Linux Audio

Check our new training course

Loading...
v5.9
   1// SPDX-License-Identifier: GPL-2.0-or-later
   2/*
   3 * INET		An implementation of the TCP/IP protocol suite for the LINUX
   4 *		operating system.  INET is implemented using the  BSD Socket
   5 *		interface as the means of communication with the user level.
   6 *
   7 *		Implementation of the Transmission Control Protocol(TCP).
   8 *
   9 *		IPv4 specific functions
  10 *
 
  11 *		code split from:
  12 *		linux/ipv4/tcp.c
  13 *		linux/ipv4/tcp_input.c
  14 *		linux/ipv4/tcp_output.c
  15 *
  16 *		See tcp.c for author information
 
 
 
 
 
  17 */
  18
  19/*
  20 * Changes:
  21 *		David S. Miller	:	New socket lookup architecture.
  22 *					This code is dedicated to John Dyson.
  23 *		David S. Miller :	Change semantics of established hash,
  24 *					half is devoted to TIME_WAIT sockets
  25 *					and the rest go in the other half.
  26 *		Andi Kleen :		Add support for syncookies and fixed
  27 *					some bugs: ip options weren't passed to
  28 *					the TCP layer, missed a check for an
  29 *					ACK bit.
  30 *		Andi Kleen :		Implemented fast path mtu discovery.
  31 *	     				Fixed many serious bugs in the
  32 *					request_sock handling and moved
  33 *					most of it into the af independent code.
  34 *					Added tail drop and some other bugfixes.
  35 *					Added new listen semantics.
  36 *		Mike McLagan	:	Routing by source
  37 *	Juan Jose Ciarlante:		ip_dynaddr bits
  38 *		Andi Kleen:		various fixes.
  39 *	Vitaly E. Lavrov	:	Transparent proxy revived after year
  40 *					coma.
  41 *	Andi Kleen		:	Fix new listen.
  42 *	Andi Kleen		:	Fix accept error reporting.
  43 *	YOSHIFUJI Hideaki @USAGI and:	Support IPV6_V6ONLY socket option, which
  44 *	Alexey Kuznetsov		allow both IPv4 and IPv6 sockets to bind
  45 *					a single port at the same time.
  46 */
  47
  48#define pr_fmt(fmt) "TCP: " fmt
  49
  50#include <linux/bottom_half.h>
  51#include <linux/types.h>
  52#include <linux/fcntl.h>
  53#include <linux/module.h>
  54#include <linux/random.h>
  55#include <linux/cache.h>
  56#include <linux/jhash.h>
  57#include <linux/init.h>
  58#include <linux/times.h>
  59#include <linux/slab.h>
  60
  61#include <net/net_namespace.h>
  62#include <net/icmp.h>
  63#include <net/inet_hashtables.h>
  64#include <net/tcp.h>
  65#include <net/transp_v6.h>
  66#include <net/ipv6.h>
  67#include <net/inet_common.h>
  68#include <net/timewait_sock.h>
  69#include <net/xfrm.h>
 
  70#include <net/secure_seq.h>
 
  71#include <net/busy_poll.h>
  72
  73#include <linux/inet.h>
  74#include <linux/ipv6.h>
  75#include <linux/stddef.h>
  76#include <linux/proc_fs.h>
  77#include <linux/seq_file.h>
  78#include <linux/inetdevice.h>
  79#include <linux/btf_ids.h>
  80
  81#include <crypto/hash.h>
  82#include <linux/scatterlist.h>
  83
  84#include <trace/events/tcp.h>
 
 
 
  85
  86#ifdef CONFIG_TCP_MD5SIG
  87static int tcp_v4_md5_hash_hdr(char *md5_hash, const struct tcp_md5sig_key *key,
  88			       __be32 daddr, __be32 saddr, const struct tcphdr *th);
  89#endif
  90
  91struct inet_hashinfo tcp_hashinfo;
  92EXPORT_SYMBOL(tcp_hashinfo);
  93
  94static u32 tcp_v4_init_seq(const struct sk_buff *skb)
  95{
  96	return secure_tcp_seq(ip_hdr(skb)->daddr,
  97			      ip_hdr(skb)->saddr,
  98			      tcp_hdr(skb)->dest,
  99			      tcp_hdr(skb)->source);
 100}
 101
 102static u32 tcp_v4_init_ts_off(const struct net *net, const struct sk_buff *skb)
 103{
 104	return secure_tcp_ts_off(net, ip_hdr(skb)->daddr, ip_hdr(skb)->saddr);
 105}
 106
 107int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp)
 108{
 109	const struct inet_timewait_sock *tw = inet_twsk(sktw);
 110	const struct tcp_timewait_sock *tcptw = tcp_twsk(sktw);
 111	struct tcp_sock *tp = tcp_sk(sk);
 112	int reuse = sock_net(sk)->ipv4.sysctl_tcp_tw_reuse;
 113
 114	if (reuse == 2) {
 115		/* Still does not detect *everything* that goes through
 116		 * lo, since we require a loopback src or dst address
 117		 * or direct binding to 'lo' interface.
 118		 */
 119		bool loopback = false;
 120		if (tw->tw_bound_dev_if == LOOPBACK_IFINDEX)
 121			loopback = true;
 122#if IS_ENABLED(CONFIG_IPV6)
 123		if (tw->tw_family == AF_INET6) {
 124			if (ipv6_addr_loopback(&tw->tw_v6_daddr) ||
 125			    ipv6_addr_v4mapped_loopback(&tw->tw_v6_daddr) ||
 126			    ipv6_addr_loopback(&tw->tw_v6_rcv_saddr) ||
 127			    ipv6_addr_v4mapped_loopback(&tw->tw_v6_rcv_saddr))
 128				loopback = true;
 129		} else
 130#endif
 131		{
 132			if (ipv4_is_loopback(tw->tw_daddr) ||
 133			    ipv4_is_loopback(tw->tw_rcv_saddr))
 134				loopback = true;
 135		}
 136		if (!loopback)
 137			reuse = 0;
 138	}
 139
 140	/* With PAWS, it is safe from the viewpoint
 141	   of data integrity. Even without PAWS it is safe provided sequence
 142	   spaces do not overlap i.e. at data rates <= 80Mbit/sec.
 143
 144	   Actually, the idea is close to VJ's one, only timestamp cache is
 145	   held not per host, but per port pair and TW bucket is used as state
 146	   holder.
 147
 148	   If TW bucket has been already destroyed we fall back to VJ's scheme
 149	   and use initial timestamp retrieved from peer table.
 150	 */
 151	if (tcptw->tw_ts_recent_stamp &&
 152	    (!twp || (reuse && time_after32(ktime_get_seconds(),
 153					    tcptw->tw_ts_recent_stamp)))) {
 154		/* In case of repair and re-using TIME-WAIT sockets we still
 155		 * want to be sure that it is safe as above but honor the
 156		 * sequence numbers and time stamps set as part of the repair
 157		 * process.
 158		 *
 159		 * Without this check re-using a TIME-WAIT socket with TCP
 160		 * repair would accumulate a -1 on the repair assigned
 161		 * sequence number. The first time it is reused the sequence
 162		 * is -1, the second time -2, etc. This fixes that issue
 163		 * without appearing to create any others.
 164		 */
 165		if (likely(!tp->repair)) {
 166			u32 seq = tcptw->tw_snd_nxt + 65535 + 2;
 167
 168			if (!seq)
 169				seq = 1;
 170			WRITE_ONCE(tp->write_seq, seq);
 171			tp->rx_opt.ts_recent	   = tcptw->tw_ts_recent;
 172			tp->rx_opt.ts_recent_stamp = tcptw->tw_ts_recent_stamp;
 173		}
 174		sock_hold(sktw);
 175		return 1;
 176	}
 177
 178	return 0;
 179}
 180EXPORT_SYMBOL_GPL(tcp_twsk_unique);
 181
 182static int tcp_v4_pre_connect(struct sock *sk, struct sockaddr *uaddr,
 183			      int addr_len)
 184{
 185	/* This check is replicated from tcp_v4_connect() and intended to
 186	 * prevent BPF program called below from accessing bytes that are out
 187	 * of the bound specified by user in addr_len.
 188	 */
 189	if (addr_len < sizeof(struct sockaddr_in))
 190		return -EINVAL;
 191
 192	sock_owned_by_me(sk);
 193
 194	return BPF_CGROUP_RUN_PROG_INET4_CONNECT(sk, uaddr);
 195}
 196
 197/* This will initiate an outgoing connection. */
 198int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len)
 199{
 200	struct sockaddr_in *usin = (struct sockaddr_in *)uaddr;
 201	struct inet_sock *inet = inet_sk(sk);
 202	struct tcp_sock *tp = tcp_sk(sk);
 203	__be16 orig_sport, orig_dport;
 204	__be32 daddr, nexthop;
 205	struct flowi4 *fl4;
 206	struct rtable *rt;
 207	int err;
 208	struct ip_options_rcu *inet_opt;
 209	struct inet_timewait_death_row *tcp_death_row = &sock_net(sk)->ipv4.tcp_death_row;
 210
 211	if (addr_len < sizeof(struct sockaddr_in))
 212		return -EINVAL;
 213
 214	if (usin->sin_family != AF_INET)
 215		return -EAFNOSUPPORT;
 216
 217	nexthop = daddr = usin->sin_addr.s_addr;
 218	inet_opt = rcu_dereference_protected(inet->inet_opt,
 219					     lockdep_sock_is_held(sk));
 220	if (inet_opt && inet_opt->opt.srr) {
 221		if (!daddr)
 222			return -EINVAL;
 223		nexthop = inet_opt->opt.faddr;
 224	}
 225
 226	orig_sport = inet->inet_sport;
 227	orig_dport = usin->sin_port;
 228	fl4 = &inet->cork.fl.u.ip4;
 229	rt = ip_route_connect(fl4, nexthop, inet->inet_saddr,
 230			      RT_CONN_FLAGS(sk), sk->sk_bound_dev_if,
 231			      IPPROTO_TCP,
 232			      orig_sport, orig_dport, sk);
 233	if (IS_ERR(rt)) {
 234		err = PTR_ERR(rt);
 235		if (err == -ENETUNREACH)
 236			IP_INC_STATS(sock_net(sk), IPSTATS_MIB_OUTNOROUTES);
 237		return err;
 238	}
 239
 240	if (rt->rt_flags & (RTCF_MULTICAST | RTCF_BROADCAST)) {
 241		ip_rt_put(rt);
 242		return -ENETUNREACH;
 243	}
 244
 245	if (!inet_opt || !inet_opt->opt.srr)
 246		daddr = fl4->daddr;
 247
 248	if (!inet->inet_saddr)
 249		inet->inet_saddr = fl4->saddr;
 250	sk_rcv_saddr_set(sk, inet->inet_saddr);
 251
 252	if (tp->rx_opt.ts_recent_stamp && inet->inet_daddr != daddr) {
 253		/* Reset inherited state */
 254		tp->rx_opt.ts_recent	   = 0;
 255		tp->rx_opt.ts_recent_stamp = 0;
 256		if (likely(!tp->repair))
 257			WRITE_ONCE(tp->write_seq, 0);
 258	}
 259
 
 
 
 
 260	inet->inet_dport = usin->sin_port;
 261	sk_daddr_set(sk, daddr);
 262
 263	inet_csk(sk)->icsk_ext_hdr_len = 0;
 264	if (inet_opt)
 265		inet_csk(sk)->icsk_ext_hdr_len = inet_opt->opt.optlen;
 266
 267	tp->rx_opt.mss_clamp = TCP_MSS_DEFAULT;
 268
 269	/* Socket identity is still unknown (sport may be zero).
 270	 * However we set state to SYN-SENT and not releasing socket
 271	 * lock select source port, enter ourselves into the hash tables and
 272	 * complete initialization after this.
 273	 */
 274	tcp_set_state(sk, TCP_SYN_SENT);
 275	err = inet_hash_connect(tcp_death_row, sk);
 276	if (err)
 277		goto failure;
 278
 279	sk_set_txhash(sk);
 280
 281	rt = ip_route_newports(fl4, rt, orig_sport, orig_dport,
 282			       inet->inet_sport, inet->inet_dport, sk);
 283	if (IS_ERR(rt)) {
 284		err = PTR_ERR(rt);
 285		rt = NULL;
 286		goto failure;
 287	}
 288	/* OK, now commit destination to socket.  */
 289	sk->sk_gso_type = SKB_GSO_TCPV4;
 290	sk_setup_caps(sk, &rt->dst);
 291	rt = NULL;
 292
 293	if (likely(!tp->repair)) {
 294		if (!tp->write_seq)
 295			WRITE_ONCE(tp->write_seq,
 296				   secure_tcp_seq(inet->inet_saddr,
 297						  inet->inet_daddr,
 298						  inet->inet_sport,
 299						  usin->sin_port));
 300		tp->tsoffset = secure_tcp_ts_off(sock_net(sk),
 301						 inet->inet_saddr,
 302						 inet->inet_daddr);
 303	}
 304
 305	inet->inet_id = prandom_u32();
 
 
 
 
 306
 307	if (tcp_fastopen_defer_connect(sk, &err))
 308		return err;
 309	if (err)
 310		goto failure;
 311
 312	err = tcp_connect(sk);
 313
 
 314	if (err)
 315		goto failure;
 316
 317	return 0;
 318
 319failure:
 320	/*
 321	 * This unhashes the socket and releases the local port,
 322	 * if necessary.
 323	 */
 324	tcp_set_state(sk, TCP_CLOSE);
 325	ip_rt_put(rt);
 326	sk->sk_route_caps = 0;
 327	inet->inet_dport = 0;
 328	return err;
 329}
 330EXPORT_SYMBOL(tcp_v4_connect);
 331
 332/*
 333 * This routine reacts to ICMP_FRAG_NEEDED mtu indications as defined in RFC1191.
 334 * It can be called through tcp_release_cb() if socket was owned by user
 335 * at the time tcp_v4_err() was called to handle ICMP message.
 336 */
 337void tcp_v4_mtu_reduced(struct sock *sk)
 338{
 339	struct inet_sock *inet = inet_sk(sk);
 340	struct dst_entry *dst;
 341	u32 mtu;
 
 342
 343	if ((1 << sk->sk_state) & (TCPF_LISTEN | TCPF_CLOSE))
 344		return;
 345	mtu = tcp_sk(sk)->mtu_info;
 346	dst = inet_csk_update_pmtu(sk, mtu);
 347	if (!dst)
 348		return;
 349
 350	/* Something is about to be wrong... Remember soft error
 351	 * for the case, if this connection will not able to recover.
 352	 */
 353	if (mtu < dst_mtu(dst) && ip_dont_fragment(sk, dst))
 354		sk->sk_err_soft = EMSGSIZE;
 355
 356	mtu = dst_mtu(dst);
 357
 358	if (inet->pmtudisc != IP_PMTUDISC_DONT &&
 359	    ip_sk_accept_pmtu(sk) &&
 360	    inet_csk(sk)->icsk_pmtu_cookie > mtu) {
 361		tcp_sync_mss(sk, mtu);
 362
 363		/* Resend the TCP packet because it's
 364		 * clear that the old packet has been
 365		 * dropped. This is the new "fast" path mtu
 366		 * discovery.
 367		 */
 368		tcp_simple_retransmit(sk);
 369	} /* else let the usual retransmit timer handle it */
 370}
 371EXPORT_SYMBOL(tcp_v4_mtu_reduced);
 372
 373static void do_redirect(struct sk_buff *skb, struct sock *sk)
 374{
 375	struct dst_entry *dst = __sk_dst_check(sk, 0);
 376
 377	if (dst)
 378		dst->ops->redirect(dst, sk, skb);
 379}
 380
 381
 382/* handle ICMP messages on TCP_NEW_SYN_RECV request sockets */
 383void tcp_req_err(struct sock *sk, u32 seq, bool abort)
 384{
 385	struct request_sock *req = inet_reqsk(sk);
 386	struct net *net = sock_net(sk);
 387
 388	/* ICMPs are not backlogged, hence we cannot get
 389	 * an established socket here.
 390	 */
 391	if (seq != tcp_rsk(req)->snt_isn) {
 392		__NET_INC_STATS(net, LINUX_MIB_OUTOFWINDOWICMPS);
 393	} else if (abort) {
 394		/*
 395		 * Still in SYN_RECV, just remove it silently.
 396		 * There is no good way to pass the error to the newly
 397		 * created socket, and POSIX does not want network
 398		 * errors returned from accept().
 399		 */
 400		inet_csk_reqsk_queue_drop(req->rsk_listener, req);
 401		tcp_listendrop(req->rsk_listener);
 402	}
 403	reqsk_put(req);
 404}
 405EXPORT_SYMBOL(tcp_req_err);
 406
 407/* TCP-LD (RFC 6069) logic */
 408void tcp_ld_RTO_revert(struct sock *sk, u32 seq)
 409{
 410	struct inet_connection_sock *icsk = inet_csk(sk);
 411	struct tcp_sock *tp = tcp_sk(sk);
 412	struct sk_buff *skb;
 413	s32 remaining;
 414	u32 delta_us;
 415
 416	if (sock_owned_by_user(sk))
 417		return;
 418
 419	if (seq != tp->snd_una  || !icsk->icsk_retransmits ||
 420	    !icsk->icsk_backoff)
 421		return;
 422
 423	skb = tcp_rtx_queue_head(sk);
 424	if (WARN_ON_ONCE(!skb))
 425		return;
 426
 427	icsk->icsk_backoff--;
 428	icsk->icsk_rto = tp->srtt_us ? __tcp_set_rto(tp) : TCP_TIMEOUT_INIT;
 429	icsk->icsk_rto = inet_csk_rto_backoff(icsk, TCP_RTO_MAX);
 430
 431	tcp_mstamp_refresh(tp);
 432	delta_us = (u32)(tp->tcp_mstamp - tcp_skb_timestamp_us(skb));
 433	remaining = icsk->icsk_rto - usecs_to_jiffies(delta_us);
 434
 435	if (remaining > 0) {
 436		inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
 437					  remaining, TCP_RTO_MAX);
 438	} else {
 439		/* RTO revert clocked out retransmission.
 440		 * Will retransmit now.
 441		 */
 442		tcp_retransmit_timer(sk);
 443	}
 444}
 445EXPORT_SYMBOL(tcp_ld_RTO_revert);
 446
 447/*
 448 * This routine is called by the ICMP module when it gets some
 449 * sort of error condition.  If err < 0 then the socket should
 450 * be closed and the error returned to the user.  If err > 0
 451 * it's just the icmp type << 8 | icmp code.  After adjustment
 452 * header points to the first 8 bytes of the tcp header.  We need
 453 * to find the appropriate port.
 454 *
 455 * The locking strategy used here is very "optimistic". When
 456 * someone else accesses the socket the ICMP is just dropped
 457 * and for some paths there is no check at all.
 458 * A more general error queue to queue errors for later handling
 459 * is probably better.
 460 *
 461 */
 462
 463int tcp_v4_err(struct sk_buff *skb, u32 info)
 464{
 465	const struct iphdr *iph = (const struct iphdr *)skb->data;
 466	struct tcphdr *th = (struct tcphdr *)(skb->data + (iph->ihl << 2));
 
 467	struct tcp_sock *tp;
 468	struct inet_sock *inet;
 469	const int type = icmp_hdr(skb)->type;
 470	const int code = icmp_hdr(skb)->code;
 471	struct sock *sk;
 472	struct request_sock *fastopen;
 473	u32 seq, snd_una;
 
 
 474	int err;
 475	struct net *net = dev_net(skb->dev);
 
 
 
 
 
 476
 477	sk = __inet_lookup_established(net, &tcp_hashinfo, iph->daddr,
 478				       th->dest, iph->saddr, ntohs(th->source),
 479				       inet_iif(skb), 0);
 480	if (!sk) {
 481		__ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
 482		return -ENOENT;
 483	}
 484	if (sk->sk_state == TCP_TIME_WAIT) {
 485		inet_twsk_put(inet_twsk(sk));
 486		return 0;
 487	}
 488	seq = ntohl(th->seq);
 489	if (sk->sk_state == TCP_NEW_SYN_RECV) {
 490		tcp_req_err(sk, seq, type == ICMP_PARAMETERPROB ||
 491				     type == ICMP_TIME_EXCEEDED ||
 492				     (type == ICMP_DEST_UNREACH &&
 493				      (code == ICMP_NET_UNREACH ||
 494				       code == ICMP_HOST_UNREACH)));
 495		return 0;
 496	}
 497
 498	bh_lock_sock(sk);
 499	/* If too many ICMPs get dropped on busy
 500	 * servers this needs to be solved differently.
 501	 * We do take care of PMTU discovery (RFC1191) special case :
 502	 * we can receive locally generated ICMP messages while socket is held.
 503	 */
 504	if (sock_owned_by_user(sk)) {
 505		if (!(type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED))
 506			__NET_INC_STATS(net, LINUX_MIB_LOCKDROPPEDICMPS);
 507	}
 508	if (sk->sk_state == TCP_CLOSE)
 509		goto out;
 510
 511	if (unlikely(iph->ttl < inet_sk(sk)->min_ttl)) {
 512		__NET_INC_STATS(net, LINUX_MIB_TCPMINTTLDROP);
 513		goto out;
 514	}
 515
 
 516	tp = tcp_sk(sk);
 517	/* XXX (TFO) - tp->snd_una should be ISN (tcp_create_openreq_child() */
 518	fastopen = rcu_dereference(tp->fastopen_rsk);
 519	snd_una = fastopen ? tcp_rsk(fastopen)->snt_isn : tp->snd_una;
 520	if (sk->sk_state != TCP_LISTEN &&
 521	    !between(seq, snd_una, tp->snd_nxt)) {
 522		__NET_INC_STATS(net, LINUX_MIB_OUTOFWINDOWICMPS);
 
 
 523		goto out;
 524	}
 525
 526	switch (type) {
 527	case ICMP_REDIRECT:
 528		if (!sock_owned_by_user(sk))
 529			do_redirect(skb, sk);
 530		goto out;
 531	case ICMP_SOURCE_QUENCH:
 532		/* Just silently ignore these. */
 533		goto out;
 534	case ICMP_PARAMETERPROB:
 535		err = EPROTO;
 536		break;
 537	case ICMP_DEST_UNREACH:
 538		if (code > NR_ICMP_UNREACH)
 539			goto out;
 540
 541		if (code == ICMP_FRAG_NEEDED) { /* PMTU discovery (RFC1191) */
 542			/* We are not interested in TCP_LISTEN and open_requests
 543			 * (SYN-ACKs send out by Linux are always <576bytes so
 544			 * they should go through unfragmented).
 545			 */
 546			if (sk->sk_state == TCP_LISTEN)
 547				goto out;
 548
 549			tp->mtu_info = info;
 550			if (!sock_owned_by_user(sk)) {
 551				tcp_v4_mtu_reduced(sk);
 552			} else {
 553				if (!test_and_set_bit(TCP_MTU_REDUCED_DEFERRED, &sk->sk_tsq_flags))
 554					sock_hold(sk);
 555			}
 556			goto out;
 557		}
 558
 559		err = icmp_err_convert[code].errno;
 560		/* check if this ICMP message allows revert of backoff.
 561		 * (see RFC 6069)
 562		 */
 563		if (!fastopen &&
 564		    (code == ICMP_NET_UNREACH || code == ICMP_HOST_UNREACH))
 565			tcp_ld_RTO_revert(sk, seq);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 566		break;
 567	case ICMP_TIME_EXCEEDED:
 568		err = EHOSTUNREACH;
 569		break;
 570	default:
 571		goto out;
 572	}
 573
 
 
 
 
 
 
 
 
 574	switch (sk->sk_state) {
 575	case TCP_SYN_SENT:
 576	case TCP_SYN_RECV:
 577		/* Only in fast or simultaneous open. If a fast open socket is
 578		 * is already accepted it is treated as a connected one below.
 
