1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * Copyright (c) 2018 Chelsio Communications, Inc.
4  *
5  * Written by: Atul Gupta (atul.gupta@chelsio.com)
6  */
7 
8 #include <linux/module.h>
9 #include <linux/list.h>
10 #include <linux/workqueue.h>
11 #include <linux/skbuff.h>
12 #include <linux/timer.h>
13 #include <linux/notifier.h>
14 #include <linux/inetdevice.h>
15 #include <linux/ip.h>
16 #include <linux/tcp.h>
17 #include <linux/sched/signal.h>
18 #include <linux/kallsyms.h>
19 #include <linux/kprobes.h>
20 #include <linux/if_vlan.h>
21 #include <linux/ipv6.h>
22 #include <net/ipv6.h>
23 #include <net/transp_v6.h>
24 #include <net/ip6_route.h>
25 #include <net/inet_common.h>
26 #include <net/tcp.h>
27 #include <net/dst.h>
28 #include <net/tls.h>
29 #include <net/addrconf.h>
30 #include <net/secure_seq.h>
31 
32 #include "chtls.h"
33 #include "chtls_cm.h"
34 #include "clip_tbl.h"
35 #include "t4_tcb.h"
36 
37 /*
38  * State transitions and actions for close.  Note that if we are in SYN_SENT
39  * we remain in that state as we cannot control a connection while it's in
40  * SYN_SENT; such connections are allowed to establish and are then aborted.
41  */
42 static unsigned char new_state[16] = {
43 	/* current state:     new state:      action: */
44 	/* (Invalid)       */ TCP_CLOSE,
45 	/* TCP_ESTABLISHED */ TCP_FIN_WAIT1 | TCP_ACTION_FIN,
46 	/* TCP_SYN_SENT    */ TCP_SYN_SENT,
47 	/* TCP_SYN_RECV    */ TCP_FIN_WAIT1 | TCP_ACTION_FIN,
48 	/* TCP_FIN_WAIT1   */ TCP_FIN_WAIT1,
49 	/* TCP_FIN_WAIT2   */ TCP_FIN_WAIT2,
50 	/* TCP_TIME_WAIT   */ TCP_CLOSE,
51 	/* TCP_CLOSE       */ TCP_CLOSE,
52 	/* TCP_CLOSE_WAIT  */ TCP_LAST_ACK | TCP_ACTION_FIN,
53 	/* TCP_LAST_ACK    */ TCP_LAST_ACK,
54 	/* TCP_LISTEN      */ TCP_CLOSE,
55 	/* TCP_CLOSING     */ TCP_CLOSING,
56 };
57 
58 static struct chtls_sock *chtls_sock_create(struct chtls_dev *cdev)
59 {
60 	struct chtls_sock *csk = kzalloc(sizeof(*csk), GFP_ATOMIC);
61 
62 	if (!csk)
63 		return NULL;
64 
65 	csk->txdata_skb_cache = alloc_skb(TXDATA_SKB_LEN, GFP_ATOMIC);
66 	if (!csk->txdata_skb_cache) {
67 		kfree(csk);
68 		return NULL;
69 	}
70 
71 	kref_init(&csk->kref);
72 	csk->cdev = cdev;
73 	skb_queue_head_init(&csk->txq);
74 	csk->wr_skb_head = NULL;
75 	csk->wr_skb_tail = NULL;
76 	csk->mss = MAX_MSS;
77 	csk->tlshws.ofld = 1;
78 	csk->tlshws.txkey = -1;
79 	csk->tlshws.rxkey = -1;
80 	csk->tlshws.mfs = TLS_MFS;
81 	skb_queue_head_init(&csk->tlshws.sk_recv_queue);
82 	return csk;
83 }
84 
85 static void chtls_sock_release(struct kref *ref)
86 {
87 	struct chtls_sock *csk =
88 		container_of(ref, struct chtls_sock, kref);
89 
90 	kfree(csk);
91 }
92 
93 static struct net_device *chtls_find_netdev(struct chtls_dev *cdev,
94 					    struct sock *sk)
95 {
96 	struct adapter *adap = pci_get_drvdata(cdev->pdev);
97 	struct net_device *ndev = cdev->ports[0];
98 #if IS_ENABLED(CONFIG_IPV6)
99 	struct net_device *temp;
100 	int addr_type;
101 #endif
102 	int i;
103 
104 	switch (sk->sk_family) {
105 	case PF_INET:
106 		if (likely(!inet_sk(sk)->inet_rcv_saddr))
107 			return ndev;
108 		ndev = __ip_dev_find(&init_net, inet_sk(sk)->inet_rcv_saddr, false);
109 		break;
110 #if IS_ENABLED(CONFIG_IPV6)
111 	case PF_INET6:
112 		addr_type = ipv6_addr_type(&sk->sk_v6_rcv_saddr);
113 		if (likely(addr_type == IPV6_ADDR_ANY))
114 			return ndev;
115 
116 		for_each_netdev_rcu(&init_net, temp) {
117 			if (ipv6_chk_addr(&init_net, (struct in6_addr *)
118 					  &sk->sk_v6_rcv_saddr, temp, 1)) {
119 				ndev = temp;
120 				break;
121 			}
122 		}
123 	break;
124 #endif
125 	default:
126 		return NULL;
127 	}
128 
129 	if (!ndev)
130 		return NULL;
131 
132 	if (is_vlan_dev(ndev))
133 		ndev = vlan_dev_real_dev(ndev);
134 
135 	for_each_port(adap, i)
136 		if (cdev->ports[i] == ndev)
137 			return ndev;
138 	return NULL;
139 }
140 
141 static void assign_rxopt(struct sock *sk, unsigned int opt)
142 {
143 	const struct chtls_dev *cdev;
144 	struct chtls_sock *csk;
145 	struct tcp_sock *tp;
146 
147 	csk = rcu_dereference_sk_user_data(sk);
148 	tp = tcp_sk(sk);
149 
150 	cdev = csk->cdev;
151 	tp->tcp_header_len           = sizeof(struct tcphdr);
152 	tp->rx_opt.mss_clamp         = cdev->mtus[TCPOPT_MSS_G(opt)] - 40;
153 	tp->mss_cache                = tp->rx_opt.mss_clamp;
154 	tp->rx_opt.tstamp_ok         = TCPOPT_TSTAMP_G(opt);
155 	tp->rx_opt.snd_wscale        = TCPOPT_SACK_G(opt);
156 	tp->rx_opt.wscale_ok         = TCPOPT_WSCALE_OK_G(opt);
157 	SND_WSCALE(tp)               = TCPOPT_SND_WSCALE_G(opt);
158 	if (!tp->rx_opt.wscale_ok)
159 		tp->rx_opt.rcv_wscale = 0;
160 	if (tp->rx_opt.tstamp_ok) {
161 		tp->tcp_header_len += TCPOLEN_TSTAMP_ALIGNED;
162 		tp->rx_opt.mss_clamp -= TCPOLEN_TSTAMP_ALIGNED;
163 	} else if (csk->opt2 & TSTAMPS_EN_F) {
164 		csk->opt2 &= ~TSTAMPS_EN_F;
165 		csk->mtu_idx = TCPOPT_MSS_G(opt);
166 	}
167 }
168 
169 static void chtls_purge_receive_queue(struct sock *sk)
170 {
171 	struct sk_buff *skb;
172 
173 	while ((skb = __skb_dequeue(&sk->sk_receive_queue)) != NULL) {
174 		skb_dst_set(skb, (void *)NULL);
175 		kfree_skb(skb);
176 	}
177 }
178 
179 static void chtls_purge_write_queue(struct sock *sk)
180 {
181 	struct chtls_sock *csk = rcu_dereference_sk_user_data(sk);
182 	struct sk_buff *skb;
183 
184 	while ((skb = __skb_dequeue(&csk->txq))) {
185 		sk->sk_wmem_queued -= skb->truesize;
186 		__kfree_skb(skb);
187 	}
188 }
189 
190 static void chtls_purge_recv_queue(struct sock *sk)
191 {
192 	struct chtls_sock *csk = rcu_dereference_sk_user_data(sk);
193 	struct chtls_hws *tlsk = &csk->tlshws;
194 	struct sk_buff *skb;
195 
196 	while ((skb = __skb_dequeue(&tlsk->sk_recv_queue)) != NULL) {
197 		skb_dst_set(skb, NULL);
198 		kfree_skb(skb);
199 	}
200 }
201 
202 static void abort_arp_failure(void *handle, struct sk_buff *skb)
203 {
204 	struct cpl_abort_req *req = cplhdr(skb);
205 	struct chtls_dev *cdev;
206 
207 	cdev = (struct chtls_dev *)handle;
208 	req->cmd = CPL_ABORT_NO_RST;
209 	cxgb4_ofld_send(cdev->lldi->ports[0], skb);
210 }
211 
212 static struct sk_buff *alloc_ctrl_skb(struct sk_buff *skb, int len)
213 {
214 	if (likely(skb && !skb_shared(skb) && !skb_cloned(skb))) {
215 		__skb_trim(skb, 0);
216 		refcount_inc(&skb->users);
217 	} else {
218 		skb = alloc_skb(len, GFP_KERNEL | __GFP_NOFAIL);
219 	}
220 	return skb;
221 }
222 
223 static void chtls_send_abort(struct sock *sk, int mode, struct sk_buff *skb)
224 {
225 	struct cpl_abort_req *req;
226 	struct chtls_sock *csk;
227 	struct tcp_sock *tp;
228 
229 	csk = rcu_dereference_sk_user_data(sk);
230 	tp = tcp_sk(sk);
231 
232 	if (!skb)
233 		skb = alloc_ctrl_skb(csk->txdata_skb_cache, sizeof(*req));
234 
235 	req = (struct cpl_abort_req *)skb_put(skb, sizeof(*req));
236 	INIT_TP_WR_CPL(req, CPL_ABORT_REQ, csk->tid);
237 	skb_set_queue_mapping(skb, (csk->txq_idx << 1) | CPL_PRIORITY_DATA);
238 	req->rsvd0 = htonl(tp->snd_nxt);
239 	req->rsvd1 = !csk_flag_nochk(csk, CSK_TX_DATA_SENT);
240 	req->cmd = mode;
241 	t4_set_arp_err_handler(skb, csk->cdev, abort_arp_failure);
242 	send_or_defer(sk, tp, skb, mode == CPL_ABORT_SEND_RST);
243 }
244 
245 static void chtls_send_reset(struct sock *sk, int mode, struct sk_buff *skb)
246 {
247 	struct chtls_sock *csk = rcu_dereference_sk_user_data(sk);
248 
249 	if (unlikely(csk_flag_nochk(csk, CSK_ABORT_SHUTDOWN) ||
250 		     !csk->cdev)) {
251 		if (sk->sk_state == TCP_SYN_RECV)
252 			csk_set_flag(csk, CSK_RST_ABORTED);
253 		goto out;
254 	}
255 
256 	if (!csk_flag_nochk(csk, CSK_TX_DATA_SENT)) {
257 		struct tcp_sock *tp = tcp_sk(sk);
258 
259 		if (send_tx_flowc_wr(sk, 0, tp->snd_nxt, tp->rcv_nxt) < 0)
260 			WARN_ONCE(1, "send tx flowc error");
261 		csk_set_flag(csk, CSK_TX_DATA_SENT);
262 	}
263 
264 	csk_set_flag(csk, CSK_ABORT_RPL_PENDING);
265 	chtls_purge_write_queue(sk);
266 
267 	csk_set_flag(csk, CSK_ABORT_SHUTDOWN);
268 	if (sk->sk_state != TCP_SYN_RECV)
269 		chtls_send_abort(sk, mode, skb);
270 	else
271 		chtls_set_tcb_field_rpl_skb(sk, TCB_T_FLAGS_W,
272 					    TCB_T_FLAGS_V(TCB_T_FLAGS_M), 0,
273 					    TCB_FIELD_COOKIE_TFLAG, 1);
274 
275 	return;
276 out:
277 	kfree_skb(skb);
278 }
279 
280 static void release_tcp_port(struct sock *sk)
281 {
282 	if (inet_csk(sk)->icsk_bind_hash)
283 		inet_put_port(sk);
284 }
285 
286 static void tcp_uncork(struct sock *sk)
287 {
288 	struct tcp_sock *tp = tcp_sk(sk);
289 
290 	if (tp->nonagle & TCP_NAGLE_CORK) {
291 		tp->nonagle &= ~TCP_NAGLE_CORK;
292 		chtls_tcp_push(sk, 0);
293 	}
294 }
295 
296 static void chtls_close_conn(struct sock *sk)
297 {
298 	struct cpl_close_con_req *req;
299 	struct chtls_sock *csk;
300 	struct sk_buff *skb;
301 	unsigned int tid;
302 	unsigned int len;
303 
304 	len = roundup(sizeof(struct cpl_close_con_req), 16);
305 	csk = rcu_dereference_sk_user_data(sk);
306 	tid = csk->tid;
307 
308 	skb = alloc_skb(len, GFP_KERNEL | __GFP_NOFAIL);
309 	req = (struct cpl_close_con_req *)__skb_put(skb, len);
310 	memset(req, 0, len);
311 	req->wr.wr_hi = htonl(FW_WR_OP_V(FW_TP_WR) |
312 			      FW_WR_IMMDLEN_V(sizeof(*req) -
313 					      sizeof(req->wr)));
314 	req->wr.wr_mid = htonl(FW_WR_LEN16_V(DIV_ROUND_UP(sizeof(*req), 16)) |
315 			       FW_WR_FLOWID_V(tid));
316 
317 	OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_CLOSE_CON_REQ, tid));
318 
319 	tcp_uncork(sk);
320 	skb_entail(sk, skb, ULPCB_FLAG_NO_HDR | ULPCB_FLAG_NO_APPEND);
321 	if (sk->sk_state != TCP_SYN_SENT)
322 		chtls_push_frames(csk, 1);
323 }
324 
325 /*
326  * Perform a state transition during close and return the actions indicated
327  * for the transition.  Do not make this function inline, the main reason
328  * it exists at all is to avoid multiple inlining of tcp_set_state.
