Linux Audio

Check our new training course

Linux BSP upgrade and security maintenance

Need help to get security updates for your Linux BSP?
Loading...
v5.14.15
   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
  61#include <net/net_namespace.h>
  62#include <net/icmp.h>
  63#include <net/inet_hashtables.h>
  64#include <net/tcp.h>
  65#include <net/transp_v6.h>
  66#include <net/ipv6.h>
  67#include <net/inet_common.h>
  68#include <net/timewait_sock.h>
  69#include <net/xfrm.h>
  70#include <net/secure_seq.h>
  71#include <net/busy_poll.h>
  72
  73#include <linux/inet.h>
  74#include <linux/ipv6.h>
  75#include <linux/stddef.h>
  76#include <linux/proc_fs.h>
  77#include <linux/seq_file.h>
  78#include <linux/inetdevice.h>
  79#include <linux/btf_ids.h>
  80
  81#include <crypto/hash.h>
  82#include <linux/scatterlist.h>
  83
  84#include <trace/events/tcp.h>
  85
  86#ifdef CONFIG_TCP_MD5SIG
  87static int tcp_v4_md5_hash_hdr(char *md5_hash, const struct tcp_md5sig_key *key,
  88			       __be32 daddr, __be32 saddr, const struct tcphdr *th);
  89#endif
  90
  91struct inet_hashinfo tcp_hashinfo;
  92EXPORT_SYMBOL(tcp_hashinfo);
  93
  94static u32 tcp_v4_init_seq(const struct sk_buff *skb)
  95{
  96	return secure_tcp_seq(ip_hdr(skb)->daddr,
  97			      ip_hdr(skb)->saddr,
  98			      tcp_hdr(skb)->dest,
  99			      tcp_hdr(skb)->source);
 100}
 101
 102static u32 tcp_v4_init_ts_off(const struct net *net, const struct sk_buff *skb)
 103{
 104	return secure_tcp_ts_off(net, ip_hdr(skb)->daddr, ip_hdr(skb)->saddr);
 105}
 106
 107int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp)
 108{
 109	const struct inet_timewait_sock *tw = inet_twsk(sktw);
 110	const struct tcp_timewait_sock *tcptw = tcp_twsk(sktw);
 111	struct tcp_sock *tp = tcp_sk(sk);
 112	int reuse = sock_net(sk)->ipv4.sysctl_tcp_tw_reuse;
 113
 114	if (reuse == 2) {
 115		/* Still does not detect *everything* that goes through
 116		 * lo, since we require a loopback src or dst address
 117		 * or direct binding to 'lo' interface.
 118		 */
 119		bool loopback = false;
 120		if (tw->tw_bound_dev_if == LOOPBACK_IFINDEX)
 121			loopback = true;
 122#if IS_ENABLED(CONFIG_IPV6)
 123		if (tw->tw_family == AF_INET6) {
 124			if (ipv6_addr_loopback(&tw->tw_v6_daddr) ||
 125			    ipv6_addr_v4mapped_loopback(&tw->tw_v6_daddr) ||
 126			    ipv6_addr_loopback(&tw->tw_v6_rcv_saddr) ||
 127			    ipv6_addr_v4mapped_loopback(&tw->tw_v6_rcv_saddr))
 128				loopback = true;
 129		} else
 130#endif
 131		{
 132			if (ipv4_is_loopback(tw->tw_daddr) ||
 133			    ipv4_is_loopback(tw->tw_rcv_saddr))
 134				loopback = true;
 135		}
 136		if (!loopback)
 137			reuse = 0;
 138	}
 139
 140	/* With PAWS, it is safe from the viewpoint
 141	   of data integrity. Even without PAWS it is safe provided sequence
 142	   spaces do not overlap i.e. at data rates <= 80Mbit/sec.
 143
 144	   Actually, the idea is close to VJ's one, only timestamp cache is
 145	   held not per host, but per port pair and TW bucket is used as state
 146	   holder.
 147
 148	   If TW bucket has been already destroyed we fall back to VJ's scheme
 149	   and use initial timestamp retrieved from peer table.
 150	 */
 151	if (tcptw->tw_ts_recent_stamp &&
 152	    (!twp || (reuse && time_after32(ktime_get_seconds(),
 153					    tcptw->tw_ts_recent_stamp)))) {
 154		/* In case of repair and re-using TIME-WAIT sockets we still
 155		 * want to be sure that it is safe as above but honor the
 156		 * sequence numbers and time stamps set as part of the repair
 157		 * process.
 158		 *
 159		 * Without this check re-using a TIME-WAIT socket with TCP
 160		 * repair would accumulate a -1 on the repair assigned
 161		 * sequence number. The first time it is reused the sequence
 162		 * is -1, the second time -2, etc. This fixes that issue
 163		 * without appearing to create any others.
 164		 */
 165		if (likely(!tp->repair)) {
 166			u32 seq = tcptw->tw_snd_nxt + 65535 + 2;
 167
 168			if (!seq)
 169				seq = 1;
 170			WRITE_ONCE(tp->write_seq, seq);
 171			tp->rx_opt.ts_recent	   = tcptw->tw_ts_recent;
 172			tp->rx_opt.ts_recent_stamp = tcptw->tw_ts_recent_stamp;
 173		}
 174		sock_hold(sktw);
 175		return 1;
 176	}
 177
 178	return 0;
 179}
 180EXPORT_SYMBOL_GPL(tcp_twsk_unique);
 181
 182static int tcp_v4_pre_connect(struct sock *sk, struct sockaddr *uaddr,
 183			      int addr_len)
 184{
 185	/* This check is replicated from tcp_v4_connect() and intended to
 186	 * prevent BPF program called below from accessing bytes that are out
 187	 * of the bound specified by user in addr_len.
 188	 */
 189	if (addr_len < sizeof(struct sockaddr_in))
 190		return -EINVAL;
 191
 192	sock_owned_by_me(sk);
 193
 194	return BPF_CGROUP_RUN_PROG_INET4_CONNECT(sk, uaddr);
 195}
 196
 197/* This will initiate an outgoing connection. */
 198int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len)
 199{
 200	struct sockaddr_in *usin = (struct sockaddr_in *)uaddr;
 201	struct inet_sock *inet = inet_sk(sk);
 202	struct tcp_sock *tp = tcp_sk(sk);
 203	__be16 orig_sport, orig_dport;
 204	__be32 daddr, nexthop;
 205	struct flowi4 *fl4;
 206	struct rtable *rt;
 207	int err;
 208	struct ip_options_rcu *inet_opt;
 209	struct inet_timewait_death_row *tcp_death_row = &sock_net(sk)->ipv4.tcp_death_row;
 210
 211	if (addr_len < sizeof(struct sockaddr_in))
 212		return -EINVAL;
 213
 214	if (usin->sin_family != AF_INET)
 215		return -EAFNOSUPPORT;
 216
 217	nexthop = daddr = usin->sin_addr.s_addr;
 218	inet_opt = rcu_dereference_protected(inet->inet_opt,
 219					     lockdep_sock_is_held(sk));
 220	if (inet_opt && inet_opt->opt.srr) {
 221		if (!daddr)
 222			return -EINVAL;
 223		nexthop = inet_opt->opt.faddr;
 224	}
 225
 226	orig_sport = inet->inet_sport;
 227	orig_dport = usin->sin_port;
 228	fl4 = &inet->cork.fl.u.ip4;
 229	rt = ip_route_connect(fl4, nexthop, inet->inet_saddr,
 230			      RT_CONN_FLAGS(sk), sk->sk_bound_dev_if,
 231			      IPPROTO_TCP,
 232			      orig_sport, orig_dport, sk);
 233	if (IS_ERR(rt)) {
 234		err = PTR_ERR(rt);
 235		if (err == -ENETUNREACH)
 236			IP_INC_STATS(sock_net(sk), IPSTATS_MIB_OUTNOROUTES);
 237		return err;
 238	}
 239
 240	if (rt->rt_flags & (RTCF_MULTICAST | RTCF_BROADCAST)) {
 241		ip_rt_put(rt);
 242		return -ENETUNREACH;
 243	}
 244
 245	if (!inet_opt || !inet_opt->opt.srr)
 246		daddr = fl4->daddr;
 247
 248	if (!inet->inet_saddr)
 249		inet->inet_saddr = fl4->saddr;
 250	sk_rcv_saddr_set(sk, inet->inet_saddr);
 251
 252	if (tp->rx_opt.ts_recent_stamp && inet->inet_daddr != daddr) {
 253		/* Reset inherited state */
 254		tp->rx_opt.ts_recent	   = 0;
 255		tp->rx_opt.ts_recent_stamp = 0;
 256		if (likely(!tp->repair))
 257			WRITE_ONCE(tp->write_seq, 0);
 258	}
 259
 260	inet->inet_dport = usin->sin_port;
 261	sk_daddr_set(sk, daddr);
 262
 263	inet_csk(sk)->icsk_ext_hdr_len = 0;
 264	if (inet_opt)
 265		inet_csk(sk)->icsk_ext_hdr_len = inet_opt->opt.optlen;
 266
 267	tp->rx_opt.mss_clamp = TCP_MSS_DEFAULT;
 268
 269	/* Socket identity is still unknown (sport may be zero).
 270	 * However we set state to SYN-SENT and not releasing socket
 271	 * lock select source port, enter ourselves into the hash tables and
 272	 * complete initialization after this.
 273	 */
 274	tcp_set_state(sk, TCP_SYN_SENT);
 275	err = inet_hash_connect(tcp_death_row, sk);
 276	if (err)
 277		goto failure;
 278
 279	sk_set_txhash(sk);
 280
 281	rt = ip_route_newports(fl4, rt, orig_sport, orig_dport,
 282			       inet->inet_sport, inet->inet_dport, sk);
 283	if (IS_ERR(rt)) {
 284		err = PTR_ERR(rt);
 285		rt = NULL;
 286		goto failure;
 287	}
 288	/* OK, now commit destination to socket.  */
 289	sk->sk_gso_type = SKB_GSO_TCPV4;
 290	sk_setup_caps(sk, &rt->dst);
 291	rt = NULL;
 292
 293	if (likely(!tp->repair)) {
 294		if (!tp->write_seq)
 295			WRITE_ONCE(tp->write_seq,
 296				   secure_tcp_seq(inet->inet_saddr,
 297						  inet->inet_daddr,
 298						  inet->inet_sport,
 299						  usin->sin_port));
 300		tp->tsoffset = secure_tcp_ts_off(sock_net(sk),
 301						 inet->inet_saddr,
 302						 inet->inet_daddr);
 303	}
 304
 305	inet->inet_id = prandom_u32();
 306
 307	if (tcp_fastopen_defer_connect(sk, &err))
 308		return err;
 309	if (err)
 310		goto failure;
 311
 312	err = tcp_connect(sk);
 313
 314	if (err)
 315		goto failure;
 316
 317	return 0;
 318
 319failure:
 320	/*
 321	 * This unhashes the socket and releases the local port,
 322	 * if necessary.
 323	 */
 324	tcp_set_state(sk, TCP_CLOSE);
 325	ip_rt_put(rt);
 326	sk->sk_route_caps = 0;
 327	inet->inet_dport = 0;
 328	return err;
 329}
 330EXPORT_SYMBOL(tcp_v4_connect);
 331
 332/*
 333 * This routine reacts to ICMP_FRAG_NEEDED mtu indications as defined in RFC1191.
 334 * It can be called through tcp_release_cb() if socket was owned by user
 335 * at the time tcp_v4_err() was called to handle ICMP message.
 336 */
 337void tcp_v4_mtu_reduced(struct sock *sk)
 338{
 339	struct inet_sock *inet = inet_sk(sk);
 340	struct dst_entry *dst;
 341	u32 mtu;
 342
 343	if ((1 << sk->sk_state) & (TCPF_LISTEN | TCPF_CLOSE))
 344		return;
 345	mtu = READ_ONCE(tcp_sk(sk)->mtu_info);
 346	dst = inet_csk_update_pmtu(sk, mtu);
 347	if (!dst)
 348		return;
 349
 350	/* Something is about to be wrong... Remember soft error
 351	 * for the case, if this connection will not able to recover.
 352	 */
 353	if (mtu < dst_mtu(dst) && ip_dont_fragment(sk, dst))
 354		sk->sk_err_soft = EMSGSIZE;
 355
 356	mtu = dst_mtu(dst);
 357
 358	if (inet->pmtudisc != IP_PMTUDISC_DONT &&
 359	    ip_sk_accept_pmtu(sk) &&
 360	    inet_csk(sk)->icsk_pmtu_cookie > mtu) {
 361		tcp_sync_mss(sk, mtu);
 362
 363		/* Resend the TCP packet because it's
 364		 * clear that the old packet has been
 365		 * dropped. This is the new "fast" path mtu
 366		 * discovery.
 367		 */
 368		tcp_simple_retransmit(sk);
 369	} /* else let the usual retransmit timer handle it */
 370}
 371EXPORT_SYMBOL(tcp_v4_mtu_reduced);
 372
 373static void do_redirect(struct sk_buff *skb, struct sock *sk)
 374{
 375	struct dst_entry *dst = __sk_dst_check(sk, 0);
 376
 377	if (dst)
 378		dst->ops->redirect(dst, sk, skb);
 379}
 380
 381
 382/* handle ICMP messages on TCP_NEW_SYN_RECV request sockets */
 383void tcp_req_err(struct sock *sk, u32 seq, bool abort)
 384{
 385	struct request_sock *req = inet_reqsk(sk);
 386	struct net *net = sock_net(sk);
 387
 388	/* ICMPs are not backlogged, hence we cannot get
 389	 * an established socket here.
 390	 */
 391	if (seq != tcp_rsk(req)->snt_isn) {
 392		__NET_INC_STATS(net, LINUX_MIB_OUTOFWINDOWICMPS);
 393	} else if (abort) {
 394		/*
 395		 * Still in SYN_RECV, just remove it silently.
 396		 * There is no good way to pass the error to the newly
 397		 * created socket, and POSIX does not want network
 398		 * errors returned from accept().
 399		 */
 400		inet_csk_reqsk_queue_drop(req->rsk_listener, req);
 401		tcp_listendrop(req->rsk_listener);
 402	}
 403	reqsk_put(req);
 404}
 405EXPORT_SYMBOL(tcp_req_err);
 406
 407/* TCP-LD (RFC 6069) logic */
 408void tcp_ld_RTO_revert(struct sock *sk, u32 seq)
 409{
 410	struct inet_connection_sock *icsk = inet_csk(sk);
 411	struct tcp_sock *tp = tcp_sk(sk);
 412	struct sk_buff *skb;
 413	s32 remaining;
 414	u32 delta_us;
 415
 416	if (sock_owned_by_user(sk))
 417		return;
 418
 419	if (seq != tp->snd_una  || !icsk->icsk_retransmits ||
 420	    !icsk->icsk_backoff)
 421		return;
 422
 423	skb = tcp_rtx_queue_head(sk);
 424	if (WARN_ON_ONCE(!skb))
 425		return;
 426
 427	icsk->icsk_backoff--;
 428	icsk->icsk_rto = tp->srtt_us ? __tcp_set_rto(tp) : TCP_TIMEOUT_INIT;
 429	icsk->icsk_rto = inet_csk_rto_backoff(icsk, TCP_RTO_MAX);
 430
 431	tcp_mstamp_refresh(tp);
 432	delta_us = (u32)(tp->tcp_mstamp - tcp_skb_timestamp_us(skb));
 433	remaining = icsk->icsk_rto - usecs_to_jiffies(delta_us);
 434
 435	if (remaining > 0) {
 436		inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
 437					  remaining, TCP_RTO_MAX);
 438	} else {
 439		/* RTO revert clocked out retransmission.
 440		 * Will retransmit now.
 441		 */
 442		tcp_retransmit_timer(sk);
 443	}
 444}
 445EXPORT_SYMBOL(tcp_ld_RTO_revert);
 446
 447/*
 448 * This routine is called by the ICMP module when it gets some
 449 * sort of error condition.  If err < 0 then the socket should
 450 * be closed and the error returned to the user.  If err > 0
 451 * it's just the icmp type << 8 | icmp code.  After adjustment
 452 * header points to the first 8 bytes of the tcp header.  We need
 453 * to find the appropriate port.
 454 *
 455 * The locking strategy used here is very "optimistic". When
 456 * someone else accesses the socket the ICMP is just dropped
 457 * and for some paths there is no check at all.
 458 * A more general error queue to queue errors for later handling
 459 * is probably better.
 460 *
 461 */
 462
 463int tcp_v4_err(struct sk_buff *skb, u32 info)
 464{
 465	const struct iphdr *iph = (const struct iphdr *)skb->data;
 466	struct tcphdr *th = (struct tcphdr *)(skb->data + (iph->ihl << 2));
 467	struct tcp_sock *tp;
 468	struct inet_sock *inet;
 469	const int type = icmp_hdr(skb)->type;
 470	const int code = icmp_hdr(skb)->code;
 471	struct sock *sk;
 472	struct request_sock *fastopen;
 473	u32 seq, snd_una;
 474	int err;
 475	struct net *net = dev_net(skb->dev);
 476
 477	sk = __inet_lookup_established(net, &tcp_hashinfo, iph->daddr,
 478				       th->dest, iph->saddr, ntohs(th->source),
 479				       inet_iif(skb), 0);
 480	if (!sk) {
 481		__ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
 482		return -ENOENT;
 483	}
 484	if (sk->sk_state == TCP_TIME_WAIT) {
 485		inet_twsk_put(inet_twsk(sk));
 486		return 0;
 487	}
 488	seq = ntohl(th->seq);
 489	if (sk->sk_state == TCP_NEW_SYN_RECV) {
 490		tcp_req_err(sk, seq, type == ICMP_PARAMETERPROB ||
 491				     type == ICMP_TIME_EXCEEDED ||
 492				     (type == ICMP_DEST_UNREACH &&
 493				      (code == ICMP_NET_UNREACH ||
 494				       code == ICMP_HOST_UNREACH)));
 495		return 0;
 496	}
 497
 498	bh_lock_sock(sk);
 499	/* If too many ICMPs get dropped on busy
 500	 * servers this needs to be solved differently.
 501	 * We do take care of PMTU discovery (RFC1191) special case :
 502	 * we can receive locally generated ICMP messages while socket is held.
 503	 */
 504	if (sock_owned_by_user(sk)) {
 505		if (!(type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED))
 506			__NET_INC_STATS(net, LINUX_MIB_LOCKDROPPEDICMPS);
 507	}
 508	if (sk->sk_state == TCP_CLOSE)
 509		goto out;
 510
 511	if (unlikely(iph->ttl < inet_sk(sk)->min_ttl)) {
 512		__NET_INC_STATS(net, LINUX_MIB_TCPMINTTLDROP);
 513		goto out;
 514	}
 515
 516	tp = tcp_sk(sk);
 517	/* XXX (TFO) - tp->snd_una should be ISN (tcp_create_openreq_child() */
 518	fastopen = rcu_dereference(tp->fastopen_rsk);
 519	snd_una = fastopen ? tcp_rsk(fastopen)->snt_isn : tp->snd_una;
 520	if (sk->sk_state != TCP_LISTEN &&
 521	    !between(seq, snd_una, tp->snd_nxt)) {
 522		__NET_INC_STATS(net, LINUX_MIB_OUTOFWINDOWICMPS);
 523		goto out;
 524	}
 525
 526	switch (type) {
 527	case ICMP_REDIRECT:
 528		if (!sock_owned_by_user(sk))
 529			do_redirect(skb, sk);
 530		goto out;
 531	case ICMP_SOURCE_QUENCH:
 532		/* Just silently ignore these. */
 533		goto out;
 534	case ICMP_PARAMETERPROB:
 535		err = EPROTO;
 536		break;
 537	case ICMP_DEST_UNREACH:
 538		if (code > NR_ICMP_UNREACH)
 539			goto out;
 540
 541		if (code == ICMP_FRAG_NEEDED) { /* PMTU discovery (RFC1191) */
 542			/* We are not interested in TCP_LISTEN and open_requests
 543			 * (SYN-ACKs send out by Linux are always <576bytes so
 544			 * they should go through unfragmented).
 545			 */
 546			if (sk->sk_state == TCP_LISTEN)
 547				goto out;
 548
 549			WRITE_ONCE(tp->mtu_info, info);
 550			if (!sock_owned_by_user(sk)) {
 551				tcp_v4_mtu_reduced(sk);
 552			} else {
 553				if (!test_and_set_bit(TCP_MTU_REDUCED_DEFERRED, &sk->sk_tsq_flags))
 554					sock_hold(sk);
 555			}
 556			goto out;
 557		}
 558
 559		err = icmp_err_convert[code].errno;
 560		/* check if this ICMP message allows revert of backoff.
 561		 * (see RFC 6069)
 562		 */
 563		if (!fastopen &&
 564		    (code == ICMP_NET_UNREACH || code == ICMP_HOST_UNREACH))
 565			tcp_ld_RTO_revert(sk, seq);
 566		break;
 567	case ICMP_TIME_EXCEEDED:
 568		err = EHOSTUNREACH;
 569		break;
 570	default:
 571		goto out;
 572	}
 573
 574	switch (sk->sk_state) {
 575	case TCP_SYN_SENT:
 576	case TCP_SYN_RECV:
 577		/* Only in fast or simultaneous open. If a fast open socket is
 578		 * already accepted it is treated as a connected one below.
 579		 */
 580		if (fastopen && !fastopen->sk)
 581			break;
 582
 583		ip_icmp_error(sk, skb, err, th->dest, info, (u8 *)th);
 584
 585		if (!sock_owned_by_user(sk)) {
 586			sk->sk_err = err;
 587
 588			sk_error_report(sk);
 589
 590			tcp_done(sk);
 591		} else {
 592			sk->sk_err_soft = err;
 593		}
 594		goto out;
 595	}
 596
 597	/* If we've already connected we will keep trying
 598	 * until we time out, or the user gives up.
 599	 *
 600	 * rfc1122 4.2.3.9 allows to consider as hard errors
 601	 * only PROTO_UNREACH and PORT_UNREACH (well, FRAG_FAILED too,
 602	 * but it is obsoleted by pmtu discovery).
 603	 *
 604	 * Note, that in modern internet, where routing is unreliable
 605	 * and in each dark corner broken firewalls sit, sending random
 606	 * errors ordered by their masters even this two messages finally lose
 607	 * their original sense (even Linux sends invalid PORT_UNREACHs)
 608	 *
 609	 * Now we are in compliance with RFCs.
 610	 *							--ANK (980905)
 611	 */
 612
 613	inet = inet_sk(sk);
 614	if (!sock_owned_by_user(sk) && inet->recverr) {
 615		sk->sk_err = err;
 616		sk_error_report(sk);
 617	} else	{ /* Only an error on timeout */
 618		sk->sk_err_soft = err;
 619	}
 620
 621out:
 622	bh_unlock_sock(sk);
 623	sock_put(sk);
 624	return 0;
 625}
 626
 627void __tcp_v4_send_check(struct sk_buff *skb, __be32 saddr, __be32 daddr)
 628{
 629	struct tcphdr *th = tcp_hdr(skb);
 630
 631	th->check = ~tcp_v4_check(skb->len, saddr, daddr, 0);
 632	skb->csum_start = skb_transport_header(skb) - skb->head;
 633	skb->csum_offset = offsetof(struct tcphdr, check);
 634}
 635
 636/* This routine computes an IPv4 TCP checksum. */
 637void tcp_v4_send_check(struct sock *sk, struct sk_buff *skb)
 638{
 639	const struct inet_sock *inet = inet_sk(sk);
 640
 641	__tcp_v4_send_check(skb, inet->inet_saddr, inet->inet_daddr);
 642}
 643EXPORT_SYMBOL(tcp_v4_send_check);
 644
 645/*
 646 *	This routine will send an RST to the other tcp.
 647 *
 648 *	Someone asks: why I NEVER use socket parameters (TOS, TTL etc.)
 649 *		      for reset.
 650 *	Answer: if a packet caused RST, it is not for a socket
 651 *		existing in our system, if it is matched to a socket,
 652 *		it is just duplicate segment or bug in other side's TCP.
 653 *		So that we build reply only basing on parameters
 654 *		arrived with segment.
 655 *	Exception: precedence violation. We do not implement it in any case.
 656 */
 657
 658#ifdef CONFIG_TCP_MD5SIG
 659#define OPTION_BYTES TCPOLEN_MD5SIG_ALIGNED
 660#else
 661#define OPTION_BYTES sizeof(__be32)
 662#endif
 663
 664static void tcp_v4_send_reset(const struct sock *sk, struct sk_buff *skb)
 665{
 666	const struct tcphdr *th = tcp_hdr(skb);
 667	struct {
 668		struct tcphdr th;
 669		__be32 opt[OPTION_BYTES / sizeof(__be32)];
 
