• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 /*
2  * Copyright (c) 2018 Chelsio Communications, Inc.
3  *
4  * This program is free software; you can redistribute it and/or modify
5  * it under the terms of the GNU General Public License version 2 as
6  * published by the Free Software Foundation.
7  *
8  * Written by: Atul Gupta (atul.gupta@chelsio.com)
9  */
10 
11 #include <linux/module.h>
12 #include <linux/list.h>
13 #include <linux/workqueue.h>
14 #include <linux/skbuff.h>
15 #include <linux/timer.h>
16 #include <linux/notifier.h>
17 #include <linux/inetdevice.h>
18 #include <linux/ip.h>
19 #include <linux/tcp.h>
20 #include <linux/sched/signal.h>
21 #include <linux/kallsyms.h>
22 #include <linux/kprobes.h>
23 #include <linux/if_vlan.h>
24 #include <net/tcp.h>
25 #include <net/dst.h>
26 
27 #include "chtls.h"
28 #include "chtls_cm.h"
29 
30 /*
31  * State transitions and actions for close.  Note that if we are in SYN_SENT
32  * we remain in that state as we cannot control a connection while it's in
33  * SYN_SENT; such connections are allowed to establish and are then aborted.
34  */
35 static unsigned char new_state[16] = {
36 	/* current state:     new state:      action: */
37 	/* (Invalid)       */ TCP_CLOSE,
38 	/* TCP_ESTABLISHED */ TCP_FIN_WAIT1 | TCP_ACTION_FIN,
39 	/* TCP_SYN_SENT    */ TCP_SYN_SENT,
40 	/* TCP_SYN_RECV    */ TCP_FIN_WAIT1 | TCP_ACTION_FIN,
41 	/* TCP_FIN_WAIT1   */ TCP_FIN_WAIT1,
42 	/* TCP_FIN_WAIT2   */ TCP_FIN_WAIT2,
43 	/* TCP_TIME_WAIT   */ TCP_CLOSE,
44 	/* TCP_CLOSE       */ TCP_CLOSE,
45 	/* TCP_CLOSE_WAIT  */ TCP_LAST_ACK | TCP_ACTION_FIN,
46 	/* TCP_LAST_ACK    */ TCP_LAST_ACK,
47 	/* TCP_LISTEN      */ TCP_CLOSE,
48 	/* TCP_CLOSING     */ TCP_CLOSING,
49 };
50 
chtls_sock_create(struct chtls_dev * cdev)51 static struct chtls_sock *chtls_sock_create(struct chtls_dev *cdev)
52 {
53 	struct chtls_sock *csk = kzalloc(sizeof(*csk), GFP_ATOMIC);
54 
55 	if (!csk)
56 		return NULL;
57 
58 	csk->txdata_skb_cache = alloc_skb(TXDATA_SKB_LEN, GFP_ATOMIC);
59 	if (!csk->txdata_skb_cache) {
60 		kfree(csk);
61 		return NULL;
62 	}
63 
64 	kref_init(&csk->kref);
65 	csk->cdev = cdev;
66 	skb_queue_head_init(&csk->txq);
67 	csk->wr_skb_head = NULL;
68 	csk->wr_skb_tail = NULL;
69 	csk->mss = MAX_MSS;
70 	csk->tlshws.ofld = 1;
71 	csk->tlshws.txkey = -1;
72 	csk->tlshws.rxkey = -1;
73 	csk->tlshws.mfs = TLS_MFS;
74 	skb_queue_head_init(&csk->tlshws.sk_recv_queue);
75 	return csk;
76 }
77 
chtls_sock_release(struct kref * ref)78 static void chtls_sock_release(struct kref *ref)
79 {
80 	struct chtls_sock *csk =
81 		container_of(ref, struct chtls_sock, kref);
82 
83 	kfree(csk);
84 }
85 
chtls_ipv4_netdev(struct chtls_dev * cdev,struct sock * sk)86 static struct net_device *chtls_ipv4_netdev(struct chtls_dev *cdev,
87 					    struct sock *sk)
88 {
89 	struct net_device *ndev = cdev->ports[0];
90 
91 	if (likely(!inet_sk(sk)->inet_rcv_saddr))
92 		return ndev;
93 
94 	ndev = ip_dev_find(&init_net, inet_sk(sk)->inet_rcv_saddr);
95 	if (!ndev)
96 		return NULL;
97 
98 	if (is_vlan_dev(ndev))
99 		return vlan_dev_real_dev(ndev);
100 	return ndev;
101 }
102 
assign_rxopt(struct sock * sk,unsigned int opt)103 static void assign_rxopt(struct sock *sk, unsigned int opt)
104 {
105 	const struct chtls_dev *cdev;
106 	struct chtls_sock *csk;
107 	struct tcp_sock *tp;
108 
109 	csk = rcu_dereference_sk_user_data(sk);
110 	tp = tcp_sk(sk);
111 
112 	cdev = csk->cdev;
113 	tp->tcp_header_len           = sizeof(struct tcphdr);
114 	tp->rx_opt.mss_clamp         = cdev->mtus[TCPOPT_MSS_G(opt)] - 40;
115 	tp->mss_cache                = tp->rx_opt.mss_clamp;
116 	tp->rx_opt.tstamp_ok         = TCPOPT_TSTAMP_G(opt);
117 	tp->rx_opt.snd_wscale        = TCPOPT_SACK_G(opt);
118 	tp->rx_opt.wscale_ok         = TCPOPT_WSCALE_OK_G(opt);
119 	SND_WSCALE(tp)               = TCPOPT_SND_WSCALE_G(opt);
120 	if (!tp->rx_opt.wscale_ok)
121 		tp->rx_opt.rcv_wscale = 0;
122 	if (tp->rx_opt.tstamp_ok) {
123 		tp->tcp_header_len += TCPOLEN_TSTAMP_ALIGNED;
124 		tp->rx_opt.mss_clamp -= TCPOLEN_TSTAMP_ALIGNED;
125 	} else if (csk->opt2 & TSTAMPS_EN_F) {
126 		csk->opt2 &= ~TSTAMPS_EN_F;
127 		csk->mtu_idx = TCPOPT_MSS_G(opt);
128 	}
129 }
130 
chtls_purge_receive_queue(struct sock * sk)131 static void chtls_purge_receive_queue(struct sock *sk)
132 {
133 	struct sk_buff *skb;
134 
135 	while ((skb = __skb_dequeue(&sk->sk_receive_queue)) != NULL) {
136 		skb_dst_set(skb, (void *)NULL);
137 		kfree_skb(skb);
138 	}
139 }
140 
chtls_purge_write_queue(struct sock * sk)141 static void chtls_purge_write_queue(struct sock *sk)
142 {
143 	struct chtls_sock *csk = rcu_dereference_sk_user_data(sk);
144 	struct sk_buff *skb;
145 
146 	while ((skb = __skb_dequeue(&csk->txq))) {
147 		sk->sk_wmem_queued -= skb->truesize;
148 		__kfree_skb(skb);
149 	}
150 }
151 
chtls_purge_recv_queue(struct sock * sk)152 static void chtls_purge_recv_queue(struct sock *sk)
153 {
154 	struct chtls_sock *csk = rcu_dereference_sk_user_data(sk);
155 	struct chtls_hws *tlsk = &csk->tlshws;
156 	struct sk_buff *skb;
157 
158 	while ((skb = __skb_dequeue(&tlsk->sk_recv_queue)) != NULL) {
159 		skb_dst_set(skb, NULL);
160 		kfree_skb(skb);
161 	}
162 }
163 
abort_arp_failure(void * handle,struct sk_buff * skb)164 static void abort_arp_failure(void *handle, struct sk_buff *skb)
165 {
166 	struct cpl_abort_req *req = cplhdr(skb);
167 	struct chtls_dev *cdev;
168 
169 	cdev = (struct chtls_dev *)handle;
170 	req->cmd = CPL_ABORT_NO_RST;
171 	cxgb4_ofld_send(cdev->lldi->ports[0], skb);
172 }
173 
alloc_ctrl_skb(struct sk_buff * skb,int len)174 static struct sk_buff *alloc_ctrl_skb(struct sk_buff *skb, int len)
175 {
176 	if (likely(skb && !skb_shared(skb) && !skb_cloned(skb))) {
177 		__skb_trim(skb, 0);
178 		refcount_add(2, &skb->users);
179 	} else {
180 		skb = alloc_skb(len, GFP_KERNEL | __GFP_NOFAIL);
181 	}
182 	return skb;
183 }
184 
chtls_send_abort(struct sock * sk,int mode,struct sk_buff * skb)185 static void chtls_send_abort(struct sock *sk, int mode, struct sk_buff *skb)
186 {
187 	struct cpl_abort_req *req;
188 	struct chtls_sock *csk;
189 	struct tcp_sock *tp;
190 
191 	csk = rcu_dereference_sk_user_data(sk);
192 	tp = tcp_sk(sk);
193 
194 	if (!skb)
195 		skb = alloc_ctrl_skb(csk->txdata_skb_cache, sizeof(*req));
196 
197 	req = (struct cpl_abort_req *)skb_put(skb, sizeof(*req));
198 	INIT_TP_WR_CPL(req, CPL_ABORT_REQ, csk->tid);
199 	skb_set_queue_mapping(skb, (csk->txq_idx << 1) | CPL_PRIORITY_DATA);
200 	req->rsvd0 = htonl(tp->snd_nxt);
201 	req->rsvd1 = !csk_flag_nochk(csk, CSK_TX_DATA_SENT);
202 	req->cmd = mode;
203 	t4_set_arp_err_handler(skb, csk->cdev, abort_arp_failure);
204 	send_or_defer(sk, tp, skb, mode == CPL_ABORT_SEND_RST);
205 }
206 
chtls_send_reset(struct sock * sk,int mode,struct sk_buff * skb)207 static void chtls_send_reset(struct sock *sk, int mode, struct sk_buff *skb)
208 {
209 	struct chtls_sock *csk = rcu_dereference_sk_user_data(sk);
210 
211 	if (unlikely(csk_flag_nochk(csk, CSK_ABORT_SHUTDOWN) ||
212 		     !csk->cdev)) {
213 		if (sk->sk_state == TCP_SYN_RECV)
214 			csk_set_flag(csk, CSK_RST_ABORTED);
215 		goto out;
216 	}
217 
218 	if (!csk_flag_nochk(csk, CSK_TX_DATA_SENT)) {
219 		struct tcp_sock *tp = tcp_sk(sk);
220 
221 		if (send_tx_flowc_wr(sk, 0, tp->snd_nxt, tp->rcv_nxt) < 0)
222 			WARN_ONCE(1, "send tx flowc error");
223 		csk_set_flag(csk, CSK_TX_DATA_SENT);
224 	}
225 
226 	csk_set_flag(csk, CSK_ABORT_RPL_PENDING);
227 	chtls_purge_write_queue(sk);
228 
229 	csk_set_flag(csk, CSK_ABORT_SHUTDOWN);
230 	if (sk->sk_state != TCP_SYN_RECV)
231 		chtls_send_abort(sk, mode, skb);
232 	else
233 		goto out;
234 
235 	return;
236 out:
237 	if (skb)
238 		kfree_skb(skb);
239 }
240 
release_tcp_port(struct sock * sk)241 static void release_tcp_port(struct sock *sk)
242 {
243 	if (inet_csk(sk)->icsk_bind_hash)
244 		inet_put_port(sk);
245 }
246 
tcp_uncork(struct sock * sk)247 static void tcp_uncork(struct sock *sk)
248 {
249 	struct tcp_sock *tp = tcp_sk(sk);
250 
251 	if (tp->nonagle & TCP_NAGLE_CORK) {
252 		tp->nonagle &= ~TCP_NAGLE_CORK;
253 		chtls_tcp_push(sk, 0);
254 	}
255 }
256 
chtls_close_conn(struct sock * sk)257 static void chtls_close_conn(struct sock *sk)
258 {
259 	struct cpl_close_con_req *req;
260 	struct chtls_sock *csk;
261 	struct sk_buff *skb;
262 	unsigned int tid;
263 	unsigned int len;
264 
265 	len = roundup(sizeof(struct cpl_close_con_req), 16);
266 	csk = rcu_dereference_sk_user_data(sk);
267 	tid = csk->tid;
268 
269 	skb = alloc_skb(len, GFP_KERNEL | __GFP_NOFAIL);
270 	req = (struct cpl_close_con_req *)__skb_put(skb, len);
271 	memset(req, 0, len);
272 	req->wr.wr_hi = htonl(FW_WR_OP_V(FW_TP_WR) |
273 			      FW_WR_IMMDLEN_V(sizeof(*req) -
274 					      sizeof(req->wr)));
275 	req->wr.wr_mid = htonl(FW_WR_LEN16_V(DIV_ROUND_UP(sizeof(*req), 16)) |
276 			       FW_WR_FLOWID_V(tid));
277 
278 	OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_CLOSE_CON_REQ, tid));
279 
280 	tcp_uncork(sk);
281 	skb_entail(sk, skb, ULPCB_FLAG_NO_HDR | ULPCB_FLAG_NO_APPEND);
282 	if (sk->sk_state != TCP_SYN_SENT)
283 		chtls_push_frames(csk, 1);
284 }
285 
286 /*
287  * Perform a state transition during close and return the actions indicated
288  * for the transition.  Do not make this function inline, the main reason
289  * it exists at all is to avoid multiple inlining of tcp_set_state.