329  */
330 static int make_close_transition(struct sock *sk)
331 {
332 	int next = (int)new_state[sk->sk_state];
333 
334 	tcp_set_state(sk, next & TCP_STATE_MASK);
335 	return next & TCP_ACTION_FIN;
336 }
337 
338 void chtls_close(struct sock *sk, long timeout)
339 {
340 	int data_lost, prev_state;
341 	struct chtls_sock *csk;
342 
343 	csk = rcu_dereference_sk_user_data(sk);
344 
345 	lock_sock(sk);
346 	sk->sk_shutdown |= SHUTDOWN_MASK;
347 
348 	data_lost = skb_queue_len(&sk->sk_receive_queue);
349 	data_lost |= skb_queue_len(&csk->tlshws.sk_recv_queue);
350 	chtls_purge_recv_queue(sk);
351 	chtls_purge_receive_queue(sk);
352 
353 	if (sk->sk_state == TCP_CLOSE) {
354 		goto wait;
355 	} else if (data_lost || sk->sk_state == TCP_SYN_SENT) {
356 		chtls_send_reset(sk, CPL_ABORT_SEND_RST, NULL);
357 		release_tcp_port(sk);
358 		goto unlock;
359 	} else if (sock_flag(sk, SOCK_LINGER) && !sk->sk_lingertime) {
360 		sk->sk_prot->disconnect(sk, 0);
361 	} else if (make_close_transition(sk)) {
362 		chtls_close_conn(sk);
363 	}
364 wait:
365 	if (timeout)
366 		sk_stream_wait_close(sk, timeout);
367 
368 unlock:
369 	prev_state = sk->sk_state;
370 	sock_hold(sk);
371 	sock_orphan(sk);
372 
373 	release_sock(sk);
374 
375 	local_bh_disable();
376 	bh_lock_sock(sk);
377 
378 	if (prev_state != TCP_CLOSE && sk->sk_state == TCP_CLOSE)
379 		goto out;
380 
381 	if (sk->sk_state == TCP_FIN_WAIT2 && tcp_sk(sk)->linger2 < 0 &&
382 	    !csk_flag(sk, CSK_ABORT_SHUTDOWN)) {
383 		struct sk_buff *skb;
384 
385 		skb = alloc_skb(sizeof(struct cpl_abort_req), GFP_ATOMIC);
386 		if (skb)
387 			chtls_send_reset(sk, CPL_ABORT_SEND_RST, skb);
388 	}
389 
390 	if (sk->sk_state == TCP_CLOSE)
391 		inet_csk_destroy_sock(sk);
392 
393 out:
394 	bh_unlock_sock(sk);
395 	local_bh_enable();
396 	sock_put(sk);
397 }
398 
399 /*
400  * Wait until a socket enters on of the given states.
401  */
402 static int wait_for_states(struct sock *sk, unsigned int states)
403 {
404 	DECLARE_WAITQUEUE(wait, current);
405 	struct socket_wq _sk_wq;
406 	long current_timeo;
407 	int err = 0;
408 
409 	current_timeo = 200;
410 
411 	/*
412 	 * We want this to work even when there's no associated struct socket.
413 	 * In that case we provide a temporary wait_queue_head_t.
414 	 */
415 	if (!sk->sk_wq) {
416 		init_waitqueue_head(&_sk_wq.wait);
417 		_sk_wq.fasync_list = NULL;
418 		init_rcu_head_on_stack(&_sk_wq.rcu);
419 		RCU_INIT_POINTER(sk->sk_wq, &_sk_wq);
420 	}
421 
422 	add_wait_queue(sk_sleep(sk), &wait);
423 	while (!sk_in_state(sk, states)) {
424 		if (!current_timeo) {
425 			err = -EBUSY;
426 			break;
427 		}
428 		if (signal_pending(current)) {
429 			err = sock_intr_errno(current_timeo);
430 			break;
431 		}
432 		set_current_state(TASK_UNINTERRUPTIBLE);
433 		release_sock(sk);
434 		if (!sk_in_state(sk, states))
435 			current_timeo = schedule_timeout(current_timeo);
436 		__set_current_state(TASK_RUNNING);
437 		lock_sock(sk);
438 	}
439 	remove_wait_queue(sk_sleep(sk), &wait);
440 
441 	if (rcu_dereference(sk->sk_wq) == &_sk_wq)
442 		sk->sk_wq = NULL;
443 	return err;
444 }
445 
446 int chtls_disconnect(struct sock *sk, int flags)
447 {
448 	struct tcp_sock *tp;
449 	int err;
450 
451 	tp = tcp_sk(sk);
452 	chtls_purge_recv_queue(sk);
453 	chtls_purge_receive_queue(sk);
454 	chtls_purge_write_queue(sk);
455 
456 	if (sk->sk_state != TCP_CLOSE) {
457 		sk->sk_err = ECONNRESET;
458 		chtls_send_reset(sk, CPL_ABORT_SEND_RST, NULL);
459 		err = wait_for_states(sk, TCPF_CLOSE);
460 		if (err)
461 			return err;
462 	}
463 	chtls_purge_recv_queue(sk);
464 	chtls_purge_receive_queue(sk);
465 	tp->max_window = 0xFFFF << (tp->rx_opt.snd_wscale);
466 	return tcp_disconnect(sk, flags);
467 }
468 
469 #define SHUTDOWN_ELIGIBLE_STATE (TCPF_ESTABLISHED | \
470 				 TCPF_SYN_RECV | TCPF_CLOSE_WAIT)
471 void chtls_shutdown(struct sock *sk, int how)
472 {
473 	if ((how & SEND_SHUTDOWN) &&
474 	    sk_in_state(sk, SHUTDOWN_ELIGIBLE_STATE) &&
475 	    make_close_transition(sk))
476 		chtls_close_conn(sk);
477 }
478 
479 void chtls_destroy_sock(struct sock *sk)
480 {
481 	struct chtls_sock *csk;
482 
483 	csk = rcu_dereference_sk_user_data(sk);
484 	chtls_purge_recv_queue(sk);
485 	csk->ulp_mode = ULP_MODE_NONE;
486 	chtls_purge_write_queue(sk);
487 	free_tls_keyid(sk);
488 	kref_put(&csk->kref, chtls_sock_release);
489 	if (sk->sk_family == AF_INET)
490 		sk->sk_prot = &tcp_prot;
491 #if IS_ENABLED(CONFIG_IPV6)
492 	else
493 		sk->sk_prot = &tcpv6_prot;
494 #endif
495 	sk->sk_prot->destroy(sk);
496 }
497 
498 static void reset_listen_child(struct sock *child)
499 {
500 	struct chtls_sock *csk = rcu_dereference_sk_user_data(child);
501 	struct sk_buff *skb;
502 
503 	skb = alloc_ctrl_skb(csk->txdata_skb_cache,
504 			     sizeof(struct cpl_abort_req));
505 
506 	chtls_send_reset(child, CPL_ABORT_SEND_RST, skb);
507 	sock_orphan(child);
508 	INC_ORPHAN_COUNT(child);
509 	if (child->sk_state == TCP_CLOSE)
510 		inet_csk_destroy_sock(child);
511 }
512 
513 static void chtls_disconnect_acceptq(struct sock *listen_sk)
514 {
515 	struct request_sock **pprev;
516 
517 	pprev = ACCEPT_QUEUE(listen_sk);
518 	while (*pprev) {
519 		struct request_sock *req = *pprev;
520 
521 		if (req->rsk_ops == &chtls_rsk_ops ||
522 		    req->rsk_ops == &chtls_rsk_opsv6) {
523 			struct sock *child = req->sk;
524 
525 			*pprev = req->dl_next;
526 			sk_acceptq_removed(listen_sk);
527 			reqsk_put(req);
528 			sock_hold(child);
529 			local_bh_disable();
530 			bh_lock_sock(child);
531 			release_tcp_port(child);
532 			reset_listen_child(child);
533 			bh_unlock_sock(child);
534 			local_bh_enable();
535 			sock_put(child);
536 		} else {
537 			pprev = &req->dl_next;
538 		}
539 	}
540 }
541 
542 static int listen_hashfn(const struct sock *sk)
543 {
544 	return ((unsigned long)sk >> 10) & (LISTEN_INFO_HASH_SIZE - 1);
545 }
546 
547 static struct listen_info *listen_hash_add(struct chtls_dev *cdev,
548 					   struct sock *sk,
549 					   unsigned int stid)
550 {
551 	struct listen_info *p = kmalloc(sizeof(*p), GFP_KERNEL);
552 
553 	if (p) {
554 		int key = listen_hashfn(sk);
555 
556 		p->sk = sk;
557 		p->stid = stid;
558 		spin_lock(&cdev->listen_lock);
559 		p->next = cdev->listen_hash_tab[key];
560 		cdev->listen_hash_tab[key] = p;
561 		spin_unlock(&cdev->listen_lock);
562 	}
563 	return p;
564 }
565 
566 static int listen_hash_find(struct chtls_dev *cdev,
567 			    struct sock *sk)
568 {
569 	struct listen_info *p;
570 	int stid = -1;
571 	int key;
572 
573 	key = listen_hashfn(sk);
574 
575 	spin_lock(&cdev->listen_lock);
576 	for (p = cdev->listen_hash_tab[key]; p; p = p->next)
577 		if (p->sk == sk) {
578 			stid = p->stid;
579 			break;
580 		}
581 	spin_unlock(&cdev->listen_lock);
582 	return stid;
583 }
584 
585 static int listen_hash_del(struct chtls_dev *cdev,
586 			   struct sock *sk)
587 {
588 	struct listen_info *p, **prev;
589 	int stid = -1;
590 	int key;
591 
592 	key = listen_hashfn(sk);
593 	prev = &cdev->listen_hash_tab[key];
594 
595 	spin_lock(&cdev->listen_lock);
596 	for (p = *prev; p; prev = &p->next, p = p->next)
597 		if (p->sk == sk) {
598 			stid = p->stid;
599 			*prev = p->next;
600 			kfree(p);
601 			break;
602 		}
603 	spin_unlock(&cdev->listen_lock);
604 	return stid;
605 }
606 
607 static void cleanup_syn_rcv_conn(struct sock *child, struct sock *parent)
608 {
609 	struct request_sock *req;
610 	struct chtls_sock *csk;
611 
612 	csk = rcu_dereference_sk_user_data(child);
613 	req = csk->passive_reap_next;
614 
615 	reqsk_queue_removed(&inet_csk(parent)->icsk_accept_queue, req);
616 	__skb_unlink((struct sk_buff *)&csk->synq, &csk->listen_ctx->synq);
617 	chtls_reqsk_free(req);
618 	csk->passive_reap_next = NULL;
619 }
620 
621 static void chtls_reset_synq(struct listen_ctx *listen_ctx)
622 {
623 	struct sock *listen_sk = listen_ctx->lsk;
624 
625 	while (!skb_queue_empty(&listen_ctx->synq)) {
626 		struct chtls_sock *csk =
627 			container_of((struct synq *)skb_peek
