xref: /linux/net/ipv4/tcp_ipv4.c (revision adda5401807882f5575364442cd583462fc40ae4)
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3  * INET		An implementation of the TCP/IP protocol suite for the LINUX
4  *		operating system.  INET is implemented using the  BSD Socket
5  *		interface as the means of communication with the user level.
6  *
7  *		Implementation of the Transmission Control Protocol(TCP).
8  *
9  *		IPv4 specific functions
10  *
11  *		code split from:
12  *		linux/ipv4/tcp.c
13  *		linux/ipv4/tcp_input.c
14  *		linux/ipv4/tcp_output.c
15  *
16  *		See tcp.c for author information
17  */
18 
19 /*
20  * Changes:
21  *		David S. Miller	:	New socket lookup architecture.
22  *					This code is dedicated to John Dyson.
23  *		David S. Miller :	Change semantics of established hash,
24  *					half is devoted to TIME_WAIT sockets
25  *					and the rest go in the other half.
26  *		Andi Kleen :		Add support for syncookies and fixed
27  *					some bugs: ip options weren't passed to
28  *					the TCP layer, missed a check for an
29  *					ACK bit.
30  *		Andi Kleen :		Implemented fast path mtu discovery.
31  *	     				Fixed many serious bugs in the
32  *					request_sock handling and moved
33  *					most of it into the af independent code.
34  *					Added tail drop and some other bugfixes.
35  *					Added new listen semantics.
36  *		Mike McLagan	:	Routing by source
37  *	Juan Jose Ciarlante:		ip_dynaddr bits
38  *		Andi Kleen:		various fixes.
39  *	Vitaly E. Lavrov	:	Transparent proxy revived after year
40  *					coma.
41  *	Andi Kleen		:	Fix new listen.
42  *	Andi Kleen		:	Fix accept error reporting.
43  *	YOSHIFUJI Hideaki @USAGI and:	Support IPV6_V6ONLY socket option, which
44  *	Alexey Kuznetsov		allow both IPv4 and IPv6 sockets to bind
45  *					a single port at the same time.
46  */
47 
48 #define pr_fmt(fmt) "TCP: " fmt
49 
50 #include <linux/bottom_half.h>
51 #include <linux/types.h>
52 #include <linux/fcntl.h>
53 #include <linux/module.h>
54 #include <linux/random.h>
55 #include <linux/cache.h>
56 #include <linux/jhash.h>
57 #include <linux/init.h>
58 #include <linux/times.h>
59 #include <linux/slab.h>
60 #include <linux/sched.h>
61 
62 #include <net/net_namespace.h>
63 #include <net/icmp.h>
64 #include <net/inet_hashtables.h>
65 #include <net/tcp.h>
66 #include <net/transp_v6.h>
67 #include <net/ipv6.h>
68 #include <net/inet_common.h>
69 #include <net/timewait_sock.h>
70 #include <net/xfrm.h>
71 #include <net/secure_seq.h>
72 #include <net/busy_poll.h>
73 
74 #include <linux/inet.h>
75 #include <linux/ipv6.h>
76 #include <linux/stddef.h>
77 #include <linux/proc_fs.h>
78 #include <linux/seq_file.h>
79 #include <linux/inetdevice.h>
80 #include <linux/btf_ids.h>
81 
82 #include <crypto/hash.h>
83 #include <linux/scatterlist.h>
84 
85 #include <trace/events/tcp.h>
86 
87 #ifdef CONFIG_TCP_MD5SIG
88 static int tcp_v4_md5_hash_hdr(char *md5_hash, const struct tcp_md5sig_key *key,
89 			       __be32 daddr, __be32 saddr, const struct tcphdr *th);
90 #endif
91 
92 struct inet_hashinfo tcp_hashinfo;
93 EXPORT_SYMBOL(tcp_hashinfo);
94 
95 static DEFINE_PER_CPU(struct sock *, ipv4_tcp_sk);
96 
97 static u32 tcp_v4_init_seq(const struct sk_buff *skb)
98 {
99 	return secure_tcp_seq(ip_hdr(skb)->daddr,
100 			      ip_hdr(skb)->saddr,
101 			      tcp_hdr(skb)->dest,
102 			      tcp_hdr(skb)->source);
103 }
104 
105 static u32 tcp_v4_init_ts_off(const struct net *net, const struct sk_buff *skb)
106 {
107 	return secure_tcp_ts_off(net, ip_hdr(skb)->daddr, ip_hdr(skb)->saddr);
108 }
109 
110 int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp)
111 {
112 	int reuse = READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_tw_reuse);
113 	const struct inet_timewait_sock *tw = inet_twsk(sktw);
114 	const struct tcp_timewait_sock *tcptw = tcp_twsk(sktw);
115 	struct tcp_sock *tp = tcp_sk(sk);
116 
117 	if (reuse == 2) {
118 		/* Still does not detect *everything* that goes through
119 		 * lo, since we require a loopback src or dst address
120 		 * or direct binding to 'lo' interface.
121 		 */
122 		bool loopback = false;
123 		if (tw->tw_bound_dev_if == LOOPBACK_IFINDEX)
124 			loopback = true;
125 #if IS_ENABLED(CONFIG_IPV6)
126 		if (tw->tw_family == AF_INET6) {
127 			if (ipv6_addr_loopback(&tw->tw_v6_daddr) ||
128 			    ipv6_addr_v4mapped_loopback(&tw->tw_v6_daddr) ||
129 			    ipv6_addr_loopback(&tw->tw_v6_rcv_saddr) ||
130 			    ipv6_addr_v4mapped_loopback(&tw->tw_v6_rcv_saddr))
131 				loopback = true;
132 		} else
133 #endif
134 		{
135 			if (ipv4_is_loopback(tw->tw_daddr) ||
136 			    ipv4_is_loopback(tw->tw_rcv_saddr))
137 				loopback = true;
138 		}
139 		if (!loopback)
140 			reuse = 0;
141 	}
142 
143 	/* With PAWS, it is safe from the viewpoint
144 	   of data integrity. Even without PAWS it is safe provided sequence
145 	   spaces do not overlap i.e. at data rates <= 80Mbit/sec.
146 
147 	   Actually, the idea is close to VJ's one, only timestamp cache is
148 	   held not per host, but per port pair and TW bucket is used as state
149 	   holder.
150 
151 	   If TW bucket has been already destroyed we fall back to VJ's scheme
152 	   and use initial timestamp retrieved from peer table.
153 	 */
154 	if (tcptw->tw_ts_recent_stamp &&
155 	    (!twp || (reuse && time_after32(ktime_get_seconds(),
156 					    tcptw->tw_ts_recent_stamp)))) {
157 		/* In case of repair and re-using TIME-WAIT sockets we still
158 		 * want to be sure that it is safe as above but honor the
159 		 * sequence numbers and time stamps set as part of the repair
160 		 * process.
161 		 *
162 		 * Without this check re-using a TIME-WAIT socket with TCP
163 		 * repair would accumulate a -1 on the repair assigned
164 		 * sequence number. The first time it is reused the sequence
165 		 * is -1, the second time -2, etc. This fixes that issue
166 		 * without appearing to create any others.
167 		 */
168 		if (likely(!tp->repair)) {
169 			u32 seq = tcptw->tw_snd_nxt + 65535 + 2;
170 
171 			if (!seq)
172 				seq = 1;
173 			WRITE_ONCE(tp->write_seq, seq);
174 			tp->rx_opt.ts_recent	   = tcptw->tw_ts_recent;
175 			tp->rx_opt.ts_recent_stamp = tcptw->tw_ts_recent_stamp;
176 		}
177 		sock_hold(sktw);
178 		return 1;
179 	}
180 
181 	return 0;
182 }
183 EXPORT_SYMBOL_GPL(tcp_twsk_unique);
184 
185 static int tcp_v4_pre_connect(struct sock *sk, struct sockaddr *uaddr,
186 			      int addr_len)
187 {
188 	/* This check is replicated from tcp_v4_connect() and intended to
189 	 * prevent BPF program called below from accessing bytes that are out
190 	 * of the bound specified by user in addr_len.
191 	 */
192 	if (addr_len < sizeof(struct sockaddr_in))
193 		return -EINVAL;
194 
195 	sock_owned_by_me(sk);
196 
197 	return BPF_CGROUP_RUN_PROG_INET4_CONNECT(sk, uaddr, &addr_len);
198 }
199 
200 /* This will initiate an outgoing connection. */
201 int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len)
202 {
203 	struct sockaddr_in *usin = (struct sockaddr_in *)uaddr;
204 	struct inet_timewait_death_row *tcp_death_row;
205 	struct inet_sock *inet = inet_sk(sk);
206 	struct tcp_sock *tp = tcp_sk(sk);
207 	struct ip_options_rcu *inet_opt;
208 	struct net *net = sock_net(sk);
209 	__be16 orig_sport, orig_dport;
210 	__be32 daddr, nexthop;
211 	struct flowi4 *fl4;
212 	struct rtable *rt;
213 	int err;
214 
215 	if (addr_len < sizeof(struct sockaddr_in))
216 		return -EINVAL;
217 
218 	if (usin->sin_family != AF_INET)
219 		return -EAFNOSUPPORT;
220 
221 	nexthop = daddr = usin->sin_addr.s_addr;
222 	inet_opt = rcu_dereference_protected(inet->inet_opt,
223 					     lockdep_sock_is_held(sk));
224 	if (inet_opt && inet_opt->opt.srr) {
225 		if (!daddr)
226 			return -EINVAL;
227 		nexthop = inet_opt->opt.faddr;
228 	}
229 
230 	orig_sport = inet->inet_sport;
231 	orig_dport = usin->sin_port;
232 	fl4 = &inet->cork.fl.u.ip4;
233 	rt = ip_route_connect(fl4, nexthop, inet->inet_saddr,
234 			      sk->sk_bound_dev_if, IPPROTO_TCP, orig_sport,
235 			      orig_dport, sk);
236 	if (IS_ERR(rt)) {
237 		err = PTR_ERR(rt);
238 		if (err == -ENETUNREACH)
239 			IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES);
240 		return err;
241 	}
242 
243 	if (rt->rt_flags & (RTCF_MULTICAST | RTCF_BROADCAST)) {
244 		ip_rt_put(rt);
245 		return -ENETUNREACH;
246 	}
247 
248 	if (!inet_opt || !inet_opt->opt.srr)
249 		daddr = fl4->daddr;
250 
251 	tcp_death_row = &sock_net(sk)->ipv4.tcp_death_row;
252 
253 	if (!inet->inet_saddr) {
254 		err = inet_bhash2_update_saddr(sk,  &fl4->saddr, AF_INET);
255 		if (err) {
256 			ip_rt_put(rt);
257 			return err;
258 		}
259 	} else {
260 		sk_rcv_saddr_set(sk, inet->inet_saddr);
261 	}
262 
263 	if (tp->rx_opt.ts_recent_stamp && inet->inet_daddr != daddr) {
264 		/* Reset inherited state */
265 		tp->rx_opt.ts_recent	   = 0;
266 		tp->rx_opt.ts_recent_stamp = 0;
267 		if (likely(!tp->repair))
268 			WRITE_ONCE(tp->write_seq, 0);
269 	}
270 
271 	inet->inet_dport = usin->sin_port;
272 	sk_daddr_set(sk, daddr);
273 
274 	inet_csk(sk)->icsk_ext_hdr_len = 0;
275 	if (inet_opt)
276 		inet_csk(sk)->icsk_ext_hdr_len = inet_opt->opt.optlen;
277 
278 	tp->rx_opt.mss_clamp = TCP_MSS_DEFAULT;
279 
280 	/* Socket identity is still unknown (sport may be zero).
281 	 * However we set state to SYN-SENT and not releasing socket
282 	 * lock select source port, enter ourselves into the hash tables and
283 	 * complete initialization after this.
284 	 */
285 	tcp_set_state(sk, TCP_SYN_SENT);
286 	err = inet_hash_connect(tcp_death_row, sk);
287 	if (err)
288 		goto failure;
289 
290 	sk_set_txhash(sk);
291 
292 	rt = ip_route_newports(fl4, rt, orig_sport, orig_dport,
293 			       inet->inet_sport, inet->inet_dport, sk);
294 	if (IS_ERR(rt)) {
295 		err = PTR_ERR(rt);
296 		rt = NULL;
297 		goto failure;
298 	}
299 	tp->tcp_usec_ts = dst_tcp_usec_ts(&rt->dst);
300 	/* OK, now commit destination to socket.  */
301 	sk->sk_gso_type = SKB_GSO_TCPV4;
302 	sk_setup_caps(sk, &rt->dst);
303 	rt = NULL;
304 
305 	if (likely(!tp->repair)) {
306 		if (!tp->write_seq)
307 			WRITE_ONCE(tp->write_seq,
308 				   secure_tcp_seq(inet->inet_saddr,
309 						  inet->inet_daddr,
310 						  inet->inet_sport,
311 						  usin->sin_port));
312 		WRITE_ONCE(tp->tsoffset,
313 			   secure_tcp_ts_off(net, inet->inet_saddr,
314 					     inet->inet_daddr));
315 	}
316 
317 	atomic_set(&inet->inet_id, get_random_u16());
318 
319 	if (tcp_fastopen_defer_connect(sk, &err))
320 		return err;
321 	if (err)
322 		goto failure;
323 
324 	err = tcp_connect(sk);
325 
326 	if (err)
327 		goto failure;
328 
329 	return 0;
330 
331 failure:
332 	/*
333 	 * This unhashes the socket and releases the local port,
334 	 * if necessary.
335 	 */
336 	tcp_set_state(sk, TCP_CLOSE);
337 	inet_bhash2_reset_saddr(sk);
338 	ip_rt_put(rt);
339 	sk->sk_route_caps = 0;
340 	inet->inet_dport = 0;
341 	return err;
342 }
343 EXPORT_SYMBOL(tcp_v4_connect);
344 
345 /*
346  * This routine reacts to ICMP_FRAG_NEEDED mtu indications as defined in RFC1191.
347  * It can be called through tcp_release_cb() if socket was owned by user
348  * at the time tcp_v4_err() was called to handle ICMP message.
349  */
350 void tcp_v4_mtu_reduced(struct sock *sk)
351 {
352 	struct inet_sock *inet = inet_sk(sk);
353 	struct dst_entry *dst;
354 	u32 mtu;
355 
356 	if ((1 << sk->sk_state) & (TCPF_LISTEN | TCPF_CLOSE))
357 		return;
358 	mtu = READ_ONCE(tcp_sk(sk)->mtu_info);
359 	dst = inet_csk_update_pmtu(sk, mtu);
360 	if (!dst)
361 		return;
362 
363 	/* Something is about to be wrong... Remember soft error
364 	 * for the case, if this connection will not able to recover.
365 	 */
366 	if (mtu < dst_mtu(dst) && ip_dont_fragment(sk, dst))
367 		WRITE_ONCE(sk->sk_err_soft, EMSGSIZE);
368 
369 	mtu = dst_mtu(dst);
370 
371 	if (inet->pmtudisc != IP_PMTUDISC_DONT &&
372 	    ip_sk_accept_pmtu(sk) &&
373 	    inet_csk(sk)->icsk_pmtu_cookie > mtu) {
374 		tcp_sync_mss(sk, mtu);
375 
376 		/* Resend the TCP packet because it's
377 		 * clear that the old packet has been
378 		 * dropped. This is the new "fast" path mtu
379 		 * discovery.
380 		 */
381 		tcp_simple_retransmit(sk);
382 	} /* else let the usual retransmit timer handle it */
383 }
384 EXPORT_SYMBOL(tcp_v4_mtu_reduced);
385 
386 static void do_redirect(struct sk_buff *skb, struct sock *sk)
387 {
388 	struct dst_entry *dst = __sk_dst_check(sk, 0);
389 
390 	if (dst)
391 		dst->ops->redirect(dst, sk, skb);
392 }
393 
394 
395 /* handle ICMP messages on TCP_NEW_SYN_RECV request sockets */
396 void tcp_req_err(struct sock *sk, u32 seq, bool abort)
397 {
398 	struct request_sock *req = inet_reqsk(sk);
399 	struct net *net = sock_net(sk);
400 
401 	/* ICMPs are not backlogged, hence we cannot get
402 	 * an established socket here.
403 	 */
404 	if (seq != tcp_rsk(req)->snt_isn) {
405 		__NET_INC_STATS(net, LINUX_MIB_OUTOFWINDOWICMPS);
406 	} else if (abort) {
407 		/*
408 		 * Still in SYN_RECV, just remove it silently.
409 		 * There is no good way to pass the error to the newly
410 		 * created socket, and POSIX does not want network
411 		 * errors returned from accept().
412 		 */
413 		inet_csk_reqsk_queue_drop(req->rsk_listener, req);
414 		tcp_listendrop(req->rsk_listener);
415 	}
416 	reqsk_put(req);
417 }
418 EXPORT_SYMBOL(tcp_req_err);
419 
420 /* TCP-LD (RFC 6069) logic */
421 void tcp_ld_RTO_revert(struct sock *sk, u32 seq)
422 {
423 	struct inet_connection_sock *icsk = inet_csk(sk);
424 	struct tcp_sock *tp = tcp_sk(sk);
425 	struct sk_buff *skb;
426 	s32 remaining;
427 	u32 delta_us;
428 
429 	if (sock_owned_by_user(sk))
430 		return;
431 
432 	if (seq != tp->snd_una  || !icsk->icsk_retransmits ||
433 	    !icsk->icsk_backoff)
434 		return;
435 
436 	skb = tcp_rtx_queue_head(sk);
437 	if (WARN_ON_ONCE(!skb))
438 		return;
439 
440 	icsk->icsk_backoff--;
441 	icsk->icsk_rto = tp->srtt_us ? __tcp_set_rto(tp) : TCP_TIMEOUT_INIT;
442 	icsk->icsk_rto = inet_csk_rto_backoff(icsk, TCP_RTO_MAX);
443 
444 	tcp_mstamp_refresh(tp);
445 	delta_us = (u32)(tp->tcp_mstamp - tcp_skb_timestamp_us(skb));
446 	remaining = icsk->icsk_rto - usecs_to_jiffies(delta_us);
447 
448 	if (remaining > 0) {
449 		inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
450 					  remaining, TCP_RTO_MAX);
451 	} else {
452 		/* RTO revert clocked out retransmission.
453 		 * Will retransmit now.
454 		 */
455 		tcp_retransmit_timer(sk);
456 	}
457 }
458 EXPORT_SYMBOL(tcp_ld_RTO_revert);
459 
460 /*
461  * This routine is called by the ICMP module when it gets some
462  * sort of error condition.  If err < 0 then the socket should
463  * be closed and the error returned to the user.  If err > 0
464  * it's just the icmp type << 8 | icmp code.  After adjustment
465  * header points to the first 8 bytes of the tcp header.  We need
466  * to find the appropriate port.
467  *
468  * The locking strategy used here is very "optimistic". When
469  * someone else accesses the socket the ICMP is just dropped
470  * and for some paths there is no check at all.
471  * A more general error queue to queue errors for later handling
472  * is probably better.
