m, 0, mpalen, (void *)mpa);
1429 	free(mpa, M_CXGBE);
1430 
1431 	err = -sosend(ep->com.so, NULL, NULL, m, NULL, MSG_DONTWAIT,
1432 			ep->com.thread);
1433 	if (err) {
1434 		CTR3(KTR_IW_CXGBE, "%s:smr3 ep: %p , error: %d",
1435 				__func__, ep, err);
1436 		goto err;
1437 	}
1438 
1439 	START_EP_TIMER(ep);
1440 	ep->com.state = MPA_REQ_SENT;
1441 	ep->mpa_attr.initiator = 1;
1442 	CTR3(KTR_IW_CXGBE, "%s:smrE %p, error: %d", __func__, ep, err);
1443 	return 0;
1444 err:
1445 	connect_reply_upcall(ep, err);
1446 	CTR3(KTR_IW_CXGBE, "%s:smrE %p, error: %d", __func__, ep, err);
1447 	return err;
1448 }
1449 
1450 static int send_mpa_reject(struct c4iw_ep *ep, const void *pdata, u8 plen)
1451 {
1452 	int mpalen ;
1453 	struct mpa_message *mpa;
1454 	struct mpa_v2_conn_params mpa_v2_params;
1455 	struct mbuf *m;
1456 	int err;
1457 
1458 	CTR4(KTR_IW_CXGBE, "%s:smrejB %p %u %d", __func__, ep, ep->hwtid,
1459 	    ep->plen);
1460 
1461 	mpalen = sizeof(*mpa) + plen;
1462 
1463 	if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) {
1464 
1465 		mpalen += sizeof(struct mpa_v2_conn_params);
1466 		CTR4(KTR_IW_CXGBE, "%s:smrej1 %p %u %d", __func__, ep,
1467 		    ep->mpa_attr.version, mpalen);
1468 	}
1469 
1470 	mpa = malloc(mpalen, M_CXGBE, M_NOWAIT);
1471 	if (mpa == NULL)
1472 		return (-ENOMEM);
1473 
1474 	memset(mpa, 0, mpalen);
1475 	memcpy(mpa->key, MPA_KEY_REP, sizeof(mpa->key));
1476 	mpa->flags = MPA_REJECT;
1477 	mpa->revision = mpa_rev;
1478 	mpa->private_data_size = htons(plen);
1479 
1480 	if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) {
1481 
1482 		mpa->flags |= MPA_ENHANCED_RDMA_CONN;
1483 		mpa->private_data_size = htons(ntohs(mpa->private_data_size) +
1484 					    sizeof(struct mpa_v2_conn_params));
1485 		mpa_v2_params.ird = htons(((u16)ep->ird) |
1486 				(peer2peer ? MPA_V2_PEER2PEER_MODEL :
1487 				 0));
1488 		mpa_v2_params.ord = htons(((u16)ep->ord) | (peer2peer ?
1489 					(p2p_type ==
1490 					 FW_RI_INIT_P2PTYPE_RDMA_WRITE ?
1491 					 MPA_V2_RDMA_WRITE_RTR : p2p_type ==
1492 					 FW_RI_INIT_P2PTYPE_READ_REQ ?
1493 					 MPA_V2_RDMA_READ_RTR : 0) : 0));
1494 		memcpy(mpa->private_data, &mpa_v2_params,
1495 				sizeof(struct mpa_v2_conn_params));
1496 
1497 		if (ep->plen)
1498 			memcpy(mpa->private_data +
1499 				sizeof(struct mpa_v2_conn_params), pdata, plen);
1500 		CTR5(KTR_IW_CXGBE, "%s:smrej3 %p %d %d %d", __func__, ep,
1501 		    mpa_v2_params.ird, mpa_v2_params.ord, ep->plen);
1502 	} else
1503 		if (plen)
1504 			memcpy(mpa->private_data, pdata, plen);
1505 
1506 	m = m_getm(NULL, mpalen, M_NOWAIT, MT_DATA);
1507 	if (m == NULL) {
1508 		free(mpa, M_CXGBE);
1509 		return (-ENOMEM);
1510 	}
1511 	m_copyback(m, 0, mpalen, (void *)mpa);
1512 	free(mpa, M_CXGBE);
1513 
1514 	err = -sosend(ep->com.so, NULL, NULL, m, NULL, MSG_DONTWAIT, ep->com.thread);
1515 	if (!err)
1516 		ep->snd_seq += mpalen;
1517 	CTR4(KTR_IW_CXGBE, "%s:smrejE %p %u %d", __func__, ep, ep->hwtid, err);
1518 	return err;
1519 }
1520 
1521 static int send_mpa_reply(struct c4iw_ep *ep, const void *pdata, u8 plen)
1522 {
1523 	int mpalen;
1524 	struct mpa_message *mpa;
1525 	struct mbuf *m;
1526 	struct mpa_v2_conn_params mpa_v2_params;
1527 	int err;
1528 
1529 	CTR2(KTR_IW_CXGBE, "%s:smrepB %p", __func__, ep);
1530 
1531 	mpalen = sizeof(*mpa) + plen;
1532 
1533 	if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) {
1534 
1535 		CTR3(KTR_IW_CXGBE, "%s:smrep1 %p %d", __func__, ep,
1536 		    ep->mpa_attr.version);
1537 		mpalen += sizeof(struct mpa_v2_conn_params);
1538 	}
1539 
1540 	mpa = malloc(mpalen, M_CXGBE, M_NOWAIT);
1541 	if (mpa == NULL)
1542 		return (-ENOMEM);
1543 
1544 	memset(mpa, 0, sizeof(*mpa));
1545 	memcpy(mpa->key, MPA_KEY_REP, sizeof(mpa->key));
1546 	mpa->flags = (ep->mpa_attr.crc_enabled ? MPA_CRC : 0) |
1547 		(markers_enabled ? MPA_MARKERS : 0);
1548 	mpa->revision = ep->mpa_attr.version;
1549 	mpa->private_data_size = htons(plen);
1550 
1551 	if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) {
1552 
1553 		mpa->flags |= MPA_ENHANCED_RDMA_CONN;
1554 		mpa->private_data_size +=
1555 			htons(sizeof(struct mpa_v2_conn_params));
1556 		mpa_v2_params.ird = htons((u16)ep->ird);
1557 		mpa_v2_params.ord = htons((u16)ep->ord);
1558 		CTR5(KTR_IW_CXGBE, "%s:smrep3 %p %d %d %d", __func__, ep,
1559 		    ep->mpa_attr.version, mpa_v2_params.ird, mpa_v2_params.ord);
1560 
1561 		if (peer2peer && (ep->mpa_attr.p2p_type !=
1562 			FW_RI_INIT_P2PTYPE_DISABLED)) {
1563 
1564 			mpa_v2_params.ird |= htons(MPA_V2_PEER2PEER_MODEL);
1565 
1566 			if (p2p_type == FW_RI_INIT_P2PTYPE_RDMA_WRITE) {
1567 
1568 				mpa_v2_params.ord |=
1569 					htons(MPA_V2_RDMA_WRITE_RTR);
1570 				CTR5(KTR_IW_CXGBE, "%s:smrep4 %p %d %d %d",
1571 				    __func__, ep, p2p_type, mpa_v2_params.ird,
1572 				    mpa_v2_params.ord);
1573 			}
1574 			else if (p2p_type == FW_RI_INIT_P2PTYPE_READ_REQ) {
1575 
1576 				mpa_v2_params.ord |=
1577 					htons(MPA_V2_RDMA_READ_RTR);
1578 				CTR5(KTR_IW_CXGBE, "%s:smrep5 %p %d %d %d",
1579 				    __func__, ep, p2p_type, mpa_v2_params.ird,
1580 				    mpa_v2_params.ord);
1581 			}
1582 		}
1583 
1584 		memcpy(mpa->private_data, &mpa_v2_params,
1585 			sizeof(struct mpa_v2_conn_params));
1586 
1587 		if (ep->plen)
1588 			memcpy(mpa->private_data +
1589 				sizeof(struct mpa_v2_conn_params), pdata, plen);
1590 	} else
1591 		if (plen)
1592 			memcpy(mpa->private_data, pdata, plen);
1593 
1594 	m = m_getm(NULL, mpalen, M_NOWAIT, MT_DATA);
1595 	if (m == NULL) {
1596 		free(mpa, M_CXGBE);
1597 		return (-ENOMEM);
1598 	}
1599 	m_copyback(m, 0, mpalen, (void *)mpa);
1600 	free(mpa, M_CXGBE);
1601 
1602 
1603 	ep->com.state = MPA_REP_SENT;
1604 	ep->snd_seq += mpalen;
1605 	err = -sosend(ep->com.so, NULL, NULL, m, NULL, MSG_DONTWAIT,
1606 			ep->com.thread);
1607 	CTR3(KTR_IW_CXGBE, "%s:smrepE %p %d", __func__, ep, err);
1608 	return err;
1609 }
1610 
1611 
1612 
1613 static void close_complete_upcall(struct c4iw_ep *ep, int status)
1614 {
1615 	struct iw_cm_event event;
1616 
1617 	CTR2(KTR_IW_CXGBE, "%s:ccuB %p", __func__, ep);
1618 	memset(&event, 0, sizeof(event));
1619 	event.event = IW_CM_EVENT_CLOSE;
1620 	event.status = status;
1621 
1622 	if (ep->com.cm_id) {
1623 
1624 		CTR2(KTR_IW_CXGBE, "%s:ccu1 %1", __func__, ep);
1625 		ep->com.cm_id->event_handler(ep->com.cm_id, &event);
1626 		deref_cm_id(&ep->com);
1627 		set_bit(CLOSE_UPCALL, &ep->com.history);
1628 	}
1629 	CTR2(KTR_IW_CXGBE, "%s:ccuE %p", __func__, ep);
1630 }
1631 
1632 static int
1633 send_abort(struct c4iw_ep *ep)
1634 {
1635 	struct socket *so = ep->com.so;
1636 	struct sockopt sopt;
1637 	int rc;
1638 	struct linger l;
1639 
1640 	CTR5(KTR_IW_CXGBE, "%s ep %p so %p state %s tid %d", __func__, ep, so,
1641 	    states[ep->com.state], ep->hwtid);
1642 
1643 	l.l_onoff = 1;
1644 	l.l_linger = 0;
1645 
1646 	/* linger_time of 0 forces RST to be sent */
1647 	sopt.sopt_dir = SOPT_SET;
1648 	sopt.sopt_level = SOL_SOCKET;
1649 	sopt.sopt_name = SO_LINGER;
1650 	sopt.sopt_val = (caddr_t)&l;
1651 	sopt.sopt_valsize = sizeof l;
1652 	sopt.sopt_td = NULL;
1653 	rc = -sosetopt(so, &sopt);
1654 	if (rc != 0) {
1655 		log(LOG_ERR, "%s: sosetopt(%p, linger = 0) failed with %d.\n",
1656 		    __func__, so, rc);
1657 	}
1658 
1659 	uninit_iwarp_socket(so);
1660 	soclose(so);
1661 	set_bit(ABORT_CONN, &ep->com.history);
1662 
1663 	/*
1664 	 * TBD: iw_cxgbe driver should receive ABORT reply for every ABORT
1665 	 * request it has sent. But the current TOE driver is not propagating
1666 	 * this ABORT reply event (via do_abort_rpl) to iw_cxgbe. So as a work-
1667 	 * around de-refererece 'ep' here instead of doing it in abort_rpl()
1668 	 * handler(not yet implemented) of iw_cxgbe driver.