628 				(&listen_ctx->synq), struct chtls_sock, synq);
629 		struct sock *child = csk->sk;
630 
631 		cleanup_syn_rcv_conn(child, listen_sk);
632 		sock_hold(child);
633 		local_bh_disable();
634 		bh_lock_sock(child);
635 		release_tcp_port(child);
636 		reset_listen_child(child);
637 		bh_unlock_sock(child);
638 		local_bh_enable();
639 		sock_put(child);
640 	}
641 }
642 
643 int chtls_listen_start(struct chtls_dev *cdev, struct sock *sk)
644 {
645 	struct net_device *ndev;
646 #if IS_ENABLED(CONFIG_IPV6)
647 	bool clip_valid = false;
648 #endif
649 	struct listen_ctx *ctx;
650 	struct adapter *adap;
651 	struct port_info *pi;
652 	int ret = 0;
653 	int stid;
654 
655 	rcu_read_lock();
656 	ndev = chtls_find_netdev(cdev, sk);
657 	rcu_read_unlock();
658 	if (!ndev)
659 		return -EBADF;
660 
661 	pi = netdev_priv(ndev);
662 	adap = pi->adapter;
663 	if (!(adap->flags & CXGB4_FULL_INIT_DONE))
664 		return -EBADF;
665 
666 	if (listen_hash_find(cdev, sk) >= 0)   /* already have it */
667 		return -EADDRINUSE;
668 
669 	ctx = kmalloc(sizeof(*ctx), GFP_KERNEL);
670 	if (!ctx)
671 		return -ENOMEM;
672 
673 	__module_get(THIS_MODULE);
674 	ctx->lsk = sk;
675 	ctx->cdev = cdev;
676 	ctx->state = T4_LISTEN_START_PENDING;
677 	skb_queue_head_init(&ctx->synq);
678 
679 	stid = cxgb4_alloc_stid(cdev->tids, sk->sk_family, ctx);
680 	if (stid < 0)
681 		goto free_ctx;
682 
683 	sock_hold(sk);
684 	if (!listen_hash_add(cdev, sk, stid))
685 		goto free_stid;
686 
687 	if (sk->sk_family == PF_INET) {
688 		ret = cxgb4_create_server(ndev, stid,
689 					  inet_sk(sk)->inet_rcv_saddr,
690 					  inet_sk(sk)->inet_sport, 0,
691 					  cdev->lldi->rxq_ids[0]);
692 #if IS_ENABLED(CONFIG_IPV6)
693 	} else {
694 		int addr_type;
695 
696 		addr_type = ipv6_addr_type(&sk->sk_v6_rcv_saddr);
697 		if (addr_type != IPV6_ADDR_ANY) {
698 			ret = cxgb4_clip_get(ndev, (const u32 *)
699 					     &sk->sk_v6_rcv_saddr, 1);
700 			if (ret)
701 				goto del_hash;
702 			clip_valid = true;
703 		}
704 		ret = cxgb4_create_server6(ndev, stid,
705 					   &sk->sk_v6_rcv_saddr,
706 					   inet_sk(sk)->inet_sport,
707 					   cdev->lldi->rxq_ids[0]);
708 #endif
709 	}
710 	if (ret > 0)
711 		ret = net_xmit_errno(ret);
712 	if (ret)
713 		goto del_hash;
714 	return 0;
715 del_hash:
716 #if IS_ENABLED(CONFIG_IPV6)
717 	if (clip_valid)
718 		cxgb4_clip_release(ndev, (const u32 *)&sk->sk_v6_rcv_saddr, 1);
719 #endif
720 	listen_hash_del(cdev, sk);
721 free_stid:
722 	cxgb4_free_stid(cdev->tids, stid, sk->sk_family);
723 	sock_put(sk);
724 free_ctx:
725 	kfree(ctx);
726 	module_put(THIS_MODULE);
727 	return -EBADF;
728 }
729 
730 void chtls_listen_stop(struct chtls_dev *cdev, struct sock *sk)
731 {
732 	struct listen_ctx *listen_ctx;
733 	int stid;
734 
735 	stid = listen_hash_del(cdev, sk);
736 	if (stid < 0)
737 		return;
738 
739 	listen_ctx = (struct listen_ctx *)lookup_stid(cdev->tids, stid);
740 	chtls_reset_synq(listen_ctx);
741 
742 	cxgb4_remove_server(cdev->lldi->ports[0], stid,
743 			    cdev->lldi->rxq_ids[0], sk->sk_family == PF_INET6);
744 
745 #if IS_ENABLED(CONFIG_IPV6)
746 	if (sk->sk_family == PF_INET6) {
747 		struct net_device *ndev = chtls_find_netdev(cdev, sk);
748 		int addr_type = 0;
749 
750 		addr_type = ipv6_addr_type((const struct in6_addr *)
751 					  &sk->sk_v6_rcv_saddr);
752 		if (addr_type != IPV6_ADDR_ANY)
753 			cxgb4_clip_release(ndev, (const u32 *)
754 					   &sk->sk_v6_rcv_saddr, 1);
755 	}
756 #endif
757 	chtls_disconnect_acceptq(sk);
758 }
759 
760 static int chtls_pass_open_rpl(struct chtls_dev *cdev, struct sk_buff *skb)
761 {
762 	struct cpl_pass_open_rpl *rpl = cplhdr(skb) + RSS_HDR;
763 	unsigned int stid = GET_TID(rpl);
764 	struct listen_ctx *listen_ctx;
765 
766 	listen_ctx = (struct listen_ctx *)lookup_stid(cdev->tids, stid);
767 	if (!listen_ctx)
768 		return CPL_RET_BUF_DONE;
769 
770 	if (listen_ctx->state == T4_LISTEN_START_PENDING) {
771 		listen_ctx->state = T4_LISTEN_STARTED;
772 		return CPL_RET_BUF_DONE;
773 	}
774 
775 	if (rpl->status != CPL_ERR_NONE) {
776 		pr_info("Unexpected PASS_OPEN_RPL status %u for STID %u\n",
777 			rpl->status, stid);
778 	} else {
779 		cxgb4_free_stid(cdev->tids, stid, listen_ctx->lsk->sk_family);
780 		sock_put(listen_ctx->lsk);
781 		kfree(listen_ctx);
782 		module_put(THIS_MODULE);
783 	}
784 	return CPL_RET_BUF_DONE;
785 }
786 
787 static int chtls_close_listsrv_rpl(struct chtls_dev *cdev, struct sk_buff *skb)
788 {
789 	struct cpl_close_listsvr_rpl *rpl = cplhdr(skb) + RSS_HDR;
790 	struct listen_ctx *listen_ctx;
791 	unsigned int stid;
792 	void *data;
793 
794 	stid = GET_TID(rpl);
795 	data = lookup_stid(cdev->tids, stid);
796 	listen_ctx = (struct listen_ctx *)data;
797 
798 	if (rpl->status != CPL_ERR_NONE) {
799 		pr_info("Unexpected CLOSE_LISTSRV_RPL status %u for STID %u\n",
800 			rpl->status, stid);
801 	} else {
802 		cxgb4_free_stid(cdev->tids, stid, listen_ctx->lsk->sk_family);
803 		sock_put(listen_ctx->lsk);
804 		kfree(listen_ctx);
805 		module_put(THIS_MODULE);
806 	}
807 	return CPL_RET_BUF_DONE;
808 }
809 
810 static void chtls_purge_wr_queue(struct sock *sk)
811 {
812 	struct sk_buff *skb;
813 
814 	while ((skb = dequeue_wr(sk)) != NULL)
815 		kfree_skb(skb);
816 }
817 
818 static void chtls_release_resources(struct sock *sk)
819 {
820 	struct chtls_sock *csk = rcu_dereference_sk_user_data(sk);
821 	struct chtls_dev *cdev = csk->cdev;
822 	unsigned int tid = csk->tid;
823 	struct tid_info *tids;
824 
825 	if (!cdev)
826 		return;
827 
828 	tids = cdev->tids;
829 	kfree_skb(csk->txdata_skb_cache);
830 	csk->txdata_skb_cache = NULL;
831 
832 	if (csk->wr_credits != csk->wr_max_credits) {
833 		chtls_purge_wr_queue(sk);
834 		chtls_reset_wr_list(csk);
835 	}
836 
837 	if (csk->l2t_entry) {
838 		cxgb4_l2t_release(csk->l2t_entry);
839 		csk->l2t_entry = NULL;
840 	}
841 
842 	if (sk->sk_state != TCP_SYN_SENT) {
843 		cxgb4_remove_tid(tids, csk->port_id, tid, sk->sk_family);
844 		sock_put(sk);
845 	}
846 }
847 
848 static void chtls_conn_done(struct sock *sk)
849 {
850 	if (sock_flag(sk, SOCK_DEAD))
851 		chtls_purge_receive_queue(sk);
852 	sk_wakeup_sleepers(sk, 0);
853 	tcp_done(sk);
854 }
855 
856 static void do_abort_syn_rcv(struct sock *child, struct sock *parent)
857 {
858 	/*
859 	 * If the server is still open we clean up the child connection,
860 	 * otherwise the server already did the clean up as it was purging
861 	 * its SYN queue and the skb was just sitting in its backlog.
862 	 */
863 	if (likely(parent->sk_state == TCP_LISTEN)) {
864 		cleanup_syn_rcv_conn(child, parent);
865 		/* Without the below call to sock_orphan,
866 		 * we leak the socket resource with syn_flood test
867 		 * as inet_csk_destroy_sock will not be called
868 		 * in tcp_done since SOCK_DEAD flag is not set.
869 		 * Kernel handles this differently where new socket is
870 		 * created only after 3 way handshake is done.
871 		 */
872 		sock_orphan(child);
873 		INC_ORPHAN_COUNT(child);
874 		chtls_release_resources(child);
875 		chtls_conn_done(child);
876 	} else {
877 		if (csk_flag(child, CSK_RST_ABORTED)) {
878 			chtls_release_resources(child);
879 			chtls_conn_done(child);
880 		}
881 	}
882 }
883 
884 static void pass_open_abort(struct sock *child, struct sock *parent,
885 			    struct sk_buff *skb)
886 {
887 	do_abort_syn_rcv(child, parent);
888 	kfree_skb(skb);
889 }
890 
891 static void bl_pass_open_abort(struct sock *lsk, struct sk_buff *skb)
892 {
893 	pass_open_abort(skb->sk, lsk, skb);
894 }
895 
896 static void chtls_pass_open_arp_failure(struct sock *sk,
897 					struct sk_buff *skb)
898 {
899 	const struct request_sock *oreq;
900 	struct chtls_sock *csk;
901 	struct chtls_dev *cdev;
902 	struct sock *parent;
903 	void *data;
904 
905 	csk = rcu_dereference_sk_user_data(sk);
906 	cdev = csk->cdev;
907 
908 	/*
909 	 * If the connection is being aborted due to the parent listening
910 	 * socket going away there's nothing to do, the ABORT_REQ will close
911 	 * the connection.
