xref: /linux/net/ipv4/tcp_ipv4.c (revision 0340dc4c82590d8735c58cf904a8aa1173273ab5)
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 		if (sk_fullsock(sk))
872 			trace_tcp_send_reset(sk, skb);
873 	}
874 
875 	BUILD_BUG_ON(offsetof(struct sock, sk_bound_dev_if) !=
876 		     offsetof(struct inet_timewait_sock, tw_bound_dev_if));
877 
878 	arg.tos = ip_hdr(skb)->tos;
879 	arg.uid = sock_net_uid(net, sk && sk_fullsock(sk) ? sk : NULL);
880 	local_bh_disable();
881 	ctl_sk = this_cpu_read(ipv4_tcp_sk);
882 	sock_net_set(ctl_sk, net);
883 	if (sk) {
884 		ctl_sk->sk_mark = (sk->sk_state == TCP_TIME_WAIT) ?
885 				   inet_twsk(sk)->tw_mark : sk->sk_mark;
886 		ctl_sk->sk_priority = (sk->sk_state == TCP_TIME_WAIT) ?
887 				   inet_twsk(sk)->tw_priority : READ_ONCE(sk->sk_priority);
888 		transmit_time = tcp_transmit_time(sk);
889 		xfrm_sk_clone_policy(ctl_sk, sk);
890 		txhash = (sk->sk_state == TCP_TIME_WAIT) ?
891 			 inet_twsk(sk)->tw_txhash : sk->sk_txhash;
892 	} else {
893 		ctl_sk->sk_mark = 0;
894 		ctl_sk->sk_priority = 0;
895 	}
896 	ip_send_unicast_reply(ctl_sk,
897 			      skb, &TCP_SKB_CB(skb)->header.h4.opt,
898 			      ip_hdr(skb)->saddr, ip_hdr(skb)->daddr,
899 			      &arg, arg.iov[0].iov_len,
900 			      transmit_time, txhash);
901 
902 	xfrm_sk_free_policy(ctl_sk);
903 	sock_net_set(ctl_sk, &init_net);
904 	__TCP_INC_STATS(net, TCP_MIB_OUTSEGS);
905 	__TCP_INC_STATS(net, TCP_MIB_OUTRSTS);
906 	local_bh_enable();
907 
908 #ifdef CONFIG_TCP_MD5SIG
909 out:
910 	rcu_read_unlock();
911 #endif
912 }
913 
914 /* The code following below sending ACKs in SYN-RECV and TIME-WAIT states
915    outside socket context is ugly, certainly. What can I do?
916  */
917 
918 static void tcp_v4_send_ack(const struct sock *sk,
919 			    struct sk_buff *skb, u32 seq, u32 ack,
920 			    u32 win, u32 tsval, u32 tsecr, int oif,
921 			    struct tcp_key *key,
922 			    int reply_flags, u8 tos, u32 txhash)
923 {
924 	const struct tcphdr *th = tcp_hdr(skb);
925 	struct {
926 		struct tcphdr th;
927 		__be32 opt[(MAX_TCP_OPTION_SPACE  >> 2)];
928 	} rep;
929 	struct net *net = sock_net(sk);
930 	struct ip_reply_arg arg;
931 	struct sock *ctl_sk;
932 	u64 transmit_time;
933 
934 	memset(&rep.th, 0, sizeof(struct tcphdr));
935 	memset(&arg, 0, sizeof(arg));
936 
937 	arg.iov[0].iov_base = (unsigned char *)&rep;
938 	arg.iov[0].iov_len  = sizeof(rep.th);
939 	if (tsecr) {
940 		rep.opt[0] = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
941 				   (TCPOPT_TIMESTAMP << 8) |
942 				   TCPOLEN_TIMESTAMP);
943 		rep.opt[1] = htonl(tsval);
944 		rep.opt[2] = htonl(tsecr);
945 		arg.iov[0].iov_len += TCPOLEN_TSTAMP_ALIGNED;
946 	}
947 
948 	/* Swap the send and the receive. */
949 	rep.th.dest    = th->source;
950 	rep.th.source  = th->dest;
951 	rep.th.doff    = arg.iov[0].iov_len / 4;
952 	rep.th.seq     = htonl(seq);
953 	rep.th.ack_seq = htonl(ack);
954 	rep.th.ack     = 1;
955 	rep.th.window  = htons(win);
956 
957 #ifdef CONFIG_TCP_MD5SIG
958 	if (tcp_key_is_md5(key)) {
959 		int offset = (tsecr) ? 3 : 0;
960 
961 		rep.opt[offset++] = htonl((TCPOPT_NOP << 24) |
962 					  (TCPOPT_NOP << 16) |
963 					  (TCPOPT_MD5SIG << 8) |
964 					  TCPOLEN_MD5SIG);
965 		arg.iov[0].iov_len += TCPOLEN_MD5SIG_ALIGNED;
966 		rep.th.doff = arg.iov[0].iov_len/4;
967 
968 		tcp_v4_md5_hash_hdr((__u8 *) &rep.opt[offset],
969 				    key->md5_key, ip_hdr(skb)->saddr,
970 				    ip_hdr(skb)->daddr, &rep.th);
971 	}
972 #endif
973 #ifdef CONFIG_TCP_AO
974 	if (tcp_key_is_ao(key)) {
975 		int offset = (tsecr) ? 3 : 0;
976 
977 		rep.opt[offset++] = htonl((TCPOPT_AO << 24) |
978 					  (tcp_ao_len(key->ao_key) << 16) |
979 					  (key->ao_key->sndid << 8) |
980 					  key->rcv_next);
981 		arg.iov[0].iov_len += tcp_ao_len_aligned(key->ao_key);
982 		rep.th.doff = arg.iov[0].iov_len / 4;
983 
984 		tcp_ao_hash_hdr(AF_INET, (char *)&rep.opt[offset],
985 				key->ao_key, key->traffic_key,
986 				(union tcp_ao_addr *)&ip_hdr(skb)->saddr,
987 				(union tcp_ao_addr *)&ip_hdr(skb)->daddr,
988 				&rep.th, key->sne);
989 	}
990 #endif
991 	arg.flags = reply_flags;
992 	arg.csum = csum_tcpudp_nofold(ip_hdr(skb)->daddr,
993 				      ip_hdr(skb)->saddr, /* XXX */
994 				      arg.iov[0].iov_len, IPPROTO_TCP, 0);
995 	arg.csumoffset = offsetof(struct tcphdr, check) / 2;
996 	if (oif)
997 		arg.bound_dev_if = oif;
998 	arg.tos = tos;
999 	arg.uid = sock_net_uid(net, sk_fullsock(sk) ? sk : NULL);
1000 	local_bh_disable();
1001 	ctl_sk = this_cpu_read(ipv4_tcp_sk);
1002 	sock_net_set(ctl_sk, net);
1003 	ctl_sk->sk_mark = (sk->sk_state == TCP_TIME_WAIT) ?
1004 			   inet_twsk(sk)->tw_mark : READ_ONCE(sk->sk_mark);
1005 	ctl_sk->sk_priority = (sk->sk_state == TCP_TIME_WAIT) ?
1006 			   inet_twsk(sk)->tw_priority : READ_ONCE(sk->sk_priority);
1007 	transmit_time = tcp_transmit_time(sk);
1008 	ip_send_unicast_reply(ctl_sk,
1009 			      skb, &TCP_SKB_CB(skb)->header.h4.opt,
1010 			      ip_hdr(skb)->saddr, ip_hdr(skb)->daddr,
1011 			      &arg, arg.iov[0].iov_len,
1012 			      transmit_time, txhash);
1013 
1014 	sock_net_set(ctl_sk, &init_net);
1015 	__TCP_INC_STATS(net, TCP_MIB_OUTSEGS);
1016 	local_bh_enable();
1017 }
1018 
1019 static void tcp_v4_timewait_ack(struct sock *sk, struct sk_buff *skb)
1020 {
1021 	struct inet_timewait_sock *tw = inet_twsk(sk);
1022 	struct tcp_timewait_sock *tcptw = tcp_twsk(sk);
1023 	struct tcp_key key = {};
1024 #ifdef CONFIG_TCP_AO
1025 	struct tcp_ao_info *ao_info;
1026 
1027 	if (static_branch_unlikely(&tcp_ao_needed.key)) {
1028 		/* FIXME: the segment to-be-acked is not verified yet */
1029 		ao_info = rcu_dereference(tcptw->ao_info);
1030 		if (ao_info) {
1031 			const struct tcp_ao_hdr *aoh;
1032 
1033 			if (tcp_parse_auth_options(tcp_hdr(skb), NULL, &aoh)) {
1034 				inet_twsk_put(tw);
1035 				return;
1036 			}
1037 
1038 			if (aoh)
1039 				key.ao_key = tcp_ao_established_key(ao_info, aoh->rnext_keyid, -1);
1040 		}
1041 	}
1042 	if (key.ao_key) {
1043 		struct tcp_ao_key *rnext_key;
1044 
1045 		key.traffic_key = snd_other_key(key.ao_key);
1046 		key.sne = READ_ONCE(ao_info->snd_sne);
1047 		rnext_key = READ_ONCE(ao_info->rnext_key);
1048 		key.rcv_next = rnext_key->rcvid;
1049 		key.type = TCP_KEY_AO;
1050 #else
1051 	if (0) {
1052 #endif
1053 #ifdef CONFIG_TCP_MD5SIG
1054 	} else if (static_branch_unlikely(&tcp_md5_needed.key)) {
1055 		key.md5_key = tcp_twsk_md5_key(tcptw);
1056 		if (key.md5_key)
1057 			key.type = TCP_KEY_MD5;
1058 #endif
1059 	}
1060 
1061 	tcp_v4_send_ack(sk, skb,
1062 			tcptw->tw_snd_nxt, tcptw->tw_rcv_nxt,
1063 			tcptw->tw_rcv_wnd >> tw->tw_rcv_wscale,
1064 			tcp_tw_tsval(tcptw),
1065 			tcptw->tw_ts_recent,
1066 			tw->tw_bound_dev_if, &key,
1067 			tw->tw_transparent ? IP_REPLY_ARG_NOSRCCHECK : 0,
1068 			tw->tw_tos,
1069 			tw->tw_txhash);
1070 
1071 	inet_twsk_put(tw);
1072 }
1073 
1074 static void tcp_v4_reqsk_send_ack(const struct sock *sk, struct sk_buff *skb,
1075 				  struct request_sock *req)
1076 {
1077 	struct tcp_key key = {};
1078 
1079 	/* sk->sk_state == TCP_LISTEN -> for regular TCP_SYN_RECV
1080 	 * sk->sk_state == TCP_SYN_RECV -> for Fast Open.
1081 	 */
1082 	u32 seq = (sk->sk_state == TCP_LISTEN) ? tcp_rsk(req)->snt_isn + 1 :
1083 					     tcp_sk(sk)->snd_nxt;
1084 
1085 #ifdef CONFIG_TCP_AO
1086 	if (static_branch_unlikely(&tcp_ao_needed.key) &&
1087 	    tcp_rsk_used_ao(req)) {
1088 		const union tcp_md5_addr *addr;
1089 		const struct tcp_ao_hdr *aoh;
1090 		int l3index;
1091 
1092 		/* Invalid TCP option size or twice included auth */
1093 		if (tcp_parse_auth_options(tcp_hdr(skb), NULL, &aoh))
1094 			return;
1095 		if (!aoh)
1096 			return;
1097 
1098 		addr = (union tcp_md5_addr *)&ip_hdr(skb)->saddr;
1099 		l3index = tcp_v4_sdif(skb) ? inet_iif(skb) : 0;
1100 		key.ao_key = tcp_ao_do_lookup(sk, l3index, addr, AF_INET,
1101 					      aoh->rnext_keyid, -1);
1102 		if (unlikely(!key.ao_key)) {
1103 			/* Send ACK with any matching MKT for the peer */
1104 			key.ao_key = tcp_ao_do_lookup(sk, l3index, addr, AF_INET, -1, -1);
1105 			/* Matching key disappeared (user removed the key?)
1106 			 * let the handshake timeout.
1107 			 */
1108 			if (!key.ao_key) {
1109 				net_info_ratelimited("TCP-AO key for (%pI4, %d)->(%pI4, %d) suddenly disappeared, won't ACK new connection\n",
1110 						     addr,
1111 						     ntohs(tcp_hdr(skb)->source),
1112 						     &ip_hdr(skb)->daddr,
1113 						     ntohs(tcp_hdr(skb)->dest));
1114 				return;
1115 			}
1116 		}
1117 		key.traffic_key = kmalloc(tcp_ao_digest_size(key.ao_key), GFP_ATOMIC);
1118 		if (!key.traffic_key)
1119 			return;
1120 
1121 		key.type = TCP_KEY_AO;
1122 		key.rcv_next = aoh->keyid;
1123 		tcp_v4_ao_calc_key_rsk(key.ao_key, key.traffic_key, req);
1124 #else
1125 	if (0) {
1126 #endif
1127 #ifdef CONFIG_TCP_MD5SIG
1128 	} else if (static_branch_unlikely(&tcp_md5_needed.key)) {
1129 		const union tcp_md5_addr *addr;
1130 		int l3index;
1131 
1132 		addr = (union tcp_md5_addr *)&ip_hdr(skb)->saddr;
1133 		l3index = tcp_v4_sdif(skb) ? inet_iif(skb) : 0;
1134 		key.md5_key = tcp_md5_do_lookup(sk, l3index, addr, AF_INET);
1135 		if (key.md5_key)
1136 			key.type = TCP_KEY_MD5;
1137 #endif
1138 	}
1139 
1140 	/* RFC 7323 2.3
1141 	 * The window field (SEG.WND) of every outgoing segment, with the
1142 	 * exception of <SYN> segments, MUST be right-shifted by
1143 	 * Rcv.Wind.Shift bits:
1144 	 */
1145 	tcp_v4_send_ack(sk, skb, seq,
1146 			tcp_rsk(req)->rcv_nxt,
1147 			req->rsk_rcv_wnd >> inet_rsk(req)->rcv_wscale,
1148 			tcp_rsk_tsval(tcp_rsk(req)),
1149 			READ_ONCE(req->ts_recent),
1150 			0, &key,
1151 			inet_rsk(req)->no_srccheck ? IP_REPLY_ARG_NOSRCCHECK : 0,
1152 			ip_hdr(skb)->tos,
1153 			READ_ONCE(tcp_rsk(req)->txhash));
1154 	if (tcp_key_is_ao(&key))
1155 		kfree(key.traffic_key);
1156 }
1157 
1158 /*
1159  *	Send a SYN-ACK after having received a SYN.
1160  *	This still operates on a request_sock only, not on a big
1161  *	socket.