290  */
make_close_transition(struct sock * sk)291 static int make_close_transition(struct sock *sk)
292 {
293 	int next = (int)new_state[sk->sk_state];
294 
295 	tcp_set_state(sk, next & TCP_STATE_MASK);
296 	return next & TCP_ACTION_FIN;
297 }
298 
chtls_close(struct sock * sk,long timeout)299 void chtls_close(struct sock *sk, long timeout)
300 {
301 	int data_lost, prev_state;
302 	struct chtls_sock *csk;
303 
304 	csk = rcu_dereference_sk_user_data(sk);
305 
306 	lock_sock(sk);
307 	sk->sk_shutdown |= SHUTDOWN_MASK;
308 
309 	data_lost = skb_queue_len(&sk->sk_receive_queue);
310 	data_lost |= skb_queue_len(&csk->tlshws.sk_recv_queue);
311 	chtls_purge_recv_queue(sk);
312 	chtls_purge_receive_queue(sk);
313 
314 	if (sk->sk_state == TCP_CLOSE) {
315 		goto wait;
316 	} else if (data_lost || sk->sk_state == TCP_SYN_SENT) {
317 		chtls_send_reset(sk, CPL_ABORT_SEND_RST, NULL);
318 		release_tcp_port(sk);
319 		goto unlock;
320 	} else if (sock_flag(sk, SOCK_LINGER) && !sk->sk_lingertime) {
321 		sk->sk_prot->disconnect(sk, 0);
322 	} else if (make_close_transition(sk)) {
323 		chtls_close_conn(sk);
324 	}
325 wait:
326 	if (timeout)
327 		sk_stream_wait_close(sk, timeout);
328 
329 unlock:
330 	prev_state = sk->sk_state;
331 	sock_hold(sk);
332 	sock_orphan(sk);
333 
334 	release_sock(sk);
335 
336 	local_bh_disable();
337 	bh_lock_sock(sk);
338 
339 	if (prev_state != TCP_CLOSE && sk->sk_state == TCP_CLOSE)
340 		goto out;
341 
342 	if (sk->sk_state == TCP_FIN_WAIT2 && tcp_sk(sk)->linger2 < 0 &&
343 	    !csk_flag(sk, CSK_ABORT_SHUTDOWN)) {
344 		struct sk_buff *skb;
345 
346 		skb = alloc_skb(sizeof(struct cpl_abort_req), GFP_ATOMIC);
347 		if (skb)
348 			chtls_send_reset(sk, CPL_ABORT_SEND_RST, skb);
349 	}
350 
351 	if (sk->sk_state == TCP_CLOSE)
352 		inet_csk_destroy_sock(sk);
353 
354 out:
355 	bh_unlock_sock(sk);
356 	local_bh_enable();
357 	sock_put(sk);
358 }
359 
360 /*
361  * Wait until a socket enters on of the given states.
362  */
wait_for_states(struct sock * sk,unsigned int states)363 static int wait_for_states(struct sock *sk, unsigned int states)
364 {
365 	DECLARE_WAITQUEUE(wait, current);
366 	struct socket_wq _sk_wq;
367 	long current_timeo;
368 	int err = 0;
369 
370 	current_timeo = 200;
371 
372 	/*
373 	 * We want this to work even when there's no associated struct socket.
374 	 * In that case we provide a temporary wait_queue_head_t.
375 	 */
376 	if (!sk->sk_wq) {
377 		init_waitqueue_head(&_sk_wq.wait);
378 		_sk_wq.fasync_list = NULL;
379 		init_rcu_head_on_stack(&_sk_wq.rcu);
380 		RCU_INIT_POINTER(sk->sk_wq, &_sk_wq);
381 	}
382 
383 	add_wait_queue(sk_sleep(sk), &wait);
384 	while (!sk_in_state(sk, states)) {
385 		if (!current_timeo) {
386 			err = -EBUSY;
387 			break;
388 		}
389 		if (signal_pending(current)) {
390 			err = sock_intr_errno(current_timeo);
391 			break;
392 		}
393 		set_current_state(TASK_UNINTERRUPTIBLE);
394 		release_sock(sk);
395 		if (!sk_in_state(sk, states))
396 			current_timeo = schedule_timeout(current_timeo);
397 		__set_current_state(TASK_RUNNING);
398 		lock_sock(sk);
399 	}
400 	remove_wait_queue(sk_sleep(sk), &wait);
401 
402 	if (rcu_dereference(sk->sk_wq) == &_sk_wq)
403 		sk->sk_wq = NULL;
404 	return err;
405 }
406 
chtls_disconnect(struct sock * sk,int flags)407 int chtls_disconnect(struct sock *sk, int flags)
408 {
409 	struct chtls_sock *csk;
410 	struct tcp_sock *tp;
411 	int err;
412 
413 	tp = tcp_sk(sk);
414 	csk = rcu_dereference_sk_user_data(sk);
415 	chtls_purge_recv_queue(sk);
416 	chtls_purge_receive_queue(sk);
417 	chtls_purge_write_queue(sk);
418 
419 	if (sk->sk_state != TCP_CLOSE) {
420 		sk->sk_err = ECONNRESET;
421 		chtls_send_reset(sk, CPL_ABORT_SEND_RST, NULL);
422 		err = wait_for_states(sk, TCPF_CLOSE);
423 		if (err)
424 			return err;
425 	}
426 	chtls_purge_recv_queue(sk);
427 	chtls_purge_receive_queue(sk);
428 	tp->max_window = 0xFFFF << (tp->rx_opt.snd_wscale);
429 	return tcp_disconnect(sk, flags);
430 }
431 
432 #define SHUTDOWN_ELIGIBLE_STATE (TCPF_ESTABLISHED | \
433 				 TCPF_SYN_RECV | TCPF_CLOSE_WAIT)
chtls_shutdown(struct sock * sk,int how)434 void chtls_shutdown(struct sock *sk, int how)
435 {
436 	if ((how & SEND_SHUTDOWN) &&
437 	    sk_in_state(sk, SHUTDOWN_ELIGIBLE_STATE) &&
438 	    make_close_transition(sk))
439 		chtls_close_conn(sk);
440 }
441 
chtls_destroy_sock(struct sock * sk)442 void chtls_destroy_sock(struct sock *sk)
443 {
444 	struct chtls_sock *csk;
445 
446 	csk = rcu_dereference_sk_user_data(sk);
447 	chtls_purge_recv_queue(sk);
448 	csk->ulp_mode = ULP_MODE_NONE;
449 	chtls_purge_write_queue(sk);
450 	free_tls_keyid(sk);
451 	kref_put(&csk->kref, chtls_sock_release);
452 	sk->sk_prot = &tcp_prot;
453 	sk->sk_prot->destroy(sk);
454 }
455 
reset_listen_child(struct sock * child)456 static void reset_listen_child(struct sock *child)
457 {
458 	struct chtls_sock *csk = rcu_dereference_sk_user_data(child);
459 	struct sk_buff *skb;
460 
461 	skb = alloc_ctrl_skb(csk->txdata_skb_cache,
462 			     sizeof(struct cpl_abort_req));
463 
464 	chtls_send_reset(child, CPL_ABORT_SEND_RST, skb);
465 	sock_orphan(child);
466 	INC_ORPHAN_COUNT(child);
467 	if (child->sk_state == TCP_CLOSE)
468 		inet_csk_destroy_sock(child);
469 }
470 
chtls_disconnect_acceptq(struct sock * listen_sk)471 static void chtls_disconnect_acceptq(struct sock *listen_sk)
472 {
473 	struct request_sock **pprev;
474 
475 	pprev = ACCEPT_QUEUE(listen_sk);
476 	while (*pprev) {
477 		struct request_sock *req = *pprev;
478 
479 		if (req->rsk_ops == &chtls_rsk_ops) {
480 			struct sock *child = req->sk;
481 
482 			*pprev = req->dl_next;
483 			sk_acceptq_removed(listen_sk);
484 			reqsk_put(req);
485 			sock_hold(child);
486 			local_bh_disable();
487 			bh_lock_sock(child);
488 			release_tcp_port(child);
489 			reset_listen_child(child);
490 			bh_unlock_sock(child);
491 			local_bh_enable();
492 			sock_put(child);
493 		} else {
494 			pprev = &req->dl_next;
495 		}
496 	}
497 }
498 
listen_hashfn(const struct sock * sk)499 static int listen_hashfn(const struct sock *sk)
500 {
501 	return ((unsigned long)sk >> 10) & (LISTEN_INFO_HASH_SIZE - 1);
502 }
503 
listen_hash_add(struct chtls_dev * cdev,struct sock * sk,unsigned int stid)504 static struct listen_info *listen_hash_add(struct chtls_dev *cdev,
505 					   struct sock *sk,
506 					   unsigned int stid)
507 {
508 	struct listen_info *p = kmalloc(sizeof(*p), GFP_KERNEL);
509 
510 	if (p) {
511 		int key = listen_hashfn(sk);
512 
513 		p->sk = sk;
514 		p->stid = stid;
515 		spin_lock(&cdev->listen_lock);
516 		p->next = cdev->listen_hash_tab[key];
517 		cdev->listen_hash_tab[key] = p;
518 		spin_unlock(&cdev->listen_lock);
519 	}
520 	return p;
521 }
522 
listen_hash_find(struct chtls_dev * cdev,struct sock * sk)523 static int listen_hash_find(struct chtls_dev *cdev,
524 			    struct sock *sk)
525 {
526 	struct listen_info *p;
527 	int stid = -1;
528 	int key;
529 
530 	key = listen_hashfn(sk);
531 
532 	spin_lock(&cdev->listen_lock);
533 	for (p = cdev->listen_hash_tab[key]; p; p = p->next)
534 		if (p->sk == sk) {
535 			stid = p->stid;
536 			break;
537 		}
538 	spin_unlock(&cdev->listen_lock);
539 	return stid;
540 }
541 
listen_hash_del(struct chtls_dev * cdev,struct sock * sk)542 static int listen_hash_del(struct chtls_dev *cdev,
543 			   struct sock *sk)
544 {
545 	struct listen_info *p, **prev;
546 	int stid = -1;
547 	int key;
548 
549 	key = listen_hashfn(sk);
550 	prev = &cdev->listen_hash_tab[key];
551 
552 	spin_lock(&cdev->listen_lock);
553 	for (p = *prev; p; prev = &p->next, p = p->next)
554 		if (p->sk == sk) {
555 			stid = p->stid;
556 			*prev = p->next;
557 			kfree(p);
558 			break;
559 		}
560 	spin_unlock(&cdev->listen_lock);
561 	return stid;
562 }
563 
cleanup_syn_rcv_conn(struct sock * child,struct sock * parent)564 static void cleanup_syn_rcv_conn(struct sock *child, struct sock *parent)
565 {
566 	struct request_sock *req;
567 	struct chtls_sock *csk;
568 
569 	csk = rcu_dereference_sk_user_data(child);