1669 	 */
1670 	release_ep_resources(ep);
1671 	ep->com.state = DEAD;
1672 
1673 	return (0);
1674 }
1675 
1676 static void peer_close_upcall(struct c4iw_ep *ep)
1677 {
1678 	struct iw_cm_event event;
1679 
1680 	CTR2(KTR_IW_CXGBE, "%s:pcuB %p", __func__, ep);
1681 	memset(&event, 0, sizeof(event));
1682 	event.event = IW_CM_EVENT_DISCONNECT;
1683 
1684 	if (ep->com.cm_id) {
1685 
1686 		CTR2(KTR_IW_CXGBE, "%s:pcu1 %p", __func__, ep);
1687 		ep->com.cm_id->event_handler(ep->com.cm_id, &event);
1688 		set_bit(DISCONN_UPCALL, &ep->com.history);
1689 	}
1690 	CTR2(KTR_IW_CXGBE, "%s:pcuE %p", __func__, ep);
1691 }
1692 
1693 static void peer_abort_upcall(struct c4iw_ep *ep)
1694 {
1695 	struct iw_cm_event event;
1696 
1697 	CTR2(KTR_IW_CXGBE, "%s:pauB %p", __func__, ep);
1698 	memset(&event, 0, sizeof(event));
1699 	event.event = IW_CM_EVENT_CLOSE;
1700 	event.status = -ECONNRESET;
1701 
1702 	if (ep->com.cm_id) {
1703 
1704 		CTR2(KTR_IW_CXGBE, "%s:pau1 %p", __func__, ep);
1705 		ep->com.cm_id->event_handler(ep->com.cm_id, &event);
1706 		deref_cm_id(&ep->com);
1707 		set_bit(ABORT_UPCALL, &ep->com.history);
1708 	}
1709 	CTR2(KTR_IW_CXGBE, "%s:pauE %p", __func__, ep);
1710 }
1711 
1712 static void connect_reply_upcall(struct c4iw_ep *ep, int status)
1713 {
1714 	struct iw_cm_event event;
1715 
1716 	CTR3(KTR_IW_CXGBE, "%s:cruB %p, status: %d", __func__, ep, status);
1717 	memset(&event, 0, sizeof(event));
1718 	event.event = IW_CM_EVENT_CONNECT_REPLY;
1719 	event.status = ((status == -ECONNABORTED) || (status == -EPIPE)) ?
1720 					-ECONNRESET : status;
1721 	event.local_addr = ep->com.local_addr;
1722 	event.remote_addr = ep->com.remote_addr;
1723 
1724 	if ((status == 0) || (status == -ECONNREFUSED)) {
1725 
1726 		if (!ep->tried_with_mpa_v1) {
1727 
1728 			CTR2(KTR_IW_CXGBE, "%s:cru1 %p", __func__, ep);
1729 			/* this means MPA_v2 is used */
1730 			event.ord = ep->ird;
1731 			event.ird = ep->ord;
1732 			event.private_data_len = ep->plen -
1733 				sizeof(struct mpa_v2_conn_params);
1734 			event.private_data = ep->mpa_pkt +
1735 				sizeof(struct mpa_message) +
1736 				sizeof(struct mpa_v2_conn_params);
1737 		} else {
1738 
1739 			CTR2(KTR_IW_CXGBE, "%s:cru2 %p", __func__, ep);
1740 			/* this means MPA_v1 is used */
1741 			event.ord = c4iw_max_read_depth;
1742 			event.ird = c4iw_max_read_depth;
1743 			event.private_data_len = ep->plen;
1744 			event.private_data = ep->mpa_pkt +
1745 				sizeof(struct mpa_message);
1746 		}
1747 	}
1748 
1749 	if (ep->com.cm_id) {
1750 
1751 		CTR2(KTR_IW_CXGBE, "%s:cru3 %p", __func__, ep);
1752 		set_bit(CONN_RPL_UPCALL, &ep->com.history);
1753 		ep->com.cm_id->event_handler(ep->com.cm_id, &event);
1754 	}
1755 
1756 	if(status == -ECONNABORTED) {
1757 
1758 		CTR3(KTR_IW_CXGBE, "%s:cruE %p %d", __func__, ep, status);
1759 		return;
1760 	}
1761 
1762 	if (status < 0) {
1763 
1764 		CTR3(KTR_IW_CXGBE, "%s:cru4 %p %d", __func__, ep, status);
1765 		deref_cm_id(&ep->com);
1766 	}
1767 
1768 	CTR2(KTR_IW_CXGBE, "%s:cruE %p", __func__, ep);
1769 }
1770 
1771 static int connect_request_upcall(struct c4iw_ep *ep)
1772 {
1773 	struct iw_cm_event event;
1774 	int ret;
1775 
1776 	CTR3(KTR_IW_CXGBE, "%s: ep %p, mpa_v1 %d", __func__, ep,
1777 	    ep->tried_with_mpa_v1);
1778 
1779 	memset(&event, 0, sizeof(event));
1780 	event.event = IW_CM_EVENT_CONNECT_REQUEST;
1781 	event.local_addr = ep->com.local_addr;
1782 	event.remote_addr = ep->com.remote_addr;
1783 	event.provider_data = ep;
1784 
1785 	if (!ep->tried_with_mpa_v1) {
1786 		/* this means MPA_v2 is used */
1787 		event.ord = ep->ord;
1788 		event.ird = ep->ird;
1789 		event.private_data_len = ep->plen -
1790 			sizeof(struct mpa_v2_conn_params);
1791 		event.private_data = ep->mpa_pkt + sizeof(struct mpa_message) +
1792 			sizeof(struct mpa_v2_conn_params);
1793 	} else {
1794 
1795 		/* this means MPA_v1 is used. Send max supported */
1796 		event.ord = c4iw_max_read_depth;
1797 		event.ird = c4iw_max_read_depth;
1798 		event.private_data_len = ep->plen;
1799 		event.private_data = ep->mpa_pkt + sizeof(struct mpa_message);
1800 	}
1801 
1802 	c4iw_get_ep(&ep->com);
1803 	ret = ep->parent_ep->com.cm_id->event_handler(ep->parent_ep->com.cm_id,
1804 	    &event);
1805 	if(ret) {
1806 		CTR3(KTR_IW_CXGBE, "%s: ep %p, Failure while notifying event to"
1807 			" IWCM, err:%d", __func__, ep, ret);
1808 		c4iw_put_ep(&ep->com);
1809 	} else
1810 		/* Dereference parent_ep only in success case.
1811 		 * In case of failure, parent_ep is dereferenced by the caller
1812 		 * of process_mpa_request().
