• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 /*
2  * Copyright (c) 2009-2014 Chelsio, Inc. All rights reserved.
3  *
4  * This software is available to you under a choice of one of two
5  * licenses.  You may choose to be licensed under the terms of the GNU
6  * General Public License (GPL) Version 2, available from the file
7  * COPYING in the main directory of this source tree, or the
8  * OpenIB.org BSD license below:
9  *
10  *     Redistribution and use in source and binary forms, with or
11  *     without modification, are permitted provided that the following
12  *     conditions are met:
13  *
14  *      - Redistributions of source code must retain the above
15  *	  copyright notice, this list of conditions and the following
16  *	  disclaimer.
17  *
18  *      - Redistributions in binary form must reproduce the above
19  *	  copyright notice, this list of conditions and the following
20  *	  disclaimer in the documentation and/or other materials
21  *	  provided with the distribution.
22  *
23  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
24  * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
25  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
26  * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
27  * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
28  * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
29  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
30  * SOFTWARE.
31  */
32 #include <linux/module.h>
33 #include <linux/list.h>
34 #include <linux/workqueue.h>
35 #include <linux/skbuff.h>
36 #include <linux/timer.h>
37 #include <linux/notifier.h>
38 #include <linux/inetdevice.h>
39 #include <linux/ip.h>
40 #include <linux/tcp.h>
41 #include <linux/if_vlan.h>
42 
43 #include <net/neighbour.h>
44 #include <net/netevent.h>
45 #include <net/route.h>
46 #include <net/tcp.h>
47 #include <net/ip6_route.h>
48 #include <net/addrconf.h>
49 
50 #include <rdma/ib_addr.h>
51 
52 #include <libcxgb_cm.h>
53 #include "iw_cxgb4.h"
54 #include "clip_tbl.h"
55 
56 static char *states[] = {
57 	"idle",
58 	"listen",
59 	"connecting",
60 	"mpa_wait_req",
61 	"mpa_req_sent",
62 	"mpa_req_rcvd",
63 	"mpa_rep_sent",
64 	"fpdu_mode",
65 	"aborting",
66 	"closing",
67 	"moribund",
68 	"dead",
69 	NULL,
70 };
71 
72 static int nocong;
73 module_param(nocong, int, 0644);
74 MODULE_PARM_DESC(nocong, "Turn of congestion control (default=0)");
75 
76 static int enable_ecn;
77 module_param(enable_ecn, int, 0644);
78 MODULE_PARM_DESC(enable_ecn, "Enable ECN (default=0/disabled)");
79 
80 static int dack_mode;
81 module_param(dack_mode, int, 0644);
82 MODULE_PARM_DESC(dack_mode, "Delayed ack mode (default=0)");
83 
84 uint c4iw_max_read_depth = 32;
85 module_param(c4iw_max_read_depth, int, 0644);
86 MODULE_PARM_DESC(c4iw_max_read_depth,
87 		 "Per-connection max ORD/IRD (default=32)");
88 
89 static int enable_tcp_timestamps;
90 module_param(enable_tcp_timestamps, int, 0644);
91 MODULE_PARM_DESC(enable_tcp_timestamps, "Enable tcp timestamps (default=0)");
92 
93 static int enable_tcp_sack;
94 module_param(enable_tcp_sack, int, 0644);
95 MODULE_PARM_DESC(enable_tcp_sack, "Enable tcp SACK (default=0)");
96 
97 static int enable_tcp_window_scaling = 1;
98 module_param(enable_tcp_window_scaling, int, 0644);
99 MODULE_PARM_DESC(enable_tcp_window_scaling,
100 		 "Enable tcp window scaling (default=1)");
101 
102 static int peer2peer = 1;
103 module_param(peer2peer, int, 0644);
104 MODULE_PARM_DESC(peer2peer, "Support peer2peer ULPs (default=1)");
105 
106 static int p2p_type = FW_RI_INIT_P2PTYPE_READ_REQ;
107 module_param(p2p_type, int, 0644);
108 MODULE_PARM_DESC(p2p_type, "RDMAP opcode to use for the RTR message: "
109 			   "1=RDMA_READ 0=RDMA_WRITE (default 1)");
110 
111 static int ep_timeout_secs = 60;
112 module_param(ep_timeout_secs, int, 0644);
113 MODULE_PARM_DESC(ep_timeout_secs, "CM Endpoint operation timeout "
114 				   "in seconds (default=60)");
115 
116 static int mpa_rev = 2;
117 module_param(mpa_rev, int, 0644);
118 MODULE_PARM_DESC(mpa_rev, "MPA Revision, 0 supports amso1100, "
119 		"1 is RFC5044 spec compliant, 2 is IETF MPA Peer Connect Draft"
120 		" compliant (default=2)");
121 
122 static int markers_enabled;
123 module_param(markers_enabled, int, 0644);
124 MODULE_PARM_DESC(markers_enabled, "Enable MPA MARKERS (default(0)=disabled)");
125 
126 static int crc_enabled = 1;
127 module_param(crc_enabled, int, 0644);
128 MODULE_PARM_DESC(crc_enabled, "Enable MPA CRC (default(1)=enabled)");
129 
130 static int rcv_win = 256 * 1024;
131 module_param(rcv_win, int, 0644);
132 MODULE_PARM_DESC(rcv_win, "TCP receive window in bytes (default=256KB)");
133 
134 static int snd_win = 128 * 1024;
135 module_param(snd_win, int, 0644);
136 MODULE_PARM_DESC(snd_win, "TCP send window in bytes (default=128KB)");
137 
138 static struct workqueue_struct *workq;
139 
140 static struct sk_buff_head rxq;
141 
142 static struct sk_buff *get_skb(struct sk_buff *skb, int len, gfp_t gfp);
143 static void ep_timeout(struct timer_list *t);
144 static void connect_reply_upcall(struct c4iw_ep *ep, int status);
145 static int sched(struct c4iw_dev *dev, struct sk_buff *skb);
146 
147 static LIST_HEAD(timeout_list);
148 static DEFINE_SPINLOCK(timeout_lock);
149 
deref_cm_id(struct c4iw_ep_common * epc)150 static void deref_cm_id(struct c4iw_ep_common *epc)
151 {
152 	epc->cm_id->rem_ref(epc->cm_id);
153 	epc->cm_id = NULL;
154 	set_bit(CM_ID_DEREFED, &epc->history);
155 }
156 
ref_cm_id(struct c4iw_ep_common * epc)157 static void ref_cm_id(struct c4iw_ep_common *epc)
158 {
159 	set_bit(CM_ID_REFED, &epc->history);
160 	epc->cm_id->add_ref(epc->cm_id);
161 }
162 
deref_qp(struct c4iw_ep * ep)163 static void deref_qp(struct c4iw_ep *ep)
164 {
165 	c4iw_qp_rem_ref(&ep->com.qp->ibqp);
166 	clear_bit(QP_REFERENCED, &ep->com.flags);
167 	set_bit(QP_DEREFED, &ep->com.history);
168 }
169 
ref_qp(struct c4iw_ep * ep)170 static void ref_qp(struct c4iw_ep *ep)
171 {
172 	set_bit(QP_REFERENCED, &ep->com.flags);
173 	set_bit(QP_REFED, &ep->com.history);
174 	c4iw_qp_add_ref(&ep->com.qp->ibqp);
175 }
176 
start_ep_timer(struct c4iw_ep * ep)177 static void start_ep_timer(struct c4iw_ep *ep)
178 {
179 	pr_debug("ep %p\n", ep);
180 	if (timer_pending(&ep->timer)) {
181 		pr_err("%s timer already started! ep %p\n",
182 		       __func__, ep);
183 		return;
184 	}
185 	clear_bit(TIMEOUT, &ep->com.flags);
186 	c4iw_get_ep(&ep->com);
187 	ep->timer.expires = jiffies + ep_timeout_secs * HZ;
188 	add_timer(&ep->timer);
189 }
190 
stop_ep_timer(struct c4iw_ep * ep)191 static int stop_ep_timer(struct c4iw_ep *ep)
192 {
193 	pr_debug("ep %p stopping\n", ep);
194 	del_timer_sync(&ep->timer);
195 	if (!test_and_set_bit(TIMEOUT, &ep->com.flags)) {
196 		c4iw_put_ep(&ep->com);
197 		return 0;
198 	}
199 	return 1;
200 }
201 
c4iw_l2t_send(struct c4iw_rdev * rdev,struct sk_buff * skb,struct l2t_entry * l2e)202 static int c4iw_l2t_send(struct c4iw_rdev *rdev, struct sk_buff *skb,
203 		  struct l2t_entry *l2e)
204 {
205 	int	error = 0;
206 
207 	if (c4iw_fatal_error(rdev)) {
208 		kfree_skb(skb);
209 		pr_err("%s - device in error state - dropping\n", __func__);
210 		return -EIO;
211 	}
212 	error = cxgb4_l2t_send(rdev->lldi.ports[0], skb, l2e);
213 	if (error < 0)
214 		kfree_skb(skb);
215 	else if (error == NET_XMIT_DROP)
216 		return -ENOMEM;
217 	return error < 0 ? error : 0;
218 }
219 
c4iw_ofld_send(struct c4iw_rdev * rdev,struct sk_buff * skb)220 int c4iw_ofld_send(struct c4iw_rdev *rdev, struct sk_buff *skb)
221 {
222 	int	error = 0;
223 
224 	if (c4iw_fatal_error(rdev)) {
225 		kfree_skb(skb);
226 		pr_err("%s - device in error state - dropping\n", __func__);
227 		return -EIO;
228 	}
229 	error = cxgb4_ofld_send(rdev->lldi.ports[0], skb);
230 	if (error < 0)
231 		kfree_skb(skb);
232 	return error < 0 ? error : 0;
233 }
234 
release_tid(struct c4iw_rdev * rdev,u32 hwtid,struct sk_buff * skb)235 static void release_tid(struct c4iw_rdev *rdev, u32 hwtid, struct sk_buff *skb)
236 {
237 	u32 len = roundup(sizeof(struct cpl_tid_release), 16);
238 
239 	skb = get_skb(skb, len, GFP_KERNEL);
240 	if (!skb)
241 		return;
242 
243 	cxgb_mk_tid_release(skb, len, hwtid, 0);
244 	c4iw_ofld_send(rdev, skb);
245 	return;
246 }
247 
set_emss(struct c4iw_ep * ep,u16 opt)248 static void set_emss(struct c4iw_ep *ep, u16 opt)
249 {
250 	ep->emss = ep->com.dev->rdev.lldi.mtus[TCPOPT_MSS_G(opt)] -
251 		   ((AF_INET == ep->com.remote_addr.ss_family) ?
252 		    sizeof(struct iphdr) : sizeof(struct ipv6hdr)) -
253 		   sizeof(struct tcphdr);
254 	ep->mss = ep->emss;
255 	if (TCPOPT_TSTAMP_G(opt))
256 		ep->emss -= round_up(TCPOLEN_TIMESTAMP, 4);
257 	if (ep->emss < 128)
258 		ep->emss = 128;
259 	if (ep->emss & 7)
260 		pr_debug("Warning: misaligned mtu idx %u mss %u emss=%u\n",
261 			 TCPOPT_MSS_G(opt), ep->mss, ep->emss);
262 	pr_debug("mss_idx %u mss %u emss=%u\n", TCPOPT_MSS_G(opt), ep->mss,
263 		 ep->emss);
264 }
265 
state_read(struct c4iw_ep_common * epc)266 static enum c4iw_ep_state state_read(struct c4iw_ep_common *epc)
267 {
268 	enum c4iw_ep_state state;
269 
270 	mutex_lock(&epc->mutex);
271 	state = epc->state;
272 	mutex_unlock(&epc->mutex);
273 	return state;
274 }
275 
__state_set(struct c4iw_ep_common * epc,enum c4iw_ep_state new)276 static void __state_set(struct c4iw_ep_common *epc, enum c4iw_ep_state new)
277 {
278 	epc->state = new;
279 }
280 
state_set(struct c4iw_ep_common * epc,enum c4iw_ep_state new)281 static void state_set(struct c4iw_ep_common *epc, enum c4iw_ep_state new)
282 {
283 	mutex_lock(&epc->mutex);
284 	pr_debug("%s -> %s\n", states[epc->state], states[new]);
285 	__state_set(epc, new);
286 	mutex_unlock(&epc->mutex);
287 	return;
288 }
289 
alloc_ep_skb_list(struct sk_buff_head * ep_skb_list,int size)290 static int alloc_ep_skb_list(struct sk_buff_head *ep_skb_list, int size)
291 {
292 	struct sk_buff *skb;
293 	unsigned int i;
294 	size_t len;
295 
296 	len = roundup(sizeof(union cpl_wr_size), 16);
297 	for (i = 0; i < size; i++) {
298 		skb = alloc_skb(len, GFP_KERNEL);
299 		if (!skb)
300 			goto fail;
301 		skb_queue_tail(ep_skb_list, skb);
302 	}
303 	return 0;
304 fail:
305 	skb_queue_purge(ep_skb_list);
306 	return -ENOMEM;
307 }
308 
alloc_ep(int size,gfp_t gfp)309 static void *alloc_ep(int size, gfp_t gfp)
310 {
311 	struct c4iw_ep_common *epc;
312 
313 	epc = kzalloc(size, gfp);
314 	if (epc) {
315 		epc->wr_waitp = c4iw_alloc_wr_wait(gfp);
316 		if (!epc->wr_waitp) {
317 			kfree(epc);
318 			epc = NULL;
319 			goto out;
320 		}
321 		kref_init(&epc->kref);
322 		mutex_init(&epc->mutex);
323 		c4iw_init_wr_wait(epc->wr_waitp);
324 	}
325 	pr_debug("alloc ep %p\n", epc);
326 out:
327 	return epc;
328 }
329 
remove_ep_tid(struct c4iw_ep * ep)330 static void remove_ep_tid(struct c4iw_ep *ep)
331 {
332 	unsigned long flags;
333 
334 	xa_lock_irqsave(&ep->com.dev->hwtids, flags);
335 	__xa_erase(&ep->com.dev->hwtids, ep->hwtid);
336 	if (xa_empty(&ep->com.dev->hwtids))
337 		wake_up(&ep->com.dev->wait);
338 	xa_unlock_irqrestore(&ep->com.dev->hwtids, flags);
339 }
340 
insert_ep_tid(struct c4iw_ep * ep)341 static int insert_ep_tid(struct c4iw_ep *ep)
342 {
343 	unsigned long flags;
344 	int err;
345 
346 	xa_lock_irqsave(&ep->com.dev->hwtids, flags);
347 	err = __xa_insert(&ep->com.dev->hwtids, ep->hwtid, ep, GFP_KERNEL);
348 	xa_unlock_irqrestore(&ep->com.dev->hwtids, flags);
349 
350 	return err;
351 }
352 
353 /*
354  * Atomically lookup the ep ptr given the tid and grab a reference on the ep.
355  */
get_ep_from_tid(struct c4iw_dev * dev,unsigned int tid)356 static struct c4iw_ep *get_ep_from_tid(struct c4iw_dev *dev, unsigned int tid)
357 {
358 	struct c4iw_ep *ep;
359 	unsigned long flags;
360 
361 	xa_lock_irqsave(&dev->hwtids, flags);
362 	ep = xa_load(&dev->hwtids, tid);
363 	if (ep)
364 		c4iw_get_ep(&ep->com);
365 	xa_unlock_irqrestore(&dev->hwtids, flags);
366 	return ep;
367 }
368 
369 /*
370  * Atomically lookup the ep ptr given the stid and grab a reference on the ep.
371  */
get_ep_from_stid(struct c4iw_dev * dev,unsigned int stid)372 static struct c4iw_listen_ep *get_ep_from_stid(struct c4iw_dev *dev,
373 					       unsigned int stid)
374 {
375 	struct c4iw_listen_ep *ep;
376 	unsigned long flags;
377 
378 	xa_lock_irqsave(&dev->stids, flags);
379 	ep = xa_load(&dev->stids, stid);
380 	if (ep)
381 		c4iw_get_ep(&ep->com);
382 	xa_unlock_irqrestore(&dev->stids, flags);
383 	return ep;
384 }
385 
_c4iw_free_ep(struct kref * kref)386 void _c4iw_free_ep(struct kref *kref)
387 {
388 	struct c4iw_ep *ep;
389 
390 	ep = container_of(kref, struct c4iw_ep, com.kref);
391 	pr_debug("ep %p state %s\n", ep, states[ep->com.state]);
392 	if (test_bit(QP_REFERENCED, &ep->com.flags))
393 		deref_qp(ep);
394 	if (test_bit(RELEASE_RESOURCES, &ep->com.flags)) {
395 		if (ep->com.remote_addr.ss_family == AF_INET6) {
396 			struct sockaddr_in6 *sin6 =
397 					(struct sockaddr_in6 *)
398 					&ep->com.local_addr;
399 
400 			cxgb4_clip_release(
401 					ep->com.dev->rdev.lldi.ports[0],
402 					(const u32 *)&sin6->sin6_addr.s6_addr,
403 					1);
404 		}
405 		cxgb4_remove_tid(ep->com.dev->rdev.lldi.tids, 0, ep->hwtid,
406 				 ep->com.local_addr.ss_family);
407 		dst_release(ep->dst);
408 		cxgb4_l2t_release(ep->l2t);
409 		kfree_skb(ep->mpa_skb);
410 	}
411 	if (!skb_queue_empty(&ep->com.ep_skb_list))
412 		skb_queue_purge(&ep->com.ep_skb_list);
413 	c4iw_put_wr_wait(ep->com.wr_waitp);
414 	kfree(ep);
415 }
416 
release_ep_resources(struct c4iw_ep * ep)417 static void release_ep_resources(struct c4iw_ep *ep)
418 {
419 	set_bit(RELEASE_RESOURCES, &ep->com.flags);
420 
421 	/*
422 	 * If we have a hwtid, then remove it from the idr table
423 	 * so lookups will no longer find this endpoint.  Otherwise
424 	 * we have a race where one thread finds the ep ptr just
425 	 * before the other thread is freeing the ep memory.
426 	 */
427 	if (ep->hwtid != -1)
428 		remove_ep_tid(ep);
429 	c4iw_put_ep(&ep->com);
430 }
431 
status2errno(int status)432 static int status2errno(int status)
433 {
434 	switch (status) {
435 	case CPL_ERR_NONE:
436 		return 0;
437 	case CPL_ERR_CONN_RESET:
438 		return -ECONNRESET;
439 	case CPL_ERR_ARP_MISS:
440 		return -EHOSTUNREACH;
441 	case CPL_ERR_CONN_TIMEDOUT:
442 		return -ETIMEDOUT;
443 	case CPL_ERR_TCAM_FULL:
444 		return -ENOMEM;
445 	case CPL_ERR_CONN_EXIST:
446 		return -EADDRINUSE;
447 	default:
448 		return -EIO;
449 	}
450 }
451 
452 /*
453  * Try and reuse skbs already allocated...
454  */
get_skb(struct sk_buff * skb,int len,gfp_t gfp)455 static struct sk_buff *get_skb(struct sk_buff *skb, int len, gfp_t gfp)
456 {
457 	if (skb && !skb_is_nonlinear(skb) && !skb_cloned(skb)) {
458 		skb_trim(skb, 0);
459 		skb_get(skb);
460 		skb_reset_transport_header(skb);
461 	} else {
462 		skb = alloc_skb(len, gfp);
463 		if (!skb)
464 			return NULL;
465 	}
466 	t4_set_arp_err_handler(skb, NULL, NULL);
467 	return skb;
468 }
469 
get_real_dev(struct net_device * egress_dev)470 static struct net_device *get_real_dev(struct net_device *egress_dev)
471 {
472 	return rdma_vlan_dev_real_dev(egress_dev) ? : egress_dev;
473 }
474 
arp_failure_discard(void * handle,struct sk_buff * skb)475 static void arp_failure_discard(void *handle, struct sk_buff *skb)
476 {
477 	pr_err("ARP failure\n");
478 	kfree_skb(skb);
479 }
480 
mpa_start_arp_failure(void * handle,struct sk_buff * skb)481 static void mpa_start_arp_failure(void *handle, struct sk_buff *skb)
482 {
483 	pr_err("ARP failure during MPA Negotiation - Closing Connection\n");
484 }
485 
486 enum {
487 	NUM_FAKE_CPLS = 2,
488 	FAKE_CPL_PUT_EP_SAFE = NUM_CPL_CMDS + 0,
489 	FAKE_CPL_PASS_PUT_EP_SAFE = NUM_CPL_CMDS + 1,
490 };
491 
_put_ep_safe(struct c4iw_dev * dev,struct sk_buff * skb)492 static int _put_ep_safe(struct c4iw_dev *dev, struct sk_buff *skb)
493 {
494 	struct c4iw_ep *ep;
495 
496 	ep = *((struct c4iw_ep **)(skb->cb + 2 * sizeof(void *)));
497 	release_ep_resources(ep);
498 	return 0;
499 }
500 
_put_pass_ep_safe(struct c4iw_dev * dev,struct sk_buff * skb)501 static int _put_pass_ep_safe(struct c4iw_dev *dev, struct sk_buff *skb)
502 {
503 	struct c4iw_ep *ep;
504 
505 	ep = *((struct c4iw_ep **)(skb->cb + 2 * sizeof(void *)));
506 	c4iw_put_ep(&ep->parent_ep->com);
507 	release_ep_resources(ep);
508 	return 0;
509 }
510 
511 /*
512  * Fake up a special CPL opcode and call sched() so process_work() will call
513  * _put_ep_safe() in a safe context to free the ep resources.  This is needed
514  * because ARP error handlers are called in an ATOMIC context, and
515  * _c4iw_free_ep() needs to block.
516  */
queue_arp_failure_cpl(struct c4iw_ep * ep,struct sk_buff * skb,int cpl)517 static void queue_arp_failure_cpl(struct c4iw_ep *ep, struct sk_buff *skb,
518 				  int cpl)
519 {
520 	struct cpl_act_establish *rpl = cplhdr(skb);
521 
522 	/* Set our special ARP_FAILURE opcode */
523 	rpl->ot.opcode = cpl;
524 
525 	/*
526 	 * Save ep in the skb->cb area, after where sched() will save the dev
527 	 * ptr.
528 	 */
529 	*((struct c4iw_ep **)(skb->cb + 2 * sizeof(void *))) = ep;
530 	sched(ep->com.dev, skb);
531 }
532 
533 /* Handle an ARP failure for an accept */
pass_accept_rpl_arp_failure(void * handle,struct sk_buff * skb)534 static void pass_accept_rpl_arp_failure(void *handle, struct sk_buff *skb)
535 {
536 	struct c4iw_ep *ep = handle;
537 
538 	pr_err("ARP failure during accept - tid %u - dropping connection\n",
539 	       ep->hwtid);
540 
541 	__state_set(&ep->com, DEAD);
542 	queue_arp_failure_cpl(ep, skb, FAKE_CPL_PASS_PUT_EP_SAFE);
543 }
544 
545 /*
546  * Handle an ARP failure for an active open.
547  */
act_open_req_arp_failure(void * handle,struct sk_buff * skb)548 static void act_open_req_arp_failure(void *handle, struct sk_buff *skb)
549 {
550 	struct c4iw_ep *ep = handle;
551 
552 	pr_err("ARP failure during connect\n");
553 	connect_reply_upcall(ep, -EHOSTUNREACH);
554 	__state_set(&ep->com, DEAD);
555 	if (ep->com.remote_addr.ss_family == AF_INET6) {
556 		struct sockaddr_in6 *sin6 =
557 			(struct sockaddr_in6 *)&ep->com.local_addr;
558 		cxgb4_clip_release(ep->com.dev->rdev.lldi.ports[0],
559 				   (const u32 *)&sin6->sin6_addr.s6_addr, 1);
560 	}
561 	xa_erase_irq(&ep->com.dev->atids, ep->atid);
562 	cxgb4_free_atid(ep->com.dev->rdev.lldi.tids, ep->atid);
563 	queue_arp_failure_cpl(ep, skb, FAKE_CPL_PUT_EP_SAFE);
564 }
565 
566 /*
567  * Handle an ARP failure for a CPL_ABORT_REQ.  Change it into a no RST variant
568  * and send it along.