1162  */
1163 static int tcp_v4_send_synack(const struct sock *sk, struct dst_entry *dst,
1164 			      struct flowi *fl,
1165 			      struct request_sock *req,
1166 			      struct tcp_fastopen_cookie *foc,
1167 			      enum tcp_synack_type synack_type,
1168 			      struct sk_buff *syn_skb)
1169 {
1170 	const struct inet_request_sock *ireq = inet_rsk(req);
1171 	struct flowi4 fl4;
1172 	int err = -1;
1173 	struct sk_buff *skb;
1174 	u8 tos;
1175 
1176 	/* First, grab a route. */
1177 	if (!dst && (dst = inet_csk_route_req(sk, &fl4, req)) == NULL)
1178 		return -1;
1179 
1180 	skb = tcp_make_synack(sk, dst, req, foc, synack_type, syn_skb);
1181 
1182 	if (skb) {
1183 		__tcp_v4_send_check(skb, ireq->ir_loc_addr, ireq->ir_rmt_addr);
1184 
1185 		tos = READ_ONCE(inet_sk(sk)->tos);
1186 
1187 		if (READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_reflect_tos))
1188 			tos = (tcp_rsk(req)->syn_tos & ~INET_ECN_MASK) |
1189 			      (tos & INET_ECN_MASK);
1190 
1191 		if (!INET_ECN_is_capable(tos) &&
1192 		    tcp_bpf_ca_needs_ecn((struct sock *)req))
1193 			tos |= INET_ECN_ECT_0;
1194 
1195 		rcu_read_lock();
1196 		err = ip_build_and_send_pkt(skb, sk, ireq->ir_loc_addr,
1197 					    ireq->ir_rmt_addr,
1198 					    rcu_dereference(ireq->ireq_opt),
1199 					    tos);
1200 		rcu_read_unlock();
1201 		err = net_xmit_eval(err);
1202 	}
1203 
1204 	return err;
1205 }
1206 
1207 /*
1208  *	IPv4 request_sock destructor.
1209  */
1210 static void tcp_v4_reqsk_destructor(struct request_sock *req)
1211 {
1212 	kfree(rcu_dereference_protected(inet_rsk(req)->ireq_opt, 1));
1213 }
1214 
1215 #ifdef CONFIG_TCP_MD5SIG
1216 /*
1217  * RFC2385 MD5 checksumming requires a mapping of
1218  * IP address->MD5 Key.
1219  * We need to maintain these in the sk structure.
1220  */
1221 
1222 DEFINE_STATIC_KEY_DEFERRED_FALSE(tcp_md5_needed, HZ);
1223 EXPORT_SYMBOL(tcp_md5_needed);
1224 
1225 static bool better_md5_match(struct tcp_md5sig_key *old, struct tcp_md5sig_key *new)
1226 {
1227 	if (!old)
1228 		return true;
1229 
1230 	/* l3index always overrides non-l3index */
1231 	if (old->l3index && new->l3index == 0)
1232 		return false;
1233 	if (old->l3index == 0 && new->l3index)
1234 		return true;
1235 
1236 	return old->prefixlen < new->prefixlen;
1237 }
1238 
1239 /* Find the Key structure for an address.  */
1240 struct tcp_md5sig_key *__tcp_md5_do_lookup(const struct sock *sk, int l3index,
1241 					   const union tcp_md5_addr *addr,
1242 					   int family, bool any_l3index)
1243 {
1244 	const struct tcp_sock *tp = tcp_sk(sk);
1245 	struct tcp_md5sig_key *key;
1246 	const struct tcp_md5sig_info *md5sig;
1247 	__be32 mask;
1248 	struct tcp_md5sig_key *best_match = NULL;
1249 	bool match;
1250 
1251 	/* caller either holds rcu_read_lock() or socket lock */
1252 	md5sig = rcu_dereference_check(tp->md5sig_info,
1253 				       lockdep_sock_is_held(sk));
1254 	if (!md5sig)
1255 		return NULL;
1256 
1257 	hlist_for_each_entry_rcu(key, &md5sig->head, node,
1258 				 lockdep_sock_is_held(sk)) {
1259 		if (key->family != family)
1260 			continue;
1261 		if (!any_l3index && key->flags & TCP_MD5SIG_FLAG_IFINDEX &&
1262 		    key->l3index != l3index)
1263 			continue;
1264 		if (family == AF_INET) {
1265 			mask = inet_make_mask(key->prefixlen);
1266 			match = (key->addr.a4.s_addr & mask) ==
1267 				(addr->a4.s_addr & mask);
1268 #if IS_ENABLED(CONFIG_IPV6)
1269 		} else if (family == AF_INET6) {
1270 			match = ipv6_prefix_equal(&key->addr.a6, &addr->a6,
1271 						  key->prefixlen);
1272 #endif
1273 		} else {
1274 			match = false;
1275 		}
1276 
1277 		if (match && better_md5_match(best_match, key))
1278 			best_match = key;
1279 	}
1280 	return best_match;
1281 }
1282 EXPORT_SYMBOL(__tcp_md5_do_lookup);
1283 
1284 static struct tcp_md5sig_key *tcp_md5_do_lookup_exact(const struct sock *sk,
1285 						      const union tcp_md5_addr *addr,
1286 						      int family, u8 prefixlen,
1287 						      int l3index, u8 flags)
1288 {
1289 	const struct tcp_sock *tp = tcp_sk(sk);
1290 	struct tcp_md5sig_key *key;
1291 	unsigned int size = sizeof(struct in_addr);
1292 	const struct tcp_md5sig_info *md5sig;
1293 
1294 	/* caller either holds rcu_read_lock() or socket lock */
1295 	md5sig = rcu_dereference_check(tp->md5sig_info,
1296 				       lockdep_sock_is_held(sk));
1297 	if (!md5sig)
1298 		return NULL;
1299 #if IS_ENABLED(CONFIG_IPV6)
1300 	if (family == AF_INET6)
1301 		size = sizeof(struct in6_addr);
1302 #endif
1303 	hlist_for_each_entry_rcu(key, &md5sig->head, node,
1304 				 lockdep_sock_is_held(sk)) {
1305 		if (key->family != family)
1306 			continue;
1307 		if ((key->flags & TCP_MD5SIG_FLAG_IFINDEX) != (flags & TCP_MD5SIG_FLAG_IFINDEX))
1308 			continue;
1309 		if (key->l3index != l3index)
1310 			continue;
1311 		if (!memcmp(&key->addr, addr, size) &&
1312 		    key->prefixlen == prefixlen)
1313 			return key;
1314 	}
1315 	return NULL;
1316 }
1317 
1318 struct tcp_md5sig_key *tcp_v4_md5_lookup(const struct sock *sk,
1319 					 const struct sock *addr_sk)
1320 {
1321 	const union tcp_md5_addr *addr;
1322 	int l3index;
1323 
1324 	l3index = l3mdev_master_ifindex_by_index(sock_net(sk),
1325 						 addr_sk->sk_bound_dev_if);
1326 	addr = (const union tcp_md5_addr *)&addr_sk->sk_daddr;
1327 	return tcp_md5_do_lookup(sk, l3index, addr, AF_INET);
1328 }
1329 EXPORT_SYMBOL(tcp_v4_md5_lookup);
1330 
1331 static int tcp_md5sig_info_add(struct sock *sk, gfp_t gfp)
1332 {
1333 	struct tcp_sock *tp = tcp_sk(sk);
1334 	struct tcp_md5sig_info *md5sig;
1335 
1336 	md5sig = kmalloc(sizeof(*md5sig), gfp);
1337 	if (!md5sig)
1338 		return -ENOMEM;
1339 
1340 	sk_gso_disable(sk);
1341 	INIT_HLIST_HEAD(&md5sig->head);
1342 	rcu_assign_pointer(tp->md5sig_info, md5sig);
1343 	return 0;
1344 }
1345 
1346 /* This can be called on a newly created socket, from other files */
1347 static int __tcp_md5_do_add(struct sock *sk, const union tcp_md5_addr *addr,
1348 			    int family, u8 prefixlen, int l3index, u8 flags,
1349 			    const u8 *newkey, u8 newkeylen, gfp_t gfp)
1350 {
1351 	/* Add Key to the list */
1352 	struct tcp_md5sig_key *key;
1353 	struct tcp_sock *tp = tcp_sk(sk);
1354 	struct tcp_md5sig_info *md5sig;
1355 
1356 	key = tcp_md5_do_lookup_exact(sk, addr, family, prefixlen, l3index, flags);
1357 	if (key) {
1358 		/* Pre-existing entry - just update that one.
1359 		 * Note that the key might be used concurrently.
1360 		 * data_race() is telling kcsan that we do not care of
1361 		 * key mismatches, since changing MD5 key on live flows
1362 		 * can lead to packet drops.
1363 		 */
1364 		data_race(memcpy(key->key, newkey, newkeylen));
1365 
1366 		/* Pairs with READ_ONCE() in tcp_md5_hash_key().
1367 		 * Also note that a reader could catch new key->keylen value
1368 		 * but old key->key[], this is the reason we use __GFP_ZERO
1369 		 * at sock_kmalloc() time below these lines.
1370 		 */
1371 		WRITE_ONCE(key->keylen, newkeylen);
1372 
1373 		return 0;
1374 	}
1375 
1376 	md5sig = rcu_dereference_protected(tp->md5sig_info,
1377 					   lockdep_sock_is_held(sk));
1378 
1379 	key = sock_kmalloc(sk, sizeof(*key), gfp | __GFP_ZERO);
1380 	if (!key)
1381 		return -ENOMEM;
1382 
1383 	memcpy(key->key, newkey, newkeylen);
1384 	key->keylen = newkeylen;
1385 	key->family = family;
1386 	key->prefixlen = prefixlen;
1387 	key->l3index = l3index;
1388 	key->flags = flags;
1389 	memcpy(&key->addr, addr,
1390 	       (IS_ENABLED(CONFIG_IPV6) && family == AF_INET6) ? sizeof(struct in6_addr) :
1391 								 sizeof(struct in_addr));
1392 	hlist_add_head_rcu(&key->node, &md5sig->head);
1393 	return 0;
1394 }
1395 
1396 int tcp_md5_do_add(struct sock *sk, const union tcp_md5_addr *addr,
1397 		   int family, u8 prefixlen, int l3index, u8 flags,
1398 		   const u8 *newkey, u8 newkeylen)
1399 {
1400 	struct tcp_sock *tp = tcp_sk(sk);
1401 
1402 	if (!rcu_dereference_protected(tp->md5sig_info, lockdep_sock_is_held(sk))) {
1403 		if (tcp_md5_alloc_sigpool())
1404 			return -ENOMEM;
1405 
1406 		if (tcp_md5sig_info_add(sk, GFP_KERNEL)) {
1407 			tcp_md5_release_sigpool();
1408 			return -ENOMEM;
1409 		}
1410 
1411 		if (!static_branch_inc(&tcp_md5_needed.key)) {
1412 			struct tcp_md5sig_info *md5sig;
1413 
1414 			md5sig = rcu_dereference_protected(tp->md5sig_info, lockdep_sock_is_held(sk));
1415 			rcu_assign_pointer(tp->md5sig_info, NULL);
1416 			kfree_rcu(md5sig, rcu);
1417 			tcp_md5_release_sigpool();
1418 			return -EUSERS;
1419 		}
1420 	}
1421 
1422 	return __tcp_md5_do_add(sk, addr, family, prefixlen, l3index, flags,
1423 				newkey, newkeylen, GFP_KERNEL);
1424 }
1425 EXPORT_SYMBOL(tcp_md5_do_add);
1426 
1427 int tcp_md5_key_copy(struct sock *sk, const union tcp_md5_addr *addr,
1428 		     int family, u8 prefixlen, int l3index,
1429 		     struct tcp_md5sig_key *key)
1430 {
1431 	struct tcp_sock *tp = tcp_sk(sk);
1432 
1433 	if (!rcu_dereference_protected(tp->md5sig_info, lockdep_sock_is_held(sk))) {
1434 		tcp_md5_add_sigpool();
1435 
1436 		if (tcp_md5sig_info_add(sk, sk_gfp_mask(sk, GFP_ATOMIC))) {
1437 			tcp_md5_release_sigpool();
1438 			return -ENOMEM;
1439 		}
1440 
1441 		if (!static_key_fast_inc_not_disabled(&tcp_md5_needed.key.key)) {
1442 			struct tcp_md5sig_info *md5sig;
1443 
1444 			md5sig = rcu_dereference_protected(tp->md5sig_info, lockdep_sock_is_held(sk));
1445 			net_warn_ratelimited("Too many TCP-MD5 keys in the system\n");
1446 			rcu_assign_pointer(tp->md5sig_info, NULL);
1447 			kfree_rcu(md5sig, rcu);
1448 			tcp_md5_release_sigpool();
1449 			return -EUSERS;
1450 		}
1451 	}
1452 
1453 	return __tcp_md5_do_add(sk, addr, family, prefixlen, l3index,
1454 				key->flags, key->key, key->keylen,
1455 				sk_gfp_mask(sk, GFP_ATOMIC));
1456 }
1457 EXPORT_SYMBOL(tcp_md5_key_copy);
1458 
1459 int tcp_md5_do_del(struct sock *sk, const union tcp_md5_addr *addr, int family,
1460 		   u8 prefixlen, int l3index, u8 flags)
1461 {
1462 	struct tcp_md5sig_key *key;
1463 
1464 	key = tcp_md5_do_lookup_exact(sk, addr, family, prefixlen, l3index, flags);
1465 	if (!key)
1466 		return -ENOENT;
1467 	hlist_del_rcu(&key->node);
1468 	atomic_sub(sizeof(*key), &sk->sk_omem_alloc);
1469 	kfree_rcu(key, rcu);
1470 	return 0;
1471 }
1472 EXPORT_SYMBOL(tcp_md5_do_del);
1473 
1474 void tcp_clear_md5_list(struct sock *sk)
1475 {
1476 	struct tcp_sock *tp = tcp_sk(sk);
1477 	struct tcp_md5sig_key *key;
1478 	struct hlist_node *n;
1479 	struct tcp_md5sig_info *md5sig;
1480 
1481 	md5sig = rcu_dereference_protected(tp->md5sig_info, 1);
1482 
1483 	hlist_for_each_entry_safe(key, n, &md5sig->head, node) {
1484 		hlist_del_rcu(&key->node);
1485 		atomic_sub(sizeof(*key), &sk->sk_omem_alloc);
1486 		kfree_rcu(key, rcu);
1487 	}
1488 }
1489 
1490 static int tcp_v4_parse_md5_keys(struct sock *sk, int optname,
1491 				 sockptr_t optval, int optlen)
1492 {
1493 	struct tcp_md5sig cmd;
1494 	struct sockaddr_in *sin = (struct sockaddr_in *)&cmd.tcpm_addr;
1495 	const union tcp_md5_addr *addr;
1496 	u8 prefixlen = 32;
1497 	int l3index = 0;
1498 	bool l3flag;
1499 	u8 flags;
1500 
1501 	if (optlen < sizeof(cmd))
1502 		return -EINVAL;
1503 
1504 	if (copy_from_sockptr(&cmd, optval, sizeof(cmd)))
1505 		return -EFAULT;
1506 
1507 	if (sin->sin_family != AF_INET)
1508 		return -EINVAL;
1509 
1510 	flags = cmd.tcpm_flags & TCP_MD5SIG_FLAG_IFINDEX;
1511 	l3flag = cmd.tcpm_flags & TCP_MD5SIG_FLAG_IFINDEX;
1512 
1513 	if (optname == TCP_MD5SIG_EXT &&
1514 	    cmd.tcpm_flags & TCP_MD5SIG_FLAG_PREFIX) {
1515 		prefixlen = cmd.tcpm_prefixlen;
1516 		if (prefixlen > 32)
1517 			return -EINVAL;
1518 	}
1519 
1520 	if (optname == TCP_MD5SIG_EXT && cmd.tcpm_ifindex &&
1521 	    cmd.tcpm_flags & TCP_MD5SIG_FLAG_IFINDEX) {
1522 		struct net_device *dev;
1523 
1524 		rcu_read_lock();
1525 		dev = dev_get_by_index_rcu(sock_net(sk), cmd.tcpm_ifindex);
1526 		if (dev && netif_is_l3_master(dev))
1527 			l3index = dev->ifindex;
1528 
1529 		rcu_read_unlock();
1530 
1531 		/* ok to reference set/not set outside of rcu;
1532 		 * right now device MUST be an L3 master
1533 		 */
1534 		if (!dev || !l3index)
1535 			return -EINVAL;
1536 	}
1537 
1538 	addr = (union tcp_md5_addr *)&sin->sin_addr.s_addr;
1539 
1540 	if (!cmd.tcpm_keylen)
1541 		return tcp_md5_do_del(sk, addr, AF_INET, prefixlen, l3index, flags);
1542 
1543 	if (cmd.tcpm_keylen > TCP_MD5SIG_MAXKEYLEN)
1544 		return -EINVAL;
1545 
1546 	/* Don't allow keys for peers that have a matching TCP-AO key.