473  *
474  */
475 
476 int tcp_v4_err(struct sk_buff *skb, u32 info)
477 {
478 	const struct iphdr *iph = (const struct iphdr *)skb->data;
479 	struct tcphdr *th = (struct tcphdr *)(skb->data + (iph->ihl << 2));
480 	struct tcp_sock *tp;
481 	const int type = icmp_hdr(skb)->type;
482 	const int code = icmp_hdr(skb)->code;
483 	struct sock *sk;
484 	struct request_sock *fastopen;
485 	u32 seq, snd_una;
486 	int err;
487 	struct net *net = dev_net(skb->dev);
488 
489 	sk = __inet_lookup_established(net, net->ipv4.tcp_death_row.hashinfo,
490 				       iph->daddr, th->dest, iph->saddr,
491 				       ntohs(th->source), inet_iif(skb), 0);
492 	if (!sk) {
493 		__ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
494 		return -ENOENT;
495 	}
496 	if (sk->sk_state == TCP_TIME_WAIT) {
497 		/* To increase the counter of ignored icmps for TCP-AO */
498 		tcp_ao_ignore_icmp(sk, AF_INET, type, code);
499 		inet_twsk_put(inet_twsk(sk));
500 		return 0;
501 	}
502 	seq = ntohl(th->seq);
503 	if (sk->sk_state == TCP_NEW_SYN_RECV) {
504 		tcp_req_err(sk, seq, type == ICMP_PARAMETERPROB ||
505 				     type == ICMP_TIME_EXCEEDED ||
506 				     (type == ICMP_DEST_UNREACH &&
507 				      (code == ICMP_NET_UNREACH ||
508 				       code == ICMP_HOST_UNREACH)));
509 		return 0;
510 	}
511 
512 	if (tcp_ao_ignore_icmp(sk, AF_INET, type, code)) {
513 		sock_put(sk);
514 		return 0;
515 	}
516 
517 	bh_lock_sock(sk);
518 	/* If too many ICMPs get dropped on busy
519 	 * servers this needs to be solved differently.
520 	 * We do take care of PMTU discovery (RFC1191) special case :
521 	 * we can receive locally generated ICMP messages while socket is held.
522 	 */
523 	if (sock_owned_by_user(sk)) {
524 		if (!(type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED))
525 			__NET_INC_STATS(net, LINUX_MIB_LOCKDROPPEDICMPS);
526 	}
527 	if (sk->sk_state == TCP_CLOSE)
528 		goto out;
529 
530 	if (static_branch_unlikely(&ip4_min_ttl)) {
531 		/* min_ttl can be changed concurrently from do_ip_setsockopt() */
532 		if (unlikely(iph->ttl < READ_ONCE(inet_sk(sk)->min_ttl))) {
533 			__NET_INC_STATS(net, LINUX_MIB_TCPMINTTLDROP);
534 			goto out;
535 		}
536 	}
537 
538 	tp = tcp_sk(sk);
539 	/* XXX (TFO) - tp->snd_una should be ISN (tcp_create_openreq_child() */
540 	fastopen = rcu_dereference(tp->fastopen_rsk);
541 	snd_una = fastopen ? tcp_rsk(fastopen)->snt_isn : tp->snd_una;
542 	if (sk->sk_state != TCP_LISTEN &&
543 	    !between(seq, snd_una, tp->snd_nxt)) {
544 		__NET_INC_STATS(net, LINUX_MIB_OUTOFWINDOWICMPS);
545 		goto out;
546 	}
547 
548 	switch (type) {
549 	case ICMP_REDIRECT:
550 		if (!sock_owned_by_user(sk))
551 			do_redirect(skb, sk);
552 		goto out;
553 	case ICMP_SOURCE_QUENCH:
554 		/* Just silently ignore these. */
555 		goto out;
556 	case ICMP_PARAMETERPROB:
557 		err = EPROTO;
558 		break;
559 	case ICMP_DEST_UNREACH:
560 		if (code > NR_ICMP_UNREACH)
561 			goto out;
562 
563 		if (code == ICMP_FRAG_NEEDED) { /* PMTU discovery (RFC1191) */
564 			/* We are not interested in TCP_LISTEN and open_requests
565 			 * (SYN-ACKs send out by Linux are always <576bytes so
566 			 * they should go through unfragmented).
567 			 */
568 			if (sk->sk_state == TCP_LISTEN)
569 				goto out;
570 
571 			WRITE_ONCE(tp->mtu_info, info);
572 			if (!sock_owned_by_user(sk)) {
573 				tcp_v4_mtu_reduced(sk);
574 			} else {
575 				if (!test_and_set_bit(TCP_MTU_REDUCED_DEFERRED, &sk->sk_tsq_flags))
576 					sock_hold(sk);
577 			}
578 			goto out;
579 		}
580 
581 		err = icmp_err_convert[code].errno;
582 		/* check if this ICMP message allows revert of backoff.
583 		 * (see RFC 6069)
584 		 */
585 		if (!fastopen &&
586 		    (code == ICMP_NET_UNREACH || code == ICMP_HOST_UNREACH))
587 			tcp_ld_RTO_revert(sk, seq);
588 		break;
589 	case ICMP_TIME_EXCEEDED:
590 		err = EHOSTUNREACH;
591 		break;
592 	default:
593 		goto out;
594 	}
595 
596 	switch (sk->sk_state) {
597 	case TCP_SYN_SENT:
598 	case TCP_SYN_RECV:
599 		/* Only in fast or simultaneous open. If a fast open socket is
600 		 * already accepted it is treated as a connected one below.
601 		 */
602 		if (fastopen && !fastopen->sk)
603 			break;
604 
605 		ip_icmp_error(sk, skb, err, th->dest, info, (u8 *)th);
606 
607 		if (!sock_owned_by_user(sk)) {
608 			WRITE_ONCE(sk->sk_err, err);
609 
610 			sk_error_report(sk);
611 
612 			tcp_done(sk);
613 		} else {
614 			WRITE_ONCE(sk->sk_err_soft, err);
615 		}
616 		goto out;
617 	}
618 
619 	/* If we've already connected we will keep trying
620 	 * until we time out, or the user gives up.
621 	 *
622 	 * rfc1122 4.2.3.9 allows to consider as hard errors
623 	 * only PROTO_UNREACH and PORT_UNREACH (well, FRAG_FAILED too,
624 	 * but it is obsoleted by pmtu discovery).
625 	 *
626 	 * Note, that in modern internet, where routing is unreliable
627 	 * and in each dark corner broken firewalls sit, sending random
628 	 * errors ordered by their masters even this two messages finally lose
629 	 * their original sense (even Linux sends invalid PORT_UNREACHs)
630 	 *
631 	 * Now we are in compliance with RFCs.
632 	 *							--ANK (980905)
633 	 */
634 
635 	if (!sock_owned_by_user(sk) &&
636 	    inet_test_bit(RECVERR, sk)) {
637 		WRITE_ONCE(sk->sk_err, err);
638 		sk_error_report(sk);
639 	} else	{ /* Only an error on timeout */
640 		WRITE_ONCE(sk->sk_err_soft, err);
641 	}
642 
643 out:
644 	bh_unlock_sock(sk);
645 	sock_put(sk);
646 	return 0;
647 }
648 
649 void __tcp_v4_send_check(struct sk_buff *skb, __be32 saddr, __be32 daddr)
650 {
651 	struct tcphdr *th = tcp_hdr(skb);
652 
653 	th->check = ~tcp_v4_check(skb->len, saddr, daddr, 0);
654 	skb->csum_start = skb_transport_header(skb) - skb->head;
655 	skb->csum_offset = offsetof(struct tcphdr, check);
656 }
657 
658 /* This routine computes an IPv4 TCP checksum. */
659 void tcp_v4_send_check(struct sock *sk, struct sk_buff *skb)
660 {
661 	const struct inet_sock *inet = inet_sk(sk);
662 
663 	__tcp_v4_send_check(skb, inet->inet_saddr, inet->inet_daddr);
664 }
665 EXPORT_SYMBOL(tcp_v4_send_check);
666 
667 #define REPLY_OPTIONS_LEN      (MAX_TCP_OPTION_SPACE / sizeof(__be32))
668 
669 static bool tcp_v4_ao_sign_reset(const struct sock *sk, struct sk_buff *skb,
670 				 const struct tcp_ao_hdr *aoh,
671 				 struct ip_reply_arg *arg, struct tcphdr *reply,
672 				 __be32 reply_options[REPLY_OPTIONS_LEN])
673 {
674 #ifdef CONFIG_TCP_AO
675 	int sdif = tcp_v4_sdif(skb);
676 	int dif = inet_iif(skb);
677 	int l3index = sdif ? dif : 0;
678 	bool allocated_traffic_key;
679 	struct tcp_ao_key *key;
680 	char *traffic_key;
681 	bool drop = true;
682 	u32 ao_sne = 0;
683 	u8 keyid;
684 
685 	rcu_read_lock();
686 	if (tcp_ao_prepare_reset(sk, skb, aoh, l3index, ntohl(reply->seq),
687 				 &key, &traffic_key, &allocated_traffic_key,
688 				 &keyid, &ao_sne))
689 		goto out;
690 
691 	reply_options[0] = htonl((TCPOPT_AO << 24) | (tcp_ao_len(key) << 16) |
692 				 (aoh->rnext_keyid << 8) | keyid);
693 	arg->iov[0].iov_len += tcp_ao_len_aligned(key);
694 	reply->doff = arg->iov[0].iov_len / 4;
695 
696 	if (tcp_ao_hash_hdr(AF_INET, (char *)&reply_options[1],
697 			    key, traffic_key,
698 			    (union tcp_ao_addr *)&ip_hdr(skb)->saddr,
699 			    (union tcp_ao_addr *)&ip_hdr(skb)->daddr,
700 			    reply, ao_sne))
701 		goto out;
702 	drop = false;
703 out:
704 	rcu_read_unlock();
705 	if (allocated_traffic_key)
706 		kfree(traffic_key);
707 	return drop;
708 #else
709 	return true;
710 #endif
711 }
712 
713 /*
714  *	This routine will send an RST to the other tcp.
715  *
716  *	Someone asks: why I NEVER use socket parameters (TOS, TTL etc.)
717  *		      for reset.
718  *	Answer: if a packet caused RST, it is not for a socket
719  *		existing in our system, if it is matched to a socket,
720  *		it is just duplicate segment or bug in other side's TCP.
721  *		So that we build reply only basing on parameters
722  *		arrived with segment.
723  *	Exception: precedence violation. We do not implement it in any case.
724  */
725 
726 static void tcp_v4_send_reset(const struct sock *sk, struct sk_buff *skb)
727 {
728 	const struct tcphdr *th = tcp_hdr(skb);
729 	struct {
730 		struct tcphdr th;
731 		__be32 opt[REPLY_OPTIONS_LEN];
732 	} rep;
733 	const __u8 *md5_hash_location = NULL;
734 	const struct tcp_ao_hdr *aoh;
735 	struct ip_reply_arg arg;
736 #ifdef CONFIG_TCP_MD5SIG
737 	struct tcp_md5sig_key *key = NULL;
738 	unsigned char newhash[16];
739 	struct sock *sk1 = NULL;
740 	int genhash;
741 #endif
742 	u64 transmit_time = 0;
743 	struct sock *ctl_sk;
744 	struct net *net;
745 	u32 txhash = 0;
746 
747 	/* Never send a reset in response to a reset. */
748 	if (th->rst)
749 		return;
750 
751 	/* If sk not NULL, it means we did a successful lookup and incoming
752 	 * route had to be correct. prequeue might have dropped our dst.
753 	 */
754 	if (!sk && skb_rtable(skb)->rt_type != RTN_LOCAL)
755 		return;
756 
757 	/* Swap the send and the receive. */
758 	memset(&rep, 0, sizeof(rep));
759 	rep.th.dest   = th->source;
760 	rep.th.source = th->dest;
761 	rep.th.doff   = sizeof(struct tcphdr) / 4;
762 	rep.th.rst    = 1;
763 
764 	if (th->ack) {
765 		rep.th.seq = th->ack_seq;
766 	} else {
767 		rep.th.ack = 1;
768 		rep.th.ack_seq = htonl(ntohl(th->seq) + th->syn + th->fin +
769 				       skb->len - (th->doff << 2));
770 	}
771 
772 	memset(&arg, 0, sizeof(arg));
773 	arg.iov[0].iov_base = (unsigned char *)&rep;
774 	arg.iov[0].iov_len  = sizeof(rep.th);
775 
776 	net = sk ? sock_net(sk) : dev_net(skb_dst(skb)->dev);
777 
778 	/* Invalid TCP option size or twice included auth */
779 	if (tcp_parse_auth_options(tcp_hdr(skb), &md5_hash_location, &aoh))
780 		return;
781 
782 	if (aoh && tcp_v4_ao_sign_reset(sk, skb, aoh, &arg, &rep.th, rep.opt))
783 		return;
784 
785 #ifdef CONFIG_TCP_MD5SIG
786 	rcu_read_lock();
787 	if (sk && sk_fullsock(sk)) {
788 		const union tcp_md5_addr *addr;
789 		int l3index;
790 
791 		/* sdif set, means packet ingressed via a device
792 		 * in an L3 domain and inet_iif is set to it.
793 		 */
794 		l3index = tcp_v4_sdif(skb) ? inet_iif(skb) : 0;
795 		addr = (union tcp_md5_addr *)&ip_hdr(skb)->saddr;
796 		key = tcp_md5_do_lookup(sk, l3index, addr, AF_INET);
797 	} else if (md5_hash_location) {
798 		const union tcp_md5_addr *addr;
799 		int sdif = tcp_v4_sdif(skb);
800 		int dif = inet_iif(skb);
801 		int l3index;
802 
803 		/*
804 		 * active side is lost. Try to find listening socket through
805 		 * source port, and then find md5 key through listening socket.
806 		 * we are not loose security here:
807 		 * Incoming packet is checked with md5 hash with finding key,
808 		 * no RST generated if md5 hash doesn't match.
809 		 */
810 		sk1 = __inet_lookup_listener(net, net->ipv4.tcp_death_row.hashinfo,
811 					     NULL, 0, ip_hdr(skb)->saddr,
812 					     th->source, ip_hdr(skb)->daddr,
813 					     ntohs(th->source), dif, sdif);
814 		/* don't send rst if it can't find key */
815 		if (!sk1)
816 			goto out;
817 
818 		/* sdif set, means packet ingressed via a device
819 		 * in an L3 domain and dif is set to it.
820 		 */
821 		l3index = sdif ? dif : 0;
822 		addr = (union tcp_md5_addr *)&ip_hdr(skb)->saddr;
823 		key = tcp_md5_do_lookup(sk1, l3index, addr, AF_INET);
824 		if (!key)
825 			goto out;
826 
827 
828 		genhash = tcp_v4_md5_hash_skb(newhash, key, NULL, skb);
829 		if (genhash || memcmp(md5_hash_location, newhash, 16) != 0)
830 			goto out;
831 
832 	}
833 
834 	if (key) {
835 		rep.opt[0] = htonl((TCPOPT_NOP << 24) |
836 				   (TCPOPT_NOP << 16) |
837 				   (TCPOPT_MD5SIG << 8) |
838 				   TCPOLEN_MD5SIG);
839 		/* Update length and the length the header thinks exists */
840 		arg.iov[0].iov_len += TCPOLEN_MD5SIG_ALIGNED;
841 		rep.th.doff = arg.iov[0].iov_len / 4;
842 
843 		tcp_v4_md5_hash_hdr((__u8 *) &rep.opt[1],
844 				     key, ip_hdr(skb)->saddr,
845 				     ip_hdr(skb)->daddr, &rep.th);
846 	}
847 #endif
848 	/* Can't co-exist with TCPMD5, hence check rep.opt[0] */
849 	if (rep.opt[0] == 0) {
850 		__be32 mrst = mptcp_reset_option(skb);
851 
852 		if (mrst) {
853 			rep.opt[0] = mrst;
854 			arg.iov[0].iov_len += sizeof(mrst);
855 			rep.th.doff = arg.iov[0].iov_len / 4;
856 		}
857 	}
858 
859 	arg.csum = csum_tcpudp_nofold(ip_hdr(skb)->daddr,
860 				      ip_hdr(skb)->saddr, /* XXX */
861 				      arg.iov[0].iov_len, IPPROTO_TCP, 0);
862 	arg.csumoffset = offsetof(struct tcphdr, check) / 2;
863 	arg.flags = (sk && inet_sk_transparent(sk)) ? IP_REPLY_ARG_NOSRCCHECK : 0;
864 
865 	/* When socket is gone, all binding information is lost.
866 	 * routing might fail in this case. No choice here, if we choose to force
867 	 * input interface, we will misroute in case of asymmetric route.
868 	 */
869 	if (sk)
870 		arg.bound_dev_if = sk->sk_bound_dev_if;
871 
872 	trace_tcp_send_reset(sk, skb);
873 
874 	BUILD_BUG_ON(offsetof(struct sock, sk_bound_dev_if) !=
875 		     offsetof(struct inet_timewait_sock, tw_bound_dev_if));
876 
877 	arg.tos = ip_hdr(skb)->tos;
878 	arg.uid = sock_net_uid(net, sk && sk_fullsock(sk) ? sk : NULL);
879 	local_bh_disable();
880 	ctl_sk = this_cpu_read(ipv4_tcp_sk);
881 	sock_net_set(ctl_sk, net);
882 	if (sk) {
883 		ctl_sk->sk_mark = (sk->sk_state == TCP_TIME_WAIT) ?
884 				   inet_twsk(sk)->tw_mark : sk->sk_mark;
885 		ctl_sk->sk_priority = (sk->sk_state == TCP_TIME_WAIT) ?
886 				   inet_twsk(sk)->tw_priority : READ_ONCE(sk->sk_priority);
887 		transmit_time = tcp_transmit_time(sk);
888 		xfrm_sk_clone_policy(ctl_sk, sk);
889 		txhash = (sk->sk_state == TCP_TIME_WAIT) ?
890 			 inet_twsk(sk)->tw_txhash : sk->sk_txhash;
891 	} else {
892 		ctl_sk->sk_mark = 0;
893 		ctl_sk->sk_priority = 0;
894 	}
895 	ip_send_unicast_reply(ctl_sk,
896 			      skb, &TCP_SKB_CB(skb)->header.h4.opt,
897 			      ip_hdr(skb)->saddr, ip_hdr(skb)->daddr,
898 			      &arg, arg.iov[0].iov_len,
899 			      transmit_time, txhash);
900 
901 	xfrm_sk_free_policy(ctl_sk);
902 	sock_net_set(ctl_sk, &init_net);
903 	__TCP_INC_STATS(net, TCP_MIB_OUTSEGS);
904 	__TCP_INC_STATS(net, TCP_MIB_OUTRSTS);
905 	local_bh_enable();
906 
907 #ifdef CONFIG_TCP_MD5SIG
908 out:
909 	rcu_read_unlock();
910 #endif
911 }
912 
913 /* The code following below sending ACKs in SYN-RECV and TIME-WAIT states
914    outside socket context is ugly, certainly. What can I do?
