Linux Audio

Check our new training course

Loading...
v6.13.7
   1// SPDX-License-Identifier: GPL-2.0-or-later
   2/*
   3 * INET		An implementation of the TCP/IP protocol suite for the LINUX
   4 *		operating system.  INET is implemented using the  BSD Socket
   5 *		interface as the means of communication with the user level.
   6 *
   7 *		Implementation of the Transmission Control Protocol(TCP).
   8 *
   9 *		IPv4 specific functions
  10 *
  11 *		code split from:
  12 *		linux/ipv4/tcp.c
  13 *		linux/ipv4/tcp_input.c
  14 *		linux/ipv4/tcp_output.c
  15 *
  16 *		See tcp.c for author information
  17 */
  18
  19/*
  20 * Changes:
  21 *		David S. Miller	:	New socket lookup architecture.
  22 *					This code is dedicated to John Dyson.
  23 *		David S. Miller :	Change semantics of established hash,
  24 *					half is devoted to TIME_WAIT sockets
  25 *					and the rest go in the other half.
  26 *		Andi Kleen :		Add support for syncookies and fixed
  27 *					some bugs: ip options weren't passed to
  28 *					the TCP layer, missed a check for an
  29 *					ACK bit.
  30 *		Andi Kleen :		Implemented fast path mtu discovery.
  31 *	     				Fixed many serious bugs in the
  32 *					request_sock handling and moved
  33 *					most of it into the af independent code.
  34 *					Added tail drop and some other bugfixes.
  35 *					Added new listen semantics.
  36 *		Mike McLagan	:	Routing by source
  37 *	Juan Jose Ciarlante:		ip_dynaddr bits
  38 *		Andi Kleen:		various fixes.
  39 *	Vitaly E. Lavrov	:	Transparent proxy revived after year
  40 *					coma.
  41 *	Andi Kleen		:	Fix new listen.
  42 *	Andi Kleen		:	Fix accept error reporting.
  43 *	YOSHIFUJI Hideaki @USAGI and:	Support IPV6_V6ONLY socket option, which
  44 *	Alexey Kuznetsov		allow both IPv4 and IPv6 sockets to bind
  45 *					a single port at the same time.
  46 */
  47
  48#define pr_fmt(fmt) "TCP: " fmt
  49
  50#include <linux/bottom_half.h>
  51#include <linux/types.h>
  52#include <linux/fcntl.h>
  53#include <linux/module.h>
  54#include <linux/random.h>
  55#include <linux/cache.h>
  56#include <linux/jhash.h>
  57#include <linux/init.h>
  58#include <linux/times.h>
  59#include <linux/slab.h>
  60#include <linux/sched.h>
  61
  62#include <net/net_namespace.h>
  63#include <net/icmp.h>
  64#include <net/inet_hashtables.h>
  65#include <net/tcp.h>
  66#include <net/transp_v6.h>
  67#include <net/ipv6.h>
  68#include <net/inet_common.h>
  69#include <net/timewait_sock.h>
  70#include <net/xfrm.h>
  71#include <net/secure_seq.h>
  72#include <net/busy_poll.h>
  73#include <net/rstreason.h>
  74
  75#include <linux/inet.h>
  76#include <linux/ipv6.h>
  77#include <linux/stddef.h>
  78#include <linux/proc_fs.h>
  79#include <linux/seq_file.h>
  80#include <linux/inetdevice.h>
  81#include <linux/btf_ids.h>
  82#include <linux/skbuff_ref.h>
  83
  84#include <crypto/hash.h>
  85#include <linux/scatterlist.h>
  86
  87#include <trace/events/tcp.h>
  88
  89#ifdef CONFIG_TCP_MD5SIG
  90static int tcp_v4_md5_hash_hdr(char *md5_hash, const struct tcp_md5sig_key *key,
  91			       __be32 daddr, __be32 saddr, const struct tcphdr *th);
  92#endif
  93
  94struct inet_hashinfo tcp_hashinfo;
  95EXPORT_SYMBOL(tcp_hashinfo);
  96
  97static DEFINE_PER_CPU(struct sock_bh_locked, ipv4_tcp_sk) = {
  98	.bh_lock = INIT_LOCAL_LOCK(bh_lock),
  99};
 100
 101static DEFINE_MUTEX(tcp_exit_batch_mutex);
 102
 103static u32 tcp_v4_init_seq(const struct sk_buff *skb)
 104{
 105	return secure_tcp_seq(ip_hdr(skb)->daddr,
 106			      ip_hdr(skb)->saddr,
 107			      tcp_hdr(skb)->dest,
 108			      tcp_hdr(skb)->source);
 109}
 110
 111static u32 tcp_v4_init_ts_off(const struct net *net, const struct sk_buff *skb)
 112{
 113	return secure_tcp_ts_off(net, ip_hdr(skb)->daddr, ip_hdr(skb)->saddr);
 114}
 115
 116int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp)
 117{
 118	int reuse = READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_tw_reuse);
 119	const struct inet_timewait_sock *tw = inet_twsk(sktw);
 120	const struct tcp_timewait_sock *tcptw = tcp_twsk(sktw);
 121	struct tcp_sock *tp = tcp_sk(sk);
 122	int ts_recent_stamp;
 123
 124	if (READ_ONCE(tw->tw_substate) == TCP_FIN_WAIT2)
 125		reuse = 0;
 126
 127	if (reuse == 2) {
 128		/* Still does not detect *everything* that goes through
 129		 * lo, since we require a loopback src or dst address
 130		 * or direct binding to 'lo' interface.
 131		 */
 132		bool loopback = false;
 133		if (tw->tw_bound_dev_if == LOOPBACK_IFINDEX)
 134			loopback = true;
 135#if IS_ENABLED(CONFIG_IPV6)
 136		if (tw->tw_family == AF_INET6) {
 137			if (ipv6_addr_loopback(&tw->tw_v6_daddr) ||
 138			    ipv6_addr_v4mapped_loopback(&tw->tw_v6_daddr) ||
 139			    ipv6_addr_loopback(&tw->tw_v6_rcv_saddr) ||
 140			    ipv6_addr_v4mapped_loopback(&tw->tw_v6_rcv_saddr))
 141				loopback = true;
 142		} else
 143#endif
 144		{
 145			if (ipv4_is_loopback(tw->tw_daddr) ||
 146			    ipv4_is_loopback(tw->tw_rcv_saddr))
 147				loopback = true;
 148		}
 149		if (!loopback)
 150			reuse = 0;
 151	}
 152
 153	/* With PAWS, it is safe from the viewpoint
 154	   of data integrity. Even without PAWS it is safe provided sequence
 155	   spaces do not overlap i.e. at data rates <= 80Mbit/sec.
 156
 157	   Actually, the idea is close to VJ's one, only timestamp cache is
 158	   held not per host, but per port pair and TW bucket is used as state
 159	   holder.
 160
 161	   If TW bucket has been already destroyed we fall back to VJ's scheme
 162	   and use initial timestamp retrieved from peer table.
 163	 */
 164	ts_recent_stamp = READ_ONCE(tcptw->tw_ts_recent_stamp);
 165	if (ts_recent_stamp &&
 166	    (!twp || (reuse && time_after32(ktime_get_seconds(),
 167					    ts_recent_stamp)))) {
 168		/* inet_twsk_hashdance_schedule() sets sk_refcnt after putting twsk
 169		 * and releasing the bucket lock.
 170		 */
 171		if (unlikely(!refcount_inc_not_zero(&sktw->sk_refcnt)))
 172			return 0;
 173
 174		/* In case of repair and re-using TIME-WAIT sockets we still
 175		 * want to be sure that it is safe as above but honor the
 176		 * sequence numbers and time stamps set as part of the repair
 177		 * process.
 178		 *
 179		 * Without this check re-using a TIME-WAIT socket with TCP
 180		 * repair would accumulate a -1 on the repair assigned
 181		 * sequence number. The first time it is reused the sequence
 182		 * is -1, the second time -2, etc. This fixes that issue
 183		 * without appearing to create any others.
 184		 */
 185		if (likely(!tp->repair)) {
 186			u32 seq = tcptw->tw_snd_nxt + 65535 + 2;
 187
 188			if (!seq)
 189				seq = 1;
 190			WRITE_ONCE(tp->write_seq, seq);
 191			tp->rx_opt.ts_recent	   = READ_ONCE(tcptw->tw_ts_recent);
 192			tp->rx_opt.ts_recent_stamp = ts_recent_stamp;
 193		}
 194
 195		return 1;
 196	}
 197
 198	return 0;
 199}
 200EXPORT_SYMBOL_GPL(tcp_twsk_unique);
 201
 202static int tcp_v4_pre_connect(struct sock *sk, struct sockaddr *uaddr,
 203			      int addr_len)
 204{
 205	/* This check is replicated from tcp_v4_connect() and intended to
 206	 * prevent BPF program called below from accessing bytes that are out
 207	 * of the bound specified by user in addr_len.
 208	 */
 209	if (addr_len < sizeof(struct sockaddr_in))
 210		return -EINVAL;
 211
 212	sock_owned_by_me(sk);
 213
 214	return BPF_CGROUP_RUN_PROG_INET4_CONNECT(sk, uaddr, &addr_len);
 215}
 216
 217/* This will initiate an outgoing connection. */
 218int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len)
 219{
 220	struct sockaddr_in *usin = (struct sockaddr_in *)uaddr;
 221	struct inet_timewait_death_row *tcp_death_row;
 222	struct inet_sock *inet = inet_sk(sk);
 223	struct tcp_sock *tp = tcp_sk(sk);
 224	struct ip_options_rcu *inet_opt;
 225	struct net *net = sock_net(sk);
 226	__be16 orig_sport, orig_dport;
 227	__be32 daddr, nexthop;
 228	struct flowi4 *fl4;
 229	struct rtable *rt;
 230	int err;
 231
 232	if (addr_len < sizeof(struct sockaddr_in))
 233		return -EINVAL;
 234
 235	if (usin->sin_family != AF_INET)
 236		return -EAFNOSUPPORT;
 237
 238	nexthop = daddr = usin->sin_addr.s_addr;
 239	inet_opt = rcu_dereference_protected(inet->inet_opt,
 240					     lockdep_sock_is_held(sk));
 241	if (inet_opt && inet_opt->opt.srr) {
 242		if (!daddr)
 243			return -EINVAL;
 244		nexthop = inet_opt->opt.faddr;
 245	}
 246
 247	orig_sport = inet->inet_sport;
 248	orig_dport = usin->sin_port;
 249	fl4 = &inet->cork.fl.u.ip4;
 250	rt = ip_route_connect(fl4, nexthop, inet->inet_saddr,
 251			      sk->sk_bound_dev_if, IPPROTO_TCP, orig_sport,
 252			      orig_dport, sk);
 253	if (IS_ERR(rt)) {
 254		err = PTR_ERR(rt);
 255		if (err == -ENETUNREACH)
 256			IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES);
 257		return err;
 258	}
 259
 260	if (rt->rt_flags & (RTCF_MULTICAST | RTCF_BROADCAST)) {
 261		ip_rt_put(rt);
 262		return -ENETUNREACH;
 263	}
 264
 265	if (!inet_opt || !inet_opt->opt.srr)
 266		daddr = fl4->daddr;
 267
 268	tcp_death_row = &sock_net(sk)->ipv4.tcp_death_row;
 269
 270	if (!inet->inet_saddr) {
 271		err = inet_bhash2_update_saddr(sk,  &fl4->saddr, AF_INET);
 272		if (err) {
 273			ip_rt_put(rt);
 274			return err;
 275		}
 276	} else {
 277		sk_rcv_saddr_set(sk, inet->inet_saddr);
 278	}
 279
 280	if (tp->rx_opt.ts_recent_stamp && inet->inet_daddr != daddr) {
 281		/* Reset inherited state */
 282		tp->rx_opt.ts_recent	   = 0;
 283		tp->rx_opt.ts_recent_stamp = 0;
 284		if (likely(!tp->repair))
 285			WRITE_ONCE(tp->write_seq, 0);
 286	}
 287
 288	inet->inet_dport = usin->sin_port;
 289	sk_daddr_set(sk, daddr);
 290
 291	inet_csk(sk)->icsk_ext_hdr_len = 0;
 292	if (inet_opt)
 293		inet_csk(sk)->icsk_ext_hdr_len = inet_opt->opt.optlen;
 294
 295	tp->rx_opt.mss_clamp = TCP_MSS_DEFAULT;
 296
 297	/* Socket identity is still unknown (sport may be zero).
 298	 * However we set state to SYN-SENT and not releasing socket
 299	 * lock select source port, enter ourselves into the hash tables and
 300	 * complete initialization after this.
 301	 */
 302	tcp_set_state(sk, TCP_SYN_SENT);
 303	err = inet_hash_connect(tcp_death_row, sk);
 304	if (err)
 305		goto failure;
 306
 307	sk_set_txhash(sk);
 308
 309	rt = ip_route_newports(fl4, rt, orig_sport, orig_dport,
 310			       inet->inet_sport, inet->inet_dport, sk);
 311	if (IS_ERR(rt)) {
 312		err = PTR_ERR(rt);
 313		rt = NULL;
 314		goto failure;
 315	}
 316	tp->tcp_usec_ts = dst_tcp_usec_ts(&rt->dst);
 317	/* OK, now commit destination to socket.  */
 318	sk->sk_gso_type = SKB_GSO_TCPV4;
 319	sk_setup_caps(sk, &rt->dst);
 320	rt = NULL;
 321
 322	if (likely(!tp->repair)) {
 323		if (!tp->write_seq)
 324			WRITE_ONCE(tp->write_seq,
 325				   secure_tcp_seq(inet->inet_saddr,
 326						  inet->inet_daddr,
 327						  inet->inet_sport,
 328						  usin->sin_port));
 329		WRITE_ONCE(tp->tsoffset,
 330			   secure_tcp_ts_off(net, inet->inet_saddr,
 331					     inet->inet_daddr));
 332	}
 333
 334	atomic_set(&inet->inet_id, get_random_u16());
 335
 336	if (tcp_fastopen_defer_connect(sk, &err))
 337		return err;
 338	if (err)
 339		goto failure;
 340
 341	err = tcp_connect(sk);
 342
 343	if (err)
 344		goto failure;
 345
 346	return 0;
 347
 348failure:
 349	/*
 350	 * This unhashes the socket and releases the local port,
 351	 * if necessary.
 352	 */
 353	tcp_set_state(sk, TCP_CLOSE);
 354	inet_bhash2_reset_saddr(sk);
 355	ip_rt_put(rt);
 356	sk->sk_route_caps = 0;
 357	inet->inet_dport = 0;
 358	return err;
 359}
 360EXPORT_SYMBOL(tcp_v4_connect);
 361
 362/*
 363 * This routine reacts to ICMP_FRAG_NEEDED mtu indications as defined in RFC1191.
 364 * It can be called through tcp_release_cb() if socket was owned by user
 365 * at the time tcp_v4_err() was called to handle ICMP message.
 366 */
 367void tcp_v4_mtu_reduced(struct sock *sk)
 368{
 369	struct inet_sock *inet = inet_sk(sk);
 370	struct dst_entry *dst;
 371	u32 mtu;
 372
 373	if ((1 << sk->sk_state) & (TCPF_LISTEN | TCPF_CLOSE))
 374		return;
 375	mtu = READ_ONCE(tcp_sk(sk)->mtu_info);
 376	dst = inet_csk_update_pmtu(sk, mtu);
 377	if (!dst)
 378		return;
 379
 380	/* Something is about to be wrong... Remember soft error
 381	 * for the case, if this connection will not able to recover.
 382	 */
 383	if (mtu < dst_mtu(dst) && ip_dont_fragment(sk, dst))
 384		WRITE_ONCE(sk->sk_err_soft, EMSGSIZE);
 385
 386	mtu = dst_mtu(dst);
 387
 388	if (inet->pmtudisc != IP_PMTUDISC_DONT &&
 389	    ip_sk_accept_pmtu(sk) &&
 390	    inet_csk(sk)->icsk_pmtu_cookie > mtu) {
 391		tcp_sync_mss(sk, mtu);
 392
 393		/* Resend the TCP packet because it's
 394		 * clear that the old packet has been
 395		 * dropped. This is the new "fast" path mtu
 396		 * discovery.
 397		 */
 398		tcp_simple_retransmit(sk);
 399	} /* else let the usual retransmit timer handle it */
 400}
 401EXPORT_SYMBOL(tcp_v4_mtu_reduced);
 402
 403static void do_redirect(struct sk_buff *skb, struct sock *sk)
 404{
 405	struct dst_entry *dst = __sk_dst_check(sk, 0);
 406
 407	if (dst)
 408		dst->ops->redirect(dst, sk, skb);
 409}
 410
 411
 412/* handle ICMP messages on TCP_NEW_SYN_RECV request sockets */
 413void tcp_req_err(struct sock *sk, u32 seq, bool abort)
 414{
 415	struct request_sock *req = inet_reqsk(sk);
 416	struct net *net = sock_net(sk);
 417
 418	/* ICMPs are not backlogged, hence we cannot get
 419	 * an established socket here.
 420	 */
 421	if (seq != tcp_rsk(req)->snt_isn) {
 422		__NET_INC_STATS(net, LINUX_MIB_OUTOFWINDOWICMPS);
 423	} else if (abort) {
 424		/*
 425		 * Still in SYN_RECV, just remove it silently.
 426		 * There is no good way to pass the error to the newly
 427		 * created socket, and POSIX does not want network
 428		 * errors returned from accept().
 429		 */
 430		inet_csk_reqsk_queue_drop(req->rsk_listener, req);
 431		tcp_listendrop(req->rsk_listener);
 432	}
 433	reqsk_put(req);
 434}
 435EXPORT_SYMBOL(tcp_req_err);
 436
 437/* TCP-LD (RFC 6069) logic */
 438void tcp_ld_RTO_revert(struct sock *sk, u32 seq)
 439{
 440	struct inet_connection_sock *icsk = inet_csk(sk);
 441	struct tcp_sock *tp = tcp_sk(sk);
 442	struct sk_buff *skb;
 443	s32 remaining;
 444	u32 delta_us;
 445
 446	if (sock_owned_by_user(sk))
 447		return;
 448
 449	if (seq != tp->snd_una  || !icsk->icsk_retransmits ||
 450	    !icsk->icsk_backoff)
 451		return;
 452
 453	skb = tcp_rtx_queue_head(sk);
 454	if (WARN_ON_ONCE(!skb))
 455		return;
 456
 457	icsk->icsk_backoff--;
 458	icsk->icsk_rto = tp->srtt_us ? __tcp_set_rto(tp) : TCP_TIMEOUT_INIT;
 459	icsk->icsk_rto = inet_csk_rto_backoff(icsk, TCP_RTO_MAX);
 460
 461	tcp_mstamp_refresh(tp);
 462	delta_us = (u32)(tp->tcp_mstamp - tcp_skb_timestamp_us(skb));
 463	remaining = icsk->icsk_rto - usecs_to_jiffies(delta_us);
 464
 465	if (remaining > 0) {
 466		inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
 467					  remaining, TCP_RTO_MAX);
 468	} else {
 469		/* RTO revert clocked out retransmission.
 470		 * Will retransmit now.
 471		 */
 472		tcp_retransmit_timer(sk);
 473	}
 474}
 475EXPORT_SYMBOL(tcp_ld_RTO_revert);
 476
 477/*
 478 * This routine is called by the ICMP module when it gets some
 479 * sort of error condition.  If err < 0 then the socket should
 480 * be closed and the error returned to the user.  If err > 0
 481 * it's just the icmp type << 8 | icmp code.  After adjustment
 482 * header points to the first 8 bytes of the tcp header.  We need
 483 * to find the appropriate port.
 484 *
 485 * The locking strategy used here is very "optimistic". When
 486 * someone else accesses the socket the ICMP is just dropped
 487 * and for some paths there is no check at all.
 488 * A more general error queue to queue errors for later handling
 489 * is probably better.
 490 *
 491 */
 492
 493int tcp_v4_err(struct sk_buff *skb, u32 info)
 494{
 495	const struct iphdr *iph = (const struct iphdr *)skb->data;
 496	struct tcphdr *th = (struct tcphdr *)(skb->data + (iph->ihl << 2));
 497	struct tcp_sock *tp;
 498	const int type = icmp_hdr(skb)->type;
 499	const int code = icmp_hdr(skb)->code;
 500	struct sock *sk;
 501	struct request_sock *fastopen;
 502	u32 seq, snd_una;
 503	int err;
 504	struct net *net = dev_net(skb->dev);
 505
 506	sk = __inet_lookup_established(net, net->ipv4.tcp_death_row.hashinfo,
 507				       iph->daddr, th->dest, iph->saddr,
 508				       ntohs(th->source), inet_iif(skb), 0);
 509	if (!sk) {
 510		__ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
 511		return -ENOENT;
 512	}
 513	if (sk->sk_state == TCP_TIME_WAIT) {
 514		/* To increase the counter of ignored icmps for TCP-AO */
 515		tcp_ao_ignore_icmp(sk, AF_INET, type, code);
 516		inet_twsk_put(inet_twsk(sk));
 517		return 0;
 518	}
 519	seq = ntohl(th->seq);
 520	if (sk->sk_state == TCP_NEW_SYN_RECV) {
 521		tcp_req_err(sk, seq, type == ICMP_PARAMETERPROB ||
 522				     type == ICMP_TIME_EXCEEDED ||
 523				     (type == ICMP_DEST_UNREACH &&
 524				      (code == ICMP_NET_UNREACH ||
 525				       code == ICMP_HOST_UNREACH)));
 526		return 0;
 527	}
 528
 529	if (tcp_ao_ignore_icmp(sk, AF_INET, type, code)) {
 530		sock_put(sk);
 531		return 0;
 532	}
 533
 534	bh_lock_sock(sk);
 535	/* If too many ICMPs get dropped on busy
 536	 * servers this needs to be solved differently.
 537	 * We do take care of PMTU discovery (RFC1191) special case :
 538	 * we can receive locally generated ICMP messages while socket is held.
 539	 */
 540	if (sock_owned_by_user(sk)) {
 541		if (!(type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED))
 542			__NET_INC_STATS(net, LINUX_MIB_LOCKDROPPEDICMPS);
 543	}
 544	if (sk->sk_state == TCP_CLOSE)
 545		goto out;
 546
 547	if (static_branch_unlikely(&ip4_min_ttl)) {
 548		/* min_ttl can be changed concurrently from do_ip_setsockopt() */
 549		if (unlikely(iph->ttl < READ_ONCE(inet_sk(sk)->min_ttl))) {
 550			__NET_INC_STATS(net, LINUX_MIB_TCPMINTTLDROP);
 551			goto out;
 552		}
 553	}
 554
 555	tp = tcp_sk(sk);
 556	/* XXX (TFO) - tp->snd_una should be ISN (tcp_create_openreq_child() */
 557	fastopen = rcu_dereference(tp->fastopen_rsk);
 558	snd_una = fastopen ? tcp_rsk(fastopen)->snt_isn : tp->snd_una;
 559	if (sk->sk_state != TCP_LISTEN &&
 560	    !between(seq, snd_una, tp->snd_nxt)) {
 561		__NET_INC_STATS(net, LINUX_MIB_OUTOFWINDOWICMPS);
 562		goto out;
 563	}
 564
 565	switch (type) {
 566	case ICMP_REDIRECT:
 567		if (!sock_owned_by_user(sk))
 568			do_redirect(skb, sk);
 569		goto out;
 570	case ICMP_SOURCE_QUENCH:
 571		/* Just silently ignore these. */
 572		goto out;
 573	case ICMP_PARAMETERPROB:
 574		err = EPROTO;
 575		break;
 576	case ICMP_DEST_UNREACH:
 577		if (code > NR_ICMP_UNREACH)
 578			goto out;
 579
 580		if (code == ICMP_FRAG_NEEDED) { /* PMTU discovery (RFC1191) */
 581			/* We are not interested in TCP_LISTEN and open_requests
 582			 * (SYN-ACKs send out by Linux are always <576bytes so
 583			 * they should go through unfragmented).
 584			 */
 585			if (sk->sk_state == TCP_LISTEN)
 586				goto out;
 587
 588			WRITE_ONCE(tp->mtu_info, info);
 589			if (!sock_owned_by_user(sk)) {
 590				tcp_v4_mtu_reduced(sk);
 591			} else {
 592				if (!test_and_set_bit(TCP_MTU_REDUCED_DEFERRED, &sk->sk_tsq_flags))
 593					sock_hold(sk);
 594			}
 595			goto out;
 596		}
 597
 598		err = icmp_err_convert[code].errno;
 599		/* check if this ICMP message allows revert of backoff.
 600		 * (see RFC 6069)
 601		 */
 602		if (!fastopen &&
 603		    (code == ICMP_NET_UNREACH || code == ICMP_HOST_UNREACH))
 604			tcp_ld_RTO_revert(sk, seq);
 605		break;
 606	case ICMP_TIME_EXCEEDED:
 607		err = EHOSTUNREACH;
 608		break;
 609	default:
 610		goto out;
 611	}
 612
 613	switch (sk->sk_state) {
 614	case TCP_SYN_SENT:
 615	case TCP_SYN_RECV:
 616		/* Only in fast or simultaneous open. If a fast open socket is
 617		 * already accepted it is treated as a connected one below.
 618		 */
 619		if (fastopen && !fastopen->sk)
 620			break;
 621
 622		ip_icmp_error(sk, skb, err, th->dest, info, (u8 *)th);
 623
 624		if (!sock_owned_by_user(sk))
 625			tcp_done_with_error(sk, err);
 626		else
 
 
 
 
 627			WRITE_ONCE(sk->sk_err_soft, err);
 
 628		goto out;
 629	}
 630
 631	/* If we've already connected we will keep trying
 632	 * until we time out, or the user gives up.
 633	 *
 634	 * rfc1122 4.2.3.9 allows to consider as hard errors
 635	 * only PROTO_UNREACH and PORT_UNREACH (well, FRAG_FAILED too,
 636	 * but it is obsoleted by pmtu discovery).
 637	 *
 638	 * Note, that in modern internet, where routing is unreliable
 639	 * and in each dark corner broken firewalls sit, sending random
 640	 * errors ordered by their masters even this two messages finally lose
 641	 * their original sense (even Linux sends invalid PORT_UNREACHs)
 642	 *
 643	 * Now we are in compliance with RFCs.
 644	 *							--ANK (980905)
 645	 */
 646
 647	if (!sock_owned_by_user(sk) &&
 648	    inet_test_bit(RECVERR, sk)) {
 649		WRITE_ONCE(sk->sk_err, err);
 650		sk_error_report(sk);
 651	} else	{ /* Only an error on timeout */
 652		WRITE_ONCE(sk->sk_err_soft, err);
 653	}
 654
 655out:
 656	bh_unlock_sock(sk);
 657	sock_put(sk);
 658	return 0;
 659}
 660
 661void __tcp_v4_send_check(struct sk_buff *skb, __be32 saddr, __be32 daddr)
 662{
 663	struct tcphdr *th = tcp_hdr(skb);
 664
 665	th->check = ~tcp_v4_check(skb->len, saddr, daddr, 0);
 666	skb->csum_start = skb_transport_header(skb) - skb->head;
 667	skb->csum_offset = offsetof(struct tcphdr, check);
 668}
 669
 670/* This routine computes an IPv4 TCP checksum. */
 671void tcp_v4_send_check(struct sock *sk, struct sk_buff *skb)
 672{
 673	const struct inet_sock *inet = inet_sk(sk);
 674
 675	__tcp_v4_send_check(skb, inet->inet_saddr, inet->inet_daddr);
 676}
 677EXPORT_SYMBOL(tcp_v4_send_check);
 678
 679#define REPLY_OPTIONS_LEN      (MAX_TCP_OPTION_SPACE / sizeof(__be32))
 680
 681static bool tcp_v4_ao_sign_reset(const struct sock *sk, struct sk_buff *skb,
 682				 const struct tcp_ao_hdr *aoh,
 683				 struct ip_reply_arg *arg, struct tcphdr *reply,
 684				 __be32 reply_options[REPLY_OPTIONS_LEN])
 685{
 686#ifdef CONFIG_TCP_AO
 687	int sdif = tcp_v4_sdif(skb);
 688	int dif = inet_iif(skb);
 689	int l3index = sdif ? dif : 0;
 690	bool allocated_traffic_key;
 691	struct tcp_ao_key *key;
 692	char *traffic_key;
 693	bool drop = true;
 694	u32 ao_sne = 0;
 695	u8 keyid;
 696
 697	rcu_read_lock();
 698	if (tcp_ao_prepare_reset(sk, skb, aoh, l3index, ntohl(reply->seq),
 699				 &key, &traffic_key, &allocated_traffic_key,
 700				 &keyid, &ao_sne))
 701		goto out;
 702
 703	reply_options[0] = htonl((TCPOPT_AO << 24) | (tcp_ao_len(key) << 16) |
 704				 (aoh->rnext_keyid << 8) | keyid);
 705	arg->iov[0].iov_len += tcp_ao_len_aligned(key);
 706	reply->doff = arg->iov[0].iov_len / 4;
 707
 708	if (tcp_ao_hash_hdr(AF_INET, (char *)&reply_options[1],
 709			    key, traffic_key,
 710			    (union tcp_ao_addr *)&ip_hdr(skb)->saddr,
 711			    (union tcp_ao_addr *)&ip_hdr(skb)->daddr,
 712			    reply, ao_sne))
 713		goto out;
 714	drop = false;
 715out:
 716	rcu_read_unlock();
 717	if (allocated_traffic_key)
 718		kfree(traffic_key);
 719	return drop;
 720#else
 721	return true;
 722#endif
 723}
 724
 725/*
 726 *	This routine will send an RST to the other tcp.
 727 *
 728 *	Someone asks: why I NEVER use socket parameters (TOS, TTL etc.)
 729 *		      for reset.
 730 *	Answer: if a packet caused RST, it is not for a socket
 731 *		existing in our system, if it is matched to a socket,
 732 *		it is just duplicate segment or bug in other side's TCP.
 733 *		So that we build reply only basing on parameters
 734 *		arrived with segment.
 735 *	Exception: precedence violation. We do not implement it in any case.
 736 */
 737
 738static void tcp_v4_send_reset(const struct sock *sk, struct sk_buff *skb,
 739			      enum sk_rst_reason reason)
 740{
 741	const struct tcphdr *th = tcp_hdr(skb);
 742	struct {
 743		struct tcphdr th;
 744		__be32 opt[REPLY_OPTIONS_LEN];
 745	} rep;
 746	const __u8 *md5_hash_location = NULL;
 747	const struct tcp_ao_hdr *aoh;
 748	struct ip_reply_arg arg;
 749#ifdef CONFIG_TCP_MD5SIG
 750	struct tcp_md5sig_key *key = NULL;
 751	unsigned char newhash[16];
 752	struct sock *sk1 = NULL;
 753	int genhash;
 754#endif
 755	u64 transmit_time = 0;
 756	struct sock *ctl_sk;
 757	struct net *net;
 758	u32 txhash = 0;
 759
 760	/* Never send a reset in response to a reset. */
 761	if (th->rst)
 762		return;
 763
 764	/* If sk not NULL, it means we did a successful lookup and incoming
 765	 * route had to be correct. prequeue might have dropped our dst.
 766	 */
 767	if (!sk && skb_rtable(skb)->rt_type != RTN_LOCAL)
 768		return;
 769
 770	/* Swap the send and the receive. */
 771	memset(&rep, 0, sizeof(rep));
 772	rep.th.dest   = th->source;
 773	rep.th.source = th->dest;
 774	rep.th.doff   = sizeof(struct tcphdr) / 4;
 775	rep.th.rst    = 1;
 776
 777	if (th->ack) {
 778		rep.th.seq = th->ack_seq;
 779	} else {
 780		rep.th.ack = 1;
 781		rep.th.ack_seq = htonl(ntohl(th->seq) + th->syn + th->fin +
 782				       skb->len - (th->doff << 2));
 783	}
 784
 785	memset(&arg, 0, sizeof(arg));
 786	arg.iov[0].iov_base = (unsigned char *)&rep;
 787	arg.iov[0].iov_len  = sizeof(rep.th);
 788
 789	net = sk ? sock_net(sk) : dev_net(skb_dst(skb)->dev);
 790
 791	/* Invalid TCP option size or twice included auth */
 792	if (tcp_parse_auth_options(tcp_hdr(skb), &md5_hash_location, &aoh))
 793		return;
 794
 795	if (aoh && tcp_v4_ao_sign_reset(sk, skb, aoh, &arg, &rep.th, rep.opt))
 796		return;
 797
 798#ifdef CONFIG_TCP_MD5SIG
 799	rcu_read_lock();
 800	if (sk && sk_fullsock(sk)) {
 801		const union tcp_md5_addr *addr;
 802		int l3index;
 803
 804		/* sdif set, means packet ingressed via a device
 805		 * in an L3 domain and inet_iif is set to it.
 806		 */
 807		l3index = tcp_v4_sdif(skb) ? inet_iif(skb) : 0;
 808		addr = (union tcp_md5_addr *)&ip_hdr(skb)->saddr;
 809		key = tcp_md5_do_lookup(sk, l3index, addr, AF_INET);
 810	} else if (md5_hash_location) {
 811		const union tcp_md5_addr *addr;
 812		int sdif = tcp_v4_sdif(skb);
 813		int dif = inet_iif(skb);
 814		int l3index;
 815
 816		/*
 817		 * active side is lost. Try to find listening socket through
 818		 * source port, and then find md5 key through listening socket.
 819		 * we are not loose security here:
 820		 * Incoming packet is checked with md5 hash with finding key,
 821		 * no RST generated if md5 hash doesn't match.
 822		 */
 823		sk1 = __inet_lookup_listener(net, net->ipv4.tcp_death_row.hashinfo,
 824					     NULL, 0, ip_hdr(skb)->saddr,
 825					     th->source, ip_hdr(skb)->daddr,
 826					     ntohs(th->source), dif, sdif);
 827		/* don't send rst if it can't find key */
 828		if (!sk1)
 829			goto out;
 830
 831		/* sdif set, means packet ingressed via a device
 832		 * in an L3 domain and dif is set to it.
 833		 */
 834		l3index = sdif ? dif : 0;
 835		addr = (union tcp_md5_addr *)&ip_hdr(skb)->saddr;
 836		key = tcp_md5_do_lookup(sk1, l3index, addr, AF_INET);
 837		if (!key)
 838			goto out;
 839
 840
 841		genhash = tcp_v4_md5_hash_skb(newhash, key, NULL, skb);
 842		if (genhash || memcmp(md5_hash_location, newhash, 16) != 0)
 843			goto out;
 844
 845	}
 846
 847	if (key) {
 848		rep.opt[0] = htonl((TCPOPT_NOP << 24) |
 849				   (TCPOPT_NOP << 16) |
 850				   (TCPOPT_MD5SIG << 8) |
 851				   TCPOLEN_MD5SIG);
 852		/* Update length and the length the header thinks exists */
 853		arg.iov[0].iov_len += TCPOLEN_MD5SIG_ALIGNED;
 854		rep.th.doff = arg.iov[0].iov_len / 4;
 855
 856		tcp_v4_md5_hash_hdr((__u8 *) &rep.opt[1],
 857				     key, ip_hdr(skb)->saddr,
 858				     ip_hdr(skb)->daddr, &rep.th);
 859	}
 860#endif
 861	/* Can't co-exist with TCPMD5, hence check rep.opt[0] */
 862	if (rep.opt[0] == 0) {
 863		__be32 mrst = mptcp_reset_option(skb);
 864
 865		if (mrst) {
 866			rep.opt[0] = mrst;
 867			arg.iov[0].iov_len += sizeof(mrst);
 868			rep.th.doff = arg.iov[0].iov_len / 4;
 869		}
 870	}
 871
 872	arg.csum = csum_tcpudp_nofold(ip_hdr(skb)->daddr,
 873				      ip_hdr(skb)->saddr, /* XXX */
 874				      arg.iov[0].iov_len, IPPROTO_TCP, 0);
 875	arg.csumoffset = offsetof(struct tcphdr, check) / 2;
 876	arg.flags = (sk && inet_sk_transparent(sk)) ? IP_REPLY_ARG_NOSRCCHECK : 0;
 877
 878	/* When socket is gone, all binding information is lost.
 879	 * routing might fail in this case. No choice here, if we choose to force
 880	 * input interface, we will misroute in case of asymmetric route.
 881	 */
 882	if (sk)
 883		arg.bound_dev_if = sk->sk_bound_dev_if;
 884
 885	trace_tcp_send_reset(sk, skb, reason);
 