912 	 */
913 	if (csk_flag(sk, CSK_ABORT_RPL_PENDING)) {
914 		kfree_skb(skb);
915 		return;
916 	}
917 
918 	oreq = csk->passive_reap_next;
919 	data = lookup_stid(cdev->tids, oreq->ts_recent);
920 	parent = ((struct listen_ctx *)data)->lsk;
921 
922 	bh_lock_sock(parent);
923 	if (!sock_owned_by_user(parent)) {
924 		pass_open_abort(sk, parent, skb);
925 	} else {
926 		BLOG_SKB_CB(skb)->backlog_rcv = bl_pass_open_abort;
927 		__sk_add_backlog(parent, skb);
928 	}
929 	bh_unlock_sock(parent);
930 }
931 
932 static void chtls_accept_rpl_arp_failure(void *handle,
933 					 struct sk_buff *skb)
934 {
935 	struct sock *sk = (struct sock *)handle;
936 
937 	sock_hold(sk);
938 	process_cpl_msg(chtls_pass_open_arp_failure, sk, skb);
939 	sock_put(sk);
940 }
941 
942 static unsigned int chtls_select_mss(const struct chtls_sock *csk,
943 				     unsigned int pmtu,
944 				     struct cpl_pass_accept_req *req)
945 {
946 	struct chtls_dev *cdev;
947 	struct dst_entry *dst;
948 	unsigned int tcpoptsz;
949 	unsigned int iphdrsz;
950 	unsigned int mtu_idx;
951 	struct tcp_sock *tp;
952 	unsigned int mss;
953 	struct sock *sk;
954 
955 	mss = ntohs(req->tcpopt.mss);
956 	sk = csk->sk;
957 	dst = __sk_dst_get(sk);
958 	cdev = csk->cdev;
959 	tp = tcp_sk(sk);
960 	tcpoptsz = 0;
961 
962 #if IS_ENABLED(CONFIG_IPV6)
963 	if (sk->sk_family == AF_INET6)
964 		iphdrsz = sizeof(struct ipv6hdr) + sizeof(struct tcphdr);
965 	else
966 #endif
967 		iphdrsz = sizeof(struct iphdr) + sizeof(struct tcphdr);
968 	if (req->tcpopt.tstamp)
969 		tcpoptsz += round_up(TCPOLEN_TIMESTAMP, 4);
970 
971 	tp->advmss = dst_metric_advmss(dst);
972 	if (USER_MSS(tp) && tp->advmss > USER_MSS(tp))
973 		tp->advmss = USER_MSS(tp);
974 	if (tp->advmss > pmtu - iphdrsz)
975 		tp->advmss = pmtu - iphdrsz;
976 	if (mss && tp->advmss > mss)
977 		tp->advmss = mss;
978 
979 	tp->advmss = cxgb4_best_aligned_mtu(cdev->lldi->mtus,
980 					    iphdrsz + tcpoptsz,
981 					    tp->advmss - tcpoptsz,
982 					    8, &mtu_idx);
983 	tp->advmss -= iphdrsz;
984 
985 	inet_csk(sk)->icsk_pmtu_cookie = pmtu;
986 	return mtu_idx;
987 }
988 
989 static unsigned int select_rcv_wscale(int space, int wscale_ok, int win_clamp)
990 {
991 	int wscale = 0;
992 
993 	if (space > MAX_RCV_WND)
994 		space = MAX_RCV_WND;
995 	if (win_clamp && win_clamp < space)
996 		space = win_clamp;
997 
998 	if (wscale_ok) {
999 		while (wscale < 14 && (65535 << wscale) < space)
1000 			wscale++;
1001 	}
1002 	return wscale;
1003 }
1004 
1005 static void chtls_pass_accept_rpl(struct sk_buff *skb,
1006 				  struct cpl_pass_accept_req *req,
1007 				  unsigned int tid)
1008 
1009 {
1010 	struct cpl_t5_pass_accept_rpl *rpl5;
1011 	struct cxgb4_lld_info *lldi;
1012 	const struct tcphdr *tcph;
1013 	const struct tcp_sock *tp;
1014 	struct chtls_sock *csk;
1015 	unsigned int len;
1016 	struct sock *sk;
1017 	u32 opt2, hlen;
1018 	u64 opt0;
1019 
1020 	sk = skb->sk;
1021 	tp = tcp_sk(sk);
1022 	csk = sk->sk_user_data;
1023 	csk->tid = tid;
1024 	lldi = csk->cdev->lldi;
1025 	len = roundup(sizeof(*rpl5), 16);
1026 
1027 	rpl5 = __skb_put_zero(skb, len);
1028 	INIT_TP_WR(rpl5, tid);
1029 
1030 	OPCODE_TID(rpl5) = cpu_to_be32(MK_OPCODE_TID(CPL_PASS_ACCEPT_RPL,
1031 						     csk->tid));
1032 	csk->mtu_idx = chtls_select_mss(csk, dst_mtu(__sk_dst_get(sk)),
1033 					req);
1034 	opt0 = TCAM_BYPASS_F |
1035 	       WND_SCALE_V(RCV_WSCALE(tp)) |
1036 	       MSS_IDX_V(csk->mtu_idx) |
1037 	       L2T_IDX_V(csk->l2t_entry->idx) |
1038 	       NAGLE_V(!(tp->nonagle & TCP_NAGLE_OFF)) |
1039 	       TX_CHAN_V(csk->tx_chan) |
1040 	       SMAC_SEL_V(csk->smac_idx) |
1041 	       DSCP_V(csk->tos >> 2) |
1042 	       ULP_MODE_V(ULP_MODE_TLS) |
1043 	       RCV_BUFSIZ_V(min(tp->rcv_wnd >> 10, RCV_BUFSIZ_M));
1044 
1045 	opt2 = RX_CHANNEL_V(0) |
1046 		RSS_QUEUE_VALID_F | RSS_QUEUE_V(csk->rss_qid);
1047 
1048 	if (!is_t5(lldi->adapter_type))
1049 		opt2 |= RX_FC_DISABLE_F;
1050 	if (req->tcpopt.tstamp)
1051 		opt2 |= TSTAMPS_EN_F;
1052 	if (req->tcpopt.sack)
1053 		opt2 |= SACK_EN_F;
1054 	hlen = ntohl(req->hdr_len);
1055 
1056 	tcph = (struct tcphdr *)((u8 *)(req + 1) +
1057 			T6_ETH_HDR_LEN_G(hlen) + T6_IP_HDR_LEN_G(hlen));
1058 	if (tcph->ece && tcph->cwr)
1059 		opt2 |= CCTRL_ECN_V(1);
1060 	opt2 |= CONG_CNTRL_V(CONG_ALG_NEWRENO);
1061 	opt2 |= T5_ISS_F;
1062 	opt2 |= T5_OPT_2_VALID_F;
1063 	opt2 |= WND_SCALE_EN_V(WSCALE_OK(tp));
1064 	rpl5->opt0 = cpu_to_be64(opt0);
1065 	rpl5->opt2 = cpu_to_be32(opt2);
1066 	rpl5->iss = cpu_to_be32((get_random_u32() & ~7UL) - 1);
1067 	set_wr_txq(skb, CPL_PRIORITY_SETUP, csk->port_id);
1068 	t4_set_arp_err_handler(skb, sk, chtls_accept_rpl_arp_failure);
1069 	cxgb4_l2t_send(csk->egress_dev, skb, csk->l2t_entry);
1070 }
1071 
1072 static void inet_inherit_port(struct sock *lsk, struct sock *newsk)
1073 {
1074 	local_bh_disable();
1075 	__inet_inherit_port(lsk, newsk);
1076 	local_bh_enable();
1077 }
1078 
1079 static int chtls_backlog_rcv(struct sock *sk, struct sk_buff *skb)
1080 {
1081 	if (skb->protocol) {
1082 		kfree_skb(skb);
1083 		return 0;
1084 	}
1085 	BLOG_SKB_CB(skb)->backlog_rcv(sk, skb);
1086 	return 0;
1087 }
1088 
1089 static void chtls_set_tcp_window(struct chtls_sock *csk)
1090 {
1091 	struct net_device *ndev = csk->egress_dev;
1092 	struct port_info *pi = netdev_priv(ndev);
1093 	unsigned int linkspeed;
1094 	u8 scale;
1095 
1096 	linkspeed = pi->link_cfg.speed;
1097 	scale = linkspeed / SPEED_10000;
1098 #define CHTLS_10G_RCVWIN (256 * 1024)
1099 	csk->rcv_win = CHTLS_10G_RCVWIN;
1100 	if (scale)
1101 		csk->rcv_win *= scale;
1102 #define CHTLS_10G_SNDWIN (256 * 1024)
1103 	csk->snd_win = CHTLS_10G_SNDWIN;
1104 	if (scale)
1105 		csk->snd_win *= scale;
1106 }
1107 
1108 static struct sock *chtls_recv_sock(struct sock *lsk,
1109 				    struct request_sock *oreq,
1110 				    void *network_hdr,
1111 				    const struct cpl_pass_accept_req *req,
1112 				    struct chtls_dev *cdev)
1113 {
1114 	struct adapter *adap = pci_get_drvdata(cdev->pdev);
1115 	struct neighbour *n = NULL;
1116 	struct inet_sock *newinet;
1117 	const struct iphdr *iph;
1118 	struct tls_context *ctx;
1119 	struct net_device *ndev;
1120 	struct chtls_sock *csk;
1121 	struct dst_entry *dst;
1122 	struct tcp_sock *tp;
1123 	struct sock *newsk;
1124 	bool found = false;
1125 	u16 port_id;
1126 	int rxq_idx;
1127 	int step, i;
1128 
1129 	iph = (const struct iphdr *)network_hdr;
1130 	newsk = tcp_create_openreq_child(lsk, oreq, cdev->askb);
1131 	if (!newsk)
1132 		goto free_oreq;
1133 
1134 	if (lsk->sk_family == AF_INET) {
1135 		dst = inet_csk_route_child_sock(lsk, newsk, oreq);
1136 		if (!dst)
1137 			goto free_sk;
1138 
1139 		n = dst_neigh_lookup(dst, &iph->saddr);
1140 #if IS_ENABLED(CONFIG_IPV6)
1141 	} else {
1142 		const struct ipv6hdr *ip6h;
1143 		struct flowi6 fl6;
1144 
1145 		ip6h = (const struct ipv6hdr *)network_hdr;
1146 		memset(&fl6, 0, sizeof(fl6));
1147 		fl6.flowi6_proto = IPPROTO_TCP;
1148 		fl6.saddr = ip6h->daddr;
1149 		fl6.daddr = ip6h->saddr;
1150 		fl6.fl6_dport = inet_rsk(oreq)->ir_rmt_port;
1151 		fl6.fl6_sport = htons(inet_rsk(oreq)->ir_num);
1152 		security_req_classify_flow(oreq, flowi6_to_flowi_common(&fl6));
1153 		dst = ip6_dst_lookup_flow(sock_net(lsk), lsk, &fl6, NULL);
1154 		if (IS_ERR(dst))
1155 			goto free_sk;
1156 		n = dst_neigh_lookup(dst, &ip6h->saddr);
1157 #endif
1158 	}
1159 	if (!n || !n->dev)
1160 		goto free_dst;
1161 
1162 	ndev = n->dev;
1163 	if (is_vlan_dev(ndev))
1164 		ndev = vlan_dev_real_dev(ndev);
1165 
1166 	for_each_port(adap, i)
1167 		if (cdev->ports[i] == ndev)
1168 			found = true;
1169 
1170 	if (!found)
1171 		goto free_dst;
1172 
1173 	port_id = cxgb4_port_idx(ndev);
1174 
1175 	csk = chtls_sock_create(cdev);
1176 	if (!csk)
1177 		goto free_dst;
1178 
1179 	csk->l2t_entry = cxgb4_l2t_get(cdev->lldi->l2t, n, ndev, 0);
1180 	if (!csk->l2t_entry)
1181 		goto free_csk;
1182 
1183 	newsk->sk_user_data = csk;
1184 	newsk->sk_backlog_rcv = chtls_backlog_rcv;
1185 
1186 	tp = tcp_sk(newsk);
1187 	newinet = inet_sk(newsk);
1188 
1189 	if (iph->version == 0x4) {
1190 		newinet->inet_daddr = iph->saddr;
1191 		newinet->inet_rcv_saddr = iph->daddr;
1192 		newinet->inet_saddr = iph->daddr;
1193 #if IS_ENABLED(CONFIG_IPV6)
1194 	} else {
1195 		struct tcp6_sock *newtcp6sk = (struct tcp6_sock *)newsk;
1196 		struct inet_request_sock *treq = inet_rsk(oreq);
1197 		struct ipv6_pinfo *newnp = inet6_sk(newsk);
1198 		struct ipv6_pinfo *np = inet6_sk(lsk);
1199 
1200 		inet_sk(newsk)->pinet6 = &newtcp6sk->inet6;
1201 		memcpy(newnp, np, sizeof(struct ipv6_pinfo));
1202 		newsk->sk_v6_daddr = treq->ir_v6_rmt_addr;
1203 		newsk->sk_v6_rcv_saddr = treq->ir_v6_loc_addr;
1204 		inet6_sk(newsk)->saddr = treq->ir_v6_loc_addr;
1205 		newnp->ipv6_fl_list = NULL;
1206 		newnp->pktoptions = NULL;
1207 		newsk->sk_bound_dev_if = treq->ir_iif;
1208 		newinet->inet_opt = NULL;
1209 		newinet->inet_daddr = LOOPBACK4_IPV6;
1210 		newinet->inet_saddr = LOOPBACK4_IPV6;
1211 #endif
1212 	}
1213 
1214 	oreq->ts_recent = PASS_OPEN_TID_G(ntohl(req->tos_stid));
1215 	sk_setup_caps(newsk, dst);
1216 	ctx = tls_get_ctx(lsk);
1217 	newsk->sk_destruct = ctx->sk_destruct;
1218 	newsk->sk_prot_creator = lsk->sk_prot_creator;
1219 	csk->sk = newsk;
1220 	csk->passive_reap_next = oreq;
1221 	csk->tx_chan = cxgb4_port_chan(ndev);
1222 	csk->port_id = port_id;
1223 	csk->egress_dev = ndev;
1224 	csk->tos = PASS_OPEN_TOS_G(ntohl(req->tos_stid));
1225 	chtls_set_tcp_window(csk);
1226 	tp->rcv_wnd = csk->rcv_win;
1227 	csk->sndbuf = csk->snd_win;
1228 	csk->ulp_mode = ULP_MODE_TLS;
1229 	step = cdev->lldi->nrxq / cdev->lldi->nchan;
1230 	rxq_idx = port_id * step;
1231 	rxq_idx += cdev->round_robin_cnt++ % step;
1232 	csk->rss_qid = cdev->lldi->rxq_ids[rxq_idx];
1233 	csk->txq_idx = (rxq_idx < cdev->lldi->ntxq) ? rxq_idx :
1234 			port_id * step;
1235 	csk->sndbuf = newsk->sk_sndbuf;
1236 	csk->smac_idx = ((struct port_info *)netdev_priv(ndev))->smt_idx;
1237 	RCV_WSCALE(tp) = select_rcv_wscale(tcp_full_space(newsk),
1238 					   READ_ONCE(sock_net(newsk)->
1239 						     ipv4.sysctl_tcp_window_scaling),
1240 					   tp->window_clamp);
1241 	neigh_release(n);
1242 	inet_inherit_port(lsk, newsk);
1243 	csk_set_flag(csk, CSK_CONN_INLINE);
1244 	bh_unlock_sock(newsk); /* tcp_create_openreq_child ->sk_clone_lock */
1245 
1246 	return newsk;
1247 free_csk:
1248 	chtls_sock_release(&csk->kref);
1249 free_dst:
1250 	if (n)
1251 		neigh_release(n);
1252 	dst_release(dst);
1253 free_sk:
1254 	inet_csk_prepare_forced_close(newsk);
1255 	tcp_done(newsk);
1256 free_oreq:
1257 	chtls_reqsk_free(oreq);
1258 	return NULL;
1259 }
1260 
1261 /*
1262  * Populate a TID_RELEASE WR.  The skb must be already propely sized.