915  */
916 
917 static void tcp_v4_send_ack(const struct sock *sk,
918 			    struct sk_buff *skb, u32 seq, u32 ack,
919 			    u32 win, u32 tsval, u32 tsecr, int oif,
920 			    struct tcp_key *key,
921 			    int reply_flags, u8 tos, u32 txhash)
922 {
923 	const struct tcphdr *th = tcp_hdr(skb);
924 	struct {
925 		struct tcphdr th;
926 		__be32 opt[(MAX_TCP_OPTION_SPACE  >> 2)];
927 	} rep;
928 	struct net *net = sock_net(sk);
929 	struct ip_reply_arg arg;
930 	struct sock *ctl_sk;
931 	u64 transmit_time;
932 
933 	memset(&rep.th, 0, sizeof(struct tcphdr));
934 	memset(&arg, 0, sizeof(arg));
935 
936 	arg.iov[0].iov_base = (unsigned char *)&rep;
937 	arg.iov[0].iov_len  = sizeof(rep.th);
938 	if (tsecr) {
939 		rep.opt[0] = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
940 				   (TCPOPT_TIMESTAMP << 8) |
941 				   TCPOLEN_TIMESTAMP);
942 		rep.opt[1] = htonl(tsval);
943 		rep.opt[2] = htonl(tsecr);
944 		arg.iov[0].iov_len += TCPOLEN_TSTAMP_ALIGNED;
945 	}
946 
947 	/* Swap the send and the receive. */
948 	rep.th.dest    = th->source;
949 	rep.th.source  = th->dest;
950 	rep.th.doff    = arg.iov[0].iov_len / 4;
951 	rep.th.seq     = htonl(seq);
952 	rep.th.ack_seq = htonl(ack);
953 	rep.th.ack     = 1;
954 	rep.th.window  = htons(win);
955 
956 #ifdef CONFIG_TCP_MD5SIG
957 	if (tcp_key_is_md5(key)) {
958 		int offset = (tsecr) ? 3 : 0;
959 
960 		rep.opt[offset++] = htonl((TCPOPT_NOP << 24) |
961 					  (TCPOPT_NOP << 16) |
962 					  (TCPOPT_MD5SIG << 8) |
963 					  TCPOLEN_MD5SIG);
964 		arg.iov[0].iov_len += TCPOLEN_MD5SIG_ALIGNED;
965 		rep.th.doff = arg.iov[0].iov_len/4;
966 
967 		tcp_v4_md5_hash_hdr((__u8 *) &rep.opt[offset],
968 				    key->md5_key, ip_hdr(skb)->saddr,
969 				    ip_hdr(skb)->daddr, &rep.th);
970 	}
971 #endif
972 #ifdef CONFIG_TCP_AO
973 	if (tcp_key_is_ao(key)) {
974 		int offset = (tsecr) ? 3 : 0;
975 
976 		rep.opt[offset++] = htonl((TCPOPT_AO << 24) |
977 					  (tcp_ao_len(key->ao_key) << 16) |
978 					  (key->ao_key->sndid << 8) |
979 					  key->rcv_next);
980 		arg.iov[0].iov_len += tcp_ao_len_aligned(key->ao_key);
981 		rep.th.doff = arg.iov[0].iov_len / 4;
982 
983 		tcp_ao_hash_hdr(AF_INET, (char *)&rep.opt[offset],
984 				key->ao_key, key->traffic_key,
985 				(union tcp_ao_addr *)&ip_hdr(skb)->saddr,
986 				(union tcp_ao_addr *)&ip_hdr(skb)->daddr,
987 				&rep.th, key->sne);
988 	}
989 #endif
990 	arg.flags = reply_flags;
991 	arg.csum = csum_tcpudp_nofold(ip_hdr(skb)->daddr,
992 				      ip_hdr(skb)->saddr, /* XXX */
993 				      arg.iov[0].iov_len, IPPROTO_TCP, 0);
994 	arg.csumoffset = offsetof(struct tcphdr, check) / 2;
995 	if (oif)
996 		arg.bound_dev_if = oif;
997 	arg.tos = tos;
998 	arg.uid = sock_net_uid(net, sk_fullsock(sk) ? sk : NULL);
999 	local_bh_disable();
1000 	ctl_sk = this_cpu_read(ipv4_tcp_sk);
1001 	sock_net_set(ctl_sk, net);
1002 	ctl_sk->sk_mark = (sk->sk_state == TCP_TIME_WAIT) ?
1003 			   inet_twsk(sk)->tw_mark : READ_ONCE(sk->sk_mark);
1004 	ctl_sk->sk_priority = (sk->sk_state == TCP_TIME_WAIT) ?
1005 			   inet_twsk(sk)->tw_priority : READ_ONCE(sk->sk_priority);
1006 	transmit_time = tcp_transmit_time(sk);
1007 	ip_send_unicast_reply(ctl_sk,
1008 			      skb, &TCP_SKB_CB(skb)->header.h4.opt,
1009 			      ip_hdr(skb)->saddr, ip_hdr(skb)->daddr,
1010 			      &arg, arg.iov[0].iov_len,
1011 			      transmit_time, txhash);
1012 
1013 	sock_net_set(ctl_sk, &init_net);
1014 	__TCP_INC_STATS(net, TCP_MIB_OUTSEGS);
1015 	local_bh_enable();
1016 }
1017 
1018 static void tcp_v4_timewait_ack(struct sock *sk, struct sk_buff *skb)
1019 {
1020 	struct inet_timewait_sock *tw = inet_twsk(sk);
1021 	struct tcp_timewait_sock *tcptw = tcp_twsk(sk);
1022 	struct tcp_key key = {};
1023 #ifdef CONFIG_TCP_AO
1024 	struct tcp_ao_info *ao_info;
1025 
1026 	if (static_branch_unlikely(&tcp_ao_needed.key)) {
1027 		/* FIXME: the segment to-be-acked is not verified yet */
1028 		ao_info = rcu_dereference(tcptw->ao_info);
1029 		if (ao_info) {
1030 			const struct tcp_ao_hdr *aoh;
1031 
1032 			if (tcp_parse_auth_options(tcp_hdr(skb), NULL, &aoh)) {
1033 				inet_twsk_put(tw);
1034 				return;
1035 			}
1036 
1037 			if (aoh)
1038 				key.ao_key = tcp_ao_established_key(ao_info, aoh->rnext_keyid, -1);
1039 		}
1040 	}
1041 	if (key.ao_key) {
1042 		struct tcp_ao_key *rnext_key;
1043 
1044 		key.traffic_key = snd_other_key(key.ao_key);
1045 		key.sne = READ_ONCE(ao_info->snd_sne);
1046 		rnext_key = READ_ONCE(ao_info->rnext_key);
1047 		key.rcv_next = rnext_key->rcvid;
1048 		key.type = TCP_KEY_AO;
1049 #else
1050 	if (0) {
1051 #endif
1052 #ifdef CONFIG_TCP_MD5SIG
1053 	} else if (static_branch_unlikely(&tcp_md5_needed.key)) {
1054 		key.md5_key = tcp_twsk_md5_key(tcptw);
1055 		if (key.md5_key)
1056 			key.type = TCP_KEY_MD5;
1057 #endif
1058 	}
1059 
1060 	tcp_v4_send_ack(sk, skb,
1061 			tcptw->tw_snd_nxt, tcptw->tw_rcv_nxt,
1062 			tcptw->tw_rcv_wnd >> tw->tw_rcv_wscale,
1063 			tcp_tw_tsval(tcptw),
1064 			tcptw->tw_ts_recent,
1065 			tw->tw_bound_dev_if, &key,
1066 			tw->tw_transparent ? IP_REPLY_ARG_NOSRCCHECK : 0,
1067 			tw->tw_tos,
1068 			tw->tw_txhash);
1069 
1070 	inet_twsk_put(tw);
1071 }
1072 
1073 static void tcp_v4_reqsk_send_ack(const struct sock *sk, struct sk_buff *skb,
1074 				  struct request_sock *req)
1075 {
1076 	struct tcp_key key = {};
1077 
1078 	/* sk->sk_state == TCP_LISTEN -> for regular TCP_SYN_RECV
1079 	 * sk->sk_state == TCP_SYN_RECV -> for Fast Open.
1080 	 */
1081 	u32 seq = (sk->sk_state == TCP_LISTEN) ? tcp_rsk(req)->snt_isn + 1 :
1082 					     tcp_sk(sk)->snd_nxt;
1083 
1084 #ifdef CONFIG_TCP_AO
1085 	if (static_branch_unlikely(&tcp_ao_needed.key) &&
1086 	    tcp_rsk_used_ao(req)) {
1087 		const union tcp_md5_addr *addr;
1088 		const struct tcp_ao_hdr *aoh;
1089 		int l3index;
1090 
1091 		/* Invalid TCP option size or twice included auth */
1092 		if (tcp_parse_auth_options(tcp_hdr(skb), NULL, &aoh))
1093 			return;
1094 		if (!aoh)
1095 			return;
1096 
1097 		addr = (union tcp_md5_addr *)&ip_hdr(skb)->saddr;
1098 		l3index = tcp_v4_sdif(skb) ? inet_iif(skb) : 0;
1099 		key.ao_key = tcp_ao_do_lookup(sk, l3index, addr, AF_INET,
1100 					      aoh->rnext_keyid, -1);
1101 		if (unlikely(!key.ao_key)) {
1102 			/* Send ACK with any matching MKT for the peer */
1103 			key.ao_key = tcp_ao_do_lookup(sk, l3index, addr, AF_INET, -1, -1);
1104 			/* Matching key disappeared (user removed the key?)
1105 			 * let the handshake timeout.
1106 			 */
1107 			if (!key.ao_key) {
1108 				net_info_ratelimited("TCP-AO key for (%pI4, %d)->(%pI4, %d) suddenly disappeared, won't ACK new connection\n",
1109 						     addr,
1110 						     ntohs(tcp_hdr(skb)->source),
1111 						     &ip_hdr(skb)->daddr,
1112 						     ntohs(tcp_hdr(skb)->dest));
1113 				return;
1114 			}
1115 		}
1116 		key.traffic_key = kmalloc(tcp_ao_digest_size(key.ao_key), GFP_ATOMIC);
1117 		if (!key.traffic_key)
1118 			return;
1119 
1120 		key.type = TCP_KEY_AO;
1121 		key.rcv_next = aoh->keyid;
1122 		tcp_v4_ao_calc_key_rsk(key.ao_key, key.traffic_key, req);
1123 #else
1124 	if (0) {
1125 #endif
1126 #ifdef CONFIG_TCP_MD5SIG
1127 	} else if (static_branch_unlikely(&tcp_md5_needed.key)) {
1128 		const union tcp_md5_addr *addr;
1129 		int l3index;
1130 
1131 		addr = (union tcp_md5_addr *)&ip_hdr(skb)->saddr;
1132 		l3index = tcp_v4_sdif(skb) ? inet_iif(skb) : 0;
1133 		key.md5_key = tcp_md5_do_lookup(sk, l3index, addr, AF_INET);
1134 		if (key.md5_key)
1135 			key.type = TCP_KEY_MD5;
1136 #endif
1137 	}
1138 
1139 	/* RFC 7323 2.3
1140 	 * The window field (SEG.WND) of every outgoing segment, with the
1141 	 * exception of <SYN> segments, MUST be right-shifted by
1142 	 * Rcv.Wind.Shift bits:
1143 	 */
1144 	tcp_v4_send_ack(sk, skb, seq,
1145 			tcp_rsk(req)->rcv_nxt,
1146 			req->rsk_rcv_wnd >> inet_rsk(req)->rcv_wscale,
1147 			tcp_rsk_tsval(tcp_rsk(req)),
1148 			READ_ONCE(req->ts_recent),
1149 			0, &key,
1150 			inet_rsk(req)->no_srccheck ? IP_REPLY_ARG_NOSRCCHECK : 0,
1151 			ip_hdr(skb)->tos,
1152 			READ_ONCE(tcp_rsk(req)->txhash));
1153 	if (tcp_key_is_ao(&key))
1154 		kfree(key.traffic_key);
1155 }
1156 
1157 /*
1158  *	Send a SYN-ACK after having received a SYN.
1159  *	This still operates on a request_sock only, not on a big
1160  *	socket.
1161  */
1162 static int tcp_v4_send_synack(const struct sock *sk, struct dst_entry *dst,
1163 			      struct flowi *fl,
1164 			      struct request_sock *req,
1165 			      struct tcp_fastopen_cookie *foc,
1166 			      enum tcp_synack_type synack_type,
1167 			      struct sk_buff *syn_skb)
1168 {
1169 	const struct inet_request_sock *ireq = inet_rsk(req);
1170 	struct flowi4 fl4;
1171 	int err = -1;
1172 	struct sk_buff *skb;
1173 	u8 tos;
1174 
1175 	/* First, grab a route. */
1176 	if (!dst && (dst = inet_csk_route_req(sk, &fl4, req)) == NULL)
1177 		return -1;
1178 
1179 	skb = tcp_make_synack(sk, dst, req, foc, synack_type, syn_skb);
1180 
1181 	if (skb) {
1182 		__tcp_v4_send_check(skb, ireq->ir_loc_addr, ireq->ir_rmt_addr);
1183 
1184 		tos = READ_ONCE(inet_sk(sk)->tos);
1185 
1186 		if (READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_reflect_tos))
1187 			tos = (tcp_rsk(req)->syn_tos & ~INET_ECN_MASK) |
1188 			      (tos & INET_ECN_MASK);
1189 
1190 		if (!INET_ECN_is_capable(tos) &&
1191 		    tcp_bpf_ca_needs_ecn((struct sock *)req))
1192 			tos |= INET_ECN_ECT_0;
1193 
1194 		rcu_read_lock();
1195 		err = ip_build_and_send_pkt(skb, sk, ireq->ir_loc_addr,
1196 					    ireq->ir_rmt_addr,
1197 					    rcu_dereference(ireq->ireq_opt),
1198 					    tos);
1199 		rcu_read_unlock();
1200 		err = net_xmit_eval(err);
1201 	}
1202 
1203 	return err;
1204 }
1205 
1206 /*
1207  *	IPv4 request_sock destructor.
1208  */
1209 static void tcp_v4_reqsk_destructor(struct request_sock *req)
1210 {
1211 	kfree(rcu_dereference_protected(inet_rsk(req)->ireq_opt, 1));
1212 }
1213 
1214 #ifdef CONFIG_TCP_MD5SIG
1215 /*
1216  * RFC2385 MD5 checksumming requires a mapping of
1217  * IP address->MD5 Key.
1218  * We need to maintain these in the sk structure.
1219  */
1220 
1221 DEFINE_STATIC_KEY_DEFERRED_FALSE(tcp_md5_needed, HZ);
1222 EXPORT_SYMBOL(tcp_md5_needed);
1223 
1224 static bool better_md5_match(struct tcp_md5sig_key *old, struct tcp_md5sig_key *new)
1225 {
1226 	if (!old)
1227 		return true;
1228 
1229 	/* l3index always overrides non-l3index */
1230 	if (old->l3index && new->l3index == 0)
1231 		return false;
1232 	if (old->l3index == 0 && new->l3index)
1233 		return true;
1234 
1235 	return old->prefixlen < new->prefixlen;
1236 }
1237 
1238 /* Find the Key structure for an address.  */
1239 struct tcp_md5sig_key *__tcp_md5_do_lookup(const struct sock *sk, int l3index,
1240 					   const union tcp_md5_addr *addr,
1241 					   int family, bool any_l3index)
1242 {
1243 	const struct tcp_sock *tp = tcp_sk(sk);
1244 	struct tcp_md5sig_key *key;
1245 	const struct tcp_md5sig_info *md5sig;
1246 	__be32 mask;
1247 	struct tcp_md5sig_key *best_match = NULL;
1248 	bool match;
1249 
1250 	/* caller either holds rcu_read_lock() or socket lock */
1251 	md5sig = rcu_dereference_check(tp->md5sig_info,
1252 				       lockdep_sock_is_held(sk));
1253 	if (!md5sig)
1254 		return NULL;
1255 
1256 	hlist_for_each_entry_rcu(key, &md5sig->head, node,
1257 				 lockdep_sock_is_held(sk)) {
1258 		if (key->family != family)
1259 			continue;
1260 		if (!any_l3index && key->flags & TCP_MD5SIG_FLAG_IFINDEX &&
1261 		    key->l3index != l3index)
1262 			continue;
1263 		if (family == AF_INET) {
1264 			mask = inet_make_mask(key->prefixlen);
1265 			match = (key->addr.a4.s_addr & mask) ==
1266 				(addr->a4.s_addr & mask);
1267 #if IS_ENABLED(CONFIG_IPV6)
1268 		} else if (family == AF_INET6) {
1269 			match = ipv6_prefix_equal(&key->addr.a6, &addr->a6,
1270 						  key->prefixlen);
1271 #endif
1272 		} else {
1273 			match = false;
1274 		}
1275 
1276 		if (match && better_md5_match(best_match, key))
1277 			best_match = key;
1278 	}
1279 	return best_match;
1280 }
1281 EXPORT_SYMBOL(__tcp_md5_do_lookup);
1282 
1283 static struct tcp_md5sig_key *tcp_md5_do_lookup_exact(const struct sock *sk,
1284 						      const union tcp_md5_addr *addr,
1285 						      int family, u8 prefixlen,
1286 						      int l3index, u8 flags)
1287 {
1288 	const struct tcp_sock *tp = tcp_sk(sk);
1289 	struct tcp_md5sig_key *key;
1290 	unsigned int size = sizeof(struct in_addr);
1291 	const struct tcp_md5sig_info *md5sig;
1292 
1293 	/* caller either holds rcu_read_lock() or socket lock */
1294 	md5sig = rcu_dereference_check(tp->md5sig_info,
1295 				       lockdep_sock_is_held(sk));
1296 	if (!md5sig)
1297 		return NULL;
1298 #if IS_ENABLED(CONFIG_IPV6)
1299 	if (family == AF_INET6)
1300 		size = sizeof(struct in6_addr);
1301 #endif
1302 	hlist_for_each_entry_rcu(key, &md5sig->head, node,
1303 				 lockdep_sock_is_held(sk)) {
1304 		if (key->family != family)
1305 			continue;
1306 		if ((key->flags & TCP_MD5SIG_FLAG_IFINDEX) != (flags & TCP_MD5SIG_FLAG_IFINDEX))
1307 			continue;
1308 		if (key->l3index != l3index)
1309 			continue;
1310 		if (!memcmp(&key->addr, addr, size) &&
1311 		    key->prefixlen == prefixlen)
1312 			return key;
1313 	}
1314 	return NULL;
1315 }
1316 
1317 struct tcp_md5sig_key *tcp_v4_md5_lookup(const struct sock *sk,
1318 					 const struct sock *addr_sk)
1319 {
1320 	const union tcp_md5_addr *addr;
1321 	int l3index;
1322 
1323 	l3index = l3mdev_master_ifindex_by_index(sock_net(sk),
1324 						 addr_sk->sk_bound_dev_if);
1325 	addr = (const union tcp_md5_addr *)&addr_sk->sk_daddr;
1326 	return tcp_md5_do_lookup(sk, l3index, addr, AF_INET);
1327 }
1328 EXPORT_SYMBOL(tcp_v4_md5_lookup);
1329 
1330 static int tcp_md5sig_info_add(struct sock *sk, gfp_t gfp)
1331 {
1332 	struct tcp_sock *tp = tcp_sk(sk);
1333 	struct tcp_md5sig_info *md5sig;
1334 
1335 	md5sig = kmalloc(sizeof(*md5sig), gfp);
1336 	if (!md5sig)
1337 		return -ENOMEM;
1338 
1339 	sk_gso_disable(sk);
1340 	INIT_HLIST_HEAD(&md5sig->head);
1341 	rcu_assign_pointer(tp->md5sig_info, md5sig);
1342 	return 0;
1343 }
1344 
1345 /* This can be called on a newly created socket, from other files */
1346 static int __tcp_md5_do_add(struct sock *sk, const union tcp_md5_addr *addr,
1347 			    int family, u8 prefixlen, int l3index, u8 flags,
1348 			    const u8 *newkey, u8 newkeylen, gfp_t gfp)
1349 {
1350 	/* Add Key to the list */
1351 	struct tcp_md5sig_key *key;
1352 	struct tcp_sock *tp = tcp_sk(sk);
1353 	struct tcp_md5sig_info *md5sig;
1354 
1355 	key = tcp_md5_do_lookup_exact(sk, addr, family, prefixlen, l3index, flags);
1356 	if (key) {
1357 		/* Pre-existing entry - just update that one.
1358 		 * Note that the key might be used concurrently.
1359 		 * data_race() is telling kcsan that we do not care of
1360 		 * key mismatches, since changing MD5 key on live flows
1361 		 * can lead to packet drops.
1362 		 */
1363 		data_race(memcpy(key->key, newkey, newkeylen));
1364 
1365 		/* Pairs with READ_ONCE() in tcp_md5_hash_key().