570 	req = csk->passive_reap_next;
571 
572 	reqsk_queue_removed(&inet_csk(parent)->icsk_accept_queue, req);
573 	__skb_unlink((struct sk_buff *)&csk->synq, &csk->listen_ctx->synq);
574 	chtls_reqsk_free(req);
575 	csk->passive_reap_next = NULL;
576 }
577 
chtls_reset_synq(struct listen_ctx * listen_ctx)578 static void chtls_reset_synq(struct listen_ctx *listen_ctx)
579 {
580 	struct sock *listen_sk = listen_ctx->lsk;
581 
582 	while (!skb_queue_empty(&listen_ctx->synq)) {
583 		struct chtls_sock *csk =
584 			container_of((struct synq *)__skb_dequeue
585 				(&listen_ctx->synq), struct chtls_sock, synq);
586 		struct sock *child = csk->sk;
587 
588 		cleanup_syn_rcv_conn(child, listen_sk);
589 		sock_hold(child);
590 		local_bh_disable();
591 		bh_lock_sock(child);
592 		release_tcp_port(child);
593 		reset_listen_child(child);
594 		bh_unlock_sock(child);
595 		local_bh_enable();
596 		sock_put(child);
597 	}
598 }
599 
chtls_listen_start(struct chtls_dev * cdev,struct sock * sk)600 int chtls_listen_start(struct chtls_dev *cdev, struct sock *sk)
601 {
602 	struct net_device *ndev;
603 	struct listen_ctx *ctx;
604 	struct adapter *adap;
605 	struct port_info *pi;
606 	int stid;
607 	int ret;
608 
609 	if (sk->sk_family != PF_INET)
610 		return -EAGAIN;
611 
612 	rcu_read_lock();
613 	ndev = chtls_ipv4_netdev(cdev, sk);
614 	rcu_read_unlock();
615 	if (!ndev)
616 		return -EBADF;
617 
618 	pi = netdev_priv(ndev);
619 	adap = pi->adapter;
620 	if (!(adap->flags & FULL_INIT_DONE))
621 		return -EBADF;
622 
623 	if (listen_hash_find(cdev, sk) >= 0)   /* already have it */
624 		return -EADDRINUSE;
625 
626 	ctx = kmalloc(sizeof(*ctx), GFP_KERNEL);
627 	if (!ctx)
628 		return -ENOMEM;
629 
630 	__module_get(THIS_MODULE);
631 	ctx->lsk = sk;
632 	ctx->cdev = cdev;
633 	ctx->state = T4_LISTEN_START_PENDING;
634 	skb_queue_head_init(&ctx->synq);
635 
636 	stid = cxgb4_alloc_stid(cdev->tids, sk->sk_family, ctx);
637 	if (stid < 0)
638 		goto free_ctx;
639 
640 	sock_hold(sk);
641 	if (!listen_hash_add(cdev, sk, stid))
642 		goto free_stid;
643 
644 	ret = cxgb4_create_server(ndev, stid,
645 				  inet_sk(sk)->inet_rcv_saddr,
646 				  inet_sk(sk)->inet_sport, 0,
647 				  cdev->lldi->rxq_ids[0]);
648 	if (ret > 0)
649 		ret = net_xmit_errno(ret);
650 	if (ret)
651 		goto del_hash;
652 	return 0;
653 del_hash:
654 	listen_hash_del(cdev, sk);
655 free_stid:
656 	cxgb4_free_stid(cdev->tids, stid, sk->sk_family);
657 	sock_put(sk);
658 free_ctx:
659 	kfree(ctx);
660 	module_put(THIS_MODULE);
661 	return -EBADF;
662 }
663 
chtls_listen_stop(struct chtls_dev * cdev,struct sock * sk)664 void chtls_listen_stop(struct chtls_dev *cdev, struct sock *sk)
665 {
666 	struct listen_ctx *listen_ctx;
667 	int stid;
668 
669 	stid = listen_hash_del(cdev, sk);
670 	if (stid < 0)
671 		return;
672 
673 	listen_ctx = (struct listen_ctx *)lookup_stid(cdev->tids, stid);
674 	chtls_reset_synq(listen_ctx);
675 
676 	cxgb4_remove_server(cdev->lldi->ports[0], stid,
677 			    cdev->lldi->rxq_ids[0], 0);
678 	chtls_disconnect_acceptq(sk);
679 }
680 
chtls_pass_open_rpl(struct chtls_dev * cdev,struct sk_buff * skb)681 static int chtls_pass_open_rpl(struct chtls_dev *cdev, struct sk_buff *skb)
682 {
683 	struct cpl_pass_open_rpl *rpl = cplhdr(skb) + RSS_HDR;
684 	unsigned int stid = GET_TID(rpl);
685 	struct listen_ctx *listen_ctx;
686 
687 	listen_ctx = (struct listen_ctx *)lookup_stid(cdev->tids, stid);
688 	if (!listen_ctx)
689 		return CPL_RET_BUF_DONE;
690 
691 	if (listen_ctx->state == T4_LISTEN_START_PENDING) {
692 		listen_ctx->state = T4_LISTEN_STARTED;
693 		return CPL_RET_BUF_DONE;
694 	}
695 
696 	if (rpl->status != CPL_ERR_NONE) {
697 		pr_info("Unexpected PASS_OPEN_RPL status %u for STID %u\n",
698 			rpl->status, stid);
699 	} else {
700 		cxgb4_free_stid(cdev->tids, stid, listen_ctx->lsk->sk_family);
701 		sock_put(listen_ctx->lsk);
702 		kfree(listen_ctx);
703 		module_put(THIS_MODULE);
704 	}
705 	return CPL_RET_BUF_DONE;
706 }
707 
chtls_close_listsrv_rpl(struct chtls_dev * cdev,struct sk_buff * skb)708 static int chtls_close_listsrv_rpl(struct chtls_dev *cdev, struct sk_buff *skb)
709 {
710 	struct cpl_close_listsvr_rpl *rpl = cplhdr(skb) + RSS_HDR;
711 	struct listen_ctx *listen_ctx;
712 	unsigned int stid;
713 	void *data;
714 
715 	stid = GET_TID(rpl);
716 	data = lookup_stid(cdev->tids, stid);
717 	listen_ctx = (struct listen_ctx *)data;
718 
719 	if (rpl->status != CPL_ERR_NONE) {
720 		pr_info("Unexpected CLOSE_LISTSRV_RPL status %u for STID %u\n",
721 			rpl->status, stid);
722 	} else {
723 		cxgb4_free_stid(cdev->tids, stid, listen_ctx->lsk->sk_family);
724 		sock_put(listen_ctx->lsk);
725 		kfree(listen_ctx);
726 		module_put(THIS_MODULE);
727 	}
728 	return CPL_RET_BUF_DONE;
729 }
730 
chtls_purge_wr_queue(struct sock * sk)731 static void chtls_purge_wr_queue(struct sock *sk)
732 {
733 	struct sk_buff *skb;
734 
735 	while ((skb = dequeue_wr(sk)) != NULL)
736 		kfree_skb(skb);
737 }
738 
chtls_release_resources(struct sock * sk)739 static void chtls_release_resources(struct sock *sk)
740 {
741 	struct chtls_sock *csk = rcu_dereference_sk_user_data(sk);
742 	struct chtls_dev *cdev = csk->cdev;
743 	unsigned int tid = csk->tid;
744 	struct tid_info *tids;
745 
746 	if (!cdev)
747 		return;
748 
749 	tids = cdev->tids;
750 	kfree_skb(csk->txdata_skb_cache);
751 	csk->txdata_skb_cache = NULL;
752 
753 	if (csk->wr_credits != csk->wr_max_credits) {
754 		chtls_purge_wr_queue(sk);
755 		chtls_reset_wr_list(csk);
756 	}
757 
758 	if (csk->l2t_entry) {
759 		cxgb4_l2t_release(csk->l2t_entry);
760 		csk->l2t_entry = NULL;
761 	}
762 
763 	cxgb4_remove_tid(tids, csk->port_id, tid, sk->sk_family);
764 	sock_put(sk);
765 }
766 
chtls_conn_done(struct sock * sk)767 static void chtls_conn_done(struct sock *sk)
768 {
769 	if (sock_flag(sk, SOCK_DEAD))
770 		chtls_purge_receive_queue(sk);
771 	sk_wakeup_sleepers(sk, 0);
772 	tcp_done(sk);
773 }
774 
do_abort_syn_rcv(struct sock * child,struct sock * parent)775 static void do_abort_syn_rcv(struct sock *child, struct sock *parent)
776 {
777 	/*
778 	 * If the server is still open we clean up the child connection,
779 	 * otherwise the server already did the clean up as it was purging
780 	 * its SYN queue and the skb was just sitting in its backlog.
781 	 */
782 	if (likely(parent->sk_state == TCP_LISTEN)) {
783 		cleanup_syn_rcv_conn(child, parent);
784 		/* Without the below call to sock_orphan,
785 		 * we leak the socket resource with syn_flood test
786 		 * as inet_csk_destroy_sock will not be called
787 		 * in tcp_done since SOCK_DEAD flag is not set.
788 		 * Kernel handles this differently where new socket is
789 		 * created only after 3 way handshake is done.
790 		 */
791 		sock_orphan(child);
792 		percpu_counter_inc((child)->sk_prot->orphan_count);
793 		chtls_release_resources(child);
794 		chtls_conn_done(child);
795 	} else {
796 		if (csk_flag(child, CSK_RST_ABORTED)) {
797 			chtls_release_resources(child);
798 			chtls_conn_done(child);
799 		}
800 	}
801 }
802 
pass_open_abort(struct sock * child,struct sock * parent,struct sk_buff * skb)803 static void pass_open_abort(struct sock *child, struct sock *parent,
804 			    struct sk_buff *skb)
805 {
806 	do_abort_syn_rcv(child, parent);
807 	kfree_skb(skb);
808 }
809 
bl_pass_open_abort(struct sock * lsk,struct sk_buff * skb)810 static void bl_pass_open_abort(struct sock *lsk, struct sk_buff *skb)
811 {
812 	pass_open_abort(skb->sk, lsk, skb);
813 }
814 
chtls_pass_open_arp_failure(struct sock * sk,struct sk_buff * skb)815 static void chtls_pass_open_arp_failure(struct sock *sk,
816 					struct sk_buff *skb)
817 {
818 	const struct request_sock *oreq;
819 	struct chtls_sock *csk;
820 	struct chtls_dev *cdev;
821 	struct sock *parent;
822 	void *data;
823 
824 	csk = rcu_dereference_sk_user_data(sk);
825 	cdev = csk->cdev;
826 
827 	/*
828 	 * If the connection is being aborted due to the parent listening
829 	 * socket going away there's nothing to do, the ABORT_REQ will close
830 	 * the connection.