1263  */
1264 static  void mk_tid_release(struct sk_buff *skb,
1265 			    unsigned int chan, unsigned int tid)
1266 {
1267 	struct cpl_tid_release *req;
1268 	unsigned int len;
1269 
1270 	len = roundup(sizeof(struct cpl_tid_release), 16);
1271 	req = (struct cpl_tid_release *)__skb_put(skb, len);
1272 	memset(req, 0, len);
1273 	set_wr_txq(skb, CPL_PRIORITY_SETUP, chan);
1274 	INIT_TP_WR_CPL(req, CPL_TID_RELEASE, tid);
1275 }
1276 
1277 static int chtls_get_module(struct sock *sk)
1278 {
1279 	struct inet_connection_sock *icsk = inet_csk(sk);
1280 
1281 	if (!try_module_get(icsk->icsk_ulp_ops->owner))
1282 		return -1;
1283 
1284 	return 0;
1285 }
1286 
1287 static void chtls_pass_accept_request(struct sock *sk,
1288 				      struct sk_buff *skb)
1289 {
1290 	struct cpl_t5_pass_accept_rpl *rpl;
1291 	struct cpl_pass_accept_req *req;
1292 	struct listen_ctx *listen_ctx;
1293 	struct vlan_ethhdr *vlan_eh;
1294 	struct request_sock *oreq;
1295 	struct sk_buff *reply_skb;
1296 	struct chtls_sock *csk;
1297 	struct chtls_dev *cdev;
1298 	struct ipv6hdr *ip6h;
1299 	struct tcphdr *tcph;
1300 	struct sock *newsk;
1301 	struct ethhdr *eh;
1302 	struct iphdr *iph;
1303 	void *network_hdr;
1304 	unsigned int stid;
1305 	unsigned int len;
1306 	unsigned int tid;
1307 	bool th_ecn, ect;
1308 	__u8 ip_dsfield; /* IPv4 tos or IPv6 dsfield */
1309 	u16 eth_hdr_len;
1310 	bool ecn_ok;
1311 
1312 	req = cplhdr(skb) + RSS_HDR;
1313 	tid = GET_TID(req);
1314 	cdev = BLOG_SKB_CB(skb)->cdev;
1315 	newsk = lookup_tid(cdev->tids, tid);
1316 	stid = PASS_OPEN_TID_G(ntohl(req->tos_stid));
1317 	if (newsk) {
1318 		pr_info("tid (%d) already in use\n", tid);
1319 		return;
1320 	}
1321 
1322 	len = roundup(sizeof(*rpl), 16);
1323 	reply_skb = alloc_skb(len, GFP_ATOMIC);
1324 	if (!reply_skb) {
1325 		cxgb4_remove_tid(cdev->tids, 0, tid, sk->sk_family);
1326 		kfree_skb(skb);
1327 		return;
1328 	}
1329 
1330 	if (sk->sk_state != TCP_LISTEN)
1331 		goto reject;
1332 
1333 	if (inet_csk_reqsk_queue_is_full(sk))
1334 		goto reject;
1335 
1336 	if (sk_acceptq_is_full(sk))
1337 		goto reject;
1338 
1339 
1340 	eth_hdr_len = T6_ETH_HDR_LEN_G(ntohl(req->hdr_len));
1341 	if (eth_hdr_len == ETH_HLEN) {
1342 		eh = (struct ethhdr *)(req + 1);
1343 		iph = (struct iphdr *)(eh + 1);
1344 		ip6h = (struct ipv6hdr *)(eh + 1);
1345 		network_hdr = (void *)(eh + 1);
1346 	} else {
1347 		vlan_eh = (struct vlan_ethhdr *)(req + 1);
1348 		iph = (struct iphdr *)(vlan_eh + 1);
1349 		ip6h = (struct ipv6hdr *)(vlan_eh + 1);
1350 		network_hdr = (void *)(vlan_eh + 1);
1351 	}
1352 
1353 	if (iph->version == 0x4) {
1354 		tcph = (struct tcphdr *)(iph + 1);
1355 		skb_set_network_header(skb, (void *)iph - (void *)req);
1356 		oreq = inet_reqsk_alloc(&chtls_rsk_ops, sk, true);
1357 	} else {
1358 		tcph = (struct tcphdr *)(ip6h + 1);
1359 		skb_set_network_header(skb, (void *)ip6h - (void *)req);
1360 		oreq = inet_reqsk_alloc(&chtls_rsk_opsv6, sk, false);
1361 	}
1362 
1363 	if (!oreq)
1364 		goto reject;
1365 
1366 	oreq->rsk_rcv_wnd = 0;
1367 	oreq->rsk_window_clamp = 0;
1368 	oreq->syncookie = 0;
1369 	oreq->mss = 0;
1370 	oreq->ts_recent = 0;
1371 
1372 	tcp_rsk(oreq)->tfo_listener = false;
1373 	tcp_rsk(oreq)->rcv_isn = ntohl(tcph->seq);
1374 	chtls_set_req_port(oreq, tcph->source, tcph->dest);
1375 	if (iph->version == 0x4) {
1376 		chtls_set_req_addr(oreq, iph->daddr, iph->saddr);
1377 		ip_dsfield = ipv4_get_dsfield(iph);
1378 #if IS_ENABLED(CONFIG_IPV6)
1379 	} else {
1380 		inet_rsk(oreq)->ir_v6_rmt_addr = ipv6_hdr(skb)->saddr;
1381 		inet_rsk(oreq)->ir_v6_loc_addr = ipv6_hdr(skb)->daddr;
1382 		ip_dsfield = ipv6_get_dsfield(ipv6_hdr(skb));
1383 #endif
1384 	}
1385 	if (req->tcpopt.wsf <= 14 &&
1386 	    READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_window_scaling)) {
1387 		inet_rsk(oreq)->wscale_ok = 1;
1388 		inet_rsk(oreq)->snd_wscale = req->tcpopt.wsf;
1389 	}
1390 	inet_rsk(oreq)->ir_iif = sk->sk_bound_dev_if;
1391 	th_ecn = tcph->ece && tcph->cwr;
1392 	if (th_ecn) {
1393 		ect = !INET_ECN_is_not_ect(ip_dsfield);
1394 		ecn_ok = READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_ecn);
1395 		if ((!ect && ecn_ok) || tcp_ca_needs_ecn(sk))
1396 			inet_rsk(oreq)->ecn_ok = 1;
1397 	}
1398 
1399 	newsk = chtls_recv_sock(sk, oreq, network_hdr, req, cdev);
1400 	if (!newsk)
1401 		goto reject;
1402 
1403 	if (chtls_get_module(newsk))
1404 		goto reject;
1405 	inet_csk_reqsk_queue_added(sk);
1406 	reply_skb->sk = newsk;
1407 	chtls_install_cpl_ops(newsk);
1408 	cxgb4_insert_tid(cdev->tids, newsk, tid, newsk->sk_family);
1409 	csk = rcu_dereference_sk_user_data(newsk);
1410 	listen_ctx = (struct listen_ctx *)lookup_stid(cdev->tids, stid);
1411 	csk->listen_ctx = listen_ctx;
1412 	__skb_queue_tail(&listen_ctx->synq, (struct sk_buff *)&csk->synq);
1413 	chtls_pass_accept_rpl(reply_skb, req, tid);
1414 	kfree_skb(skb);
1415 	return;
1416 
1417 reject:
1418 	mk_tid_release(reply_skb, 0, tid);
1419 	cxgb4_ofld_send(cdev->lldi->ports[0], reply_skb);
1420 	kfree_skb(skb);
1421 }
1422 
1423 /*
1424  * Handle a CPL_PASS_ACCEPT_REQ message.
1425  */
1426 static int chtls_pass_accept_req(struct chtls_dev *cdev, struct sk_buff *skb)
1427 {
1428 	struct cpl_pass_accept_req *req = cplhdr(skb) + RSS_HDR;
1429 	struct listen_ctx *ctx;
1430 	unsigned int stid;
1431 	unsigned int tid;
1432 	struct sock *lsk;
1433 	void *data;
1434 
1435 	stid = PASS_OPEN_TID_G(ntohl(req->tos_stid));
1436 	tid = GET_TID(req);
1437 
1438 	data = lookup_stid(cdev->tids, stid);
1439 	if (!data)
1440 		return 1;
1441 
1442 	ctx = (struct listen_ctx *)data;
1443 	lsk = ctx->lsk;
1444 
1445 	if (unlikely(tid_out_of_range(cdev->tids, tid))) {
1446 		pr_info("passive open TID %u too large\n", tid);
1447 		return 1;
1448 	}
1449 
1450 	BLOG_SKB_CB(skb)->cdev = cdev;
1451 	process_cpl_msg(chtls_pass_accept_request, lsk, skb);
1452 	return 0;
1453 }
1454 
1455 /*
1456  * Completes some final bits of initialization for just established connections
1457  * and changes their state to TCP_ESTABLISHED.
1458  *
1459  * snd_isn here is the ISN after the SYN, i.e., the true ISN + 1.
1460  */
1461 static void make_established(struct sock *sk, u32 snd_isn, unsigned int opt)
1462 {
1463 	struct tcp_sock *tp = tcp_sk(sk);
1464 
1465 	tp->pushed_seq = snd_isn;
1466 	tp->write_seq = snd_isn;
1467 	tp->snd_nxt = snd_isn;
1468 	tp->snd_una = snd_isn;
1469 	inet_sk(sk)->inet_id = get_random_u16();
1470 	assign_rxopt(sk, opt);
1471 
1472 	if (tp->rcv_wnd > (RCV_BUFSIZ_M << 10))
1473 		tp->rcv_wup -= tp->rcv_wnd - (RCV_BUFSIZ_M << 10);
1474 
1475 	smp_mb();
1476 	tcp_set_state(sk, TCP_ESTABLISHED);
1477 }
1478 
1479 static void chtls_abort_conn(struct sock *sk, struct sk_buff *skb)
1480 {
1481 	struct sk_buff *abort_skb;
1482 
1483 	abort_skb = alloc_skb(sizeof(struct cpl_abort_req), GFP_ATOMIC);
1484 	if (abort_skb)
1485 		chtls_send_reset(sk, CPL_ABORT_SEND_RST, abort_skb);
1486 }
1487 
1488 static struct sock *reap_list;
1489 static DEFINE_SPINLOCK(reap_list_lock);
1490 
1491 /*
1492  * Process the reap list.