 
 670	} rep;
 671	struct ip_reply_arg arg;
 672#ifdef CONFIG_TCP_MD5SIG
 673	struct tcp_md5sig_key *key = NULL;
 674	const __u8 *hash_location = NULL;
 675	unsigned char newhash[16];
 676	int genhash;
 677	struct sock *sk1 = NULL;
 678#endif
 679	u64 transmit_time = 0;
 680	struct sock *ctl_sk;
 681	struct net *net;
 682
 683	/* Never send a reset in response to a reset. */
 684	if (th->rst)
 685		return;
 686
 687	/* If sk not NULL, it means we did a successful lookup and incoming
 688	 * route had to be correct. prequeue might have dropped our dst.
 689	 */
 690	if (!sk && skb_rtable(skb)->rt_type != RTN_LOCAL)
 691		return;
 692
 693	/* Swap the send and the receive. */
 694	memset(&rep, 0, sizeof(rep));
 695	rep.th.dest   = th->source;
 696	rep.th.source = th->dest;
 697	rep.th.doff   = sizeof(struct tcphdr) / 4;
 698	rep.th.rst    = 1;
 699
 700	if (th->ack) {
 701		rep.th.seq = th->ack_seq;
 702	} else {
 703		rep.th.ack = 1;
 704		rep.th.ack_seq = htonl(ntohl(th->seq) + th->syn + th->fin +
 705				       skb->len - (th->doff << 2));
 706	}
 707
 708	memset(&arg, 0, sizeof(arg));
 709	arg.iov[0].iov_base = (unsigned char *)&rep;
 710	arg.iov[0].iov_len  = sizeof(rep.th);
 711
 712	net = sk ? sock_net(sk) : dev_net(skb_dst(skb)->dev);
 713#ifdef CONFIG_TCP_MD5SIG
 714	rcu_read_lock();
 715	hash_location = tcp_parse_md5sig_option(th);
 716	if (sk && sk_fullsock(sk)) {
 717		const union tcp_md5_addr *addr;
 718		int l3index;
 719
 720		/* sdif set, means packet ingressed via a device
 721		 * in an L3 domain and inet_iif is set to it.
 722		 */
 723		l3index = tcp_v4_sdif(skb) ? inet_iif(skb) : 0;
 724		addr = (union tcp_md5_addr *)&ip_hdr(skb)->saddr;
 725		key = tcp_md5_do_lookup(sk, l3index, addr, AF_INET);
 726	} else if (hash_location) {
 727		const union tcp_md5_addr *addr;
 728		int sdif = tcp_v4_sdif(skb);
 729		int dif = inet_iif(skb);
 730		int l3index;
 731
 732		/*
 733		 * active side is lost. Try to find listening socket through
 734		 * source port, and then find md5 key through listening socket.
 735		 * we are not loose security here:
 736		 * Incoming packet is checked with md5 hash with finding key,
 737		 * no RST generated if md5 hash doesn't match.
 738		 */
 739		sk1 = __inet_lookup_listener(net, &tcp_hashinfo, NULL, 0,
 740					     ip_hdr(skb)->saddr,
 741					     th->source, ip_hdr(skb)->daddr,
 742					     ntohs(th->source), dif, sdif);
 743		/* don't send rst if it can't find key */
 744		if (!sk1)
 745			goto out;
 746
 747		/* sdif set, means packet ingressed via a device
 748		 * in an L3 domain and dif is set to it.
 749		 */
 750		l3index = sdif ? dif : 0;
 751		addr = (union tcp_md5_addr *)&ip_hdr(skb)->saddr;
 752		key = tcp_md5_do_lookup(sk1, l3index, addr, AF_INET);
 753		if (!key)
 754			goto out;
 755
 756
 757		genhash = tcp_v4_md5_hash_skb(newhash, key, NULL, skb);
 758		if (genhash || memcmp(hash_location, newhash, 16) != 0)
 759			goto out;
 760
 761	}
 762
 763	if (key) {
 764		rep.opt[0] = htonl((TCPOPT_NOP << 24) |
 765				   (TCPOPT_NOP << 16) |
 766				   (TCPOPT_MD5SIG << 8) |
 767				   TCPOLEN_MD5SIG);
 768		/* Update length and the length the header thinks exists */
 769		arg.iov[0].iov_len += TCPOLEN_MD5SIG_ALIGNED;
 770		rep.th.doff = arg.iov[0].iov_len / 4;
 771
 772		tcp_v4_md5_hash_hdr((__u8 *) &rep.opt[1],
 773				     key, ip_hdr(skb)->saddr,
 774				     ip_hdr(skb)->daddr, &rep.th);
 775	}
 776#endif
 777	/* Can't co-exist with TCPMD5, hence check rep.opt[0] */
 778	if (rep.opt[0] == 0) {
 779		__be32 mrst = mptcp_reset_option(skb);
 780
 781		if (mrst) {
 782			rep.opt[0] = mrst;
 783			arg.iov[0].iov_len += sizeof(mrst);
 784			rep.th.doff = arg.iov[0].iov_len / 4;
 785		}
 786	}
 787
 788	arg.csum = csum_tcpudp_nofold(ip_hdr(skb)->daddr,
 789				      ip_hdr(skb)->saddr, /* XXX */
 790				      arg.iov[0].iov_len, IPPROTO_TCP, 0);
 791	arg.csumoffset = offsetof(struct tcphdr, check) / 2;
 792	arg.flags = (sk && inet_sk_transparent(sk)) ? IP_REPLY_ARG_NOSRCCHECK : 0;
 793
 794	/* When socket is gone, all binding information is lost.
 795	 * routing might fail in this case. No choice here, if we choose to force
 796	 * input interface, we will misroute in case of asymmetric route.
 797	 */
 798	if (sk) {
 799		arg.bound_dev_if = sk->sk_bound_dev_if;
 800		if (sk_fullsock(sk))
 801			trace_tcp_send_reset(sk, skb);
 802	}
 803
 804	BUILD_BUG_ON(offsetof(struct sock, sk_bound_dev_if) !=
 805		     offsetof(struct inet_timewait_sock, tw_bound_dev_if));
 806
 807	arg.tos = ip_hdr(skb)->tos;
 808	arg.uid = sock_net_uid(net, sk && sk_fullsock(sk) ? sk : NULL);
 809	local_bh_disable();
 810	ctl_sk = this_cpu_read(*net->ipv4.tcp_sk);
 811	if (sk) {
 812		ctl_sk->sk_mark = (sk->sk_state == TCP_TIME_WAIT) ?
 813				   inet_twsk(sk)->tw_mark : sk->sk_mark;
 814		ctl_sk->sk_priority = (sk->sk_state == TCP_TIME_WAIT) ?
 815				   inet_twsk(sk)->tw_priority : sk->sk_priority;
 816		transmit_time = tcp_transmit_time(sk);
 817	}
 818	ip_send_unicast_reply(ctl_sk,
 819			      skb, &TCP_SKB_CB(skb)->header.h4.opt,
 820			      ip_hdr(skb)->saddr, ip_hdr(skb)->daddr,
 821			      &arg, arg.iov[0].iov_len,
 822			      transmit_time);
 823
 824	ctl_sk->sk_mark = 0;
 825	__TCP_INC_STATS(net, TCP_MIB_OUTSEGS);
 826	__TCP_INC_STATS(net, TCP_MIB_OUTRSTS);
 827	local_bh_enable();
 828
 829#ifdef CONFIG_TCP_MD5SIG
 830out:
 831	rcu_read_unlock();
 832#endif
 833}
 834
 835/* The code following below sending ACKs in SYN-RECV and TIME-WAIT states
 836   outside socket context is ugly, certainly. What can I do?
 837 */
 838
 839static void tcp_v4_send_ack(const struct sock *sk,
 840			    struct sk_buff *skb, u32 seq, u32 ack,
 841			    u32 win, u32 tsval, u32 tsecr, int oif,
 842			    struct tcp_md5sig_key *key,
 843			    int reply_flags, u8 tos)
 844{
 845	const struct tcphdr *th = tcp_hdr(skb);
 846	struct {
 847		struct tcphdr th;
 848		__be32 opt[(TCPOLEN_TSTAMP_ALIGNED >> 2)
 849#ifdef CONFIG_TCP_MD5SIG
 850			   + (TCPOLEN_MD5SIG_ALIGNED >> 2)
 851#endif
 852			];
 853	} rep;
 854	struct net *net = sock_net(sk);
 855	struct ip_reply_arg arg;
 856	struct sock *ctl_sk;
 857	u64 transmit_time;
 858
 859	memset(&rep.th, 0, sizeof(struct tcphdr));
 860	memset(&arg, 0, sizeof(arg));
 861
 862	arg.iov[0].iov_base = (unsigned char *)&rep;
 863	arg.iov[0].iov_len  = sizeof(rep.th);
 864	if (tsecr) {
 865		rep.opt[0] = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
 866				   (TCPOPT_TIMESTAMP << 8) |
 867				   TCPOLEN_TIMESTAMP);
 868		rep.opt[1] = htonl(tsval);
 869		rep.opt[2] = htonl(tsecr);
 870		arg.iov[0].iov_len += TCPOLEN_TSTAMP_ALIGNED;
 871	}
 872
 873	/* Swap the send and the receive. */
 874	rep.th.dest    = th->source;
 875	rep.th.source  = th->dest;
 876	rep.th.doff    = arg.iov[0].iov_len / 4;
 877	rep.th.seq     = htonl(seq);
 878	rep.th.ack_seq = htonl(ack);
 879	rep.th.ack     = 1;
 880	rep.th.window  = htons(win);
 881
 882#ifdef CONFIG_TCP_MD5SIG
 883	if (key) {
 884		int offset = (tsecr) ? 3 : 0;
 885
 886		rep.opt[offset++] = htonl((TCPOPT_NOP << 24) |
 887					  (TCPOPT_NOP << 16) |
 888					  (TCPOPT_MD5SIG << 8) |
 889					  TCPOLEN_MD5SIG);
 890		arg.iov[0].iov_len += TCPOLEN_MD5SIG_ALIGNED;
 891		rep.th.doff = arg.iov[0].iov_len/4;
 892
 893		tcp_v4_md5_hash_hdr((__u8 *) &rep.opt[offset],
 894				    key, ip_hdr(skb)->saddr,
 895				    ip_hdr(skb)->daddr, &rep.th);
 896	}
 897#endif
 898	arg.flags = reply_flags;
 899	arg.csum = csum_tcpudp_nofold(ip_hdr(skb)->daddr,
 900				      ip_hdr(skb)->saddr, /* XXX */
 901				      arg.iov[0].iov_len, IPPROTO_TCP, 0);
 902	arg.csumoffset = offsetof(struct tcphdr, check) / 2;
 903	if (oif)
 904		arg.bound_dev_if = oif;
 905	arg.tos = tos;
 906	arg.uid = sock_net_uid(net, sk_fullsock(sk) ? sk : NULL);
 907	local_bh_disable();
 908	ctl_sk = this_cpu_read(*net->ipv4.tcp_sk);
 909	ctl_sk->sk_mark = (sk->sk_state == TCP_TIME_WAIT) ?
 910			   inet_twsk(sk)->tw_mark : sk->sk_mark;
 911	ctl_sk->sk_priority = (sk->sk_state == TCP_TIME_WAIT) ?
 912			   inet_twsk(sk)->tw_priority : sk->sk_priority;
 913	transmit_time = tcp_transmit_time(sk);
 914	ip_send_unicast_reply(ctl_sk,
 915			      skb, &TCP_SKB_CB(skb)->header.h4.opt,
 916			      ip_hdr(skb)->saddr, ip_hdr(skb)->daddr,
 917			      &arg, arg.iov[0].iov_len,
 918			      transmit_time);
 919
 920	ctl_sk->sk_mark = 0;
 921	__TCP_INC_STATS(net, TCP_MIB_OUTSEGS);
 922	local_bh_enable();
 923}
 924
 925static void tcp_v4_timewait_ack(struct sock *sk, struct sk_buff *skb)
 926{
 927	struct inet_timewait_sock *tw = inet_twsk(sk);
 928	struct tcp_timewait_sock *tcptw = tcp_twsk(sk);
 929
 930	tcp_v4_send_ack(sk, skb,
 931			tcptw->tw_snd_nxt, tcptw->tw_rcv_nxt,
 932			tcptw->tw_rcv_wnd >> tw->tw_rcv_wscale,
 933			tcp_time_stamp_raw() + tcptw->tw_ts_offset,
 934			tcptw->tw_ts_recent,
 935			tw->tw_bound_dev_if,
 936			tcp_twsk_md5_key(tcptw),
 937			tw->tw_transparent ? IP_REPLY_ARG_NOSRCCHECK : 0,
 938			tw->tw_tos
 939			);
 940
 941	inet_twsk_put(tw);
 942}
 943
 944static void tcp_v4_reqsk_send_ack(const struct sock *sk, struct sk_buff *skb,
 945				  struct request_sock *req)
 946{
 947	const union tcp_md5_addr *addr;
 948	int l3index;
 949
 950	/* sk->sk_state == TCP_LISTEN -> for regular TCP_SYN_RECV
 951	 * sk->sk_state == TCP_SYN_RECV -> for Fast Open.
 952	 */
 953	u32 seq = (sk->sk_state == TCP_LISTEN) ? tcp_rsk(req)->snt_isn + 1 :
 954					     tcp_sk(sk)->snd_nxt;
 955
 956	/* RFC 7323 2.3
 957	 * The window field (SEG.WND) of every outgoing segment, with the
 958	 * exception of <SYN> segments, MUST be right-shifted by
 959	 * Rcv.Wind.Shift bits:
 960	 */
 961	addr = (union tcp_md5_addr *)&ip_hdr(skb)->saddr;
 962	l3index = tcp_v4_sdif(skb) ? inet_iif(skb) : 0;
 963	tcp_v4_send_ack(sk, skb, seq,
 964			tcp_rsk(req)->rcv_nxt,
 965			req->rsk_rcv_wnd >> inet_rsk(req)->rcv_wscale,
 966			tcp_time_stamp_raw() + tcp_rsk(req)->ts_off,
 967			req->ts_recent,
 968			0,
 969			tcp_md5_do_lookup(sk, l3index, addr, AF_INET),
 970			inet_rsk(req)->no_srccheck ? IP_REPLY_ARG_NOSRCCHECK : 0,
 971			ip_hdr(skb)->tos);
 972}
 973
 974/*
 975 *	Send a SYN-ACK after having received a SYN.
 976 *	This still operates on a request_sock only, not on a big
 977 *	socket.
 978 */
 979static int tcp_v4_send_synack(const struct sock *sk, struct dst_entry *dst,
 980			      struct flowi *fl,
 981			      struct request_sock *req,
 982			      struct tcp_fastopen_cookie *foc,
 983			      enum tcp_synack_type synack_type,
 984			      struct sk_buff *syn_skb)
 985{
 986	const struct inet_request_sock *ireq = inet_rsk(req);
 987	struct flowi4 fl4;
 988	int err = -1;
 989	struct sk_buff *skb;
 990	u8 tos;
 991
 992	/* First, grab a route. */
 993	if (!dst && (dst = inet_csk_route_req(sk, &fl4, req)) == NULL)
 994		return -1;
 995
 996	skb = tcp_make_synack(sk, dst, req, foc, synack_type, syn_skb);
 997
 998	if (skb) {
 999		__tcp_v4_send_check(skb, ireq->ir_loc_addr, ireq->ir_rmt_addr);
1000
1001		tos = sock_net(sk)->ipv4.sysctl_tcp_reflect_tos ?
1002				(tcp_rsk(req)->syn_tos & ~INET_ECN_MASK) |
1003				(inet_sk(sk)->tos & INET_ECN_MASK) :
1004				inet_sk(sk)->tos;
1005
1006		if (!INET_ECN_is_capable(tos) &&
1007		    tcp_bpf_ca_needs_ecn((struct sock *)req))
1008			tos |= INET_ECN_ECT_0;
1009
1010		rcu_read_lock();
1011		err = ip_build_and_send_pkt(skb, sk, ireq->ir_loc_addr,
1012					    ireq->ir_rmt_addr,
1013					    rcu_dereference(ireq->ireq_opt),
1014					    tos);
1015		rcu_read_unlock();
1016		err = net_xmit_eval(err);
1017	}
1018
1019	return err;
1020}
1021
1022/*
1023 *	IPv4 request_sock destructor.
1024 */
1025static void tcp_v4_reqsk_destructor(struct request_sock *req)
1026{
1027	kfree(rcu_dereference_protected(inet_rsk(req)->ireq_opt, 1));
1028}
1029
1030#ifdef CONFIG_TCP_MD5SIG
1031/*
1032 * RFC2385 MD5 checksumming requires a mapping of
1033 * IP address->MD5 Key.
1034 * We need to maintain these in the sk structure.
1035 */
1036
1037DEFINE_STATIC_KEY_FALSE(tcp_md5_needed);
1038EXPORT_SYMBOL(tcp_md5_needed);
1039
1040static bool better_md5_match(struct tcp_md5sig_key *old, struct tcp_md5sig_key *new)
1041{
1042	if (!old)
1043		return true;
1044
1045	/* l3index always overrides non-l3index */
1046	if (old->l3index && new->l3index == 0)
1047		return false;
1048	if (old->l3index == 0 && new->l3index)
1049		return true;
1050
1051	return old->prefixlen < new->prefixlen;
1052}
1053
1054/* Find the Key structure for an address.  */
1055struct tcp_md5sig_key *__tcp_md5_do_lookup(const struct sock *sk, int l3index,
1056					   const union tcp_md5_addr *addr,
1057					   int family)
1058{
1059	const struct tcp_sock *tp = tcp_sk(sk);
1060	struct tcp_md5sig_key *key;
1061	const struct tcp_md5sig_info *md5sig;
1062	__be32 mask;
1063	struct tcp_md5sig_key *best_match = NULL;
1064	bool match;
1065
1066	/* caller either holds rcu_read_lock() or socket lock */
1067	md5sig = rcu_dereference_check(tp->md5sig_info,
1068				       lockdep_sock_is_held(sk));
1069	if (!md5sig)
1070		return NULL;
1071
1072	hlist_for_each_entry_rcu(key, &md5sig->head, node,
1073				 lockdep_sock_is_held(sk)) {
1074		if (key->family != family)
1075			continue;
1076		if (key->l3index && key->l3index != l3index)
1077			continue;
1078		if (family == AF_INET) {
1079			mask = inet_make_mask(key->prefixlen);
1080			match = (key->addr.a4.s_addr & mask) ==
1081				(addr->a4.s_addr & mask);
1082#if IS_ENABLED(CONFIG_IPV6)
1083		} else if (family == AF_INET6) {
1084			match = ipv6_prefix_equal(&key->addr.a6, &addr->a6,
1085						  key->prefixlen);
1086#endif
1087		} else {
1088			match = false;
1089		}
1090
1091		if (match && better_md5_match(best_match, key))
 
1092			best_match = key;
1093	}
1094	return best_match;
1095}
1096EXPORT_SYMBOL(__tcp_md5_do_lookup);
1097
1098static struct tcp_md5sig_key *tcp_md5_do_lookup_exact(const struct sock *sk,
1099						      const union tcp_md5_addr *addr,
1100						      int family, u8 prefixlen,
1101						      int l3index)
1102{
1103	const struct tcp_sock *tp = tcp_sk(sk);
1104	struct tcp_md5sig_key *key;
1105	unsigned int size = sizeof(struct in_addr);
1106	const struct tcp_md5sig_info *md5sig;
1107
1108	/* caller either holds rcu_read_lock() or socket lock */
1109	md5sig = rcu_dereference_check(tp->md5sig_info,
1110				       lockdep_sock_is_held(sk));
1111	if (!md5sig)
1112		return NULL;
1113#if IS_ENABLED(CONFIG_IPV6)
1114	if (family == AF_INET6)
1115		size = sizeof(struct in6_addr);
1116#endif
1117	hlist_for_each_entry_rcu(key, &md5sig->head, node,
1118				 lockdep_sock_is_held(sk)) {
1119		if (key->family != family)
1120			continue;
1121		if (key->l3index != l3index)
1122			continue;
1123		if (!memcmp(&key->addr, addr, size) &&
1124		    key->prefixlen == prefixlen)
1125			return key;
1126	}
1127	return NULL;
1128}
1129
1130struct tcp_md5sig_key *tcp_v4_md5_lookup(const struct sock *sk,
1131					 const struct sock *addr_sk)
1132{
1133	const union tcp_md5_addr *addr;
1134	int l3index;
1135
1136	l3index = l3mdev_master_ifindex_by_index(sock_net(sk),
1137						 addr_sk->sk_bound_dev_if);
1138	addr = (const union tcp_md5_addr *)&addr_sk->sk_daddr;
1139	return tcp_md5_do_lookup(sk, l3index, addr, AF_INET);
1140}
1141EXPORT_SYMBOL(tcp_v4_md5_lookup);
1142
1143/* This can be called on a newly created socket, from other files */
1144int tcp_md5_do_add(struct sock *sk, const union tcp_md5_addr *addr,
1145		   int family, u8 prefixlen, int l3index,
1146		   const u8 *newkey, u8 newkeylen, gfp_t gfp)
1147{
1148	/* Add Key to the list */
1149	struct tcp_md5sig_key *key;
1150	struct tcp_sock *tp = tcp_sk(sk);
1151	struct tcp_md5sig_info *md5sig;
1152
1153	key = tcp_md5_do_lookup_exact(sk, addr, family, prefixlen, l3index);
1154	if (key) {
1155		/* Pre-existing entry - just update that one.
1156		 * Note that the key might be used concurrently.
1157		 * data_race() is telling kcsan that we do not care of
1158		 * key mismatches, since changing MD5 key on live flows
1159		 * can lead to packet drops.
1160		 */
1161		data_race(memcpy(key->key, newkey, newkeylen));
1162
1163		/* Pairs with READ_ONCE() in tcp_md5_hash_key().
1164		 * Also note that a reader could catch new key->keylen value
1165		 * but old key->key[], this is the reason we use __GFP_ZERO
1166		 * at sock_kmalloc() time below these lines.
1167		 */
1168		WRITE_ONCE(key->keylen, newkeylen);
1169
1170		return 0;
1171	}
1172
1173	md5sig = rcu_dereference_protected(tp->md5sig_info,
1174					   lockdep_sock_is_held(sk));
1175	if (!md5sig) {
1176		md5sig = kmalloc(sizeof(*md5sig), gfp);
1177		if (!md5sig)
1178			return -ENOMEM;
1179
1180		sk_nocaps_add(sk, NETIF_F_GSO_MASK);
1181		INIT_HLIST_HEAD(&md5sig->head);
1182		rcu_assign_pointer(tp->md5sig_info, md5sig);
1183	}
1184
1185	key = sock_kmalloc(sk, sizeof(*key), gfp | __GFP_ZERO);
1186	if (!key)
1187		return -ENOMEM;
1188	if (!tcp_alloc_md5sig_pool()) {
1189		sock_kfree_s(sk, key, sizeof(*key));
1190		return -ENOMEM;
1191	}
1192
1193	memcpy(key->key, newkey, newkeylen);
1194	key->keylen = newkeylen;
1195	key->family = family;
1196	key->prefixlen = prefixlen;
1197	key->l3index = l3index;
1198	memcpy(&key->addr, addr,
1199	       (family == AF_INET6) ? sizeof(struct in6_addr) :
1200				      sizeof(struct in_addr));
1201	hlist_add_head_rcu(&key->node, &md5sig->head);
1202	return 0;
1203}
1204EXPORT_SYMBOL(tcp_md5_do_add);
1205
1206int tcp_md5_do_del(struct sock *sk, const union tcp_md5_addr *addr, int family,
1207		   u8 prefixlen, int l3index)
1208{
1209	struct tcp_md5sig_key *key;
1210
1211	key = tcp_md5_do_lookup_exact(sk, addr, family, prefixlen, l3index);
1212	if (!key)
1213		return -ENOENT;
1214	hlist_del_rcu(&key->node);
1215	atomic_sub(sizeof(*key), &sk->sk_omem_alloc);
1216	kfree_rcu(key, rcu);
1217	return 0;
1218}
1219EXPORT_SYMBOL(tcp_md5_do_del);
1220
1221static void tcp_clear_md5_list(struct sock *sk)
1222{
1223	struct tcp_sock *tp = tcp_sk(sk);
1224	struct tcp_md5sig_key *key;
1225	struct hlist_node *n;
1226	struct tcp_md5sig_info *md5sig;
1227
1228	md5sig = rcu_dereference_protected(tp->md5sig_info, 1);
1229
1230	hlist_for_each_entry_safe(key, n, &md5sig->head, node) {
1231		hlist_del_rcu(&key->node);
1232		atomic_sub(sizeof(*key), &sk->sk_omem_alloc);
1233		kfree_rcu(key, rcu);
1234	}
1235}
1236
1237static int tcp_v4_parse_md5_keys(struct sock *sk, int optname,
1238				 sockptr_t optval, int optlen)
1239{
1240	struct tcp_md5sig cmd;
1241	struct sockaddr_in *sin = (struct sockaddr_in *)&cmd.tcpm_addr;
1242	const union tcp_md5_addr *addr;
1243	u8 prefixlen = 32;
1244	int l3index = 0;
1245
1246	if (optlen < sizeof(cmd))
1247		return -EINVAL;
1248
1249	if (copy_from_sockptr(&cmd, optval, sizeof(cmd)))
1250		return -EFAULT;
1251
1252	if (sin->sin_family != AF_INET)
1253		return -EINVAL;
1254
1255	if (optname == TCP_MD5SIG_EXT &&
1256	    cmd.tcpm_flags & TCP_MD5SIG_FLAG_PREFIX) {
1257		prefixlen = cmd.tcpm_prefixlen;
1258		if (prefixlen > 32)
1259			return -EINVAL;
1260	}
1261
1262	if (optname == TCP_MD5SIG_EXT &&
1263	    cmd.tcpm_flags & TCP_MD5SIG_FLAG_IFINDEX) {
1264		struct net_device *dev;
1265
1266		rcu_read_lock();
1267		dev = dev_get_by_index_rcu(sock_net(sk), cmd.tcpm_ifindex);
1268		if (dev && netif_is_l3_master(dev))
1269			l3index = dev->ifindex;
1270
1271		rcu_read_unlock();
1272
1273		/* ok to reference set/not set outside of rcu;
1274		 * right now device MUST be an L3 master
1275		 */
1276		if (!dev || !l3index)
1277			return -EINVAL;
1278	}
1279
1280	addr = (union tcp_md5_addr *)&sin->sin_addr.s_addr;
1281
1282	if (!cmd.tcpm_keylen)
1283		return tcp_md5_do_del(sk, addr, AF_INET, prefixlen, l3index);
1284
1285	if (cmd.tcpm_keylen > TCP_MD5SIG_MAXKEYLEN)
1286		return -EINVAL;
1287
1288	return tcp_md5_do_add(sk, addr, AF_INET, prefixlen, l3index,
1289			      cmd.tcpm_key, cmd.tcpm_keylen, GFP_KERNEL);
1290}
1291
1292static int tcp_v4_md5_hash_headers(struct tcp_md5sig_pool *hp,
1293				   __be32 daddr, __be32 saddr,
1294				   const struct tcphdr *th, int nbytes)
1295{
1296	struct tcp4_pseudohdr *bp;
1297	struct scatterlist sg;
1298	struct tcphdr *_th;
1299
1300	bp = hp->scratch;
1301	bp->saddr = saddr;
1302	bp->daddr = daddr;
1303	bp->pad = 0;
1304	bp->protocol = IPPROTO_TCP;
1305	bp->len = cpu_to_be16(nbytes);
1306
1307	_th = (struct tcphdr *)(bp + 1);
1308	memcpy(_th, th, sizeof(*th));
1309	_th->check = 0;
1310
1311	sg_init_one(&sg, bp, sizeof(*bp) + sizeof(*th));
1312	ahash_request_set_crypt(hp->md5_req, &sg, NULL,
1313				sizeof(*bp) + sizeof(*th));
1314	return crypto_ahash_update(hp->md5_req);
1315}
1316
1317static int tcp_v4_md5_hash_hdr(char *md5_hash, const struct tcp_md5sig_key *key,
1318			       __be32 daddr, __be32 saddr, const struct tcphdr *th)
1319{
1320	struct tcp_md5sig_pool *hp;
1321	struct ahash_request *req;
1322
1323	hp = tcp_get_md5sig_pool();
1324	if (!hp)
1325		goto clear_hash_noput;
1326	req = hp->md5_req;
1327
1328	if (crypto_ahash_init(req))
1329		goto clear_hash;
1330	if (tcp_v4_md5_hash_headers(hp, daddr, saddr, th, th->doff << 2))
1331		goto clear_hash;
1332	if (tcp_md5_hash_key(hp, key))
1333		goto clear_hash;
1334	ahash_request_set_crypt(req, NULL, md5_hash, 0);
1335	if (crypto_ahash_final(req))
1336		goto clear_hash;
1337
1338	tcp_put_md5sig_pool();
1339	return 0;
1340
1341clear_hash:
1342	tcp_put_md5sig_pool();
1343clear_hash_noput:
1344	memset(md5_hash, 0, 16);
1345	return 1;
1346}
1347
1348int tcp_v4_md5_hash_skb(char *md5_hash, const struct tcp_md5sig_key *key,
1349			const struct sock *sk,
1350			const struct sk_buff *skb)
1351{
1352	struct tcp_md5sig_pool *hp;
1353	struct ahash_request *req;
1354	const struct tcphdr *th = tcp_hdr(skb);
1355	__be32 saddr, daddr;
1356
1357	if (sk) { /* valid for establish/request sockets */
1358		saddr = sk->sk_rcv_saddr;
1359		daddr = sk->sk_daddr;
1360	} else {
1361		const struct iphdr *iph = ip_hdr(skb);
1362		saddr = iph->saddr;
1363		daddr = iph->daddr;
1364	}
1365
1366	hp = tcp_get_md5sig_pool();
1367	if (!hp)
1368		goto clear_hash_noput;
1369	req = hp->md5_req;
1370
1371	if (crypto_ahash_init(req))
1372		goto clear_hash;
1373
1374	if (tcp_v4_md5_hash_headers(hp, daddr, saddr, th, skb->len))
1375		goto clear_hash;
1376	if (tcp_md5_hash_skb_data(hp, skb, th->doff << 2))
1377		goto clear_hash;
1378	if (tcp_md5_hash_key(hp, key))
1379		goto clear_hash;
1380	ahash_request_set_crypt(req, NULL, md5_hash, 0);
1381	if (crypto_ahash_final(req))
1382		goto clear_hash;
1383
1384	tcp_put_md5sig_pool();
1385	return 0;
1386
1387clear_hash:
1388	tcp_put_md5sig_pool();
1389clear_hash_noput:
1390	memset(md5_hash, 0, 16);
1391	return 1;
1392}
1393EXPORT_SYMBOL(tcp_v4_md5_hash_skb);
1394
1395#endif
1396
1397/* Called with rcu_read_lock() */
1398static bool tcp_v4_inbound_md5_hash(const struct sock *sk,
1399				    const struct sk_buff *skb,
1400				    int dif, int sdif)
1401{
1402#ifdef CONFIG_TCP_MD5SIG
1403	/*
1404	 * This gets called for each TCP segment that arrives
1405	 * so we want to be efficient.
1406	 * We have 3 drop cases:
1407	 * o No MD5 hash and one expected.
1408	 * o MD5 hash and we're not expecting one.
1409	 * o MD5 hash and its wrong.
1410	 */
1411	const __u8 *hash_location = NULL;
1412	struct tcp_md5sig_key *hash_expected;
1413	const struct iphdr *iph = ip_hdr(skb);
1414	const struct tcphdr *th = tcp_hdr(skb);
1415	const union tcp_md5_addr *addr;
1416	unsigned char newhash[16];
1417	int genhash, l3index;
1418
1419	/* sdif set, means packet ingressed via a device
1420	 * in an L3 domain and dif is set to the l3mdev
1421	 */
1422	l3index = sdif ? dif : 0;
1423
1424	addr = (union tcp_md5_addr *)&iph->saddr;
1425	hash_expected = tcp_md5_do_lookup(sk, l3index, addr, AF_INET);
1426	hash_location = tcp_parse_md5sig_option(th);
1427
1428	/* We've parsed the options - do we have a hash? */
1429	if (!hash_expected && !hash_location)
1430		return false;
1431
1432	if (hash_expected && !hash_location) {
1433		NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPMD5NOTFOUND);
1434		return true;
1435	}
1436
1437	if (!hash_expected && hash_location) {
1438		NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPMD5UNEXPECTED);
1439		return true;
1440	}
1441
1442	/* Okay, so this is hash_expected and hash_location -
1443	 * so we need to calculate the checksum.
1444	 */
1445	genhash = tcp_v4_md5_hash_skb(newhash,
1446				      hash_expected,
1447				      NULL, skb);
1448
1449	if (genhash || memcmp(hash_location, newhash, 16) != 0) {
1450		NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPMD5FAILURE);
1451		net_info_ratelimited("MD5 Hash failed for (%pI4, %d)->(%pI4, %d)%s L3 index %d\n",
1452				     &iph->saddr, ntohs(th->source),
1453				     &iph->daddr, ntohs(th->dest),
1454				     genhash ? " tcp_v4_calc_md5_hash failed"
1455				     : "", l3index);
1456		return true;
1457	}
1458	return false;
1459#endif
1460	return false;
1461}
1462
1463static void tcp_v4_init_req(struct request_sock *req,
1464			    const struct sock *sk_listener,
1465			    struct sk_buff *skb)
1466{
1467	struct inet_request_sock *ireq = inet_rsk(req);
1468	struct net *net = sock_net(sk_listener);
1469
1470	sk_rcv_saddr_set(req_to_sk(req), ip_hdr(skb)->daddr);
1471	sk_daddr_set(req_to_sk(req), ip_hdr(skb)->saddr);
1472	RCU_INIT_POINTER(ireq->ireq_opt, tcp_v4_save_options(net, skb));
1473}
1474
1475static struct dst_entry *tcp_v4_route_req(const struct sock *sk,
1476					  struct sk_buff *skb,
1477					  struct flowi *fl,
1478					  struct request_sock *req)
1479{
1480	tcp_v4_init_req(req, sk, skb);
1481
1482	if (security_inet_conn_request(sk, skb, req))
1483		return NULL;
1484
1485	return inet_csk_route_req(sk, &fl->u.ip4, req);
1486}
1487
1488struct request_sock_ops tcp_request_sock_ops __read_mostly = {
1489	.family		=	PF_INET,
1490	.obj_size	=	sizeof(struct tcp_request_sock),
1491	.rtx_syn_ack	=	tcp_rtx_synack,
1492	.send_ack	=	tcp_v4_reqsk_send_ack,
1493	.destructor	=	tcp_v4_reqsk_destructor,
1494	.send_reset	=	tcp_v4_send_reset,
1495	.syn_ack_timeout =	tcp_syn_ack_timeout,
1496};
1497
1498const struct tcp_request_sock_ops tcp_request_sock_ipv4_ops = {
1499	.mss_clamp	=	TCP_MSS_DEFAULT,
1500#ifdef CONFIG_TCP_MD5SIG
1501	.req_md5_lookup	=	tcp_v4_md5_lookup,
1502	.calc_md5_hash	=	tcp_v4_md5_hash_skb,
1503#endif
 