 
 
 
 
 
 
 
 579		 */
 580		if (fastopen && !fastopen->sk)
 581			break;
 
 
 
 
 582
 583		ip_icmp_error(sk, skb, err, th->dest, info, (u8 *)th);
 
 
 
 
 
 
 
 
 584
 
 
 
 
 
 585		if (!sock_owned_by_user(sk)) {
 586			sk->sk_err = err;
 587
 588			sk->sk_error_report(sk);
 589
 590			tcp_done(sk);
 591		} else {
 592			sk->sk_err_soft = err;
 593		}
 594		goto out;
 595	}
 596
 597	/* If we've already connected we will keep trying
 598	 * until we time out, or the user gives up.
 599	 *
 600	 * rfc1122 4.2.3.9 allows to consider as hard errors
 601	 * only PROTO_UNREACH and PORT_UNREACH (well, FRAG_FAILED too,
 602	 * but it is obsoleted by pmtu discovery).
 603	 *
 604	 * Note, that in modern internet, where routing is unreliable
 605	 * and in each dark corner broken firewalls sit, sending random
 606	 * errors ordered by their masters even this two messages finally lose
 607	 * their original sense (even Linux sends invalid PORT_UNREACHs)
 608	 *
 609	 * Now we are in compliance with RFCs.
 610	 *							--ANK (980905)
 611	 */
 612
 613	inet = inet_sk(sk);
 614	if (!sock_owned_by_user(sk) && inet->recverr) {
 615		sk->sk_err = err;
 616		sk->sk_error_report(sk);
 617	} else	{ /* Only an error on timeout */
 618		sk->sk_err_soft = err;
 619	}
 620
 621out:
 622	bh_unlock_sock(sk);
 623	sock_put(sk);
 624	return 0;
 625}
 626
 627void __tcp_v4_send_check(struct sk_buff *skb, __be32 saddr, __be32 daddr)
 628{
 629	struct tcphdr *th = tcp_hdr(skb);
 630
 631	th->check = ~tcp_v4_check(skb->len, saddr, daddr, 0);
 632	skb->csum_start = skb_transport_header(skb) - skb->head;
 633	skb->csum_offset = offsetof(struct tcphdr, check);
 
 
 
 
 
 
 
 634}
 635
 636/* This routine computes an IPv4 TCP checksum. */
 637void tcp_v4_send_check(struct sock *sk, struct sk_buff *skb)
 638{
 639	const struct inet_sock *inet = inet_sk(sk);
 640
 641	__tcp_v4_send_check(skb, inet->inet_saddr, inet->inet_daddr);
 642}
 643EXPORT_SYMBOL(tcp_v4_send_check);
 644
 645/*
 646 *	This routine will send an RST to the other tcp.
 647 *
 648 *	Someone asks: why I NEVER use socket parameters (TOS, TTL etc.)
 649 *		      for reset.
 650 *	Answer: if a packet caused RST, it is not for a socket
 651 *		existing in our system, if it is matched to a socket,
 652 *		it is just duplicate segment or bug in other side's TCP.
 653 *		So that we build reply only basing on parameters
 654 *		arrived with segment.
 655 *	Exception: precedence violation. We do not implement it in any case.
 656 */
 657
 658static void tcp_v4_send_reset(const struct sock *sk, struct sk_buff *skb)
 659{
 660	const struct tcphdr *th = tcp_hdr(skb);
 661	struct {
 662		struct tcphdr th;
 663#ifdef CONFIG_TCP_MD5SIG
 664		__be32 opt[(TCPOLEN_MD5SIG_ALIGNED >> 2)];
 665#endif
 666	} rep;
 667	struct ip_reply_arg arg;
 668#ifdef CONFIG_TCP_MD5SIG
 669	struct tcp_md5sig_key *key = NULL;
 670	const __u8 *hash_location = NULL;
 671	unsigned char newhash[16];
 672	int genhash;
 673	struct sock *sk1 = NULL;
 674#endif
 675	u64 transmit_time = 0;
 676	struct sock *ctl_sk;
 677	struct net *net;
 678
 679	/* Never send a reset in response to a reset. */
 680	if (th->rst)
 681		return;
 682
 683	/* If sk not NULL, it means we did a successful lookup and incoming
 684	 * route had to be correct. prequeue might have dropped our dst.
 685	 */
 686	if (!sk && skb_rtable(skb)->rt_type != RTN_LOCAL)
 687		return;
 688
 689	/* Swap the send and the receive. */
 690	memset(&rep, 0, sizeof(rep));
 691	rep.th.dest   = th->source;
 692	rep.th.source = th->dest;
 693	rep.th.doff   = sizeof(struct tcphdr) / 4;
 694	rep.th.rst    = 1;
 695
 696	if (th->ack) {
 697		rep.th.seq = th->ack_seq;
 698	} else {
 699		rep.th.ack = 1;
 700		rep.th.ack_seq = htonl(ntohl(th->seq) + th->syn + th->fin +
 701				       skb->len - (th->doff << 2));
 702	}
 703
 704	memset(&arg, 0, sizeof(arg));
 705	arg.iov[0].iov_base = (unsigned char *)&rep;
 706	arg.iov[0].iov_len  = sizeof(rep.th);
 707
 708	net = sk ? sock_net(sk) : dev_net(skb_dst(skb)->dev);
 709#ifdef CONFIG_TCP_MD5SIG
 710	rcu_read_lock();
 711	hash_location = tcp_parse_md5sig_option(th);
 712	if (sk && sk_fullsock(sk)) {
 713		const union tcp_md5_addr *addr;
 714		int l3index;
 715
 716		/* sdif set, means packet ingressed via a device
 717		 * in an L3 domain and inet_iif is set to it.
 718		 */
 719		l3index = tcp_v4_sdif(skb) ? inet_iif(skb) : 0;
 720		addr = (union tcp_md5_addr *)&ip_hdr(skb)->saddr;
 721		key = tcp_md5_do_lookup(sk, l3index, addr, AF_INET);
 722	} else if (hash_location) {
 723		const union tcp_md5_addr *addr;
 724		int sdif = tcp_v4_sdif(skb);
 725		int dif = inet_iif(skb);
 726		int l3index;
 727
 728		/*
 729		 * active side is lost. Try to find listening socket through
 730		 * source port, and then find md5 key through listening socket.
 731		 * we are not loose security here:
 732		 * Incoming packet is checked with md5 hash with finding key,
 733		 * no RST generated if md5 hash doesn't match.
 734		 */
 735		sk1 = __inet_lookup_listener(net, &tcp_hashinfo, NULL, 0,
 736					     ip_hdr(skb)->saddr,
 737					     th->source, ip_hdr(skb)->daddr,
 738					     ntohs(th->source), dif, sdif);
 739		/* don't send rst if it can't find key */
 740		if (!sk1)
 741			goto out;
 742
 743		/* sdif set, means packet ingressed via a device
 744		 * in an L3 domain and dif is set to it.
 745		 */
 746		l3index = sdif ? dif : 0;
 747		addr = (union tcp_md5_addr *)&ip_hdr(skb)->saddr;
 748		key = tcp_md5_do_lookup(sk1, l3index, addr, AF_INET);
 749		if (!key)
 750			goto out;
 751
 752
 753		genhash = tcp_v4_md5_hash_skb(newhash, key, NULL, skb);
 754		if (genhash || memcmp(hash_location, newhash, 16) != 0)
 755			goto out;
 756
 
 
 
 757	}
 758
 759	if (key) {
 760		rep.opt[0] = htonl((TCPOPT_NOP << 24) |
 761				   (TCPOPT_NOP << 16) |
 762				   (TCPOPT_MD5SIG << 8) |
 763				   TCPOLEN_MD5SIG);
 764		/* Update length and the length the header thinks exists */
 765		arg.iov[0].iov_len += TCPOLEN_MD5SIG_ALIGNED;
 766		rep.th.doff = arg.iov[0].iov_len / 4;
 767
 768		tcp_v4_md5_hash_hdr((__u8 *) &rep.opt[1],
 769				     key, ip_hdr(skb)->saddr,
 770				     ip_hdr(skb)->daddr, &rep.th);
 771	}
 772#endif
 773	arg.csum = csum_tcpudp_nofold(ip_hdr(skb)->daddr,
 774				      ip_hdr(skb)->saddr, /* XXX */
 775				      arg.iov[0].iov_len, IPPROTO_TCP, 0);
 776	arg.csumoffset = offsetof(struct tcphdr, check) / 2;
 777	arg.flags = (sk && inet_sk_transparent(sk)) ? IP_REPLY_ARG_NOSRCCHECK : 0;
 778
 779	/* When socket is gone, all binding information is lost.
 780	 * routing might fail in this case. No choice here, if we choose to force
 781	 * input interface, we will misroute in case of asymmetric route.
 782	 */
 783	if (sk) {
 784		arg.bound_dev_if = sk->sk_bound_dev_if;
 785		if (sk_fullsock(sk))
 786			trace_tcp_send_reset(sk, skb);
 787	}
 788
 789	BUILD_BUG_ON(offsetof(struct sock, sk_bound_dev_if) !=
 790		     offsetof(struct inet_timewait_sock, tw_bound_dev_if));
 791
 
 792	arg.tos = ip_hdr(skb)->tos;
 793	arg.uid = sock_net_uid(net, sk && sk_fullsock(sk) ? sk : NULL);
 794	local_bh_disable();
 795	ctl_sk = this_cpu_read(*net->ipv4.tcp_sk);
 796	if (sk) {
 797		ctl_sk->sk_mark = (sk->sk_state == TCP_TIME_WAIT) ?
 798				   inet_twsk(sk)->tw_mark : sk->sk_mark;
 799		ctl_sk->sk_priority = (sk->sk_state == TCP_TIME_WAIT) ?
 800				   inet_twsk(sk)->tw_priority : sk->sk_priority;
 801		transmit_time = tcp_transmit_time(sk);
 802	}
 803	ip_send_unicast_reply(ctl_sk,
 804			      skb, &TCP_SKB_CB(skb)->header.h4.opt,
 805			      ip_hdr(skb)->saddr, ip_hdr(skb)->daddr,
 806			      &arg, arg.iov[0].iov_len,
 807			      transmit_time);
 808
 809	ctl_sk->sk_mark = 0;
 810	__TCP_INC_STATS(net, TCP_MIB_OUTSEGS);
 811	__TCP_INC_STATS(net, TCP_MIB_OUTRSTS);
 812	local_bh_enable();
 813
 814#ifdef CONFIG_TCP_MD5SIG
 815out:
 816	rcu_read_unlock();
 
 
 
 817#endif
 818}
 819
 820/* The code following below sending ACKs in SYN-RECV and TIME-WAIT states
 821   outside socket context is ugly, certainly. What can I do?
 822 */
 823
 824static void tcp_v4_send_ack(const struct sock *sk,
 825			    struct sk_buff *skb, u32 seq, u32 ack,
 826			    u32 win, u32 tsval, u32 tsecr, int oif,
 827			    struct tcp_md5sig_key *key,
 828			    int reply_flags, u8 tos)
 829{
 830	const struct tcphdr *th = tcp_hdr(skb);
 831	struct {
 832		struct tcphdr th;
 833		__be32 opt[(TCPOLEN_TSTAMP_ALIGNED >> 2)
 834#ifdef CONFIG_TCP_MD5SIG
 835			   + (TCPOLEN_MD5SIG_ALIGNED >> 2)
 836#endif
 837			];
 838	} rep;
 839	struct net *net = sock_net(sk);
 840	struct ip_reply_arg arg;
 841	struct sock *ctl_sk;
 842	u64 transmit_time;
 843
 844	memset(&rep.th, 0, sizeof(struct tcphdr));
 845	memset(&arg, 0, sizeof(arg));
 846
 847	arg.iov[0].iov_base = (unsigned char *)&rep;
 848	arg.iov[0].iov_len  = sizeof(rep.th);
 849	if (tsecr) {
 850		rep.opt[0] = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
 851				   (TCPOPT_TIMESTAMP << 8) |
 852				   TCPOLEN_TIMESTAMP);
 853		rep.opt[1] = htonl(tsval);
 854		rep.opt[2] = htonl(tsecr);
 855		arg.iov[0].iov_len += TCPOLEN_TSTAMP_ALIGNED;
 856	}
 857
 858	/* Swap the send and the receive. */
 859	rep.th.dest    = th->source;
 860	rep.th.source  = th->dest;
 861	rep.th.doff    = arg.iov[0].iov_len / 4;
 862	rep.th.seq     = htonl(seq);
 863	rep.th.ack_seq = htonl(ack);
 864	rep.th.ack     = 1;
 865	rep.th.window  = htons(win);
 866
 867#ifdef CONFIG_TCP_MD5SIG
 868	if (key) {
 869		int offset = (tsecr) ? 3 : 0;
 870
 871		rep.opt[offset++] = htonl((TCPOPT_NOP << 24) |
 872					  (TCPOPT_NOP << 16) |
 873					  (TCPOPT_MD5SIG << 8) |
 874					  TCPOLEN_MD5SIG);
 875		arg.iov[0].iov_len += TCPOLEN_MD5SIG_ALIGNED;
 876		rep.th.doff = arg.iov[0].iov_len/4;
 877
 878		tcp_v4_md5_hash_hdr((__u8 *) &rep.opt[offset],
 879				    key, ip_hdr(skb)->saddr,
 880				    ip_hdr(skb)->daddr, &rep.th);
 881	}
 882#endif
 883	arg.flags = reply_flags;
 884	arg.csum = csum_tcpudp_nofold(ip_hdr(skb)->daddr,
 885				      ip_hdr(skb)->saddr, /* XXX */
 886				      arg.iov[0].iov_len, IPPROTO_TCP, 0);
 887	arg.csumoffset = offsetof(struct tcphdr, check) / 2;
 888	if (oif)
 889		arg.bound_dev_if = oif;
 890	arg.tos = tos;
 891	arg.uid = sock_net_uid(net, sk_fullsock(sk) ? sk : NULL);
 892	local_bh_disable();
 893	ctl_sk = this_cpu_read(*net->ipv4.tcp_sk);
 894	ctl_sk->sk_mark = (sk->sk_state == TCP_TIME_WAIT) ?
 895			   inet_twsk(sk)->tw_mark : sk->sk_mark;
 896	ctl_sk->sk_priority = (sk->sk_state == TCP_TIME_WAIT) ?
 897			   inet_twsk(sk)->tw_priority : sk->sk_priority;
 898	transmit_time = tcp_transmit_time(sk);
 899	ip_send_unicast_reply(ctl_sk,
 900			      skb, &TCP_SKB_CB(skb)->header.h4.opt,
 901			      ip_hdr(skb)->saddr, ip_hdr(skb)->daddr,
 902			      &arg, arg.iov[0].iov_len,
 903			      transmit_time);
 904
 905	ctl_sk->sk_mark = 0;
 906	__TCP_INC_STATS(net, TCP_MIB_OUTSEGS);
 907	local_bh_enable();
 908}
 909
 910static void tcp_v4_timewait_ack(struct sock *sk, struct sk_buff *skb)
 911{
 912	struct inet_timewait_sock *tw = inet_twsk(sk);
 913	struct tcp_timewait_sock *tcptw = tcp_twsk(sk);
 914
 915	tcp_v4_send_ack(sk, skb,
 916			tcptw->tw_snd_nxt, tcptw->tw_rcv_nxt,
 917			tcptw->tw_rcv_wnd >> tw->tw_rcv_wscale,
 918			tcp_time_stamp_raw() + tcptw->tw_ts_offset,
 919			tcptw->tw_ts_recent,
 920			tw->tw_bound_dev_if,
 921			tcp_twsk_md5_key(tcptw),
 922			tw->tw_transparent ? IP_REPLY_ARG_NOSRCCHECK : 0,
 923			tw->tw_tos
 924			);
 925
 926	inet_twsk_put(tw);
 927}
 928
 929static void tcp_v4_reqsk_send_ack(const struct sock *sk, struct sk_buff *skb,
 930				  struct request_sock *req)
 931{
 932	const union tcp_md5_addr *addr;
 933	int l3index;
 934
 935	/* sk->sk_state == TCP_LISTEN -> for regular TCP_SYN_RECV
 936	 * sk->sk_state == TCP_SYN_RECV -> for Fast Open.
 937	 */
 938	u32 seq = (sk->sk_state == TCP_LISTEN) ? tcp_rsk(req)->snt_isn + 1 :
 939					     tcp_sk(sk)->snd_nxt;
 940
 941	/* RFC 7323 2.3
 942	 * The window field (SEG.WND) of every outgoing segment, with the
 943	 * exception of <SYN> segments, MUST be right-shifted by
 944	 * Rcv.Wind.Shift bits:
 945	 */
 946	addr = (union tcp_md5_addr *)&ip_hdr(skb)->saddr;
 947	l3index = tcp_v4_sdif(skb) ? inet_iif(skb) : 0;
 948	tcp_v4_send_ack(sk, skb, seq,
 949			tcp_rsk(req)->rcv_nxt,
 950			req->rsk_rcv_wnd >> inet_rsk(req)->rcv_wscale,
 951			tcp_time_stamp_raw() + tcp_rsk(req)->ts_off,
 952			req->ts_recent,
 953			0,
 954			tcp_md5_do_lookup(sk, l3index, addr, AF_INET),
 
 955			inet_rsk(req)->no_srccheck ? IP_REPLY_ARG_NOSRCCHECK : 0,
 956			ip_hdr(skb)->tos);
 957}
 958
 959/*
 960 *	Send a SYN-ACK after having received a SYN.
 961 *	This still operates on a request_sock only, not on a big
 962 *	socket.
 963 */
 964static int tcp_v4_send_synack(const struct sock *sk, struct dst_entry *dst,
 965			      struct flowi *fl,
 966			      struct request_sock *req,
 967			      struct tcp_fastopen_cookie *foc,
 968			      enum tcp_synack_type synack_type)
 969{
 970	const struct inet_request_sock *ireq = inet_rsk(req);
 971	struct flowi4 fl4;
 972	int err = -1;
 973	struct sk_buff *skb;
 974
 975	/* First, grab a route. */
 976	if (!dst && (dst = inet_csk_route_req(sk, &fl4, req)) == NULL)
 977		return -1;
 978
 979	skb = tcp_make_synack(sk, dst, req, foc, synack_type);
 980
 981	if (skb) {
 982		__tcp_v4_send_check(skb, ireq->ir_loc_addr, ireq->ir_rmt_addr);
 983
 984		rcu_read_lock();
 985		err = ip_build_and_send_pkt(skb, sk, ireq->ir_loc_addr,
 986					    ireq->ir_rmt_addr,
 987					    rcu_dereference(ireq->ireq_opt));
 988		rcu_read_unlock();
 989		err = net_xmit_eval(err);
 
 
 990	}
 991
 992	return err;
 993}
 994
 
 
 
 
 
 
 
 
 
 
 
 995/*
 996 *	IPv4 request_sock destructor.
 997 */
 998static void tcp_v4_reqsk_destructor(struct request_sock *req)
 999{
1000	kfree(rcu_dereference_protected(inet_rsk(req)->ireq_opt, 1));
1001}
1002
1003#ifdef CONFIG_TCP_MD5SIG
1004/*
1005 * RFC2385 MD5 checksumming requires a mapping of
1006 * IP address->MD5 Key.
1007 * We need to maintain these in the sk structure.
1008 */
 
 
 
 
 
 
 
1009
1010DEFINE_STATIC_KEY_FALSE(tcp_md5_needed);
1011EXPORT_SYMBOL(tcp_md5_needed);
 
 
 
 
 
 
1012
1013/* Find the Key structure for an address.  */
1014struct tcp_md5sig_key *__tcp_md5_do_lookup(const struct sock *sk, int l3index,
1015					   const union tcp_md5_addr *addr,
1016					   int family)
 
 
 
 
 
 
 
 
 
 
1017{
1018	const struct tcp_sock *tp = tcp_sk(sk);
1019	struct tcp_md5sig_key *key;
1020	const struct tcp_md5sig_info *md5sig;
1021	__be32 mask;
1022	struct tcp_md5sig_key *best_match = NULL;
1023	bool match;
1024
1025	/* caller either holds rcu_read_lock() or socket lock */
1026	md5sig = rcu_dereference_check(tp->md5sig_info,
1027				       lockdep_sock_is_held(sk));
1028	if (!md5sig)
1029		return NULL;
1030
1031	hlist_for_each_entry_rcu(key, &md5sig->head, node,
1032				 lockdep_sock_is_held(sk)) {
1033		if (key->family != family)
1034			continue;
1035		if (key->l3index && key->l3index != l3index)
1036			continue;
1037		if (family == AF_INET) {
1038			mask = inet_make_mask(key->prefixlen);
1039			match = (key->addr.a4.s_addr & mask) ==
1040				(addr->a4.s_addr & mask);
1041#if IS_ENABLED(CONFIG_IPV6)
1042		} else if (family == AF_INET6) {
1043			match = ipv6_prefix_equal(&key->addr.a6, &addr->a6,
1044						  key->prefixlen);
1045#endif
1046		} else {
1047			match = false;
1048		}
1049
1050		if (match && (!best_match ||
1051			      key->prefixlen > best_match->prefixlen))
1052			best_match = key;
1053	}
1054	return best_match;
1055}
1056EXPORT_SYMBOL(__tcp_md5_do_lookup);
1057
1058static struct tcp_md5sig_key *tcp_md5_do_lookup_exact(const struct sock *sk,
1059						      const union tcp_md5_addr *addr,
1060						      int family, u8 prefixlen,
1061						      int l3index)
 
 
 
 
 
 
 
1062{
1063	const struct tcp_sock *tp = tcp_sk(sk);
1064	struct tcp_md5sig_key *key;
1065	unsigned int size = sizeof(struct in_addr);
1066	const struct tcp_md5sig_info *md5sig;
1067
1068	/* caller either holds rcu_read_lock() or socket lock */
1069	md5sig = rcu_dereference_check(tp->md5sig_info,
1070				       lockdep_sock_is_held(sk));
 
1071	if (!md5sig)
1072		return NULL;
1073#if IS_ENABLED(CONFIG_IPV6)
1074	if (family == AF_INET6)
1075		size = sizeof(struct in6_addr);
1076#endif
1077	hlist_for_each_entry_rcu(key, &md5sig->head, node,
1078				 lockdep_sock_is_held(sk)) {
1079		if (key->family != family)
1080			continue;
1081		if (key->l3index && key->l3index != l3index)
1082			continue;
1083		if (!memcmp(&key->addr, addr, size) &&
1084		    key->prefixlen == prefixlen)
1085			return key;
1086	}
1087	return NULL;
1088}
 
1089
1090struct tcp_md5sig_key *tcp_v4_md5_lookup(const struct sock *sk,
1091					 const struct sock *addr_sk)
1092{
1093	const union tcp_md5_addr *addr;
1094	int l3index;
1095
1096	l3index = l3mdev_master_ifindex_by_index(sock_net(sk),
1097						 addr_sk->sk_bound_dev_if);
1098	addr = (const union tcp_md5_addr *)&addr_sk->sk_daddr;
1099	return tcp_md5_do_lookup(sk, l3index, addr, AF_INET);
1100}
1101EXPORT_SYMBOL(tcp_v4_md5_lookup);
1102
 
 
 
 
 
 
 
 
 
1103/* This can be called on a newly created socket, from other files */
1104int tcp_md5_do_add(struct sock *sk, const union tcp_md5_addr *addr,
1105		   int family, u8 prefixlen, int l3index,
1106		   const u8 *newkey, u8 newkeylen, gfp_t gfp)
1107{
1108	/* Add Key to the list */
1109	struct tcp_md5sig_key *key;
1110	struct tcp_sock *tp = tcp_sk(sk);
1111	struct tcp_md5sig_info *md5sig;
1112
1113	key = tcp_md5_do_lookup_exact(sk, addr, family, prefixlen, l3index);
1114	if (key) {
1115		/* Pre-existing entry - just update that one.
1116		 * Note that the key might be used concurrently.
1117		 * data_race() is telling kcsan that we do not care of
1118		 * key mismatches, since changing MD5 key on live flows
1119		 * can lead to packet drops.
1120		 */
1121		data_race(memcpy(key->key, newkey, newkeylen));
1122
1123		/* Pairs with READ_ONCE() in tcp_md5_hash_key().
1124		 * Also note that a reader could catch new key->keylen value
1125		 * but old key->key[], this is the reason we use __GFP_ZERO
1126		 * at sock_kmalloc() time below these lines.
1127		 */
1128		WRITE_ONCE(key->keylen, newkeylen);
1129
1130		return 0;
1131	}
1132
1133	md5sig = rcu_dereference_protected(tp->md5sig_info,
1134					   lockdep_sock_is_held(sk));
1135	if (!md5sig) {
1136		md5sig = kmalloc(sizeof(*md5sig), gfp);
1137		if (!md5sig)
1138			return -ENOMEM;
1139
1140		sk_nocaps_add(sk, NETIF_F_GSO_MASK);
1141		INIT_HLIST_HEAD(&md5sig->head);
1142		rcu_assign_pointer(tp->md5sig_info, md5sig);
1143	}
1144
1145	key = sock_kmalloc(sk, sizeof(*key), gfp | __GFP_ZERO);
1146	if (!key)
1147		return -ENOMEM;
1148	if (!tcp_alloc_md5sig_pool()) {
1149		sock_kfree_s(sk, key, sizeof(*key));
1150		return -ENOMEM;
1151	}
1152
1153	memcpy(key->key, newkey, newkeylen);
1154	key->keylen = newkeylen;
1155	key->family = family;
1156	key->prefixlen = prefixlen;
1157	key->l3index = l3index;
1158	memcpy(&key->addr, addr,
1159	       (family == AF_INET6) ? sizeof(struct in6_addr) :
1160				      sizeof(struct in_addr));
1161	hlist_add_head_rcu(&key->node, &md5sig->head);
1162	return 0;
1163}
1164EXPORT_SYMBOL(tcp_md5_do_add);
1165
1166int tcp_md5_do_del(struct sock *sk, const union tcp_md5_addr *addr, int family,
1167		   u8 prefixlen, int l3index)
1168{
1169	struct tcp_md5sig_key *key;
1170
1171	key = tcp_md5_do_lookup_exact(sk, addr, family, prefixlen, l3index);
1172	if (!key)
1173		return -ENOENT;
1174	hlist_del_rcu(&key->node);
1175	atomic_sub(sizeof(*key), &sk->sk_omem_alloc);
1176	kfree_rcu(key, rcu);
1177	return 0;
1178}
1179EXPORT_SYMBOL(tcp_md5_do_del);
1180
1181static void tcp_clear_md5_list(struct sock *sk)
1182{
1183	struct tcp_sock *tp = tcp_sk(sk);
1184	struct tcp_md5sig_key *key;
1185	struct hlist_node *n;
1186	struct tcp_md5sig_info *md5sig;
1187
1188	md5sig = rcu_dereference_protected(tp->md5sig_info, 1);
1189
1190	hlist_for_each_entry_safe(key, n, &md5sig->head, node) {
1191		hlist_del_rcu(&key->node);
1192		atomic_sub(sizeof(*key), &sk->sk_omem_alloc);
1193		kfree_rcu(key, rcu);
1194	}
1195}
1196
1197static int tcp_v4_parse_md5_keys(struct sock *sk, int optname,
1198				 sockptr_t optval, int optlen)
1199{
1200	struct tcp_md5sig cmd;
1201	struct sockaddr_in *sin = (struct sockaddr_in *)&cmd.tcpm_addr;
1202	const union tcp_md5_addr *addr;
1203	u8 prefixlen = 32;
1204	int l3index = 0;
1205
1206	if (optlen < sizeof(cmd))
1207		return -EINVAL;
1208
1209	if (copy_from_sockptr(&cmd, optval, sizeof(cmd)))
1210		return -EFAULT;
1211
1212	if (sin->sin_family != AF_INET)
1213		return -EINVAL;
1214
1215	if (optname == TCP_MD5SIG_EXT &&
1216	    cmd.tcpm_flags & TCP_MD5SIG_FLAG_PREFIX) {
1217		prefixlen = cmd.tcpm_prefixlen;
1218		if (prefixlen > 32)
1219			return -EINVAL;
1220	}
1221
1222	if (optname == TCP_MD5SIG_EXT &&
1223	    cmd.tcpm_flags & TCP_MD5SIG_FLAG_IFINDEX) {
1224		struct net_device *dev;
1225
1226		rcu_read_lock();
1227		dev = dev_get_by_index_rcu(sock_net(sk), cmd.tcpm_ifindex);
1228		if (dev && netif_is_l3_master(dev))
1229			l3index = dev->ifindex;
1230
1231		rcu_read_unlock();
1232
1233		/* ok to reference set/not set outside of rcu;
1234		 * right now device MUST be an L3 master
1235		 */
1236		if (!dev || !l3index)
1237			return -EINVAL;
1238	}
1239
1240	addr = (union tcp_md5_addr *)&sin->sin_addr.s_addr;
1241
1242	if (!cmd.tcpm_keylen)
1243		return tcp_md5_do_del(sk, addr, AF_INET, prefixlen, l3index);
1244
1245	if (cmd.tcpm_keylen > TCP_MD5SIG_MAXKEYLEN)
1246		return -EINVAL;
1247
1248	return tcp_md5_do_add(sk, addr, AF_INET, prefixlen, l3index,
1249			      cmd.tcpm_key, cmd.tcpm_keylen, GFP_KERNEL);
 