1493  */
1494 DECLARE_TASK_FUNC(process_reap_list, task_param)
1495 {
1496 	spin_lock_bh(&reap_list_lock);
1497 	while (reap_list) {
1498 		struct sock *sk = reap_list;
1499 		struct chtls_sock *csk = rcu_dereference_sk_user_data(sk);
1500 
1501 		reap_list = csk->passive_reap_next;
1502 		csk->passive_reap_next = NULL;
1503 		spin_unlock(&reap_list_lock);
1504 		sock_hold(sk);
1505 
1506 		bh_lock_sock(sk);
1507 		chtls_abort_conn(sk, NULL);
1508 		sock_orphan(sk);
1509 		if (sk->sk_state == TCP_CLOSE)
1510 			inet_csk_destroy_sock(sk);
1511 		bh_unlock_sock(sk);
1512 		sock_put(sk);
1513 		spin_lock(&reap_list_lock);
1514 	}
1515 	spin_unlock_bh(&reap_list_lock);
1516 }
1517 
1518 static DECLARE_WORK(reap_task, process_reap_list);
1519 
1520 static void add_to_reap_list(struct sock *sk)
1521 {
1522 	struct chtls_sock *csk = sk->sk_user_data;
1523 
1524 	local_bh_disable();
1525 	release_tcp_port(sk); /* release the port immediately */
1526 
1527 	spin_lock(&reap_list_lock);
1528 	csk->passive_reap_next = reap_list;
1529 	reap_list = sk;
1530 	if (!csk->passive_reap_next)
1531 		schedule_work(&reap_task);
1532 	spin_unlock(&reap_list_lock);
1533 	local_bh_enable();
1534 }
1535 
1536 static void add_pass_open_to_parent(struct sock *child, struct sock *lsk,
1537 				    struct chtls_dev *cdev)
1538 {
1539 	struct request_sock *oreq;
1540 	struct chtls_sock *csk;
1541 
1542 	if (lsk->sk_state != TCP_LISTEN)
1543 		return;
1544 
1545 	csk = child->sk_user_data;
1546 	oreq = csk->passive_reap_next;
1547 	csk->passive_reap_next = NULL;
1548 
1549 	reqsk_queue_removed(&inet_csk(lsk)->icsk_accept_queue, oreq);
1550 	__skb_unlink((struct sk_buff *)&csk->synq, &csk->listen_ctx->synq);
1551 
1552 	if (sk_acceptq_is_full(lsk)) {
1553 		chtls_reqsk_free(oreq);
1554 		add_to_reap_list(child);
1555 	} else {
1556 		refcount_set(&oreq->rsk_refcnt, 1);
1557 		inet_csk_reqsk_queue_add(lsk, oreq, child);
1558 		lsk->sk_data_ready(lsk);
1559 	}
1560 }
1561 
1562 static void bl_add_pass_open_to_parent(struct sock *lsk, struct sk_buff *skb)
1563 {
1564 	struct sock *child = skb->sk;
1565 
1566 	skb->sk = NULL;
1567 	add_pass_open_to_parent(child, lsk, BLOG_SKB_CB(skb)->cdev);
1568 	kfree_skb(skb);
1569 }
1570 
1571 static int chtls_pass_establish(struct chtls_dev *cdev, struct sk_buff *skb)
1572 {
1573 	struct cpl_pass_establish *req = cplhdr(skb) + RSS_HDR;
1574 	struct chtls_sock *csk;
1575 	struct sock *lsk, *sk;
1576 	unsigned int hwtid;
1577 
1578 	hwtid = GET_TID(req);
1579 	sk = lookup_tid(cdev->tids, hwtid);
1580 	if (!sk)
1581 		return (CPL_RET_UNKNOWN_TID | CPL_RET_BUF_DONE);
1582 
1583 	bh_lock_sock(sk);
1584 	if (unlikely(sock_owned_by_user(sk))) {
1585 		kfree_skb(skb);
1586 	} else {
1587 		unsigned int stid;
1588 		void *data;
1589 
1590 		csk = sk->sk_user_data;
1591 		csk->wr_max_credits = 64;
1592 		csk->wr_credits = 64;
1593 		csk->wr_unacked = 0;
1594 		make_established(sk, ntohl(req->snd_isn), ntohs(req->tcp_opt));
1595 		stid = PASS_OPEN_TID_G(ntohl(req->tos_stid));
1596 		sk->sk_state_change(sk);
1597 		if (unlikely(sk->sk_socket))
1598 			sk_wake_async(sk, 0, POLL_OUT);
1599 
1600 		data = lookup_stid(cdev->tids, stid);
1601 		if (!data) {
1602 			/* listening server close */
1603 			kfree_skb(skb);
1604 			goto unlock;
1605 		}
1606 		lsk = ((struct listen_ctx *)data)->lsk;
1607 
1608 		bh_lock_sock(lsk);
1609 		if (unlikely(skb_queue_empty(&csk->listen_ctx->synq))) {
1610 			/* removed from synq */
1611 			bh_unlock_sock(lsk);
1612 			kfree_skb(skb);
1613 			goto unlock;
1614 		}
1615 
1616 		if (likely(!sock_owned_by_user(lsk))) {
1617 			kfree_skb(skb);
1618 			add_pass_open_to_parent(sk, lsk, cdev);
1619 		} else {
1620 			skb->sk = sk;
1621 			BLOG_SKB_CB(skb)->cdev = cdev;
1622 			BLOG_SKB_CB(skb)->backlog_rcv =
1623 				bl_add_pass_open_to_parent;
1624 			__sk_add_backlog(lsk, skb);
1625 		}
1626 		bh_unlock_sock(lsk);
1627 	}
1628 unlock:
1629 	bh_unlock_sock(sk);
1630 	return 0;
1631 }
1632 
1633 /*
1634  * Handle receipt of an urgent pointer.
1635  */
1636 static void handle_urg_ptr(struct sock *sk, u32 urg_seq)
1637 {
1638 	struct tcp_sock *tp = tcp_sk(sk);
1639 
1640 	urg_seq--;
1641 	if (tp->urg_data && !after(urg_seq, tp->urg_seq))
1642 		return;	/* duplicate pointer */
1643 
1644 	sk_send_sigurg(sk);
1645 	if (tp->urg_seq == tp->copied_seq && tp->urg_data &&
1646 	    !sock_flag(sk, SOCK_URGINLINE) &&
1647 	    tp->copied_seq != tp->rcv_nxt) {
1648 		struct sk_buff *skb = skb_peek(&sk->sk_receive_queue);
1649 
1650 		tp->copied_seq++;
1651 		if (skb && tp->copied_seq - ULP_SKB_CB(skb)->seq >= skb->len)
1652 			chtls_free_skb(sk, skb);
1653 	}
1654 
1655 	tp->urg_data = TCP_URG_NOTYET;
1656 	tp->urg_seq = urg_seq;
1657 }
1658 
1659 static void check_sk_callbacks(struct chtls_sock *csk)
1660 {
1661 	struct sock *sk = csk->sk;
1662 
1663 	if (unlikely(sk->sk_user_data &&
1664 		     !csk_flag_nochk(csk, CSK_CALLBACKS_CHKD)))
1665 		csk_set_flag(csk, CSK_CALLBACKS_CHKD);
1666 }
1667 
1668 /*
1669  * Handles Rx data that arrives in a state where the socket isn't accepting
1670  * new data.
1671  */
1672 static void handle_excess_rx(struct sock *sk, struct sk_buff *skb)
1673 {
1674 	if (!csk_flag(sk, CSK_ABORT_SHUTDOWN))
1675 		chtls_abort_conn(sk, skb);
1676 
1677 	kfree_skb(skb);
1678 }
1679 
1680 static void chtls_recv_data(struct sock *sk, struct sk_buff *skb)
1681 {
1682 	struct cpl_rx_data *hdr = cplhdr(skb) + RSS_HDR;
1683 	struct chtls_sock *csk;
1684 	struct tcp_sock *tp;
1685 
1686 	csk = rcu_dereference_sk_user_data(sk);
1687 	tp = tcp_sk(sk);
1688 
1689 	if (unlikely(sk->sk_shutdown & RCV_SHUTDOWN)) {
1690 		handle_excess_rx(sk, skb);
1691 		return;
1692 	}
1693 
1694 	ULP_SKB_CB(skb)->seq = ntohl(hdr->seq);
1695 	ULP_SKB_CB(skb)->psh = hdr->psh;
1696 	skb_ulp_mode(skb) = ULP_MODE_NONE;
1697 
1698 	skb_reset_transport_header(skb);
1699 	__skb_pull(skb, sizeof(*hdr) + RSS_HDR);
1700 	if (!skb->data_len)
1701 		__skb_trim(skb, ntohs(hdr->len));
1702 
1703 	if (unlikely(hdr->urg))
1704 		handle_urg_ptr(sk, tp->rcv_nxt + ntohs(hdr->urg));
1705 	if (unlikely(tp->urg_data == TCP_URG_NOTYET &&
1706 		     tp->urg_seq - tp->rcv_nxt < skb->len))
1707 		tp->urg_data = TCP_URG_VALID |
1708 			       skb->data[tp->urg_seq - tp->rcv_nxt];
1709 
1710 	if (unlikely(hdr->dack_mode != csk->delack_mode)) {
1711 		csk->delack_mode = hdr->dack_mode;
1712 		csk->delack_seq = tp->rcv_nxt;
1713 	}
1714 
1715 	tcp_hdr(skb)->fin = 0;
1716 	tp->rcv_nxt += skb->len;
1717 
1718 	__skb_queue_tail(&sk->sk_receive_queue, skb);
1719 
1720 	if (!sock_flag(sk, SOCK_DEAD)) {
1721 		check_sk_callbacks(csk);
1722 		sk->sk_data_ready(sk);
1723 	}
1724 }
1725 
1726 static int chtls_rx_data(struct chtls_dev *cdev, struct sk_buff *skb)
1727 {
1728 	struct cpl_rx_data *req = cplhdr(skb) + RSS_HDR;
1729 	unsigned int hwtid = GET_TID(req);
1730 	struct sock *sk;
1731 
1732 	sk = lookup_tid(cdev->tids, hwtid);
1733 	if (unlikely(!sk)) {
1734 		pr_err("can't find conn. for hwtid %u.\n", hwtid);
1735 		return -EINVAL;
1736 	}
1737 	skb_dst_set(skb, NULL);
1738 	process_cpl_msg(chtls_recv_data, sk, skb);
1739 	return 0;
1740 }
1741 
1742 static void chtls_recv_pdu(struct sock *sk, struct sk_buff *skb)
1743 {
1744 	struct cpl_tls_data *hdr = cplhdr(skb);
1745 	struct chtls_sock *csk;
1746 	struct chtls_hws *tlsk;
1747 	struct tcp_sock *tp;
1748 
1749 	csk = rcu_dereference_sk_user_data(sk);
1750 	tlsk = &csk->tlshws;
1751 	tp = tcp_sk(sk);
1752 
1753 	if (unlikely(sk->sk_shutdown & RCV_SHUTDOWN)) {
1754 		handle_excess_rx(sk, skb);
1755 		return;
1756 	}
1757 
1758 	ULP_SKB_CB(skb)->seq = ntohl(hdr->seq);
1759 	ULP_SKB_CB(skb)->flags = 0;
1760 	skb_ulp_mode(skb) = ULP_MODE_TLS;
1761 
1762 	skb_reset_transport_header(skb);
1763 	__skb_pull(skb, sizeof(*hdr));
1764 	if (!skb->data_len)
1765 		__skb_trim(skb,
1766 			   CPL_TLS_DATA_LENGTH_G(ntohl(hdr->length_pkd)));
1767 
1768 	if (unlikely(tp->urg_data == TCP_URG_NOTYET && tp->urg_seq -
1769 		     tp->rcv_nxt < skb->len))
1770 		tp->urg_data = TCP_URG_VALID |
1771 			       skb->data[tp->urg_seq - tp->rcv_nxt];
1772 
1773 	tcp_hdr(skb)->fin = 0;
1774 	tlsk->pldlen = CPL_TLS_DATA_LENGTH_G(ntohl(hdr->length_pkd));
1775 	__skb_queue_tail(&tlsk->sk_recv_queue, skb);
1776 }
1777 
1778 static int chtls_rx_pdu(struct chtls_dev *cdev, struct sk_buff *skb)