1547 	 * See the comment in tcp_ao_add_cmd()
1548 	 */
1549 	if (tcp_ao_required(sk, addr, AF_INET, l3flag ? l3index : -1, false))
1550 		return -EKEYREJECTED;
1551 
1552 	return tcp_md5_do_add(sk, addr, AF_INET, prefixlen, l3index, flags,
1553 			      cmd.tcpm_key, cmd.tcpm_keylen);
1554 }
1555 
1556 static int tcp_v4_md5_hash_headers(struct tcp_sigpool *hp,
1557 				   __be32 daddr, __be32 saddr,
1558 				   const struct tcphdr *th, int nbytes)
1559 {
1560 	struct tcp4_pseudohdr *bp;
1561 	struct scatterlist sg;
1562 	struct tcphdr *_th;
1563 
1564 	bp = hp->scratch;
1565 	bp->saddr = saddr;
1566 	bp->daddr = daddr;
1567 	bp->pad = 0;
1568 	bp->protocol = IPPROTO_TCP;
1569 	bp->len = cpu_to_be16(nbytes);
1570 
1571 	_th = (struct tcphdr *)(bp + 1);
1572 	memcpy(_th, th, sizeof(*th));
1573 	_th->check = 0;
1574 
1575 	sg_init_one(&sg, bp, sizeof(*bp) + sizeof(*th));
1576 	ahash_request_set_crypt(hp->req, &sg, NULL,
1577 				sizeof(*bp) + sizeof(*th));
1578 	return crypto_ahash_update(hp->req);
1579 }
1580 
1581 static int tcp_v4_md5_hash_hdr(char *md5_hash, const struct tcp_md5sig_key *key,
1582 			       __be32 daddr, __be32 saddr, const struct tcphdr *th)
1583 {
1584 	struct tcp_sigpool hp;
1585 
1586 	if (tcp_sigpool_start(tcp_md5_sigpool_id, &hp))
1587 		goto clear_hash_nostart;
1588 
1589 	if (crypto_ahash_init(hp.req))
1590 		goto clear_hash;
1591 	if (tcp_v4_md5_hash_headers(&hp, daddr, saddr, th, th->doff << 2))
1592 		goto clear_hash;
1593 	if (tcp_md5_hash_key(&hp, key))
1594 		goto clear_hash;
1595 	ahash_request_set_crypt(hp.req, NULL, md5_hash, 0);
1596 	if (crypto_ahash_final(hp.req))
1597 		goto clear_hash;
1598 
1599 	tcp_sigpool_end(&hp);
1600 	return 0;
1601 
1602 clear_hash:
1603 	tcp_sigpool_end(&hp);
1604 clear_hash_nostart:
1605 	memset(md5_hash, 0, 16);
1606 	return 1;
1607 }
1608 
1609 int tcp_v4_md5_hash_skb(char *md5_hash, const struct tcp_md5sig_key *key,
1610 			const struct sock *sk,
1611 			const struct sk_buff *skb)
1612 {
1613 	const struct tcphdr *th = tcp_hdr(skb);
1614 	struct tcp_sigpool hp;
1615 	__be32 saddr, daddr;
1616 
1617 	if (sk) { /* valid for establish/request sockets */
1618 		saddr = sk->sk_rcv_saddr;
1619 		daddr = sk->sk_daddr;
1620 	} else {
1621 		const struct iphdr *iph = ip_hdr(skb);
1622 		saddr = iph->saddr;
1623 		daddr = iph->daddr;
1624 	}
1625 
1626 	if (tcp_sigpool_start(tcp_md5_sigpool_id, &hp))
1627 		goto clear_hash_nostart;
1628 
1629 	if (crypto_ahash_init(hp.req))
1630 		goto clear_hash;
1631 
1632 	if (tcp_v4_md5_hash_headers(&hp, daddr, saddr, th, skb->len))
1633 		goto clear_hash;
1634 	if (tcp_sigpool_hash_skb_data(&hp, skb, th->doff << 2))
1635 		goto clear_hash;
1636 	if (tcp_md5_hash_key(&hp, key))
1637 		goto clear_hash;
1638 	ahash_request_set_crypt(hp.req, NULL, md5_hash, 0);
1639 	if (crypto_ahash_final(hp.req))
1640 		goto clear_hash;
1641 
1642 	tcp_sigpool_end(&hp);
1643 	return 0;
1644 
1645 clear_hash:
1646 	tcp_sigpool_end(&hp);
1647 clear_hash_nostart:
1648 	memset(md5_hash, 0, 16);
1649 	return 1;
1650 }
1651 EXPORT_SYMBOL(tcp_v4_md5_hash_skb);
1652 
1653 #endif
1654 
1655 static void tcp_v4_init_req(struct request_sock *req,
1656 			    const struct sock *sk_listener,
1657 			    struct sk_buff *skb)
1658 {
1659 	struct inet_request_sock *ireq = inet_rsk(req);
1660 	struct net *net = sock_net(sk_listener);
1661 
1662 	sk_rcv_saddr_set(req_to_sk(req), ip_hdr(skb)->daddr);
1663 	sk_daddr_set(req_to_sk(req), ip_hdr(skb)->saddr);
1664 	RCU_INIT_POINTER(ireq->ireq_opt, tcp_v4_save_options(net, skb));
1665 }
1666 
1667 static struct dst_entry *tcp_v4_route_req(const struct sock *sk,
1668 					  struct sk_buff *skb,
1669 					  struct flowi *fl,
1670 					  struct request_sock *req)
1671 {
1672 	tcp_v4_init_req(req, sk, skb);
1673 
1674 	if (security_inet_conn_request(sk, skb, req))
1675 		return NULL;
1676 
1677 	return inet_csk_route_req(sk, &fl->u.ip4, req);
1678 }
1679 
1680 struct request_sock_ops tcp_request_sock_ops __read_mostly = {
1681 	.family		=	PF_INET,
1682 	.obj_size	=	sizeof(struct tcp_request_sock),
1683 	.rtx_syn_ack	=	tcp_rtx_synack,
1684 	.send_ack	=	tcp_v4_reqsk_send_ack,
1685 	.destructor	=	tcp_v4_reqsk_destructor,
1686 	.send_reset	=	tcp_v4_send_reset,
1687 	.syn_ack_timeout =	tcp_syn_ack_timeout,
1688 };
1689 
1690 const struct tcp_request_sock_ops tcp_request_sock_ipv4_ops = {
1691 	.mss_clamp	=	TCP_MSS_DEFAULT,
1692 #ifdef CONFIG_TCP_MD5SIG
1693 	.req_md5_lookup	=	tcp_v4_md5_lookup,
1694 	.calc_md5_hash	=	tcp_v4_md5_hash_skb,
1695 #endif
1696 #ifdef CONFIG_TCP_AO
1697 	.ao_lookup	=	tcp_v4_ao_lookup_rsk,
1698 	.ao_calc_key	=	tcp_v4_ao_calc_key_rsk,
1699 	.ao_synack_hash	=	tcp_v4_ao_synack_hash,
1700 #endif
1701 #ifdef CONFIG_SYN_COOKIES
1702 	.cookie_init_seq =	cookie_v4_init_sequence,
1703 #endif
1704 	.route_req	=	tcp_v4_route_req,
1705 	.init_seq	=	tcp_v4_init_seq,
1706 	.init_ts_off	=	tcp_v4_init_ts_off,
1707 	.send_synack	=	tcp_v4_send_synack,
1708 };
1709 
1710 int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb)
1711 {
1712 	/* Never answer to SYNs send to broadcast or multicast */
1713 	if (skb_rtable(skb)->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))
1714 		goto drop;
1715 
1716 	return tcp_conn_request(&tcp_request_sock_ops,
1717 				&tcp_request_sock_ipv4_ops, sk, skb);
1718 
1719 drop:
1720 	tcp_listendrop(sk);
1721 	return 0;
1722 }
1723 EXPORT_SYMBOL(tcp_v4_conn_request);
1724 
1725 
1726 /*
1727  * The three way handshake has completed - we got a valid synack -
1728  * now create the new socket.
1729  */
1730 struct sock *tcp_v4_syn_recv_sock(const struct sock *sk, struct sk_buff *skb,
1731 				  struct request_sock *req,
1732 				  struct dst_entry *dst,
1733 				  struct request_sock *req_unhash,
1734 				  bool *own_req)
1735 {
1736 	struct inet_request_sock *ireq;
1737 	bool found_dup_sk = false;
1738 	struct inet_sock *newinet;
1739 	struct tcp_sock *newtp;
1740 	struct sock *newsk;
1741 #ifdef CONFIG_TCP_MD5SIG
1742 	const union tcp_md5_addr *addr;
1743 	struct tcp_md5sig_key *key;
1744 	int l3index;
1745 #endif
1746 	struct ip_options_rcu *inet_opt;
1747 
1748 	if (sk_acceptq_is_full(sk))
1749 		goto exit_overflow;
1750 
1751 	newsk = tcp_create_openreq_child(sk, req, skb);
1752 	if (!newsk)
1753 		goto exit_nonewsk;
1754 
1755 	newsk->sk_gso_type = SKB_GSO_TCPV4;
1756 	inet_sk_rx_dst_set(newsk, skb);
1757 
1758 	newtp		      = tcp_sk(newsk);
1759 	newinet		      = inet_sk(newsk);
1760 	ireq		      = inet_rsk(req);
1761 	sk_daddr_set(newsk, ireq->ir_rmt_addr);
1762 	sk_rcv_saddr_set(newsk, ireq->ir_loc_addr);
1763 	newsk->sk_bound_dev_if = ireq->ir_iif;
1764 	newinet->inet_saddr   = ireq->ir_loc_addr;
1765 	inet_opt	      = rcu_dereference(ireq->ireq_opt);
1766 	RCU_INIT_POINTER(newinet->inet_opt, inet_opt);
1767 	newinet->mc_index     = inet_iif(skb);
1768 	newinet->mc_ttl	      = ip_hdr(skb)->ttl;
1769 	newinet->rcv_tos      = ip_hdr(skb)->tos;
1770 	inet_csk(newsk)->icsk_ext_hdr_len = 0;
1771 	if (inet_opt)
1772 		inet_csk(newsk)->icsk_ext_hdr_len = inet_opt->opt.optlen;
1773 	atomic_set(&newinet->inet_id, get_random_u16());
1774 
1775 	/* Set ToS of the new socket based upon the value of incoming SYN.
1776 	 * ECT bits are set later in tcp_init_transfer().
1777 	 */
1778 	if (READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_reflect_tos))
1779 		newinet->tos = tcp_rsk(req)->syn_tos & ~INET_ECN_MASK;
1780 
1781 	if (!dst) {
1782 		dst = inet_csk_route_child_sock(sk, newsk, req);
1783 		if (!dst)
1784 			goto put_and_exit;
1785 	} else {
1786 		/* syncookie case : see end of cookie_v4_check() */
1787 	}
1788 	sk_setup_caps(newsk, dst);
1789 
1790 	tcp_ca_openreq_child(newsk, dst);
1791 
1792 	tcp_sync_mss(newsk, dst_mtu(dst));
1793 	newtp->advmss = tcp_mss_clamp(tcp_sk(sk), dst_metric_advmss(dst));
1794 
1795 	tcp_initialize_rcv_mss(newsk);
1796 
1797 #ifdef CONFIG_TCP_MD5SIG
1798 	l3index = l3mdev_master_ifindex_by_index(sock_net(sk), ireq->ir_iif);
1799 	/* Copy over the MD5 key from the original socket */
1800 	addr = (union tcp_md5_addr *)&newinet->inet_daddr;
1801 	key = tcp_md5_do_lookup(sk, l3index, addr, AF_INET);
1802 	if (key && !tcp_rsk_used_ao(req)) {
1803 		if (tcp_md5_key_copy(newsk, addr, AF_INET, 32, l3index, key))
1804 			goto put_and_exit;
1805 		sk_gso_disable(newsk);
1806 	}
1807 #endif
1808 #ifdef CONFIG_TCP_AO
1809 	if (tcp_ao_copy_all_matching(sk, newsk, req, skb, AF_INET))
1810 		goto put_and_exit; /* OOM, release back memory */
1811 #endif
1812 
1813 	if (__inet_inherit_port(sk, newsk) < 0)
1814 		goto put_and_exit;
1815 	*own_req = inet_ehash_nolisten(newsk, req_to_sk(req_unhash),
1816 				       &found_dup_sk);
1817 	if (likely(*own_req)) {
1818 		tcp_move_syn(newtp, req);
1819 		ireq->ireq_opt = NULL;
1820 	} else {
1821 		newinet->inet_opt = NULL;
1822 
1823 		if (!req_unhash && found_dup_sk) {
1824 			/* This code path should only be executed in the
1825 			 * syncookie case only
1826 			 */
1827 			bh_unlock_sock(newsk);
1828 			sock_put(newsk);
1829 			newsk = NULL;
1830 		}
1831 	}
1832 	return newsk;
1833 
1834 exit_overflow:
1835 	NET_INC_STATS(sock_net(sk), LINUX_MIB_LISTENOVERFLOWS);
1836 exit_nonewsk:
1837 	dst_release(dst);
1838 exit:
1839 	tcp_listendrop(sk);
1840 	return NULL;
1841 put_and_exit:
1842 	newinet->inet_opt = NULL;
1843 	inet_csk_prepare_forced_close(newsk);
1844 	tcp_done(newsk);
1845 	goto exit;
1846 }
1847 EXPORT_SYMBOL(tcp_v4_syn_recv_sock);
1848 
1849 static struct sock *tcp_v4_cookie_check(struct sock *sk, struct sk_buff *skb)
1850 {
1851 #ifdef CONFIG_SYN_COOKIES
1852 	const struct tcphdr *th = tcp_hdr(skb);
1853 
1854 	if (!th->syn)
1855 		sk = cookie_v4_check(sk, skb);
1856 #endif
1857 	return sk;
1858 }
1859 
1860 u16 tcp_v4_get_syncookie(struct sock *sk, struct iphdr *iph,
1861 			 struct tcphdr *th, u32 *cookie)
1862 {
1863 	u16 mss = 0;
1864 #ifdef CONFIG_SYN_COOKIES
1865 	mss = tcp_get_syncookie_mss(&tcp_request_sock_ops,
1866 				    &tcp_request_sock_ipv4_ops, sk, th);
1867 	if (mss) {
1868 		*cookie = __cookie_v4_init_sequence(iph, th, &mss);
1869 		tcp_synq_overflow(sk);
1870 	}
1871 #endif
1872 	return mss;
1873 }
1874 
1875 INDIRECT_CALLABLE_DECLARE(struct dst_entry *ipv4_dst_check(struct dst_entry *,
1876 							   u32));
1877 /* The socket must have it's spinlock held when we get
1878  * here, unless it is a TCP_LISTEN socket.