1250}
1251
1252static int tcp_v4_md5_hash_headers(struct tcp_md5sig_pool *hp,
1253				   __be32 daddr, __be32 saddr,
1254				   const struct tcphdr *th, int nbytes)
1255{
1256	struct tcp4_pseudohdr *bp;
1257	struct scatterlist sg;
1258	struct tcphdr *_th;
1259
1260	bp = hp->scratch;
 
 
 
 
 
 
1261	bp->saddr = saddr;
1262	bp->daddr = daddr;
1263	bp->pad = 0;
1264	bp->protocol = IPPROTO_TCP;
1265	bp->len = cpu_to_be16(nbytes);
1266
1267	_th = (struct tcphdr *)(bp + 1);
1268	memcpy(_th, th, sizeof(*th));
1269	_th->check = 0;
1270
1271	sg_init_one(&sg, bp, sizeof(*bp) + sizeof(*th));
1272	ahash_request_set_crypt(hp->md5_req, &sg, NULL,
1273				sizeof(*bp) + sizeof(*th));
1274	return crypto_ahash_update(hp->md5_req);
1275}
1276
1277static int tcp_v4_md5_hash_hdr(char *md5_hash, const struct tcp_md5sig_key *key,
1278			       __be32 daddr, __be32 saddr, const struct tcphdr *th)
1279{
1280	struct tcp_md5sig_pool *hp;
1281	struct ahash_request *req;
1282
1283	hp = tcp_get_md5sig_pool();
1284	if (!hp)
1285		goto clear_hash_noput;
1286	req = hp->md5_req;
1287
1288	if (crypto_ahash_init(req))
 
 
1289		goto clear_hash;
1290	if (tcp_v4_md5_hash_headers(hp, daddr, saddr, th, th->doff << 2))
1291		goto clear_hash;
1292	if (tcp_md5_hash_key(hp, key))
1293		goto clear_hash;
1294	ahash_request_set_crypt(req, NULL, md5_hash, 0);
1295	if (crypto_ahash_final(req))
1296		goto clear_hash;
1297
1298	tcp_put_md5sig_pool();
1299	return 0;
1300
1301clear_hash:
1302	tcp_put_md5sig_pool();
1303clear_hash_noput:
1304	memset(md5_hash, 0, 16);
1305	return 1;
1306}
1307
1308int tcp_v4_md5_hash_skb(char *md5_hash, const struct tcp_md5sig_key *key,
1309			const struct sock *sk,
1310			const struct sk_buff *skb)
1311{
1312	struct tcp_md5sig_pool *hp;
1313	struct ahash_request *req;
1314	const struct tcphdr *th = tcp_hdr(skb);
1315	__be32 saddr, daddr;
1316
1317	if (sk) { /* valid for establish/request sockets */
1318		saddr = sk->sk_rcv_saddr;
1319		daddr = sk->sk_daddr;
 
 
 
1320	} else {
1321		const struct iphdr *iph = ip_hdr(skb);
1322		saddr = iph->saddr;
1323		daddr = iph->daddr;
1324	}
1325
1326	hp = tcp_get_md5sig_pool();
1327	if (!hp)
1328		goto clear_hash_noput;
1329	req = hp->md5_req;
1330
1331	if (crypto_ahash_init(req))
1332		goto clear_hash;
1333
1334	if (tcp_v4_md5_hash_headers(hp, daddr, saddr, th, skb->len))
 
 
1335		goto clear_hash;
1336	if (tcp_md5_hash_skb_data(hp, skb, th->doff << 2))
1337		goto clear_hash;
1338	if (tcp_md5_hash_key(hp, key))
1339		goto clear_hash;
1340	ahash_request_set_crypt(req, NULL, md5_hash, 0);
1341	if (crypto_ahash_final(req))
1342		goto clear_hash;
1343
1344	tcp_put_md5sig_pool();
1345	return 0;
1346
1347clear_hash:
1348	tcp_put_md5sig_pool();
1349clear_hash_noput:
1350	memset(md5_hash, 0, 16);
1351	return 1;
1352}
1353EXPORT_SYMBOL(tcp_v4_md5_hash_skb);
1354
1355#endif
1356
1357/* Called with rcu_read_lock() */
1358static bool tcp_v4_inbound_md5_hash(const struct sock *sk,
1359				    const struct sk_buff *skb,
1360				    int dif, int sdif)
1361{
1362#ifdef CONFIG_TCP_MD5SIG
1363	/*
1364	 * This gets called for each TCP segment that arrives
1365	 * so we want to be efficient.
1366	 * We have 3 drop cases:
1367	 * o No MD5 hash and one expected.
1368	 * o MD5 hash and we're not expecting one.
1369	 * o MD5 hash and its wrong.
1370	 */
1371	const __u8 *hash_location = NULL;
1372	struct tcp_md5sig_key *hash_expected;
1373	const struct iphdr *iph = ip_hdr(skb);
1374	const struct tcphdr *th = tcp_hdr(skb);
1375	const union tcp_md5_addr *addr;
1376	unsigned char newhash[16];
1377	int genhash, l3index;
1378
1379	/* sdif set, means packet ingressed via a device
1380	 * in an L3 domain and dif is set to the l3mdev
1381	 */
1382	l3index = sdif ? dif : 0;
1383
1384	addr = (union tcp_md5_addr *)&iph->saddr;
1385	hash_expected = tcp_md5_do_lookup(sk, l3index, addr, AF_INET);
1386	hash_location = tcp_parse_md5sig_option(th);
1387
1388	/* We've parsed the options - do we have a hash? */
1389	if (!hash_expected && !hash_location)
1390		return false;
1391
1392	if (hash_expected && !hash_location) {
1393		NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPMD5NOTFOUND);
1394		return true;
1395	}
1396
1397	if (!hash_expected && hash_location) {
1398		NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPMD5UNEXPECTED);
1399		return true;
1400	}
1401
1402	/* Okay, so this is hash_expected and hash_location -
1403	 * so we need to calculate the checksum.
1404	 */
1405	genhash = tcp_v4_md5_hash_skb(newhash,
1406				      hash_expected,
1407				      NULL, skb);
1408
1409	if (genhash || memcmp(hash_location, newhash, 16) != 0) {
1410		NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPMD5FAILURE);
1411		net_info_ratelimited("MD5 Hash failed for (%pI4, %d)->(%pI4, %d)%s L3 index %d\n",
1412				     &iph->saddr, ntohs(th->source),
1413				     &iph->daddr, ntohs(th->dest),
1414				     genhash ? " tcp_v4_calc_md5_hash failed"
1415				     : "", l3index);
1416		return true;
1417	}
1418	return false;
1419#endif
1420	return false;
1421}
1422
1423static void tcp_v4_init_req(struct request_sock *req,
1424			    const struct sock *sk_listener,
1425			    struct sk_buff *skb)
1426{
1427	struct inet_request_sock *ireq = inet_rsk(req);
1428	struct net *net = sock_net(sk_listener);
1429
1430	sk_rcv_saddr_set(req_to_sk(req), ip_hdr(skb)->daddr);
1431	sk_daddr_set(req_to_sk(req), ip_hdr(skb)->saddr);
1432	RCU_INIT_POINTER(ireq->ireq_opt, tcp_v4_save_options(net, skb));
1433}
1434
1435static struct dst_entry *tcp_v4_route_req(const struct sock *sk,
1436					  struct flowi *fl,
1437					  const struct request_sock *req)
1438{
1439	return inet_csk_route_req(sk, &fl->u.ip4, req);
1440}
1441
1442struct request_sock_ops tcp_request_sock_ops __read_mostly = {
1443	.family		=	PF_INET,
1444	.obj_size	=	sizeof(struct tcp_request_sock),
1445	.rtx_syn_ack	=	tcp_rtx_synack,
1446	.send_ack	=	tcp_v4_reqsk_send_ack,
1447	.destructor	=	tcp_v4_reqsk_destructor,
1448	.send_reset	=	tcp_v4_send_reset,
1449	.syn_ack_timeout =	tcp_syn_ack_timeout,
1450};
1451
1452const struct tcp_request_sock_ops tcp_request_sock_ipv4_ops = {
1453	.mss_clamp	=	TCP_MSS_DEFAULT,
1454#ifdef CONFIG_TCP_MD5SIG
1455	.req_md5_lookup	=	tcp_v4_md5_lookup,
 
1456	.calc_md5_hash	=	tcp_v4_md5_hash_skb,
1457#endif
1458	.init_req	=	tcp_v4_init_req,
1459#ifdef CONFIG_SYN_COOKIES
1460	.cookie_init_seq =	cookie_v4_init_sequence,
1461#endif
1462	.route_req	=	tcp_v4_route_req,
1463	.init_seq	=	tcp_v4_init_seq,
1464	.init_ts_off	=	tcp_v4_init_ts_off,
1465	.send_synack	=	tcp_v4_send_synack,
1466};
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1467
1468int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb)
1469{
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1470	/* Never answer to SYNs send to broadcast or multicast */
1471	if (skb_rtable(skb)->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))
1472		goto drop;
1473
1474	return tcp_conn_request(&tcp_request_sock_ops,
1475				&tcp_request_sock_ipv4_ops, sk, skb);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1476
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1477drop:
1478	tcp_listendrop(sk);
1479	return 0;
1480}
1481EXPORT_SYMBOL(tcp_v4_conn_request);
1482
1483
1484/*
1485 * The three way handshake has completed - we got a valid synack -
1486 * now create the new socket.
1487 */
1488struct sock *tcp_v4_syn_recv_sock(const struct sock *sk, struct sk_buff *skb,
1489				  struct request_sock *req,
1490				  struct dst_entry *dst,
1491				  struct request_sock *req_unhash,
1492				  bool *own_req)
1493{
1494	struct inet_request_sock *ireq;
1495	struct inet_sock *newinet;
1496	struct tcp_sock *newtp;
1497	struct sock *newsk;
1498#ifdef CONFIG_TCP_MD5SIG
1499	const union tcp_md5_addr *addr;
1500	struct tcp_md5sig_key *key;
1501	int l3index;
1502#endif
1503	struct ip_options_rcu *inet_opt;
1504
1505	if (sk_acceptq_is_full(sk))
1506		goto exit_overflow;
1507
1508	newsk = tcp_create_openreq_child(sk, req, skb);
1509	if (!newsk)
1510		goto exit_nonewsk;
1511
1512	newsk->sk_gso_type = SKB_GSO_TCPV4;
1513	inet_sk_rx_dst_set(newsk, skb);
1514
1515	newtp		      = tcp_sk(newsk);
1516	newinet		      = inet_sk(newsk);
1517	ireq		      = inet_rsk(req);
1518	sk_daddr_set(newsk, ireq->ir_rmt_addr);
1519	sk_rcv_saddr_set(newsk, ireq->ir_loc_addr);
1520	newsk->sk_bound_dev_if = ireq->ir_iif;
1521	newinet->inet_saddr   = ireq->ir_loc_addr;
1522	inet_opt	      = rcu_dereference(ireq->ireq_opt);
1523	RCU_INIT_POINTER(newinet->inet_opt, inet_opt);
1524	newinet->mc_index     = inet_iif(skb);
1525	newinet->mc_ttl	      = ip_hdr(skb)->ttl;
1526	newinet->rcv_tos      = ip_hdr(skb)->tos;
1527	inet_csk(newsk)->icsk_ext_hdr_len = 0;
1528	if (inet_opt)
1529		inet_csk(newsk)->icsk_ext_hdr_len = inet_opt->opt.optlen;
1530	newinet->inet_id = prandom_u32();
1531
1532	if (!dst) {
1533		dst = inet_csk_route_child_sock(sk, newsk, req);
1534		if (!dst)
1535			goto put_and_exit;
1536	} else {
1537		/* syncookie case : see end of cookie_v4_check() */
1538	}
1539	sk_setup_caps(newsk, dst);
1540
1541	tcp_ca_openreq_child(newsk, dst);
1542
1543	tcp_sync_mss(newsk, dst_mtu(dst));
1544	newtp->advmss = tcp_mss_clamp(tcp_sk(sk), dst_metric_advmss(dst));
 
 
 
1545
1546	tcp_initialize_rcv_mss(newsk);
1547
1548#ifdef CONFIG_TCP_MD5SIG
1549	l3index = l3mdev_master_ifindex_by_index(sock_net(sk), ireq->ir_iif);
1550	/* Copy over the MD5 key from the original socket */
1551	addr = (union tcp_md5_addr *)&newinet->inet_daddr;
1552	key = tcp_md5_do_lookup(sk, l3index, addr, AF_INET);
1553	if (key) {
1554		/*
1555		 * We're using one, so create a matching key
1556		 * on the newsk structure. If we fail to get
1557		 * memory, then we end up not copying the key
1558		 * across. Shucks.
1559		 */
1560		tcp_md5_do_add(newsk, addr, AF_INET, 32, l3index,
1561			       key->key, key->keylen, GFP_ATOMIC);
1562		sk_nocaps_add(newsk, NETIF_F_GSO_MASK);
1563	}
1564#endif
1565
1566	if (__inet_inherit_port(sk, newsk) < 0)
1567		goto put_and_exit;
1568	*own_req = inet_ehash_nolisten(newsk, req_to_sk(req_unhash));
1569	if (likely(*own_req)) {
1570		tcp_move_syn(newtp, req);
1571		ireq->ireq_opt = NULL;
1572	} else {
1573		newinet->inet_opt = NULL;
1574	}
1575	return newsk;
1576
1577exit_overflow:
1578	NET_INC_STATS(sock_net(sk), LINUX_MIB_LISTENOVERFLOWS);
1579exit_nonewsk:
1580	dst_release(dst);
1581exit:
1582	tcp_listendrop(sk);
1583	return NULL;
1584put_and_exit:
1585	newinet->inet_opt = NULL;
1586	inet_csk_prepare_forced_close(newsk);
1587	tcp_done(newsk);
1588	goto exit;
1589}
1590EXPORT_SYMBOL(tcp_v4_syn_recv_sock);
1591
1592static struct sock *tcp_v4_cookie_check(struct sock *sk, struct sk_buff *skb)
1593{
1594#ifdef CONFIG_SYN_COOKIES
1595	const struct tcphdr *th = tcp_hdr(skb);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1596
 
1597	if (!th->syn)
1598		sk = cookie_v4_check(sk, skb);
1599#endif
1600	return sk;
1601}
1602
1603u16 tcp_v4_get_syncookie(struct sock *sk, struct iphdr *iph,
1604			 struct tcphdr *th, u32 *cookie)
1605{
1606	u16 mss = 0;
1607#ifdef CONFIG_SYN_COOKIES
1608	mss = tcp_get_syncookie_mss(&tcp_request_sock_ops,
1609				    &tcp_request_sock_ipv4_ops, sk, th);
1610	if (mss) {
1611		*cookie = __cookie_v4_init_sequence(iph, th, &mss);
1612		tcp_synq_overflow(sk);
 
1613	}
1614#endif
1615	return mss;
 
 
 
 
 
 
1616}
1617
 
1618/* The socket must have it's spinlock held when we get
1619 * here, unless it is a TCP_LISTEN socket.
1620 *
1621 * We have a potential double-lock case here, so even when
1622 * doing backlog processing we use the BH locking scheme.
1623 * This is because we cannot sleep with the original spinlock
1624 * held.
1625 */
1626int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb)
1627{
1628	struct sock *rsk;
 
 
 
 
 
 
 
 
 
 
1629
1630	if (sk->sk_state == TCP_ESTABLISHED) { /* Fast path */
1631		struct dst_entry *dst = sk->sk_rx_dst;
1632
1633		sock_rps_save_rxhash(sk, skb);
1634		sk_mark_napi_id(sk, skb);
1635		if (dst) {
1636			if (inet_sk(sk)->rx_dst_ifindex != skb->skb_iif ||
1637			    !dst->ops->check(dst, 0)) {
1638				dst_release(dst);
1639				sk->sk_rx_dst = NULL;
1640			}
1641		}
1642		tcp_rcv_established(sk, skb);
1643		return 0;
1644	}
1645
1646	if (tcp_checksum_complete(skb))
1647		goto csum_err;
1648
1649	if (sk->sk_state == TCP_LISTEN) {
1650		struct sock *nsk = tcp_v4_cookie_check(sk, skb);
1651
1652		if (!nsk)
1653			goto discard;
 
1654		if (nsk != sk) {
 
1655			if (tcp_child_process(sk, nsk, skb)) {
1656				rsk = nsk;
1657				goto reset;
1658			}
1659			return 0;
1660		}
1661	} else
1662		sock_rps_save_rxhash(sk, skb);
1663
1664	if (tcp_rcv_state_process(sk, skb)) {
1665		rsk = sk;
1666		goto reset;
1667	}
1668	return 0;
1669
1670reset:
1671	tcp_v4_send_reset(rsk, skb);
1672discard:
1673	kfree_skb(skb);
1674	/* Be careful here. If this function gets more complicated and
1675	 * gcc suffers from register pressure on the x86, sk (in %ebx)
1676	 * might be destroyed here. This current version compiles correctly,
1677	 * but you have been warned.
1678	 */
1679	return 0;
1680
1681csum_err:
1682	TCP_INC_STATS(sock_net(sk), TCP_MIB_CSUMERRORS);
1683	TCP_INC_STATS(sock_net(sk), TCP_MIB_INERRS);
1684	goto discard;
1685}
1686EXPORT_SYMBOL(tcp_v4_do_rcv);
1687
1688int tcp_v4_early_demux(struct sk_buff *skb)
1689{
1690	const struct iphdr *iph;
1691	const struct tcphdr *th;
1692	struct sock *sk;
1693
1694	if (skb->pkt_type != PACKET_HOST)
1695		return 0;
1696
1697	if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct tcphdr)))
1698		return 0;
1699
1700	iph = ip_hdr(skb);
1701	th = tcp_hdr(skb);
1702
1703	if (th->doff < sizeof(struct tcphdr) / 4)
1704		return 0;
1705
1706	sk = __inet_lookup_established(dev_net(skb->dev), &tcp_hashinfo,
1707				       iph->saddr, th->source,
1708				       iph->daddr, ntohs(th->dest),
1709				       skb->skb_iif, inet_sdif(skb));
1710	if (sk) {
1711		skb->sk = sk;
1712		skb->destructor = sock_edemux;
1713		if (sk_fullsock(sk)) {
1714			struct dst_entry *dst = READ_ONCE(sk->sk_rx_dst);
1715
1716			if (dst)
1717				dst = dst_check(dst, 0);
1718			if (dst &&
1719			    inet_sk(sk)->rx_dst_ifindex == skb->skb_iif)
1720				skb_dst_set_noref(skb, dst);
1721		}
1722	}
1723	return 0;
1724}
1725
1726bool tcp_add_backlog(struct sock *sk, struct sk_buff *skb)
 
 
 
 
 
 
 
1727{
1728	u32 limit = READ_ONCE(sk->sk_rcvbuf) + READ_ONCE(sk->sk_sndbuf);
1729	struct skb_shared_info *shinfo;
1730	const struct tcphdr *th;
1731	struct tcphdr *thtail;
1732	struct sk_buff *tail;
1733	unsigned int hdrlen;
1734	bool fragstolen;
1735	u32 gso_segs;
1736	int delta;
1737
1738	/* In case all data was pulled from skb frags (in __pskb_pull_tail()),
1739	 * we can fix skb->truesize to its real value to avoid future drops.
1740	 * This is valid because skb is not yet charged to the socket.
1741	 * It has been noticed pure SACK packets were sometimes dropped
1742	 * (if cooked by drivers without copybreak feature).
1743	 */
1744	skb_condense(skb);
1745
1746	skb_dst_drop(skb);
1747
1748	if (unlikely(tcp_checksum_complete(skb))) {
1749		bh_unlock_sock(sk);
1750		__TCP_INC_STATS(sock_net(sk), TCP_MIB_CSUMERRORS);
1751		__TCP_INC_STATS(sock_net(sk), TCP_MIB_INERRS);
1752		return true;
1753	}
1754
1755	/* Attempt coalescing to last skb in backlog, even if we are
1756	 * above the limits.
1757	 * This is okay because skb capacity is limited to MAX_SKB_FRAGS.
1758	 */
1759	th = (const struct tcphdr *)skb->data;
1760	hdrlen = th->doff * 4;
1761	shinfo = skb_shinfo(skb);
1762
1763	if (!shinfo->gso_size)
1764		shinfo->gso_size = skb->len - hdrlen;
1765
1766	if (!shinfo->gso_segs)
1767		shinfo->gso_segs = 1;
1768
1769	tail = sk->sk_backlog.tail;
1770	if (!tail)
1771		goto no_coalesce;
1772	thtail = (struct tcphdr *)tail->data;
1773
1774	if (TCP_SKB_CB(tail)->end_seq != TCP_SKB_CB(skb)->seq ||
1775	    TCP_SKB_CB(tail)->ip_dsfield != TCP_SKB_CB(skb)->ip_dsfield ||
1776	    ((TCP_SKB_CB(tail)->tcp_flags |
1777	      TCP_SKB_CB(skb)->tcp_flags) & (TCPHDR_SYN | TCPHDR_RST | TCPHDR_URG)) ||
1778	    !((TCP_SKB_CB(tail)->tcp_flags &
1779	      TCP_SKB_CB(skb)->tcp_flags) & TCPHDR_ACK) ||
1780	    ((TCP_SKB_CB(tail)->tcp_flags ^
1781	      TCP_SKB_CB(skb)->tcp_flags) & (TCPHDR_ECE | TCPHDR_CWR)) ||
1782#ifdef CONFIG_TLS_DEVICE
1783	    tail->decrypted != skb->decrypted ||
1784#endif
1785	    thtail->doff != th->doff ||
1786	    memcmp(thtail + 1, th + 1, hdrlen - sizeof(*th)))
1787		goto no_coalesce;
1788
1789	__skb_pull(skb, hdrlen);
1790	if (skb_try_coalesce(tail, skb, &fragstolen, &delta)) {
1791		TCP_SKB_CB(tail)->end_seq = TCP_SKB_CB(skb)->end_seq;
1792
1793		if (likely(!before(TCP_SKB_CB(skb)->ack_seq, TCP_SKB_CB(tail)->ack_seq))) {
1794			TCP_SKB_CB(tail)->ack_seq = TCP_SKB_CB(skb)->ack_seq;
1795			thtail->window = th->window;
1796		}
1797
1798		/* We have to update both TCP_SKB_CB(tail)->tcp_flags and
1799		 * thtail->fin, so that the fast path in tcp_rcv_established()
1800		 * is not entered if we append a packet with a FIN.
1801		 * SYN, RST, URG are not present.
1802		 * ACK is set on both packets.
1803		 * PSH : we do not really care in TCP stack,
1804		 *       at least for 'GRO' packets.
1805		 */
1806		thtail->fin |= th->fin;
1807		TCP_SKB_CB(tail)->tcp_flags |= TCP_SKB_CB(skb)->tcp_flags;
1808
1809		if (TCP_SKB_CB(skb)->has_rxtstamp) {
1810			TCP_SKB_CB(tail)->has_rxtstamp = true;
1811			tail->tstamp = skb->tstamp;
1812			skb_hwtstamps(tail)->hwtstamp = skb_hwtstamps(skb)->hwtstamp;
1813		}
1814
1815		/* Not as strict as GRO. We only need to carry mss max value */
1816		skb_shinfo(tail)->gso_size = max(shinfo->gso_size,
1817						 skb_shinfo(tail)->gso_size);
1818
1819		gso_segs = skb_shinfo(tail)->gso_segs + shinfo->gso_segs;
1820		skb_shinfo(tail)->gso_segs = min_t(u32, gso_segs, 0xFFFF);
1821
1822		sk->sk_backlog.len += delta;
1823		__NET_INC_STATS(sock_net(sk),
1824				LINUX_MIB_TCPBACKLOGCOALESCE);
1825		kfree_skb_partial(skb, fragstolen);
1826		return false;
1827	}
1828	__skb_push(skb, hdrlen);
1829
1830no_coalesce:
1831	/* Only socket owner can try to collapse/prune rx queues
1832	 * to reduce memory overhead, so add a little headroom here.
1833	 * Few sockets backlog are possibly concurrently non empty.
1834	 */
1835	limit += 64*1024;
 