1504#ifdef CONFIG_SYN_COOKIES
1505	.cookie_init_seq =	cookie_v4_init_sequence,
1506#endif
1507	.route_req	=	tcp_v4_route_req,
1508	.init_seq	=	tcp_v4_init_seq,
1509	.init_ts_off	=	tcp_v4_init_ts_off,
1510	.send_synack	=	tcp_v4_send_synack,
1511};
1512
1513int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb)
1514{
1515	/* Never answer to SYNs send to broadcast or multicast */
1516	if (skb_rtable(skb)->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))
1517		goto drop;
1518
1519	return tcp_conn_request(&tcp_request_sock_ops,
1520				&tcp_request_sock_ipv4_ops, sk, skb);
1521
1522drop:
1523	tcp_listendrop(sk);
1524	return 0;
1525}
1526EXPORT_SYMBOL(tcp_v4_conn_request);
1527
1528
1529/*
1530 * The three way handshake has completed - we got a valid synack -
1531 * now create the new socket.
1532 */
1533struct sock *tcp_v4_syn_recv_sock(const struct sock *sk, struct sk_buff *skb,
1534				  struct request_sock *req,
1535				  struct dst_entry *dst,
1536				  struct request_sock *req_unhash,
1537				  bool *own_req)
1538{
1539	struct inet_request_sock *ireq;
1540	bool found_dup_sk = false;
1541	struct inet_sock *newinet;
1542	struct tcp_sock *newtp;
1543	struct sock *newsk;
1544#ifdef CONFIG_TCP_MD5SIG
1545	const union tcp_md5_addr *addr;
1546	struct tcp_md5sig_key *key;
1547	int l3index;
1548#endif
1549	struct ip_options_rcu *inet_opt;
1550
1551	if (sk_acceptq_is_full(sk))
1552		goto exit_overflow;
1553
1554	newsk = tcp_create_openreq_child(sk, req, skb);
1555	if (!newsk)
1556		goto exit_nonewsk;
1557
1558	newsk->sk_gso_type = SKB_GSO_TCPV4;
1559	inet_sk_rx_dst_set(newsk, skb);
1560
1561	newtp		      = tcp_sk(newsk);
1562	newinet		      = inet_sk(newsk);
1563	ireq		      = inet_rsk(req);
1564	sk_daddr_set(newsk, ireq->ir_rmt_addr);
1565	sk_rcv_saddr_set(newsk, ireq->ir_loc_addr);
1566	newsk->sk_bound_dev_if = ireq->ir_iif;
1567	newinet->inet_saddr   = ireq->ir_loc_addr;
1568	inet_opt	      = rcu_dereference(ireq->ireq_opt);
1569	RCU_INIT_POINTER(newinet->inet_opt, inet_opt);
1570	newinet->mc_index     = inet_iif(skb);
1571	newinet->mc_ttl	      = ip_hdr(skb)->ttl;
1572	newinet->rcv_tos      = ip_hdr(skb)->tos;
1573	inet_csk(newsk)->icsk_ext_hdr_len = 0;
1574	if (inet_opt)
1575		inet_csk(newsk)->icsk_ext_hdr_len = inet_opt->opt.optlen;
1576	newinet->inet_id = prandom_u32();
1577
1578	/* Set ToS of the new socket based upon the value of incoming SYN.
1579	 * ECT bits are set later in tcp_init_transfer().
1580	 */
1581	if (sock_net(sk)->ipv4.sysctl_tcp_reflect_tos)
1582		newinet->tos = tcp_rsk(req)->syn_tos & ~INET_ECN_MASK;
1583
1584	if (!dst) {
1585		dst = inet_csk_route_child_sock(sk, newsk, req);
1586		if (!dst)
1587			goto put_and_exit;
1588	} else {
1589		/* syncookie case : see end of cookie_v4_check() */
1590	}
1591	sk_setup_caps(newsk, dst);
1592
1593	tcp_ca_openreq_child(newsk, dst);
1594
1595	tcp_sync_mss(newsk, dst_mtu(dst));
1596	newtp->advmss = tcp_mss_clamp(tcp_sk(sk), dst_metric_advmss(dst));
1597
1598	tcp_initialize_rcv_mss(newsk);
1599
1600#ifdef CONFIG_TCP_MD5SIG
1601	l3index = l3mdev_master_ifindex_by_index(sock_net(sk), ireq->ir_iif);
1602	/* Copy over the MD5 key from the original socket */
1603	addr = (union tcp_md5_addr *)&newinet->inet_daddr;
1604	key = tcp_md5_do_lookup(sk, l3index, addr, AF_INET);
1605	if (key) {
1606		/*
1607		 * We're using one, so create a matching key
1608		 * on the newsk structure. If we fail to get
1609		 * memory, then we end up not copying the key
1610		 * across. Shucks.
1611		 */
1612		tcp_md5_do_add(newsk, addr, AF_INET, 32, l3index,
1613			       key->key, key->keylen, GFP_ATOMIC);
1614		sk_nocaps_add(newsk, NETIF_F_GSO_MASK);
1615	}
1616#endif
1617
1618	if (__inet_inherit_port(sk, newsk) < 0)
1619		goto put_and_exit;
1620	*own_req = inet_ehash_nolisten(newsk, req_to_sk(req_unhash),
1621				       &found_dup_sk);
1622	if (likely(*own_req)) {
1623		tcp_move_syn(newtp, req);
1624		ireq->ireq_opt = NULL;
1625	} else {
1626		newinet->inet_opt = NULL;
1627
1628		if (!req_unhash && found_dup_sk) {
1629			/* This code path should only be executed in the
1630			 * syncookie case only
1631			 */
1632			bh_unlock_sock(newsk);
1633			sock_put(newsk);
1634			newsk = NULL;
1635		}
1636	}
1637	return newsk;
1638
1639exit_overflow:
1640	NET_INC_STATS(sock_net(sk), LINUX_MIB_LISTENOVERFLOWS);
1641exit_nonewsk:
1642	dst_release(dst);
1643exit:
1644	tcp_listendrop(sk);
1645	return NULL;
1646put_and_exit:
1647	newinet->inet_opt = NULL;
1648	inet_csk_prepare_forced_close(newsk);
1649	tcp_done(newsk);
1650	goto exit;
1651}
1652EXPORT_SYMBOL(tcp_v4_syn_recv_sock);
1653
1654static struct sock *tcp_v4_cookie_check(struct sock *sk, struct sk_buff *skb)
1655{
1656#ifdef CONFIG_SYN_COOKIES
1657	const struct tcphdr *th = tcp_hdr(skb);
1658
1659	if (!th->syn)
1660		sk = cookie_v4_check(sk, skb);
1661#endif
1662	return sk;
1663}
1664
1665u16 tcp_v4_get_syncookie(struct sock *sk, struct iphdr *iph,
1666			 struct tcphdr *th, u32 *cookie)
1667{
1668	u16 mss = 0;
1669#ifdef CONFIG_SYN_COOKIES
1670	mss = tcp_get_syncookie_mss(&tcp_request_sock_ops,
1671				    &tcp_request_sock_ipv4_ops, sk, th);
1672	if (mss) {
1673		*cookie = __cookie_v4_init_sequence(iph, th, &mss);
1674		tcp_synq_overflow(sk);
1675	}
1676#endif
1677	return mss;
1678}
1679
1680INDIRECT_CALLABLE_DECLARE(struct dst_entry *ipv4_dst_check(struct dst_entry *,
1681							   u32));
1682/* The socket must have it's spinlock held when we get
1683 * here, unless it is a TCP_LISTEN socket.
1684 *
1685 * We have a potential double-lock case here, so even when
1686 * doing backlog processing we use the BH locking scheme.
1687 * This is because we cannot sleep with the original spinlock
1688 * held.
1689 */
1690int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb)
1691{
1692	struct sock *rsk;
1693
1694	if (sk->sk_state == TCP_ESTABLISHED) { /* Fast path */
1695		struct dst_entry *dst = sk->sk_rx_dst;
1696
1697		sock_rps_save_rxhash(sk, skb);
1698		sk_mark_napi_id(sk, skb);
1699		if (dst) {
1700			if (inet_sk(sk)->rx_dst_ifindex != skb->skb_iif ||
1701			    !INDIRECT_CALL_1(dst->ops->check, ipv4_dst_check,
1702					     dst, 0)) {
1703				dst_release(dst);
1704				sk->sk_rx_dst = NULL;
1705			}
1706		}
1707		tcp_rcv_established(sk, skb);
1708		return 0;
1709	}
1710
1711	if (tcp_checksum_complete(skb))
1712		goto csum_err;
1713
1714	if (sk->sk_state == TCP_LISTEN) {
1715		struct sock *nsk = tcp_v4_cookie_check(sk, skb);
1716
1717		if (!nsk)
1718			goto discard;
1719		if (nsk != sk) {
1720			if (tcp_child_process(sk, nsk, skb)) {
1721				rsk = nsk;
1722				goto reset;
1723			}
1724			return 0;
1725		}
1726	} else
1727		sock_rps_save_rxhash(sk, skb);
1728
1729	if (tcp_rcv_state_process(sk, skb)) {
1730		rsk = sk;
1731		goto reset;
1732	}
1733	return 0;
1734
1735reset:
1736	tcp_v4_send_reset(rsk, skb);
1737discard:
1738	kfree_skb(skb);
1739	/* Be careful here. If this function gets more complicated and
1740	 * gcc suffers from register pressure on the x86, sk (in %ebx)
1741	 * might be destroyed here. This current version compiles correctly,
1742	 * but you have been warned.
1743	 */
1744	return 0;
1745
1746csum_err:
1747	trace_tcp_bad_csum(skb);
1748	TCP_INC_STATS(sock_net(sk), TCP_MIB_CSUMERRORS);
1749	TCP_INC_STATS(sock_net(sk), TCP_MIB_INERRS);
1750	goto discard;
1751}
1752EXPORT_SYMBOL(tcp_v4_do_rcv);
1753
1754int tcp_v4_early_demux(struct sk_buff *skb)
1755{
1756	const struct iphdr *iph;
1757	const struct tcphdr *th;
1758	struct sock *sk;
1759
1760	if (skb->pkt_type != PACKET_HOST)
1761		return 0;
1762
1763	if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct tcphdr)))
1764		return 0;
1765
1766	iph = ip_hdr(skb);
1767	th = tcp_hdr(skb);
1768
1769	if (th->doff < sizeof(struct tcphdr) / 4)
1770		return 0;
1771
1772	sk = __inet_lookup_established(dev_net(skb->dev), &tcp_hashinfo,
1773				       iph->saddr, th->source,
1774				       iph->daddr, ntohs(th->dest),
1775				       skb->skb_iif, inet_sdif(skb));
1776	if (sk) {
1777		skb->sk = sk;
1778		skb->destructor = sock_edemux;
1779		if (sk_fullsock(sk)) {
1780			struct dst_entry *dst = READ_ONCE(sk->sk_rx_dst);
1781
1782			if (dst)
1783				dst = dst_check(dst, 0);
1784			if (dst &&
1785			    inet_sk(sk)->rx_dst_ifindex == skb->skb_iif)
1786				skb_dst_set_noref(skb, dst);
1787		}
1788	}
1789	return 0;
1790}
1791
1792bool tcp_add_backlog(struct sock *sk, struct sk_buff *skb)
1793{
1794	u32 limit = READ_ONCE(sk->sk_rcvbuf) + READ_ONCE(sk->sk_sndbuf);
1795	u32 tail_gso_size, tail_gso_segs;
1796	struct skb_shared_info *shinfo;
1797	const struct tcphdr *th;
1798	struct tcphdr *thtail;
1799	struct sk_buff *tail;
1800	unsigned int hdrlen;
1801	bool fragstolen;
1802	u32 gso_segs;
1803	u32 gso_size;
1804	int delta;
1805
1806	/* In case all data was pulled from skb frags (in __pskb_pull_tail()),
1807	 * we can fix skb->truesize to its real value to avoid future drops.
1808	 * This is valid because skb is not yet charged to the socket.
1809	 * It has been noticed pure SACK packets were sometimes dropped
1810	 * (if cooked by drivers without copybreak feature).
1811	 */
1812	skb_condense(skb);
1813
1814	skb_dst_drop(skb);
1815
1816	if (unlikely(tcp_checksum_complete(skb))) {
1817		bh_unlock_sock(sk);
1818		trace_tcp_bad_csum(skb);
1819		__TCP_INC_STATS(sock_net(sk), TCP_MIB_CSUMERRORS);
1820		__TCP_INC_STATS(sock_net(sk), TCP_MIB_INERRS);
1821		return true;
1822	}
1823
1824	/* Attempt coalescing to last skb in backlog, even if we are
1825	 * above the limits.
1826	 * This is okay because skb capacity is limited to MAX_SKB_FRAGS.
1827	 */
1828	th = (const struct tcphdr *)skb->data;
1829	hdrlen = th->doff * 4;
 
 
 
 
 
 
 
1830
1831	tail = sk->sk_backlog.tail;
1832	if (!tail)
1833		goto no_coalesce;
1834	thtail = (struct tcphdr *)tail->data;
1835
1836	if (TCP_SKB_CB(tail)->end_seq != TCP_SKB_CB(skb)->seq ||
1837	    TCP_SKB_CB(tail)->ip_dsfield != TCP_SKB_CB(skb)->ip_dsfield ||
1838	    ((TCP_SKB_CB(tail)->tcp_flags |
1839	      TCP_SKB_CB(skb)->tcp_flags) & (TCPHDR_SYN | TCPHDR_RST | TCPHDR_URG)) ||
1840	    !((TCP_SKB_CB(tail)->tcp_flags &
1841	      TCP_SKB_CB(skb)->tcp_flags) & TCPHDR_ACK) ||
1842	    ((TCP_SKB_CB(tail)->tcp_flags ^
1843	      TCP_SKB_CB(skb)->tcp_flags) & (TCPHDR_ECE | TCPHDR_CWR)) ||
1844#ifdef CONFIG_TLS_DEVICE
1845	    tail->decrypted != skb->decrypted ||
1846#endif
1847	    thtail->doff != th->doff ||
1848	    memcmp(thtail + 1, th + 1, hdrlen - sizeof(*th)))
1849		goto no_coalesce;
1850
1851	__skb_pull(skb, hdrlen);
1852
1853	shinfo = skb_shinfo(skb);
1854	gso_size = shinfo->gso_size ?: skb->len;
1855	gso_segs = shinfo->gso_segs ?: 1;
1856
1857	shinfo = skb_shinfo(tail);
1858	tail_gso_size = shinfo->gso_size ?: (tail->len - hdrlen);
1859	tail_gso_segs = shinfo->gso_segs ?: 1;
1860
1861	if (skb_try_coalesce(tail, skb, &fragstolen, &delta)) {
1862		TCP_SKB_CB(tail)->end_seq = TCP_SKB_CB(skb)->end_seq;
1863
1864		if (likely(!before(TCP_SKB_CB(skb)->ack_seq, TCP_SKB_CB(tail)->ack_seq))) {
1865			TCP_SKB_CB(tail)->ack_seq = TCP_SKB_CB(skb)->ack_seq;
1866			thtail->window = th->window;
1867		}
1868
1869		/* We have to update both TCP_SKB_CB(tail)->tcp_flags and
1870		 * thtail->fin, so that the fast path in tcp_rcv_established()
1871		 * is not entered if we append a packet with a FIN.
1872		 * SYN, RST, URG are not present.
1873		 * ACK is set on both packets.
1874		 * PSH : we do not really care in TCP stack,
1875		 *       at least for 'GRO' packets.
1876		 */
1877		thtail->fin |= th->fin;
1878		TCP_SKB_CB(tail)->tcp_flags |= TCP_SKB_CB(skb)->tcp_flags;
1879
1880		if (TCP_SKB_CB(skb)->has_rxtstamp) {
1881			TCP_SKB_CB(tail)->has_rxtstamp = true;
1882			tail->tstamp = skb->tstamp;
1883			skb_hwtstamps(tail)->hwtstamp = skb_hwtstamps(skb)->hwtstamp;
1884		}
1885
1886		/* Not as strict as GRO. We only need to carry mss max value */
1887		shinfo->gso_size = max(gso_size, tail_gso_size);
1888		shinfo->gso_segs = min_t(u32, gso_segs + tail_gso_segs, 0xFFFF);
 
 
 
1889
1890		sk->sk_backlog.len += delta;
1891		__NET_INC_STATS(sock_net(sk),
1892				LINUX_MIB_TCPBACKLOGCOALESCE);
1893		kfree_skb_partial(skb, fragstolen);
1894		return false;
1895	}
1896	__skb_push(skb, hdrlen);
1897
1898no_coalesce:
1899	/* Only socket owner can try to collapse/prune rx queues
1900	 * to reduce memory overhead, so add a little headroom here.
1901	 * Few sockets backlog are possibly concurrently non empty.
1902	 */
1903	limit += 64*1024;
1904
1905	if (unlikely(sk_add_backlog(sk, skb, limit))) {
1906		bh_unlock_sock(sk);
1907		__NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPBACKLOGDROP);
1908		return true;
1909	}
1910	return false;
1911}
1912EXPORT_SYMBOL(tcp_add_backlog);
1913
1914int tcp_filter(struct sock *sk, struct sk_buff *skb)
1915{
1916	struct tcphdr *th = (struct tcphdr *)skb->data;
1917
1918	return sk_filter_trim_cap(sk, skb, th->doff * 4);
1919}
1920EXPORT_SYMBOL(tcp_filter);
1921
1922static void tcp_v4_restore_cb(struct sk_buff *skb)
1923{
1924	memmove(IPCB(skb), &TCP_SKB_CB(skb)->header.h4,
1925		sizeof(struct inet_skb_parm));
1926}
1927
1928static void tcp_v4_fill_cb(struct sk_buff *skb, const struct iphdr *iph,
1929			   const struct tcphdr *th)
1930{
1931	/* This is tricky : We move IPCB at its correct location into TCP_SKB_CB()
1932	 * barrier() makes sure compiler wont play fool^Waliasing games.
1933	 */
1934	memmove(&TCP_SKB_CB(skb)->header.h4, IPCB(skb),
1935		sizeof(struct inet_skb_parm));
1936	barrier();
1937
1938	TCP_SKB_CB(skb)->seq = ntohl(th->seq);
1939	TCP_SKB_CB(skb)->end_seq = (TCP_SKB_CB(skb)->seq + th->syn + th->fin +
1940				    skb->len - th->doff * 4);
1941	TCP_SKB_CB(skb)->ack_seq = ntohl(th->ack_seq);
1942	TCP_SKB_CB(skb)->tcp_flags = tcp_flag_byte(th);
1943	TCP_SKB_CB(skb)->tcp_tw_isn = 0;
1944	TCP_SKB_CB(skb)->ip_dsfield = ipv4_get_dsfield(iph);
1945	TCP_SKB_CB(skb)->sacked	 = 0;
1946	TCP_SKB_CB(skb)->has_rxtstamp =
1947			skb->tstamp || skb_hwtstamps(skb)->hwtstamp;
1948}
1949
1950/*
1951 *	From tcp_input.c
1952 */
1953
1954int tcp_v4_rcv(struct sk_buff *skb)
1955{
1956	struct net *net = dev_net(skb->dev);
1957	struct sk_buff *skb_to_free;
1958	int sdif = inet_sdif(skb);
1959	int dif = inet_iif(skb);
1960	const struct iphdr *iph;
1961	const struct tcphdr *th;
1962	bool refcounted;
1963	struct sock *sk;
1964	int ret;
1965
1966	if (skb->pkt_type != PACKET_HOST)
1967		goto discard_it;
1968
1969	/* Count it even if it's bad */
1970	__TCP_INC_STATS(net, TCP_MIB_INSEGS);
1971
1972	if (!pskb_may_pull(skb, sizeof(struct tcphdr)))
1973		goto discard_it;
1974
1975	th = (const struct tcphdr *)skb->data;
1976
1977	if (unlikely(th->doff < sizeof(struct tcphdr) / 4))
1978		goto bad_packet;
1979	if (!pskb_may_pull(skb, th->doff * 4))
1980		goto discard_it;
1981
1982	/* An explanation is required here, I think.
1983	 * Packet length and doff are validated by header prediction,
1984	 * provided case of th->doff==0 is eliminated.
1985	 * So, we defer the checks. */
1986
1987	if (skb_checksum_init(skb, IPPROTO_TCP, inet_compute_pseudo))
1988		goto csum_error;
1989
1990	th = (const struct tcphdr *)skb->data;
1991	iph = ip_hdr(skb);
1992lookup:
1993	sk = __inet_lookup_skb(&tcp_hashinfo, skb, __tcp_hdrlen(th), th->source,
1994			       th->dest, sdif, &refcounted);
1995	if (!sk)
1996		goto no_tcp_socket;
1997
1998process:
1999	if (sk->sk_state == TCP_TIME_WAIT)
2000		goto do_time_wait;
2001
2002	if (sk->sk_state == TCP_NEW_SYN_RECV) {
2003		struct request_sock *req = inet_reqsk(sk);
2004		bool req_stolen = false;
2005		struct sock *nsk;
2006
2007		sk = req->rsk_listener;
2008		if (unlikely(tcp_v4_inbound_md5_hash(sk, skb, dif, sdif))) {
2009			sk_drops_add(sk, skb);
2010			reqsk_put(req);
2011			goto discard_it;
2012		}
2013		if (tcp_checksum_complete(skb)) {
2014			reqsk_put(req);
2015			goto csum_error;
2016		}
2017		if (unlikely(sk->sk_state != TCP_LISTEN)) {
2018			nsk = reuseport_migrate_sock(sk, req_to_sk(req), skb);
2019			if (!nsk) {
2020				inet_csk_reqsk_queue_drop_and_put(sk, req);
2021				goto lookup;
2022			}
2023			sk = nsk;
2024			/* reuseport_migrate_sock() has already held one sk_refcnt
2025			 * before returning.
2026			 */
2027		} else {
2028			/* We own a reference on the listener, increase it again
2029			 * as we might lose it too soon.
2030			 */
2031			sock_hold(sk);
2032		}
 
 
 