569  */
abort_arp_failure(void * handle,struct sk_buff * skb)570 static void abort_arp_failure(void *handle, struct sk_buff *skb)
571 {
572 	int ret;
573 	struct c4iw_ep *ep = handle;
574 	struct c4iw_rdev *rdev = &ep->com.dev->rdev;
575 	struct cpl_abort_req *req = cplhdr(skb);
576 
577 	pr_debug("rdev %p\n", rdev);
578 	req->cmd = CPL_ABORT_NO_RST;
579 	skb_get(skb);
580 	ret = c4iw_ofld_send(rdev, skb);
581 	if (ret) {
582 		__state_set(&ep->com, DEAD);
583 		queue_arp_failure_cpl(ep, skb, FAKE_CPL_PUT_EP_SAFE);
584 	} else
585 		kfree_skb(skb);
586 }
587 
send_flowc(struct c4iw_ep * ep)588 static int send_flowc(struct c4iw_ep *ep)
589 {
590 	struct fw_flowc_wr *flowc;
591 	struct sk_buff *skb = skb_dequeue(&ep->com.ep_skb_list);
592 	u16 vlan = ep->l2t->vlan;
593 	int nparams;
594 	int flowclen, flowclen16;
595 
596 	if (WARN_ON(!skb))
597 		return -ENOMEM;
598 
599 	if (vlan == CPL_L2T_VLAN_NONE)
600 		nparams = 9;
601 	else
602 		nparams = 10;
603 
604 	flowclen = offsetof(struct fw_flowc_wr, mnemval[nparams]);
605 	flowclen16 = DIV_ROUND_UP(flowclen, 16);
606 	flowclen = flowclen16 * 16;
607 
608 	flowc = __skb_put(skb, flowclen);
609 	memset(flowc, 0, flowclen);
610 
611 	flowc->op_to_nparams = cpu_to_be32(FW_WR_OP_V(FW_FLOWC_WR) |
612 					   FW_FLOWC_WR_NPARAMS_V(nparams));
613 	flowc->flowid_len16 = cpu_to_be32(FW_WR_LEN16_V(flowclen16) |
614 					  FW_WR_FLOWID_V(ep->hwtid));
615 
616 	flowc->mnemval[0].mnemonic = FW_FLOWC_MNEM_PFNVFN;
617 	flowc->mnemval[0].val = cpu_to_be32(FW_PFVF_CMD_PFN_V
618 					    (ep->com.dev->rdev.lldi.pf));
619 	flowc->mnemval[1].mnemonic = FW_FLOWC_MNEM_CH;
620 	flowc->mnemval[1].val = cpu_to_be32(ep->tx_chan);
621 	flowc->mnemval[2].mnemonic = FW_FLOWC_MNEM_PORT;
622 	flowc->mnemval[2].val = cpu_to_be32(ep->tx_chan);
623 	flowc->mnemval[3].mnemonic = FW_FLOWC_MNEM_IQID;
624 	flowc->mnemval[3].val = cpu_to_be32(ep->rss_qid);
625 	flowc->mnemval[4].mnemonic = FW_FLOWC_MNEM_SNDNXT;
626 	flowc->mnemval[4].val = cpu_to_be32(ep->snd_seq);
627 	flowc->mnemval[5].mnemonic = FW_FLOWC_MNEM_RCVNXT;
628 	flowc->mnemval[5].val = cpu_to_be32(ep->rcv_seq);
629 	flowc->mnemval[6].mnemonic = FW_FLOWC_MNEM_SNDBUF;
630 	flowc->mnemval[6].val = cpu_to_be32(ep->snd_win);
631 	flowc->mnemval[7].mnemonic = FW_FLOWC_MNEM_MSS;
632 	flowc->mnemval[7].val = cpu_to_be32(ep->emss);
633 	flowc->mnemval[8].mnemonic = FW_FLOWC_MNEM_RCV_SCALE;
634 	flowc->mnemval[8].val = cpu_to_be32(ep->snd_wscale);
635 	if (nparams == 10) {
636 		u16 pri;
637 		pri = (vlan & VLAN_PRIO_MASK) >> VLAN_PRIO_SHIFT;
638 		flowc->mnemval[9].mnemonic = FW_FLOWC_MNEM_SCHEDCLASS;
639 		flowc->mnemval[9].val = cpu_to_be32(pri);
640 	}
641 
642 	set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
643 	return c4iw_ofld_send(&ep->com.dev->rdev, skb);
644 }
645 
send_halfclose(struct c4iw_ep * ep)646 static int send_halfclose(struct c4iw_ep *ep)
647 {
648 	struct sk_buff *skb = skb_dequeue(&ep->com.ep_skb_list);
649 	u32 wrlen = roundup(sizeof(struct cpl_close_con_req), 16);
650 
651 	pr_debug("ep %p tid %u\n", ep, ep->hwtid);
652 	if (WARN_ON(!skb))
653 		return -ENOMEM;
654 
655 	cxgb_mk_close_con_req(skb, wrlen, ep->hwtid, ep->txq_idx,
656 			      NULL, arp_failure_discard);
657 
658 	return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
659 }
660 
read_tcb(struct c4iw_ep * ep)661 static void read_tcb(struct c4iw_ep *ep)
662 {
663 	struct sk_buff *skb;
664 	struct cpl_get_tcb *req;
665 	int wrlen = roundup(sizeof(*req), 16);
666 
667 	skb = get_skb(NULL, sizeof(*req), GFP_KERNEL);
668 	if (WARN_ON(!skb))
669 		return;
670 
671 	set_wr_txq(skb, CPL_PRIORITY_CONTROL, ep->ctrlq_idx);
672 	req = (struct cpl_get_tcb *) skb_put(skb, wrlen);
673 	memset(req, 0, wrlen);
674 	INIT_TP_WR(req, ep->hwtid);
675 	OPCODE_TID(req) = cpu_to_be32(MK_OPCODE_TID(CPL_GET_TCB, ep->hwtid));
676 	req->reply_ctrl = htons(REPLY_CHAN_V(0) | QUEUENO_V(ep->rss_qid));
677 
678 	/*
679 	 * keep a ref on the ep so the tcb is not unlocked before this
680 	 * cpl completes. The ref is released in read_tcb_rpl().
681 	 */
682 	c4iw_get_ep(&ep->com);
683 	if (WARN_ON(c4iw_ofld_send(&ep->com.dev->rdev, skb)))
684 		c4iw_put_ep(&ep->com);
685 }
686 
send_abort_req(struct c4iw_ep * ep)687 static int send_abort_req(struct c4iw_ep *ep)
688 {
689 	u32 wrlen = roundup(sizeof(struct cpl_abort_req), 16);
690 	struct sk_buff *req_skb = skb_dequeue(&ep->com.ep_skb_list);
691 
692 	pr_debug("ep %p tid %u\n", ep, ep->hwtid);
693 	if (WARN_ON(!req_skb))
694 		return -ENOMEM;
695 
696 	cxgb_mk_abort_req(req_skb, wrlen, ep->hwtid, ep->txq_idx,
697 			  ep, abort_arp_failure);
698 
699 	return c4iw_l2t_send(&ep->com.dev->rdev, req_skb, ep->l2t);
700 }
701 
send_abort(struct c4iw_ep * ep)702 static int send_abort(struct c4iw_ep *ep)
703 {
704 	if (!ep->com.qp || !ep->com.qp->srq) {
705 		send_abort_req(ep);
706 		return 0;
707 	}
708 	set_bit(ABORT_REQ_IN_PROGRESS, &ep->com.flags);
709 	read_tcb(ep);
710 	return 0;
711 }
712 
send_connect(struct c4iw_ep * ep)713 static int send_connect(struct c4iw_ep *ep)
714 {
715 	struct cpl_act_open_req *req = NULL;
716 	struct cpl_t5_act_open_req *t5req = NULL;
717 	struct cpl_t6_act_open_req *t6req = NULL;
718 	struct cpl_act_open_req6 *req6 = NULL;
719 	struct cpl_t5_act_open_req6 *t5req6 = NULL;
720 	struct cpl_t6_act_open_req6 *t6req6 = NULL;
721 	struct sk_buff *skb;
722 	u64 opt0;
723 	u32 opt2;
724 	unsigned int mtu_idx;
725 	u32 wscale;
726 	int win, sizev4, sizev6, wrlen;
727 	struct sockaddr_in *la = (struct sockaddr_in *)
728 				 &ep->com.local_addr;
729 	struct sockaddr_in *ra = (struct sockaddr_in *)
730 				 &ep->com.remote_addr;
731 	struct sockaddr_in6 *la6 = (struct sockaddr_in6 *)
732 				   &ep->com.local_addr;
733 	struct sockaddr_in6 *ra6 = (struct sockaddr_in6 *)
734 				   &ep->com.remote_addr;
735 	int ret;
736 	enum chip_type adapter_type = ep->com.dev->rdev.lldi.adapter_type;
737 	u32 isn = (prandom_u32() & ~7UL) - 1;
738 	struct net_device *netdev;
739 	u64 params;
740 
741 	netdev = ep->com.dev->rdev.lldi.ports[0];
742 
743 	switch (CHELSIO_CHIP_VERSION(adapter_type)) {
744 	case CHELSIO_T4:
745 		sizev4 = sizeof(struct cpl_act_open_req);
746 		sizev6 = sizeof(struct cpl_act_open_req6);
747 		break;
748 	case CHELSIO_T5:
749 		sizev4 = sizeof(struct cpl_t5_act_open_req);
750 		sizev6 = sizeof(struct cpl_t5_act_open_req6);
751 		break;
752 	case CHELSIO_T6:
753 		sizev4 = sizeof(struct cpl_t6_act_open_req);
754 		sizev6 = sizeof(struct cpl_t6_act_open_req6);
755 		break;
756 	default:
757 		pr_err("T%d Chip is not supported\n",
758 		       CHELSIO_CHIP_VERSION(adapter_type));
759 		return -EINVAL;
760 	}
761 
762 	wrlen = (ep->com.remote_addr.ss_family == AF_INET) ?
763 			roundup(sizev4, 16) :
764 			roundup(sizev6, 16);
765 
766 	pr_debug("ep %p atid %u\n", ep, ep->atid);
767 
768 	skb = get_skb(NULL, wrlen, GFP_KERNEL);
769 	if (!skb) {
770 		pr_err("%s - failed to alloc skb\n", __func__);
771 		return -ENOMEM;
772 	}
773 	set_wr_txq(skb, CPL_PRIORITY_SETUP, ep->ctrlq_idx);
774 
775 	cxgb_best_mtu(ep->com.dev->rdev.lldi.mtus, ep->mtu, &mtu_idx,
776 		      enable_tcp_timestamps,
777 		      (ep->com.remote_addr.ss_family == AF_INET) ? 0 : 1);
778 	wscale = cxgb_compute_wscale(rcv_win);
779 
780 	/*
781 	 * Specify the largest window that will fit in opt0. The
782 	 * remainder will be specified in the rx_data_ack.
783 	 */
784 	win = ep->rcv_win >> 10;
785 	if (win > RCV_BUFSIZ_M)
786 		win = RCV_BUFSIZ_M;
787 
788 	opt0 = (nocong ? NO_CONG_F : 0) |
789 	       KEEP_ALIVE_F |
790 	       DELACK_F |
791 	       WND_SCALE_V(wscale) |
792 	       MSS_IDX_V(mtu_idx) |
793 	       L2T_IDX_V(ep->l2t->idx) |
794 	       TX_CHAN_V(ep->tx_chan) |
795 	       SMAC_SEL_V(ep->smac_idx) |
796 	       DSCP_V(ep->tos >> 2) |
797 	       ULP_MODE_V(ULP_MODE_TCPDDP) |
798 	       RCV_BUFSIZ_V(win);
799 	opt2 = RX_CHANNEL_V(0) |
800 	       CCTRL_ECN_V(enable_ecn) |
801 	       RSS_QUEUE_VALID_F | RSS_QUEUE_V(ep->rss_qid);
802 	if (enable_tcp_timestamps)
803 		opt2 |= TSTAMPS_EN_F;
804 	if (enable_tcp_sack)
805 		opt2 |= SACK_EN_F;
806 	if (wscale && enable_tcp_window_scaling)
807 		opt2 |= WND_SCALE_EN_F;
808 	if (CHELSIO_CHIP_VERSION(adapter_type) > CHELSIO_T4) {
809 		if (peer2peer)
810 			isn += 4;
811 
812 		opt2 |= T5_OPT_2_VALID_F;
813 		opt2 |= CONG_CNTRL_V(CONG_ALG_TAHOE);
814 		opt2 |= T5_ISS_F;
815 	}
816 
817 	params = cxgb4_select_ntuple(netdev, ep->l2t);
818 
819 	if (ep->com.remote_addr.ss_family == AF_INET6)
820 		cxgb4_clip_get(ep->com.dev->rdev.lldi.ports[0],
821 			       (const u32 *)&la6->sin6_addr.s6_addr, 1);
822 
823 	t4_set_arp_err_handler(skb, ep, act_open_req_arp_failure);
824 
825 	if (ep->com.remote_addr.ss_family == AF_INET) {
826 		switch (CHELSIO_CHIP_VERSION(adapter_type)) {
827 		case CHELSIO_T4:
828 			req = skb_put(skb, wrlen);
829 			INIT_TP_WR(req, 0);
830 			break;
831 		case CHELSIO_T5:
832 			t5req = skb_put(skb, wrlen);
833 			INIT_TP_WR(t5req, 0);
834 			req = (struct cpl_act_open_req *)t5req;
835 			break;
836 		case CHELSIO_T6:
837 			t6req = skb_put(skb, wrlen);
838 			INIT_TP_WR(t6req, 0);
839 			req = (struct cpl_act_open_req *)t6req;
840 			t5req = (struct cpl_t5_act_open_req *)t6req;
841 			break;
842 		default:
843 			pr_err("T%d Chip is not supported\n",
844 			       CHELSIO_CHIP_VERSION(adapter_type));
845 			ret = -EINVAL;
846 			goto clip_release;
847 		}
848 
849 		OPCODE_TID(req) = cpu_to_be32(MK_OPCODE_TID(CPL_ACT_OPEN_REQ,
850 					((ep->rss_qid<<14) | ep->atid)));
851 		req->local_port = la->sin_port;
852 		req->peer_port = ra->sin_port;
853 		req->local_ip = la->sin_addr.s_addr;
854 		req->peer_ip = ra->sin_addr.s_addr;
855 		req->opt0 = cpu_to_be64(opt0);
856 
857 		if (is_t4(ep->com.dev->rdev.lldi.adapter_type)) {
858 			req->params = cpu_to_be32(params);
859 			req->opt2 = cpu_to_be32(opt2);
860 		} else {
861 			if (is_t5(ep->com.dev->rdev.lldi.adapter_type)) {
862 				t5req->params =
863 					  cpu_to_be64(FILTER_TUPLE_V(params));
864 				t5req->rsvd = cpu_to_be32(isn);
865 				pr_debug("snd_isn %u\n", t5req->rsvd);
866 				t5req->opt2 = cpu_to_be32(opt2);
867 			} else {
868 				t6req->params =
869 					  cpu_to_be64(FILTER_TUPLE_V(params));
870 				t6req->rsvd = cpu_to_be32(isn);
871 				pr_debug("snd_isn %u\n", t6req->rsvd);
872 				t6req->opt2 = cpu_to_be32(opt2);
873 			}
874 		}
875 	} else {
876 		switch (CHELSIO_CHIP_VERSION(adapter_type)) {
877 		case CHELSIO_T4:
878 			req6 = skb_put(skb, wrlen);
879 			INIT_TP_WR(req6, 0);
880 			break;
881 		case CHELSIO_T5:
882 			t5req6 = skb_put(skb, wrlen);
883 			INIT_TP_WR(t5req6, 0);
884 			req6 = (struct cpl_act_open_req6 *)t5req6;
885 			break;
886 		case CHELSIO_T6:
887 			t6req6 = skb_put(skb, wrlen);
888 			INIT_TP_WR(t6req6, 0);
889 			req6 = (struct cpl_act_open_req6 *)t6req6;
890 			t5req6 = (struct cpl_t5_act_open_req6 *)t6req6;
891 			break;
892 		default:
893 			pr_err("T%d Chip is not supported\n",
894 			       CHELSIO_CHIP_VERSION(adapter_type));
895 			ret = -EINVAL;
896 			goto clip_release;
897 		}
898 
899 		OPCODE_TID(req6) = cpu_to_be32(MK_OPCODE_TID(CPL_ACT_OPEN_REQ6,
900 					((ep->rss_qid<<14)|ep->atid)));
901 		req6->local_port = la6->sin6_port;
902 		req6->peer_port = ra6->sin6_port;
903 		req6->local_ip_hi = *((__be64 *)(la6->sin6_addr.s6_addr));
904 		req6->local_ip_lo = *((__be64 *)(la6->sin6_addr.s6_addr + 8));
905 		req6->peer_ip_hi = *((__be64 *)(ra6->sin6_addr.s6_addr));
906 		req6->peer_ip_lo = *((__be64 *)(ra6->sin6_addr.s6_addr + 8));
907 		req6->opt0 = cpu_to_be64(opt0);
908 
909 		if (is_t4(ep->com.dev->rdev.lldi.adapter_type)) {
910 			req6->params = cpu_to_be32(cxgb4_select_ntuple(netdev,
911 								      ep->l2t));
912 			req6->opt2 = cpu_to_be32(opt2);
913 		} else {
914 			if (is_t5(ep->com.dev->rdev.lldi.adapter_type)) {
915 				t5req6->params =
916 					    cpu_to_be64(FILTER_TUPLE_V(params));
917 				t5req6->rsvd = cpu_to_be32(isn);
918 				pr_debug("snd_isn %u\n", t5req6->rsvd);
919 				t5req6->opt2 = cpu_to_be32(opt2);
920 			} else {
921 				t6req6->params =
922 					    cpu_to_be64(FILTER_TUPLE_V(params));
923 				t6req6->rsvd = cpu_to_be32(isn);
924 				pr_debug("snd_isn %u\n", t6req6->rsvd);
925 				t6req6->opt2 = cpu_to_be32(opt2);
926 			}
927 
928 		}
929 	}
930 
931 	set_bit(ACT_OPEN_REQ, &ep->com.history);
932 	ret = c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
933 clip_release:
934 	if (ret && ep->com.remote_addr.ss_family == AF_INET6)
935 		cxgb4_clip_release(ep->com.dev->rdev.lldi.ports[0],
936 				   (const u32 *)&la6->sin6_addr.s6_addr, 1);
937 	return ret;
938 }
939 
send_mpa_req(struct c4iw_ep * ep,struct sk_buff * skb,u8 mpa_rev_to_use)940 static int send_mpa_req(struct c4iw_ep *ep, struct sk_buff *skb,
941 			u8 mpa_rev_to_use)
942 {
943 	int mpalen, wrlen, ret;
944 	struct fw_ofld_tx_data_wr *req;
945 	struct mpa_message *mpa;
946 	struct mpa_v2_conn_params mpa_v2_params;
947 
948 	pr_debug("ep %p tid %u pd_len %d\n",
949 		 ep, ep->hwtid, ep->plen);
950 
951 	mpalen = sizeof(*mpa) + ep->plen;
952 	if (mpa_rev_to_use == 2)
953 		mpalen += sizeof(struct mpa_v2_conn_params);
954 	wrlen = roundup(mpalen + sizeof(*req), 16);
955 	skb = get_skb(skb, wrlen, GFP_KERNEL);
956 	if (!skb) {
957 		connect_reply_upcall(ep, -ENOMEM);
958 		return -ENOMEM;
959 	}
960 	set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
961 
962 	req = skb_put_zero(skb, wrlen);
963 	req->op_to_immdlen = cpu_to_be32(
964 		FW_WR_OP_V(FW_OFLD_TX_DATA_WR) |
965 		FW_WR_COMPL_F |
966 		FW_WR_IMMDLEN_V(mpalen));
967 	req->flowid_len16 = cpu_to_be32(
968 		FW_WR_FLOWID_V(ep->hwtid) |
969 		FW_WR_LEN16_V(wrlen >> 4));
970 	req->plen = cpu_to_be32(mpalen);
971 	req->tunnel_to_proxy = cpu_to_be32(
972 		FW_OFLD_TX_DATA_WR_FLUSH_F |
973 		FW_OFLD_TX_DATA_WR_SHOVE_F);
974 
975 	mpa = (struct mpa_message *)(req + 1);
976 	memcpy(mpa->key, MPA_KEY_REQ, sizeof(mpa->key));
977 
978 	mpa->flags = 0;
979 	if (crc_enabled)
980 		mpa->flags |= MPA_CRC;
981 	if (markers_enabled) {
982 		mpa->flags |= MPA_MARKERS;
983 		ep->mpa_attr.recv_marker_enabled = 1;
984 	} else {
985 		ep->mpa_attr.recv_marker_enabled = 0;
986 	}
987 	if (mpa_rev_to_use == 2)
988 		mpa->flags |= MPA_ENHANCED_RDMA_CONN;
989 
990 	mpa->private_data_size = htons(ep->plen);
991 	mpa->revision = mpa_rev_to_use;
992 	if (mpa_rev_to_use == 1) {
993 		ep->tried_with_mpa_v1 = 1;
994 		ep->retry_with_mpa_v1 = 0;
995 	}
996 
997 	if (mpa_rev_to_use == 2) {
998 		mpa->private_data_size =
999 			htons(ntohs(mpa->private_data_size) +
1000 			      sizeof(struct mpa_v2_conn_params));
1001 		pr_debug("initiator ird %u ord %u\n", ep->ird,
1002 			 ep->ord);
1003 		mpa_v2_params.ird = htons((u16)ep->ird);
1004 		mpa_v2_params.ord = htons((u16)ep->ord);
1005 
1006 		if (peer2peer) {
1007 			mpa_v2_params.ird |= htons(MPA_V2_PEER2PEER_MODEL);
1008 			if (p2p_type == FW_RI_INIT_P2PTYPE_RDMA_WRITE)
1009 				mpa_v2_params.ord |=
1010 					htons(MPA_V2_RDMA_WRITE_RTR);
1011 			else if (p2p_type == FW_RI_INIT_P2PTYPE_READ_REQ)
1012 				mpa_v2_params.ord |=
1013 					htons(MPA_V2_RDMA_READ_RTR);
1014 		}
1015 		memcpy(mpa->private_data, &mpa_v2_params,
1016 		       sizeof(struct mpa_v2_conn_params));
1017 
1018 		if (ep->plen)
1019 			memcpy(mpa->private_data +
1020 			       sizeof(struct mpa_v2_conn_params),
1021 			       ep->mpa_pkt + sizeof(*mpa), ep->plen);
1022 	} else
1023 		if (ep->plen)
1024 			memcpy(mpa->private_data,
1025 					ep->mpa_pkt + sizeof(*mpa), ep->plen);
1026 
1027 	/*
1028 	 * Reference the mpa skb.  This ensures the data area
1029 	 * will remain in memory until the hw acks the tx.
1030 	 * Function fw4_ack() will deref it.
1031 	 */
1032 	skb_get(skb);
1033 	t4_set_arp_err_handler(skb, NULL, arp_failure_discard);
1034 	ep->mpa_skb = skb;
1035 	ret = c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
1036 	if (ret)
1037 		return ret;
1038 	start_ep_timer(ep);
1039 	__state_set(&ep->com, MPA_REQ_SENT);
1040 	ep->mpa_attr.initiator = 1;
1041 	ep->snd_seq += mpalen;
1042 	return ret;
1043 }
1044 
send_mpa_reject(struct c4iw_ep * ep,const void * pdata,u8 plen)1045 static int send_mpa_reject(struct c4iw_ep *ep, const void *pdata, u8 plen)
1046 {
1047 	int mpalen, wrlen;
1048 	struct fw_ofld_tx_data_wr *req;
1049 	struct mpa_message *mpa;
1050 	struct sk_buff *skb;
1051 	struct mpa_v2_conn_params mpa_v2_params;
1052 
1053 	pr_debug("ep %p tid %u pd_len %d\n",
1054 		 ep, ep->hwtid, ep->plen);
1055 
1056 	mpalen = sizeof(*mpa) + plen;
1057 	if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn)
1058 		mpalen += sizeof(struct mpa_v2_conn_params);
1059 	wrlen = roundup(mpalen + sizeof(*req), 16);
1060 
1061 	skb = get_skb(NULL, wrlen, GFP_KERNEL);
1062 	if (!skb) {
1063 		pr_err("%s - cannot alloc skb!\n", __func__);
1064 		return -ENOMEM;
1065 	}
1066 	set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
1067 
1068 	req = skb_put_zero(skb, wrlen);
1069 	req->op_to_immdlen = cpu_to_be32(
1070 		FW_WR_OP_V(FW_OFLD_TX_DATA_WR) |
1071 		FW_WR_COMPL_F |
1072 		FW_WR_IMMDLEN_V(mpalen));
1073 	req->flowid_len16 = cpu_to_be32(
1074 		FW_WR_FLOWID_V(ep->hwtid) |
1075 		FW_WR_LEN16_V(wrlen >> 4));
1076 	req->plen = cpu_to_be32(mpalen);
1077 	req->tunnel_to_proxy = cpu_to_be32(
1078 		FW_OFLD_TX_DATA_WR_FLUSH_F |
1079 		FW_OFLD_TX_DATA_WR_SHOVE_F);
1080 
1081 	mpa = (struct mpa_message *)(req + 1);
1082 	memset(mpa, 0, sizeof(*mpa));
1083 	memcpy(mpa->key, MPA_KEY_REP, sizeof(mpa->key));
1084 	mpa->flags = MPA_REJECT;
1085 	mpa->revision = ep->mpa_attr.version;
1086 	mpa->private_data_size = htons(plen);
1087 
1088 	if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) {
1089 		mpa->flags |= MPA_ENHANCED_RDMA_CONN;
1090 		mpa->private_data_size =
1091 			htons(ntohs(mpa->private_data_size) +
1092 			      sizeof(struct mpa_v2_conn_params));
1093 		mpa_v2_params.ird = htons(((u16)ep->ird) |
1094 					  (peer2peer ? MPA_V2_PEER2PEER_MODEL :
1095 					   0));
1096 		mpa_v2_params.ord = htons(((u16)ep->ord) | (peer2peer ?
1097 					  (p2p_type ==
1098 					   FW_RI_INIT_P2PTYPE_RDMA_WRITE ?
1099 					   MPA_V2_RDMA_WRITE_RTR : p2p_type ==
1100 					   FW_RI_INIT_P2PTYPE_READ_REQ ?
1101 					   MPA_V2_RDMA_READ_RTR : 0) : 0));
1102 		memcpy(mpa->private_data, &mpa_v2_params,
1103 		       sizeof(struct mpa_v2_conn_params));
1104 
1105 		if (ep->plen)
1106 			memcpy(mpa->private_data +
1107 			       sizeof(struct mpa_v2_conn_params), pdata, plen);
1108 	} else
1109 		if (plen)
1110 			memcpy(mpa->private_data, pdata, plen);
1111 
1112 	/*
1113 	 * Reference the mpa skb again.  This ensures the data area
1114 	 * will remain in memory until the hw acks the tx.
1115 	 * Function fw4_ack() will deref it.