 
 
 
 
 
 
1836
1837	if (unlikely(sk_add_backlog(sk, skb, limit))) {
1838		bh_unlock_sock(sk);
1839		__NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPBACKLOGDROP);
1840		return true;
 
 
 
 
1841	}
1842	return false;
1843}
1844EXPORT_SYMBOL(tcp_add_backlog);
1845
1846int tcp_filter(struct sock *sk, struct sk_buff *skb)
1847{
1848	struct tcphdr *th = (struct tcphdr *)skb->data;
1849
1850	return sk_filter_trim_cap(sk, skb, th->doff * 4);
1851}
1852EXPORT_SYMBOL(tcp_filter);
1853
1854static void tcp_v4_restore_cb(struct sk_buff *skb)
1855{
1856	memmove(IPCB(skb), &TCP_SKB_CB(skb)->header.h4,
1857		sizeof(struct inet_skb_parm));
1858}
1859
1860static void tcp_v4_fill_cb(struct sk_buff *skb, const struct iphdr *iph,
1861			   const struct tcphdr *th)
1862{
1863	/* This is tricky : We move IPCB at its correct location into TCP_SKB_CB()
1864	 * barrier() makes sure compiler wont play fool^Waliasing games.
1865	 */
1866	memmove(&TCP_SKB_CB(skb)->header.h4, IPCB(skb),
1867		sizeof(struct inet_skb_parm));
1868	barrier();
1869
1870	TCP_SKB_CB(skb)->seq = ntohl(th->seq);
1871	TCP_SKB_CB(skb)->end_seq = (TCP_SKB_CB(skb)->seq + th->syn + th->fin +
1872				    skb->len - th->doff * 4);
1873	TCP_SKB_CB(skb)->ack_seq = ntohl(th->ack_seq);
1874	TCP_SKB_CB(skb)->tcp_flags = tcp_flag_byte(th);
1875	TCP_SKB_CB(skb)->tcp_tw_isn = 0;
1876	TCP_SKB_CB(skb)->ip_dsfield = ipv4_get_dsfield(iph);
1877	TCP_SKB_CB(skb)->sacked	 = 0;
1878	TCP_SKB_CB(skb)->has_rxtstamp =
1879			skb->tstamp || skb_hwtstamps(skb)->hwtstamp;
1880}
 
1881
1882/*
1883 *	From tcp_input.c
1884 */
1885
1886int tcp_v4_rcv(struct sk_buff *skb)
1887{
1888	struct net *net = dev_net(skb->dev);
1889	struct sk_buff *skb_to_free;
1890	int sdif = inet_sdif(skb);
1891	int dif = inet_iif(skb);
1892	const struct iphdr *iph;
1893	const struct tcphdr *th;
1894	bool refcounted;
1895	struct sock *sk;
1896	int ret;
 
1897
1898	if (skb->pkt_type != PACKET_HOST)
1899		goto discard_it;
1900
1901	/* Count it even if it's bad */
1902	__TCP_INC_STATS(net, TCP_MIB_INSEGS);
1903
1904	if (!pskb_may_pull(skb, sizeof(struct tcphdr)))
1905		goto discard_it;
1906
1907	th = (const struct tcphdr *)skb->data;
1908
1909	if (unlikely(th->doff < sizeof(struct tcphdr) / 4))
1910		goto bad_packet;
1911	if (!pskb_may_pull(skb, th->doff * 4))
1912		goto discard_it;
1913
1914	/* An explanation is required here, I think.
1915	 * Packet length and doff are validated by header prediction,
1916	 * provided case of th->doff==0 is eliminated.
1917	 * So, we defer the checks. */
1918
1919	if (skb_checksum_init(skb, IPPROTO_TCP, inet_compute_pseudo))
1920		goto csum_error;
1921
1922	th = (const struct tcphdr *)skb->data;
1923	iph = ip_hdr(skb);
1924lookup:
1925	sk = __inet_lookup_skb(&tcp_hashinfo, skb, __tcp_hdrlen(th), th->source,
1926			       th->dest, sdif, &refcounted);
 
 
 
 
 
 
1927	if (!sk)
1928		goto no_tcp_socket;
1929
1930process:
1931	if (sk->sk_state == TCP_TIME_WAIT)
1932		goto do_time_wait;
1933
1934	if (sk->sk_state == TCP_NEW_SYN_RECV) {
1935		struct request_sock *req = inet_reqsk(sk);
1936		bool req_stolen = false;
1937		struct sock *nsk;
1938
1939		sk = req->rsk_listener;
1940		if (unlikely(tcp_v4_inbound_md5_hash(sk, skb, dif, sdif))) {
1941			sk_drops_add(sk, skb);
1942			reqsk_put(req);
1943			goto discard_it;
1944		}
1945		if (tcp_checksum_complete(skb)) {
1946			reqsk_put(req);
1947			goto csum_error;
1948		}
1949		if (unlikely(sk->sk_state != TCP_LISTEN)) {
1950			inet_csk_reqsk_queue_drop_and_put(sk, req);
1951			goto lookup;
1952		}
1953		/* We own a reference on the listener, increase it again
1954		 * as we might lose it too soon.
1955		 */
1956		sock_hold(sk);
1957		refcounted = true;
1958		nsk = NULL;
1959		if (!tcp_filter(sk, skb)) {
1960			th = (const struct tcphdr *)skb->data;
1961			iph = ip_hdr(skb);
1962			tcp_v4_fill_cb(skb, iph, th);
1963			nsk = tcp_check_req(sk, skb, req, false, &req_stolen);
1964		}
1965		if (!nsk) {
1966			reqsk_put(req);
1967			if (req_stolen) {
1968				/* Another cpu got exclusive access to req
1969				 * and created a full blown socket.
1970				 * Try to feed this packet to this socket
1971				 * instead of discarding it.
1972				 */
1973				tcp_v4_restore_cb(skb);
1974				sock_put(sk);
1975				goto lookup;
1976			}
1977			goto discard_and_relse;
1978		}
1979		if (nsk == sk) {
1980			reqsk_put(req);
1981			tcp_v4_restore_cb(skb);
1982		} else if (tcp_child_process(sk, nsk, skb)) {
1983			tcp_v4_send_reset(nsk, skb);
1984			goto discard_and_relse;
1985		} else {
1986			sock_put(sk);
1987			return 0;
1988		}
1989	}
1990	if (unlikely(iph->ttl < inet_sk(sk)->min_ttl)) {
1991		__NET_INC_STATS(net, LINUX_MIB_TCPMINTTLDROP);
1992		goto discard_and_relse;
1993	}
1994
1995	if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
1996		goto discard_and_relse;
 
1997
1998	if (tcp_v4_inbound_md5_hash(sk, skb, dif, sdif))
1999		goto discard_and_relse;
2000
2001	nf_reset_ct(skb);
2002
2003	if (tcp_filter(sk, skb))
2004		goto discard_and_relse;
2005	th = (const struct tcphdr *)skb->data;
2006	iph = ip_hdr(skb);
2007	tcp_v4_fill_cb(skb, iph, th);
2008
2009	skb->dev = NULL;
2010
2011	if (sk->sk_state == TCP_LISTEN) {
2012		ret = tcp_v4_do_rcv(sk, skb);
2013		goto put_and_return;
2014	}
2015
2016	sk_incoming_cpu_update(sk);
2017
2018	bh_lock_sock_nested(sk);
2019	tcp_segs_in(tcp_sk(sk), skb);
2020	ret = 0;
2021	if (!sock_owned_by_user(sk)) {
2022		skb_to_free = sk->sk_rx_skb_cache;
2023		sk->sk_rx_skb_cache = NULL;
2024		ret = tcp_v4_do_rcv(sk, skb);
2025	} else {
2026		if (tcp_add_backlog(sk, skb))
2027			goto discard_and_relse;
2028		skb_to_free = NULL;
 
 
 
 
 
 
 
 
 
 
2029	}
2030	bh_unlock_sock(sk);
2031	if (skb_to_free)
2032		__kfree_skb(skb_to_free);
2033
2034put_and_return:
2035	if (refcounted)
2036		sock_put(sk);
2037
2038	return ret;
2039
2040no_tcp_socket:
2041	if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
2042		goto discard_it;
2043
2044	tcp_v4_fill_cb(skb, iph, th);
2045
2046	if (tcp_checksum_complete(skb)) {
2047csum_error:
2048		__TCP_INC_STATS(net, TCP_MIB_CSUMERRORS);
2049bad_packet:
2050		__TCP_INC_STATS(net, TCP_MIB_INERRS);
2051	} else {
2052		tcp_v4_send_reset(NULL, skb);
2053	}
2054
2055discard_it:
2056	/* Discard frame. */
2057	kfree_skb(skb);
2058	return 0;
2059
2060discard_and_relse:
2061	sk_drops_add(sk, skb);
2062	if (refcounted)
2063		sock_put(sk);
2064	goto discard_it;
2065
2066do_time_wait:
2067	if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) {
2068		inet_twsk_put(inet_twsk(sk));
2069		goto discard_it;
2070	}
2071
2072	tcp_v4_fill_cb(skb, iph, th);
2073
 
 
2074	if (tcp_checksum_complete(skb)) {
2075		inet_twsk_put(inet_twsk(sk));
2076		goto csum_error;
2077	}
2078	switch (tcp_timewait_state_process(inet_twsk(sk), skb, th)) {
2079	case TCP_TW_SYN: {
2080		struct sock *sk2 = inet_lookup_listener(dev_net(skb->dev),
2081							&tcp_hashinfo, skb,
2082							__tcp_hdrlen(th),
2083							iph->saddr, th->source,
2084							iph->daddr, th->dest,
2085							inet_iif(skb),
2086							sdif);
2087		if (sk2) {
2088			inet_twsk_deschedule_put(inet_twsk(sk));
 
2089			sk = sk2;
2090			tcp_v4_restore_cb(skb);
2091			refcounted = false;
2092			goto process;
2093		}
 
2094	}
2095		/* to ACK */
2096		fallthrough;
2097	case TCP_TW_ACK:
2098		tcp_v4_timewait_ack(sk, skb);
2099		break;
2100	case TCP_TW_RST:
2101		tcp_v4_send_reset(sk, skb);
2102		inet_twsk_deschedule_put(inet_twsk(sk));
2103		goto discard_it;
2104	case TCP_TW_SUCCESS:;
2105	}
2106	goto discard_it;
2107}
2108
2109static struct timewait_sock_ops tcp_timewait_sock_ops = {
2110	.twsk_obj_size	= sizeof(struct tcp_timewait_sock),
2111	.twsk_unique	= tcp_twsk_unique,
2112	.twsk_destructor= tcp_twsk_destructor,
2113};
2114
2115void inet_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb)
2116{
2117	struct dst_entry *dst = skb_dst(skb);
2118
2119	if (dst && dst_hold_safe(dst)) {
2120		sk->sk_rx_dst = dst;
2121		inet_sk(sk)->rx_dst_ifindex = skb->skb_iif;
2122	}
2123}
2124EXPORT_SYMBOL(inet_sk_rx_dst_set);
2125
2126const struct inet_connection_sock_af_ops ipv4_specific = {
2127	.queue_xmit	   = ip_queue_xmit,
2128	.send_check	   = tcp_v4_send_check,
2129	.rebuild_header	   = inet_sk_rebuild_header,
2130	.sk_rx_dst_set	   = inet_sk_rx_dst_set,
2131	.conn_request	   = tcp_v4_conn_request,
2132	.syn_recv_sock	   = tcp_v4_syn_recv_sock,
2133	.net_header_len	   = sizeof(struct iphdr),
2134	.setsockopt	   = ip_setsockopt,
2135	.getsockopt	   = ip_getsockopt,
2136	.addr2sockaddr	   = inet_csk_addr2sockaddr,
2137	.sockaddr_len	   = sizeof(struct sockaddr_in),
2138	.mtu_reduced	   = tcp_v4_mtu_reduced,
 
 
 
 
2139};
2140EXPORT_SYMBOL(ipv4_specific);
2141
2142#ifdef CONFIG_TCP_MD5SIG
2143static const struct tcp_sock_af_ops tcp_sock_ipv4_specific = {
2144	.md5_lookup		= tcp_v4_md5_lookup,
2145	.calc_md5_hash		= tcp_v4_md5_hash_skb,
2146	.md5_parse		= tcp_v4_parse_md5_keys,
2147};
2148#endif
2149
2150/* NOTE: A lot of things set to zero explicitly by call to
2151 *       sk_alloc() so need not be done here.
2152 */
2153static int tcp_v4_init_sock(struct sock *sk)
2154{
2155	struct inet_connection_sock *icsk = inet_csk(sk);
2156
2157	tcp_init_sock(sk);
2158
2159	icsk->icsk_af_ops = &ipv4_specific;
2160
2161#ifdef CONFIG_TCP_MD5SIG
2162	tcp_sk(sk)->af_specific = &tcp_sock_ipv4_specific;
2163#endif
2164
2165	return 0;
2166}
2167
2168void tcp_v4_destroy_sock(struct sock *sk)
2169{
2170	struct tcp_sock *tp = tcp_sk(sk);
2171
2172	trace_tcp_destroy_sock(sk);
2173
2174	tcp_clear_xmit_timers(sk);
2175
2176	tcp_cleanup_congestion_control(sk);
2177
2178	tcp_cleanup_ulp(sk);
2179
2180	/* Cleanup up the write buffer. */
2181	tcp_write_queue_purge(sk);
2182
2183	/* Check if we want to disable active TFO */
2184	tcp_fastopen_active_disable_ofo_check(sk);
2185
2186	/* Cleans up our, hopefully empty, out_of_order_queue. */
2187	skb_rbtree_purge(&tp->out_of_order_queue);
2188
2189#ifdef CONFIG_TCP_MD5SIG
2190	/* Clean up the MD5 key list, if any */
2191	if (tp->md5sig_info) {
2192		tcp_clear_md5_list(sk);
2193		kfree_rcu(rcu_dereference_protected(tp->md5sig_info, 1), rcu);
2194		tp->md5sig_info = NULL;
2195	}
2196#endif
2197
 
 
 
 
 
 
 
 
2198	/* Clean up a referenced TCP bind bucket. */
2199	if (inet_csk(sk)->icsk_bind_hash)
2200		inet_put_port(sk);
2201
2202	BUG_ON(rcu_access_pointer(tp->fastopen_rsk));
2203
2204	/* If socket is aborted during connect operation */
2205	tcp_free_fastopen_req(tp);
2206	tcp_fastopen_destroy_cipher(sk);
2207	tcp_saved_syn_free(tp);
2208
2209	sk_sockets_allocated_dec(sk);
 
2210}
2211EXPORT_SYMBOL(tcp_v4_destroy_sock);
2212
2213#ifdef CONFIG_PROC_FS
2214/* Proc filesystem TCP sock list dumping. */
2215
2216/*
2217 * Get next listener socket follow cur.  If cur is NULL, get first socket
2218 * starting from bucket given in st->bucket; when st->bucket is zero the
2219 * very first socket in the hash table is returned.
2220 */
2221static void *listening_get_next(struct seq_file *seq, void *cur)
2222{
2223	struct tcp_seq_afinfo *afinfo;
2224	struct tcp_iter_state *st = seq->private;
2225	struct net *net = seq_file_net(seq);
2226	struct inet_listen_hashbucket *ilb;
2227	struct hlist_nulls_node *node;
2228	struct sock *sk = cur;
2229
2230	if (st->bpf_seq_afinfo)
2231		afinfo = st->bpf_seq_afinfo;
2232	else
2233		afinfo = PDE_DATA(file_inode(seq->file));
2234
2235	if (!sk) {
2236get_head:
2237		ilb = &tcp_hashinfo.listening_hash[st->bucket];
2238		spin_lock(&ilb->lock);
2239		sk = sk_nulls_head(&ilb->nulls_head);
2240		st->offset = 0;
2241		goto get_sk;
2242	}
2243	ilb = &tcp_hashinfo.listening_hash[st->bucket];
2244	++st->num;
2245	++st->offset;
2246
2247	sk = sk_nulls_next(sk);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2248get_sk:
2249	sk_nulls_for_each_from(sk, node) {
2250		if (!net_eq(sock_net(sk), net))
2251			continue;
2252		if (afinfo->family == AF_UNSPEC ||
2253		    sk->sk_family == afinfo->family)
2254			return sk;
 
 
 
 
 
 
 
 
 
 
 
 
2255	}
2256	spin_unlock(&ilb->lock);
2257	st->offset = 0;
2258	if (++st->bucket < INET_LHTABLE_SIZE)
2259		goto get_head;
2260	return NULL;
 
 
 
 
 
 
2261}
2262
2263static void *listening_get_idx(struct seq_file *seq, loff_t *pos)
2264{
2265	struct tcp_iter_state *st = seq->private;
2266	void *rc;
2267
2268	st->bucket = 0;
2269	st->offset = 0;
2270	rc = listening_get_next(seq, NULL);
2271
2272	while (rc && *pos) {
2273		rc = listening_get_next(seq, rc);
2274		--*pos;
2275	}
2276	return rc;
2277}
2278
2279static inline bool empty_bucket(const struct tcp_iter_state *st)
2280{
2281	return hlist_nulls_empty(&tcp_hashinfo.ehash[st->bucket].chain);
2282}
2283
2284/*
2285 * Get first established socket starting from bucket given in st->bucket.
2286 * If st->bucket is zero, the very first socket in the hash is returned.
2287 */
2288static void *established_get_first(struct seq_file *seq)
2289{
2290	struct tcp_seq_afinfo *afinfo;
2291	struct tcp_iter_state *st = seq->private;
2292	struct net *net = seq_file_net(seq);
2293	void *rc = NULL;
2294
2295	if (st->bpf_seq_afinfo)
2296		afinfo = st->bpf_seq_afinfo;
2297	else
2298		afinfo = PDE_DATA(file_inode(seq->file));
2299
2300	st->offset = 0;
2301	for (; st->bucket <= tcp_hashinfo.ehash_mask; ++st->bucket) {
2302		struct sock *sk;
2303		struct hlist_nulls_node *node;
2304		spinlock_t *lock = inet_ehash_lockp(&tcp_hashinfo, st->bucket);
2305
2306		/* Lockless fast path for the common case of empty buckets */
2307		if (empty_bucket(st))
2308			continue;
2309
2310		spin_lock_bh(lock);
2311		sk_nulls_for_each(sk, node, &tcp_hashinfo.ehash[st->bucket].chain) {
2312			if ((afinfo->family != AF_UNSPEC &&
2313			     sk->sk_family != afinfo->family) ||
2314			    !net_eq(sock_net(sk), net)) {
2315				continue;
2316			}
2317			rc = sk;
2318			goto out;
2319		}
2320		spin_unlock_bh(lock);
2321	}
2322out:
2323	return rc;
2324}
2325
2326static void *established_get_next(struct seq_file *seq, void *cur)
2327{
2328	struct tcp_seq_afinfo *afinfo;
2329	struct sock *sk = cur;
2330	struct hlist_nulls_node *node;
2331	struct tcp_iter_state *st = seq->private;
2332	struct net *net = seq_file_net(seq);
2333
2334	if (st->bpf_seq_afinfo)
2335		afinfo = st->bpf_seq_afinfo;
2336	else
2337		afinfo = PDE_DATA(file_inode(seq->file));
2338
2339	++st->num;
2340	++st->offset;
2341
2342	sk = sk_nulls_next(sk);
2343
2344	sk_nulls_for_each_from(sk, node) {
2345		if ((afinfo->family == AF_UNSPEC ||
2346		     sk->sk_family == afinfo->family) &&
2347		    net_eq(sock_net(sk), net))
2348			return sk;
2349	}
2350
2351	spin_unlock_bh(inet_ehash_lockp(&tcp_hashinfo, st->bucket));
2352	++st->bucket;
2353	return established_get_first(seq);
2354}
2355
2356static void *established_get_idx(struct seq_file *seq, loff_t pos)
2357{
2358	struct tcp_iter_state *st = seq->private;
2359	void *rc;
2360
2361	st->bucket = 0;
2362	rc = established_get_first(seq);
2363
2364	while (rc && pos) {
2365		rc = established_get_next(seq, rc);
2366		--pos;
2367	}
2368	return rc;
2369}
2370
2371static void *tcp_get_idx(struct seq_file *seq, loff_t pos)
2372{
2373	void *rc;
2374	struct tcp_iter_state *st = seq->private;
2375
2376	st->state = TCP_SEQ_STATE_LISTENING;
2377	rc	  = listening_get_idx(seq, &pos);
2378
2379	if (!rc) {
2380		st->state = TCP_SEQ_STATE_ESTABLISHED;
2381		rc	  = established_get_idx(seq, pos);
2382	}
2383
2384	return rc;
2385}
2386
2387static void *tcp_seek_last_pos(struct seq_file *seq)
2388{
2389	struct tcp_iter_state *st = seq->private;
2390	int offset = st->offset;
2391	int orig_num = st->num;
2392	void *rc = NULL;
2393
2394	switch (st->state) {
 
2395	case TCP_SEQ_STATE_LISTENING:
2396		if (st->bucket >= INET_LHTABLE_SIZE)
2397			break;
2398		st->state = TCP_SEQ_STATE_LISTENING;
2399		rc = listening_get_next(seq, NULL);
2400		while (offset-- && rc)
2401			rc = listening_get_next(seq, rc);
2402		if (rc)
2403			break;
2404		st->bucket = 0;
2405		st->state = TCP_SEQ_STATE_ESTABLISHED;
2406		fallthrough;
2407	case TCP_SEQ_STATE_ESTABLISHED:
2408		if (st->bucket > tcp_hashinfo.ehash_mask)
2409			break;
2410		rc = established_get_first(seq);
2411		while (offset-- && rc)
2412			rc = established_get_next(seq, rc);
2413	}
2414
2415	st->num = orig_num;
2416
2417	return rc;
2418}
2419
2420void *tcp_seq_start(struct seq_file *seq, loff_t *pos)
2421{
2422	struct tcp_iter_state *st = seq->private;
2423	void *rc;
2424
2425	if (*pos && *pos == st->last_pos) {
2426		rc = tcp_seek_last_pos(seq);
2427		if (rc)
2428			goto out;
2429	}
2430
2431	st->state = TCP_SEQ_STATE_LISTENING;
2432	st->num = 0;
2433	st->bucket = 0;
2434	st->offset = 0;
2435	rc = *pos ? tcp_get_idx(seq, *pos - 1) : SEQ_START_TOKEN;
2436
2437out:
2438	st->last_pos = *pos;
2439	return rc;
2440}
2441EXPORT_SYMBOL(tcp_seq_start);
2442
2443void *tcp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2444{
2445	struct tcp_iter_state *st = seq->private;
2446	void *rc = NULL;
2447
2448	if (v == SEQ_START_TOKEN) {
2449		rc = tcp_get_idx(seq, 0);
2450		goto out;
2451	}
2452
2453	switch (st->state) {
 
2454	case TCP_SEQ_STATE_LISTENING:
2455		rc = listening_get_next(seq, v);
2456		if (!rc) {
2457			st->state = TCP_SEQ_STATE_ESTABLISHED;
2458			st->bucket = 0;
2459			st->offset = 0;
2460			rc	  = established_get_first(seq);
2461		}
2462		break;
2463	case TCP_SEQ_STATE_ESTABLISHED:
2464		rc = established_get_next(seq, v);
2465		break;
2466	}
2467out:
2468	++*pos;
2469	st->last_pos = *pos;
2470	return rc;
2471}
2472EXPORT_SYMBOL(tcp_seq_next);
2473
2474void tcp_seq_stop(struct seq_file *seq, void *v)
2475{
2476	struct tcp_iter_state *st = seq->private;
2477
2478	switch (st->state) {
 