 
2033		refcounted = true;
2034		nsk = NULL;
2035		if (!tcp_filter(sk, skb)) {
2036			th = (const struct tcphdr *)skb->data;
2037			iph = ip_hdr(skb);
2038			tcp_v4_fill_cb(skb, iph, th);
2039			nsk = tcp_check_req(sk, skb, req, false, &req_stolen);
2040		}
2041		if (!nsk) {
2042			reqsk_put(req);
2043			if (req_stolen) {
2044				/* Another cpu got exclusive access to req
2045				 * and created a full blown socket.
2046				 * Try to feed this packet to this socket
2047				 * instead of discarding it.
2048				 */
2049				tcp_v4_restore_cb(skb);
2050				sock_put(sk);
2051				goto lookup;
2052			}
2053			goto discard_and_relse;
2054		}
2055		if (nsk == sk) {
2056			reqsk_put(req);
2057			tcp_v4_restore_cb(skb);
2058		} else if (tcp_child_process(sk, nsk, skb)) {
2059			tcp_v4_send_reset(nsk, skb);
2060			goto discard_and_relse;
2061		} else {
2062			sock_put(sk);
2063			return 0;
2064		}
2065	}
2066	if (unlikely(iph->ttl < inet_sk(sk)->min_ttl)) {
2067		__NET_INC_STATS(net, LINUX_MIB_TCPMINTTLDROP);
2068		goto discard_and_relse;
2069	}
2070
2071	if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
2072		goto discard_and_relse;
2073
2074	if (tcp_v4_inbound_md5_hash(sk, skb, dif, sdif))
2075		goto discard_and_relse;
2076
2077	nf_reset_ct(skb);
2078
2079	if (tcp_filter(sk, skb))
2080		goto discard_and_relse;
2081	th = (const struct tcphdr *)skb->data;
2082	iph = ip_hdr(skb);
2083	tcp_v4_fill_cb(skb, iph, th);
2084
2085	skb->dev = NULL;
2086
2087	if (sk->sk_state == TCP_LISTEN) {
2088		ret = tcp_v4_do_rcv(sk, skb);
2089		goto put_and_return;
2090	}
2091
2092	sk_incoming_cpu_update(sk);
2093
2094	bh_lock_sock_nested(sk);
2095	tcp_segs_in(tcp_sk(sk), skb);
2096	ret = 0;
2097	if (!sock_owned_by_user(sk)) {
2098		skb_to_free = sk->sk_rx_skb_cache;
2099		sk->sk_rx_skb_cache = NULL;
2100		ret = tcp_v4_do_rcv(sk, skb);
2101	} else {
2102		if (tcp_add_backlog(sk, skb))
2103			goto discard_and_relse;
2104		skb_to_free = NULL;
2105	}
2106	bh_unlock_sock(sk);
2107	if (skb_to_free)
2108		__kfree_skb(skb_to_free);
2109
2110put_and_return:
2111	if (refcounted)
2112		sock_put(sk);
2113
2114	return ret;
2115
2116no_tcp_socket:
2117	if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
2118		goto discard_it;
2119
2120	tcp_v4_fill_cb(skb, iph, th);
2121
2122	if (tcp_checksum_complete(skb)) {
2123csum_error:
2124		trace_tcp_bad_csum(skb);
2125		__TCP_INC_STATS(net, TCP_MIB_CSUMERRORS);
2126bad_packet:
2127		__TCP_INC_STATS(net, TCP_MIB_INERRS);
2128	} else {
2129		tcp_v4_send_reset(NULL, skb);
2130	}
2131
2132discard_it:
2133	/* Discard frame. */
2134	kfree_skb(skb);
2135	return 0;
2136
2137discard_and_relse:
2138	sk_drops_add(sk, skb);
2139	if (refcounted)
2140		sock_put(sk);
2141	goto discard_it;
2142
2143do_time_wait:
2144	if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) {
2145		inet_twsk_put(inet_twsk(sk));
2146		goto discard_it;
2147	}
2148
2149	tcp_v4_fill_cb(skb, iph, th);
2150
2151	if (tcp_checksum_complete(skb)) {
2152		inet_twsk_put(inet_twsk(sk));
2153		goto csum_error;
2154	}
2155	switch (tcp_timewait_state_process(inet_twsk(sk), skb, th)) {
2156	case TCP_TW_SYN: {
2157		struct sock *sk2 = inet_lookup_listener(dev_net(skb->dev),
2158							&tcp_hashinfo, skb,
2159							__tcp_hdrlen(th),
2160							iph->saddr, th->source,
2161							iph->daddr, th->dest,
2162							inet_iif(skb),
2163							sdif);
2164		if (sk2) {
2165			inet_twsk_deschedule_put(inet_twsk(sk));
2166			sk = sk2;
2167			tcp_v4_restore_cb(skb);
2168			refcounted = false;
2169			goto process;
2170		}
2171	}
2172		/* to ACK */
2173		fallthrough;
2174	case TCP_TW_ACK:
2175		tcp_v4_timewait_ack(sk, skb);
2176		break;
2177	case TCP_TW_RST:
2178		tcp_v4_send_reset(sk, skb);
2179		inet_twsk_deschedule_put(inet_twsk(sk));
2180		goto discard_it;
2181	case TCP_TW_SUCCESS:;
2182	}
2183	goto discard_it;
2184}
2185
2186static struct timewait_sock_ops tcp_timewait_sock_ops = {
2187	.twsk_obj_size	= sizeof(struct tcp_timewait_sock),
2188	.twsk_unique	= tcp_twsk_unique,
2189	.twsk_destructor= tcp_twsk_destructor,
2190};
2191
2192void inet_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb)
2193{
2194	struct dst_entry *dst = skb_dst(skb);
2195
2196	if (dst && dst_hold_safe(dst)) {
2197		sk->sk_rx_dst = dst;
2198		inet_sk(sk)->rx_dst_ifindex = skb->skb_iif;
2199	}
2200}
2201EXPORT_SYMBOL(inet_sk_rx_dst_set);
2202
2203const struct inet_connection_sock_af_ops ipv4_specific = {
2204	.queue_xmit	   = ip_queue_xmit,
2205	.send_check	   = tcp_v4_send_check,
2206	.rebuild_header	   = inet_sk_rebuild_header,
2207	.sk_rx_dst_set	   = inet_sk_rx_dst_set,
2208	.conn_request	   = tcp_v4_conn_request,
2209	.syn_recv_sock	   = tcp_v4_syn_recv_sock,
2210	.net_header_len	   = sizeof(struct iphdr),
2211	.setsockopt	   = ip_setsockopt,
2212	.getsockopt	   = ip_getsockopt,
2213	.addr2sockaddr	   = inet_csk_addr2sockaddr,
2214	.sockaddr_len	   = sizeof(struct sockaddr_in),
2215	.mtu_reduced	   = tcp_v4_mtu_reduced,
2216};
2217EXPORT_SYMBOL(ipv4_specific);
2218
2219#ifdef CONFIG_TCP_MD5SIG
2220static const struct tcp_sock_af_ops tcp_sock_ipv4_specific = {
2221	.md5_lookup		= tcp_v4_md5_lookup,
2222	.calc_md5_hash		= tcp_v4_md5_hash_skb,
2223	.md5_parse		= tcp_v4_parse_md5_keys,
2224};
2225#endif
2226
2227/* NOTE: A lot of things set to zero explicitly by call to
2228 *       sk_alloc() so need not be done here.
2229 */
2230static int tcp_v4_init_sock(struct sock *sk)
2231{
2232	struct inet_connection_sock *icsk = inet_csk(sk);
2233
2234	tcp_init_sock(sk);
2235
2236	icsk->icsk_af_ops = &ipv4_specific;
2237
2238#ifdef CONFIG_TCP_MD5SIG
2239	tcp_sk(sk)->af_specific = &tcp_sock_ipv4_specific;
2240#endif
2241
2242	return 0;
2243}
2244
2245void tcp_v4_destroy_sock(struct sock *sk)
2246{
2247	struct tcp_sock *tp = tcp_sk(sk);
2248
2249	trace_tcp_destroy_sock(sk);
2250
2251	tcp_clear_xmit_timers(sk);
2252
2253	tcp_cleanup_congestion_control(sk);
2254
2255	tcp_cleanup_ulp(sk);
2256
2257	/* Cleanup up the write buffer. */
2258	tcp_write_queue_purge(sk);
2259
2260	/* Check if we want to disable active TFO */
2261	tcp_fastopen_active_disable_ofo_check(sk);
2262
2263	/* Cleans up our, hopefully empty, out_of_order_queue. */
2264	skb_rbtree_purge(&tp->out_of_order_queue);
2265
2266#ifdef CONFIG_TCP_MD5SIG
2267	/* Clean up the MD5 key list, if any */
2268	if (tp->md5sig_info) {
2269		tcp_clear_md5_list(sk);
2270		kfree_rcu(rcu_dereference_protected(tp->md5sig_info, 1), rcu);
2271		tp->md5sig_info = NULL;
2272	}
2273#endif
2274
2275	/* Clean up a referenced TCP bind bucket. */
2276	if (inet_csk(sk)->icsk_bind_hash)
2277		inet_put_port(sk);
2278
2279	BUG_ON(rcu_access_pointer(tp->fastopen_rsk));
2280
2281	/* If socket is aborted during connect operation */
2282	tcp_free_fastopen_req(tp);
2283	tcp_fastopen_destroy_cipher(sk);
2284	tcp_saved_syn_free(tp);
2285
2286	sk_sockets_allocated_dec(sk);
2287}
2288EXPORT_SYMBOL(tcp_v4_destroy_sock);
2289
2290#ifdef CONFIG_PROC_FS
2291/* Proc filesystem TCP sock list dumping. */
2292
2293/*
2294 * Get next listener socket follow cur.  If cur is NULL, get first socket
2295 * starting from bucket given in st->bucket; when st->bucket is zero the
2296 * very first socket in the hash table is returned.
2297 */
2298static void *listening_get_next(struct seq_file *seq, void *cur)
2299{
2300	struct tcp_seq_afinfo *afinfo;
2301	struct tcp_iter_state *st = seq->private;
2302	struct net *net = seq_file_net(seq);
2303	struct inet_listen_hashbucket *ilb;
2304	struct hlist_nulls_node *node;
2305	struct sock *sk = cur;
2306
2307	if (st->bpf_seq_afinfo)
2308		afinfo = st->bpf_seq_afinfo;
2309	else
2310		afinfo = PDE_DATA(file_inode(seq->file));
2311
2312	if (!sk) {
2313get_head:
2314		ilb = &tcp_hashinfo.listening_hash[st->bucket];
2315		spin_lock(&ilb->lock);
2316		sk = sk_nulls_head(&ilb->nulls_head);
2317		st->offset = 0;
2318		goto get_sk;
2319	}
2320	ilb = &tcp_hashinfo.listening_hash[st->bucket];
2321	++st->num;
2322	++st->offset;
2323
2324	sk = sk_nulls_next(sk);
2325get_sk:
2326	sk_nulls_for_each_from(sk, node) {
2327		if (!net_eq(sock_net(sk), net))
2328			continue;
2329		if (afinfo->family == AF_UNSPEC ||
2330		    sk->sk_family == afinfo->family)
2331			return sk;
2332	}
2333	spin_unlock(&ilb->lock);
2334	st->offset = 0;
2335	if (++st->bucket < INET_LHTABLE_SIZE)
2336		goto get_head;
2337	return NULL;
2338}
2339
2340static void *listening_get_idx(struct seq_file *seq, loff_t *pos)
2341{
2342	struct tcp_iter_state *st = seq->private;
2343	void *rc;
2344
2345	st->bucket = 0;
2346	st->offset = 0;
2347	rc = listening_get_next(seq, NULL);
2348
2349	while (rc && *pos) {
2350		rc = listening_get_next(seq, rc);
2351		--*pos;
2352	}
2353	return rc;
2354}
2355
2356static inline bool empty_bucket(const struct tcp_iter_state *st)
2357{
2358	return hlist_nulls_empty(&tcp_hashinfo.ehash[st->bucket].chain);
2359}
2360
2361/*
2362 * Get first established socket starting from bucket given in st->bucket.
2363 * If st->bucket is zero, the very first socket in the hash is returned.
2364 */
2365static void *established_get_first(struct seq_file *seq)
2366{
2367	struct tcp_seq_afinfo *afinfo;
2368	struct tcp_iter_state *st = seq->private;
2369	struct net *net = seq_file_net(seq);
2370	void *rc = NULL;
2371
2372	if (st->bpf_seq_afinfo)
2373		afinfo = st->bpf_seq_afinfo;
2374	else
2375		afinfo = PDE_DATA(file_inode(seq->file));
2376
2377	st->offset = 0;
2378	for (; st->bucket <= tcp_hashinfo.ehash_mask; ++st->bucket) {
2379		struct sock *sk;
2380		struct hlist_nulls_node *node;
2381		spinlock_t *lock = inet_ehash_lockp(&tcp_hashinfo, st->bucket);
2382
2383		/* Lockless fast path for the common case of empty buckets */
2384		if (empty_bucket(st))
2385			continue;
2386
2387		spin_lock_bh(lock);
2388		sk_nulls_for_each(sk, node, &tcp_hashinfo.ehash[st->bucket].chain) {
2389			if ((afinfo->family != AF_UNSPEC &&
2390			     sk->sk_family != afinfo->family) ||
2391			    !net_eq(sock_net(sk), net)) {
2392				continue;
2393			}
2394			rc = sk;
2395			goto out;
2396		}
2397		spin_unlock_bh(lock);
2398	}
2399out:
2400	return rc;
2401}
2402
2403static void *established_get_next(struct seq_file *seq, void *cur)
2404{
2405	struct tcp_seq_afinfo *afinfo;
2406	struct sock *sk = cur;
2407	struct hlist_nulls_node *node;
2408	struct tcp_iter_state *st = seq->private;
2409	struct net *net = seq_file_net(seq);
2410
2411	if (st->bpf_seq_afinfo)
2412		afinfo = st->bpf_seq_afinfo;
2413	else
2414		afinfo = PDE_DATA(file_inode(seq->file));
2415
2416	++st->num;
2417	++st->offset;
2418
2419	sk = sk_nulls_next(sk);
2420
2421	sk_nulls_for_each_from(sk, node) {
2422		if ((afinfo->family == AF_UNSPEC ||
2423		     sk->sk_family == afinfo->family) &&
2424		    net_eq(sock_net(sk), net))
2425			return sk;
2426	}
2427
2428	spin_unlock_bh(inet_ehash_lockp(&tcp_hashinfo, st->bucket));
2429	++st->bucket;
2430	return established_get_first(seq);
2431}
2432
2433static void *established_get_idx(struct seq_file *seq, loff_t pos)
2434{
2435	struct tcp_iter_state *st = seq->private;
2436	void *rc;
2437
2438	st->bucket = 0;
2439	rc = established_get_first(seq);
2440
2441	while (rc && pos) {
2442		rc = established_get_next(seq, rc);
2443		--pos;
2444	}
2445	return rc;
2446}
2447
2448static void *tcp_get_idx(struct seq_file *seq, loff_t pos)
2449{
2450	void *rc;
2451	struct tcp_iter_state *st = seq->private;
2452
2453	st->state = TCP_SEQ_STATE_LISTENING;
2454	rc	  = listening_get_idx(seq, &pos);
2455
2456	if (!rc) {
2457		st->state = TCP_SEQ_STATE_ESTABLISHED;
2458		rc	  = established_get_idx(seq, pos);
2459	}
2460
2461	return rc;
2462}
2463
2464static void *tcp_seek_last_pos(struct seq_file *seq)
2465{
2466	struct tcp_iter_state *st = seq->private;
2467	int bucket = st->bucket;
2468	int offset = st->offset;
2469	int orig_num = st->num;
2470	void *rc = NULL;
2471
2472	switch (st->state) {
2473	case TCP_SEQ_STATE_LISTENING:
2474		if (st->bucket >= INET_LHTABLE_SIZE)
2475			break;
2476		st->state = TCP_SEQ_STATE_LISTENING;
2477		rc = listening_get_next(seq, NULL);
2478		while (offset-- && rc && bucket == st->bucket)
2479			rc = listening_get_next(seq, rc);
2480		if (rc)
2481			break;
2482		st->bucket = 0;
2483		st->state = TCP_SEQ_STATE_ESTABLISHED;
2484		fallthrough;
2485	case TCP_SEQ_STATE_ESTABLISHED:
2486		if (st->bucket > tcp_hashinfo.ehash_mask)
2487			break;
2488		rc = established_get_first(seq);
2489		while (offset-- && rc && bucket == st->bucket)
2490			rc = established_get_next(seq, rc);
2491	}
2492
2493	st->num = orig_num;
2494
2495	return rc;
2496}
2497
2498void *tcp_seq_start(struct seq_file *seq, loff_t *pos)
2499{
2500	struct tcp_iter_state *st = seq->private;
2501	void *rc;
2502
2503	if (*pos && *pos == st->last_pos) {
2504		rc = tcp_seek_last_pos(seq);
2505		if (rc)
2506			goto out;
2507	}
2508
2509	st->state = TCP_SEQ_STATE_LISTENING;
2510	st->num = 0;
2511	st->bucket = 0;
2512	st->offset = 0;
2513	rc = *pos ? tcp_get_idx(seq, *pos - 1) : SEQ_START_TOKEN;
2514
2515out:
2516	st->last_pos = *pos;
2517	return rc;
2518}
2519EXPORT_SYMBOL(tcp_seq_start);
2520
2521void *tcp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2522{
2523	struct tcp_iter_state *st = seq->private;
2524	void *rc = NULL;
2525
2526	if (v == SEQ_START_TOKEN) {
2527		rc = tcp_get_idx(seq, 0);
2528		goto out;
2529	}
2530
2531	switch (st->state) {
2532	case TCP_SEQ_STATE_LISTENING:
2533		rc = listening_get_next(seq, v);
2534		if (!rc) {
2535			st->state = TCP_SEQ_STATE_ESTABLISHED;
2536			st->bucket = 0;
2537			st->offset = 0;
2538			rc	  = established_get_first(seq);
2539		}
2540		break;
2541	case TCP_SEQ_STATE_ESTABLISHED:
2542		rc = established_get_next(seq, v);
2543		break;
2544	}
2545out:
2546	++*pos;
2547	st->last_pos = *pos;
2548	return rc;
2549}
2550EXPORT_SYMBOL(tcp_seq_next);
2551
2552void tcp_seq_stop(struct seq_file *seq, void *v)
2553{
2554	struct tcp_iter_state *st = seq->private;
2555
2556	switch (st->state) {
2557	case TCP_SEQ_STATE_LISTENING:
2558		if (v != SEQ_START_TOKEN)
2559			spin_unlock(&tcp_hashinfo.listening_hash[st->bucket].lock);
2560		break;
2561	case TCP_SEQ_STATE_ESTABLISHED:
2562		if (v)
2563			spin_unlock_bh(inet_ehash_lockp(&tcp_hashinfo, st->bucket));
2564		break;
2565	}
2566}
2567EXPORT_SYMBOL(tcp_seq_stop);
2568
2569static void get_openreq4(const struct request_sock *req,
2570			 struct seq_file *f, int i)
2571{
2572	const struct inet_request_sock *ireq = inet_rsk(req);
2573	long delta = req->rsk_timer.expires - jiffies;
2574
2575	seq_printf(f, "%4d: %08X:%04X %08X:%04X"
2576		" %02X %08X:%08X %02X:%08lX %08X %5u %8d %u %d %pK",
2577		i,
2578		ireq->ir_loc_addr,
2579		ireq->ir_num,
2580		ireq->ir_rmt_addr,
2581		ntohs(ireq->ir_rmt_port),
2582		TCP_SYN_RECV,
2583		0, 0, /* could print option size, but that is af dependent. */
2584		1,    /* timers active (only the expire timer) */
2585		jiffies_delta_to_clock_t(delta),
2586		req->num_timeout,
2587		from_kuid_munged(seq_user_ns(f),
2588				 sock_i_uid(req->rsk_listener)),
2589		0,  /* non standard timer */
2590		0, /* open_requests have no inode */
2591		0,
2592		req);
2593}
2594
2595static void get_tcp4_sock(struct sock *sk, struct seq_file *f, int i)
2596{
2597	int timer_active;
2598	unsigned long timer_expires;
2599	const struct tcp_sock *tp = tcp_sk(sk);
2600	const struct inet_connection_sock *icsk = inet_csk(sk);
2601	const struct inet_sock *inet = inet_sk(sk);
2602	const struct fastopen_queue *fastopenq = &icsk->icsk_accept_queue.fastopenq;
2603	__be32 dest = inet->inet_daddr;
2604	__be32 src = inet->inet_rcv_saddr;
2605	__u16 destp = ntohs(inet->inet_dport);
2606	__u16 srcp = ntohs(inet->inet_sport);
2607	int rx_queue;
2608	int state;
2609
2610	if (icsk->icsk_pending == ICSK_TIME_RETRANS ||
2611	    icsk->icsk_pending == ICSK_TIME_REO_TIMEOUT ||
2612	    icsk->icsk_pending == ICSK_TIME_LOSS_PROBE) {
2613		timer_active	= 1;
2614		timer_expires	= icsk->icsk_timeout;
2615	} else if (icsk->icsk_pending == ICSK_TIME_PROBE0) {
2616		timer_active	= 4;
2617		timer_expires	= icsk->icsk_timeout;
2618	} else if (timer_pending(&sk->sk_timer)) {
2619		timer_active	= 2;
2620		timer_expires	= sk->sk_timer.expires;
2621	} else {
2622		timer_active	= 0;
2623		timer_expires = jiffies;
2624	}
2625
2626	state = inet_sk_state_load(sk);
2627	if (state == TCP_LISTEN)
2628		rx_queue = READ_ONCE(sk->sk_ack_backlog);
2629	else
2630		/* Because we don't lock the socket,
2631		 * we might find a transient negative value.
2632		 */
2633		rx_queue = max_t(int, READ_ONCE(tp->rcv_nxt) -
2634				      READ_ONCE(tp->copied_seq), 0);
2635
2636	seq_printf(f, "%4d: %08X:%04X %08X:%04X %02X %08X:%08X %02X:%08lX "
2637			"%08X %5u %8d %lu %d %pK %lu %lu %u %u %d",
2638		i, src, srcp, dest, destp, state,
2639		READ_ONCE(tp->write_seq) - tp->snd_una,
2640		rx_queue,
2641		timer_active,
2642		jiffies_delta_to_clock_t(timer_expires - jiffies),
2643		icsk->icsk_retransmits,
2644		from_kuid_munged(seq_user_ns(f), sock_i_uid(sk)),
2645		icsk->icsk_probes_out,
2646		sock_i_ino(sk),
2647		refcount_read(&sk->sk_refcnt), sk,
2648		jiffies_to_clock_t(icsk->icsk_rto),
2649		jiffies_to_clock_t(icsk->icsk_ack.ato),
2650		(icsk->icsk_ack.quick << 1) | inet_csk_in_pingpong_mode(sk),
2651		tp->snd_cwnd,
2652		state == TCP_LISTEN ?
2653		    fastopenq->max_qlen :
2654		    (tcp_in_initial_slowstart(tp) ? -1 : tp->snd_ssthresh));
2655}
2656
2657static void get_timewait4_sock(const struct inet_timewait_sock *tw,
2658			       struct seq_file *f, int i)
2659{
2660	long delta = tw->tw_timer.expires - jiffies;
2661	__be32 dest, src;
2662	__u16 destp, srcp;
2663
2664	dest  = tw->tw_daddr;
2665	src   = tw->tw_rcv_saddr;
2666	destp = ntohs(tw->tw_dport);
2667	srcp  = ntohs(tw->tw_sport);
2668
2669	seq_printf(f, "%4d: %08X:%04X %08X:%04X"
2670		" %02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %pK",
2671		i, src, srcp, dest, destp, tw->tw_substate, 0, 0,
2672		3, jiffies_delta_to_clock_t(delta), 0, 0, 0, 0,
2673		refcount_read(&tw->tw_refcnt), tw);
2674}
2675
2676#define TMPSZ 150
2677
2678static int tcp4_seq_show(struct seq_file *seq, void *v)
2679{
2680	struct tcp_iter_state *st;
2681	struct sock *sk = v;
2682
2683	seq_setwidth(seq, TMPSZ - 1);
2684	if (v == SEQ_START_TOKEN) {
2685		seq_puts(seq, "  sl  local_address rem_address   st tx_queue "
2686			   "rx_queue tr tm->when retrnsmt   uid  timeout "
2687			   "inode");
2688		goto out;
2689	}
2690	st = seq->private;
2691
2692	if (sk->sk_state == TCP_TIME_WAIT)
2693		get_timewait4_sock(v, seq, st->num);
2694	else if (sk->sk_state == TCP_NEW_SYN_RECV)
2695		get_openreq4(v, seq, st->num);
2696	else
2697		get_tcp4_sock(v, seq, st->num);
2698out:
2699	seq_pad(seq, '\n');
2700	return 0;
2701}
2702
2703#ifdef CONFIG_BPF_SYSCALL
2704struct bpf_iter__tcp {
2705	__bpf_md_ptr(struct bpf_iter_meta *, meta);
2706	__bpf_md_ptr(struct sock_common *, sk_common);
2707	uid_t uid __aligned(8);
2708};
2709
2710static int tcp_prog_seq_show(struct bpf_prog *prog, struct bpf_iter_meta *meta,
2711			     struct sock_common *sk_common, uid_t uid)
2712{
2713	struct bpf_iter__tcp ctx;
2714
2715	meta->seq_num--;  /* skip SEQ_START_TOKEN */
2716	ctx.meta = meta;
2717	ctx.sk_common = sk_common;
2718	ctx.uid = uid;
2719	return bpf_iter_run_prog(prog, &ctx);
2720}
2721
2722static int bpf_iter_tcp_seq_show(struct seq_file *seq, void *v)
2723{
2724	struct bpf_iter_meta meta;
2725	struct bpf_prog *prog;
2726	struct sock *sk = v;
2727	uid_t uid;
2728
2729	if (v == SEQ_START_TOKEN)
2730		return 0;
2731
2732	if (sk->sk_state == TCP_TIME_WAIT) {
2733		uid = 0;
2734	} else if (sk->sk_state == TCP_NEW_SYN_RECV) {
2735		const struct request_sock *req = v;
2736
2737		uid = from_kuid_munged(seq_user_ns(seq),
2738				       sock_i_uid(req->rsk_listener));
2739	} else {
2740		uid = from_kuid_munged(seq_user_ns(seq), sock_i_uid(sk));
2741	}
2742
2743	meta.seq = seq;
2744	prog = bpf_iter_get_info(&meta, false);
2745	return tcp_prog_seq_show(prog, &meta, v, uid);
2746}
2747
2748static void bpf_iter_tcp_seq_stop(struct seq_file *seq, void *v)
2749{
2750	struct bpf_iter_meta meta;
2751	struct bpf_prog *prog;
2752
2753	if (!v) {
2754		meta.seq = seq;
2755		prog = bpf_iter_get_info(&meta, true);
2756		if (prog)
2757			(void)tcp_prog_seq_show(prog, &meta, v, 0);
2758	}
2759
2760	tcp_seq_stop(seq, v);
2761}
2762
2763static const struct seq_operations bpf_iter_tcp_seq_ops = {
2764	.show		= bpf_iter_tcp_seq_show,
2765	.start		= tcp_seq_start,
2766	.next		= tcp_seq_next,
2767	.stop		= bpf_iter_tcp_seq_stop,
2768};
2769#endif
2770
2771static const struct seq_operations tcp4_seq_ops = {
2772	.show		= tcp4_seq_show,
2773	.start		= tcp_seq_start,
2774	.next		= tcp_seq_next,
2775	.stop		= tcp_seq_stop,
2776};
2777
2778static struct tcp_seq_afinfo tcp4_seq_afinfo = {
2779	.family		= AF_INET,
2780};
2781
2782static int __net_init tcp4_proc_init_net(struct net *net)
2783{
2784	if (!proc_create_net_data("tcp", 0444, net->proc_net, &tcp4_seq_ops,
2785			sizeof(struct tcp_iter_state), &tcp4_seq_afinfo))
2786		return -ENOMEM;
2787	return 0;
2788}
2789
2790static void __net_exit tcp4_proc_exit_net(struct net *net)
2791{
2792	remove_proc_entry("tcp", net->proc_net);
2793}
2794
2795static struct pernet_operations tcp4_net_ops = {
2796	.init = tcp4_proc_init_net,
2797	.exit = tcp4_proc_exit_net,
2798};
2799
2800int __init tcp4_proc_init(void)
2801{
2802	return register_pernet_subsys(&tcp4_net_ops);
2803}
2804
2805void tcp4_proc_exit(void)
2806{
2807	unregister_pernet_subsys(&tcp4_net_ops);
2808}
2809#endif /* CONFIG_PROC_FS */
2810
2811/* @wake is one when sk_stream_write_space() calls us.
2812 * This sends EPOLLOUT only if notsent_bytes is half the limit.
2813 * This mimics the strategy used in sock_def_write_space().
2814 */
2815bool tcp_stream_memory_free(const struct sock *sk, int wake)
2816{
2817	const struct tcp_sock *tp = tcp_sk(sk);
2818	u32 notsent_bytes = READ_ONCE(tp->write_seq) -
2819			    READ_ONCE(tp->snd_nxt);
2820
2821	return (notsent_bytes << wake) < tcp_notsent_lowat(tp);
2822}
2823EXPORT_SYMBOL(tcp_stream_memory_free);
2824
2825struct proto tcp_prot = {
2826	.name			= "TCP",
2827	.owner			= THIS_MODULE,
2828	.close			= tcp_close,
2829	.pre_connect		= tcp_v4_pre_connect,
2830	.connect		= tcp_v4_connect,
2831	.disconnect		= tcp_disconnect,
2832	.accept			= inet_csk_accept,
2833	.ioctl			= tcp_ioctl,
2834	.init			= tcp_v4_init_sock,
2835	.destroy		= tcp_v4_destroy_sock,
2836	.shutdown		= tcp_shutdown,
2837	.setsockopt		= tcp_setsockopt,
2838	.getsockopt		= tcp_getsockopt,
2839	.bpf_bypass_getsockopt	= tcp_bpf_bypass_getsockopt,
2840	.keepalive		= tcp_set_keepalive,
2841	.recvmsg		= tcp_recvmsg,
2842	.sendmsg		= tcp_sendmsg,
2843	.sendpage		= tcp_sendpage,
2844	.backlog_rcv		= tcp_v4_do_rcv,
2845	.release_cb		= tcp_release_cb,
2846	.hash			= inet_hash,
2847	.unhash			= inet_unhash,
2848	.get_port		= inet_csk_get_port,
2849#ifdef CONFIG_BPF_SYSCALL
2850	.psock_update_sk_prot	= tcp_bpf_update_proto,
2851#endif
2852	.enter_memory_pressure	= tcp_enter_memory_pressure,
2853	.leave_memory_pressure	= tcp_leave_memory_pressure,
2854	.stream_memory_free	= tcp_stream_memory_free,
2855	.sockets_allocated	= &tcp_sockets_allocated,
2856	.orphan_count		= &tcp_orphan_count,
2857	.memory_allocated	= &tcp_memory_allocated,
2858	.memory_pressure	= &tcp_memory_pressure,
2859	.sysctl_mem		= sysctl_tcp_mem,
2860	.sysctl_wmem_offset	= offsetof(struct net, ipv4.sysctl_tcp_wmem),
2861	.sysctl_rmem_offset	= offsetof(struct net, ipv4.sysctl_tcp_rmem),
2862	.max_header		= MAX_TCP_HEADER,
2863	.obj_size		= sizeof(struct tcp_sock),
2864	.slab_flags		= SLAB_TYPESAFE_BY_RCU,
2865	.twsk_prot		= &tcp_timewait_sock_ops,
2866	.rsk_prot		= &tcp_request_sock_ops,
2867	.h.hashinfo		= &tcp_hashinfo,
2868	.no_autobind		= true,
2869	.diag_destroy		= tcp_abort,
2870};
2871EXPORT_SYMBOL(tcp_prot);
2872
2873static void __net_exit tcp_sk_exit(struct net *net)
2874{
2875	int cpu;
2876
2877	if (net->ipv4.tcp_congestion_control)
2878		bpf_module_put(net->ipv4.tcp_congestion_control,
2879			       net->ipv4.tcp_congestion_control->owner);
2880
2881	for_each_possible_cpu(cpu)
2882		inet_ctl_sock_destroy(*per_cpu_ptr(net->ipv4.tcp_sk, cpu));
2883	free_percpu(net->ipv4.tcp_sk);
2884}
2885
2886static int __net_init tcp_sk_init(struct net *net)
2887{
2888	int res, cpu, cnt;
2889
2890	net->ipv4.tcp_sk = alloc_percpu(struct sock *);
2891	if (!net->ipv4.tcp_sk)
2892		return -ENOMEM;
2893
2894	for_each_possible_cpu(cpu) {
2895		struct sock *sk;
2896
2897		res = inet_ctl_sock_create(&sk, PF_INET, SOCK_RAW,
2898					   IPPROTO_TCP, net);
2899		if (res)
2900			goto fail;
2901		sock_set_flag(sk, SOCK_USE_WRITE_QUEUE);
2902
2903		/* Please enforce IP_DF and IPID==0 for RST and
2904		 * ACK sent in SYN-RECV and TIME-WAIT state.
2905		 */
2906		inet_sk(sk)->pmtudisc = IP_PMTUDISC_DO;
2907
2908		*per_cpu_ptr(net->ipv4.tcp_sk, cpu) = sk;
2909	}
2910
2911	net->ipv4.sysctl_tcp_ecn = 2;
2912	net->ipv4.sysctl_tcp_ecn_fallback = 1;
2913
2914	net->ipv4.sysctl_tcp_base_mss = TCP_BASE_MSS;
2915	net->ipv4.sysctl_tcp_min_snd_mss = TCP_MIN_SND_MSS;
2916	net->ipv4.sysctl_tcp_probe_threshold = TCP_PROBE_THRESHOLD;
2917	net->ipv4.sysctl_tcp_probe_interval = TCP_PROBE_INTERVAL;
2918	net->ipv4.sysctl_tcp_mtu_probe_floor = TCP_MIN_SND_MSS;
2919
2920	net->ipv4.sysctl_tcp_keepalive_time = TCP_KEEPALIVE_TIME;
2921	net->ipv4.sysctl_tcp_keepalive_probes = TCP_KEEPALIVE_PROBES;
2922	net->ipv4.sysctl_tcp_keepalive_intvl = TCP_KEEPALIVE_INTVL;
2923
2924	net->ipv4.sysctl_tcp_syn_retries = TCP_SYN_RETRIES;
2925	net->ipv4.sysctl_tcp_synack_retries = TCP_SYNACK_RETRIES;
2926	net->ipv4.sysctl_tcp_syncookies = 1;
2927	net->ipv4.sysctl_tcp_reordering = TCP_FASTRETRANS_THRESH;
2928	net->ipv4.sysctl_tcp_retries1 = TCP_RETR1;
2929	net->ipv4.sysctl_tcp_retries2 = TCP_RETR2;
2930	net->ipv4.sysctl_tcp_orphan_retries = 0;
2931	net->ipv4.sysctl_tcp_fin_timeout = TCP_FIN_TIMEOUT;
2932	net->ipv4.sysctl_tcp_notsent_lowat = UINT_MAX;
2933	net->ipv4.sysctl_tcp_tw_reuse = 2;
2934	net->ipv4.sysctl_tcp_no_ssthresh_metrics_save = 1;
2935
2936	cnt = tcp_hashinfo.ehash_mask + 1;
2937	net->ipv4.tcp_death_row.sysctl_max_tw_buckets = cnt / 2;
2938	net->ipv4.tcp_death_row.hashinfo = &tcp_hashinfo;
2939
2940	net->ipv4.sysctl_max_syn_backlog = max(128, cnt / 128);
2941	net->ipv4.sysctl_tcp_sack = 1;
2942	net->ipv4.sysctl_tcp_window_scaling = 1;
2943	net->ipv4.sysctl_tcp_timestamps = 1;
2944	net->ipv4.sysctl_tcp_early_retrans = 3;
2945	net->ipv4.sysctl_tcp_recovery = TCP_RACK_LOSS_DETECTION;
2946	net->ipv4.sysctl_tcp_slow_start_after_idle = 1; /* By default, RFC2861 behavior.  */
2947	net->ipv4.sysctl_tcp_retrans_collapse = 1;
2948	net->ipv4.sysctl_tcp_max_reordering = 300;
2949	net->ipv4.sysctl_tcp_dsack = 1;
2950	net->ipv4.sysctl_tcp_app_win = 31;
2951	net->ipv4.sysctl_tcp_adv_win_scale = 1;
2952	net->ipv4.sysctl_tcp_frto = 2;
2953	net->ipv4.sysctl_tcp_moderate_rcvbuf = 1;
2954	/* This limits the percentage of the congestion window which we
2955	 * will allow a single TSO frame to consume.  Building TSO frames
2956	 * which are too large can cause TCP streams to be bursty.
2957	 */
2958	net->ipv4.sysctl_tcp_tso_win_divisor = 3;
2959	/* Default TSQ limit of 16 TSO segments */
2960	net->ipv4.sysctl_tcp_limit_output_bytes = 16 * 65536;
2961	/* rfc5961 challenge ack rate limiting */
2962	net->ipv4.sysctl_tcp_challenge_ack_limit = 1000;
2963	net->ipv4.sysctl_tcp_min_tso_segs = 2;
2964	net->ipv4.sysctl_tcp_min_rtt_wlen = 300;
2965	net->ipv4.sysctl_tcp_autocorking = 1;
2966	net->ipv4.sysctl_tcp_invalid_ratelimit = HZ/2;
2967	net->ipv4.sysctl_tcp_pacing_ss_ratio = 200;
2968	net->ipv4.sysctl_tcp_pacing_ca_ratio = 120;
2969	if (net != &init_net) {
2970		memcpy(net->ipv4.sysctl_tcp_rmem,
2971		       init_net.ipv4.sysctl_tcp_rmem,
2972		       sizeof(init_net.ipv4.sysctl_tcp_rmem));
2973		memcpy(net->ipv4.sysctl_tcp_wmem,
2974		       init_net.ipv4.sysctl_tcp_wmem,
2975		       sizeof(init_net.ipv4.sysctl_tcp_wmem));
2976	}
2977	net->ipv4.sysctl_tcp_comp_sack_delay_ns = NSEC_PER_MSEC;
2978	net->ipv4.sysctl_tcp_comp_sack_slack_ns = 100 * NSEC_PER_USEC;
2979	net->ipv4.sysctl_tcp_comp_sack_nr = 44;
2980	net->ipv4.sysctl_tcp_fastopen = TFO_CLIENT_ENABLE;
2981	spin_lock_init(&net->ipv4.tcp_fastopen_ctx_lock);
2982	net->ipv4.sysctl_tcp_fastopen_blackhole_timeout = 0;
2983	atomic_set(&net->ipv4.tfo_active_disable_times, 0);
2984
2985	/* Reno is always built in */
2986	if (!net_eq(net, &init_net) &&
2987	    bpf_try_module_get(init_net.ipv4.tcp_congestion_control,
2988			       init_net.ipv4.tcp_congestion_control->owner))
2989		net->ipv4.tcp_congestion_control = init_net.ipv4.tcp_congestion_control;
2990	else
2991		net->ipv4.tcp_congestion_control = &tcp_reno;
2992
2993	return 0;
2994fail:
2995	tcp_sk_exit(net);
2996
2997	return res;
2998}
2999
3000static void __net_exit tcp_sk_exit_batch(struct list_head *net_exit_list)
3001{
3002	struct net *net;
3003
3004	inet_twsk_purge(&tcp_hashinfo, AF_INET);
3005
3006	list_for_each_entry(net, net_exit_list, exit_list)
3007		tcp_fastopen_ctx_destroy(net);
3008}
3009
3010static struct pernet_operations __net_initdata tcp_sk_ops = {
3011       .init	   = tcp_sk_init,
3012       .exit	   = tcp_sk_exit,
3013       .exit_batch = tcp_sk_exit_batch,
3014};
3015
3016#if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
3017DEFINE_BPF_ITER_FUNC(tcp, struct bpf_iter_meta *meta,
3018		     struct sock_common *sk_common, uid_t uid)
3019
3020static int bpf_iter_init_tcp(void *priv_data, struct bpf_iter_aux_info *aux)
3021{
3022	struct tcp_iter_state *st = priv_data;
3023	struct tcp_seq_afinfo *afinfo;
3024	int ret;
3025
3026	afinfo = kmalloc(sizeof(*afinfo), GFP_USER | __GFP_NOWARN);
3027	if (!afinfo)
3028		return -ENOMEM;
3029
3030	afinfo->family = AF_UNSPEC;
3031	st->bpf_seq_afinfo = afinfo;
3032	ret = bpf_iter_init_seq_net(priv_data, aux);
3033	if (ret)
3034		kfree(afinfo);
3035	return ret;
3036}
3037
3038static void bpf_iter_fini_tcp(void *priv_data)
3039{
3040	struct tcp_iter_state *st = priv_data;
3041
3042	kfree(st->bpf_seq_afinfo);
3043	bpf_iter_fini_seq_net(priv_data);
3044}
3045
3046static const struct bpf_iter_seq_info tcp_seq_info = {
3047	.seq_ops		= &bpf_iter_tcp_seq_ops,
3048	.init_seq_private	= bpf_iter_init_tcp,
3049	.fini_seq_private	= bpf_iter_fini_tcp,
3050	.seq_priv_size		= sizeof(struct tcp_iter_state),
3051};
3052
3053static struct bpf_iter_reg tcp_reg_info = {
3054	.target			= "tcp",
3055	.ctx_arg_info_size	= 1,
3056	.ctx_arg_info		= {
3057		{ offsetof(struct bpf_iter__tcp, sk_common),
3058		  PTR_TO_BTF_ID_OR_NULL },
3059	},
3060	.seq_info		= &tcp_seq_info,
3061};
3062
3063static void __init bpf_iter_register(void)
3064{
3065	tcp_reg_info.ctx_arg_info[0].btf_id = btf_sock_ids[BTF_SOCK_TYPE_SOCK_COMMON];
3066	if (bpf_iter_reg_target(&tcp_reg_info))
3067		pr_warn("Warning: could not register bpf iterator tcp\n");
3068}
3069
3070#endif
3071
3072void __init tcp_v4_init(void)
3073{
3074	if (register_pernet_subsys(&tcp_sk_ops))
3075		panic("Failed to create the TCP control socket.\n");
3076
3077#if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
3078	bpf_iter_register();
3079#endif
3080}
v5.9
   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
  61#include <net/net_namespace.h>
  62#include <net/icmp.h>
  63#include <net/inet_hashtables.h>
  64#include <net/tcp.h>
  65#include <net/transp_v6.h>
  66#include <net/ipv6.h>
  67#include <net/inet_common.h>
  68#include <net/timewait_sock.h>
  69#include <net/xfrm.h>
  70#include <net/secure_seq.h>
  71#include <net/busy_poll.h>
  72
  73#include <linux/inet.h>
  74#include <linux/ipv6.h>
  75#include <linux/stddef.h>
  76#include <linux/proc_fs.h>
  77#include <linux/seq_file.h>
  78#include <linux/inetdevice.h>
  79#include <linux/btf_ids.h>
  80
  81#include <crypto/hash.h>
  82#include <linux/scatterlist.h>
  83
  84#include <trace/events/tcp.h>
  85
  86#ifdef CONFIG_TCP_MD5SIG
  87static int tcp_v4_md5_hash_hdr(char *md5_hash, const struct tcp_md5sig_key *key,
  88			       __be32 daddr, __be32 saddr, const struct tcphdr *th);
  89#endif
  90
  91struct inet_hashinfo tcp_hashinfo;
  92EXPORT_SYMBOL(tcp_hashinfo);
  93
  94static u32 tcp_v4_init_seq(const struct sk_buff *skb)
  95{
  96	return secure_tcp_seq(ip_hdr(skb)->daddr,
  97			      ip_hdr(skb)->saddr,
  98			      tcp_hdr(skb)->dest,
  99			      tcp_hdr(skb)->source);
 100}
 101
 102static u32 tcp_v4_init_ts_off(const struct net *net, const struct sk_buff *skb)
 103{
 104	return secure_tcp_ts_off(net, ip_hdr(skb)->daddr, ip_hdr(skb)->saddr);
 105}
 106
 107int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp)
 108{
 109	const struct inet_timewait_sock *tw = inet_twsk(sktw);
 110	const struct tcp_timewait_sock *tcptw = tcp_twsk(sktw);
 111	struct tcp_sock *tp = tcp_sk(sk);
 112	int reuse = sock_net(sk)->ipv4.sysctl_tcp_tw_reuse;
 113
 114	if (reuse == 2) {
 115		/* Still does not detect *everything* that goes through
 116		 * lo, since we require a loopback src or dst address
 117		 * or direct binding to 'lo' interface.
 118		 */
 119		bool loopback = false;
 120		if (tw->tw_bound_dev_if == LOOPBACK_IFINDEX)
 121			loopback = true;
 122#if IS_ENABLED(CONFIG_IPV6)
 123		if (tw->tw_family == AF_INET6) {
 124			if (ipv6_addr_loopback(&tw->tw_v6_daddr) ||
 125			    ipv6_addr_v4mapped_loopback(&tw->tw_v6_daddr) ||
 126			    ipv6_addr_loopback(&tw->tw_v6_rcv_saddr) ||
 127			    ipv6_addr_v4mapped_loopback(&tw->tw_v6_rcv_saddr))
 128				loopback = true;
 129		} else
 130#endif
 131		{
 132			if (ipv4_is_loopback(tw->tw_daddr) ||
 133			    ipv4_is_loopback(tw->tw_rcv_saddr))
 134				loopback = true;
 135		}
 136		if (!loopback)
 137			reuse = 0;
 138	}
 139
 140	/* With PAWS, it is safe from the viewpoint
 141	   of data integrity. Even without PAWS it is safe provided sequence
 142	   spaces do not overlap i.e. at data rates <= 80Mbit/sec.
 143
 144	   Actually, the idea is close to VJ's one, only timestamp cache is
 145	   held not per host, but per port pair and TW bucket is used as state
 146	   holder.
 147
 148	   If TW bucket has been already destroyed we fall back to VJ's scheme
 149	   and use initial timestamp retrieved from peer table.
 150	 */
 151	if (tcptw->tw_ts_recent_stamp &&
 152	    (!twp || (reuse && time_after32(ktime_get_seconds(),
 153					    tcptw->tw_ts_recent_stamp)))) {
 154		/* In case of repair and re-using TIME-WAIT sockets we still
 155		 * want to be sure that it is safe as above but honor the
 156		 * sequence numbers and time stamps set as part of the repair
 157		 * process.
 158		 *
 159		 * Without this check re-using a TIME-WAIT socket with TCP
 160		 * repair would accumulate a -1 on the repair assigned
 161		 * sequence number. The first time it is reused the sequence
 162		 * is -1, the second time -2, etc. This fixes that issue
 163		 * without appearing to create any others.
 164		 */
 165		if (likely(!tp->repair)) {
 166			u32 seq = tcptw->tw_snd_nxt + 65535 + 2;
 167
 168			if (!seq)
 169				seq = 1;
 170			WRITE_ONCE(tp->write_seq, seq);
 171			tp->rx_opt.ts_recent	   = tcptw->tw_ts_recent;
 172			tp->rx_opt.ts_recent_stamp = tcptw->tw_ts_recent_stamp;
 173		}
 174		sock_hold(sktw);
 175		return 1;
 176	}
 177
 178	return 0;
 179}
 180EXPORT_SYMBOL_GPL(tcp_twsk_unique);
 181
 182static int tcp_v4_pre_connect(struct sock *sk, struct sockaddr *uaddr,
 183			      int addr_len)
 184{
 185	/* This check is replicated from tcp_v4_connect() and intended to
 186	 * prevent BPF program called below from accessing bytes that are out
 187	 * of the bound specified by user in addr_len.
 188	 */
 189	if (addr_len < sizeof(struct sockaddr_in))
 190		return -EINVAL;
 191
 192	sock_owned_by_me(sk);
 193
 194	return BPF_CGROUP_RUN_PROG_INET4_CONNECT(sk, uaddr);
 195}
 196
 197/* This will initiate an outgoing connection. */
 198int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len)
 199{
 200	struct sockaddr_in *usin = (struct sockaddr_in *)uaddr;
 201	struct inet_sock *inet = inet_sk(sk);
 202	struct tcp_sock *tp = tcp_sk(sk);
 203	__be16 orig_sport, orig_dport;
 204	__be32 daddr, nexthop;
 205	struct flowi4 *fl4;
 206	struct rtable *rt;
 207	int err;
 208	struct ip_options_rcu *inet_opt;
 209	struct inet_timewait_death_row *tcp_death_row = &sock_net(sk)->ipv4.tcp_death_row;
 210
 211	if (addr_len < sizeof(struct sockaddr_in))
 212		return -EINVAL;
 213
 214	if (usin->sin_family != AF_INET)
 215		return -EAFNOSUPPORT;
 216
 217	nexthop = daddr = usin->sin_addr.s_addr;
 218	inet_opt = rcu_dereference_protected(inet->inet_opt,
 219					     lockdep_sock_is_held(sk));
 220	if (inet_opt && inet_opt->opt.srr) {
 221		if (!daddr)
 222			return -EINVAL;
 223		nexthop = inet_opt->opt.faddr;
 224	}
 225
 226	orig_sport = inet->inet_sport;
 227	orig_dport = usin->sin_port;
 228	fl4 = &inet->cork.fl.u.ip4;
 229	rt = ip_route_connect(fl4, nexthop, inet->inet_saddr,
 230			      RT_CONN_FLAGS(sk), sk->sk_bound_dev_if,
 231			      IPPROTO_TCP,
 232			      orig_sport, orig_dport, sk);
 233	if (IS_ERR(rt)) {
 234		err = PTR_ERR(rt);
 235		if (err == -ENETUNREACH)
 236			IP_INC_STATS(sock_net(sk), IPSTATS_MIB_OUTNOROUTES);
 237		return err;
 238	}
 239
 240	if (rt->rt_flags & (RTCF_MULTICAST | RTCF_BROADCAST)) {
 241		ip_rt_put(rt);
 242		return -ENETUNREACH;
 243	}
 244
 245	if (!inet_opt || !inet_opt->opt.srr)
 246		daddr = fl4->daddr;
 247
 248	if (!inet->inet_saddr)
 249		inet->inet_saddr = fl4->saddr;
 250	sk_rcv_saddr_set(sk, inet->inet_saddr);
 251
 252	if (tp->rx_opt.ts_recent_stamp && inet->inet_daddr != daddr) {
 253		/* Reset inherited state */
 254		tp->rx_opt.ts_recent	   = 0;
 255		tp->rx_opt.ts_recent_stamp = 0;
 256		if (likely(!tp->repair))
 257			WRITE_ONCE(tp->write_seq, 0);
 258	}
 259
 260	inet->inet_dport = usin->sin_port;
 261	sk_daddr_set(sk, daddr);
 262
 263	inet_csk(sk)->icsk_ext_hdr_len = 0;
 264	if (inet_opt)
 265		inet_csk(sk)->icsk_ext_hdr_len = inet_opt->opt.optlen;
 266
 267	tp->rx_opt.mss_clamp = TCP_MSS_DEFAULT;
 268
 269	/* Socket identity is still unknown (sport may be zero).
 270	 * However we set state to SYN-SENT and not releasing socket
 271	 * lock select source port, enter ourselves into the hash tables and
 272	 * complete initialization after this.
 273	 */
 274	tcp_set_state(sk, TCP_SYN_SENT);
 275	err = inet_hash_connect(tcp_death_row, sk);
 276	if (err)
 277		goto failure;
 278
 279	sk_set_txhash(sk);
 280
 281	rt = ip_route_newports(fl4, rt, orig_sport, orig_dport,
 282			       inet->inet_sport, inet->inet_dport, sk);
 283	if (IS_ERR(rt)) {
 284		err = PTR_ERR(rt);
 285		rt = NULL;
 286		goto failure;
 287	}
 288	/* OK, now commit destination to socket.  */
 289	sk->sk_gso_type = SKB_GSO_TCPV4;
 290	sk_setup_caps(sk, &rt->dst);
 291	rt = NULL;
 292
 293	if (likely(!tp->repair)) {
 294		if (!tp->write_seq)
 295			WRITE_ONCE(tp->write_seq,
 296				   secure_tcp_seq(inet->inet_saddr,
 297						  inet->inet_daddr,
 298						  inet->inet_sport,
 299						  usin->sin_port));
 300		tp->tsoffset = secure_tcp_ts_off(sock_net(sk),
 301						 inet->inet_saddr,
 302						 inet->inet_daddr);
 303	}
 304
 305	inet->inet_id = prandom_u32();
 306
 307	if (tcp_fastopen_defer_connect(sk, &err))
 308		return err;
 309	if (err)
 310		goto failure;
 311
 312	err = tcp_connect(sk);
 313
 314	if (err)
 315		goto failure;
 316
 317	return 0;
 318
 319failure:
 320	/*
 321	 * This unhashes the socket and releases the local port,
 322	 * if necessary.
 323	 */
 324	tcp_set_state(sk, TCP_CLOSE);
 325	ip_rt_put(rt);
 326	sk->sk_route_caps = 0;
 327	inet->inet_dport = 0;
 328	return err;
 329}
 330EXPORT_SYMBOL(tcp_v4_connect);
 331
 332/*
 333 * This routine reacts to ICMP_FRAG_NEEDED mtu indications as defined in RFC1191.
 334 * It can be called through tcp_release_cb() if socket was owned by user
 335 * at the time tcp_v4_err() was called to handle ICMP message.
 336 */
 337void tcp_v4_mtu_reduced(struct sock *sk)
 338{
 339	struct inet_sock *inet = inet_sk(sk);
 340	struct dst_entry *dst;
 341	u32 mtu;
 342
 343	if ((1 << sk->sk_state) & (TCPF_LISTEN | TCPF_CLOSE))
 344		return;
 345	mtu = tcp_sk(sk)->mtu_info;
 346	dst = inet_csk_update_pmtu(sk, mtu);
 347	if (!dst)
 348		return;
 349
 350	/* Something is about to be wrong... Remember soft error
 351	 * for the case, if this connection will not able to recover.
 352	 */
 353	if (mtu < dst_mtu(dst) && ip_dont_fragment(sk, dst))
 354		sk->sk_err_soft = EMSGSIZE;
 355
 356	mtu = dst_mtu(dst);
 357
 358	if (inet->pmtudisc != IP_PMTUDISC_DONT &&
 359	    ip_sk_accept_pmtu(sk) &&
 360	    inet_csk(sk)->icsk_pmtu_cookie > mtu) {
 361		tcp_sync_mss(sk, mtu);
 362
 363		/* Resend the TCP packet because it's
 364		 * clear that the old packet has been
 365		 * dropped. This is the new "fast" path mtu
 366		 * discovery.
 367		 */
 368		tcp_simple_retransmit(sk);
 369	} /* else let the usual retransmit timer handle it */
 370}
 371EXPORT_SYMBOL(tcp_v4_mtu_reduced);
 372
 373static void do_redirect(struct sk_buff *skb, struct sock *sk)
 374{
 375	struct dst_entry *dst = __sk_dst_check(sk, 0);
 376
 377	if (dst)
 378		dst->ops->redirect(dst, sk, skb);
 379}
 380
 381
 382/* handle ICMP messages on TCP_NEW_SYN_RECV request sockets */
 383void tcp_req_err(struct sock *sk, u32 seq, bool abort)
 384{
 385	struct request_sock *req = inet_reqsk(sk);
 386	struct net *net = sock_net(sk);
 387
 388	/* ICMPs are not backlogged, hence we cannot get
 389	 * an established socket here.
 390	 */
 391	if (seq != tcp_rsk(req)->snt_isn) {
 392		__NET_INC_STATS(net, LINUX_MIB_OUTOFWINDOWICMPS);
 393	} else if (abort) {
 394		/*
 395		 * Still in SYN_RECV, just remove it silently.
 396		 * There is no good way to pass the error to the newly
 397		 * created socket, and POSIX does not want network
 398		 * errors returned from accept().
 399		 */
 400		inet_csk_reqsk_queue_drop(req->rsk_listener, req);
 401		tcp_listendrop(req->rsk_listener);
 402	}
 403	reqsk_put(req);
 404}
 405EXPORT_SYMBOL(tcp_req_err);
 406
 407/* TCP-LD (RFC 6069) logic */
 408void tcp_ld_RTO_revert(struct sock *sk, u32 seq)
 409{
 410	struct inet_connection_sock *icsk = inet_csk(sk);
 411	struct tcp_sock *tp = tcp_sk(sk);
 412	struct sk_buff *skb;
 413	s32 remaining;
 414	u32 delta_us;
 415
 416	if (sock_owned_by_user(sk))
 417		return;
 418
 419	if (seq != tp->snd_una  || !icsk->icsk_retransmits ||
 420	    !icsk->icsk_backoff)
 421		return;
 422
 423	skb = tcp_rtx_queue_head(sk);
 424	if (WARN_ON_ONCE(!skb))
 425		return;
 426
 427	icsk->icsk_backoff--;
 428	icsk->icsk_rto = tp->srtt_us ? __tcp_set_rto(tp) : TCP_TIMEOUT_INIT;
 429	icsk->icsk_rto = inet_csk_rto_backoff(icsk, TCP_RTO_MAX);
 430
 431	tcp_mstamp_refresh(tp);
 432	delta_us = (u32)(tp->tcp_mstamp - tcp_skb_timestamp_us(skb));
 433	remaining = icsk->icsk_rto - usecs_to_jiffies(delta_us);
 434
 435	if (remaining > 0) {
 436		inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
 437					  remaining, TCP_RTO_MAX);
 438	} else {
 439		/* RTO revert clocked out retransmission.
 440		 * Will retransmit now.
 441		 */
 442		tcp_retransmit_timer(sk);
 443	}
 444}
 445EXPORT_SYMBOL(tcp_ld_RTO_revert);
 446
 447/*
 448 * This routine is called by the ICMP module when it gets some
 449 * sort of error condition.  If err < 0 then the socket should
 450 * be closed and the error returned to the user.  If err > 0
 451 * it's just the icmp type << 8 | icmp code.  After adjustment
 452 * header points to the first 8 bytes of the tcp header.  We need
 453 * to find the appropriate port.
 454 *
 455 * The locking strategy used here is very "optimistic". When
 456 * someone else accesses the socket the ICMP is just dropped
 457 * and for some paths there is no check at all.
 458 * A more general error queue to queue errors for later handling
 459 * is probably better.
 460 *
 461 */
 462
 463int tcp_v4_err(struct sk_buff *skb, u32 info)
 464{
 465	const struct iphdr *iph = (const struct iphdr *)skb->data;
 466	struct tcphdr *th = (struct tcphdr *)(skb->data + (iph->ihl << 2));
 467	struct tcp_sock *tp;
 468	struct inet_sock *inet;
 469	const int type = icmp_hdr(skb)->type;
 470	const int code = icmp_hdr(skb)->code;
 471	struct sock *sk;
 472	struct request_sock *fastopen;
 473	u32 seq, snd_una;
 474	int err;
 475	struct net *net = dev_net(skb->dev);
 476
 477	sk = __inet_lookup_established(net, &tcp_hashinfo, iph->daddr,
 478				       th->dest, iph->saddr, ntohs(th->source),
 479				       inet_iif(skb), 0);
 480	if (!sk) {
 481		__ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
 482		return -ENOENT;
 483	}
 484	if (sk->sk_state == TCP_TIME_WAIT) {
 485		inet_twsk_put(inet_twsk(sk));
 486		return 0;
 487	}
 488	seq = ntohl(th->seq);
 489	if (sk->sk_state == TCP_NEW_SYN_RECV) {
 490		tcp_req_err(sk, seq, type == ICMP_PARAMETERPROB ||
 491				     type == ICMP_TIME_EXCEEDED ||
 492				     (type == ICMP_DEST_UNREACH &&
 493				      (code == ICMP_NET_UNREACH ||
 494				       code == ICMP_HOST_UNREACH)));
 495		return 0;
 496	}
 497
 498	bh_lock_sock(sk);
 499	/* If too many ICMPs get dropped on busy
 500	 * servers this needs to be solved differently.
 501	 * We do take care of PMTU discovery (RFC1191) special case :
 502	 * we can receive locally generated ICMP messages while socket is held.
 503	 */
 504	if (sock_owned_by_user(sk)) {
 505		if (!(type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED))
 506			__NET_INC_STATS(net, LINUX_MIB_LOCKDROPPEDICMPS);
 507	}
 508	if (sk->sk_state == TCP_CLOSE)
 509		goto out;
 510
 511	if (unlikely(iph->ttl < inet_sk(sk)->min_ttl)) {
 512		__NET_INC_STATS(net, LINUX_MIB_TCPMINTTLDROP);
 513		goto out;
 514	}
 515
 516	tp = tcp_sk(sk);
 517	/* XXX (TFO) - tp->snd_una should be ISN (tcp_create_openreq_child() */
 518	fastopen = rcu_dereference(tp->fastopen_rsk);
 519	snd_una = fastopen ? tcp_rsk(fastopen)->snt_isn : tp->snd_una;
 520	if (sk->sk_state != TCP_LISTEN &&
 521	    !between(seq, snd_una, tp->snd_nxt)) {
 522		__NET_INC_STATS(net, LINUX_MIB_OUTOFWINDOWICMPS);
 523		goto out;
 524	}
 525
 526	switch (type) {
 527	case ICMP_REDIRECT:
 528		if (!sock_owned_by_user(sk))
 529			do_redirect(skb, sk);
 530		goto out;
 531	case ICMP_SOURCE_QUENCH:
 532		/* Just silently ignore these. */
 533		goto out;
 534	case ICMP_PARAMETERPROB:
 535		err = EPROTO;
 536		break;
 537	case ICMP_DEST_UNREACH:
 538		if (code > NR_ICMP_UNREACH)
 539			goto out;
 540
 541		if (code == ICMP_FRAG_NEEDED) { /* PMTU discovery (RFC1191) */
 542			/* We are not interested in TCP_LISTEN and open_requests
 543			 * (SYN-ACKs send out by Linux are always <576bytes so
 544			 * they should go through unfragmented).
 545			 */
 546			if (sk->sk_state == TCP_LISTEN)
 547				goto out;
 548
 549			tp->mtu_info = info;
 550			if (!sock_owned_by_user(sk)) {
 551				tcp_v4_mtu_reduced(sk);
 552			} else {
 553				if (!test_and_set_bit(TCP_MTU_REDUCED_DEFERRED, &sk->sk_tsq_flags))
 554					sock_hold(sk);
 555			}
 556			goto out;
 557		}
 558
 559		err = icmp_err_convert[code].errno;
 560		/* check if this ICMP message allows revert of backoff.
 561		 * (see RFC 6069)
 562		 */
 563		if (!fastopen &&
 564		    (code == ICMP_NET_UNREACH || code == ICMP_HOST_UNREACH))
 565			tcp_ld_RTO_revert(sk, seq);
 566		break;
 567	case ICMP_TIME_EXCEEDED:
 568		err = EHOSTUNREACH;
 569		break;
 570	default:
 571		goto out;
 572	}
 573
 574	switch (sk->sk_state) {
 575	case TCP_SYN_SENT:
 576	case TCP_SYN_RECV:
 577		/* Only in fast or simultaneous open. If a fast open socket is
 578		 * is already accepted it is treated as a connected one below.
 579		 */
 580		if (fastopen && !fastopen->sk)
 581			break;
 582
 583		ip_icmp_error(sk, skb, err, th->dest, info, (u8 *)th);
 584
 585		if (!sock_owned_by_user(sk)) {
 586			sk->sk_err = err;
 587
 588			sk->sk_error_report(sk);
 589
 590			tcp_done(sk);
 591		} else {
 592			sk->sk_err_soft = err;
 593		}
 594		goto out;
 595	}
 596
 597	/* If we've already connected we will keep trying
 598	 * until we time out, or the user gives up.
 599	 *
 600	 * rfc1122 4.2.3.9 allows to consider as hard errors
 601	 * only PROTO_UNREACH and PORT_UNREACH (well, FRAG_FAILED too,
 602	 * but it is obsoleted by pmtu discovery).
 603	 *
 604	 * Note, that in modern internet, where routing is unreliable
 605	 * and in each dark corner broken firewalls sit, sending random
 606	 * errors ordered by their masters even this two messages finally lose
 607	 * their original sense (even Linux sends invalid PORT_UNREACHs)
 608	 *
 609	 * Now we are in compliance with RFCs.
 610	 *							--ANK (980905)
 611	 */
 612
 613	inet = inet_sk(sk);
 614	if (!sock_owned_by_user(sk) && inet->recverr) {
 615		sk->sk_err = err;
 616		sk->sk_error_report(sk);
 617	} else	{ /* Only an error on timeout */
 618		sk->sk_err_soft = err;
 619	}
 620
 621out:
 622	bh_unlock_sock(sk);
 623	sock_put(sk);
 624	return 0;
 625}
 626
 627void __tcp_v4_send_check(struct sk_buff *skb, __be32 saddr, __be32 daddr)
 628{
 629	struct tcphdr *th = tcp_hdr(skb);
 630
 631	th->check = ~tcp_v4_check(skb->len, saddr, daddr, 0);
 632	skb->csum_start = skb_transport_header(skb) - skb->head;
 633	skb->csum_offset = offsetof(struct tcphdr, check);
 634}
 635
 636/* This routine computes an IPv4 TCP checksum. */
 637void tcp_v4_send_check(struct sock *sk, struct sk_buff *skb)
 638{
 639	const struct inet_sock *inet = inet_sk(sk);
 640
 641	__tcp_v4_send_check(skb, inet->inet_saddr, inet->inet_daddr);
 642}
 643EXPORT_SYMBOL(tcp_v4_send_check);
 644
 645/*
 646 *	This routine will send an RST to the other tcp.
 647 *
 648 *	Someone asks: why I NEVER use socket parameters (TOS, TTL etc.)
 649 *		      for reset.
 650 *	Answer: if a packet caused RST, it is not for a socket
 651 *		existing in our system, if it is matched to a socket,
 652 *		it is just duplicate segment or bug in other side's TCP.
 653 *		So that we build reply only basing on parameters
 654 *		arrived with segment.
 655 *	Exception: precedence violation. We do not implement it in any case.
 656 */
 657
 