 886
 887	BUILD_BUG_ON(offsetof(struct sock, sk_bound_dev_if) !=
 888		     offsetof(struct inet_timewait_sock, tw_bound_dev_if));
 889
 890	arg.tos = ip_hdr(skb)->tos;
 891	arg.uid = sock_net_uid(net, sk && sk_fullsock(sk) ? sk : NULL);
 892	local_bh_disable();
 893	local_lock_nested_bh(&ipv4_tcp_sk.bh_lock);
 894	ctl_sk = this_cpu_read(ipv4_tcp_sk.sock);
 895
 896	sock_net_set(ctl_sk, net);
 897	if (sk) {
 898		ctl_sk->sk_mark = (sk->sk_state == TCP_TIME_WAIT) ?
 899				   inet_twsk(sk)->tw_mark : READ_ONCE(sk->sk_mark);
 900		ctl_sk->sk_priority = (sk->sk_state == TCP_TIME_WAIT) ?
 901				   inet_twsk(sk)->tw_priority : READ_ONCE(sk->sk_priority);
 902		transmit_time = tcp_transmit_time(sk);
 903		xfrm_sk_clone_policy(ctl_sk, sk);
 904		txhash = (sk->sk_state == TCP_TIME_WAIT) ?
 905			 inet_twsk(sk)->tw_txhash : sk->sk_txhash;
 906	} else {
 907		ctl_sk->sk_mark = 0;
 908		ctl_sk->sk_priority = 0;
 909	}
 910	ip_send_unicast_reply(ctl_sk, sk,
 911			      skb, &TCP_SKB_CB(skb)->header.h4.opt,
 912			      ip_hdr(skb)->saddr, ip_hdr(skb)->daddr,
 913			      &arg, arg.iov[0].iov_len,
 914			      transmit_time, txhash);
 915
 916	xfrm_sk_free_policy(ctl_sk);
 917	sock_net_set(ctl_sk, &init_net);
 918	__TCP_INC_STATS(net, TCP_MIB_OUTSEGS);
 919	__TCP_INC_STATS(net, TCP_MIB_OUTRSTS);
 920	local_unlock_nested_bh(&ipv4_tcp_sk.bh_lock);
 921	local_bh_enable();
 922
 923#ifdef CONFIG_TCP_MD5SIG
 924out:
 925	rcu_read_unlock();
 926#endif
 927}
 928
 929/* The code following below sending ACKs in SYN-RECV and TIME-WAIT states
 930   outside socket context is ugly, certainly. What can I do?
 931 */
 932
 933static void tcp_v4_send_ack(const struct sock *sk,
 934			    struct sk_buff *skb, u32 seq, u32 ack,
 935			    u32 win, u32 tsval, u32 tsecr, int oif,
 936			    struct tcp_key *key,
 937			    int reply_flags, u8 tos, u32 txhash)
 938{
 939	const struct tcphdr *th = tcp_hdr(skb);
 940	struct {
 941		struct tcphdr th;
 942		__be32 opt[(MAX_TCP_OPTION_SPACE  >> 2)];
 943	} rep;
 944	struct net *net = sock_net(sk);
 945	struct ip_reply_arg arg;
 946	struct sock *ctl_sk;
 947	u64 transmit_time;
 948
 949	memset(&rep.th, 0, sizeof(struct tcphdr));
 950	memset(&arg, 0, sizeof(arg));
 951
 952	arg.iov[0].iov_base = (unsigned char *)&rep;
 953	arg.iov[0].iov_len  = sizeof(rep.th);
 954	if (tsecr) {
 955		rep.opt[0] = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
 956				   (TCPOPT_TIMESTAMP << 8) |
 957				   TCPOLEN_TIMESTAMP);
 958		rep.opt[1] = htonl(tsval);
 959		rep.opt[2] = htonl(tsecr);
 960		arg.iov[0].iov_len += TCPOLEN_TSTAMP_ALIGNED;
 961	}
 962
 963	/* Swap the send and the receive. */
 964	rep.th.dest    = th->source;
 965	rep.th.source  = th->dest;
 966	rep.th.doff    = arg.iov[0].iov_len / 4;
 967	rep.th.seq     = htonl(seq);
 968	rep.th.ack_seq = htonl(ack);
 969	rep.th.ack     = 1;
 970	rep.th.window  = htons(win);
 971
 972#ifdef CONFIG_TCP_MD5SIG
 973	if (tcp_key_is_md5(key)) {
 974		int offset = (tsecr) ? 3 : 0;
 975
 976		rep.opt[offset++] = htonl((TCPOPT_NOP << 24) |
 977					  (TCPOPT_NOP << 16) |
 978					  (TCPOPT_MD5SIG << 8) |
 979					  TCPOLEN_MD5SIG);
 980		arg.iov[0].iov_len += TCPOLEN_MD5SIG_ALIGNED;
 981		rep.th.doff = arg.iov[0].iov_len/4;
 982
 983		tcp_v4_md5_hash_hdr((__u8 *) &rep.opt[offset],
 984				    key->md5_key, ip_hdr(skb)->saddr,
 985				    ip_hdr(skb)->daddr, &rep.th);
 986	}
 987#endif
 988#ifdef CONFIG_TCP_AO
 989	if (tcp_key_is_ao(key)) {
 990		int offset = (tsecr) ? 3 : 0;
 991
 992		rep.opt[offset++] = htonl((TCPOPT_AO << 24) |
 993					  (tcp_ao_len(key->ao_key) << 16) |
 994					  (key->ao_key->sndid << 8) |
 995					  key->rcv_next);
 996		arg.iov[0].iov_len += tcp_ao_len_aligned(key->ao_key);
 997		rep.th.doff = arg.iov[0].iov_len / 4;
 998
 999		tcp_ao_hash_hdr(AF_INET, (char *)&rep.opt[offset],
1000				key->ao_key, key->traffic_key,
1001				(union tcp_ao_addr *)&ip_hdr(skb)->saddr,
1002				(union tcp_ao_addr *)&ip_hdr(skb)->daddr,
1003				&rep.th, key->sne);
1004	}
1005#endif
1006	arg.flags = reply_flags;
1007	arg.csum = csum_tcpudp_nofold(ip_hdr(skb)->daddr,
1008				      ip_hdr(skb)->saddr, /* XXX */
1009				      arg.iov[0].iov_len, IPPROTO_TCP, 0);
1010	arg.csumoffset = offsetof(struct tcphdr, check) / 2;
1011	if (oif)
1012		arg.bound_dev_if = oif;
1013	arg.tos = tos;
1014	arg.uid = sock_net_uid(net, sk_fullsock(sk) ? sk : NULL);
1015	local_bh_disable();
1016	local_lock_nested_bh(&ipv4_tcp_sk.bh_lock);
1017	ctl_sk = this_cpu_read(ipv4_tcp_sk.sock);
1018	sock_net_set(ctl_sk, net);
1019	ctl_sk->sk_mark = (sk->sk_state == TCP_TIME_WAIT) ?
1020			   inet_twsk(sk)->tw_mark : READ_ONCE(sk->sk_mark);
1021	ctl_sk->sk_priority = (sk->sk_state == TCP_TIME_WAIT) ?
1022			   inet_twsk(sk)->tw_priority : READ_ONCE(sk->sk_priority);
1023	transmit_time = tcp_transmit_time(sk);
1024	ip_send_unicast_reply(ctl_sk, sk,
1025			      skb, &TCP_SKB_CB(skb)->header.h4.opt,
1026			      ip_hdr(skb)->saddr, ip_hdr(skb)->daddr,
1027			      &arg, arg.iov[0].iov_len,
1028			      transmit_time, txhash);
1029
1030	sock_net_set(ctl_sk, &init_net);
1031	__TCP_INC_STATS(net, TCP_MIB_OUTSEGS);
1032	local_unlock_nested_bh(&ipv4_tcp_sk.bh_lock);
1033	local_bh_enable();
1034}
1035
1036static void tcp_v4_timewait_ack(struct sock *sk, struct sk_buff *skb)
1037{
1038	struct inet_timewait_sock *tw = inet_twsk(sk);
1039	struct tcp_timewait_sock *tcptw = tcp_twsk(sk);
1040	struct tcp_key key = {};
1041#ifdef CONFIG_TCP_AO
1042	struct tcp_ao_info *ao_info;
1043
1044	if (static_branch_unlikely(&tcp_ao_needed.key)) {
1045		/* FIXME: the segment to-be-acked is not verified yet */
1046		ao_info = rcu_dereference(tcptw->ao_info);
1047		if (ao_info) {
1048			const struct tcp_ao_hdr *aoh;
1049
1050			if (tcp_parse_auth_options(tcp_hdr(skb), NULL, &aoh)) {
1051				inet_twsk_put(tw);
1052				return;
1053			}
1054
1055			if (aoh)
1056				key.ao_key = tcp_ao_established_key(sk, ao_info,
1057								    aoh->rnext_keyid, -1);
1058		}
1059	}
1060	if (key.ao_key) {
1061		struct tcp_ao_key *rnext_key;
1062
1063		key.traffic_key = snd_other_key(key.ao_key);
1064		key.sne = READ_ONCE(ao_info->snd_sne);
1065		rnext_key = READ_ONCE(ao_info->rnext_key);
1066		key.rcv_next = rnext_key->rcvid;
1067		key.type = TCP_KEY_AO;
1068#else
1069	if (0) {
1070#endif
1071	} else if (static_branch_tcp_md5()) {
 
1072		key.md5_key = tcp_twsk_md5_key(tcptw);
1073		if (key.md5_key)
1074			key.type = TCP_KEY_MD5;
 
1075	}
1076
1077	tcp_v4_send_ack(sk, skb,
1078			tcptw->tw_snd_nxt, READ_ONCE(tcptw->tw_rcv_nxt),
1079			tcptw->tw_rcv_wnd >> tw->tw_rcv_wscale,
1080			tcp_tw_tsval(tcptw),
1081			READ_ONCE(tcptw->tw_ts_recent),
1082			tw->tw_bound_dev_if, &key,
1083			tw->tw_transparent ? IP_REPLY_ARG_NOSRCCHECK : 0,
1084			tw->tw_tos,
1085			tw->tw_txhash);
1086
1087	inet_twsk_put(tw);
1088}
1089
1090static void tcp_v4_reqsk_send_ack(const struct sock *sk, struct sk_buff *skb,
1091				  struct request_sock *req)
1092{
1093	struct tcp_key key = {};
1094
1095	/* sk->sk_state == TCP_LISTEN -> for regular TCP_SYN_RECV
1096	 * sk->sk_state == TCP_SYN_RECV -> for Fast Open.
1097	 */
1098	u32 seq = (sk->sk_state == TCP_LISTEN) ? tcp_rsk(req)->snt_isn + 1 :
1099					     tcp_sk(sk)->snd_nxt;
1100
1101#ifdef CONFIG_TCP_AO
1102	if (static_branch_unlikely(&tcp_ao_needed.key) &&
1103	    tcp_rsk_used_ao(req)) {
1104		const union tcp_md5_addr *addr;
1105		const struct tcp_ao_hdr *aoh;
1106		int l3index;
1107
1108		/* Invalid TCP option size or twice included auth */
1109		if (tcp_parse_auth_options(tcp_hdr(skb), NULL, &aoh))
1110			return;
1111		if (!aoh)
1112			return;
1113
1114		addr = (union tcp_md5_addr *)&ip_hdr(skb)->saddr;
1115		l3index = tcp_v4_sdif(skb) ? inet_iif(skb) : 0;
1116		key.ao_key = tcp_ao_do_lookup(sk, l3index, addr, AF_INET,
1117					      aoh->rnext_keyid, -1);
1118		if (unlikely(!key.ao_key)) {
1119			/* Send ACK with any matching MKT for the peer */
1120			key.ao_key = tcp_ao_do_lookup(sk, l3index, addr, AF_INET, -1, -1);
1121			/* Matching key disappeared (user removed the key?)
1122			 * let the handshake timeout.
1123			 */
1124			if (!key.ao_key) {
1125				net_info_ratelimited("TCP-AO key for (%pI4, %d)->(%pI4, %d) suddenly disappeared, won't ACK new connection\n",
1126						     addr,
1127						     ntohs(tcp_hdr(skb)->source),
1128						     &ip_hdr(skb)->daddr,
1129						     ntohs(tcp_hdr(skb)->dest));
1130				return;
1131			}
1132		}
1133		key.traffic_key = kmalloc(tcp_ao_digest_size(key.ao_key), GFP_ATOMIC);
1134		if (!key.traffic_key)
1135			return;
1136
1137		key.type = TCP_KEY_AO;
1138		key.rcv_next = aoh->keyid;
1139		tcp_v4_ao_calc_key_rsk(key.ao_key, key.traffic_key, req);
1140#else
1141	if (0) {
1142#endif
1143	} else if (static_branch_tcp_md5()) {
 
1144		const union tcp_md5_addr *addr;
1145		int l3index;
1146
1147		addr = (union tcp_md5_addr *)&ip_hdr(skb)->saddr;
1148		l3index = tcp_v4_sdif(skb) ? inet_iif(skb) : 0;
1149		key.md5_key = tcp_md5_do_lookup(sk, l3index, addr, AF_INET);
1150		if (key.md5_key)
1151			key.type = TCP_KEY_MD5;
 
1152	}
1153
 
 
 
 
 
1154	tcp_v4_send_ack(sk, skb, seq,
1155			tcp_rsk(req)->rcv_nxt,
1156			tcp_synack_window(req) >> inet_rsk(req)->rcv_wscale,
1157			tcp_rsk_tsval(tcp_rsk(req)),
1158			READ_ONCE(req->ts_recent),
1159			0, &key,
1160			inet_rsk(req)->no_srccheck ? IP_REPLY_ARG_NOSRCCHECK : 0,
1161			ip_hdr(skb)->tos,
1162			READ_ONCE(tcp_rsk(req)->txhash));
1163	if (tcp_key_is_ao(&key))
1164		kfree(key.traffic_key);
1165}
1166
1167/*
1168 *	Send a SYN-ACK after having received a SYN.
1169 *	This still operates on a request_sock only, not on a big
1170 *	socket.
1171 */
1172static int tcp_v4_send_synack(const struct sock *sk, struct dst_entry *dst,
1173			      struct flowi *fl,
1174			      struct request_sock *req,
1175			      struct tcp_fastopen_cookie *foc,
1176			      enum tcp_synack_type synack_type,
1177			      struct sk_buff *syn_skb)
1178{
1179	const struct inet_request_sock *ireq = inet_rsk(req);
1180	struct flowi4 fl4;
1181	int err = -1;
1182	struct sk_buff *skb;
1183	u8 tos;
1184
1185	/* First, grab a route. */
1186	if (!dst && (dst = inet_csk_route_req(sk, &fl4, req)) == NULL)
1187		return -1;
1188
1189	skb = tcp_make_synack(sk, dst, req, foc, synack_type, syn_skb);
1190
1191	if (skb) {
1192		__tcp_v4_send_check(skb, ireq->ir_loc_addr, ireq->ir_rmt_addr);
1193
1194		tos = READ_ONCE(inet_sk(sk)->tos);
1195
1196		if (READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_reflect_tos))
1197			tos = (tcp_rsk(req)->syn_tos & ~INET_ECN_MASK) |
1198			      (tos & INET_ECN_MASK);
1199
1200		if (!INET_ECN_is_capable(tos) &&
1201		    tcp_bpf_ca_needs_ecn((struct sock *)req))
1202			tos |= INET_ECN_ECT_0;
1203
1204		rcu_read_lock();
1205		err = ip_build_and_send_pkt(skb, sk, ireq->ir_loc_addr,
1206					    ireq->ir_rmt_addr,
1207					    rcu_dereference(ireq->ireq_opt),
1208					    tos);
1209		rcu_read_unlock();
1210		err = net_xmit_eval(err);
1211	}
1212
1213	return err;
1214}
1215
1216/*
1217 *	IPv4 request_sock destructor.
1218 */
1219static void tcp_v4_reqsk_destructor(struct request_sock *req)
1220{
1221	kfree(rcu_dereference_protected(inet_rsk(req)->ireq_opt, 1));
1222}
1223
1224#ifdef CONFIG_TCP_MD5SIG
1225/*
1226 * RFC2385 MD5 checksumming requires a mapping of
1227 * IP address->MD5 Key.
1228 * We need to maintain these in the sk structure.
1229 */
1230
1231DEFINE_STATIC_KEY_DEFERRED_FALSE(tcp_md5_needed, HZ);
1232EXPORT_SYMBOL(tcp_md5_needed);
1233
1234static bool better_md5_match(struct tcp_md5sig_key *old, struct tcp_md5sig_key *new)
1235{
1236	if (!old)
1237		return true;
1238
1239	/* l3index always overrides non-l3index */
1240	if (old->l3index && new->l3index == 0)
1241		return false;
1242	if (old->l3index == 0 && new->l3index)
1243		return true;
1244
1245	return old->prefixlen < new->prefixlen;
1246}
1247
1248/* Find the Key structure for an address.  */
1249struct tcp_md5sig_key *__tcp_md5_do_lookup(const struct sock *sk, int l3index,
1250					   const union tcp_md5_addr *addr,
1251					   int family, bool any_l3index)
1252{
1253	const struct tcp_sock *tp = tcp_sk(sk);
1254	struct tcp_md5sig_key *key;
1255	const struct tcp_md5sig_info *md5sig;
1256	__be32 mask;
1257	struct tcp_md5sig_key *best_match = NULL;
1258	bool match;
1259
1260	/* caller either holds rcu_read_lock() or socket lock */
1261	md5sig = rcu_dereference_check(tp->md5sig_info,
1262				       lockdep_sock_is_held(sk));
1263	if (!md5sig)
1264		return NULL;
1265
1266	hlist_for_each_entry_rcu(key, &md5sig->head, node,
1267				 lockdep_sock_is_held(sk)) {
1268		if (key->family != family)
1269			continue;
1270		if (!any_l3index && key->flags & TCP_MD5SIG_FLAG_IFINDEX &&
1271		    key->l3index != l3index)
1272			continue;
1273		if (family == AF_INET) {
1274			mask = inet_make_mask(key->prefixlen);
1275			match = (key->addr.a4.s_addr & mask) ==
1276				(addr->a4.s_addr & mask);
1277#if IS_ENABLED(CONFIG_IPV6)
1278		} else if (family == AF_INET6) {
1279			match = ipv6_prefix_equal(&key->addr.a6, &addr->a6,
1280						  key->prefixlen);
1281#endif
1282		} else {
1283			match = false;
1284		}
1285
1286		if (match && better_md5_match(best_match, key))
1287			best_match = key;
1288	}
1289	return best_match;
1290}
1291EXPORT_SYMBOL(__tcp_md5_do_lookup);
1292
1293static struct tcp_md5sig_key *tcp_md5_do_lookup_exact(const struct sock *sk,
1294						      const union tcp_md5_addr *addr,
1295						      int family, u8 prefixlen,
1296						      int l3index, u8 flags)
1297{
1298	const struct tcp_sock *tp = tcp_sk(sk);
1299	struct tcp_md5sig_key *key;
1300	unsigned int size = sizeof(struct in_addr);
1301	const struct tcp_md5sig_info *md5sig;
1302
1303	/* caller either holds rcu_read_lock() or socket lock */
1304	md5sig = rcu_dereference_check(tp->md5sig_info,
1305				       lockdep_sock_is_held(sk));
1306	if (!md5sig)
1307		return NULL;
1308#if IS_ENABLED(CONFIG_IPV6)
1309	if (family == AF_INET6)
1310		size = sizeof(struct in6_addr);
1311#endif
1312	hlist_for_each_entry_rcu(key, &md5sig->head, node,
1313				 lockdep_sock_is_held(sk)) {
1314		if (key->family != family)
1315			continue;
1316		if ((key->flags & TCP_MD5SIG_FLAG_IFINDEX) != (flags & TCP_MD5SIG_FLAG_IFINDEX))
1317			continue;
1318		if (key->l3index != l3index)
1319			continue;
1320		if (!memcmp(&key->addr, addr, size) &&
1321		    key->prefixlen == prefixlen)
1322			return key;
1323	}
1324	return NULL;
1325}
1326
1327struct tcp_md5sig_key *tcp_v4_md5_lookup(const struct sock *sk,
1328					 const struct sock *addr_sk)
1329{
1330	const union tcp_md5_addr *addr;
1331	int l3index;
1332
1333	l3index = l3mdev_master_ifindex_by_index(sock_net(sk),
1334						 addr_sk->sk_bound_dev_if);
1335	addr = (const union tcp_md5_addr *)&addr_sk->sk_daddr;
1336	return tcp_md5_do_lookup(sk, l3index, addr, AF_INET);
1337}
1338EXPORT_SYMBOL(tcp_v4_md5_lookup);
1339
1340static int tcp_md5sig_info_add(struct sock *sk, gfp_t gfp)
1341{
1342	struct tcp_sock *tp = tcp_sk(sk);
1343	struct tcp_md5sig_info *md5sig;
1344
1345	md5sig = kmalloc(sizeof(*md5sig), gfp);
1346	if (!md5sig)
1347		return -ENOMEM;
1348
1349	sk_gso_disable(sk);
1350	INIT_HLIST_HEAD(&md5sig->head);
1351	rcu_assign_pointer(tp->md5sig_info, md5sig);
1352	return 0;
1353}
1354
1355/* This can be called on a newly created socket, from other files */
1356static int __tcp_md5_do_add(struct sock *sk, const union tcp_md5_addr *addr,
1357			    int family, u8 prefixlen, int l3index, u8 flags,
1358			    const u8 *newkey, u8 newkeylen, gfp_t gfp)
1359{
1360	/* Add Key to the list */
1361	struct tcp_md5sig_key *key;
1362	struct tcp_sock *tp = tcp_sk(sk);
1363	struct tcp_md5sig_info *md5sig;
1364
1365	key = tcp_md5_do_lookup_exact(sk, addr, family, prefixlen, l3index, flags);
1366	if (key) {
1367		/* Pre-existing entry - just update that one.
1368		 * Note that the key might be used concurrently.
1369		 * data_race() is telling kcsan that we do not care of
1370		 * key mismatches, since changing MD5 key on live flows
1371		 * can lead to packet drops.
1372		 */
1373		data_race(memcpy(key->key, newkey, newkeylen));
1374
1375		/* Pairs with READ_ONCE() in tcp_md5_hash_key().
1376		 * Also note that a reader could catch new key->keylen value
1377		 * but old key->key[], this is the reason we use __GFP_ZERO
1378		 * at sock_kmalloc() time below these lines.
1379		 */
1380		WRITE_ONCE(key->keylen, newkeylen);
1381
1382		return 0;
1383	}
1384
1385	md5sig = rcu_dereference_protected(tp->md5sig_info,
1386					   lockdep_sock_is_held(sk));
1387
1388	key = sock_kmalloc(sk, sizeof(*key), gfp | __GFP_ZERO);
1389	if (!key)
1390		return -ENOMEM;
1391
1392	memcpy(key->key, newkey, newkeylen);
1393	key->keylen = newkeylen;
1394	key->family = family;
1395	key->prefixlen = prefixlen;
1396	key->l3index = l3index;
1397	key->flags = flags;
1398	memcpy(&key->addr, addr,
1399	       (IS_ENABLED(CONFIG_IPV6) && family == AF_INET6) ? sizeof(struct in6_addr) :
1400								 sizeof(struct in_addr));
1401	hlist_add_head_rcu(&key->node, &md5sig->head);
1402	return 0;
1403}
1404
1405int tcp_md5_do_add(struct sock *sk, const union tcp_md5_addr *addr,
1406		   int family, u8 prefixlen, int l3index, u8 flags,
1407		   const u8 *newkey, u8 newkeylen)
1408{
1409	struct tcp_sock *tp = tcp_sk(sk);
1410
1411	if (!rcu_dereference_protected(tp->md5sig_info, lockdep_sock_is_held(sk))) {
1412		if (tcp_md5_alloc_sigpool())
1413			return -ENOMEM;
1414
1415		if (tcp_md5sig_info_add(sk, GFP_KERNEL)) {
1416			tcp_md5_release_sigpool();
1417			return -ENOMEM;
1418		}
1419
1420		if (!static_branch_inc(&tcp_md5_needed.key)) {
1421			struct tcp_md5sig_info *md5sig;
1422
1423			md5sig = rcu_dereference_protected(tp->md5sig_info, lockdep_sock_is_held(sk));
1424			rcu_assign_pointer(tp->md5sig_info, NULL);
1425			kfree_rcu(md5sig, rcu);
1426			tcp_md5_release_sigpool();
1427			return -EUSERS;
1428		}
1429	}
1430
1431	return __tcp_md5_do_add(sk, addr, family, prefixlen, l3index, flags,
1432				newkey, newkeylen, GFP_KERNEL);
1433}
1434EXPORT_SYMBOL(tcp_md5_do_add);
1435
1436int tcp_md5_key_copy(struct sock *sk, const union tcp_md5_addr *addr,
1437		     int family, u8 prefixlen, int l3index,
1438		     struct tcp_md5sig_key *key)
1439{
1440	struct tcp_sock *tp = tcp_sk(sk);
1441
1442	if (!rcu_dereference_protected(tp->md5sig_info, lockdep_sock_is_held(sk))) {
1443		tcp_md5_add_sigpool();
1444
1445		if (tcp_md5sig_info_add(sk, sk_gfp_mask(sk, GFP_ATOMIC))) {
1446			tcp_md5_release_sigpool();
1447			return -ENOMEM;
1448		}
1449
1450		if (!static_key_fast_inc_not_disabled(&tcp_md5_needed.key.key)) {
1451			struct tcp_md5sig_info *md5sig;
1452
1453			md5sig = rcu_dereference_protected(tp->md5sig_info, lockdep_sock_is_held(sk));
1454			net_warn_ratelimited("Too many TCP-MD5 keys in the system\n");
1455			rcu_assign_pointer(tp->md5sig_info, NULL);
1456			kfree_rcu(md5sig, rcu);
1457			tcp_md5_release_sigpool();
1458			return -EUSERS;
1459		}
1460	}
1461
1462	return __tcp_md5_do_add(sk, addr, family, prefixlen, l3index,
1463				key->flags, key->key, key->keylen,
1464				sk_gfp_mask(sk, GFP_ATOMIC));
1465}
1466EXPORT_SYMBOL(tcp_md5_key_copy);
1467
1468int tcp_md5_do_del(struct sock *sk, const union tcp_md5_addr *addr, int family,
1469		   u8 prefixlen, int l3index, u8 flags)
1470{
1471	struct tcp_md5sig_key *key;
1472
1473	key = tcp_md5_do_lookup_exact(sk, addr, family, prefixlen, l3index, flags);
1474	if (!key)
1475		return -ENOENT;
1476	hlist_del_rcu(&key->node);
1477	atomic_sub(sizeof(*key), &sk->sk_omem_alloc);
1478	kfree_rcu(key, rcu);
1479	return 0;
1480}
1481EXPORT_SYMBOL(tcp_md5_do_del);
1482
1483void tcp_clear_md5_list(struct sock *sk)
1484{
1485	struct tcp_sock *tp = tcp_sk(sk);
1486	struct tcp_md5sig_key *key;
1487	struct hlist_node *n;
1488	struct tcp_md5sig_info *md5sig;
1489
1490	md5sig = rcu_dereference_protected(tp->md5sig_info, 1);
1491
1492	hlist_for_each_entry_safe(key, n, &md5sig->head, node) {
1493		hlist_del_rcu(&key->node);
1494		atomic_sub(sizeof(*key), &sk->sk_omem_alloc);
1495		kfree_rcu(key, rcu);
1496	}
1497}
1498
1499static int tcp_v4_parse_md5_keys(struct sock *sk, int optname,
1500				 sockptr_t optval, int optlen)
1501{
1502	struct tcp_md5sig cmd;
1503	struct sockaddr_in *sin = (struct sockaddr_in *)&cmd.tcpm_addr;
1504	const union tcp_md5_addr *addr;
1505	u8 prefixlen = 32;
1506	int l3index = 0;
1507	bool l3flag;
1508	u8 flags;
1509
1510	if (optlen < sizeof(cmd))
1511		return -EINVAL;
1512
1513	if (copy_from_sockptr(&cmd, optval, sizeof(cmd)))
1514		return -EFAULT;
1515
1516	if (sin->sin_family != AF_INET)
1517		return -EINVAL;
1518
1519	flags = cmd.tcpm_flags & TCP_MD5SIG_FLAG_IFINDEX;
1520	l3flag = cmd.tcpm_flags & TCP_MD5SIG_FLAG_IFINDEX;
1521
1522	if (optname == TCP_MD5SIG_EXT &&
1523	    cmd.tcpm_flags & TCP_MD5SIG_FLAG_PREFIX) {
1524		prefixlen = cmd.tcpm_prefixlen;
1525		if (prefixlen > 32)
1526			return -EINVAL;
1527	}
1528
1529	if (optname == TCP_MD5SIG_EXT && cmd.tcpm_ifindex &&
1530	    cmd.tcpm_flags & TCP_MD5SIG_FLAG_IFINDEX) {
1531		struct net_device *dev;
1532
1533		rcu_read_lock();
1534		dev = dev_get_by_index_rcu(sock_net(sk), cmd.tcpm_ifindex);
1535		if (dev && netif_is_l3_master(dev))
1536			l3index = dev->ifindex;
1537
1538		rcu_read_unlock();
1539
1540		/* ok to reference set/not set outside of rcu;
1541		 * right now device MUST be an L3 master
1542		 */
1543		if (!dev || !l3index)
1544			return -EINVAL;
1545	}
1546
1547	addr = (union tcp_md5_addr *)&sin->sin_addr.s_addr;
1548
1549	if (!cmd.tcpm_keylen)
1550		return tcp_md5_do_del(sk, addr, AF_INET, prefixlen, l3index, flags);
1551
1552	if (cmd.tcpm_keylen > TCP_MD5SIG_MAXKEYLEN)
1553		return -EINVAL;
1554
1555	/* Don't allow keys for peers that have a matching TCP-AO key.
1556	 * See the comment in tcp_ao_add_cmd()
1557	 */
1558	if (tcp_ao_required(sk, addr, AF_INET, l3flag ? l3index : -1, false))
1559		return -EKEYREJECTED;
1560
1561	return tcp_md5_do_add(sk, addr, AF_INET, prefixlen, l3index, flags,
1562			      cmd.tcpm_key, cmd.tcpm_keylen);
1563}
1564
1565static int tcp_v4_md5_hash_headers(struct tcp_sigpool *hp,
1566				   __be32 daddr, __be32 saddr,
1567				   const struct tcphdr *th, int nbytes)
1568{
1569	struct tcp4_pseudohdr *bp;
1570	struct scatterlist sg;
1571	struct tcphdr *_th;
1572
1573	bp = hp->scratch;
1574	bp->saddr = saddr;
1575	bp->daddr = daddr;
1576	bp->pad = 0;
1577	bp->protocol = IPPROTO_TCP;
1578	bp->len = cpu_to_be16(nbytes);
1579
1580	_th = (struct tcphdr *)(bp + 1);
1581	memcpy(_th, th, sizeof(*th));
1582	_th->check = 0;
1583
1584	sg_init_one(&sg, bp, sizeof(*bp) + sizeof(*th));
1585	ahash_request_set_crypt(hp->req, &sg, NULL,
1586				sizeof(*bp) + sizeof(*th));
1587	return crypto_ahash_update(hp->req);
1588}
1589
1590static int tcp_v4_md5_hash_hdr(char *md5_hash, const struct tcp_md5sig_key *key,
1591			       __be32 daddr, __be32 saddr, const struct tcphdr *th)
1592{
1593	struct tcp_sigpool hp;
1594
1595	if (tcp_sigpool_start(tcp_md5_sigpool_id, &hp))
1596		goto clear_hash_nostart;
1597
1598	if (crypto_ahash_init(hp.req))
1599		goto clear_hash;
1600	if (tcp_v4_md5_hash_headers(&hp, daddr, saddr, th, th->doff << 2))
1601		goto clear_hash;
1602	if (tcp_md5_hash_key(&hp, key))
1603		goto clear_hash;
1604	ahash_request_set_crypt(hp.req, NULL, md5_hash, 0);
1605	if (crypto_ahash_final(hp.req))
1606		goto clear_hash;
1607
1608	tcp_sigpool_end(&hp);
1609	return 0;
1610
1611clear_hash:
1612	tcp_sigpool_end(&hp);
1613clear_hash_nostart:
1614	memset(md5_hash, 0, 16);
1615	return 1;
1616}
1617
1618int tcp_v4_md5_hash_skb(char *md5_hash, const struct tcp_md5sig_key *key,
1619			const struct sock *sk,
1620			const struct sk_buff *skb)
1621{
1622	const struct tcphdr *th = tcp_hdr(skb);
1623	struct tcp_sigpool hp;
1624	__be32 saddr, daddr;
1625
1626	if (sk) { /* valid for establish/request sockets */
1627		saddr = sk->sk_rcv_saddr;
1628		daddr = sk->sk_daddr;
1629	} else {
1630		const struct iphdr *iph = ip_hdr(skb);
1631		saddr = iph->saddr;
1632		daddr = iph->daddr;
1633	}
1634
1635	if (tcp_sigpool_start(tcp_md5_sigpool_id, &hp))
1636		goto clear_hash_nostart;
1637
1638	if (crypto_ahash_init(hp.req))
1639		goto clear_hash;
1640
1641	if (tcp_v4_md5_hash_headers(&hp, daddr, saddr, th, skb->len))
1642		goto clear_hash;
1643	if (tcp_sigpool_hash_skb_data(&hp, skb, th->doff << 2))
1644		goto clear_hash;
1645	if (tcp_md5_hash_key(&hp, key))
1646		goto clear_hash;
1647	ahash_request_set_crypt(hp.req, NULL, md5_hash, 0);
1648	if (crypto_ahash_final(hp.req))
1649		goto clear_hash;
1650
1651	tcp_sigpool_end(&hp);
1652	return 0;
1653
1654clear_hash:
1655	tcp_sigpool_end(&hp);
1656clear_hash_nostart:
1657	memset(md5_hash, 0, 16);
1658	return 1;
1659}
1660EXPORT_SYMBOL(tcp_v4_md5_hash_skb);
1661
1662#endif
1663
1664static void tcp_v4_init_req(struct request_sock *req,
1665			    const struct sock *sk_listener,
1666			    struct sk_buff *skb)
1667{
1668	struct inet_request_sock *ireq = inet_rsk(req);
1669	struct net *net = sock_net(sk_listener);
1670
1671	sk_rcv_saddr_set(req_to_sk(req), ip_hdr(skb)->daddr);
1672	sk_daddr_set(req_to_sk(req), ip_hdr(skb)->saddr);
1673	RCU_INIT_POINTER(ireq->ireq_opt, tcp_v4_save_options(net, skb));
1674}
1675
1676static struct dst_entry *tcp_v4_route_req(const struct sock *sk,
1677					  struct sk_buff *skb,
1678					  struct flowi *fl,
1679					  struct request_sock *req,
1680					  u32 tw_isn)
1681{
1682	tcp_v4_init_req(req, sk, skb);
1683
1684	if (security_inet_conn_request(sk, skb, req))
1685		return NULL;
1686
1687	return inet_csk_route_req(sk, &fl->u.ip4, req);
1688}
1689
1690struct request_sock_ops tcp_request_sock_ops __read_mostly = {
1691	.family		=	PF_INET,
1692	.obj_size	=	sizeof(struct tcp_request_sock),
1693	.rtx_syn_ack	=	tcp_rtx_synack,
1694	.send_ack	=	tcp_v4_reqsk_send_ack,
1695	.destructor	=	tcp_v4_reqsk_destructor,
1696	.send_reset	=	tcp_v4_send_reset,
1697	.syn_ack_timeout =	tcp_syn_ack_timeout,
1698};
1699
1700const struct tcp_request_sock_ops tcp_request_sock_ipv4_ops = {
1701	.mss_clamp	=	TCP_MSS_DEFAULT,
1702#ifdef CONFIG_TCP_MD5SIG
1703	.req_md5_lookup	=	tcp_v4_md5_lookup,
1704	.calc_md5_hash	=	tcp_v4_md5_hash_skb,
1705#endif
1706#ifdef CONFIG_TCP_AO
1707	.ao_lookup	=	tcp_v4_ao_lookup_rsk,
1708	.ao_calc_key	=	tcp_v4_ao_calc_key_rsk,
1709	.ao_synack_hash	=	tcp_v4_ao_synack_hash,
1710#endif
1711#ifdef CONFIG_SYN_COOKIES
1712	.cookie_init_seq =	cookie_v4_init_sequence,
1713#endif
1714	.route_req	=	tcp_v4_route_req,
1715	.init_seq	=	tcp_v4_init_seq,
1716	.init_ts_off	=	tcp_v4_init_ts_off,
1717	.send_synack	=	tcp_v4_send_synack,
1718};
1719
1720int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb)
1721{
1722	/* Never answer to SYNs send to broadcast or multicast */
1723	if (skb_rtable(skb)->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))
1724		goto drop;
1725
1726	return tcp_conn_request(&tcp_request_sock_ops,
1727				&tcp_request_sock_ipv4_ops, sk, skb);
1728
1729drop:
1730	tcp_listendrop(sk);
1731	return 0;
1732}
1733EXPORT_SYMBOL(tcp_v4_conn_request);
1734
1735
1736/*
1737 * The three way handshake has completed - we got a valid synack -
1738 * now create the new socket.
1739 */
1740struct sock *tcp_v4_syn_recv_sock(const struct sock *sk, struct sk_buff *skb,
1741				  struct request_sock *req,
1742				  struct dst_entry *dst,
1743				  struct request_sock *req_unhash,
1744				  bool *own_req)
1745{
1746	struct inet_request_sock *ireq;
1747	bool found_dup_sk = false;
1748	struct inet_sock *newinet;
1749	struct tcp_sock *newtp;
1750	struct sock *newsk;
1751#ifdef CONFIG_TCP_MD5SIG
1752	const union tcp_md5_addr *addr;
1753	struct tcp_md5sig_key *key;
1754	int l3index;
1755#endif
1756	struct ip_options_rcu *inet_opt;
1757
1758	if (sk_acceptq_is_full(sk))
1759		goto exit_overflow;
1760
1761	newsk = tcp_create_openreq_child(sk, req, skb);
1762	if (!newsk)
1763		goto exit_nonewsk;
1764
1765	newsk->sk_gso_type = SKB_GSO_TCPV4;
1766	inet_sk_rx_dst_set(newsk, skb);
1767
1768	newtp		      = tcp_sk(newsk);
1769	newinet		      = inet_sk(newsk);
1770	ireq		      = inet_rsk(req);
1771	sk_daddr_set(newsk, ireq->ir_rmt_addr);
1772	sk_rcv_saddr_set(newsk, ireq->ir_loc_addr);
1773	newsk->sk_bound_dev_if = ireq->ir_iif;
1774	newinet->inet_saddr   = ireq->ir_loc_addr;
1775	inet_opt	      = rcu_dereference(ireq->ireq_opt);
1776	RCU_INIT_POINTER(newinet->inet_opt, inet_opt);
1777	newinet->mc_index     = inet_iif(skb);
1778	newinet->mc_ttl	      = ip_hdr(skb)->ttl;
1779	newinet->rcv_tos      = ip_hdr(skb)->tos;
1780	inet_csk(newsk)->icsk_ext_hdr_len = 0;
1781	if (inet_opt)
1782		inet_csk(newsk)->icsk_ext_hdr_len = inet_opt->opt.optlen;
1783	atomic_set(&newinet->inet_id, get_random_u16());
1784
1785	/* Set ToS of the new socket based upon the value of incoming SYN.
1786	 * ECT bits are set later in tcp_init_transfer().
1787	 */
1788	if (READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_reflect_tos))
1789		newinet->tos = tcp_rsk(req)->syn_tos & ~INET_ECN_MASK;
1790
1791	if (!dst) {
1792		dst = inet_csk_route_child_sock(sk, newsk, req);
1793		if (!dst)
1794			goto put_and_exit;
1795	} else {
1796		/* syncookie case : see end of cookie_v4_check() */
1797	}
1798	sk_setup_caps(newsk, dst);
1799
1800	tcp_ca_openreq_child(newsk, dst);
1801
1802	tcp_sync_mss(newsk, dst_mtu(dst));
1803	newtp->advmss = tcp_mss_clamp(tcp_sk(sk), dst_metric_advmss(dst));
1804
1805	tcp_initialize_rcv_mss(newsk);
1806
1807#ifdef CONFIG_TCP_MD5SIG
1808	l3index = l3mdev_master_ifindex_by_index(sock_net(sk), ireq->ir_iif);
1809	/* Copy over the MD5 key from the original socket */
1810	addr = (union tcp_md5_addr *)&newinet->inet_daddr;
1811	key = tcp_md5_do_lookup(sk, l3index, addr, AF_INET);
1812	if (key && !tcp_rsk_used_ao(req)) {
1813		if (tcp_md5_key_copy(newsk, addr, AF_INET, 32, l3index, key))
1814			goto put_and_exit;
1815		sk_gso_disable(newsk);
1816	}
1817#endif
1818#ifdef CONFIG_TCP_AO
1819	if (tcp_ao_copy_all_matching(sk, newsk, req, skb, AF_INET))
1820		goto put_and_exit; /* OOM, release back memory */
1821#endif
1822
1823	if (__inet_inherit_port(sk, newsk) < 0)
1824		goto put_and_exit;
1825	*own_req = inet_ehash_nolisten(newsk, req_to_sk(req_unhash),
1826				       &found_dup_sk);
1827	if (likely(*own_req)) {
1828		tcp_move_syn(newtp, req);
1829		ireq->ireq_opt = NULL;
1830	} else {
1831		newinet->inet_opt = NULL;
1832
1833		if (!req_unhash && found_dup_sk) {
1834			/* This code path should only be executed in the
1835			 * syncookie case only
1836			 */
1837			bh_unlock_sock(newsk);
1838			sock_put(newsk);
1839			newsk = NULL;
1840		}
1841	}
1842	return newsk;
1843
1844exit_overflow:
1845	NET_INC_STATS(sock_net(sk), LINUX_MIB_LISTENOVERFLOWS);
1846exit_nonewsk:
1847	dst_release(dst);
1848exit:
1849	tcp_listendrop(sk);
1850	return NULL;
1851put_and_exit:
1852	newinet->inet_opt = NULL;
1853	inet_csk_prepare_forced_close(newsk);
1854	tcp_done(newsk);
1855	goto exit;
1856}
1857EXPORT_SYMBOL(tcp_v4_syn_recv_sock);
1858
1859static struct sock *tcp_v4_cookie_check(struct sock *sk, struct sk_buff *skb)
1860{
1861#ifdef CONFIG_SYN_COOKIES
1862	const struct tcphdr *th = tcp_hdr(skb);
1863
1864	if (!th->syn)
1865		sk = cookie_v4_check(sk, skb);
1866#endif
1867	return sk;
1868}
1869
1870u16 tcp_v4_get_syncookie(struct sock *sk, struct iphdr *iph,
1871			 struct tcphdr *th, u32 *cookie)
1872{
1873	u16 mss = 0;
1874#ifdef CONFIG_SYN_COOKIES
1875	mss = tcp_get_syncookie_mss(&tcp_request_sock_ops,
1876				    &tcp_request_sock_ipv4_ops, sk, th);
1877	if (mss) {
1878		*cookie = __cookie_v4_init_sequence(iph, th, &mss);
1879		tcp_synq_overflow(sk);
1880	}
1881#endif
1882	return mss;
1883}
1884
1885INDIRECT_CALLABLE_DECLARE(struct dst_entry *ipv4_dst_check(struct dst_entry *,
1886							   u32));
1887/* The socket must have it's spinlock held when we get
1888 * here, unless it is a TCP_LISTEN socket.
1889 *
1890 * We have a potential double-lock case here, so even when
1891 * doing backlog processing we use the BH locking scheme.
1892 * This is because we cannot sleep with the original spinlock
1893 * held.
1894 */
1895int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb)
1896{
1897	enum skb_drop_reason reason;
1898	struct sock *rsk;
1899
1900	if (sk->sk_state == TCP_ESTABLISHED) { /* Fast path */
1901		struct dst_entry *dst;
1902
1903		dst = rcu_dereference_protected(sk->sk_rx_dst,
1904						lockdep_sock_is_held(sk));
1905
1906		sock_rps_save_rxhash(sk, skb);
1907		sk_mark_napi_id(sk, skb);
1908		if (dst) {
1909			if (sk->sk_rx_dst_ifindex != skb->skb_iif ||
1910			    !INDIRECT_CALL_1(dst->ops->check, ipv4_dst_check,
1911					     dst, 0)) {
1912				RCU_INIT_POINTER(sk->sk_rx_dst, NULL);
1913				dst_release(dst);
1914			}
1915		}
1916		tcp_rcv_established(sk, skb);
1917		return 0;
1918	}
1919
 