1116 	 */
1117 	skb_get(skb);
1118 	set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
1119 	t4_set_arp_err_handler(skb, NULL, mpa_start_arp_failure);
1120 	ep->mpa_skb = skb;
1121 	ep->snd_seq += mpalen;
1122 	return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
1123 }
1124 
send_mpa_reply(struct c4iw_ep * ep,const void * pdata,u8 plen)1125 static int send_mpa_reply(struct c4iw_ep *ep, const void *pdata, u8 plen)
1126 {
1127 	int mpalen, wrlen;
1128 	struct fw_ofld_tx_data_wr *req;
1129 	struct mpa_message *mpa;
1130 	struct sk_buff *skb;
1131 	struct mpa_v2_conn_params mpa_v2_params;
1132 
1133 	pr_debug("ep %p tid %u pd_len %d\n",
1134 		 ep, ep->hwtid, ep->plen);
1135 
1136 	mpalen = sizeof(*mpa) + plen;
1137 	if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn)
1138 		mpalen += sizeof(struct mpa_v2_conn_params);
1139 	wrlen = roundup(mpalen + sizeof(*req), 16);
1140 
1141 	skb = get_skb(NULL, wrlen, GFP_KERNEL);
1142 	if (!skb) {
1143 		pr_err("%s - cannot alloc skb!\n", __func__);
1144 		return -ENOMEM;
1145 	}
1146 	set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
1147 
1148 	req = skb_put_zero(skb, wrlen);
1149 	req->op_to_immdlen = cpu_to_be32(
1150 		FW_WR_OP_V(FW_OFLD_TX_DATA_WR) |
1151 		FW_WR_COMPL_F |
1152 		FW_WR_IMMDLEN_V(mpalen));
1153 	req->flowid_len16 = cpu_to_be32(
1154 		FW_WR_FLOWID_V(ep->hwtid) |
1155 		FW_WR_LEN16_V(wrlen >> 4));
1156 	req->plen = cpu_to_be32(mpalen);
1157 	req->tunnel_to_proxy = cpu_to_be32(
1158 		FW_OFLD_TX_DATA_WR_FLUSH_F |
1159 		FW_OFLD_TX_DATA_WR_SHOVE_F);
1160 
1161 	mpa = (struct mpa_message *)(req + 1);
1162 	memset(mpa, 0, sizeof(*mpa));
1163 	memcpy(mpa->key, MPA_KEY_REP, sizeof(mpa->key));
1164 	mpa->flags = 0;
1165 	if (ep->mpa_attr.crc_enabled)
1166 		mpa->flags |= MPA_CRC;
1167 	if (ep->mpa_attr.recv_marker_enabled)
1168 		mpa->flags |= MPA_MARKERS;
1169 	mpa->revision = ep->mpa_attr.version;
1170 	mpa->private_data_size = htons(plen);
1171 
1172 	if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) {
1173 		mpa->flags |= MPA_ENHANCED_RDMA_CONN;
1174 		mpa->private_data_size =
1175 			htons(ntohs(mpa->private_data_size) +
1176 			      sizeof(struct mpa_v2_conn_params));
1177 		mpa_v2_params.ird = htons((u16)ep->ird);
1178 		mpa_v2_params.ord = htons((u16)ep->ord);
1179 		if (peer2peer && (ep->mpa_attr.p2p_type !=
1180 					FW_RI_INIT_P2PTYPE_DISABLED)) {
1181 			mpa_v2_params.ird |= htons(MPA_V2_PEER2PEER_MODEL);
1182 
1183 			if (p2p_type == FW_RI_INIT_P2PTYPE_RDMA_WRITE)
1184 				mpa_v2_params.ord |=
1185 					htons(MPA_V2_RDMA_WRITE_RTR);
1186 			else if (p2p_type == FW_RI_INIT_P2PTYPE_READ_REQ)
1187 				mpa_v2_params.ord |=
1188 					htons(MPA_V2_RDMA_READ_RTR);
1189 		}
1190 
1191 		memcpy(mpa->private_data, &mpa_v2_params,
1192 		       sizeof(struct mpa_v2_conn_params));
1193 
1194 		if (ep->plen)
1195 			memcpy(mpa->private_data +
1196 			       sizeof(struct mpa_v2_conn_params), pdata, plen);
1197 	} else
1198 		if (plen)
1199 			memcpy(mpa->private_data, pdata, plen);
1200 
1201 	/*
1202 	 * Reference the mpa skb.  This ensures the data area
1203 	 * will remain in memory until the hw acks the tx.
1204 	 * Function fw4_ack() will deref it.
1205 	 */
1206 	skb_get(skb);
1207 	t4_set_arp_err_handler(skb, NULL, mpa_start_arp_failure);
1208 	ep->mpa_skb = skb;
1209 	__state_set(&ep->com, MPA_REP_SENT);
1210 	ep->snd_seq += mpalen;
1211 	return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
1212 }
1213 
act_establish(struct c4iw_dev * dev,struct sk_buff * skb)1214 static int act_establish(struct c4iw_dev *dev, struct sk_buff *skb)
1215 {
1216 	struct c4iw_ep *ep;
1217 	struct cpl_act_establish *req = cplhdr(skb);
1218 	unsigned short tcp_opt = ntohs(req->tcp_opt);
1219 	unsigned int tid = GET_TID(req);
1220 	unsigned int atid = TID_TID_G(ntohl(req->tos_atid));
1221 	struct tid_info *t = dev->rdev.lldi.tids;
1222 	int ret;
1223 
1224 	ep = lookup_atid(t, atid);
1225 
1226 	pr_debug("ep %p tid %u snd_isn %u rcv_isn %u\n", ep, tid,
1227 		 be32_to_cpu(req->snd_isn), be32_to_cpu(req->rcv_isn));
1228 
1229 	mutex_lock(&ep->com.mutex);
1230 	dst_confirm(ep->dst);
1231 
1232 	/* setup the hwtid for this connection */
1233 	ep->hwtid = tid;
1234 	cxgb4_insert_tid(t, ep, tid, ep->com.local_addr.ss_family);
1235 	insert_ep_tid(ep);
1236 
1237 	ep->snd_seq = be32_to_cpu(req->snd_isn);
1238 	ep->rcv_seq = be32_to_cpu(req->rcv_isn);
1239 	ep->snd_wscale = TCPOPT_SND_WSCALE_G(tcp_opt);
1240 
1241 	set_emss(ep, tcp_opt);
1242 
1243 	/* dealloc the atid */
1244 	xa_erase_irq(&ep->com.dev->atids, atid);
1245 	cxgb4_free_atid(t, atid);
1246 	set_bit(ACT_ESTAB, &ep->com.history);
1247 
1248 	/* start MPA negotiation */
1249 	ret = send_flowc(ep);
1250 	if (ret)
1251 		goto err;
1252 	if (ep->retry_with_mpa_v1)
1253 		ret = send_mpa_req(ep, skb, 1);
1254 	else
1255 		ret = send_mpa_req(ep, skb, mpa_rev);
1256 	if (ret)
1257 		goto err;
1258 	mutex_unlock(&ep->com.mutex);
1259 	return 0;
1260 err:
1261 	mutex_unlock(&ep->com.mutex);
1262 	connect_reply_upcall(ep, -ENOMEM);
1263 	c4iw_ep_disconnect(ep, 0, GFP_KERNEL);
1264 	return 0;
1265 }
1266 
close_complete_upcall(struct c4iw_ep * ep,int status)1267 static void close_complete_upcall(struct c4iw_ep *ep, int status)
1268 {
1269 	struct iw_cm_event event;
1270 
1271 	pr_debug("ep %p tid %u\n", ep, ep->hwtid);
1272 	memset(&event, 0, sizeof(event));
1273 	event.event = IW_CM_EVENT_CLOSE;
1274 	event.status = status;
1275 	if (ep->com.cm_id) {
1276 		pr_debug("close complete delivered ep %p cm_id %p tid %u\n",
1277 			 ep, ep->com.cm_id, ep->hwtid);
1278 		ep->com.cm_id->event_handler(ep->com.cm_id, &event);
1279 		deref_cm_id(&ep->com);
1280 		set_bit(CLOSE_UPCALL, &ep->com.history);
1281 	}
1282 }
1283 
peer_close_upcall(struct c4iw_ep * ep)1284 static void peer_close_upcall(struct c4iw_ep *ep)
1285 {
1286 	struct iw_cm_event event;
1287 
1288 	pr_debug("ep %p tid %u\n", ep, ep->hwtid);
1289 	memset(&event, 0, sizeof(event));
1290 	event.event = IW_CM_EVENT_DISCONNECT;
1291 	if (ep->com.cm_id) {
1292 		pr_debug("peer close delivered ep %p cm_id %p tid %u\n",
1293 			 ep, ep->com.cm_id, ep->hwtid);
1294 		ep->com.cm_id->event_handler(ep->com.cm_id, &event);
1295 		set_bit(DISCONN_UPCALL, &ep->com.history);
1296 	}
1297 }
1298 
peer_abort_upcall(struct c4iw_ep * ep)1299 static void peer_abort_upcall(struct c4iw_ep *ep)
1300 {
1301 	struct iw_cm_event event;
1302 
1303 	pr_debug("ep %p tid %u\n", ep, ep->hwtid);
1304 	memset(&event, 0, sizeof(event));
1305 	event.event = IW_CM_EVENT_CLOSE;
1306 	event.status = -ECONNRESET;
1307 	if (ep->com.cm_id) {
1308 		pr_debug("abort delivered ep %p cm_id %p tid %u\n", ep,
1309 			 ep->com.cm_id, ep->hwtid);
1310 		ep->com.cm_id->event_handler(ep->com.cm_id, &event);
1311 		deref_cm_id(&ep->com);
1312 		set_bit(ABORT_UPCALL, &ep->com.history);
1313 	}
1314 }
1315 
connect_reply_upcall(struct c4iw_ep * ep,int status)1316 static void connect_reply_upcall(struct c4iw_ep *ep, int status)
1317 {
1318 	struct iw_cm_event event;
1319 
1320 	pr_debug("ep %p tid %u status %d\n",
1321 		 ep, ep->hwtid, status);
1322 	memset(&event, 0, sizeof(event));
1323 	event.event = IW_CM_EVENT_CONNECT_REPLY;
1324 	event.status = status;
1325 	memcpy(&event.local_addr, &ep->com.local_addr,
1326 	       sizeof(ep->com.local_addr));
1327 	memcpy(&event.remote_addr, &ep->com.remote_addr,
1328 	       sizeof(ep->com.remote_addr));
1329 
1330 	if ((status == 0) || (status == -ECONNREFUSED)) {
1331 		if (!ep->tried_with_mpa_v1) {
1332 			/* this means MPA_v2 is used */
1333 			event.ord = ep->ird;
1334 			event.ird = ep->ord;
1335 			event.private_data_len = ep->plen -
1336 				sizeof(struct mpa_v2_conn_params);
1337 			event.private_data = ep->mpa_pkt +
1338 				sizeof(struct mpa_message) +
1339 				sizeof(struct mpa_v2_conn_params);
1340 		} else {
1341 			/* this means MPA_v1 is used */
1342 			event.ord = cur_max_read_depth(ep->com.dev);
1343 			event.ird = cur_max_read_depth(ep->com.dev);
1344 			event.private_data_len = ep->plen;
1345 			event.private_data = ep->mpa_pkt +
1346 				sizeof(struct mpa_message);
1347 		}
1348 	}
1349 
1350 	pr_debug("ep %p tid %u status %d\n", ep,
1351 		 ep->hwtid, status);
1352 	set_bit(CONN_RPL_UPCALL, &ep->com.history);
1353 	ep->com.cm_id->event_handler(ep->com.cm_id, &event);
1354 
1355 	if (status < 0)
1356 		deref_cm_id(&ep->com);
1357 }
1358 
connect_request_upcall(struct c4iw_ep * ep)1359 static int connect_request_upcall(struct c4iw_ep *ep)
1360 {
1361 	struct iw_cm_event event;
1362 	int ret;
1363 
1364 	pr_debug("ep %p tid %u\n", ep, ep->hwtid);
1365 	memset(&event, 0, sizeof(event));
1366 	event.event = IW_CM_EVENT_CONNECT_REQUEST;
1367 	memcpy(&event.local_addr, &ep->com.local_addr,
1368 	       sizeof(ep->com.local_addr));
1369 	memcpy(&event.remote_addr, &ep->com.remote_addr,
1370 	       sizeof(ep->com.remote_addr));
1371 	event.provider_data = ep;
1372 	if (!ep->tried_with_mpa_v1) {
1373 		/* this means MPA_v2 is used */
1374 		event.ord = ep->ord;
1375 		event.ird = ep->ird;
1376 		event.private_data_len = ep->plen -
1377 			sizeof(struct mpa_v2_conn_params);
1378 		event.private_data = ep->mpa_pkt + sizeof(struct mpa_message) +
1379 			sizeof(struct mpa_v2_conn_params);
1380 	} else {
1381 		/* this means MPA_v1 is used. Send max supported */
1382 		event.ord = cur_max_read_depth(ep->com.dev);
1383 		event.ird = cur_max_read_depth(ep->com.dev);
1384 		event.private_data_len = ep->plen;
1385 		event.private_data = ep->mpa_pkt + sizeof(struct mpa_message);
1386 	}
1387 	c4iw_get_ep(&ep->com);
1388 	ret = ep->parent_ep->com.cm_id->event_handler(ep->parent_ep->com.cm_id,
1389 						      &event);
1390 	if (ret)
1391 		c4iw_put_ep(&ep->com);
1392 	set_bit(CONNREQ_UPCALL, &ep->com.history);
1393 	c4iw_put_ep(&ep->parent_ep->com);
1394 	return ret;
1395 }
1396 
established_upcall(struct c4iw_ep * ep)1397 static void established_upcall(struct c4iw_ep *ep)
1398 {
1399 	struct iw_cm_event event;
1400 
1401 	pr_debug("ep %p tid %u\n", ep, ep->hwtid);
1402 	memset(&event, 0, sizeof(event));
1403 	event.event = IW_CM_EVENT_ESTABLISHED;
1404 	event.ird = ep->ord;
1405 	event.ord = ep->ird;
1406 	if (ep->com.cm_id) {
1407 		pr_debug("ep %p tid %u\n", ep, ep->hwtid);
1408 		ep->com.cm_id->event_handler(ep->com.cm_id, &event);
1409 		set_bit(ESTAB_UPCALL, &ep->com.history);
1410 	}
1411 }
1412 
update_rx_credits(struct c4iw_ep * ep,u32 credits)1413 static int update_rx_credits(struct c4iw_ep *ep, u32 credits)
1414 {
1415 	struct sk_buff *skb;
1416 	u32 wrlen = roundup(sizeof(struct cpl_rx_data_ack), 16);
1417 	u32 credit_dack;
1418 
1419 	pr_debug("ep %p tid %u credits %u\n",
1420 		 ep, ep->hwtid, credits);
1421 	skb = get_skb(NULL, wrlen, GFP_KERNEL);
1422 	if (!skb) {
1423 		pr_err("update_rx_credits - cannot alloc skb!\n");
1424 		return 0;
1425 	}
1426 
1427 	/*
1428 	 * If we couldn't specify the entire rcv window at connection setup
1429 	 * due to the limit in the number of bits in the RCV_BUFSIZ field,
1430 	 * then add the overage in to the credits returned.
1431 	 */
1432 	if (ep->rcv_win > RCV_BUFSIZ_M * 1024)
1433 		credits += ep->rcv_win - RCV_BUFSIZ_M * 1024;
1434 
1435 	credit_dack = credits | RX_FORCE_ACK_F | RX_DACK_CHANGE_F |
1436 		      RX_DACK_MODE_V(dack_mode);
1437 
1438 	cxgb_mk_rx_data_ack(skb, wrlen, ep->hwtid, ep->ctrlq_idx,
1439 			    credit_dack);
1440 
1441 	c4iw_ofld_send(&ep->com.dev->rdev, skb);
1442 	return credits;
1443 }
1444 
1445 #define RELAXED_IRD_NEGOTIATION 1
1446 
1447 /*
1448  * process_mpa_reply - process streaming mode MPA reply
1449  *
1450  * Returns:
1451  *
1452  * 0 upon success indicating a connect request was delivered to the ULP
1453  * or the mpa request is incomplete but valid so far.
1454  *
1455  * 1 if a failure requires the caller to close the connection.
1456  *
1457  * 2 if a failure requires the caller to abort the connection.
1458  */
process_mpa_reply(struct c4iw_ep * ep,struct sk_buff * skb)1459 static int process_mpa_reply(struct c4iw_ep *ep, struct sk_buff *skb)
1460 {
1461 	struct mpa_message *mpa;
1462 	struct mpa_v2_conn_params *mpa_v2_params;
1463 	u16 plen;
1464 	u16 resp_ird, resp_ord;
1465 	u8 rtr_mismatch = 0, insuff_ird = 0;
1466 	struct c4iw_qp_attributes attrs;
1467 	enum c4iw_qp_attr_mask mask;
1468 	int err;
1469 	int disconnect = 0;
1470 
1471 	pr_debug("ep %p tid %u\n", ep, ep->hwtid);
1472 
1473 	/*
1474 	 * If we get more than the supported amount of private data
1475 	 * then we must fail this connection.
1476 	 */
1477 	if (ep->mpa_pkt_len + skb->len > sizeof(ep->mpa_pkt)) {
1478 		err = -EINVAL;
1479 		goto err_stop_timer;
1480 	}
1481 
1482 	/*
1483 	 * copy the new data into our accumulation buffer.
1484 	 */
1485 	skb_copy_from_linear_data(skb, &(ep->mpa_pkt[ep->mpa_pkt_len]),
1486 				  skb->len);
1487 	ep->mpa_pkt_len += skb->len;
1488 
1489 	/*
1490 	 * if we don't even have the mpa message, then bail.
1491 	 */
1492 	if (ep->mpa_pkt_len < sizeof(*mpa))
1493 		return 0;
1494 	mpa = (struct mpa_message *) ep->mpa_pkt;
1495 
1496 	/* Validate MPA header. */
1497 	if (mpa->revision > mpa_rev) {
1498 		pr_err("%s MPA version mismatch. Local = %d, Received = %d\n",
1499 		       __func__, mpa_rev, mpa->revision);
1500 		err = -EPROTO;
1501 		goto err_stop_timer;
1502 	}
1503 	if (memcmp(mpa->key, MPA_KEY_REP, sizeof(mpa->key))) {
1504 		err = -EPROTO;
1505 		goto err_stop_timer;
1506 	}
1507 
1508 	plen = ntohs(mpa->private_data_size);
1509 
1510 	/*
1511 	 * Fail if there's too much private data.
1512 	 */
1513 	if (plen > MPA_MAX_PRIVATE_DATA) {
1514 		err = -EPROTO;
1515 		goto err_stop_timer;
1516 	}
1517 
1518 	/*
1519 	 * If plen does not account for pkt size
1520 	 */
1521 	if (ep->mpa_pkt_len > (sizeof(*mpa) + plen)) {
1522 		err = -EPROTO;
1523 		goto err_stop_timer;
1524 	}
1525 
1526 	ep->plen = (u8) plen;
1527 
1528 	/*
1529 	 * If we don't have all the pdata yet, then bail.
1530 	 * We'll continue process when more data arrives.
1531 	 */
1532 	if (ep->mpa_pkt_len < (sizeof(*mpa) + plen))
1533 		return 0;
1534 
1535 	if (mpa->flags & MPA_REJECT) {
1536 		err = -ECONNREFUSED;
1537 		goto err_stop_timer;
1538 	}
1539 
1540 	/*
1541 	 * Stop mpa timer.  If it expired, then
1542 	 * we ignore the MPA reply.  process_timeout()
1543 	 * will abort the connection.
1544 	 */
1545 	if (stop_ep_timer(ep))
1546 		return 0;
1547 
1548 	/*
1549 	 * If we get here we have accumulated the entire mpa
1550 	 * start reply message including private data. And
1551 	 * the MPA header is valid.
1552 	 */
1553 	__state_set(&ep->com, FPDU_MODE);
1554 	ep->mpa_attr.crc_enabled = (mpa->flags & MPA_CRC) | crc_enabled ? 1 : 0;
1555 	ep->mpa_attr.xmit_marker_enabled = mpa->flags & MPA_MARKERS ? 1 : 0;
1556 	ep->mpa_attr.version = mpa->revision;
1557 	ep->mpa_attr.p2p_type = FW_RI_INIT_P2PTYPE_DISABLED;
1558 
1559 	if (mpa->revision == 2) {
1560 		ep->mpa_attr.enhanced_rdma_conn =
1561 			mpa->flags & MPA_ENHANCED_RDMA_CONN ? 1 : 0;
1562 		if (ep->mpa_attr.enhanced_rdma_conn) {
1563 			mpa_v2_params = (struct mpa_v2_conn_params *)
1564 				(ep->mpa_pkt + sizeof(*mpa));
1565 			resp_ird = ntohs(mpa_v2_params->ird) &
1566 				MPA_V2_IRD_ORD_MASK;
1567 			resp_ord = ntohs(mpa_v2_params->ord) &
1568 				MPA_V2_IRD_ORD_MASK;
1569 			pr_debug("responder ird %u ord %u ep ird %u ord %u\n",
1570 				 resp_ird, resp_ord, ep->ird, ep->ord);
1571 
1572 			/*
1573 			 * This is a double-check. Ideally, below checks are
1574 			 * not required since ird/ord stuff has been taken
1575 			 * care of in c4iw_accept_cr
1576 			 */
1577 			if (ep->ird < resp_ord) {
1578 				if (RELAXED_IRD_NEGOTIATION && resp_ord <=
1579 				    ep->com.dev->rdev.lldi.max_ordird_qp)
1580 					ep->ird = resp_ord;
1581 				else
1582 					insuff_ird = 1;
1583 			} else if (ep->ird > resp_ord) {
1584 				ep->ird = resp_ord;
1585 			}
1586 			if (ep->ord > resp_ird) {
1587 				if (RELAXED_IRD_NEGOTIATION)
1588 					ep->ord = resp_ird;
1589 				else
1590 					insuff_ird = 1;
1591 			}
1592 			if (insuff_ird) {
1593 				err = -ENOMEM;
1594 				ep->ird = resp_ord;
1595 				ep->ord = resp_ird;
1596 			}
1597 
1598 			if (ntohs(mpa_v2_params->ird) &
1599 					MPA_V2_PEER2PEER_MODEL) {
1600 				if (ntohs(mpa_v2_params->ord) &
1601 						MPA_V2_RDMA_WRITE_RTR)
1602 					ep->mpa_attr.p2p_type =
1603 						FW_RI_INIT_P2PTYPE_RDMA_WRITE;
1604 				else if (ntohs(mpa_v2_params->ord) &
1605 						MPA_V2_RDMA_READ_RTR)
1606 					ep->mpa_attr.p2p_type =
1607 						FW_RI_INIT_P2PTYPE_READ_REQ;
1608 			}
1609 		}
1610 	} else if (mpa->revision == 1)
1611 		if (peer2peer)
1612 			ep->mpa_attr.p2p_type = p2p_type;
1613 
1614 	pr_debug("crc_enabled=%d, recv_marker_enabled=%d, xmit_marker_enabled=%d, version=%d p2p_type=%d local-p2p_type = %d\n",
1615 		 ep->mpa_attr.crc_enabled,
1616 		 ep->mpa_attr.recv_marker_enabled,
1617 		 ep->mpa_attr.xmit_marker_enabled, ep->mpa_attr.version,
1618 		 ep->mpa_attr.p2p_type, p2p_type);
1619 
1620 	/*
1621 	 * If responder's RTR does not match with that of initiator, assign
1622 	 * FW_RI_INIT_P2PTYPE_DISABLED in mpa attributes so that RTR is not
1623 	 * generated when moving QP to RTS state.
1624 	 * A TERM message will be sent after QP has moved to RTS state
1625 	 */
1626 	if ((ep->mpa_attr.version == 2) && peer2peer &&
1627 			(ep->mpa_attr.p2p_type != p2p_type)) {
1628 		ep->mpa_attr.p2p_type = FW_RI_INIT_P2PTYPE_DISABLED;
1629 		rtr_mismatch = 1;
1630 	}
1631 
1632 	attrs.mpa_attr = ep->mpa_attr;
1633 	attrs.max_ird = ep->ird;
1634 	attrs.max_ord = ep->ord;
1635 	attrs.llp_stream_handle = ep;
1636 	attrs.next_state = C4IW_QP_STATE_RTS;
1637 
1638 	mask = C4IW_QP_ATTR_NEXT_STATE |
1639 	    C4IW_QP_ATTR_LLP_STREAM_HANDLE | C4IW_QP_ATTR_MPA_ATTR |
1640 	    C4IW_QP_ATTR_MAX_IRD | C4IW_QP_ATTR_MAX_ORD;
1641 
1642 	/* bind QP and TID with INIT_WR */
1643 	err = c4iw_modify_qp(ep->com.qp->rhp,
1644 			     ep->com.qp, mask, &attrs, 1);
1645 	if (err)
1646 		goto err;
1647 
1648 	/*
1649 	 * If responder's RTR requirement did not match with what initiator
1650 	 * supports, generate TERM message
1651 	 */
1652 	if (rtr_mismatch) {
1653 		pr_err("%s: RTR mismatch, sending TERM\n", __func__);
1654 		attrs.layer_etype = LAYER_MPA | DDP_LLP;
1655 		attrs.ecode = MPA_NOMATCH_RTR;
1656 		attrs.next_state = C4IW_QP_STATE_TERMINATE;
1657 		attrs.send_term = 1;
1658 		err = c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
1659 				C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
1660 		err = -ENOMEM;
1661 		disconnect = 1;
1662 		goto out;
1663 	}
1664 
1665 	/*
1666 	 * Generate TERM if initiator IRD is not sufficient for responder
1667 	 * provided ORD. Currently, we do the same behaviour even when
1668 	 * responder provided IRD is also not sufficient as regards to
1669 	 * initiator ORD.
1670 	 */
1671 	if (insuff_ird) {
1672 		pr_err("%s: Insufficient IRD, sending TERM\n", __func__);
1673 		attrs.layer_etype = LAYER_MPA | DDP_LLP;
1674 		attrs.ecode = MPA_INSUFF_IRD;
1675 		attrs.next_state = C4IW_QP_STATE_TERMINATE;
1676 		attrs.send_term = 1;
1677 		err = c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
1678 				C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
1679 		err = -ENOMEM;
1680 		disconnect = 1;
1681 		goto out;
1682 	}
1683 	goto out;
1684 err_stop_timer:
1685 	stop_ep_timer(ep);
1686 err:
1687 	disconnect = 2;
1688 out:
1689 	connect_reply_upcall(ep, err);
1690 	return disconnect;
1691 }
1692 
1693 /*
1694  * process_mpa_request - process streaming mode MPA request
1695  *
1696  * Returns:
1697  *
1698  * 0 upon success indicating a connect request was delivered to the ULP
1699  * or the mpa request is incomplete but valid so far.
1700  *
1701  * 1 if a failure requires the caller to close the connection.