1366 		 * Also note that a reader could catch new key->keylen value
1367 		 * but old key->key[], this is the reason we use __GFP_ZERO
1368 		 * at sock_kmalloc() time below these lines.
1369 		 */
1370 		WRITE_ONCE(key->keylen, newkeylen);
1371 
1372 		return 0;
1373 	}
1374 
1375 	md5sig = rcu_dereference_protected(tp->md5sig_info,
1376 					   lockdep_sock_is_held(sk));
1377 
1378 	key = sock_kmalloc(sk, sizeof(*key), gfp | __GFP_ZERO);
1379 	if (!key)
1380 		return -ENOMEM;
1381 
1382 	memcpy(key->key, newkey, newkeylen);
1383 	key->keylen = newkeylen;
1384 	key->family = family;
1385 	key->prefixlen = prefixlen;
1386 	key->l3index = l3index;
1387 	key->flags = flags;
1388 	memcpy(&key->addr, addr,
1389 	       (IS_ENABLED(CONFIG_IPV6) && family == AF_INET6) ? sizeof(struct in6_addr) :
1390 								 sizeof(struct in_addr));
1391 	hlist_add_head_rcu(&key->node, &md5sig->head);
1392 	return 0;
1393 }
1394 
1395 int tcp_md5_do_add(struct sock *sk, const union tcp_md5_addr *addr,
1396 		   int family, u8 prefixlen, int l3index, u8 flags,
1397 		   const u8 *newkey, u8 newkeylen)
1398 {
1399 	struct tcp_sock *tp = tcp_sk(sk);
1400 
1401 	if (!rcu_dereference_protected(tp->md5sig_info, lockdep_sock_is_held(sk))) {
1402 		if (tcp_md5_alloc_sigpool())
1403 			return -ENOMEM;
1404 
1405 		if (tcp_md5sig_info_add(sk, GFP_KERNEL)) {
1406 			tcp_md5_release_sigpool();
1407 			return -ENOMEM;
1408 		}
1409 
1410 		if (!static_branch_inc(&tcp_md5_needed.key)) {
1411 			struct tcp_md5sig_info *md5sig;
1412 
1413 			md5sig = rcu_dereference_protected(tp->md5sig_info, lockdep_sock_is_held(sk));
1414 			rcu_assign_pointer(tp->md5sig_info, NULL);
1415 			kfree_rcu(md5sig, rcu);
1416 			tcp_md5_release_sigpool();
1417 			return -EUSERS;
1418 		}
1419 	}
1420 
1421 	return __tcp_md5_do_add(sk, addr, family, prefixlen, l3index, flags,
1422 				newkey, newkeylen, GFP_KERNEL);
1423 }
1424 EXPORT_SYMBOL(tcp_md5_do_add);
1425 
1426 int tcp_md5_key_copy(struct sock *sk, const union tcp_md5_addr *addr,
1427 		     int family, u8 prefixlen, int l3index,
1428 		     struct tcp_md5sig_key *key)
1429 {
1430 	struct tcp_sock *tp = tcp_sk(sk);
1431 
1432 	if (!rcu_dereference_protected(tp->md5sig_info, lockdep_sock_is_held(sk))) {
1433 		tcp_md5_add_sigpool();
1434 
1435 		if (tcp_md5sig_info_add(sk, sk_gfp_mask(sk, GFP_ATOMIC))) {
1436 			tcp_md5_release_sigpool();
1437 			return -ENOMEM;
1438 		}
1439 
1440 		if (!static_key_fast_inc_not_disabled(&tcp_md5_needed.key.key)) {
1441 			struct tcp_md5sig_info *md5sig;
1442 
1443 			md5sig = rcu_dereference_protected(tp->md5sig_info, lockdep_sock_is_held(sk));
1444 			net_warn_ratelimited("Too many TCP-MD5 keys in the system\n");
1445 			rcu_assign_pointer(tp->md5sig_info, NULL);
1446 			kfree_rcu(md5sig, rcu);
1447 			tcp_md5_release_sigpool();
1448 			return -EUSERS;
1449 		}
1450 	}
1451 
1452 	return __tcp_md5_do_add(sk, addr, family, prefixlen, l3index,
1453 				key->flags, key->key, key->keylen,
1454 				sk_gfp_mask(sk, GFP_ATOMIC));
1455 }
1456 EXPORT_SYMBOL(tcp_md5_key_copy);
1457 
1458 int tcp_md5_do_del(struct sock *sk, const union tcp_md5_addr *addr, int family,
1459 		   u8 prefixlen, int l3index, u8 flags)
1460 {
1461 	struct tcp_md5sig_key *key;
1462 
1463 	key = tcp_md5_do_lookup_exact(sk, addr, family, prefixlen, l3index, flags);
1464 	if (!key)
1465 		return -ENOENT;
1466 	hlist_del_rcu(&key->node);
1467 	atomic_sub(sizeof(*key), &sk->sk_omem_alloc);
1468 	kfree_rcu(key, rcu);
1469 	return 0;
1470 }
1471 EXPORT_SYMBOL(tcp_md5_do_del);
1472 
1473 void tcp_clear_md5_list(struct sock *sk)
1474 {
1475 	struct tcp_sock *tp = tcp_sk(sk);
1476 	struct tcp_md5sig_key *key;
1477 	struct hlist_node *n;
1478 	struct tcp_md5sig_info *md5sig;
1479 
1480 	md5sig = rcu_dereference_protected(tp->md5sig_info, 1);
1481 
1482 	hlist_for_each_entry_safe(key, n, &md5sig->head, node) {
1483 		hlist_del_rcu(&key->node);
1484 		atomic_sub(sizeof(*key), &sk->sk_omem_alloc);
1485 		kfree_rcu(key, rcu);
1486 	}
1487 }
1488 
1489 static int tcp_v4_parse_md5_keys(struct sock *sk, int optname,
1490 				 sockptr_t optval, int optlen)
1491 {
1492 	struct tcp_md5sig cmd;
1493 	struct sockaddr_in *sin = (struct sockaddr_in *)&cmd.tcpm_addr;
1494 	const union tcp_md5_addr *addr;
1495 	u8 prefixlen = 32;
1496 	int l3index = 0;
1497 	bool l3flag;
1498 	u8 flags;
1499 
1500 	if (optlen < sizeof(cmd))
1501 		return -EINVAL;
1502 
1503 	if (copy_from_sockptr(&cmd, optval, sizeof(cmd)))
1504 		return -EFAULT;
1505 
1506 	if (sin->sin_family != AF_INET)
1507 		return -EINVAL;
1508 
1509 	flags = cmd.tcpm_flags & TCP_MD5SIG_FLAG_IFINDEX;
1510 	l3flag = cmd.tcpm_flags & TCP_MD5SIG_FLAG_IFINDEX;
1511 
1512 	if (optname == TCP_MD5SIG_EXT &&
1513 	    cmd.tcpm_flags & TCP_MD5SIG_FLAG_PREFIX) {
1514 		prefixlen = cmd.tcpm_prefixlen;
1515 		if (prefixlen > 32)
1516 			return -EINVAL;
1517 	}
1518 
1519 	if (optname == TCP_MD5SIG_EXT && cmd.tcpm_ifindex &&
1520 	    cmd.tcpm_flags & TCP_MD5SIG_FLAG_IFINDEX) {
1521 		struct net_device *dev;
1522 
1523 		rcu_read_lock();
1524 		dev = dev_get_by_index_rcu(sock_net(sk), cmd.tcpm_ifindex);
1525 		if (dev && netif_is_l3_master(dev))
1526 			l3index = dev->ifindex;
1527 
1528 		rcu_read_unlock();
1529 
1530 		/* ok to reference set/not set outside of rcu;
1531 		 * right now device MUST be an L3 master
1532 		 */
1533 		if (!dev || !l3index)
1534 			return -EINVAL;
1535 	}
1536 
1537 	addr = (union tcp_md5_addr *)&sin->sin_addr.s_addr;
1538 
1539 	if (!cmd.tcpm_keylen)
1540 		return tcp_md5_do_del(sk, addr, AF_INET, prefixlen, l3index, flags);
1541 
1542 	if (cmd.tcpm_keylen > TCP_MD5SIG_MAXKEYLEN)
1543 		return -EINVAL;
1544 
1545 	/* Don't allow keys for peers that have a matching TCP-AO key.
1546 	 * See the comment in tcp_ao_add_cmd()
1547 	 */
1548 	if (tcp_ao_required(sk, addr, AF_INET, l3flag ? l3index : -1, false))
1549 		return -EKEYREJECTED;
1550 
1551 	return tcp_md5_do_add(sk, addr, AF_INET, prefixlen, l3index, flags,
1552 			      cmd.tcpm_key, cmd.tcpm_keylen);
1553 }
1554 
1555 static int tcp_v4_md5_hash_headers(struct tcp_sigpool *hp,
1556 				   __be32 daddr, __be32 saddr,
1557 				   const struct tcphdr *th, int nbytes)
1558 {
1559 	struct tcp4_pseudohdr *bp;
1560 	struct scatterlist sg;
1561 	struct tcphdr *_th;
1562 
1563 	bp = hp->scratch;
1564 	bp->saddr = saddr;
1565 	bp->daddr = daddr;
1566 	bp->pad = 0;
1567 	bp->protocol = IPPROTO_TCP;
1568 	bp->len = cpu_to_be16(nbytes);
1569 
1570 	_th = (struct tcphdr *)(bp + 1);
1571 	memcpy(_th, th, sizeof(*th));
1572 	_th->check = 0;
1573 
1574 	sg_init_one(&sg, bp, sizeof(*bp) + sizeof(*th));
1575 	ahash_request_set_crypt(hp->req, &sg, NULL,
1576 				sizeof(*bp) + sizeof(*th));
1577 	return crypto_ahash_update(hp->req);
1578 }
1579 
1580 static int tcp_v4_md5_hash_hdr(char *md5_hash, const struct tcp_md5sig_key *key,
1581 			       __be32 daddr, __be32 saddr, const struct tcphdr *th)
1582 {
1583 	struct tcp_sigpool hp;
1584 
1585 	if (tcp_sigpool_start(tcp_md5_sigpool_id, &hp))
1586 		goto clear_hash_nostart;
1587 
1588 	if (crypto_ahash_init(hp.req))
1589 		goto clear_hash;
1590 	if (tcp_v4_md5_hash_headers(&hp, daddr, saddr, th, th->doff << 2))
1591 		goto clear_hash;
1592 	if (tcp_md5_hash_key(&hp, key))
1593 		goto clear_hash;
1594 	ahash_request_set_crypt(hp.req, NULL, md5_hash, 0);
1595 	if (crypto_ahash_final(hp.req))
1596 		goto clear_hash;
1597 
1598 	tcp_sigpool_end(&hp);
1599 	return 0;
1600 
1601 clear_hash:
1602 	tcp_sigpool_end(&hp);
1603 clear_hash_nostart:
1604 	memset(md5_hash, 0, 16);
1605 	return 1;
1606 }
1607 
1608 int tcp_v4_md5_hash_skb(char *md5_hash, const struct tcp_md5sig_key *key,
1609 			const struct sock *sk,
1610 			const struct sk_buff *skb)
1611 {
1612 	const struct tcphdr *th = tcp_hdr(skb);
1613 	struct tcp_sigpool hp;
1614 	__be32 saddr, daddr;
1615 
1616 	if (sk) { /* valid for establish/request sockets */
1617 		saddr = sk->sk_rcv_saddr;
1618 		daddr = sk->sk_daddr;
1619 	} else {
1620 		const struct iphdr *iph = ip_hdr(skb);
1621 		saddr = iph->saddr;
1622 		daddr = iph->daddr;
1623 	}
1624 
1625 	if (tcp_sigpool_start(tcp_md5_sigpool_id, &hp))
1626 		goto clear_hash_nostart;
1627 
1628 	if (crypto_ahash_init(hp.req))
1629 		goto clear_hash;
1630 
1631 	if (tcp_v4_md5_hash_headers(&hp, daddr, saddr, th, skb->len))
1632 		goto clear_hash;
1633 	if (tcp_sigpool_hash_skb_data(&hp, skb, th->doff << 2))
1634 		goto clear_hash;
1635 	if (tcp_md5_hash_key(&hp, key))
1636 		goto clear_hash;
1637 	ahash_request_set_crypt(hp.req, NULL, md5_hash, 0);
1638 	if (crypto_ahash_final(hp.req))
1639 		goto clear_hash;
1640 
1641 	tcp_sigpool_end(&hp);
1642 	return 0;
1643 
1644 clear_hash:
1645 	tcp_sigpool_end(&hp);
1646 clear_hash_nostart:
1647 	memset(md5_hash, 0, 16);
1648 	return 1;
1649 }
1650 EXPORT_SYMBOL(tcp_v4_md5_hash_skb);
1651 
1652 #endif
1653 
1654 static void tcp_v4_init_req(struct request_sock *req,
1655 			    const struct sock *sk_listener,
1656 			    struct sk_buff *skb)
1657 {
1658 	struct inet_request_sock *ireq = inet_rsk(req);
1659 	struct net *net = sock_net(sk_listener);
1660 
1661 	sk_rcv_saddr_set(req_to_sk(req), ip_hdr(skb)->daddr);
1662 	sk_daddr_set(req_to_sk(req), ip_hdr(skb)->saddr);
1663 	RCU_INIT_POINTER(ireq->ireq_opt, tcp_v4_save_options(net, skb));
1664 }
1665 
1666 static struct dst_entry *tcp_v4_route_req(const struct sock *sk,
1667 					  struct sk_buff *skb,
1668 					  struct flowi *fl,
1669 					  struct request_sock *req)
1670 {
1671 	tcp_v4_init_req(req, sk, skb);
1672 
1673 	if (security_inet_conn_request(sk, skb, req))
1674 		return NULL;
1675 
1676 	return inet_csk_route_req(sk, &fl->u.ip4, req);
1677 }
1678 
1679 struct request_sock_ops tcp_request_sock_ops __read_mostly = {
1680 	.family		=	PF_INET,
1681 	.obj_size	=	sizeof(struct tcp_request_sock),
1682 	.rtx_syn_ack	=	tcp_rtx_synack,
1683 	.send_ack	=	tcp_v4_reqsk_send_ack,
1684 	.destructor	=	tcp_v4_reqsk_destructor,
1685 	.send_reset	=	tcp_v4_send_reset,
1686 	.syn_ack_timeout =	tcp_syn_ack_timeout,
1687 };
1688 
1689 const struct tcp_request_sock_ops tcp_request_sock_ipv4_ops = {
1690 	.mss_clamp	=	TCP_MSS_DEFAULT,
1691 #ifdef CONFIG_TCP_MD5SIG
1692 	.req_md5_lookup	=	tcp_v4_md5_lookup,
1693 	.calc_md5_hash	=	tcp_v4_md5_hash_skb,
1694 #endif
1695 #ifdef CONFIG_TCP_AO
1696 	.ao_lookup	=	tcp_v4_ao_lookup_rsk,
1697 	.ao_calc_key	=	tcp_v4_ao_calc_key_rsk,
1698 	.ao_synack_hash	=	tcp_v4_ao_synack_hash,
1699 #endif
1700 #ifdef CONFIG_SYN_COOKIES
1701 	.cookie_init_seq =	cookie_v4_init_sequence,
1702 #endif
1703 	.route_req	=	tcp_v4_route_req,
1704 	.init_seq	=	tcp_v4_init_seq,
1705 	.init_ts_off	=	tcp_v4_init_ts_off,
1706 	.send_synack	=	tcp_v4_send_synack,
1707 };
1708 
1709 int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb)
1710 {
1711 	/* Never answer to SYNs send to broadcast or multicast */
1712 	if (skb_rtable(skb)->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))
1713 		goto drop;
1714 
1715 	return tcp_conn_request(&tcp_request_sock_ops,
1716 				&tcp_request_sock_ipv4_ops, sk, skb);
1717 
1718 drop:
1719 	tcp_listendrop(sk);
1720 	return 0;
1721 }
1722 EXPORT_SYMBOL(tcp_v4_conn_request);
1723 
1724 
1725 /*
1726  * The three way handshake has completed - we got a valid synack -
1727  * now create the new socket.
1728  */
1729 struct sock *tcp_v4_syn_recv_sock(const struct sock *sk, struct sk_buff *skb,
1730 				  struct request_sock *req,
1731 				  struct dst_entry *dst,
1732 				  struct request_sock *req_unhash,
1733 				  bool *own_req)
1734 {
1735 	struct inet_request_sock *ireq;
1736 	bool found_dup_sk = false;
1737 	struct inet_sock *newinet;
1738 	struct tcp_sock *newtp;
1739 	struct sock *newsk;
1740 #ifdef CONFIG_TCP_MD5SIG
1741 	const union tcp_md5_addr *addr;
1742 	struct tcp_md5sig_key *key;
1743 	int l3index;
1744 #endif
1745 	struct ip_options_rcu *inet_opt;
1746 
1747 	if (sk_acceptq_is_full(sk))
1748 		goto exit_overflow;
1749 
1750 	newsk = tcp_create_openreq_child(sk, req, skb);
1751 	if (!newsk)
1752 		goto exit_nonewsk;
1753 
1754 	newsk->sk_gso_type = SKB_GSO_TCPV4;
1755 	inet_sk_rx_dst_set(newsk, skb);
1756 
1757 	newtp		      = tcp_sk(newsk);
1758 	newinet		      = inet_sk(newsk);
1759 	ireq		      = inet_rsk(req);
1760 	sk_daddr_set(newsk, ireq->ir_rmt_addr);
1761 	sk_rcv_saddr_set(newsk, ireq->ir_loc_addr);
1762 	newsk->sk_bound_dev_if = ireq->ir_iif;
1763 	newinet->inet_saddr   = ireq->ir_loc_addr;
1764 	inet_opt	      = rcu_dereference(ireq->ireq_opt);
1765 	RCU_INIT_POINTER(newinet->inet_opt, inet_opt);
1766 	newinet->mc_index     = inet_iif(skb);
1767 	newinet->mc_ttl	      = ip_hdr(skb)->ttl;
1768 	newinet->rcv_tos      = ip_hdr(skb)->tos;
1769 	inet_csk(newsk)->icsk_ext_hdr_len = 0;
1770 	if (inet_opt)
1771 		inet_csk(newsk)->icsk_ext_hdr_len = inet_opt->opt.optlen;
1772 	atomic_set(&newinet->inet_id, get_random_u16());
1773 
1774 	/* Set ToS of the new socket based upon the value of incoming SYN.
1775 	 * ECT bits are set later in tcp_init_transfer().