1813 		 */
1814 		c4iw_put_ep(&ep->parent_ep->com);
1815 
1816 	set_bit(CONNREQ_UPCALL, &ep->com.history);
1817 	return ret;
1818 }
1819 
1820 static void established_upcall(struct c4iw_ep *ep)
1821 {
1822 	struct iw_cm_event event;
1823 
1824 	CTR2(KTR_IW_CXGBE, "%s:euB %p", __func__, ep);
1825 	memset(&event, 0, sizeof(event));
1826 	event.event = IW_CM_EVENT_ESTABLISHED;
1827 	event.ird = ep->ord;
1828 	event.ord = ep->ird;
1829 
1830 	if (ep->com.cm_id) {
1831 
1832 		CTR2(KTR_IW_CXGBE, "%s:eu1 %p", __func__, ep);
1833 		ep->com.cm_id->event_handler(ep->com.cm_id, &event);
1834 		set_bit(ESTAB_UPCALL, &ep->com.history);
1835 	}
1836 	CTR2(KTR_IW_CXGBE, "%s:euE %p", __func__, ep);
1837 }
1838 
1839 
1840 #define RELAXED_IRD_NEGOTIATION 1
1841 
1842 /*
1843  * process_mpa_reply - process streaming mode MPA reply
1844  *
1845  * Returns:
1846  *
1847  * 0 upon success indicating a connect request was delivered to the ULP
1848  * or the mpa request is incomplete but valid so far.
1849  *
1850  * 1 if a failure requires the caller to close the connection.
1851  *
1852  * 2 if a failure requires the caller to abort the connection.
1853  */
1854 static int process_mpa_reply(struct c4iw_ep *ep)
1855 {
1856 	struct mpa_message *mpa;
1857 	struct mpa_v2_conn_params *mpa_v2_params;
1858 	u16 plen;
1859 	u16 resp_ird, resp_ord;
1860 	u8 rtr_mismatch = 0, insuff_ird = 0;
1861 	struct c4iw_qp_attributes attrs = {0};
1862 	enum c4iw_qp_attr_mask mask;
1863 	int err;
1864 	struct mbuf *top, *m;
1865 	int flags = MSG_DONTWAIT;
1866 	struct uio uio;
1867 	int disconnect = 0;
1868 
1869 	CTR2(KTR_IW_CXGBE, "%s:pmrB %p", __func__, ep);
1870 
1871 	/*
1872 	 * Stop mpa timer.  If it expired, then
1873 	 * we ignore the MPA reply.  process_timeout()
1874 	 * will abort the connection.
1875 	 */
1876 	if (STOP_EP_TIMER(ep))
1877 		return 0;
1878 
1879 	uio.uio_resid = 1000000;
1880 	uio.uio_td = ep->com.thread;
1881 	err = soreceive(ep->com.so, NULL, &uio, &top, NULL, &flags);
1882 
1883 	if (err) {
1884 
1885 		if (err == EWOULDBLOCK) {
1886 
1887 			CTR2(KTR_IW_CXGBE, "%s:pmr1 %p", __func__, ep);
1888 			START_EP_TIMER(ep);
1889 			return 0;
1890 		}
1891 		err = -err;
1892 		CTR2(KTR_IW_CXGBE, "%s:pmr2 %p", __func__, ep);
1893 		goto err;
1894 	}
1895 
1896 	if (ep->com.so->so_rcv.sb_mb) {
1897 
1898 		CTR2(KTR_IW_CXGBE, "%s:pmr3 %p", __func__, ep);
1899 		printf("%s data after soreceive called! so %p sb_mb %p top %p\n",
1900 		       __func__, ep->com.so, ep->com.so->so_rcv.sb_mb, top);
1901 	}
1902 
1903 	m = top;
1904 
1905 	do {
1906 
1907 		CTR2(KTR_IW_CXGBE, "%s:pmr4 %p", __func__, ep);
1908 		/*
1909 		 * If we get more than the supported amount of private data
1910 		 * then we must fail this connection.
1911 		 */
1912 		if (ep->mpa_pkt_len + m->m_len > sizeof(ep->mpa_pkt)) {
1913 
1914 			CTR3(KTR_IW_CXGBE, "%s:pmr5 %p %d", __func__, ep,
1915 			    ep->mpa_pkt_len + m->m_len);
1916 			err = (-EINVAL);
1917 			goto err_stop_timer;
1918 		}
1919 
1920 		/*
1921 		 * copy the new data into our accumulation buffer.
1922 		 */
1923 		m_copydata(m, 0, m->m_len, &(ep->mpa_pkt[ep->mpa_pkt_len]));
1924 		ep->mpa_pkt_len += m->m_len;
1925 		if (!m->m_next)
1926 			m = m->m_nextpkt;
1927 		else
1928 			m = m->m_next;
1929 	} while (m);
1930 
1931 	m_freem(top);
1932 	/*
1933 	 * if we don't even have the mpa message, then bail.
1934 	 */
1935 	if (ep->mpa_pkt_len < sizeof(*mpa)) {
1936 		return 0;
1937 	}
1938 	mpa = (struct mpa_message *) ep->mpa_pkt;
1939 
1940 	/* Validate MPA header. */
1941 	if (mpa->revision > mpa_rev) {
1942 
1943 		CTR4(KTR_IW_CXGBE, "%s:pmr6 %p %d %d", __func__, ep,
1944 		    mpa->revision, mpa_rev);
1945 		printk(KERN_ERR MOD "%s MPA version mismatch. Local = %d, "
1946 				" Received = %d\n", __func__, mpa_rev, mpa->revision);
1947 		err = -EPROTO;
1948 		goto err_stop_timer;
1949 	}
1950 
1951 	if (memcmp(mpa->key, MPA_KEY_REP, sizeof(mpa->key))) {
1952 
1953 		CTR2(KTR_IW_CXGBE, "%s:pmr7 %p", __func__, ep);
1954 		err = -EPROTO;
1955 		goto err_stop_timer;
1956 	}
1957 
1958 	plen = ntohs(mpa->private_data_size);
1959 
1960 	/*
1961 	 * Fail if there's too much private data.
1962 	 */
1963 	if (plen > MPA_MAX_PRIVATE_DATA) {
1964 
1965 		CTR2(KTR_IW_CXGBE, "%s:pmr8 %p", __func__, ep);
1966 		err = -EPROTO;
1967 		goto err_stop_timer;
1968 	}
1969 
1970 	/*
1971 	 * If plen does not account for pkt size
1972 	 */
1973 	if (ep->mpa_pkt_len > (sizeof(*mpa) + plen)) {
1974 
1975 		CTR2(KTR_IW_CXGBE, "%s:pmr9 %p", __func__, ep);
1976 		STOP_EP_TIMER(ep);
1977 		err = -EPROTO;
1978 		goto err_stop_timer;
1979 	}
1980 
1981 	ep->plen = (u8) plen;
1982 
1983 	/*
1984 	 * If we don't have all the pdata yet, then bail.
1985 	 * We'll continue process when more data arrives.
1986 	 */
1987 	if (ep->mpa_pkt_len < (sizeof(*mpa) + plen)) {
1988 
1989 		CTR2(KTR_IW_CXGBE, "%s:pmra %p", __func__, ep);
1990 		return 0;
1991 	}
1992 
1993 	if (mpa->flags & MPA_REJECT) {
1994 
1995 		CTR2(KTR_IW_CXGBE, "%s:pmrb %p", __func__, ep);
1996 		err = -ECONNREFUSED;
1997 		goto err_stop_timer;
1998 	}
1999 
2000 	/*
2001 	 * If we get here we have accumulated the entire mpa
2002 	 * start reply message including private data. And
2003 	 * the MPA header is valid.