1920	if (tcp_checksum_complete(skb))
1921		goto csum_err;
1922
1923	if (sk->sk_state == TCP_LISTEN) {
1924		struct sock *nsk = tcp_v4_cookie_check(sk, skb);
1925
1926		if (!nsk)
1927			return 0;
1928		if (nsk != sk) {
1929			reason = tcp_child_process(sk, nsk, skb);
1930			if (reason) {
1931				rsk = nsk;
1932				goto reset;
1933			}
1934			return 0;
1935		}
1936	} else
1937		sock_rps_save_rxhash(sk, skb);
1938
1939	reason = tcp_rcv_state_process(sk, skb);
1940	if (reason) {
1941		rsk = sk;
1942		goto reset;
1943	}
1944	return 0;
1945
1946reset:
1947	tcp_v4_send_reset(rsk, skb, sk_rst_convert_drop_reason(reason));
1948discard:
1949	sk_skb_reason_drop(sk, skb, reason);
1950	/* Be careful here. If this function gets more complicated and
1951	 * gcc suffers from register pressure on the x86, sk (in %ebx)
1952	 * might be destroyed here. This current version compiles correctly,
1953	 * but you have been warned.
1954	 */
1955	return 0;
1956
1957csum_err:
1958	reason = SKB_DROP_REASON_TCP_CSUM;
1959	trace_tcp_bad_csum(skb);
1960	TCP_INC_STATS(sock_net(sk), TCP_MIB_CSUMERRORS);
1961	TCP_INC_STATS(sock_net(sk), TCP_MIB_INERRS);
1962	goto discard;
1963}
1964EXPORT_SYMBOL(tcp_v4_do_rcv);
1965
1966int tcp_v4_early_demux(struct sk_buff *skb)
1967{
1968	struct net *net = dev_net(skb->dev);
1969	const struct iphdr *iph;
1970	const struct tcphdr *th;
1971	struct sock *sk;
1972
1973	if (skb->pkt_type != PACKET_HOST)
1974		return 0;
1975
1976	if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct tcphdr)))
1977		return 0;
1978
1979	iph = ip_hdr(skb);
1980	th = tcp_hdr(skb);
1981
1982	if (th->doff < sizeof(struct tcphdr) / 4)
1983		return 0;
1984
1985	sk = __inet_lookup_established(net, net->ipv4.tcp_death_row.hashinfo,
1986				       iph->saddr, th->source,
1987				       iph->daddr, ntohs(th->dest),
1988				       skb->skb_iif, inet_sdif(skb));
1989	if (sk) {
1990		skb->sk = sk;
1991		skb->destructor = sock_edemux;
1992		if (sk_fullsock(sk)) {
1993			struct dst_entry *dst = rcu_dereference(sk->sk_rx_dst);
1994
1995			if (dst)
1996				dst = dst_check(dst, 0);
1997			if (dst &&
1998			    sk->sk_rx_dst_ifindex == skb->skb_iif)
1999				skb_dst_set_noref(skb, dst);
2000		}
2001	}
2002	return 0;
2003}
2004
2005bool tcp_add_backlog(struct sock *sk, struct sk_buff *skb,
2006		     enum skb_drop_reason *reason)
2007{
2008	u32 tail_gso_size, tail_gso_segs;
2009	struct skb_shared_info *shinfo;
2010	const struct tcphdr *th;
2011	struct tcphdr *thtail;
2012	struct sk_buff *tail;
2013	unsigned int hdrlen;
2014	bool fragstolen;
2015	u32 gso_segs;
2016	u32 gso_size;
2017	u64 limit;
2018	int delta;
2019
2020	/* In case all data was pulled from skb frags (in __pskb_pull_tail()),
2021	 * we can fix skb->truesize to its real value to avoid future drops.
2022	 * This is valid because skb is not yet charged to the socket.
2023	 * It has been noticed pure SACK packets were sometimes dropped
2024	 * (if cooked by drivers without copybreak feature).
2025	 */
2026	skb_condense(skb);
2027
2028	tcp_cleanup_skb(skb);
2029
2030	if (unlikely(tcp_checksum_complete(skb))) {
2031		bh_unlock_sock(sk);
2032		trace_tcp_bad_csum(skb);
2033		*reason = SKB_DROP_REASON_TCP_CSUM;
2034		__TCP_INC_STATS(sock_net(sk), TCP_MIB_CSUMERRORS);
2035		__TCP_INC_STATS(sock_net(sk), TCP_MIB_INERRS);
2036		return true;
2037	}
2038
2039	/* Attempt coalescing to last skb in backlog, even if we are
2040	 * above the limits.
2041	 * This is okay because skb capacity is limited to MAX_SKB_FRAGS.
2042	 */
2043	th = (const struct tcphdr *)skb->data;
2044	hdrlen = th->doff * 4;
2045
2046	tail = sk->sk_backlog.tail;
2047	if (!tail)
2048		goto no_coalesce;
2049	thtail = (struct tcphdr *)tail->data;
2050
2051	if (TCP_SKB_CB(tail)->end_seq != TCP_SKB_CB(skb)->seq ||
2052	    TCP_SKB_CB(tail)->ip_dsfield != TCP_SKB_CB(skb)->ip_dsfield ||
2053	    ((TCP_SKB_CB(tail)->tcp_flags |
2054	      TCP_SKB_CB(skb)->tcp_flags) & (TCPHDR_SYN | TCPHDR_RST | TCPHDR_URG)) ||
2055	    !((TCP_SKB_CB(tail)->tcp_flags &
2056	      TCP_SKB_CB(skb)->tcp_flags) & TCPHDR_ACK) ||
2057	    ((TCP_SKB_CB(tail)->tcp_flags ^
2058	      TCP_SKB_CB(skb)->tcp_flags) & (TCPHDR_ECE | TCPHDR_CWR)) ||
2059	    !tcp_skb_can_collapse_rx(tail, skb) ||
 
 
 
2060	    thtail->doff != th->doff ||
2061	    memcmp(thtail + 1, th + 1, hdrlen - sizeof(*th)))
2062		goto no_coalesce;
2063
2064	__skb_pull(skb, hdrlen);
2065
2066	shinfo = skb_shinfo(skb);
2067	gso_size = shinfo->gso_size ?: skb->len;
2068	gso_segs = shinfo->gso_segs ?: 1;
2069
2070	shinfo = skb_shinfo(tail);
2071	tail_gso_size = shinfo->gso_size ?: (tail->len - hdrlen);
2072	tail_gso_segs = shinfo->gso_segs ?: 1;
2073
2074	if (skb_try_coalesce(tail, skb, &fragstolen, &delta)) {
2075		TCP_SKB_CB(tail)->end_seq = TCP_SKB_CB(skb)->end_seq;
2076
2077		if (likely(!before(TCP_SKB_CB(skb)->ack_seq, TCP_SKB_CB(tail)->ack_seq))) {
2078			TCP_SKB_CB(tail)->ack_seq = TCP_SKB_CB(skb)->ack_seq;
2079			thtail->window = th->window;
2080		}
2081
2082		/* We have to update both TCP_SKB_CB(tail)->tcp_flags and
2083		 * thtail->fin, so that the fast path in tcp_rcv_established()
2084		 * is not entered if we append a packet with a FIN.
2085		 * SYN, RST, URG are not present.
2086		 * ACK is set on both packets.
2087		 * PSH : we do not really care in TCP stack,
2088		 *       at least for 'GRO' packets.
2089		 */
2090		thtail->fin |= th->fin;
2091		TCP_SKB_CB(tail)->tcp_flags |= TCP_SKB_CB(skb)->tcp_flags;
2092
2093		if (TCP_SKB_CB(skb)->has_rxtstamp) {
2094			TCP_SKB_CB(tail)->has_rxtstamp = true;
2095			tail->tstamp = skb->tstamp;
2096			skb_hwtstamps(tail)->hwtstamp = skb_hwtstamps(skb)->hwtstamp;
2097		}
2098
2099		/* Not as strict as GRO. We only need to carry mss max value */
2100		shinfo->gso_size = max(gso_size, tail_gso_size);
2101		shinfo->gso_segs = min_t(u32, gso_segs + tail_gso_segs, 0xFFFF);
2102
2103		sk->sk_backlog.len += delta;
2104		__NET_INC_STATS(sock_net(sk),
2105				LINUX_MIB_TCPBACKLOGCOALESCE);
2106		kfree_skb_partial(skb, fragstolen);
2107		return false;
2108	}
2109	__skb_push(skb, hdrlen);
2110
2111no_coalesce:
2112	/* sk->sk_backlog.len is reset only at the end of __release_sock().
2113	 * Both sk->sk_backlog.len and sk->sk_rmem_alloc could reach
2114	 * sk_rcvbuf in normal conditions.
2115	 */
2116	limit = ((u64)READ_ONCE(sk->sk_rcvbuf)) << 1;
2117
2118	limit += ((u32)READ_ONCE(sk->sk_sndbuf)) >> 1;
2119
2120	/* Only socket owner can try to collapse/prune rx queues
2121	 * to reduce memory overhead, so add a little headroom here.
2122	 * Few sockets backlog are possibly concurrently non empty.
2123	 */
2124	limit += 64 * 1024;
2125
2126	limit = min_t(u64, limit, UINT_MAX);
2127
2128	if (unlikely(sk_add_backlog(sk, skb, limit))) {
2129		bh_unlock_sock(sk);
2130		*reason = SKB_DROP_REASON_SOCKET_BACKLOG;
2131		__NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPBACKLOGDROP);
2132		return true;
2133	}
2134	return false;
2135}
2136EXPORT_SYMBOL(tcp_add_backlog);
2137
2138int tcp_filter(struct sock *sk, struct sk_buff *skb)
2139{
2140	struct tcphdr *th = (struct tcphdr *)skb->data;
2141
2142	return sk_filter_trim_cap(sk, skb, th->doff * 4);
2143}
2144EXPORT_SYMBOL(tcp_filter);
2145
2146static void tcp_v4_restore_cb(struct sk_buff *skb)
2147{
2148	memmove(IPCB(skb), &TCP_SKB_CB(skb)->header.h4,
2149		sizeof(struct inet_skb_parm));
2150}
2151
2152static void tcp_v4_fill_cb(struct sk_buff *skb, const struct iphdr *iph,
2153			   const struct tcphdr *th)
2154{
2155	/* This is tricky : We move IPCB at its correct location into TCP_SKB_CB()
2156	 * barrier() makes sure compiler wont play fool^Waliasing games.
2157	 */
2158	memmove(&TCP_SKB_CB(skb)->header.h4, IPCB(skb),
2159		sizeof(struct inet_skb_parm));
2160	barrier();
2161
2162	TCP_SKB_CB(skb)->seq = ntohl(th->seq);
2163	TCP_SKB_CB(skb)->end_seq = (TCP_SKB_CB(skb)->seq + th->syn + th->fin +
2164				    skb->len - th->doff * 4);
2165	TCP_SKB_CB(skb)->ack_seq = ntohl(th->ack_seq);
2166	TCP_SKB_CB(skb)->tcp_flags = tcp_flag_byte(th);
 
2167	TCP_SKB_CB(skb)->ip_dsfield = ipv4_get_dsfield(iph);
2168	TCP_SKB_CB(skb)->sacked	 = 0;
2169	TCP_SKB_CB(skb)->has_rxtstamp =
2170			skb->tstamp || skb_hwtstamps(skb)->hwtstamp;
2171}
2172
2173/*
2174 *	From tcp_input.c
2175 */
2176
2177int tcp_v4_rcv(struct sk_buff *skb)
2178{
2179	struct net *net = dev_net(skb->dev);
2180	enum skb_drop_reason drop_reason;
2181	int sdif = inet_sdif(skb);
2182	int dif = inet_iif(skb);
2183	const struct iphdr *iph;
2184	const struct tcphdr *th;
2185	struct sock *sk = NULL;
2186	bool refcounted;
 
2187	int ret;
2188	u32 isn;
2189
2190	drop_reason = SKB_DROP_REASON_NOT_SPECIFIED;
2191	if (skb->pkt_type != PACKET_HOST)
2192		goto discard_it;
2193
2194	/* Count it even if it's bad */
2195	__TCP_INC_STATS(net, TCP_MIB_INSEGS);
2196
2197	if (!pskb_may_pull(skb, sizeof(struct tcphdr)))
2198		goto discard_it;
2199
2200	th = (const struct tcphdr *)skb->data;
2201
2202	if (unlikely(th->doff < sizeof(struct tcphdr) / 4)) {
2203		drop_reason = SKB_DROP_REASON_PKT_TOO_SMALL;
2204		goto bad_packet;
2205	}
2206	if (!pskb_may_pull(skb, th->doff * 4))
2207		goto discard_it;
2208
2209	/* An explanation is required here, I think.
2210	 * Packet length and doff are validated by header prediction,
2211	 * provided case of th->doff==0 is eliminated.
2212	 * So, we defer the checks. */
2213
2214	if (skb_checksum_init(skb, IPPROTO_TCP, inet_compute_pseudo))
2215		goto csum_error;
2216
2217	th = (const struct tcphdr *)skb->data;
2218	iph = ip_hdr(skb);
2219lookup:
2220	sk = __inet_lookup_skb(net->ipv4.tcp_death_row.hashinfo,
2221			       skb, __tcp_hdrlen(th), th->source,
2222			       th->dest, sdif, &refcounted);
2223	if (!sk)
2224		goto no_tcp_socket;
2225
 
2226	if (sk->sk_state == TCP_TIME_WAIT)
2227		goto do_time_wait;
2228
2229	if (sk->sk_state == TCP_NEW_SYN_RECV) {
2230		struct request_sock *req = inet_reqsk(sk);
2231		bool req_stolen = false;
2232		struct sock *nsk;
2233
2234		sk = req->rsk_listener;
2235		if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
2236			drop_reason = SKB_DROP_REASON_XFRM_POLICY;
2237		else
2238			drop_reason = tcp_inbound_hash(sk, req, skb,
2239						       &iph->saddr, &iph->daddr,
2240						       AF_INET, dif, sdif);
2241		if (unlikely(drop_reason)) {
2242			sk_drops_add(sk, skb);
2243			reqsk_put(req);
2244			goto discard_it;
2245		}
2246		if (tcp_checksum_complete(skb)) {
2247			reqsk_put(req);
2248			goto csum_error;
2249		}
2250		if (unlikely(sk->sk_state != TCP_LISTEN)) {
2251			nsk = reuseport_migrate_sock(sk, req_to_sk(req), skb);
2252			if (!nsk) {
2253				inet_csk_reqsk_queue_drop_and_put(sk, req);
2254				goto lookup;
2255			}
2256			sk = nsk;
2257			/* reuseport_migrate_sock() has already held one sk_refcnt
2258			 * before returning.
2259			 */
2260		} else {
2261			/* We own a reference on the listener, increase it again
2262			 * as we might lose it too soon.
2263			 */
2264			sock_hold(sk);
2265		}
2266		refcounted = true;
2267		nsk = NULL;
2268		if (!tcp_filter(sk, skb)) {
2269			th = (const struct tcphdr *)skb->data;
2270			iph = ip_hdr(skb);
2271			tcp_v4_fill_cb(skb, iph, th);
2272			nsk = tcp_check_req(sk, skb, req, false, &req_stolen);
2273		} else {
2274			drop_reason = SKB_DROP_REASON_SOCKET_FILTER;
2275		}
2276		if (!nsk) {
2277			reqsk_put(req);
2278			if (req_stolen) {
2279				/* Another cpu got exclusive access to req
2280				 * and created a full blown socket.
2281				 * Try to feed this packet to this socket
2282				 * instead of discarding it.
2283				 */
2284				tcp_v4_restore_cb(skb);
2285				sock_put(sk);
2286				goto lookup;
2287			}
2288			goto discard_and_relse;
2289		}
2290		nf_reset_ct(skb);
2291		if (nsk == sk) {
2292			reqsk_put(req);
2293			tcp_v4_restore_cb(skb);
 
 
 
2294		} else {
2295			drop_reason = tcp_child_process(sk, nsk, skb);
2296			if (drop_reason) {
2297				enum sk_rst_reason rst_reason;
2298
2299				rst_reason = sk_rst_convert_drop_reason(drop_reason);
2300				tcp_v4_send_reset(nsk, skb, rst_reason);
2301				goto discard_and_relse;
2302			}
2303			sock_put(sk);
2304			return 0;
2305		}
2306	}
2307
2308process:
2309	if (static_branch_unlikely(&ip4_min_ttl)) {
2310		/* min_ttl can be changed concurrently from do_ip_setsockopt() */
2311		if (unlikely(iph->ttl < READ_ONCE(inet_sk(sk)->min_ttl))) {
2312			__NET_INC_STATS(net, LINUX_MIB_TCPMINTTLDROP);
2313			drop_reason = SKB_DROP_REASON_TCP_MINTTL;
2314			goto discard_and_relse;
2315		}
2316	}
2317
2318	if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb)) {
2319		drop_reason = SKB_DROP_REASON_XFRM_POLICY;
2320		goto discard_and_relse;
2321	}
2322
2323	drop_reason = tcp_inbound_hash(sk, NULL, skb, &iph->saddr, &iph->daddr,
2324				       AF_INET, dif, sdif);
2325	if (drop_reason)
2326		goto discard_and_relse;
2327
2328	nf_reset_ct(skb);
2329
2330	if (tcp_filter(sk, skb)) {
2331		drop_reason = SKB_DROP_REASON_SOCKET_FILTER;
2332		goto discard_and_relse;
2333	}
2334	th = (const struct tcphdr *)skb->data;
2335	iph = ip_hdr(skb);
2336	tcp_v4_fill_cb(skb, iph, th);
2337
2338	skb->dev = NULL;
2339
2340	if (sk->sk_state == TCP_LISTEN) {
2341		ret = tcp_v4_do_rcv(sk, skb);
2342		goto put_and_return;
2343	}
2344
2345	sk_incoming_cpu_update(sk);
2346
2347	bh_lock_sock_nested(sk);
2348	tcp_segs_in(tcp_sk(sk), skb);
2349	ret = 0;
2350	if (!sock_owned_by_user(sk)) {
2351		ret = tcp_v4_do_rcv(sk, skb);
2352	} else {
2353		if (tcp_add_backlog(sk, skb, &drop_reason))
2354			goto discard_and_relse;
2355	}
2356	bh_unlock_sock(sk);
2357
2358put_and_return:
2359	if (refcounted)
2360		sock_put(sk);
2361
2362	return ret;
2363
2364no_tcp_socket:
2365	drop_reason = SKB_DROP_REASON_NO_SOCKET;
2366	if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
2367		goto discard_it;
2368
2369	tcp_v4_fill_cb(skb, iph, th);
2370
2371	if (tcp_checksum_complete(skb)) {
2372csum_error:
2373		drop_reason = SKB_DROP_REASON_TCP_CSUM;
2374		trace_tcp_bad_csum(skb);
2375		__TCP_INC_STATS(net, TCP_MIB_CSUMERRORS);
2376bad_packet:
2377		__TCP_INC_STATS(net, TCP_MIB_INERRS);
2378	} else {
2379		tcp_v4_send_reset(NULL, skb, sk_rst_convert_drop_reason(drop_reason));
2380	}
2381
2382discard_it:
2383	SKB_DR_OR(drop_reason, NOT_SPECIFIED);
2384	/* Discard frame. */
2385	sk_skb_reason_drop(sk, skb, drop_reason);
2386	return 0;
2387
2388discard_and_relse:
2389	sk_drops_add(sk, skb);
2390	if (refcounted)
2391		sock_put(sk);
2392	goto discard_it;
2393
2394do_time_wait:
2395	if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) {
2396		drop_reason = SKB_DROP_REASON_XFRM_POLICY;
2397		inet_twsk_put(inet_twsk(sk));
2398		goto discard_it;
2399	}
2400
2401	tcp_v4_fill_cb(skb, iph, th);
2402
2403	if (tcp_checksum_complete(skb)) {
2404		inet_twsk_put(inet_twsk(sk));
2405		goto csum_error;
2406	}
2407	switch (tcp_timewait_state_process(inet_twsk(sk), skb, th, &isn)) {
2408	case TCP_TW_SYN: {
2409		struct sock *sk2 = inet_lookup_listener(net,
2410							net->ipv4.tcp_death_row.hashinfo,
2411							skb, __tcp_hdrlen(th),
2412							iph->saddr, th->source,
2413							iph->daddr, th->dest,
2414							inet_iif(skb),
2415							sdif);
2416		if (sk2) {
2417			inet_twsk_deschedule_put(inet_twsk(sk));
2418			sk = sk2;
2419			tcp_v4_restore_cb(skb);
2420			refcounted = false;
2421			__this_cpu_write(tcp_tw_isn, isn);
2422			goto process;
2423		}
2424	}
2425		/* to ACK */
2426		fallthrough;
2427	case TCP_TW_ACK:
2428		tcp_v4_timewait_ack(sk, skb);
2429		break;
2430	case TCP_TW_RST:
2431		tcp_v4_send_reset(sk, skb, SK_RST_REASON_TCP_TIMEWAIT_SOCKET);
2432		inet_twsk_deschedule_put(inet_twsk(sk));
2433		goto discard_it;
2434	case TCP_TW_SUCCESS:;
2435	}
2436	goto discard_it;
2437}
2438
2439static struct timewait_sock_ops tcp_timewait_sock_ops = {
2440	.twsk_obj_size	= sizeof(struct tcp_timewait_sock),
 