1776 	 */
1777 	if (READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_reflect_tos))
1778 		newinet->tos = tcp_rsk(req)->syn_tos & ~INET_ECN_MASK;
1779 
1780 	if (!dst) {
1781 		dst = inet_csk_route_child_sock(sk, newsk, req);
1782 		if (!dst)
1783 			goto put_and_exit;
1784 	} else {
1785 		/* syncookie case : see end of cookie_v4_check() */
1786 	}
1787 	sk_setup_caps(newsk, dst);
1788 
1789 	tcp_ca_openreq_child(newsk, dst);
1790 
1791 	tcp_sync_mss(newsk, dst_mtu(dst));
1792 	newtp->advmss = tcp_mss_clamp(tcp_sk(sk), dst_metric_advmss(dst));
1793 
1794 	tcp_initialize_rcv_mss(newsk);
1795 
1796 #ifdef CONFIG_TCP_MD5SIG
1797 	l3index = l3mdev_master_ifindex_by_index(sock_net(sk), ireq->ir_iif);
1798 	/* Copy over the MD5 key from the original socket */
1799 	addr = (union tcp_md5_addr *)&newinet->inet_daddr;
1800 	key = tcp_md5_do_lookup(sk, l3index, addr, AF_INET);
1801 	if (key && !tcp_rsk_used_ao(req)) {
1802 		if (tcp_md5_key_copy(newsk, addr, AF_INET, 32, l3index, key))
1803 			goto put_and_exit;
1804 		sk_gso_disable(newsk);
1805 	}
1806 #endif
1807 #ifdef CONFIG_TCP_AO
1808 	if (tcp_ao_copy_all_matching(sk, newsk, req, skb, AF_INET))
1809 		goto put_and_exit; /* OOM, release back memory */
1810 #endif
1811 
1812 	if (__inet_inherit_port(sk, newsk) < 0)
1813 		goto put_and_exit;
1814 	*own_req = inet_ehash_nolisten(newsk, req_to_sk(req_unhash),
1815 				       &found_dup_sk);
1816 	if (likely(*own_req)) {
1817 		tcp_move_syn(newtp, req);
1818 		ireq->ireq_opt = NULL;
1819 	} else {
1820 		newinet->inet_opt = NULL;
1821 
1822 		if (!req_unhash && found_dup_sk) {
1823 			/* This code path should only be executed in the
1824 			 * syncookie case only
1825 			 */
1826 			bh_unlock_sock(newsk);
1827 			sock_put(newsk);
1828 			newsk = NULL;
1829 		}
1830 	}
1831 	return newsk;
1832 
1833 exit_overflow:
1834 	NET_INC_STATS(sock_net(sk), LINUX_MIB_LISTENOVERFLOWS);
1835 exit_nonewsk:
1836 	dst_release(dst);
1837 exit:
1838 	tcp_listendrop(sk);
1839 	return NULL;
1840 put_and_exit:
1841 	newinet->inet_opt = NULL;
1842 	inet_csk_prepare_forced_close(newsk);
1843 	tcp_done(newsk);
1844 	goto exit;
1845 }
1846 EXPORT_SYMBOL(tcp_v4_syn_recv_sock);
1847 
1848 static struct sock *tcp_v4_cookie_check(struct sock *sk, struct sk_buff *skb)
1849 {
1850 #ifdef CONFIG_SYN_COOKIES
1851 	const struct tcphdr *th = tcp_hdr(skb);
1852 
1853 	if (!th->syn)
1854 		sk = cookie_v4_check(sk, skb);
1855 #endif
1856 	return sk;
1857 }
1858 
1859 u16 tcp_v4_get_syncookie(struct sock *sk, struct iphdr *iph,
1860 			 struct tcphdr *th, u32 *cookie)
1861 {
1862 	u16 mss = 0;
1863 #ifdef CONFIG_SYN_COOKIES
1864 	mss = tcp_get_syncookie_mss(&tcp_request_sock_ops,
1865 				    &tcp_request_sock_ipv4_ops, sk, th);
1866 	if (mss) {
1867 		*cookie = __cookie_v4_init_sequence(iph, th, &mss);
1868 		tcp_synq_overflow(sk);
1869 	}
1870 #endif
1871 	return mss;
1872 }
1873 
1874 INDIRECT_CALLABLE_DECLARE(struct dst_entry *ipv4_dst_check(struct dst_entry *,
1875 							   u32));
1876 /* The socket must have it's spinlock held when we get
1877  * here, unless it is a TCP_LISTEN socket.
1878  *
1879  * We have a potential double-lock case here, so even when
1880  * doing backlog processing we use the BH locking scheme.
1881  * This is because we cannot sleep with the original spinlock
1882  * held.
1883  */
1884 int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb)
1885 {
1886 	enum skb_drop_reason reason;
1887 	struct sock *rsk;
1888 
1889 	if (sk->sk_state == TCP_ESTABLISHED) { /* Fast path */
1890 		struct dst_entry *dst;
1891 
1892 		dst = rcu_dereference_protected(sk->sk_rx_dst,
1893 						lockdep_sock_is_held(sk));
1894 
1895 		sock_rps_save_rxhash(sk, skb);
1896 		sk_mark_napi_id(sk, skb);
1897 		if (dst) {
1898 			if (sk->sk_rx_dst_ifindex != skb->skb_iif ||
1899 			    !INDIRECT_CALL_1(dst->ops->check, ipv4_dst_check,
1900 					     dst, 0)) {
1901 				RCU_INIT_POINTER(sk->sk_rx_dst, NULL);
1902 				dst_release(dst);
1903 			}
1904 		}
1905 		tcp_rcv_established(sk, skb);
1906 		return 0;
1907 	}
1908 
1909 	if (tcp_checksum_complete(skb))
1910 		goto csum_err;
1911 
1912 	if (sk->sk_state == TCP_LISTEN) {
1913 		struct sock *nsk = tcp_v4_cookie_check(sk, skb);
1914 
1915 		if (!nsk)
1916 			return 0;
1917 		if (nsk != sk) {
1918 			reason = tcp_child_process(sk, nsk, skb);
1919 			if (reason) {
1920 				rsk = nsk;
1921 				goto reset;
1922 			}
1923 			return 0;
1924 		}
1925 	} else
1926 		sock_rps_save_rxhash(sk, skb);
1927 
1928 	reason = tcp_rcv_state_process(sk, skb);
1929 	if (reason) {
1930 		rsk = sk;
1931 		goto reset;
1932 	}
1933 	return 0;
1934 
1935 reset:
1936 	tcp_v4_send_reset(rsk, skb);
1937 discard:
1938 	kfree_skb_reason(skb, reason);
1939 	/* Be careful here. If this function gets more complicated and
1940 	 * gcc suffers from register pressure on the x86, sk (in %ebx)
1941 	 * might be destroyed here. This current version compiles correctly,
1942 	 * but you have been warned.
1943 	 */
1944 	return 0;
1945 
1946 csum_err:
1947 	reason = SKB_DROP_REASON_TCP_CSUM;
1948 	trace_tcp_bad_csum(skb);
1949 	TCP_INC_STATS(sock_net(sk), TCP_MIB_CSUMERRORS);
1950 	TCP_INC_STATS(sock_net(sk), TCP_MIB_INERRS);
1951 	goto discard;
1952 }
1953 EXPORT_SYMBOL(tcp_v4_do_rcv);
1954 
1955 int tcp_v4_early_demux(struct sk_buff *skb)
1956 {
1957 	struct net *net = dev_net(skb->dev);
1958 	const struct iphdr *iph;
1959 	const struct tcphdr *th;
1960 	struct sock *sk;
1961 
1962 	if (skb->pkt_type != PACKET_HOST)
1963 		return 0;
1964 
1965 	if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct tcphdr)))
1966 		return 0;
1967 
1968 	iph = ip_hdr(skb);
1969 	th = tcp_hdr(skb);
1970 
1971 	if (th->doff < sizeof(struct tcphdr) / 4)
1972 		return 0;
1973 
1974 	sk = __inet_lookup_established(net, net->ipv4.tcp_death_row.hashinfo,
1975 				       iph->saddr, th->source,
1976 				       iph->daddr, ntohs(th->dest),
1977 				       skb->skb_iif, inet_sdif(skb));
1978 	if (sk) {
1979 		skb->sk = sk;
1980 		skb->destructor = sock_edemux;
1981 		if (sk_fullsock(sk)) {
1982 			struct dst_entry *dst = rcu_dereference(sk->sk_rx_dst);
1983 
1984 			if (dst)
1985 				dst = dst_check(dst, 0);
1986 			if (dst &&
1987 			    sk->sk_rx_dst_ifindex == skb->skb_iif)
1988 				skb_dst_set_noref(skb, dst);
1989 		}
1990 	}
1991 	return 0;
1992 }
1993 
1994 bool tcp_add_backlog(struct sock *sk, struct sk_buff *skb,
1995 		     enum skb_drop_reason *reason)
1996 {
1997 	u32 limit, tail_gso_size, tail_gso_segs;
1998 	struct skb_shared_info *shinfo;
1999 	const struct tcphdr *th;
2000 	struct tcphdr *thtail;
2001 	struct sk_buff *tail;
2002 	unsigned int hdrlen;
2003 	bool fragstolen;
2004 	u32 gso_segs;
2005 	u32 gso_size;
2006 	int delta;
2007 
2008 	/* In case all data was pulled from skb frags (in __pskb_pull_tail()),
2009 	 * we can fix skb->truesize to its real value to avoid future drops.
2010 	 * This is valid because skb is not yet charged to the socket.
2011 	 * It has been noticed pure SACK packets were sometimes dropped
2012 	 * (if cooked by drivers without copybreak feature).
2013 	 */
2014 	skb_condense(skb);
2015 
2016 	skb_dst_drop(skb);
2017 
2018 	if (unlikely(tcp_checksum_complete(skb))) {
2019 		bh_unlock_sock(sk);
2020 		trace_tcp_bad_csum(skb);
2021 		*reason = SKB_DROP_REASON_TCP_CSUM;
2022 		__TCP_INC_STATS(sock_net(sk), TCP_MIB_CSUMERRORS);
2023 		__TCP_INC_STATS(sock_net(sk), TCP_MIB_INERRS);
2024 		return true;
2025 	}
2026 
2027 	/* Attempt coalescing to last skb in backlog, even if we are
2028 	 * above the limits.
2029 	 * This is okay because skb capacity is limited to MAX_SKB_FRAGS.
2030 	 */
2031 	th = (const struct tcphdr *)skb->data;
2032 	hdrlen = th->doff * 4;
2033 
2034 	tail = sk->sk_backlog.tail;
2035 	if (!tail)
2036 		goto no_coalesce;
2037 	thtail = (struct tcphdr *)tail->data;
2038 
2039 	if (TCP_SKB_CB(tail)->end_seq != TCP_SKB_CB(skb)->seq ||
2040 	    TCP_SKB_CB(tail)->ip_dsfield != TCP_SKB_CB(skb)->ip_dsfield ||
2041 	    ((TCP_SKB_CB(tail)->tcp_flags |
2042 	      TCP_SKB_CB(skb)->tcp_flags) & (TCPHDR_SYN | TCPHDR_RST | TCPHDR_URG)) ||
2043 	    !((TCP_SKB_CB(tail)->tcp_flags &
2044 	      TCP_SKB_CB(skb)->tcp_flags) & TCPHDR_ACK) ||
2045 	    ((TCP_SKB_CB(tail)->tcp_flags ^
2046 	      TCP_SKB_CB(skb)->tcp_flags) & (TCPHDR_ECE | TCPHDR_CWR)) ||
2047 #ifdef CONFIG_TLS_DEVICE
2048 	    tail->decrypted != skb->decrypted ||
2049 #endif
2050 	    !mptcp_skb_can_collapse(tail, skb) ||
2051 	    thtail->doff != th->doff ||
2052 	    memcmp(thtail + 1, th + 1, hdrlen - sizeof(*th)))
2053 		goto no_coalesce;
2054 
2055 	__skb_pull(skb, hdrlen);
2056 
2057 	shinfo = skb_shinfo(skb);
2058 	gso_size = shinfo->gso_size ?: skb->len;
2059 	gso_segs = shinfo->gso_segs ?: 1;
2060 
2061 	shinfo = skb_shinfo(tail);
2062 	tail_gso_size = shinfo->gso_size ?: (tail->len - hdrlen);
2063 	tail_gso_segs = shinfo->gso_segs ?: 1;
2064 
2065 	if (skb_try_coalesce(tail, skb, &fragstolen, &delta)) {
2066 		TCP_SKB_CB(tail)->end_seq = TCP_SKB_CB(skb)->end_seq;
2067 
2068 		if (likely(!before(TCP_SKB_CB(skb)->ack_seq, TCP_SKB_CB(tail)->ack_seq))) {
2069 			TCP_SKB_CB(tail)->ack_seq = TCP_SKB_CB(skb)->ack_seq;
2070 			thtail->window = th->window;
2071 		}
2072 
2073 		/* We have to update both TCP_SKB_CB(tail)->tcp_flags and
2074 		 * thtail->fin, so that the fast path in tcp_rcv_established()
2075 		 * is not entered if we append a packet with a FIN.
2076 		 * SYN, RST, URG are not present.
2077 		 * ACK is set on both packets.
2078 		 * PSH : we do not really care in TCP stack,
2079 		 *       at least for 'GRO' packets.
2080 		 */
2081 		thtail->fin |= th->fin;
2082 		TCP_SKB_CB(tail)->tcp_flags |= TCP_SKB_CB(skb)->tcp_flags;
2083 
2084 		if (TCP_SKB_CB(skb)->has_rxtstamp) {
2085 			TCP_SKB_CB(tail)->has_rxtstamp = true;
2086 			tail->tstamp = skb->tstamp;
2087 			skb_hwtstamps(tail)->hwtstamp = skb_hwtstamps(skb)->hwtstamp;
2088 		}
2089 
2090 		/* Not as strict as GRO. We only need to carry mss max value */
2091 		shinfo->gso_size = max(gso_size, tail_gso_size);
2092 		shinfo->gso_segs = min_t(u32, gso_segs + tail_gso_segs, 0xFFFF);
2093 
2094 		sk->sk_backlog.len += delta;
2095 		__NET_INC_STATS(sock_net(sk),
2096 				LINUX_MIB_TCPBACKLOGCOALESCE);
2097 		kfree_skb_partial(skb, fragstolen);
2098 		return false;
2099 	}
2100 	__skb_push(skb, hdrlen);
2101 
2102 no_coalesce:
2103 	limit = (u32)READ_ONCE(sk->sk_rcvbuf) + (u32)(READ_ONCE(sk->sk_sndbuf) >> 1);
2104 
2105 	/* Only socket owner can try to collapse/prune rx queues
2106 	 * to reduce memory overhead, so add a little headroom here.
2107 	 * Few sockets backlog are possibly concurrently non empty.
2108 	 */
2109 	limit += 64 * 1024;
2110 
2111 	if (unlikely(sk_add_backlog(sk, skb, limit))) {
2112 		bh_unlock_sock(sk);
2113 		*reason = SKB_DROP_REASON_SOCKET_BACKLOG;
2114 		__NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPBACKLOGDROP);
2115 		return true;
2116 	}
2117 	return false;
2118 }
2119 EXPORT_SYMBOL(tcp_add_backlog);
2120 
2121 int tcp_filter(struct sock *sk, struct sk_buff *skb)
2122 {
2123 	struct tcphdr *th = (struct tcphdr *)skb->data;
2124 
2125 	return sk_filter_trim_cap(sk, skb, th->doff * 4);
2126 }
2127 EXPORT_SYMBOL(tcp_filter);
2128 
2129 static void tcp_v4_restore_cb(struct sk_buff *skb)
2130 {
2131 	memmove(IPCB(skb), &TCP_SKB_CB(skb)->header.h4,
2132 		sizeof(struct inet_skb_parm));
2133 }
2134 
2135 static void tcp_v4_fill_cb(struct sk_buff *skb, const struct iphdr *iph,
2136 			   const struct tcphdr *th)
2137 {
2138 	/* This is tricky : We move IPCB at its correct location into TCP_SKB_CB()
2139 	 * barrier() makes sure compiler wont play fool^Waliasing games.
2140 	 */
2141 	memmove(&TCP_SKB_CB(skb)->header.h4, IPCB(skb),
2142 		sizeof(struct inet_skb_parm));
2143 	barrier();
2144 
2145 	TCP_SKB_CB(skb)->seq = ntohl(th->seq);
2146 	TCP_SKB_CB(skb)->end_seq = (TCP_SKB_CB(skb)->seq + th->syn + th->fin +
2147 				    skb->len - th->doff * 4);
2148 	TCP_SKB_CB(skb)->ack_seq = ntohl(th->ack_seq);
2149 	TCP_SKB_CB(skb)->tcp_flags = tcp_flag_byte(th);
2150 	TCP_SKB_CB(skb)->tcp_tw_isn = 0;
2151 	TCP_SKB_CB(skb)->ip_dsfield = ipv4_get_dsfield(iph);
2152 	TCP_SKB_CB(skb)->sacked	 = 0;
2153 	TCP_SKB_CB(skb)->has_rxtstamp =
2154 			skb->tstamp || skb_hwtstamps(skb)->hwtstamp;
2155 }
2156 
2157 /*
2158  *	From tcp_input.c
2159  */
2160 
2161 int tcp_v4_rcv(struct sk_buff *skb)
2162 {
2163 	struct net *net = dev_net(skb->dev);
2164 	enum skb_drop_reason drop_reason;
2165 	int sdif = inet_sdif(skb);
2166 	int dif = inet_iif(skb);
2167 	const struct iphdr *iph;
2168 	const struct tcphdr *th;
2169 	bool refcounted;
2170 	struct sock *sk;
2171 	int ret;
2172 
2173 	drop_reason = SKB_DROP_REASON_NOT_SPECIFIED;
2174 	if (skb->pkt_type != PACKET_HOST)
2175 		goto discard_it;
2176 
2177 	/* Count it even if it's bad */
2178 	__TCP_INC_STATS(net, TCP_MIB_INSEGS);
2179 
2180 	if (!pskb_may_pull(skb, sizeof(struct tcphdr)))
2181 		goto discard_it;
2182 
2183 	th = (const struct tcphdr *)skb->data;
2184 
2185 	if (unlikely(th->doff < sizeof(struct tcphdr) / 4)) {
2186 		drop_reason = SKB_DROP_REASON_PKT_TOO_SMALL;
2187 		goto bad_packet;
2188 	}
2189 	if (!pskb_may_pull(skb, th->doff * 4))
2190 		goto discard_it;
2191 
2192 	/* An explanation is required here, I think.
2193 	 * Packet length and doff are validated by header prediction,
2194 	 * provided case of th->doff==0 is eliminated.
2195 	 * So, we defer the checks. */
2196 
2197 	if (skb_checksum_init(skb, IPPROTO_TCP, inet_compute_pseudo))
2198 		goto csum_error;
2199 
2200 	th = (const struct tcphdr *)skb->data;
2201 	iph = ip_hdr(skb);
2202 lookup:
2203 	sk = __inet_lookup_skb(net->ipv4.tcp_death_row.hashinfo,
2204 			       skb, __tcp_hdrlen(th), th->source,
2205 			       th->dest, sdif, &refcounted);
2206 	if (!sk)
2207 		goto no_tcp_socket;
2208 
2209 process:
2210 	if (sk->sk_state == TCP_TIME_WAIT)
2211 		goto do_time_wait;
2212 
2213 	if (sk->sk_state == TCP_NEW_SYN_RECV) {
2214 		struct request_sock *req = inet_reqsk(sk);
2215 		bool req_stolen = false;
2216 		struct sock *nsk;
2217 
2218 		sk = req->rsk_listener;
2219 		if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
2220 			drop_reason = SKB_DROP_REASON_XFRM_POLICY;
2221 		else
2222 			drop_reason = tcp_inbound_hash(sk, req, skb,
2223 						       &iph->saddr, &iph->daddr,
2224 						       AF_INET, dif, sdif);
2225 		if (unlikely(drop_reason)) {
2226 			sk_drops_add(sk, skb);
2227 			reqsk_put(req);
2228 			goto discard_it;
2229 		}
2230 		if (tcp_checksum_complete(skb)) {
2231 			reqsk_put(req);
2232 			goto csum_error;
2233 		}
2234 		if (unlikely(sk->sk_state != TCP_LISTEN)) {
2235 			nsk = reuseport_migrate_sock(sk, req_to_sk(req), skb);
2236 			if (!nsk) {
2237 				inet_csk_reqsk_queue_drop_and_put(sk, req);
2238 				goto lookup;
2239 			}
2240 			sk = nsk;
2241 			/* reuseport_migrate_sock() has already held one sk_refcnt
2242 			 * before returning.
2243 			 */
2244 		} else {
2245 			/* We own a reference on the listener, increase it again
2246 			 * as we might lose it too soon.