831 	 */
832 	if (csk_flag(sk, CSK_ABORT_RPL_PENDING)) {
833 		kfree_skb(skb);
834 		return;
835 	}
836 
837 	oreq = csk->passive_reap_next;
838 	data = lookup_stid(cdev->tids, oreq->ts_recent);
839 	parent = ((struct listen_ctx *)data)->lsk;
840 
841 	bh_lock_sock(parent);
842 	if (!sock_owned_by_user(parent)) {
843 		pass_open_abort(sk, parent, skb);
844 	} else {
845 		BLOG_SKB_CB(skb)->backlog_rcv = bl_pass_open_abort;
846 		__sk_add_backlog(parent, skb);
847 	}
848 	bh_unlock_sock(parent);
849 }
850 
chtls_accept_rpl_arp_failure(void * handle,struct sk_buff * skb)851 static void chtls_accept_rpl_arp_failure(void *handle,
852 					 struct sk_buff *skb)
853 {
854 	struct sock *sk = (struct sock *)handle;
855 
856 	sock_hold(sk);
857 	process_cpl_msg(chtls_pass_open_arp_failure, sk, skb);
858 	sock_put(sk);
859 }
860 
chtls_select_mss(const struct chtls_sock * csk,unsigned int pmtu,struct cpl_pass_accept_req * req)861 static unsigned int chtls_select_mss(const struct chtls_sock *csk,
862 				     unsigned int pmtu,
863 				     struct cpl_pass_accept_req *req)
864 {
865 	struct chtls_dev *cdev;
866 	struct dst_entry *dst;
867 	unsigned int tcpoptsz;
868 	unsigned int iphdrsz;
869 	unsigned int mtu_idx;
870 	struct tcp_sock *tp;
871 	unsigned int mss;
872 	struct sock *sk;
873 
874 	mss = ntohs(req->tcpopt.mss);
875 	sk = csk->sk;
876 	dst = __sk_dst_get(sk);
877 	cdev = csk->cdev;
878 	tp = tcp_sk(sk);
879 	tcpoptsz = 0;
880 
881 	iphdrsz = sizeof(struct iphdr) + sizeof(struct tcphdr);
882 	if (req->tcpopt.tstamp)
883 		tcpoptsz += round_up(TCPOLEN_TIMESTAMP, 4);
884 
885 	tp->advmss = dst_metric_advmss(dst);
886 	if (USER_MSS(tp) && tp->advmss > USER_MSS(tp))
887 		tp->advmss = USER_MSS(tp);
888 	if (tp->advmss > pmtu - iphdrsz)
889 		tp->advmss = pmtu - iphdrsz;
890 	if (mss && tp->advmss > mss)
891 		tp->advmss = mss;
892 
893 	tp->advmss = cxgb4_best_aligned_mtu(cdev->lldi->mtus,
894 					    iphdrsz + tcpoptsz,
895 					    tp->advmss - tcpoptsz,
896 					    8, &mtu_idx);
897 	tp->advmss -= iphdrsz;
898 
899 	inet_csk(sk)->icsk_pmtu_cookie = pmtu;
900 	return mtu_idx;
901 }
902 
select_rcv_wnd(struct chtls_sock * csk)903 static unsigned int select_rcv_wnd(struct chtls_sock *csk)
904 {
905 	unsigned int rcvwnd;
906 	unsigned int wnd;
907 	struct sock *sk;
908 
909 	sk = csk->sk;
910 	wnd = tcp_full_space(sk);
911 
912 	if (wnd < MIN_RCV_WND)
913 		wnd = MIN_RCV_WND;
914 
915 	rcvwnd = MAX_RCV_WND;
916 
917 	csk_set_flag(csk, CSK_UPDATE_RCV_WND);
918 	return min(wnd, rcvwnd);
919 }
920 
select_rcv_wscale(int space,int wscale_ok,int win_clamp)921 static unsigned int select_rcv_wscale(int space, int wscale_ok, int win_clamp)
922 {
923 	int wscale = 0;
924 
925 	if (space > MAX_RCV_WND)
926 		space = MAX_RCV_WND;
927 	if (win_clamp && win_clamp < space)
928 		space = win_clamp;
929 
930 	if (wscale_ok) {
931 		while (wscale < 14 && (65535 << wscale) < space)
932 			wscale++;
933 	}
934 	return wscale;
935 }
936 
chtls_pass_accept_rpl(struct sk_buff * skb,struct cpl_pass_accept_req * req,unsigned int tid)937 static void chtls_pass_accept_rpl(struct sk_buff *skb,
938 				  struct cpl_pass_accept_req *req,
939 				  unsigned int tid)
940 
941 {
942 	struct cpl_t5_pass_accept_rpl *rpl5;
943 	struct cxgb4_lld_info *lldi;
944 	const struct tcphdr *tcph;
945 	const struct tcp_sock *tp;
946 	struct chtls_sock *csk;
947 	unsigned int len;
948 	struct sock *sk;
949 	u32 opt2, hlen;
950 	u64 opt0;
951 
952 	sk = skb->sk;
953 	tp = tcp_sk(sk);
954 	csk = sk->sk_user_data;
955 	csk->tid = tid;
956 	lldi = csk->cdev->lldi;
957 	len = roundup(sizeof(*rpl5), 16);
958 
959 	rpl5 = __skb_put_zero(skb, len);
960 	INIT_TP_WR(rpl5, tid);
961 
962 	OPCODE_TID(rpl5) = cpu_to_be32(MK_OPCODE_TID(CPL_PASS_ACCEPT_RPL,
963 						     csk->tid));
964 	csk->mtu_idx = chtls_select_mss(csk, dst_mtu(__sk_dst_get(sk)),
965 					req);
966 	opt0 = TCAM_BYPASS_F |
967 	       WND_SCALE_V((tp)->rx_opt.rcv_wscale) |
968 	       MSS_IDX_V(csk->mtu_idx) |
969 	       L2T_IDX_V(csk->l2t_entry->idx) |
970 	       NAGLE_V(!(tp->nonagle & TCP_NAGLE_OFF)) |
971 	       TX_CHAN_V(csk->tx_chan) |
972 	       SMAC_SEL_V(csk->smac_idx) |
973 	       DSCP_V(csk->tos >> 2) |
974 	       ULP_MODE_V(ULP_MODE_TLS) |
975 	       RCV_BUFSIZ_V(min(tp->rcv_wnd >> 10, RCV_BUFSIZ_M));
976 
977 	opt2 = RX_CHANNEL_V(0) |
978 		RSS_QUEUE_VALID_F | RSS_QUEUE_V(csk->rss_qid);
979 
980 	if (!is_t5(lldi->adapter_type))
981 		opt2 |= RX_FC_DISABLE_F;
982 	if (req->tcpopt.tstamp)
983 		opt2 |= TSTAMPS_EN_F;
984 	if (req->tcpopt.sack)
985 		opt2 |= SACK_EN_F;
986 	hlen = ntohl(req->hdr_len);
987 
988 	tcph = (struct tcphdr *)((u8 *)(req + 1) +
989 			T6_ETH_HDR_LEN_G(hlen) + T6_IP_HDR_LEN_G(hlen));
990 	if (tcph->ece && tcph->cwr)
991 		opt2 |= CCTRL_ECN_V(1);
992 	opt2 |= CONG_CNTRL_V(CONG_ALG_NEWRENO);
993 	opt2 |= T5_ISS_F;
994 	opt2 |= T5_OPT_2_VALID_F;
995 	rpl5->opt0 = cpu_to_be64(opt0);
996 	rpl5->opt2 = cpu_to_be32(opt2);
997 	rpl5->iss = cpu_to_be32((prandom_u32() & ~7UL) - 1);
998 	set_wr_txq(skb, CPL_PRIORITY_SETUP, csk->port_id);
999 	t4_set_arp_err_handler(skb, sk, chtls_accept_rpl_arp_failure);
1000 	cxgb4_l2t_send(csk->egress_dev, skb, csk->l2t_entry);
1001 }
1002 
inet_inherit_port(struct inet_hashinfo * hash_info,struct sock * lsk,struct sock * newsk)1003 static void inet_inherit_port(struct inet_hashinfo *hash_info,
1004 			      struct sock *lsk, struct sock *newsk)
1005 {
1006 	local_bh_disable();
1007 	__inet_inherit_port(lsk, newsk);
1008 	local_bh_enable();
1009 }
1010 
chtls_backlog_rcv(struct sock * sk,struct sk_buff * skb)1011 static int chtls_backlog_rcv(struct sock *sk, struct sk_buff *skb)
1012 {
1013 	if (skb->protocol) {
1014 		kfree_skb(skb);
1015 		return 0;
1016 	}
1017 	BLOG_SKB_CB(skb)->backlog_rcv(sk, skb);
1018 	return 0;
1019 }
1020 
chtls_recv_sock(struct sock * lsk,struct request_sock * oreq,void * network_hdr,const struct cpl_pass_accept_req * req,struct chtls_dev * cdev)1021 static struct sock *chtls_recv_sock(struct sock *lsk,
1022 				    struct request_sock *oreq,
1023 				    void *network_hdr,
1024 				    const struct cpl_pass_accept_req *req,
1025 				    struct chtls_dev *cdev)
1026 {
1027 	const struct tcphdr *tcph;
1028 	struct inet_sock *newinet;
1029 	const struct iphdr *iph;
1030 	struct net_device *ndev;
1031 	struct chtls_sock *csk;
1032 	struct dst_entry *dst;
1033 	struct neighbour *n;
1034 	struct tcp_sock *tp;
1035 	struct sock *newsk;
1036 	u16 port_id;
1037 	int rxq_idx;
1038 	int step;
1039 
1040 	iph = (const struct iphdr *)network_hdr;
1041 	newsk = tcp_create_openreq_child(lsk, oreq, cdev->askb);
1042 	if (!newsk)
1043 		goto free_oreq;
1044 
1045 	dst = inet_csk_route_child_sock(lsk, newsk, oreq);
1046 	if (!dst)
1047 		goto free_sk;
1048 
1049 	tcph = (struct tcphdr *)(iph + 1);
1050 	n = dst_neigh_lookup(dst, &iph->saddr);
1051 	if (!n)
1052 		goto free_sk;
1053 
1054 	ndev = n->dev;
1055 	if (!ndev)
1056 		goto free_dst;
1057 	if (is_vlan_dev(ndev))
1058 		ndev = vlan_dev_real_dev(ndev);
1059 
1060 	port_id = cxgb4_port_idx(ndev);
1061 
1062 	csk = chtls_sock_create(cdev);
1063 	if (!csk)
1064 		goto free_dst;
1065 
1066 	csk->l2t_entry = cxgb4_l2t_get(cdev->lldi->l2t, n, ndev, 0);
1067 	if (!csk->l2t_entry)
1068 		goto free_csk;
1069 
1070 	newsk->sk_user_data = csk;
1071 	newsk->sk_backlog_rcv = chtls_backlog_rcv;
1072 
1073 	tp = tcp_sk(newsk);
1074 	newinet = inet_sk(newsk);
1075 
1076 	newinet->inet_daddr = iph->saddr;
1077 	newinet->inet_rcv_saddr = iph->daddr;
1078 	newinet->inet_saddr = iph->daddr;
1079 
1080 	oreq->ts_recent = PASS_OPEN_TID_G(ntohl(req->tos_stid));
1081 	sk_setup_caps(newsk, dst);
1082 	csk->sk = newsk;
1083 	csk->passive_reap_next = oreq;
1084 	csk->tx_chan = cxgb4_port_chan(ndev);
1085 	csk->port_id = port_id;
1086 	csk->egress_dev = ndev;
1087 	csk->tos = PASS_OPEN_TOS_G(ntohl(req->tos_stid));
1088 	csk->ulp_mode = ULP_MODE_TLS;
1089 	step = cdev->lldi->nrxq / cdev->lldi->nchan;
1090 	csk->rss_qid = cdev->lldi->rxq_ids[port_id * step];
1091 	rxq_idx = port_id * step;
1092 	csk->txq_idx = (rxq_idx < cdev->lldi->ntxq) ? rxq_idx :
1093 			port_id * step;
1094 	csk->sndbuf = newsk->sk_sndbuf;
1095 	csk->smac_idx = cxgb4_tp_smt_idx(cdev->lldi->adapter_type,
1096 					 cxgb4_port_viid(ndev));
1097 	tp->rcv_wnd = select_rcv_wnd(csk);
1098 	RCV_WSCALE(tp) = select_rcv_wscale(tcp_full_space(newsk),
1099 					   WSCALE_OK(tp),
1100 					   tp->window_clamp);
1101 	neigh_release(n);
1102 	inet_inherit_port(&tcp_hashinfo, lsk, newsk);
1103 	csk_set_flag(csk, CSK_CONN_INLINE);
1104 	bh_unlock_sock(newsk); /* tcp_create_openreq_child ->sk_clone_lock */
1105 
1106 	return newsk;
1107 free_csk:
1108 	chtls_sock_release(&csk->kref);
1109 free_dst:
1110 	dst_release(dst);
1111 free_sk:
1112 	inet_csk_prepare_forced_close(newsk);
1113 	tcp_done(newsk);
1114 free_oreq:
1115 	chtls_reqsk_free(oreq);
1116 	return NULL;
1117 }
1118 
1119 /*
1120  * Populate a TID_RELEASE WR.  The skb must be already propely sized.