2004 	 */
2005 	ep->com.state = FPDU_MODE;
2006 	ep->mpa_attr.crc_enabled = (mpa->flags & MPA_CRC) | crc_enabled ? 1 : 0;
2007 	ep->mpa_attr.recv_marker_enabled = markers_enabled;
2008 	ep->mpa_attr.xmit_marker_enabled = mpa->flags & MPA_MARKERS ? 1 : 0;
2009 	ep->mpa_attr.version = mpa->revision;
2010 	ep->mpa_attr.p2p_type = FW_RI_INIT_P2PTYPE_DISABLED;
2011 
2012 	if (mpa->revision == 2) {
2013 
2014 		CTR2(KTR_IW_CXGBE, "%s:pmrc %p", __func__, ep);
2015 		ep->mpa_attr.enhanced_rdma_conn =
2016 			mpa->flags & MPA_ENHANCED_RDMA_CONN ? 1 : 0;
2017 
2018 		if (ep->mpa_attr.enhanced_rdma_conn) {
2019 
2020 			CTR2(KTR_IW_CXGBE, "%s:pmrd %p", __func__, ep);
2021 			mpa_v2_params = (struct mpa_v2_conn_params *)
2022 				(ep->mpa_pkt + sizeof(*mpa));
2023 			resp_ird = ntohs(mpa_v2_params->ird) &
2024 				MPA_V2_IRD_ORD_MASK;
2025 			resp_ord = ntohs(mpa_v2_params->ord) &
2026 				MPA_V2_IRD_ORD_MASK;
2027 
2028 			/*
2029 			 * This is a double-check. Ideally, below checks are
2030 			 * not required since ird/ord stuff has been taken
2031 			 * care of in c4iw_accept_cr
2032 			 */
2033 			if (ep->ird < resp_ord) {
2034 				if (RELAXED_IRD_NEGOTIATION && resp_ord <=
2035 				   ep->com.dev->rdev.adap->params.max_ordird_qp)
2036 					ep->ird = resp_ord;
2037 				else
2038 					insuff_ird = 1;
2039 			} else if (ep->ird > resp_ord) {
2040 				ep->ird = resp_ord;
2041 			}
2042 			if (ep->ord > resp_ird) {
2043 				if (RELAXED_IRD_NEGOTIATION)
2044 					ep->ord = resp_ird;
2045 				else
2046 					insuff_ird = 1;
2047 			}
2048 			if (insuff_ird) {
2049 				err = -ENOMEM;
2050 				ep->ird = resp_ord;
2051 				ep->ord = resp_ird;
2052 			}
2053 
2054 			if (ntohs(mpa_v2_params->ird) &
2055 				MPA_V2_PEER2PEER_MODEL) {
2056 
2057 				CTR2(KTR_IW_CXGBE, "%s:pmrf %p", __func__, ep);
2058 				if (ntohs(mpa_v2_params->ord) &
2059 					MPA_V2_RDMA_WRITE_RTR) {
2060 
2061 					CTR2(KTR_IW_CXGBE, "%s:pmrg %p", __func__, ep);
2062 					ep->mpa_attr.p2p_type =
2063 						FW_RI_INIT_P2PTYPE_RDMA_WRITE;
2064 				}
2065 				else if (ntohs(mpa_v2_params->ord) &
2066 					MPA_V2_RDMA_READ_RTR) {
2067 
2068 					CTR2(KTR_IW_CXGBE, "%s:pmrh %p", __func__, ep);
2069 					ep->mpa_attr.p2p_type =
2070 						FW_RI_INIT_P2PTYPE_READ_REQ;
2071 				}
2072 			}
2073 		}
2074 	} else {
2075 
2076 		CTR2(KTR_IW_CXGBE, "%s:pmri %p", __func__, ep);
2077 
2078 		if (mpa->revision == 1) {
2079 
2080 			CTR2(KTR_IW_CXGBE, "%s:pmrj %p", __func__, ep);
2081 
2082 			if (peer2peer) {
2083 
2084 				CTR2(KTR_IW_CXGBE, "%s:pmrk %p", __func__, ep);
2085 				ep->mpa_attr.p2p_type = p2p_type;
2086 			}
2087 		}
2088 	}
2089 
2090 	if (set_tcpinfo(ep)) {
2091 
2092 		CTR2(KTR_IW_CXGBE, "%s:pmrl %p", __func__, ep);
2093 		printf("%s set_tcpinfo error\n", __func__);
2094 		err = -ECONNRESET;
2095 		goto err;
2096 	}
2097 
2098 	CTR6(KTR_IW_CXGBE, "%s - crc_enabled = %d, recv_marker_enabled = %d, "
2099 	    "xmit_marker_enabled = %d, version = %d p2p_type = %d", __func__,
2100 	    ep->mpa_attr.crc_enabled, ep->mpa_attr.recv_marker_enabled,
2101 	    ep->mpa_attr.xmit_marker_enabled, ep->mpa_attr.version,
2102 	    ep->mpa_attr.p2p_type);
2103 
2104 	/*
2105 	 * If responder's RTR does not match with that of initiator, assign
2106 	 * FW_RI_INIT_P2PTYPE_DISABLED in mpa attributes so that RTR is not
2107 	 * generated when moving QP to RTS state.
2108 	 * A TERM message will be sent after QP has moved to RTS state
2109 	 */
2110 	if ((ep->mpa_attr.version == 2) && peer2peer &&
2111 		(ep->mpa_attr.p2p_type != p2p_type)) {
2112 
2113 		CTR2(KTR_IW_CXGBE, "%s:pmrm %p", __func__, ep);
2114 		ep->mpa_attr.p2p_type = FW_RI_INIT_P2PTYPE_DISABLED;
2115 		rtr_mismatch = 1;
2116 	}
2117 
2118 
2119 	//ep->ofld_txq = TOEPCB(ep->com.so)->ofld_txq;
2120 	attrs.mpa_attr = ep->mpa_attr;
2121 	attrs.max_ird = ep->ird;
2122 	attrs.max_ord = ep->ord;
2123 	attrs.llp_stream_handle = ep;
2124 	attrs.next_state = C4IW_QP_STATE_RTS;
2125 
2126 	mask = C4IW_QP_ATTR_NEXT_STATE |
2127 		C4IW_QP_ATTR_LLP_STREAM_HANDLE | C4IW_QP_ATTR_MPA_ATTR |
2128 		C4IW_QP_ATTR_MAX_IRD | C4IW_QP_ATTR_MAX_ORD;
2129 
2130 	/* bind QP and TID with INIT_WR */
2131 	err = c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp, mask, &attrs, 1);
2132 
2133 	if (err) {
2134 
2135 		CTR2(KTR_IW_CXGBE, "%s:pmrn %p", __func__, ep);
2136 		goto err;
2137 	}
2138 
2139 	/*
2140 	 * If responder's RTR requirement did not match with what initiator
2141 	 * supports, generate TERM message
2142 	 */
2143 	if (rtr_mismatch) {
2144 
2145 		CTR2(KTR_IW_CXGBE, "%s:pmro %p", __func__, ep);
2146 		printk(KERN_ERR "%s: RTR mismatch, sending TERM\n", __func__);
2147 		attrs.layer_etype = LAYER_MPA | DDP_LLP;
2148 		attrs.ecode = MPA_NOMATCH_RTR;
2149 		attrs.next_state = C4IW_QP_STATE_TERMINATE;
2150 		attrs.send_term = 1;
2151 		err = c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
2152 			C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
2153 		err = -ENOMEM;
2154 		disconnect = 1;
2155 		goto out;
2156 	}
2157 
2158 	/*
2159 	 * Generate TERM if initiator IRD is not sufficient for responder
2160 	 * provided ORD. Currently, we do the same behaviour even when
2161 	 * responder provided IRD is also not sufficient as regards to
2162 	 * initiator ORD.
2163 	 */
2164 	if (insuff_ird) {
2165 
2166 		CTR2(KTR_IW_CXGBE, "%s:pmrp %p", __func__, ep);
2167 		printk(KERN_ERR "%s: Insufficient IRD, sending TERM\n",
2168 				__func__);
2169 		attrs.layer_etype = LAYER_MPA | DDP_LLP;
2170 		attrs.ecode = MPA_INSUFF_IRD;
2171 		attrs.next_state = C4IW_QP_STATE_TERMINATE;
2172 		attrs.send_term = 1;
2173 		err = c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
2174 			C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
2175 		err = -ENOMEM;
2176 		disconnect = 1;
2177 		goto out;
2178 	}
2179 	goto out;
2180 err_stop_timer:
2181 	STOP_EP_TIMER(ep);
2182 err:
2183 	disconnect = 2;
2184 out:
2185 	connect_reply_upcall(ep, err);
2186 	CTR2(KTR_IW_CXGBE, "%s:pmrE %p", __func__, ep);
2187 	return disconnect;
2188 }
2189 
2190 /*
2191  * process_mpa_request - process streaming mode MPA request
2192  *
2193  * Returns:
2194  *
2195  * 0 upon success indicating a connect request was delivered to the ULP
2196  * or the mpa request is incomplete but valid so far.
2197  *
2198  * 1 if a failure requires the caller to close the connection.
2199  *
2200  * 2 if a failure requires the caller to abort the connection.
2201  */
2202 static int
2203 process_mpa_request(struct c4iw_ep *ep)
2204 {
2205 	struct mpa_message *mpa;
2206 	struct mpa_v2_conn_params *mpa_v2_params;
2207 	u16 plen;
2208 	int flags = MSG_DONTWAIT;
2209 	int rc;
2210 	struct iovec iov;
2211 	struct uio uio;
2212 	enum c4iw_ep_state state = ep->com.state;
2213 
2214 	CTR3(KTR_IW_CXGBE, "%s: ep %p, state %s", __func__, ep, states[state]);
2215 
2216 	if (state != MPA_REQ_WAIT)
2217 		return 0;
2218 
2219 	iov.iov_base = &ep->mpa_pkt[ep->mpa_pkt_len];
2220 	iov.iov_len = sizeof(ep->mpa_pkt) - ep->mpa_pkt_len;
2221 	uio.uio_iov = &iov;
2222 	uio.uio_iovcnt = 1;
2223 	uio.uio_offset = 0;
2224 	uio.uio_resid = sizeof(ep->mpa_pkt) - ep->mpa_pkt_len;
2225 	uio.uio_segflg = UIO_SYSSPACE;
2226 	uio.uio_rw = UIO_READ;
2227 	uio.uio_td = NULL; /* uio.uio_td = ep->com.thread; */
2228 
2229 	rc = soreceive(ep->com.so, NULL, &uio, NULL, NULL, &flags);
2230 	if (rc == EAGAIN)
2231 		return 0;
2232 	else if (rc)
2233 		goto err_stop_timer;
2234 
2235 	KASSERT(uio.uio_offset > 0, ("%s: sorecieve on so %p read no data",
2236 	    __func__, ep->com.so));
2237 	ep->mpa_pkt_len += uio.uio_offset;
2238 
2239 	/*
2240 	 * If we get more than the supported amount of private data then we must
2241 	 * fail this connection.  XXX: check so_rcv->sb_cc, or peek with another
2242 	 * soreceive, or increase the size of mpa_pkt by 1 and abort if the last
2243 	 * byte is filled by the soreceive above.