1879  *
1880  * We have a potential double-lock case here, so even when
1881  * doing backlog processing we use the BH locking scheme.
1882  * This is because we cannot sleep with the original spinlock
1883  * held.
1884  */
1885 int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb)
1886 {
1887 	enum skb_drop_reason reason;
1888 	struct sock *rsk;
1889 
1890 	if (sk->sk_state == TCP_ESTABLISHED) { /* Fast path */
1891 		struct dst_entry *dst;
1892 
1893 		dst = rcu_dereference_protected(sk->sk_rx_dst,
1894 						lockdep_sock_is_held(sk));
1895 
1896 		sock_rps_save_rxhash(sk, skb);
1897 		sk_mark_napi_id(sk, skb);
1898 		if (dst) {
1899 			if (sk->sk_rx_dst_ifindex != skb->skb_iif ||
1900 			    !INDIRECT_CALL_1(dst->ops->check, ipv4_dst_check,
1901 					     dst, 0)) {
1902 				RCU_INIT_POINTER(sk->sk_rx_dst, NULL);
1903 				dst_release(dst);
1904 			}
1905 		}
1906 		tcp_rcv_established(sk, skb);
1907 		return 0;
1908 	}
1909 
1910 	if (tcp_checksum_complete(skb))
1911 		goto csum_err;
1912 
1913 	if (sk->sk_state == TCP_LISTEN) {
1914 		struct sock *nsk = tcp_v4_cookie_check(sk, skb);
1915 
1916 		if (!nsk)
1917 			return 0;
1918 		if (nsk != sk) {
1919 			reason = tcp_child_process(sk, nsk, skb);
1920 			if (reason) {
1921 				rsk = nsk;
1922 				goto reset;
1923 			}
1924 			return 0;
1925 		}
1926 	} else
1927 		sock_rps_save_rxhash(sk, skb);
1928 
1929 	reason = tcp_rcv_state_process(sk, skb);
1930 	if (reason) {
1931 		rsk = sk;
1932 		goto reset;
1933 	}
1934 	return 0;
1935 
1936 reset:
1937 	tcp_v4_send_reset(rsk, skb);
1938 discard:
1939 	kfree_skb_reason(skb, reason);
1940 	/* Be careful here. If this function gets more complicated and
1941 	 * gcc suffers from register pressure on the x86, sk (in %ebx)
1942 	 * might be destroyed here. This current version compiles correctly,
1943 	 * but you have been warned.
1944 	 */
1945 	return 0;
1946 
1947 csum_err:
1948 	reason = SKB_DROP_REASON_TCP_CSUM;
1949 	trace_tcp_bad_csum(skb);
1950 	TCP_INC_STATS(sock_net(sk), TCP_MIB_CSUMERRORS);
1951 	TCP_INC_STATS(sock_net(sk), TCP_MIB_INERRS);
1952 	goto discard;
1953 }
1954 EXPORT_SYMBOL(tcp_v4_do_rcv);
1955 
1956 int tcp_v4_early_demux(struct sk_buff *skb)
1957 {
1958 	struct net *net = dev_net(skb->dev);
1959 	const struct iphdr *iph;
1960 	const struct tcphdr *th;
1961 	struct sock *sk;
1962 
1963 	if (skb->pkt_type != PACKET_HOST)
1964 		return 0;
1965 
1966 	if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct tcphdr)))
1967 		return 0;
1968 
1969 	iph = ip_hdr(skb);
1970 	th = tcp_hdr(skb);
1971 
1972 	if (th->doff < sizeof(struct tcphdr) / 4)
1973 		return 0;
1974 
1975 	sk = __inet_lookup_established(net, net->ipv4.tcp_death_row.hashinfo,
1976 				       iph->saddr, th->source,
1977 				       iph->daddr, ntohs(th->dest),
1978 				       skb->skb_iif, inet_sdif(skb));
1979 	if (sk) {
1980 		skb->sk = sk;
1981 		skb->destructor = sock_edemux;
1982 		if (sk_fullsock(sk)) {
1983 			struct dst_entry *dst = rcu_dereference(sk->sk_rx_dst);
1984 
1985 			if (dst)
1986 				dst = dst_check(dst, 0);
1987 			if (dst &&
1988 			    sk->sk_rx_dst_ifindex == skb->skb_iif)
1989 				skb_dst_set_noref(skb, dst);
1990 		}
1991 	}
1992 	return 0;
1993 }
1994 
1995 bool tcp_add_backlog(struct sock *sk, struct sk_buff *skb,
1996 		     enum skb_drop_reason *reason)
1997 {
1998 	u32 limit, tail_gso_size, tail_gso_segs;
1999 	struct skb_shared_info *shinfo;
2000 	const struct tcphdr *th;
2001 	struct tcphdr *thtail;
2002 	struct sk_buff *tail;
2003 	unsigned int hdrlen;
2004 	bool fragstolen;
2005 	u32 gso_segs;
2006 	u32 gso_size;
2007 	int delta;
2008 
2009 	/* In case all data was pulled from skb frags (in __pskb_pull_tail()),
2010 	 * we can fix skb->truesize to its real value to avoid future drops.
2011 	 * This is valid because skb is not yet charged to the socket.
2012 	 * It has been noticed pure SACK packets were sometimes dropped
2013 	 * (if cooked by drivers without copybreak feature).
2014 	 */
2015 	skb_condense(skb);
2016 
2017 	skb_dst_drop(skb);
2018 
2019 	if (unlikely(tcp_checksum_complete(skb))) {
2020 		bh_unlock_sock(sk);
2021 		trace_tcp_bad_csum(skb);
2022 		*reason = SKB_DROP_REASON_TCP_CSUM;
2023 		__TCP_INC_STATS(sock_net(sk), TCP_MIB_CSUMERRORS);
2024 		__TCP_INC_STATS(sock_net(sk), TCP_MIB_INERRS);
2025 		return true;
2026 	}
2027 
2028 	/* Attempt coalescing to last skb in backlog, even if we are
2029 	 * above the limits.
2030 	 * This is okay because skb capacity is limited to MAX_SKB_FRAGS.
2031 	 */
2032 	th = (const struct tcphdr *)skb->data;
2033 	hdrlen = th->doff * 4;
2034 
2035 	tail = sk->sk_backlog.tail;
2036 	if (!tail)
2037 		goto no_coalesce;
2038 	thtail = (struct tcphdr *)tail->data;
2039 
2040 	if (TCP_SKB_CB(tail)->end_seq != TCP_SKB_CB(skb)->seq ||
2041 	    TCP_SKB_CB(tail)->ip_dsfield != TCP_SKB_CB(skb)->ip_dsfield ||
2042 	    ((TCP_SKB_CB(tail)->tcp_flags |
2043 	      TCP_SKB_CB(skb)->tcp_flags) & (TCPHDR_SYN | TCPHDR_RST | TCPHDR_URG)) ||
2044 	    !((TCP_SKB_CB(tail)->tcp_flags &
2045 	      TCP_SKB_CB(skb)->tcp_flags) & TCPHDR_ACK) ||
2046 	    ((TCP_SKB_CB(tail)->tcp_flags ^
2047 	      TCP_SKB_CB(skb)->tcp_flags) & (TCPHDR_ECE | TCPHDR_CWR)) ||
2048 #ifdef CONFIG_TLS_DEVICE
2049 	    tail->decrypted != skb->decrypted ||
2050 #endif
2051 	    !mptcp_skb_can_collapse(tail, skb) ||
2052 	    thtail->doff != th->doff ||
2053 	    memcmp(thtail + 1, th + 1, hdrlen - sizeof(*th)))
2054 		goto no_coalesce;
2055 
2056 	__skb_pull(skb, hdrlen);
2057 
2058 	shinfo = skb_shinfo(skb);
2059 	gso_size = shinfo->gso_size ?: skb->len;
2060 	gso_segs = shinfo->gso_segs ?: 1;
2061 
2062 	shinfo = skb_shinfo(tail);
2063 	tail_gso_size = shinfo->gso_size ?: (tail->len - hdrlen);
2064 	tail_gso_segs = shinfo->gso_segs ?: 1;
2065 
2066 	if (skb_try_coalesce(tail, skb, &fragstolen, &delta)) {
2067 		TCP_SKB_CB(tail)->end_seq = TCP_SKB_CB(skb)->end_seq;
2068 
2069 		if (likely(!before(TCP_SKB_CB(skb)->ack_seq, TCP_SKB_CB(tail)->ack_seq))) {
2070 			TCP_SKB_CB(tail)->ack_seq = TCP_SKB_CB(skb)->ack_seq;
2071 			thtail->window = th->window;
2072 		}
2073 
2074 		/* We have to update both TCP_SKB_CB(tail)->tcp_flags and
2075 		 * thtail->fin, so that the fast path in tcp_rcv_established()
2076 		 * is not entered if we append a packet with a FIN.
2077 		 * SYN, RST, URG are not present.
2078 		 * ACK is set on both packets.
2079 		 * PSH : we do not really care in TCP stack,
2080 		 *       at least for 'GRO' packets.
2081 		 */
2082 		thtail->fin |= th->fin;
2083 		TCP_SKB_CB(tail)->tcp_flags |= TCP_SKB_CB(skb)->tcp_flags;
2084 
2085 		if (TCP_SKB_CB(skb)->has_rxtstamp) {
2086 			TCP_SKB_CB(tail)->has_rxtstamp = true;
2087 			tail->tstamp = skb->tstamp;
2088 			skb_hwtstamps(tail)->hwtstamp = skb_hwtstamps(skb)->hwtstamp;
2089 		}
2090 
2091 		/* Not as strict as GRO. We only need to carry mss max value */
2092 		shinfo->gso_size = max(gso_size, tail_gso_size);
2093 		shinfo->gso_segs = min_t(u32, gso_segs + tail_gso_segs, 0xFFFF);
2094 
2095 		sk->sk_backlog.len += delta;
2096 		__NET_INC_STATS(sock_net(sk),
2097 				LINUX_MIB_TCPBACKLOGCOALESCE);
2098 		kfree_skb_partial(skb, fragstolen);
2099 		return false;
2100 	}
2101 	__skb_push(skb, hdrlen);
2102 
2103 no_coalesce:
2104 	limit = (u32)READ_ONCE(sk->sk_rcvbuf) + (u32)(READ_ONCE(sk->sk_sndbuf) >> 1);
2105 
2106 	/* Only socket owner can try to collapse/prune rx queues
2107 	 * to reduce memory overhead, so add a little headroom here.
2108 	 * Few sockets backlog are possibly concurrently non empty.
2109 	 */
2110 	limit += 64 * 1024;
2111 
2112 	if (unlikely(sk_add_backlog(sk, skb, limit))) {
2113 		bh_unlock_sock(sk);
2114 		*reason = SKB_DROP_REASON_SOCKET_BACKLOG;
2115 		__NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPBACKLOGDROP);
2116 		return true;
2117 	}
2118 	return false;
2119 }
2120 EXPORT_SYMBOL(tcp_add_backlog);
2121 
2122 int tcp_filter(struct sock *sk, struct sk_buff *skb)
2123 {
2124 	struct tcphdr *th = (struct tcphdr *)skb->data;
2125 
2126 	return sk_filter_trim_cap(sk, skb, th->doff * 4);
2127 }
2128 EXPORT_SYMBOL(tcp_filter);
2129 
2130 static void tcp_v4_restore_cb(struct sk_buff *skb)
2131 {
2132 	memmove(IPCB(skb), &TCP_SKB_CB(skb)->header.h4,
2133 		sizeof(struct inet_skb_parm));
2134 }
2135 
2136 static void tcp_v4_fill_cb(struct sk_buff *skb, const struct iphdr *iph,
2137 			   const struct tcphdr *th)
2138 {
2139 	/* This is tricky : We move IPCB at its correct location into TCP_SKB_CB()
2140 	 * barrier() makes sure compiler wont play fool^Waliasing games.
2141 	 */
2142 	memmove(&TCP_SKB_CB(skb)->header.h4, IPCB(skb),
2143 		sizeof(struct inet_skb_parm));
2144 	barrier();
2145 
2146 	TCP_SKB_CB(skb)->seq = ntohl(th->seq);
2147 	TCP_SKB_CB(skb)->end_seq = (TCP_SKB_CB(skb)->seq + th->syn + th->fin +
2148 				    skb->len - th->doff * 4);
2149 	TCP_SKB_CB(skb)->ack_seq = ntohl(th->ack_seq);
2150 	TCP_SKB_CB(skb)->tcp_flags = tcp_flag_byte(th);
2151 	TCP_SKB_CB(skb)->tcp_tw_isn = 0;
2152 	TCP_SKB_CB(skb)->ip_dsfield = ipv4_get_dsfield(iph);
2153 	TCP_SKB_CB(skb)->sacked	 = 0;
2154 	TCP_SKB_CB(skb)->has_rxtstamp =
2155 			skb->tstamp || skb_hwtstamps(skb)->hwtstamp;
2156 }
2157 
2158 /*
2159  *	From tcp_input.c
2160  */
2161 
2162 int tcp_v4_rcv(struct sk_buff *skb)
2163 {
2164 	struct net *net = dev_net(skb->dev);
2165 	enum skb_drop_reason drop_reason;
2166 	int sdif = inet_sdif(skb);
2167 	int dif = inet_iif(skb);
2168 	const struct iphdr *iph;
2169 	const struct tcphdr *th;
2170 	bool refcounted;
2171 	struct sock *sk;
2172 	int ret;
2173 
2174 	drop_reason = SKB_DROP_REASON_NOT_SPECIFIED;
2175 	if (skb->pkt_type != PACKET_HOST)
2176 		goto discard_it;
2177 
2178 	/* Count it even if it's bad */
2179 	__TCP_INC_STATS(net, TCP_MIB_INSEGS);
2180 
2181 	if (!pskb_may_pull(skb, sizeof(struct tcphdr)))
2182 		goto discard_it;
2183 
2184 	th = (const struct tcphdr *)skb->data;
2185 
2186 	if (unlikely(th->doff < sizeof(struct tcphdr) / 4)) {
2187 		drop_reason = SKB_DROP_REASON_PKT_TOO_SMALL;
2188 		goto bad_packet;
2189 	}
2190 	if (!pskb_may_pull(skb, th->doff * 4))
2191 		goto discard_it;
2192 
2193 	/* An explanation is required here, I think.
2194 	 * Packet length and doff are validated by header prediction,
2195 	 * provided case of th->doff==0 is eliminated.