1121  */
mk_tid_release(struct sk_buff * skb,unsigned int chan,unsigned int tid)1122 static  void mk_tid_release(struct sk_buff *skb,
1123 			    unsigned int chan, unsigned int tid)
1124 {
1125 	struct cpl_tid_release *req;
1126 	unsigned int len;
1127 
1128 	len = roundup(sizeof(struct cpl_tid_release), 16);
1129 	req = (struct cpl_tid_release *)__skb_put(skb, len);
1130 	memset(req, 0, len);
1131 	set_wr_txq(skb, CPL_PRIORITY_SETUP, chan);
1132 	INIT_TP_WR_CPL(req, CPL_TID_RELEASE, tid);
1133 }
1134 
chtls_get_module(struct sock * sk)1135 static int chtls_get_module(struct sock *sk)
1136 {
1137 	struct inet_connection_sock *icsk = inet_csk(sk);
1138 
1139 	if (!try_module_get(icsk->icsk_ulp_ops->owner))
1140 		return -1;
1141 
1142 	return 0;
1143 }
1144 
chtls_pass_accept_request(struct sock * sk,struct sk_buff * skb)1145 static void chtls_pass_accept_request(struct sock *sk,
1146 				      struct sk_buff *skb)
1147 {
1148 	struct cpl_t5_pass_accept_rpl *rpl;
1149 	struct cpl_pass_accept_req *req;
1150 	struct listen_ctx *listen_ctx;
1151 	struct request_sock *oreq;
1152 	struct sk_buff *reply_skb;
1153 	struct chtls_sock *csk;
1154 	struct chtls_dev *cdev;
1155 	struct tcphdr *tcph;
1156 	struct sock *newsk;
1157 	struct ethhdr *eh;
1158 	struct iphdr *iph;
1159 	void *network_hdr;
1160 	unsigned int stid;
1161 	unsigned int len;
1162 	unsigned int tid;
1163 
1164 	req = cplhdr(skb) + RSS_HDR;
1165 	tid = GET_TID(req);
1166 	cdev = BLOG_SKB_CB(skb)->cdev;
1167 	newsk = lookup_tid(cdev->tids, tid);
1168 	stid = PASS_OPEN_TID_G(ntohl(req->tos_stid));
1169 	if (newsk) {
1170 		pr_info("tid (%d) already in use\n", tid);
1171 		return;
1172 	}
1173 
1174 	len = roundup(sizeof(*rpl), 16);
1175 	reply_skb = alloc_skb(len, GFP_ATOMIC);
1176 	if (!reply_skb) {
1177 		cxgb4_remove_tid(cdev->tids, 0, tid, sk->sk_family);
1178 		kfree_skb(skb);
1179 		return;
1180 	}
1181 
1182 	if (sk->sk_state != TCP_LISTEN)
1183 		goto reject;
1184 
1185 	if (inet_csk_reqsk_queue_is_full(sk))
1186 		goto reject;
1187 
1188 	if (sk_acceptq_is_full(sk))
1189 		goto reject;
1190 
1191 	oreq = inet_reqsk_alloc(&chtls_rsk_ops, sk, true);
1192 	if (!oreq)
1193 		goto reject;
1194 
1195 	oreq->rsk_rcv_wnd = 0;
1196 	oreq->rsk_window_clamp = 0;
1197 	oreq->cookie_ts = 0;
1198 	oreq->mss = 0;
1199 	oreq->ts_recent = 0;
1200 
1201 	eh = (struct ethhdr *)(req + 1);
1202 	iph = (struct iphdr *)(eh + 1);
1203 	if (iph->version != 0x4)
1204 		goto free_oreq;
1205 
1206 	network_hdr = (void *)(eh + 1);
1207 	tcph = (struct tcphdr *)(iph + 1);
1208 
1209 	tcp_rsk(oreq)->tfo_listener = false;
1210 	tcp_rsk(oreq)->rcv_isn = ntohl(tcph->seq);
1211 	chtls_set_req_port(oreq, tcph->source, tcph->dest);
1212 	inet_rsk(oreq)->ecn_ok = 0;
1213 	chtls_set_req_addr(oreq, iph->daddr, iph->saddr);
1214 	if (req->tcpopt.wsf <= 14) {
1215 		inet_rsk(oreq)->wscale_ok = 1;
1216 		inet_rsk(oreq)->snd_wscale = req->tcpopt.wsf;
1217 	}
1218 	inet_rsk(oreq)->ir_iif = sk->sk_bound_dev_if;
1219 
1220 	newsk = chtls_recv_sock(sk, oreq, network_hdr, req, cdev);
1221 	if (!newsk)
1222 		goto reject;
1223 
1224 	if (chtls_get_module(newsk))
1225 		goto reject;
1226 	inet_csk_reqsk_queue_added(sk);
1227 	reply_skb->sk = newsk;
1228 	chtls_install_cpl_ops(newsk);
1229 	cxgb4_insert_tid(cdev->tids, newsk, tid, newsk->sk_family);
1230 	csk = rcu_dereference_sk_user_data(newsk);
1231 	listen_ctx = (struct listen_ctx *)lookup_stid(cdev->tids, stid);
1232 	csk->listen_ctx = listen_ctx;
1233 	__skb_queue_tail(&listen_ctx->synq, (struct sk_buff *)&csk->synq);
1234 	chtls_pass_accept_rpl(reply_skb, req, tid);
1235 	kfree_skb(skb);
1236 	return;
1237 
1238 free_oreq:
1239 	chtls_reqsk_free(oreq);
1240 reject:
1241 	mk_tid_release(reply_skb, 0, tid);
1242 	cxgb4_ofld_send(cdev->lldi->ports[0], reply_skb);
1243 	kfree_skb(skb);
1244 }
1245 
1246 /*
1247  * Handle a CPL_PASS_ACCEPT_REQ message.
1248  */
chtls_pass_accept_req(struct chtls_dev * cdev,struct sk_buff * skb)1249 static int chtls_pass_accept_req(struct chtls_dev *cdev, struct sk_buff *skb)
1250 {
1251 	struct cpl_pass_accept_req *req = cplhdr(skb) + RSS_HDR;
1252 	struct listen_ctx *ctx;
1253 	unsigned int stid;
1254 	unsigned int tid;
1255 	struct sock *lsk;
1256 	void *data;
1257 
1258 	stid = PASS_OPEN_TID_G(ntohl(req->tos_stid));
1259 	tid = GET_TID(req);
1260 
1261 	data = lookup_stid(cdev->tids, stid);
1262 	if (!data)
1263 		return 1;
1264 
1265 	ctx = (struct listen_ctx *)data;
1266 	lsk = ctx->lsk;
1267 
1268 	if (unlikely(tid >= cdev->tids->ntids)) {
1269 		pr_info("passive open TID %u too large\n", tid);
1270 		return 1;
1271 	}
1272 
1273 	BLOG_SKB_CB(skb)->cdev = cdev;
1274 	process_cpl_msg(chtls_pass_accept_request, lsk, skb);
1275 	return 0;
1276 }
1277 
1278 /*
1279  * Completes some final bits of initialization for just established connections
1280  * and changes their state to TCP_ESTABLISHED.
1281  *
1282  * snd_isn here is the ISN after the SYN, i.e., the true ISN + 1.
1283  */
make_established(struct sock * sk,u32 snd_isn,unsigned int opt)1284 static void make_established(struct sock *sk, u32 snd_isn, unsigned int opt)
1285 {
1286 	struct tcp_sock *tp = tcp_sk(sk);
1287 
1288 	tp->pushed_seq = snd_isn;
1289 	tp->write_seq = snd_isn;
1290 	tp->snd_nxt = snd_isn;
1291 	tp->snd_una = snd_isn;
1292 	inet_sk(sk)->inet_id = prandom_u32();
1293 	assign_rxopt(sk, opt);
1294 
1295 	if (tp->rcv_wnd > (RCV_BUFSIZ_M << 10))
1296 		tp->rcv_wup -= tp->rcv_wnd - (RCV_BUFSIZ_M << 10);
1297 
1298 	smp_mb();
1299 	tcp_set_state(sk, TCP_ESTABLISHED);
1300 }
1301 
chtls_abort_conn(struct sock * sk,struct sk_buff * skb)1302 static void chtls_abort_conn(struct sock *sk, struct sk_buff *skb)
1303 {
1304 	struct sk_buff *abort_skb;
1305 
1306 	abort_skb = alloc_skb(sizeof(struct cpl_abort_req), GFP_ATOMIC);
1307 	if (abort_skb)
1308 		chtls_send_reset(sk, CPL_ABORT_SEND_RST, abort_skb);
1309 }
1310 
1311 static struct sock *reap_list;
1312 static DEFINE_SPINLOCK(reap_list_lock);
1313 
1314 /*
1315  * Process the reap list.
1316  */
DECLARE_TASK_FUNC(process_reap_list,task_param)1317 DECLARE_TASK_FUNC(process_reap_list, task_param)
1318 {
1319 	spin_lock_bh(&reap_list_lock);
1320 	while (reap_list) {
1321 		struct sock *sk = reap_list;
1322 		struct chtls_sock *csk = rcu_dereference_sk_user_data(sk);
1323 
1324 		reap_list = csk->passive_reap_next;
1325 		csk->passive_reap_next = NULL;
1326 		spin_unlock(&reap_list_lock);
1327 		sock_hold(sk);
1328 
1329 		bh_lock_sock(sk);
1330 		chtls_abort_conn(sk, NULL);
1331 		sock_orphan(sk);
1332 		if (sk->sk_state == TCP_CLOSE)
1333 			inet_csk_destroy_sock(sk);
1334 		bh_unlock_sock(sk);
1335 		sock_put(sk);
1336 		spin_lock(&reap_list_lock);
1337 	}
1338 	spin_unlock_bh(&reap_list_lock);
1339 }
1340 
1341 static DECLARE_WORK(reap_task, process_reap_list);
1342 
add_to_reap_list(struct sock * sk)1343 static void add_to_reap_list(struct sock *sk)
1344 {
1345 	struct chtls_sock *csk = sk->sk_user_data;
1346 
1347 	local_bh_disable();
1348 	release_tcp_port(sk); /* release the port immediately */
1349 
1350 	spin_lock(&reap_list_lock);
1351 	csk->passive_reap_next = reap_list;
1352 	reap_list = sk;
1353 	if (!csk->passive_reap_next)
1354 		schedule_work(&reap_task);
1355 	spin_unlock(&reap_list_lock);
1356 	local_bh_enable();
1357 }
1358 
add_pass_open_to_parent(struct sock * child,struct sock * lsk,struct chtls_dev * cdev)1359 static void add_pass_open_to_parent(struct sock *child, struct sock *lsk,
1360 				    struct chtls_dev *cdev)
1361 {
1362 	struct request_sock *oreq;
1363 	struct chtls_sock *csk;
1364 
1365 	if (lsk->sk_state != TCP_LISTEN)
1366 		return;
1367 
1368 	csk = child->sk_user_data;
1369 	oreq = csk->passive_reap_next;
1370 	csk->passive_reap_next = NULL;
1371 
1372 	reqsk_queue_removed(&inet_csk(lsk)->icsk_accept_queue, oreq);
1373 	__skb_unlink((struct sk_buff *)&csk->synq, &csk->listen_ctx->synq);
1374 
1375 	if (sk_acceptq_is_full(lsk)) {
1376 		chtls_reqsk_free(oreq);
1377 		add_to_reap_list(child);
1378 	} else {
1379 		refcount_set(&oreq->rsk_refcnt, 1);
1380 		inet_csk_reqsk_queue_add(lsk, oreq, child);
1381 		lsk->sk_data_ready(lsk);
1382 	}
1383 }
1384 
bl_add_pass_open_to_parent(struct sock * lsk,struct sk_buff * skb)1385 static void bl_add_pass_open_to_parent(struct sock *lsk, struct sk_buff *skb)
1386 {
1387 	struct sock *child = skb->sk;
1388 
1389 	skb->sk = NULL;
1390 	add_pass_open_to_parent(child, lsk, BLOG_SKB_CB(skb)->cdev);
1391 	kfree_skb(skb);
1392 }
1393 
chtls_pass_establish(struct chtls_dev * cdev,struct sk_buff * skb)1394 static int chtls_pass_establish(struct chtls_dev *cdev, struct sk_buff *skb)
1395 {
1396 	struct cpl_pass_establish *req = cplhdr(skb) + RSS_HDR;
1397 	struct chtls_sock *csk;
1398 	struct sock *lsk, *sk;
1399 	unsigned int hwtid;
1400 
1401 	hwtid = GET_TID(req);
1402 	sk = lookup_tid(cdev->tids, hwtid);
1403 	if (!sk)
1404 		return (CPL_RET_UNKNOWN_TID | CPL_RET_BUF_DONE);
1405 
1406 	bh_lock_sock(sk);
1407 	if (unlikely(sock_owned_by_user(sk))) {
1408 		kfree_skb(skb);
1409 	} else {
1410 		unsigned int stid;
1411 		void *data;
1412 
1413 		csk = sk->sk_user_data;
1414 		csk->wr_max_credits = 64;
1415 		csk->wr_credits = 64;
1416 		csk->wr_unacked = 0;
1417 		make_established(sk, ntohl(req->snd_isn), ntohs(req->tcp_opt));
1418 		stid = PASS_OPEN_TID_G(ntohl(req->tos_stid));
1419 		sk->sk_state_change(sk);
1420 		if (unlikely(sk->sk_socket))
1421 			sk_wake_async(sk, 0, POLL_OUT);
1422 
1423 		data = lookup_stid(cdev->tids, stid);
1424 		lsk = ((struct listen_ctx *)data)->lsk;
1425 
1426 		bh_lock_sock(lsk);
1427 		if (unlikely(skb_queue_empty(&csk->listen_ctx->synq))) {
1428 			/* removed from synq */
1429 			bh_unlock_sock(lsk);
1430 			kfree_skb(skb);
1431 			goto unlock;
1432 		}
1433 
1434 		if (likely(!sock_owned_by_user(lsk))) {
1435 			kfree_skb(skb);
1436 			add_pass_open_to_parent(sk, lsk, cdev);
1437 		} else {
1438 			skb->sk = sk;
1439 			BLOG_SKB_CB(skb)->cdev = cdev;
1440 			BLOG_SKB_CB(skb)->backlog_rcv =
1441 				bl_add_pass_open_to_parent;
1442 			__sk_add_backlog(lsk, skb);
1443 		}
1444 		bh_unlock_sock(lsk);
1445 	}
1446 unlock:
1447 	bh_unlock_sock(sk);
1448 	return 0;
1449 }
1450 
1451 /*
1452  * Handle receipt of an urgent pointer.