 
 
 
 
2479	case TCP_SEQ_STATE_LISTENING:
2480		if (v != SEQ_START_TOKEN)
2481			spin_unlock(&tcp_hashinfo.listening_hash[st->bucket].lock);
2482		break;
2483	case TCP_SEQ_STATE_ESTABLISHED:
2484		if (v)
2485			spin_unlock_bh(inet_ehash_lockp(&tcp_hashinfo, st->bucket));
2486		break;
2487	}
2488}
2489EXPORT_SYMBOL(tcp_seq_stop);
2490
2491static void get_openreq4(const struct request_sock *req,
2492			 struct seq_file *f, int i)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2493{
2494	const struct inet_request_sock *ireq = inet_rsk(req);
2495	long delta = req->rsk_timer.expires - jiffies;
2496
2497	seq_printf(f, "%4d: %08X:%04X %08X:%04X"
2498		" %02X %08X:%08X %02X:%08lX %08X %5u %8d %u %d %pK",
2499		i,
2500		ireq->ir_loc_addr,
2501		ireq->ir_num,
2502		ireq->ir_rmt_addr,
2503		ntohs(ireq->ir_rmt_port),
2504		TCP_SYN_RECV,
2505		0, 0, /* could print option size, but that is af dependent. */
2506		1,    /* timers active (only the expire timer) */
2507		jiffies_delta_to_clock_t(delta),
2508		req->num_timeout,
2509		from_kuid_munged(seq_user_ns(f),
2510				 sock_i_uid(req->rsk_listener)),
2511		0,  /* non standard timer */
2512		0, /* open_requests have no inode */
2513		0,
2514		req);
2515}
2516
2517static void get_tcp4_sock(struct sock *sk, struct seq_file *f, int i)
2518{
2519	int timer_active;
2520	unsigned long timer_expires;
2521	const struct tcp_sock *tp = tcp_sk(sk);
2522	const struct inet_connection_sock *icsk = inet_csk(sk);
2523	const struct inet_sock *inet = inet_sk(sk);
2524	const struct fastopen_queue *fastopenq = &icsk->icsk_accept_queue.fastopenq;
2525	__be32 dest = inet->inet_daddr;
2526	__be32 src = inet->inet_rcv_saddr;
2527	__u16 destp = ntohs(inet->inet_dport);
2528	__u16 srcp = ntohs(inet->inet_sport);
2529	int rx_queue;
2530	int state;
2531
2532	if (icsk->icsk_pending == ICSK_TIME_RETRANS ||
2533	    icsk->icsk_pending == ICSK_TIME_REO_TIMEOUT ||
2534	    icsk->icsk_pending == ICSK_TIME_LOSS_PROBE) {
2535		timer_active	= 1;
2536		timer_expires	= icsk->icsk_timeout;
2537	} else if (icsk->icsk_pending == ICSK_TIME_PROBE0) {
2538		timer_active	= 4;
2539		timer_expires	= icsk->icsk_timeout;
2540	} else if (timer_pending(&sk->sk_timer)) {
2541		timer_active	= 2;
2542		timer_expires	= sk->sk_timer.expires;
2543	} else {
2544		timer_active	= 0;
2545		timer_expires = jiffies;
2546	}
2547
2548	state = inet_sk_state_load(sk);
2549	if (state == TCP_LISTEN)
2550		rx_queue = READ_ONCE(sk->sk_ack_backlog);
2551	else
2552		/* Because we don't lock the socket,
2553		 * we might find a transient negative value.
2554		 */
2555		rx_queue = max_t(int, READ_ONCE(tp->rcv_nxt) -
2556				      READ_ONCE(tp->copied_seq), 0);
2557
2558	seq_printf(f, "%4d: %08X:%04X %08X:%04X %02X %08X:%08X %02X:%08lX "
2559			"%08X %5u %8d %lu %d %pK %lu %lu %u %u %d",
2560		i, src, srcp, dest, destp, state,
2561		READ_ONCE(tp->write_seq) - tp->snd_una,
2562		rx_queue,
2563		timer_active,
2564		jiffies_delta_to_clock_t(timer_expires - jiffies),
2565		icsk->icsk_retransmits,
2566		from_kuid_munged(seq_user_ns(f), sock_i_uid(sk)),
2567		icsk->icsk_probes_out,
2568		sock_i_ino(sk),
2569		refcount_read(&sk->sk_refcnt), sk,
2570		jiffies_to_clock_t(icsk->icsk_rto),
2571		jiffies_to_clock_t(icsk->icsk_ack.ato),
2572		(icsk->icsk_ack.quick << 1) | inet_csk_in_pingpong_mode(sk),
2573		tp->snd_cwnd,
2574		state == TCP_LISTEN ?
2575		    fastopenq->max_qlen :
2576		    (tcp_in_initial_slowstart(tp) ? -1 : tp->snd_ssthresh));
2577}
2578
2579static void get_timewait4_sock(const struct inet_timewait_sock *tw,
2580			       struct seq_file *f, int i)
2581{
2582	long delta = tw->tw_timer.expires - jiffies;
2583	__be32 dest, src;
2584	__u16 destp, srcp;
 
2585
2586	dest  = tw->tw_daddr;
2587	src   = tw->tw_rcv_saddr;
2588	destp = ntohs(tw->tw_dport);
2589	srcp  = ntohs(tw->tw_sport);
2590
2591	seq_printf(f, "%4d: %08X:%04X %08X:%04X"
2592		" %02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %pK",
2593		i, src, srcp, dest, destp, tw->tw_substate, 0, 0,
2594		3, jiffies_delta_to_clock_t(delta), 0, 0, 0, 0,
2595		refcount_read(&tw->tw_refcnt), tw);
2596}
2597
2598#define TMPSZ 150
2599
2600static int tcp4_seq_show(struct seq_file *seq, void *v)
2601{
2602	struct tcp_iter_state *st;
2603	struct sock *sk = v;
2604
2605	seq_setwidth(seq, TMPSZ - 1);
2606	if (v == SEQ_START_TOKEN) {
2607		seq_puts(seq, "  sl  local_address rem_address   st tx_queue "
2608			   "rx_queue tr tm->when retrnsmt   uid  timeout "
2609			   "inode");
2610		goto out;
2611	}
2612	st = seq->private;
2613
2614	if (sk->sk_state == TCP_TIME_WAIT)
2615		get_timewait4_sock(v, seq, st->num);
2616	else if (sk->sk_state == TCP_NEW_SYN_RECV)
2617		get_openreq4(v, seq, st->num);
2618	else
2619		get_tcp4_sock(v, seq, st->num);
 
 
 
 
 
 
2620out:
2621	seq_pad(seq, '\n');
2622	return 0;
2623}
2624
2625#ifdef CONFIG_BPF_SYSCALL
2626struct bpf_iter__tcp {
2627	__bpf_md_ptr(struct bpf_iter_meta *, meta);
2628	__bpf_md_ptr(struct sock_common *, sk_common);
2629	uid_t uid __aligned(8);
2630};
2631
2632static int tcp_prog_seq_show(struct bpf_prog *prog, struct bpf_iter_meta *meta,
2633			     struct sock_common *sk_common, uid_t uid)
2634{
2635	struct bpf_iter__tcp ctx;
2636
2637	meta->seq_num--;  /* skip SEQ_START_TOKEN */
2638	ctx.meta = meta;
2639	ctx.sk_common = sk_common;
2640	ctx.uid = uid;
2641	return bpf_iter_run_prog(prog, &ctx);
2642}
2643
2644static int bpf_iter_tcp_seq_show(struct seq_file *seq, void *v)
2645{
2646	struct bpf_iter_meta meta;
2647	struct bpf_prog *prog;
2648	struct sock *sk = v;
2649	uid_t uid;
2650
2651	if (v == SEQ_START_TOKEN)
2652		return 0;
2653
2654	if (sk->sk_state == TCP_TIME_WAIT) {
2655		uid = 0;
2656	} else if (sk->sk_state == TCP_NEW_SYN_RECV) {
2657		const struct request_sock *req = v;
2658
2659		uid = from_kuid_munged(seq_user_ns(seq),
2660				       sock_i_uid(req->rsk_listener));
2661	} else {
2662		uid = from_kuid_munged(seq_user_ns(seq), sock_i_uid(sk));
2663	}
2664
2665	meta.seq = seq;
2666	prog = bpf_iter_get_info(&meta, false);
2667	return tcp_prog_seq_show(prog, &meta, v, uid);
2668}
2669
2670static void bpf_iter_tcp_seq_stop(struct seq_file *seq, void *v)
2671{
2672	struct bpf_iter_meta meta;
2673	struct bpf_prog *prog;
2674
2675	if (!v) {
2676		meta.seq = seq;
2677		prog = bpf_iter_get_info(&meta, true);
2678		if (prog)
2679			(void)tcp_prog_seq_show(prog, &meta, v, 0);
2680	}
2681
2682	tcp_seq_stop(seq, v);
2683}
2684
2685static const struct seq_operations bpf_iter_tcp_seq_ops = {
2686	.show		= bpf_iter_tcp_seq_show,
2687	.start		= tcp_seq_start,
2688	.next		= tcp_seq_next,
2689	.stop		= bpf_iter_tcp_seq_stop,
2690};
2691#endif
2692
2693static const struct seq_operations tcp4_seq_ops = {
2694	.show		= tcp4_seq_show,
2695	.start		= tcp_seq_start,
2696	.next		= tcp_seq_next,
2697	.stop		= tcp_seq_stop,
2698};
2699
2700static struct tcp_seq_afinfo tcp4_seq_afinfo = {
 
2701	.family		= AF_INET,
 
 
 
 
2702};
2703
2704static int __net_init tcp4_proc_init_net(struct net *net)
2705{
2706	if (!proc_create_net_data("tcp", 0444, net->proc_net, &tcp4_seq_ops,
2707			sizeof(struct tcp_iter_state), &tcp4_seq_afinfo))
2708		return -ENOMEM;
2709	return 0;
2710}
2711
2712static void __net_exit tcp4_proc_exit_net(struct net *net)
2713{
2714	remove_proc_entry("tcp", net->proc_net);
2715}
2716
2717static struct pernet_operations tcp4_net_ops = {
2718	.init = tcp4_proc_init_net,
2719	.exit = tcp4_proc_exit_net,
2720};
2721
2722int __init tcp4_proc_init(void)
2723{
2724	return register_pernet_subsys(&tcp4_net_ops);
2725}
2726
2727void tcp4_proc_exit(void)
2728{
2729	unregister_pernet_subsys(&tcp4_net_ops);
2730}
2731#endif /* CONFIG_PROC_FS */
2732
2733struct proto tcp_prot = {
2734	.name			= "TCP",
2735	.owner			= THIS_MODULE,
2736	.close			= tcp_close,
2737	.pre_connect		= tcp_v4_pre_connect,
2738	.connect		= tcp_v4_connect,
2739	.disconnect		= tcp_disconnect,
2740	.accept			= inet_csk_accept,
2741	.ioctl			= tcp_ioctl,
2742	.init			= tcp_v4_init_sock,
2743	.destroy		= tcp_v4_destroy_sock,
2744	.shutdown		= tcp_shutdown,
2745	.setsockopt		= tcp_setsockopt,
2746	.getsockopt		= tcp_getsockopt,
2747	.keepalive		= tcp_set_keepalive,
2748	.recvmsg		= tcp_recvmsg,
2749	.sendmsg		= tcp_sendmsg,
2750	.sendpage		= tcp_sendpage,
2751	.backlog_rcv		= tcp_v4_do_rcv,
2752	.release_cb		= tcp_release_cb,
 
2753	.hash			= inet_hash,
2754	.unhash			= inet_unhash,
2755	.get_port		= inet_csk_get_port,
2756	.enter_memory_pressure	= tcp_enter_memory_pressure,
2757	.leave_memory_pressure	= tcp_leave_memory_pressure,
2758	.stream_memory_free	= tcp_stream_memory_free,
2759	.sockets_allocated	= &tcp_sockets_allocated,
2760	.orphan_count		= &tcp_orphan_count,
2761	.memory_allocated	= &tcp_memory_allocated,
2762	.memory_pressure	= &tcp_memory_pressure,
2763	.sysctl_mem		= sysctl_tcp_mem,
2764	.sysctl_wmem_offset	= offsetof(struct net, ipv4.sysctl_tcp_wmem),
2765	.sysctl_rmem_offset	= offsetof(struct net, ipv4.sysctl_tcp_rmem),
2766	.max_header		= MAX_TCP_HEADER,
2767	.obj_size		= sizeof(struct tcp_sock),
2768	.slab_flags		= SLAB_TYPESAFE_BY_RCU,
2769	.twsk_prot		= &tcp_timewait_sock_ops,
2770	.rsk_prot		= &tcp_request_sock_ops,
2771	.h.hashinfo		= &tcp_hashinfo,
2772	.no_autobind		= true,
2773	.diag_destroy		= tcp_abort,
 
 
 
 
 
 
 
 
2774};
2775EXPORT_SYMBOL(tcp_prot);
2776
2777static void __net_exit tcp_sk_exit(struct net *net)
2778{
2779	int cpu;
2780
2781	if (net->ipv4.tcp_congestion_control)
2782		bpf_module_put(net->ipv4.tcp_congestion_control,
2783			       net->ipv4.tcp_congestion_control->owner);
2784
2785	for_each_possible_cpu(cpu)
2786		inet_ctl_sock_destroy(*per_cpu_ptr(net->ipv4.tcp_sk, cpu));
2787	free_percpu(net->ipv4.tcp_sk);
2788}
2789
2790static int __net_init tcp_sk_init(struct net *net)
2791{
2792	int res, cpu, cnt;
2793
2794	net->ipv4.tcp_sk = alloc_percpu(struct sock *);
2795	if (!net->ipv4.tcp_sk)
2796		return -ENOMEM;
2797
2798	for_each_possible_cpu(cpu) {
2799		struct sock *sk;
2800
2801		res = inet_ctl_sock_create(&sk, PF_INET, SOCK_RAW,
2802					   IPPROTO_TCP, net);
2803		if (res)
2804			goto fail;
2805		sock_set_flag(sk, SOCK_USE_WRITE_QUEUE);
2806
2807		/* Please enforce IP_DF and IPID==0 for RST and
2808		 * ACK sent in SYN-RECV and TIME-WAIT state.
2809		 */
2810		inet_sk(sk)->pmtudisc = IP_PMTUDISC_DO;
2811
2812		*per_cpu_ptr(net->ipv4.tcp_sk, cpu) = sk;
2813	}
2814
2815	net->ipv4.sysctl_tcp_ecn = 2;
2816	net->ipv4.sysctl_tcp_ecn_fallback = 1;
2817
2818	net->ipv4.sysctl_tcp_base_mss = TCP_BASE_MSS;
2819	net->ipv4.sysctl_tcp_min_snd_mss = TCP_MIN_SND_MSS;
2820	net->ipv4.sysctl_tcp_probe_threshold = TCP_PROBE_THRESHOLD;
2821	net->ipv4.sysctl_tcp_probe_interval = TCP_PROBE_INTERVAL;
2822	net->ipv4.sysctl_tcp_mtu_probe_floor = TCP_MIN_SND_MSS;
2823
2824	net->ipv4.sysctl_tcp_keepalive_time = TCP_KEEPALIVE_TIME;
2825	net->ipv4.sysctl_tcp_keepalive_probes = TCP_KEEPALIVE_PROBES;
2826	net->ipv4.sysctl_tcp_keepalive_intvl = TCP_KEEPALIVE_INTVL;
2827
2828	net->ipv4.sysctl_tcp_syn_retries = TCP_SYN_RETRIES;
2829	net->ipv4.sysctl_tcp_synack_retries = TCP_SYNACK_RETRIES;
2830	net->ipv4.sysctl_tcp_syncookies = 1;
2831	net->ipv4.sysctl_tcp_reordering = TCP_FASTRETRANS_THRESH;
2832	net->ipv4.sysctl_tcp_retries1 = TCP_RETR1;
2833	net->ipv4.sysctl_tcp_retries2 = TCP_RETR2;
2834	net->ipv4.sysctl_tcp_orphan_retries = 0;
2835	net->ipv4.sysctl_tcp_fin_timeout = TCP_FIN_TIMEOUT;
2836	net->ipv4.sysctl_tcp_notsent_lowat = UINT_MAX;
2837	net->ipv4.sysctl_tcp_tw_reuse = 2;
2838	net->ipv4.sysctl_tcp_no_ssthresh_metrics_save = 1;
2839
2840	cnt = tcp_hashinfo.ehash_mask + 1;
2841	net->ipv4.tcp_death_row.sysctl_max_tw_buckets = cnt / 2;
2842	net->ipv4.tcp_death_row.hashinfo = &tcp_hashinfo;
2843
2844	net->ipv4.sysctl_max_syn_backlog = max(128, cnt / 128);
2845	net->ipv4.sysctl_tcp_sack = 1;
2846	net->ipv4.sysctl_tcp_window_scaling = 1;
2847	net->ipv4.sysctl_tcp_timestamps = 1;
2848	net->ipv4.sysctl_tcp_early_retrans = 3;
2849	net->ipv4.sysctl_tcp_recovery = TCP_RACK_LOSS_DETECTION;
2850	net->ipv4.sysctl_tcp_slow_start_after_idle = 1; /* By default, RFC2861 behavior.  */
2851	net->ipv4.sysctl_tcp_retrans_collapse = 1;
2852	net->ipv4.sysctl_tcp_max_reordering = 300;
2853	net->ipv4.sysctl_tcp_dsack = 1;
2854	net->ipv4.sysctl_tcp_app_win = 31;
2855	net->ipv4.sysctl_tcp_adv_win_scale = 1;
2856	net->ipv4.sysctl_tcp_frto = 2;
2857	net->ipv4.sysctl_tcp_moderate_rcvbuf = 1;
2858	/* This limits the percentage of the congestion window which we
2859	 * will allow a single TSO frame to consume.  Building TSO frames
2860	 * which are too large can cause TCP streams to be bursty.
2861	 */
2862	net->ipv4.sysctl_tcp_tso_win_divisor = 3;
2863	/* Default TSQ limit of 16 TSO segments */
2864	net->ipv4.sysctl_tcp_limit_output_bytes = 16 * 65536;
2865	/* rfc5961 challenge ack rate limiting */
2866	net->ipv4.sysctl_tcp_challenge_ack_limit = 1000;
2867	net->ipv4.sysctl_tcp_min_tso_segs = 2;
2868	net->ipv4.sysctl_tcp_min_rtt_wlen = 300;
2869	net->ipv4.sysctl_tcp_autocorking = 1;
2870	net->ipv4.sysctl_tcp_invalid_ratelimit = HZ/2;
2871	net->ipv4.sysctl_tcp_pacing_ss_ratio = 200;
2872	net->ipv4.sysctl_tcp_pacing_ca_ratio = 120;
2873	if (net != &init_net) {
2874		memcpy(net->ipv4.sysctl_tcp_rmem,
2875		       init_net.ipv4.sysctl_tcp_rmem,
2876		       sizeof(init_net.ipv4.sysctl_tcp_rmem));
2877		memcpy(net->ipv4.sysctl_tcp_wmem,
2878		       init_net.ipv4.sysctl_tcp_wmem,
2879		       sizeof(init_net.ipv4.sysctl_tcp_wmem));
2880	}
2881	net->ipv4.sysctl_tcp_comp_sack_delay_ns = NSEC_PER_MSEC;
2882	net->ipv4.sysctl_tcp_comp_sack_slack_ns = 100 * NSEC_PER_USEC;
2883	net->ipv4.sysctl_tcp_comp_sack_nr = 44;
2884	net->ipv4.sysctl_tcp_fastopen = TFO_CLIENT_ENABLE;
2885	spin_lock_init(&net->ipv4.tcp_fastopen_ctx_lock);
2886	net->ipv4.sysctl_tcp_fastopen_blackhole_timeout = 60 * 60;
2887	atomic_set(&net->ipv4.tfo_active_disable_times, 0);
2888
2889	/* Reno is always built in */
2890	if (!net_eq(net, &init_net) &&
2891	    bpf_try_module_get(init_net.ipv4.tcp_congestion_control,
2892			       init_net.ipv4.tcp_congestion_control->owner))
2893		net->ipv4.tcp_congestion_control = init_net.ipv4.tcp_congestion_control;
2894	else
2895		net->ipv4.tcp_congestion_control = &tcp_reno;
2896
2897	return 0;
2898fail:
2899	tcp_sk_exit(net);
2900
2901	return res;
 
2902}
2903
2904static void __net_exit tcp_sk_exit_batch(struct list_head *net_exit_list)
2905{
2906	struct net *net;
2907
2908	inet_twsk_purge(&tcp_hashinfo, AF_INET);
2909
2910	list_for_each_entry(net, net_exit_list, exit_list)
2911		tcp_fastopen_ctx_destroy(net);
2912}
2913
2914static struct pernet_operations __net_initdata tcp_sk_ops = {
2915       .init	   = tcp_sk_init,
2916       .exit	   = tcp_sk_exit,
2917       .exit_batch = tcp_sk_exit_batch,
2918};
2919
2920#if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
2921DEFINE_BPF_ITER_FUNC(tcp, struct bpf_iter_meta *meta,
2922		     struct sock_common *sk_common, uid_t uid)
2923
2924static int bpf_iter_init_tcp(void *priv_data, struct bpf_iter_aux_info *aux)
2925{
2926	struct tcp_iter_state *st = priv_data;
2927	struct tcp_seq_afinfo *afinfo;
2928	int ret;
2929
2930	afinfo = kmalloc(sizeof(*afinfo), GFP_USER | __GFP_NOWARN);
2931	if (!afinfo)
2932		return -ENOMEM;
2933
2934	afinfo->family = AF_UNSPEC;
2935	st->bpf_seq_afinfo = afinfo;
2936	ret = bpf_iter_init_seq_net(priv_data, aux);
2937	if (ret)
2938		kfree(afinfo);
2939	return ret;
2940}
2941
2942static void bpf_iter_fini_tcp(void *priv_data)
2943{
2944	struct tcp_iter_state *st = priv_data;
2945
2946	kfree(st->bpf_seq_afinfo);
2947	bpf_iter_fini_seq_net(priv_data);
2948}
2949
2950static const struct bpf_iter_seq_info tcp_seq_info = {
2951	.seq_ops		= &bpf_iter_tcp_seq_ops,
2952	.init_seq_private	= bpf_iter_init_tcp,
2953	.fini_seq_private	= bpf_iter_fini_tcp,
2954	.seq_priv_size		= sizeof(struct tcp_iter_state),
2955};
2956
2957static struct bpf_iter_reg tcp_reg_info = {
2958	.target			= "tcp",
2959	.ctx_arg_info_size	= 1,
2960	.ctx_arg_info		= {
2961		{ offsetof(struct bpf_iter__tcp, sk_common),
2962		  PTR_TO_BTF_ID_OR_NULL },
2963	},
2964	.seq_info		= &tcp_seq_info,
2965};
2966
2967static void __init bpf_iter_register(void)
2968{
2969	tcp_reg_info.ctx_arg_info[0].btf_id = btf_sock_ids[BTF_SOCK_TYPE_SOCK_COMMON];
2970	if (bpf_iter_reg_target(&tcp_reg_info))
2971		pr_warn("Warning: could not register bpf iterator tcp\n");
2972}
2973
2974#endif
2975
2976void __init tcp_v4_init(void)
2977{
 
2978	if (register_pernet_subsys(&tcp_sk_ops))
2979		panic("Failed to create the TCP control socket.\n");
2980
2981#if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
2982	bpf_iter_register();
2983#endif
2984}
v3.15
 