2441	.twsk_destructor= tcp_twsk_destructor,
2442};
2443
2444void inet_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb)
2445{
2446	struct dst_entry *dst = skb_dst(skb);
2447
2448	if (dst && dst_hold_safe(dst)) {
2449		rcu_assign_pointer(sk->sk_rx_dst, dst);
2450		sk->sk_rx_dst_ifindex = skb->skb_iif;
2451	}
2452}
2453EXPORT_SYMBOL(inet_sk_rx_dst_set);
2454
2455const struct inet_connection_sock_af_ops ipv4_specific = {
2456	.queue_xmit	   = ip_queue_xmit,
2457	.send_check	   = tcp_v4_send_check,
2458	.rebuild_header	   = inet_sk_rebuild_header,
2459	.sk_rx_dst_set	   = inet_sk_rx_dst_set,
2460	.conn_request	   = tcp_v4_conn_request,
2461	.syn_recv_sock	   = tcp_v4_syn_recv_sock,
2462	.net_header_len	   = sizeof(struct iphdr),
2463	.setsockopt	   = ip_setsockopt,
2464	.getsockopt	   = ip_getsockopt,
2465	.addr2sockaddr	   = inet_csk_addr2sockaddr,
2466	.sockaddr_len	   = sizeof(struct sockaddr_in),
2467	.mtu_reduced	   = tcp_v4_mtu_reduced,
2468};
2469EXPORT_SYMBOL(ipv4_specific);
2470
2471#if defined(CONFIG_TCP_MD5SIG) || defined(CONFIG_TCP_AO)
2472static const struct tcp_sock_af_ops tcp_sock_ipv4_specific = {
2473#ifdef CONFIG_TCP_MD5SIG
2474	.md5_lookup		= tcp_v4_md5_lookup,
2475	.calc_md5_hash		= tcp_v4_md5_hash_skb,
2476	.md5_parse		= tcp_v4_parse_md5_keys,
2477#endif
2478#ifdef CONFIG_TCP_AO
2479	.ao_lookup		= tcp_v4_ao_lookup,
2480	.calc_ao_hash		= tcp_v4_ao_hash_skb,
2481	.ao_parse		= tcp_v4_parse_ao,
2482	.ao_calc_key_sk		= tcp_v4_ao_calc_key_sk,
2483#endif
2484};
2485#endif
2486
2487/* NOTE: A lot of things set to zero explicitly by call to
2488 *       sk_alloc() so need not be done here.
2489 */
2490static int tcp_v4_init_sock(struct sock *sk)
2491{
2492	struct inet_connection_sock *icsk = inet_csk(sk);
2493
2494	tcp_init_sock(sk);
2495
2496	icsk->icsk_af_ops = &ipv4_specific;
2497
2498#if defined(CONFIG_TCP_MD5SIG) || defined(CONFIG_TCP_AO)
2499	tcp_sk(sk)->af_specific = &tcp_sock_ipv4_specific;
2500#endif
2501
2502	return 0;
2503}
2504
2505#ifdef CONFIG_TCP_MD5SIG
2506static void tcp_md5sig_info_free_rcu(struct rcu_head *head)
2507{
2508	struct tcp_md5sig_info *md5sig;
2509
2510	md5sig = container_of(head, struct tcp_md5sig_info, rcu);
2511	kfree(md5sig);
2512	static_branch_slow_dec_deferred(&tcp_md5_needed);
2513	tcp_md5_release_sigpool();
2514}
2515#endif
2516
2517static void tcp_release_user_frags(struct sock *sk)
2518{
2519#ifdef CONFIG_PAGE_POOL
2520	unsigned long index;
2521	void *netmem;
2522
2523	xa_for_each(&sk->sk_user_frags, index, netmem)
2524		WARN_ON_ONCE(!napi_pp_put_page((__force netmem_ref)netmem));
2525#endif
2526}
2527
2528void tcp_v4_destroy_sock(struct sock *sk)
2529{
2530	struct tcp_sock *tp = tcp_sk(sk);
2531
2532	tcp_release_user_frags(sk);
2533
2534	xa_destroy(&sk->sk_user_frags);
2535
2536	trace_tcp_destroy_sock(sk);
2537
2538	tcp_clear_xmit_timers(sk);
2539
2540	tcp_cleanup_congestion_control(sk);
2541
2542	tcp_cleanup_ulp(sk);
2543
2544	/* Cleanup up the write buffer. */
2545	tcp_write_queue_purge(sk);
2546
2547	/* Check if we want to disable active TFO */
2548	tcp_fastopen_active_disable_ofo_check(sk);
2549
2550	/* Cleans up our, hopefully empty, out_of_order_queue. */
2551	skb_rbtree_purge(&tp->out_of_order_queue);
2552
2553#ifdef CONFIG_TCP_MD5SIG
2554	/* Clean up the MD5 key list, if any */
2555	if (tp->md5sig_info) {
2556		struct tcp_md5sig_info *md5sig;
2557
2558		md5sig = rcu_dereference_protected(tp->md5sig_info, 1);
2559		tcp_clear_md5_list(sk);
2560		call_rcu(&md5sig->rcu, tcp_md5sig_info_free_rcu);
2561		rcu_assign_pointer(tp->md5sig_info, NULL);
2562	}
2563#endif
2564	tcp_ao_destroy_sock(sk, false);
2565
2566	/* Clean up a referenced TCP bind bucket. */
2567	if (inet_csk(sk)->icsk_bind_hash)
2568		inet_put_port(sk);
2569
2570	BUG_ON(rcu_access_pointer(tp->fastopen_rsk));
2571
2572	/* If socket is aborted during connect operation */
2573	tcp_free_fastopen_req(tp);
2574	tcp_fastopen_destroy_cipher(sk);
2575	tcp_saved_syn_free(tp);
2576
2577	sk_sockets_allocated_dec(sk);
2578}
2579EXPORT_SYMBOL(tcp_v4_destroy_sock);
2580
2581#ifdef CONFIG_PROC_FS
2582/* Proc filesystem TCP sock list dumping. */
2583
2584static unsigned short seq_file_family(const struct seq_file *seq);
2585
2586static bool seq_sk_match(struct seq_file *seq, const struct sock *sk)
2587{
2588	unsigned short family = seq_file_family(seq);
2589
2590	/* AF_UNSPEC is used as a match all */
2591	return ((family == AF_UNSPEC || family == sk->sk_family) &&
2592		net_eq(sock_net(sk), seq_file_net(seq)));
2593}
2594
2595/* Find a non empty bucket (starting from st->bucket)
2596 * and return the first sk from it.
2597 */
2598static void *listening_get_first(struct seq_file *seq)
2599{
2600	struct inet_hashinfo *hinfo = seq_file_net(seq)->ipv4.tcp_death_row.hashinfo;
2601	struct tcp_iter_state *st = seq->private;
2602
2603	st->offset = 0;
2604	for (; st->bucket <= hinfo->lhash2_mask; st->bucket++) {
2605		struct inet_listen_hashbucket *ilb2;
2606		struct hlist_nulls_node *node;
2607		struct sock *sk;
2608
2609		ilb2 = &hinfo->lhash2[st->bucket];
2610		if (hlist_nulls_empty(&ilb2->nulls_head))
2611			continue;
2612
2613		spin_lock(&ilb2->lock);
2614		sk_nulls_for_each(sk, node, &ilb2->nulls_head) {
2615			if (seq_sk_match(seq, sk))
2616				return sk;
2617		}
2618		spin_unlock(&ilb2->lock);
2619	}
2620
2621	return NULL;
2622}
2623
2624/* Find the next sk of "cur" within the same bucket (i.e. st->bucket).
2625 * If "cur" is the last one in the st->bucket,
2626 * call listening_get_first() to return the first sk of the next
2627 * non empty bucket.
2628 */
2629static void *listening_get_next(struct seq_file *seq, void *cur)
2630{
2631	struct tcp_iter_state *st = seq->private;
2632	struct inet_listen_hashbucket *ilb2;
2633	struct hlist_nulls_node *node;
2634	struct inet_hashinfo *hinfo;
2635	struct sock *sk = cur;
2636
2637	++st->num;
2638	++st->offset;
2639
2640	sk = sk_nulls_next(sk);
2641	sk_nulls_for_each_from(sk, node) {
2642		if (seq_sk_match(seq, sk))
2643			return sk;
2644	}
2645
2646	hinfo = seq_file_net(seq)->ipv4.tcp_death_row.hashinfo;
2647	ilb2 = &hinfo->lhash2[st->bucket];
2648	spin_unlock(&ilb2->lock);
2649	++st->bucket;
2650	return listening_get_first(seq);
2651}
2652
2653static void *listening_get_idx(struct seq_file *seq, loff_t *pos)
2654{
2655	struct tcp_iter_state *st = seq->private;
2656	void *rc;
2657
2658	st->bucket = 0;
2659	st->offset = 0;
2660	rc = listening_get_first(seq);
2661
2662	while (rc && *pos) {
2663		rc = listening_get_next(seq, rc);
2664		--*pos;
2665	}
2666	return rc;
2667}
2668
2669static inline bool empty_bucket(struct inet_hashinfo *hinfo,
2670				const struct tcp_iter_state *st)
2671{
2672	return hlist_nulls_empty(&hinfo->ehash[st->bucket].chain);
2673}
2674
2675/*
2676 * Get first established socket starting from bucket given in st->bucket.
2677 * If st->bucket is zero, the very first socket in the hash is returned.
2678 */
2679static void *established_get_first(struct seq_file *seq)
2680{
2681	struct inet_hashinfo *hinfo = seq_file_net(seq)->ipv4.tcp_death_row.hashinfo;
2682	struct tcp_iter_state *st = seq->private;
2683
2684	st->offset = 0;
2685	for (; st->bucket <= hinfo->ehash_mask; ++st->bucket) {
2686		struct sock *sk;
2687		struct hlist_nulls_node *node;
2688		spinlock_t *lock = inet_ehash_lockp(hinfo, st->bucket);
2689
2690		cond_resched();
2691
2692		/* Lockless fast path for the common case of empty buckets */
2693		if (empty_bucket(hinfo, st))
2694			continue;
2695
2696		spin_lock_bh(lock);
2697		sk_nulls_for_each(sk, node, &hinfo->ehash[st->bucket].chain) {
2698			if (seq_sk_match(seq, sk))
2699				return sk;
2700		}
2701		spin_unlock_bh(lock);
2702	}
2703
2704	return NULL;
2705}
2706
2707static void *established_get_next(struct seq_file *seq, void *cur)
2708{
2709	struct inet_hashinfo *hinfo = seq_file_net(seq)->ipv4.tcp_death_row.hashinfo;
2710	struct tcp_iter_state *st = seq->private;
2711	struct hlist_nulls_node *node;
2712	struct sock *sk = cur;
2713
2714	++st->num;
2715	++st->offset;
2716
2717	sk = sk_nulls_next(sk);
2718
2719	sk_nulls_for_each_from(sk, node) {
2720		if (seq_sk_match(seq, sk))
2721			return sk;
2722	}
2723
2724	spin_unlock_bh(inet_ehash_lockp(hinfo, st->bucket));
2725	++st->bucket;
2726	return established_get_first(seq);
2727}
2728
2729static void *established_get_idx(struct seq_file *seq, loff_t pos)
2730{
2731	struct tcp_iter_state *st = seq->private;
2732	void *rc;
2733
2734	st->bucket = 0;
2735	rc = established_get_first(seq);
2736
2737	while (rc && pos) {
2738		rc = established_get_next(seq, rc);
2739		--pos;
2740	}
2741	return rc;
2742}
2743
2744static void *tcp_get_idx(struct seq_file *seq, loff_t pos)
2745{
2746	void *rc;
2747	struct tcp_iter_state *st = seq->private;
2748
2749	st->state = TCP_SEQ_STATE_LISTENING;
2750	rc	  = listening_get_idx(seq, &pos);
2751
2752	if (!rc) {
2753		st->state = TCP_SEQ_STATE_ESTABLISHED;
2754		rc	  = established_get_idx(seq, pos);
2755	}
2756
2757	return rc;
2758}
2759
2760static void *tcp_seek_last_pos(struct seq_file *seq)
2761{
2762	struct inet_hashinfo *hinfo = seq_file_net(seq)->ipv4.tcp_death_row.hashinfo;
2763	struct tcp_iter_state *st = seq->private;
2764	int bucket = st->bucket;
2765	int offset = st->offset;
2766	int orig_num = st->num;
2767	void *rc = NULL;
2768
2769	switch (st->state) {
2770	case TCP_SEQ_STATE_LISTENING:
2771		if (st->bucket > hinfo->lhash2_mask)
2772			break;
2773		rc = listening_get_first(seq);
2774		while (offset-- && rc && bucket == st->bucket)
2775			rc = listening_get_next(seq, rc);
2776		if (rc)
2777			break;
2778		st->bucket = 0;
2779		st->state = TCP_SEQ_STATE_ESTABLISHED;
2780		fallthrough;
2781	case TCP_SEQ_STATE_ESTABLISHED:
2782		if (st->bucket > hinfo->ehash_mask)
2783			break;
2784		rc = established_get_first(seq);
2785		while (offset-- && rc && bucket == st->bucket)
2786			rc = established_get_next(seq, rc);
2787	}
2788
2789	st->num = orig_num;
2790
2791	return rc;
2792}
2793
2794void *tcp_seq_start(struct seq_file *seq, loff_t *pos)
2795{
2796	struct tcp_iter_state *st = seq->private;
2797	void *rc;
2798
2799	if (*pos && *pos == st->last_pos) {
2800		rc = tcp_seek_last_pos(seq);
2801		if (rc)
2802			goto out;
2803	}
2804
2805	st->state = TCP_SEQ_STATE_LISTENING;
2806	st->num = 0;
2807	st->bucket = 0;
2808	st->offset = 0;
2809	rc = *pos ? tcp_get_idx(seq, *pos - 1) : SEQ_START_TOKEN;
2810
2811out:
2812	st->last_pos = *pos;
2813	return rc;
2814}
2815EXPORT_SYMBOL(tcp_seq_start);
2816
2817void *tcp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2818{
2819	struct tcp_iter_state *st = seq->private;
2820	void *rc = NULL;
2821
2822	if (v == SEQ_START_TOKEN) {
2823		rc = tcp_get_idx(seq, 0);
2824		goto out;
2825	}
2826
2827	switch (st->state) {
2828	case TCP_SEQ_STATE_LISTENING:
2829		rc = listening_get_next(seq, v);
2830		if (!rc) {
2831			st->state = TCP_SEQ_STATE_ESTABLISHED;
2832			st->bucket = 0;
2833			st->offset = 0;
2834			rc	  = established_get_first(seq);
2835		}
2836		break;
2837	case TCP_SEQ_STATE_ESTABLISHED:
2838		rc = established_get_next(seq, v);
2839		break;
2840	}
2841out:
2842	++*pos;
2843	st->last_pos = *pos;
2844	return rc;
2845}
2846EXPORT_SYMBOL(tcp_seq_next);
2847
2848void tcp_seq_stop(struct seq_file *seq, void *v)
2849{
2850	struct inet_hashinfo *hinfo = seq_file_net(seq)->ipv4.tcp_death_row.hashinfo;
2851	struct tcp_iter_state *st = seq->private;
2852
2853	switch (st->state) {
2854	case TCP_SEQ_STATE_LISTENING:
2855		if (v != SEQ_START_TOKEN)
2856			spin_unlock(&hinfo->lhash2[st->bucket].lock);
2857		break;
2858	case TCP_SEQ_STATE_ESTABLISHED:
2859		if (v)
2860			spin_unlock_bh(inet_ehash_lockp(hinfo, st->bucket));
2861		break;
2862	}
2863}
2864EXPORT_SYMBOL(tcp_seq_stop);
2865
2866static void get_openreq4(const struct request_sock *req,
2867			 struct seq_file *f, int i)
2868{
2869	const struct inet_request_sock *ireq = inet_rsk(req);
2870	long delta = req->rsk_timer.expires - jiffies;
2871
2872	seq_printf(f, "%4d: %08X:%04X %08X:%04X"
2873		" %02X %08X:%08X %02X:%08lX %08X %5u %8d %u %d %pK",
2874		i,
2875		ireq->ir_loc_addr,
2876		ireq->ir_num,
2877		ireq->ir_rmt_addr,
2878		ntohs(ireq->ir_rmt_port),
2879		TCP_SYN_RECV,
2880		0, 0, /* could print option size, but that is af dependent. */
2881		1,    /* timers active (only the expire timer) */
2882		jiffies_delta_to_clock_t(delta),
2883		req->num_timeout,
2884		from_kuid_munged(seq_user_ns(f),
2885				 sock_i_uid(req->rsk_listener)),
2886		0,  /* non standard timer */
2887		0, /* open_requests have no inode */
2888		0,
2889		req);
2890}
2891
2892static void get_tcp4_sock(struct sock *sk, struct seq_file *f, int i)
2893{
2894	int timer_active;
2895	unsigned long timer_expires;
2896	const struct tcp_sock *tp = tcp_sk(sk);
2897	const struct inet_connection_sock *icsk = inet_csk(sk);
2898	const struct inet_sock *inet = inet_sk(sk);
2899	const struct fastopen_queue *fastopenq = &icsk->icsk_accept_queue.fastopenq;
2900	__be32 dest = inet->inet_daddr;
2901	__be32 src = inet->inet_rcv_saddr;
2902	__u16 destp = ntohs(inet->inet_dport);
2903	__u16 srcp = ntohs(inet->inet_sport);
2904	u8 icsk_pending;
2905	int rx_queue;
2906	int state;
2907
2908	icsk_pending = smp_load_acquire(&icsk->icsk_pending);
2909	if (icsk_pending == ICSK_TIME_RETRANS ||
2910	    icsk_pending == ICSK_TIME_REO_TIMEOUT ||
2911	    icsk_pending == ICSK_TIME_LOSS_PROBE) {
2912		timer_active	= 1;
2913		timer_expires	= icsk->icsk_timeout;
2914	} else if (icsk_pending == ICSK_TIME_PROBE0) {
2915		timer_active	= 4;
2916		timer_expires	= icsk->icsk_timeout;
2917	} else if (timer_pending(&sk->sk_timer)) {
2918		timer_active	= 2;
2919		timer_expires	= sk->sk_timer.expires;
2920	} else {
2921		timer_active	= 0;
2922		timer_expires = jiffies;
2923	}
2924
2925	state = inet_sk_state_load(sk);
2926	if (state == TCP_LISTEN)
2927		rx_queue = READ_ONCE(sk->sk_ack_backlog);
2928	else
2929		/* Because we don't lock the socket,
2930		 * we might find a transient negative value.
2931		 */
2932		rx_queue = max_t(int, READ_ONCE(tp->rcv_nxt) -
2933				      READ_ONCE(tp->copied_seq), 0);
2934
2935	seq_printf(f, "%4d: %08X:%04X %08X:%04X %02X %08X:%08X %02X:%08lX "
2936			"%08X %5u %8d %lu %d %pK %lu %lu %u %u %d",
2937		i, src, srcp, dest, destp, state,
2938		READ_ONCE(tp->write_seq) - tp->snd_una,
2939		rx_queue,
2940		timer_active,
2941		jiffies_delta_to_clock_t(timer_expires - jiffies),
2942		icsk->icsk_retransmits,
2943		from_kuid_munged(seq_user_ns(f), sock_i_uid(sk)),
2944		icsk->icsk_probes_out,
2945		sock_i_ino(sk),
2946		refcount_read(&sk->sk_refcnt), sk,
2947		jiffies_to_clock_t(icsk->icsk_rto),
2948		jiffies_to_clock_t(icsk->icsk_ack.ato),
2949		(icsk->icsk_ack.quick << 1) | inet_csk_in_pingpong_mode(sk),
2950		tcp_snd_cwnd(tp),
2951		state == TCP_LISTEN ?
2952		    fastopenq->max_qlen :
2953		    (tcp_in_initial_slowstart(tp) ? -1 : tp->snd_ssthresh));
2954}
2955
2956static void get_timewait4_sock(const struct inet_timewait_sock *tw,
2957			       struct seq_file *f, int i)
2958{
2959	long delta = tw->tw_timer.expires - jiffies;
2960	__be32 dest, src;
2961	__u16 destp, srcp;
2962
2963	dest  = tw->tw_daddr;
2964	src   = tw->tw_rcv_saddr;
2965	destp = ntohs(tw->tw_dport);
2966	srcp  = ntohs(tw->tw_sport);
2967
2968	seq_printf(f, "%4d: %08X:%04X %08X:%04X"
2969		" %02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %pK",
2970		i, src, srcp, dest, destp, READ_ONCE(tw->tw_substate), 0, 0,
2971		3, jiffies_delta_to_clock_t(delta), 0, 0, 0, 0,
2972		refcount_read(&tw->tw_refcnt), tw);
2973}
2974
2975#define TMPSZ 150
2976
2977static int tcp4_seq_show(struct seq_file *seq, void *v)
2978{
2979	struct tcp_iter_state *st;
2980	struct sock *sk = v;
2981
2982	seq_setwidth(seq, TMPSZ - 1);
2983	if (v == SEQ_START_TOKEN) {
2984		seq_puts(seq, "  sl  local_address rem_address   st tx_queue "
2985			   "rx_queue tr tm->when retrnsmt   uid  timeout "
2986			   "inode");
2987		goto out;
2988	}
2989	st = seq->private;
2990
2991	if (sk->sk_state == TCP_TIME_WAIT)
2992		get_timewait4_sock(v, seq, st->num);
2993	else if (sk->sk_state == TCP_NEW_SYN_RECV)
2994		get_openreq4(v, seq, st->num);
2995	else
2996		get_tcp4_sock(v, seq, st->num);
2997out:
2998	seq_pad(seq, '\n');
2999	return 0;
3000}
3001
3002#ifdef CONFIG_BPF_SYSCALL
3003struct bpf_tcp_iter_state {
3004	struct tcp_iter_state state;
3005	unsigned int cur_sk;
3006	unsigned int end_sk;
3007	unsigned int max_sk;
3008	struct sock **batch;
3009	bool st_bucket_done;
3010};
3011
3012struct bpf_iter__tcp {
3013	__bpf_md_ptr(struct bpf_iter_meta *, meta);
3014	__bpf_md_ptr(struct sock_common *, sk_common);
3015	uid_t uid __aligned(8);
3016};
3017
3018static int tcp_prog_seq_show(struct bpf_prog *prog, struct bpf_iter_meta *meta,
3019			     struct sock_common *sk_common, uid_t uid)
3020{
3021	struct bpf_iter__tcp ctx;
3022
3023	meta->seq_num--;  /* skip SEQ_START_TOKEN */
3024	ctx.meta = meta;
3025	ctx.sk_common = sk_common;
3026	ctx.uid = uid;
3027	return bpf_iter_run_prog(prog, &ctx);
3028}
3029
3030static void bpf_iter_tcp_put_batch(struct bpf_tcp_iter_state *iter)
3031{
3032	while (iter->cur_sk < iter->end_sk)
3033		sock_gen_put(iter->batch[iter->cur_sk++]);
3034}
3035
3036static int bpf_iter_tcp_realloc_batch(struct bpf_tcp_iter_state *iter,
3037				      unsigned int new_batch_sz)
3038{
3039	struct sock **new_batch;
3040
3041	new_batch = kvmalloc(sizeof(*new_batch) * new_batch_sz,
3042			     GFP_USER | __GFP_NOWARN);
3043	if (!new_batch)
3044		return -ENOMEM;
3045
3046	bpf_iter_tcp_put_batch(iter);
3047	kvfree(iter->batch);
3048	iter->batch = new_batch;
3049	iter->max_sk = new_batch_sz;
3050
3051	return 0;
3052}
3053
3054static unsigned int bpf_iter_tcp_listening_batch(struct seq_file *seq,
3055						 struct sock *start_sk)
3056{
3057	struct inet_hashinfo *hinfo = seq_file_net(seq)->ipv4.tcp_death_row.hashinfo;
3058	struct bpf_tcp_iter_state *iter = seq->private;
3059	struct tcp_iter_state *st = &iter->state;
3060	struct hlist_nulls_node *node;
3061	unsigned int expected = 1;
3062	struct sock *sk;
3063
3064	sock_hold(start_sk);
3065	iter->batch[iter->end_sk++] = start_sk;
3066
3067	sk = sk_nulls_next(start_sk);
3068	sk_nulls_for_each_from(sk, node) {
3069		if (seq_sk_match(seq, sk)) {
3070			if (iter->end_sk < iter->max_sk) {
3071				sock_hold(sk);
3072				iter->batch[iter->end_sk++] = sk;
3073			}
3074			expected++;
3075		}
3076	}
3077	spin_unlock(&hinfo->lhash2[st->bucket].lock);
3078
3079	return expected;
3080}
3081
3082static unsigned int bpf_iter_tcp_established_batch(struct seq_file *seq,
3083						   struct sock *start_sk)
3084{
3085	struct inet_hashinfo *hinfo = seq_file_net(seq)->ipv4.tcp_death_row.hashinfo;
3086	struct bpf_tcp_iter_state *iter = seq->private;
3087	struct tcp_iter_state *st = &iter->state;
3088	struct hlist_nulls_node *node;
3089	unsigned int expected = 1;
3090	struct sock *sk;
3091
3092	sock_hold(start_sk);
3093	iter->batch[iter->end_sk++] = start_sk;
3094
3095	sk = sk_nulls_next(start_sk);
3096	sk_nulls_for_each_from(sk, node) {
3097		if (seq_sk_match(seq, sk)) {
3098			if (iter->end_sk < iter->max_sk) {
3099				sock_hold(sk);
3100				iter->batch[iter->end_sk++] = sk;
3101			}
3102			expected++;
3103		}
3104	}
3105	spin_unlock_bh(inet_ehash_lockp(hinfo, st->bucket));
3106
3107	return expected;
3108}
3109
3110static struct sock *bpf_iter_tcp_batch(struct seq_file *seq)
3111{
3112	struct inet_hashinfo *hinfo = seq_file_net(seq)->ipv4.tcp_death_row.hashinfo;
3113	struct bpf_tcp_iter_state *iter = seq->private;
3114	struct tcp_iter_state *st = &iter->state;
3115	unsigned int expected;
3116	bool resized = false;
3117	struct sock *sk;
3118
3119	/* The st->bucket is done.  Directly advance to the next
3120	 * bucket instead of having the tcp_seek_last_pos() to skip
3121	 * one by one in the current bucket and eventually find out
3122	 * it has to advance to the next bucket.
3123	 */
3124	if (iter->st_bucket_done) {
3125		st->offset = 0;
3126		st->bucket++;
3127		if (st->state == TCP_SEQ_STATE_LISTENING &&
3128		    st->bucket > hinfo->lhash2_mask) {
3129			st->state = TCP_SEQ_STATE_ESTABLISHED;
3130			st->bucket = 0;
3131		}
3132	}
3133
3134again:
3135	/* Get a new batch */
3136	iter->cur_sk = 0;
3137	iter->end_sk = 0;
3138	iter->st_bucket_done = false;
3139
3140	sk = tcp_seek_last_pos(seq);
3141	if (!sk)
3142		return NULL; /* Done */
3143
3144	if (st->state == TCP_SEQ_STATE_LISTENING)
3145		expected = bpf_iter_tcp_listening_batch(seq, sk);
3146	else
3147		expected = bpf_iter_tcp_established_batch(seq, sk);
3148
3149	if (iter->end_sk == expected) {
3150		iter->st_bucket_done = true;
3151		return sk;
3152	}
3153
3154	if (!resized && !bpf_iter_tcp_realloc_batch(iter, expected * 3 / 2)) {
3155		resized = true;
3156		goto again;
3157	}
3158
3159	return sk;
3160}
3161
3162static void *bpf_iter_tcp_seq_start(struct seq_file *seq, loff_t *pos)
3163{
3164	/* bpf iter does not support lseek, so it always
3165	 * continue from where it was stop()-ped.
3166	 */
3167	if (*pos)
3168		return bpf_iter_tcp_batch(seq);
3169
3170	return SEQ_START_TOKEN;
3171}
3172
3173static void *bpf_iter_tcp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3174{
3175	struct bpf_tcp_iter_state *iter = seq->private;
3176	struct tcp_iter_state *st = &iter->state;
3177	struct sock *sk;
3178
3179	/* Whenever seq_next() is called, the iter->cur_sk is
3180	 * done with seq_show(), so advance to the next sk in
3181	 * the batch.
3182	 */
3183	if (iter->cur_sk < iter->end_sk) {
3184		/* Keeping st->num consistent in tcp_iter_state.
3185		 * bpf_iter_tcp does not use st->num.
3186		 * meta.seq_num is used instead.
3187		 */
3188		st->num++;
3189		/* Move st->offset to the next sk in the bucket such that
3190		 * the future start() will resume at st->offset in
3191		 * st->bucket.  See tcp_seek_last_pos().
3192		 */
3193		st->offset++;
3194		sock_gen_put(iter->batch[iter->cur_sk++]);
3195	}
3196
3197	if (iter->cur_sk < iter->end_sk)
3198		sk = iter->batch[iter->cur_sk];
3199	else
3200		sk = bpf_iter_tcp_batch(seq);
3201
3202	++*pos;
3203	/* Keeping st->last_pos consistent in tcp_iter_state.
3204	 * bpf iter does not do lseek, so st->last_pos always equals to *pos.
3205	 */
3206	st->last_pos = *pos;
3207	return sk;
3208}
3209
3210static int bpf_iter_tcp_seq_show(struct seq_file *seq, void *v)
3211{
3212	struct bpf_iter_meta meta;
3213	struct bpf_prog *prog;
3214	struct sock *sk = v;
3215	uid_t uid;
3216	int ret;
3217
3218	if (v == SEQ_START_TOKEN)
3219		return 0;
3220
3221	if (sk_fullsock(sk))
3222		lock_sock(sk);
3223
3224	if (unlikely(sk_unhashed(sk))) {
3225		ret = SEQ_SKIP;
3226		goto unlock;
3227	}
3228
3229	if (sk->sk_state == TCP_TIME_WAIT) {
3230		uid = 0;
3231	} else if (sk->sk_state == TCP_NEW_SYN_RECV) {
3232		const struct request_sock *req = v;
3233
3234		uid = from_kuid_munged(seq_user_ns(seq),
3235				       sock_i_uid(req->rsk_listener));
3236	} else {
3237		uid = from_kuid_munged(seq_user_ns(seq), sock_i_uid(sk));
3238	}
3239
3240	meta.seq = seq;
3241	prog = bpf_iter_get_info(&meta, false);
3242	ret = tcp_prog_seq_show(prog, &meta, v, uid);
3243
3244unlock:
3245	if (sk_fullsock(sk))
3246		release_sock(sk);
3247	return ret;
3248
3249}
3250
3251static void bpf_iter_tcp_seq_stop(struct seq_file *seq, void *v)
3252{
3253	struct bpf_tcp_iter_state *iter = seq->private;
3254	struct bpf_iter_meta meta;
3255	struct bpf_prog *prog;
3256
3257	if (!v) {
3258		meta.seq = seq;
3259		prog = bpf_iter_get_info(&meta, true);
3260		if (prog)
3261			(void)tcp_prog_seq_show(prog, &meta, v, 0);
3262	}
3263
3264	if (iter->cur_sk < iter->end_sk) {
3265		bpf_iter_tcp_put_batch(iter);
3266		iter->st_bucket_done = false;
3267	}
3268}
3269
3270static const struct seq_operations bpf_iter_tcp_seq_ops = {
3271	.show		= bpf_iter_tcp_seq_show,
3272	.start		= bpf_iter_tcp_seq_start,
3273	.next		= bpf_iter_tcp_seq_next,
3274	.stop		= bpf_iter_tcp_seq_stop,
3275};
3276#endif
3277static unsigned short seq_file_family(const struct seq_file *seq)
3278{
3279	const struct tcp_seq_afinfo *afinfo;
3280
3281#ifdef CONFIG_BPF_SYSCALL
3282	/* Iterated from bpf_iter.  Let the bpf prog to filter instead. */
3283	if (seq->op == &bpf_iter_tcp_seq_ops)
3284		return AF_UNSPEC;
3285#endif
3286
3287	/* Iterated from proc fs */
3288	afinfo = pde_data(file_inode(seq->file));
3289	return afinfo->family;
3290}
3291
3292static const struct seq_operations tcp4_seq_ops = {
3293	.show		= tcp4_seq_show,
3294	.start		= tcp_seq_start,
3295	.next		= tcp_seq_next,
3296	.stop		= tcp_seq_stop,
3297};
3298
3299static struct tcp_seq_afinfo tcp4_seq_afinfo = {
3300	.family		= AF_INET,
3301};
3302
3303static int __net_init tcp4_proc_init_net(struct net *net)
3304{
3305	if (!proc_create_net_data("tcp", 0444, net->proc_net, &tcp4_seq_ops,
3306			sizeof(struct tcp_iter_state), &tcp4_seq_afinfo))
3307		return -ENOMEM;
3308	return 0;
3309}
3310
3311static void __net_exit tcp4_proc_exit_net(struct net *net)
3312{
3313	remove_proc_entry("tcp", net->proc_net);
3314}
3315
3316static struct pernet_operations tcp4_net_ops = {
3317	.init = tcp4_proc_init_net,
3318	.exit = tcp4_proc_exit_net,
3319};
3320
3321int __init tcp4_proc_init(void)
3322{
3323	return register_pernet_subsys(&tcp4_net_ops);
3324}
3325
3326void tcp4_proc_exit(void)
3327{
3328	unregister_pernet_subsys(&tcp4_net_ops);
3329}
3330#endif /* CONFIG_PROC_FS */
3331
3332/* @wake is one when sk_stream_write_space() calls us.
3333 * This sends EPOLLOUT only if notsent_bytes is half the limit.
3334 * This mimics the strategy used in sock_def_write_space().
3335 */
3336bool tcp_stream_memory_free(const struct sock *sk, int wake)
3337{
3338	const struct tcp_sock *tp = tcp_sk(sk);
3339	u32 notsent_bytes = READ_ONCE(tp->write_seq) -
3340			    READ_ONCE(tp->snd_nxt);
3341
3342	return (notsent_bytes << wake) < tcp_notsent_lowat(tp);
3343}
3344EXPORT_SYMBOL(tcp_stream_memory_free);
3345
3346struct proto tcp_prot = {
3347	.name			= "TCP",
3348	.owner			= THIS_MODULE,
3349	.close			= tcp_close,
3350	.pre_connect		= tcp_v4_pre_connect,
3351	.connect		= tcp_v4_connect,
3352	.disconnect		= tcp_disconnect,
3353	.accept			= inet_csk_accept,
3354	.ioctl			= tcp_ioctl,
3355	.init			= tcp_v4_init_sock,
3356	.destroy		= tcp_v4_destroy_sock,
3357	.shutdown		= tcp_shutdown,
3358	.setsockopt		= tcp_setsockopt,
3359	.getsockopt		= tcp_getsockopt,
3360	.bpf_bypass_getsockopt	= tcp_bpf_bypass_getsockopt,
3361	.keepalive		= tcp_set_keepalive,
3362	.recvmsg		= tcp_recvmsg,
3363	.sendmsg		= tcp_sendmsg,
3364	.splice_eof		= tcp_splice_eof,
3365	.backlog_rcv		= tcp_v4_do_rcv,
3366	.release_cb		= tcp_release_cb,
3367	.hash			= inet_hash,
3368	.unhash			= inet_unhash,
3369	.get_port		= inet_csk_get_port,
3370	.put_port		= inet_put_port,
3371#ifdef CONFIG_BPF_SYSCALL
3372	.psock_update_sk_prot	= tcp_bpf_update_proto,
3373#endif
3374	.enter_memory_pressure	= tcp_enter_memory_pressure,
3375	.leave_memory_pressure	= tcp_leave_memory_pressure,
3376	.stream_memory_free	= tcp_stream_memory_free,
3377	.sockets_allocated	= &tcp_sockets_allocated,
3378	.orphan_count		= &tcp_orphan_count,
3379
3380	.memory_allocated	= &tcp_memory_allocated,
3381	.per_cpu_fw_alloc	= &tcp_memory_per_cpu_fw_alloc,
3382
3383	.memory_pressure	= &tcp_memory_pressure,
3384	.sysctl_mem		= sysctl_tcp_mem,
3385	.sysctl_wmem_offset	= offsetof(struct net, ipv4.sysctl_tcp_wmem),
3386	.sysctl_rmem_offset	= offsetof(struct net, ipv4.sysctl_tcp_rmem),
3387	.max_header		= MAX_TCP_HEADER,
3388	.obj_size		= sizeof(struct tcp_sock),
3389	.slab_flags		= SLAB_TYPESAFE_BY_RCU,
3390	.twsk_prot		= &tcp_timewait_sock_ops,
3391	.rsk_prot		= &tcp_request_sock_ops,
3392	.h.hashinfo		= NULL,
3393	.no_autobind		= true,
3394	.diag_destroy		= tcp_abort,
3395};
3396EXPORT_SYMBOL(tcp_prot);
3397
3398static void __net_exit tcp_sk_exit(struct net *net)
3399{
3400	if (net->ipv4.tcp_congestion_control)
3401		bpf_module_put(net->ipv4.tcp_congestion_control,
3402			       net->ipv4.tcp_congestion_control->owner);
3403}
3404
3405static void __net_init tcp_set_hashinfo(struct net *net)
3406{
3407	struct inet_hashinfo *hinfo;
3408	unsigned int ehash_entries;
3409	struct net *old_net;
3410
3411	if (net_eq(net, &init_net))
3412		goto fallback;
3413
3414	old_net = current->nsproxy->net_ns;
3415	ehash_entries = READ_ONCE(old_net->ipv4.sysctl_tcp_child_ehash_entries);
3416	if (!ehash_entries)
3417		goto fallback;
3418
3419	ehash_entries = roundup_pow_of_two(ehash_entries);
3420	hinfo = inet_pernet_hashinfo_alloc(&tcp_hashinfo, ehash_entries);
3421	if (!hinfo) {
3422		pr_warn("Failed to allocate TCP ehash (entries: %u) "
3423			"for a netns, fallback to the global one\n",
3424			ehash_entries);
3425fallback:
3426		hinfo = &tcp_hashinfo;
3427		ehash_entries = tcp_hashinfo.ehash_mask + 1;
3428	}
3429
3430	net->ipv4.tcp_death_row.hashinfo = hinfo;
3431	net->ipv4.tcp_death_row.sysctl_max_tw_buckets = ehash_entries / 2;
3432	net->ipv4.sysctl_max_syn_backlog = max(128U, ehash_entries / 128);
3433}
3434
3435static int __net_init tcp_sk_init(struct net *net)
3436{
3437	net->ipv4.sysctl_tcp_ecn = 2;
3438	net->ipv4.sysctl_tcp_ecn_fallback = 1;
3439
3440	net->ipv4.sysctl_tcp_base_mss = TCP_BASE_MSS;
3441	net->ipv4.sysctl_tcp_min_snd_mss = TCP_MIN_SND_MSS;
3442	net->ipv4.sysctl_tcp_probe_threshold = TCP_PROBE_THRESHOLD;
3443	net->ipv4.sysctl_tcp_probe_interval = TCP_PROBE_INTERVAL;
3444	net->ipv4.sysctl_tcp_mtu_probe_floor = TCP_MIN_SND_MSS;
3445
3446	net->ipv4.sysctl_tcp_keepalive_time = TCP_KEEPALIVE_TIME;
3447	net->ipv4.sysctl_tcp_keepalive_probes = TCP_KEEPALIVE_PROBES;
3448	net->ipv4.sysctl_tcp_keepalive_intvl = TCP_KEEPALIVE_INTVL;
3449
3450	net->ipv4.sysctl_tcp_syn_retries = TCP_SYN_RETRIES;
3451	net->ipv4.sysctl_tcp_synack_retries = TCP_SYNACK_RETRIES;
3452	net->ipv4.sysctl_tcp_syncookies = 1;
3453	net->ipv4.sysctl_tcp_reordering = TCP_FASTRETRANS_THRESH;
3454	net->ipv4.sysctl_tcp_retries1 = TCP_RETR1;
3455	net->ipv4.sysctl_tcp_retries2 = TCP_RETR2;
3456	net->ipv4.sysctl_tcp_orphan_retries = 0;
3457	net->ipv4.sysctl_tcp_fin_timeout = TCP_FIN_TIMEOUT;
3458	net->ipv4.sysctl_tcp_notsent_lowat = UINT_MAX;
3459	net->ipv4.sysctl_tcp_tw_reuse = 2;
3460	net->ipv4.sysctl_tcp_no_ssthresh_metrics_save = 1;
3461
3462	refcount_set(&net->ipv4.tcp_death_row.tw_refcount, 1);
3463	tcp_set_hashinfo(net);
3464
3465	net->ipv4.sysctl_tcp_sack = 1;
3466	net->ipv4.sysctl_tcp_window_scaling = 1;
3467	net->ipv4.sysctl_tcp_timestamps = 1;
3468	net->ipv4.sysctl_tcp_early_retrans = 3;
3469	net->ipv4.sysctl_tcp_recovery = TCP_RACK_LOSS_DETECTION;
3470	net->ipv4.sysctl_tcp_slow_start_after_idle = 1; /* By default, RFC2861 behavior.  */
3471	net->ipv4.sysctl_tcp_retrans_collapse = 1;
3472	net->ipv4.sysctl_tcp_max_reordering = 300;
3473	net->ipv4.sysctl_tcp_dsack = 1;
3474	net->ipv4.sysctl_tcp_app_win = 31;
3475	net->ipv4.sysctl_tcp_adv_win_scale = 1;
3476	net->ipv4.sysctl_tcp_frto = 2;
3477	net->ipv4.sysctl_tcp_moderate_rcvbuf = 1;
3478	/* This limits the percentage of the congestion window which we
3479	 * will allow a single TSO frame to consume.  Building TSO frames
3480	 * which are too large can cause TCP streams to be bursty.
3481	 */
3482	net->ipv4.sysctl_tcp_tso_win_divisor = 3;
3483	/* Default TSQ limit of 16 TSO segments */
3484	net->ipv4.sysctl_tcp_limit_output_bytes = 16 * 65536;
3485
3486	/* rfc5961 challenge ack rate limiting, per net-ns, disabled by default. */
3487	net->ipv4.sysctl_tcp_challenge_ack_limit = INT_MAX;
3488
3489	net->ipv4.sysctl_tcp_min_tso_segs = 2;
3490	net->ipv4.sysctl_tcp_tso_rtt_log = 9;  /* 2^9 = 512 usec */
3491	net->ipv4.sysctl_tcp_min_rtt_wlen = 300;
3492	net->ipv4.sysctl_tcp_autocorking = 1;
3493	net->ipv4.sysctl_tcp_invalid_ratelimit = HZ/2;
3494	net->ipv4.sysctl_tcp_pacing_ss_ratio = 200;
3495	net->ipv4.sysctl_tcp_pacing_ca_ratio = 120;
3496	if (net != &init_net) {
3497		memcpy(net->ipv4.sysctl_tcp_rmem,
3498		       init_net.ipv4.sysctl_tcp_rmem,
3499		       sizeof(init_net.ipv4.sysctl_tcp_rmem));
3500		memcpy(net->ipv4.sysctl_tcp_wmem,
3501		       init_net.ipv4.sysctl_tcp_wmem,
3502		       sizeof(init_net.ipv4.sysctl_tcp_wmem));
3503	}
3504	net->ipv4.sysctl_tcp_comp_sack_delay_ns = NSEC_PER_MSEC;
3505	net->ipv4.sysctl_tcp_comp_sack_slack_ns = 100 * NSEC_PER_USEC;
3506	net->ipv4.sysctl_tcp_comp_sack_nr = 44;
3507	net->ipv4.sysctl_tcp_backlog_ack_defer = 1;
3508	net->ipv4.sysctl_tcp_fastopen = TFO_CLIENT_ENABLE;
3509	net->ipv4.sysctl_tcp_fastopen_blackhole_timeout = 0;
3510	atomic_set(&net->ipv4.tfo_active_disable_times, 0);
3511
3512	/* Set default values for PLB */
3513	net->ipv4.sysctl_tcp_plb_enabled = 0; /* Disabled by default */
3514	net->ipv4.sysctl_tcp_plb_idle_rehash_rounds = 3;
3515	net->ipv4.sysctl_tcp_plb_rehash_rounds = 12;
3516	net->ipv4.sysctl_tcp_plb_suspend_rto_sec = 60;
3517	/* Default congestion threshold for PLB to mark a round is 50% */
3518	net->ipv4.sysctl_tcp_plb_cong_thresh = (1 << TCP_PLB_SCALE) / 2;
3519
3520	/* Reno is always built in */
3521	if (!net_eq(net, &init_net) &&
3522	    bpf_try_module_get(init_net.ipv4.tcp_congestion_control,
3523			       init_net.ipv4.tcp_congestion_control->owner))
3524		net->ipv4.tcp_congestion_control = init_net.ipv4.tcp_congestion_control;
3525	else
3526		net->ipv4.tcp_congestion_control = &tcp_reno;
3527
3528	net->ipv4.sysctl_tcp_syn_linear_timeouts = 4;
3529	net->ipv4.sysctl_tcp_shrink_window = 0;
3530
3531	net->ipv4.sysctl_tcp_pingpong_thresh = 1;
3532	net->ipv4.sysctl_tcp_rto_min_us = jiffies_to_usecs(TCP_RTO_MIN);
3533
3534	return 0;
3535}
3536
3537static void __net_exit tcp_sk_exit_batch(struct list_head *net_exit_list)
3538{
3539	struct net *net;
3540
3541	/* make sure concurrent calls to tcp_sk_exit_batch from net_cleanup_work
3542	 * and failed setup_net error unwinding path are serialized.
3543	 *
3544	 * tcp_twsk_purge() handles twsk in any dead netns, not just those in
3545	 * net_exit_list, the thread that dismantles a particular twsk must
3546	 * do so without other thread progressing to refcount_dec_and_test() of
3547	 * tcp_death_row.tw_refcount.
3548	 */
3549	mutex_lock(&tcp_exit_batch_mutex);
3550
3551	tcp_twsk_purge(net_exit_list);
3552
3553	list_for_each_entry(net, net_exit_list, exit_list) {
3554		inet_pernet_hashinfo_free(net->ipv4.tcp_death_row.hashinfo);
3555		WARN_ON_ONCE(!refcount_dec_and_test(&net->ipv4.tcp_death_row.tw_refcount));
3556		tcp_fastopen_ctx_destroy(net);
3557	}
3558
3559	mutex_unlock(&tcp_exit_batch_mutex);
3560}
3561
3562static struct pernet_operations __net_initdata tcp_sk_ops = {
3563       .init	   = tcp_sk_init,
3564       .exit	   = tcp_sk_exit,
3565       .exit_batch = tcp_sk_exit_batch,
3566};
3567
3568#if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
3569DEFINE_BPF_ITER_FUNC(tcp, struct bpf_iter_meta *meta,
3570		     struct sock_common *sk_common, uid_t uid)
3571
3572#define INIT_BATCH_SZ 16
3573
3574static int bpf_iter_init_tcp(void *priv_data, struct bpf_iter_aux_info *aux)
3575{
3576	struct bpf_tcp_iter_state *iter = priv_data;
3577	int err;
3578
3579	err = bpf_iter_init_seq_net(priv_data, aux);
3580	if (err)
3581		return err;
3582
3583	err = bpf_iter_tcp_realloc_batch(iter, INIT_BATCH_SZ);
3584	if (err) {
3585		bpf_iter_fini_seq_net(priv_data);
3586		return err;
3587	}
3588
3589	return 0;
3590}
3591
3592static void bpf_iter_fini_tcp(void *priv_data)
3593{
3594	struct bpf_tcp_iter_state *iter = priv_data;
3595
3596	bpf_iter_fini_seq_net(priv_data);
3597	kvfree(iter->batch);
3598}
3599
3600static const struct bpf_iter_seq_info tcp_seq_info = {
3601	.seq_ops		= &bpf_iter_tcp_seq_ops,
3602	.init_seq_private	= bpf_iter_init_tcp,
3603	.fini_seq_private	= bpf_iter_fini_tcp,
3604	.seq_priv_size		= sizeof(struct bpf_tcp_iter_state),
3605};
3606
3607static const struct bpf_func_proto *
3608bpf_iter_tcp_get_func_proto(enum bpf_func_id func_id,
3609			    const struct bpf_prog *prog)
3610{
3611	switch (func_id) {
3612	case BPF_FUNC_setsockopt:
3613		return &bpf_sk_setsockopt_proto;
3614	case BPF_FUNC_getsockopt:
3615		return &bpf_sk_getsockopt_proto;
3616	default:
3617		return NULL;
3618	}
3619}
3620
3621static struct bpf_iter_reg tcp_reg_info = {
3622	.target			= "tcp",
3623	.ctx_arg_info_size	= 1,
3624	.ctx_arg_info		= {
3625		{ offsetof(struct bpf_iter__tcp, sk_common),
3626		  PTR_TO_BTF_ID_OR_NULL | PTR_TRUSTED },
3627	},
3628	.get_func_proto		= bpf_iter_tcp_get_func_proto,
3629	.seq_info		= &tcp_seq_info,
3630};
3631
3632static void __init bpf_iter_register(void)
3633{
3634	tcp_reg_info.ctx_arg_info[0].btf_id = btf_sock_ids[BTF_SOCK_TYPE_SOCK_COMMON];
3635	if (bpf_iter_reg_target(&tcp_reg_info))
3636		pr_warn("Warning: could not register bpf iterator tcp\n");
3637}
3638
3639#endif
3640
3641void __init tcp_v4_init(void)
3642{
3643	int cpu, res;
3644
3645	for_each_possible_cpu(cpu) {
3646		struct sock *sk;
3647
3648		res = inet_ctl_sock_create(&sk, PF_INET, SOCK_RAW,
3649					   IPPROTO_TCP, &init_net);
3650		if (res)
3651			panic("Failed to create the TCP control socket.\n");
3652		sock_set_flag(sk, SOCK_USE_WRITE_QUEUE);
3653
3654		/* Please enforce IP_DF and IPID==0 for RST and
3655		 * ACK sent in SYN-RECV and TIME-WAIT state.
3656		 */
3657		inet_sk(sk)->pmtudisc = IP_PMTUDISC_DO;
3658
3659		sk->sk_clockid = CLOCK_MONOTONIC;
3660
3661		per_cpu(ipv4_tcp_sk.sock, cpu) = sk;
3662	}
3663	if (register_pernet_subsys(&tcp_sk_ops))
3664		panic("Failed to create the TCP control socket.\n");
3665
3666#if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
3667	bpf_iter_register();
3668#endif
3669}
v6.8
   1// SPDX-License-Identifier: GPL-2.0-or-later
   2/*
   3 * INET		An implementation of the TCP/IP protocol suite for the LINUX
   4 *		operating system.  INET is implemented using the  BSD Socket
   5 *		interface as the means of communication with the user level.
   6 *
   7 *		Implementation of the Transmission Control Protocol(TCP).
   8 *
   9 *		IPv4 specific functions
  10 *
  11 *		code split from:
  12 *		linux/ipv4/tcp.c
  13 *		linux/ipv4/tcp_input.c
  14 *		linux/ipv4/tcp_output.c
  15 *
  16 *		See tcp.c for author information
  17 */
  18
  19/*
  20 * Changes:
  21 *		David S. Miller	:	New socket lookup architecture.
  22 *					This code is dedicated to John Dyson.
  23 *		David S. Miller :	Change semantics of established hash,
  24 *					half is devoted to TIME_WAIT sockets
  25 *					and the rest go in the other half.
  26 *		Andi Kleen :		Add support for syncookies and fixed
  27 *					some bugs: ip options weren't passed to
  28 *					the TCP layer, missed a check for an
  29 *					ACK bit.
  30 *		Andi Kleen :		Implemented fast path mtu discovery.
  31 *	     				Fixed many serious bugs in the
  32 *					request_sock handling and moved
  33 *					most of it into the af independent code.
  34 *					Added tail drop and some other bugfixes.
  35 *					Added new listen semantics.
  36 *		Mike McLagan	:	Routing by source
  37 *	Juan Jose Ciarlante:		ip_dynaddr bits
  38 *		Andi Kleen:		various fixes.
  39 *	Vitaly E. Lavrov	:	Transparent proxy revived after year
  40 *					coma.
  41 *	Andi Kleen		:	Fix new listen.
  42 *	Andi Kleen		:	Fix accept error reporting.
  43 *	YOSHIFUJI Hideaki @USAGI and:	Support IPV6_V6ONLY socket option, which
  44 *	Alexey Kuznetsov		allow both IPv4 and IPv6 sockets to bind
  45 *					a single port at the same time.
  46 */
  47
  48#define pr_fmt(fmt) "TCP: " fmt
  49
  50#include <linux/bottom_half.h>
  51#include <linux/types.h>
  52#include <linux/fcntl.h>
  53#include <linux/module.h>
  54#include <linux/random.h>
  55#include <linux/cache.h>
  56#include <linux/jhash.h>
  57#include <linux/init.h>
  58#include <linux/times.h>
  59#include <linux/slab.h>
  60#include <linux/sched.h>
  61
  62#include <net/net_namespace.h>
  63#include <net/icmp.h>
  64#include <net/inet_hashtables.h>
  65#include <net/tcp.h>
  66#include <net/transp_v6.h>
  67#include <net/ipv6.h>
  68#include <net/inet_common.h>
  69#include <net/timewait_sock.h>
  70#include <net/xfrm.h>
  71#include <net/secure_seq.h>
  72#include <net/busy_poll.h>
 