1702  *
1703  * 2 if a failure requires the caller to abort the connection.
1704  */
process_mpa_request(struct c4iw_ep * ep,struct sk_buff * skb)1705 static int process_mpa_request(struct c4iw_ep *ep, struct sk_buff *skb)
1706 {
1707 	struct mpa_message *mpa;
1708 	struct mpa_v2_conn_params *mpa_v2_params;
1709 	u16 plen;
1710 
1711 	pr_debug("ep %p tid %u\n", ep, ep->hwtid);
1712 
1713 	/*
1714 	 * If we get more than the supported amount of private data
1715 	 * then we must fail this connection.
1716 	 */
1717 	if (ep->mpa_pkt_len + skb->len > sizeof(ep->mpa_pkt))
1718 		goto err_stop_timer;
1719 
1720 	pr_debug("enter (%s line %u)\n", __FILE__, __LINE__);
1721 
1722 	/*
1723 	 * Copy the new data into our accumulation buffer.
1724 	 */
1725 	skb_copy_from_linear_data(skb, &(ep->mpa_pkt[ep->mpa_pkt_len]),
1726 				  skb->len);
1727 	ep->mpa_pkt_len += skb->len;
1728 
1729 	/*
1730 	 * If we don't even have the mpa message, then bail.
1731 	 * We'll continue process when more data arrives.
1732 	 */
1733 	if (ep->mpa_pkt_len < sizeof(*mpa))
1734 		return 0;
1735 
1736 	pr_debug("enter (%s line %u)\n", __FILE__, __LINE__);
1737 	mpa = (struct mpa_message *) ep->mpa_pkt;
1738 
1739 	/*
1740 	 * Validate MPA Header.
1741 	 */
1742 	if (mpa->revision > mpa_rev) {
1743 		pr_err("%s MPA version mismatch. Local = %d, Received = %d\n",
1744 		       __func__, mpa_rev, mpa->revision);
1745 		goto err_stop_timer;
1746 	}
1747 
1748 	if (memcmp(mpa->key, MPA_KEY_REQ, sizeof(mpa->key)))
1749 		goto err_stop_timer;
1750 
1751 	plen = ntohs(mpa->private_data_size);
1752 
1753 	/*
1754 	 * Fail if there's too much private data.
1755 	 */
1756 	if (plen > MPA_MAX_PRIVATE_DATA)
1757 		goto err_stop_timer;
1758 
1759 	/*
1760 	 * If plen does not account for pkt size
1761 	 */
1762 	if (ep->mpa_pkt_len > (sizeof(*mpa) + plen))
1763 		goto err_stop_timer;
1764 	ep->plen = (u8) plen;
1765 
1766 	/*
1767 	 * If we don't have all the pdata yet, then bail.
1768 	 */
1769 	if (ep->mpa_pkt_len < (sizeof(*mpa) + plen))
1770 		return 0;
1771 
1772 	/*
1773 	 * If we get here we have accumulated the entire mpa
1774 	 * start reply message including private data.
1775 	 */
1776 	ep->mpa_attr.initiator = 0;
1777 	ep->mpa_attr.crc_enabled = (mpa->flags & MPA_CRC) | crc_enabled ? 1 : 0;
1778 	ep->mpa_attr.recv_marker_enabled = markers_enabled;
1779 	ep->mpa_attr.xmit_marker_enabled = mpa->flags & MPA_MARKERS ? 1 : 0;
1780 	ep->mpa_attr.version = mpa->revision;
1781 	if (mpa->revision == 1)
1782 		ep->tried_with_mpa_v1 = 1;
1783 	ep->mpa_attr.p2p_type = FW_RI_INIT_P2PTYPE_DISABLED;
1784 
1785 	if (mpa->revision == 2) {
1786 		ep->mpa_attr.enhanced_rdma_conn =
1787 			mpa->flags & MPA_ENHANCED_RDMA_CONN ? 1 : 0;
1788 		if (ep->mpa_attr.enhanced_rdma_conn) {
1789 			mpa_v2_params = (struct mpa_v2_conn_params *)
1790 				(ep->mpa_pkt + sizeof(*mpa));
1791 			ep->ird = ntohs(mpa_v2_params->ird) &
1792 				MPA_V2_IRD_ORD_MASK;
1793 			ep->ird = min_t(u32, ep->ird,
1794 					cur_max_read_depth(ep->com.dev));
1795 			ep->ord = ntohs(mpa_v2_params->ord) &
1796 				MPA_V2_IRD_ORD_MASK;
1797 			ep->ord = min_t(u32, ep->ord,
1798 					cur_max_read_depth(ep->com.dev));
1799 			pr_debug("initiator ird %u ord %u\n",
1800 				 ep->ird, ep->ord);
1801 			if (ntohs(mpa_v2_params->ird) & MPA_V2_PEER2PEER_MODEL)
1802 				if (peer2peer) {
1803 					if (ntohs(mpa_v2_params->ord) &
1804 							MPA_V2_RDMA_WRITE_RTR)
1805 						ep->mpa_attr.p2p_type =
1806 						FW_RI_INIT_P2PTYPE_RDMA_WRITE;
1807 					else if (ntohs(mpa_v2_params->ord) &
1808 							MPA_V2_RDMA_READ_RTR)
1809 						ep->mpa_attr.p2p_type =
1810 						FW_RI_INIT_P2PTYPE_READ_REQ;
1811 				}
1812 		}
1813 	} else if (mpa->revision == 1)
1814 		if (peer2peer)
1815 			ep->mpa_attr.p2p_type = p2p_type;
1816 
1817 	pr_debug("crc_enabled=%d, recv_marker_enabled=%d, xmit_marker_enabled=%d, version=%d p2p_type=%d\n",
1818 		 ep->mpa_attr.crc_enabled, ep->mpa_attr.recv_marker_enabled,
1819 		 ep->mpa_attr.xmit_marker_enabled, ep->mpa_attr.version,
1820 		 ep->mpa_attr.p2p_type);
1821 
1822 	__state_set(&ep->com, MPA_REQ_RCVD);
1823 
1824 	/* drive upcall */
1825 	mutex_lock_nested(&ep->parent_ep->com.mutex, SINGLE_DEPTH_NESTING);
1826 	if (ep->parent_ep->com.state != DEAD) {
1827 		if (connect_request_upcall(ep))
1828 			goto err_unlock_parent;
1829 	} else {
1830 		goto err_unlock_parent;
1831 	}
1832 	mutex_unlock(&ep->parent_ep->com.mutex);
1833 	return 0;
1834 
1835 err_unlock_parent:
1836 	mutex_unlock(&ep->parent_ep->com.mutex);
1837 	goto err_out;
1838 err_stop_timer:
1839 	(void)stop_ep_timer(ep);
1840 err_out:
1841 	return 2;
1842 }
1843 
rx_data(struct c4iw_dev * dev,struct sk_buff * skb)1844 static int rx_data(struct c4iw_dev *dev, struct sk_buff *skb)
1845 {
1846 	struct c4iw_ep *ep;
1847 	struct cpl_rx_data *hdr = cplhdr(skb);
1848 	unsigned int dlen = ntohs(hdr->len);
1849 	unsigned int tid = GET_TID(hdr);
1850 	__u8 status = hdr->status;
1851 	int disconnect = 0;
1852 
1853 	ep = get_ep_from_tid(dev, tid);
1854 	if (!ep)
1855 		return 0;
1856 	pr_debug("ep %p tid %u dlen %u\n", ep, ep->hwtid, dlen);
1857 	skb_pull(skb, sizeof(*hdr));
1858 	skb_trim(skb, dlen);
1859 	mutex_lock(&ep->com.mutex);
1860 
1861 	switch (ep->com.state) {
1862 	case MPA_REQ_SENT:
1863 		update_rx_credits(ep, dlen);
1864 		ep->rcv_seq += dlen;
1865 		disconnect = process_mpa_reply(ep, skb);
1866 		break;
1867 	case MPA_REQ_WAIT:
1868 		update_rx_credits(ep, dlen);
1869 		ep->rcv_seq += dlen;
1870 		disconnect = process_mpa_request(ep, skb);
1871 		break;
1872 	case FPDU_MODE: {
1873 		struct c4iw_qp_attributes attrs;
1874 
1875 		update_rx_credits(ep, dlen);
1876 		if (status)
1877 			pr_err("%s Unexpected streaming data." \
1878 			       " qpid %u ep %p state %d tid %u status %d\n",
1879 			       __func__, ep->com.qp->wq.sq.qid, ep,
1880 			       ep->com.state, ep->hwtid, status);
1881 		attrs.next_state = C4IW_QP_STATE_TERMINATE;
1882 		c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
1883 			       C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
1884 		disconnect = 1;
1885 		break;
1886 	}
1887 	default:
1888 		break;
1889 	}
1890 	mutex_unlock(&ep->com.mutex);
1891 	if (disconnect)
1892 		c4iw_ep_disconnect(ep, disconnect == 2, GFP_KERNEL);
1893 	c4iw_put_ep(&ep->com);
1894 	return 0;
1895 }
1896 
complete_cached_srq_buffers(struct c4iw_ep * ep,u32 srqidx)1897 static void complete_cached_srq_buffers(struct c4iw_ep *ep, u32 srqidx)
1898 {
1899 	enum chip_type adapter_type;
1900 
1901 	adapter_type = ep->com.dev->rdev.lldi.adapter_type;
1902 
1903 	/*
1904 	 * If this TCB had a srq buffer cached, then we must complete
1905 	 * it. For user mode, that means saving the srqidx in the
1906 	 * user/kernel status page for this qp.  For kernel mode, just
1907 	 * synthesize the CQE now.
1908 	 */
1909 	if (CHELSIO_CHIP_VERSION(adapter_type) > CHELSIO_T5 && srqidx) {
1910 		if (ep->com.qp->ibqp.uobject)
1911 			t4_set_wq_in_error(&ep->com.qp->wq, srqidx);
1912 		else
1913 			c4iw_flush_srqidx(ep->com.qp, srqidx);
1914 	}
1915 }
1916 
abort_rpl(struct c4iw_dev * dev,struct sk_buff * skb)1917 static int abort_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
1918 {
1919 	u32 srqidx;
1920 	struct c4iw_ep *ep;
1921 	struct cpl_abort_rpl_rss6 *rpl = cplhdr(skb);
1922 	int release = 0;
1923 	unsigned int tid = GET_TID(rpl);
1924 
1925 	ep = get_ep_from_tid(dev, tid);
1926 	if (!ep) {
1927 		pr_warn("Abort rpl to freed endpoint\n");
1928 		return 0;
1929 	}
1930 
1931 	if (ep->com.qp && ep->com.qp->srq) {
1932 		srqidx = ABORT_RSS_SRQIDX_G(be32_to_cpu(rpl->srqidx_status));
1933 		complete_cached_srq_buffers(ep, srqidx ? srqidx : ep->srqe_idx);
1934 	}
1935 
1936 	pr_debug("ep %p tid %u\n", ep, ep->hwtid);
1937 	mutex_lock(&ep->com.mutex);
1938 	switch (ep->com.state) {
1939 	case ABORTING:
1940 		c4iw_wake_up_noref(ep->com.wr_waitp, -ECONNRESET);
1941 		__state_set(&ep->com, DEAD);
1942 		release = 1;
1943 		break;
1944 	default:
1945 		pr_err("%s ep %p state %d\n", __func__, ep, ep->com.state);
1946 		break;
1947 	}
1948 	mutex_unlock(&ep->com.mutex);
1949 
1950 	if (release) {
1951 		close_complete_upcall(ep, -ECONNRESET);
1952 		release_ep_resources(ep);
1953 	}
1954 	c4iw_put_ep(&ep->com);
1955 	return 0;
1956 }
1957 
send_fw_act_open_req(struct c4iw_ep * ep,unsigned int atid)1958 static int send_fw_act_open_req(struct c4iw_ep *ep, unsigned int atid)
1959 {
1960 	struct sk_buff *skb;
1961 	struct fw_ofld_connection_wr *req;
1962 	unsigned int mtu_idx;
1963 	u32 wscale;
1964 	struct sockaddr_in *sin;
1965 	int win;
1966 
1967 	skb = get_skb(NULL, sizeof(*req), GFP_KERNEL);
1968 	if (!skb)
1969 		return -ENOMEM;
1970 
1971 	req = __skb_put_zero(skb, sizeof(*req));
1972 	req->op_compl = htonl(WR_OP_V(FW_OFLD_CONNECTION_WR));
1973 	req->len16_pkd = htonl(FW_WR_LEN16_V(DIV_ROUND_UP(sizeof(*req), 16)));
1974 	req->le.filter = cpu_to_be32(cxgb4_select_ntuple(
1975 				     ep->com.dev->rdev.lldi.ports[0],
1976 				     ep->l2t));
1977 	sin = (struct sockaddr_in *)&ep->com.local_addr;
1978 	req->le.lport = sin->sin_port;
1979 	req->le.u.ipv4.lip = sin->sin_addr.s_addr;
1980 	sin = (struct sockaddr_in *)&ep->com.remote_addr;
1981 	req->le.pport = sin->sin_port;
1982 	req->le.u.ipv4.pip = sin->sin_addr.s_addr;
1983 	req->tcb.t_state_to_astid =
1984 			htonl(FW_OFLD_CONNECTION_WR_T_STATE_V(TCP_SYN_SENT) |
1985 			FW_OFLD_CONNECTION_WR_ASTID_V(atid));
1986 	req->tcb.cplrxdataack_cplpassacceptrpl =
1987 			htons(FW_OFLD_CONNECTION_WR_CPLRXDATAACK_F);
1988 	req->tcb.tx_max = (__force __be32) jiffies;
1989 	req->tcb.rcv_adv = htons(1);
1990 	cxgb_best_mtu(ep->com.dev->rdev.lldi.mtus, ep->mtu, &mtu_idx,
1991 		      enable_tcp_timestamps,
1992 		      (ep->com.remote_addr.ss_family == AF_INET) ? 0 : 1);
1993 	wscale = cxgb_compute_wscale(rcv_win);
1994 
1995 	/*
1996 	 * Specify the largest window that will fit in opt0. The
1997 	 * remainder will be specified in the rx_data_ack.
1998 	 */
1999 	win = ep->rcv_win >> 10;
2000 	if (win > RCV_BUFSIZ_M)
2001 		win = RCV_BUFSIZ_M;
2002 
2003 	req->tcb.opt0 = (__force __be64) (TCAM_BYPASS_F |
2004 		(nocong ? NO_CONG_F : 0) |
2005 		KEEP_ALIVE_F |
2006 		DELACK_F |
2007 		WND_SCALE_V(wscale) |
2008 		MSS_IDX_V(mtu_idx) |
2009 		L2T_IDX_V(ep->l2t->idx) |
2010 		TX_CHAN_V(ep->tx_chan) |
2011 		SMAC_SEL_V(ep->smac_idx) |
2012 		DSCP_V(ep->tos >> 2) |
2013 		ULP_MODE_V(ULP_MODE_TCPDDP) |
2014 		RCV_BUFSIZ_V(win));
2015 	req->tcb.opt2 = (__force __be32) (PACE_V(1) |
2016 		TX_QUEUE_V(ep->com.dev->rdev.lldi.tx_modq[ep->tx_chan]) |
2017 		RX_CHANNEL_V(0) |
2018 		CCTRL_ECN_V(enable_ecn) |
2019 		RSS_QUEUE_VALID_F | RSS_QUEUE_V(ep->rss_qid));
2020 	if (enable_tcp_timestamps)
2021 		req->tcb.opt2 |= (__force __be32)TSTAMPS_EN_F;
2022 	if (enable_tcp_sack)
2023 		req->tcb.opt2 |= (__force __be32)SACK_EN_F;
2024 	if (wscale && enable_tcp_window_scaling)
2025 		req->tcb.opt2 |= (__force __be32)WND_SCALE_EN_F;
2026 	req->tcb.opt0 = cpu_to_be64((__force u64)req->tcb.opt0);
2027 	req->tcb.opt2 = cpu_to_be32((__force u32)req->tcb.opt2);
2028 	set_wr_txq(skb, CPL_PRIORITY_CONTROL, ep->ctrlq_idx);
2029 	set_bit(ACT_OFLD_CONN, &ep->com.history);
2030 	return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
2031 }
2032 
2033 /*
2034  * Some of the error codes above implicitly indicate that there is no TID
2035  * allocated with the result of an ACT_OPEN.  We use this predicate to make
2036  * that explicit.
2037  */
act_open_has_tid(int status)2038 static inline int act_open_has_tid(int status)
2039 {
2040 	return (status != CPL_ERR_TCAM_PARITY &&
2041 		status != CPL_ERR_TCAM_MISS &&
2042 		status != CPL_ERR_TCAM_FULL &&
2043 		status != CPL_ERR_CONN_EXIST_SYNRECV &&
2044 		status != CPL_ERR_CONN_EXIST);
2045 }
2046 
neg_adv_str(unsigned int status)2047 static char *neg_adv_str(unsigned int status)
2048 {
2049 	switch (status) {
2050 	case CPL_ERR_RTX_NEG_ADVICE:
2051 		return "Retransmit timeout";
2052 	case CPL_ERR_PERSIST_NEG_ADVICE:
2053 		return "Persist timeout";
2054 	case CPL_ERR_KEEPALV_NEG_ADVICE:
2055 		return "Keepalive timeout";
2056 	default:
2057 		return "Unknown";
2058 	}
2059 }
2060 
set_tcp_window(struct c4iw_ep * ep,struct port_info * pi)2061 static void set_tcp_window(struct c4iw_ep *ep, struct port_info *pi)
2062 {
2063 	ep->snd_win = snd_win;
2064 	ep->rcv_win = rcv_win;
2065 	pr_debug("snd_win %d rcv_win %d\n",
2066 		 ep->snd_win, ep->rcv_win);
2067 }
2068 
2069 #define ACT_OPEN_RETRY_COUNT 2
2070 
import_ep(struct c4iw_ep * ep,int iptype,__u8 * peer_ip,struct dst_entry * dst,struct c4iw_dev * cdev,bool clear_mpa_v1,enum chip_type adapter_type,u8 tos)2071 static int import_ep(struct c4iw_ep *ep, int iptype, __u8 *peer_ip,
2072 		     struct dst_entry *dst, struct c4iw_dev *cdev,
2073 		     bool clear_mpa_v1, enum chip_type adapter_type, u8 tos)
2074 {
2075 	struct neighbour *n;
2076 	int err, step;
2077 	struct net_device *pdev;
2078 
2079 	n = dst_neigh_lookup(dst, peer_ip);
2080 	if (!n)
2081 		return -ENODEV;
2082 
2083 	rcu_read_lock();
2084 	err = -ENOMEM;
2085 	if (n->dev->flags & IFF_LOOPBACK) {
2086 		if (iptype == 4)
2087 			pdev = ip_dev_find(&init_net, *(__be32 *)peer_ip);
2088 		else if (IS_ENABLED(CONFIG_IPV6))
2089 			for_each_netdev(&init_net, pdev) {
2090 				if (ipv6_chk_addr(&init_net,
2091 						  (struct in6_addr *)peer_ip,
2092 						  pdev, 1))
2093 					break;
2094 			}
2095 		else
2096 			pdev = NULL;
2097 
2098 		if (!pdev) {
2099 			err = -ENODEV;
2100 			goto out;
2101 		}
2102 		ep->l2t = cxgb4_l2t_get(cdev->rdev.lldi.l2t,
2103 					n, pdev, rt_tos2priority(tos));
2104 		if (!ep->l2t) {
2105 			dev_put(pdev);
2106 			goto out;
2107 		}
2108 		ep->mtu = pdev->mtu;
2109 		ep->tx_chan = cxgb4_port_chan(pdev);
2110 		ep->smac_idx = ((struct port_info *)netdev_priv(pdev))->smt_idx;
2111 		step = cdev->rdev.lldi.ntxq /
2112 			cdev->rdev.lldi.nchan;
2113 		ep->txq_idx = cxgb4_port_idx(pdev) * step;
2114 		step = cdev->rdev.lldi.nrxq /
2115 			cdev->rdev.lldi.nchan;
2116 		ep->ctrlq_idx = cxgb4_port_idx(pdev);
2117 		ep->rss_qid = cdev->rdev.lldi.rxq_ids[
2118 			cxgb4_port_idx(pdev) * step];
2119 		set_tcp_window(ep, (struct port_info *)netdev_priv(pdev));
2120 		dev_put(pdev);
2121 	} else {
2122 		pdev = get_real_dev(n->dev);
2123 		ep->l2t = cxgb4_l2t_get(cdev->rdev.lldi.l2t,
2124 					n, pdev, rt_tos2priority(tos));
2125 		if (!ep->l2t)
2126 			goto out;
2127 		ep->mtu = dst_mtu(dst);
2128 		ep->tx_chan = cxgb4_port_chan(pdev);
2129 		ep->smac_idx = ((struct port_info *)netdev_priv(pdev))->smt_idx;
2130 		step = cdev->rdev.lldi.ntxq /
2131 			cdev->rdev.lldi.nchan;
2132 		ep->txq_idx = cxgb4_port_idx(pdev) * step;
2133 		ep->ctrlq_idx = cxgb4_port_idx(pdev);
2134 		step = cdev->rdev.lldi.nrxq /
2135 			cdev->rdev.lldi.nchan;
2136 		ep->rss_qid = cdev->rdev.lldi.rxq_ids[
2137 			cxgb4_port_idx(pdev) * step];
2138 		set_tcp_window(ep, (struct port_info *)netdev_priv(pdev));
2139 
2140 		if (clear_mpa_v1) {
2141 			ep->retry_with_mpa_v1 = 0;
2142 			ep->tried_with_mpa_v1 = 0;
2143 		}
2144 	}
2145 	err = 0;
2146 out:
2147 	rcu_read_unlock();
2148 
2149 	neigh_release(n);
2150 
2151 	return err;
2152 }
2153 
c4iw_reconnect(struct c4iw_ep * ep)2154 static int c4iw_reconnect(struct c4iw_ep *ep)
2155 {
2156 	int err = 0;
2157 	int size = 0;
2158 	struct sockaddr_in *laddr = (struct sockaddr_in *)
2159 				    &ep->com.cm_id->m_local_addr;
2160 	struct sockaddr_in *raddr = (struct sockaddr_in *)
2161 				    &ep->com.cm_id->m_remote_addr;
2162 	struct sockaddr_in6 *laddr6 = (struct sockaddr_in6 *)
2163 				      &ep->com.cm_id->m_local_addr;
2164 	struct sockaddr_in6 *raddr6 = (struct sockaddr_in6 *)
2165 				      &ep->com.cm_id->m_remote_addr;
2166 	int iptype;
2167 	__u8 *ra;
2168 
2169 	pr_debug("qp %p cm_id %p\n", ep->com.qp, ep->com.cm_id);
2170 	c4iw_init_wr_wait(ep->com.wr_waitp);
2171 
2172 	/* When MPA revision is different on nodes, the node with MPA_rev=2
2173 	 * tries to reconnect with MPA_rev 1 for the same EP through
2174 	 * c4iw_reconnect(), where the same EP is assigned with new tid for
2175 	 * further connection establishment. As we are using the same EP pointer
2176 	 * for reconnect, few skbs are used during the previous c4iw_connect(),
2177 	 * which leaves the EP with inadequate skbs for further
2178 	 * c4iw_reconnect(), Further causing a crash due to an empty
2179 	 * skb_list() during peer_abort(). Allocate skbs which is already used.
2180 	 */
2181 	size = (CN_MAX_CON_BUF - skb_queue_len(&ep->com.ep_skb_list));
2182 	if (alloc_ep_skb_list(&ep->com.ep_skb_list, size)) {
2183 		err = -ENOMEM;
2184 		goto fail1;
2185 	}
2186 
2187 	/*
2188 	 * Allocate an active TID to initiate a TCP connection.
2189 	 */
2190 	ep->atid = cxgb4_alloc_atid(ep->com.dev->rdev.lldi.tids, ep);
2191 	if (ep->atid == -1) {
2192 		pr_err("%s - cannot alloc atid\n", __func__);
2193 		err = -ENOMEM;
2194 		goto fail2;
2195 	}
2196 	err = xa_insert_irq(&ep->com.dev->atids, ep->atid, ep, GFP_KERNEL);
2197 	if (err)
2198 		goto fail2a;
2199 
2200 	/* find a route */
2201 	if (ep->com.cm_id->m_local_addr.ss_family == AF_INET) {
2202 		ep->dst = cxgb_find_route(&ep->com.dev->rdev.lldi, get_real_dev,
2203 					  laddr->sin_addr.s_addr,
2204 					  raddr->sin_addr.s_addr,
2205 					  laddr->sin_port,
2206 					  raddr->sin_port, ep->com.cm_id->tos);
2207 		iptype = 4;
2208 		ra = (__u8 *)&raddr->sin_addr;
2209 	} else {
2210 		ep->dst = cxgb_find_route6(&ep->com.dev->rdev.lldi,
2211 					   get_real_dev,
2212 					   laddr6->sin6_addr.s6_addr,
2213 					   raddr6->sin6_addr.s6_addr,
2214 					   laddr6->sin6_port,
2215 					   raddr6->sin6_port,
2216 					   ep->com.cm_id->tos,
2217 					   raddr6->sin6_scope_id);
2218 		iptype = 6;
2219 		ra = (__u8 *)&raddr6->sin6_addr;
2220 	}
2221 	if (!ep->dst) {
2222 		pr_err("%s - cannot find route\n", __func__);
2223 		err = -EHOSTUNREACH;
2224 		goto fail3;
2225 	}
2226 	err = import_ep(ep, iptype, ra, ep->dst, ep->com.dev, false,
2227 			ep->com.dev->rdev.lldi.adapter_type,
2228 			ep->com.cm_id->tos);
2229 	if (err) {
2230 		pr_err("%s - cannot alloc l2e\n", __func__);
2231 		goto fail4;
2232 	}
2233 
2234 	pr_debug("txq_idx %u tx_chan %u smac_idx %u rss_qid %u l2t_idx %u\n",
2235 		 ep->txq_idx, ep->tx_chan, ep->smac_idx, ep->rss_qid,
2236 		 ep->l2t->idx);
2237 
2238 	state_set(&ep->com, CONNECTING);
2239 	ep->tos = ep->com.cm_id->tos;
2240 
2241 	/* send connect request to rnic */
2242 	err = send_connect(ep);
2243 	if (!err)
2244 		goto out;
2245 
2246 	cxgb4_l2t_release(ep->l2t);
2247 fail4:
2248 	dst_release(ep->dst);
2249 fail3:
2250 	xa_erase_irq(&ep->com.dev->atids, ep->atid);
2251 fail2a:
2252 	cxgb4_free_atid(ep->com.dev->rdev.lldi.tids, ep->atid);
2253 fail2:
2254 	/*
2255 	 * remember to send notification to upper layer.