   1/*
   2 * INET		An implementation of the TCP/IP protocol suite for the LINUX
   3 *		operating system.  INET is implemented using the  BSD Socket
   4 *		interface as the means of communication with the user level.
   5 *
   6 *		Implementation of the Transmission Control Protocol(TCP).
   7 *
   8 *		IPv4 specific functions
   9 *
  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 *	This program is free software; you can redistribute it and/or
  19 *      modify it under the terms of the GNU General Public License
  20 *      as published by the Free Software Foundation; either version
  21 *      2 of the License, or (at your option) any later version.
  22 */
  23
  24/*
  25 * Changes:
  26 *		David S. Miller	:	New socket lookup architecture.
  27 *					This code is dedicated to John Dyson.
  28 *		David S. Miller :	Change semantics of established hash,
  29 *					half is devoted to TIME_WAIT sockets
  30 *					and the rest go in the other half.
  31 *		Andi Kleen :		Add support for syncookies and fixed
  32 *					some bugs: ip options weren't passed to
  33 *					the TCP layer, missed a check for an
  34 *					ACK bit.
  35 *		Andi Kleen :		Implemented fast path mtu discovery.
  36 *	     				Fixed many serious bugs in the
  37 *					request_sock handling and moved
  38 *					most of it into the af independent code.
  39 *					Added tail drop and some other bugfixes.
  40 *					Added new listen semantics.
  41 *		Mike McLagan	:	Routing by source
  42 *	Juan Jose Ciarlante:		ip_dynaddr bits
  43 *		Andi Kleen:		various fixes.
  44 *	Vitaly E. Lavrov	:	Transparent proxy revived after year
  45 *					coma.
  46 *	Andi Kleen		:	Fix new listen.
  47 *	Andi Kleen		:	Fix accept error reporting.
  48 *	YOSHIFUJI Hideaki @USAGI and:	Support IPV6_V6ONLY socket option, which
  49 *	Alexey Kuznetsov		allow both IPv4 and IPv6 sockets to bind
  50 *					a single port at the same time.
  51 */
  52
  53#define pr_fmt(fmt) "TCP: " fmt
  54
  55#include <linux/bottom_half.h>
  56#include <linux/types.h>
  57#include <linux/fcntl.h>
  58#include <linux/module.h>
  59#include <linux/random.h>
  60#include <linux/cache.h>
  61#include <linux/jhash.h>
  62#include <linux/init.h>
  63#include <linux/times.h>
  64#include <linux/slab.h>
  65
  66#include <net/net_namespace.h>
  67#include <net/icmp.h>
  68#include <net/inet_hashtables.h>
  69#include <net/tcp.h>
  70#include <net/transp_v6.h>
  71#include <net/ipv6.h>
  72#include <net/inet_common.h>
  73#include <net/timewait_sock.h>
  74#include <net/xfrm.h>
  75#include <net/netdma.h>
  76#include <net/secure_seq.h>
  77#include <net/tcp_memcontrol.h>
  78#include <net/busy_poll.h>
  79
  80#include <linux/inet.h>
  81#include <linux/ipv6.h>
  82#include <linux/stddef.h>
  83#include <linux/proc_fs.h>
  84#include <linux/seq_file.h>
 
 
  85
  86#include <linux/crypto.h>
  87#include <linux/scatterlist.h>
  88
  89int sysctl_tcp_tw_reuse __read_mostly;
  90int sysctl_tcp_low_latency __read_mostly;
  91EXPORT_SYMBOL(sysctl_tcp_low_latency);
  92
  93
  94#ifdef CONFIG_TCP_MD5SIG
  95static int tcp_v4_md5_hash_hdr(char *md5_hash, const struct tcp_md5sig_key *key,
  96			       __be32 daddr, __be32 saddr, const struct tcphdr *th);
  97#endif
  98
  99struct inet_hashinfo tcp_hashinfo;
 100EXPORT_SYMBOL(tcp_hashinfo);
 101
 102static inline __u32 tcp_v4_init_sequence(const struct sk_buff *skb)
 103{
 104	return secure_tcp_sequence_number(ip_hdr(skb)->daddr,
 105					  ip_hdr(skb)->saddr,
 106					  tcp_hdr(skb)->dest,
 107					  tcp_hdr(skb)->source);
 
 
 
 
 
 108}
 109
 110int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp)
 111{
 
 112	const struct tcp_timewait_sock *tcptw = tcp_twsk(sktw);
 113	struct tcp_sock *tp = tcp_sk(sk);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 114
 115	/* With PAWS, it is safe from the viewpoint
 116	   of data integrity. Even without PAWS it is safe provided sequence
 117	   spaces do not overlap i.e. at data rates <= 80Mbit/sec.
 118
 119	   Actually, the idea is close to VJ's one, only timestamp cache is
 120	   held not per host, but per port pair and TW bucket is used as state
 121	   holder.
 122
 123	   If TW bucket has been already destroyed we fall back to VJ's scheme
 124	   and use initial timestamp retrieved from peer table.
 125	 */
 126	if (tcptw->tw_ts_recent_stamp &&
 127	    (twp == NULL || (sysctl_tcp_tw_reuse &&
 128			     get_seconds() - tcptw->tw_ts_recent_stamp > 1))) {
 129		tp->write_seq = tcptw->tw_snd_nxt + 65535 + 2;
 130		if (tp->write_seq == 0)
 131			tp->write_seq = 1;
 132		tp->rx_opt.ts_recent	   = tcptw->tw_ts_recent;
 133		tp->rx_opt.ts_recent_stamp = tcptw->tw_ts_recent_stamp;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 134		sock_hold(sktw);
 135		return 1;
 136	}
 137
 138	return 0;
 139}
 140EXPORT_SYMBOL_GPL(tcp_twsk_unique);
 141
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 142/* This will initiate an outgoing connection. */
 143int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len)
 144{
 145	struct sockaddr_in *usin = (struct sockaddr_in *)uaddr;
 146	struct inet_sock *inet = inet_sk(sk);
 147	struct tcp_sock *tp = tcp_sk(sk);
 148	__be16 orig_sport, orig_dport;
 149	__be32 daddr, nexthop;
 150	struct flowi4 *fl4;
 151	struct rtable *rt;
 152	int err;
 153	struct ip_options_rcu *inet_opt;
 
 154
 155	if (addr_len < sizeof(struct sockaddr_in))
 156		return -EINVAL;
 157
 158	if (usin->sin_family != AF_INET)
 159		return -EAFNOSUPPORT;
 160
 161	nexthop = daddr = usin->sin_addr.s_addr;
 162	inet_opt = rcu_dereference_protected(inet->inet_opt,
 163					     sock_owned_by_user(sk));
 164	if (inet_opt && inet_opt->opt.srr) {
 165		if (!daddr)
 166			return -EINVAL;
 167		nexthop = inet_opt->opt.faddr;
 168	}
 169
 170	orig_sport = inet->inet_sport;
 171	orig_dport = usin->sin_port;
 172	fl4 = &inet->cork.fl.u.ip4;
 173	rt = ip_route_connect(fl4, nexthop, inet->inet_saddr,
 174			      RT_CONN_FLAGS(sk), sk->sk_bound_dev_if,
 175			      IPPROTO_TCP,
 176			      orig_sport, orig_dport, sk);
 177	if (IS_ERR(rt)) {
 178		err = PTR_ERR(rt);
 179		if (err == -ENETUNREACH)
 180			IP_INC_STATS(sock_net(sk), IPSTATS_MIB_OUTNOROUTES);
 181		return err;
 182	}
 183
 184	if (rt->rt_flags & (RTCF_MULTICAST | RTCF_BROADCAST)) {
 185		ip_rt_put(rt);
 186		return -ENETUNREACH;
 187	}
 188
 189	if (!inet_opt || !inet_opt->opt.srr)
 190		daddr = fl4->daddr;
 191
 192	if (!inet->inet_saddr)
 193		inet->inet_saddr = fl4->saddr;
 194	inet->inet_rcv_saddr = inet->inet_saddr;
 195
 196	if (tp->rx_opt.ts_recent_stamp && inet->inet_daddr != daddr) {
 197		/* Reset inherited state */
 198		tp->rx_opt.ts_recent	   = 0;
 199		tp->rx_opt.ts_recent_stamp = 0;
 200		if (likely(!tp->repair))
 201			tp->write_seq	   = 0;
 202	}
 203
 204	if (tcp_death_row.sysctl_tw_recycle &&
 205	    !tp->rx_opt.ts_recent_stamp && fl4->daddr == daddr)
 206		tcp_fetch_timewait_stamp(sk, &rt->dst);
 207
 208	inet->inet_dport = usin->sin_port;
 209	inet->inet_daddr = daddr;
 210
 211	inet_csk(sk)->icsk_ext_hdr_len = 0;
 212	if (inet_opt)
 213		inet_csk(sk)->icsk_ext_hdr_len = inet_opt->opt.optlen;
 214
 215	tp->rx_opt.mss_clamp = TCP_MSS_DEFAULT;
 216
 217	/* Socket identity is still unknown (sport may be zero).
 218	 * However we set state to SYN-SENT and not releasing socket
 219	 * lock select source port, enter ourselves into the hash tables and
 220	 * complete initialization after this.
 221	 */
 222	tcp_set_state(sk, TCP_SYN_SENT);
 223	err = inet_hash_connect(&tcp_death_row, sk);
 224	if (err)
 225		goto failure;
 226
 
 
 227	rt = ip_route_newports(fl4, rt, orig_sport, orig_dport,
 228			       inet->inet_sport, inet->inet_dport, sk);
 229	if (IS_ERR(rt)) {
 230		err = PTR_ERR(rt);
 231		rt = NULL;
 232		goto failure;
 233	}
 234	/* OK, now commit destination to socket.  */
 235	sk->sk_gso_type = SKB_GSO_TCPV4;
 236	sk_setup_caps(sk, &rt->dst);
 
 
 
 
 
 
 
 
 
 
 
 
 
 237
 238	if (!tp->write_seq && likely(!tp->repair))
 239		tp->write_seq = secure_tcp_sequence_number(inet->inet_saddr,
 240							   inet->inet_daddr,
 241							   inet->inet_sport,
 242							   usin->sin_port);
 243
 244	inet->inet_id = tp->write_seq ^ jiffies;
 
 
 
 245
 246	err = tcp_connect(sk);
 247
 248	rt = NULL;
 249	if (err)
 250		goto failure;
 251
 252	return 0;
 253
 254failure:
 255	/*
 256	 * This unhashes the socket and releases the local port,
 257	 * if necessary.
 258	 */
 259	tcp_set_state(sk, TCP_CLOSE);
 260	ip_rt_put(rt);
 261	sk->sk_route_caps = 0;
 262	inet->inet_dport = 0;
 263	return err;
 264}
 265EXPORT_SYMBOL(tcp_v4_connect);
 266
 267/*
 268 * This routine reacts to ICMP_FRAG_NEEDED mtu indications as defined in RFC1191.
 269 * It can be called through tcp_release_cb() if socket was owned by user
 270 * at the time tcp_v4_err() was called to handle ICMP message.
 271 */
 272static void tcp_v4_mtu_reduced(struct sock *sk)
 273{
 
 274	struct dst_entry *dst;
 275	struct inet_sock *inet = inet_sk(sk);
 276	u32 mtu = tcp_sk(sk)->mtu_info;
 277
 
 
 
 278	dst = inet_csk_update_pmtu(sk, mtu);
 279	if (!dst)
 280		return;
 281
 282	/* Something is about to be wrong... Remember soft error
 283	 * for the case, if this connection will not able to recover.
 284	 */
 285	if (mtu < dst_mtu(dst) && ip_dont_fragment(sk, dst))
 286		sk->sk_err_soft = EMSGSIZE;
 287
 288	mtu = dst_mtu(dst);
 289
 290	if (inet->pmtudisc != IP_PMTUDISC_DONT &&
 291	    ip_sk_accept_pmtu(sk) &&
 292	    inet_csk(sk)->icsk_pmtu_cookie > mtu) {
 293		tcp_sync_mss(sk, mtu);
 294
 295		/* Resend the TCP packet because it's
 296		 * clear that the old packet has been
 297		 * dropped. This is the new "fast" path mtu
 298		 * discovery.
 299		 */
 300		tcp_simple_retransmit(sk);
 301	} /* else let the usual retransmit timer handle it */
 302}
 
 303
 304static void do_redirect(struct sk_buff *skb, struct sock *sk)
 305{
 306	struct dst_entry *dst = __sk_dst_check(sk, 0);
 307
 308	if (dst)
 309		dst->ops->redirect(dst, sk, skb);
 310}
 311
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 312/*
 313 * This routine is called by the ICMP module when it gets some
 314 * sort of error condition.  If err < 0 then the socket should
 315 * be closed and the error returned to the user.  If err > 0
 316 * it's just the icmp type << 8 | icmp code.  After adjustment
 317 * header points to the first 8 bytes of the tcp header.  We need
 318 * to find the appropriate port.
 319 *
 320 * The locking strategy used here is very "optimistic". When
 321 * someone else accesses the socket the ICMP is just dropped
 322 * and for some paths there is no check at all.
 323 * A more general error queue to queue errors for later handling
 324 * is probably better.
 325 *
 326 */
 327
 328void tcp_v4_err(struct sk_buff *icmp_skb, u32 info)
 329{
 330	const struct iphdr *iph = (const struct iphdr *)icmp_skb->data;
 331	struct tcphdr *th = (struct tcphdr *)(icmp_skb->data + (iph->ihl << 2));
 332	struct inet_connection_sock *icsk;
 333	struct tcp_sock *tp;
 334	struct inet_sock *inet;
 335	const int type = icmp_hdr(icmp_skb)->type;
 336	const int code = icmp_hdr(icmp_skb)->code;
 337	struct sock *sk;
 338	struct sk_buff *skb;
 339	struct request_sock *req;
 340	__u32 seq;
 341	__u32 remaining;
 342	int err;
 343	struct net *net = dev_net(icmp_skb->dev);
 344
 345	if (icmp_skb->len < (iph->ihl << 2) + 8) {
 346		ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS);
 347		return;
 348	}
 349
 350	sk = inet_lookup(net, &tcp_hashinfo, iph->daddr, th->dest,
 351			iph->saddr, th->source, inet_iif(icmp_skb));
 
 352	if (!sk) {
 353		ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS);
 354		return;
 355	}
 356	if (sk->sk_state == TCP_TIME_WAIT) {
 357		inet_twsk_put(inet_twsk(sk));
 358		return;
 
 
 
 
 
 
 
 
 
 359	}
 360
 361	bh_lock_sock(sk);
 362	/* If too many ICMPs get dropped on busy
 363	 * servers this needs to be solved differently.
 364	 * We do take care of PMTU discovery (RFC1191) special case :
 365	 * we can receive locally generated ICMP messages while socket is held.
 366	 */
 367	if (sock_owned_by_user(sk)) {
 368		if (!(type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED))
 369			NET_INC_STATS_BH(net, LINUX_MIB_LOCKDROPPEDICMPS);
 370	}
 371	if (sk->sk_state == TCP_CLOSE)
 372		goto out;
 373
 374	if (unlikely(iph->ttl < inet_sk(sk)->min_ttl)) {
 375		NET_INC_STATS_BH(net, LINUX_MIB_TCPMINTTLDROP);
 376		goto out;
 377	}
 378
 379	icsk = inet_csk(sk);
 380	tp = tcp_sk(sk);
 381	req = tp->fastopen_rsk;
 382	seq = ntohl(th->seq);
 
 383	if (sk->sk_state != TCP_LISTEN &&
 384	    !between(seq, tp->snd_una, tp->snd_nxt) &&
 385	    (req == NULL || seq != tcp_rsk(req)->snt_isn)) {
 386		/* For a Fast Open socket, allow seq to be snt_isn. */
 387		NET_INC_STATS_BH(net, LINUX_MIB_OUTOFWINDOWICMPS);
 388		goto out;
 389	}
 390
 391	switch (type) {
 392	case ICMP_REDIRECT:
 393		do_redirect(icmp_skb, sk);
 
 394		goto out;
 395	case ICMP_SOURCE_QUENCH:
 396		/* Just silently ignore these. */
 397		goto out;
 398	case ICMP_PARAMETERPROB:
 399		err = EPROTO;
 400		break;
 401	case ICMP_DEST_UNREACH:
 402		if (code > NR_ICMP_UNREACH)
 403			goto out;
 404
 405		if (code == ICMP_FRAG_NEEDED) { /* PMTU discovery (RFC1191) */
 406			/* We are not interested in TCP_LISTEN and open_requests
 407			 * (SYN-ACKs send out by Linux are always <576bytes so
 408			 * they should go through unfragmented).
 409			 */
 410			if (sk->sk_state == TCP_LISTEN)
 411				goto out;
 412
 413			tp->mtu_info = info;
 414			if (!sock_owned_by_user(sk)) {
 415				tcp_v4_mtu_reduced(sk);
 416			} else {
 417				if (!test_and_set_bit(TCP_MTU_REDUCED_DEFERRED, &tp->tsq_flags))
 418					sock_hold(sk);
 419			}
 420			goto out;
 421		}
 422
 423		err = icmp_err_convert[code].errno;
 424		/* check if icmp_skb allows revert of backoff
 425		 * (see draft-zimmermann-tcp-lcd) */
 426		if (code != ICMP_NET_UNREACH && code != ICMP_HOST_UNREACH)
 427			break;
 428		if (seq != tp->snd_una  || !icsk->icsk_retransmits ||
 429		    !icsk->icsk_backoff)
 430			break;
 431
 432		/* XXX (TFO) - revisit the following logic for TFO */
 433
 434		if (sock_owned_by_user(sk))
 435			break;
 436
 437		icsk->icsk_backoff--;
 438		inet_csk(sk)->icsk_rto = (tp->srtt_us ? __tcp_set_rto(tp) :
 439			TCP_TIMEOUT_INIT) << icsk->icsk_backoff;
 440		tcp_bound_rto(sk);
 441
 442		skb = tcp_write_queue_head(sk);
 443		BUG_ON(!skb);
 444
 445		remaining = icsk->icsk_rto - min(icsk->icsk_rto,
 446				tcp_time_stamp - TCP_SKB_CB(skb)->when);
 447
 448		if (remaining) {
 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			tcp_retransmit_timer(sk);
 455		}
 456
 457		break;
 458	case ICMP_TIME_EXCEEDED:
 459		err = EHOSTUNREACH;
 460		break;
 461	default:
 462		goto out;
 463	}
 464
 465	/* XXX (TFO) - if it's a TFO socket and has been accepted, rather
 466	 * than following the TCP_SYN_RECV case and closing the socket,
 467	 * we ignore the ICMP error and keep trying like a fully established
 468	 * socket. Is this the right thing to do?
 469	 */
 470	if (req && req->sk == NULL)
 471		goto out;
 472
 473	switch (sk->sk_state) {
 474		struct request_sock *req, **prev;
 475	case TCP_LISTEN:
 476		if (sock_owned_by_user(sk))
 477			goto out;
 478
 479		req = inet_csk_search_req(sk, &prev, th->dest,
 480					  iph->daddr, iph->saddr);
 481		if (!req)
 482			goto out;
 483
 484		/* ICMPs are not backlogged, hence we cannot get
 485		   an established socket here.
 486		 */
 487		WARN_ON(req->sk);
 488
 489		if (seq != tcp_rsk(req)->snt_isn) {
 490			NET_INC_STATS_BH(net, LINUX_MIB_OUTOFWINDOWICMPS);
 491			goto out;
 492		}
 493
 494		/*
 495		 * Still in SYN_RECV, just remove it silently.
 496		 * There is no good way to pass the error to the newly
 497		 * created socket, and POSIX does not want network
 498		 * errors returned from accept().
 499		 */
 500		inet_csk_reqsk_queue_drop(sk, req, prev);
 501		NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENDROPS);
 502		goto out;
 503
 504	case TCP_SYN_SENT:
 505	case TCP_SYN_RECV:  /* Cannot happen.
 506			       It can f.e. if SYNs crossed,
 507			       or Fast Open.
 508			     */
 509		if (!sock_owned_by_user(sk)) {
 510			sk->sk_err = err;
 511
 512			sk->sk_error_report(sk);
 513
 514			tcp_done(sk);
 515		} else {
 516			sk->sk_err_soft = err;
 517		}
 518		goto out;
 519	}
 520
 521	/* If we've already connected we will keep trying
 522	 * until we time out, or the user gives up.
 523	 *
 524	 * rfc1122 4.2.3.9 allows to consider as hard errors
 525	 * only PROTO_UNREACH and PORT_UNREACH (well, FRAG_FAILED too,
 526	 * but it is obsoleted by pmtu discovery).
 527	 *
 528	 * Note, that in modern internet, where routing is unreliable
 529	 * and in each dark corner broken firewalls sit, sending random
 530	 * errors ordered by their masters even this two messages finally lose
 531	 * their original sense (even Linux sends invalid PORT_UNREACHs)
 532	 *
 533	 * Now we are in compliance with RFCs.
 534	 *							--ANK (980905)
 535	 */
 536
 537	inet = inet_sk(sk);
 538	if (!sock_owned_by_user(sk) && inet->recverr) {
 539		sk->sk_err = err;
 540		sk->sk_error_report(sk);
 541	} else	{ /* Only an error on timeout */
 542		sk->sk_err_soft = err;
 543	}
 544
 545out:
 546	bh_unlock_sock(sk);
 547	sock_put(sk);
 
 548}
 549
 550void __tcp_v4_send_check(struct sk_buff *skb, __be32 saddr, __be32 daddr)
 551{
 552	struct tcphdr *th = tcp_hdr(skb);
 553
 554	if (skb->ip_summed == CHECKSUM_PARTIAL) {
 555		th->check = ~tcp_v4_check(skb->len, saddr, daddr, 0);
 556		skb->csum_start = skb_transport_header(skb) - skb->head;
 557		skb->csum_offset = offsetof(struct tcphdr, check);
 558	} else {
 559		th->check = tcp_v4_check(skb->len, saddr, daddr,
 560					 csum_partial(th,
 561						      th->doff << 2,
 562						      skb->csum));
 563	}
 564}
 565
 566/* This routine computes an IPv4 TCP checksum. */
 567void tcp_v4_send_check(struct sock *sk, struct sk_buff *skb)
 568{
 569	const struct inet_sock *inet = inet_sk(sk);
 570
 571	__tcp_v4_send_check(skb, inet->inet_saddr, inet->inet_daddr);
 572}
 573EXPORT_SYMBOL(tcp_v4_send_check);
 574
 575/*
 576 *	This routine will send an RST to the other tcp.
 577 *
 578 *	Someone asks: why I NEVER use socket parameters (TOS, TTL etc.)
 579 *		      for reset.
 580 *	Answer: if a packet caused RST, it is not for a socket
 581 *		existing in our system, if it is matched to a socket,
 582 *		it is just duplicate segment or bug in other side's TCP.
 583 *		So that we build reply only basing on parameters
 584 *		arrived with segment.
 585 *	Exception: precedence violation. We do not implement it in any case.
 586 */
 587
 588static void tcp_v4_send_reset(struct sock *sk, struct sk_buff *skb)
 589{
 590	const struct tcphdr *th = tcp_hdr(skb);
 591	struct {
 592		struct tcphdr th;
 593#ifdef CONFIG_TCP_MD5SIG
 594		__be32 opt[(TCPOLEN_MD5SIG_ALIGNED >> 2)];
 595#endif
 596	} rep;
 597	struct ip_reply_arg arg;
 598#ifdef CONFIG_TCP_MD5SIG
 599	struct tcp_md5sig_key *key;
 600	const __u8 *hash_location = NULL;
 601	unsigned char newhash[16];
 602	int genhash;
 603	struct sock *sk1 = NULL;
 604#endif
 
 
 605	struct net *net;
 606
 607	/* Never send a reset in response to a reset. */
 608	if (th->rst)
 609		return;
 610
 611	if (skb_rtable(skb)->rt_type != RTN_LOCAL)
 
 
 
 612		return;
 613
 614	/* Swap the send and the receive. */
 615	memset(&rep, 0, sizeof(rep));
 616	rep.th.dest   = th->source;
 617	rep.th.source = th->dest;
 618	rep.th.doff   = sizeof(struct tcphdr) / 4;
 619	rep.th.rst    = 1;
 620
 621	if (th->ack) {
 622		rep.th.seq = th->ack_seq;
 623	} else {
 624		rep.th.ack = 1;
 625		rep.th.ack_seq = htonl(ntohl(th->seq) + th->syn + th->fin +
 626				       skb->len - (th->doff << 2));
 627	}
 628
 629	memset(&arg, 0, sizeof(arg));
 630	arg.iov[0].iov_base = (unsigned char *)&rep;
 631	arg.iov[0].iov_len  = sizeof(rep.th);
 632
 
 633#ifdef CONFIG_TCP_MD5SIG
 
 634	hash_location = tcp_parse_md5sig_option(th);
 635	if (!sk && hash_location) {
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 636		/*
 637		 * active side is lost. Try to find listening socket through
 638		 * source port, and then find md5 key through listening socket.
 639		 * we are not loose security here:
 640		 * Incoming packet is checked with md5 hash with finding key,
 641		 * no RST generated if md5 hash doesn't match.
 642		 */
 643		sk1 = __inet_lookup_listener(dev_net(skb_dst(skb)->dev),
 644					     &tcp_hashinfo, ip_hdr(skb)->saddr,
 645					     th->source, ip_hdr(skb)->daddr,
 646					     ntohs(th->source), inet_iif(skb));
 647		/* don't send rst if it can't find key */
 648		if (!sk1)
 649			return;
 650		rcu_read_lock();
 651		key = tcp_md5_do_lookup(sk1, (union tcp_md5_addr *)
 652					&ip_hdr(skb)->saddr, AF_INET);
 
 
 
 
 653		if (!key)
 654			goto release_sk1;
 