2244 	 */
2245 
2246 	/* Don't even have the MPA message.  Wait for more data to arrive. */
2247 	if (ep->mpa_pkt_len < sizeof(*mpa))
2248 		return 0;
2249 	mpa = (struct mpa_message *) ep->mpa_pkt;
2250 
2251 	/*
2252 	 * Validate MPA Header.
2253 	 */
2254 	if (mpa->revision > mpa_rev) {
2255 		log(LOG_ERR, "%s: MPA version mismatch. Local = %d,"
2256 		    " Received = %d\n", __func__, mpa_rev, mpa->revision);
2257 		goto err_stop_timer;
2258 	}
2259 
2260 	if (memcmp(mpa->key, MPA_KEY_REQ, sizeof(mpa->key)))
2261 		goto err_stop_timer;
2262 
2263 	/*
2264 	 * Fail if there's too much private data.
2265 	 */
2266 	plen = ntohs(mpa->private_data_size);
2267 	if (plen > MPA_MAX_PRIVATE_DATA)
2268 		goto err_stop_timer;
2269 
2270 	/*
2271 	 * If plen does not account for pkt size
2272 	 */
2273 	if (ep->mpa_pkt_len > (sizeof(*mpa) + plen))
2274 		goto err_stop_timer;
2275 
2276 	ep->plen = (u8) plen;
2277 
2278 	/*
2279 	 * If we don't have all the pdata yet, then bail.
2280 	 */
2281 	if (ep->mpa_pkt_len < (sizeof(*mpa) + plen))
2282 		return 0;
2283 
2284 	/*
2285 	 * If we get here we have accumulated the entire mpa
2286 	 * start reply message including private data.
2287 	 */
2288 	ep->mpa_attr.initiator = 0;
2289 	ep->mpa_attr.crc_enabled = (mpa->flags & MPA_CRC) | crc_enabled ? 1 : 0;
2290 	ep->mpa_attr.recv_marker_enabled = markers_enabled;
2291 	ep->mpa_attr.xmit_marker_enabled = mpa->flags & MPA_MARKERS ? 1 : 0;
2292 	ep->mpa_attr.version = mpa->revision;
2293 	if (mpa->revision == 1)
2294 		ep->tried_with_mpa_v1 = 1;
2295 	ep->mpa_attr.p2p_type = FW_RI_INIT_P2PTYPE_DISABLED;
2296 
2297 	if (mpa->revision == 2) {
2298 		ep->mpa_attr.enhanced_rdma_conn =
2299 		    mpa->flags & MPA_ENHANCED_RDMA_CONN ? 1 : 0;
2300 		if (ep->mpa_attr.enhanced_rdma_conn) {
2301 			mpa_v2_params = (struct mpa_v2_conn_params *)
2302 				(ep->mpa_pkt + sizeof(*mpa));
2303 			ep->ird = ntohs(mpa_v2_params->ird) &
2304 				MPA_V2_IRD_ORD_MASK;
2305 			ep->ird = min_t(u32, ep->ird,
2306 					cur_max_read_depth(ep->com.dev));
2307 			ep->ord = ntohs(mpa_v2_params->ord) &
2308 				MPA_V2_IRD_ORD_MASK;
2309 			ep->ord = min_t(u32, ep->ord,
2310 					cur_max_read_depth(ep->com.dev));
2311 			CTR3(KTR_IW_CXGBE, "%s initiator ird %u ord %u",
2312 				 __func__, ep->ird, ep->ord);
2313 			if (ntohs(mpa_v2_params->ird) & MPA_V2_PEER2PEER_MODEL)
2314 				if (peer2peer) {
2315 					if (ntohs(mpa_v2_params->ord) &
2316 							MPA_V2_RDMA_WRITE_RTR)
2317 						ep->mpa_attr.p2p_type =
2318 						FW_RI_INIT_P2PTYPE_RDMA_WRITE;
2319 					else if (ntohs(mpa_v2_params->ord) &
2320 							MPA_V2_RDMA_READ_RTR)
2321 						ep->mpa_attr.p2p_type =
2322 						FW_RI_INIT_P2PTYPE_READ_REQ;
2323 				}
2324 		}
2325 	} else if (mpa->revision == 1 && peer2peer)
2326 		ep->mpa_attr.p2p_type = p2p_type;
2327 
2328 	if (set_tcpinfo(ep))
2329 		goto err_stop_timer;
2330 
2331 	CTR5(KTR_IW_CXGBE, "%s: crc_enabled = %d, recv_marker_enabled = %d, "
2332 	    "xmit_marker_enabled = %d, version = %d", __func__,
2333 	    ep->mpa_attr.crc_enabled, ep->mpa_attr.recv_marker_enabled,
2334 	    ep->mpa_attr.xmit_marker_enabled, ep->mpa_attr.version);
2335 
2336 	ep->com.state = MPA_REQ_RCVD;
2337 	STOP_EP_TIMER(ep);
2338 
2339 	/* drive upcall */
2340 	if (ep->parent_ep->com.state != DEAD)
2341 		if (connect_request_upcall(ep))
2342 			goto err_out;
2343 	return 0;
2344 
2345 err_stop_timer:
2346 	STOP_EP_TIMER(ep);
2347 err_out:
2348 	return 2;
2349 }
2350 
2351 /*
2352  * Upcall from the adapter indicating data has been transmitted.
2353  * For us its just the single MPA request or reply.  We can now free
2354  * the skb holding the mpa message.
2355  */
2356 int c4iw_reject_cr(struct iw_cm_id *cm_id, const void *pdata, u8 pdata_len)
2357 {
2358 #ifdef KTR
2359 	int err;
2360 #endif
2361 	struct c4iw_ep *ep = to_ep(cm_id);
2362 	int abort = 0;
2363 
2364 	mutex_lock(&ep->com.mutex);
2365 	CTR2(KTR_IW_CXGBE, "%s:crcB %p", __func__, ep);
2366 
2367 	if ((ep->com.state == DEAD) ||
2368 			(ep->com.state != MPA_REQ_RCVD)) {
2369 
2370 		CTR2(KTR_IW_CXGBE, "%s:crc1 %p", __func__, ep);
2371 		mutex_unlock(&ep->com.mutex);
2372 		c4iw_put_ep(&ep->com);
2373 		return -ECONNRESET;
2374 	}
2375 	set_bit(ULP_REJECT, &ep->com.history);
2376 
2377 	if (mpa_rev == 0) {
2378 
2379 		CTR2(KTR_IW_CXGBE, "%s:crc2 %p", __func__, ep);
2380 		abort = 1;
2381 	}
2382 	else {
2383 
2384 		CTR2(KTR_IW_CXGBE, "%s:crc3 %p", __func__, ep);
2385 		abort = send_mpa_reject(ep, pdata, pdata_len);
2386 	}
2387 	STOP_EP_TIMER(ep);
2388 #ifdef KTR
2389 	err = c4iw_ep_disconnect(ep, abort != 0, GFP_KERNEL);
2390 #else
2391 	c4iw_ep_disconnect(ep, abort != 0, GFP_KERNEL);
2392 #endif
2393 	mutex_unlock(&ep->com.mutex);
2394 	c4iw_put_ep(&ep->com);
2395 	CTR3(KTR_IW_CXGBE, "%s:crc4 %p, err: %d", __func__, ep, err);
2396 	return 0;
2397 }
2398 
2399 int c4iw_accept_cr(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param)
2400 {
2401 	int err;
2402 	struct c4iw_qp_attributes attrs = {0};
2403 	enum c4iw_qp_attr_mask mask;
2404 	struct c4iw_ep *ep = to_ep(cm_id);