2247 			 */
2248 			sock_hold(sk);
2249 		}
2250 		refcounted = true;
2251 		nsk = NULL;
2252 		if (!tcp_filter(sk, skb)) {
2253 			th = (const struct tcphdr *)skb->data;
2254 			iph = ip_hdr(skb);
2255 			tcp_v4_fill_cb(skb, iph, th);
2256 			nsk = tcp_check_req(sk, skb, req, false, &req_stolen);
2257 		} else {
2258 			drop_reason = SKB_DROP_REASON_SOCKET_FILTER;
2259 		}
2260 		if (!nsk) {
2261 			reqsk_put(req);
2262 			if (req_stolen) {
2263 				/* Another cpu got exclusive access to req
2264 				 * and created a full blown socket.
2265 				 * Try to feed this packet to this socket
2266 				 * instead of discarding it.
2267 				 */
2268 				tcp_v4_restore_cb(skb);
2269 				sock_put(sk);
2270 				goto lookup;
2271 			}
2272 			goto discard_and_relse;
2273 		}
2274 		nf_reset_ct(skb);
2275 		if (nsk == sk) {
2276 			reqsk_put(req);
2277 			tcp_v4_restore_cb(skb);
2278 		} else {
2279 			drop_reason = tcp_child_process(sk, nsk, skb);
2280 			if (drop_reason) {
2281 				tcp_v4_send_reset(nsk, skb);
2282 				goto discard_and_relse;
2283 			}
2284 			sock_put(sk);
2285 			return 0;
2286 		}
2287 	}
2288 
2289 	if (static_branch_unlikely(&ip4_min_ttl)) {
2290 		/* min_ttl can be changed concurrently from do_ip_setsockopt() */
2291 		if (unlikely(iph->ttl < READ_ONCE(inet_sk(sk)->min_ttl))) {
2292 			__NET_INC_STATS(net, LINUX_MIB_TCPMINTTLDROP);
2293 			drop_reason = SKB_DROP_REASON_TCP_MINTTL;
2294 			goto discard_and_relse;
2295 		}
2296 	}
2297 
2298 	if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb)) {
2299 		drop_reason = SKB_DROP_REASON_XFRM_POLICY;
2300 		goto discard_and_relse;
2301 	}
2302 
2303 	drop_reason = tcp_inbound_hash(sk, NULL, skb, &iph->saddr, &iph->daddr,
2304 				       AF_INET, dif, sdif);
2305 	if (drop_reason)
2306 		goto discard_and_relse;
2307 
2308 	nf_reset_ct(skb);
2309 
2310 	if (tcp_filter(sk, skb)) {
2311 		drop_reason = SKB_DROP_REASON_SOCKET_FILTER;
2312 		goto discard_and_relse;
2313 	}
2314 	th = (const struct tcphdr *)skb->data;
2315 	iph = ip_hdr(skb);
2316 	tcp_v4_fill_cb(skb, iph, th);
2317 
2318 	skb->dev = NULL;
2319 
2320 	if (sk->sk_state == TCP_LISTEN) {
2321 		ret = tcp_v4_do_rcv(sk, skb);
2322 		goto put_and_return;
2323 	}
2324 
2325 	sk_incoming_cpu_update(sk);
2326 
2327 	bh_lock_sock_nested(sk);
2328 	tcp_segs_in(tcp_sk(sk), skb);
2329 	ret = 0;
2330 	if (!sock_owned_by_user(sk)) {
2331 		ret = tcp_v4_do_rcv(sk, skb);
2332 	} else {
2333 		if (tcp_add_backlog(sk, skb, &drop_reason))
2334 			goto discard_and_relse;
2335 	}
2336 	bh_unlock_sock(sk);
2337 
2338 put_and_return:
2339 	if (refcounted)
2340 		sock_put(sk);
2341 
2342 	return ret;
2343 
2344 no_tcp_socket:
2345 	drop_reason = SKB_DROP_REASON_NO_SOCKET;
2346 	if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
2347 		goto discard_it;
2348 
2349 	tcp_v4_fill_cb(skb, iph, th);
2350 
2351 	if (tcp_checksum_complete(skb)) {
2352 csum_error:
2353 		drop_reason = SKB_DROP_REASON_TCP_CSUM;
2354 		trace_tcp_bad_csum(skb);
2355 		__TCP_INC_STATS(net, TCP_MIB_CSUMERRORS);
2356 bad_packet:
2357 		__TCP_INC_STATS(net, TCP_MIB_INERRS);
2358 	} else {
2359 		tcp_v4_send_reset(NULL, skb);
2360 	}
2361 
2362 discard_it:
2363 	SKB_DR_OR(drop_reason, NOT_SPECIFIED);
2364 	/* Discard frame. */
2365 	kfree_skb_reason(skb, drop_reason);
2366 	return 0;
2367 
2368 discard_and_relse:
2369 	sk_drops_add(sk, skb);
2370 	if (refcounted)
2371 		sock_put(sk);
2372 	goto discard_it;
2373 
2374 do_time_wait:
2375 	if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) {
2376 		drop_reason = SKB_DROP_REASON_XFRM_POLICY;
2377 		inet_twsk_put(inet_twsk(sk));
2378 		goto discard_it;
2379 	}
2380 
2381 	tcp_v4_fill_cb(skb, iph, th);
2382 
2383 	if (tcp_checksum_complete(skb)) {
2384 		inet_twsk_put(inet_twsk(sk));
2385 		goto csum_error;
2386 	}
2387 	switch (tcp_timewait_state_process(inet_twsk(sk), skb, th)) {
2388 	case TCP_TW_SYN: {
2389 		struct sock *sk2 = inet_lookup_listener(net,
2390 							net->ipv4.tcp_death_row.hashinfo,
2391 							skb, __tcp_hdrlen(th),
2392 							iph->saddr, th->source,
2393 							iph->daddr, th->dest,
2394 							inet_iif(skb),
2395 							sdif);
2396 		if (sk2) {
2397 			inet_twsk_deschedule_put(inet_twsk(sk));
2398 			sk = sk2;
2399 			tcp_v4_restore_cb(skb);
2400 			refcounted = false;
2401 			goto process;
2402 		}
2403 	}
2404 		/* to ACK */
2405 		fallthrough;
2406 	case TCP_TW_ACK:
2407 		tcp_v4_timewait_ack(sk, skb);
2408 		break;
2409 	case TCP_TW_RST:
2410 		tcp_v4_send_reset(sk, skb);
2411 		inet_twsk_deschedule_put(inet_twsk(sk));
2412 		goto discard_it;
2413 	case TCP_TW_SUCCESS:;
2414 	}
2415 	goto discard_it;
2416 }
2417 
2418 static struct timewait_sock_ops tcp_timewait_sock_ops = {
2419 	.twsk_obj_size	= sizeof(struct tcp_timewait_sock),
2420 	.twsk_unique	= tcp_twsk_unique,
2421 	.twsk_destructor= tcp_twsk_destructor,
2422 };
2423 
2424 void inet_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb)
2425 {
2426 	struct dst_entry *dst = skb_dst(skb);
2427 
2428 	if (dst && dst_hold_safe(dst)) {
2429 		rcu_assign_pointer(sk->sk_rx_dst, dst);
2430 		sk->sk_rx_dst_ifindex = skb->skb_iif;
2431 	}
2432 }
2433 EXPORT_SYMBOL(inet_sk_rx_dst_set);
2434 
2435 const struct inet_connection_sock_af_ops ipv4_specific = {
2436 	.queue_xmit	   = ip_queue_xmit,
2437 	.send_check	   = tcp_v4_send_check,
2438 	.rebuild_header	   = inet_sk_rebuild_header,
2439 	.sk_rx_dst_set	   = inet_sk_rx_dst_set,
2440 	.conn_request	   = tcp_v4_conn_request,
2441 	.syn_recv_sock	   = tcp_v4_syn_recv_sock,
2442 	.net_header_len	   = sizeof(struct iphdr),
2443 	.setsockopt	   = ip_setsockopt,
2444 	.getsockopt	   = ip_getsockopt,
2445 	.addr2sockaddr	   = inet_csk_addr2sockaddr,
2446 	.sockaddr_len	   = sizeof(struct sockaddr_in),
2447 	.mtu_reduced	   = tcp_v4_mtu_reduced,
2448 };
2449 EXPORT_SYMBOL(ipv4_specific);
2450 
2451 #if defined(CONFIG_TCP_MD5SIG) || defined(CONFIG_TCP_AO)
2452 static const struct tcp_sock_af_ops tcp_sock_ipv4_specific = {
2453 #ifdef CONFIG_TCP_MD5SIG
2454 	.md5_lookup		= tcp_v4_md5_lookup,
2455 	.calc_md5_hash		= tcp_v4_md5_hash_skb,
2456 	.md5_parse		= tcp_v4_parse_md5_keys,
2457 #endif
2458 #ifdef CONFIG_TCP_AO
2459 	.ao_lookup		= tcp_v4_ao_lookup,
2460 	.calc_ao_hash		= tcp_v4_ao_hash_skb,
2461 	.ao_parse		= tcp_v4_parse_ao,
2462 	.ao_calc_key_sk		= tcp_v4_ao_calc_key_sk,
2463 #endif
2464 };
2465 #endif
2466 
2467 /* NOTE: A lot of things set to zero explicitly by call to
2468  *       sk_alloc() so need not be done here.
2469  */
2470 static int tcp_v4_init_sock(struct sock *sk)
2471 {
2472 	struct inet_connection_sock *icsk = inet_csk(sk);
2473 
2474 	tcp_init_sock(sk);
2475 
2476 	icsk->icsk_af_ops = &ipv4_specific;
2477 
2478 #if defined(CONFIG_TCP_MD5SIG) || defined(CONFIG_TCP_AO)
2479 	tcp_sk(sk)->af_specific = &tcp_sock_ipv4_specific;
2480 #endif
2481 
2482 	return 0;
2483 }
2484 
2485 #ifdef CONFIG_TCP_MD5SIG
2486 static void tcp_md5sig_info_free_rcu(struct rcu_head *head)
2487 {
2488 	struct tcp_md5sig_info *md5sig;
2489 
2490 	md5sig = container_of(head, struct tcp_md5sig_info, rcu);
2491 	kfree(md5sig);
2492 	static_branch_slow_dec_deferred(&tcp_md5_needed);
2493 	tcp_md5_release_sigpool();
2494 }
2495 #endif
2496 
2497 void tcp_v4_destroy_sock(struct sock *sk)
2498 {
2499 	struct tcp_sock *tp = tcp_sk(sk);
2500 
2501 	trace_tcp_destroy_sock(sk);
2502 
2503 	tcp_clear_xmit_timers(sk);
2504 
2505 	tcp_cleanup_congestion_control(sk);
2506 
2507 	tcp_cleanup_ulp(sk);
2508 
2509 	/* Cleanup up the write buffer. */
2510 	tcp_write_queue_purge(sk);
2511 
2512 	/* Check if we want to disable active TFO */
2513 	tcp_fastopen_active_disable_ofo_check(sk);
2514 
2515 	/* Cleans up our, hopefully empty, out_of_order_queue. */
2516 	skb_rbtree_purge(&tp->out_of_order_queue);
2517 
2518 #ifdef CONFIG_TCP_MD5SIG
2519 	/* Clean up the MD5 key list, if any */
2520 	if (tp->md5sig_info) {
2521 		struct tcp_md5sig_info *md5sig;
2522 
2523 		md5sig = rcu_dereference_protected(tp->md5sig_info, 1);
2524 		tcp_clear_md5_list(sk);
2525 		call_rcu(&md5sig->rcu, tcp_md5sig_info_free_rcu);
2526 		rcu_assign_pointer(tp->md5sig_info, NULL);
2527 	}
2528 #endif
2529 	tcp_ao_destroy_sock(sk, false);
2530 
2531 	/* Clean up a referenced TCP bind bucket. */
2532 	if (inet_csk(sk)->icsk_bind_hash)
2533 		inet_put_port(sk);
2534 
2535 	BUG_ON(rcu_access_pointer(tp->fastopen_rsk));
2536 
2537 	/* If socket is aborted during connect operation */
2538 	tcp_free_fastopen_req(tp);
2539 	tcp_fastopen_destroy_cipher(sk);
2540 	tcp_saved_syn_free(tp);
2541 
2542 	sk_sockets_allocated_dec(sk);
2543 }
2544 EXPORT_SYMBOL(tcp_v4_destroy_sock);
2545 
2546 #ifdef CONFIG_PROC_FS
2547 /* Proc filesystem TCP sock list dumping. */
2548 
2549 static unsigned short seq_file_family(const struct seq_file *seq);
2550 
2551 static bool seq_sk_match(struct seq_file *seq, const struct sock *sk)
2552 {
2553 	unsigned short family = seq_file_family(seq);
2554 
2555 	/* AF_UNSPEC is used as a match all */
2556 	return ((family == AF_UNSPEC || family == sk->sk_family) &&
2557 		net_eq(sock_net(sk), seq_file_net(seq)));
2558 }
2559 
2560 /* Find a non empty bucket (starting from st->bucket)
2561  * and return the first sk from it.
2562  */
2563 static void *listening_get_first(struct seq_file *seq)
2564 {
2565 	struct inet_hashinfo *hinfo = seq_file_net(seq)->ipv4.tcp_death_row.hashinfo;
2566 	struct tcp_iter_state *st = seq->private;
2567 
2568 	st->offset = 0;
2569 	for (; st->bucket <= hinfo->lhash2_mask; st->bucket++) {
2570 		struct inet_listen_hashbucket *ilb2;
2571 		struct hlist_nulls_node *node;
2572 		struct sock *sk;
2573 
2574 		ilb2 = &hinfo->lhash2[st->bucket];
2575 		if (hlist_nulls_empty(&ilb2->nulls_head))
2576 			continue;
2577 
2578 		spin_lock(&ilb2->lock);
2579 		sk_nulls_for_each(sk, node, &ilb2->nulls_head) {
2580 			if (seq_sk_match(seq, sk))
2581 				return sk;
2582 		}
2583 		spin_unlock(&ilb2->lock);
2584 	}
2585 
2586 	return NULL;
2587 }
2588 
2589 /* Find the next sk of "cur" within the same bucket (i.e. st->bucket).
2590  * If "cur" is the last one in the st->bucket,
2591  * call listening_get_first() to return the first sk of the next
2592  * non empty bucket.
2593  */
2594 static void *listening_get_next(struct seq_file *seq, void *cur)
2595 {
2596 	struct tcp_iter_state *st = seq->private;
2597 	struct inet_listen_hashbucket *ilb2;
2598 	struct hlist_nulls_node *node;
2599 	struct inet_hashinfo *hinfo;
2600 	struct sock *sk = cur;
2601 
2602 	++st->num;
2603 	++st->offset;
2604 
2605 	sk = sk_nulls_next(sk);
2606 	sk_nulls_for_each_from(sk, node) {
2607 		if (seq_sk_match(seq, sk))
2608 			return sk;
2609 	}
2610 
2611 	hinfo = seq_file_net(seq)->ipv4.tcp_death_row.hashinfo;
2612 	ilb2 = &hinfo->lhash2[st->bucket];
2613 	spin_unlock(&ilb2->lock);
2614 	++st->bucket;
2615 	return listening_get_first(seq);
2616 }
2617 
2618 static void *listening_get_idx(struct seq_file *seq, loff_t *pos)
2619 {
2620 	struct tcp_iter_state *st = seq->private;
2621 	void *rc;
2622 
2623 	st->bucket = 0;
2624 	st->offset = 0;
2625 	rc = listening_get_first(seq);
2626 
2627 	while (rc && *pos) {
2628 		rc = listening_get_next(seq, rc);
2629 		--*pos;
2630 	}
2631 	return rc;
2632 }
2633 
2634 static inline bool empty_bucket(struct inet_hashinfo *hinfo,
2635 				const struct tcp_iter_state *st)
2636 {
2637 	return hlist_nulls_empty(&hinfo->ehash[st->bucket].chain);
2638 }
2639 
2640 /*
2641  * Get first established socket starting from bucket given in st->bucket.
2642  * If st->bucket is zero, the very first socket in the hash is returned.