2196 	 * So, we defer the checks. */
2197 
2198 	if (skb_checksum_init(skb, IPPROTO_TCP, inet_compute_pseudo))
2199 		goto csum_error;
2200 
2201 	th = (const struct tcphdr *)skb->data;
2202 	iph = ip_hdr(skb);
2203 lookup:
2204 	sk = __inet_lookup_skb(net->ipv4.tcp_death_row.hashinfo,
2205 			       skb, __tcp_hdrlen(th), th->source,
2206 			       th->dest, sdif, &refcounted);
2207 	if (!sk)
2208 		goto no_tcp_socket;
2209 
2210 process:
2211 	if (sk->sk_state == TCP_TIME_WAIT)
2212 		goto do_time_wait;
2213 
2214 	if (sk->sk_state == TCP_NEW_SYN_RECV) {
2215 		struct request_sock *req = inet_reqsk(sk);
2216 		bool req_stolen = false;
2217 		struct sock *nsk;
2218 
2219 		sk = req->rsk_listener;
2220 		if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
2221 			drop_reason = SKB_DROP_REASON_XFRM_POLICY;
2222 		else
2223 			drop_reason = tcp_inbound_hash(sk, req, skb,
2224 						       &iph->saddr, &iph->daddr,
2225 						       AF_INET, dif, sdif);
2226 		if (unlikely(drop_reason)) {
2227 			sk_drops_add(sk, skb);
2228 			reqsk_put(req);
2229 			goto discard_it;
2230 		}
2231 		if (tcp_checksum_complete(skb)) {
2232 			reqsk_put(req);
2233 			goto csum_error;
2234 		}
2235 		if (unlikely(sk->sk_state != TCP_LISTEN)) {
2236 			nsk = reuseport_migrate_sock(sk, req_to_sk(req), skb);
2237 			if (!nsk) {
2238 				inet_csk_reqsk_queue_drop_and_put(sk, req);
2239 				goto lookup;
2240 			}
2241 			sk = nsk;
2242 			/* reuseport_migrate_sock() has already held one sk_refcnt
2243 			 * before returning.
2244 			 */
2245 		} else {
2246 			/* We own a reference on the listener, increase it again
2247 			 * as we might lose it too soon.
2248 			 */
2249 			sock_hold(sk);
2250 		}
2251 		refcounted = true;
2252 		nsk = NULL;
2253 		if (!tcp_filter(sk, skb)) {
2254 			th = (const struct tcphdr *)skb->data;
2255 			iph = ip_hdr(skb);
2256 			tcp_v4_fill_cb(skb, iph, th);
2257 			nsk = tcp_check_req(sk, skb, req, false, &req_stolen);
2258 		} else {
2259 			drop_reason = SKB_DROP_REASON_SOCKET_FILTER;
2260 		}
2261 		if (!nsk) {
2262 			reqsk_put(req);
2263 			if (req_stolen) {
2264 				/* Another cpu got exclusive access to req
2265 				 * and created a full blown socket.
2266 				 * Try to feed this packet to this socket
2267 				 * instead of discarding it.
2268 				 */
2269 				tcp_v4_restore_cb(skb);
2270 				sock_put(sk);
2271 				goto lookup;
2272 			}
2273 			goto discard_and_relse;
2274 		}
2275 		nf_reset_ct(skb);
2276 		if (nsk == sk) {
2277 			reqsk_put(req);
2278 			tcp_v4_restore_cb(skb);
2279 		} else {
2280 			drop_reason = tcp_child_process(sk, nsk, skb);
2281 			if (drop_reason) {
2282 				tcp_v4_send_reset(nsk, skb);
2283 				goto discard_and_relse;
2284 			}
2285 			sock_put(sk);
2286 			return 0;
2287 		}
2288 	}
2289 
2290 	if (static_branch_unlikely(&ip4_min_ttl)) {
2291 		/* min_ttl can be changed concurrently from do_ip_setsockopt() */
2292 		if (unlikely(iph->ttl < READ_ONCE(inet_sk(sk)->min_ttl))) {
2293 			__NET_INC_STATS(net, LINUX_MIB_TCPMINTTLDROP);
2294 			drop_reason = SKB_DROP_REASON_TCP_MINTTL;
2295 			goto discard_and_relse;
2296 		}
2297 	}
2298 
2299 	if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb)) {
2300 		drop_reason = SKB_DROP_REASON_XFRM_POLICY;
2301 		goto discard_and_relse;
2302 	}
2303 
2304 	drop_reason = tcp_inbound_hash(sk, NULL, skb, &iph->saddr, &iph->daddr,
2305 				       AF_INET, dif, sdif);
2306 	if (drop_reason)
2307 		goto discard_and_relse;
2308 
2309 	nf_reset_ct(skb);
2310 
2311 	if (tcp_filter(sk, skb)) {
2312 		drop_reason = SKB_DROP_REASON_SOCKET_FILTER;
2313 		goto discard_and_relse;
2314 	}
2315 	th = (const struct tcphdr *)skb->data;
2316 	iph = ip_hdr(skb);
2317 	tcp_v4_fill_cb(skb, iph, th);
2318 
2319 	skb->dev = NULL;
2320 
2321 	if (sk->sk_state == TCP_LISTEN) {
2322 		ret = tcp_v4_do_rcv(sk, skb);
2323 		goto put_and_return;
2324 	}
2325 
2326 	sk_incoming_cpu_update(sk);
2327 
2328 	bh_lock_sock_nested(sk);
2329 	tcp_segs_in(tcp_sk(sk), skb);
2330 	ret = 0;
2331 	if (!sock_owned_by_user(sk)) {
2332 		ret = tcp_v4_do_rcv(sk, skb);
2333 	} else {
2334 		if (tcp_add_backlog(sk, skb, &drop_reason))
2335 			goto discard_and_relse;
2336 	}
2337 	bh_unlock_sock(sk);
2338 
2339 put_and_return:
2340 	if (refcounted)
2341 		sock_put(sk);
2342 
2343 	return ret;
2344 
2345 no_tcp_socket:
2346 	drop_reason = SKB_DROP_REASON_NO_SOCKET;
2347 	if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
2348 		goto discard_it;
2349 
2350 	tcp_v4_fill_cb(skb, iph, th);
2351 
2352 	if (tcp_checksum_complete(skb)) {
2353 csum_error:
2354 		drop_reason = SKB_DROP_REASON_TCP_CSUM;
2355 		trace_tcp_bad_csum(skb);
2356 		__TCP_INC_STATS(net, TCP_MIB_CSUMERRORS);
2357 bad_packet:
2358 		__TCP_INC_STATS(net, TCP_MIB_INERRS);
2359 	} else {
2360 		tcp_v4_send_reset(NULL, skb);
2361 	}
2362 
2363 discard_it:
2364 	SKB_DR_OR(drop_reason, NOT_SPECIFIED);
2365 	/* Discard frame. */
2366 	kfree_skb_reason(skb, drop_reason);
2367 	return 0;
2368 
2369 discard_and_relse:
2370 	sk_drops_add(sk, skb);
2371 	if (refcounted)
2372 		sock_put(sk);
2373 	goto discard_it;
2374 
2375 do_time_wait:
2376 	if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) {
2377 		drop_reason = SKB_DROP_REASON_XFRM_POLICY;
2378 		inet_twsk_put(inet_twsk(sk));
2379 		goto discard_it;
2380 	}
2381 
2382 	tcp_v4_fill_cb(skb, iph, th);
2383 
2384 	if (tcp_checksum_complete(skb)) {
2385 		inet_twsk_put(inet_twsk(sk));
2386 		goto csum_error;
2387 	}
2388 	switch (tcp_timewait_state_process(inet_twsk(sk), skb, th)) {
2389 	case TCP_TW_SYN: {
2390 		struct sock *sk2 = inet_lookup_listener(net,
2391 							net->ipv4.tcp_death_row.hashinfo,
2392 							skb, __tcp_hdrlen(th),
2393 							iph->saddr, th->source,
2394 							iph->daddr, th->dest,
2395 							inet_iif(skb),
2396 							sdif);
2397 		if (sk2) {
2398 			inet_twsk_deschedule_put(inet_twsk(sk));
2399 			sk = sk2;
2400 			tcp_v4_restore_cb(skb);
2401 			refcounted = false;
2402 			goto process;
2403 		}
2404 	}
2405 		/* to ACK */
2406 		fallthrough;
2407 	case TCP_TW_ACK:
2408 		tcp_v4_timewait_ack(sk, skb);
2409 		break;
2410 	case TCP_TW_RST:
2411 		tcp_v4_send_reset(sk, skb);
2412 		inet_twsk_deschedule_put(inet_twsk(sk));
2413 		goto discard_it;
2414 	case TCP_TW_SUCCESS:;
2415 	}
2416 	goto discard_it;
2417 }
2418 
2419 static struct timewait_sock_ops tcp_timewait_sock_ops = {
2420 	.twsk_obj_size	= sizeof(struct tcp_timewait_sock),
2421 	.twsk_unique	= tcp_twsk_unique,
2422 	.twsk_destructor= tcp_twsk_destructor,
2423 };
2424 
2425 void inet_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb)
2426 {
2427 	struct dst_entry *dst = skb_dst(skb);
2428 
2429 	if (dst && dst_hold_safe(dst)) {
2430 		rcu_assign_pointer(sk->sk_rx_dst, dst);
2431 		sk->sk_rx_dst_ifindex = skb->skb_iif;
2432 	}
2433 }
2434 EXPORT_SYMBOL(inet_sk_rx_dst_set);
2435 
2436 const struct inet_connection_sock_af_ops ipv4_specific = {
2437 	.queue_xmit	   = ip_queue_xmit,
2438 	.send_check	   = tcp_v4_send_check,
2439 	.rebuild_header	   = inet_sk_rebuild_header,
2440 	.sk_rx_dst_set	   = inet_sk_rx_dst_set,
2441 	.conn_request	   = tcp_v4_conn_request,
2442 	.syn_recv_sock	   = tcp_v4_syn_recv_sock,
2443 	.net_header_len	   = sizeof(struct iphdr),
2444 	.setsockopt	   = ip_setsockopt,
2445 	.getsockopt	   = ip_getsockopt,
2446 	.addr2sockaddr	   = inet_csk_addr2sockaddr,
2447 	.sockaddr_len	   = sizeof(struct sockaddr_in),
2448 	.mtu_reduced	   = tcp_v4_mtu_reduced,
2449 };
2450 EXPORT_SYMBOL(ipv4_specific);
2451 
2452 #if defined(CONFIG_TCP_MD5SIG) || defined(CONFIG_TCP_AO)
2453 static const struct tcp_sock_af_ops tcp_sock_ipv4_specific = {
2454 #ifdef CONFIG_TCP_MD5SIG
2455 	.md5_lookup		= tcp_v4_md5_lookup,
2456 	.calc_md5_hash		= tcp_v4_md5_hash_skb,
2457 	.md5_parse		= tcp_v4_parse_md5_keys,
2458 #endif
2459 #ifdef CONFIG_TCP_AO
2460 	.ao_lookup		= tcp_v4_ao_lookup,
2461 	.calc_ao_hash		= tcp_v4_ao_hash_skb,
2462 	.ao_parse		= tcp_v4_parse_ao,
2463 	.ao_calc_key_sk		= tcp_v4_ao_calc_key_sk,
2464 #endif
2465 };
2466 #endif
2467 
2468 /* NOTE: A lot of things set to zero explicitly by call to
2469  *       sk_alloc() so need not be done here.
2470  */
2471 static int tcp_v4_init_sock(struct sock *sk)
2472 {
2473 	struct inet_connection_sock *icsk = inet_csk(sk);
2474 
2475 	tcp_init_sock(sk);
2476 
2477 	icsk->icsk_af_ops = &ipv4_specific;
2478 
2479 #if defined(CONFIG_TCP_MD5SIG) || defined(CONFIG_TCP_AO)
2480 	tcp_sk(sk)->af_specific = &tcp_sock_ipv4_specific;
2481 #endif
2482 
2483 	return 0;
2484 }
2485 
2486 #ifdef CONFIG_TCP_MD5SIG
2487 static void tcp_md5sig_info_free_rcu(struct rcu_head *head)
2488 {
2489 	struct tcp_md5sig_info *md5sig;
2490 
2491 	md5sig = container_of(head, struct tcp_md5sig_info, rcu);
2492 	kfree(md5sig);
2493 	static_branch_slow_dec_deferred(&tcp_md5_needed);
2494 	tcp_md5_release_sigpool();
2495 }
2496 #endif
2497 
2498 void tcp_v4_destroy_sock(struct sock *sk)
2499 {
2500 	struct tcp_sock *tp = tcp_sk(sk);
2501 
2502 	trace_tcp_destroy_sock(sk);
2503 
2504 	tcp_clear_xmit_timers(sk);
2505 
2506 	tcp_cleanup_congestion_control(sk);
2507 
2508 	tcp_cleanup_ulp(sk);
2509 
2510 	/* Cleanup up the write buffer. */
2511 	tcp_write_queue_purge(sk);
2512 
2513 	/* Check if we want to disable active TFO */
2514 	tcp_fastopen_active_disable_ofo_check(sk);
2515 
2516 	/* Cleans up our, hopefully empty, out_of_order_queue. */
2517 	skb_rbtree_purge(&tp->out_of_order_queue);
2518 
2519 #ifdef CONFIG_TCP_MD5SIG
2520 	/* Clean up the MD5 key list, if any */
2521 	if (tp->md5sig_info) {
2522 		struct tcp_md5sig_info *md5sig;
2523 
2524 		md5sig = rcu_dereference_protected(tp->md5sig_info, 1);
2525 		tcp_clear_md5_list(sk);
2526 		call_rcu(&md5sig->rcu, tcp_md5sig_info_free_rcu);
2527 		rcu_assign_pointer(tp->md5sig_info, NULL);
2528 	}
2529 #endif
2530 	tcp_ao_destroy_sock(sk, false);
2531 
2532 	/* Clean up a referenced TCP bind bucket. */
2533 	if (inet_csk(sk)->icsk_bind_hash)
2534 		inet_put_port(sk);
2535 
2536 	BUG_ON(rcu_access_pointer(tp->fastopen_rsk));
2537 
2538 	/* If socket is aborted during connect operation */
2539 	tcp_free_fastopen_req(tp);
2540 	tcp_fastopen_destroy_cipher(sk);
2541 	tcp_saved_syn_free(tp);
2542 
2543 	sk_sockets_allocated_dec(sk);
2544 }
2545 EXPORT_SYMBOL(tcp_v4_destroy_sock);
2546 
2547 #ifdef CONFIG_PROC_FS
2548 /* Proc filesystem TCP sock list dumping. */
2549 
2550 static unsigned short seq_file_family(const struct seq_file *seq);
2551 
2552 static bool seq_sk_match(struct seq_file *seq, const struct sock *sk)
2553 {
2554 	unsigned short family = seq_file_family(seq);
2555 
2556 	/* AF_UNSPEC is used as a match all */
2557 	return ((family == AF_UNSPEC || family == sk->sk_family) &&
2558 		net_eq(sock_net(sk), seq_file_net(seq)));
2559 }
2560 
2561 /* Find a non empty bucket (starting from st->bucket)
2562  * and return the first sk from it.