2405 	struct c4iw_dev *h = to_c4iw_dev(cm_id->device);
2406 	struct c4iw_qp *qp = get_qhp(h, conn_param->qpn);
2407 	int abort = 0;
2408 
2409 	mutex_lock(&ep->com.mutex);
2410 	CTR2(KTR_IW_CXGBE, "%s:cacB %p", __func__, ep);
2411 
2412 	if ((ep->com.state == DEAD) ||
2413 			(ep->com.state != MPA_REQ_RCVD)) {
2414 
2415 		CTR2(KTR_IW_CXGBE, "%s:cac1 %p", __func__, ep);
2416 		err = -ECONNRESET;
2417 		goto err_out;
2418 	}
2419 
2420 	BUG_ON(!qp);
2421 
2422 	set_bit(ULP_ACCEPT, &ep->com.history);
2423 
2424 	if ((conn_param->ord > c4iw_max_read_depth) ||
2425 		(conn_param->ird > c4iw_max_read_depth)) {
2426 
2427 		CTR2(KTR_IW_CXGBE, "%s:cac2 %p", __func__, ep);
2428 		err = -EINVAL;
2429 		goto err_abort;
2430 	}
2431 
2432 	if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) {
2433 
2434 		CTR2(KTR_IW_CXGBE, "%s:cac3 %p", __func__, ep);
2435 
2436 		if (conn_param->ord > ep->ird) {
2437 			if (RELAXED_IRD_NEGOTIATION) {
2438 				conn_param->ord = ep->ird;
2439 			} else {
2440 				ep->ird = conn_param->ird;
2441 				ep->ord = conn_param->ord;
2442 				send_mpa_reject(ep, conn_param->private_data,
2443 						conn_param->private_data_len);
2444 				err = -ENOMEM;
2445 				goto err_abort;
2446 			}
2447 		}
2448 		if (conn_param->ird < ep->ord) {
2449 			if (RELAXED_IRD_NEGOTIATION &&
2450 			    ep->ord <= h->rdev.adap->params.max_ordird_qp) {
2451 				conn_param->ird = ep->ord;
2452 			} else {
2453 				err = -ENOMEM;
2454 				goto err_abort;
2455 			}
2456 		}
2457 	}
2458 	ep->ird = conn_param->ird;
2459 	ep->ord = conn_param->ord;
2460 
2461 	if (ep->mpa_attr.version == 1) {
2462 		if (peer2peer && ep->ird == 0)
2463 			ep->ird = 1;
2464 	} else {
2465 		if (peer2peer &&
2466 		    (ep->mpa_attr.p2p_type != FW_RI_INIT_P2PTYPE_DISABLED) &&
2467 		    (p2p_type == FW_RI_INIT_P2PTYPE_READ_REQ) && ep->ird == 0)
2468 			ep->ird = 1;
2469 	}
2470 
2471 	CTR4(KTR_IW_CXGBE, "%s %d ird %d ord %d", __func__, __LINE__,
2472 			ep->ird, ep->ord);
2473 
2474 	ep->com.cm_id = cm_id;
2475 	ref_cm_id(&ep->com);
2476 	ep->com.qp = qp;
2477 	ref_qp(ep);
2478 	//ep->ofld_txq = TOEPCB(ep->com.so)->ofld_txq;
2479 
2480 	/* bind QP to EP and move to RTS */
2481 	attrs.mpa_attr = ep->mpa_attr;
2482 	attrs.max_ird = ep->ird;
2483 	attrs.max_ord = ep->ord;
2484 	attrs.llp_stream_handle = ep;
2485 	attrs.next_state = C4IW_QP_STATE_RTS;
2486 
2487 	/* bind QP and TID with INIT_WR */
2488 	mask = C4IW_QP_ATTR_NEXT_STATE |
2489 		C4IW_QP_ATTR_LLP_STREAM_HANDLE |
2490 		C4IW_QP_ATTR_MPA_ATTR |
2491 		C4IW_QP_ATTR_MAX_IRD |
2492 		C4IW_QP_ATTR_MAX_ORD;
2493 
2494 	err = c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp, mask, &attrs, 1);
2495 	if (err) {
2496 		CTR3(KTR_IW_CXGBE, "%s:caca %p, err: %d", __func__, ep, err);
2497 		goto err_defef_cm_id;
2498 	}
2499 
2500 	err = send_mpa_reply(ep, conn_param->private_data,
2501 			conn_param->private_data_len);
2502 	if (err) {
2503 		CTR3(KTR_IW_CXGBE, "%s:cacb %p, err: %d", __func__, ep, err);
2504 		goto err_defef_cm_id;
2505 	}
2506 
2507 	ep->com.state = FPDU_MODE;
2508 	established_upcall(ep);
2509 	mutex_unlock(&ep->com.mutex);
2510 	c4iw_put_ep(&ep->com);
2511 	CTR2(KTR_IW_CXGBE, "%s:cacE %p", __func__, ep);
2512 	return 0;
2513 err_defef_cm_id:
2514 	deref_cm_id(&ep->com);
2515 err_abort:
2516 	abort = 1;
2517 err_out:
2518 	if (abort)
2519 		c4iw_ep_disconnect(ep, 1, GFP_KERNEL);
2520 	mutex_unlock(&ep->com.mutex);
2521 	c4iw_put_ep(&ep->com);
2522 	CTR2(KTR_IW_CXGBE, "%s:cacE err %p", __func__, ep);
2523 	return err;
2524 }
2525 
2526 static int
2527 c4iw_sock_create(struct sockaddr_storage *laddr, struct socket **so)
2528 {
2529 	int ret;
2530 	int size, on;
2531 	struct socket *sock = NULL;
2532 	struct sockopt sopt;
2533 
2534 	ret = sock_create_kern(laddr->ss_family,
2535 			SOCK_STREAM, IPPROTO_TCP, &sock);
2536 	if (ret) {
2537 		CTR2(KTR_IW_CXGBE, "%s:Failed to create TCP socket. err %d",
2538 				__func__, ret);
2539 		return ret;
2540 	}
2541 
2542 	if (reuseaddr) {
2543 		bzero(&sopt, sizeof(struct sockopt));
2544 		sopt.sopt_dir = SOPT_SET;
2545 		sopt.sopt_level = SOL_SOCKET;
2546 		sopt.sopt_name = SO_REUSEADDR;
2547 		on = 1;
2548 		sopt.sopt_val = &on;
2549 		sopt.sopt_valsize = sizeof(on);
2550 		ret = -sosetopt(sock, &sopt);
2551 		if (ret != 0) {
2552 			log(LOG_ERR, "%s: sosetopt(%p, SO_REUSEADDR) "
2553 				"failed with %d.\n", __func__, sock, ret);
2554 		}
2555 		bzero(&sopt, sizeof(struct sockopt));
2556 		sopt.sopt_dir = SOPT_SET;
2557 		sopt.sopt_level = SOL_SOCKET;
2558 		sopt.sopt_name = SO_REUSEPORT;
2559 		on = 1;
2560 		sopt.sopt_val = &on;
2561 		sopt.sopt_valsize = sizeof(on);
2562 		ret = -sosetopt(sock, &sopt);
2563 		if (ret != 0) {
2564 			log(LOG_ERR, "%s: sosetopt(%p, SO_REUSEPORT) "
2565 				"failed with %d.\n", __func__, sock, ret);
2566 		}
2567 	}
2568 
2569 	ret = -sobind(sock, (struct sockaddr *)laddr, curthread);
2570 	if (ret) {
2571 		CTR2(KTR_IW_CXGBE, "%s:Failed to bind socket. err %p",
2572 				__func__, ret);
2573 		sock_release(sock);
2574 		return ret;
2575 	}
2576 
2577 	size = laddr->ss_family == AF_INET6 ?