  73
  74#include <linux/inet.h>
  75#include <linux/ipv6.h>
  76#include <linux/stddef.h>
  77#include <linux/proc_fs.h>
  78#include <linux/seq_file.h>
  79#include <linux/inetdevice.h>
  80#include <linux/btf_ids.h>
 
  81
  82#include <crypto/hash.h>
  83#include <linux/scatterlist.h>
  84
  85#include <trace/events/tcp.h>
  86
  87#ifdef CONFIG_TCP_MD5SIG
  88static int tcp_v4_md5_hash_hdr(char *md5_hash, const struct tcp_md5sig_key *key,
  89			       __be32 daddr, __be32 saddr, const struct tcphdr *th);
  90#endif
  91
  92struct inet_hashinfo tcp_hashinfo;
  93EXPORT_SYMBOL(tcp_hashinfo);
  94
  95static DEFINE_PER_CPU(struct sock *, ipv4_tcp_sk);
 
 
 
 
  96
  97static u32 tcp_v4_init_seq(const struct sk_buff *skb)
  98{
  99	return secure_tcp_seq(ip_hdr(skb)->daddr,
 100			      ip_hdr(skb)->saddr,
 101			      tcp_hdr(skb)->dest,
 102			      tcp_hdr(skb)->source);
 103}
 104
 105static u32 tcp_v4_init_ts_off(const struct net *net, const struct sk_buff *skb)
 106{
 107	return secure_tcp_ts_off(net, ip_hdr(skb)->daddr, ip_hdr(skb)->saddr);
 108}
 109
 110int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp)
 111{
 112	int reuse = READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_tw_reuse);
 113	const struct inet_timewait_sock *tw = inet_twsk(sktw);
 114	const struct tcp_timewait_sock *tcptw = tcp_twsk(sktw);
 115	struct tcp_sock *tp = tcp_sk(sk);
 
 
 
 
 116
 117	if (reuse == 2) {
 118		/* Still does not detect *everything* that goes through
 119		 * lo, since we require a loopback src or dst address
 120		 * or direct binding to 'lo' interface.
 121		 */
 122		bool loopback = false;
 123		if (tw->tw_bound_dev_if == LOOPBACK_IFINDEX)
 124			loopback = true;
 125#if IS_ENABLED(CONFIG_IPV6)
 126		if (tw->tw_family == AF_INET6) {
 127			if (ipv6_addr_loopback(&tw->tw_v6_daddr) ||
 128			    ipv6_addr_v4mapped_loopback(&tw->tw_v6_daddr) ||
 129			    ipv6_addr_loopback(&tw->tw_v6_rcv_saddr) ||
 130			    ipv6_addr_v4mapped_loopback(&tw->tw_v6_rcv_saddr))
 131				loopback = true;
 132		} else
 133#endif
 134		{
 135			if (ipv4_is_loopback(tw->tw_daddr) ||
 136			    ipv4_is_loopback(tw->tw_rcv_saddr))
 137				loopback = true;
 138		}
 139		if (!loopback)
 140			reuse = 0;
 141	}
 142
 143	/* With PAWS, it is safe from the viewpoint
 144	   of data integrity. Even without PAWS it is safe provided sequence
 145	   spaces do not overlap i.e. at data rates <= 80Mbit/sec.
 146
 147	   Actually, the idea is close to VJ's one, only timestamp cache is
 148	   held not per host, but per port pair and TW bucket is used as state
 149	   holder.
 150
 151	   If TW bucket has been already destroyed we fall back to VJ's scheme
 152	   and use initial timestamp retrieved from peer table.
 153	 */
 154	if (tcptw->tw_ts_recent_stamp &&
 
 155	    (!twp || (reuse && time_after32(ktime_get_seconds(),
 156					    tcptw->tw_ts_recent_stamp)))) {
 
 
 
 
 
 
 157		/* In case of repair and re-using TIME-WAIT sockets we still
 158		 * want to be sure that it is safe as above but honor the
 159		 * sequence numbers and time stamps set as part of the repair
 160		 * process.
 161		 *
 162		 * Without this check re-using a TIME-WAIT socket with TCP
 163		 * repair would accumulate a -1 on the repair assigned
 164		 * sequence number. The first time it is reused the sequence
 165		 * is -1, the second time -2, etc. This fixes that issue
 166		 * without appearing to create any others.
 167		 */
 168		if (likely(!tp->repair)) {
 169			u32 seq = tcptw->tw_snd_nxt + 65535 + 2;
 170
 171			if (!seq)
 172				seq = 1;
 173			WRITE_ONCE(tp->write_seq, seq);
 174			tp->rx_opt.ts_recent	   = tcptw->tw_ts_recent;
 175			tp->rx_opt.ts_recent_stamp = tcptw->tw_ts_recent_stamp;
 176		}
 177		sock_hold(sktw);
 178		return 1;
 179	}
 180
 181	return 0;
 182}
 183EXPORT_SYMBOL_GPL(tcp_twsk_unique);
 184
 185static int tcp_v4_pre_connect(struct sock *sk, struct sockaddr *uaddr,
 186			      int addr_len)
 187{
 188	/* This check is replicated from tcp_v4_connect() and intended to
 189	 * prevent BPF program called below from accessing bytes that are out
 190	 * of the bound specified by user in addr_len.
 191	 */
 192	if (addr_len < sizeof(struct sockaddr_in))
 193		return -EINVAL;
 194
 195	sock_owned_by_me(sk);
 196
 197	return BPF_CGROUP_RUN_PROG_INET4_CONNECT(sk, uaddr, &addr_len);
 198}
 199
 200/* This will initiate an outgoing connection. */
 201int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len)
 202{
 203	struct sockaddr_in *usin = (struct sockaddr_in *)uaddr;
 204	struct inet_timewait_death_row *tcp_death_row;
 205	struct inet_sock *inet = inet_sk(sk);
 206	struct tcp_sock *tp = tcp_sk(sk);
 207	struct ip_options_rcu *inet_opt;
 208	struct net *net = sock_net(sk);
 209	__be16 orig_sport, orig_dport;
 210	__be32 daddr, nexthop;
 211	struct flowi4 *fl4;
 212	struct rtable *rt;
 213	int err;
 214
 215	if (addr_len < sizeof(struct sockaddr_in))
 216		return -EINVAL;
 217
 218	if (usin->sin_family != AF_INET)
 219		return -EAFNOSUPPORT;
 220
 221	nexthop = daddr = usin->sin_addr.s_addr;
 222	inet_opt = rcu_dereference_protected(inet->inet_opt,
 223					     lockdep_sock_is_held(sk));
 224	if (inet_opt && inet_opt->opt.srr) {
 225		if (!daddr)
 226			return -EINVAL;
 227		nexthop = inet_opt->opt.faddr;
 228	}
 229
 230	orig_sport = inet->inet_sport;
 231	orig_dport = usin->sin_port;
 232	fl4 = &inet->cork.fl.u.ip4;
 233	rt = ip_route_connect(fl4, nexthop, inet->inet_saddr,
 234			      sk->sk_bound_dev_if, IPPROTO_TCP, orig_sport,
 235			      orig_dport, sk);
 236	if (IS_ERR(rt)) {
 237		err = PTR_ERR(rt);
 238		if (err == -ENETUNREACH)
 239			IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES);
 240		return err;
 241	}
 242
 243	if (rt->rt_flags & (RTCF_MULTICAST | RTCF_BROADCAST)) {
 244		ip_rt_put(rt);
 245		return -ENETUNREACH;
 246	}
 247
 248	if (!inet_opt || !inet_opt->opt.srr)
 249		daddr = fl4->daddr;
 250
 251	tcp_death_row = &sock_net(sk)->ipv4.tcp_death_row;
 252
 253	if (!inet->inet_saddr) {
 254		err = inet_bhash2_update_saddr(sk,  &fl4->saddr, AF_INET);
 255		if (err) {
 256			ip_rt_put(rt);
 257			return err;
 258		}
 259	} else {
 260		sk_rcv_saddr_set(sk, inet->inet_saddr);
 261	}
 262
 263	if (tp->rx_opt.ts_recent_stamp && inet->inet_daddr != daddr) {
 264		/* Reset inherited state */
 265		tp->rx_opt.ts_recent	   = 0;
 266		tp->rx_opt.ts_recent_stamp = 0;
 267		if (likely(!tp->repair))
 268			WRITE_ONCE(tp->write_seq, 0);
 269	}
 270
 271	inet->inet_dport = usin->sin_port;
 272	sk_daddr_set(sk, daddr);
 273
 274	inet_csk(sk)->icsk_ext_hdr_len = 0;
 275	if (inet_opt)
 276		inet_csk(sk)->icsk_ext_hdr_len = inet_opt->opt.optlen;
 277
 278	tp->rx_opt.mss_clamp = TCP_MSS_DEFAULT;
 279
 280	/* Socket identity is still unknown (sport may be zero).
 281	 * However we set state to SYN-SENT and not releasing socket
 282	 * lock select source port, enter ourselves into the hash tables and
 283	 * complete initialization after this.
 284	 */
 285	tcp_set_state(sk, TCP_SYN_SENT);
 286	err = inet_hash_connect(tcp_death_row, sk);
 287	if (err)
 288		goto failure;
 289
 290	sk_set_txhash(sk);
 291
 292	rt = ip_route_newports(fl4, rt, orig_sport, orig_dport,
 293			       inet->inet_sport, inet->inet_dport, sk);
 294	if (IS_ERR(rt)) {
 295		err = PTR_ERR(rt);
 296		rt = NULL;
 297		goto failure;
 298	}
 299	tp->tcp_usec_ts = dst_tcp_usec_ts(&rt->dst);
 300	/* OK, now commit destination to socket.  */
 301	sk->sk_gso_type = SKB_GSO_TCPV4;
 302	sk_setup_caps(sk, &rt->dst);
 303	rt = NULL;
 304
 305	if (likely(!tp->repair)) {
 306		if (!tp->write_seq)
 307			WRITE_ONCE(tp->write_seq,
 308				   secure_tcp_seq(inet->inet_saddr,
 309						  inet->inet_daddr,
 310						  inet->inet_sport,
 311						  usin->sin_port));
 312		WRITE_ONCE(tp->tsoffset,
 313			   secure_tcp_ts_off(net, inet->inet_saddr,
 314					     inet->inet_daddr));
 315	}
 316
 317	atomic_set(&inet->inet_id, get_random_u16());
 318
 319	if (tcp_fastopen_defer_connect(sk, &err))
 320		return err;
 321	if (err)
 322		goto failure;
 323
 324	err = tcp_connect(sk);
 325
 326	if (err)
 327		goto failure;
 328
 329	return 0;
 330
 331failure:
 332	/*
 333	 * This unhashes the socket and releases the local port,
 334	 * if necessary.
 335	 */
 336	tcp_set_state(sk, TCP_CLOSE);
 337	inet_bhash2_reset_saddr(sk);
 338	ip_rt_put(rt);
 339	sk->sk_route_caps = 0;
 340	inet->inet_dport = 0;
 341	return err;
 342}
 343EXPORT_SYMBOL(tcp_v4_connect);
 344
 345/*
 346 * This routine reacts to ICMP_FRAG_NEEDED mtu indications as defined in RFC1191.
 347 * It can be called through tcp_release_cb() if socket was owned by user
 348 * at the time tcp_v4_err() was called to handle ICMP message.
 349 */
 350void tcp_v4_mtu_reduced(struct sock *sk)
 351{
 352	struct inet_sock *inet = inet_sk(sk);
 353	struct dst_entry *dst;
 354	u32 mtu;
 355
 356	if ((1 << sk->sk_state) & (TCPF_LISTEN | TCPF_CLOSE))
 357		return;
 358	mtu = READ_ONCE(tcp_sk(sk)->mtu_info);
 359	dst = inet_csk_update_pmtu(sk, mtu);
 360	if (!dst)
 361		return;
 362
 363	/* Something is about to be wrong... Remember soft error
 364	 * for the case, if this connection will not able to recover.
 365	 */
 366	if (mtu < dst_mtu(dst) && ip_dont_fragment(sk, dst))
 367		WRITE_ONCE(sk->sk_err_soft, EMSGSIZE);
 368
 369	mtu = dst_mtu(dst);
 370
 371	if (inet->pmtudisc != IP_PMTUDISC_DONT &&
 372	    ip_sk_accept_pmtu(sk) &&
 373	    inet_csk(sk)->icsk_pmtu_cookie > mtu) {
 374		tcp_sync_mss(sk, mtu);
 375
 376		/* Resend the TCP packet because it's
 377		 * clear that the old packet has been
 378		 * dropped. This is the new "fast" path mtu
 379		 * discovery.
 380		 */
 381		tcp_simple_retransmit(sk);
 382	} /* else let the usual retransmit timer handle it */
 383}
 384EXPORT_SYMBOL(tcp_v4_mtu_reduced);
 385
 386static void do_redirect(struct sk_buff *skb, struct sock *sk)
 387{
 388	struct dst_entry *dst = __sk_dst_check(sk, 0);
 389
 390	if (dst)
 391		dst->ops->redirect(dst, sk, skb);
 392}
 393
 394
 395/* handle ICMP messages on TCP_NEW_SYN_RECV request sockets */
 396void tcp_req_err(struct sock *sk, u32 seq, bool abort)
 397{
 398	struct request_sock *req = inet_reqsk(sk);
 399	struct net *net = sock_net(sk);
 400
 401	/* ICMPs are not backlogged, hence we cannot get
 402	 * an established socket here.
 403	 */
 404	if (seq != tcp_rsk(req)->snt_isn) {
 405		__NET_INC_STATS(net, LINUX_MIB_OUTOFWINDOWICMPS);
 406	} else if (abort) {
 407		/*
 408		 * Still in SYN_RECV, just remove it silently.
 409		 * There is no good way to pass the error to the newly
 410		 * created socket, and POSIX does not want network
 411		 * errors returned from accept().
 412		 */
 413		inet_csk_reqsk_queue_drop(req->rsk_listener, req);
 414		tcp_listendrop(req->rsk_listener);
 415	}
 416	reqsk_put(req);
 417}
 418EXPORT_SYMBOL(tcp_req_err);
 419
 420/* TCP-LD (RFC 6069) logic */
 421void tcp_ld_RTO_revert(struct sock *sk, u32 seq)
 422{
 423	struct inet_connection_sock *icsk = inet_csk(sk);
 424	struct tcp_sock *tp = tcp_sk(sk);
 425	struct sk_buff *skb;
 426	s32 remaining;
 427	u32 delta_us;
 428
 429	if (sock_owned_by_user(sk))
 430		return;
 431
 432	if (seq != tp->snd_una  || !icsk->icsk_retransmits ||
 433	    !icsk->icsk_backoff)
 434		return;
 435
 436	skb = tcp_rtx_queue_head(sk);
 437	if (WARN_ON_ONCE(!skb))
 438		return;
 439
 440	icsk->icsk_backoff--;
 441	icsk->icsk_rto = tp->srtt_us ? __tcp_set_rto(tp) : TCP_TIMEOUT_INIT;
 442	icsk->icsk_rto = inet_csk_rto_backoff(icsk, TCP_RTO_MAX);
 443
 444	tcp_mstamp_refresh(tp);
 445	delta_us = (u32)(tp->tcp_mstamp - tcp_skb_timestamp_us(skb));
 446	remaining = icsk->icsk_rto - usecs_to_jiffies(delta_us);
 447
 448	if (remaining > 0) {
 449		inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
 450					  remaining, TCP_RTO_MAX);
 451	} else {
 452		/* RTO revert clocked out retransmission.
 453		 * Will retransmit now.
 454		 */
 455		tcp_retransmit_timer(sk);
 456	}
 457}
 458EXPORT_SYMBOL(tcp_ld_RTO_revert);
 459
 460/*
 461 * This routine is called by the ICMP module when it gets some
 462 * sort of error condition.  If err < 0 then the socket should
 463 * be closed and the error returned to the user.  If err > 0
 464 * it's just the icmp type << 8 | icmp code.  After adjustment
 465 * header points to the first 8 bytes of the tcp header.  We need
 466 * to find the appropriate port.
 467 *
 468 * The locking strategy used here is very "optimistic". When
 469 * someone else accesses the socket the ICMP is just dropped
 470 * and for some paths there is no check at all.
 471 * A more general error queue to queue errors for later handling
 472 * is probably better.
 473 *
 474 */
 475
 476int tcp_v4_err(struct sk_buff *skb, u32 info)
 477{
 478	const struct iphdr *iph = (const struct iphdr *)skb->data;
 479	struct tcphdr *th = (struct tcphdr *)(skb->data + (iph->ihl << 2));
 480	struct tcp_sock *tp;
 481	const int type = icmp_hdr(skb)->type;
 482	const int code = icmp_hdr(skb)->code;
 483	struct sock *sk;
 484	struct request_sock *fastopen;
 485	u32 seq, snd_una;
 486	int err;
 487	struct net *net = dev_net(skb->dev);
 488
 489	sk = __inet_lookup_established(net, net->ipv4.tcp_death_row.hashinfo,
 490				       iph->daddr, th->dest, iph->saddr,
 491				       ntohs(th->source), inet_iif(skb), 0);
 492	if (!sk) {
 493		__ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
 494		return -ENOENT;
 495	}
 496	if (sk->sk_state == TCP_TIME_WAIT) {
 497		/* To increase the counter of ignored icmps for TCP-AO */
 498		tcp_ao_ignore_icmp(sk, AF_INET, type, code);
 499		inet_twsk_put(inet_twsk(sk));
 500		return 0;
 501	}
 502	seq = ntohl(th->seq);
 503	if (sk->sk_state == TCP_NEW_SYN_RECV) {
 504		tcp_req_err(sk, seq, type == ICMP_PARAMETERPROB ||
 505				     type == ICMP_TIME_EXCEEDED ||
 506				     (type == ICMP_DEST_UNREACH &&
 507				      (code == ICMP_NET_UNREACH ||
 508				       code == ICMP_HOST_UNREACH)));
 509		return 0;
 510	}
 511
 512	if (tcp_ao_ignore_icmp(sk, AF_INET, type, code)) {
 513		sock_put(sk);
 514		return 0;
 515	}
 516
 517	bh_lock_sock(sk);
 518	/* If too many ICMPs get dropped on busy
 519	 * servers this needs to be solved differently.
 520	 * We do take care of PMTU discovery (RFC1191) special case :
 521	 * we can receive locally generated ICMP messages while socket is held.
 522	 */
 523	if (sock_owned_by_user(sk)) {
 524		if (!(type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED))
 525			__NET_INC_STATS(net, LINUX_MIB_LOCKDROPPEDICMPS);
 526	}
 527	if (sk->sk_state == TCP_CLOSE)
 528		goto out;
 529
 530	if (static_branch_unlikely(&ip4_min_ttl)) {
 531		/* min_ttl can be changed concurrently from do_ip_setsockopt() */
 532		if (unlikely(iph->ttl < READ_ONCE(inet_sk(sk)->min_ttl))) {
 533			__NET_INC_STATS(net, LINUX_MIB_TCPMINTTLDROP);
 534			goto out;
 535		}
 536	}
 537
 538	tp = tcp_sk(sk);
 539	/* XXX (TFO) - tp->snd_una should be ISN (tcp_create_openreq_child() */
 540	fastopen = rcu_dereference(tp->fastopen_rsk);
 541	snd_una = fastopen ? tcp_rsk(fastopen)->snt_isn : tp->snd_una;
 542	if (sk->sk_state != TCP_LISTEN &&
 543	    !between(seq, snd_una, tp->snd_nxt)) {
 544		__NET_INC_STATS(net, LINUX_MIB_OUTOFWINDOWICMPS);
 545		goto out;
 546	}
 547
 548	switch (type) {
 549	case ICMP_REDIRECT:
 550		if (!sock_owned_by_user(sk))
 551			do_redirect(skb, sk);
 552		goto out;
 553	case ICMP_SOURCE_QUENCH:
 554		/* Just silently ignore these. */
 555		goto out;
 556	case ICMP_PARAMETERPROB:
 557		err = EPROTO;
 558		break;
 559	case ICMP_DEST_UNREACH:
 560		if (code > NR_ICMP_UNREACH)
 561			goto out;
 562
 563		if (code == ICMP_FRAG_NEEDED) { /* PMTU discovery (RFC1191) */
 564			/* We are not interested in TCP_LISTEN and open_requests
 565			 * (SYN-ACKs send out by Linux are always <576bytes so
 566			 * they should go through unfragmented).
 567			 */
 568			if (sk->sk_state == TCP_LISTEN)
 569				goto out;
 570
 571			WRITE_ONCE(tp->mtu_info, info);
 572			if (!sock_owned_by_user(sk)) {
 573				tcp_v4_mtu_reduced(sk);
 574			} else {
 575				if (!test_and_set_bit(TCP_MTU_REDUCED_DEFERRED, &sk->sk_tsq_flags))
 576					sock_hold(sk);
 577			}
 578			goto out;
 579		}
 580
 581		err = icmp_err_convert[code].errno;
 582		/* check if this ICMP message allows revert of backoff.
 583		 * (see RFC 6069)
 584		 */
 585		if (!fastopen &&
 586		    (code == ICMP_NET_UNREACH || code == ICMP_HOST_UNREACH))
 587			tcp_ld_RTO_revert(sk, seq);
 588		break;
 589	case ICMP_TIME_EXCEEDED:
 590		err = EHOSTUNREACH;
 591		break;
 592	default:
 593		goto out;
 594	}
 595
 596	switch (sk->sk_state) {
 597	case TCP_SYN_SENT:
 598	case TCP_SYN_RECV:
 599		/* Only in fast or simultaneous open. If a fast open socket is
 600		 * already accepted it is treated as a connected one below.
 601		 */
 602		if (fastopen && !fastopen->sk)
 603			break;
 604
 605		ip_icmp_error(sk, skb, err, th->dest, info, (u8 *)th);
 606
 607		if (!sock_owned_by_user(sk)) {
 608			WRITE_ONCE(sk->sk_err, err);
 609
 610			sk_error_report(sk);
 611
 612			tcp_done(sk);
 613		} else {
 614			WRITE_ONCE(sk->sk_err_soft, err);
 615		}
 616		goto out;
 617	}
 618
 619	/* If we've already connected we will keep trying
 620	 * until we time out, or the user gives up.
 621	 *
 622	 * rfc1122 4.2.3.9 allows to consider as hard errors
 623	 * only PROTO_UNREACH and PORT_UNREACH (well, FRAG_FAILED too,
 624	 * but it is obsoleted by pmtu discovery).
 625	 *
 626	 * Note, that in modern internet, where routing is unreliable
 627	 * and in each dark corner broken firewalls sit, sending random
 628	 * errors ordered by their masters even this two messages finally lose
 629	 * their original sense (even Linux sends invalid PORT_UNREACHs)
 630	 *
 631	 * Now we are in compliance with RFCs.
 632	 *							--ANK (980905)
 633	 */
 634
 635	if (!sock_owned_by_user(sk) &&
 636	    inet_test_bit(RECVERR, sk)) {
 637		WRITE_ONCE(sk->sk_err, err);
 638		sk_error_report(sk);
 639	} else	{ /* Only an error on timeout */
 640		WRITE_ONCE(sk->sk_err_soft, err);
 641	}
 642
 643out:
 644	bh_unlock_sock(sk);
 645	sock_put(sk);
 646	return 0;
 647}
 648
 649void __tcp_v4_send_check(struct sk_buff *skb, __be32 saddr, __be32 daddr)
 650{
 651	struct tcphdr *th = tcp_hdr(skb);
 652
 653	th->check = ~tcp_v4_check(skb->len, saddr, daddr, 0);
 654	skb->csum_start = skb_transport_header(skb) - skb->head;
 655	skb->csum_offset = offsetof(struct tcphdr, check);
 656}
 657
 658/* This routine computes an IPv4 TCP checksum. */
 659void tcp_v4_send_check(struct sock *sk, struct sk_buff *skb)
 660{
 661	const struct inet_sock *inet = inet_sk(sk);
 662
 663	__tcp_v4_send_check(skb, inet->inet_saddr, inet->inet_daddr);
 664}
 665EXPORT_SYMBOL(tcp_v4_send_check);
 666
 667#define REPLY_OPTIONS_LEN      (MAX_TCP_OPTION_SPACE / sizeof(__be32))
 668
 669static bool tcp_v4_ao_sign_reset(const struct sock *sk, struct sk_buff *skb,
 670				 const struct tcp_ao_hdr *aoh,
 671				 struct ip_reply_arg *arg, struct tcphdr *reply,
 672				 __be32 reply_options[REPLY_OPTIONS_LEN])
 673{
 674#ifdef CONFIG_TCP_AO
 675	int sdif = tcp_v4_sdif(skb);
 676	int dif = inet_iif(skb);
 677	int l3index = sdif ? dif : 0;
 678	bool allocated_traffic_key;
 679	struct tcp_ao_key *key;
 680	char *traffic_key;
 681	bool drop = true;
 682	u32 ao_sne = 0;
 683	u8 keyid;
 684
 685	rcu_read_lock();
 686	if (tcp_ao_prepare_reset(sk, skb, aoh, l3index, ntohl(reply->seq),
 687				 &key, &traffic_key, &allocated_traffic_key,
 688				 &keyid, &ao_sne))
 689		goto out;
 690
 691	reply_options[0] = htonl((TCPOPT_AO << 24) | (tcp_ao_len(key) << 16) |
 692				 (aoh->rnext_keyid << 8) | keyid);
 693	arg->iov[0].iov_len += tcp_ao_len_aligned(key);
 694	reply->doff = arg->iov[0].iov_len / 4;
 695
 696	if (tcp_ao_hash_hdr(AF_INET, (char *)&reply_options[1],
 697			    key, traffic_key,
 698			    (union tcp_ao_addr *)&ip_hdr(skb)->saddr,
 699			    (union tcp_ao_addr *)&ip_hdr(skb)->daddr,
 700			    reply, ao_sne))
 701		goto out;
 702	drop = false;
 703out:
 704	rcu_read_unlock();
 705	if (allocated_traffic_key)
 706		kfree(traffic_key);
 707	return drop;
 708#else
 709	return true;
 710#endif
 711}
 712
 713/*
 714 *	This routine will send an RST to the other tcp.
 715 *
 716 *	Someone asks: why I NEVER use socket parameters (TOS, TTL etc.)
 717 *		      for reset.
 718 *	Answer: if a packet caused RST, it is not for a socket
 719 *		existing in our system, if it is matched to a socket,
 720 *		it is just duplicate segment or bug in other side's TCP.
 721 *		So that we build reply only basing on parameters
 722 *		arrived with segment.
 723 *	Exception: precedence violation. We do not implement it in any case.
 724 */
 725
 726static void tcp_v4_send_reset(const struct sock *sk, struct sk_buff *skb)
 
 727{
 728	const struct tcphdr *th = tcp_hdr(skb);
 729	struct {
 730		struct tcphdr th;
 731		__be32 opt[REPLY_OPTIONS_LEN];
 732	} rep;
 733	const __u8 *md5_hash_location = NULL;
 734	const struct tcp_ao_hdr *aoh;
 735	struct ip_reply_arg arg;
 736#ifdef CONFIG_TCP_MD5SIG
 737	struct tcp_md5sig_key *key = NULL;
 738	unsigned char newhash[16];
 739	struct sock *sk1 = NULL;
 740	int genhash;
 741#endif
 742	u64 transmit_time = 0;
 743	struct sock *ctl_sk;
 744	struct net *net;
 745	u32 txhash = 0;
 746
 747	/* Never send a reset in response to a reset. */
 748	if (th->rst)
 749		return;
 750
 751	/* If sk not NULL, it means we did a successful lookup and incoming
 752	 * route had to be correct. prequeue might have dropped our dst.
 753	 */
 754	if (!sk && skb_rtable(skb)->rt_type != RTN_LOCAL)
 755		return;
 756
 757	/* Swap the send and the receive. */
 758	memset(&rep, 0, sizeof(rep));
 759	rep.th.dest   = th->source;
 760	rep.th.source = th->dest;
 761	rep.th.doff   = sizeof(struct tcphdr) / 4;
 762	rep.th.rst    = 1;
 763
 764	if (th->ack) {
 765		rep.th.seq = th->ack_seq;
 766	} else {
 767		rep.th.ack = 1;
 768		rep.th.ack_seq = htonl(ntohl(th->seq) + th->syn + th->fin +
 769				       skb->len - (th->doff << 2));
 770	}
 771
 772	memset(&arg, 0, sizeof(arg));
 773	arg.iov[0].iov_base = (unsigned char *)&rep;
 774	arg.iov[0].iov_len  = sizeof(rep.th);
 775
 776	net = sk ? sock_net(sk) : dev_net(skb_dst(skb)->dev);
 777
 778	/* Invalid TCP option size or twice included auth */
 779	if (tcp_parse_auth_options(tcp_hdr(skb), &md5_hash_location, &aoh))
 780		return;
 781
 782	if (aoh && tcp_v4_ao_sign_reset(sk, skb, aoh, &arg, &rep.th, rep.opt))
 783		return;
 784
 785#ifdef CONFIG_TCP_MD5SIG
 786	rcu_read_lock();
 787	if (sk && sk_fullsock(sk)) {
 788		const union tcp_md5_addr *addr;
 789		int l3index;
 790
 791		/* sdif set, means packet ingressed via a device
 792		 * in an L3 domain and inet_iif is set to it.
 793		 */
 794		l3index = tcp_v4_sdif(skb) ? inet_iif(skb) : 0;
 795		addr = (union tcp_md5_addr *)&ip_hdr(skb)->saddr;
 796		key = tcp_md5_do_lookup(sk, l3index, addr, AF_INET);
 797	} else if (md5_hash_location) {
 798		const union tcp_md5_addr *addr;
 799		int sdif = tcp_v4_sdif(skb);
 800		int dif = inet_iif(skb);
 801		int l3index;
 802
 803		/*
 804		 * active side is lost. Try to find listening socket through
 805		 * source port, and then find md5 key through listening socket.
 806		 * we are not loose security here:
 807		 * Incoming packet is checked with md5 hash with finding key,
 808		 * no RST generated if md5 hash doesn't match.
 809		 */
 810		sk1 = __inet_lookup_listener(net, net->ipv4.tcp_death_row.hashinfo,
 811					     NULL, 0, ip_hdr(skb)->saddr,
 812					     th->source, ip_hdr(skb)->daddr,
 813					     ntohs(th->source), dif, sdif);
 814		/* don't send rst if it can't find key */
 815		if (!sk1)
 816			goto out;
 817
 818		/* sdif set, means packet ingressed via a device
 819		 * in an L3 domain and dif is set to it.
 820		 */
 821		l3index = sdif ? dif : 0;
 822		addr = (union tcp_md5_addr *)&ip_hdr(skb)->saddr;
 823		key = tcp_md5_do_lookup(sk1, l3index, addr, AF_INET);
 824		if (!key)
 825			goto out;
 826
 827
 828		genhash = tcp_v4_md5_hash_skb(newhash, key, NULL, skb);
 829		if (genhash || memcmp(md5_hash_location, newhash, 16) != 0)
 830			goto out;
 831
 832	}
 833
 834	if (key) {
 835		rep.opt[0] = htonl((TCPOPT_NOP << 24) |
 836				   (TCPOPT_NOP << 16) |
 837				   (TCPOPT_MD5SIG << 8) |
 838				   TCPOLEN_MD5SIG);
 839		/* Update length and the length the header thinks exists */
 840		arg.iov[0].iov_len += TCPOLEN_MD5SIG_ALIGNED;
 841		rep.th.doff = arg.iov[0].iov_len / 4;
 842
 843		tcp_v4_md5_hash_hdr((__u8 *) &rep.opt[1],
 844				     key, ip_hdr(skb)->saddr,
 845				     ip_hdr(skb)->daddr, &rep.th);
 846	}
 847#endif
 848	/* Can't co-exist with TCPMD5, hence check rep.opt[0] */
 849	if (rep.opt[0] == 0) {
 850		__be32 mrst = mptcp_reset_option(skb);
 851
 852		if (mrst) {
 853			rep.opt[0] = mrst;
 854			arg.iov[0].iov_len += sizeof(mrst);
 855			rep.th.doff = arg.iov[0].iov_len / 4;
 856		}
 857	}
 858
 859	arg.csum = csum_tcpudp_nofold(ip_hdr(skb)->daddr,
 860				      ip_hdr(skb)->saddr, /* XXX */
 861				      arg.iov[0].iov_len, IPPROTO_TCP, 0);
 862	arg.csumoffset = offsetof(struct tcphdr, check) / 2;
 863	arg.flags = (sk && inet_sk_transparent(sk)) ? IP_REPLY_ARG_NOSRCCHECK : 0;
 864
 865	/* When socket is gone, all binding information is lost.
 866	 * routing might fail in this case. No choice here, if we choose to force
 867	 * input interface, we will misroute in case of asymmetric route.
 868	 */
 869	if (sk) {
 870		arg.bound_dev_if = sk->sk_bound_dev_if;
 871		if (sk_fullsock(sk))
 872			trace_tcp_send_reset(sk, skb);
 873	}
 874
 875	BUILD_BUG_ON(offsetof(struct sock, sk_bound_dev_if) !=
 876		     offsetof(struct inet_timewait_sock, tw_bound_dev_if));
 877
 878	arg.tos = ip_hdr(skb)->tos;
 879	arg.uid = sock_net_uid(net, sk && sk_fullsock(sk) ? sk : NULL);
 880	local_bh_disable();
 881	ctl_sk = this_cpu_read(ipv4_tcp_sk);
 