2256 	 * We are in here so the upper layer is not aware that this is
2257 	 * re-connect attempt and so, upper layer is still waiting for
2258 	 * response of 1st connect request.
2259 	 */
2260 	connect_reply_upcall(ep, -ECONNRESET);
2261 fail1:
2262 	c4iw_put_ep(&ep->com);
2263 out:
2264 	return err;
2265 }
2266 
act_open_rpl(struct c4iw_dev * dev,struct sk_buff * skb)2267 static int act_open_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
2268 {
2269 	struct c4iw_ep *ep;
2270 	struct cpl_act_open_rpl *rpl = cplhdr(skb);
2271 	unsigned int atid = TID_TID_G(AOPEN_ATID_G(
2272 				      ntohl(rpl->atid_status)));
2273 	struct tid_info *t = dev->rdev.lldi.tids;
2274 	int status = AOPEN_STATUS_G(ntohl(rpl->atid_status));
2275 	struct sockaddr_in *la;
2276 	struct sockaddr_in *ra;
2277 	struct sockaddr_in6 *la6;
2278 	struct sockaddr_in6 *ra6;
2279 	int ret = 0;
2280 
2281 	ep = lookup_atid(t, atid);
2282 	la = (struct sockaddr_in *)&ep->com.local_addr;
2283 	ra = (struct sockaddr_in *)&ep->com.remote_addr;
2284 	la6 = (struct sockaddr_in6 *)&ep->com.local_addr;
2285 	ra6 = (struct sockaddr_in6 *)&ep->com.remote_addr;
2286 
2287 	pr_debug("ep %p atid %u status %u errno %d\n", ep, atid,
2288 		 status, status2errno(status));
2289 
2290 	if (cxgb_is_neg_adv(status)) {
2291 		pr_debug("Connection problems for atid %u status %u (%s)\n",
2292 			 atid, status, neg_adv_str(status));
2293 		ep->stats.connect_neg_adv++;
2294 		mutex_lock(&dev->rdev.stats.lock);
2295 		dev->rdev.stats.neg_adv++;
2296 		mutex_unlock(&dev->rdev.stats.lock);
2297 		return 0;
2298 	}
2299 
2300 	set_bit(ACT_OPEN_RPL, &ep->com.history);
2301 
2302 	/*
2303 	 * Log interesting failures.
2304 	 */
2305 	switch (status) {
2306 	case CPL_ERR_CONN_RESET:
2307 	case CPL_ERR_CONN_TIMEDOUT:
2308 		break;
2309 	case CPL_ERR_TCAM_FULL:
2310 		mutex_lock(&dev->rdev.stats.lock);
2311 		dev->rdev.stats.tcam_full++;
2312 		mutex_unlock(&dev->rdev.stats.lock);
2313 		if (ep->com.local_addr.ss_family == AF_INET &&
2314 		    dev->rdev.lldi.enable_fw_ofld_conn) {
2315 			ret = send_fw_act_open_req(ep, TID_TID_G(AOPEN_ATID_G(
2316 						   ntohl(rpl->atid_status))));
2317 			if (ret)
2318 				goto fail;
2319 			return 0;
2320 		}
2321 		break;
2322 	case CPL_ERR_CONN_EXIST:
2323 		if (ep->retry_count++ < ACT_OPEN_RETRY_COUNT) {
2324 			set_bit(ACT_RETRY_INUSE, &ep->com.history);
2325 			if (ep->com.remote_addr.ss_family == AF_INET6) {
2326 				struct sockaddr_in6 *sin6 =
2327 						(struct sockaddr_in6 *)
2328 						&ep->com.local_addr;
2329 				cxgb4_clip_release(
2330 						ep->com.dev->rdev.lldi.ports[0],
2331 						(const u32 *)
2332 						&sin6->sin6_addr.s6_addr, 1);
2333 			}
2334 			xa_erase_irq(&ep->com.dev->atids, atid);
2335 			cxgb4_free_atid(t, atid);
2336 			dst_release(ep->dst);
2337 			cxgb4_l2t_release(ep->l2t);
2338 			c4iw_reconnect(ep);
2339 			return 0;
2340 		}
2341 		break;
2342 	default:
2343 		if (ep->com.local_addr.ss_family == AF_INET) {
2344 			pr_info("Active open failure - atid %u status %u errno %d %pI4:%u->%pI4:%u\n",
2345 				atid, status, status2errno(status),
2346 				&la->sin_addr.s_addr, ntohs(la->sin_port),
2347 				&ra->sin_addr.s_addr, ntohs(ra->sin_port));
2348 		} else {
2349 			pr_info("Active open failure - atid %u status %u errno %d %pI6:%u->%pI6:%u\n",
2350 				atid, status, status2errno(status),
2351 				la6->sin6_addr.s6_addr, ntohs(la6->sin6_port),
2352 				ra6->sin6_addr.s6_addr, ntohs(ra6->sin6_port));
2353 		}
2354 		break;
2355 	}
2356 
2357 fail:
2358 	connect_reply_upcall(ep, status2errno(status));
2359 	state_set(&ep->com, DEAD);
2360 
2361 	if (ep->com.remote_addr.ss_family == AF_INET6) {
2362 		struct sockaddr_in6 *sin6 =
2363 			(struct sockaddr_in6 *)&ep->com.local_addr;
2364 		cxgb4_clip_release(ep->com.dev->rdev.lldi.ports[0],
2365 				   (const u32 *)&sin6->sin6_addr.s6_addr, 1);
2366 	}
2367 	if (status && act_open_has_tid(status))
2368 		cxgb4_remove_tid(ep->com.dev->rdev.lldi.tids, 0, GET_TID(rpl),
2369 				 ep->com.local_addr.ss_family);
2370 
2371 	xa_erase_irq(&ep->com.dev->atids, atid);
2372 	cxgb4_free_atid(t, atid);
2373 	dst_release(ep->dst);
2374 	cxgb4_l2t_release(ep->l2t);
2375 	c4iw_put_ep(&ep->com);
2376 
2377 	return 0;
2378 }
2379 
pass_open_rpl(struct c4iw_dev * dev,struct sk_buff * skb)2380 static int pass_open_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
2381 {
2382 	struct cpl_pass_open_rpl *rpl = cplhdr(skb);
2383 	unsigned int stid = GET_TID(rpl);
2384 	struct c4iw_listen_ep *ep = get_ep_from_stid(dev, stid);
2385 
2386 	if (!ep) {
2387 		pr_warn("%s stid %d lookup failure!\n", __func__, stid);
2388 		goto out;
2389 	}
2390 	pr_debug("ep %p status %d error %d\n", ep,
2391 		 rpl->status, status2errno(rpl->status));
2392 	c4iw_wake_up_noref(ep->com.wr_waitp, status2errno(rpl->status));
2393 	c4iw_put_ep(&ep->com);
2394 out:
2395 	return 0;
2396 }
2397 
close_listsrv_rpl(struct c4iw_dev * dev,struct sk_buff * skb)2398 static int close_listsrv_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
2399 {
2400 	struct cpl_close_listsvr_rpl *rpl = cplhdr(skb);
2401 	unsigned int stid = GET_TID(rpl);
2402 	struct c4iw_listen_ep *ep = get_ep_from_stid(dev, stid);
2403 
2404 	if (!ep) {
2405 		pr_warn("%s stid %d lookup failure!\n", __func__, stid);
2406 		goto out;
2407 	}
2408 	pr_debug("ep %p\n", ep);
2409 	c4iw_wake_up_noref(ep->com.wr_waitp, status2errno(rpl->status));
2410 	c4iw_put_ep(&ep->com);
2411 out:
2412 	return 0;
2413 }
2414 
accept_cr(struct c4iw_ep * ep,struct sk_buff * skb,struct cpl_pass_accept_req * req)2415 static int accept_cr(struct c4iw_ep *ep, struct sk_buff *skb,
2416 		     struct cpl_pass_accept_req *req)
2417 {
2418 	struct cpl_pass_accept_rpl *rpl;
2419 	unsigned int mtu_idx;
2420 	u64 opt0;
2421 	u32 opt2;
2422 	u32 wscale;
2423 	struct cpl_t5_pass_accept_rpl *rpl5 = NULL;
2424 	int win;
2425 	enum chip_type adapter_type = ep->com.dev->rdev.lldi.adapter_type;
2426 
2427 	pr_debug("ep %p tid %u\n", ep, ep->hwtid);
2428 	cxgb_best_mtu(ep->com.dev->rdev.lldi.mtus, ep->mtu, &mtu_idx,
2429 		      enable_tcp_timestamps && req->tcpopt.tstamp,
2430 		      (ep->com.remote_addr.ss_family == AF_INET) ? 0 : 1);
2431 	wscale = cxgb_compute_wscale(rcv_win);
2432 
2433 	/*
2434 	 * Specify the largest window that will fit in opt0. The
2435 	 * remainder will be specified in the rx_data_ack.
2436 	 */
2437 	win = ep->rcv_win >> 10;
2438 	if (win > RCV_BUFSIZ_M)
2439 		win = RCV_BUFSIZ_M;
2440 	opt0 = (nocong ? NO_CONG_F : 0) |
2441 	       KEEP_ALIVE_F |
2442 	       DELACK_F |
2443 	       WND_SCALE_V(wscale) |
2444 	       MSS_IDX_V(mtu_idx) |
2445 	       L2T_IDX_V(ep->l2t->idx) |
2446 	       TX_CHAN_V(ep->tx_chan) |
2447 	       SMAC_SEL_V(ep->smac_idx) |
2448 	       DSCP_V(ep->tos >> 2) |
2449 	       ULP_MODE_V(ULP_MODE_TCPDDP) |
2450 	       RCV_BUFSIZ_V(win);
2451 	opt2 = RX_CHANNEL_V(0) |
2452 	       RSS_QUEUE_VALID_F | RSS_QUEUE_V(ep->rss_qid);
2453 
2454 	if (enable_tcp_timestamps && req->tcpopt.tstamp)
2455 		opt2 |= TSTAMPS_EN_F;
2456 	if (enable_tcp_sack && req->tcpopt.sack)
2457 		opt2 |= SACK_EN_F;
2458 	if (wscale && enable_tcp_window_scaling)
2459 		opt2 |= WND_SCALE_EN_F;
2460 	if (enable_ecn) {
2461 		const struct tcphdr *tcph;
2462 		u32 hlen = ntohl(req->hdr_len);
2463 
2464 		if (CHELSIO_CHIP_VERSION(adapter_type) <= CHELSIO_T5)
2465 			tcph = (const void *)(req + 1) + ETH_HDR_LEN_G(hlen) +
2466 				IP_HDR_LEN_G(hlen);
2467 		else
2468 			tcph = (const void *)(req + 1) +
2469 				T6_ETH_HDR_LEN_G(hlen) + T6_IP_HDR_LEN_G(hlen);
2470 		if (tcph->ece && tcph->cwr)
2471 			opt2 |= CCTRL_ECN_V(1);
2472 	}
2473 
2474 	skb_get(skb);
2475 	rpl = cplhdr(skb);
2476 	if (!is_t4(adapter_type)) {
2477 		skb_trim(skb, roundup(sizeof(*rpl5), 16));
2478 		rpl5 = (void *)rpl;
2479 		INIT_TP_WR(rpl5, ep->hwtid);
2480 	} else {
2481 		skb_trim(skb, sizeof(*rpl));
2482 		INIT_TP_WR(rpl, ep->hwtid);
2483 	}
2484 	OPCODE_TID(rpl) = cpu_to_be32(MK_OPCODE_TID(CPL_PASS_ACCEPT_RPL,
2485 						    ep->hwtid));
2486 
2487 	if (CHELSIO_CHIP_VERSION(adapter_type) > CHELSIO_T4) {
2488 		u32 isn = (prandom_u32() & ~7UL) - 1;
2489 		opt2 |= T5_OPT_2_VALID_F;
2490 		opt2 |= CONG_CNTRL_V(CONG_ALG_TAHOE);
2491 		opt2 |= T5_ISS_F;
2492 		rpl5 = (void *)rpl;
2493 		memset(&rpl5->iss, 0, roundup(sizeof(*rpl5)-sizeof(*rpl), 16));
2494 		if (peer2peer)
2495 			isn += 4;
2496 		rpl5->iss = cpu_to_be32(isn);
2497 		pr_debug("iss %u\n", be32_to_cpu(rpl5->iss));
2498 	}
2499 
2500 	rpl->opt0 = cpu_to_be64(opt0);
2501 	rpl->opt2 = cpu_to_be32(opt2);
2502 	set_wr_txq(skb, CPL_PRIORITY_SETUP, ep->ctrlq_idx);
2503 	t4_set_arp_err_handler(skb, ep, pass_accept_rpl_arp_failure);
2504 
2505 	return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
2506 }
2507 
reject_cr(struct c4iw_dev * dev,u32 hwtid,struct sk_buff * skb)2508 static void reject_cr(struct c4iw_dev *dev, u32 hwtid, struct sk_buff *skb)
2509 {
2510 	pr_debug("c4iw_dev %p tid %u\n", dev, hwtid);
2511 	skb_trim(skb, sizeof(struct cpl_tid_release));
2512 	release_tid(&dev->rdev, hwtid, skb);
2513 	return;
2514 }
2515 
pass_accept_req(struct c4iw_dev * dev,struct sk_buff * skb)2516 static int pass_accept_req(struct c4iw_dev *dev, struct sk_buff *skb)
2517 {
2518 	struct c4iw_ep *child_ep = NULL, *parent_ep;
2519 	struct cpl_pass_accept_req *req = cplhdr(skb);
2520 	unsigned int stid = PASS_OPEN_TID_G(ntohl(req->tos_stid));
2521 	struct tid_info *t = dev->rdev.lldi.tids;
2522 	unsigned int hwtid = GET_TID(req);
2523 	struct dst_entry *dst;
2524 	__u8 local_ip[16], peer_ip[16];
2525 	__be16 local_port, peer_port;
2526 	struct sockaddr_in6 *sin6;
2527 	int err;
2528 	u16 peer_mss = ntohs(req->tcpopt.mss);
2529 	int iptype;
2530 	unsigned short hdrs;
2531 	u8 tos;
2532 
2533 	parent_ep = (struct c4iw_ep *)get_ep_from_stid(dev, stid);
2534 	if (!parent_ep) {
2535 		pr_err("%s connect request on invalid stid %d\n",
2536 		       __func__, stid);
2537 		goto reject;
2538 	}
2539 
2540 	if (state_read(&parent_ep->com) != LISTEN) {
2541 		pr_err("%s - listening ep not in LISTEN\n", __func__);
2542 		goto reject;
2543 	}
2544 
2545 	if (parent_ep->com.cm_id->tos_set)
2546 		tos = parent_ep->com.cm_id->tos;
2547 	else
2548 		tos = PASS_OPEN_TOS_G(ntohl(req->tos_stid));
2549 
2550 	cxgb_get_4tuple(req, parent_ep->com.dev->rdev.lldi.adapter_type,
2551 			&iptype, local_ip, peer_ip, &local_port, &peer_port);
2552 
2553 	/* Find output route */
2554 	if (iptype == 4)  {
2555 		pr_debug("parent ep %p hwtid %u laddr %pI4 raddr %pI4 lport %d rport %d peer_mss %d\n"
2556 			 , parent_ep, hwtid,
2557 			 local_ip, peer_ip, ntohs(local_port),
2558 			 ntohs(peer_port), peer_mss);
2559 		dst = cxgb_find_route(&dev->rdev.lldi, get_real_dev,
2560 				      *(__be32 *)local_ip, *(__be32 *)peer_ip,
2561 				      local_port, peer_port, tos);
2562 	} else {
2563 		pr_debug("parent ep %p hwtid %u laddr %pI6 raddr %pI6 lport %d rport %d peer_mss %d\n"
2564 			 , parent_ep, hwtid,
2565 			 local_ip, peer_ip, ntohs(local_port),
2566 			 ntohs(peer_port), peer_mss);
2567 		dst = cxgb_find_route6(&dev->rdev.lldi, get_real_dev,
2568 				local_ip, peer_ip, local_port, peer_port,
2569 				tos,
2570 				((struct sockaddr_in6 *)
2571 				 &parent_ep->com.local_addr)->sin6_scope_id);
2572 	}
2573 	if (!dst) {
2574 		pr_err("%s - failed to find dst entry!\n", __func__);
2575 		goto reject;
2576 	}
2577 
2578 	child_ep = alloc_ep(sizeof(*child_ep), GFP_KERNEL);
2579 	if (!child_ep) {
2580 		pr_err("%s - failed to allocate ep entry!\n", __func__);
2581 		dst_release(dst);
2582 		goto reject;
2583 	}
2584 
2585 	err = import_ep(child_ep, iptype, peer_ip, dst, dev, false,
2586 			parent_ep->com.dev->rdev.lldi.adapter_type, tos);
2587 	if (err) {
2588 		pr_err("%s - failed to allocate l2t entry!\n", __func__);
2589 		dst_release(dst);
2590 		kfree(child_ep);
2591 		goto reject;
2592 	}
2593 
2594 	hdrs = ((iptype == 4) ? sizeof(struct iphdr) : sizeof(struct ipv6hdr)) +
2595 	       sizeof(struct tcphdr) +
2596 	       ((enable_tcp_timestamps && req->tcpopt.tstamp) ? 12 : 0);
2597 	if (peer_mss && child_ep->mtu > (peer_mss + hdrs))
2598 		child_ep->mtu = peer_mss + hdrs;
2599 
2600 	skb_queue_head_init(&child_ep->com.ep_skb_list);
2601 	if (alloc_ep_skb_list(&child_ep->com.ep_skb_list, CN_MAX_CON_BUF))
2602 		goto fail;
2603 
2604 	state_set(&child_ep->com, CONNECTING);
2605 	child_ep->com.dev = dev;
2606 	child_ep->com.cm_id = NULL;
2607 
2608 	if (iptype == 4) {
2609 		struct sockaddr_in *sin = (struct sockaddr_in *)
2610 			&child_ep->com.local_addr;
2611 
2612 		sin->sin_family = AF_INET;
2613 		sin->sin_port = local_port;
2614 		sin->sin_addr.s_addr = *(__be32 *)local_ip;
2615 
2616 		sin = (struct sockaddr_in *)&child_ep->com.local_addr;
2617 		sin->sin_family = AF_INET;
2618 		sin->sin_port = ((struct sockaddr_in *)
2619 				 &parent_ep->com.local_addr)->sin_port;
2620 		sin->sin_addr.s_addr = *(__be32 *)local_ip;
2621 
2622 		sin = (struct sockaddr_in *)&child_ep->com.remote_addr;
2623 		sin->sin_family = AF_INET;
2624 		sin->sin_port = peer_port;
2625 		sin->sin_addr.s_addr = *(__be32 *)peer_ip;
2626 	} else {
2627 		sin6 = (struct sockaddr_in6 *)&child_ep->com.local_addr;
2628 		sin6->sin6_family = PF_INET6;
2629 		sin6->sin6_port = local_port;
2630 		memcpy(sin6->sin6_addr.s6_addr, local_ip, 16);
2631 
2632 		sin6 = (struct sockaddr_in6 *)&child_ep->com.local_addr;
2633 		sin6->sin6_family = PF_INET6;
2634 		sin6->sin6_port = ((struct sockaddr_in6 *)
2635 				   &parent_ep->com.local_addr)->sin6_port;
2636 		memcpy(sin6->sin6_addr.s6_addr, local_ip, 16);
2637 
2638 		sin6 = (struct sockaddr_in6 *)&child_ep->com.remote_addr;
2639 		sin6->sin6_family = PF_INET6;
2640 		sin6->sin6_port = peer_port;
2641 		memcpy(sin6->sin6_addr.s6_addr, peer_ip, 16);
2642 	}
2643 
2644 	c4iw_get_ep(&parent_ep->com);
2645 	child_ep->parent_ep = parent_ep;
2646 	child_ep->tos = tos;
2647 	child_ep->dst = dst;
2648 	child_ep->hwtid = hwtid;
2649 
2650 	pr_debug("tx_chan %u smac_idx %u rss_qid %u\n",
2651 		 child_ep->tx_chan, child_ep->smac_idx, child_ep->rss_qid);
2652 
2653 	timer_setup(&child_ep->timer, ep_timeout, 0);
2654 	cxgb4_insert_tid(t, child_ep, hwtid,
2655 			 child_ep->com.local_addr.ss_family);
2656 	insert_ep_tid(child_ep);
2657 	if (accept_cr(child_ep, skb, req)) {
2658 		c4iw_put_ep(&parent_ep->com);
2659 		release_ep_resources(child_ep);
2660 	} else {
2661 		set_bit(PASS_ACCEPT_REQ, &child_ep->com.history);
2662 	}
2663 	if (iptype == 6) {
2664 		sin6 = (struct sockaddr_in6 *)&child_ep->com.local_addr;
2665 		cxgb4_clip_get(child_ep->com.dev->rdev.lldi.ports[0],
2666 			       (const u32 *)&sin6->sin6_addr.s6_addr, 1);
2667 	}
2668 	goto out;
2669 fail:
2670 	c4iw_put_ep(&child_ep->com);
2671 reject:
2672 	reject_cr(dev, hwtid, skb);
2673 out:
2674 	if (parent_ep)
2675 		c4iw_put_ep(&parent_ep->com);
2676 	return 0;
2677 }
2678 
pass_establish(struct c4iw_dev * dev,struct sk_buff * skb)2679 static int pass_establish(struct c4iw_dev *dev, struct sk_buff *skb)
2680 {
2681 	struct c4iw_ep *ep;
2682 	struct cpl_pass_establish *req = cplhdr(skb);
2683 	unsigned int tid = GET_TID(req);
2684 	int ret;
2685 	u16 tcp_opt = ntohs(req->tcp_opt);
2686 
2687 	ep = get_ep_from_tid(dev, tid);
2688 	if (!ep)
2689 		return 0;
2690 
2691 	pr_debug("ep %p tid %u\n", ep, ep->hwtid);
2692 	ep->snd_seq = be32_to_cpu(req->snd_isn);
2693 	ep->rcv_seq = be32_to_cpu(req->rcv_isn);
2694 	ep->snd_wscale = TCPOPT_SND_WSCALE_G(tcp_opt);
2695 
2696 	pr_debug("ep %p hwtid %u tcp_opt 0x%02x\n", ep, tid, tcp_opt);
2697 
2698 	set_emss(ep, tcp_opt);
2699 
2700 	dst_confirm(ep->dst);
2701 	mutex_lock(&ep->com.mutex);
2702 	ep->com.state = MPA_REQ_WAIT;
2703 	start_ep_timer(ep);
2704 	set_bit(PASS_ESTAB, &ep->com.history);
2705 	ret = send_flowc(ep);
2706 	mutex_unlock(&ep->com.mutex);
2707 	if (ret)
2708 		c4iw_ep_disconnect(ep, 1, GFP_KERNEL);
2709 	c4iw_put_ep(&ep->com);
2710 
2711 	return 0;
2712 }
2713 
peer_close(struct c4iw_dev * dev,struct sk_buff * skb)2714 static int peer_close(struct c4iw_dev *dev, struct sk_buff *skb)
2715 {
2716 	struct cpl_peer_close *hdr = cplhdr(skb);
2717 	struct c4iw_ep *ep;
2718 	struct c4iw_qp_attributes attrs;
2719 	int disconnect = 1;
2720 	int release = 0;
2721 	unsigned int tid = GET_TID(hdr);
2722 	int ret;
2723 
2724 	ep = get_ep_from_tid(dev, tid);
2725 	if (!ep)
2726 		return 0;
2727 
2728 	pr_debug("ep %p tid %u\n", ep, ep->hwtid);
2729 	dst_confirm(ep->dst);
2730 
2731 	set_bit(PEER_CLOSE, &ep->com.history);
2732 	mutex_lock(&ep->com.mutex);
2733 	switch (ep->com.state) {
2734 	case MPA_REQ_WAIT:
2735 		__state_set(&ep->com, CLOSING);
2736 		break;
2737 	case MPA_REQ_SENT:
2738 		__state_set(&ep->com, CLOSING);
2739 		connect_reply_upcall(ep, -ECONNRESET);
2740 		break;
2741 	case MPA_REQ_RCVD:
2742 
2743 		/*
2744 		 * We're gonna mark this puppy DEAD, but keep
2745 		 * the reference on it until the ULP accepts or
2746 		 * rejects the CR. Also wake up anyone waiting
2747 		 * in rdma connection migration (see c4iw_accept_cr()).