 
 
 
 
 
 658static void tcp_v4_send_reset(const struct sock *sk, struct sk_buff *skb)
 659{
 660	const struct tcphdr *th = tcp_hdr(skb);
 661	struct {
 662		struct tcphdr th;
 663#ifdef CONFIG_TCP_MD5SIG
 664		__be32 opt[(TCPOLEN_MD5SIG_ALIGNED >> 2)];
 665#endif
 666	} rep;
 667	struct ip_reply_arg arg;
 668#ifdef CONFIG_TCP_MD5SIG
 669	struct tcp_md5sig_key *key = NULL;
 670	const __u8 *hash_location = NULL;
 671	unsigned char newhash[16];
 672	int genhash;
 673	struct sock *sk1 = NULL;
 674#endif
 675	u64 transmit_time = 0;
 676	struct sock *ctl_sk;
 677	struct net *net;
 678
 679	/* Never send a reset in response to a reset. */
 680	if (th->rst)
 681		return;
 682
 683	/* If sk not NULL, it means we did a successful lookup and incoming
 684	 * route had to be correct. prequeue might have dropped our dst.
 685	 */
 686	if (!sk && skb_rtable(skb)->rt_type != RTN_LOCAL)
 687		return;
 688
 689	/* Swap the send and the receive. */
 690	memset(&rep, 0, sizeof(rep));
 691	rep.th.dest   = th->source;
 692	rep.th.source = th->dest;
 693	rep.th.doff   = sizeof(struct tcphdr) / 4;
 694	rep.th.rst    = 1;
 695
 696	if (th->ack) {
 697		rep.th.seq = th->ack_seq;
 698	} else {
 699		rep.th.ack = 1;
 700		rep.th.ack_seq = htonl(ntohl(th->seq) + th->syn + th->fin +
 701				       skb->len - (th->doff << 2));
 702	}
 703
 704	memset(&arg, 0, sizeof(arg));
 705	arg.iov[0].iov_base = (unsigned char *)&rep;
 706	arg.iov[0].iov_len  = sizeof(rep.th);
 707
 708	net = sk ? sock_net(sk) : dev_net(skb_dst(skb)->dev);
 709#ifdef CONFIG_TCP_MD5SIG
 710	rcu_read_lock();
 711	hash_location = tcp_parse_md5sig_option(th);
 712	if (sk && sk_fullsock(sk)) {
 713		const union tcp_md5_addr *addr;
 714		int l3index;
 715
 716		/* sdif set, means packet ingressed via a device
 717		 * in an L3 domain and inet_iif is set to it.
 718		 */
 719		l3index = tcp_v4_sdif(skb) ? inet_iif(skb) : 0;
 720		addr = (union tcp_md5_addr *)&ip_hdr(skb)->saddr;
 721		key = tcp_md5_do_lookup(sk, l3index, addr, AF_INET);
 722	} else if (hash_location) {
 723		const union tcp_md5_addr *addr;
 724		int sdif = tcp_v4_sdif(skb);
 725		int dif = inet_iif(skb);
 726		int l3index;
 727
 728		/*
 729		 * active side is lost. Try to find listening socket through
 730		 * source port, and then find md5 key through listening socket.
 731		 * we are not loose security here:
 732		 * Incoming packet is checked with md5 hash with finding key,
 733		 * no RST generated if md5 hash doesn't match.
 734		 */
 735		sk1 = __inet_lookup_listener(net, &tcp_hashinfo, NULL, 0,
 736					     ip_hdr(skb)->saddr,
 737					     th->source, ip_hdr(skb)->daddr,
 738					     ntohs(th->source), dif, sdif);
 739		/* don't send rst if it can't find key */
 740		if (!sk1)
 741			goto out;
 742
 743		/* sdif set, means packet ingressed via a device
 744		 * in an L3 domain and dif is set to it.
 745		 */
 746		l3index = sdif ? dif : 0;
 747		addr = (union tcp_md5_addr *)&ip_hdr(skb)->saddr;
 748		key = tcp_md5_do_lookup(sk1, l3index, addr, AF_INET);
 749		if (!key)
 750			goto out;
 751
 752
 753		genhash = tcp_v4_md5_hash_skb(newhash, key, NULL, skb);
 754		if (genhash || memcmp(hash_location, newhash, 16) != 0)
 755			goto out;
 756
 757	}
 758
 759	if (key) {
 760		rep.opt[0] = htonl((TCPOPT_NOP << 24) |
 761				   (TCPOPT_NOP << 16) |
 762				   (TCPOPT_MD5SIG << 8) |
 763				   TCPOLEN_MD5SIG);
 764		/* Update length and the length the header thinks exists */
 765		arg.iov[0].iov_len += TCPOLEN_MD5SIG_ALIGNED;
 766		rep.th.doff = arg.iov[0].iov_len / 4;
 767
 768		tcp_v4_md5_hash_hdr((__u8 *) &rep.opt[1],
 769				     key, ip_hdr(skb)->saddr,
 770				     ip_hdr(skb)->daddr, &rep.th);
 771	}
 772#endif
 