1779 {
1780 	struct cpl_tls_data *req = cplhdr(skb);
1781 	unsigned int hwtid = GET_TID(req);
1782 	struct sock *sk;
1783 
1784 	sk = lookup_tid(cdev->tids, hwtid);
1785 	if (unlikely(!sk)) {
1786 		pr_err("can't find conn. for hwtid %u.\n", hwtid);
1787 		return -EINVAL;
1788 	}
1789 	skb_dst_set(skb, NULL);
1790 	process_cpl_msg(chtls_recv_pdu, sk, skb);
1791 	return 0;
1792 }
1793 
1794 static void chtls_set_hdrlen(struct sk_buff *skb, unsigned int nlen)
1795 {
1796 	struct tlsrx_cmp_hdr *tls_cmp_hdr = cplhdr(skb);
1797 
1798 	skb->hdr_len = ntohs((__force __be16)tls_cmp_hdr->length);
1799 	tls_cmp_hdr->length = ntohs((__force __be16)nlen);
1800 }
1801 
1802 static void chtls_rx_hdr(struct sock *sk, struct sk_buff *skb)
1803 {
1804 	struct tlsrx_cmp_hdr *tls_hdr_pkt;
1805 	struct cpl_rx_tls_cmp *cmp_cpl;
1806 	struct sk_buff *skb_rec;
1807 	struct chtls_sock *csk;
1808 	struct chtls_hws *tlsk;
1809 	struct tcp_sock *tp;
1810 
1811 	cmp_cpl = cplhdr(skb);
1812 	csk = rcu_dereference_sk_user_data(sk);
1813 	tlsk = &csk->tlshws;
1814 	tp = tcp_sk(sk);
1815 
1816 	ULP_SKB_CB(skb)->seq = ntohl(cmp_cpl->seq);
1817 	ULP_SKB_CB(skb)->flags = 0;
1818 
1819 	skb_reset_transport_header(skb);
1820 	__skb_pull(skb, sizeof(*cmp_cpl));
1821 	tls_hdr_pkt = (struct tlsrx_cmp_hdr *)skb->data;
1822 	if (tls_hdr_pkt->res_to_mac_error & TLSRX_HDR_PKT_ERROR_M)
1823 		tls_hdr_pkt->type = CONTENT_TYPE_ERROR;
1824 	if (!skb->data_len)
1825 		__skb_trim(skb, TLS_HEADER_LENGTH);
1826 
1827 	tp->rcv_nxt +=
1828 		CPL_RX_TLS_CMP_PDULENGTH_G(ntohl(cmp_cpl->pdulength_length));
1829 
1830 	ULP_SKB_CB(skb)->flags |= ULPCB_FLAG_TLS_HDR;
1831 	skb_rec = __skb_dequeue(&tlsk->sk_recv_queue);
1832 	if (!skb_rec) {
1833 		__skb_queue_tail(&sk->sk_receive_queue, skb);
1834 	} else {
1835 		chtls_set_hdrlen(skb, tlsk->pldlen);
1836 		tlsk->pldlen = 0;
1837 		__skb_queue_tail(&sk->sk_receive_queue, skb);
1838 		__skb_queue_tail(&sk->sk_receive_queue, skb_rec);
1839 	}
1840 
1841 	if (!sock_flag(sk, SOCK_DEAD)) {
1842 		check_sk_callbacks(csk);
1843 		sk->sk_data_ready(sk);
1844 	}
1845 }
1846 
1847 static int chtls_rx_cmp(struct chtls_dev *cdev, struct sk_buff *skb)
1848 {
1849 	struct cpl_rx_tls_cmp *req = cplhdr(skb);
1850 	unsigned int hwtid = GET_TID(req);
1851 	struct sock *sk;
1852 
1853 	sk = lookup_tid(cdev->tids, hwtid);
1854 	if (unlikely(!sk)) {
1855 		pr_err("can't find conn. for hwtid %u.\n", hwtid);
1856 		return -EINVAL;
1857 	}
1858 	skb_dst_set(skb, NULL);
1859 	process_cpl_msg(chtls_rx_hdr, sk, skb);
1860 
1861 	return 0;
1862 }
1863 
1864 static void chtls_timewait(struct sock *sk)
1865 {
1866 	struct tcp_sock *tp = tcp_sk(sk);
1867 
1868 	tp->rcv_nxt++;
1869 	tp->rx_opt.ts_recent_stamp = ktime_get_seconds();
1870 	tp->srtt_us = 0;
1871 	tcp_time_wait(sk, TCP_TIME_WAIT, 0);
1872 }
1873 
1874 static void chtls_peer_close(struct sock *sk, struct sk_buff *skb)
1875 {
1876 	struct chtls_sock *csk = rcu_dereference_sk_user_data(sk);
1877 
1878 	if (csk_flag_nochk(csk, CSK_ABORT_RPL_PENDING))
1879 		goto out;
1880 
1881 	sk->sk_shutdown |= RCV_SHUTDOWN;
1882 	sock_set_flag(sk, SOCK_DONE);
1883 
1884 	switch (sk->sk_state) {
1885 	case TCP_SYN_RECV:
1886 	case TCP_ESTABLISHED:
1887 		tcp_set_state(sk, TCP_CLOSE_WAIT);
1888 		break;
1889 	case TCP_FIN_WAIT1:
1890 		tcp_set_state(sk, TCP_CLOSING);
1891 		break;
1892 	case TCP_FIN_WAIT2:
1893 		chtls_release_resources(sk);
1894 		if (csk_flag_nochk(csk, CSK_ABORT_RPL_PENDING))
1895 			chtls_conn_done(sk);
1896 		else
1897 			chtls_timewait(sk);
1898 		break;
1899 	default:
1900 		pr_info("cpl_peer_close in bad state %d\n", sk->sk_state);
1901 	}
1902 
1903 	if (!sock_flag(sk, SOCK_DEAD)) {
1904 		sk->sk_state_change(sk);
1905 		/* Do not send POLL_HUP for half duplex close. */
1906 
1907 		if ((sk->sk_shutdown & SEND_SHUTDOWN) ||
1908 		    sk->sk_state == TCP_CLOSE)
1909 			sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_HUP);
1910 		else
1911 			sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN);
1912 	}
1913 out:
1914 	kfree_skb(skb);
1915 }
1916 
1917 static void chtls_close_con_rpl(struct sock *sk, struct sk_buff *skb)
1918 {
1919 	struct cpl_close_con_rpl *rpl = cplhdr(skb) + RSS_HDR;
1920 	struct chtls_sock *csk;
1921 	struct tcp_sock *tp;
1922 
1923 	csk = rcu_dereference_sk_user_data(sk);
1924 
1925 	if (csk_flag_nochk(csk, CSK_ABORT_RPL_PENDING))
1926 		goto out;
1927 
1928 	tp = tcp_sk(sk);
1929 
1930 	tp->snd_una = ntohl(rpl->snd_nxt) - 1;  /* exclude FIN */
1931 
1932 	switch (sk->sk_state) {
1933 	case TCP_CLOSING:
1934 		chtls_release_resources(sk);
1935 		if (csk_flag_nochk(csk, CSK_ABORT_RPL_PENDING))
1936 			chtls_conn_done(sk);
1937 		else
1938 			chtls_timewait(sk);
1939 		break;
1940 	case TCP_LAST_ACK:
1941 		chtls_release_resources(sk);
1942 		chtls_conn_done(sk);
1943 		break;
1944 	case TCP_FIN_WAIT1:
1945 		tcp_set_state(sk, TCP_FIN_WAIT2);
1946 		sk->sk_shutdown |= SEND_SHUTDOWN;
1947 
1948 		if (!sock_flag(sk, SOCK_DEAD))
1949 			sk->sk_state_change(sk);
1950 		else if (tcp_sk(sk)->linger2 < 0 &&
1951 			 !csk_flag_nochk(csk, CSK_ABORT_SHUTDOWN))
1952 			chtls_abort_conn(sk, skb);
1953 		else if (csk_flag_nochk(csk, CSK_TX_DATA_SENT))
1954 			chtls_set_quiesce_ctrl(sk, 0);
1955 		break;
1956 	default:
1957 		pr_info("close_con_rpl in bad state %d\n", sk->sk_state);
1958 	}
1959 out:
1960 	kfree_skb(skb);
1961 }
1962 
1963 static struct sk_buff *get_cpl_skb(struct sk_buff *skb,
1964 				   size_t len, gfp_t gfp)
1965 {
1966 	if (likely(!skb_is_nonlinear(skb) && !skb_cloned(skb))) {
1967 		WARN_ONCE(skb->len < len, "skb alloc error");
1968 		__skb_trim(skb, len);
1969 		skb_get(skb);
1970 	} else {
1971 		skb = alloc_skb(len, gfp);
1972 		if (skb)
1973 			__skb_put(skb, len);
1974 	}
1975 	return skb;
1976 }
1977 
1978 static void set_abort_rpl_wr(struct sk_buff *skb, unsigned int tid,
1979 			     int cmd)
1980 {
1981 	struct cpl_abort_rpl *rpl = cplhdr(skb);
1982 
1983 	INIT_TP_WR_CPL(rpl, CPL_ABORT_RPL, tid);
1984 	rpl->cmd = cmd;
1985 }
1986 
1987 static void send_defer_abort_rpl(struct chtls_dev *cdev, struct sk_buff *skb)
1988 {
1989 	struct cpl_abort_req_rss *req = cplhdr(skb);
1990 	struct sk_buff *reply_skb;
1991 
1992 	reply_skb = alloc_skb(sizeof(struct cpl_abort_rpl),
1993 			      GFP_KERNEL | __GFP_NOFAIL);
1994 	__skb_put(reply_skb, sizeof(struct cpl_abort_rpl));
1995 	set_abort_rpl_wr(reply_skb, GET_TID(req),
1996 			 (req->status & CPL_ABORT_NO_RST));
1997 	set_wr_txq(reply_skb, CPL_PRIORITY_DATA, req->status >> 1);
1998 	cxgb4_ofld_send(cdev->lldi->ports[0], reply_skb);
1999 	kfree_skb(skb);
2000 }
2001 
2002 /*
2003  * Add an skb to the deferred skb queue for processing from process context.
2004  */
2005 static void t4_defer_reply(struct sk_buff *skb, struct chtls_dev *cdev,
2006 			   defer_handler_t handler)
2007 {
2008 	DEFERRED_SKB_CB(skb)->handler = handler;
2009 	spin_lock_bh(&cdev->deferq.lock);
2010 	__skb_queue_tail(&cdev->deferq, skb);
2011 	if (skb_queue_len(&cdev->deferq) == 1)
2012 		schedule_work(&cdev->deferq_task);
2013 	spin_unlock_bh(&cdev->deferq.lock);
2014 }
2015 
2016 static void chtls_send_abort_rpl(struct sock *sk, struct sk_buff *skb,
2017 				 struct chtls_dev *cdev,
2018 				 int status, int queue)
2019 {
2020 	struct cpl_abort_req_rss *req = cplhdr(skb) + RSS_HDR;
2021 	struct sk_buff *reply_skb;
2022 	struct chtls_sock *csk;
2023 	unsigned int tid;
2024 
2025 	csk = rcu_dereference_sk_user_data(sk);
2026 	tid = GET_TID(req);
2027 
2028 	reply_skb = get_cpl_skb(skb, sizeof(struct cpl_abort_rpl), gfp_any());
2029 	if (!reply_skb) {
2030 		req->status = (queue << 1) | status;
2031 		t4_defer_reply(skb, cdev, send_defer_abort_rpl);
2032 		return;
2033 	}
2034 
2035 	set_abort_rpl_wr(reply_skb, tid, status);
2036 	kfree_skb(skb);
2037 	set_wr_txq(reply_skb, CPL_PRIORITY_DATA, queue);
2038 	if (csk_conn_inline(csk)) {
2039 		struct l2t_entry *e = csk->l2t_entry;
2040 
2041 		if (e && sk->sk_state != TCP_SYN_RECV) {
2042 			cxgb4_l2t_send(csk->egress_dev, reply_skb, e);
2043 			return;
2044 		}
2045 	}
2046 	cxgb4_ofld_send(cdev->lldi->ports[0], reply_skb);
2047 }
2048 
2049 /*
2050  * This is run from a listener's backlog to abort a child connection in
2051  * SYN_RCV state (i.e., one on the listener's SYN queue).