2748 		 */
2749 		__state_set(&ep->com, CLOSING);
2750 		pr_debug("waking up ep %p tid %u\n", ep, ep->hwtid);
2751 		c4iw_wake_up_noref(ep->com.wr_waitp, -ECONNRESET);
2752 		break;
2753 	case MPA_REP_SENT:
2754 		__state_set(&ep->com, CLOSING);
2755 		pr_debug("waking up ep %p tid %u\n", ep, ep->hwtid);
2756 		c4iw_wake_up_noref(ep->com.wr_waitp, -ECONNRESET);
2757 		break;
2758 	case FPDU_MODE:
2759 		start_ep_timer(ep);
2760 		__state_set(&ep->com, CLOSING);
2761 		attrs.next_state = C4IW_QP_STATE_CLOSING;
2762 		ret = c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
2763 				       C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
2764 		if (ret != -ECONNRESET) {
2765 			peer_close_upcall(ep);
2766 			disconnect = 1;
2767 		}
2768 		break;
2769 	case ABORTING:
2770 		disconnect = 0;
2771 		break;
2772 	case CLOSING:
2773 		__state_set(&ep->com, MORIBUND);
2774 		disconnect = 0;
2775 		break;
2776 	case MORIBUND:
2777 		(void)stop_ep_timer(ep);
2778 		if (ep->com.cm_id && ep->com.qp) {
2779 			attrs.next_state = C4IW_QP_STATE_IDLE;
2780 			c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
2781 				       C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
2782 		}
2783 		close_complete_upcall(ep, 0);
2784 		__state_set(&ep->com, DEAD);
2785 		release = 1;
2786 		disconnect = 0;
2787 		break;
2788 	case DEAD:
2789 		disconnect = 0;
2790 		break;
2791 	default:
2792 		WARN_ONCE(1, "Bad endpoint state %u\n", ep->com.state);
2793 	}
2794 	mutex_unlock(&ep->com.mutex);
2795 	if (disconnect)
2796 		c4iw_ep_disconnect(ep, 0, GFP_KERNEL);
2797 	if (release)
2798 		release_ep_resources(ep);
2799 	c4iw_put_ep(&ep->com);
2800 	return 0;
2801 }
2802 
finish_peer_abort(struct c4iw_dev * dev,struct c4iw_ep * ep)2803 static void finish_peer_abort(struct c4iw_dev *dev, struct c4iw_ep *ep)
2804 {
2805 	complete_cached_srq_buffers(ep, ep->srqe_idx);
2806 	if (ep->com.cm_id && ep->com.qp) {
2807 		struct c4iw_qp_attributes attrs;
2808 
2809 		attrs.next_state = C4IW_QP_STATE_ERROR;
2810 		c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
2811 			       C4IW_QP_ATTR_NEXT_STATE,	&attrs, 1);
2812 	}
2813 	peer_abort_upcall(ep);
2814 	release_ep_resources(ep);
2815 	c4iw_put_ep(&ep->com);
2816 }
2817 
peer_abort(struct c4iw_dev * dev,struct sk_buff * skb)2818 static int peer_abort(struct c4iw_dev *dev, struct sk_buff *skb)
2819 {
2820 	struct cpl_abort_req_rss6 *req = cplhdr(skb);
2821 	struct c4iw_ep *ep;
2822 	struct sk_buff *rpl_skb;
2823 	struct c4iw_qp_attributes attrs;
2824 	int ret;
2825 	int release = 0;
2826 	unsigned int tid = GET_TID(req);
2827 	u8 status;
2828 	u32 srqidx;
2829 
2830 	u32 len = roundup(sizeof(struct cpl_abort_rpl), 16);
2831 
2832 	ep = get_ep_from_tid(dev, tid);
2833 	if (!ep)
2834 		return 0;
2835 
2836 	status = ABORT_RSS_STATUS_G(be32_to_cpu(req->srqidx_status));
2837 
2838 	if (cxgb_is_neg_adv(status)) {
2839 		pr_debug("Negative advice on abort- tid %u status %d (%s)\n",
2840 			 ep->hwtid, status, neg_adv_str(status));
2841 		ep->stats.abort_neg_adv++;
2842 		mutex_lock(&dev->rdev.stats.lock);
2843 		dev->rdev.stats.neg_adv++;
2844 		mutex_unlock(&dev->rdev.stats.lock);
2845 		goto deref_ep;
2846 	}
2847 
2848 	pr_debug("ep %p tid %u state %u\n", ep, ep->hwtid,
2849 		 ep->com.state);
2850 	set_bit(PEER_ABORT, &ep->com.history);
2851 
2852 	/*
2853 	 * Wake up any threads in rdma_init() or rdma_fini().
2854 	 * However, this is not needed if com state is just
2855 	 * MPA_REQ_SENT
2856 	 */
2857 	if (ep->com.state != MPA_REQ_SENT)
2858 		c4iw_wake_up_noref(ep->com.wr_waitp, -ECONNRESET);
2859 
2860 	mutex_lock(&ep->com.mutex);
2861 	switch (ep->com.state) {
2862 	case CONNECTING:
2863 		c4iw_put_ep(&ep->parent_ep->com);
2864 		break;
2865 	case MPA_REQ_WAIT:
2866 		(void)stop_ep_timer(ep);
2867 		break;
2868 	case MPA_REQ_SENT:
2869 		(void)stop_ep_timer(ep);
2870 		if (status != CPL_ERR_CONN_RESET || mpa_rev == 1 ||
2871 		    (mpa_rev == 2 && ep->tried_with_mpa_v1))
2872 			connect_reply_upcall(ep, -ECONNRESET);
2873 		else {
2874 			/*
2875 			 * we just don't send notification upwards because we
2876 			 * want to retry with mpa_v1 without upper layers even
2877 			 * knowing it.
2878 			 *
2879 			 * do some housekeeping so as to re-initiate the
2880 			 * connection
2881 			 */
2882 			pr_info("%s: mpa_rev=%d. Retrying with mpav1\n",
2883 				__func__, mpa_rev);
2884 			ep->retry_with_mpa_v1 = 1;
2885 		}
2886 		break;
2887 	case MPA_REP_SENT:
2888 		break;
2889 	case MPA_REQ_RCVD:
2890 		break;
2891 	case MORIBUND:
2892 	case CLOSING:
2893 		stop_ep_timer(ep);
2894 		fallthrough;
2895 	case FPDU_MODE:
2896 		if (ep->com.qp && ep->com.qp->srq) {
2897 			srqidx = ABORT_RSS_SRQIDX_G(
2898 					be32_to_cpu(req->srqidx_status));
2899 			if (srqidx) {
2900 				complete_cached_srq_buffers(ep, srqidx);
2901 			} else {
2902 				/* Hold ep ref until finish_peer_abort() */
2903 				c4iw_get_ep(&ep->com);
2904 				__state_set(&ep->com, ABORTING);
2905 				set_bit(PEER_ABORT_IN_PROGRESS, &ep->com.flags);
2906 				read_tcb(ep);
2907 				break;
2908 
2909 			}
2910 		}
2911 
2912 		if (ep->com.cm_id && ep->com.qp) {
2913 			attrs.next_state = C4IW_QP_STATE_ERROR;
2914 			ret = c4iw_modify_qp(ep->com.qp->rhp,
2915 				     ep->com.qp, C4IW_QP_ATTR_NEXT_STATE,
2916 				     &attrs, 1);
2917 			if (ret)
2918 				pr_err("%s - qp <- error failed!\n", __func__);
2919 		}
2920 		peer_abort_upcall(ep);
2921 		break;
2922 	case ABORTING:
2923 		break;
2924 	case DEAD:
2925 		pr_warn("%s PEER_ABORT IN DEAD STATE!!!!\n", __func__);
2926 		mutex_unlock(&ep->com.mutex);
2927 		goto deref_ep;
2928 	default:
2929 		WARN_ONCE(1, "Bad endpoint state %u\n", ep->com.state);
2930 		break;
2931 	}
2932 	dst_confirm(ep->dst);
2933 	if (ep->com.state != ABORTING) {
2934 		__state_set(&ep->com, DEAD);
2935 		/* we don't release if we want to retry with mpa_v1 */
2936 		if (!ep->retry_with_mpa_v1)
2937 			release = 1;
2938 	}
2939 	mutex_unlock(&ep->com.mutex);
2940 
2941 	rpl_skb = skb_dequeue(&ep->com.ep_skb_list);
2942 	if (WARN_ON(!rpl_skb)) {
2943 		release = 1;
2944 		goto out;
2945 	}
2946 
2947 	cxgb_mk_abort_rpl(rpl_skb, len, ep->hwtid, ep->txq_idx);
2948 
2949 	c4iw_ofld_send(&ep->com.dev->rdev, rpl_skb);
2950 out:
2951 	if (release)
2952 		release_ep_resources(ep);
2953 	else if (ep->retry_with_mpa_v1) {
2954 		if (ep->com.remote_addr.ss_family == AF_INET6) {
2955 			struct sockaddr_in6 *sin6 =
2956 					(struct sockaddr_in6 *)
2957 					&ep->com.local_addr;
2958 			cxgb4_clip_release(
2959 					ep->com.dev->rdev.lldi.ports[0],
2960 					(const u32 *)&sin6->sin6_addr.s6_addr,
2961 					1);
2962 		}
2963 		xa_erase_irq(&ep->com.dev->hwtids, ep->hwtid);
2964 		cxgb4_remove_tid(ep->com.dev->rdev.lldi.tids, 0, ep->hwtid,
2965 				 ep->com.local_addr.ss_family);
2966 		dst_release(ep->dst);
2967 		cxgb4_l2t_release(ep->l2t);
2968 		c4iw_reconnect(ep);
2969 	}
2970 
2971 deref_ep:
2972 	c4iw_put_ep(&ep->com);
2973 	/* Dereferencing ep, referenced in peer_abort_intr() */
2974 	c4iw_put_ep(&ep->com);
2975 	return 0;
2976 }
2977 
close_con_rpl(struct c4iw_dev * dev,struct sk_buff * skb)2978 static int close_con_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
2979 {
2980 	struct c4iw_ep *ep;
2981 	struct c4iw_qp_attributes attrs;
2982 	struct cpl_close_con_rpl *rpl = cplhdr(skb);
2983 	int release = 0;
2984 	unsigned int tid = GET_TID(rpl);
2985 
2986 	ep = get_ep_from_tid(dev, tid);
2987 	if (!ep)
2988 		return 0;
2989 
2990 	pr_debug("ep %p tid %u\n", ep, ep->hwtid);
2991 
2992 	/* The cm_id may be null if we failed to connect */
2993 	mutex_lock(&ep->com.mutex);
2994 	set_bit(CLOSE_CON_RPL, &ep->com.history);
2995 	switch (ep->com.state) {
2996 	case CLOSING:
2997 		__state_set(&ep->com, MORIBUND);
2998 		break;
2999 	case MORIBUND:
3000 		(void)stop_ep_timer(ep);
3001 		if ((ep->com.cm_id) && (ep->com.qp)) {
3002 			attrs.next_state = C4IW_QP_STATE_IDLE;
3003 			c4iw_modify_qp(ep->com.qp->rhp,
3004 					     ep->com.qp,
3005 					     C4IW_QP_ATTR_NEXT_STATE,
3006 					     &attrs, 1);
3007 		}
3008 		close_complete_upcall(ep, 0);
3009 		__state_set(&ep->com, DEAD);
3010 		release = 1;
3011 		break;
3012 	case ABORTING:
3013 	case DEAD:
3014 		break;
3015 	default:
3016 		WARN_ONCE(1, "Bad endpoint state %u\n", ep->com.state);
3017 		break;
3018 	}
3019 	mutex_unlock(&ep->com.mutex);
3020 	if (release)
3021 		release_ep_resources(ep);
3022 	c4iw_put_ep(&ep->com);
3023 	return 0;
3024 }
3025 
terminate(struct c4iw_dev * dev,struct sk_buff * skb)3026 static int terminate(struct c4iw_dev *dev, struct sk_buff *skb)
3027 {
3028 	struct cpl_rdma_terminate *rpl = cplhdr(skb);
3029 	unsigned int tid = GET_TID(rpl);
3030 	struct c4iw_ep *ep;
3031 	struct c4iw_qp_attributes attrs;
3032 
3033 	ep = get_ep_from_tid(dev, tid);
3034 
3035 	if (ep) {
3036 		if (ep->com.qp) {
3037 			pr_warn("TERM received tid %u qpid %u\n", tid,
3038 				ep->com.qp->wq.sq.qid);
3039 			attrs.next_state = C4IW_QP_STATE_TERMINATE;
3040 			c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
3041 				       C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
3042 		}
3043 
3044 		/* As per draft-hilland-iwarp-verbs-v1.0, sec 6.2.3,
3045 		 * when entering the TERM state the RNIC MUST initiate a CLOSE.
3046 		 */
3047 		c4iw_ep_disconnect(ep, 1, GFP_KERNEL);
3048 		c4iw_put_ep(&ep->com);
3049 	} else
3050 		pr_warn("TERM received tid %u no ep/qp\n", tid);
3051 
3052 	return 0;
3053 }
3054 
3055 /*
3056  * Upcall from the adapter indicating data has been transmitted.
3057  * For us its just the single MPA request or reply.  We can now free
3058  * the skb holding the mpa message.
3059  */
fw4_ack(struct c4iw_dev * dev,struct sk_buff * skb)3060 static int fw4_ack(struct c4iw_dev *dev, struct sk_buff *skb)
3061 {
3062 	struct c4iw_ep *ep;
3063 	struct cpl_fw4_ack *hdr = cplhdr(skb);
3064 	u8 credits = hdr->credits;
3065 	unsigned int tid = GET_TID(hdr);
3066 
3067 
3068 	ep = get_ep_from_tid(dev, tid);
3069 	if (!ep)
3070 		return 0;
3071 	pr_debug("ep %p tid %u credits %u\n",
3072 		 ep, ep->hwtid, credits);
3073 	if (credits == 0) {
3074 		pr_debug("0 credit ack ep %p tid %u state %u\n",
3075 			 ep, ep->hwtid, state_read(&ep->com));
3076 		goto out;
3077 	}
3078 
3079 	dst_confirm(ep->dst);
3080 	if (ep->mpa_skb) {
3081 		pr_debug("last streaming msg ack ep %p tid %u state %u initiator %u freeing skb\n",
3082 			 ep, ep->hwtid, state_read(&ep->com),
3083 			 ep->mpa_attr.initiator ? 1 : 0);
3084 		mutex_lock(&ep->com.mutex);
3085 		kfree_skb(ep->mpa_skb);
3086 		ep->mpa_skb = NULL;
3087 		if (test_bit(STOP_MPA_TIMER, &ep->com.flags))
3088 			stop_ep_timer(ep);
3089 		mutex_unlock(&ep->com.mutex);
3090 	}
3091 out:
3092 	c4iw_put_ep(&ep->com);
3093 	return 0;
3094 }
3095 
c4iw_reject_cr(struct iw_cm_id * cm_id,const void * pdata,u8 pdata_len)3096 int c4iw_reject_cr(struct iw_cm_id *cm_id, const void *pdata, u8 pdata_len)
3097 {
3098 	int abort;
3099 	struct c4iw_ep *ep = to_ep(cm_id);
3100 
3101 	pr_debug("ep %p tid %u\n", ep, ep->hwtid);
3102 
3103 	mutex_lock(&ep->com.mutex);
3104 	if (ep->com.state != MPA_REQ_RCVD) {
3105 		mutex_unlock(&ep->com.mutex);
3106 		c4iw_put_ep(&ep->com);
3107 		return -ECONNRESET;
3108 	}
3109 	set_bit(ULP_REJECT, &ep->com.history);
3110 	if (mpa_rev == 0)
3111 		abort = 1;
3112 	else
3113 		abort = send_mpa_reject(ep, pdata, pdata_len);
3114 	mutex_unlock(&ep->com.mutex);
3115 
3116 	stop_ep_timer(ep);
3117 	c4iw_ep_disconnect(ep, abort != 0, GFP_KERNEL);
3118 	c4iw_put_ep(&ep->com);
3119 	return 0;
3120 }
3121 
c4iw_accept_cr(struct iw_cm_id * cm_id,struct iw_cm_conn_param * conn_param)3122 int c4iw_accept_cr(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param)
3123 {
3124 	int err;
3125 	struct c4iw_qp_attributes attrs;
3126 	enum c4iw_qp_attr_mask mask;
3127 	struct c4iw_ep *ep = to_ep(cm_id);
3128 	struct c4iw_dev *h = to_c4iw_dev(cm_id->device);
3129 	struct c4iw_qp *qp = get_qhp(h, conn_param->qpn);
3130 	int abort = 0;
3131 
3132 	pr_debug("ep %p tid %u\n", ep, ep->hwtid);
3133 
3134 	mutex_lock(&ep->com.mutex);
3135 	if (ep->com.state != MPA_REQ_RCVD) {
3136 		err = -ECONNRESET;
3137 		goto err_out;
3138 	}
3139 
3140 	if (!qp) {
3141 		err = -EINVAL;
3142 		goto err_out;
3143 	}
3144 
3145 	set_bit(ULP_ACCEPT, &ep->com.history);
3146 	if ((conn_param->ord > cur_max_read_depth(ep->com.dev)) ||
3147 	    (conn_param->ird > cur_max_read_depth(ep->com.dev))) {
3148 		err = -EINVAL;
3149 		goto err_abort;
3150 	}
3151 
3152 	if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) {
3153 		if (conn_param->ord > ep->ird) {
3154 			if (RELAXED_IRD_NEGOTIATION) {
3155 				conn_param->ord = ep->ird;
3156 			} else {
3157 				ep->ird = conn_param->ird;
3158 				ep->ord = conn_param->ord;
3159 				send_mpa_reject(ep, conn_param->private_data,
3160 						conn_param->private_data_len);
3161 				err = -ENOMEM;
3162 				goto err_abort;
3163 			}
3164 		}
3165 		if (conn_param->ird < ep->ord) {
3166 			if (RELAXED_IRD_NEGOTIATION &&
3167 			    ep->ord <= h->rdev.lldi.max_ordird_qp) {
3168 				conn_param->ird = ep->ord;
3169 			} else {
3170 				err = -ENOMEM;
3171 				goto err_abort;
3172 			}
3173 		}
3174 	}
3175 	ep->ird = conn_param->ird;
3176 	ep->ord = conn_param->ord;
3177 
3178 	if (ep->mpa_attr.version == 1) {
3179 		if (peer2peer && ep->ird == 0)
3180 			ep->ird = 1;
3181 	} else {
3182 		if (peer2peer &&
3183 		    (ep->mpa_attr.p2p_type != FW_RI_INIT_P2PTYPE_DISABLED) &&
3184 		    (p2p_type == FW_RI_INIT_P2PTYPE_READ_REQ) && ep->ird == 0)
3185 			ep->ird = 1;
3186 	}
3187 
3188 	pr_debug("ird %d ord %d\n", ep->ird, ep->ord);
3189 
3190 	ep->com.cm_id = cm_id;
3191 	ref_cm_id(&ep->com);
3192 	ep->com.qp = qp;
3193 	ref_qp(ep);
3194 
3195 	/* bind QP to EP and move to RTS */
3196 	attrs.mpa_attr = ep->mpa_attr;
3197 	attrs.max_ird = ep->ird;
3198 	attrs.max_ord = ep->ord;
3199 	attrs.llp_stream_handle = ep;
3200 	attrs.next_state = C4IW_QP_STATE_RTS;
3201 
3202 	/* bind QP and TID with INIT_WR */
3203 	mask = C4IW_QP_ATTR_NEXT_STATE |
3204 			     C4IW_QP_ATTR_LLP_STREAM_HANDLE |
3205 			     C4IW_QP_ATTR_MPA_ATTR |
3206 			     C4IW_QP_ATTR_MAX_IRD |
3207 			     C4IW_QP_ATTR_MAX_ORD;
3208 
3209 	err = c4iw_modify_qp(ep->com.qp->rhp,
3210 			     ep->com.qp, mask, &attrs, 1);
3211 	if (err)
3212 		goto err_deref_cm_id;
3213 
3214 	set_bit(STOP_MPA_TIMER, &ep->com.flags);
3215 	err = send_mpa_reply(ep, conn_param->private_data,
3216 			     conn_param->private_data_len);
3217 	if (err)
3218 		goto err_deref_cm_id;
3219 
3220 	__state_set(&ep->com, FPDU_MODE);
3221 	established_upcall(ep);
3222 	mutex_unlock(&ep->com.mutex);
3223 	c4iw_put_ep(&ep->com);
3224 	return 0;
3225 err_deref_cm_id:
3226 	deref_cm_id(&ep->com);
3227 err_abort:
3228 	abort = 1;
3229 err_out:
3230 	mutex_unlock(&ep->com.mutex);
3231 	if (abort)
3232 		c4iw_ep_disconnect(ep, 1, GFP_KERNEL);
3233 	c4iw_put_ep(&ep->com);
3234 	return err;
3235 }
3236 
pick_local_ipaddrs(struct c4iw_dev * dev,struct iw_cm_id * cm_id)3237 static int pick_local_ipaddrs(struct c4iw_dev *dev, struct iw_cm_id *cm_id)
3238 {
3239 	struct in_device *ind;
3240 	int found = 0;
3241 	struct sockaddr_in *laddr = (struct sockaddr_in *)&cm_id->m_local_addr;
3242 	struct sockaddr_in *raddr = (struct sockaddr_in *)&cm_id->m_remote_addr;
3243 	const struct in_ifaddr *ifa;
3244 
3245 	ind = in_dev_get(dev->rdev.lldi.ports[0]);
3246 	if (!ind)
3247 		return -EADDRNOTAVAIL;
3248 	rcu_read_lock();
3249 	in_dev_for_each_ifa_rcu(ifa, ind) {
3250 		if (ifa->ifa_flags & IFA_F_SECONDARY)
3251 			continue;
3252 		laddr->sin_addr.s_addr = ifa->ifa_address;
3253 		raddr->sin_addr.s_addr = ifa->ifa_address;
3254 		found = 1;
3255 		break;
3256 	}
3257 	rcu_read_unlock();
3258 
3259 	in_dev_put(ind);
3260 	return found ? 0 : -EADDRNOTAVAIL;
3261 }
3262 
get_lladdr(struct net_device * dev,struct in6_addr * addr,unsigned char banned_flags)3263 static int get_lladdr(struct net_device *dev, struct in6_addr *addr,
3264 		      unsigned char banned_flags)
3265 {
3266 	struct inet6_dev *idev;
3267 	int err = -EADDRNOTAVAIL;
3268 
3269 	rcu_read_lock();
3270 	idev = __in6_dev_get(dev);
3271 	if (idev != NULL) {
3272 		struct inet6_ifaddr *ifp;
3273 
3274 		read_lock_bh(&idev->lock);
3275 		list_for_each_entry(ifp, &idev->addr_list, if_list) {
3276 			if (ifp->scope == IFA_LINK &&
3277 			    !(ifp->flags & banned_flags)) {
3278 				memcpy(addr, &ifp->addr, 16);
3279 				err = 0;
3280 				break;
3281 			}
3282 		}
3283 		read_unlock_bh(&idev->lock);
3284 	}
3285 	rcu_read_unlock();
3286 	return err;
3287 }
3288 
pick_local_ip6addrs(struct c4iw_dev * dev,struct iw_cm_id * cm_id)3289 static int pick_local_ip6addrs(struct c4iw_dev *dev, struct iw_cm_id *cm_id)
3290 {
3291 	struct in6_addr addr;
3292 	struct sockaddr_in6 *la6 = (struct sockaddr_in6 *)&cm_id->m_local_addr;
3293 	struct sockaddr_in6 *ra6 = (struct sockaddr_in6 *)&cm_id->m_remote_addr;
3294 
3295 	if (!get_lladdr(dev->rdev.lldi.ports[0], &addr, IFA_F_TENTATIVE)) {
3296 		memcpy(la6->sin6_addr.s6_addr, &addr, 16);
3297 		memcpy(ra6->sin6_addr.s6_addr, &addr, 16);
3298 		return 0;
3299 	}
3300 	return -EADDRNOTAVAIL;
3301 }
3302 
c4iw_connect(struct iw_cm_id * cm_id,struct iw_cm_conn_param * conn_param)3303 int c4iw_connect(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param)
3304 {
3305 	struct c4iw_dev *dev = to_c4iw_dev(cm_id->device);
3306 	struct c4iw_ep *ep;
3307 	int err = 0;
3308 	struct sockaddr_in *laddr;
3309 	struct sockaddr_in *raddr;
3310 	struct sockaddr_in6 *laddr6;
3311 	struct sockaddr_in6 *raddr6;
3312 	__u8 *ra;
3313 	int iptype;
3314 
3315 	if ((conn_param->ord > cur_max_read_depth(dev)) ||
3316 	    (conn_param->ird > cur_max_read_depth(dev))) {
3317 		err = -EINVAL;
3318 		goto out;
3319 	}
3320 	ep = alloc_ep(sizeof(*ep), GFP_KERNEL);
3321 	if (!ep) {
3322 		pr_err("%s - cannot alloc ep\n", __func__);
3323 		err = -ENOMEM;
3324 		goto out;
3325 	}
3326 
3327 	skb_queue_head_init(&ep->com.ep_skb_list);
3328 	if (alloc_ep_skb_list(&ep->com.ep_skb_list, CN_MAX_CON_BUF)) {
3329 		err = -ENOMEM;
3330 		goto fail1;
3331 	}
3332 
3333 	timer_setup(&ep->timer, ep_timeout, 0);
3334 	ep->plen = conn_param->private_data_len;
3335 	if (ep->plen)
3336 		memcpy(ep->mpa_pkt + sizeof(struct mpa_message),
3337 		       conn_param->private_data, ep->plen);
3338 	ep->ird = conn_param->ird;
3339 	ep->ord = conn_param->ord;
3340 
3341 	if (peer2peer && ep->ord == 0)
3342 		ep->ord = 1;
3343 
3344 	ep->com.cm_id = cm_id;
3345 	ref_cm_id(&ep->com);
3346 	cm_id->provider_data = ep;
3347 	ep->com.dev = dev;
3348 	ep->com.qp = get_qhp(dev, conn_param->qpn);
3349 	if (!ep->com.qp) {
3350 		pr_warn("%s qpn 0x%x not found!\n", __func__, conn_param->qpn);
3351 		err = -EINVAL;
3352 		goto fail2;
3353 	}
3354 	ref_qp(ep);
3355 	pr_debug("qpn 0x%x qp %p cm_id %p\n", conn_param->qpn,
3356 		 ep->com.qp, cm_id);
3357 
3358 	/*
3359 	 * Allocate an active TID to initiate a TCP connection.