2563  */
2564 static void *listening_get_first(struct seq_file *seq)
2565 {
2566 	struct inet_hashinfo *hinfo = seq_file_net(seq)->ipv4.tcp_death_row.hashinfo;
2567 	struct tcp_iter_state *st = seq->private;
2568 
2569 	st->offset = 0;
2570 	for (; st->bucket <= hinfo->lhash2_mask; st->bucket++) {
2571 		struct inet_listen_hashbucket *ilb2;
2572 		struct hlist_nulls_node *node;
2573 		struct sock *sk;
2574 
2575 		ilb2 = &hinfo->lhash2[st->bucket];
2576 		if (hlist_nulls_empty(&ilb2->nulls_head))
2577 			continue;
2578 
2579 		spin_lock(&ilb2->lock);
2580 		sk_nulls_for_each(sk, node, &ilb2->nulls_head) {
2581 			if (seq_sk_match(seq, sk))
2582 				return sk;
2583 		}
2584 		spin_unlock(&ilb2->lock);
2585 	}
2586 
2587 	return NULL;
2588 }
2589 
2590 /* Find the next sk of "cur" within the same bucket (i.e. st->bucket).
2591  * If "cur" is the last one in the st->bucket,
2592  * call listening_get_first() to return the first sk of the next
2593  * non empty bucket.
2594  */
2595 static void *listening_get_next(struct seq_file *seq, void *cur)
2596 {
2597 	struct tcp_iter_state *st = seq->private;
2598 	struct inet_listen_hashbucket *ilb2;
2599 	struct hlist_nulls_node *node;
2600 	struct inet_hashinfo *hinfo;
2601 	struct sock *sk = cur;
2602 
2603 	++st->num;
2604 	++st->offset;
2605 
2606 	sk = sk_nulls_next(sk);
2607 	sk_nulls_for_each_from(sk, node) {
2608 		if (seq_sk_match(seq, sk))
2609 			return sk;
2610 	}
2611 
2612 	hinfo = seq_file_net(seq)->ipv4.tcp_death_row.hashinfo;
2613 	ilb2 = &hinfo->lhash2[st->bucket];
2614 	spin_unlock(&ilb2->lock);
2615 	++st->bucket;
2616 	return listening_get_first(seq);
2617 }
2618 
2619 static void *listening_get_idx(struct seq_file *seq, loff_t *pos)
2620 {
2621 	struct tcp_iter_state *st = seq->private;
2622 	void *rc;
2623 
2624 	st->bucket = 0;
2625 	st->offset = 0;
2626 	rc = listening_get_first(seq);
2627 
2628 	while (rc && *pos) {
2629 		rc = listening_get_next(seq, rc);
2630 		--*pos;
2631 	}
2632 	return rc;
2633 }
2634 
2635 static inline bool empty_bucket(struct inet_hashinfo *hinfo,
2636 				const struct tcp_iter_state *st)
2637 {
2638 	return hlist_nulls_empty(&hinfo->ehash[st->bucket].chain);
2639 }
2640 
2641 /*
2642  * Get first established socket starting from bucket given in st->bucket.
2643  * If st->bucket is zero, the very first socket in the hash is returned.
2644  */
2645 static void *established_get_first(struct seq_file *seq)
2646 {
2647 	struct inet_hashinfo *hinfo = seq_file_net(seq)->ipv4.tcp_death_row.hashinfo;
2648 	struct tcp_iter_state *st = seq->private;
2649 
2650 	st->offset = 0;
2651 	for (; st->bucket <= hinfo->ehash_mask; ++st->bucket) {
2652 		struct sock *sk;
2653 		struct hlist_nulls_node *node;
2654 		spinlock_t *lock = inet_ehash_lockp(hinfo, st->bucket);
2655 
2656 		cond_resched();
2657 
2658 		/* Lockless fast path for the common case of empty buckets */
2659 		if (empty_bucket(hinfo, st))
2660 			continue;
2661 
2662 		spin_lock_bh(lock);
2663 		sk_nulls_for_each(sk, node, &hinfo->ehash[st->bucket].chain) {
2664 			if (seq_sk_match(seq, sk))
2665 				return sk;
2666 		}
2667 		spin_unlock_bh(lock);
2668 	}
2669 
2670 	return NULL;
2671 }
2672 
2673 static void *established_get_next(struct seq_file *seq, void *cur)
2674 {
2675 	struct inet_hashinfo *hinfo = seq_file_net(seq)->ipv4.tcp_death_row.hashinfo;
2676 	struct tcp_iter_state *st = seq->private;
2677 	struct hlist_nulls_node *node;
2678 	struct sock *sk = cur;
2679 
2680 	++st->num;
2681 	++st->offset;
2682 
2683 	sk = sk_nulls_next(sk);
2684 
2685 	sk_nulls_for_each_from(sk, node) {
2686 		if (seq_sk_match(seq, sk))
2687 			return sk;
2688 	}
2689 
2690 	spin_unlock_bh(inet_ehash_lockp(hinfo, st->bucket));
2691 	++st->bucket;
2692 	return established_get_first(seq);
2693 }
2694 
2695 static void *established_get_idx(struct seq_file *seq, loff_t pos)
2696 {
2697 	struct tcp_iter_state *st = seq->private;
2698 	void *rc;
2699 
2700 	st->bucket = 0;
2701 	rc = established_get_first(seq);
2702 
2703 	while (rc && pos) {
2704 		rc = established_get_next(seq, rc);
2705 		--pos;
2706 	}
2707 	return rc;
2708 }
2709 
2710 static void *tcp_get_idx(struct seq_file *seq, loff_t pos)
2711 {
2712 	void *rc;
2713 	struct tcp_iter_state *st = seq->private;
2714 
2715 	st->state = TCP_SEQ_STATE_LISTENING;
2716 	rc	  = listening_get_idx(seq, &pos);
2717 
2718 	if (!rc) {
2719 		st->state = TCP_SEQ_STATE_ESTABLISHED;
2720 		rc	  = established_get_idx(seq, pos);
2721 	}
2722 
2723 	return rc;
2724 }
2725 
2726 static void *tcp_seek_last_pos(struct seq_file *seq)
2727 {
2728 	struct inet_hashinfo *hinfo = seq_file_net(seq)->ipv4.tcp_death_row.hashinfo;
2729 	struct tcp_iter_state *st = seq->private;
2730 	int bucket = st->bucket;
2731 	int offset = st->offset;
2732 	int orig_num = st->num;
2733 	void *rc = NULL;
2734 
2735 	switch (st->state) {
2736 	case TCP_SEQ_STATE_LISTENING:
2737 		if (st->bucket > hinfo->lhash2_mask)
2738 			break;
2739 		rc = listening_get_first(seq);
2740 		while (offset-- && rc && bucket == st->bucket)
2741 			rc = listening_get_next(seq, rc);
2742 		if (rc)
2743 			break;
2744 		st->bucket = 0;
2745 		st->state = TCP_SEQ_STATE_ESTABLISHED;
2746 		fallthrough;
2747 	case TCP_SEQ_STATE_ESTABLISHED:
2748 		if (st->bucket > hinfo->ehash_mask)
2749 			break;
2750 		rc = established_get_first(seq);
2751 		while (offset-- && rc && bucket == st->bucket)
2752 			rc = established_get_next(seq, rc);
2753 	}
2754 
2755 	st->num = orig_num;
2756 
2757 	return rc;
2758 }
2759 
2760 void *tcp_seq_start(struct seq_file *seq, loff_t *pos)
2761 {
2762 	struct tcp_iter_state *st = seq->private;
2763 	void *rc;
2764 
2765 	if (*pos && *pos == st->last_pos) {
2766 		rc = tcp_seek_last_pos(seq);
2767 		if (rc)
2768 			goto out;
2769 	}
2770 
2771 	st->state = TCP_SEQ_STATE_LISTENING;
2772 	st->num = 0;
2773 	st->bucket = 0;
2774 	st->offset = 0;
2775 	rc = *pos ? tcp_get_idx(seq, *pos - 1) : SEQ_START_TOKEN;
2776 
2777 out:
2778 	st->last_pos = *pos;
2779 	return rc;
2780 }
2781 EXPORT_SYMBOL(tcp_seq_start);
2782 
2783 void *tcp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2784 {
2785 	struct tcp_iter_state *st = seq->private;
2786 	void *rc = NULL;
2787 
2788 	if (v == SEQ_START_TOKEN) {
2789 		rc = tcp_get_idx(seq, 0);
2790 		goto out;
2791 	}
2792 
2793 	switch (st->state) {
2794 	case TCP_SEQ_STATE_LISTENING:
2795 		rc = listening_get_next(seq, v);
2796 		if (!rc) {
2797 			st->state = TCP_SEQ_STATE_ESTABLISHED;
2798 			st->bucket = 0;
2799 			st->offset = 0;
2800 			rc	  = established_get_first(seq);
2801 		}
2802 		break;
2803 	case TCP_SEQ_STATE_ESTABLISHED:
2804 		rc = established_get_next(seq, v);
2805 		break;
2806 	}
2807 out:
2808 	++*pos;
2809 	st->last_pos = *pos;
2810 	return rc;
2811 }
2812 EXPORT_SYMBOL(tcp_seq_next);
2813 
2814 void tcp_seq_stop(struct seq_file *seq, void *v)
2815 {
2816 	struct inet_hashinfo *hinfo = seq_file_net(seq)->ipv4.tcp_death_row.hashinfo;
2817 	struct tcp_iter_state *st = seq->private;
2818 
2819 	switch (st->state) {
2820 	case TCP_SEQ_STATE_LISTENING:
2821 		if (v != SEQ_START_TOKEN)
2822 			spin_unlock(&hinfo->lhash2[st->bucket].lock);
2823 		break;
2824 	case TCP_SEQ_STATE_ESTABLISHED:
2825 		if (v)
2826 			spin_unlock_bh(inet_ehash_lockp(hinfo, st->bucket));
2827 		break;
2828 	}
2829 }
2830 EXPORT_SYMBOL(tcp_seq_stop);
2831 
2832 static void get_openreq4(const struct request_sock *req,
2833 			 struct seq_file *f, int i)
2834 {
2835 	const struct inet_request_sock *ireq = inet_rsk(req);
2836 	long delta = req->rsk_timer.expires - jiffies;
2837 
2838 	seq_printf(f, "%4d: %08X:%04X %08X:%04X"
2839 		" %02X %08X:%08X %02X:%08lX %08X %5u %8d %u %d %pK",
2840 		i,
2841 		ireq->ir_loc_addr,
2842 		ireq->ir_num,
2843 		ireq->ir_rmt_addr,
2844 		ntohs(ireq->ir_rmt_port),
2845 		TCP_SYN_RECV,
2846 		0, 0, /* could print option size, but that is af dependent. */
2847 		1,    /* timers active (only the expire timer) */
2848 		jiffies_delta_to_clock_t(delta),
2849 		req->num_timeout,
2850 		from_kuid_munged(seq_user_ns(f),
2851 				 sock_i_uid(req->rsk_listener)),
2852 		0,  /* non standard timer */
2853 		0, /* open_requests have no inode */
2854 		0,
2855 		req);
2856 }
2857 
2858 static void get_tcp4_sock(struct sock *sk, struct seq_file *f, int i)
2859 {
2860 	int timer_active;
2861 	unsigned long timer_expires;
2862 	const struct tcp_sock *tp = tcp_sk(sk);
2863 	const struct inet_connection_sock *icsk = inet_csk(sk);
2864 	const struct inet_sock *inet = inet_sk(sk);
2865 	const struct fastopen_queue *fastopenq = &icsk->icsk_accept_queue.fastopenq;
2866 	__be32 dest = inet->inet_daddr;
2867 	__be32 src = inet->inet_rcv_saddr;
2868 	__u16 destp = ntohs(inet->inet_dport);
2869 	__u16 srcp = ntohs(inet->inet_sport);
2870 	int rx_queue;
2871 	int state;
2872 
2873 	if (icsk->icsk_pending == ICSK_TIME_RETRANS ||
2874 	    icsk->icsk_pending == ICSK_TIME_REO_TIMEOUT ||
2875 	    icsk->icsk_pending == ICSK_TIME_LOSS_PROBE) {
2876 		timer_active	= 1;
2877 		timer_expires	= icsk->icsk_timeout;
2878 	} else if (icsk->icsk_pending == ICSK_TIME_PROBE0) {
2879 		timer_active	= 4;
2880 		timer_expires	= icsk->icsk_timeout;
2881 	} else if (timer_pending(&sk->sk_timer)) {
2882 		timer_active	= 2;
2883 		timer_expires	= sk->sk_timer.expires;
2884 	} else {
2885 		timer_active	= 0;
2886 		timer_expires = jiffies;
2887 	}
2888 
2889 	state = inet_sk_state_load(sk);
2890 	if (state == TCP_LISTEN)
2891 		rx_queue = READ_ONCE(sk->sk_ack_backlog);
2892 	else
2893 		/* Because we don't lock the socket,
2894 		 * we might find a transient negative value.
2895 		 */
2896 		rx_queue = max_t(int, READ_ONCE(tp->rcv_nxt) -
2897 				      READ_ONCE(tp->copied_seq), 0);
2898 
2899 	seq_printf(f, "%4d: %08X:%04X %08X:%04X %02X %08X:%08X %02X:%08lX "
2900 			"%08X %5u %8d %lu %d %pK %lu %lu %u %u %d",
2901 		i, src, srcp, dest, destp, state,
2902 		READ_ONCE(tp->write_seq) - tp->snd_una,
2903 		rx_queue,
2904 		timer_active,
2905 		jiffies_delta_to_clock_t(timer_expires - jiffies),
2906 		icsk->icsk_retransmits,
2907 		from_kuid_munged(seq_user_ns(f), sock_i_uid(sk)),
2908 		icsk->icsk_probes_out,
2909 		sock_i_ino(sk),
2910 		refcount_read(&sk->sk_refcnt), sk,
2911 		jiffies_to_clock_t(icsk->icsk_rto),
2912 		jiffies_to_clock_t(icsk->icsk_ack.ato),
2913 		(icsk->icsk_ack.quick << 1) | inet_csk_in_pingpong_mode(sk),
2914 		tcp_snd_cwnd(tp),
2915 		state == TCP_LISTEN ?