2578 		sizeof(struct sockaddr_in6) : sizeof(struct sockaddr_in);
2579 	ret = sock_getname(sock, (struct sockaddr *)laddr, &size, 0);
2580 	if (ret) {
2581 		CTR2(KTR_IW_CXGBE, "%s:sock_getname failed. err %p",
2582 				__func__, ret);
2583 		sock_release(sock);
2584 		return ret;
2585 	}
2586 
2587 	*so = sock;
2588 	return 0;
2589 }
2590 
2591 int c4iw_connect(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param)
2592 {
2593 	int err = 0;
2594 	struct c4iw_dev *dev = to_c4iw_dev(cm_id->device);
2595 	struct c4iw_ep *ep = NULL;
2596 	struct ifnet    *nh_ifp;        /* Logical egress interface */
2597 	struct epoch_tracker et;
2598 #ifdef VIMAGE
2599 	struct rdma_cm_id *rdma_id = (struct rdma_cm_id*)cm_id->context;
2600 	struct vnet *vnet = rdma_id->route.addr.dev_addr.net;
2601 #endif
2602 
2603 	CTR2(KTR_IW_CXGBE, "%s:ccB %p", __func__, cm_id);
2604 
2605 
2606 	if ((conn_param->ord > c4iw_max_read_depth) ||
2607 		(conn_param->ird > c4iw_max_read_depth)) {
2608 
2609 		CTR2(KTR_IW_CXGBE, "%s:cc1 %p", __func__, cm_id);
2610 		err = -EINVAL;
2611 		goto out;
2612 	}
2613 	ep = alloc_ep(sizeof(*ep), GFP_KERNEL);
2614 	cm_id->provider_data = ep;
2615 
2616 	init_timer(&ep->timer);
2617 	ep->plen = conn_param->private_data_len;
2618 
2619 	if (ep->plen) {
2620 
2621 		CTR2(KTR_IW_CXGBE, "%s:cc3 %p", __func__, ep);
2622 		memcpy(ep->mpa_pkt + sizeof(struct mpa_message),
2623 				conn_param->private_data, ep->plen);
2624 	}
2625 	ep->ird = conn_param->ird;
2626 	ep->ord = conn_param->ord;
2627 
2628 	if (peer2peer && ep->ord == 0) {
2629 
2630 		CTR2(KTR_IW_CXGBE, "%s:cc4 %p", __func__, ep);
2631 		ep->ord = 1;
2632 	}
2633 
2634 	ep->com.dev = dev;
2635 	ep->com.cm_id = cm_id;
2636 	ref_cm_id(&ep->com);
2637 	ep->com.qp = get_qhp(dev, conn_param->qpn);
2638 
2639 	if (!ep->com.qp) {
2640 
2641 		CTR2(KTR_IW_CXGBE, "%s:cc5 %p", __func__, ep);
2642 		err = -EINVAL;
2643 		goto fail;
2644 	}
2645 	ref_qp(ep);
2646 	ep->com.thread = curthread;
2647 
2648 	NET_EPOCH_ENTER(et);
2649 	CURVNET_SET(vnet);
2650 	err = get_ifnet_from_raddr(&cm_id->remote_addr, &nh_ifp);
2651 	CURVNET_RESTORE();
2652 	NET_EPOCH_EXIT(et);
2653 
2654 	if (err) {
2655 
2656 		CTR2(KTR_IW_CXGBE, "%s:cc7 %p", __func__, ep);
2657 		printk(KERN_ERR MOD "%s - cannot find route.\n", __func__);
2658 		err = EHOSTUNREACH;
2659 		return err;
2660 	}
2661 
2662 	if (!(nh_ifp->if_capenable & IFCAP_TOE) ||
2663 	    TOEDEV(nh_ifp) == NULL) {
2664 		err = -ENOPROTOOPT;
2665 		goto fail;
2666 	}
2667 	ep->com.state = CONNECTING;
2668 	ep->tos = 0;
2669 	ep->com.local_addr = cm_id->local_addr;
2670 	ep->com.remote_addr = cm_id->remote_addr;
2671 
2672 	err = c4iw_sock_create(&cm_id->local_addr, &ep->com.so);
2673 	if (err)
2674 		goto fail;
2675 
2676 	setiwsockopt(ep->com.so);
2677 	init_iwarp_socket(ep->com.so, &ep->com);
2678 	err = -soconnect(ep->com.so, (struct sockaddr *)&ep->com.remote_addr,
2679 		ep->com.thread);
2680 	if (err)
2681 		goto fail_free_so;
2682 	CTR2(KTR_IW_CXGBE, "%s:ccE, ep %p", __func__, ep);
2683 	return 0;
2684 
2685 fail_free_so:
2686 	uninit_iwarp_socket(ep->com.so);
2687 	ep->com.state = DEAD;
2688 	sock_release(ep->com.so);
2689 fail:
2690 	deref_cm_id(&ep->com);
2691 	c4iw_put_ep(&ep->com);
2692 	ep = NULL;
2693 out:
2694 	CTR2(KTR_IW_CXGBE, "%s:ccE Error %d", __func__, err);
2695 	return err;
2696 }
2697 
2698 /*
2699  * iwcm->create_listen.  Returns -errno on failure.
2700  */
2701 int
2702 c4iw_create_listen(struct iw_cm_id *cm_id, int backlog)
2703 {
2704 	struct c4iw_dev *dev = to_c4iw_dev(cm_id->device);
2705 	struct c4iw_listen_ep *lep = NULL;
2706 	struct listen_port_info *port_info = NULL;
2707 	int rc = 0;
2708 
2709 	CTR3(KTR_IW_CXGBE, "%s: cm_id %p, backlog %s", __func__, cm_id,
2710 			backlog);
2711 	if (c4iw_fatal_error(&dev->rdev)) {
2712 		CTR2(KTR_IW_CXGBE, "%s: cm_id %p, fatal error", __func__,
2713 			       cm_id);
2714 		return -EIO;
2715 	}
2716 	lep = alloc_ep(sizeof(*lep), GFP_KERNEL);
2717 	lep->com.cm_id = cm_id;
2718 	ref_cm_id(&lep->com);
2719 	lep->com.dev = dev;
2720 	lep->backlog = backlog;
2721 	lep->com.local_addr = cm_id->local_addr;
2722 	lep->com.thread = curthread;
2723 	cm_id->provider_data = lep;
2724 	lep->com.state = LISTEN;
2725 
2726 	/* In case of INDADDR_ANY, ibcore creates cmid for each device and
2727 	 * invokes iw_cxgbe listener callbacks assuming that iw_cxgbe creates
2728 	 * HW listeners for each device seperately. But toecore expects single
2729 	 * solisten() call with INADDR_ANY address to create HW listeners on
2730 	 * all devices for a given port number. So iw_cxgbe driver calls
2731 	 * solisten() only once for INADDR_ANY(usually done at first time
2732 	 * listener callback from ibcore). And all the subsequent INADDR_ANY
2733 	 * listener callbacks from ibcore(for the same port address) do not
2734 	 * invoke solisten() as first listener callback has already created
2735 	 * listeners for all other devices(via solisten).
2736 	 */
2737 	if (c4iw_any_addr((struct sockaddr *)&lep->com.local_addr)) {
2738 		port_info = add_ep_to_listenlist(lep);
2739 		/* skip solisten() if refcnt > 1, as the listeners were
2740 		 * already created by 'Master lep'
2741 		 */
2742 		if (port_info->refcnt > 1) {
2743 			/* As there will be only one listener socket for a TCP
2744 			 * port, copy Master lep's socket pointer to other lep's
2745 			 * that are belonging to same TCP port.
2746 			 */
2747 			struct c4iw_listen_ep *head_lep =
2748 					container_of(port_info->lep_list.next,
2749 					struct c4iw_listen_ep, listen_ep_list);
2750 			lep->com.so =  head_lep->com.so;
2751 			goto out;
2752 		}
2753 	}
2754 	rc = c4iw_sock_create(&cm_id->local_addr, &lep->com.so);
2755 	if (rc) {
2756 		CTR2(KTR_IW_CXGBE, "%s:Failed to create socket. err %d",
2757 				__func__, rc);
2758 		goto fail;
2759 	}
2760 
2761 	rc = -solisten(lep->com.so, backlog, curthread);
2762 	if (rc) {
2763 		CTR3(KTR_IW_CXGBE, "%s:Failed to listen on sock:%p. err %d",
2764 				__func__, lep->com.so, rc);
2765 		goto fail_free_so;
2766 	}
2767 	init_iwarp_socket(lep->com.so, &lep->com);
2768 out:
2769 	return 0;
2770 
2771 fail_free_so:
2772 	sock_release(lep->com.so);
2773 fail:
2774 	if (port_info)
2775 		rem_ep_from_listenlist(lep);
2776 	deref_cm_id(&lep->com);
2777 	c4iw_put_ep(&lep->com);
2778 	return rc;
2779 }
2780 
2781 int
2782 c4iw_destroy_listen(struct iw_cm_id *cm_id)
2783 {
2784 	struct c4iw_listen_ep *lep = to_listen_ep(cm_id);
2785 
2786 	mutex_lock(&lep->com.mutex);
2787 	CTR3(KTR_IW_CXGBE, "%s: cm_id %p, state %s", __func__, cm_id,
2788 	    states[lep->com.state]);
2789 
2790 	lep->com.state = DEAD;
2791 	if (c4iw_any_addr((struct sockaddr *)&lep->com.local_addr)) {
2792 		/* if no refcount then close listen socket */
2793 		if (!rem_ep_from_listenlist(lep))
2794 			close_socket(lep->com.so);
2795 	} else
2796 		close_socket(lep->com.so);
2797 	deref_cm_id(&lep->com);
2798 	mutex_unlock(&lep->com.mutex);
2799 	c4iw_put_ep(&lep->com);
2800 	return 0;
2801 }
2802 
2803 int __c4iw_ep_disconnect(struct c4iw_ep *ep, int abrupt, gfp_t gfp)
2804 {
2805 	int ret;
2806 	mutex_lock(&ep->com.mutex);
2807 	ret = c4iw_ep_disconnect(ep, abrupt, gfp);
2808 	mutex_unlock(&ep->com.mutex);
2809 	return ret;
2810 }
2811 
2812 int c4iw_ep_disconnect(struct c4iw_ep *ep, int abrupt, gfp_t gfp)
2813 {
2814 	int ret = 0;
2815 	int close = 0;
2816 	struct c4iw_rdev *rdev;
2817 
2818 
2819 	CTR2(KTR_IW_CXGBE, "%s:cedB %p", __func__, ep);
2820 
2821 	rdev = &ep->com.dev->rdev;
2822 
2823 	if (c4iw_fatal_error(rdev)) {
2824 		CTR3(KTR_IW_CXGBE, "%s:ced1 fatal error %p %s", __func__, ep,
2825 					states[ep->com.state]);
2826 		if (ep->com.state != DEAD) {
2827 			send_abort(ep);
2828 			ep->com.state = DEAD;
2829 		}
2830 		close_complete_upcall(ep, -ECONNRESET);
2831 		return ECONNRESET;
2832 	}
2833 	CTR3(KTR_IW_CXGBE, "%s:ced2 %p %s", __func__, ep,
2834 	    states[ep->com.state]);
2835 
2836 	/*
2837 	 * Ref the ep here in case we have fatal errors causing the
2838 	 * ep to be released and freed.