 655
 656		genhash = tcp_v4_md5_hash_skb(newhash, key, NULL, NULL, skb);
 657		if (genhash || memcmp(hash_location, newhash, 16) != 0)
 658			goto release_sk1;
 659	} else {
 660		key = sk ? tcp_md5_do_lookup(sk, (union tcp_md5_addr *)
 661					     &ip_hdr(skb)->saddr,
 662					     AF_INET) : NULL;
 663	}
 664
 665	if (key) {
 666		rep.opt[0] = htonl((TCPOPT_NOP << 24) |
 667				   (TCPOPT_NOP << 16) |
 668				   (TCPOPT_MD5SIG << 8) |
 669				   TCPOLEN_MD5SIG);
 670		/* Update length and the length the header thinks exists */
 671		arg.iov[0].iov_len += TCPOLEN_MD5SIG_ALIGNED;
 672		rep.th.doff = arg.iov[0].iov_len / 4;
 673
 674		tcp_v4_md5_hash_hdr((__u8 *) &rep.opt[1],
 675				     key, ip_hdr(skb)->saddr,
 676				     ip_hdr(skb)->daddr, &rep.th);
 677	}
 678#endif
 679	arg.csum = csum_tcpudp_nofold(ip_hdr(skb)->daddr,
 680				      ip_hdr(skb)->saddr, /* XXX */
 681				      arg.iov[0].iov_len, IPPROTO_TCP, 0);
 682	arg.csumoffset = offsetof(struct tcphdr, check) / 2;
 683	arg.flags = (sk && inet_sk(sk)->transparent) ? IP_REPLY_ARG_NOSRCCHECK : 0;
 
 684	/* When socket is gone, all binding information is lost.
 685	 * routing might fail in this case. No choice here, if we choose to force
 686	 * input interface, we will misroute in case of asymmetric route.
 687	 */
 688	if (sk)
 689		arg.bound_dev_if = sk->sk_bound_dev_if;
 
 
 
 
 
 
 690
 691	net = dev_net(skb_dst(skb)->dev);
 692	arg.tos = ip_hdr(skb)->tos;
 693	ip_send_unicast_reply(net, skb, ip_hdr(skb)->saddr,
 694			      ip_hdr(skb)->daddr, &arg, arg.iov[0].iov_len);
 695
 696	TCP_INC_STATS_BH(net, TCP_MIB_OUTSEGS);
 697	TCP_INC_STATS_BH(net, TCP_MIB_OUTRSTS);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 698
 699#ifdef CONFIG_TCP_MD5SIG
 700release_sk1:
 701	if (sk1) {
 702		rcu_read_unlock();
 703		sock_put(sk1);
 704	}
 705#endif
 706}
 707
 708/* The code following below sending ACKs in SYN-RECV and TIME-WAIT states
 709   outside socket context is ugly, certainly. What can I do?
 710 */
 711
 712static void tcp_v4_send_ack(struct sk_buff *skb, u32 seq, u32 ack,
 
 713			    u32 win, u32 tsval, u32 tsecr, int oif,
 714			    struct tcp_md5sig_key *key,
 715			    int reply_flags, u8 tos)
 716{
 717	const struct tcphdr *th = tcp_hdr(skb);
 718	struct {
 719		struct tcphdr th;
 720		__be32 opt[(TCPOLEN_TSTAMP_ALIGNED >> 2)
 721#ifdef CONFIG_TCP_MD5SIG
 722			   + (TCPOLEN_MD5SIG_ALIGNED >> 2)
 723#endif
 724			];
 725	} rep;
 
 726	struct ip_reply_arg arg;
 727	struct net *net = dev_net(skb_dst(skb)->dev);
 
 728
 729	memset(&rep.th, 0, sizeof(struct tcphdr));
 730	memset(&arg, 0, sizeof(arg));
 731
 732	arg.iov[0].iov_base = (unsigned char *)&rep;
 733	arg.iov[0].iov_len  = sizeof(rep.th);
 734	if (tsecr) {
 735		rep.opt[0] = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
 736				   (TCPOPT_TIMESTAMP << 8) |
 737				   TCPOLEN_TIMESTAMP);
 738		rep.opt[1] = htonl(tsval);
 739		rep.opt[2] = htonl(tsecr);
 740		arg.iov[0].iov_len += TCPOLEN_TSTAMP_ALIGNED;
 741	}
 742
 743	/* Swap the send and the receive. */
 744	rep.th.dest    = th->source;
 745	rep.th.source  = th->dest;
 746	rep.th.doff    = arg.iov[0].iov_len / 4;
 747	rep.th.seq     = htonl(seq);
 748	rep.th.ack_seq = htonl(ack);
 749	rep.th.ack     = 1;
 750	rep.th.window  = htons(win);
 751
 752#ifdef CONFIG_TCP_MD5SIG
 753	if (key) {
 754		int offset = (tsecr) ? 3 : 0;
 755
 756		rep.opt[offset++] = htonl((TCPOPT_NOP << 24) |
 757					  (TCPOPT_NOP << 16) |
 758					  (TCPOPT_MD5SIG << 8) |
 759					  TCPOLEN_MD5SIG);
 760		arg.iov[0].iov_len += TCPOLEN_MD5SIG_ALIGNED;
 761		rep.th.doff = arg.iov[0].iov_len/4;
 762
 763		tcp_v4_md5_hash_hdr((__u8 *) &rep.opt[offset],
 764				    key, ip_hdr(skb)->saddr,
 765				    ip_hdr(skb)->daddr, &rep.th);
 766	}
 767#endif
 768	arg.flags = reply_flags;
 769	arg.csum = csum_tcpudp_nofold(ip_hdr(skb)->daddr,
 770				      ip_hdr(skb)->saddr, /* XXX */
 771				      arg.iov[0].iov_len, IPPROTO_TCP, 0);
 772	arg.csumoffset = offsetof(struct tcphdr, check) / 2;
 773	if (oif)
 774		arg.bound_dev_if = oif;
 775	arg.tos = tos;
 776	ip_send_unicast_reply(net, skb, ip_hdr(skb)->saddr,
 777			      ip_hdr(skb)->daddr, &arg, arg.iov[0].iov_len);
 778
 779	TCP_INC_STATS_BH(net, TCP_MIB_OUTSEGS);
 
 
 
 
 
 
 
 
 
 
 
 
 
 780}
 781
 782static void tcp_v4_timewait_ack(struct sock *sk, struct sk_buff *skb)
 783{
 784	struct inet_timewait_sock *tw = inet_twsk(sk);
 785	struct tcp_timewait_sock *tcptw = tcp_twsk(sk);
 786
 787	tcp_v4_send_ack(skb, tcptw->tw_snd_nxt, tcptw->tw_rcv_nxt,
 
 788			tcptw->tw_rcv_wnd >> tw->tw_rcv_wscale,
 789			tcp_time_stamp + tcptw->tw_ts_offset,
 790			tcptw->tw_ts_recent,
 791			tw->tw_bound_dev_if,
 792			tcp_twsk_md5_key(tcptw),
 793			tw->tw_transparent ? IP_REPLY_ARG_NOSRCCHECK : 0,
 794			tw->tw_tos
 795			);
 796
 797	inet_twsk_put(tw);
 798}
 799
 800static void tcp_v4_reqsk_send_ack(struct sock *sk, struct sk_buff *skb,
 801				  struct request_sock *req)
 802{
 
 
 
 803	/* sk->sk_state == TCP_LISTEN -> for regular TCP_SYN_RECV
 804	 * sk->sk_state == TCP_SYN_RECV -> for Fast Open.
 805	 */
 806	tcp_v4_send_ack(skb, (sk->sk_state == TCP_LISTEN) ?
 807			tcp_rsk(req)->snt_isn + 1 : tcp_sk(sk)->snd_nxt,
 808			tcp_rsk(req)->rcv_nxt, req->rcv_wnd,
 809			tcp_time_stamp,
 
 
 
 
 
 
 
 
 
 
 810			req->ts_recent,
 811			0,
 812			tcp_md5_do_lookup(sk, (union tcp_md5_addr *)&ip_hdr(skb)->daddr,
 813					  AF_INET),
 814			inet_rsk(req)->no_srccheck ? IP_REPLY_ARG_NOSRCCHECK : 0,
 815			ip_hdr(skb)->tos);
 816}
 817
 818/*
 819 *	Send a SYN-ACK after having received a SYN.
 820 *	This still operates on a request_sock only, not on a big
 821 *	socket.
 822 */
 823static int tcp_v4_send_synack(struct sock *sk, struct dst_entry *dst,
 
 824			      struct request_sock *req,
 825			      u16 queue_mapping)
 
 826{
 827	const struct inet_request_sock *ireq = inet_rsk(req);
 828	struct flowi4 fl4;
 829	int err = -1;
 830	struct sk_buff *skb;
 831
 832	/* First, grab a route. */
 833	if (!dst && (dst = inet_csk_route_req(sk, &fl4, req)) == NULL)
 834		return -1;
 835
 836	skb = tcp_make_synack(sk, dst, req, NULL);
 837
 838	if (skb) {
 839		__tcp_v4_send_check(skb, ireq->ir_loc_addr, ireq->ir_rmt_addr);
 840
 841		skb_set_queue_mapping(skb, queue_mapping);
 842		err = ip_build_and_send_pkt(skb, sk, ireq->ir_loc_addr,
 843					    ireq->ir_rmt_addr,
 844					    ireq->opt);
 
 845		err = net_xmit_eval(err);
 846		if (!tcp_rsk(req)->snt_synack && !err)
 847			tcp_rsk(req)->snt_synack = tcp_time_stamp;
 848	}
 849
 850	return err;
 851}
 852
 853static int tcp_v4_rtx_synack(struct sock *sk, struct request_sock *req)
 854{
 855	int res = tcp_v4_send_synack(sk, NULL, req, 0);
 856
 857	if (!res) {
 858		TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_RETRANSSEGS);
 859		NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPSYNRETRANS);
 860	}
 861	return res;
 862}
 863
 864/*
 865 *	IPv4 request_sock destructor.
 866 */
 867static void tcp_v4_reqsk_destructor(struct request_sock *req)
 868{
 869	kfree(inet_rsk(req)->opt);
 870}
 871
 
 872/*
 873 * Return true if a syncookie should be sent
 
 
 874 */
 875bool tcp_syn_flood_action(struct sock *sk,
 876			 const struct sk_buff *skb,
 877			 const char *proto)
 878{
 879	const char *msg = "Dropping request";
 880	bool want_cookie = false;
 881	struct listen_sock *lopt;
 882
 883#ifdef CONFIG_SYN_COOKIES
 884	if (sysctl_tcp_syncookies) {
 885		msg = "Sending cookies";
 886		want_cookie = true;
 887		NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPREQQFULLDOCOOKIES);
 888	} else
 889#endif
 890		NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPREQQFULLDROP);
 891
 892	lopt = inet_csk(sk)->icsk_accept_queue.listen_opt;
 893	if (!lopt->synflood_warned && sysctl_tcp_syncookies != 2) {
 894		lopt->synflood_warned = 1;
 895		pr_info("%s: Possible SYN flooding on port %d. %s.  Check SNMP counters.\n",
 896			proto, ntohs(tcp_hdr(skb)->dest), msg);
 897	}
 898	return want_cookie;
 899}
 900EXPORT_SYMBOL(tcp_syn_flood_action);
 901
 902/*
 903 * Save and compile IPv4 options into the request_sock if needed.
 904 */
 905static struct ip_options_rcu *tcp_v4_save_options(struct sk_buff *skb)
 906{
 907	const struct ip_options *opt = &(IPCB(skb)->opt);
 908	struct ip_options_rcu *dopt = NULL;
 
 
 
 
 909
 910	if (opt && opt->optlen) {
 911		int opt_size = sizeof(*dopt) + opt->optlen;
 
 
 
 912
 913		dopt = kmalloc(opt_size, GFP_ATOMIC);
 914		if (dopt) {
 915			if (ip_options_echo(&dopt->opt, skb)) {
 916				kfree(dopt);
 917				dopt = NULL;
 918			}
 
 
 
 
 
 
 
 
 
 
 
 919		}
 
 
 
 
 920	}
 921	return dopt;
 922}
 
 923
 924#ifdef CONFIG_TCP_MD5SIG
 925/*
 926 * RFC2385 MD5 checksumming requires a mapping of
 927 * IP address->MD5 Key.
 928 * We need to maintain these in the sk structure.
 929 */
 930
 931/* Find the Key structure for an address.  */
 932struct tcp_md5sig_key *tcp_md5_do_lookup(struct sock *sk,
 933					 const union tcp_md5_addr *addr,
 934					 int family)
 935{
 936	struct tcp_sock *tp = tcp_sk(sk);
 937	struct tcp_md5sig_key *key;
 938	unsigned int size = sizeof(struct in_addr);
 939	struct tcp_md5sig_info *md5sig;
 940
 941	/* caller either holds rcu_read_lock() or socket lock */
 942	md5sig = rcu_dereference_check(tp->md5sig_info,
 943				       sock_owned_by_user(sk) ||
 944				       lockdep_is_held(&sk->sk_lock.slock));
 945	if (!md5sig)
 946		return NULL;
 947#if IS_ENABLED(CONFIG_IPV6)
 948	if (family == AF_INET6)
 949		size = sizeof(struct in6_addr);
 950#endif
 951	hlist_for_each_entry_rcu(key, &md5sig->head, node) {
 
 952		if (key->family != family)
 953			continue;
 954		if (!memcmp(&key->addr, addr, size))
 
 
 
 955			return key;
 956	}
 957	return NULL;
 958}
 959EXPORT_SYMBOL(tcp_md5_do_lookup);
 960
 961struct tcp_md5sig_key *tcp_v4_md5_lookup(struct sock *sk,
 962					 struct sock *addr_sk)
 963{
 964	union tcp_md5_addr *addr;
 
 965
 966	addr = (union tcp_md5_addr *)&inet_sk(addr_sk)->inet_daddr;
 967	return tcp_md5_do_lookup(sk, addr, AF_INET);
 
 
 968}
 969EXPORT_SYMBOL(tcp_v4_md5_lookup);
 970
 971static struct tcp_md5sig_key *tcp_v4_reqsk_md5_lookup(struct sock *sk,
 972						      struct request_sock *req)
 973{
 974	union tcp_md5_addr *addr;
 975
 976	addr = (union tcp_md5_addr *)&inet_rsk(req)->ir_rmt_addr;
 977	return tcp_md5_do_lookup(sk, addr, AF_INET);
 978}
 979
 980/* This can be called on a newly created socket, from other files */
 981int tcp_md5_do_add(struct sock *sk, const union tcp_md5_addr *addr,
 982		   int family, const u8 *newkey, u8 newkeylen, gfp_t gfp)
 
 983{
 984	/* Add Key to the list */
 985	struct tcp_md5sig_key *key;
 986	struct tcp_sock *tp = tcp_sk(sk);
 987	struct tcp_md5sig_info *md5sig;
 988
 989	key = tcp_md5_do_lookup(sk, addr, family);
 990	if (key) {
 991		/* Pre-existing entry - just update that one. */
 992		memcpy(key->key, newkey, newkeylen);
 993		key->keylen = newkeylen;
 
 
 
 
 
 
 
 
 
 
 
 
 994		return 0;
 995	}
 996
 997	md5sig = rcu_dereference_protected(tp->md5sig_info,
 998					   sock_owned_by_user(sk));
 999	if (!md5sig) {
1000		md5sig = kmalloc(sizeof(*md5sig), gfp);
1001		if (!md5sig)
1002			return -ENOMEM;
1003
1004		sk_nocaps_add(sk, NETIF_F_GSO_MASK);
1005		INIT_HLIST_HEAD(&md5sig->head);
1006		rcu_assign_pointer(tp->md5sig_info, md5sig);
1007	}
1008
1009	key = sock_kmalloc(sk, sizeof(*key), gfp);
1010	if (!key)
1011		return -ENOMEM;
1012	if (!tcp_alloc_md5sig_pool()) {
1013		sock_kfree_s(sk, key, sizeof(*key));
1014		return -ENOMEM;
1015	}
1016
1017	memcpy(key->key, newkey, newkeylen);
1018	key->keylen = newkeylen;
1019	key->family = family;
 
 
1020	memcpy(&key->addr, addr,
1021	       (family == AF_INET6) ? sizeof(struct in6_addr) :
1022				      sizeof(struct in_addr));
1023	hlist_add_head_rcu(&key->node, &md5sig->head);
1024	return 0;
1025}
1026EXPORT_SYMBOL(tcp_md5_do_add);
1027
1028int tcp_md5_do_del(struct sock *sk, const union tcp_md5_addr *addr, int family)
 
1029{
1030	struct tcp_md5sig_key *key;
1031
1032	key = tcp_md5_do_lookup(sk, addr, family);
1033	if (!key)
1034		return -ENOENT;
1035	hlist_del_rcu(&key->node);
1036	atomic_sub(sizeof(*key), &sk->sk_omem_alloc);
1037	kfree_rcu(key, rcu);
1038	return 0;
1039}
1040EXPORT_SYMBOL(tcp_md5_do_del);
1041
1042static void tcp_clear_md5_list(struct sock *sk)
1043{
1044	struct tcp_sock *tp = tcp_sk(sk);
1045	struct tcp_md5sig_key *key;
1046	struct hlist_node *n;
1047	struct tcp_md5sig_info *md5sig;
1048
1049	md5sig = rcu_dereference_protected(tp->md5sig_info, 1);
1050
1051	hlist_for_each_entry_safe(key, n, &md5sig->head, node) {
1052		hlist_del_rcu(&key->node);
1053		atomic_sub(sizeof(*key), &sk->sk_omem_alloc);
1054		kfree_rcu(key, rcu);
1055	}
1056}
1057
1058static int tcp_v4_parse_md5_keys(struct sock *sk, char __user *optval,
1059				 int optlen)
1060{
1061	struct tcp_md5sig cmd;
1062	struct sockaddr_in *sin = (struct sockaddr_in *)&cmd.tcpm_addr;
 
 
 
1063
1064	if (optlen < sizeof(cmd))
1065		return -EINVAL;
1066
1067	if (copy_from_user(&cmd, optval, sizeof(cmd)))
1068		return -EFAULT;
1069
1070	if (sin->sin_family != AF_INET)
1071		return -EINVAL;
1072
1073	if (!cmd.tcpm_key || !cmd.tcpm_keylen)
1074		return tcp_md5_do_del(sk, (union tcp_md5_addr *)&sin->sin_addr.s_addr,
1075				      AF_INET);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1076
1077	if (cmd.tcpm_keylen > TCP_MD5SIG_MAXKEYLEN)
1078		return -EINVAL;
1079
1080	return tcp_md5_do_add(sk, (union tcp_md5_addr *)&sin->sin_addr.s_addr,
1081			      AF_INET, cmd.tcpm_key, cmd.tcpm_keylen,
1082			      GFP_KERNEL);
1083}
1084
1085static int tcp_v4_md5_hash_pseudoheader(struct tcp_md5sig_pool *hp,
1086					__be32 daddr, __be32 saddr, int nbytes)
 
1087{
1088	struct tcp4_pseudohdr *bp;
1089	struct scatterlist sg;
 
1090
1091	bp = &hp->md5_blk.ip4;
1092
1093	/*
1094	 * 1. the TCP pseudo-header (in the order: source IP address,
1095	 * destination IP address, zero-padded protocol number, and
1096	 * segment length)
1097	 */
1098	bp->saddr = saddr;
1099	bp->daddr = daddr;
1100	bp->pad = 0;
1101	bp->protocol = IPPROTO_TCP;
1102	bp->len = cpu_to_be16(nbytes);
1103
1104	sg_init_one(&sg, bp, sizeof(*bp));
1105	return crypto_hash_update(&hp->md5_desc, &sg, sizeof(*bp));
 
 
 
 
 
 
1106}
1107
1108static int tcp_v4_md5_hash_hdr(char *md5_hash, const struct tcp_md5sig_key *key,
1109			       __be32 daddr, __be32 saddr, const struct tcphdr *th)
1110{
1111	struct tcp_md5sig_pool *hp;
1112	struct hash_desc *desc;
1113
1114	hp = tcp_get_md5sig_pool();
1115	if (!hp)
1116		goto clear_hash_noput;
1117	desc = &hp->md5_desc;
1118
1119	if (crypto_hash_init(desc))
1120		goto clear_hash;
1121	if (tcp_v4_md5_hash_pseudoheader(hp, daddr, saddr, th->doff << 2))
1122		goto clear_hash;
1123	if (tcp_md5_hash_header(hp, th))
1124		goto clear_hash;
1125	if (tcp_md5_hash_key(hp, key))
1126		goto clear_hash;
1127	if (crypto_hash_final(desc, md5_hash))
 
1128		goto clear_hash;
1129
1130	tcp_put_md5sig_pool();
1131	return 0;
1132
1133clear_hash:
1134	tcp_put_md5sig_pool();
1135clear_hash_noput:
1136	memset(md5_hash, 0, 16);
1137	return 1;
1138}
1139
1140int tcp_v4_md5_hash_skb(char *md5_hash, struct tcp_md5sig_key *key,
1141			const struct sock *sk, const struct request_sock *req,
1142			const struct sk_buff *skb)
1143{
1144	struct tcp_md5sig_pool *hp;
1145	struct hash_desc *desc;
1146	const struct tcphdr *th = tcp_hdr(skb);
1147	__be32 saddr, daddr;
1148
1149	if (sk) {
1150		saddr = inet_sk(sk)->inet_saddr;
1151		daddr = inet_sk(sk)->inet_daddr;
1152	} else if (req) {
1153		saddr = inet_rsk(req)->ir_loc_addr;
1154		daddr = inet_rsk(req)->ir_rmt_addr;
1155	} else {
1156		const struct iphdr *iph = ip_hdr(skb);
1157		saddr = iph->saddr;
1158		daddr = iph->daddr;
1159	}
1160
1161	hp = tcp_get_md5sig_pool();
1162	if (!hp)
1163		goto clear_hash_noput;
1164	desc = &hp->md5_desc;
1165
1166	if (crypto_hash_init(desc))
1167		goto clear_hash;
1168
1169	if (tcp_v4_md5_hash_pseudoheader(hp, daddr, saddr, skb->len))
1170		goto clear_hash;
1171	if (tcp_md5_hash_header(hp, th))
1172		goto clear_hash;
1173	if (tcp_md5_hash_skb_data(hp, skb, th->doff << 2))
1174		goto clear_hash;
1175	if (tcp_md5_hash_key(hp, key))
1176		goto clear_hash;
1177	if (crypto_hash_final(desc, md5_hash))
 
1178		goto clear_hash;
1179
1180	tcp_put_md5sig_pool();
1181	return 0;
1182
1183clear_hash:
1184	tcp_put_md5sig_pool();
1185clear_hash_noput:
1186	memset(md5_hash, 0, 16);
1187	return 1;
1188}
1189EXPORT_SYMBOL(tcp_v4_md5_hash_skb);
1190
1191static bool tcp_v4_inbound_md5_hash(struct sock *sk, const struct sk_buff *skb)
 
 
 
 
 
1192{
 
1193	/*
1194	 * This gets called for each TCP segment that arrives
1195	 * so we want to be efficient.
1196	 * We have 3 drop cases:
1197	 * o No MD5 hash and one expected.
1198	 * o MD5 hash and we're not expecting one.
1199	 * o MD5 hash and its wrong.
1200	 */
1201	const __u8 *hash_location = NULL;
1202	struct tcp_md5sig_key *hash_expected;
1203	const struct iphdr *iph = ip_hdr(skb);
1204	const struct tcphdr *th = tcp_hdr(skb);
1205	int genhash;
1206	unsigned char newhash[16];
 
 
 
 
 
 
1207
1208	hash_expected = tcp_md5_do_lookup(sk, (union tcp_md5_addr *)&iph->saddr,
1209					  AF_INET);
1210	hash_location = tcp_parse_md5sig_option(th);
1211
1212	/* We've parsed the options - do we have a hash? */
1213	if (!hash_expected && !hash_location)
1214		return false;
1215
1216	if (hash_expected && !hash_location) {
1217		NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPMD5NOTFOUND);
1218		return true;
1219	}
1220
1221	if (!hash_expected && hash_location) {
1222		NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPMD5UNEXPECTED);
1223		return true;
1224	}
1225
1226	/* Okay, so this is hash_expected and hash_location -
1227	 * so we need to calculate the checksum.
1228	 */
1229	genhash = tcp_v4_md5_hash_skb(newhash,
1230				      hash_expected,
1231				      NULL, NULL, skb);
1232
1233	if (genhash || memcmp(hash_location, newhash, 16) != 0) {
1234		net_info_ratelimited("MD5 Hash failed for (%pI4, %d)->(%pI4, %d)%s\n",
 
1235				     &iph->saddr, ntohs(th->source),
1236				     &iph->daddr, ntohs(th->dest),
1237				     genhash ? " tcp_v4_calc_md5_hash failed"
1238				     : "");
1239		return true;
1240	}
1241	return false;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1242}
1243
1244#endif
 
 
 
 
 
1245
1246struct request_sock_ops tcp_request_sock_ops __read_mostly = {
1247	.family		=	PF_INET,
1248	.obj_size	=	sizeof(struct tcp_request_sock),
1249	.rtx_syn_ack	=	tcp_v4_rtx_synack,
1250	.send_ack	=	tcp_v4_reqsk_send_ack,
1251	.destructor	=	tcp_v4_reqsk_destructor,
1252	.send_reset	=	tcp_v4_send_reset,
1253	.syn_ack_timeout = 	tcp_syn_ack_timeout,
1254};
1255
 
 
1256#ifdef CONFIG_TCP_MD5SIG
1257static const struct tcp_request_sock_ops tcp_request_sock_ipv4_ops = {
1258	.md5_lookup	=	tcp_v4_reqsk_md5_lookup,
1259	.calc_md5_hash	=	tcp_v4_md5_hash_skb,
 