2916 		    fastopenq->max_qlen :
2917 		    (tcp_in_initial_slowstart(tp) ? -1 : tp->snd_ssthresh));
2918 }
2919 
2920 static void get_timewait4_sock(const struct inet_timewait_sock *tw,
2921 			       struct seq_file *f, int i)
2922 {
2923 	long delta = tw->tw_timer.expires - jiffies;
2924 	__be32 dest, src;
2925 	__u16 destp, srcp;
2926 
2927 	dest  = tw->tw_daddr;
2928 	src   = tw->tw_rcv_saddr;
2929 	destp = ntohs(tw->tw_dport);
2930 	srcp  = ntohs(tw->tw_sport);
2931 
2932 	seq_printf(f, "%4d: %08X:%04X %08X:%04X"
2933 		" %02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %pK",
2934 		i, src, srcp, dest, destp, tw->tw_substate, 0, 0,
2935 		3, jiffies_delta_to_clock_t(delta), 0, 0, 0, 0,
2936 		refcount_read(&tw->tw_refcnt), tw);
2937 }
2938 
2939 #define TMPSZ 150
2940 
2941 static int tcp4_seq_show(struct seq_file *seq, void *v)
2942 {
2943 	struct tcp_iter_state *st;
2944 	struct sock *sk = v;
2945 
2946 	seq_setwidth(seq, TMPSZ - 1);
2947 	if (v == SEQ_START_TOKEN) {
2948 		seq_puts(seq, "  sl  local_address rem_address   st tx_queue "
2949 			   "rx_queue tr tm->when retrnsmt   uid  timeout "
2950 			   "inode");
2951 		goto out;
2952 	}
2953 	st = seq->private;
2954 
2955 	if (sk->sk_state == TCP_TIME_WAIT)
2956 		get_timewait4_sock(v, seq, st->num);
2957 	else if (sk->sk_state == TCP_NEW_SYN_RECV)
2958 		get_openreq4(v, seq, st->num);
2959 	else
2960 		get_tcp4_sock(v, seq, st->num);
2961 out:
2962 	seq_pad(seq, '\n');
2963 	return 0;
2964 }
2965 
2966 #ifdef CONFIG_BPF_SYSCALL
2967 struct bpf_tcp_iter_state {
2968 	struct tcp_iter_state state;
2969 	unsigned int cur_sk;
2970 	unsigned int end_sk;
2971 	unsigned int max_sk;
2972 	struct sock **batch;
2973 	bool st_bucket_done;
2974 };
2975 
2976 struct bpf_iter__tcp {
2977 	__bpf_md_ptr(struct bpf_iter_meta *, meta);
2978 	__bpf_md_ptr(struct sock_common *, sk_common);
2979 	uid_t uid __aligned(8);
2980 };
2981 
2982 static int tcp_prog_seq_show(struct bpf_prog *prog, struct bpf_iter_meta *meta,
2983 			     struct sock_common *sk_common, uid_t uid)
2984 {
2985 	struct bpf_iter__tcp ctx;
2986 
2987 	meta->seq_num--;  /* skip SEQ_START_TOKEN */
2988 	ctx.meta = meta;
2989 	ctx.sk_common = sk_common;
2990 	ctx.uid = uid;
2991 	return bpf_iter_run_prog(prog, &ctx);
2992 }
2993 
2994 static void bpf_iter_tcp_put_batch(struct bpf_tcp_iter_state *iter)
2995 {
2996 	while (iter->cur_sk < iter->end_sk)
2997 		sock_gen_put(iter->batch[iter->cur_sk++]);
2998 }
2999 
3000 static int bpf_iter_tcp_realloc_batch(struct bpf_tcp_iter_state *iter,
3001 				      unsigned int new_batch_sz)
3002 {
3003 	struct sock **new_batch;
3004 
3005 	new_batch = kvmalloc(sizeof(*new_batch) * new_batch_sz,
3006 			     GFP_USER | __GFP_NOWARN);
3007 	if (!new_batch)
3008 		return -ENOMEM;
3009 
3010 	bpf_iter_tcp_put_batch(iter);
3011 	kvfree(iter->batch);
3012 	iter->batch = new_batch;
3013 	iter->max_sk = new_batch_sz;
3014 
3015 	return 0;
3016 }
3017 
3018 static unsigned int bpf_iter_tcp_listening_batch(struct seq_file *seq,
3019 						 struct sock *start_sk)
3020 {
3021 	struct inet_hashinfo *hinfo = seq_file_net(seq)->ipv4.tcp_death_row.hashinfo;
3022 	struct bpf_tcp_iter_state *iter = seq->private;
3023 	struct tcp_iter_state *st = &iter->state;
3024 	struct hlist_nulls_node *node;
3025 	unsigned int expected = 1;
3026 	struct sock *sk;
3027 
3028 	sock_hold(start_sk);
3029 	iter->batch[iter->end_sk++] = start_sk;
3030 
3031 	sk = sk_nulls_next(start_sk);
3032 	sk_nulls_for_each_from(sk, node) {
3033 		if (seq_sk_match(seq, sk)) {
3034 			if (iter->end_sk < iter->max_sk) {
3035 				sock_hold(sk);
3036 				iter->batch[iter->end_sk++] = sk;
3037 			}
3038 			expected++;
3039 		}
3040 	}
3041 	spin_unlock(&hinfo->lhash2[st->bucket].lock);
3042 
3043 	return expected;
3044 }
3045 
3046 static unsigned int bpf_iter_tcp_established_batch(struct seq_file *seq,
3047 						   struct sock *start_sk)
3048 {
3049 	struct inet_hashinfo *hinfo = seq_file_net(seq)->ipv4.tcp_death_row.hashinfo;
3050 	struct bpf_tcp_iter_state *iter = seq->private;
3051 	struct tcp_iter_state *st = &iter->state;
3052 	struct hlist_nulls_node *node;
3053 	unsigned int expected = 1;
3054 	struct sock *sk;
3055 
3056 	sock_hold(start_sk);
3057 	iter->batch[iter->end_sk++] = start_sk;
3058 
3059 	sk = sk_nulls_next(start_sk);
3060 	sk_nulls_for_each_from(sk, node) {
3061 		if (seq_sk_match(seq, sk)) {
3062 			if (iter->end_sk < iter->max_sk) {
3063 				sock_hold(sk);
3064 				iter->batch[iter->end_sk++] = sk;
3065 			}
3066 			expected++;
3067 		}
3068 	}
3069 	spin_unlock_bh(inet_ehash_lockp(hinfo, st->bucket));
3070 
3071 	return expected;
3072 }
3073 
3074 static struct sock *bpf_iter_tcp_batch(struct seq_file *seq)
3075 {
3076 	struct inet_hashinfo *hinfo = seq_file_net(seq)->ipv4.tcp_death_row.hashinfo;
3077 	struct bpf_tcp_iter_state *iter = seq->private;
3078 	struct tcp_iter_state *st = &iter->state;
3079 	unsigned int expected;
3080 	bool resized = false;
3081 	struct sock *sk;
3082 
3083 	/* The st->bucket is done.  Directly advance to the next
3084 	 * bucket instead of having the tcp_seek_last_pos() to skip
3085 	 * one by one in the current bucket and eventually find out
3086 	 * it has to advance to the next bucket.
3087 	 */
3088 	if (iter->st_bucket_done) {
3089 		st->offset = 0;
3090 		st->bucket++;
3091 		if (st->state == TCP_SEQ_STATE_LISTENING &&
3092 		    st->bucket > hinfo->lhash2_mask) {
3093 			st->state = TCP_SEQ_STATE_ESTABLISHED;
3094 			st->bucket = 0;
3095 		}
3096 	}
3097 
3098 again:
3099 	/* Get a new batch */
3100 	iter->cur_sk = 0;
3101 	iter->end_sk = 0;
3102 	iter->st_bucket_done = false;
3103 
3104 	sk = tcp_seek_last_pos(seq);
3105 	if (!sk)
3106 		return NULL; /* Done */
3107 
3108 	if (st->state == TCP_SEQ_STATE_LISTENING)
3109 		expected = bpf_iter_tcp_listening_batch(seq, sk);
3110 	else
3111 		expected = bpf_iter_tcp_established_batch(seq, sk);
3112 
3113 	if (iter->end_sk == expected) {
3114 		iter->st_bucket_done = true;
3115 		return sk;
3116 	}
3117 
3118 	if (!resized && !bpf_iter_tcp_realloc_batch(iter, expected * 3 / 2)) {
3119 		resized = true;
3120 		goto again;
3121 	}
3122 
3123 	return sk;
3124 }
3125 
3126 static void *bpf_iter_tcp_seq_start(struct seq_file *seq, loff_t *pos)
3127 {
3128 	/* bpf iter does not support lseek, so it always
3129 	 * continue from where it was stop()-ped.
3130 	 */
3131 	if (*pos)
3132 		return bpf_iter_tcp_batch(seq);
3133 
3134 	return SEQ_START_TOKEN;
3135 }
3136 
3137 static void *bpf_iter_tcp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3138 {
3139 	struct bpf_tcp_iter_state *iter = seq->private;
3140 	struct tcp_iter_state *st = &iter->state;
3141 	struct sock *sk;
3142 
3143 	/* Whenever seq_next() is called, the iter->cur_sk is
3144 	 * done with seq_show(), so advance to the next sk in
3145 	 * the batch.
3146 	 */
3147 	if (iter->cur_sk < iter->end_sk) {
3148 		/* Keeping st->num consistent in tcp_iter_state.
3149 		 * bpf_iter_tcp does not use st->num.
3150 		 * meta.seq_num is used instead.
3151 		 */
3152 		st->num++;
3153 		/* Move st->offset to the next sk in the bucket such that
3154 		 * the future start() will resume at st->offset in
3155 		 * st->bucket.  See tcp_seek_last_pos().
3156 		 */
3157 		st->offset++;
3158 		sock_gen_put(iter->batch[iter->cur_sk++]);
3159 	}
3160 
3161 	if (iter->cur_sk < iter->end_sk)
3162 		sk = iter->batch[iter->cur_sk];
3163 	else
3164 		sk = bpf_iter_tcp_batch(seq);
3165 
3166 	++*pos;
3167 	/* Keeping st->last_pos consistent in tcp_iter_state.
3168 	 * bpf iter does not do lseek, so st->last_pos always equals to *pos.
3169 	 */
3170 	st->last_pos = *pos;
3171 	return sk;
3172 }
3173 
3174 static int bpf_iter_tcp_seq_show(struct seq_file *seq, void *v)
3175 {
3176 	struct bpf_iter_meta meta;
3177 	struct bpf_prog *prog;
3178 	struct sock *sk = v;
3179 	uid_t uid;
3180 	int ret;
3181 
3182 	if (v == SEQ_START_TOKEN)
3183 		return 0;
3184 
3185 	if (sk_fullsock(sk))
3186 		lock_sock(sk);
3187 
3188 	if (unlikely(sk_unhashed(sk))) {
3189 		ret = SEQ_SKIP;
3190 		goto unlock;
3191 	}
3192 
3193 	if (sk->sk_state == TCP_TIME_WAIT) {
3194 		uid = 0;
3195 	} else if (sk->sk_state == TCP_NEW_SYN_RECV) {
3196 		const struct request_sock *req = v;
3197 
3198 		uid = from_kuid_munged(seq_user_ns(seq),
3199 				       sock_i_uid(req->rsk_listener));
3200 	} else {
3201 		uid = from_kuid_munged(seq_user_ns(seq), sock_i_uid(sk));
3202 	}
3203 
3204 	meta.seq = seq;
3205 	prog = bpf_iter_get_info(&meta, false);
3206 	ret = tcp_prog_seq_show(prog, &meta, v, uid);
3207 
3208 unlock:
3209 	if (sk_fullsock(sk))
3210 		release_sock(sk);
3211 	return ret;
3212 
3213 }
3214 
3215 static void bpf_iter_tcp_seq_stop(struct seq_file *seq, void *v)
3216 {
3217 	struct bpf_tcp_iter_state *iter = seq->private;
3218 	struct bpf_iter_meta meta;
3219 	struct bpf_prog *prog;
3220 
3221 	if (!v) {
3222 		meta.seq = seq;
3223 		prog = bpf_iter_get_info(&meta, true);
3224 		if (prog)
3225 			(void)tcp_prog_seq_show(prog, &meta, v, 0);
3226 	}
3227 
3228 	if (iter->cur_sk < iter->end_sk) {
3229 		bpf_iter_tcp_put_batch(iter);
3230 		iter->st_bucket_done = false;
3231 	}
3232 }
3233 
3234 static const struct seq_operations bpf_iter_tcp_seq_ops = {
3235 	.show		= bpf_iter_tcp_seq_show,
3236 	.start		= bpf_iter_tcp_seq_start,
3237 	.next		= bpf_iter_tcp_seq_next,
3238 	.stop		= bpf_iter_tcp_seq_stop,
3239 };
3240 #endif
3241 static unsigned short seq_file_family(const struct seq_file *seq)
3242 {
3243 	const struct tcp_seq_afinfo *afinfo;
3244 
3245 #ifdef CONFIG_BPF_SYSCALL
3246 	/* Iterated from bpf_iter.  Let the bpf prog to filter instead. */
3247 	if (seq->op == &bpf_iter_tcp_seq_ops)
3248 		return AF_UNSPEC;
3249 #endif
3250 
3251 	/* Iterated from proc fs */
3252 	afinfo = pde_data(file_inode(seq->file));
3253 	return afinfo->family;
3254 }
3255 
3256 static const struct seq_operations tcp4_seq_ops = {
3257 	.show		= tcp4_seq_show,
3258 	.start		= tcp_seq_start,
3259 	.next		= tcp_seq_next,
3260 	.stop		= tcp_seq_stop,
3261 };
3262 
3263 static struct tcp_seq_afinfo tcp4_seq_afinfo = {
3264 	.family		= AF_INET,
3265 };
3266 
3267 static int __net_init tcp4_proc_init_net(struct net *net)
3268 {
3269 	if (!proc_create_net_data("tcp", 0444, net->proc_net, &tcp4_seq_ops,
3270 			sizeof(struct tcp_iter_state), &tcp4_seq_afinfo))
3271 		return -ENOMEM;
3272 	return 0;
3273 }
3274 
3275 static void __net_exit tcp4_proc_exit_net(struct net *net)
3276 {
3277 	remove_proc_entry("tcp", net->proc_net);
3278 }
3279 
3280 static struct pernet_operations tcp4_net_ops = {
3281 	.init = tcp4_proc_init_net,
3282 	.exit = tcp4_proc_exit_net,
3283 };
3284 
3285 int __init tcp4_proc_init(void)
3286 {
3287 	return register_pernet_subsys(&tcp4_net_ops);
3288 }
3289 
3290 void tcp4_proc_exit(void)
3291 {
3292 	unregister_pernet_subsys(&tcp4_net_ops);
3293 }
3294 #endif /* CONFIG_PROC_FS */
3295 
3296 /* @wake is one when sk_stream_write_space() calls us.
3297  * This sends EPOLLOUT only if notsent_bytes is half the limit.
3298  * This mimics the strategy used in sock_def_write_space().