 
 882	sock_net_set(ctl_sk, net);
 883	if (sk) {
 884		ctl_sk->sk_mark = (sk->sk_state == TCP_TIME_WAIT) ?
 885				   inet_twsk(sk)->tw_mark : sk->sk_mark;
 886		ctl_sk->sk_priority = (sk->sk_state == TCP_TIME_WAIT) ?
 887				   inet_twsk(sk)->tw_priority : READ_ONCE(sk->sk_priority);
 888		transmit_time = tcp_transmit_time(sk);
 889		xfrm_sk_clone_policy(ctl_sk, sk);
 890		txhash = (sk->sk_state == TCP_TIME_WAIT) ?
 891			 inet_twsk(sk)->tw_txhash : sk->sk_txhash;
 892	} else {
 893		ctl_sk->sk_mark = 0;
 894		ctl_sk->sk_priority = 0;
 895	}
 896	ip_send_unicast_reply(ctl_sk,
 897			      skb, &TCP_SKB_CB(skb)->header.h4.opt,
 898			      ip_hdr(skb)->saddr, ip_hdr(skb)->daddr,
 899			      &arg, arg.iov[0].iov_len,
 900			      transmit_time, txhash);
 901
 902	xfrm_sk_free_policy(ctl_sk);
 903	sock_net_set(ctl_sk, &init_net);
 904	__TCP_INC_STATS(net, TCP_MIB_OUTSEGS);
 905	__TCP_INC_STATS(net, TCP_MIB_OUTRSTS);
 
 906	local_bh_enable();
 907
 908#ifdef CONFIG_TCP_MD5SIG
 909out:
 910	rcu_read_unlock();
 911#endif
 912}
 913
 914/* The code following below sending ACKs in SYN-RECV and TIME-WAIT states
 915   outside socket context is ugly, certainly. What can I do?
 916 */
 917
 918static void tcp_v4_send_ack(const struct sock *sk,
 919			    struct sk_buff *skb, u32 seq, u32 ack,
 920			    u32 win, u32 tsval, u32 tsecr, int oif,
 921			    struct tcp_key *key,
 922			    int reply_flags, u8 tos, u32 txhash)
 923{
 924	const struct tcphdr *th = tcp_hdr(skb);
 925	struct {
 926		struct tcphdr th;
 927		__be32 opt[(MAX_TCP_OPTION_SPACE  >> 2)];
 928	} rep;
 929	struct net *net = sock_net(sk);
 930	struct ip_reply_arg arg;
 931	struct sock *ctl_sk;
 932	u64 transmit_time;
 933
 934	memset(&rep.th, 0, sizeof(struct tcphdr));
 935	memset(&arg, 0, sizeof(arg));
 936
 937	arg.iov[0].iov_base = (unsigned char *)&rep;
 938	arg.iov[0].iov_len  = sizeof(rep.th);
 939	if (tsecr) {
 940		rep.opt[0] = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
 941				   (TCPOPT_TIMESTAMP << 8) |
 942				   TCPOLEN_TIMESTAMP);
 943		rep.opt[1] = htonl(tsval);
 944		rep.opt[2] = htonl(tsecr);
 945		arg.iov[0].iov_len += TCPOLEN_TSTAMP_ALIGNED;
 946	}
 947
 948	/* Swap the send and the receive. */
 949	rep.th.dest    = th->source;
 950	rep.th.source  = th->dest;
 951	rep.th.doff    = arg.iov[0].iov_len / 4;
 952	rep.th.seq     = htonl(seq);
 953	rep.th.ack_seq = htonl(ack);
 954	rep.th.ack     = 1;
 955	rep.th.window  = htons(win);
 956
 957#ifdef CONFIG_TCP_MD5SIG
 958	if (tcp_key_is_md5(key)) {
 959		int offset = (tsecr) ? 3 : 0;
 960
 961		rep.opt[offset++] = htonl((TCPOPT_NOP << 24) |
 962					  (TCPOPT_NOP << 16) |
 963					  (TCPOPT_MD5SIG << 8) |
 964					  TCPOLEN_MD5SIG);
 965		arg.iov[0].iov_len += TCPOLEN_MD5SIG_ALIGNED;
 966		rep.th.doff = arg.iov[0].iov_len/4;
 967
 968		tcp_v4_md5_hash_hdr((__u8 *) &rep.opt[offset],
 969				    key->md5_key, ip_hdr(skb)->saddr,
 970				    ip_hdr(skb)->daddr, &rep.th);
 971	}
 972#endif
 973#ifdef CONFIG_TCP_AO
 974	if (tcp_key_is_ao(key)) {
 975		int offset = (tsecr) ? 3 : 0;
 976
 977		rep.opt[offset++] = htonl((TCPOPT_AO << 24) |
 978					  (tcp_ao_len(key->ao_key) << 16) |
 979					  (key->ao_key->sndid << 8) |
 980					  key->rcv_next);
 981		arg.iov[0].iov_len += tcp_ao_len_aligned(key->ao_key);
 982		rep.th.doff = arg.iov[0].iov_len / 4;
 983
 984		tcp_ao_hash_hdr(AF_INET, (char *)&rep.opt[offset],
 985				key->ao_key, key->traffic_key,
 986				(union tcp_ao_addr *)&ip_hdr(skb)->saddr,
 987				(union tcp_ao_addr *)&ip_hdr(skb)->daddr,
 988				&rep.th, key->sne);
 989	}
 990#endif
 991	arg.flags = reply_flags;
 992	arg.csum = csum_tcpudp_nofold(ip_hdr(skb)->daddr,
 993				      ip_hdr(skb)->saddr, /* XXX */
 994				      arg.iov[0].iov_len, IPPROTO_TCP, 0);
 995	arg.csumoffset = offsetof(struct tcphdr, check) / 2;
 996	if (oif)
 997		arg.bound_dev_if = oif;
 998	arg.tos = tos;
 999	arg.uid = sock_net_uid(net, sk_fullsock(sk) ? sk : NULL);
1000	local_bh_disable();
1001	ctl_sk = this_cpu_read(ipv4_tcp_sk);
 
1002	sock_net_set(ctl_sk, net);
1003	ctl_sk->sk_mark = (sk->sk_state == TCP_TIME_WAIT) ?
1004			   inet_twsk(sk)->tw_mark : READ_ONCE(sk->sk_mark);
1005	ctl_sk->sk_priority = (sk->sk_state == TCP_TIME_WAIT) ?
1006			   inet_twsk(sk)->tw_priority : READ_ONCE(sk->sk_priority);
1007	transmit_time = tcp_transmit_time(sk);
1008	ip_send_unicast_reply(ctl_sk,
1009			      skb, &TCP_SKB_CB(skb)->header.h4.opt,
1010			      ip_hdr(skb)->saddr, ip_hdr(skb)->daddr,
1011			      &arg, arg.iov[0].iov_len,
1012			      transmit_time, txhash);
1013
1014	sock_net_set(ctl_sk, &init_net);
1015	__TCP_INC_STATS(net, TCP_MIB_OUTSEGS);
 
1016	local_bh_enable();
1017}
1018
1019static void tcp_v4_timewait_ack(struct sock *sk, struct sk_buff *skb)
1020{
1021	struct inet_timewait_sock *tw = inet_twsk(sk);
1022	struct tcp_timewait_sock *tcptw = tcp_twsk(sk);
1023	struct tcp_key key = {};
1024#ifdef CONFIG_TCP_AO
1025	struct tcp_ao_info *ao_info;
1026
1027	if (static_branch_unlikely(&tcp_ao_needed.key)) {
1028		/* FIXME: the segment to-be-acked is not verified yet */
1029		ao_info = rcu_dereference(tcptw->ao_info);
1030		if (ao_info) {
1031			const struct tcp_ao_hdr *aoh;
1032
1033			if (tcp_parse_auth_options(tcp_hdr(skb), NULL, &aoh)) {
1034				inet_twsk_put(tw);
1035				return;
1036			}
1037
1038			if (aoh)
1039				key.ao_key = tcp_ao_established_key(ao_info, aoh->rnext_keyid, -1);
 
1040		}
1041	}
1042	if (key.ao_key) {
1043		struct tcp_ao_key *rnext_key;
1044
1045		key.traffic_key = snd_other_key(key.ao_key);
1046		key.sne = READ_ONCE(ao_info->snd_sne);
1047		rnext_key = READ_ONCE(ao_info->rnext_key);
1048		key.rcv_next = rnext_key->rcvid;
1049		key.type = TCP_KEY_AO;
1050#else
1051	if (0) {
1052#endif
1053#ifdef CONFIG_TCP_MD5SIG
1054	} else if (static_branch_unlikely(&tcp_md5_needed.key)) {
1055		key.md5_key = tcp_twsk_md5_key(tcptw);
1056		if (key.md5_key)
1057			key.type = TCP_KEY_MD5;
1058#endif
1059	}
1060
1061	tcp_v4_send_ack(sk, skb,
1062			tcptw->tw_snd_nxt, tcptw->tw_rcv_nxt,
1063			tcptw->tw_rcv_wnd >> tw->tw_rcv_wscale,
1064			tcp_tw_tsval(tcptw),
1065			tcptw->tw_ts_recent,
1066			tw->tw_bound_dev_if, &key,
1067			tw->tw_transparent ? IP_REPLY_ARG_NOSRCCHECK : 0,
1068			tw->tw_tos,
1069			tw->tw_txhash);
1070
1071	inet_twsk_put(tw);
1072}
1073
1074static void tcp_v4_reqsk_send_ack(const struct sock *sk, struct sk_buff *skb,
1075				  struct request_sock *req)
1076{
1077	struct tcp_key key = {};
1078
1079	/* sk->sk_state == TCP_LISTEN -> for regular TCP_SYN_RECV
1080	 * sk->sk_state == TCP_SYN_RECV -> for Fast Open.
1081	 */
1082	u32 seq = (sk->sk_state == TCP_LISTEN) ? tcp_rsk(req)->snt_isn + 1 :
1083					     tcp_sk(sk)->snd_nxt;
1084
1085#ifdef CONFIG_TCP_AO
1086	if (static_branch_unlikely(&tcp_ao_needed.key) &&
1087	    tcp_rsk_used_ao(req)) {
1088		const union tcp_md5_addr *addr;
1089		const struct tcp_ao_hdr *aoh;
1090		int l3index;
1091
1092		/* Invalid TCP option size or twice included auth */
1093		if (tcp_parse_auth_options(tcp_hdr(skb), NULL, &aoh))
1094			return;
1095		if (!aoh)
1096			return;
1097
1098		addr = (union tcp_md5_addr *)&ip_hdr(skb)->saddr;
1099		l3index = tcp_v4_sdif(skb) ? inet_iif(skb) : 0;
1100		key.ao_key = tcp_ao_do_lookup(sk, l3index, addr, AF_INET,
1101					      aoh->rnext_keyid, -1);
1102		if (unlikely(!key.ao_key)) {
1103			/* Send ACK with any matching MKT for the peer */
1104			key.ao_key = tcp_ao_do_lookup(sk, l3index, addr, AF_INET, -1, -1);
1105			/* Matching key disappeared (user removed the key?)
1106			 * let the handshake timeout.
1107			 */
1108			if (!key.ao_key) {
1109				net_info_ratelimited("TCP-AO key for (%pI4, %d)->(%pI4, %d) suddenly disappeared, won't ACK new connection\n",
1110						     addr,
1111						     ntohs(tcp_hdr(skb)->source),
1112						     &ip_hdr(skb)->daddr,
1113						     ntohs(tcp_hdr(skb)->dest));
1114				return;
1115			}
1116		}
1117		key.traffic_key = kmalloc(tcp_ao_digest_size(key.ao_key), GFP_ATOMIC);
1118		if (!key.traffic_key)
1119			return;
1120
1121		key.type = TCP_KEY_AO;
1122		key.rcv_next = aoh->keyid;
1123		tcp_v4_ao_calc_key_rsk(key.ao_key, key.traffic_key, req);
1124#else
1125	if (0) {
1126#endif
1127#ifdef CONFIG_TCP_MD5SIG
1128	} else if (static_branch_unlikely(&tcp_md5_needed.key)) {
1129		const union tcp_md5_addr *addr;
1130		int l3index;
1131
1132		addr = (union tcp_md5_addr *)&ip_hdr(skb)->saddr;
1133		l3index = tcp_v4_sdif(skb) ? inet_iif(skb) : 0;
1134		key.md5_key = tcp_md5_do_lookup(sk, l3index, addr, AF_INET);
1135		if (key.md5_key)
1136			key.type = TCP_KEY_MD5;
1137#endif
1138	}
1139
1140	/* RFC 7323 2.3
1141	 * The window field (SEG.WND) of every outgoing segment, with the
1142	 * exception of <SYN> segments, MUST be right-shifted by
1143	 * Rcv.Wind.Shift bits:
1144	 */
1145	tcp_v4_send_ack(sk, skb, seq,
1146			tcp_rsk(req)->rcv_nxt,
1147			req->rsk_rcv_wnd >> inet_rsk(req)->rcv_wscale,
1148			tcp_rsk_tsval(tcp_rsk(req)),
1149			READ_ONCE(req->ts_recent),
1150			0, &key,
1151			inet_rsk(req)->no_srccheck ? IP_REPLY_ARG_NOSRCCHECK : 0,
1152			ip_hdr(skb)->tos,
1153			READ_ONCE(tcp_rsk(req)->txhash));
1154	if (tcp_key_is_ao(&key))
1155		kfree(key.traffic_key);
1156}
1157
1158/*
1159 *	Send a SYN-ACK after having received a SYN.
1160 *	This still operates on a request_sock only, not on a big
1161 *	socket.
1162 */
1163static int tcp_v4_send_synack(const struct sock *sk, struct dst_entry *dst,
1164			      struct flowi *fl,
1165			      struct request_sock *req,
1166			      struct tcp_fastopen_cookie *foc,
1167			      enum tcp_synack_type synack_type,
1168			      struct sk_buff *syn_skb)
1169{
1170	const struct inet_request_sock *ireq = inet_rsk(req);
1171	struct flowi4 fl4;
1172	int err = -1;
1173	struct sk_buff *skb;
1174	u8 tos;
1175
1176	/* First, grab a route. */
1177	if (!dst && (dst = inet_csk_route_req(sk, &fl4, req)) == NULL)
1178		return -1;
1179
1180	skb = tcp_make_synack(sk, dst, req, foc, synack_type, syn_skb);
1181
1182	if (skb) {
1183		__tcp_v4_send_check(skb, ireq->ir_loc_addr, ireq->ir_rmt_addr);
1184
1185		tos = READ_ONCE(inet_sk(sk)->tos);
1186
1187		if (READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_reflect_tos))
1188			tos = (tcp_rsk(req)->syn_tos & ~INET_ECN_MASK) |
1189			      (tos & INET_ECN_MASK);
1190
1191		if (!INET_ECN_is_capable(tos) &&
1192		    tcp_bpf_ca_needs_ecn((struct sock *)req))
1193			tos |= INET_ECN_ECT_0;
1194
1195		rcu_read_lock();
1196		err = ip_build_and_send_pkt(skb, sk, ireq->ir_loc_addr,
1197					    ireq->ir_rmt_addr,
1198					    rcu_dereference(ireq->ireq_opt),
1199					    tos);
1200		rcu_read_unlock();
1201		err = net_xmit_eval(err);
1202	}
1203
1204	return err;
1205}
1206
1207/*
1208 *	IPv4 request_sock destructor.
1209 */
1210static void tcp_v4_reqsk_destructor(struct request_sock *req)
1211{
1212	kfree(rcu_dereference_protected(inet_rsk(req)->ireq_opt, 1));
1213}
1214
1215#ifdef CONFIG_TCP_MD5SIG
1216/*
1217 * RFC2385 MD5 checksumming requires a mapping of
1218 * IP address->MD5 Key.
1219 * We need to maintain these in the sk structure.
1220 */
1221
1222DEFINE_STATIC_KEY_DEFERRED_FALSE(tcp_md5_needed, HZ);
1223EXPORT_SYMBOL(tcp_md5_needed);
1224
1225static bool better_md5_match(struct tcp_md5sig_key *old, struct tcp_md5sig_key *new)
1226{
1227	if (!old)
1228		return true;
1229
1230	/* l3index always overrides non-l3index */
1231	if (old->l3index && new->l3index == 0)
1232		return false;
1233	if (old->l3index == 0 && new->l3index)
1234		return true;
1235
1236	return old->prefixlen < new->prefixlen;
1237}
1238
1239/* Find the Key structure for an address.  */
1240struct tcp_md5sig_key *__tcp_md5_do_lookup(const struct sock *sk, int l3index,
1241					   const union tcp_md5_addr *addr,
1242					   int family, bool any_l3index)
1243{
1244	const struct tcp_sock *tp = tcp_sk(sk);
1245	struct tcp_md5sig_key *key;
1246	const struct tcp_md5sig_info *md5sig;
1247	__be32 mask;
1248	struct tcp_md5sig_key *best_match = NULL;
1249	bool match;
1250
1251	/* caller either holds rcu_read_lock() or socket lock */
1252	md5sig = rcu_dereference_check(tp->md5sig_info,
1253				       lockdep_sock_is_held(sk));
1254	if (!md5sig)
1255		return NULL;
1256
1257	hlist_for_each_entry_rcu(key, &md5sig->head, node,
1258				 lockdep_sock_is_held(sk)) {
1259		if (key->family != family)
1260			continue;
1261		if (!any_l3index && key->flags & TCP_MD5SIG_FLAG_IFINDEX &&
1262		    key->l3index != l3index)
1263			continue;
1264		if (family == AF_INET) {
1265			mask = inet_make_mask(key->prefixlen);
1266			match = (key->addr.a4.s_addr & mask) ==
1267				(addr->a4.s_addr & mask);
1268#if IS_ENABLED(CONFIG_IPV6)
1269		} else if (family == AF_INET6) {
1270			match = ipv6_prefix_equal(&key->addr.a6, &addr->a6,
1271						  key->prefixlen);
1272#endif
1273		} else {
1274			match = false;
1275		}
1276
1277		if (match && better_md5_match(best_match, key))
1278			best_match = key;
1279	}
1280	return best_match;
1281}
1282EXPORT_SYMBOL(__tcp_md5_do_lookup);
1283
1284static struct tcp_md5sig_key *tcp_md5_do_lookup_exact(const struct sock *sk,
1285						      const union tcp_md5_addr *addr,
1286						      int family, u8 prefixlen,
1287						      int l3index, u8 flags)
1288{
1289	const struct tcp_sock *tp = tcp_sk(sk);
1290	struct tcp_md5sig_key *key;
1291	unsigned int size = sizeof(struct in_addr);
1292	const struct tcp_md5sig_info *md5sig;
1293
1294	/* caller either holds rcu_read_lock() or socket lock */
1295	md5sig = rcu_dereference_check(tp->md5sig_info,
1296				       lockdep_sock_is_held(sk));
1297	if (!md5sig)
1298		return NULL;
1299#if IS_ENABLED(CONFIG_IPV6)
1300	if (family == AF_INET6)
1301		size = sizeof(struct in6_addr);
1302#endif
1303	hlist_for_each_entry_rcu(key, &md5sig->head, node,
1304				 lockdep_sock_is_held(sk)) {
1305		if (key->family != family)
1306			continue;
1307		if ((key->flags & TCP_MD5SIG_FLAG_IFINDEX) != (flags & TCP_MD5SIG_FLAG_IFINDEX))
1308			continue;
1309		if (key->l3index != l3index)
1310			continue;
1311		if (!memcmp(&key->addr, addr, size) &&
1312		    key->prefixlen == prefixlen)
1313			return key;
1314	}
1315	return NULL;
1316}
1317
1318struct tcp_md5sig_key *tcp_v4_md5_lookup(const struct sock *sk,
1319					 const struct sock *addr_sk)
1320{
1321	const union tcp_md5_addr *addr;
1322	int l3index;
1323
1324	l3index = l3mdev_master_ifindex_by_index(sock_net(sk),
1325						 addr_sk->sk_bound_dev_if);
1326	addr = (const union tcp_md5_addr *)&addr_sk->sk_daddr;
1327	return tcp_md5_do_lookup(sk, l3index, addr, AF_INET);
1328}
1329EXPORT_SYMBOL(tcp_v4_md5_lookup);
1330
1331static int tcp_md5sig_info_add(struct sock *sk, gfp_t gfp)
1332{
1333	struct tcp_sock *tp = tcp_sk(sk);
1334	struct tcp_md5sig_info *md5sig;
1335
1336	md5sig = kmalloc(sizeof(*md5sig), gfp);
1337	if (!md5sig)
1338		return -ENOMEM;
1339
1340	sk_gso_disable(sk);
1341	INIT_HLIST_HEAD(&md5sig->head);
1342	rcu_assign_pointer(tp->md5sig_info, md5sig);
1343	return 0;
1344}
1345
1346/* This can be called on a newly created socket, from other files */
1347static int __tcp_md5_do_add(struct sock *sk, const union tcp_md5_addr *addr,
1348			    int family, u8 prefixlen, int l3index, u8 flags,
1349			    const u8 *newkey, u8 newkeylen, gfp_t gfp)
1350{
1351	/* Add Key to the list */
1352	struct tcp_md5sig_key *key;
1353	struct tcp_sock *tp = tcp_sk(sk);
1354	struct tcp_md5sig_info *md5sig;
1355
1356	key = tcp_md5_do_lookup_exact(sk, addr, family, prefixlen, l3index, flags);
1357	if (key) {
1358		/* Pre-existing entry - just update that one.
1359		 * Note that the key might be used concurrently.
1360		 * data_race() is telling kcsan that we do not care of
1361		 * key mismatches, since changing MD5 key on live flows
1362		 * can lead to packet drops.
1363		 */
1364		data_race(memcpy(key->key, newkey, newkeylen));
1365
1366		/* Pairs with READ_ONCE() in tcp_md5_hash_key().
1367		 * Also note that a reader could catch new key->keylen value
1368		 * but old key->key[], this is the reason we use __GFP_ZERO
1369		 * at sock_kmalloc() time below these lines.
1370		 */
1371		WRITE_ONCE(key->keylen, newkeylen);
1372
1373		return 0;
1374	}
1375
1376	md5sig = rcu_dereference_protected(tp->md5sig_info,
1377					   lockdep_sock_is_held(sk));
1378
1379	key = sock_kmalloc(sk, sizeof(*key), gfp | __GFP_ZERO);
1380	if (!key)
1381		return -ENOMEM;
1382
1383	memcpy(key->key, newkey, newkeylen);
1384	key->keylen = newkeylen;
1385	key->family = family;
1386	key->prefixlen = prefixlen;
1387	key->l3index = l3index;
1388	key->flags = flags;
1389	memcpy(&key->addr, addr,
1390	       (IS_ENABLED(CONFIG_IPV6) && family == AF_INET6) ? sizeof(struct in6_addr) :
1391								 sizeof(struct in_addr));
1392	hlist_add_head_rcu(&key->node, &md5sig->head);
1393	return 0;
1394}
1395
1396int tcp_md5_do_add(struct sock *sk, const union tcp_md5_addr *addr,
1397		   int family, u8 prefixlen, int l3index, u8 flags,
1398		   const u8 *newkey, u8 newkeylen)
1399{
1400	struct tcp_sock *tp = tcp_sk(sk);
1401
1402	if (!rcu_dereference_protected(tp->md5sig_info, lockdep_sock_is_held(sk))) {
1403		if (tcp_md5_alloc_sigpool())
1404			return -ENOMEM;
1405
1406		if (tcp_md5sig_info_add(sk, GFP_KERNEL)) {
1407			tcp_md5_release_sigpool();
1408			return -ENOMEM;
1409		}
1410
1411		if (!static_branch_inc(&tcp_md5_needed.key)) {
1412			struct tcp_md5sig_info *md5sig;
1413
1414			md5sig = rcu_dereference_protected(tp->md5sig_info, lockdep_sock_is_held(sk));
1415			rcu_assign_pointer(tp->md5sig_info, NULL);
1416			kfree_rcu(md5sig, rcu);
1417			tcp_md5_release_sigpool();
1418			return -EUSERS;
1419		}
1420	}
1421
1422	return __tcp_md5_do_add(sk, addr, family, prefixlen, l3index, flags,
1423				newkey, newkeylen, GFP_KERNEL);
1424}
1425EXPORT_SYMBOL(tcp_md5_do_add);
1426
1427int tcp_md5_key_copy(struct sock *sk, const union tcp_md5_addr *addr,
1428		     int family, u8 prefixlen, int l3index,
1429		     struct tcp_md5sig_key *key)
1430{
1431	struct tcp_sock *tp = tcp_sk(sk);
1432
1433	if (!rcu_dereference_protected(tp->md5sig_info, lockdep_sock_is_held(sk))) {
1434		tcp_md5_add_sigpool();
1435
1436		if (tcp_md5sig_info_add(sk, sk_gfp_mask(sk, GFP_ATOMIC))) {
1437			tcp_md5_release_sigpool();
1438			return -ENOMEM;
1439		}
1440
1441		if (!static_key_fast_inc_not_disabled(&tcp_md5_needed.key.key)) {
1442			struct tcp_md5sig_info *md5sig;
1443
1444			md5sig = rcu_dereference_protected(tp->md5sig_info, lockdep_sock_is_held(sk));
1445			net_warn_ratelimited("Too many TCP-MD5 keys in the system\n");
1446			rcu_assign_pointer(tp->md5sig_info, NULL);
1447			kfree_rcu(md5sig, rcu);
1448			tcp_md5_release_sigpool();
1449			return -EUSERS;
1450		}
1451	}
1452
1453	return __tcp_md5_do_add(sk, addr, family, prefixlen, l3index,
1454				key->flags, key->key, key->keylen,
1455				sk_gfp_mask(sk, GFP_ATOMIC));
1456}
1457EXPORT_SYMBOL(tcp_md5_key_copy);
1458
1459int tcp_md5_do_del(struct sock *sk, const union tcp_md5_addr *addr, int family,
1460		   u8 prefixlen, int l3index, u8 flags)
1461{
1462	struct tcp_md5sig_key *key;
1463
1464	key = tcp_md5_do_lookup_exact(sk, addr, family, prefixlen, l3index, flags);
1465	if (!key)
1466		return -ENOENT;
1467	hlist_del_rcu(&key->node);
1468	atomic_sub(sizeof(*key), &sk->sk_omem_alloc);
1469	kfree_rcu(key, rcu);
1470	return 0;
1471}
1472EXPORT_SYMBOL(tcp_md5_do_del);
1473
1474void tcp_clear_md5_list(struct sock *sk)
1475{
1476	struct tcp_sock *tp = tcp_sk(sk);
1477	struct tcp_md5sig_key *key;
1478	struct hlist_node *n;
1479	struct tcp_md5sig_info *md5sig;
1480
1481	md5sig = rcu_dereference_protected(tp->md5sig_info, 1);
1482
1483	hlist_for_each_entry_safe(key, n, &md5sig->head, node) {
1484		hlist_del_rcu(&key->node);
1485		atomic_sub(sizeof(*key), &sk->sk_omem_alloc);
1486		kfree_rcu(key, rcu);
1487	}
1488}
1489
1490static int tcp_v4_parse_md5_keys(struct sock *sk, int optname,
1491				 sockptr_t optval, int optlen)
1492{
1493	struct tcp_md5sig cmd;
1494	struct sockaddr_in *sin = (struct sockaddr_in *)&cmd.tcpm_addr;
1495	const union tcp_md5_addr *addr;
1496	u8 prefixlen = 32;
1497	int l3index = 0;
1498	bool l3flag;
1499	u8 flags;
1500
1501	if (optlen < sizeof(cmd))
1502		return -EINVAL;
1503
1504	if (copy_from_sockptr(&cmd, optval, sizeof(cmd)))
1505		return -EFAULT;
1506
1507	if (sin->sin_family != AF_INET)
1508		return -EINVAL;
1509
1510	flags = cmd.tcpm_flags & TCP_MD5SIG_FLAG_IFINDEX;
1511	l3flag = cmd.tcpm_flags & TCP_MD5SIG_FLAG_IFINDEX;
1512
1513	if (optname == TCP_MD5SIG_EXT &&
1514	    cmd.tcpm_flags & TCP_MD5SIG_FLAG_PREFIX) {
1515		prefixlen = cmd.tcpm_prefixlen;
1516		if (prefixlen > 32)
1517			return -EINVAL;
1518	}
1519
1520	if (optname == TCP_MD5SIG_EXT && cmd.tcpm_ifindex &&
1521	    cmd.tcpm_flags & TCP_MD5SIG_FLAG_IFINDEX) {
1522		struct net_device *dev;
1523
1524		rcu_read_lock();
1525		dev = dev_get_by_index_rcu(sock_net(sk), cmd.tcpm_ifindex);
1526		if (dev && netif_is_l3_master(dev))
1527			l3index = dev->ifindex;
1528
1529		rcu_read_unlock();
1530
1531		/* ok to reference set/not set outside of rcu;
1532		 * right now device MUST be an L3 master
1533		 */
1534		if (!dev || !l3index)
1535			return -EINVAL;
1536	}
1537
1538	addr = (union tcp_md5_addr *)&sin->sin_addr.s_addr;
1539
1540	if (!cmd.tcpm_keylen)
1541		return tcp_md5_do_del(sk, addr, AF_INET, prefixlen, l3index, flags);
1542
1543	if (cmd.tcpm_keylen > TCP_MD5SIG_MAXKEYLEN)
1544		return -EINVAL;
1545
1546	/* Don't allow keys for peers that have a matching TCP-AO key.
1547	 * See the comment in tcp_ao_add_cmd()
1548	 */
1549	if (tcp_ao_required(sk, addr, AF_INET, l3flag ? l3index : -1, false))
1550		return -EKEYREJECTED;
1551
1552	return tcp_md5_do_add(sk, addr, AF_INET, prefixlen, l3index, flags,
1553			      cmd.tcpm_key, cmd.tcpm_keylen);
1554}
1555
1556static int tcp_v4_md5_hash_headers(struct tcp_sigpool *hp,
1557				   __be32 daddr, __be32 saddr,
1558				   const struct tcphdr *th, int nbytes)
1559{
1560	struct tcp4_pseudohdr *bp;
1561	struct scatterlist sg;
1562	struct tcphdr *_th;
1563
1564	bp = hp->scratch;
1565	bp->saddr = saddr;
1566	bp->daddr = daddr;
1567	bp->pad = 0;
1568	bp->protocol = IPPROTO_TCP;
1569	bp->len = cpu_to_be16(nbytes);
1570
1571	_th = (struct tcphdr *)(bp + 1);
1572	memcpy(_th, th, sizeof(*th));
1573	_th->check = 0;
1574
1575	sg_init_one(&sg, bp, sizeof(*bp) + sizeof(*th));
1576	ahash_request_set_crypt(hp->req, &sg, NULL,
1577				sizeof(*bp) + sizeof(*th));
1578	return crypto_ahash_update(hp->req);
1579}
1580
1581static int tcp_v4_md5_hash_hdr(char *md5_hash, const struct tcp_md5sig_key *key,
1582			       __be32 daddr, __be32 saddr, const struct tcphdr *th)
1583{
1584	struct tcp_sigpool hp;
1585
1586	if (tcp_sigpool_start(tcp_md5_sigpool_id, &hp))
1587		goto clear_hash_nostart;
1588
1589	if (crypto_ahash_init(hp.req))
1590		goto clear_hash;
1591	if (tcp_v4_md5_hash_headers(&hp, daddr, saddr, th, th->doff << 2))
1592		goto clear_hash;
1593	if (tcp_md5_hash_key(&hp, key))
1594		goto clear_hash;
1595	ahash_request_set_crypt(hp.req, NULL, md5_hash, 0);
1596	if (crypto_ahash_final(hp.req))
1597		goto clear_hash;
1598
1599	tcp_sigpool_end(&hp);
1600	return 0;
1601
1602clear_hash:
1603	tcp_sigpool_end(&hp);
1604clear_hash_nostart:
1605	memset(md5_hash, 0, 16);
1606	return 1;
1607}
1608
1609int tcp_v4_md5_hash_skb(char *md5_hash, const struct tcp_md5sig_key *key,
1610			const struct sock *sk,
1611			const struct sk_buff *skb)
1612{
1613	const struct tcphdr *th = tcp_hdr(skb);
1614	struct tcp_sigpool hp;
1615	__be32 saddr, daddr;
1616
1617	if (sk) { /* valid for establish/request sockets */
1618		saddr = sk->sk_rcv_saddr;
1619		daddr = sk->sk_daddr;
1620	} else {
1621		const struct iphdr *iph = ip_hdr(skb);
1622		saddr = iph->saddr;
1623		daddr = iph->daddr;
1624	}
1625
1626	if (tcp_sigpool_start(tcp_md5_sigpool_id, &hp))
1627		goto clear_hash_nostart;
1628
1629	if (crypto_ahash_init(hp.req))
1630		goto clear_hash;
1631
1632	if (tcp_v4_md5_hash_headers(&hp, daddr, saddr, th, skb->len))
1633		goto clear_hash;
1634	if (tcp_sigpool_hash_skb_data(&hp, skb, th->doff << 2))
1635		goto clear_hash;
1636	if (tcp_md5_hash_key(&hp, key))
1637		goto clear_hash;
1638	ahash_request_set_crypt(hp.req, NULL, md5_hash, 0);
1639	if (crypto_ahash_final(hp.req))
1640		goto clear_hash;
1641
1642	tcp_sigpool_end(&hp);
1643	return 0;
1644
1645clear_hash:
1646	tcp_sigpool_end(&hp);
1647clear_hash_nostart:
1648	memset(md5_hash, 0, 16);
1649	return 1;
1650}
1651EXPORT_SYMBOL(tcp_v4_md5_hash_skb);
1652
1653#endif
1654
1655static void tcp_v4_init_req(struct request_sock *req,
1656			    const struct sock *sk_listener,
1657			    struct sk_buff *skb)
1658{
1659	struct inet_request_sock *ireq = inet_rsk(req);
1660	struct net *net = sock_net(sk_listener);
1661
1662	sk_rcv_saddr_set(req_to_sk(req), ip_hdr(skb)->daddr);
1663	sk_daddr_set(req_to_sk(req), ip_hdr(skb)->saddr);
1664	RCU_INIT_POINTER(ireq->ireq_opt, tcp_v4_save_options(net, skb));
1665}
1666
1667static struct dst_entry *tcp_v4_route_req(const struct sock *sk,
1668					  struct sk_buff *skb,
1669					  struct flowi *fl,
1670					  struct request_sock *req)
 