2839 	 */
2840 	c4iw_get_ep(&ep->com);
2841 	switch (ep->com.state) {
2842 
2843 		case MPA_REQ_WAIT:
2844 		case MPA_REQ_SENT:
2845 		case MPA_REQ_RCVD:
2846 		case MPA_REP_SENT:
2847 		case FPDU_MODE:
2848 			close = 1;
2849 			if (abrupt)
2850 				ep->com.state = ABORTING;
2851 			else {
2852 				ep->com.state = CLOSING;
2853 				START_EP_TIMER(ep);
2854 			}
2855 			set_bit(CLOSE_SENT, &ep->com.flags);
2856 			break;
2857 
2858 		case CLOSING:
2859 
2860 			if (!test_and_set_bit(CLOSE_SENT, &ep->com.flags)) {
2861 
2862 				close = 1;
2863 				if (abrupt) {
2864 					STOP_EP_TIMER(ep);
2865 					ep->com.state = ABORTING;
2866 				} else
2867 					ep->com.state = MORIBUND;
2868 			}
2869 			break;
2870 
2871 		case MORIBUND:
2872 		case ABORTING:
2873 		case DEAD:
2874 			CTR3(KTR_IW_CXGBE,
2875 			    "%s ignoring disconnect ep %p state %u", __func__,
2876 			    ep, ep->com.state);
2877 			break;
2878 
2879 		default:
2880 			BUG();
2881 			break;
2882 	}
2883 
2884 
2885 	if (close) {
2886 
2887 		CTR2(KTR_IW_CXGBE, "%s:ced3 %p", __func__, ep);
2888 
2889 		if (abrupt) {
2890 
2891 			CTR2(KTR_IW_CXGBE, "%s:ced4 %p", __func__, ep);
2892 			set_bit(EP_DISC_ABORT, &ep->com.history);
2893 			close_complete_upcall(ep, -ECONNRESET);
2894 			send_abort(ep);
2895 		} else {
2896 
2897 			CTR2(KTR_IW_CXGBE, "%s:ced5 %p", __func__, ep);
2898 			set_bit(EP_DISC_CLOSE, &ep->com.history);
2899 
2900 			if (!ep->parent_ep)
2901 				ep->com.state = MORIBUND;
2902 
2903 			CURVNET_SET(ep->com.so->so_vnet);
2904 			ret = sodisconnect(ep->com.so);
2905 			CURVNET_RESTORE();
2906 			if (ret) {
2907 				CTR2(KTR_IW_CXGBE, "%s:ced6 %p", __func__, ep);
2908 				STOP_EP_TIMER(ep);
2909 				send_abort(ep);
2910 				ep->com.state = DEAD;
2911 				close_complete_upcall(ep, -ECONNRESET);
2912 				set_bit(EP_DISC_FAIL, &ep->com.history);
2913 				if (ep->com.qp) {
2914 					struct c4iw_qp_attributes attrs = {0};
2915 
2916 					attrs.next_state = C4IW_QP_STATE_ERROR;
2917 					ret = c4iw_modify_qp(
2918 							ep->com.dev, ep->com.qp,
2919 							C4IW_QP_ATTR_NEXT_STATE,
2920 							&attrs, 1);
2921 					CTR3(KTR_IW_CXGBE, "%s:ced7 %p ret %d",
2922 						__func__, ep, ret);
2923 				}
2924 			}
2925 		}
2926 	}
2927 	c4iw_put_ep(&ep->com);
2928 	CTR2(KTR_IW_CXGBE, "%s:cedE %p", __func__, ep);
2929 	return ret;
2930 }
2931 
2932 #ifdef C4IW_EP_REDIRECT
2933 int c4iw_ep_redirect(void *ctx, struct dst_entry *old, struct dst_entry *new,
2934 		struct l2t_entry *l2t)
2935 {
2936 	struct c4iw_ep *ep = ctx;
2937 
2938 	if (ep->dst != old)
2939 		return 0;
2940 
2941 	PDBG("%s ep %p redirect to dst %p l2t %p\n", __func__, ep, new,
2942 			l2t);
2943 	dst_hold(new);
2944 	cxgb4_l2t_release(ep->l2t);
2945 	ep->l2t = l2t;
2946 	dst_release(old);
2947 	ep->dst = new;
2948 	return 1;
2949 }
2950 #endif
2951 
2952 
2953 
2954 static void ep_timeout(unsigned long arg)
2955 {
2956 	struct c4iw_ep *ep = (struct c4iw_ep *)arg;
2957 
2958 	if (!test_and_set_bit(TIMEOUT, &ep->com.flags)) {
2959 
2960 		/*
2961 		 * Only insert if it is not already on the list.
2962 		 */
2963 		if (!(ep->com.ep_events & C4IW_EVENT_TIMEOUT)) {
2964 			CTR2(KTR_IW_CXGBE, "%s:et1 %p", __func__, ep);
2965 			add_ep_to_req_list(ep, C4IW_EVENT_TIMEOUT);
2966 		}
2967 	}
2968 }
2969 
2970 static int fw6_wr_rpl(struct adapter *sc, const __be64 *rpl)
2971 {
2972 	uint64_t val = be64toh(*rpl);
2973 	int ret;
2974 	struct c4iw_wr_wait *wr_waitp;
2975 
2976 	ret = (int)((val >> 8) & 0xff);
2977 	wr_waitp = (struct c4iw_wr_wait *)rpl[1];
2978 	CTR3(KTR_IW_CXGBE, "%s wr_waitp %p ret %u", __func__, wr_waitp, ret);
2979 	if (wr_waitp)
2980 		c4iw_wake_up(wr_waitp, ret ? -ret : 0);
2981 
2982 	return (0);
2983 }
2984 
2985 static int fw6_cqe_handler(struct adapter *sc, const __be64 *rpl)
2986 {
2987 	struct cqe_list_entry *cle;
2988 	unsigned long flag;
2989 
2990 	cle = malloc(sizeof(*cle), M_CXGBE, M_NOWAIT);
2991 	cle->rhp = sc->iwarp_softc;
2992 	cle->err_cqe = *(const struct t4_cqe *)(&rpl[0]);
2993 
2994 	spin_lock_irqsave(&err_cqe_lock, flag);
2995 	list_add_tail(&cle->entry, &err_cqe_list);
2996 	queue_work(c4iw_taskq, &c4iw_task);
2997 	spin_unlock_irqrestore(&err_cqe_lock, flag);
2998 
2999 	return (0);
3000 }
3001 
3002 static int
3003 process_terminate(struct c4iw_ep *ep)
3004 {
3005 	struct c4iw_qp_attributes attrs = {0};
3006 
3007 	CTR2(KTR_IW_CXGBE, "%s:tB %p %d", __func__, ep);
3008 
3009 	if (ep && ep->com.qp) {
3010 
3011 		printk(KERN_WARNING MOD "TERM received tid %u qpid %u\n",
3012 				ep->hwtid, ep->com.qp->wq.sq.qid);
3013 		attrs.next_state = C4IW_QP_STATE_TERMINATE;
3014 		c4iw_modify_qp(ep->com.dev, ep->com.qp, C4IW_QP_ATTR_NEXT_STATE, &attrs,
3015 				1);
3016 	} else
3017 		printk(KERN_WARNING MOD "TERM received tid %u no ep/qp\n",
3018 								ep->hwtid);
3019 	CTR2(KTR_IW_CXGBE, "%s:tE %p %d", __func__, ep);
3020 
3021 	return 0;
3022 }
3023 
3024 int __init c4iw_cm_init(void)
3025 {
3026 
3027 	t4_register_cpl_handler(CPL_RDMA_TERMINATE, terminate);
3028 	t4_register_fw_msg_handler(FW6_TYPE_WR_RPL, fw6_wr_rpl);
3029 	t4_register_fw_msg_handler(FW6_TYPE_CQE, fw6_cqe_handler);
3030 	t4_register_an_handler(c4iw_ev_handler);
3031 
3032 	TAILQ_INIT(&req_list);
3033 	spin_lock_init(&req_lock);
3034 	INIT_LIST_HEAD(&err_cqe_list);
3035 	spin_lock_init(&err_cqe_lock);
3036 
3037 	INIT_WORK(&c4iw_task, process_req);
3038 
3039 	c4iw_taskq = create_singlethread_workqueue("iw_cxgbe");
3040 	if (!c4iw_taskq)
3041 		return -ENOMEM;
3042 
3043 	return 0;
3044 }
3045 
3046 void __exit c4iw_cm_term(void)
3047 {
3048 	WARN_ON(!TAILQ_EMPTY(&req_list));
3049 	WARN_ON(!list_empty(&err_cqe_list));
3050 	flush_workqueue(c4iw_taskq);
3051 	destroy_workqueue(c4iw_taskq);
3052 
3053 	t4_register_cpl_handler(CPL_RDMA_TERMINATE, NULL);
3054 	t4_register_fw_msg_handler(FW6_TYPE_WR_RPL, NULL);
3055 	t4_register_fw_msg_handler(FW6_TYPE_CQE, NULL);
3056 	t4_register_an_handler(NULL);
3057 }
3058 #endif
3059