3299  */
3300 bool tcp_stream_memory_free(const struct sock *sk, int wake)
3301 {
3302 	const struct tcp_sock *tp = tcp_sk(sk);
3303 	u32 notsent_bytes = READ_ONCE(tp->write_seq) -
3304 			    READ_ONCE(tp->snd_nxt);
3305 
3306 	return (notsent_bytes << wake) < tcp_notsent_lowat(tp);
3307 }
3308 EXPORT_SYMBOL(tcp_stream_memory_free);
3309 
3310 struct proto tcp_prot = {
3311 	.name			= "TCP",
3312 	.owner			= THIS_MODULE,
3313 	.close			= tcp_close,
3314 	.pre_connect		= tcp_v4_pre_connect,
3315 	.connect		= tcp_v4_connect,
3316 	.disconnect		= tcp_disconnect,
3317 	.accept			= inet_csk_accept,
3318 	.ioctl			= tcp_ioctl,
3319 	.init			= tcp_v4_init_sock,
3320 	.destroy		= tcp_v4_destroy_sock,
3321 	.shutdown		= tcp_shutdown,
3322 	.setsockopt		= tcp_setsockopt,
3323 	.getsockopt		= tcp_getsockopt,
3324 	.bpf_bypass_getsockopt	= tcp_bpf_bypass_getsockopt,
3325 	.keepalive		= tcp_set_keepalive,
3326 	.recvmsg		= tcp_recvmsg,
3327 	.sendmsg		= tcp_sendmsg,
3328 	.splice_eof		= tcp_splice_eof,
3329 	.backlog_rcv		= tcp_v4_do_rcv,
3330 	.release_cb		= tcp_release_cb,
3331 	.hash			= inet_hash,
3332 	.unhash			= inet_unhash,
3333 	.get_port		= inet_csk_get_port,
3334 	.put_port		= inet_put_port,
3335 #ifdef CONFIG_BPF_SYSCALL
3336 	.psock_update_sk_prot	= tcp_bpf_update_proto,
3337 #endif
3338 	.enter_memory_pressure	= tcp_enter_memory_pressure,
3339 	.leave_memory_pressure	= tcp_leave_memory_pressure,
3340 	.stream_memory_free	= tcp_stream_memory_free,
3341 	.sockets_allocated	= &tcp_sockets_allocated,
3342 	.orphan_count		= &tcp_orphan_count,
3343 
3344 	.memory_allocated	= &tcp_memory_allocated,
3345 	.per_cpu_fw_alloc	= &tcp_memory_per_cpu_fw_alloc,
3346 
3347 	.memory_pressure	= &tcp_memory_pressure,
3348 	.sysctl_mem		= sysctl_tcp_mem,
3349 	.sysctl_wmem_offset	= offsetof(struct net, ipv4.sysctl_tcp_wmem),
3350 	.sysctl_rmem_offset	= offsetof(struct net, ipv4.sysctl_tcp_rmem),
3351 	.max_header		= MAX_TCP_HEADER,
3352 	.obj_size		= sizeof(struct tcp_sock),
3353 	.slab_flags		= SLAB_TYPESAFE_BY_RCU,
3354 	.twsk_prot		= &tcp_timewait_sock_ops,
3355 	.rsk_prot		= &tcp_request_sock_ops,
3356 	.h.hashinfo		= NULL,
3357 	.no_autobind		= true,
3358 	.diag_destroy		= tcp_abort,
3359 };
3360 EXPORT_SYMBOL(tcp_prot);
3361 
3362 static void __net_exit tcp_sk_exit(struct net *net)
3363 {
3364 	if (net->ipv4.tcp_congestion_control)
3365 		bpf_module_put(net->ipv4.tcp_congestion_control,
3366 			       net->ipv4.tcp_congestion_control->owner);
3367 }
3368 
3369 static void __net_init tcp_set_hashinfo(struct net *net)
3370 {
3371 	struct inet_hashinfo *hinfo;
3372 	unsigned int ehash_entries;
3373 	struct net *old_net;
3374 
3375 	if (net_eq(net, &init_net))
3376 		goto fallback;
3377 
3378 	old_net = current->nsproxy->net_ns;
3379 	ehash_entries = READ_ONCE(old_net->ipv4.sysctl_tcp_child_ehash_entries);
3380 	if (!ehash_entries)
3381 		goto fallback;
3382 
3383 	ehash_entries = roundup_pow_of_two(ehash_entries);
3384 	hinfo = inet_pernet_hashinfo_alloc(&tcp_hashinfo, ehash_entries);
3385 	if (!hinfo) {
3386 		pr_warn("Failed to allocate TCP ehash (entries: %u) "
3387 			"for a netns, fallback to the global one\n",
3388 			ehash_entries);
3389 fallback:
3390 		hinfo = &tcp_hashinfo;
3391 		ehash_entries = tcp_hashinfo.ehash_mask + 1;
3392 	}
3393 
3394 	net->ipv4.tcp_death_row.hashinfo = hinfo;
3395 	net->ipv4.tcp_death_row.sysctl_max_tw_buckets = ehash_entries / 2;
3396 	net->ipv4.sysctl_max_syn_backlog = max(128U, ehash_entries / 128);
3397 }
3398 
3399 static int __net_init tcp_sk_init(struct net *net)
3400 {
3401 	net->ipv4.sysctl_tcp_ecn = 2;
3402 	net->ipv4.sysctl_tcp_ecn_fallback = 1;
3403 
3404 	net->ipv4.sysctl_tcp_base_mss = TCP_BASE_MSS;
3405 	net->ipv4.sysctl_tcp_min_snd_mss = TCP_MIN_SND_MSS;
3406 	net->ipv4.sysctl_tcp_probe_threshold = TCP_PROBE_THRESHOLD;
3407 	net->ipv4.sysctl_tcp_probe_interval = TCP_PROBE_INTERVAL;
3408 	net->ipv4.sysctl_tcp_mtu_probe_floor = TCP_MIN_SND_MSS;
3409 
3410 	net->ipv4.sysctl_tcp_keepalive_time = TCP_KEEPALIVE_TIME;
3411 	net->ipv4.sysctl_tcp_keepalive_probes = TCP_KEEPALIVE_PROBES;
3412 	net->ipv4.sysctl_tcp_keepalive_intvl = TCP_KEEPALIVE_INTVL;
3413 
3414 	net->ipv4.sysctl_tcp_syn_retries = TCP_SYN_RETRIES;
3415 	net->ipv4.sysctl_tcp_synack_retries = TCP_SYNACK_RETRIES;
3416 	net->ipv4.sysctl_tcp_syncookies = 1;
3417 	net->ipv4.sysctl_tcp_reordering = TCP_FASTRETRANS_THRESH;
3418 	net->ipv4.sysctl_tcp_retries1 = TCP_RETR1;
3419 	net->ipv4.sysctl_tcp_retries2 = TCP_RETR2;
3420 	net->ipv4.sysctl_tcp_orphan_retries = 0;
3421 	net->ipv4.sysctl_tcp_fin_timeout = TCP_FIN_TIMEOUT;
3422 	net->ipv4.sysctl_tcp_notsent_lowat = UINT_MAX;
3423 	net->ipv4.sysctl_tcp_tw_reuse = 2;
3424 	net->ipv4.sysctl_tcp_no_ssthresh_metrics_save = 1;
3425 
3426 	refcount_set(&net->ipv4.tcp_death_row.tw_refcount, 1);
3427 	tcp_set_hashinfo(net);
3428 
3429 	net->ipv4.sysctl_tcp_sack = 1;
3430 	net->ipv4.sysctl_tcp_window_scaling = 1;
3431 	net->ipv4.sysctl_tcp_timestamps = 1;
3432 	net->ipv4.sysctl_tcp_early_retrans = 3;
3433 	net->ipv4.sysctl_tcp_recovery = TCP_RACK_LOSS_DETECTION;
3434 	net->ipv4.sysctl_tcp_slow_start_after_idle = 1; /* By default, RFC2861 behavior.  */
3435 	net->ipv4.sysctl_tcp_retrans_collapse = 1;
3436 	net->ipv4.sysctl_tcp_max_reordering = 300;
3437 	net->ipv4.sysctl_tcp_dsack = 1;
3438 	net->ipv4.sysctl_tcp_app_win = 31;
3439 	net->ipv4.sysctl_tcp_adv_win_scale = 1;
3440 	net->ipv4.sysctl_tcp_frto = 2;
3441 	net->ipv4.sysctl_tcp_moderate_rcvbuf = 1;
3442 	/* This limits the percentage of the congestion window which we
3443 	 * will allow a single TSO frame to consume.  Building TSO frames
3444 	 * which are too large can cause TCP streams to be bursty.
3445 	 */
3446 	net->ipv4.sysctl_tcp_tso_win_divisor = 3;
3447 	/* Default TSQ limit of 16 TSO segments */
3448 	net->ipv4.sysctl_tcp_limit_output_bytes = 16 * 65536;
3449 
3450 	/* rfc5961 challenge ack rate limiting, per net-ns, disabled by default. */
3451 	net->ipv4.sysctl_tcp_challenge_ack_limit = INT_MAX;
3452 
3453 	net->ipv4.sysctl_tcp_min_tso_segs = 2;
3454 	net->ipv4.sysctl_tcp_tso_rtt_log = 9;  /* 2^9 = 512 usec */
3455 	net->ipv4.sysctl_tcp_min_rtt_wlen = 300;
3456 	net->ipv4.sysctl_tcp_autocorking = 1;
3457 	net->ipv4.sysctl_tcp_invalid_ratelimit = HZ/2;
3458 	net->ipv4.sysctl_tcp_pacing_ss_ratio = 200;
3459 	net->ipv4.sysctl_tcp_pacing_ca_ratio = 120;
3460 	if (net != &init_net) {
3461 		memcpy(net->ipv4.sysctl_tcp_rmem,
3462 		       init_net.ipv4.sysctl_tcp_rmem,
3463 		       sizeof(init_net.ipv4.sysctl_tcp_rmem));
3464 		memcpy(net->ipv4.sysctl_tcp_wmem,
3465 		       init_net.ipv4.sysctl_tcp_wmem,
3466 		       sizeof(init_net.ipv4.sysctl_tcp_wmem));
3467 	}
3468 	net->ipv4.sysctl_tcp_comp_sack_delay_ns = NSEC_PER_MSEC;
3469 	net->ipv4.sysctl_tcp_comp_sack_slack_ns = 100 * NSEC_PER_USEC;
3470 	net->ipv4.sysctl_tcp_comp_sack_nr = 44;
3471 	net->ipv4.sysctl_tcp_backlog_ack_defer = 1;
3472 	net->ipv4.sysctl_tcp_fastopen = TFO_CLIENT_ENABLE;
3473 	net->ipv4.sysctl_tcp_fastopen_blackhole_timeout = 0;
3474 	atomic_set(&net->ipv4.tfo_active_disable_times, 0);
3475 
3476 	/* Set default values for PLB */
3477 	net->ipv4.sysctl_tcp_plb_enabled = 0; /* Disabled by default */
3478 	net->ipv4.sysctl_tcp_plb_idle_rehash_rounds = 3;
3479 	net->ipv4.sysctl_tcp_plb_rehash_rounds = 12;
3480 	net->ipv4.sysctl_tcp_plb_suspend_rto_sec = 60;
3481 	/* Default congestion threshold for PLB to mark a round is 50% */
3482 	net->ipv4.sysctl_tcp_plb_cong_thresh = (1 << TCP_PLB_SCALE) / 2;
3483 
3484 	/* Reno is always built in */
3485 	if (!net_eq(net, &init_net) &&
3486 	    bpf_try_module_get(init_net.ipv4.tcp_congestion_control,
3487 			       init_net.ipv4.tcp_congestion_control->owner))
3488 		net->ipv4.tcp_congestion_control = init_net.ipv4.tcp_congestion_control;
3489 	else
3490 		net->ipv4.tcp_congestion_control = &tcp_reno;
3491 
3492 	net->ipv4.sysctl_tcp_syn_linear_timeouts = 4;
3493 	net->ipv4.sysctl_tcp_shrink_window = 0;
3494 
3495 	net->ipv4.sysctl_tcp_pingpong_thresh = 1;
3496 
3497 	return 0;
3498 }
3499 
3500 static void __net_exit tcp_sk_exit_batch(struct list_head *net_exit_list)
3501 {
3502 	struct net *net;
3503 
3504 	tcp_twsk_purge(net_exit_list, AF_INET);
3505 
3506 	list_for_each_entry(net, net_exit_list, exit_list) {
3507 		inet_pernet_hashinfo_free(net->ipv4.tcp_death_row.hashinfo);
3508 		WARN_ON_ONCE(!refcount_dec_and_test(&net->ipv4.tcp_death_row.tw_refcount));
3509 		tcp_fastopen_ctx_destroy(net);
3510 	}
3511 }
3512 
3513 static struct pernet_operations __net_initdata tcp_sk_ops = {
3514        .init	   = tcp_sk_init,
3515        .exit	   = tcp_sk_exit,
3516        .exit_batch = tcp_sk_exit_batch,
3517 };
3518 
3519 #if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
3520 DEFINE_BPF_ITER_FUNC(tcp, struct bpf_iter_meta *meta,
3521 		     struct sock_common *sk_common, uid_t uid)
3522 
3523 #define INIT_BATCH_SZ 16
3524 
3525 static int bpf_iter_init_tcp(void *priv_data, struct bpf_iter_aux_info *aux)
3526 {
3527 	struct bpf_tcp_iter_state *iter = priv_data;
3528 	int err;
3529 
3530 	err = bpf_iter_init_seq_net(priv_data, aux);
3531 	if (err)
3532 		return err;
3533 
3534 	err = bpf_iter_tcp_realloc_batch(iter, INIT_BATCH_SZ);
3535 	if (err) {
3536 		bpf_iter_fini_seq_net(priv_data);
3537 		return err;
3538 	}
3539 
3540 	return 0;
3541 }
3542 
3543 static void bpf_iter_fini_tcp(void *priv_data)
3544 {
3545 	struct bpf_tcp_iter_state *iter = priv_data;
3546 
3547 	bpf_iter_fini_seq_net(priv_data);
3548 	kvfree(iter->batch);
3549 }
3550 
3551 static const struct bpf_iter_seq_info tcp_seq_info = {
3552 	.seq_ops		= &bpf_iter_tcp_seq_ops,
3553 	.init_seq_private	= bpf_iter_init_tcp,
3554 	.fini_seq_private	= bpf_iter_fini_tcp,
3555 	.seq_priv_size		= sizeof(struct bpf_tcp_iter_state),
3556 };
3557 
3558 static const struct bpf_func_proto *
3559 bpf_iter_tcp_get_func_proto(enum bpf_func_id func_id,
3560 			    const struct bpf_prog *prog)
3561 {
3562 	switch (func_id) {
3563 	case BPF_FUNC_setsockopt:
3564 		return &bpf_sk_setsockopt_proto;
3565 	case BPF_FUNC_getsockopt:
3566 		return &bpf_sk_getsockopt_proto;
3567 	default:
3568 		return NULL;
3569 	}
3570 }
3571 
3572 static struct bpf_iter_reg tcp_reg_info = {
3573 	.target			= "tcp",
3574 	.ctx_arg_info_size	= 1,
3575 	.ctx_arg_info		= {
3576 		{ offsetof(struct bpf_iter__tcp, sk_common),
3577 		  PTR_TO_BTF_ID_OR_NULL | PTR_TRUSTED },
3578 	},
3579 	.get_func_proto		= bpf_iter_tcp_get_func_proto,
3580 	.seq_info		= &tcp_seq_info,
3581 };
3582 
3583 static void __init bpf_iter_register(void)
3584 {
3585 	tcp_reg_info.ctx_arg_info[0].btf_id = btf_sock_ids[BTF_SOCK_TYPE_SOCK_COMMON];
3586 	if (bpf_iter_reg_target(&tcp_reg_info))
3587 		pr_warn("Warning: could not register bpf iterator tcp\n");
3588 }
3589 
3590 #endif
3591 
3592 void __init tcp_v4_init(void)
3593 {
3594 	int cpu, res;
3595 
3596 	for_each_possible_cpu(cpu) {
3597 		struct sock *sk;
3598 
3599 		res = inet_ctl_sock_create(&sk, PF_INET, SOCK_RAW,
3600 					   IPPROTO_TCP, &init_net);
3601 		if (res)
3602 			panic("Failed to create the TCP control socket.\n");
3603 		sock_set_flag(sk, SOCK_USE_WRITE_QUEUE);
3604 
3605 		/* Please enforce IP_DF and IPID==0 for RST and
3606 		 * ACK sent in SYN-RECV and TIME-WAIT state.
3607 		 */
3608 		inet_sk(sk)->pmtudisc = IP_PMTUDISC_DO;
3609 
3610 		per_cpu(ipv4_tcp_sk, cpu) = sk;
3611 	}
3612 	if (register_pernet_subsys(&tcp_sk_ops))
3613 		panic("Failed to create the TCP control socket.\n");
3614 
3615 #if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
3616 	bpf_iter_register();
3617 #endif
3618 }
3619