 
 
 
 
 
 
 
 
 
 
 773	arg.csum = csum_tcpudp_nofold(ip_hdr(skb)->daddr,
 774				      ip_hdr(skb)->saddr, /* XXX */
 775				      arg.iov[0].iov_len, IPPROTO_TCP, 0);
 776	arg.csumoffset = offsetof(struct tcphdr, check) / 2;
 777	arg.flags = (sk && inet_sk_transparent(sk)) ? IP_REPLY_ARG_NOSRCCHECK : 0;
 778
 779	/* When socket is gone, all binding information is lost.
 780	 * routing might fail in this case. No choice here, if we choose to force
 781	 * input interface, we will misroute in case of asymmetric route.
 782	 */
 783	if (sk) {
 784		arg.bound_dev_if = sk->sk_bound_dev_if;
 785		if (sk_fullsock(sk))
 786			trace_tcp_send_reset(sk, skb);
 787	}
 788
 789	BUILD_BUG_ON(offsetof(struct sock, sk_bound_dev_if) !=
 790		     offsetof(struct inet_timewait_sock, tw_bound_dev_if));
 791
 792	arg.tos = ip_hdr(skb)->tos;
 793	arg.uid = sock_net_uid(net, sk && sk_fullsock(sk) ? sk : NULL);
 794	local_bh_disable();
 795	ctl_sk = this_cpu_read(*net->ipv4.tcp_sk);
 796	if (sk) {
 797		ctl_sk->sk_mark = (sk->sk_state == TCP_TIME_WAIT) ?
 798				   inet_twsk(sk)->tw_mark : sk->sk_mark;
 799		ctl_sk->sk_priority = (sk->sk_state == TCP_TIME_WAIT) ?
 800				   inet_twsk(sk)->tw_priority : sk->sk_priority;
 801		transmit_time = tcp_transmit_time(sk);
 802	}
 803	ip_send_unicast_reply(ctl_sk,
 804			      skb, &TCP_SKB_CB(skb)->header.h4.opt,
 805			      ip_hdr(skb)->saddr, ip_hdr(skb)->daddr,
 806			      &arg, arg.iov[0].iov_len,
 807			      transmit_time);
 808
 809	ctl_sk->sk_mark = 0;
 810	__TCP_INC_STATS(net, TCP_MIB_OUTSEGS);
 811	__TCP_INC_STATS(net, TCP_MIB_OUTRSTS);
 812	local_bh_enable();
 813
 814#ifdef CONFIG_TCP_MD5SIG
 815out:
 816	rcu_read_unlock();
 817#endif
 818}
 819
 820/* The code following below sending ACKs in SYN-RECV and TIME-WAIT states
 821   outside socket context is ugly, certainly. What can I do?
 822 */
 823
 824static void tcp_v4_send_ack(const struct sock *sk,
 825			    struct sk_buff *skb, u32 seq, u32 ack,
 826			    u32 win, u32 tsval, u32 tsecr, int oif,
 827			    struct tcp_md5sig_key *key,
 828			    int reply_flags, u8 tos)
 829{
 830	const struct tcphdr *th = tcp_hdr(skb);
 831	struct {
 832		struct tcphdr th;
 833		__be32 opt[(TCPOLEN_TSTAMP_ALIGNED >> 2)
 834#ifdef CONFIG_TCP_MD5SIG
 835			   + (TCPOLEN_MD5SIG_ALIGNED >> 2)
 836#endif
 837			];
 838	} rep;
 839	struct net *net = sock_net(sk);
 840	struct ip_reply_arg arg;
 841	struct sock *ctl_sk;
 842	u64 transmit_time;
 843
 844	memset(&rep.th, 0, sizeof(struct tcphdr));
 845	memset(&arg, 0, sizeof(arg));
 846
 847	arg.iov[0].iov_base = (unsigned char *)&rep;
 848	arg.iov[0].iov_len  = sizeof(rep.th);
 849	if (tsecr) {
 850		rep.opt[0] = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
 851				   (TCPOPT_TIMESTAMP << 8) |
 852				   TCPOLEN_TIMESTAMP);
 853		rep.opt[1] = htonl(tsval);
 854		rep.opt[2] = htonl(tsecr);
 855		arg.iov[0].iov_len += TCPOLEN_TSTAMP_ALIGNED;
 856	}
 857
 858	/* Swap the send and the receive. */
 859	rep.th.dest    = th->source;
 860	rep.th.source  = th->dest;
 861	rep.th.doff    = arg.iov[0].iov_len / 4;
 862	rep.th.seq     = htonl(seq);
 863	rep.th.ack_seq = htonl(ack);
 864	rep.th.ack     = 1;
 865	rep.th.window  = htons(win);
 866
 867#ifdef CONFIG_TCP_MD5SIG
 868	if (key) {
 869		int offset = (tsecr) ? 3 : 0;
 870
 871		rep.opt[offset++] = htonl((TCPOPT_NOP << 24) |
 872					  (TCPOPT_NOP << 16) |
 873					  (TCPOPT_MD5SIG << 8) |
 874					  TCPOLEN_MD5SIG);
 875		arg.iov[0].iov_len += TCPOLEN_MD5SIG_ALIGNED;
 876		rep.th.doff = arg.iov[0].iov_len/4;
 877
 878		tcp_v4_md5_hash_hdr((__u8 *) &rep.opt[offset],
 879				    key, ip_hdr(skb)->saddr,
 880				    ip_hdr(skb)->daddr, &rep.th);
 881	}
 882#endif
 883	arg.flags = reply_flags;
 884	arg.csum = csum_tcpudp_nofold(ip_hdr(skb)->daddr,
 885				      ip_hdr(skb)->saddr, /* XXX */
 886				      arg.iov[0].iov_len, IPPROTO_TCP, 0);
 887	arg.csumoffset = offsetof(struct tcphdr, check) / 2;
 888	if (oif)
 889		arg.bound_dev_if = oif;
 890	arg.tos = tos;
 891	arg.uid = sock_net_uid(net, sk_fullsock(sk) ? sk : NULL);
 892	local_bh_disable();
 893	ctl_sk = this_cpu_read(*net->ipv4.tcp_sk);
 894	ctl_sk->sk_mark = (sk->sk_state == TCP_TIME_WAIT) ?
 895			   inet_twsk(sk)->tw_mark : sk->sk_mark;
 896	ctl_sk->sk_priority = (sk->sk_state == TCP_TIME_WAIT) ?
 897			   inet_twsk(sk)->tw_priority : sk->sk_priority;
 898	transmit_time = tcp_transmit_time(sk);
 899	ip_send_unicast_reply(ctl_sk,
 900			      skb, &TCP_SKB_CB(skb)->header.h4.opt,
 901			      ip_hdr(skb)->saddr, ip_hdr(skb)->daddr,
 902			      &arg, arg.iov[0].iov_len,
 903			      transmit_time);
 904
 905	ctl_sk->sk_mark = 0;
 906	__TCP_INC_STATS(net, TCP_MIB_OUTSEGS);
 907	local_bh_enable();
 908}
 909
 910static void tcp_v4_timewait_ack(struct sock *sk, struct sk_buff *skb)
 911{
 912	struct inet_timewait_sock *tw = inet_twsk(sk);
 913	struct tcp_timewait_sock *tcptw = tcp_twsk(sk);
 914
 915	tcp_v4_send_ack(sk, skb,
 916			tcptw->tw_snd_nxt, tcptw->tw_rcv_nxt,
 917			tcptw->tw_rcv_wnd >> tw->tw_rcv_wscale,
 918			tcp_time_stamp_raw() + tcptw->tw_ts_offset,
 919			tcptw->tw_ts_recent,
 920			tw->tw_bound_dev_if,
 921			tcp_twsk_md5_key(tcptw),
 922			tw->tw_transparent ? IP_REPLY_ARG_NOSRCCHECK : 0,
 923			tw->tw_tos
 924			);
 925
 926	inet_twsk_put(tw);
 927}
 928
 929static void tcp_v4_reqsk_send_ack(const struct sock *sk, struct sk_buff *skb,
 930				  struct request_sock *req)
 931{
 932	const union tcp_md5_addr *addr;
 933	int l3index;
 934
 935	/* sk->sk_state == TCP_LISTEN -> for regular TCP_SYN_RECV
 936	 * sk->sk_state == TCP_SYN_RECV -> for Fast Open.
 937	 */
 938	u32 seq = (sk->sk_state == TCP_LISTEN) ? tcp_rsk(req)->snt_isn + 1 :
 939					     tcp_sk(sk)->snd_nxt;
 940
 941	/* RFC 7323 2.3
 942	 * The window field (SEG.WND) of every outgoing segment, with the
 943	 * exception of <SYN> segments, MUST be right-shifted by
 944	 * Rcv.Wind.Shift bits:
 945	 */
 946	addr = (union tcp_md5_addr *)&ip_hdr(skb)->saddr;
 947	l3index = tcp_v4_sdif(skb) ? inet_iif(skb) : 0;
 948	tcp_v4_send_ack(sk, skb, seq,
 949			tcp_rsk(req)->rcv_nxt,
 950			req->rsk_rcv_wnd >> inet_rsk(req)->rcv_wscale,
 951			tcp_time_stamp_raw() + tcp_rsk(req)->ts_off,
 952			req->ts_recent,
 953			0,
 954			tcp_md5_do_lookup(sk, l3index, addr, AF_INET),
 955			inet_rsk(req)->no_srccheck ? IP_REPLY_ARG_NOSRCCHECK : 0,
 956			ip_hdr(skb)->tos);
 957}
 958
 959/*
 960 *	Send a SYN-ACK after having received a SYN.
 961 *	This still operates on a request_sock only, not on a big
 962 *	socket.
 963 */
 964static int tcp_v4_send_synack(const struct sock *sk, struct dst_entry *dst,
 965			      struct flowi *fl,
 966			      struct request_sock *req,
 967			      struct tcp_fastopen_cookie *foc,
 968			      enum tcp_synack_type synack_type)
 
 969{
 970	const struct inet_request_sock *ireq = inet_rsk(req);
 971	struct flowi4 fl4;
 972	int err = -1;
 973	struct sk_buff *skb;
 
 974
 975	/* First, grab a route. */
 976	if (!dst && (dst = inet_csk_route_req(sk, &fl4, req)) == NULL)
 977		return -1;
 978
 979	skb = tcp_make_synack(sk, dst, req, foc, synack_type);
 980
 981	if (skb) {
 982		__tcp_v4_send_check(skb, ireq->ir_loc_addr, ireq->ir_rmt_addr);
 983
 
 
 
 
 
 
 
 
 
 984		rcu_read_lock();
 985		err = ip_build_and_send_pkt(skb, sk, ireq->ir_loc_addr,
 986					    ireq->ir_rmt_addr,
 987					    rcu_dereference(ireq->ireq_opt));
 
 988		rcu_read_unlock();
 989		err = net_xmit_eval(err);
 990	}
 991
 992	return err;
 993}
 994
 995/*
 996 *	IPv4 request_sock destructor.
 997 */
 998static void tcp_v4_reqsk_destructor(struct request_sock *req)
 999{
1000	kfree(rcu_dereference_protected(inet_rsk(req)->ireq_opt, 1));
1001}
1002
1003#ifdef CONFIG_TCP_MD5SIG
1004/*
1005 * RFC2385 MD5 checksumming requires a mapping of
1006 * IP address->MD5 Key.
1007 * We need to maintain these in the sk structure.
1008 */
1009
1010DEFINE_STATIC_KEY_FALSE(tcp_md5_needed);
1011EXPORT_SYMBOL(tcp_md5_needed);
1012
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1013/* Find the Key structure for an address.  */
1014struct tcp_md5sig_key *__tcp_md5_do_lookup(const struct sock *sk, int l3index,
1015					   const union tcp_md5_addr *addr,
1016					   int family)
1017{
1018	const struct tcp_sock *tp = tcp_sk(sk);
1019	struct tcp_md5sig_key *key;
1020	const struct tcp_md5sig_info *md5sig;
1021	__be32 mask;
1022	struct tcp_md5sig_key *best_match = NULL;
1023	bool match;
1024
1025	/* caller either holds rcu_read_lock() or socket lock */
1026	md5sig = rcu_dereference_check(tp->md5sig_info,
1027				       lockdep_sock_is_held(sk));
1028	if (!md5sig)
1029		return NULL;
1030
1031	hlist_for_each_entry_rcu(key, &md5sig->head, node,
1032				 lockdep_sock_is_held(sk)) {
1033		if (key->family != family)
1034			continue;
1035		if (key->l3index && key->l3index != l3index)
1036			continue;
1037		if (family == AF_INET) {
1038			mask = inet_make_mask(key->prefixlen);
1039			match = (key->addr.a4.s_addr & mask) ==
1040				(addr->a4.s_addr & mask);
1041#if IS_ENABLED(CONFIG_IPV6)
1042		} else if (family == AF_INET6) {
1043			match = ipv6_prefix_equal(&key->addr.a6, &addr->a6,
1044						  key->prefixlen);
1045#endif
1046		} else {
1047			match = false;
1048		}
1049
1050		if (match && (!best_match ||
1051			      key->prefixlen > best_match->prefixlen))
1052			best_match = key;
1053	}
1054	return best_match;
1055}
1056EXPORT_SYMBOL(__tcp_md5_do_lookup);
1057
1058static struct tcp_md5sig_key *tcp_md5_do_lookup_exact(const struct sock *sk,
1059						      const union tcp_md5_addr *addr,
1060						      int family, u8 prefixlen,
1061						      int l3index)
1062{
1063	const struct tcp_sock *tp = tcp_sk(sk);
1064	struct tcp_md5sig_key *key;
1065	unsigned int size = sizeof(struct in_addr);
1066	const struct tcp_md5sig_info *md5sig;
1067
1068	/* caller either holds rcu_read_lock() or socket lock */
1069	md5sig = rcu_dereference_check(tp->md5sig_info,
1070				       lockdep_sock_is_held(sk));
1071	if (!md5sig)
1072		return NULL;
1073#if IS_ENABLED(CONFIG_IPV6)
1074	if (family == AF_INET6)
1075		size = sizeof(struct in6_addr);
1076#endif
1077	hlist_for_each_entry_rcu(key, &md5sig->head, node,
1078				 lockdep_sock_is_held(sk)) {
1079		if (key->family != family)
1080			continue;
1081		if (key->l3index && key->l3index != l3index)
1082			continue;
1083		if (!memcmp(&key->addr, addr, size) &&
1084		    key->prefixlen == prefixlen)
1085			return key;
1086	}
1087	return NULL;
1088}
1089
1090struct tcp_md5sig_key *tcp_v4_md5_lookup(const struct sock *sk,
1091					 const struct sock *addr_sk)
1092{
1093	const union tcp_md5_addr *addr;
1094	int l3index;
1095
1096	l3index = l3mdev_master_ifindex_by_index(sock_net(sk),
1097						 addr_sk->sk_bound_dev_if);
1098	addr = (const union tcp_md5_addr *)&addr_sk->sk_daddr;
1099	return tcp_md5_do_lookup(sk, l3index, addr, AF_INET);
1100}
1101EXPORT_SYMBOL(tcp_v4_md5_lookup);
1102
1103/* This can be called on a newly created socket, from other files */
1104int tcp_md5_do_add(struct sock *sk, const union tcp_md5_addr *addr,
1105		   int family, u8 prefixlen, int l3index,
1106		   const u8 *newkey, u8 newkeylen, gfp_t gfp)
1107{
1108	/* Add Key to the list */
1109	struct tcp_md5sig_key *key;
1110	struct tcp_sock *tp = tcp_sk(sk);
1111	struct tcp_md5sig_info *md5sig;
1112
1113	key = tcp_md5_do_lookup_exact(sk, addr, family, prefixlen, l3index);
1114	if (key) {
1115		/* Pre-existing entry - just update that one.
1116		 * Note that the key might be used concurrently.
1117		 * data_race() is telling kcsan that we do not care of
1118		 * key mismatches, since changing MD5 key on live flows
1119		 * can lead to packet drops.
1120		 */
1121		data_race(memcpy(key->key, newkey, newkeylen));
1122
1123		/* Pairs with READ_ONCE() in tcp_md5_hash_key().
1124		 * Also note that a reader could catch new key->keylen value
1125		 * but old key->key[], this is the reason we use __GFP_ZERO
1126		 * at sock_kmalloc() time below these lines.
1127		 */
1128		WRITE_ONCE(key->keylen, newkeylen);
1129
1130		return 0;
1131	}
1132
1133	md5sig = rcu_dereference_protected(tp->md5sig_info,
1134					   lockdep_sock_is_held(sk));
1135	if (!md5sig) {
1136		md5sig = kmalloc(sizeof(*md5sig), gfp);
1137		if (!md5sig)
1138			return -ENOMEM;
1139
1140		sk_nocaps_add(sk, NETIF_F_GSO_MASK);
1141		INIT_HLIST_HEAD(&md5sig->head);
1142		rcu_assign_pointer(tp->md5sig_info, md5sig);
1143	}
1144
1145	key = sock_kmalloc(sk, sizeof(*key), gfp | __GFP_ZERO);
1146	if (!key)
1147		return -ENOMEM;
1148	if (!tcp_alloc_md5sig_pool()) {
1149		sock_kfree_s(sk, key, sizeof(*key));
1150		return -ENOMEM;
1151	}
1152
1153	memcpy(key->key, newkey, newkeylen);
1154	key->keylen = newkeylen;
1155	key->family = family;
1156	key->prefixlen = prefixlen;
1157	key->l3index = l3index;
1158	memcpy(&key->addr, addr,
1159	       (family == AF_INET6) ? sizeof(struct in6_addr) :
1160				      sizeof(struct in_addr));
1161	hlist_add_head_rcu(&key->node, &md5sig->head);
1162	return 0;
1163}
1164EXPORT_SYMBOL(tcp_md5_do_add);
1165
1166int tcp_md5_do_del(struct sock *sk, const union tcp_md5_addr *addr, int family,
1167		   u8 prefixlen, int l3index)
1168{
1169	struct tcp_md5sig_key *key;
1170
1171	key = tcp_md5_do_lookup_exact(sk, addr, family, prefixlen, l3index);
1172	if (!key)
1173		return -ENOENT;
1174	hlist_del_rcu(&key->node);
1175	atomic_sub(sizeof(*key), &sk->sk_omem_alloc);
1176	kfree_rcu(key, rcu);
1177	return 0;
1178}
1179EXPORT_SYMBOL(tcp_md5_do_del);
1180
1181static void tcp_clear_md5_list(struct sock *sk)
1182{
1183	struct tcp_sock *tp = tcp_sk(sk);
1184	struct tcp_md5sig_key *key;
1185	struct hlist_node *n;
1186	struct tcp_md5sig_info *md5sig;
1187
1188	md5sig = rcu_dereference_protected(tp->md5sig_info, 1);
1189
1190	hlist_for_each_entry_safe(key, n, &md5sig->head, node) {
1191		hlist_del_rcu(&key->node);
1192		atomic_sub(sizeof(*key), &sk->sk_omem_alloc);
1193		kfree_rcu(key, rcu);
1194	}
1195}
1196
1197static int tcp_v4_parse_md5_keys(struct sock *sk, int optname,
1198				 sockptr_t optval, int optlen)
1199{
1200	struct tcp_md5sig cmd;
1201	struct sockaddr_in *sin = (struct sockaddr_in *)&cmd.tcpm_addr;
1202	const union tcp_md5_addr *addr;
1203	u8 prefixlen = 32;
1204	int l3index = 0;
1205
1206	if (optlen < sizeof(cmd))
1207		return -EINVAL;
1208
1209	if (copy_from_sockptr(&cmd, optval, sizeof(cmd)))
1210		return -EFAULT;
1211
1212	if (sin->sin_family != AF_INET)
1213		return -EINVAL;
1214
1215	if (optname == TCP_MD5SIG_EXT &&
1216	    cmd.tcpm_flags & TCP_MD5SIG_FLAG_PREFIX) {
1217		prefixlen = cmd.tcpm_prefixlen;
1218		if (prefixlen > 32)
1219			return -EINVAL;
1220	}
1221
1222	if (optname == TCP_MD5SIG_EXT &&
1223	    cmd.tcpm_flags & TCP_MD5SIG_FLAG_IFINDEX) {
1224		struct net_device *dev;
1225
1226		rcu_read_lock();
1227		dev = dev_get_by_index_rcu(sock_net(sk), cmd.tcpm_ifindex);
1228		if (dev && netif_is_l3_master(dev))
1229			l3index = dev->ifindex;
1230
1231		rcu_read_unlock();
1232
1233		/* ok to reference set/not set outside of rcu;
1234		 * right now device MUST be an L3 master
1235		 */
1236		if (!dev || !l3index)
1237			return -EINVAL;
1238	}
1239
1240	addr = (union tcp_md5_addr *)&sin->sin_addr.s_addr;
1241
1242	if (!cmd.tcpm_keylen)
1243		return tcp_md5_do_del(sk, addr, AF_INET, prefixlen, l3index);
1244
1245	if (cmd.tcpm_keylen > TCP_MD5SIG_MAXKEYLEN)
1246		return -EINVAL;
1247
1248	return tcp_md5_do_add(sk, addr, AF_INET, prefixlen, l3index,
1249			      cmd.tcpm_key, cmd.tcpm_keylen, GFP_KERNEL);
1250}
1251
1252static int tcp_v4_md5_hash_headers(struct tcp_md5sig_pool *hp,
1253				   __be32 daddr, __be32 saddr,
1254				   const struct tcphdr *th, int nbytes)
1255{
1256	struct tcp4_pseudohdr *bp;
1257	struct scatterlist sg;
1258	struct tcphdr *_th;
1259
1260	bp = hp->scratch;
1261	bp->saddr = saddr;
1262	bp->daddr = daddr;
1263	bp->pad = 0;
1264	bp->protocol = IPPROTO_TCP;
1265	bp->len = cpu_to_be16(nbytes);
1266
1267	_th = (struct tcphdr *)(bp + 1);
1268	memcpy(_th, th, sizeof(*th));
1269	_th->check = 0;
1270
1271	sg_init_one(&sg, bp, sizeof(*bp) + sizeof(*th));
1272	ahash_request_set_crypt(hp->md5_req, &sg, NULL,
1273				sizeof(*bp) + sizeof(*th));
1274	return crypto_ahash_update(hp->md5_req);
1275}
1276
1277static int tcp_v4_md5_hash_hdr(char *md5_hash, const struct tcp_md5sig_key *key,
1278			       __be32 daddr, __be32 saddr, const struct tcphdr *th)
1279{
1280	struct tcp_md5sig_pool *hp;
1281	struct ahash_request *req;
1282
1283	hp = tcp_get_md5sig_pool();
1284	if (!hp)
1285		goto clear_hash_noput;
1286	req = hp->md5_req;
1287
1288	if (crypto_ahash_init(req))
1289		goto clear_hash;
1290	if (tcp_v4_md5_hash_headers(hp, daddr, saddr, th, th->doff << 2))
1291		goto clear_hash;
1292	if (tcp_md5_hash_key(hp, key))
1293		goto clear_hash;
1294	ahash_request_set_crypt(req, NULL, md5_hash, 0);
1295	if (crypto_ahash_final(req))
1296		goto clear_hash;
1297
1298	tcp_put_md5sig_pool();
1299	return 0;
1300
1301clear_hash:
1302	tcp_put_md5sig_pool();
1303clear_hash_noput:
1304	memset(md5_hash, 0, 16);
1305	return 1;
1306}
1307
1308int tcp_v4_md5_hash_skb(char *md5_hash, const struct tcp_md5sig_key *key,
1309			const struct sock *sk,
1310			const struct sk_buff *skb)
1311{
1312	struct tcp_md5sig_pool *hp;
1313	struct ahash_request *req;
1314	const struct tcphdr *th = tcp_hdr(skb);
1315	__be32 saddr, daddr;
1316
1317	if (sk) { /* valid for establish/request sockets */
1318		saddr = sk->sk_rcv_saddr;
1319		daddr = sk->sk_daddr;
1320	} else {
1321		const struct iphdr *iph = ip_hdr(skb);
1322		saddr = iph->saddr;
1323		daddr = iph->daddr;
1324	}
1325
1326	hp = tcp_get_md5sig_pool();
1327	if (!hp)
1328		goto clear_hash_noput;
1329	req = hp->md5_req;
1330
1331	if (crypto_ahash_init(req))
1332		goto clear_hash;
1333
1334	if (tcp_v4_md5_hash_headers(hp, daddr, saddr, th, skb->len))
1335		goto clear_hash;
1336	if (tcp_md5_hash_skb_data(hp, skb, th->doff << 2))
1337		goto clear_hash;
1338	if (tcp_md5_hash_key(hp, key))
1339		goto clear_hash;
1340	ahash_request_set_crypt(req, NULL, md5_hash, 0);
1341	if (crypto_ahash_final(req))
1342		goto clear_hash;
1343
1344	tcp_put_md5sig_pool();
1345	return 0;
1346
1347clear_hash:
1348	tcp_put_md5sig_pool();
1349clear_hash_noput:
1350	memset(md5_hash, 0, 16);
1351	return 1;
1352}
1353EXPORT_SYMBOL(tcp_v4_md5_hash_skb);
1354
1355#endif
1356
1357/* Called with rcu_read_lock() */
1358static bool tcp_v4_inbound_md5_hash(const struct sock *sk,
1359				    const struct sk_buff *skb,
1360				    int dif, int sdif)
1361{
1362#ifdef CONFIG_TCP_MD5SIG
1363	/*
1364	 * This gets called for each TCP segment that arrives
1365	 * so we want to be efficient.
1366	 * We have 3 drop cases:
1367	 * o No MD5 hash and one expected.
1368	 * o MD5 hash and we're not expecting one.
1369	 * o MD5 hash and its wrong.
1370	 */
1371	const __u8 *hash_location = NULL;
1372	struct tcp_md5sig_key *hash_expected;
1373	const struct iphdr *iph = ip_hdr(skb);
1374	const struct tcphdr *th = tcp_hdr(skb);
1375	const union tcp_md5_addr *addr;
1376	unsigned char newhash[16];
1377	int genhash, l3index;
1378
1379	/* sdif set, means packet ingressed via a device
1380	 * in an L3 domain and dif is set to the l3mdev
1381	 */
1382	l3index = sdif ? dif : 0;
1383
1384	addr = (union tcp_md5_addr *)&iph->saddr;
1385	hash_expected = tcp_md5_do_lookup(sk, l3index, addr, AF_INET);
1386	hash_location = tcp_parse_md5sig_option(th);
1387
1388	/* We've parsed the options - do we have a hash? */
1389	if (!hash_expected && !hash_location)
1390		return false;
1391
1392	if (hash_expected && !hash_location) {
1393		NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPMD5NOTFOUND);
1394		return true;
1395	}
1396
1397	if (!hash_expected && hash_location) {
1398		NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPMD5UNEXPECTED);
1399		return true;
1400	}
1401
1402	/* Okay, so this is hash_expected and hash_location -
1403	 * so we need to calculate the checksum.
1404	 */
1405	genhash = tcp_v4_md5_hash_skb(newhash,
1406				      hash_expected,
1407				      NULL, skb);
1408
1409	if (genhash || memcmp(hash_location, newhash, 16) != 0) {
1410		NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPMD5FAILURE);
1411		net_info_ratelimited("MD5 Hash failed for (%pI4, %d)->(%pI4, %d)%s L3 index %d\n",
1412				     &iph->saddr, ntohs(th->source),
1413				     &iph->daddr, ntohs(th->dest),
1414				     genhash ? " tcp_v4_calc_md5_hash failed"
1415				     : "", l3index);
1416		return true;
1417	}
1418	return false;
1419#endif
1420	return false;
1421}
1422
1423static void tcp_v4_init_req(struct request_sock *req,
1424			    const struct sock *sk_listener,
1425			    struct sk_buff *skb)
1426{
1427	struct inet_request_sock *ireq = inet_rsk(req);
1428	struct net *net = sock_net(sk_listener);
1429
1430	sk_rcv_saddr_set(req_to_sk(req), ip_hdr(skb)->daddr);
1431	sk_daddr_set(req_to_sk(req), ip_hdr(skb)->saddr);
1432	RCU_INIT_POINTER(ireq->ireq_opt, tcp_v4_save_options(net, skb));
1433}
1434
1435static struct dst_entry *tcp_v4_route_req(const struct sock *sk,
 
1436					  struct flowi *fl,
1437					  const struct request_sock *req)
1438{
 
 
 
 
 
1439	return inet_csk_route_req(sk, &fl->u.ip4, req);
1440}
1441
1442struct request_sock_ops tcp_request_sock_ops __read_mostly = {
1443	.family		=	PF_INET,
1444	.obj_size	=	sizeof(struct tcp_request_sock),
1445	.rtx_syn_ack	=	tcp_rtx_synack,
1446	.send_ack	=	tcp_v4_reqsk_send_ack,
1447	.destructor	=	tcp_v4_reqsk_destructor,
1448	.send_reset	=	tcp_v4_send_reset,
1449	.syn_ack_timeout =	tcp_syn_ack_timeout,
1450};
1451
1452const struct tcp_request_sock_ops tcp_request_sock_ipv4_ops = {
1453	.mss_clamp	=	TCP_MSS_DEFAULT,
1454#ifdef CONFIG_TCP_MD5SIG
1455	.req_md5_lookup	=	tcp_v4_md5_lookup,
1456	.calc_md5_hash	=	tcp_v4_md5_hash_skb,
1457#endif
1458	.init_req	=	tcp_v4_init_req,
1459#ifdef CONFIG_SYN_COOKIES
1460	.cookie_init_seq =	cookie_v4_init_sequence,
1461#endif
1462	.route_req	=	tcp_v4_route_req,
1463	.init_seq	=	tcp_v4_init_seq,
1464	.init_ts_off	=	tcp_v4_init_ts_off,
1465	.send_synack	=	tcp_v4_send_synack,
1466};
1467
1468int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb)
1469{
1470	/* Never answer to SYNs send to broadcast or multicast */
1471	if (skb_rtable(skb)->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))
1472		goto drop;
1473
1474	return tcp_conn_request(&tcp_request_sock_ops,
1475				&tcp_request_sock_ipv4_ops, sk, skb);
1476
1477drop:
1478	tcp_listendrop(sk);
1479	return 0;
1480}
1481EXPORT_SYMBOL(tcp_v4_conn_request);
1482
1483
1484/*
1485 * The three way handshake has completed - we got a valid synack -
1486 * now create the new socket.
1487 */
1488struct sock *tcp_v4_syn_recv_sock(const struct sock *sk, struct sk_buff *skb,
1489				  struct request_sock *req,
1490				  struct dst_entry *dst,
1491				  struct request_sock *req_unhash,
1492				  bool *own_req)
1493{
1494	struct inet_request_sock *ireq;
 