3360 	 */
3361 	ep->atid = cxgb4_alloc_atid(dev->rdev.lldi.tids, ep);
3362 	if (ep->atid == -1) {
3363 		pr_err("%s - cannot alloc atid\n", __func__);
3364 		err = -ENOMEM;
3365 		goto fail2;
3366 	}
3367 	err = xa_insert_irq(&dev->atids, ep->atid, ep, GFP_KERNEL);
3368 	if (err)
3369 		goto fail5;
3370 
3371 	memcpy(&ep->com.local_addr, &cm_id->m_local_addr,
3372 	       sizeof(ep->com.local_addr));
3373 	memcpy(&ep->com.remote_addr, &cm_id->m_remote_addr,
3374 	       sizeof(ep->com.remote_addr));
3375 
3376 	laddr = (struct sockaddr_in *)&ep->com.local_addr;
3377 	raddr = (struct sockaddr_in *)&ep->com.remote_addr;
3378 	laddr6 = (struct sockaddr_in6 *)&ep->com.local_addr;
3379 	raddr6 = (struct sockaddr_in6 *) &ep->com.remote_addr;
3380 
3381 	if (cm_id->m_remote_addr.ss_family == AF_INET) {
3382 		iptype = 4;
3383 		ra = (__u8 *)&raddr->sin_addr;
3384 
3385 		/*
3386 		 * Handle loopback requests to INADDR_ANY.
3387 		 */
3388 		if (raddr->sin_addr.s_addr == htonl(INADDR_ANY)) {
3389 			err = pick_local_ipaddrs(dev, cm_id);
3390 			if (err)
3391 				goto fail3;
3392 		}
3393 
3394 		/* find a route */
3395 		pr_debug("saddr %pI4 sport 0x%x raddr %pI4 rport 0x%x\n",
3396 			 &laddr->sin_addr, ntohs(laddr->sin_port),
3397 			 ra, ntohs(raddr->sin_port));
3398 		ep->dst = cxgb_find_route(&dev->rdev.lldi, get_real_dev,
3399 					  laddr->sin_addr.s_addr,
3400 					  raddr->sin_addr.s_addr,
3401 					  laddr->sin_port,
3402 					  raddr->sin_port, cm_id->tos);
3403 	} else {
3404 		iptype = 6;
3405 		ra = (__u8 *)&raddr6->sin6_addr;
3406 
3407 		/*
3408 		 * Handle loopback requests to INADDR_ANY.
3409 		 */
3410 		if (ipv6_addr_type(&raddr6->sin6_addr) == IPV6_ADDR_ANY) {
3411 			err = pick_local_ip6addrs(dev, cm_id);
3412 			if (err)
3413 				goto fail3;
3414 		}
3415 
3416 		/* find a route */
3417 		pr_debug("saddr %pI6 sport 0x%x raddr %pI6 rport 0x%x\n",
3418 			 laddr6->sin6_addr.s6_addr,
3419 			 ntohs(laddr6->sin6_port),
3420 			 raddr6->sin6_addr.s6_addr, ntohs(raddr6->sin6_port));
3421 		ep->dst = cxgb_find_route6(&dev->rdev.lldi, get_real_dev,
3422 					   laddr6->sin6_addr.s6_addr,
3423 					   raddr6->sin6_addr.s6_addr,
3424 					   laddr6->sin6_port,
3425 					   raddr6->sin6_port, cm_id->tos,
3426 					   raddr6->sin6_scope_id);
3427 	}
3428 	if (!ep->dst) {
3429 		pr_err("%s - cannot find route\n", __func__);
3430 		err = -EHOSTUNREACH;
3431 		goto fail3;
3432 	}
3433 
3434 	err = import_ep(ep, iptype, ra, ep->dst, ep->com.dev, true,
3435 			ep->com.dev->rdev.lldi.adapter_type, cm_id->tos);
3436 	if (err) {
3437 		pr_err("%s - cannot alloc l2e\n", __func__);
3438 		goto fail4;
3439 	}
3440 
3441 	pr_debug("txq_idx %u tx_chan %u smac_idx %u rss_qid %u l2t_idx %u\n",
3442 		 ep->txq_idx, ep->tx_chan, ep->smac_idx, ep->rss_qid,
3443 		 ep->l2t->idx);
3444 
3445 	state_set(&ep->com, CONNECTING);
3446 	ep->tos = cm_id->tos;
3447 
3448 	/* send connect request to rnic */
3449 	err = send_connect(ep);
3450 	if (!err)
3451 		goto out;
3452 
3453 	cxgb4_l2t_release(ep->l2t);
3454 fail4:
3455 	dst_release(ep->dst);
3456 fail3:
3457 	xa_erase_irq(&ep->com.dev->atids, ep->atid);
3458 fail5:
3459 	cxgb4_free_atid(ep->com.dev->rdev.lldi.tids, ep->atid);
3460 fail2:
3461 	skb_queue_purge(&ep->com.ep_skb_list);
3462 	deref_cm_id(&ep->com);
3463 fail1:
3464 	c4iw_put_ep(&ep->com);
3465 out:
3466 	return err;
3467 }
3468 
create_server6(struct c4iw_dev * dev,struct c4iw_listen_ep * ep)3469 static int create_server6(struct c4iw_dev *dev, struct c4iw_listen_ep *ep)
3470 {
3471 	int err;
3472 	struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)
3473 				    &ep->com.local_addr;
3474 
3475 	if (ipv6_addr_type(&sin6->sin6_addr) != IPV6_ADDR_ANY) {
3476 		err = cxgb4_clip_get(ep->com.dev->rdev.lldi.ports[0],
3477 				     (const u32 *)&sin6->sin6_addr.s6_addr, 1);
3478 		if (err)
3479 			return err;
3480 	}
3481 	c4iw_init_wr_wait(ep->com.wr_waitp);
3482 	err = cxgb4_create_server6(ep->com.dev->rdev.lldi.ports[0],
3483 				   ep->stid, &sin6->sin6_addr,
3484 				   sin6->sin6_port,
3485 				   ep->com.dev->rdev.lldi.rxq_ids[0]);
3486 	if (!err)
3487 		err = c4iw_wait_for_reply(&ep->com.dev->rdev,
3488 					  ep->com.wr_waitp,
3489 					  0, 0, __func__);
3490 	else if (err > 0)
3491 		err = net_xmit_errno(err);
3492 	if (err) {
3493 		cxgb4_clip_release(ep->com.dev->rdev.lldi.ports[0],
3494 				   (const u32 *)&sin6->sin6_addr.s6_addr, 1);
3495 		pr_err("cxgb4_create_server6/filter failed err %d stid %d laddr %pI6 lport %d\n",
3496 		       err, ep->stid,
3497 		       sin6->sin6_addr.s6_addr, ntohs(sin6->sin6_port));
3498 	}
3499 	return err;
3500 }
3501 
create_server4(struct c4iw_dev * dev,struct c4iw_listen_ep * ep)3502 static int create_server4(struct c4iw_dev *dev, struct c4iw_listen_ep *ep)
3503 {
3504 	int err;
3505 	struct sockaddr_in *sin = (struct sockaddr_in *)
3506 				  &ep->com.local_addr;
3507 
3508 	if (dev->rdev.lldi.enable_fw_ofld_conn) {
3509 		do {
3510 			err = cxgb4_create_server_filter(
3511 				ep->com.dev->rdev.lldi.ports[0], ep->stid,
3512 				sin->sin_addr.s_addr, sin->sin_port, 0,
3513 				ep->com.dev->rdev.lldi.rxq_ids[0], 0, 0);
3514 			if (err == -EBUSY) {
3515 				if (c4iw_fatal_error(&ep->com.dev->rdev)) {
3516 					err = -EIO;
3517 					break;
3518 				}
3519 				set_current_state(TASK_UNINTERRUPTIBLE);
3520 				schedule_timeout(usecs_to_jiffies(100));
3521 			}
3522 		} while (err == -EBUSY);
3523 	} else {
3524 		c4iw_init_wr_wait(ep->com.wr_waitp);
3525 		err = cxgb4_create_server(ep->com.dev->rdev.lldi.ports[0],
3526 				ep->stid, sin->sin_addr.s_addr, sin->sin_port,
3527 				0, ep->com.dev->rdev.lldi.rxq_ids[0]);
3528 		if (!err)
3529 			err = c4iw_wait_for_reply(&ep->com.dev->rdev,
3530 						  ep->com.wr_waitp,
3531 						  0, 0, __func__);
3532 		else if (err > 0)
3533 			err = net_xmit_errno(err);
3534 	}
3535 	if (err)
3536 		pr_err("cxgb4_create_server/filter failed err %d stid %d laddr %pI4 lport %d\n"
3537 		       , err, ep->stid,
3538 		       &sin->sin_addr, ntohs(sin->sin_port));
3539 	return err;
3540 }
3541 
c4iw_create_listen(struct iw_cm_id * cm_id,int backlog)3542 int c4iw_create_listen(struct iw_cm_id *cm_id, int backlog)
3543 {
3544 	int err = 0;
3545 	struct c4iw_dev *dev = to_c4iw_dev(cm_id->device);
3546 	struct c4iw_listen_ep *ep;
3547 
3548 	might_sleep();
3549 
3550 	ep = alloc_ep(sizeof(*ep), GFP_KERNEL);
3551 	if (!ep) {
3552 		pr_err("%s - cannot alloc ep\n", __func__);
3553 		err = -ENOMEM;
3554 		goto fail1;
3555 	}
3556 	skb_queue_head_init(&ep->com.ep_skb_list);
3557 	pr_debug("ep %p\n", ep);
3558 	ep->com.cm_id = cm_id;
3559 	ref_cm_id(&ep->com);
3560 	ep->com.dev = dev;
3561 	ep->backlog = backlog;
3562 	memcpy(&ep->com.local_addr, &cm_id->m_local_addr,
3563 	       sizeof(ep->com.local_addr));
3564 
3565 	/*
3566 	 * Allocate a server TID.
3567 	 */
3568 	if (dev->rdev.lldi.enable_fw_ofld_conn &&
3569 	    ep->com.local_addr.ss_family == AF_INET)
3570 		ep->stid = cxgb4_alloc_sftid(dev->rdev.lldi.tids,
3571 					     cm_id->m_local_addr.ss_family, ep);
3572 	else
3573 		ep->stid = cxgb4_alloc_stid(dev->rdev.lldi.tids,
3574 					    cm_id->m_local_addr.ss_family, ep);
3575 
3576 	if (ep->stid == -1) {
3577 		pr_err("%s - cannot alloc stid\n", __func__);
3578 		err = -ENOMEM;
3579 		goto fail2;
3580 	}
3581 	err = xa_insert_irq(&dev->stids, ep->stid, ep, GFP_KERNEL);
3582 	if (err)
3583 		goto fail3;
3584 
3585 	state_set(&ep->com, LISTEN);
3586 	if (ep->com.local_addr.ss_family == AF_INET)
3587 		err = create_server4(dev, ep);
3588 	else
3589 		err = create_server6(dev, ep);
3590 	if (!err) {
3591 		cm_id->provider_data = ep;
3592 		goto out;
3593 	}
3594 	xa_erase_irq(&ep->com.dev->stids, ep->stid);
3595 fail3:
3596 	cxgb4_free_stid(ep->com.dev->rdev.lldi.tids, ep->stid,
3597 			ep->com.local_addr.ss_family);
3598 fail2:
3599 	deref_cm_id(&ep->com);
3600 	c4iw_put_ep(&ep->com);
3601 fail1:
3602 out:
3603 	return err;
3604 }
3605 
c4iw_destroy_listen(struct iw_cm_id * cm_id)3606 int c4iw_destroy_listen(struct iw_cm_id *cm_id)
3607 {
3608 	int err;
3609 	struct c4iw_listen_ep *ep = to_listen_ep(cm_id);
3610 
3611 	pr_debug("ep %p\n", ep);
3612 
3613 	might_sleep();
3614 	state_set(&ep->com, DEAD);
3615 	if (ep->com.dev->rdev.lldi.enable_fw_ofld_conn &&
3616 	    ep->com.local_addr.ss_family == AF_INET) {
3617 		err = cxgb4_remove_server_filter(
3618 			ep->com.dev->rdev.lldi.ports[0], ep->stid,
3619 			ep->com.dev->rdev.lldi.rxq_ids[0], false);
3620 	} else {
3621 		struct sockaddr_in6 *sin6;
3622 		c4iw_init_wr_wait(ep->com.wr_waitp);
3623 		err = cxgb4_remove_server(
3624 				ep->com.dev->rdev.lldi.ports[0], ep->stid,
3625 				ep->com.dev->rdev.lldi.rxq_ids[0],
3626 				ep->com.local_addr.ss_family == AF_INET6);
3627 		if (err)
3628 			goto done;
3629 		err = c4iw_wait_for_reply(&ep->com.dev->rdev, ep->com.wr_waitp,
3630 					  0, 0, __func__);
3631 		sin6 = (struct sockaddr_in6 *)&ep->com.local_addr;
3632 		cxgb4_clip_release(ep->com.dev->rdev.lldi.ports[0],
3633 				   (const u32 *)&sin6->sin6_addr.s6_addr, 1);
3634 	}
3635 	xa_erase_irq(&ep->com.dev->stids, ep->stid);
3636 	cxgb4_free_stid(ep->com.dev->rdev.lldi.tids, ep->stid,
3637 			ep->com.local_addr.ss_family);
3638 done:
3639 	deref_cm_id(&ep->com);
3640 	c4iw_put_ep(&ep->com);
3641 	return err;
3642 }
3643 
c4iw_ep_disconnect(struct c4iw_ep * ep,int abrupt,gfp_t gfp)3644 int c4iw_ep_disconnect(struct c4iw_ep *ep, int abrupt, gfp_t gfp)
3645 {
3646 	int ret = 0;
3647 	int close = 0;
3648 	int fatal = 0;
3649 	struct c4iw_rdev *rdev;
3650 
3651 	mutex_lock(&ep->com.mutex);
3652 
3653 	pr_debug("ep %p state %s, abrupt %d\n", ep,
3654 		 states[ep->com.state], abrupt);
3655 
3656 	/*
3657 	 * Ref the ep here in case we have fatal errors causing the
3658 	 * ep to be released and freed.
3659 	 */
3660 	c4iw_get_ep(&ep->com);
3661 
3662 	rdev = &ep->com.dev->rdev;
3663 	if (c4iw_fatal_error(rdev)) {
3664 		fatal = 1;
3665 		close_complete_upcall(ep, -EIO);
3666 		ep->com.state = DEAD;
3667 	}
3668 	switch (ep->com.state) {
3669 	case MPA_REQ_WAIT:
3670 	case MPA_REQ_SENT:
3671 	case MPA_REQ_RCVD:
3672 	case MPA_REP_SENT:
3673 	case FPDU_MODE:
3674 	case CONNECTING:
3675 		close = 1;
3676 		if (abrupt)
3677 			ep->com.state = ABORTING;
3678 		else {
3679 			ep->com.state = CLOSING;
3680 
3681 			/*
3682 			 * if we close before we see the fw4_ack() then we fix
3683 			 * up the timer state since we're reusing it.
3684 			 */
3685 			if (ep->mpa_skb &&
3686 			    test_bit(STOP_MPA_TIMER, &ep->com.flags)) {
3687 				clear_bit(STOP_MPA_TIMER, &ep->com.flags);
3688 				stop_ep_timer(ep);
3689 			}
3690 			start_ep_timer(ep);
3691 		}
3692 		set_bit(CLOSE_SENT, &ep->com.flags);
3693 		break;
3694 	case CLOSING:
3695 		if (!test_and_set_bit(CLOSE_SENT, &ep->com.flags)) {
3696 			close = 1;
3697 			if (abrupt) {
3698 				(void)stop_ep_timer(ep);
3699 				ep->com.state = ABORTING;
3700 			} else
3701 				ep->com.state = MORIBUND;
3702 		}
3703 		break;
3704 	case MORIBUND:
3705 	case ABORTING:
3706 	case DEAD:
3707 		pr_debug("ignoring disconnect ep %p state %u\n",
3708 			 ep, ep->com.state);
3709 		break;
3710 	default:
3711 		WARN_ONCE(1, "Bad endpoint state %u\n", ep->com.state);
3712 		break;
3713 	}
3714 
3715 	if (close) {
3716 		if (abrupt) {
3717 			set_bit(EP_DISC_ABORT, &ep->com.history);
3718 			ret = send_abort(ep);
3719 		} else {
3720 			set_bit(EP_DISC_CLOSE, &ep->com.history);
3721 			ret = send_halfclose(ep);
3722 		}
3723 		if (ret) {
3724 			set_bit(EP_DISC_FAIL, &ep->com.history);
3725 			if (!abrupt) {
3726 				stop_ep_timer(ep);
3727 				close_complete_upcall(ep, -EIO);
3728 			}
3729 			if (ep->com.qp) {
3730 				struct c4iw_qp_attributes attrs;
3731 
3732 				attrs.next_state = C4IW_QP_STATE_ERROR;
3733 				ret = c4iw_modify_qp(ep->com.qp->rhp,
3734 						     ep->com.qp,
3735 						     C4IW_QP_ATTR_NEXT_STATE,
3736 						     &attrs, 1);
3737 				if (ret)
3738 					pr_err("%s - qp <- error failed!\n",
3739 					       __func__);
3740 			}
3741 			fatal = 1;
3742 		}
3743 	}
3744 	mutex_unlock(&ep->com.mutex);
3745 	c4iw_put_ep(&ep->com);
3746 	if (fatal)
3747 		release_ep_resources(ep);
3748 	return ret;
3749 }
3750 
active_ofld_conn_reply(struct c4iw_dev * dev,struct sk_buff * skb,struct cpl_fw6_msg_ofld_connection_wr_rpl * req)3751 static void active_ofld_conn_reply(struct c4iw_dev *dev, struct sk_buff *skb,
3752 			struct cpl_fw6_msg_ofld_connection_wr_rpl *req)
3753 {
3754 	struct c4iw_ep *ep;
3755 	int atid = be32_to_cpu(req->tid);
3756 
3757 	ep = (struct c4iw_ep *)lookup_atid(dev->rdev.lldi.tids,
3758 					   (__force u32) req->tid);
3759 	if (!ep)
3760 		return;
3761 
3762 	switch (req->retval) {
3763 	case FW_ENOMEM:
3764 		set_bit(ACT_RETRY_NOMEM, &ep->com.history);
3765 		if (ep->retry_count++ < ACT_OPEN_RETRY_COUNT) {
3766 			send_fw_act_open_req(ep, atid);
3767 			return;
3768 		}
3769 		fallthrough;
3770 	case FW_EADDRINUSE:
3771 		set_bit(ACT_RETRY_INUSE, &ep->com.history);
3772 		if (ep->retry_count++ < ACT_OPEN_RETRY_COUNT) {
3773 			send_fw_act_open_req(ep, atid);
3774 			return;
3775 		}
3776 		break;
3777 	default:
3778 		pr_info("%s unexpected ofld conn wr retval %d\n",
3779 		       __func__, req->retval);
3780 		break;
3781 	}
3782 	pr_err("active ofld_connect_wr failure %d atid %d\n",
3783 	       req->retval, atid);
3784 	mutex_lock(&dev->rdev.stats.lock);
3785 	dev->rdev.stats.act_ofld_conn_fails++;
3786 	mutex_unlock(&dev->rdev.stats.lock);
3787 	connect_reply_upcall(ep, status2errno(req->retval));
3788 	state_set(&ep->com, DEAD);
3789 	if (ep->com.remote_addr.ss_family == AF_INET6) {
3790 		struct sockaddr_in6 *sin6 =
3791 			(struct sockaddr_in6 *)&ep->com.local_addr;
3792 		cxgb4_clip_release(ep->com.dev->rdev.lldi.ports[0],
3793 				   (const u32 *)&sin6->sin6_addr.s6_addr, 1);
3794 	}
3795 	xa_erase_irq(&dev->atids, atid);
3796 	cxgb4_free_atid(dev->rdev.lldi.tids, atid);
3797 	dst_release(ep->dst);
3798 	cxgb4_l2t_release(ep->l2t);
3799 	c4iw_put_ep(&ep->com);
3800 }
3801 
passive_ofld_conn_reply(struct c4iw_dev * dev,struct sk_buff * skb,struct cpl_fw6_msg_ofld_connection_wr_rpl * req)3802 static void passive_ofld_conn_reply(struct c4iw_dev *dev, struct sk_buff *skb,
3803 			struct cpl_fw6_msg_ofld_connection_wr_rpl *req)
3804 {
3805 	struct sk_buff *rpl_skb;
3806 	struct cpl_pass_accept_req *cpl;
3807 	int ret;
3808 
3809 	rpl_skb = (struct sk_buff *)(unsigned long)req->cookie;
3810 	if (req->retval) {
3811 		pr_err("%s passive open failure %d\n", __func__, req->retval);
3812 		mutex_lock(&dev->rdev.stats.lock);
3813 		dev->rdev.stats.pas_ofld_conn_fails++;
3814 		mutex_unlock(&dev->rdev.stats.lock);
3815 		kfree_skb(rpl_skb);
3816 	} else {
3817 		cpl = (struct cpl_pass_accept_req *)cplhdr(rpl_skb);
3818 		OPCODE_TID(cpl) = htonl(MK_OPCODE_TID(CPL_PASS_ACCEPT_REQ,
3819 					(__force u32) htonl(
3820 					(__force u32) req->tid)));
3821 		ret = pass_accept_req(dev, rpl_skb);
3822 		if (!ret)
3823 			kfree_skb(rpl_skb);
3824 	}
3825 	return;
3826 }
3827 
t4_tcb_get_field64(__be64 * tcb,u16 word)3828 static inline u64 t4_tcb_get_field64(__be64 *tcb, u16 word)
3829 {
3830 	u64 tlo = be64_to_cpu(tcb[((31 - word) / 2)]);
3831 	u64 thi = be64_to_cpu(tcb[((31 - word) / 2) - 1]);
3832 	u64 t;
3833 	u32 shift = 32;
3834 
3835 	t = (thi << shift) | (tlo >> shift);
3836 
3837 	return t;
3838 }
3839 
t4_tcb_get_field32(__be64 * tcb,u16 word,u32 mask,u32 shift)3840 static inline u32 t4_tcb_get_field32(__be64 *tcb, u16 word, u32 mask, u32 shift)
3841 {
3842 	u32 v;
3843 	u64 t = be64_to_cpu(tcb[(31 - word) / 2]);
3844 
3845 	if (word & 0x1)
3846 		shift += 32;
3847 	v = (t >> shift) & mask;
3848 	return v;
3849 }
3850 
read_tcb_rpl(struct c4iw_dev * dev,struct sk_buff * skb)3851 static int read_tcb_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
3852 {
3853 	struct cpl_get_tcb_rpl *rpl = cplhdr(skb);
3854 	__be64 *tcb = (__be64 *)(rpl + 1);
3855 	unsigned int tid = GET_TID(rpl);
3856 	struct c4iw_ep *ep;
3857 	u64 t_flags_64;
3858 	u32 rx_pdu_out;
3859 
3860 	ep = get_ep_from_tid(dev, tid);
3861 	if (!ep)
3862 		return 0;
3863 	/* Examine the TF_RX_PDU_OUT (bit 49 of the t_flags) in order to
3864 	 * determine if there's a rx PDU feedback event pending.
3865 	 *
3866 	 * If that bit is set, it means we'll need to re-read the TCB's
3867 	 * rq_start value. The final value is the one present in a TCB
3868 	 * with the TF_RX_PDU_OUT bit cleared.
3869 	 */
3870 
3871 	t_flags_64 = t4_tcb_get_field64(tcb, TCB_T_FLAGS_W);
3872 	rx_pdu_out = (t_flags_64 & TF_RX_PDU_OUT_V(1)) >> TF_RX_PDU_OUT_S;
3873 
3874 	c4iw_put_ep(&ep->com); /* from get_ep_from_tid() */
3875 	c4iw_put_ep(&ep->com); /* from read_tcb() */
3876 
3877 	/* If TF_RX_PDU_OUT bit is set, re-read the TCB */
3878 	if (rx_pdu_out) {
3879 		if (++ep->rx_pdu_out_cnt >= 2) {
3880 			WARN_ONCE(1, "tcb re-read() reached the guard limit, finishing the cleanup\n");
3881 			goto cleanup;
3882 		}
3883 		read_tcb(ep);
3884 		return 0;
3885 	}
3886 
3887 	ep->srqe_idx = t4_tcb_get_field32(tcb, TCB_RQ_START_W, TCB_RQ_START_M,
3888 					  TCB_RQ_START_S);
3889 cleanup:
3890 	pr_debug("ep %p tid %u %016x\n", ep, ep->hwtid, ep->srqe_idx);
3891 
3892 	if (test_bit(PEER_ABORT_IN_PROGRESS, &ep->com.flags))
3893 		finish_peer_abort(dev, ep);
3894 	else if (test_bit(ABORT_REQ_IN_PROGRESS, &ep->com.flags))
3895 		send_abort_req(ep);
3896 	else
3897 		WARN_ONCE(1, "unexpected state!");
3898 
3899 	return 0;
3900 }
3901 
deferred_fw6_msg(struct c4iw_dev * dev,struct sk_buff * skb)3902 static int deferred_fw6_msg(struct c4iw_dev *dev, struct sk_buff *skb)
3903 {
3904 	struct cpl_fw6_msg *rpl = cplhdr(skb);
3905 	struct cpl_fw6_msg_ofld_connection_wr_rpl *req;
3906 
3907 	switch (rpl->type) {
3908 	case FW6_TYPE_CQE:
3909 		c4iw_ev_dispatch(dev, (struct t4_cqe *)&rpl->data[0]);
3910 		break;
3911 	case FW6_TYPE_OFLD_CONNECTION_WR_RPL:
3912 		req = (struct cpl_fw6_msg_ofld_connection_wr_rpl *)rpl->data;
3913 		switch (req->t_state) {
3914 		case TCP_SYN_SENT:
3915 			active_ofld_conn_reply(dev, skb, req);
3916 			break;
3917 		case TCP_SYN_RECV:
3918 			passive_ofld_conn_reply(dev, skb, req);
3919 			break;
3920 		default:
3921 			pr_err("%s unexpected ofld conn wr state %d\n",
3922 			       __func__, req->t_state);
3923 			break;
3924 		}
3925 		break;
3926 	}
3927 	return 0;
3928 }
3929 
build_cpl_pass_accept_req(struct sk_buff * skb,int stid,u8 tos)3930 static void build_cpl_pass_accept_req(struct sk_buff *skb, int stid , u8 tos)
3931 {
3932 	__be32 l2info;
3933 	__be16 hdr_len, vlantag, len;
3934 	u16 eth_hdr_len;
3935 	int tcp_hdr_len, ip_hdr_len;
3936 	u8 intf;
3937 	struct cpl_rx_pkt *cpl = cplhdr(skb);
3938 	struct cpl_pass_accept_req *req;
3939 	struct tcp_options_received tmp_opt;
3940 	struct c4iw_dev *dev;
3941 	enum chip_type type;
3942 
3943 	dev = *((struct c4iw_dev **) (skb->cb + sizeof(void *)));
3944 	/* Store values from cpl_rx_pkt in temporary location. */
3945 	vlantag = cpl->vlan;
3946 	len = cpl->len;
3947 	l2info  = cpl->l2info;
3948 	hdr_len = cpl->hdr_len;
3949 	intf = cpl->iff;
3950 
3951 	__skb_pull(skb, sizeof(*req) + sizeof(struct rss_header));
3952 
3953 	/*
3954 	 * We need to parse the TCP options from SYN packet.