2643  */
2644 static void *established_get_first(struct seq_file *seq)
2645 {
2646 	struct inet_hashinfo *hinfo = seq_file_net(seq)->ipv4.tcp_death_row.hashinfo;
2647 	struct tcp_iter_state *st = seq->private;
2648 
2649 	st->offset = 0;
2650 	for (; st->bucket <= hinfo->ehash_mask; ++st->bucket) {
2651 		struct sock *sk;
2652 		struct hlist_nulls_node *node;
2653 		spinlock_t *lock = inet_ehash_lockp(hinfo, st->bucket);
2654 
2655 		cond_resched();
2656 
2657 		/* Lockless fast path for the common case of empty buckets */
2658 		if (empty_bucket(hinfo, st))
2659 			continue;
2660 
2661 		spin_lock_bh(lock);
2662 		sk_nulls_for_each(sk, node, &hinfo->ehash[st->bucket].chain) {
2663 			if (seq_sk_match(seq, sk))
2664 				return sk;
2665 		}
2666 		spin_unlock_bh(lock);
2667 	}
2668 
2669 	return NULL;
2670 }
2671 
2672 static void *established_get_next(struct seq_file *seq, void *cur)
2673 {
2674 	struct inet_hashinfo *hinfo = seq_file_net(seq)->ipv4.tcp_death_row.hashinfo;
2675 	struct tcp_iter_state *st = seq->private;
2676 	struct hlist_nulls_node *node;
2677 	struct sock *sk = cur;
2678 
2679 	++st->num;
2680 	++st->offset;
2681 
2682 	sk = sk_nulls_next(sk);
2683 
2684 	sk_nulls_for_each_from(sk, node) {
2685 		if (seq_sk_match(seq, sk))
2686 			return sk;
2687 	}
2688 
2689 	spin_unlock_bh(inet_ehash_lockp(hinfo, st->bucket));
2690 	++st->bucket;
2691 	return established_get_first(seq);
2692 }
2693 
2694 static void *established_get_idx(struct seq_file *seq, loff_t pos)
2695 {
2696 	struct tcp_iter_state *st = seq->private;
2697 	void *rc;
2698 
2699 	st->bucket = 0;
2700 	rc = established_get_first(seq);
2701 
2702 	while (rc && pos) {
2703 		rc = established_get_next(seq, rc);
2704 		--pos;
2705 	}
2706 	return rc;
2707 }
2708 
2709 static void *tcp_get_idx(struct seq_file *seq, loff_t pos)
2710 {
2711 	void *rc;
2712 	struct tcp_iter_state *st = seq->private;
2713 
2714 	st->state = TCP_SEQ_STATE_LISTENING;
2715 	rc	  = listening_get_idx(seq, &pos);
2716 
2717 	if (!rc) {
2718 		st->state = TCP_SEQ_STATE_ESTABLISHED;
2719 		rc	  = established_get_idx(seq, pos);
2720 	}
2721 
2722 	return rc;
2723 }
2724 
2725 static void *tcp_seek_last_pos(struct seq_file *seq)
2726 {
2727 	struct inet_hashinfo *hinfo = seq_file_net(seq)->ipv4.tcp_death_row.hashinfo;
2728 	struct tcp_iter_state *st = seq->private;
2729 	int bucket = st->bucket;
2730 	int offset = st->offset;
2731 	int orig_num = st->num;
2732 	void *rc = NULL;
2733 
2734 	switch (st->state) {
2735 	case TCP_SEQ_STATE_LISTENING:
2736 		if (st->bucket > hinfo->lhash2_mask)
2737 			break;
2738 		rc = listening_get_first(seq);
2739 		while (offset-- && rc && bucket == st->bucket)
2740 			rc = listening_get_next(seq, rc);
2741 		if (rc)
2742 			break;
2743 		st->bucket = 0;
2744 		st->state = TCP_SEQ_STATE_ESTABLISHED;
2745 		fallthrough;
2746 	case TCP_SEQ_STATE_ESTABLISHED:
2747 		if (st->bucket > hinfo->ehash_mask)
2748 			break;
2749 		rc = established_get_first(seq);
2750 		while (offset-- && rc && bucket == st->bucket)
2751 			rc = established_get_next(seq, rc);
2752 	}
2753 
2754 	st->num = orig_num;
2755 
2756 	return rc;
2757 }
2758 
2759 void *tcp_seq_start(struct seq_file *seq, loff_t *pos)
2760 {
2761 	struct tcp_iter_state *st = seq->private;
2762 	void *rc;
2763 
2764 	if (*pos && *pos == st->last_pos) {
2765 		rc = tcp_seek_last_pos(seq);
2766 		if (rc)
2767 			goto out;
2768 	}
2769 
2770 	st->state = TCP_SEQ_STATE_LISTENING;
2771 	st->num = 0;
2772 	st->bucket = 0;
2773 	st->offset = 0;
2774 	rc = *pos ? tcp_get_idx(seq, *pos - 1) : SEQ_START_TOKEN;
2775 
2776 out:
2777 	st->last_pos = *pos;
2778 	return rc;
2779 }
2780 EXPORT_SYMBOL(tcp_seq_start);
2781 
2782 void *tcp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2783 {
2784 	struct tcp_iter_state *st = seq->private;
2785 	void *rc = NULL;
2786 
2787 	if (v == SEQ_START_TOKEN) {
2788 		rc = tcp_get_idx(seq, 0);
2789 		goto out;
2790 	}
2791 
2792 	switch (st->state) {
2793 	case TCP_SEQ_STATE_LISTENING:
2794 		rc = listening_get_next(seq, v);
2795 		if (!rc) {
2796 			st->state = TCP_SEQ_STATE_ESTABLISHED;
2797 			st->bucket = 0;
2798 			st->offset = 0;
2799 			rc	  = established_get_first(seq);
2800 		}
2801 		break;
2802 	case TCP_SEQ_STATE_ESTABLISHED:
2803 		rc = established_get_next(seq, v);
2804 		break;
2805 	}
2806 out:
2807 	++*pos;
2808 	st->last_pos = *pos;
2809 	return rc;
2810 }
2811 EXPORT_SYMBOL(tcp_seq_next);
2812 
2813 void tcp_seq_stop(struct seq_file *seq, void *v)
2814 {
2815 	struct inet_hashinfo *hinfo = seq_file_net(seq)->ipv4.tcp_death_row.hashinfo;
2816 	struct tcp_iter_state *st = seq->private;
2817 
2818 	switch (st->state) {
2819 	case TCP_SEQ_STATE_LISTENING:
2820 		if (v != SEQ_START_TOKEN)
2821 			spin_unlock(&hinfo->lhash2[st->bucket].lock);
2822 		break;
2823 	case TCP_SEQ_STATE_ESTABLISHED:
2824 		if (v)
2825 			spin_unlock_bh(inet_ehash_lockp(hinfo, st->bucket));
2826 		break;
2827 	}
2828 }
2829 EXPORT_SYMBOL(tcp_seq_stop);
2830 
2831 static void get_openreq4(const struct request_sock *req,
2832 			 struct seq_file *f, int i)
2833 {
2834 	const struct inet_request_sock *ireq = inet_rsk(req);
2835 	long delta = req->rsk_timer.expires - jiffies;
2836 
2837 	seq_printf(f, "%4d: %08X:%04X %08X:%04X"
2838 		" %02X %08X:%08X %02X:%08lX %08X %5u %8d %u %d %pK",
2839 		i,
2840 		ireq->ir_loc_addr,
2841 		ireq->ir_num,
2842 		ireq->ir_rmt_addr,
2843 		ntohs(ireq->ir_rmt_port),
2844 		TCP_SYN_RECV,
2845 		0, 0, /* could print option size, but that is af dependent. */
2846 		1,    /* timers active (only the expire timer) */
2847 		jiffies_delta_to_clock_t(delta),
2848 		req->num_timeout,
2849 		from_kuid_munged(seq_user_ns(f),
2850 				 sock_i_uid(req->rsk_listener)),
2851 		0,  /* non standard timer */
2852 		0, /* open_requests have no inode */
2853 		0,
2854 		req);
2855 }
2856 
2857 static void get_tcp4_sock(struct sock *sk, struct seq_file *f, int i)
2858 {
2859 	int timer_active;
2860 	unsigned long timer_expires;
2861 	const struct tcp_sock *tp = tcp_sk(sk);
2862 	const struct inet_connection_sock *icsk = inet_csk(sk);
2863 	const struct inet_sock *inet = inet_sk(sk);
2864 	const struct fastopen_queue *fastopenq = &icsk->icsk_accept_queue.fastopenq;
2865 	__be32 dest = inet->inet_daddr;
2866 	__be32 src = inet->inet_rcv_saddr;
2867 	__u16 destp = ntohs(inet->inet_dport);
2868 	__u16 srcp = ntohs(inet->inet_sport);
2869 	int rx_queue;
2870 	int state;
2871 
2872 	if (icsk->icsk_pending == ICSK_TIME_RETRANS ||
2873 	    icsk->icsk_pending == ICSK_TIME_REO_TIMEOUT ||
2874 	    icsk->icsk_pending == ICSK_TIME_LOSS_PROBE) {
2875 		timer_active	= 1;
2876 		timer_expires	= icsk->icsk_timeout;
2877 	} else if (icsk->icsk_pending == ICSK_TIME_PROBE0) {
2878 		timer_active	= 4;
2879 		timer_expires	= icsk->icsk_timeout;
2880 	} else if (timer_pending(&sk->sk_timer)) {
2881 		timer_active	= 2;
2882 		timer_expires	= sk->sk_timer.expires;
2883 	} else {
2884 		timer_active	= 0;
2885 		timer_expires = jiffies;
2886 	}
2887 
2888 	state = inet_sk_state_load(sk);
2889 	if (state == TCP_LISTEN)
2890 		rx_queue = READ_ONCE(sk->sk_ack_backlog);
2891 	else
2892 		/* Because we don't lock the socket,
2893 		 * we might find a transient negative value.
2894 		 */
2895 		rx_queue = max_t(int, READ_ONCE(tp->rcv_nxt) -
2896 				      READ_ONCE(tp->copied_seq), 0);
2897 
2898 	seq_printf(f, "%4d: %08X:%04X %08X:%04X %02X %08X:%08X %02X:%08lX "
2899 			"%08X %5u %8d %lu %d %pK %lu %lu %u %u %d",
2900 		i, src, srcp, dest, destp, state,
2901 		READ_ONCE(tp->write_seq) - tp->snd_una,
2902 		rx_queue,
2903 		timer_active,
2904 		jiffies_delta_to_clock_t(timer_expires - jiffies),
2905 		icsk->icsk_retransmits,
2906 		from_kuid_munged(seq_user_ns(f), sock_i_uid(sk)),
2907 		icsk->icsk_probes_out,
2908 		sock_i_ino(sk),
2909 		refcount_read(&sk->sk_refcnt), sk,
2910 		jiffies_to_clock_t(icsk->icsk_rto),
2911 		jiffies_to_clock_t(icsk->icsk_ack.ato),
2912 		(icsk->icsk_ack.quick << 1) | inet_csk_in_pingpong_mode(sk),
2913 		tcp_snd_cwnd(tp),
2914 		state == TCP_LISTEN ?
2915 		    fastopenq->max_qlen :
2916 		    (tcp_in_initial_slowstart(tp) ? -1 : tp->snd_ssthresh));
2917 }
2918 
2919 static void get_timewait4_sock(const struct inet_timewait_sock *tw,
2920 			       struct seq_file *f, int i)
2921 {
2922 	long delta = tw->tw_timer.expires - jiffies;
2923 	__be32 dest, src;
2924 	__u16 destp, srcp;
2925 
2926 	dest  = tw->tw_daddr;
2927 	src   = tw->tw_rcv_saddr;
2928 	destp = ntohs(tw->tw_dport);
2929 	srcp  = ntohs(tw->tw_sport);
2930 
2931 	seq_printf(f, "%4d: %08X:%04X %08X:%04X"
2932 		" %02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %pK",
2933 		i, src, srcp, dest, destp, tw->tw_substate, 0, 0,
2934 		3, jiffies_delta_to_clock_t(delta), 0, 0, 0, 0,
2935 		refcount_read(&tw->tw_refcnt), tw);
2936 }
2937 
2938 #define TMPSZ 150
2939 
2940 static int tcp4_seq_show(struct seq_file *seq, void *v)
2941 {
2942 	struct tcp_iter_state *st;
2943 	struct sock *sk = v;
2944 
2945 	seq_setwidth(seq, TMPSZ - 1);
2946 	if (v == SEQ_START_TOKEN) {
2947 		seq_puts(seq, "  sl  local_address rem_address   st tx_queue "
2948 			   "rx_queue tr tm->when retrnsmt   uid  timeout "
2949 			   "inode");
2950 		goto out;
2951 	}
2952 	st = seq->private;
2953 
2954 	if (sk->sk_state == TCP_TIME_WAIT)
2955 		get_timewait4_sock(v, seq, st->num);
2956 	else if (sk->sk_state == TCP_NEW_SYN_RECV)
2957 		get_openreq4(v, seq, st->num);
2958 	else
2959 		get_tcp4_sock(v, seq, st->num);
2960 out:
2961 	seq_pad(seq, '\n');
2962 	return 0;
2963 }
2964 
2965 #ifdef CONFIG_BPF_SYSCALL
2966 struct bpf_tcp_iter_state {
2967 	struct tcp_iter_state state;
2968 	unsigned int cur_sk;
2969 	unsigned int end_sk;
2970 	unsigned int max_sk;
2971 	struct sock **batch;
2972 	bool st_bucket_done;
2973 };
2974 
2975 struct bpf_iter__tcp {
2976 	__bpf_md_ptr(struct bpf_iter_meta *, meta);
2977 	__bpf_md_ptr(struct sock_common *, sk_common);
2978 	uid_t uid __aligned(8);
2979 };
2980 
2981 static int tcp_prog_seq_show(struct bpf_prog *prog, struct bpf_iter_meta *meta,
2982 			     struct sock_common *sk_common, uid_t uid)
2983 {
2984 	struct bpf_iter__tcp ctx;
2985 
2986 	meta->seq_num--;  /* skip SEQ_START_TOKEN */
2987 	ctx.meta = meta;
2988 	ctx.sk_common = sk_common;
2989 	ctx.uid = uid;
2990 	return bpf_iter_run_prog(prog, &ctx);
2991 }
2992 
2993 static void bpf_iter_tcp_put_batch(struct bpf_tcp_iter_state *iter)
2994 {
2995 	while (iter->cur_sk < iter->end_sk)
2996 		sock_gen_put(iter->batch[iter->cur_sk++]);
2997 }
2998 
2999 static int bpf_iter_tcp_realloc_batch(struct bpf_tcp_iter_state *iter,
3000 				      unsigned int new_batch_sz)
3001 {
3002 	struct sock **new_batch;
3003 
3004 	new_batch = kvmalloc(sizeof(*new_batch) * new_batch_sz,
3005 			     GFP_USER | __GFP_NOWARN);
3006 	if (!new_batch)
3007 		return -ENOMEM;
3008 
3009 	bpf_iter_tcp_put_batch(iter);
3010 	kvfree(iter->batch);
3011 	iter->batch = new_batch;
3012 	iter->max_sk = new_batch_sz;
3013 
3014 	return 0;
3015 }
3016 
3017 static unsigned int bpf_iter_tcp_listening_batch(struct seq_file *seq,
3018 						 struct sock *start_sk)
3019 {
3020 	struct inet_hashinfo *hinfo = seq_file_net(seq)->ipv4.tcp_death_row.hashinfo;
3021 	struct bpf_tcp_iter_state *iter = seq->private;
3022 	struct tcp_iter_state *st = &iter->state;
3023 	struct hlist_nulls_node *node;
3024 	unsigned int expected = 1;
3025 	struct sock *sk;
3026 
3027 	sock_hold(start_sk);
3028 	iter->batch[iter->end_sk++] = start_sk;
3029 
3030 	sk = sk_nulls_next(start_sk);
3031 	sk_nulls_for_each_from(sk, node) {
3032 		if (seq_sk_match(seq, sk)) {
3033 			if (iter->end_sk < iter->max_sk) {
3034 				sock_hold(sk);
3035 				iter->batch[iter->end_sk++] = sk;
3036 			}
3037 			expected++;
3038 		}
3039 	}
3040 	spin_unlock(&hinfo->lhash2[st->bucket].lock);
3041 
3042 	return expected;
3043 }
3044 
3045 static unsigned int bpf_iter_tcp_established_batch(struct seq_file *seq,
3046 						   struct sock *start_sk)
3047 {
3048 	struct inet_hashinfo *hinfo = seq_file_net(seq)->ipv4.tcp_death_row.hashinfo;
3049 	struct bpf_tcp_iter_state *iter = seq->private;
3050 	struct tcp_iter_state *st = &iter->state;
3051 	struct hlist_nulls_node *node;
3052 	unsigned int expected = 1;
3053 	struct sock *sk;
3054 
3055 	sock_hold(start_sk);
3056 	iter->batch[iter->end_sk++] = start_sk;
3057 
3058 	sk = sk_nulls_next(start_sk);
3059 	sk_nulls_for_each_from(sk, node) {
3060 		if (seq_sk_match(seq, sk)) {
3061 			if (iter->end_sk < iter->max_sk) {
3062 				sock_hold(sk);
3063 				iter->batch[iter->end_sk++] = sk;
3064 			}
3065 			expected++;
3066 		}
3067 	}
3068 	spin_unlock_bh(inet_ehash_lockp(hinfo, st->bucket));
3069 
3070 	return expected;
3071 }
3072 
3073 static struct sock *bpf_iter_tcp_batch(struct seq_file *seq)
3074 {
3075 	struct inet_hashinfo *hinfo = seq_file_net(seq)->ipv4.tcp_death_row.hashinfo;
3076 	struct bpf_tcp_iter_state *iter = seq->private;
3077 	struct tcp_iter_state *st = &iter->state;
3078 	unsigned int expected;
3079 	bool resized = false;
3080 	struct sock *sk;
3081 
3082 	/* The st->bucket is done.  Directly advance to the next
3083 	 * bucket instead of having the tcp_seek_last_pos() to skip
3084 	 * one by one in the current bucket and eventually find out
3085 	 * it has to advance to the next bucket.
3086 	 */
3087 	if (iter->st_bucket_done) {
3088 		st->offset = 0;
3089 		st->bucket++;
3090 		if (st->state == TCP_SEQ_STATE_LISTENING &&
3091 		    st->bucket > hinfo->lhash2_mask) {
3092 			st->state = TCP_SEQ_STATE_ESTABLISHED;
3093 			st->bucket = 0;
3094 		}
3095 	}
3096 
3097 again:
3098 	/* Get a new batch */
3099 	iter->cur_sk = 0;
3100 	iter->end_sk = 0;
3101 	iter->st_bucket_done = false;
3102 
3103 	sk = tcp_seek_last_pos(seq);
3104 	if (!sk)
3105 		return NULL; /* Done */
3106 
3107 	if (st->state == TCP_SEQ_STATE_LISTENING)
3108 		expected = bpf_iter_tcp_listening_batch(seq, sk);
3109 	else
3110 		expected = bpf_iter_tcp_established_batch(seq, sk);
3111 
3112 	if (iter->end_sk == expected) {
3113 		iter->st_bucket_done = true;
3114 		return sk;
3115 	}
3116 
3117 	if (!resized && !bpf_iter_tcp_realloc_batch(iter, expected * 3 / 2)) {
3118 		resized = true;
3119 		goto again;
3120 	}
3121 
3122 	return sk;
3123 }
3124 
3125 static void *bpf_iter_tcp_seq_start(struct seq_file *seq, loff_t *pos)
3126 {
3127 	/* bpf iter does not support lseek, so it always
3128 	 * continue from where it was stop()-ped.
3129 	 */
3130 	if (*pos)
3131 		return bpf_iter_tcp_batch(seq);
3132 
3133 	return SEQ_START_TOKEN;
3134 }
3135 
3136 static void *bpf_iter_tcp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3137 {
3138 	struct bpf_tcp_iter_state *iter = seq->private;
3139 	struct tcp_iter_state *st = &iter->state;
3140 	struct sock *sk;
3141 
3142 	/* Whenever seq_next() is called, the iter->cur_sk is
3143 	 * done with seq_show(), so advance to the next sk in
3144 	 * the batch.
3145 	 */
3146 	if (iter->cur_sk < iter->end_sk) {
3147 		/* Keeping st->num consistent in tcp_iter_state.
3148 		 * bpf_iter_tcp does not use st->num.
3149 		 * meta.seq_num is used instead.
3150 		 */
3151 		st->num++;
3152 		/* Move st->offset to the next sk in the bucket such that
3153 		 * the future start() will resume at st->offset in
3154 		 * st->bucket.  See tcp_seek_last_pos().
3155 		 */
3156 		st->offset++;
3157 		sock_gen_put(iter->batch[iter->cur_sk++]);
3158 	}
3159 
3160 	if (iter->cur_sk < iter->end_sk)
3161 		sk = iter->batch[iter->cur_sk];
3162 	else
3163 		sk = bpf_iter_tcp_batch(seq);
3164 
3165 	++*pos;
3166 	/* Keeping st->last_pos consistent in tcp_iter_state.
3167 	 * bpf iter does not do lseek, so st->last_pos always equals to *pos.