1495	struct inet_sock *newinet;
1496	struct tcp_sock *newtp;
1497	struct sock *newsk;
1498#ifdef CONFIG_TCP_MD5SIG
1499	const union tcp_md5_addr *addr;
1500	struct tcp_md5sig_key *key;
1501	int l3index;
1502#endif
1503	struct ip_options_rcu *inet_opt;
1504
1505	if (sk_acceptq_is_full(sk))
1506		goto exit_overflow;
1507
1508	newsk = tcp_create_openreq_child(sk, req, skb);
1509	if (!newsk)
1510		goto exit_nonewsk;
1511
1512	newsk->sk_gso_type = SKB_GSO_TCPV4;
1513	inet_sk_rx_dst_set(newsk, skb);
1514
1515	newtp		      = tcp_sk(newsk);
1516	newinet		      = inet_sk(newsk);
1517	ireq		      = inet_rsk(req);
1518	sk_daddr_set(newsk, ireq->ir_rmt_addr);
1519	sk_rcv_saddr_set(newsk, ireq->ir_loc_addr);
1520	newsk->sk_bound_dev_if = ireq->ir_iif;
1521	newinet->inet_saddr   = ireq->ir_loc_addr;
1522	inet_opt	      = rcu_dereference(ireq->ireq_opt);
1523	RCU_INIT_POINTER(newinet->inet_opt, inet_opt);
1524	newinet->mc_index     = inet_iif(skb);
1525	newinet->mc_ttl	      = ip_hdr(skb)->ttl;
1526	newinet->rcv_tos      = ip_hdr(skb)->tos;
1527	inet_csk(newsk)->icsk_ext_hdr_len = 0;
1528	if (inet_opt)
1529		inet_csk(newsk)->icsk_ext_hdr_len = inet_opt->opt.optlen;
1530	newinet->inet_id = prandom_u32();
1531
 
 
 
 
 
 
1532	if (!dst) {
1533		dst = inet_csk_route_child_sock(sk, newsk, req);
1534		if (!dst)
1535			goto put_and_exit;
1536	} else {
1537		/* syncookie case : see end of cookie_v4_check() */
1538	}
1539	sk_setup_caps(newsk, dst);
1540
1541	tcp_ca_openreq_child(newsk, dst);
1542
1543	tcp_sync_mss(newsk, dst_mtu(dst));
1544	newtp->advmss = tcp_mss_clamp(tcp_sk(sk), dst_metric_advmss(dst));
1545
1546	tcp_initialize_rcv_mss(newsk);
1547
1548#ifdef CONFIG_TCP_MD5SIG
1549	l3index = l3mdev_master_ifindex_by_index(sock_net(sk), ireq->ir_iif);
1550	/* Copy over the MD5 key from the original socket */
1551	addr = (union tcp_md5_addr *)&newinet->inet_daddr;
1552	key = tcp_md5_do_lookup(sk, l3index, addr, AF_INET);
1553	if (key) {
1554		/*
1555		 * We're using one, so create a matching key
1556		 * on the newsk structure. If we fail to get
1557		 * memory, then we end up not copying the key
1558		 * across. Shucks.
1559		 */
1560		tcp_md5_do_add(newsk, addr, AF_INET, 32, l3index,
1561			       key->key, key->keylen, GFP_ATOMIC);
1562		sk_nocaps_add(newsk, NETIF_F_GSO_MASK);
1563	}
1564#endif
1565
1566	if (__inet_inherit_port(sk, newsk) < 0)
1567		goto put_and_exit;
1568	*own_req = inet_ehash_nolisten(newsk, req_to_sk(req_unhash));
 
1569	if (likely(*own_req)) {
1570		tcp_move_syn(newtp, req);
1571		ireq->ireq_opt = NULL;
1572	} else {
1573		newinet->inet_opt = NULL;
 
 
 
 
 
 
 
 
 
1574	}
1575	return newsk;
1576
1577exit_overflow:
1578	NET_INC_STATS(sock_net(sk), LINUX_MIB_LISTENOVERFLOWS);
1579exit_nonewsk:
1580	dst_release(dst);
1581exit:
1582	tcp_listendrop(sk);
1583	return NULL;
1584put_and_exit:
1585	newinet->inet_opt = NULL;
1586	inet_csk_prepare_forced_close(newsk);
1587	tcp_done(newsk);
1588	goto exit;
1589}
1590EXPORT_SYMBOL(tcp_v4_syn_recv_sock);
1591
1592static struct sock *tcp_v4_cookie_check(struct sock *sk, struct sk_buff *skb)
1593{
1594#ifdef CONFIG_SYN_COOKIES
1595	const struct tcphdr *th = tcp_hdr(skb);
1596
1597	if (!th->syn)
1598		sk = cookie_v4_check(sk, skb);
1599#endif
1600	return sk;
1601}
1602
1603u16 tcp_v4_get_syncookie(struct sock *sk, struct iphdr *iph,
1604			 struct tcphdr *th, u32 *cookie)
1605{
1606	u16 mss = 0;
1607#ifdef CONFIG_SYN_COOKIES
1608	mss = tcp_get_syncookie_mss(&tcp_request_sock_ops,
1609				    &tcp_request_sock_ipv4_ops, sk, th);
1610	if (mss) {
1611		*cookie = __cookie_v4_init_sequence(iph, th, &mss);
1612		tcp_synq_overflow(sk);
1613	}
1614#endif
1615	return mss;
1616}
1617
 
 
1618/* The socket must have it's spinlock held when we get
1619 * here, unless it is a TCP_LISTEN socket.
1620 *
1621 * We have a potential double-lock case here, so even when
1622 * doing backlog processing we use the BH locking scheme.
1623 * This is because we cannot sleep with the original spinlock
1624 * held.
1625 */
1626int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb)
1627{
1628	struct sock *rsk;
1629
1630	if (sk->sk_state == TCP_ESTABLISHED) { /* Fast path */
1631		struct dst_entry *dst = sk->sk_rx_dst;
1632
1633		sock_rps_save_rxhash(sk, skb);
1634		sk_mark_napi_id(sk, skb);
1635		if (dst) {
1636			if (inet_sk(sk)->rx_dst_ifindex != skb->skb_iif ||
1637			    !dst->ops->check(dst, 0)) {
 
1638				dst_release(dst);
1639				sk->sk_rx_dst = NULL;
1640			}
1641		}
1642		tcp_rcv_established(sk, skb);
1643		return 0;
1644	}
1645
1646	if (tcp_checksum_complete(skb))
1647		goto csum_err;
1648
1649	if (sk->sk_state == TCP_LISTEN) {
1650		struct sock *nsk = tcp_v4_cookie_check(sk, skb);
1651
1652		if (!nsk)
1653			goto discard;
1654		if (nsk != sk) {
1655			if (tcp_child_process(sk, nsk, skb)) {
1656				rsk = nsk;
1657				goto reset;
1658			}
1659			return 0;
1660		}
1661	} else
1662		sock_rps_save_rxhash(sk, skb);
1663
1664	if (tcp_rcv_state_process(sk, skb)) {
1665		rsk = sk;
1666		goto reset;
1667	}
1668	return 0;
1669
1670reset:
1671	tcp_v4_send_reset(rsk, skb);
1672discard:
1673	kfree_skb(skb);
1674	/* Be careful here. If this function gets more complicated and
1675	 * gcc suffers from register pressure on the x86, sk (in %ebx)
1676	 * might be destroyed here. This current version compiles correctly,
1677	 * but you have been warned.
1678	 */
1679	return 0;
1680
1681csum_err:
 
1682	TCP_INC_STATS(sock_net(sk), TCP_MIB_CSUMERRORS);
1683	TCP_INC_STATS(sock_net(sk), TCP_MIB_INERRS);
1684	goto discard;
1685}
1686EXPORT_SYMBOL(tcp_v4_do_rcv);
1687
1688int tcp_v4_early_demux(struct sk_buff *skb)
1689{
1690	const struct iphdr *iph;
1691	const struct tcphdr *th;
1692	struct sock *sk;
1693
1694	if (skb->pkt_type != PACKET_HOST)
1695		return 0;
1696
1697	if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct tcphdr)))
1698		return 0;
1699
1700	iph = ip_hdr(skb);
1701	th = tcp_hdr(skb);
1702
1703	if (th->doff < sizeof(struct tcphdr) / 4)
1704		return 0;
1705
1706	sk = __inet_lookup_established(dev_net(skb->dev), &tcp_hashinfo,
1707				       iph->saddr, th->source,
1708				       iph->daddr, ntohs(th->dest),
1709				       skb->skb_iif, inet_sdif(skb));
1710	if (sk) {
1711		skb->sk = sk;
1712		skb->destructor = sock_edemux;
1713		if (sk_fullsock(sk)) {
1714			struct dst_entry *dst = READ_ONCE(sk->sk_rx_dst);
1715
1716			if (dst)
1717				dst = dst_check(dst, 0);
1718			if (dst &&
1719			    inet_sk(sk)->rx_dst_ifindex == skb->skb_iif)
1720				skb_dst_set_noref(skb, dst);
1721		}
1722	}
1723	return 0;
1724}
1725
1726bool tcp_add_backlog(struct sock *sk, struct sk_buff *skb)
1727{
1728	u32 limit = READ_ONCE(sk->sk_rcvbuf) + READ_ONCE(sk->sk_sndbuf);
 
1729	struct skb_shared_info *shinfo;
1730	const struct tcphdr *th;
1731	struct tcphdr *thtail;
1732	struct sk_buff *tail;
1733	unsigned int hdrlen;
1734	bool fragstolen;
1735	u32 gso_segs;
 
1736	int delta;
1737
1738	/* In case all data was pulled from skb frags (in __pskb_pull_tail()),
1739	 * we can fix skb->truesize to its real value to avoid future drops.
1740	 * This is valid because skb is not yet charged to the socket.
1741	 * It has been noticed pure SACK packets were sometimes dropped
1742	 * (if cooked by drivers without copybreak feature).
1743	 */
1744	skb_condense(skb);
1745
1746	skb_dst_drop(skb);
1747
1748	if (unlikely(tcp_checksum_complete(skb))) {
1749		bh_unlock_sock(sk);
 
1750		__TCP_INC_STATS(sock_net(sk), TCP_MIB_CSUMERRORS);
1751		__TCP_INC_STATS(sock_net(sk), TCP_MIB_INERRS);
1752		return true;
1753	}
1754
1755	/* Attempt coalescing to last skb in backlog, even if we are
1756	 * above the limits.
1757	 * This is okay because skb capacity is limited to MAX_SKB_FRAGS.
1758	 */
1759	th = (const struct tcphdr *)skb->data;
1760	hdrlen = th->doff * 4;
1761	shinfo = skb_shinfo(skb);
1762
1763	if (!shinfo->gso_size)
1764		shinfo->gso_size = skb->len - hdrlen;
1765
1766	if (!shinfo->gso_segs)
1767		shinfo->gso_segs = 1;
1768
1769	tail = sk->sk_backlog.tail;
1770	if (!tail)
1771		goto no_coalesce;
1772	thtail = (struct tcphdr *)tail->data;
1773
1774	if (TCP_SKB_CB(tail)->end_seq != TCP_SKB_CB(skb)->seq ||
1775	    TCP_SKB_CB(tail)->ip_dsfield != TCP_SKB_CB(skb)->ip_dsfield ||
1776	    ((TCP_SKB_CB(tail)->tcp_flags |
1777	      TCP_SKB_CB(skb)->tcp_flags) & (TCPHDR_SYN | TCPHDR_RST | TCPHDR_URG)) ||
1778	    !((TCP_SKB_CB(tail)->tcp_flags &
1779	      TCP_SKB_CB(skb)->tcp_flags) & TCPHDR_ACK) ||
1780	    ((TCP_SKB_CB(tail)->tcp_flags ^
1781	      TCP_SKB_CB(skb)->tcp_flags) & (TCPHDR_ECE | TCPHDR_CWR)) ||
1782#ifdef CONFIG_TLS_DEVICE
1783	    tail->decrypted != skb->decrypted ||
1784#endif
1785	    thtail->doff != th->doff ||
1786	    memcmp(thtail + 1, th + 1, hdrlen - sizeof(*th)))
1787		goto no_coalesce;
1788
1789	__skb_pull(skb, hdrlen);
 
 
 
 
 
 
 
 
 
1790	if (skb_try_coalesce(tail, skb, &fragstolen, &delta)) {
1791		TCP_SKB_CB(tail)->end_seq = TCP_SKB_CB(skb)->end_seq;
1792
1793		if (likely(!before(TCP_SKB_CB(skb)->ack_seq, TCP_SKB_CB(tail)->ack_seq))) {
1794			TCP_SKB_CB(tail)->ack_seq = TCP_SKB_CB(skb)->ack_seq;
1795			thtail->window = th->window;
1796		}
1797
1798		/* We have to update both TCP_SKB_CB(tail)->tcp_flags and
1799		 * thtail->fin, so that the fast path in tcp_rcv_established()
1800		 * is not entered if we append a packet with a FIN.
1801		 * SYN, RST, URG are not present.
1802		 * ACK is set on both packets.
1803		 * PSH : we do not really care in TCP stack,
1804		 *       at least for 'GRO' packets.
1805		 */
1806		thtail->fin |= th->fin;
1807		TCP_SKB_CB(tail)->tcp_flags |= TCP_SKB_CB(skb)->tcp_flags;
1808
1809		if (TCP_SKB_CB(skb)->has_rxtstamp) {
1810			TCP_SKB_CB(tail)->has_rxtstamp = true;
1811			tail->tstamp = skb->tstamp;
1812			skb_hwtstamps(tail)->hwtstamp = skb_hwtstamps(skb)->hwtstamp;
1813		}
1814
1815		/* Not as strict as GRO. We only need to carry mss max value */
1816		skb_shinfo(tail)->gso_size = max(shinfo->gso_size,
1817						 skb_shinfo(tail)->gso_size);
1818
1819		gso_segs = skb_shinfo(tail)->gso_segs + shinfo->gso_segs;
1820		skb_shinfo(tail)->gso_segs = min_t(u32, gso_segs, 0xFFFF);
1821
1822		sk->sk_backlog.len += delta;
1823		__NET_INC_STATS(sock_net(sk),
1824				LINUX_MIB_TCPBACKLOGCOALESCE);
1825		kfree_skb_partial(skb, fragstolen);
1826		return false;
1827	}
1828	__skb_push(skb, hdrlen);
1829
1830no_coalesce:
1831	/* Only socket owner can try to collapse/prune rx queues
1832	 * to reduce memory overhead, so add a little headroom here.
1833	 * Few sockets backlog are possibly concurrently non empty.
1834	 */
1835	limit += 64*1024;
1836
1837	if (unlikely(sk_add_backlog(sk, skb, limit))) {
1838		bh_unlock_sock(sk);
1839		__NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPBACKLOGDROP);
1840		return true;
1841	}
1842	return false;
1843}
1844EXPORT_SYMBOL(tcp_add_backlog);
1845
1846int tcp_filter(struct sock *sk, struct sk_buff *skb)
1847{
1848	struct tcphdr *th = (struct tcphdr *)skb->data;
1849
1850	return sk_filter_trim_cap(sk, skb, th->doff * 4);
1851}
1852EXPORT_SYMBOL(tcp_filter);
1853
1854static void tcp_v4_restore_cb(struct sk_buff *skb)
1855{
1856	memmove(IPCB(skb), &TCP_SKB_CB(skb)->header.h4,
1857		sizeof(struct inet_skb_parm));
1858}
1859
1860static void tcp_v4_fill_cb(struct sk_buff *skb, const struct iphdr *iph,
1861			   const struct tcphdr *th)
1862{
1863	/* This is tricky : We move IPCB at its correct location into TCP_SKB_CB()
1864	 * barrier() makes sure compiler wont play fool^Waliasing games.
1865	 */
1866	memmove(&TCP_SKB_CB(skb)->header.h4, IPCB(skb),
1867		sizeof(struct inet_skb_parm));
1868	barrier();
1869
1870	TCP_SKB_CB(skb)->seq = ntohl(th->seq);
1871	TCP_SKB_CB(skb)->end_seq = (TCP_SKB_CB(skb)->seq + th->syn + th->fin +
1872				    skb->len - th->doff * 4);
1873	TCP_SKB_CB(skb)->ack_seq = ntohl(th->ack_seq);
1874	TCP_SKB_CB(skb)->tcp_flags = tcp_flag_byte(th);
1875	TCP_SKB_CB(skb)->tcp_tw_isn = 0;
1876	TCP_SKB_CB(skb)->ip_dsfield = ipv4_get_dsfield(iph);
1877	TCP_SKB_CB(skb)->sacked	 = 0;
1878	TCP_SKB_CB(skb)->has_rxtstamp =
1879			skb->tstamp || skb_hwtstamps(skb)->hwtstamp;
1880}
1881
1882/*
1883 *	From tcp_input.c
1884 */
1885
1886int tcp_v4_rcv(struct sk_buff *skb)
1887{
1888	struct net *net = dev_net(skb->dev);
1889	struct sk_buff *skb_to_free;
1890	int sdif = inet_sdif(skb);
1891	int dif = inet_iif(skb);
1892	const struct iphdr *iph;
1893	const struct tcphdr *th;
1894	bool refcounted;
1895	struct sock *sk;
1896	int ret;
1897
1898	if (skb->pkt_type != PACKET_HOST)
1899		goto discard_it;
1900
1901	/* Count it even if it's bad */
1902	__TCP_INC_STATS(net, TCP_MIB_INSEGS);
1903
1904	if (!pskb_may_pull(skb, sizeof(struct tcphdr)))
1905		goto discard_it;
1906
1907	th = (const struct tcphdr *)skb->data;
1908
1909	if (unlikely(th->doff < sizeof(struct tcphdr) / 4))
1910		goto bad_packet;
1911	if (!pskb_may_pull(skb, th->doff * 4))
1912		goto discard_it;
1913
1914	/* An explanation is required here, I think.
1915	 * Packet length and doff are validated by header prediction,
1916	 * provided case of th->doff==0 is eliminated.
1917	 * So, we defer the checks. */
1918
1919	if (skb_checksum_init(skb, IPPROTO_TCP, inet_compute_pseudo))
1920		goto csum_error;
1921
1922	th = (const struct tcphdr *)skb->data;
1923	iph = ip_hdr(skb);
1924lookup:
1925	sk = __inet_lookup_skb(&tcp_hashinfo, skb, __tcp_hdrlen(th), th->source,
1926			       th->dest, sdif, &refcounted);
1927	if (!sk)
1928		goto no_tcp_socket;
1929
1930process:
1931	if (sk->sk_state == TCP_TIME_WAIT)
1932		goto do_time_wait;
1933
1934	if (sk->sk_state == TCP_NEW_SYN_RECV) {
1935		struct request_sock *req = inet_reqsk(sk);
1936		bool req_stolen = false;
1937		struct sock *nsk;
1938
1939		sk = req->rsk_listener;
1940		if (unlikely(tcp_v4_inbound_md5_hash(sk, skb, dif, sdif))) {
1941			sk_drops_add(sk, skb);
1942			reqsk_put(req);
1943			goto discard_it;
1944		}
1945		if (tcp_checksum_complete(skb)) {
1946			reqsk_put(req);
1947			goto csum_error;
1948		}
1949		if (unlikely(sk->sk_state != TCP_LISTEN)) {
1950			inet_csk_reqsk_queue_drop_and_put(sk, req);
1951			goto lookup;
 
 
 
 
 
 
 
 
 
 
 
 
1952		}
1953		/* We own a reference on the listener, increase it again
1954		 * as we might lose it too soon.
1955		 */
1956		sock_hold(sk);
1957		refcounted = true;
1958		nsk = NULL;
1959		if (!tcp_filter(sk, skb)) {
1960			th = (const struct tcphdr *)skb->data;
1961			iph = ip_hdr(skb);
1962			tcp_v4_fill_cb(skb, iph, th);
1963			nsk = tcp_check_req(sk, skb, req, false, &req_stolen);
1964		}
1965		if (!nsk) {
1966			reqsk_put(req);
1967			if (req_stolen) {
1968				/* Another cpu got exclusive access to req
1969				 * and created a full blown socket.
1970				 * Try to feed this packet to this socket
1971				 * instead of discarding it.
1972				 */
1973				tcp_v4_restore_cb(skb);
1974				sock_put(sk);
1975				goto lookup;
1976			}
1977			goto discard_and_relse;
1978		}
1979		if (nsk == sk) {
1980			reqsk_put(req);
1981			tcp_v4_restore_cb(skb);
1982		} else if (tcp_child_process(sk, nsk, skb)) {
1983			tcp_v4_send_reset(nsk, skb);
1984			goto discard_and_relse;
1985		} else {
1986			sock_put(sk);
1987			return 0;
1988		}
1989	}
1990	if (unlikely(iph->ttl < inet_sk(sk)->min_ttl)) {
1991		__NET_INC_STATS(net, LINUX_MIB_TCPMINTTLDROP);
1992		goto discard_and_relse;
1993	}
1994
1995	if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
1996		goto discard_and_relse;
1997
1998	if (tcp_v4_inbound_md5_hash(sk, skb, dif, sdif))
1999		goto discard_and_relse;
2000
2001	nf_reset_ct(skb);
2002
2003	if (tcp_filter(sk, skb))
2004		goto discard_and_relse;
2005	th = (const struct tcphdr *)skb->data;
2006	iph = ip_hdr(skb);
2007	tcp_v4_fill_cb(skb, iph, th);
2008
2009	skb->dev = NULL;
2010
2011	if (sk->sk_state == TCP_LISTEN) {
2012		ret = tcp_v4_do_rcv(sk, skb);
2013		goto put_and_return;
2014	}
2015
2016	sk_incoming_cpu_update(sk);
2017
2018	bh_lock_sock_nested(sk);
2019	tcp_segs_in(tcp_sk(sk), skb);
2020	ret = 0;
2021	if (!sock_owned_by_user(sk)) {
2022		skb_to_free = sk->sk_rx_skb_cache;
2023		sk->sk_rx_skb_cache = NULL;
2024		ret = tcp_v4_do_rcv(sk, skb);
2025	} else {
2026		if (tcp_add_backlog(sk, skb))
2027			goto discard_and_relse;
2028		skb_to_free = NULL;
2029	}
2030	bh_unlock_sock(sk);
2031	if (skb_to_free)
2032		__kfree_skb(skb_to_free);
2033
2034put_and_return:
2035	if (refcounted)
2036		sock_put(sk);
2037
2038	return ret;
2039
2040no_tcp_socket:
2041	if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
2042		goto discard_it;
2043
2044	tcp_v4_fill_cb(skb, iph, th);
2045
2046	if (tcp_checksum_complete(skb)) {
2047csum_error:
 
2048		__TCP_INC_STATS(net, TCP_MIB_CSUMERRORS);
2049bad_packet:
2050		__TCP_INC_STATS(net, TCP_MIB_INERRS);
2051	} else {
2052		tcp_v4_send_reset(NULL, skb);
2053	}
2054
2055discard_it:
2056	/* Discard frame. */
2057	kfree_skb(skb);
2058	return 0;
2059
2060discard_and_relse:
2061	sk_drops_add(sk, skb);
2062	if (refcounted)
2063		sock_put(sk);
2064	goto discard_it;
2065
2066do_time_wait:
2067	if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) {
2068		inet_twsk_put(inet_twsk(sk));
2069		goto discard_it;
2070	}
2071
2072	tcp_v4_fill_cb(skb, iph, th);
2073
2074	if (tcp_checksum_complete(skb)) {
2075		inet_twsk_put(inet_twsk(sk));
2076		goto csum_error;
2077	}
2078	switch (tcp_timewait_state_process(inet_twsk(sk), skb, th)) {
2079	case TCP_TW_SYN: {
2080		struct sock *sk2 = inet_lookup_listener(dev_net(skb->dev),
2081							&tcp_hashinfo, skb,
2082							__tcp_hdrlen(th),
2083							iph->saddr, th->source,
2084							iph->daddr, th->dest,
2085							inet_iif(skb),
2086							sdif);
2087		if (sk2) {
2088			inet_twsk_deschedule_put(inet_twsk(sk));
2089			sk = sk2;
2090			tcp_v4_restore_cb(skb);
2091			refcounted = false;
2092			goto process;
2093		}
2094	}
2095		/* to ACK */
2096		fallthrough;
2097	case TCP_TW_ACK:
2098		tcp_v4_timewait_ack(sk, skb);
2099		break;
2100	case TCP_TW_RST:
2101		tcp_v4_send_reset(sk, skb);
2102		inet_twsk_deschedule_put(inet_twsk(sk));
2103		goto discard_it;
2104	case TCP_TW_SUCCESS:;
2105	}
2106	goto discard_it;
2107}
2108
2109static struct timewait_sock_ops tcp_timewait_sock_ops = {
2110	.twsk_obj_size	= sizeof(struct tcp_timewait_sock),
2111	.twsk_unique	= tcp_twsk_unique,
2112	.twsk_destructor= tcp_twsk_destructor,
2113};
2114
2115void inet_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb)
2116{
2117	struct dst_entry *dst = skb_dst(skb);
2118
2119	if (dst && dst_hold_safe(dst)) {
2120		sk->sk_rx_dst = dst;
2121		inet_sk(sk)->rx_dst_ifindex = skb->skb_iif;
2122	}
2123}
2124EXPORT_SYMBOL(inet_sk_rx_dst_set);
2125
2126const struct inet_connection_sock_af_ops ipv4_specific = {
2127	.queue_xmit	   = ip_queue_xmit,
2128	.send_check	   = tcp_v4_send_check,
2129	.rebuild_header	   = inet_sk_rebuild_header,
2130	.sk_rx_dst_set	   = inet_sk_rx_dst_set,
2131	.conn_request	   = tcp_v4_conn_request,
2132	.syn_recv_sock	   = tcp_v4_syn_recv_sock,
2133	.net_header_len	   = sizeof(struct iphdr),
2134	.setsockopt	   = ip_setsockopt,
2135	.getsockopt	   = ip_getsockopt,
2136	.addr2sockaddr	   = inet_csk_addr2sockaddr,
2137	.sockaddr_len	   = sizeof(struct sockaddr_in),
2138	.mtu_reduced	   = tcp_v4_mtu_reduced,
2139};
2140EXPORT_SYMBOL(ipv4_specific);
2141
2142#ifdef CONFIG_TCP_MD5SIG
2143static const struct tcp_sock_af_ops tcp_sock_ipv4_specific = {
2144	.md5_lookup		= tcp_v4_md5_lookup,
2145	.calc_md5_hash		= tcp_v4_md5_hash_skb,
2146	.md5_parse		= tcp_v4_parse_md5_keys,
2147};
2148#endif
2149
2150/* NOTE: A lot of things set to zero explicitly by call to
2151 *       sk_alloc() so need not be done here.
2152 */
2153static int tcp_v4_init_sock(struct sock *sk)
2154{
2155	struct inet_connection_sock *icsk = inet_csk(sk);
2156
2157	tcp_init_sock(sk);
2158
2159	icsk->icsk_af_ops = &ipv4_specific;
2160
2161#ifdef CONFIG_TCP_MD5SIG
2162	tcp_sk(sk)->af_specific = &tcp_sock_ipv4_specific;
2163#endif
2164
2165	return 0;
2166}
2167
2168void tcp_v4_destroy_sock(struct sock *sk)
2169{
2170	struct tcp_sock *tp = tcp_sk(sk);
2171
2172	trace_tcp_destroy_sock(sk);
2173
2174	tcp_clear_xmit_timers(sk);
2175
2176	tcp_cleanup_congestion_control(sk);
2177
2178	tcp_cleanup_ulp(sk);
2179
2180	/* Cleanup up the write buffer. */
2181	tcp_write_queue_purge(sk);
2182
2183	/* Check if we want to disable active TFO */
2184	tcp_fastopen_active_disable_ofo_check(sk);
2185
2186	/* Cleans up our, hopefully empty, out_of_order_queue. */
2187	skb_rbtree_purge(&tp->out_of_order_queue);
2188
2189#ifdef CONFIG_TCP_MD5SIG
2190	/* Clean up the MD5 key list, if any */
2191	if (tp->md5sig_info) {
2192		tcp_clear_md5_list(sk);
2193		kfree_rcu(rcu_dereference_protected(tp->md5sig_info, 1), rcu);
2194		tp->md5sig_info = NULL;
2195	}
2196#endif
2197
2198	/* Clean up a referenced TCP bind bucket. */
2199	if (inet_csk(sk)->icsk_bind_hash)
2200		inet_put_port(sk);
2201
2202	BUG_ON(rcu_access_pointer(tp->fastopen_rsk));
2203
2204	/* If socket is aborted during connect operation */
2205	tcp_free_fastopen_req(tp);
2206	tcp_fastopen_destroy_cipher(sk);
2207	tcp_saved_syn_free(tp);
2208
2209	sk_sockets_allocated_dec(sk);
2210}
2211EXPORT_SYMBOL(tcp_v4_destroy_sock);
2212
2213#ifdef CONFIG_PROC_FS
2214/* Proc filesystem TCP sock list dumping. */
2215
2216/*
2217 * Get next listener socket follow cur.  If cur is NULL, get first socket
2218 * starting from bucket given in st->bucket; when st->bucket is zero the
2219 * very first socket in the hash table is returned.
2220 */
2221static void *listening_get_next(struct seq_file *seq, void *cur)
2222{
2223	struct tcp_seq_afinfo *afinfo;
2224	struct tcp_iter_state *st = seq->private;
2225	struct net *net = seq_file_net(seq);
2226	struct inet_listen_hashbucket *ilb;
2227	struct hlist_nulls_node *node;
2228	struct sock *sk = cur;
2229
2230	if (st->bpf_seq_afinfo)
2231		afinfo = st->bpf_seq_afinfo;
2232	else
2233		afinfo = PDE_DATA(file_inode(seq->file));
2234
2235	if (!sk) {
2236get_head:
2237		ilb = &tcp_hashinfo.listening_hash[st->bucket];
2238		spin_lock(&ilb->lock);
2239		sk = sk_nulls_head(&ilb->nulls_head);
2240		st->offset = 0;
2241		goto get_sk;
2242	}
2243	ilb = &tcp_hashinfo.listening_hash[st->bucket];
2244	++st->num;
2245	++st->offset;
2246
2247	sk = sk_nulls_next(sk);
2248get_sk:
2249	sk_nulls_for_each_from(sk, node) {
2250		if (!net_eq(sock_net(sk), net))
2251			continue;
2252		if (afinfo->family == AF_UNSPEC ||
2253		    sk->sk_family == afinfo->family)
2254			return sk;
2255	}
2256	spin_unlock(&ilb->lock);
2257	st->offset = 0;
2258	if (++st->bucket < INET_LHTABLE_SIZE)
2259		goto get_head;
2260	return NULL;
2261}
2262
2263static void *listening_get_idx(struct seq_file *seq, loff_t *pos)
2264{
2265	struct tcp_iter_state *st = seq->private;
2266	void *rc;
2267
2268	st->bucket = 0;
2269	st->offset = 0;
2270	rc = listening_get_next(seq, NULL);
2271
2272	while (rc && *pos) {
2273		rc = listening_get_next(seq, rc);
2274		--*pos;
2275	}
2276	return rc;
2277}
2278
2279static inline bool empty_bucket(const struct tcp_iter_state *st)
2280{
2281	return hlist_nulls_empty(&tcp_hashinfo.ehash[st->bucket].chain);
2282}
2283
2284/*
2285 * Get first established socket starting from bucket given in st->bucket.
2286 * If st->bucket is zero, the very first socket in the hash is returned.
2287 */
2288static void *established_get_first(struct seq_file *seq)
2289{
2290	struct tcp_seq_afinfo *afinfo;
2291	struct tcp_iter_state *st = seq->private;
2292	struct net *net = seq_file_net(seq);
2293	void *rc = NULL;
2294
2295	if (st->bpf_seq_afinfo)
2296		afinfo = st->bpf_seq_afinfo;
2297	else
2298		afinfo = PDE_DATA(file_inode(seq->file));
2299
2300	st->offset = 0;
2301	for (; st->bucket <= tcp_hashinfo.ehash_mask; ++st->bucket) {
2302		struct sock *sk;
2303		struct hlist_nulls_node *node;
2304		spinlock_t *lock = inet_ehash_lockp(&tcp_hashinfo, st->bucket);
2305
2306		/* Lockless fast path for the common case of empty buckets */
2307		if (empty_bucket(st))
2308			continue;
2309
2310		spin_lock_bh(lock);
2311		sk_nulls_for_each(sk, node, &tcp_hashinfo.ehash[st->bucket].chain) {
2312			if ((afinfo->family != AF_UNSPEC &&
2313			     sk->sk_family != afinfo->family) ||
2314			    !net_eq(sock_net(sk), net)) {
2315				continue;
2316			}
2317			rc = sk;
2318			goto out;
2319		}
2320		spin_unlock_bh(lock);
2321	}
2322out:
2323	return rc;
2324}
2325
2326static void *established_get_next(struct seq_file *seq, void *cur)
2327{
2328	struct tcp_seq_afinfo *afinfo;
2329	struct sock *sk = cur;
2330	struct hlist_nulls_node *node;
2331	struct tcp_iter_state *st = seq->private;
2332	struct net *net = seq_file_net(seq);
2333
2334	if (st->bpf_seq_afinfo)
2335		afinfo = st->bpf_seq_afinfo;
2336	else
2337		afinfo = PDE_DATA(file_inode(seq->file));
2338
2339	++st->num;
2340	++st->offset;
2341
2342	sk = sk_nulls_next(sk);
2343
2344	sk_nulls_for_each_from(sk, node) {
2345		if ((afinfo->family == AF_UNSPEC ||
2346		     sk->sk_family == afinfo->family) &&
2347		    net_eq(sock_net(sk), net))
2348			return sk;
2349	}
2350
2351	spin_unlock_bh(inet_ehash_lockp(&tcp_hashinfo, st->bucket));
2352	++st->bucket;
2353	return established_get_first(seq);
2354}
2355
2356static void *established_get_idx(struct seq_file *seq, loff_t pos)
2357{
2358	struct tcp_iter_state *st = seq->private;
2359	void *rc;
2360
2361	st->bucket = 0;
2362	rc = established_get_first(seq);
2363
2364	while (rc && pos) {
2365		rc = established_get_next(seq, rc);
2366		--pos;
2367	}
2368	return rc;
2369}
2370
2371static void *tcp_get_idx(struct seq_file *seq, loff_t pos)
2372{
2373	void *rc;
2374	struct tcp_iter_state *st = seq->private;
2375
2376	st->state = TCP_SEQ_STATE_LISTENING;
2377	rc	  = listening_get_idx(seq, &pos);
2378
2379	if (!rc) {
2380		st->state = TCP_SEQ_STATE_ESTABLISHED;
2381		rc	  = established_get_idx(seq, pos);
2382	}
2383
2384	return rc;
2385}
2386
2387static void *tcp_seek_last_pos(struct seq_file *seq)
2388{
2389	struct tcp_iter_state *st = seq->private;
 