1453  */
handle_urg_ptr(struct sock * sk,u32 urg_seq)1454 static void handle_urg_ptr(struct sock *sk, u32 urg_seq)
1455 {
1456 	struct tcp_sock *tp = tcp_sk(sk);
1457 
1458 	urg_seq--;
1459 	if (tp->urg_data && !after(urg_seq, tp->urg_seq))
1460 		return;	/* duplicate pointer */
1461 
1462 	sk_send_sigurg(sk);
1463 	if (tp->urg_seq == tp->copied_seq && tp->urg_data &&
1464 	    !sock_flag(sk, SOCK_URGINLINE) &&
1465 	    tp->copied_seq != tp->rcv_nxt) {
1466 		struct sk_buff *skb = skb_peek(&sk->sk_receive_queue);
1467 
1468 		tp->copied_seq++;
1469 		if (skb && tp->copied_seq - ULP_SKB_CB(skb)->seq >= skb->len)
1470 			chtls_free_skb(sk, skb);
1471 	}
1472 
1473 	tp->urg_data = TCP_URG_NOTYET;
1474 	tp->urg_seq = urg_seq;
1475 }
1476 
check_sk_callbacks(struct chtls_sock * csk)1477 static void check_sk_callbacks(struct chtls_sock *csk)
1478 {
1479 	struct sock *sk = csk->sk;
1480 
1481 	if (unlikely(sk->sk_user_data &&
1482 		     !csk_flag_nochk(csk, CSK_CALLBACKS_CHKD)))
1483 		csk_set_flag(csk, CSK_CALLBACKS_CHKD);
1484 }
1485 
1486 /*
1487  * Handles Rx data that arrives in a state where the socket isn't accepting
1488  * new data.
1489  */
handle_excess_rx(struct sock * sk,struct sk_buff * skb)1490 static void handle_excess_rx(struct sock *sk, struct sk_buff *skb)
1491 {
1492 	if (!csk_flag(sk, CSK_ABORT_SHUTDOWN))
1493 		chtls_abort_conn(sk, skb);
1494 
1495 	kfree_skb(skb);
1496 }
1497 
chtls_recv_data(struct sock * sk,struct sk_buff * skb)1498 static void chtls_recv_data(struct sock *sk, struct sk_buff *skb)
1499 {
1500 	struct cpl_rx_data *hdr = cplhdr(skb) + RSS_HDR;
1501 	struct chtls_sock *csk;
1502 	struct tcp_sock *tp;
1503 
1504 	csk = rcu_dereference_sk_user_data(sk);
1505 	tp = tcp_sk(sk);
1506 
1507 	if (unlikely(sk->sk_shutdown & RCV_SHUTDOWN)) {
1508 		handle_excess_rx(sk, skb);
1509 		return;
1510 	}
1511 
1512 	ULP_SKB_CB(skb)->seq = ntohl(hdr->seq);
1513 	ULP_SKB_CB(skb)->psh = hdr->psh;
1514 	skb_ulp_mode(skb) = ULP_MODE_NONE;
1515 
1516 	skb_reset_transport_header(skb);
1517 	__skb_pull(skb, sizeof(*hdr) + RSS_HDR);
1518 	if (!skb->data_len)
1519 		__skb_trim(skb, ntohs(hdr->len));
1520 
1521 	if (unlikely(hdr->urg))
1522 		handle_urg_ptr(sk, tp->rcv_nxt + ntohs(hdr->urg));
1523 	if (unlikely(tp->urg_data == TCP_URG_NOTYET &&
1524 		     tp->urg_seq - tp->rcv_nxt < skb->len))
1525 		tp->urg_data = TCP_URG_VALID |
1526 			       skb->data[tp->urg_seq - tp->rcv_nxt];
1527 
1528 	if (unlikely(hdr->dack_mode != csk->delack_mode)) {
1529 		csk->delack_mode = hdr->dack_mode;
1530 		csk->delack_seq = tp->rcv_nxt;
1531 	}
1532 
1533 	tcp_hdr(skb)->fin = 0;
1534 	tp->rcv_nxt += skb->len;
1535 
1536 	__skb_queue_tail(&sk->sk_receive_queue, skb);
1537 
1538 	if (!sock_flag(sk, SOCK_DEAD)) {
1539 		check_sk_callbacks(csk);
1540 		sk->sk_data_ready(sk);
1541 	}
1542 }
1543 
chtls_rx_data(struct chtls_dev * cdev,struct sk_buff * skb)1544 static int chtls_rx_data(struct chtls_dev *cdev, struct sk_buff *skb)
1545 {
1546 	struct cpl_rx_data *req = cplhdr(skb) + RSS_HDR;
1547 	unsigned int hwtid = GET_TID(req);
1548 	struct sock *sk;
1549 
1550 	sk = lookup_tid(cdev->tids, hwtid);
1551 	if (unlikely(!sk)) {
1552 		pr_err("can't find conn. for hwtid %u.\n", hwtid);
1553 		return -EINVAL;
1554 	}
1555 	skb_dst_set(skb, NULL);
1556 	process_cpl_msg(chtls_recv_data, sk, skb);
1557 	return 0;
1558 }
1559 
chtls_recv_pdu(struct sock * sk,struct sk_buff * skb)1560 static void chtls_recv_pdu(struct sock *sk, struct sk_buff *skb)
1561 {
1562 	struct cpl_tls_data *hdr = cplhdr(skb);
1563 	struct chtls_sock *csk;
1564 	struct chtls_hws *tlsk;
1565 	struct tcp_sock *tp;
1566 
1567 	csk = rcu_dereference_sk_user_data(sk);
1568 	tlsk = &csk->tlshws;
1569 	tp = tcp_sk(sk);
1570 
1571 	if (unlikely(sk->sk_shutdown & RCV_SHUTDOWN)) {
1572 		handle_excess_rx(sk, skb);
1573 		return;
1574 	}
1575 
1576 	ULP_SKB_CB(skb)->seq = ntohl(hdr->seq);
1577 	ULP_SKB_CB(skb)->flags = 0;
1578 	skb_ulp_mode(skb) = ULP_MODE_TLS;
1579 
1580 	skb_reset_transport_header(skb);
1581 	__skb_pull(skb, sizeof(*hdr));
1582 	if (!skb->data_len)
1583 		__skb_trim(skb,
1584 			   CPL_TLS_DATA_LENGTH_G(ntohl(hdr->length_pkd)));
1585 
1586 	if (unlikely(tp->urg_data == TCP_URG_NOTYET && tp->urg_seq -
1587 		     tp->rcv_nxt < skb->len))
1588 		tp->urg_data = TCP_URG_VALID |
1589 			       skb->data[tp->urg_seq - tp->rcv_nxt];
1590 
1591 	tcp_hdr(skb)->fin = 0;
1592 	tlsk->pldlen = CPL_TLS_DATA_LENGTH_G(ntohl(hdr->length_pkd));
1593 	__skb_queue_tail(&tlsk->sk_recv_queue, skb);
1594 }
1595 
chtls_rx_pdu(struct chtls_dev * cdev,struct sk_buff * skb)1596 static int chtls_rx_pdu(struct chtls_dev *cdev, struct sk_buff *skb)
1597 {
1598 	struct cpl_tls_data *req = cplhdr(skb);
1599 	unsigned int hwtid = GET_TID(req);
1600 	struct sock *sk;
1601 
1602 	sk = lookup_tid(cdev->tids, hwtid);
1603 	if (unlikely(!sk)) {
1604 		pr_err("can't find conn. for hwtid %u.\n", hwtid);
1605 		return -EINVAL;
1606 	}
1607 	skb_dst_set(skb, NULL);
1608 	process_cpl_msg(chtls_recv_pdu, sk, skb);
1609 	return 0;
1610 }
1611 
chtls_set_hdrlen(struct sk_buff * skb,unsigned int nlen)1612 static void chtls_set_hdrlen(struct sk_buff *skb, unsigned int nlen)
1613 {
1614 	struct tlsrx_cmp_hdr *tls_cmp_hdr = cplhdr(skb);
1615 
1616 	skb->hdr_len = ntohs((__force __be16)tls_cmp_hdr->length);
1617 	tls_cmp_hdr->length = ntohs((__force __be16)nlen);
1618 }
1619 
chtls_rx_hdr(struct sock * sk,struct sk_buff * skb)1620 static void chtls_rx_hdr(struct sock *sk, struct sk_buff *skb)
1621 {
1622 	struct tlsrx_cmp_hdr *tls_hdr_pkt;
1623 	struct cpl_rx_tls_cmp *cmp_cpl;
1624 	struct sk_buff *skb_rec;
1625 	struct chtls_sock *csk;
1626 	struct chtls_hws *tlsk;
1627 	struct tcp_sock *tp;
1628 
1629 	cmp_cpl = cplhdr(skb);
1630 	csk = rcu_dereference_sk_user_data(sk);
1631 	tlsk = &csk->tlshws;
1632 	tp = tcp_sk(sk);
1633 
1634 	ULP_SKB_CB(skb)->seq = ntohl(cmp_cpl->seq);
1635 	ULP_SKB_CB(skb)->flags = 0;
1636 
1637 	skb_reset_transport_header(skb);
1638 	__skb_pull(skb, sizeof(*cmp_cpl));
1639 	tls_hdr_pkt = (struct tlsrx_cmp_hdr *)skb->data;
1640 	if (tls_hdr_pkt->res_to_mac_error & TLSRX_HDR_PKT_ERROR_M)
1641 		tls_hdr_pkt->type = CONTENT_TYPE_ERROR;
1642 	if (!skb->data_len)
1643 		__skb_trim(skb, TLS_HEADER_LENGTH);
1644 
1645 	tp->rcv_nxt +=
1646 		CPL_RX_TLS_CMP_PDULENGTH_G(ntohl(cmp_cpl->pdulength_length));
1647 
1648 	ULP_SKB_CB(skb)->flags |= ULPCB_FLAG_TLS_HDR;
1649 	skb_rec = __skb_dequeue(&tlsk->sk_recv_queue);
1650 	if (!skb_rec) {
1651 		__skb_queue_tail(&sk->sk_receive_queue, skb);
1652 	} else {
1653 		chtls_set_hdrlen(skb, tlsk->pldlen);
1654 		tlsk->pldlen = 0;
1655 		__skb_queue_tail(&sk->sk_receive_queue, skb);
1656 		__skb_queue_tail(&sk->sk_receive_queue, skb_rec);
1657 	}
1658 
1659 	if (!sock_flag(sk, SOCK_DEAD)) {
1660 		check_sk_callbacks(csk);
1661 		sk->sk_data_ready(sk);
1662 	}
1663 }
1664 
chtls_rx_cmp(struct chtls_dev * cdev,struct sk_buff * skb)1665 static int chtls_rx_cmp(struct chtls_dev *cdev, struct sk_buff *skb)
1666 {
1667 	struct cpl_rx_tls_cmp *req = cplhdr(skb);
1668 	unsigned int hwtid = GET_TID(req);
1669 	struct sock *sk;
1670 
1671 	sk = lookup_tid(cdev->tids, hwtid);
1672 	if (unlikely(!sk)) {