2052  */
2053 static void bl_abort_syn_rcv(struct sock *lsk, struct sk_buff *skb)
2054 {
2055 	struct chtls_sock *csk;
2056 	struct sock *child;
2057 	int queue;
2058 
2059 	child = skb->sk;
2060 	csk = rcu_dereference_sk_user_data(child);
2061 	queue = csk->txq_idx;
2062 
2063 	skb->sk	= NULL;
2064 	chtls_send_abort_rpl(child, skb, BLOG_SKB_CB(skb)->cdev,
2065 			     CPL_ABORT_NO_RST, queue);
2066 	do_abort_syn_rcv(child, lsk);
2067 }
2068 
2069 static int abort_syn_rcv(struct sock *sk, struct sk_buff *skb)
2070 {
2071 	const struct request_sock *oreq;
2072 	struct listen_ctx *listen_ctx;
2073 	struct chtls_sock *csk;
2074 	struct chtls_dev *cdev;
2075 	struct sock *psk;
2076 	void *ctx;
2077 
2078 	csk = sk->sk_user_data;
2079 	oreq = csk->passive_reap_next;
2080 	cdev = csk->cdev;
2081 
2082 	if (!oreq)
2083 		return -1;
2084 
2085 	ctx = lookup_stid(cdev->tids, oreq->ts_recent);
2086 	if (!ctx)
2087 		return -1;
2088 
2089 	listen_ctx = (struct listen_ctx *)ctx;
2090 	psk = listen_ctx->lsk;
2091 
2092 	bh_lock_sock(psk);
2093 	if (!sock_owned_by_user(psk)) {
2094 		int queue = csk->txq_idx;
2095 
2096 		chtls_send_abort_rpl(sk, skb, cdev, CPL_ABORT_NO_RST, queue);
2097 		do_abort_syn_rcv(sk, psk);
2098 	} else {
2099 		skb->sk = sk;
2100 		BLOG_SKB_CB(skb)->backlog_rcv = bl_abort_syn_rcv;
2101 		__sk_add_backlog(psk, skb);
2102 	}
2103 	bh_unlock_sock(psk);
2104 	return 0;
2105 }
2106 
2107 static void chtls_abort_req_rss(struct sock *sk, struct sk_buff *skb)
2108 {
2109 	const struct cpl_abort_req_rss *req = cplhdr(skb) + RSS_HDR;
2110 	struct chtls_sock *csk = sk->sk_user_data;
2111 	int rst_status = CPL_ABORT_NO_RST;
2112 	int queue = csk->txq_idx;
2113 
2114 	if (is_neg_adv(req->status)) {
2115 		kfree_skb(skb);
2116 		return;
2117 	}
2118 
2119 	csk_reset_flag(csk, CSK_ABORT_REQ_RCVD);
2120 
2121 	if (!csk_flag_nochk(csk, CSK_ABORT_SHUTDOWN) &&
2122 	    !csk_flag_nochk(csk, CSK_TX_DATA_SENT)) {
2123 		struct tcp_sock *tp = tcp_sk(sk);
2124 
2125 		if (send_tx_flowc_wr(sk, 0, tp->snd_nxt, tp->rcv_nxt) < 0)
2126 			WARN_ONCE(1, "send_tx_flowc error");
2127 		csk_set_flag(csk, CSK_TX_DATA_SENT);
2128 	}
2129 
2130 	csk_set_flag(csk, CSK_ABORT_SHUTDOWN);
2131 
2132 	if (!csk_flag_nochk(csk, CSK_ABORT_RPL_PENDING)) {
2133 		sk->sk_err = ETIMEDOUT;
2134 
2135 		if (!sock_flag(sk, SOCK_DEAD))
2136 			sk_error_report(sk);
2137 
2138 		if (sk->sk_state == TCP_SYN_RECV && !abort_syn_rcv(sk, skb))
2139 			return;
2140 
2141 	}
2142 
2143 	chtls_send_abort_rpl(sk, skb, BLOG_SKB_CB(skb)->cdev,
2144 			     rst_status, queue);
2145 	chtls_release_resources(sk);
2146 	chtls_conn_done(sk);
2147 }
2148 
2149 static void chtls_abort_rpl_rss(struct sock *sk, struct sk_buff *skb)
2150 {
2151 	struct cpl_abort_rpl_rss *rpl = cplhdr(skb) + RSS_HDR;
2152 	struct chtls_sock *csk;
2153 	struct chtls_dev *cdev;
2154 
2155 	csk = rcu_dereference_sk_user_data(sk);
2156 	cdev = csk->cdev;
2157 
2158 	if (csk_flag_nochk(csk, CSK_ABORT_RPL_PENDING)) {
2159 		csk_reset_flag(csk, CSK_ABORT_RPL_PENDING);
2160 		if (!csk_flag_nochk(csk, CSK_ABORT_REQ_RCVD)) {
2161 			if (sk->sk_state == TCP_SYN_SENT) {
2162 				cxgb4_remove_tid(cdev->tids,
2163 						 csk->port_id,
2164 						 GET_TID(rpl),
2165 						 sk->sk_family);
2166 				sock_put(sk);
2167 			}
2168 			chtls_release_resources(sk);
2169 			chtls_conn_done(sk);
2170 		}
2171 	}
2172 	kfree_skb(skb);
2173 }
2174 
2175 static int chtls_conn_cpl(struct chtls_dev *cdev, struct sk_buff *skb)
2176 {
2177 	struct cpl_peer_close *req = cplhdr(skb) + RSS_HDR;
2178 	void (*fn)(struct sock *sk, struct sk_buff *skb);
2179 	unsigned int hwtid = GET_TID(req);
2180 	struct chtls_sock *csk;
2181 	struct sock *sk;
2182 	u8 opcode;
2183 
2184 	opcode = ((const struct rss_header *)cplhdr(skb))->opcode;
2185 
2186 	sk = lookup_tid(cdev->tids, hwtid);
2187 	if (!sk)
2188 		goto rel_skb;
2189 
2190 	csk = sk->sk_user_data;
2191 
2192 	switch (opcode) {
2193 	case CPL_PEER_CLOSE:
2194 		fn = chtls_peer_close;
2195 		break;
2196 	case CPL_CLOSE_CON_RPL:
2197 		fn = chtls_close_con_rpl;
2198 		break;
2199 	case CPL_ABORT_REQ_RSS:
2200 		/*
2201 		 * Save the offload device in the skb, we may process this
2202 		 * message after the socket has closed.
2203 		 */
2204 		BLOG_SKB_CB(skb)->cdev = csk->cdev;
2205 		fn = chtls_abort_req_rss;
2206 		break;
2207 	case CPL_ABORT_RPL_RSS:
2208 		fn = chtls_abort_rpl_rss;
2209 		break;
2210 	default:
2211 		goto rel_skb;
2212 	}
2213 
2214 	process_cpl_msg(fn, sk, skb);
2215 	return 0;
2216 
2217 rel_skb:
2218 	kfree_skb(skb);
2219 	return 0;
2220 }
2221 
2222 static void chtls_rx_ack(struct sock *sk, struct sk_buff *skb)
2223 {
2224 	struct cpl_fw4_ack *hdr = cplhdr(skb) + RSS_HDR;
2225 	struct chtls_sock *csk = sk->sk_user_data;
2226 	struct tcp_sock *tp = tcp_sk(sk);
2227 	u32 credits = hdr->credits;
2228 	u32 snd_una;
2229 
2230 	snd_una = ntohl(hdr->snd_una);
2231 	csk->wr_credits += credits;
2232 
2233 	if (csk->wr_unacked > csk->wr_max_credits - csk->wr_credits)
2234 		csk->wr_unacked = csk->wr_max_credits - csk->wr_credits;
2235 
2236 	while (credits) {
2237 		struct sk_buff *pskb = csk->wr_skb_head;
2238 		u32 csum;
2239 
2240 		if (unlikely(!pskb)) {
2241 			if (csk->wr_nondata)
2242 				csk->wr_nondata -= credits;
2243 			break;
2244 		}
2245 		csum = (__force u32)pskb->csum;
2246 		if (unlikely(credits < csum)) {
2247 			pskb->csum = (__force __wsum)(csum - credits);
2248 			break;
2249 		}
2250 		dequeue_wr(sk);
2251 		credits -= csum;
2252 		kfree_skb(pskb);
2253 	}
2254 	if (hdr->seq_vld & CPL_FW4_ACK_FLAGS_SEQVAL) {
2255 		if (unlikely(before(snd_una, tp->snd_una))) {
2256 			kfree_skb(skb);
2257 			return;
2258 		}
2259 
2260 		if (tp->snd_una != snd_una) {
2261 			tp->snd_una = snd_una;
2262 			tp->rcv_tstamp = tcp_time_stamp(tp);
2263 			if (tp->snd_una == tp->snd_nxt &&
2264 			    !csk_flag_nochk(csk, CSK_TX_FAILOVER))
2265 				csk_reset_flag(csk, CSK_TX_WAIT_IDLE);
2266 		}
2267 	}
2268 
2269 	if (hdr->seq_vld & CPL_FW4_ACK_FLAGS_CH) {
2270 		unsigned int fclen16 = roundup(failover_flowc_wr_len, 16);
2271 
2272 		csk->wr_credits -= fclen16;
2273 		csk_reset_flag(csk, CSK_TX_WAIT_IDLE);
2274 		csk_reset_flag(csk, CSK_TX_FAILOVER);
2275 	}
2276 	if (skb_queue_len(&csk->txq) && chtls_push_frames(csk, 0))
2277 		sk->sk_write_space(sk);
2278 
2279 	kfree_skb(skb);
2280 }
2281 
2282 static int chtls_wr_ack(struct chtls_dev *cdev, struct sk_buff *skb)
2283 {
2284 	struct cpl_fw4_ack *rpl = cplhdr(skb) + RSS_HDR;
2285 	unsigned int hwtid = GET_TID(rpl);
2286 	struct sock *sk;
2287 
2288 	sk = lookup_tid(cdev->tids, hwtid);
2289 	if (unlikely(!sk)) {
2290 		pr_err("can't find conn. for hwtid %u.\n", hwtid);
2291 		return -EINVAL;
2292 	}
2293 	process_cpl_msg(chtls_rx_ack, sk, skb);
2294 
2295 	return 0;
2296 }
2297 
2298 static int chtls_set_tcb_rpl(struct chtls_dev *cdev, struct sk_buff *skb)
2299 {
2300 	struct cpl_set_tcb_rpl *rpl = cplhdr(skb) + RSS_HDR;
2301 	unsigned int hwtid = GET_TID(rpl);
2302 	struct sock *sk;
2303 
2304 	sk = lookup_tid(cdev->tids, hwtid);
2305 
2306 	/* return EINVAL if socket doesn't exist */
2307 	if (!sk)
2308 		return -EINVAL;
2309 
2310 	/* Reusing the skb as size of cpl_set_tcb_field structure
2311 	 * is greater than cpl_abort_req
2312 	 */
2313 	if (TCB_COOKIE_G(rpl->cookie) == TCB_FIELD_COOKIE_TFLAG)
2314 		chtls_send_abort(sk, CPL_ABORT_SEND_RST, NULL);
2315 
2316 	kfree_skb(skb);
2317 	return 0;
2318 }
2319 
2320 chtls_handler_func chtls_handlers[NUM_CPL_CMDS] = {
2321 	[CPL_PASS_OPEN_RPL]     = chtls_pass_open_rpl,
2322 	[CPL_CLOSE_LISTSRV_RPL] = chtls_close_listsrv_rpl,
2323 	[CPL_PASS_ACCEPT_REQ]   = chtls_pass_accept_req,
2324 	[CPL_PASS_ESTABLISH]    = chtls_pass_establish,
2325 	[CPL_RX_DATA]           = chtls_rx_data,
2326 	[CPL_TLS_DATA]          = chtls_rx_pdu,
2327 	[CPL_RX_TLS_CMP]        = chtls_rx_cmp,
2328 	[CPL_PEER_CLOSE]        = chtls_conn_cpl,
2329 	[CPL_CLOSE_CON_RPL]     = chtls_conn_cpl,
2330 	[CPL_ABORT_REQ_RSS]     = chtls_conn_cpl,
2331 	[CPL_ABORT_RPL_RSS]     = chtls_conn_cpl,
2332 	[CPL_FW4_ACK]		= chtls_wr_ack,
2333 	[CPL_SET_TCB_RPL]	= chtls_set_tcb_rpl,
2334 };
2335