1671{
1672	tcp_v4_init_req(req, sk, skb);
1673
1674	if (security_inet_conn_request(sk, skb, req))
1675		return NULL;
1676
1677	return inet_csk_route_req(sk, &fl->u.ip4, req);
1678}
1679
1680struct request_sock_ops tcp_request_sock_ops __read_mostly = {
1681	.family		=	PF_INET,
1682	.obj_size	=	sizeof(struct tcp_request_sock),
1683	.rtx_syn_ack	=	tcp_rtx_synack,
1684	.send_ack	=	tcp_v4_reqsk_send_ack,
1685	.destructor	=	tcp_v4_reqsk_destructor,
1686	.send_reset	=	tcp_v4_send_reset,
1687	.syn_ack_timeout =	tcp_syn_ack_timeout,
1688};
1689
1690const struct tcp_request_sock_ops tcp_request_sock_ipv4_ops = {
1691	.mss_clamp	=	TCP_MSS_DEFAULT,
1692#ifdef CONFIG_TCP_MD5SIG
1693	.req_md5_lookup	=	tcp_v4_md5_lookup,
1694	.calc_md5_hash	=	tcp_v4_md5_hash_skb,
1695#endif
1696#ifdef CONFIG_TCP_AO
1697	.ao_lookup	=	tcp_v4_ao_lookup_rsk,
1698	.ao_calc_key	=	tcp_v4_ao_calc_key_rsk,
1699	.ao_synack_hash	=	tcp_v4_ao_synack_hash,
1700#endif
1701#ifdef CONFIG_SYN_COOKIES
1702	.cookie_init_seq =	cookie_v4_init_sequence,
1703#endif
1704	.route_req	=	tcp_v4_route_req,
1705	.init_seq	=	tcp_v4_init_seq,
1706	.init_ts_off	=	tcp_v4_init_ts_off,
1707	.send_synack	=	tcp_v4_send_synack,
1708};
1709
1710int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb)
1711{
1712	/* Never answer to SYNs send to broadcast or multicast */
1713	if (skb_rtable(skb)->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))
1714		goto drop;
1715
1716	return tcp_conn_request(&tcp_request_sock_ops,
1717				&tcp_request_sock_ipv4_ops, sk, skb);
1718
1719drop:
1720	tcp_listendrop(sk);
1721	return 0;
1722}
1723EXPORT_SYMBOL(tcp_v4_conn_request);
1724
1725
1726/*
1727 * The three way handshake has completed - we got a valid synack -
1728 * now create the new socket.
1729 */
1730struct sock *tcp_v4_syn_recv_sock(const struct sock *sk, struct sk_buff *skb,
1731				  struct request_sock *req,
1732				  struct dst_entry *dst,
1733				  struct request_sock *req_unhash,
1734				  bool *own_req)
1735{
1736	struct inet_request_sock *ireq;
1737	bool found_dup_sk = false;
1738	struct inet_sock *newinet;
1739	struct tcp_sock *newtp;
1740	struct sock *newsk;
1741#ifdef CONFIG_TCP_MD5SIG
1742	const union tcp_md5_addr *addr;
1743	struct tcp_md5sig_key *key;
1744	int l3index;
1745#endif
1746	struct ip_options_rcu *inet_opt;
1747
1748	if (sk_acceptq_is_full(sk))
1749		goto exit_overflow;
1750
1751	newsk = tcp_create_openreq_child(sk, req, skb);
1752	if (!newsk)
1753		goto exit_nonewsk;
1754
1755	newsk->sk_gso_type = SKB_GSO_TCPV4;
1756	inet_sk_rx_dst_set(newsk, skb);
1757
1758	newtp		      = tcp_sk(newsk);
1759	newinet		      = inet_sk(newsk);
1760	ireq		      = inet_rsk(req);
1761	sk_daddr_set(newsk, ireq->ir_rmt_addr);
1762	sk_rcv_saddr_set(newsk, ireq->ir_loc_addr);
1763	newsk->sk_bound_dev_if = ireq->ir_iif;
1764	newinet->inet_saddr   = ireq->ir_loc_addr;
1765	inet_opt	      = rcu_dereference(ireq->ireq_opt);
1766	RCU_INIT_POINTER(newinet->inet_opt, inet_opt);
1767	newinet->mc_index     = inet_iif(skb);
1768	newinet->mc_ttl	      = ip_hdr(skb)->ttl;
1769	newinet->rcv_tos      = ip_hdr(skb)->tos;
1770	inet_csk(newsk)->icsk_ext_hdr_len = 0;
1771	if (inet_opt)
1772		inet_csk(newsk)->icsk_ext_hdr_len = inet_opt->opt.optlen;
1773	atomic_set(&newinet->inet_id, get_random_u16());
1774
1775	/* Set ToS of the new socket based upon the value of incoming SYN.
1776	 * ECT bits are set later in tcp_init_transfer().
1777	 */
1778	if (READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_reflect_tos))
1779		newinet->tos = tcp_rsk(req)->syn_tos & ~INET_ECN_MASK;
1780
1781	if (!dst) {
1782		dst = inet_csk_route_child_sock(sk, newsk, req);
1783		if (!dst)
1784			goto put_and_exit;
1785	} else {
1786		/* syncookie case : see end of cookie_v4_check() */
1787	}
1788	sk_setup_caps(newsk, dst);
1789
1790	tcp_ca_openreq_child(newsk, dst);
1791
1792	tcp_sync_mss(newsk, dst_mtu(dst));
1793	newtp->advmss = tcp_mss_clamp(tcp_sk(sk), dst_metric_advmss(dst));
1794
1795	tcp_initialize_rcv_mss(newsk);
1796
1797#ifdef CONFIG_TCP_MD5SIG
1798	l3index = l3mdev_master_ifindex_by_index(sock_net(sk), ireq->ir_iif);
1799	/* Copy over the MD5 key from the original socket */
1800	addr = (union tcp_md5_addr *)&newinet->inet_daddr;
1801	key = tcp_md5_do_lookup(sk, l3index, addr, AF_INET);
1802	if (key && !tcp_rsk_used_ao(req)) {
1803		if (tcp_md5_key_copy(newsk, addr, AF_INET, 32, l3index, key))
1804			goto put_and_exit;
1805		sk_gso_disable(newsk);
1806	}
1807#endif
1808#ifdef CONFIG_TCP_AO
1809	if (tcp_ao_copy_all_matching(sk, newsk, req, skb, AF_INET))
1810		goto put_and_exit; /* OOM, release back memory */
1811#endif
1812
1813	if (__inet_inherit_port(sk, newsk) < 0)
1814		goto put_and_exit;
1815	*own_req = inet_ehash_nolisten(newsk, req_to_sk(req_unhash),
1816				       &found_dup_sk);
1817	if (likely(*own_req)) {
1818		tcp_move_syn(newtp, req);
1819		ireq->ireq_opt = NULL;
1820	} else {
1821		newinet->inet_opt = NULL;
1822
1823		if (!req_unhash && found_dup_sk) {
1824			/* This code path should only be executed in the
1825			 * syncookie case only
1826			 */
1827			bh_unlock_sock(newsk);
1828			sock_put(newsk);
1829			newsk = NULL;
1830		}
1831	}
1832	return newsk;
1833
1834exit_overflow:
1835	NET_INC_STATS(sock_net(sk), LINUX_MIB_LISTENOVERFLOWS);
1836exit_nonewsk:
1837	dst_release(dst);
1838exit:
1839	tcp_listendrop(sk);
1840	return NULL;
1841put_and_exit:
1842	newinet->inet_opt = NULL;
1843	inet_csk_prepare_forced_close(newsk);
1844	tcp_done(newsk);
1845	goto exit;
1846}
1847EXPORT_SYMBOL(tcp_v4_syn_recv_sock);
1848
1849static struct sock *tcp_v4_cookie_check(struct sock *sk, struct sk_buff *skb)
1850{
1851#ifdef CONFIG_SYN_COOKIES
1852	const struct tcphdr *th = tcp_hdr(skb);
1853
1854	if (!th->syn)
1855		sk = cookie_v4_check(sk, skb);
1856#endif
1857	return sk;
1858}
1859
1860u16 tcp_v4_get_syncookie(struct sock *sk, struct iphdr *iph,
1861			 struct tcphdr *th, u32 *cookie)
1862{
1863	u16 mss = 0;
1864#ifdef CONFIG_SYN_COOKIES
1865	mss = tcp_get_syncookie_mss(&tcp_request_sock_ops,
1866				    &tcp_request_sock_ipv4_ops, sk, th);
1867	if (mss) {
1868		*cookie = __cookie_v4_init_sequence(iph, th, &mss);
1869		tcp_synq_overflow(sk);
1870	}
1871#endif
1872	return mss;
1873}
1874
1875INDIRECT_CALLABLE_DECLARE(struct dst_entry *ipv4_dst_check(struct dst_entry *,
1876							   u32));
1877/* The socket must have it's spinlock held when we get
1878 * here, unless it is a TCP_LISTEN socket.
1879 *
1880 * We have a potential double-lock case here, so even when
1881 * doing backlog processing we use the BH locking scheme.
1882 * This is because we cannot sleep with the original spinlock
1883 * held.
1884 */
1885int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb)
1886{
1887	enum skb_drop_reason reason;
1888	struct sock *rsk;
1889
1890	if (sk->sk_state == TCP_ESTABLISHED) { /* Fast path */
1891		struct dst_entry *dst;
1892
1893		dst = rcu_dereference_protected(sk->sk_rx_dst,
1894						lockdep_sock_is_held(sk));
1895
1896		sock_rps_save_rxhash(sk, skb);
1897		sk_mark_napi_id(sk, skb);
1898		if (dst) {
1899			if (sk->sk_rx_dst_ifindex != skb->skb_iif ||
1900			    !INDIRECT_CALL_1(dst->ops->check, ipv4_dst_check,
1901					     dst, 0)) {
1902				RCU_INIT_POINTER(sk->sk_rx_dst, NULL);
1903				dst_release(dst);
1904			}
1905		}
1906		tcp_rcv_established(sk, skb);
1907		return 0;
1908	}
1909
1910	reason = SKB_DROP_REASON_NOT_SPECIFIED;
1911	if (tcp_checksum_complete(skb))
1912		goto csum_err;
1913
1914	if (sk->sk_state == TCP_LISTEN) {
1915		struct sock *nsk = tcp_v4_cookie_check(sk, skb);
1916
1917		if (!nsk)
1918			goto discard;
1919		if (nsk != sk) {
1920			if (tcp_child_process(sk, nsk, skb)) {
 
1921				rsk = nsk;
1922				goto reset;
1923			}
1924			return 0;
1925		}
1926	} else
1927		sock_rps_save_rxhash(sk, skb);
1928
1929	if (tcp_rcv_state_process(sk, skb)) {
 
1930		rsk = sk;
1931		goto reset;
1932	}
1933	return 0;
1934
1935reset:
1936	tcp_v4_send_reset(rsk, skb);
1937discard:
1938	kfree_skb_reason(skb, reason);
1939	/* Be careful here. If this function gets more complicated and
1940	 * gcc suffers from register pressure on the x86, sk (in %ebx)
1941	 * might be destroyed here. This current version compiles correctly,
1942	 * but you have been warned.
1943	 */
1944	return 0;
1945
1946csum_err:
1947	reason = SKB_DROP_REASON_TCP_CSUM;
1948	trace_tcp_bad_csum(skb);
1949	TCP_INC_STATS(sock_net(sk), TCP_MIB_CSUMERRORS);
1950	TCP_INC_STATS(sock_net(sk), TCP_MIB_INERRS);
1951	goto discard;
1952}
1953EXPORT_SYMBOL(tcp_v4_do_rcv);
1954
1955int tcp_v4_early_demux(struct sk_buff *skb)
1956{
1957	struct net *net = dev_net(skb->dev);
1958	const struct iphdr *iph;
1959	const struct tcphdr *th;
1960	struct sock *sk;
1961
1962	if (skb->pkt_type != PACKET_HOST)
1963		return 0;
1964
1965	if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct tcphdr)))
1966		return 0;
1967
1968	iph = ip_hdr(skb);
1969	th = tcp_hdr(skb);
1970
1971	if (th->doff < sizeof(struct tcphdr) / 4)
1972		return 0;
1973
1974	sk = __inet_lookup_established(net, net->ipv4.tcp_death_row.hashinfo,
1975				       iph->saddr, th->source,
1976				       iph->daddr, ntohs(th->dest),
1977				       skb->skb_iif, inet_sdif(skb));
1978	if (sk) {
1979		skb->sk = sk;
1980		skb->destructor = sock_edemux;
1981		if (sk_fullsock(sk)) {
1982			struct dst_entry *dst = rcu_dereference(sk->sk_rx_dst);
1983
1984			if (dst)
1985				dst = dst_check(dst, 0);
1986			if (dst &&
1987			    sk->sk_rx_dst_ifindex == skb->skb_iif)
1988				skb_dst_set_noref(skb, dst);
1989		}
1990	}
1991	return 0;
1992}
1993
1994bool tcp_add_backlog(struct sock *sk, struct sk_buff *skb,
1995		     enum skb_drop_reason *reason)
1996{
1997	u32 limit, tail_gso_size, tail_gso_segs;
1998	struct skb_shared_info *shinfo;
1999	const struct tcphdr *th;
2000	struct tcphdr *thtail;
2001	struct sk_buff *tail;
2002	unsigned int hdrlen;
2003	bool fragstolen;
2004	u32 gso_segs;
2005	u32 gso_size;
 
2006	int delta;
2007
2008	/* In case all data was pulled from skb frags (in __pskb_pull_tail()),
2009	 * we can fix skb->truesize to its real value to avoid future drops.
2010	 * This is valid because skb is not yet charged to the socket.
2011	 * It has been noticed pure SACK packets were sometimes dropped
2012	 * (if cooked by drivers without copybreak feature).
2013	 */
2014	skb_condense(skb);
2015
2016	skb_dst_drop(skb);
2017
2018	if (unlikely(tcp_checksum_complete(skb))) {
2019		bh_unlock_sock(sk);
2020		trace_tcp_bad_csum(skb);
2021		*reason = SKB_DROP_REASON_TCP_CSUM;
2022		__TCP_INC_STATS(sock_net(sk), TCP_MIB_CSUMERRORS);
2023		__TCP_INC_STATS(sock_net(sk), TCP_MIB_INERRS);
2024		return true;
2025	}
2026
2027	/* Attempt coalescing to last skb in backlog, even if we are
2028	 * above the limits.
2029	 * This is okay because skb capacity is limited to MAX_SKB_FRAGS.
2030	 */
2031	th = (const struct tcphdr *)skb->data;
2032	hdrlen = th->doff * 4;
2033
2034	tail = sk->sk_backlog.tail;
2035	if (!tail)
2036		goto no_coalesce;
2037	thtail = (struct tcphdr *)tail->data;
2038
2039	if (TCP_SKB_CB(tail)->end_seq != TCP_SKB_CB(skb)->seq ||
2040	    TCP_SKB_CB(tail)->ip_dsfield != TCP_SKB_CB(skb)->ip_dsfield ||
2041	    ((TCP_SKB_CB(tail)->tcp_flags |
2042	      TCP_SKB_CB(skb)->tcp_flags) & (TCPHDR_SYN | TCPHDR_RST | TCPHDR_URG)) ||
2043	    !((TCP_SKB_CB(tail)->tcp_flags &
2044	      TCP_SKB_CB(skb)->tcp_flags) & TCPHDR_ACK) ||
2045	    ((TCP_SKB_CB(tail)->tcp_flags ^
2046	      TCP_SKB_CB(skb)->tcp_flags) & (TCPHDR_ECE | TCPHDR_CWR)) ||
2047#ifdef CONFIG_TLS_DEVICE
2048	    tail->decrypted != skb->decrypted ||
2049#endif
2050	    !mptcp_skb_can_collapse(tail, skb) ||
2051	    thtail->doff != th->doff ||
2052	    memcmp(thtail + 1, th + 1, hdrlen - sizeof(*th)))
2053		goto no_coalesce;
2054
2055	__skb_pull(skb, hdrlen);
2056
2057	shinfo = skb_shinfo(skb);
2058	gso_size = shinfo->gso_size ?: skb->len;
2059	gso_segs = shinfo->gso_segs ?: 1;
2060
2061	shinfo = skb_shinfo(tail);
2062	tail_gso_size = shinfo->gso_size ?: (tail->len - hdrlen);
2063	tail_gso_segs = shinfo->gso_segs ?: 1;
2064
2065	if (skb_try_coalesce(tail, skb, &fragstolen, &delta)) {
2066		TCP_SKB_CB(tail)->end_seq = TCP_SKB_CB(skb)->end_seq;
2067
2068		if (likely(!before(TCP_SKB_CB(skb)->ack_seq, TCP_SKB_CB(tail)->ack_seq))) {
2069			TCP_SKB_CB(tail)->ack_seq = TCP_SKB_CB(skb)->ack_seq;
2070			thtail->window = th->window;
2071		}
2072
2073		/* We have to update both TCP_SKB_CB(tail)->tcp_flags and
2074		 * thtail->fin, so that the fast path in tcp_rcv_established()
2075		 * is not entered if we append a packet with a FIN.
2076		 * SYN, RST, URG are not present.
2077		 * ACK is set on both packets.
2078		 * PSH : we do not really care in TCP stack,
2079		 *       at least for 'GRO' packets.
2080		 */
2081		thtail->fin |= th->fin;
2082		TCP_SKB_CB(tail)->tcp_flags |= TCP_SKB_CB(skb)->tcp_flags;
2083
2084		if (TCP_SKB_CB(skb)->has_rxtstamp) {
2085			TCP_SKB_CB(tail)->has_rxtstamp = true;
2086			tail->tstamp = skb->tstamp;
2087			skb_hwtstamps(tail)->hwtstamp = skb_hwtstamps(skb)->hwtstamp;
2088		}
2089
2090		/* Not as strict as GRO. We only need to carry mss max value */
2091		shinfo->gso_size = max(gso_size, tail_gso_size);
2092		shinfo->gso_segs = min_t(u32, gso_segs + tail_gso_segs, 0xFFFF);
2093
2094		sk->sk_backlog.len += delta;
2095		__NET_INC_STATS(sock_net(sk),
2096				LINUX_MIB_TCPBACKLOGCOALESCE);
2097		kfree_skb_partial(skb, fragstolen);
2098		return false;
2099	}
2100	__skb_push(skb, hdrlen);
2101
2102no_coalesce:
2103	limit = (u32)READ_ONCE(sk->sk_rcvbuf) + (u32)(READ_ONCE(sk->sk_sndbuf) >> 1);
 
 
 
 
 
 
2104
2105	/* Only socket owner can try to collapse/prune rx queues
2106	 * to reduce memory overhead, so add a little headroom here.
2107	 * Few sockets backlog are possibly concurrently non empty.
2108	 */
2109	limit += 64 * 1024;
2110
 
 
2111	if (unlikely(sk_add_backlog(sk, skb, limit))) {
2112		bh_unlock_sock(sk);
2113		*reason = SKB_DROP_REASON_SOCKET_BACKLOG;
2114		__NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPBACKLOGDROP);
2115		return true;
2116	}
2117	return false;
2118}
2119EXPORT_SYMBOL(tcp_add_backlog);
2120
2121int tcp_filter(struct sock *sk, struct sk_buff *skb)
2122{
2123	struct tcphdr *th = (struct tcphdr *)skb->data;
2124
2125	return sk_filter_trim_cap(sk, skb, th->doff * 4);
2126}
2127EXPORT_SYMBOL(tcp_filter);
2128
2129static void tcp_v4_restore_cb(struct sk_buff *skb)
2130{
2131	memmove(IPCB(skb), &TCP_SKB_CB(skb)->header.h4,
2132		sizeof(struct inet_skb_parm));
2133}
2134
2135static void tcp_v4_fill_cb(struct sk_buff *skb, const struct iphdr *iph,
2136			   const struct tcphdr *th)
2137{
2138	/* This is tricky : We move IPCB at its correct location into TCP_SKB_CB()
2139	 * barrier() makes sure compiler wont play fool^Waliasing games.
2140	 */
2141	memmove(&TCP_SKB_CB(skb)->header.h4, IPCB(skb),
2142		sizeof(struct inet_skb_parm));
2143	barrier();
2144
2145	TCP_SKB_CB(skb)->seq = ntohl(th->seq);
2146	TCP_SKB_CB(skb)->end_seq = (TCP_SKB_CB(skb)->seq + th->syn + th->fin +
2147				    skb->len - th->doff * 4);
2148	TCP_SKB_CB(skb)->ack_seq = ntohl(th->ack_seq);
2149	TCP_SKB_CB(skb)->tcp_flags = tcp_flag_byte(th);
2150	TCP_SKB_CB(skb)->tcp_tw_isn = 0;
2151	TCP_SKB_CB(skb)->ip_dsfield = ipv4_get_dsfield(iph);
2152	TCP_SKB_CB(skb)->sacked	 = 0;
2153	TCP_SKB_CB(skb)->has_rxtstamp =
2154			skb->tstamp || skb_hwtstamps(skb)->hwtstamp;
2155}
2156
2157/*
2158 *	From tcp_input.c
2159 */
2160
2161int tcp_v4_rcv(struct sk_buff *skb)
2162{
2163	struct net *net = dev_net(skb->dev);
2164	enum skb_drop_reason drop_reason;
2165	int sdif = inet_sdif(skb);
2166	int dif = inet_iif(skb);
2167	const struct iphdr *iph;
2168	const struct tcphdr *th;
 
2169	bool refcounted;
2170	struct sock *sk;
2171	int ret;
 
2172
2173	drop_reason = SKB_DROP_REASON_NOT_SPECIFIED;
2174	if (skb->pkt_type != PACKET_HOST)
2175		goto discard_it;
2176
2177	/* Count it even if it's bad */
2178	__TCP_INC_STATS(net, TCP_MIB_INSEGS);
2179
2180	if (!pskb_may_pull(skb, sizeof(struct tcphdr)))
2181		goto discard_it;
2182
2183	th = (const struct tcphdr *)skb->data;
2184
2185	if (unlikely(th->doff < sizeof(struct tcphdr) / 4)) {
2186		drop_reason = SKB_DROP_REASON_PKT_TOO_SMALL;
2187		goto bad_packet;
2188	}
2189	if (!pskb_may_pull(skb, th->doff * 4))
2190		goto discard_it;
2191
2192	/* An explanation is required here, I think.
2193	 * Packet length and doff are validated by header prediction,
2194	 * provided case of th->doff==0 is eliminated.
2195	 * So, we defer the checks. */
2196
2197	if (skb_checksum_init(skb, IPPROTO_TCP, inet_compute_pseudo))
2198		goto csum_error;
2199
2200	th = (const struct tcphdr *)skb->data;
2201	iph = ip_hdr(skb);
2202lookup:
2203	sk = __inet_lookup_skb(net->ipv4.tcp_death_row.hashinfo,
2204			       skb, __tcp_hdrlen(th), th->source,
2205			       th->dest, sdif, &refcounted);
2206	if (!sk)
2207		goto no_tcp_socket;
2208
2209process:
2210	if (sk->sk_state == TCP_TIME_WAIT)
2211		goto do_time_wait;
2212
2213	if (sk->sk_state == TCP_NEW_SYN_RECV) {
2214		struct request_sock *req = inet_reqsk(sk);
2215		bool req_stolen = false;
2216		struct sock *nsk;
2217
2218		sk = req->rsk_listener;
2219		if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
2220			drop_reason = SKB_DROP_REASON_XFRM_POLICY;
2221		else
2222			drop_reason = tcp_inbound_hash(sk, req, skb,
2223						       &iph->saddr, &iph->daddr,
2224						       AF_INET, dif, sdif);
2225		if (unlikely(drop_reason)) {
2226			sk_drops_add(sk, skb);
2227			reqsk_put(req);
2228			goto discard_it;
2229		}
2230		if (tcp_checksum_complete(skb)) {
2231			reqsk_put(req);
2232			goto csum_error;
2233		}
2234		if (unlikely(sk->sk_state != TCP_LISTEN)) {
2235			nsk = reuseport_migrate_sock(sk, req_to_sk(req), skb);
2236			if (!nsk) {
2237				inet_csk_reqsk_queue_drop_and_put(sk, req);
2238				goto lookup;
2239			}
2240			sk = nsk;
2241			/* reuseport_migrate_sock() has already held one sk_refcnt
2242			 * before returning.
2243			 */
2244		} else {
2245			/* We own a reference on the listener, increase it again
2246			 * as we might lose it too soon.
2247			 */
2248			sock_hold(sk);
2249		}
2250		refcounted = true;
2251		nsk = NULL;
2252		if (!tcp_filter(sk, skb)) {
2253			th = (const struct tcphdr *)skb->data;
2254			iph = ip_hdr(skb);
2255			tcp_v4_fill_cb(skb, iph, th);
2256			nsk = tcp_check_req(sk, skb, req, false, &req_stolen);
2257		} else {
2258			drop_reason = SKB_DROP_REASON_SOCKET_FILTER;
2259		}
2260		if (!nsk) {
2261			reqsk_put(req);
2262			if (req_stolen) {
2263				/* Another cpu got exclusive access to req
2264				 * and created a full blown socket.
2265				 * Try to feed this packet to this socket
2266				 * instead of discarding it.
2267				 */
2268				tcp_v4_restore_cb(skb);
2269				sock_put(sk);
2270				goto lookup;
2271			}
2272			goto discard_and_relse;
2273		}
2274		nf_reset_ct(skb);
2275		if (nsk == sk) {
2276			reqsk_put(req);
2277			tcp_v4_restore_cb(skb);
2278		} else if (tcp_child_process(sk, nsk, skb)) {
2279			tcp_v4_send_reset(nsk, skb);
2280			goto discard_and_relse;
2281		} else {
 
 
 
 
 
 
 
 
2282			sock_put(sk);
2283			return 0;
2284		}
2285	}
2286
 
2287	if (static_branch_unlikely(&ip4_min_ttl)) {
2288		/* min_ttl can be changed concurrently from do_ip_setsockopt() */
2289		if (unlikely(iph->ttl < READ_ONCE(inet_sk(sk)->min_ttl))) {
2290			__NET_INC_STATS(net, LINUX_MIB_TCPMINTTLDROP);
2291			drop_reason = SKB_DROP_REASON_TCP_MINTTL;
2292			goto discard_and_relse;
2293		}
2294	}
2295
2296	if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb)) {
2297		drop_reason = SKB_DROP_REASON_XFRM_POLICY;
2298		goto discard_and_relse;
2299	}
2300
2301	drop_reason = tcp_inbound_hash(sk, NULL, skb, &iph->saddr, &iph->daddr,
2302				       AF_INET, dif, sdif);
2303	if (drop_reason)
2304		goto discard_and_relse;
2305
2306	nf_reset_ct(skb);
2307
2308	if (tcp_filter(sk, skb)) {
2309		drop_reason = SKB_DROP_REASON_SOCKET_FILTER;
2310		goto discard_and_relse;
2311	}
2312	th = (const struct tcphdr *)skb->data;
2313	iph = ip_hdr(skb);
2314	tcp_v4_fill_cb(skb, iph, th);
2315
2316	skb->dev = NULL;
2317
2318	if (sk->sk_state == TCP_LISTEN) {
2319		ret = tcp_v4_do_rcv(sk, skb);
2320		goto put_and_return;
2321	}
2322
2323	sk_incoming_cpu_update(sk);
2324
2325	bh_lock_sock_nested(sk);
2326	tcp_segs_in(tcp_sk(sk), skb);
2327	ret = 0;
2328	if (!sock_owned_by_user(sk)) {
2329		ret = tcp_v4_do_rcv(sk, skb);
2330	} else {
2331		if (tcp_add_backlog(sk, skb, &drop_reason))
2332			goto discard_and_relse;
2333	}
2334	bh_unlock_sock(sk);
2335
2336put_and_return:
2337	if (refcounted)
2338		sock_put(sk);
2339
2340	return ret;
2341
2342no_tcp_socket:
2343	drop_reason = SKB_DROP_REASON_NO_SOCKET;
2344	if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
2345		goto discard_it;
2346
2347	tcp_v4_fill_cb(skb, iph, th);
2348
2349	if (tcp_checksum_complete(skb)) {
2350csum_error:
2351		drop_reason = SKB_DROP_REASON_TCP_CSUM;
2352		trace_tcp_bad_csum(skb);
2353		__TCP_INC_STATS(net, TCP_MIB_CSUMERRORS);
2354bad_packet:
2355		__TCP_INC_STATS(net, TCP_MIB_INERRS);
2356	} else {
2357		tcp_v4_send_reset(NULL, skb);
2358	}
2359
2360discard_it:
2361	SKB_DR_OR(drop_reason, NOT_SPECIFIED);
2362	/* Discard frame. */
2363	kfree_skb_reason(skb, drop_reason);
2364	return 0;
2365
2366discard_and_relse:
2367	sk_drops_add(sk, skb);
2368	if (refcounted)
2369		sock_put(sk);
2370	goto discard_it;
2371
2372do_time_wait:
2373	if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) {
2374		drop_reason = SKB_DROP_REASON_XFRM_POLICY;
2375		inet_twsk_put(inet_twsk(sk));
2376		goto discard_it;
2377	}
2378
2379	tcp_v4_fill_cb(skb, iph, th);
2380
2381	if (tcp_checksum_complete(skb)) {
2382		inet_twsk_put(inet_twsk(sk));
2383		goto csum_error;
2384	}
2385	switch (tcp_timewait_state_process(inet_twsk(sk), skb, th)) {
2386	case TCP_TW_SYN: {
2387		struct sock *sk2 = inet_lookup_listener(net,
2388							net->ipv4.tcp_death_row.hashinfo,
2389							skb, __tcp_hdrlen(th),
2390							iph->saddr, th->source,
2391							iph->daddr, th->dest,
2392							inet_iif(skb),
2393							sdif);
2394		if (sk2) {
2395			inet_twsk_deschedule_put(inet_twsk(sk));
2396			sk = sk2;
2397			tcp_v4_restore_cb(skb);
2398			refcounted = false;
 
2399			goto process;
2400		}
2401	}
2402		/* to ACK */
2403		fallthrough;
2404	case TCP_TW_ACK:
2405		tcp_v4_timewait_ack(sk, skb);
2406		break;
2407	case TCP_TW_RST:
2408		tcp_v4_send_reset(sk, skb);
2409		inet_twsk_deschedule_put(inet_twsk(sk));
2410		goto discard_it;
2411	case TCP_TW_SUCCESS:;
2412	}
2413	goto discard_it;
2414}
2415
2416static struct timewait_sock_ops tcp_timewait_sock_ops = {
2417	.twsk_obj_size	= sizeof(struct tcp_timewait_sock),
2418	.twsk_unique	= tcp_twsk_unique,
2419	.twsk_destructor= tcp_twsk_destructor,
2420};
2421
2422void inet_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb)
2423{
2424	struct dst_entry *dst = skb_dst(skb);
2425
2426	if (dst && dst_hold_safe(dst)) {
2427		rcu_assign_pointer(sk->sk_rx_dst, dst);
2428		sk->sk_rx_dst_ifindex = skb->skb_iif;
2429	}
2430}
2431EXPORT_SYMBOL(inet_sk_rx_dst_set);
2432
2433const struct inet_connection_sock_af_ops ipv4_specific = {
2434	.queue_xmit	   = ip_queue_xmit,
2435	.send_check	   = tcp_v4_send_check,
2436	.rebuild_header	   = inet_sk_rebuild_header,
2437	.sk_rx_dst_set	   = inet_sk_rx_dst_set,
2438	.conn_request	   = tcp_v4_conn_request,
2439	.syn_recv_sock	   = tcp_v4_syn_recv_sock,
2440	.net_header_len	   = sizeof(struct iphdr),
2441	.setsockopt	   = ip_setsockopt,
2442	.getsockopt	   = ip_getsockopt,
2443	.addr2sockaddr	   = inet_csk_addr2sockaddr,
2444	.sockaddr_len	   = sizeof(struct sockaddr_in),
2445	.mtu_reduced	   = tcp_v4_mtu_reduced,
2446};
2447EXPORT_SYMBOL(ipv4_specific);
2448
2449#if defined(CONFIG_TCP_MD5SIG) || defined(CONFIG_TCP_AO)
2450static const struct tcp_sock_af_ops tcp_sock_ipv4_specific = {
2451#ifdef CONFIG_TCP_MD5SIG
2452	.md5_lookup		= tcp_v4_md5_lookup,
2453	.calc_md5_hash		= tcp_v4_md5_hash_skb,
2454	.md5_parse		= tcp_v4_parse_md5_keys,
2455#endif
2456#ifdef CONFIG_TCP_AO
2457	.ao_lookup		= tcp_v4_ao_lookup,
2458	.calc_ao_hash		= tcp_v4_ao_hash_skb,
2459	.ao_parse		= tcp_v4_parse_ao,
2460	.ao_calc_key_sk		= tcp_v4_ao_calc_key_sk,
2461#endif
2462};
2463#endif
2464
2465/* NOTE: A lot of things set to zero explicitly by call to
2466 *       sk_alloc() so need not be done here.
2467 */
2468static int tcp_v4_init_sock(struct sock *sk)
2469{
2470	struct inet_connection_sock *icsk = inet_csk(sk);
2471
2472	tcp_init_sock(sk);
2473
2474	icsk->icsk_af_ops = &ipv4_specific;
2475
2476#if defined(CONFIG_TCP_MD5SIG) || defined(CONFIG_TCP_AO)
2477	tcp_sk(sk)->af_specific = &tcp_sock_ipv4_specific;
2478#endif
2479
2480	return 0;
2481}
2482
2483#ifdef CONFIG_TCP_MD5SIG
2484static void tcp_md5sig_info_free_rcu(struct rcu_head *head)
2485{
2486	struct tcp_md5sig_info *md5sig;
2487
2488	md5sig = container_of(head, struct tcp_md5sig_info, rcu);
2489	kfree(md5sig);
2490	static_branch_slow_dec_deferred(&tcp_md5_needed);
2491	tcp_md5_release_sigpool();
2492}
2493#endif
2494
 
 
 
 
 
 
 
 
 
 
 
2495void tcp_v4_destroy_sock(struct sock *sk)
2496{
2497	struct tcp_sock *tp = tcp_sk(sk);
2498
 
 
 
 
2499	trace_tcp_destroy_sock(sk);
2500
2501	tcp_clear_xmit_timers(sk);
2502
2503	tcp_cleanup_congestion_control(sk);
2504
2505	tcp_cleanup_ulp(sk);
2506
2507	/* Cleanup up the write buffer. */
2508	tcp_write_queue_purge(sk);
2509
2510	/* Check if we want to disable active TFO */
2511	tcp_fastopen_active_disable_ofo_check(sk);
2512
2513	/* Cleans up our, hopefully empty, out_of_order_queue. */
2514	skb_rbtree_purge(&tp->out_of_order_queue);
2515
2516#ifdef CONFIG_TCP_MD5SIG
2517	/* Clean up the MD5 key list, if any */
2518	if (tp->md5sig_info) {
2519		struct tcp_md5sig_info *md5sig;
2520
2521		md5sig = rcu_dereference_protected(tp->md5sig_info, 1);
2522		tcp_clear_md5_list(sk);
2523		call_rcu(&md5sig->rcu, tcp_md5sig_info_free_rcu);
2524		rcu_assign_pointer(tp->md5sig_info, NULL);
2525	}
2526#endif
2527	tcp_ao_destroy_sock(sk, false);
2528
2529	/* Clean up a referenced TCP bind bucket. */
2530	if (inet_csk(sk)->icsk_bind_hash)
2531		inet_put_port(sk);
2532
2533	BUG_ON(rcu_access_pointer(tp->fastopen_rsk));
2534
2535	/* If socket is aborted during connect operation */
2536	tcp_free_fastopen_req(tp);
2537	tcp_fastopen_destroy_cipher(sk);
2538	tcp_saved_syn_free(tp);
2539
2540	sk_sockets_allocated_dec(sk);
2541}
2542EXPORT_SYMBOL(tcp_v4_destroy_sock);
2543
2544#ifdef CONFIG_PROC_FS
2545/* Proc filesystem TCP sock list dumping. */
2546
2547static unsigned short seq_file_family(const struct seq_file *seq);
2548
2549static bool seq_sk_match(struct seq_file *seq, const struct sock *sk)
2550{
2551	unsigned short family = seq_file_family(seq);
2552
2553	/* AF_UNSPEC is used as a match all */
2554	return ((family == AF_UNSPEC || family == sk->sk_family) &&
2555		net_eq(sock_net(sk), seq_file_net(seq)));
2556}
2557
2558/* Find a non empty bucket (starting from st->bucket)
2559 * and return the first sk from it.
2560 */
2561static void *listening_get_first(struct seq_file *seq)
2562{
2563	struct inet_hashinfo *hinfo = seq_file_net(seq)->ipv4.tcp_death_row.hashinfo;
2564	struct tcp_iter_state *st = seq->private;
2565
2566	st->offset = 0;
2567	for (; st->bucket <= hinfo->lhash2_mask; st->bucket++) {
2568		struct inet_listen_hashbucket *ilb2;
2569		struct hlist_nulls_node *node;
2570		struct sock *sk;
2571
2572		ilb2 = &hinfo->lhash2[st->bucket];
2573		if (hlist_nulls_empty(&ilb2->nulls_head))
2574			continue;
2575
2576		spin_lock(&ilb2->lock);
2577		sk_nulls_for_each(sk, node, &ilb2->nulls_head) {
2578			if (seq_sk_match(seq, sk))
2579				return sk;
2580		}
2581		spin_unlock(&ilb2->lock);
2582	}
2583
2584	return NULL;
2585}
2586
2587/* Find the next sk of "cur" within the same bucket (i.e. st->bucket).
2588 * If "cur" is the last one in the st->bucket,
2589 * call listening_get_first() to return the first sk of the next
2590 * non empty bucket.
2591 */
2592static void *listening_get_next(struct seq_file *seq, void *cur)
2593{
2594	struct tcp_iter_state *st = seq->private;
2595	struct inet_listen_hashbucket *ilb2;
2596	struct hlist_nulls_node *node;
2597	struct inet_hashinfo *hinfo;
2598	struct sock *sk = cur;
2599
2600	++st->num;
2601	++st->offset;
2602
2603	sk = sk_nulls_next(sk);
2604	sk_nulls_for_each_from(sk, node) {
2605		if (seq_sk_match(seq, sk))
2606			return sk;
2607	}
2608
2609	hinfo = seq_file_net(seq)->ipv4.tcp_death_row.hashinfo;
2610	ilb2 = &hinfo->lhash2[st->bucket];
2611	spin_unlock(&ilb2->lock);
2612	++st->bucket;
2613	return listening_get_first(seq);
2614}
2615
2616static void *listening_get_idx(struct seq_file *seq, loff_t *pos)
2617{
2618	struct tcp_iter_state *st = seq->private;
2619	void *rc;
2620
2621	st->bucket = 0;
2622	st->offset = 0;
2623	rc = listening_get_first(seq);
2624
2625	while (rc && *pos) {
2626		rc = listening_get_next(seq, rc);
2627		--*pos;
2628	}
2629	return rc;
2630}
2631
2632static inline bool empty_bucket(struct inet_hashinfo *hinfo,
2633				const struct tcp_iter_state *st)
2634{
2635	return hlist_nulls_empty(&hinfo->ehash[st->bucket].chain);
2636}
2637
2638/*
2639 * Get first established socket starting from bucket given in st->bucket.
2640 * If st->bucket is zero, the very first socket in the hash is returned.
2641 */
2642static void *established_get_first(struct seq_file *seq)
2643{
2644	struct inet_hashinfo *hinfo = seq_file_net(seq)->ipv4.tcp_death_row.hashinfo;
2645	struct tcp_iter_state *st = seq->private;
2646
2647	st->offset = 0;
2648	for (; st->bucket <= hinfo->ehash_mask; ++st->bucket) {
2649		struct sock *sk;
2650		struct hlist_nulls_node *node;
2651		spinlock_t *lock = inet_ehash_lockp(hinfo, st->bucket);
2652
2653		cond_resched();
2654
2655		/* Lockless fast path for the common case of empty buckets */
2656		if (empty_bucket(hinfo, st))
2657			continue;
2658
2659		spin_lock_bh(lock);
2660		sk_nulls_for_each(sk, node, &hinfo->ehash[st->bucket].chain) {
2661			if (seq_sk_match(seq, sk))
2662				return sk;
2663		}
2664		spin_unlock_bh(lock);
2665	}
2666
2667	return NULL;
2668}
2669
2670static void *established_get_next(struct seq_file *seq, void *cur)
2671{
2672	struct inet_hashinfo *hinfo = seq_file_net(seq)->ipv4.tcp_death_row.hashinfo;
2673	struct tcp_iter_state *st = seq->private;
2674	struct hlist_nulls_node *node;
2675	struct sock *sk = cur;
2676
2677	++st->num;
2678	++st->offset;
2679
2680	sk = sk_nulls_next(sk);
2681
2682	sk_nulls_for_each_from(sk, node) {
2683		if (seq_sk_match(seq, sk))
2684			return sk;
2685	}
2686
2687	spin_unlock_bh(inet_ehash_lockp(hinfo, st->bucket));
2688	++st->bucket;
2689	return established_get_first(seq);
2690}
2691
2692static void *established_get_idx(struct seq_file *seq, loff_t pos)
2693{
2694	struct tcp_iter_state *st = seq->private;
2695	void *rc;
2696
2697	st->bucket = 0;
2698	rc = established_get_first(seq);
2699
2700	while (rc && pos) {
2701		rc = established_get_next(seq, rc);
2702		--pos;
2703	}
2704	return rc;
2705}
2706
2707static void *tcp_get_idx(struct seq_file *seq, loff_t pos)
2708{
2709	void *rc;
2710	struct tcp_iter_state *st = seq->private;
2711
2712	st->state = TCP_SEQ_STATE_LISTENING;
2713	rc	  = listening_get_idx(seq, &pos);
2714
2715	if (!rc) {
2716		st->state = TCP_SEQ_STATE_ESTABLISHED;
2717		rc	  = established_get_idx(seq, pos);
2718	}
2719
2720	return rc;
2721}
2722
2723static void *tcp_seek_last_pos(struct seq_file *seq)
2724{
2725	struct inet_hashinfo *hinfo = seq_file_net(seq)->ipv4.tcp_death_row.hashinfo;
2726	struct tcp_iter_state *st = seq->private;
2727	int bucket = st->bucket;
2728	int offset = st->offset;
2729	int orig_num = st->num;
2730	void *rc = NULL;
2731
2732	switch (st->state) {
2733	case TCP_SEQ_STATE_LISTENING:
2734		if (st->bucket > hinfo->lhash2_mask)
2735			break;
2736		rc = listening_get_first(seq);
2737		while (offset-- && rc && bucket == st->bucket)
2738			rc = listening_get_next(seq, rc);
2739		if (rc)
2740			break;
2741		st->bucket = 0;
2742		st->state = TCP_SEQ_STATE_ESTABLISHED;
2743		fallthrough;
2744	case TCP_SEQ_STATE_ESTABLISHED:
2745		if (st->bucket > hinfo->ehash_mask)
2746			break;
2747		rc = established_get_first(seq);
2748		while (offset-- && rc && bucket == st->bucket)
2749			rc = established_get_next(seq, rc);
2750	}
2751
2752	st->num = orig_num;
2753
2754	return rc;
2755}
2756
2757void *tcp_seq_start(struct seq_file *seq, loff_t *pos)
2758{
2759	struct tcp_iter_state *st = seq->private;
2760	void *rc;
2761
2762	if (*pos && *pos == st->last_pos) {
2763		rc = tcp_seek_last_pos(seq);
2764		if (rc)
2765			goto out;
2766	}
2767
2768	st->state = TCP_SEQ_STATE_LISTENING;
2769	st->num = 0;
2770	st->bucket = 0;
2771	st->offset = 0;
2772	rc = *pos ? tcp_get_idx(seq, *pos - 1) : SEQ_START_TOKEN;
2773
2774out:
2775	st->last_pos = *pos;
2776	return rc;
2777}
2778EXPORT_SYMBOL(tcp_seq_start);
2779
2780void *tcp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2781{
2782	struct tcp_iter_state *st = seq->private;
2783	void *rc = NULL;
2784
2785	if (v == SEQ_START_TOKEN) {
2786		rc = tcp_get_idx(seq, 0);
2787		goto out;
2788	}
2789
2790	switch (st->state) {
2791	case TCP_SEQ_STATE_LISTENING:
2792		rc = listening_get_next(seq, v);
2793		if (!rc) {
2794			st->state = TCP_SEQ_STATE_ESTABLISHED;
2795			st->bucket = 0;
2796			st->offset = 0;
2797			rc	  = established_get_first(seq);
2798		}
2799		break;
2800	case TCP_SEQ_STATE_ESTABLISHED:
2801		rc = established_get_next(seq, v);
2802		break;
2803	}
2804out:
2805	++*pos;
2806	st->last_pos = *pos;
2807	return rc;
2808}
2809EXPORT_SYMBOL(tcp_seq_next);
2810
2811void tcp_seq_stop(struct seq_file *seq, void *v)
2812{
2813	struct inet_hashinfo *hinfo = seq_file_net(seq)->ipv4.tcp_death_row.hashinfo;
2814	struct tcp_iter_state *st = seq->private;
2815
2816	switch (st->state) {
2817	case TCP_SEQ_STATE_LISTENING:
2818		if (v != SEQ_START_TOKEN)
2819			spin_unlock(&hinfo->lhash2[st->bucket].lock);
2820		break;
2821	case TCP_SEQ_STATE_ESTABLISHED:
2822		if (v)
2823			spin_unlock_bh(inet_ehash_lockp(hinfo, st->bucket));
2824		break;
2825	}
2826}
2827EXPORT_SYMBOL(tcp_seq_stop);
2828
2829static void get_openreq4(const struct request_sock *req,
2830			 struct seq_file *f, int i)
2831{
2832	const struct inet_request_sock *ireq = inet_rsk(req);
2833	long delta = req->rsk_timer.expires - jiffies;
2834
2835	seq_printf(f, "%4d: %08X:%04X %08X:%04X"
2836		" %02X %08X:%08X %02X:%08lX %08X %5u %8d %u %d %pK",
2837		i,
2838		ireq->ir_loc_addr,
2839		ireq->ir_num,
2840		ireq->ir_rmt_addr,
2841		ntohs(ireq->ir_rmt_port),
2842		TCP_SYN_RECV,
2843		0, 0, /* could print option size, but that is af dependent. */
2844		1,    /* timers active (only the expire timer) */
2845		jiffies_delta_to_clock_t(delta),
2846		req->num_timeout,
2847		from_kuid_munged(seq_user_ns(f),
2848				 sock_i_uid(req->rsk_listener)),
2849		0,  /* non standard timer */
2850		0, /* open_requests have no inode */
2851		0,
2852		req);
2853}
2854
2855static void get_tcp4_sock(struct sock *sk, struct seq_file *f, int i)
2856{
2857	int timer_active;
2858	unsigned long timer_expires;
2859	const struct tcp_sock *tp = tcp_sk(sk);
2860	const struct inet_connection_sock *icsk = inet_csk(sk);
2861	const struct inet_sock *inet = inet_sk(sk);
2862	const struct fastopen_queue *fastopenq = &icsk->icsk_accept_queue.fastopenq;
2863	__be32 dest = inet->inet_daddr;
2864	__be32 src = inet->inet_rcv_saddr;
2865	__u16 destp = ntohs(inet->inet_dport);
2866	__u16 srcp = ntohs(inet->inet_sport);
 