3955 	 * to generate cpl_pass_accept_req.
3956 	 */
3957 	memset(&tmp_opt, 0, sizeof(tmp_opt));
3958 	tcp_clear_options(&tmp_opt);
3959 	tcp_parse_options(&init_net, skb, &tmp_opt, 0, NULL);
3960 
3961 	req = __skb_push(skb, sizeof(*req));
3962 	memset(req, 0, sizeof(*req));
3963 	req->l2info = cpu_to_be16(SYN_INTF_V(intf) |
3964 			 SYN_MAC_IDX_V(RX_MACIDX_G(
3965 			 be32_to_cpu(l2info))) |
3966 			 SYN_XACT_MATCH_F);
3967 	type = dev->rdev.lldi.adapter_type;
3968 	tcp_hdr_len = RX_TCPHDR_LEN_G(be16_to_cpu(hdr_len));
3969 	ip_hdr_len = RX_IPHDR_LEN_G(be16_to_cpu(hdr_len));
3970 	req->hdr_len =
3971 		cpu_to_be32(SYN_RX_CHAN_V(RX_CHAN_G(be32_to_cpu(l2info))));
3972 	if (CHELSIO_CHIP_VERSION(type) <= CHELSIO_T5) {
3973 		eth_hdr_len = is_t4(type) ?
3974 				RX_ETHHDR_LEN_G(be32_to_cpu(l2info)) :
3975 				RX_T5_ETHHDR_LEN_G(be32_to_cpu(l2info));
3976 		req->hdr_len |= cpu_to_be32(TCP_HDR_LEN_V(tcp_hdr_len) |
3977 					    IP_HDR_LEN_V(ip_hdr_len) |
3978 					    ETH_HDR_LEN_V(eth_hdr_len));
3979 	} else { /* T6 and later */
3980 		eth_hdr_len = RX_T6_ETHHDR_LEN_G(be32_to_cpu(l2info));
3981 		req->hdr_len |= cpu_to_be32(T6_TCP_HDR_LEN_V(tcp_hdr_len) |
3982 					    T6_IP_HDR_LEN_V(ip_hdr_len) |
3983 					    T6_ETH_HDR_LEN_V(eth_hdr_len));
3984 	}
3985 	req->vlan = vlantag;
3986 	req->len = len;
3987 	req->tos_stid = cpu_to_be32(PASS_OPEN_TID_V(stid) |
3988 				    PASS_OPEN_TOS_V(tos));
3989 	req->tcpopt.mss = htons(tmp_opt.mss_clamp);
3990 	if (tmp_opt.wscale_ok)
3991 		req->tcpopt.wsf = tmp_opt.snd_wscale;
3992 	req->tcpopt.tstamp = tmp_opt.saw_tstamp;
3993 	if (tmp_opt.sack_ok)
3994 		req->tcpopt.sack = 1;
3995 	OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_PASS_ACCEPT_REQ, 0));
3996 	return;
3997 }
3998 
send_fw_pass_open_req(struct c4iw_dev * dev,struct sk_buff * skb,__be32 laddr,__be16 lport,__be32 raddr,__be16 rport,u32 rcv_isn,u32 filter,u16 window,u32 rss_qid,u8 port_id)3999 static void send_fw_pass_open_req(struct c4iw_dev *dev, struct sk_buff *skb,
4000 				  __be32 laddr, __be16 lport,
4001 				  __be32 raddr, __be16 rport,
4002 				  u32 rcv_isn, u32 filter, u16 window,
4003 				  u32 rss_qid, u8 port_id)
4004 {
4005 	struct sk_buff *req_skb;
4006 	struct fw_ofld_connection_wr *req;
4007 	struct cpl_pass_accept_req *cpl = cplhdr(skb);
4008 	int ret;
4009 
4010 	req_skb = alloc_skb(sizeof(struct fw_ofld_connection_wr), GFP_KERNEL);
4011 	if (!req_skb)
4012 		return;
4013 	req = __skb_put_zero(req_skb, sizeof(*req));
4014 	req->op_compl = htonl(WR_OP_V(FW_OFLD_CONNECTION_WR) | FW_WR_COMPL_F);
4015 	req->len16_pkd = htonl(FW_WR_LEN16_V(DIV_ROUND_UP(sizeof(*req), 16)));
4016 	req->le.version_cpl = htonl(FW_OFLD_CONNECTION_WR_CPL_F);
4017 	req->le.filter = (__force __be32) filter;
4018 	req->le.lport = lport;
4019 	req->le.pport = rport;
4020 	req->le.u.ipv4.lip = laddr;
4021 	req->le.u.ipv4.pip = raddr;
4022 	req->tcb.rcv_nxt = htonl(rcv_isn + 1);
4023 	req->tcb.rcv_adv = htons(window);
4024 	req->tcb.t_state_to_astid =
4025 		 htonl(FW_OFLD_CONNECTION_WR_T_STATE_V(TCP_SYN_RECV) |
4026 			FW_OFLD_CONNECTION_WR_RCV_SCALE_V(cpl->tcpopt.wsf) |
4027 			FW_OFLD_CONNECTION_WR_ASTID_V(
4028 			PASS_OPEN_TID_G(ntohl(cpl->tos_stid))));
4029 
4030 	/*
4031 	 * We store the qid in opt2 which will be used by the firmware
4032 	 * to send us the wr response.
4033 	 */
4034 	req->tcb.opt2 = htonl(RSS_QUEUE_V(rss_qid));
4035 
4036 	/*
4037 	 * We initialize the MSS index in TCB to 0xF.
4038 	 * So that when driver sends cpl_pass_accept_rpl
4039 	 * TCB picks up the correct value. If this was 0
4040 	 * TP will ignore any value > 0 for MSS index.
4041 	 */
4042 	req->tcb.opt0 = cpu_to_be64(MSS_IDX_V(0xF));
4043 	req->cookie = (uintptr_t)skb;
4044 
4045 	set_wr_txq(req_skb, CPL_PRIORITY_CONTROL, port_id);
4046 	ret = cxgb4_ofld_send(dev->rdev.lldi.ports[0], req_skb);
4047 	if (ret < 0) {
4048 		pr_err("%s - cxgb4_ofld_send error %d - dropping\n", __func__,
4049 		       ret);
4050 		kfree_skb(skb);
4051 		kfree_skb(req_skb);
4052 	}
4053 }
4054 
4055 /*
4056  * Handler for CPL_RX_PKT message. Need to handle cpl_rx_pkt
4057  * messages when a filter is being used instead of server to
4058  * redirect a syn packet. When packets hit filter they are redirected
4059  * to the offload queue and driver tries to establish the connection
4060  * using firmware work request.
4061  */
rx_pkt(struct c4iw_dev * dev,struct sk_buff * skb)4062 static int rx_pkt(struct c4iw_dev *dev, struct sk_buff *skb)
4063 {
4064 	int stid;
4065 	unsigned int filter;
4066 	struct ethhdr *eh = NULL;
4067 	struct vlan_ethhdr *vlan_eh = NULL;
4068 	struct iphdr *iph;
4069 	struct tcphdr *tcph;
4070 	struct rss_header *rss = (void *)skb->data;
4071 	struct cpl_rx_pkt *cpl = (void *)skb->data;
4072 	struct cpl_pass_accept_req *req = (void *)(rss + 1);
4073 	struct l2t_entry *e;
4074 	struct dst_entry *dst;
4075 	struct c4iw_ep *lep = NULL;
4076 	u16 window;
4077 	struct port_info *pi;
4078 	struct net_device *pdev;
4079 	u16 rss_qid, eth_hdr_len;
4080 	int step;
4081 	struct neighbour *neigh;
4082 
4083 	/* Drop all non-SYN packets */
4084 	if (!(cpl->l2info & cpu_to_be32(RXF_SYN_F)))
4085 		goto reject;
4086 
4087 	/*
4088 	 * Drop all packets which did not hit the filter.
4089 	 * Unlikely to happen.
4090 	 */
4091 	if (!(rss->filter_hit && rss->filter_tid))
4092 		goto reject;
4093 
4094 	/*
4095 	 * Calculate the server tid from filter hit index from cpl_rx_pkt.
4096 	 */
4097 	stid = (__force int) cpu_to_be32((__force u32) rss->hash_val);
4098 
4099 	lep = (struct c4iw_ep *)get_ep_from_stid(dev, stid);
4100 	if (!lep) {
4101 		pr_warn("%s connect request on invalid stid %d\n",
4102 			__func__, stid);
4103 		goto reject;
4104 	}
4105 
4106 	switch (CHELSIO_CHIP_VERSION(dev->rdev.lldi.adapter_type)) {
4107 	case CHELSIO_T4:
4108 		eth_hdr_len = RX_ETHHDR_LEN_G(be32_to_cpu(cpl->l2info));
4109 		break;
4110 	case CHELSIO_T5:
4111 		eth_hdr_len = RX_T5_ETHHDR_LEN_G(be32_to_cpu(cpl->l2info));
4112 		break;
4113 	case CHELSIO_T6:
4114 		eth_hdr_len = RX_T6_ETHHDR_LEN_G(be32_to_cpu(cpl->l2info));
4115 		break;
4116 	default:
4117 		pr_err("T%d Chip is not supported\n",
4118 		       CHELSIO_CHIP_VERSION(dev->rdev.lldi.adapter_type));
4119 		goto reject;
4120 	}
4121 
4122 	if (eth_hdr_len == ETH_HLEN) {
4123 		eh = (struct ethhdr *)(req + 1);
4124 		iph = (struct iphdr *)(eh + 1);
4125 	} else {
4126 		vlan_eh = (struct vlan_ethhdr *)(req + 1);
4127 		iph = (struct iphdr *)(vlan_eh + 1);
4128 		__vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), ntohs(cpl->vlan));
4129 	}
4130 
4131 	if (iph->version != 0x4)
4132 		goto reject;
4133 
4134 	tcph = (struct tcphdr *)(iph + 1);
4135 	skb_set_network_header(skb, (void *)iph - (void *)rss);
4136 	skb_set_transport_header(skb, (void *)tcph - (void *)rss);
4137 	skb_get(skb);
4138 
4139 	pr_debug("lip 0x%x lport %u pip 0x%x pport %u tos %d\n",
4140 		 ntohl(iph->daddr), ntohs(tcph->dest), ntohl(iph->saddr),
4141 		 ntohs(tcph->source), iph->tos);
4142 
4143 	dst = cxgb_find_route(&dev->rdev.lldi, get_real_dev,
4144 			      iph->daddr, iph->saddr, tcph->dest,
4145 			      tcph->source, iph->tos);
4146 	if (!dst) {
4147 		pr_err("%s - failed to find dst entry!\n", __func__);
4148 		goto reject;
4149 	}
4150 	neigh = dst_neigh_lookup_skb(dst, skb);
4151 
4152 	if (!neigh) {
4153 		pr_err("%s - failed to allocate neigh!\n", __func__);
4154 		goto free_dst;
4155 	}
4156 
4157 	if (neigh->dev->flags & IFF_LOOPBACK) {
4158 		pdev = ip_dev_find(&init_net, iph->daddr);
4159 		if (!pdev) {
4160 			pr_err("%s - failed to find device!\n", __func__);
4161 			goto free_dst;
4162 		}
4163 		e = cxgb4_l2t_get(dev->rdev.lldi.l2t, neigh,
4164 				    pdev, 0);
4165 		pi = (struct port_info *)netdev_priv(pdev);
4166 		dev_put(pdev);
4167 	} else {
4168 		pdev = get_real_dev(neigh->dev);
4169 		e = cxgb4_l2t_get(dev->rdev.lldi.l2t, neigh,
4170 					pdev, 0);
4171 		pi = (struct port_info *)netdev_priv(pdev);
4172 	}
4173 	neigh_release(neigh);
4174 	if (!e) {
4175 		pr_err("%s - failed to allocate l2t entry!\n",
4176 		       __func__);
4177 		goto free_dst;
4178 	}
4179 
4180 	step = dev->rdev.lldi.nrxq / dev->rdev.lldi.nchan;
4181 	rss_qid = dev->rdev.lldi.rxq_ids[pi->port_id * step];
4182 	window = (__force u16) htons((__force u16)tcph->window);
4183 
4184 	/* Calcuate filter portion for LE region. */
4185 	filter = (__force unsigned int) cpu_to_be32(cxgb4_select_ntuple(
4186 						    dev->rdev.lldi.ports[0],
4187 						    e));
4188 
4189 	/*
4190 	 * Synthesize the cpl_pass_accept_req. We have everything except the
4191 	 * TID. Once firmware sends a reply with TID we update the TID field
4192 	 * in cpl and pass it through the regular cpl_pass_accept_req path.
4193 	 */
4194 	build_cpl_pass_accept_req(skb, stid, iph->tos);
4195 	send_fw_pass_open_req(dev, skb, iph->daddr, tcph->dest, iph->saddr,
4196 			      tcph->source, ntohl(tcph->seq), filter, window,
4197 			      rss_qid, pi->port_id);
4198 	cxgb4_l2t_release(e);
4199 free_dst:
4200 	dst_release(dst);
4201 reject:
4202 	if (lep)
4203 		c4iw_put_ep(&lep->com);
4204 	return 0;
4205 }
4206 
4207 /*
4208  * These are the real handlers that are called from a
4209  * work queue.
4210  */
4211 static c4iw_handler_func work_handlers[NUM_CPL_CMDS + NUM_FAKE_CPLS] = {
4212 	[CPL_ACT_ESTABLISH] = act_establish,
4213 	[CPL_ACT_OPEN_RPL] = act_open_rpl,
4214 	[CPL_RX_DATA] = rx_data,
4215 	[CPL_ABORT_RPL_RSS] = abort_rpl,
4216 	[CPL_ABORT_RPL] = abort_rpl,
4217 	[CPL_PASS_OPEN_RPL] = pass_open_rpl,
4218 	[CPL_CLOSE_LISTSRV_RPL] = close_listsrv_rpl,
4219 	[CPL_PASS_ACCEPT_REQ] = pass_accept_req,
4220 	[CPL_PASS_ESTABLISH] = pass_establish,
4221 	[CPL_PEER_CLOSE] = peer_close,
4222 	[CPL_ABORT_REQ_RSS] = peer_abort,
4223 	[CPL_CLOSE_CON_RPL] = close_con_rpl,
4224 	[CPL_RDMA_TERMINATE] = terminate,
4225 	[CPL_FW4_ACK] = fw4_ack,
4226 	[CPL_GET_TCB_RPL] = read_tcb_rpl,
4227 	[CPL_FW6_MSG] = deferred_fw6_msg,
4228 	[CPL_RX_PKT] = rx_pkt,
4229 	[FAKE_CPL_PUT_EP_SAFE] = _put_ep_safe,
4230 	[FAKE_CPL_PASS_PUT_EP_SAFE] = _put_pass_ep_safe
4231 };
4232 
process_timeout(struct c4iw_ep * ep)4233 static void process_timeout(struct c4iw_ep *ep)
4234 {
4235 	struct c4iw_qp_attributes attrs;
4236 	int abort = 1;
4237 
4238 	mutex_lock(&ep->com.mutex);
4239 	pr_debug("ep %p tid %u state %d\n", ep, ep->hwtid, ep->com.state);
4240 	set_bit(TIMEDOUT, &ep->com.history);
4241 	switch (ep->com.state) {
4242 	case MPA_REQ_SENT:
4243 		connect_reply_upcall(ep, -ETIMEDOUT);
4244 		break;
4245 	case MPA_REQ_WAIT:
4246 	case MPA_REQ_RCVD:
4247 	case MPA_REP_SENT:
4248 	case FPDU_MODE:
4249 		break;
4250 	case CLOSING:
4251 	case MORIBUND:
4252 		if (ep->com.cm_id && ep->com.qp) {
4253 			attrs.next_state = C4IW_QP_STATE_ERROR;
4254 			c4iw_modify_qp(ep->com.qp->rhp,
4255 				     ep->com.qp, C4IW_QP_ATTR_NEXT_STATE,
4256 				     &attrs, 1);
4257 		}
4258 		close_complete_upcall(ep, -ETIMEDOUT);
4259 		break;
4260 	case ABORTING:
4261 	case DEAD:
4262 
4263 		/*
4264 		 * These states are expected if the ep timed out at the same
4265 		 * time as another thread was calling stop_ep_timer().
4266 		 * So we silently do nothing for these states.
4267 		 */
4268 		abort = 0;
4269 		break;
4270 	default:
4271 		WARN(1, "%s unexpected state ep %p tid %u state %u\n",
4272 			__func__, ep, ep->hwtid, ep->com.state);
4273 		abort = 0;
4274 	}
4275 	mutex_unlock(&ep->com.mutex);
4276 	if (abort)
4277 		c4iw_ep_disconnect(ep, 1, GFP_KERNEL);
4278 	c4iw_put_ep(&ep->com);
4279 }
4280 
process_timedout_eps(void)4281 static void process_timedout_eps(void)
4282 {
4283 	struct c4iw_ep *ep;
4284 
4285 	spin_lock_irq(&timeout_lock);
4286 	while (!list_empty(&timeout_list)) {
4287 		struct list_head *tmp;
4288 
4289 		tmp = timeout_list.next;
4290 		list_del(tmp);
4291 		tmp->next = NULL;
4292 		tmp->prev = NULL;
4293 		spin_unlock_irq(&timeout_lock);
4294 		ep = list_entry(tmp, struct c4iw_ep, entry);
4295 		process_timeout(ep);
4296 		spin_lock_irq(&timeout_lock);
4297 	}
4298 	spin_unlock_irq(&timeout_lock);
4299 }
4300 
process_work(struct work_struct * work)4301 static void process_work(struct work_struct *work)
4302 {
4303 	struct sk_buff *skb = NULL;
4304 	struct c4iw_dev *dev;
4305 	struct cpl_act_establish *rpl;
4306 	unsigned int opcode;
4307 	int ret;
4308 
4309 	process_timedout_eps();
4310 	while ((skb = skb_dequeue(&rxq))) {
4311 		rpl = cplhdr(skb);
4312 		dev = *((struct c4iw_dev **) (skb->cb + sizeof(void *)));
4313 		opcode = rpl->ot.opcode;
4314 
4315 		if (opcode >= ARRAY_SIZE(work_handlers) ||
4316 		    !work_handlers[opcode]) {
4317 			pr_err("No handler for opcode 0x%x.\n", opcode);
4318 			kfree_skb(skb);
4319 		} else {
4320 			ret = work_handlers[opcode](dev, skb);
4321 			if (!ret)
4322 				kfree_skb(skb);
4323 		}
4324 		process_timedout_eps();
4325 	}
4326 }
4327 
4328 static DECLARE_WORK(skb_work, process_work);
4329 
ep_timeout(struct timer_list * t)4330 static void ep_timeout(struct timer_list *t)
4331 {
4332 	struct c4iw_ep *ep = from_timer(ep, t, timer);
4333 	int kickit = 0;
4334 
4335 	spin_lock(&timeout_lock);
4336 	if (!test_and_set_bit(TIMEOUT, &ep->com.flags)) {
4337 		/*
4338 		 * Only insert if it is not already on the list.
4339 		 */
4340 		if (!ep->entry.next) {
4341 			list_add_tail(&ep->entry, &timeout_list);
4342 			kickit = 1;
4343 		}
4344 	}
4345 	spin_unlock(&timeout_lock);
4346 	if (kickit)
4347 		queue_work(workq, &skb_work);
4348 }
4349 
4350 /*
4351  * All the CM events are handled on a work queue to have a safe context.
4352  */
sched(struct c4iw_dev * dev,struct sk_buff * skb)4353 static int sched(struct c4iw_dev *dev, struct sk_buff *skb)
4354 {
4355 
4356 	/*
4357 	 * Save dev in the skb->cb area.
4358 	 */
4359 	*((struct c4iw_dev **) (skb->cb + sizeof(void *))) = dev;
4360 
4361 	/*
4362 	 * Queue the skb and schedule the worker thread.
4363 	 */
4364 	skb_queue_tail(&rxq, skb);
4365 	queue_work(workq, &skb_work);
4366 	return 0;
4367 }
4368 
set_tcb_rpl(struct c4iw_dev * dev,struct sk_buff * skb)4369 static int set_tcb_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
4370 {
4371 	struct cpl_set_tcb_rpl *rpl = cplhdr(skb);
4372 
4373 	if (rpl->status != CPL_ERR_NONE) {
4374 		pr_err("Unexpected SET_TCB_RPL status %u for tid %u\n",
4375 		       rpl->status, GET_TID(rpl));
4376 	}
4377 	kfree_skb(skb);
4378 	return 0;
4379 }
4380 
fw6_msg(struct c4iw_dev * dev,struct sk_buff * skb)4381 static int fw6_msg(struct c4iw_dev *dev, struct sk_buff *skb)
4382 {
4383 	struct cpl_fw6_msg *rpl = cplhdr(skb);
4384 	struct c4iw_wr_wait *wr_waitp;
4385 	int ret;
4386 
4387 	pr_debug("type %u\n", rpl->type);
4388 
4389 	switch (rpl->type) {
4390 	case FW6_TYPE_WR_RPL:
4391 		ret = (int)((be64_to_cpu(rpl->data[0]) >> 8) & 0xff);
4392 		wr_waitp = (struct c4iw_wr_wait *)(__force unsigned long) rpl->data[1];
4393 		pr_debug("wr_waitp %p ret %u\n", wr_waitp, ret);
4394 		if (wr_waitp)
4395 			c4iw_wake_up_deref(wr_waitp, ret ? -ret : 0);
4396 		kfree_skb(skb);
4397 		break;
4398 	case FW6_TYPE_CQE:
4399 	case FW6_TYPE_OFLD_CONNECTION_WR_RPL:
4400 		sched(dev, skb);
4401 		break;
4402 	default:
4403 		pr_err("%s unexpected fw6 msg type %u\n",
4404 		       __func__, rpl->type);
4405 		kfree_skb(skb);
4406 		break;
4407 	}
4408 	return 0;
4409 }
4410 
peer_abort_intr(struct c4iw_dev * dev,struct sk_buff * skb)4411 static int peer_abort_intr(struct c4iw_dev *dev, struct sk_buff *skb)
4412 {
4413 	struct cpl_abort_req_rss *req = cplhdr(skb);
4414 	struct c4iw_ep *ep;
4415 	unsigned int tid = GET_TID(req);
4416 
4417 	ep = get_ep_from_tid(dev, tid);
4418 	/* This EP will be dereferenced in peer_abort() */
4419 	if (!ep) {
4420 		pr_warn("Abort on non-existent endpoint, tid %d\n", tid);
4421 		kfree_skb(skb);
4422 		return 0;
4423 	}
4424 	if (cxgb_is_neg_adv(req->status)) {
4425 		pr_debug("Negative advice on abort- tid %u status %d (%s)\n",
4426 			 ep->hwtid, req->status,
4427 			 neg_adv_str(req->status));
4428 		goto out;
4429 	}
4430 	pr_debug("ep %p tid %u state %u\n", ep, ep->hwtid, ep->com.state);
4431 
4432 	c4iw_wake_up_noref(ep->com.wr_waitp, -ECONNRESET);
4433 out:
4434 	sched(dev, skb);
4435 	return 0;
4436 }
4437 
4438 /*
4439  * Most upcalls from the T4 Core go to sched() to
4440  * schedule the processing on a work queue.
4441  */
4442 c4iw_handler_func c4iw_handlers[NUM_CPL_CMDS] = {
4443 	[CPL_ACT_ESTABLISH] = sched,
4444 	[CPL_ACT_OPEN_RPL] = sched,
4445 	[CPL_RX_DATA] = sched,
4446 	[CPL_ABORT_RPL_RSS] = sched,
4447 	[CPL_ABORT_RPL] = sched,
4448 	[CPL_PASS_OPEN_RPL] = sched,
4449 	[CPL_CLOSE_LISTSRV_RPL] = sched,
4450 	[CPL_PASS_ACCEPT_REQ] = sched,
4451 	[CPL_PASS_ESTABLISH] = sched,
4452 	[CPL_PEER_CLOSE] = sched,
4453 	[CPL_CLOSE_CON_RPL] = sched,
4454 	[CPL_ABORT_REQ_RSS] = peer_abort_intr,
4455 	[CPL_RDMA_TERMINATE] = sched,
4456 	[CPL_FW4_ACK] = sched,
4457 	[CPL_SET_TCB_RPL] = set_tcb_rpl,
4458 	[CPL_GET_TCB_RPL] = sched,
4459 	[CPL_FW6_MSG] = fw6_msg,
4460 	[CPL_RX_PKT] = sched
4461 };
4462 
c4iw_cm_init(void)4463 int __init c4iw_cm_init(void)
4464 {
4465 	skb_queue_head_init(&rxq);
4466 
4467 	workq = alloc_ordered_workqueue("iw_cxgb4", WQ_MEM_RECLAIM);
4468 	if (!workq)
4469 		return -ENOMEM;
4470 
4471 	return 0;
4472 }
4473 
c4iw_cm_term(void)4474 void c4iw_cm_term(void)
4475 {
4476 	WARN_ON(!list_empty(&timeout_list));
4477 	flush_workqueue(workq);
4478 	destroy_workqueue(workq);
4479 }
4480