1673 		pr_err("can't find conn. for hwtid %u.\n", hwtid);
1674 		return -EINVAL;
1675 	}
1676 	skb_dst_set(skb, NULL);
1677 	process_cpl_msg(chtls_rx_hdr, sk, skb);
1678 
1679 	return 0;
1680 }
1681 
chtls_timewait(struct sock * sk)1682 static void chtls_timewait(struct sock *sk)
1683 {
1684 	struct tcp_sock *tp = tcp_sk(sk);
1685 
1686 	tp->rcv_nxt++;
1687 	tp->rx_opt.ts_recent_stamp = ktime_get_seconds();
1688 	tp->srtt_us = 0;
1689 	tcp_time_wait(sk, TCP_TIME_WAIT, 0);
1690 }
1691 
chtls_peer_close(struct sock * sk,struct sk_buff * skb)1692 static void chtls_peer_close(struct sock *sk, struct sk_buff *skb)
1693 {
1694 	struct chtls_sock *csk = rcu_dereference_sk_user_data(sk);
1695 
1696 	sk->sk_shutdown |= RCV_SHUTDOWN;
1697 	sock_set_flag(sk, SOCK_DONE);
1698 
1699 	switch (sk->sk_state) {
1700 	case TCP_SYN_RECV:
1701 	case TCP_ESTABLISHED:
1702 		tcp_set_state(sk, TCP_CLOSE_WAIT);
1703 		break;
1704 	case TCP_FIN_WAIT1:
1705 		tcp_set_state(sk, TCP_CLOSING);
1706 		break;
1707 	case TCP_FIN_WAIT2:
1708 		chtls_release_resources(sk);
1709 		if (csk_flag_nochk(csk, CSK_ABORT_RPL_PENDING))
1710 			chtls_conn_done(sk);
1711 		else
1712 			chtls_timewait(sk);
1713 		break;
1714 	default:
1715 		pr_info("cpl_peer_close in bad state %d\n", sk->sk_state);
1716 	}
1717 
1718 	if (!sock_flag(sk, SOCK_DEAD)) {
1719 		sk->sk_state_change(sk);
1720 		/* Do not send POLL_HUP for half duplex close. */
1721 
1722 		if ((sk->sk_shutdown & SEND_SHUTDOWN) ||
1723 		    sk->sk_state == TCP_CLOSE)
1724 			sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_HUP);
1725 		else
1726 			sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN);
1727 	}
1728 	kfree_skb(skb);
1729 }
1730 
chtls_close_con_rpl(struct sock * sk,struct sk_buff * skb)1731 static void chtls_close_con_rpl(struct sock *sk, struct sk_buff *skb)
1732 {
1733 	struct cpl_close_con_rpl *rpl = cplhdr(skb) + RSS_HDR;
1734 	struct chtls_sock *csk;
1735 	struct tcp_sock *tp;
1736 
1737 	csk = rcu_dereference_sk_user_data(sk);
1738 	tp = tcp_sk(sk);
1739 
1740 	tp->snd_una = ntohl(rpl->snd_nxt) - 1;  /* exclude FIN */
1741 
1742 	switch (sk->sk_state) {
1743 	case TCP_CLOSING:
1744 		chtls_release_resources(sk);
1745 		if (csk_flag_nochk(csk, CSK_ABORT_RPL_PENDING))
1746 			chtls_conn_done(sk);
1747 		else
1748 			chtls_timewait(sk);
1749 		break;
1750 	case TCP_LAST_ACK:
1751 		chtls_release_resources(sk);
1752 		chtls_conn_done(sk);
1753 		break;
1754 	case TCP_FIN_WAIT1:
1755 		tcp_set_state(sk, TCP_FIN_WAIT2);
1756 		sk->sk_shutdown |= SEND_SHUTDOWN;
1757 
1758 		if (!sock_flag(sk, SOCK_DEAD))
1759 			sk->sk_state_change(sk);
1760 		else if (tcp_sk(sk)->linger2 < 0 &&
1761 			 !csk_flag_nochk(csk, CSK_ABORT_SHUTDOWN))
1762 			chtls_abort_conn(sk, skb);
1763 		break;
1764 	default:
1765 		pr_info("close_con_rpl in bad state %d\n", sk->sk_state);
1766 	}
1767 	kfree_skb(skb);
1768 }
1769 
get_cpl_skb(struct sk_buff * skb,size_t len,gfp_t gfp)1770 static struct sk_buff *get_cpl_skb(struct sk_buff *skb,
1771 				   size_t len, gfp_t gfp)
1772 {
1773 	if (likely(!skb_is_nonlinear(skb) && !skb_cloned(skb))) {
1774 		WARN_ONCE(skb->len < len, "skb alloc error");
1775 		__skb_trim(skb, len);
1776 		skb_get(skb);
1777 	} else {
1778 		skb = alloc_skb(len, gfp);
1779 		if (skb)
1780 			__skb_put(skb, len);
1781 	}
1782 	return skb;
1783 }
1784 
set_abort_rpl_wr(struct sk_buff * skb,unsigned int tid,int cmd)1785 static void set_abort_rpl_wr(struct sk_buff *skb, unsigned int tid,
1786 			     int cmd)
1787 {
1788 	struct cpl_abort_rpl *rpl = cplhdr(skb);
1789 
1790 	INIT_TP_WR_CPL(rpl, CPL_ABORT_RPL, tid);
1791 	rpl->cmd = cmd;
1792 }
1793 
send_defer_abort_rpl(struct chtls_dev * cdev,struct sk_buff * skb)1794 static void send_defer_abort_rpl(struct chtls_dev *cdev, struct sk_buff *skb)
1795 {
1796 	struct cpl_abort_req_rss *req = cplhdr(skb);
1797 	struct sk_buff *reply_skb;
1798 
1799 	reply_skb = alloc_skb(sizeof(struct cpl_abort_rpl),
1800 			      GFP_KERNEL | __GFP_NOFAIL);
1801 	__skb_put(reply_skb, sizeof(struct cpl_abort_rpl));
1802 	set_abort_rpl_wr(reply_skb, GET_TID(req),
1803 			 (req->status & CPL_ABORT_NO_RST));
1804 	set_wr_txq(reply_skb, CPL_PRIORITY_DATA, req->status >> 1);
1805 	cxgb4_ofld_send(cdev->lldi->ports[0], reply_skb);
1806 	kfree_skb(skb);
1807 }
1808 
send_abort_rpl(struct sock * sk,struct sk_buff * skb,struct chtls_dev * cdev,int status,int queue)1809 static void send_abort_rpl(struct sock *sk, struct sk_buff *skb,
1810 			   struct chtls_dev *cdev, int status, int queue)
1811 {
1812 	struct cpl_abort_req_rss *req = cplhdr(skb);
1813 	struct sk_buff *reply_skb;
1814 	struct chtls_sock *csk;
1815 
1816 	csk = rcu_dereference_sk_user_data(sk);
1817 
1818 	reply_skb = alloc_skb(sizeof(struct cpl_abort_rpl),
1819 			      GFP_KERNEL);
1820 
1821 	if (!reply_skb) {
1822 		req->status = (queue << 1);
1823 		send_defer_abort_rpl(cdev, skb);
1824 		return;
1825 	}
1826 
1827 	set_abort_rpl_wr(reply_skb, GET_TID(req), status);
1828 	kfree_skb(skb);
1829 
1830 	set_wr_txq(reply_skb, CPL_PRIORITY_DATA, queue);
1831 	if (csk_conn_inline(csk)) {
1832 		struct l2t_entry *e = csk->l2t_entry;
1833 
1834 		if (e && sk->sk_state != TCP_SYN_RECV) {
1835 			cxgb4_l2t_send(csk->egress_dev, reply_skb, e);
1836 			return;
1837 		}
1838 	}
1839 	cxgb4_ofld_send(cdev->lldi->ports[0], reply_skb);
1840 }
1841 
1842 /*
1843  * Add an skb to the deferred skb queue for processing from process context.
1844  */
t4_defer_reply(struct sk_buff * skb,struct chtls_dev * cdev,defer_handler_t handler)1845 static void t4_defer_reply(struct sk_buff *skb, struct chtls_dev *cdev,
1846 			   defer_handler_t handler)
1847 {
1848 	DEFERRED_SKB_CB(skb)->handler = handler;
1849 	spin_lock_bh(&cdev->deferq.lock);
1850 	__skb_queue_tail(&cdev->deferq, skb);
1851 	if (skb_queue_len(&cdev->deferq) == 1)
1852 		schedule_work(&cdev->deferq_task);
1853 	spin_unlock_bh(&cdev->deferq.lock);
1854 }
1855 
chtls_send_abort_rpl(struct sock * sk,struct sk_buff * skb,struct chtls_dev * cdev,int status,int queue)1856 static void chtls_send_abort_rpl(struct sock *sk, struct sk_buff *skb,
1857 				 struct chtls_dev *cdev,
1858 				 int status, int queue)
1859 {
1860 	struct cpl_abort_req_rss *req = cplhdr(skb) + RSS_HDR;
1861 	struct sk_buff *reply_skb;
1862 	struct chtls_sock *csk;
1863 	unsigned int tid;
1864 
1865 	csk = rcu_dereference_sk_user_data(sk);
1866 	tid = GET_TID(req);
1867 
1868 	reply_skb = get_cpl_skb(skb, sizeof(struct cpl_abort_rpl), gfp_any());
1869 	if (!reply_skb) {
1870 		req->status = (queue << 1) | status;
1871 		t4_defer_reply(skb, cdev, send_defer_abort_rpl);
1872 		return;
1873 	}
1874 
1875 	set_abort_rpl_wr(reply_skb, tid, status);
1876 	set_wr_txq(reply_skb, CPL_PRIORITY_DATA, queue);
1877 	if (csk_conn_inline(csk)) {
1878 		struct l2t_entry *e = csk->l2t_entry;
1879 
1880 		if (e && sk->sk_state != TCP_SYN_RECV) {
1881 			cxgb4_l2t_send(csk->egress_dev, reply_skb, e);
1882 			return;
1883 		}
1884 	}
1885 	cxgb4_ofld_send(cdev->lldi->ports[0], reply_skb);
1886 	kfree_skb(skb);
1887 }
1888 
1889 /*
1890  * This is run from a listener's backlog to abort a child connection in
1891  * SYN_RCV state (i.e., one on the listener's SYN queue).