2867	int rx_queue;
2868	int state;
2869
2870	if (icsk->icsk_pending == ICSK_TIME_RETRANS ||
2871	    icsk->icsk_pending == ICSK_TIME_REO_TIMEOUT ||
2872	    icsk->icsk_pending == ICSK_TIME_LOSS_PROBE) {
 
2873		timer_active	= 1;
2874		timer_expires	= icsk->icsk_timeout;
2875	} else if (icsk->icsk_pending == ICSK_TIME_PROBE0) {
2876		timer_active	= 4;
2877		timer_expires	= icsk->icsk_timeout;
2878	} else if (timer_pending(&sk->sk_timer)) {
2879		timer_active	= 2;
2880		timer_expires	= sk->sk_timer.expires;
2881	} else {
2882		timer_active	= 0;
2883		timer_expires = jiffies;
2884	}
2885
2886	state = inet_sk_state_load(sk);
2887	if (state == TCP_LISTEN)
2888		rx_queue = READ_ONCE(sk->sk_ack_backlog);
2889	else
2890		/* Because we don't lock the socket,
2891		 * we might find a transient negative value.
2892		 */
2893		rx_queue = max_t(int, READ_ONCE(tp->rcv_nxt) -
2894				      READ_ONCE(tp->copied_seq), 0);
2895
2896	seq_printf(f, "%4d: %08X:%04X %08X:%04X %02X %08X:%08X %02X:%08lX "
2897			"%08X %5u %8d %lu %d %pK %lu %lu %u %u %d",
2898		i, src, srcp, dest, destp, state,
2899		READ_ONCE(tp->write_seq) - tp->snd_una,
2900		rx_queue,
2901		timer_active,
2902		jiffies_delta_to_clock_t(timer_expires - jiffies),
2903		icsk->icsk_retransmits,
2904		from_kuid_munged(seq_user_ns(f), sock_i_uid(sk)),
2905		icsk->icsk_probes_out,
2906		sock_i_ino(sk),
2907		refcount_read(&sk->sk_refcnt), sk,
2908		jiffies_to_clock_t(icsk->icsk_rto),
2909		jiffies_to_clock_t(icsk->icsk_ack.ato),
2910		(icsk->icsk_ack.quick << 1) | inet_csk_in_pingpong_mode(sk),
2911		tcp_snd_cwnd(tp),
2912		state == TCP_LISTEN ?
2913		    fastopenq->max_qlen :
2914		    (tcp_in_initial_slowstart(tp) ? -1 : tp->snd_ssthresh));
2915}
2916
2917static void get_timewait4_sock(const struct inet_timewait_sock *tw,
2918			       struct seq_file *f, int i)
2919{
2920	long delta = tw->tw_timer.expires - jiffies;
2921	__be32 dest, src;
2922	__u16 destp, srcp;
2923
2924	dest  = tw->tw_daddr;
2925	src   = tw->tw_rcv_saddr;
2926	destp = ntohs(tw->tw_dport);
2927	srcp  = ntohs(tw->tw_sport);
2928
2929	seq_printf(f, "%4d: %08X:%04X %08X:%04X"
2930		" %02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %pK",
2931		i, src, srcp, dest, destp, tw->tw_substate, 0, 0,
2932		3, jiffies_delta_to_clock_t(delta), 0, 0, 0, 0,
2933		refcount_read(&tw->tw_refcnt), tw);
2934}
2935
2936#define TMPSZ 150
2937
2938static int tcp4_seq_show(struct seq_file *seq, void *v)
2939{
2940	struct tcp_iter_state *st;
2941	struct sock *sk = v;
2942
2943	seq_setwidth(seq, TMPSZ - 1);
2944	if (v == SEQ_START_TOKEN) {
2945		seq_puts(seq, "  sl  local_address rem_address   st tx_queue "
2946			   "rx_queue tr tm->when retrnsmt   uid  timeout "
2947			   "inode");
2948		goto out;
2949	}
2950	st = seq->private;
2951
2952	if (sk->sk_state == TCP_TIME_WAIT)
2953		get_timewait4_sock(v, seq, st->num);
2954	else if (sk->sk_state == TCP_NEW_SYN_RECV)
2955		get_openreq4(v, seq, st->num);
2956	else
2957		get_tcp4_sock(v, seq, st->num);
2958out:
2959	seq_pad(seq, '\n');
2960	return 0;
2961}
2962
2963#ifdef CONFIG_BPF_SYSCALL
2964struct bpf_tcp_iter_state {
2965	struct tcp_iter_state state;
2966	unsigned int cur_sk;
2967	unsigned int end_sk;
2968	unsigned int max_sk;
2969	struct sock **batch;
2970	bool st_bucket_done;
2971};
2972
2973struct bpf_iter__tcp {
2974	__bpf_md_ptr(struct bpf_iter_meta *, meta);
2975	__bpf_md_ptr(struct sock_common *, sk_common);
2976	uid_t uid __aligned(8);
2977};
2978
2979static int tcp_prog_seq_show(struct bpf_prog *prog, struct bpf_iter_meta *meta,
2980			     struct sock_common *sk_common, uid_t uid)
2981{
2982	struct bpf_iter__tcp ctx;
2983
2984	meta->seq_num--;  /* skip SEQ_START_TOKEN */
2985	ctx.meta = meta;
2986	ctx.sk_common = sk_common;
2987	ctx.uid = uid;
2988	return bpf_iter_run_prog(prog, &ctx);
2989}
2990
2991static void bpf_iter_tcp_put_batch(struct bpf_tcp_iter_state *iter)
2992{
2993	while (iter->cur_sk < iter->end_sk)
2994		sock_gen_put(iter->batch[iter->cur_sk++]);
2995}
2996
2997static int bpf_iter_tcp_realloc_batch(struct bpf_tcp_iter_state *iter,
2998				      unsigned int new_batch_sz)
2999{
3000	struct sock **new_batch;
3001
3002	new_batch = kvmalloc(sizeof(*new_batch) * new_batch_sz,
3003			     GFP_USER | __GFP_NOWARN);
3004	if (!new_batch)
3005		return -ENOMEM;
3006
3007	bpf_iter_tcp_put_batch(iter);
3008	kvfree(iter->batch);
3009	iter->batch = new_batch;
3010	iter->max_sk = new_batch_sz;
3011
3012	return 0;
3013}
3014
3015static unsigned int bpf_iter_tcp_listening_batch(struct seq_file *seq,
3016						 struct sock *start_sk)
3017{
3018	struct inet_hashinfo *hinfo = seq_file_net(seq)->ipv4.tcp_death_row.hashinfo;
3019	struct bpf_tcp_iter_state *iter = seq->private;
3020	struct tcp_iter_state *st = &iter->state;
3021	struct hlist_nulls_node *node;
3022	unsigned int expected = 1;
3023	struct sock *sk;
3024
3025	sock_hold(start_sk);
3026	iter->batch[iter->end_sk++] = start_sk;
3027
3028	sk = sk_nulls_next(start_sk);
3029	sk_nulls_for_each_from(sk, node) {
3030		if (seq_sk_match(seq, sk)) {
3031			if (iter->end_sk < iter->max_sk) {
3032				sock_hold(sk);
3033				iter->batch[iter->end_sk++] = sk;
3034			}
3035			expected++;
3036		}
3037	}
3038	spin_unlock(&hinfo->lhash2[st->bucket].lock);
3039
3040	return expected;
3041}
3042
3043static unsigned int bpf_iter_tcp_established_batch(struct seq_file *seq,
3044						   struct sock *start_sk)
3045{
3046	struct inet_hashinfo *hinfo = seq_file_net(seq)->ipv4.tcp_death_row.hashinfo;
3047	struct bpf_tcp_iter_state *iter = seq->private;
3048	struct tcp_iter_state *st = &iter->state;
3049	struct hlist_nulls_node *node;
3050	unsigned int expected = 1;
3051	struct sock *sk;
3052
3053	sock_hold(start_sk);
3054	iter->batch[iter->end_sk++] = start_sk;
3055
3056	sk = sk_nulls_next(start_sk);
3057	sk_nulls_for_each_from(sk, node) {
3058		if (seq_sk_match(seq, sk)) {
3059			if (iter->end_sk < iter->max_sk) {
3060				sock_hold(sk);
3061				iter->batch[iter->end_sk++] = sk;
3062			}
3063			expected++;
3064		}
3065	}
3066	spin_unlock_bh(inet_ehash_lockp(hinfo, st->bucket));
3067
3068	return expected;
3069}
3070
3071static struct sock *bpf_iter_tcp_batch(struct seq_file *seq)
3072{
3073	struct inet_hashinfo *hinfo = seq_file_net(seq)->ipv4.tcp_death_row.hashinfo;
3074	struct bpf_tcp_iter_state *iter = seq->private;
3075	struct tcp_iter_state *st = &iter->state;
3076	unsigned int expected;
3077	bool resized = false;
3078	struct sock *sk;
3079
3080	/* The st->bucket is done.  Directly advance to the next
3081	 * bucket instead of having the tcp_seek_last_pos() to skip
3082	 * one by one in the current bucket and eventually find out
3083	 * it has to advance to the next bucket.
3084	 */
3085	if (iter->st_bucket_done) {
3086		st->offset = 0;
3087		st->bucket++;
3088		if (st->state == TCP_SEQ_STATE_LISTENING &&
3089		    st->bucket > hinfo->lhash2_mask) {
3090			st->state = TCP_SEQ_STATE_ESTABLISHED;
3091			st->bucket = 0;
3092		}
3093	}
3094
3095again:
3096	/* Get a new batch */
3097	iter->cur_sk = 0;
3098	iter->end_sk = 0;
3099	iter->st_bucket_done = false;
3100
3101	sk = tcp_seek_last_pos(seq);
3102	if (!sk)
3103		return NULL; /* Done */
3104
3105	if (st->state == TCP_SEQ_STATE_LISTENING)
3106		expected = bpf_iter_tcp_listening_batch(seq, sk);
3107	else
3108		expected = bpf_iter_tcp_established_batch(seq, sk);
3109
3110	if (iter->end_sk == expected) {
3111		iter->st_bucket_done = true;
3112		return sk;
3113	}
3114
3115	if (!resized && !bpf_iter_tcp_realloc_batch(iter, expected * 3 / 2)) {
3116		resized = true;
3117		goto again;
3118	}
3119
3120	return sk;
3121}
3122
3123static void *bpf_iter_tcp_seq_start(struct seq_file *seq, loff_t *pos)
3124{
3125	/* bpf iter does not support lseek, so it always
3126	 * continue from where it was stop()-ped.
3127	 */
3128	if (*pos)
3129		return bpf_iter_tcp_batch(seq);
3130
3131	return SEQ_START_TOKEN;
3132}
3133
3134static void *bpf_iter_tcp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3135{
3136	struct bpf_tcp_iter_state *iter = seq->private;
3137	struct tcp_iter_state *st = &iter->state;
3138	struct sock *sk;
3139
3140	/* Whenever seq_next() is called, the iter->cur_sk is
3141	 * done with seq_show(), so advance to the next sk in
3142	 * the batch.
3143	 */
3144	if (iter->cur_sk < iter->end_sk) {
3145		/* Keeping st->num consistent in tcp_iter_state.
3146		 * bpf_iter_tcp does not use st->num.
3147		 * meta.seq_num is used instead.
3148		 */
3149		st->num++;
3150		/* Move st->offset to the next sk in the bucket such that
3151		 * the future start() will resume at st->offset in
3152		 * st->bucket.  See tcp_seek_last_pos().
3153		 */
3154		st->offset++;
3155		sock_gen_put(iter->batch[iter->cur_sk++]);
3156	}
3157
3158	if (iter->cur_sk < iter->end_sk)
3159		sk = iter->batch[iter->cur_sk];
3160	else
3161		sk = bpf_iter_tcp_batch(seq);
3162
3163	++*pos;
3164	/* Keeping st->last_pos consistent in tcp_iter_state.
3165	 * bpf iter does not do lseek, so st->last_pos always equals to *pos.
3166	 */
3167	st->last_pos = *pos;
3168	return sk;
3169}
3170
3171static int bpf_iter_tcp_seq_show(struct seq_file *seq, void *v)
3172{
3173	struct bpf_iter_meta meta;
3174	struct bpf_prog *prog;
3175	struct sock *sk = v;
3176	uid_t uid;
3177	int ret;
3178
3179	if (v == SEQ_START_TOKEN)
3180		return 0;
3181
3182	if (sk_fullsock(sk))
3183		lock_sock(sk);
3184
3185	if (unlikely(sk_unhashed(sk))) {
3186		ret = SEQ_SKIP;
3187		goto unlock;
3188	}
3189
3190	if (sk->sk_state == TCP_TIME_WAIT) {
3191		uid = 0;
3192	} else if (sk->sk_state == TCP_NEW_SYN_RECV) {
3193		const struct request_sock *req = v;
3194
3195		uid = from_kuid_munged(seq_user_ns(seq),
3196				       sock_i_uid(req->rsk_listener));
3197	} else {
3198		uid = from_kuid_munged(seq_user_ns(seq), sock_i_uid(sk));
3199	}
3200
3201	meta.seq = seq;
3202	prog = bpf_iter_get_info(&meta, false);
3203	ret = tcp_prog_seq_show(prog, &meta, v, uid);
3204
3205unlock:
3206	if (sk_fullsock(sk))
3207		release_sock(sk);
3208	return ret;
3209
3210}
3211
3212static void bpf_iter_tcp_seq_stop(struct seq_file *seq, void *v)
3213{
3214	struct bpf_tcp_iter_state *iter = seq->private;
3215	struct bpf_iter_meta meta;
3216	struct bpf_prog *prog;
3217
3218	if (!v) {
3219		meta.seq = seq;
3220		prog = bpf_iter_get_info(&meta, true);
3221		if (prog)
3222			(void)tcp_prog_seq_show(prog, &meta, v, 0);
3223	}
3224
3225	if (iter->cur_sk < iter->end_sk) {
3226		bpf_iter_tcp_put_batch(iter);
3227		iter->st_bucket_done = false;
3228	}
3229}
3230
3231static const struct seq_operations bpf_iter_tcp_seq_ops = {
3232	.show		= bpf_iter_tcp_seq_show,
3233	.start		= bpf_iter_tcp_seq_start,
3234	.next		= bpf_iter_tcp_seq_next,
3235	.stop		= bpf_iter_tcp_seq_stop,
3236};
3237#endif
3238static unsigned short seq_file_family(const struct seq_file *seq)
3239{
3240	const struct tcp_seq_afinfo *afinfo;
3241
3242#ifdef CONFIG_BPF_SYSCALL
3243	/* Iterated from bpf_iter.  Let the bpf prog to filter instead. */
3244	if (seq->op == &bpf_iter_tcp_seq_ops)
3245		return AF_UNSPEC;
3246#endif
3247
3248	/* Iterated from proc fs */
3249	afinfo = pde_data(file_inode(seq->file));
3250	return afinfo->family;
3251}
3252
3253static const struct seq_operations tcp4_seq_ops = {
3254	.show		= tcp4_seq_show,
3255	.start		= tcp_seq_start,
3256	.next		= tcp_seq_next,
3257	.stop		= tcp_seq_stop,
3258};
3259
3260static struct tcp_seq_afinfo tcp4_seq_afinfo = {
3261	.family		= AF_INET,
3262};
3263
3264static int __net_init tcp4_proc_init_net(struct net *net)
3265{
3266	if (!proc_create_net_data("tcp", 0444, net->proc_net, &tcp4_seq_ops,
3267			sizeof(struct tcp_iter_state), &tcp4_seq_afinfo))
3268		return -ENOMEM;
3269	return 0;
3270}
3271
3272static void __net_exit tcp4_proc_exit_net(struct net *net)
3273{
3274	remove_proc_entry("tcp", net->proc_net);
3275}
3276
3277static struct pernet_operations tcp4_net_ops = {
3278	.init = tcp4_proc_init_net,
3279	.exit = tcp4_proc_exit_net,
3280};
3281
3282int __init tcp4_proc_init(void)
3283{
3284	return register_pernet_subsys(&tcp4_net_ops);
3285}
3286
3287void tcp4_proc_exit(void)
3288{
3289	unregister_pernet_subsys(&tcp4_net_ops);
3290}
3291#endif /* CONFIG_PROC_FS */
3292
3293/* @wake is one when sk_stream_write_space() calls us.
3294 * This sends EPOLLOUT only if notsent_bytes is half the limit.
3295 * This mimics the strategy used in sock_def_write_space().
3296 */
3297bool tcp_stream_memory_free(const struct sock *sk, int wake)
3298{
3299	const struct tcp_sock *tp = tcp_sk(sk);
3300	u32 notsent_bytes = READ_ONCE(tp->write_seq) -
3301			    READ_ONCE(tp->snd_nxt);
3302
3303	return (notsent_bytes << wake) < tcp_notsent_lowat(tp);
3304}
3305EXPORT_SYMBOL(tcp_stream_memory_free);
3306
3307struct proto tcp_prot = {
3308	.name			= "TCP",
3309	.owner			= THIS_MODULE,
3310	.close			= tcp_close,
3311	.pre_connect		= tcp_v4_pre_connect,
3312	.connect		= tcp_v4_connect,
3313	.disconnect		= tcp_disconnect,
3314	.accept			= inet_csk_accept,
3315	.ioctl			= tcp_ioctl,
3316	.init			= tcp_v4_init_sock,
3317	.destroy		= tcp_v4_destroy_sock,
3318	.shutdown		= tcp_shutdown,
3319	.setsockopt		= tcp_setsockopt,
3320	.getsockopt		= tcp_getsockopt,
3321	.bpf_bypass_getsockopt	= tcp_bpf_bypass_getsockopt,
3322	.keepalive		= tcp_set_keepalive,
3323	.recvmsg		= tcp_recvmsg,
3324	.sendmsg		= tcp_sendmsg,
3325	.splice_eof		= tcp_splice_eof,
3326	.backlog_rcv		= tcp_v4_do_rcv,
3327	.release_cb		= tcp_release_cb,
3328	.hash			= inet_hash,
3329	.unhash			= inet_unhash,
3330	.get_port		= inet_csk_get_port,
3331	.put_port		= inet_put_port,
3332#ifdef CONFIG_BPF_SYSCALL
3333	.psock_update_sk_prot	= tcp_bpf_update_proto,
3334#endif
3335	.enter_memory_pressure	= tcp_enter_memory_pressure,
3336	.leave_memory_pressure	= tcp_leave_memory_pressure,
3337	.stream_memory_free	= tcp_stream_memory_free,
3338	.sockets_allocated	= &tcp_sockets_allocated,
3339	.orphan_count		= &tcp_orphan_count,
3340
3341	.memory_allocated	= &tcp_memory_allocated,
3342	.per_cpu_fw_alloc	= &tcp_memory_per_cpu_fw_alloc,
3343
3344	.memory_pressure	= &tcp_memory_pressure,
3345	.sysctl_mem		= sysctl_tcp_mem,
3346	.sysctl_wmem_offset	= offsetof(struct net, ipv4.sysctl_tcp_wmem),
3347	.sysctl_rmem_offset	= offsetof(struct net, ipv4.sysctl_tcp_rmem),
3348	.max_header		= MAX_TCP_HEADER,
3349	.obj_size		= sizeof(struct tcp_sock),
3350	.slab_flags		= SLAB_TYPESAFE_BY_RCU,
3351	.twsk_prot		= &tcp_timewait_sock_ops,
3352	.rsk_prot		= &tcp_request_sock_ops,
3353	.h.hashinfo		= NULL,
3354	.no_autobind		= true,
3355	.diag_destroy		= tcp_abort,
3356};
3357EXPORT_SYMBOL(tcp_prot);
3358
3359static void __net_exit tcp_sk_exit(struct net *net)
3360{
3361	if (net->ipv4.tcp_congestion_control)
3362		bpf_module_put(net->ipv4.tcp_congestion_control,
3363			       net->ipv4.tcp_congestion_control->owner);
3364}
3365
3366static void __net_init tcp_set_hashinfo(struct net *net)
3367{
3368	struct inet_hashinfo *hinfo;
3369	unsigned int ehash_entries;
3370	struct net *old_net;
3371
3372	if (net_eq(net, &init_net))
3373		goto fallback;
3374
3375	old_net = current->nsproxy->net_ns;
3376	ehash_entries = READ_ONCE(old_net->ipv4.sysctl_tcp_child_ehash_entries);
3377	if (!ehash_entries)
3378		goto fallback;
3379
3380	ehash_entries = roundup_pow_of_two(ehash_entries);
3381	hinfo = inet_pernet_hashinfo_alloc(&tcp_hashinfo, ehash_entries);
3382	if (!hinfo) {
3383		pr_warn("Failed to allocate TCP ehash (entries: %u) "
3384			"for a netns, fallback to the global one\n",
3385			ehash_entries);
3386fallback:
3387		hinfo = &tcp_hashinfo;
3388		ehash_entries = tcp_hashinfo.ehash_mask + 1;
3389	}
3390
3391	net->ipv4.tcp_death_row.hashinfo = hinfo;
3392	net->ipv4.tcp_death_row.sysctl_max_tw_buckets = ehash_entries / 2;
3393	net->ipv4.sysctl_max_syn_backlog = max(128U, ehash_entries / 128);
3394}
3395
3396static int __net_init tcp_sk_init(struct net *net)
3397{
3398	net->ipv4.sysctl_tcp_ecn = 2;
3399	net->ipv4.sysctl_tcp_ecn_fallback = 1;
3400
3401	net->ipv4.sysctl_tcp_base_mss = TCP_BASE_MSS;
3402	net->ipv4.sysctl_tcp_min_snd_mss = TCP_MIN_SND_MSS;
3403	net->ipv4.sysctl_tcp_probe_threshold = TCP_PROBE_THRESHOLD;
3404	net->ipv4.sysctl_tcp_probe_interval = TCP_PROBE_INTERVAL;
3405	net->ipv4.sysctl_tcp_mtu_probe_floor = TCP_MIN_SND_MSS;
3406
3407	net->ipv4.sysctl_tcp_keepalive_time = TCP_KEEPALIVE_TIME;
3408	net->ipv4.sysctl_tcp_keepalive_probes = TCP_KEEPALIVE_PROBES;
3409	net->ipv4.sysctl_tcp_keepalive_intvl = TCP_KEEPALIVE_INTVL;
3410
3411	net->ipv4.sysctl_tcp_syn_retries = TCP_SYN_RETRIES;
3412	net->ipv4.sysctl_tcp_synack_retries = TCP_SYNACK_RETRIES;
3413	net->ipv4.sysctl_tcp_syncookies = 1;
3414	net->ipv4.sysctl_tcp_reordering = TCP_FASTRETRANS_THRESH;
3415	net->ipv4.sysctl_tcp_retries1 = TCP_RETR1;
3416	net->ipv4.sysctl_tcp_retries2 = TCP_RETR2;
3417	net->ipv4.sysctl_tcp_orphan_retries = 0;
3418	net->ipv4.sysctl_tcp_fin_timeout = TCP_FIN_TIMEOUT;
3419	net->ipv4.sysctl_tcp_notsent_lowat = UINT_MAX;
3420	net->ipv4.sysctl_tcp_tw_reuse = 2;
3421	net->ipv4.sysctl_tcp_no_ssthresh_metrics_save = 1;
3422
3423	refcount_set(&net->ipv4.tcp_death_row.tw_refcount, 1);
3424	tcp_set_hashinfo(net);
3425
3426	net->ipv4.sysctl_tcp_sack = 1;
3427	net->ipv4.sysctl_tcp_window_scaling = 1;
3428	net->ipv4.sysctl_tcp_timestamps = 1;
3429	net->ipv4.sysctl_tcp_early_retrans = 3;
3430	net->ipv4.sysctl_tcp_recovery = TCP_RACK_LOSS_DETECTION;
3431	net->ipv4.sysctl_tcp_slow_start_after_idle = 1; /* By default, RFC2861 behavior.  */
3432	net->ipv4.sysctl_tcp_retrans_collapse = 1;
3433	net->ipv4.sysctl_tcp_max_reordering = 300;
3434	net->ipv4.sysctl_tcp_dsack = 1;
3435	net->ipv4.sysctl_tcp_app_win = 31;
3436	net->ipv4.sysctl_tcp_adv_win_scale = 1;
3437	net->ipv4.sysctl_tcp_frto = 2;
3438	net->ipv4.sysctl_tcp_moderate_rcvbuf = 1;
3439	/* This limits the percentage of the congestion window which we
3440	 * will allow a single TSO frame to consume.  Building TSO frames
3441	 * which are too large can cause TCP streams to be bursty.
3442	 */
3443	net->ipv4.sysctl_tcp_tso_win_divisor = 3;
3444	/* Default TSQ limit of 16 TSO segments */
3445	net->ipv4.sysctl_tcp_limit_output_bytes = 16 * 65536;
3446
3447	/* rfc5961 challenge ack rate limiting, per net-ns, disabled by default. */
3448	net->ipv4.sysctl_tcp_challenge_ack_limit = INT_MAX;
3449
3450	net->ipv4.sysctl_tcp_min_tso_segs = 2;
3451	net->ipv4.sysctl_tcp_tso_rtt_log = 9;  /* 2^9 = 512 usec */
3452	net->ipv4.sysctl_tcp_min_rtt_wlen = 300;
3453	net->ipv4.sysctl_tcp_autocorking = 1;
3454	net->ipv4.sysctl_tcp_invalid_ratelimit = HZ/2;
3455	net->ipv4.sysctl_tcp_pacing_ss_ratio = 200;
3456	net->ipv4.sysctl_tcp_pacing_ca_ratio = 120;
3457	if (net != &init_net) {
3458		memcpy(net->ipv4.sysctl_tcp_rmem,
3459		       init_net.ipv4.sysctl_tcp_rmem,
3460		       sizeof(init_net.ipv4.sysctl_tcp_rmem));
3461		memcpy(net->ipv4.sysctl_tcp_wmem,
3462		       init_net.ipv4.sysctl_tcp_wmem,
3463		       sizeof(init_net.ipv4.sysctl_tcp_wmem));
3464	}
3465	net->ipv4.sysctl_tcp_comp_sack_delay_ns = NSEC_PER_MSEC;
3466	net->ipv4.sysctl_tcp_comp_sack_slack_ns = 100 * NSEC_PER_USEC;
3467	net->ipv4.sysctl_tcp_comp_sack_nr = 44;
3468	net->ipv4.sysctl_tcp_backlog_ack_defer = 1;
3469	net->ipv4.sysctl_tcp_fastopen = TFO_CLIENT_ENABLE;
3470	net->ipv4.sysctl_tcp_fastopen_blackhole_timeout = 0;
3471	atomic_set(&net->ipv4.tfo_active_disable_times, 0);
3472
3473	/* Set default values for PLB */
3474	net->ipv4.sysctl_tcp_plb_enabled = 0; /* Disabled by default */
3475	net->ipv4.sysctl_tcp_plb_idle_rehash_rounds = 3;
3476	net->ipv4.sysctl_tcp_plb_rehash_rounds = 12;
3477	net->ipv4.sysctl_tcp_plb_suspend_rto_sec = 60;
3478	/* Default congestion threshold for PLB to mark a round is 50% */
3479	net->ipv4.sysctl_tcp_plb_cong_thresh = (1 << TCP_PLB_SCALE) / 2;
3480
3481	/* Reno is always built in */
3482	if (!net_eq(net, &init_net) &&
3483	    bpf_try_module_get(init_net.ipv4.tcp_congestion_control,
3484			       init_net.ipv4.tcp_congestion_control->owner))
3485		net->ipv4.tcp_congestion_control = init_net.ipv4.tcp_congestion_control;
3486	else
3487		net->ipv4.tcp_congestion_control = &tcp_reno;
3488
3489	net->ipv4.sysctl_tcp_syn_linear_timeouts = 4;
3490	net->ipv4.sysctl_tcp_shrink_window = 0;
3491
3492	net->ipv4.sysctl_tcp_pingpong_thresh = 1;
 
3493
3494	return 0;
3495}
3496
3497static void __net_exit tcp_sk_exit_batch(struct list_head *net_exit_list)
3498{
3499	struct net *net;
3500
3501	tcp_twsk_purge(net_exit_list, AF_INET);
 
 
 
 
 
 
 
 
 
 
3502
3503	list_for_each_entry(net, net_exit_list, exit_list) {
3504		inet_pernet_hashinfo_free(net->ipv4.tcp_death_row.hashinfo);
3505		WARN_ON_ONCE(!refcount_dec_and_test(&net->ipv4.tcp_death_row.tw_refcount));
3506		tcp_fastopen_ctx_destroy(net);
3507	}
 
 
3508}
3509
3510static struct pernet_operations __net_initdata tcp_sk_ops = {
3511       .init	   = tcp_sk_init,
3512       .exit	   = tcp_sk_exit,
3513       .exit_batch = tcp_sk_exit_batch,
3514};
3515
3516#if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
3517DEFINE_BPF_ITER_FUNC(tcp, struct bpf_iter_meta *meta,
3518		     struct sock_common *sk_common, uid_t uid)
3519
3520#define INIT_BATCH_SZ 16
3521
3522static int bpf_iter_init_tcp(void *priv_data, struct bpf_iter_aux_info *aux)
3523{
3524	struct bpf_tcp_iter_state *iter = priv_data;
3525	int err;
3526
3527	err = bpf_iter_init_seq_net(priv_data, aux);
3528	if (err)
3529		return err;
3530
3531	err = bpf_iter_tcp_realloc_batch(iter, INIT_BATCH_SZ);
3532	if (err) {
3533		bpf_iter_fini_seq_net(priv_data);
3534		return err;
3535	}
3536
3537	return 0;
3538}
3539
3540static void bpf_iter_fini_tcp(void *priv_data)
3541{
3542	struct bpf_tcp_iter_state *iter = priv_data;
3543
3544	bpf_iter_fini_seq_net(priv_data);
3545	kvfree(iter->batch);
3546}
3547
3548static const struct bpf_iter_seq_info tcp_seq_info = {
3549	.seq_ops		= &bpf_iter_tcp_seq_ops,
3550	.init_seq_private	= bpf_iter_init_tcp,
3551	.fini_seq_private	= bpf_iter_fini_tcp,
3552	.seq_priv_size		= sizeof(struct bpf_tcp_iter_state),
3553};
3554
3555static const struct bpf_func_proto *
3556bpf_iter_tcp_get_func_proto(enum bpf_func_id func_id,
3557			    const struct bpf_prog *prog)
3558{
3559	switch (func_id) {
3560	case BPF_FUNC_setsockopt:
3561		return &bpf_sk_setsockopt_proto;
3562	case BPF_FUNC_getsockopt:
3563		return &bpf_sk_getsockopt_proto;
3564	default:
3565		return NULL;
3566	}
3567}
3568
3569static struct bpf_iter_reg tcp_reg_info = {
3570	.target			= "tcp",
3571	.ctx_arg_info_size	= 1,
3572	.ctx_arg_info		= {
3573		{ offsetof(struct bpf_iter__tcp, sk_common),
3574		  PTR_TO_BTF_ID_OR_NULL | PTR_TRUSTED },
3575	},
3576	.get_func_proto		= bpf_iter_tcp_get_func_proto,
3577	.seq_info		= &tcp_seq_info,
3578};
3579
3580static void __init bpf_iter_register(void)
3581{
3582	tcp_reg_info.ctx_arg_info[0].btf_id = btf_sock_ids[BTF_SOCK_TYPE_SOCK_COMMON];
3583	if (bpf_iter_reg_target(&tcp_reg_info))
3584		pr_warn("Warning: could not register bpf iterator tcp\n");
3585}
3586
3587#endif
3588
3589void __init tcp_v4_init(void)
3590{
3591	int cpu, res;
3592
3593	for_each_possible_cpu(cpu) {
3594		struct sock *sk;
3595
3596		res = inet_ctl_sock_create(&sk, PF_INET, SOCK_RAW,
3597					   IPPROTO_TCP, &init_net);
3598		if (res)
3599			panic("Failed to create the TCP control socket.\n");
3600		sock_set_flag(sk, SOCK_USE_WRITE_QUEUE);
3601
3602		/* Please enforce IP_DF and IPID==0 for RST and
3603		 * ACK sent in SYN-RECV and TIME-WAIT state.
3604		 */
3605		inet_sk(sk)->pmtudisc = IP_PMTUDISC_DO;
3606
3607		per_cpu(ipv4_tcp_sk, cpu) = sk;
 
 
3608	}
3609	if (register_pernet_subsys(&tcp_sk_ops))
3610		panic("Failed to create the TCP control socket.\n");
3611
3612#if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
3613	bpf_iter_register();
3614#endif
3615}