3168 	 */
3169 	st->last_pos = *pos;
3170 	return sk;
3171 }
3172 
3173 static int bpf_iter_tcp_seq_show(struct seq_file *seq, void *v)
3174 {
3175 	struct bpf_iter_meta meta;
3176 	struct bpf_prog *prog;
3177 	struct sock *sk = v;
3178 	uid_t uid;
3179 	int ret;
3180 
3181 	if (v == SEQ_START_TOKEN)
3182 		return 0;
3183 
3184 	if (sk_fullsock(sk))
3185 		lock_sock(sk);
3186 
3187 	if (unlikely(sk_unhashed(sk))) {
3188 		ret = SEQ_SKIP;
3189 		goto unlock;
3190 	}
3191 
3192 	if (sk->sk_state == TCP_TIME_WAIT) {
3193 		uid = 0;
3194 	} else if (sk->sk_state == TCP_NEW_SYN_RECV) {
3195 		const struct request_sock *req = v;
3196 
3197 		uid = from_kuid_munged(seq_user_ns(seq),
3198 				       sock_i_uid(req->rsk_listener));
3199 	} else {
3200 		uid = from_kuid_munged(seq_user_ns(seq), sock_i_uid(sk));
3201 	}
3202 
3203 	meta.seq = seq;
3204 	prog = bpf_iter_get_info(&meta, false);
3205 	ret = tcp_prog_seq_show(prog, &meta, v, uid);
3206 
3207 unlock:
3208 	if (sk_fullsock(sk))
3209 		release_sock(sk);
3210 	return ret;
3211 
3212 }
3213 
3214 static void bpf_iter_tcp_seq_stop(struct seq_file *seq, void *v)
3215 {
3216 	struct bpf_tcp_iter_state *iter = seq->private;
3217 	struct bpf_iter_meta meta;
3218 	struct bpf_prog *prog;
3219 
3220 	if (!v) {
3221 		meta.seq = seq;
3222 		prog = bpf_iter_get_info(&meta, true);
3223 		if (prog)
3224 			(void)tcp_prog_seq_show(prog, &meta, v, 0);
3225 	}
3226 
3227 	if (iter->cur_sk < iter->end_sk) {
3228 		bpf_iter_tcp_put_batch(iter);
3229 		iter->st_bucket_done = false;
3230 	}
3231 }
3232 
3233 static const struct seq_operations bpf_iter_tcp_seq_ops = {
3234 	.show		= bpf_iter_tcp_seq_show,
3235 	.start		= bpf_iter_tcp_seq_start,
3236 	.next		= bpf_iter_tcp_seq_next,
3237 	.stop		= bpf_iter_tcp_seq_stop,
3238 };
3239 #endif
3240 static unsigned short seq_file_family(const struct seq_file *seq)
3241 {
3242 	const struct tcp_seq_afinfo *afinfo;
3243 
3244 #ifdef CONFIG_BPF_SYSCALL
3245 	/* Iterated from bpf_iter.  Let the bpf prog to filter instead. */
3246 	if (seq->op == &bpf_iter_tcp_seq_ops)
3247 		return AF_UNSPEC;
3248 #endif
3249 
3250 	/* Iterated from proc fs */
3251 	afinfo = pde_data(file_inode(seq->file));
3252 	return afinfo->family;
3253 }
3254 
3255 static const struct seq_operations tcp4_seq_ops = {
3256 	.show		= tcp4_seq_show,
3257 	.start		= tcp_seq_start,
3258 	.next		= tcp_seq_next,
3259 	.stop		= tcp_seq_stop,
3260 };
3261 
3262 static struct tcp_seq_afinfo tcp4_seq_afinfo = {
3263 	.family		= AF_INET,
3264 };
3265 
3266 static int __net_init tcp4_proc_init_net(struct net *net)
3267 {
3268 	if (!proc_create_net_data("tcp", 0444, net->proc_net, &tcp4_seq_ops,
3269 			sizeof(struct tcp_iter_state), &tcp4_seq_afinfo))
3270 		return -ENOMEM;
3271 	return 0;
3272 }
3273 
3274 static void __net_exit tcp4_proc_exit_net(struct net *net)
3275 {
3276 	remove_proc_entry("tcp", net->proc_net);
3277 }
3278 
3279 static struct pernet_operations tcp4_net_ops = {
3280 	.init = tcp4_proc_init_net,
3281 	.exit = tcp4_proc_exit_net,
3282 };
3283 
3284 int __init tcp4_proc_init(void)
3285 {
3286 	return register_pernet_subsys(&tcp4_net_ops);
3287 }
3288 
3289 void tcp4_proc_exit(void)
3290 {
3291 	unregister_pernet_subsys(&tcp4_net_ops);
3292 }
3293 #endif /* CONFIG_PROC_FS */
3294 
3295 /* @wake is one when sk_stream_write_space() calls us.
3296  * This sends EPOLLOUT only if notsent_bytes is half the limit.
3297  * This mimics the strategy used in sock_def_write_space().
3298  */
3299 bool tcp_stream_memory_free(const struct sock *sk, int wake)
3300 {
3301 	const struct tcp_sock *tp = tcp_sk(sk);
3302 	u32 notsent_bytes = READ_ONCE(tp->write_seq) -
3303 			    READ_ONCE(tp->snd_nxt);
3304 
3305 	return (notsent_bytes << wake) < tcp_notsent_lowat(tp);
3306 }
3307 EXPORT_SYMBOL(tcp_stream_memory_free);
3308 
3309 struct proto tcp_prot = {
3310 	.name			= "TCP",
3311 	.owner			= THIS_MODULE,
3312 	.close			= tcp_close,
3313 	.pre_connect		= tcp_v4_pre_connect,
3314 	.connect		= tcp_v4_connect,
3315 	.disconnect		= tcp_disconnect,
3316 	.accept			= inet_csk_accept,
3317 	.ioctl			= tcp_ioctl,
3318 	.init			= tcp_v4_init_sock,
3319 	.destroy		= tcp_v4_destroy_sock,
3320 	.shutdown		= tcp_shutdown,
3321 	.setsockopt		= tcp_setsockopt,
3322 	.getsockopt		= tcp_getsockopt,
3323 	.bpf_bypass_getsockopt	= tcp_bpf_bypass_getsockopt,
3324 	.keepalive		= tcp_set_keepalive,
3325 	.recvmsg		= tcp_recvmsg,
3326 	.sendmsg		= tcp_sendmsg,
3327 	.splice_eof		= tcp_splice_eof,
3328 	.backlog_rcv		= tcp_v4_do_rcv,
3329 	.release_cb		= tcp_release_cb,
3330 	.hash			= inet_hash,
3331 	.unhash			= inet_unhash,
3332 	.get_port		= inet_csk_get_port,
3333 	.put_port		= inet_put_port,
3334 #ifdef CONFIG_BPF_SYSCALL
3335 	.psock_update_sk_prot	= tcp_bpf_update_proto,
3336 #endif
3337 	.enter_memory_pressure	= tcp_enter_memory_pressure,
3338 	.leave_memory_pressure	= tcp_leave_memory_pressure,
3339 	.stream_memory_free	= tcp_stream_memory_free,
3340 	.sockets_allocated	= &tcp_sockets_allocated,
3341 	.orphan_count		= &tcp_orphan_count,
3342 
3343 	.memory_allocated	= &tcp_memory_allocated,
3344 	.per_cpu_fw_alloc	= &tcp_memory_per_cpu_fw_alloc,
3345 
3346 	.memory_pressure	= &tcp_memory_pressure,
3347 	.sysctl_mem		= sysctl_tcp_mem,
3348 	.sysctl_wmem_offset	= offsetof(struct net, ipv4.sysctl_tcp_wmem),
3349 	.sysctl_rmem_offset	= offsetof(struct net, ipv4.sysctl_tcp_rmem),
3350 	.max_header		= MAX_TCP_HEADER,
3351 	.obj_size		= sizeof(struct tcp_sock),
3352 	.slab_flags		= SLAB_TYPESAFE_BY_RCU,
3353 	.twsk_prot		= &tcp_timewait_sock_ops,
3354 	.rsk_prot		= &tcp_request_sock_ops,
3355 	.h.hashinfo		= NULL,
3356 	.no_autobind		= true,
3357 	.diag_destroy		= tcp_abort,
3358 };
3359 EXPORT_SYMBOL(tcp_prot);
3360 
3361 static void __net_exit tcp_sk_exit(struct net *net)
3362 {
3363 	if (net->ipv4.tcp_congestion_control)
3364 		bpf_module_put(net->ipv4.tcp_congestion_control,
3365 			       net->ipv4.tcp_congestion_control->owner);
3366 }
3367 
3368 static void __net_init tcp_set_hashinfo(struct net *net)
3369 {
3370 	struct inet_hashinfo *hinfo;
3371 	unsigned int ehash_entries;
3372 	struct net *old_net;
3373 
3374 	if (net_eq(net, &init_net))
3375 		goto fallback;
3376 
3377 	old_net = current->nsproxy->net_ns;
3378 	ehash_entries = READ_ONCE(old_net->ipv4.sysctl_tcp_child_ehash_entries);
3379 	if (!ehash_entries)
3380 		goto fallback;
3381 
3382 	ehash_entries = roundup_pow_of_two(ehash_entries);
3383 	hinfo = inet_pernet_hashinfo_alloc(&tcp_hashinfo, ehash_entries);
3384 	if (!hinfo) {
3385 		pr_warn("Failed to allocate TCP ehash (entries: %u) "
3386 			"for a netns, fallback to the global one\n",
3387 			ehash_entries);
3388 fallback:
3389 		hinfo = &tcp_hashinfo;
3390 		ehash_entries = tcp_hashinfo.ehash_mask + 1;
3391 	}
3392 
3393 	net->ipv4.tcp_death_row.hashinfo = hinfo;
3394 	net->ipv4.tcp_death_row.sysctl_max_tw_buckets = ehash_entries / 2;
3395 	net->ipv4.sysctl_max_syn_backlog = max(128U, ehash_entries / 128);
3396 }
3397 
3398 static int __net_init tcp_sk_init(struct net *net)
3399 {
3400 	net->ipv4.sysctl_tcp_ecn = 2;
3401 	net->ipv4.sysctl_tcp_ecn_fallback = 1;
3402 
3403 	net->ipv4.sysctl_tcp_base_mss = TCP_BASE_MSS;
3404 	net->ipv4.sysctl_tcp_min_snd_mss = TCP_MIN_SND_MSS;
3405 	net->ipv4.sysctl_tcp_probe_threshold = TCP_PROBE_THRESHOLD;
3406 	net->ipv4.sysctl_tcp_probe_interval = TCP_PROBE_INTERVAL;
3407 	net->ipv4.sysctl_tcp_mtu_probe_floor = TCP_MIN_SND_MSS;
3408 
3409 	net->ipv4.sysctl_tcp_keepalive_time = TCP_KEEPALIVE_TIME;
3410 	net->ipv4.sysctl_tcp_keepalive_probes = TCP_KEEPALIVE_PROBES;
3411 	net->ipv4.sysctl_tcp_keepalive_intvl = TCP_KEEPALIVE_INTVL;
3412 
3413 	net->ipv4.sysctl_tcp_syn_retries = TCP_SYN_RETRIES;
3414 	net->ipv4.sysctl_tcp_synack_retries = TCP_SYNACK_RETRIES;
3415 	net->ipv4.sysctl_tcp_syncookies = 1;
3416 	net->ipv4.sysctl_tcp_reordering = TCP_FASTRETRANS_THRESH;
3417 	net->ipv4.sysctl_tcp_retries1 = TCP_RETR1;
3418 	net->ipv4.sysctl_tcp_retries2 = TCP_RETR2;
3419 	net->ipv4.sysctl_tcp_orphan_retries = 0;
3420 	net->ipv4.sysctl_tcp_fin_timeout = TCP_FIN_TIMEOUT;
3421 	net->ipv4.sysctl_tcp_notsent_lowat = UINT_MAX;
3422 	net->ipv4.sysctl_tcp_tw_reuse = 2;
3423 	net->ipv4.sysctl_tcp_no_ssthresh_metrics_save = 1;
3424 
3425 	refcount_set(&net->ipv4.tcp_death_row.tw_refcount, 1);
3426 	tcp_set_hashinfo(net);
3427 
3428 	net->ipv4.sysctl_tcp_sack = 1;
3429 	net->ipv4.sysctl_tcp_window_scaling = 1;
3430 	net->ipv4.sysctl_tcp_timestamps = 1;
3431 	net->ipv4.sysctl_tcp_early_retrans = 3;
3432 	net->ipv4.sysctl_tcp_recovery = TCP_RACK_LOSS_DETECTION;
3433 	net->ipv4.sysctl_tcp_slow_start_after_idle = 1; /* By default, RFC2861 behavior.  */
3434 	net->ipv4.sysctl_tcp_retrans_collapse = 1;
3435 	net->ipv4.sysctl_tcp_max_reordering = 300;
3436 	net->ipv4.sysctl_tcp_dsack = 1;
3437 	net->ipv4.sysctl_tcp_app_win = 31;
3438 	net->ipv4.sysctl_tcp_adv_win_scale = 1;
3439 	net->ipv4.sysctl_tcp_frto = 2;
3440 	net->ipv4.sysctl_tcp_moderate_rcvbuf = 1;
3441 	/* This limits the percentage of the congestion window which we
3442 	 * will allow a single TSO frame to consume.  Building TSO frames
3443 	 * which are too large can cause TCP streams to be bursty.
3444 	 */
3445 	net->ipv4.sysctl_tcp_tso_win_divisor = 3;
3446 	/* Default TSQ limit of 16 TSO segments */
3447 	net->ipv4.sysctl_tcp_limit_output_bytes = 16 * 65536;
3448 
3449 	/* rfc5961 challenge ack rate limiting, per net-ns, disabled by default. */
3450 	net->ipv4.sysctl_tcp_challenge_ack_limit = INT_MAX;
3451 
3452 	net->ipv4.sysctl_tcp_min_tso_segs = 2;
3453 	net->ipv4.sysctl_tcp_tso_rtt_log = 9;  /* 2^9 = 512 usec */
3454 	net->ipv4.sysctl_tcp_min_rtt_wlen = 300;
3455 	net->ipv4.sysctl_tcp_autocorking = 1;
3456 	net->ipv4.sysctl_tcp_invalid_ratelimit = HZ/2;
3457 	net->ipv4.sysctl_tcp_pacing_ss_ratio = 200;
3458 	net->ipv4.sysctl_tcp_pacing_ca_ratio = 120;
3459 	if (net != &init_net) {
3460 		memcpy(net->ipv4.sysctl_tcp_rmem,
3461 		       init_net.ipv4.sysctl_tcp_rmem,
3462 		       sizeof(init_net.ipv4.sysctl_tcp_rmem));
3463 		memcpy(net->ipv4.sysctl_tcp_wmem,
3464 		       init_net.ipv4.sysctl_tcp_wmem,
3465 		       sizeof(init_net.ipv4.sysctl_tcp_wmem));
3466 	}
3467 	net->ipv4.sysctl_tcp_comp_sack_delay_ns = NSEC_PER_MSEC;
3468 	net->ipv4.sysctl_tcp_comp_sack_slack_ns = 100 * NSEC_PER_USEC;
3469 	net->ipv4.sysctl_tcp_comp_sack_nr = 44;
3470 	net->ipv4.sysctl_tcp_backlog_ack_defer = 1;
3471 	net->ipv4.sysctl_tcp_fastopen = TFO_CLIENT_ENABLE;
3472 	net->ipv4.sysctl_tcp_fastopen_blackhole_timeout = 0;
3473 	atomic_set(&net->ipv4.tfo_active_disable_times, 0);
3474 
3475 	/* Set default values for PLB */
3476 	net->ipv4.sysctl_tcp_plb_enabled = 0; /* Disabled by default */
3477 	net->ipv4.sysctl_tcp_plb_idle_rehash_rounds = 3;
3478 	net->ipv4.sysctl_tcp_plb_rehash_rounds = 12;
3479 	net->ipv4.sysctl_tcp_plb_suspend_rto_sec = 60;
3480 	/* Default congestion threshold for PLB to mark a round is 50% */
3481 	net->ipv4.sysctl_tcp_plb_cong_thresh = (1 << TCP_PLB_SCALE) / 2;
3482 
3483 	/* Reno is always built in */
3484 	if (!net_eq(net, &init_net) &&
3485 	    bpf_try_module_get(init_net.ipv4.tcp_congestion_control,
3486 			       init_net.ipv4.tcp_congestion_control->owner))
3487 		net->ipv4.tcp_congestion_control = init_net.ipv4.tcp_congestion_control;
3488 	else
3489 		net->ipv4.tcp_congestion_control = &tcp_reno;
3490 
3491 	net->ipv4.sysctl_tcp_syn_linear_timeouts = 4;
3492 	net->ipv4.sysctl_tcp_shrink_window = 0;
3493 
3494 	net->ipv4.sysctl_tcp_pingpong_thresh = 1;
3495 
3496 	return 0;
3497 }
3498 
3499 static void __net_exit tcp_sk_exit_batch(struct list_head *net_exit_list)
3500 {
3501 	struct net *net;
3502 
3503 	tcp_twsk_purge(net_exit_list);
3504 
3505 	list_for_each_entry(net, net_exit_list, exit_list) {
3506 		inet_pernet_hashinfo_free(net->ipv4.tcp_death_row.hashinfo);
3507 		WARN_ON_ONCE(!refcount_dec_and_test(&net->ipv4.tcp_death_row.tw_refcount));
3508 		tcp_fastopen_ctx_destroy(net);
3509 	}
3510 }
3511 
3512 static struct pernet_operations __net_initdata tcp_sk_ops = {
3513        .init	   = tcp_sk_init,
3514        .exit	   = tcp_sk_exit,
3515        .exit_batch = tcp_sk_exit_batch,
3516 };
3517 
3518 #if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
3519 DEFINE_BPF_ITER_FUNC(tcp, struct bpf_iter_meta *meta,
3520 		     struct sock_common *sk_common, uid_t uid)
3521 
3522 #define INIT_BATCH_SZ 16
3523 
3524 static int bpf_iter_init_tcp(void *priv_data, struct bpf_iter_aux_info *aux)
3525 {
3526 	struct bpf_tcp_iter_state *iter = priv_data;
3527 	int err;
3528 
3529 	err = bpf_iter_init_seq_net(priv_data, aux);
3530 	if (err)
3531 		return err;
3532 
3533 	err = bpf_iter_tcp_realloc_batch(iter, INIT_BATCH_SZ);
3534 	if (err) {
3535 		bpf_iter_fini_seq_net(priv_data);
3536 		return err;
3537 	}
3538 
3539 	return 0;
3540 }
3541 
3542 static void bpf_iter_fini_tcp(void *priv_data)
3543 {
3544 	struct bpf_tcp_iter_state *iter = priv_data;
3545 
3546 	bpf_iter_fini_seq_net(priv_data);
3547 	kvfree(iter->batch);
3548 }
3549 
3550 static const struct bpf_iter_seq_info tcp_seq_info = {
3551 	.seq_ops		= &bpf_iter_tcp_seq_ops,
3552 	.init_seq_private	= bpf_iter_init_tcp,
3553 	.fini_seq_private	= bpf_iter_fini_tcp,
3554 	.seq_priv_size		= sizeof(struct bpf_tcp_iter_state),
3555 };
3556 
3557 static const struct bpf_func_proto *
3558 bpf_iter_tcp_get_func_proto(enum bpf_func_id func_id,
3559 			    const struct bpf_prog *prog)
3560 {
3561 	switch (func_id) {
3562 	case BPF_FUNC_setsockopt:
3563 		return &bpf_sk_setsockopt_proto;
3564 	case BPF_FUNC_getsockopt:
3565 		return &bpf_sk_getsockopt_proto;
3566 	default:
3567 		return NULL;
3568 	}
3569 }
3570 
3571 static struct bpf_iter_reg tcp_reg_info = {
3572 	.target			= "tcp",
3573 	.ctx_arg_info_size	= 1,
3574 	.ctx_arg_info		= {
3575 		{ offsetof(struct bpf_iter__tcp, sk_common),
3576 		  PTR_TO_BTF_ID_OR_NULL | PTR_TRUSTED },
3577 	},
3578 	.get_func_proto		= bpf_iter_tcp_get_func_proto,
3579 	.seq_info		= &tcp_seq_info,
3580 };
3581 
3582 static void __init bpf_iter_register(void)
3583 {
3584 	tcp_reg_info.ctx_arg_info[0].btf_id = btf_sock_ids[BTF_SOCK_TYPE_SOCK_COMMON];
3585 	if (bpf_iter_reg_target(&tcp_reg_info))
3586 		pr_warn("Warning: could not register bpf iterator tcp\n");
3587 }
3588 
3589 #endif
3590 
3591 void __init tcp_v4_init(void)
3592 {
3593 	int cpu, res;
3594 
3595 	for_each_possible_cpu(cpu) {
3596 		struct sock *sk;
3597 
3598 		res = inet_ctl_sock_create(&sk, PF_INET, SOCK_RAW,
3599 					   IPPROTO_TCP, &init_net);
3600 		if (res)
3601 			panic("Failed to create the TCP control socket.\n");
3602 		sock_set_flag(sk, SOCK_USE_WRITE_QUEUE);
3603 
3604 		/* Please enforce IP_DF and IPID==0 for RST and
3605 		 * ACK sent in SYN-RECV and TIME-WAIT state.
3606 		 */
3607 		inet_sk(sk)->pmtudisc = IP_PMTUDISC_DO;
3608 
3609 		per_cpu(ipv4_tcp_sk, cpu) = sk;
3610 	}
3611 	if (register_pernet_subsys(&tcp_sk_ops))
3612 		panic("Failed to create the TCP control socket.\n");
3613 
3614 #if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
3615 	bpf_iter_register();
3616 #endif
3617 }
3618