2390	int offset = st->offset;
2391	int orig_num = st->num;
2392	void *rc = NULL;
2393
2394	switch (st->state) {
2395	case TCP_SEQ_STATE_LISTENING:
2396		if (st->bucket >= INET_LHTABLE_SIZE)
2397			break;
2398		st->state = TCP_SEQ_STATE_LISTENING;
2399		rc = listening_get_next(seq, NULL);
2400		while (offset-- && rc)
2401			rc = listening_get_next(seq, rc);
2402		if (rc)
2403			break;
2404		st->bucket = 0;
2405		st->state = TCP_SEQ_STATE_ESTABLISHED;
2406		fallthrough;
2407	case TCP_SEQ_STATE_ESTABLISHED:
2408		if (st->bucket > tcp_hashinfo.ehash_mask)
2409			break;
2410		rc = established_get_first(seq);
2411		while (offset-- && rc)
2412			rc = established_get_next(seq, rc);
2413	}
2414
2415	st->num = orig_num;
2416
2417	return rc;
2418}
2419
2420void *tcp_seq_start(struct seq_file *seq, loff_t *pos)
2421{
2422	struct tcp_iter_state *st = seq->private;
2423	void *rc;
2424
2425	if (*pos && *pos == st->last_pos) {
2426		rc = tcp_seek_last_pos(seq);
2427		if (rc)
2428			goto out;
2429	}
2430
2431	st->state = TCP_SEQ_STATE_LISTENING;
2432	st->num = 0;
2433	st->bucket = 0;
2434	st->offset = 0;
2435	rc = *pos ? tcp_get_idx(seq, *pos - 1) : SEQ_START_TOKEN;
2436
2437out:
2438	st->last_pos = *pos;
2439	return rc;
2440}
2441EXPORT_SYMBOL(tcp_seq_start);
2442
2443void *tcp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2444{
2445	struct tcp_iter_state *st = seq->private;
2446	void *rc = NULL;
2447
2448	if (v == SEQ_START_TOKEN) {
2449		rc = tcp_get_idx(seq, 0);
2450		goto out;
2451	}
2452
2453	switch (st->state) {
2454	case TCP_SEQ_STATE_LISTENING:
2455		rc = listening_get_next(seq, v);
2456		if (!rc) {
2457			st->state = TCP_SEQ_STATE_ESTABLISHED;
2458			st->bucket = 0;
2459			st->offset = 0;
2460			rc	  = established_get_first(seq);
2461		}
2462		break;
2463	case TCP_SEQ_STATE_ESTABLISHED:
2464		rc = established_get_next(seq, v);
2465		break;
2466	}
2467out:
2468	++*pos;
2469	st->last_pos = *pos;
2470	return rc;
2471}
2472EXPORT_SYMBOL(tcp_seq_next);
2473
2474void tcp_seq_stop(struct seq_file *seq, void *v)
2475{
2476	struct tcp_iter_state *st = seq->private;
2477
2478	switch (st->state) {
2479	case TCP_SEQ_STATE_LISTENING:
2480		if (v != SEQ_START_TOKEN)
2481			spin_unlock(&tcp_hashinfo.listening_hash[st->bucket].lock);
2482		break;
2483	case TCP_SEQ_STATE_ESTABLISHED:
2484		if (v)
2485			spin_unlock_bh(inet_ehash_lockp(&tcp_hashinfo, st->bucket));
2486		break;
2487	}
2488}
2489EXPORT_SYMBOL(tcp_seq_stop);
2490
2491static void get_openreq4(const struct request_sock *req,
2492			 struct seq_file *f, int i)
2493{
2494	const struct inet_request_sock *ireq = inet_rsk(req);
2495	long delta = req->rsk_timer.expires - jiffies;
2496
2497	seq_printf(f, "%4d: %08X:%04X %08X:%04X"
2498		" %02X %08X:%08X %02X:%08lX %08X %5u %8d %u %d %pK",
2499		i,
2500		ireq->ir_loc_addr,
2501		ireq->ir_num,
2502		ireq->ir_rmt_addr,
2503		ntohs(ireq->ir_rmt_port),
2504		TCP_SYN_RECV,
2505		0, 0, /* could print option size, but that is af dependent. */
2506		1,    /* timers active (only the expire timer) */
2507		jiffies_delta_to_clock_t(delta),
2508		req->num_timeout,
2509		from_kuid_munged(seq_user_ns(f),
2510				 sock_i_uid(req->rsk_listener)),
2511		0,  /* non standard timer */
2512		0, /* open_requests have no inode */
2513		0,
2514		req);
2515}
2516
2517static void get_tcp4_sock(struct sock *sk, struct seq_file *f, int i)
2518{
2519	int timer_active;
2520	unsigned long timer_expires;
2521	const struct tcp_sock *tp = tcp_sk(sk);
2522	const struct inet_connection_sock *icsk = inet_csk(sk);
2523	const struct inet_sock *inet = inet_sk(sk);
2524	const struct fastopen_queue *fastopenq = &icsk->icsk_accept_queue.fastopenq;
2525	__be32 dest = inet->inet_daddr;
2526	__be32 src = inet->inet_rcv_saddr;
2527	__u16 destp = ntohs(inet->inet_dport);
2528	__u16 srcp = ntohs(inet->inet_sport);
2529	int rx_queue;
2530	int state;
2531
2532	if (icsk->icsk_pending == ICSK_TIME_RETRANS ||
2533	    icsk->icsk_pending == ICSK_TIME_REO_TIMEOUT ||
2534	    icsk->icsk_pending == ICSK_TIME_LOSS_PROBE) {
2535		timer_active	= 1;
2536		timer_expires	= icsk->icsk_timeout;
2537	} else if (icsk->icsk_pending == ICSK_TIME_PROBE0) {
2538		timer_active	= 4;
2539		timer_expires	= icsk->icsk_timeout;
2540	} else if (timer_pending(&sk->sk_timer)) {
2541		timer_active	= 2;
2542		timer_expires	= sk->sk_timer.expires;
2543	} else {
2544		timer_active	= 0;
2545		timer_expires = jiffies;
2546	}
2547
2548	state = inet_sk_state_load(sk);
2549	if (state == TCP_LISTEN)
2550		rx_queue = READ_ONCE(sk->sk_ack_backlog);
2551	else
2552		/* Because we don't lock the socket,
2553		 * we might find a transient negative value.
2554		 */
2555		rx_queue = max_t(int, READ_ONCE(tp->rcv_nxt) -
2556				      READ_ONCE(tp->copied_seq), 0);
2557
2558	seq_printf(f, "%4d: %08X:%04X %08X:%04X %02X %08X:%08X %02X:%08lX "
2559			"%08X %5u %8d %lu %d %pK %lu %lu %u %u %d",
2560		i, src, srcp, dest, destp, state,
2561		READ_ONCE(tp->write_seq) - tp->snd_una,
2562		rx_queue,
2563		timer_active,
2564		jiffies_delta_to_clock_t(timer_expires - jiffies),
2565		icsk->icsk_retransmits,
2566		from_kuid_munged(seq_user_ns(f), sock_i_uid(sk)),
2567		icsk->icsk_probes_out,
2568		sock_i_ino(sk),
2569		refcount_read(&sk->sk_refcnt), sk,
2570		jiffies_to_clock_t(icsk->icsk_rto),
2571		jiffies_to_clock_t(icsk->icsk_ack.ato),
2572		(icsk->icsk_ack.quick << 1) | inet_csk_in_pingpong_mode(sk),
2573		tp->snd_cwnd,
2574		state == TCP_LISTEN ?
2575		    fastopenq->max_qlen :
2576		    (tcp_in_initial_slowstart(tp) ? -1 : tp->snd_ssthresh));
2577}
2578
2579static void get_timewait4_sock(const struct inet_timewait_sock *tw,
2580			       struct seq_file *f, int i)
2581{
2582	long delta = tw->tw_timer.expires - jiffies;
2583	__be32 dest, src;
2584	__u16 destp, srcp;
2585
2586	dest  = tw->tw_daddr;
2587	src   = tw->tw_rcv_saddr;
2588	destp = ntohs(tw->tw_dport);
2589	srcp  = ntohs(tw->tw_sport);
2590
2591	seq_printf(f, "%4d: %08X:%04X %08X:%04X"
2592		" %02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %pK",
2593		i, src, srcp, dest, destp, tw->tw_substate, 0, 0,
2594		3, jiffies_delta_to_clock_t(delta), 0, 0, 0, 0,
2595		refcount_read(&tw->tw_refcnt), tw);
2596}
2597
2598#define TMPSZ 150
2599
2600static int tcp4_seq_show(struct seq_file *seq, void *v)
2601{
2602	struct tcp_iter_state *st;
2603	struct sock *sk = v;
2604
2605	seq_setwidth(seq, TMPSZ - 1);
2606	if (v == SEQ_START_TOKEN) {
2607		seq_puts(seq, "  sl  local_address rem_address   st tx_queue "
2608			   "rx_queue tr tm->when retrnsmt   uid  timeout "
2609			   "inode");
2610		goto out;
2611	}
2612	st = seq->private;
2613
2614	if (sk->sk_state == TCP_TIME_WAIT)
2615		get_timewait4_sock(v, seq, st->num);
2616	else if (sk->sk_state == TCP_NEW_SYN_RECV)
2617		get_openreq4(v, seq, st->num);
2618	else
2619		get_tcp4_sock(v, seq, st->num);
2620out:
2621	seq_pad(seq, '\n');
2622	return 0;
2623}
2624
2625#ifdef CONFIG_BPF_SYSCALL
2626struct bpf_iter__tcp {
2627	__bpf_md_ptr(struct bpf_iter_meta *, meta);
2628	__bpf_md_ptr(struct sock_common *, sk_common);
2629	uid_t uid __aligned(8);
2630};
2631
2632static int tcp_prog_seq_show(struct bpf_prog *prog, struct bpf_iter_meta *meta,
2633			     struct sock_common *sk_common, uid_t uid)
2634{
2635	struct bpf_iter__tcp ctx;
2636
2637	meta->seq_num--;  /* skip SEQ_START_TOKEN */
2638	ctx.meta = meta;
2639	ctx.sk_common = sk_common;
2640	ctx.uid = uid;
2641	return bpf_iter_run_prog(prog, &ctx);
2642}
2643
2644static int bpf_iter_tcp_seq_show(struct seq_file *seq, void *v)
2645{
2646	struct bpf_iter_meta meta;
2647	struct bpf_prog *prog;
2648	struct sock *sk = v;
2649	uid_t uid;
2650
2651	if (v == SEQ_START_TOKEN)
2652		return 0;
2653
2654	if (sk->sk_state == TCP_TIME_WAIT) {
2655		uid = 0;
2656	} else if (sk->sk_state == TCP_NEW_SYN_RECV) {
2657		const struct request_sock *req = v;
2658
2659		uid = from_kuid_munged(seq_user_ns(seq),
2660				       sock_i_uid(req->rsk_listener));
2661	} else {
2662		uid = from_kuid_munged(seq_user_ns(seq), sock_i_uid(sk));
2663	}
2664
2665	meta.seq = seq;
2666	prog = bpf_iter_get_info(&meta, false);
2667	return tcp_prog_seq_show(prog, &meta, v, uid);
2668}
2669
2670static void bpf_iter_tcp_seq_stop(struct seq_file *seq, void *v)
2671{
2672	struct bpf_iter_meta meta;
2673	struct bpf_prog *prog;
2674
2675	if (!v) {
2676		meta.seq = seq;
2677		prog = bpf_iter_get_info(&meta, true);
2678		if (prog)
2679			(void)tcp_prog_seq_show(prog, &meta, v, 0);
2680	}
2681
2682	tcp_seq_stop(seq, v);
2683}
2684
2685static const struct seq_operations bpf_iter_tcp_seq_ops = {
2686	.show		= bpf_iter_tcp_seq_show,
2687	.start		= tcp_seq_start,
2688	.next		= tcp_seq_next,
2689	.stop		= bpf_iter_tcp_seq_stop,
2690};
2691#endif
2692
2693static const struct seq_operations tcp4_seq_ops = {
2694	.show		= tcp4_seq_show,
2695	.start		= tcp_seq_start,
2696	.next		= tcp_seq_next,
2697	.stop		= tcp_seq_stop,
2698};
2699
2700static struct tcp_seq_afinfo tcp4_seq_afinfo = {
2701	.family		= AF_INET,
2702};
2703
2704static int __net_init tcp4_proc_init_net(struct net *net)
2705{
2706	if (!proc_create_net_data("tcp", 0444, net->proc_net, &tcp4_seq_ops,
2707			sizeof(struct tcp_iter_state), &tcp4_seq_afinfo))
2708		return -ENOMEM;
2709	return 0;
2710}
2711
2712static void __net_exit tcp4_proc_exit_net(struct net *net)
2713{
2714	remove_proc_entry("tcp", net->proc_net);
2715}
2716
2717static struct pernet_operations tcp4_net_ops = {
2718	.init = tcp4_proc_init_net,
2719	.exit = tcp4_proc_exit_net,
2720};
2721
2722int __init tcp4_proc_init(void)
2723{
2724	return register_pernet_subsys(&tcp4_net_ops);
2725}
2726
2727void tcp4_proc_exit(void)
2728{
2729	unregister_pernet_subsys(&tcp4_net_ops);
2730}
2731#endif /* CONFIG_PROC_FS */
2732
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2733struct proto tcp_prot = {
2734	.name			= "TCP",
2735	.owner			= THIS_MODULE,
2736	.close			= tcp_close,
2737	.pre_connect		= tcp_v4_pre_connect,
2738	.connect		= tcp_v4_connect,
2739	.disconnect		= tcp_disconnect,
2740	.accept			= inet_csk_accept,
2741	.ioctl			= tcp_ioctl,
2742	.init			= tcp_v4_init_sock,
2743	.destroy		= tcp_v4_destroy_sock,
2744	.shutdown		= tcp_shutdown,
2745	.setsockopt		= tcp_setsockopt,
2746	.getsockopt		= tcp_getsockopt,
 
2747	.keepalive		= tcp_set_keepalive,
2748	.recvmsg		= tcp_recvmsg,
2749	.sendmsg		= tcp_sendmsg,
2750	.sendpage		= tcp_sendpage,
2751	.backlog_rcv		= tcp_v4_do_rcv,
2752	.release_cb		= tcp_release_cb,
2753	.hash			= inet_hash,
2754	.unhash			= inet_unhash,
2755	.get_port		= inet_csk_get_port,
 
 
 
2756	.enter_memory_pressure	= tcp_enter_memory_pressure,
2757	.leave_memory_pressure	= tcp_leave_memory_pressure,
2758	.stream_memory_free	= tcp_stream_memory_free,
2759	.sockets_allocated	= &tcp_sockets_allocated,
2760	.orphan_count		= &tcp_orphan_count,
2761	.memory_allocated	= &tcp_memory_allocated,
2762	.memory_pressure	= &tcp_memory_pressure,
2763	.sysctl_mem		= sysctl_tcp_mem,
2764	.sysctl_wmem_offset	= offsetof(struct net, ipv4.sysctl_tcp_wmem),
2765	.sysctl_rmem_offset	= offsetof(struct net, ipv4.sysctl_tcp_rmem),
2766	.max_header		= MAX_TCP_HEADER,
2767	.obj_size		= sizeof(struct tcp_sock),
2768	.slab_flags		= SLAB_TYPESAFE_BY_RCU,
2769	.twsk_prot		= &tcp_timewait_sock_ops,
2770	.rsk_prot		= &tcp_request_sock_ops,
2771	.h.hashinfo		= &tcp_hashinfo,
2772	.no_autobind		= true,
2773	.diag_destroy		= tcp_abort,
2774};
2775EXPORT_SYMBOL(tcp_prot);
2776
2777static void __net_exit tcp_sk_exit(struct net *net)
2778{
2779	int cpu;
2780
2781	if (net->ipv4.tcp_congestion_control)
2782		bpf_module_put(net->ipv4.tcp_congestion_control,
2783			       net->ipv4.tcp_congestion_control->owner);
2784
2785	for_each_possible_cpu(cpu)
2786		inet_ctl_sock_destroy(*per_cpu_ptr(net->ipv4.tcp_sk, cpu));
2787	free_percpu(net->ipv4.tcp_sk);
2788}
2789
2790static int __net_init tcp_sk_init(struct net *net)
2791{
2792	int res, cpu, cnt;
2793
2794	net->ipv4.tcp_sk = alloc_percpu(struct sock *);
2795	if (!net->ipv4.tcp_sk)
2796		return -ENOMEM;
2797
2798	for_each_possible_cpu(cpu) {
2799		struct sock *sk;
2800
2801		res = inet_ctl_sock_create(&sk, PF_INET, SOCK_RAW,
2802					   IPPROTO_TCP, net);
2803		if (res)
2804			goto fail;
2805		sock_set_flag(sk, SOCK_USE_WRITE_QUEUE);
2806
2807		/* Please enforce IP_DF and IPID==0 for RST and
2808		 * ACK sent in SYN-RECV and TIME-WAIT state.
2809		 */
2810		inet_sk(sk)->pmtudisc = IP_PMTUDISC_DO;
2811
2812		*per_cpu_ptr(net->ipv4.tcp_sk, cpu) = sk;
2813	}
2814
2815	net->ipv4.sysctl_tcp_ecn = 2;
2816	net->ipv4.sysctl_tcp_ecn_fallback = 1;
2817
2818	net->ipv4.sysctl_tcp_base_mss = TCP_BASE_MSS;
2819	net->ipv4.sysctl_tcp_min_snd_mss = TCP_MIN_SND_MSS;
2820	net->ipv4.sysctl_tcp_probe_threshold = TCP_PROBE_THRESHOLD;
2821	net->ipv4.sysctl_tcp_probe_interval = TCP_PROBE_INTERVAL;
2822	net->ipv4.sysctl_tcp_mtu_probe_floor = TCP_MIN_SND_MSS;
2823
2824	net->ipv4.sysctl_tcp_keepalive_time = TCP_KEEPALIVE_TIME;
2825	net->ipv4.sysctl_tcp_keepalive_probes = TCP_KEEPALIVE_PROBES;
2826	net->ipv4.sysctl_tcp_keepalive_intvl = TCP_KEEPALIVE_INTVL;
2827
2828	net->ipv4.sysctl_tcp_syn_retries = TCP_SYN_RETRIES;
2829	net->ipv4.sysctl_tcp_synack_retries = TCP_SYNACK_RETRIES;
2830	net->ipv4.sysctl_tcp_syncookies = 1;
2831	net->ipv4.sysctl_tcp_reordering = TCP_FASTRETRANS_THRESH;
2832	net->ipv4.sysctl_tcp_retries1 = TCP_RETR1;
2833	net->ipv4.sysctl_tcp_retries2 = TCP_RETR2;
2834	net->ipv4.sysctl_tcp_orphan_retries = 0;
2835	net->ipv4.sysctl_tcp_fin_timeout = TCP_FIN_TIMEOUT;
2836	net->ipv4.sysctl_tcp_notsent_lowat = UINT_MAX;
2837	net->ipv4.sysctl_tcp_tw_reuse = 2;
2838	net->ipv4.sysctl_tcp_no_ssthresh_metrics_save = 1;
2839
2840	cnt = tcp_hashinfo.ehash_mask + 1;
2841	net->ipv4.tcp_death_row.sysctl_max_tw_buckets = cnt / 2;
2842	net->ipv4.tcp_death_row.hashinfo = &tcp_hashinfo;
2843
2844	net->ipv4.sysctl_max_syn_backlog = max(128, cnt / 128);
2845	net->ipv4.sysctl_tcp_sack = 1;
2846	net->ipv4.sysctl_tcp_window_scaling = 1;
2847	net->ipv4.sysctl_tcp_timestamps = 1;
2848	net->ipv4.sysctl_tcp_early_retrans = 3;
2849	net->ipv4.sysctl_tcp_recovery = TCP_RACK_LOSS_DETECTION;
2850	net->ipv4.sysctl_tcp_slow_start_after_idle = 1; /* By default, RFC2861 behavior.  */
2851	net->ipv4.sysctl_tcp_retrans_collapse = 1;
2852	net->ipv4.sysctl_tcp_max_reordering = 300;
2853	net->ipv4.sysctl_tcp_dsack = 1;
2854	net->ipv4.sysctl_tcp_app_win = 31;
2855	net->ipv4.sysctl_tcp_adv_win_scale = 1;
2856	net->ipv4.sysctl_tcp_frto = 2;
2857	net->ipv4.sysctl_tcp_moderate_rcvbuf = 1;
2858	/* This limits the percentage of the congestion window which we
2859	 * will allow a single TSO frame to consume.  Building TSO frames
2860	 * which are too large can cause TCP streams to be bursty.
2861	 */
2862	net->ipv4.sysctl_tcp_tso_win_divisor = 3;
2863	/* Default TSQ limit of 16 TSO segments */
2864	net->ipv4.sysctl_tcp_limit_output_bytes = 16 * 65536;
2865	/* rfc5961 challenge ack rate limiting */
2866	net->ipv4.sysctl_tcp_challenge_ack_limit = 1000;
2867	net->ipv4.sysctl_tcp_min_tso_segs = 2;
2868	net->ipv4.sysctl_tcp_min_rtt_wlen = 300;
2869	net->ipv4.sysctl_tcp_autocorking = 1;
2870	net->ipv4.sysctl_tcp_invalid_ratelimit = HZ/2;
2871	net->ipv4.sysctl_tcp_pacing_ss_ratio = 200;
2872	net->ipv4.sysctl_tcp_pacing_ca_ratio = 120;
2873	if (net != &init_net) {
2874		memcpy(net->ipv4.sysctl_tcp_rmem,
2875		       init_net.ipv4.sysctl_tcp_rmem,
2876		       sizeof(init_net.ipv4.sysctl_tcp_rmem));
2877		memcpy(net->ipv4.sysctl_tcp_wmem,
2878		       init_net.ipv4.sysctl_tcp_wmem,
2879		       sizeof(init_net.ipv4.sysctl_tcp_wmem));
2880	}
2881	net->ipv4.sysctl_tcp_comp_sack_delay_ns = NSEC_PER_MSEC;
2882	net->ipv4.sysctl_tcp_comp_sack_slack_ns = 100 * NSEC_PER_USEC;
2883	net->ipv4.sysctl_tcp_comp_sack_nr = 44;
2884	net->ipv4.sysctl_tcp_fastopen = TFO_CLIENT_ENABLE;
2885	spin_lock_init(&net->ipv4.tcp_fastopen_ctx_lock);
2886	net->ipv4.sysctl_tcp_fastopen_blackhole_timeout = 60 * 60;
2887	atomic_set(&net->ipv4.tfo_active_disable_times, 0);
2888
2889	/* Reno is always built in */
2890	if (!net_eq(net, &init_net) &&
2891	    bpf_try_module_get(init_net.ipv4.tcp_congestion_control,
2892			       init_net.ipv4.tcp_congestion_control->owner))
2893		net->ipv4.tcp_congestion_control = init_net.ipv4.tcp_congestion_control;
2894	else
2895		net->ipv4.tcp_congestion_control = &tcp_reno;
2896
2897	return 0;
2898fail:
2899	tcp_sk_exit(net);
2900
2901	return res;
2902}
2903
2904static void __net_exit tcp_sk_exit_batch(struct list_head *net_exit_list)
2905{
2906	struct net *net;
2907
2908	inet_twsk_purge(&tcp_hashinfo, AF_INET);
2909
2910	list_for_each_entry(net, net_exit_list, exit_list)
2911		tcp_fastopen_ctx_destroy(net);
2912}
2913
2914static struct pernet_operations __net_initdata tcp_sk_ops = {
2915       .init	   = tcp_sk_init,
2916       .exit	   = tcp_sk_exit,
2917       .exit_batch = tcp_sk_exit_batch,
2918};
2919
2920#if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
2921DEFINE_BPF_ITER_FUNC(tcp, struct bpf_iter_meta *meta,
2922		     struct sock_common *sk_common, uid_t uid)
2923
2924static int bpf_iter_init_tcp(void *priv_data, struct bpf_iter_aux_info *aux)
2925{
2926	struct tcp_iter_state *st = priv_data;
2927	struct tcp_seq_afinfo *afinfo;
2928	int ret;
2929
2930	afinfo = kmalloc(sizeof(*afinfo), GFP_USER | __GFP_NOWARN);
2931	if (!afinfo)
2932		return -ENOMEM;
2933
2934	afinfo->family = AF_UNSPEC;
2935	st->bpf_seq_afinfo = afinfo;
2936	ret = bpf_iter_init_seq_net(priv_data, aux);
2937	if (ret)
2938		kfree(afinfo);
2939	return ret;
2940}
2941
2942static void bpf_iter_fini_tcp(void *priv_data)
2943{
2944	struct tcp_iter_state *st = priv_data;
2945
2946	kfree(st->bpf_seq_afinfo);
2947	bpf_iter_fini_seq_net(priv_data);
2948}
2949
2950static const struct bpf_iter_seq_info tcp_seq_info = {
2951	.seq_ops		= &bpf_iter_tcp_seq_ops,
2952	.init_seq_private	= bpf_iter_init_tcp,
2953	.fini_seq_private	= bpf_iter_fini_tcp,
2954	.seq_priv_size		= sizeof(struct tcp_iter_state),
2955};
2956
2957static struct bpf_iter_reg tcp_reg_info = {
2958	.target			= "tcp",
2959	.ctx_arg_info_size	= 1,
2960	.ctx_arg_info		= {
2961		{ offsetof(struct bpf_iter__tcp, sk_common),
2962		  PTR_TO_BTF_ID_OR_NULL },
2963	},
2964	.seq_info		= &tcp_seq_info,
2965};
2966
2967static void __init bpf_iter_register(void)
2968{
2969	tcp_reg_info.ctx_arg_info[0].btf_id = btf_sock_ids[BTF_SOCK_TYPE_SOCK_COMMON];
2970	if (bpf_iter_reg_target(&tcp_reg_info))
2971		pr_warn("Warning: could not register bpf iterator tcp\n");
2972}
2973
2974#endif
2975
2976void __init tcp_v4_init(void)
2977{
2978	if (register_pernet_subsys(&tcp_sk_ops))
2979		panic("Failed to create the TCP control socket.\n");
2980
2981#if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
2982	bpf_iter_register();
2983#endif
2984}