1892  */
bl_abort_syn_rcv(struct sock * lsk,struct sk_buff * skb)1893 static void bl_abort_syn_rcv(struct sock *lsk, struct sk_buff *skb)
1894 {
1895 	struct chtls_sock *csk;
1896 	struct sock *child;
1897 	int queue;
1898 
1899 	child = skb->sk;
1900 	csk = rcu_dereference_sk_user_data(child);
1901 	queue = csk->txq_idx;
1902 
1903 	skb->sk	= NULL;
1904 	do_abort_syn_rcv(child, lsk);
1905 	send_abort_rpl(child, skb, BLOG_SKB_CB(skb)->cdev,
1906 		       CPL_ABORT_NO_RST, queue);
1907 }
1908 
abort_syn_rcv(struct sock * sk,struct sk_buff * skb)1909 static int abort_syn_rcv(struct sock *sk, struct sk_buff *skb)
1910 {
1911 	const struct request_sock *oreq;
1912 	struct listen_ctx *listen_ctx;
1913 	struct chtls_sock *csk;
1914 	struct chtls_dev *cdev;
1915 	struct sock *psk;
1916 	void *ctx;
1917 
1918 	csk = sk->sk_user_data;
1919 	oreq = csk->passive_reap_next;
1920 	cdev = csk->cdev;
1921 
1922 	if (!oreq)
1923 		return -1;
1924 
1925 	ctx = lookup_stid(cdev->tids, oreq->ts_recent);
1926 	if (!ctx)
1927 		return -1;
1928 
1929 	listen_ctx = (struct listen_ctx *)ctx;
1930 	psk = listen_ctx->lsk;
1931 
1932 	bh_lock_sock(psk);
1933 	if (!sock_owned_by_user(psk)) {
1934 		int queue = csk->txq_idx;
1935 
1936 		do_abort_syn_rcv(sk, psk);
1937 		send_abort_rpl(sk, skb, cdev, CPL_ABORT_NO_RST, queue);
1938 	} else {
1939 		skb->sk = sk;
1940 		BLOG_SKB_CB(skb)->backlog_rcv = bl_abort_syn_rcv;
1941 		__sk_add_backlog(psk, skb);
1942 	}
1943 	bh_unlock_sock(psk);
1944 	return 0;
1945 }
1946 
chtls_abort_req_rss(struct sock * sk,struct sk_buff * skb)1947 static void chtls_abort_req_rss(struct sock *sk, struct sk_buff *skb)
1948 {
1949 	const struct cpl_abort_req_rss *req = cplhdr(skb) + RSS_HDR;
1950 	struct chtls_sock *csk = sk->sk_user_data;
1951 	int rst_status = CPL_ABORT_NO_RST;
1952 	int queue = csk->txq_idx;
1953 
1954 	if (is_neg_adv(req->status)) {
1955 		if (sk->sk_state == TCP_SYN_RECV)
1956 			chtls_set_tcb_tflag(sk, 0, 0);
1957 
1958 		kfree_skb(skb);
1959 		return;
1960 	}
1961 
1962 	csk_reset_flag(csk, CSK_ABORT_REQ_RCVD);
1963 
1964 	if (!csk_flag_nochk(csk, CSK_ABORT_SHUTDOWN) &&
1965 	    !csk_flag_nochk(csk, CSK_TX_DATA_SENT)) {
1966 		struct tcp_sock *tp = tcp_sk(sk);
1967 
1968 		if (send_tx_flowc_wr(sk, 0, tp->snd_nxt, tp->rcv_nxt) < 0)
1969 			WARN_ONCE(1, "send_tx_flowc error");
1970 		csk_set_flag(csk, CSK_TX_DATA_SENT);
1971 	}
1972 
1973 	csk_set_flag(csk, CSK_ABORT_SHUTDOWN);
1974 
1975 	if (!csk_flag_nochk(csk, CSK_ABORT_RPL_PENDING)) {
1976 		sk->sk_err = ETIMEDOUT;
1977 
1978 		if (!sock_flag(sk, SOCK_DEAD))
1979 			sk->sk_error_report(sk);
1980 
1981 		if (sk->sk_state == TCP_SYN_RECV && !abort_syn_rcv(sk, skb))
1982 			return;
1983 
1984 		chtls_release_resources(sk);
1985 		chtls_conn_done(sk);
1986 	}
1987 
1988 	chtls_send_abort_rpl(sk, skb, csk->cdev, rst_status, queue);
1989 }
1990 
chtls_abort_rpl_rss(struct sock * sk,struct sk_buff * skb)1991 static void chtls_abort_rpl_rss(struct sock *sk, struct sk_buff *skb)
1992 {
1993 	struct cpl_abort_rpl_rss *rpl = cplhdr(skb) + RSS_HDR;
1994 	struct chtls_sock *csk;
1995 	struct chtls_dev *cdev;
1996 
1997 	csk = rcu_dereference_sk_user_data(sk);
1998 	cdev = csk->cdev;
1999 
2000 	if (csk_flag_nochk(csk, CSK_ABORT_RPL_PENDING)) {
2001 		csk_reset_flag(csk, CSK_ABORT_RPL_PENDING);
2002 		if (!csk_flag_nochk(csk, CSK_ABORT_REQ_RCVD)) {
2003 			if (sk->sk_state == TCP_SYN_SENT) {
2004 				cxgb4_remove_tid(cdev->tids,
2005 						 csk->port_id,
2006 						 GET_TID(rpl),
2007 						 sk->sk_family);
2008 				sock_put(sk);
2009 			}
2010 			chtls_release_resources(sk);
2011 			chtls_conn_done(sk);
2012 		}
2013 	}
2014 	kfree_skb(skb);
2015 }
2016 
chtls_conn_cpl(struct chtls_dev * cdev,struct sk_buff * skb)2017 static int chtls_conn_cpl(struct chtls_dev *cdev, struct sk_buff *skb)
2018 {
2019 	struct cpl_peer_close *req = cplhdr(skb) + RSS_HDR;
2020 	void (*fn)(struct sock *sk, struct sk_buff *skb);
2021 	unsigned int hwtid = GET_TID(req);
2022 	struct sock *sk;
2023 	u8 opcode;
2024 
2025 	opcode = ((const struct rss_header *)cplhdr(skb))->opcode;
2026 
2027 	sk = lookup_tid(cdev->tids, hwtid);
2028 	if (!sk)
2029 		goto rel_skb;
2030 
2031 	switch (opcode) {
2032 	case CPL_PEER_CLOSE:
2033 		fn = chtls_peer_close;
2034 		break;
2035 	case CPL_CLOSE_CON_RPL:
2036 		fn = chtls_close_con_rpl;
2037 		break;
2038 	case CPL_ABORT_REQ_RSS:
2039 		fn = chtls_abort_req_rss;
2040 		break;
2041 	case CPL_ABORT_RPL_RSS:
2042 		fn = chtls_abort_rpl_rss;
2043 		break;
2044 	default:
2045 		goto rel_skb;
2046 	}
2047 
2048 	process_cpl_msg(fn, sk, skb);
2049 	return 0;
2050 
2051 rel_skb:
2052 	kfree_skb(skb);
2053 	return 0;
2054 }
2055 
chtls_rx_ack(struct sock * sk,struct sk_buff * skb)2056 static void chtls_rx_ack(struct sock *sk, struct sk_buff *skb)
2057 {
2058 	struct cpl_fw4_ack *hdr = cplhdr(skb) + RSS_HDR;
2059 	struct chtls_sock *csk = sk->sk_user_data;
2060 	struct tcp_sock *tp = tcp_sk(sk);
2061 	u32 credits = hdr->credits;
2062 	u32 snd_una;
2063 
2064 	snd_una = ntohl(hdr->snd_una);
2065 	csk->wr_credits += credits;
2066 
2067 	if (csk->wr_unacked > csk->wr_max_credits - csk->wr_credits)
2068 		csk->wr_unacked = csk->wr_max_credits - csk->wr_credits;
2069 
2070 	while (credits) {
2071 		struct sk_buff *pskb = csk->wr_skb_head;
2072 		u32 csum;
2073 
2074 		if (unlikely(!pskb)) {
2075 			if (csk->wr_nondata)
2076 				csk->wr_nondata -= credits;
2077 			break;
2078 		}
2079 		csum = (__force u32)pskb->csum;
2080 		if (unlikely(credits < csum)) {
2081 			pskb->csum = (__force __wsum)(csum - credits);
2082 			break;
2083 		}
2084 		dequeue_wr(sk);
2085 		credits -= csum;
2086 		kfree_skb(pskb);
2087 	}
2088 	if (hdr->seq_vld & CPL_FW4_ACK_FLAGS_SEQVAL) {
2089 		if (unlikely(before(snd_una, tp->snd_una))) {
2090 			kfree_skb(skb);
2091 			return;
2092 		}
2093 
2094 		if (tp->snd_una != snd_una) {
2095 			tp->snd_una = snd_una;
2096 			tp->rcv_tstamp = tcp_time_stamp(tp);
2097 			if (tp->snd_una == tp->snd_nxt &&
2098 			    !csk_flag_nochk(csk, CSK_TX_FAILOVER))
2099 				csk_reset_flag(csk, CSK_TX_WAIT_IDLE);
2100 		}
2101 	}
2102 
2103 	if (hdr->seq_vld & CPL_FW4_ACK_FLAGS_CH) {
2104 		unsigned int fclen16 = roundup(failover_flowc_wr_len, 16);
2105 
2106 		csk->wr_credits -= fclen16;
2107 		csk_reset_flag(csk, CSK_TX_WAIT_IDLE);
2108 		csk_reset_flag(csk, CSK_TX_FAILOVER);
2109 	}
2110 	if (skb_queue_len(&csk->txq) && chtls_push_frames(csk, 0))
2111 		sk->sk_write_space(sk);
2112 
2113 	kfree_skb(skb);
2114 }
2115 
chtls_wr_ack(struct chtls_dev * cdev,struct sk_buff * skb)2116 static int chtls_wr_ack(struct chtls_dev *cdev, struct sk_buff *skb)
2117 {
2118 	struct cpl_fw4_ack *rpl = cplhdr(skb) + RSS_HDR;
2119 	unsigned int hwtid = GET_TID(rpl);
2120 	struct sock *sk;
2121 
2122 	sk = lookup_tid(cdev->tids, hwtid);
2123 	if (unlikely(!sk)) {
2124 		pr_err("can't find conn. for hwtid %u.\n", hwtid);
2125 		return -EINVAL;
2126 	}
2127 	process_cpl_msg(chtls_rx_ack, sk, skb);
2128 
2129 	return 0;
2130 }
2131 
2132 chtls_handler_func chtls_handlers[NUM_CPL_CMDS] = {
2133 	[CPL_PASS_OPEN_RPL]     = chtls_pass_open_rpl,
2134 	[CPL_CLOSE_LISTSRV_RPL] = chtls_close_listsrv_rpl,
2135 	[CPL_PASS_ACCEPT_REQ]   = chtls_pass_accept_req,
2136 	[CPL_PASS_ESTABLISH]    = chtls_pass_establish,
2137 	[CPL_RX_DATA]           = chtls_rx_data,
2138 	[CPL_TLS_DATA]          = chtls_rx_pdu,
2139 	[CPL_RX_TLS_CMP]        = chtls_rx_cmp,
2140 	[CPL_PEER_CLOSE]        = chtls_conn_cpl,
2141 	[CPL_CLOSE_CON_RPL]     = chtls_conn_cpl,
2142 	[CPL_ABORT_REQ_RSS]     = chtls_conn_cpl,
2143 	[CPL_ABORT_RPL_RSS]     = chtls_conn_cpl,
2144 	[CPL_FW4_ACK]           = chtls_wr_ack,
2145 };
2146