 
 
 
 
 
 
 
 
1260};
1261#endif
1262
1263static bool tcp_fastopen_check(struct sock *sk, struct sk_buff *skb,
1264			       struct request_sock *req,
1265			       struct tcp_fastopen_cookie *foc,
1266			       struct tcp_fastopen_cookie *valid_foc)
1267{
1268	bool skip_cookie = false;
1269	struct fastopen_queue *fastopenq;
1270
1271	if (likely(!fastopen_cookie_present(foc))) {
1272		/* See include/net/tcp.h for the meaning of these knobs */
1273		if ((sysctl_tcp_fastopen & TFO_SERVER_ALWAYS) ||
1274		    ((sysctl_tcp_fastopen & TFO_SERVER_COOKIE_NOT_REQD) &&
1275		    (TCP_SKB_CB(skb)->end_seq != TCP_SKB_CB(skb)->seq + 1)))
1276			skip_cookie = true; /* no cookie to validate */
1277		else
1278			return false;
1279	}
1280	fastopenq = inet_csk(sk)->icsk_accept_queue.fastopenq;
1281	/* A FO option is present; bump the counter. */
1282	NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPFASTOPENPASSIVE);
1283
1284	/* Make sure the listener has enabled fastopen, and we don't
1285	 * exceed the max # of pending TFO requests allowed before trying
1286	 * to validating the cookie in order to avoid burning CPU cycles
1287	 * unnecessarily.
1288	 *
1289	 * XXX (TFO) - The implication of checking the max_qlen before
1290	 * processing a cookie request is that clients can't differentiate
1291	 * between qlen overflow causing Fast Open to be disabled
1292	 * temporarily vs a server not supporting Fast Open at all.
1293	 */
1294	if ((sysctl_tcp_fastopen & TFO_SERVER_ENABLE) == 0 ||
1295	    fastopenq == NULL || fastopenq->max_qlen == 0)
1296		return false;
1297
1298	if (fastopenq->qlen >= fastopenq->max_qlen) {
1299		struct request_sock *req1;
1300		spin_lock(&fastopenq->lock);
1301		req1 = fastopenq->rskq_rst_head;
1302		if ((req1 == NULL) || time_after(req1->expires, jiffies)) {
1303			spin_unlock(&fastopenq->lock);
1304			NET_INC_STATS_BH(sock_net(sk),
1305			    LINUX_MIB_TCPFASTOPENLISTENOVERFLOW);
1306			/* Avoid bumping LINUX_MIB_TCPFASTOPENPASSIVEFAIL*/
1307			foc->len = -1;
1308			return false;
1309		}
1310		fastopenq->rskq_rst_head = req1->dl_next;
1311		fastopenq->qlen--;
1312		spin_unlock(&fastopenq->lock);
1313		reqsk_free(req1);
1314	}
1315	if (skip_cookie) {
1316		tcp_rsk(req)->rcv_nxt = TCP_SKB_CB(skb)->end_seq;
1317		return true;
1318	}
1319
1320	if (foc->len == TCP_FASTOPEN_COOKIE_SIZE) {
1321		if ((sysctl_tcp_fastopen & TFO_SERVER_COOKIE_NOT_CHKED) == 0) {
1322			tcp_fastopen_cookie_gen(ip_hdr(skb)->saddr,
1323						ip_hdr(skb)->daddr, valid_foc);
1324			if ((valid_foc->len != TCP_FASTOPEN_COOKIE_SIZE) ||
1325			    memcmp(&foc->val[0], &valid_foc->val[0],
1326			    TCP_FASTOPEN_COOKIE_SIZE) != 0)
1327				return false;
1328			valid_foc->len = -1;
1329		}
1330		/* Acknowledge the data received from the peer. */
1331		tcp_rsk(req)->rcv_nxt = TCP_SKB_CB(skb)->end_seq;
1332		return true;
1333	} else if (foc->len == 0) { /* Client requesting a cookie */
1334		tcp_fastopen_cookie_gen(ip_hdr(skb)->saddr,
1335					ip_hdr(skb)->daddr, valid_foc);
1336		NET_INC_STATS_BH(sock_net(sk),
1337		    LINUX_MIB_TCPFASTOPENCOOKIEREQD);
1338	} else {
1339		/* Client sent a cookie with wrong size. Treat it
1340		 * the same as invalid and return a valid one.
1341		 */
1342		tcp_fastopen_cookie_gen(ip_hdr(skb)->saddr,
1343					ip_hdr(skb)->daddr, valid_foc);
1344	}
1345	return false;
1346}
1347
1348static int tcp_v4_conn_req_fastopen(struct sock *sk,
1349				    struct sk_buff *skb,
1350				    struct sk_buff *skb_synack,
1351				    struct request_sock *req)
1352{
1353	struct tcp_sock *tp = tcp_sk(sk);
1354	struct request_sock_queue *queue = &inet_csk(sk)->icsk_accept_queue;
1355	const struct inet_request_sock *ireq = inet_rsk(req);
1356	struct sock *child;
1357	int err;
1358
1359	req->num_retrans = 0;
1360	req->num_timeout = 0;
1361	req->sk = NULL;
1362
1363	child = inet_csk(sk)->icsk_af_ops->syn_recv_sock(sk, skb, req, NULL);
1364	if (child == NULL) {
1365		NET_INC_STATS_BH(sock_net(sk),
1366				 LINUX_MIB_TCPFASTOPENPASSIVEFAIL);
1367		kfree_skb(skb_synack);
1368		return -1;
1369	}
1370	err = ip_build_and_send_pkt(skb_synack, sk, ireq->ir_loc_addr,
1371				    ireq->ir_rmt_addr, ireq->opt);
1372	err = net_xmit_eval(err);
1373	if (!err)
1374		tcp_rsk(req)->snt_synack = tcp_time_stamp;
1375	/* XXX (TFO) - is it ok to ignore error and continue? */
1376
1377	spin_lock(&queue->fastopenq->lock);
1378	queue->fastopenq->qlen++;
1379	spin_unlock(&queue->fastopenq->lock);
1380
1381	/* Initialize the child socket. Have to fix some values to take
1382	 * into account the child is a Fast Open socket and is created
1383	 * only out of the bits carried in the SYN packet.
1384	 */
1385	tp = tcp_sk(child);
1386
1387	tp->fastopen_rsk = req;
1388	/* Do a hold on the listner sk so that if the listener is being
1389	 * closed, the child that has been accepted can live on and still
1390	 * access listen_lock.
1391	 */
1392	sock_hold(sk);
1393	tcp_rsk(req)->listener = sk;
1394
1395	/* RFC1323: The window in SYN & SYN/ACK segments is never
1396	 * scaled. So correct it appropriately.
1397	 */
1398	tp->snd_wnd = ntohs(tcp_hdr(skb)->window);
1399
1400	/* Activate the retrans timer so that SYNACK can be retransmitted.
1401	 * The request socket is not added to the SYN table of the parent
1402	 * because it's been added to the accept queue directly.
1403	 */
1404	inet_csk_reset_xmit_timer(child, ICSK_TIME_RETRANS,
1405	    TCP_TIMEOUT_INIT, TCP_RTO_MAX);
1406
1407	/* Add the child socket directly into the accept queue */
1408	inet_csk_reqsk_queue_add(sk, req, child);
1409
1410	/* Now finish processing the fastopen child socket. */
1411	inet_csk(child)->icsk_af_ops->rebuild_header(child);
1412	tcp_init_congestion_control(child);
1413	tcp_mtup_init(child);
1414	tcp_init_metrics(child);
1415	tcp_init_buffer_space(child);
1416
1417	/* Queue the data carried in the SYN packet. We need to first
1418	 * bump skb's refcnt because the caller will attempt to free it.
1419	 *
1420	 * XXX (TFO) - we honor a zero-payload TFO request for now.
1421	 * (Any reason not to?)
1422	 */
1423	if (TCP_SKB_CB(skb)->end_seq == TCP_SKB_CB(skb)->seq + 1) {
1424		/* Don't queue the skb if there is no payload in SYN.
1425		 * XXX (TFO) - How about SYN+FIN?
1426		 */
1427		tp->rcv_nxt = TCP_SKB_CB(skb)->end_seq;
1428	} else {
1429		skb = skb_get(skb);
1430		skb_dst_drop(skb);
1431		__skb_pull(skb, tcp_hdr(skb)->doff * 4);
1432		skb_set_owner_r(skb, child);
1433		__skb_queue_tail(&child->sk_receive_queue, skb);
1434		tp->rcv_nxt = TCP_SKB_CB(skb)->end_seq;
1435		tp->syn_data_acked = 1;
1436	}
1437	sk->sk_data_ready(sk);
1438	bh_unlock_sock(child);
1439	sock_put(child);
1440	WARN_ON(req->sk == NULL);
1441	return 0;
1442}
1443
1444int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb)
1445{
1446	struct tcp_options_received tmp_opt;
1447	struct request_sock *req;
1448	struct inet_request_sock *ireq;
1449	struct tcp_sock *tp = tcp_sk(sk);
1450	struct dst_entry *dst = NULL;
1451	__be32 saddr = ip_hdr(skb)->saddr;
1452	__be32 daddr = ip_hdr(skb)->daddr;
1453	__u32 isn = TCP_SKB_CB(skb)->when;
1454	bool want_cookie = false;
1455	struct flowi4 fl4;
1456	struct tcp_fastopen_cookie foc = { .len = -1 };
1457	struct tcp_fastopen_cookie valid_foc = { .len = -1 };
1458	struct sk_buff *skb_synack;
1459	int do_fastopen;
1460
1461	/* Never answer to SYNs send to broadcast or multicast */
1462	if (skb_rtable(skb)->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))
1463		goto drop;
1464
1465	/* TW buckets are converted to open requests without
1466	 * limitations, they conserve resources and peer is
1467	 * evidently real one.
1468	 */
1469	if ((sysctl_tcp_syncookies == 2 ||
1470	     inet_csk_reqsk_queue_is_full(sk)) && !isn) {
1471		want_cookie = tcp_syn_flood_action(sk, skb, "TCP");
1472		if (!want_cookie)
1473			goto drop;
1474	}
1475
1476	/* Accept backlog is full. If we have already queued enough
1477	 * of warm entries in syn queue, drop request. It is better than
1478	 * clogging syn queue with openreqs with exponentially increasing
1479	 * timeout.
1480	 */
1481	if (sk_acceptq_is_full(sk) && inet_csk_reqsk_queue_young(sk) > 1) {
1482		NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENOVERFLOWS);
1483		goto drop;
1484	}
1485
1486	req = inet_reqsk_alloc(&tcp_request_sock_ops);
1487	if (!req)
1488		goto drop;
1489
1490#ifdef CONFIG_TCP_MD5SIG
1491	tcp_rsk(req)->af_specific = &tcp_request_sock_ipv4_ops;
1492#endif
1493
1494	tcp_clear_options(&tmp_opt);
1495	tmp_opt.mss_clamp = TCP_MSS_DEFAULT;
1496	tmp_opt.user_mss  = tp->rx_opt.user_mss;
1497	tcp_parse_options(skb, &tmp_opt, 0, want_cookie ? NULL : &foc);
1498
1499	if (want_cookie && !tmp_opt.saw_tstamp)
1500		tcp_clear_options(&tmp_opt);
1501
1502	tmp_opt.tstamp_ok = tmp_opt.saw_tstamp;
1503	tcp_openreq_init(req, &tmp_opt, skb);
1504
1505	ireq = inet_rsk(req);
1506	ireq->ir_loc_addr = daddr;
1507	ireq->ir_rmt_addr = saddr;
1508	ireq->no_srccheck = inet_sk(sk)->transparent;
1509	ireq->opt = tcp_v4_save_options(skb);
1510
1511	if (security_inet_conn_request(sk, skb, req))
1512		goto drop_and_free;
1513
1514	if (!want_cookie || tmp_opt.tstamp_ok)
1515		TCP_ECN_create_request(req, skb, sock_net(sk));
1516
1517	if (want_cookie) {
1518		isn = cookie_v4_init_sequence(sk, skb, &req->mss);
1519		req->cookie_ts = tmp_opt.tstamp_ok;
1520	} else if (!isn) {
1521		/* VJ's idea. We save last timestamp seen
1522		 * from the destination in peer table, when entering
1523		 * state TIME-WAIT, and check against it before
1524		 * accepting new connection request.
1525		 *
1526		 * If "isn" is not zero, this request hit alive
1527		 * timewait bucket, so that all the necessary checks
1528		 * are made in the function processing timewait state.
1529		 */
1530		if (tmp_opt.saw_tstamp &&
1531		    tcp_death_row.sysctl_tw_recycle &&
1532		    (dst = inet_csk_route_req(sk, &fl4, req)) != NULL &&
1533		    fl4.daddr == saddr) {
1534			if (!tcp_peer_is_proven(req, dst, true)) {
1535				NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_PAWSPASSIVEREJECTED);
1536				goto drop_and_release;
1537			}
1538		}
1539		/* Kill the following clause, if you dislike this way. */
1540		else if (!sysctl_tcp_syncookies &&
1541			 (sysctl_max_syn_backlog - inet_csk_reqsk_queue_len(sk) <
1542			  (sysctl_max_syn_backlog >> 2)) &&
1543			 !tcp_peer_is_proven(req, dst, false)) {
1544			/* Without syncookies last quarter of
1545			 * backlog is filled with destinations,
1546			 * proven to be alive.
1547			 * It means that we continue to communicate
1548			 * to destinations, already remembered
1549			 * to the moment of synflood.
1550			 */
1551			LIMIT_NETDEBUG(KERN_DEBUG pr_fmt("drop open request from %pI4/%u\n"),
1552				       &saddr, ntohs(tcp_hdr(skb)->source));
1553			goto drop_and_release;
1554		}
1555
1556		isn = tcp_v4_init_sequence(skb);
1557	}
1558	tcp_rsk(req)->snt_isn = isn;
1559
1560	if (dst == NULL) {
1561		dst = inet_csk_route_req(sk, &fl4, req);
1562		if (dst == NULL)
1563			goto drop_and_free;
1564	}
1565	do_fastopen = tcp_fastopen_check(sk, skb, req, &foc, &valid_foc);
1566
1567	/* We don't call tcp_v4_send_synack() directly because we need
1568	 * to make sure a child socket can be created successfully before
1569	 * sending back synack!
1570	 *
1571	 * XXX (TFO) - Ideally one would simply call tcp_v4_send_synack()
1572	 * (or better yet, call tcp_send_synack() in the child context
1573	 * directly, but will have to fix bunch of other code first)
1574	 * after syn_recv_sock() except one will need to first fix the
1575	 * latter to remove its dependency on the current implementation
1576	 * of tcp_v4_send_synack()->tcp_select_initial_window().
1577	 */
1578	skb_synack = tcp_make_synack(sk, dst, req,
1579	    fastopen_cookie_present(&valid_foc) ? &valid_foc : NULL);
1580
1581	if (skb_synack) {
1582		__tcp_v4_send_check(skb_synack, ireq->ir_loc_addr, ireq->ir_rmt_addr);
1583		skb_set_queue_mapping(skb_synack, skb_get_queue_mapping(skb));
1584	} else
1585		goto drop_and_free;
1586
1587	if (likely(!do_fastopen)) {
1588		int err;
1589		err = ip_build_and_send_pkt(skb_synack, sk, ireq->ir_loc_addr,
1590		     ireq->ir_rmt_addr, ireq->opt);
1591		err = net_xmit_eval(err);
1592		if (err || want_cookie)
1593			goto drop_and_free;
1594
1595		tcp_rsk(req)->snt_synack = tcp_time_stamp;
1596		tcp_rsk(req)->listener = NULL;
1597		/* Add the request_sock to the SYN table */
1598		inet_csk_reqsk_queue_hash_add(sk, req, TCP_TIMEOUT_INIT);
1599		if (fastopen_cookie_present(&foc) && foc.len != 0)
1600			NET_INC_STATS_BH(sock_net(sk),
1601			    LINUX_MIB_TCPFASTOPENPASSIVEFAIL);
1602	} else if (tcp_v4_conn_req_fastopen(sk, skb, skb_synack, req))
1603		goto drop_and_free;
1604
1605	return 0;
1606
1607drop_and_release:
1608	dst_release(dst);
1609drop_and_free:
1610	reqsk_free(req);
1611drop:
1612	NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENDROPS);
1613	return 0;
1614}
1615EXPORT_SYMBOL(tcp_v4_conn_request);
1616
1617
1618/*
1619 * The three way handshake has completed - we got a valid synack -
1620 * now create the new socket.
1621 */
1622struct sock *tcp_v4_syn_recv_sock(struct sock *sk, struct sk_buff *skb,
1623				  struct request_sock *req,
1624				  struct dst_entry *dst)
 
 
1625{
1626	struct inet_request_sock *ireq;
1627	struct inet_sock *newinet;
1628	struct tcp_sock *newtp;
1629	struct sock *newsk;
1630#ifdef CONFIG_TCP_MD5SIG
 
1631	struct tcp_md5sig_key *key;
 
1632#endif
1633	struct ip_options_rcu *inet_opt;
1634
1635	if (sk_acceptq_is_full(sk))
1636		goto exit_overflow;
1637
1638	newsk = tcp_create_openreq_child(sk, req, skb);
1639	if (!newsk)
1640		goto exit_nonewsk;
1641
1642	newsk->sk_gso_type = SKB_GSO_TCPV4;
1643	inet_sk_rx_dst_set(newsk, skb);
1644
1645	newtp		      = tcp_sk(newsk);
1646	newinet		      = inet_sk(newsk);
1647	ireq		      = inet_rsk(req);
1648	newinet->inet_daddr   = ireq->ir_rmt_addr;
1649	newinet->inet_rcv_saddr = ireq->ir_loc_addr;
1650	newinet->inet_saddr	      = ireq->ir_loc_addr;
1651	inet_opt	      = ireq->opt;
1652	rcu_assign_pointer(newinet->inet_opt, inet_opt);
1653	ireq->opt	      = NULL;
1654	newinet->mc_index     = inet_iif(skb);
1655	newinet->mc_ttl	      = ip_hdr(skb)->ttl;
1656	newinet->rcv_tos      = ip_hdr(skb)->tos;
1657	inet_csk(newsk)->icsk_ext_hdr_len = 0;
1658	if (inet_opt)
1659		inet_csk(newsk)->icsk_ext_hdr_len = inet_opt->opt.optlen;
1660	newinet->inet_id = newtp->write_seq ^ jiffies;
1661
1662	if (!dst) {
1663		dst = inet_csk_route_child_sock(sk, newsk, req);
1664		if (!dst)
1665			goto put_and_exit;
1666	} else {
1667		/* syncookie case : see end of cookie_v4_check() */
1668	}
1669	sk_setup_caps(newsk, dst);
1670
 
 
1671	tcp_sync_mss(newsk, dst_mtu(dst));
1672	newtp->advmss = dst_metric_advmss(dst);
1673	if (tcp_sk(sk)->rx_opt.user_mss &&
1674	    tcp_sk(sk)->rx_opt.user_mss < newtp->advmss)
1675		newtp->advmss = tcp_sk(sk)->rx_opt.user_mss;
1676
1677	tcp_initialize_rcv_mss(newsk);
1678
1679#ifdef CONFIG_TCP_MD5SIG
 
1680	/* Copy over the MD5 key from the original socket */
1681	key = tcp_md5_do_lookup(sk, (union tcp_md5_addr *)&newinet->inet_daddr,
1682				AF_INET);
1683	if (key != NULL) {
1684		/*
1685		 * We're using one, so create a matching key
1686		 * on the newsk structure. If we fail to get
1687		 * memory, then we end up not copying the key
1688		 * across. Shucks.
1689		 */
1690		tcp_md5_do_add(newsk, (union tcp_md5_addr *)&newinet->inet_daddr,
1691			       AF_INET, key->key, key->keylen, GFP_ATOMIC);
1692		sk_nocaps_add(newsk, NETIF_F_GSO_MASK);
1693	}
1694#endif
1695
1696	if (__inet_inherit_port(sk, newsk) < 0)
1697		goto put_and_exit;
1698	__inet_hash_nolisten(newsk, NULL);
1699
 
 
 
 
 
1700	return newsk;
1701
1702exit_overflow:
1703	NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENOVERFLOWS);
1704exit_nonewsk:
1705	dst_release(dst);
1706exit:
1707	NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENDROPS);
1708	return NULL;
1709put_and_exit:
 
1710	inet_csk_prepare_forced_close(newsk);
1711	tcp_done(newsk);
1712	goto exit;
1713}
1714EXPORT_SYMBOL(tcp_v4_syn_recv_sock);
1715
1716static struct sock *tcp_v4_hnd_req(struct sock *sk, struct sk_buff *skb)
1717{
1718	struct tcphdr *th = tcp_hdr(skb);
1719	const struct iphdr *iph = ip_hdr(skb);
1720	struct sock *nsk;
1721	struct request_sock **prev;
1722	/* Find possible connection requests. */
1723	struct request_sock *req = inet_csk_search_req(sk, &prev, th->source,
1724						       iph->saddr, iph->daddr);
1725	if (req)
1726		return tcp_check_req(sk, skb, req, prev, false);
1727
1728	nsk = inet_lookup_established(sock_net(sk), &tcp_hashinfo, iph->saddr,
1729			th->source, iph->daddr, th->dest, inet_iif(skb));
1730
1731	if (nsk) {
1732		if (nsk->sk_state != TCP_TIME_WAIT) {
1733			bh_lock_sock(nsk);
1734			return nsk;
1735		}
1736		inet_twsk_put(inet_twsk(nsk));
1737		return NULL;
1738	}
1739
1740#ifdef CONFIG_SYN_COOKIES
1741	if (!th->syn)
1742		sk = cookie_v4_check(sk, skb, &(IPCB(skb)->opt));
1743#endif
1744	return sk;
1745}
1746
1747static __sum16 tcp_v4_checksum_init(struct sk_buff *skb)
 
1748{
1749	const struct iphdr *iph = ip_hdr(skb);
1750
1751	if (skb->ip_summed == CHECKSUM_COMPLETE) {
1752		if (!tcp_v4_check(skb->len, iph->saddr,
1753				  iph->daddr, skb->csum)) {
1754			skb->ip_summed = CHECKSUM_UNNECESSARY;
1755			return 0;
1756		}
1757	}
1758
1759	skb->csum = csum_tcpudp_nofold(iph->saddr, iph->daddr,
1760				       skb->len, IPPROTO_TCP, 0);
1761
1762	if (skb->len <= 76) {
1763		return __skb_checksum_complete(skb);
1764	}
1765	return 0;
1766}
1767
1768
1769/* The socket must have it's spinlock held when we get
1770 * here.
1771 *
1772 * We have a potential double-lock case here, so even when
1773 * doing backlog processing we use the BH locking scheme.
1774 * This is because we cannot sleep with the original spinlock
1775 * held.
1776 */
1777int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb)
1778{
1779	struct sock *rsk;
1780#ifdef CONFIG_TCP_MD5SIG
1781	/*
1782	 * We really want to reject the packet as early as possible
1783	 * if:
1784	 *  o We're expecting an MD5'd packet and this is no MD5 tcp option
1785	 *  o There is an MD5 option and we're not expecting one
1786	 */
1787	if (tcp_v4_inbound_md5_hash(sk, skb))
1788		goto discard;
1789#endif
1790
1791	if (sk->sk_state == TCP_ESTABLISHED) { /* Fast path */
1792		struct dst_entry *dst = sk->sk_rx_dst;
1793
1794		sock_rps_save_rxhash(sk, skb);
 
1795		if (dst) {
1796			if (inet_sk(sk)->rx_dst_ifindex != skb->skb_iif ||
1797			    dst->ops->check(dst, 0) == NULL) {
1798				dst_release(dst);
1799				sk->sk_rx_dst = NULL;
1800			}
1801		}
1802		tcp_rcv_established(sk, skb, tcp_hdr(skb), skb->len);
1803		return 0;
1804	}
1805
1806	if (skb->len < tcp_hdrlen(skb) || tcp_checksum_complete(skb))
1807		goto csum_err;
1808
1809	if (sk->sk_state == TCP_LISTEN) {
1810		struct sock *nsk = tcp_v4_hnd_req(sk, skb);
 
1811		if (!nsk)
1812			goto discard;
1813
1814		if (nsk != sk) {
1815			sock_rps_save_rxhash(nsk, skb);
1816			if (tcp_child_process(sk, nsk, skb)) {
1817				rsk = nsk;
1818				goto reset;
1819			}
1820			return 0;
1821		}
1822	} else
1823		sock_rps_save_rxhash(sk, skb);
1824
1825	if (tcp_rcv_state_process(sk, skb, tcp_hdr(skb), skb->len)) {
1826		rsk = sk;
1827		goto reset;
1828	}
1829	return 0;
1830
1831reset:
1832	tcp_v4_send_reset(rsk, skb);
1833discard:
1834	kfree_skb(skb);
1835	/* Be careful here. If this function gets more complicated and
1836	 * gcc suffers from register pressure on the x86, sk (in %ebx)
1837	 * might be destroyed here. This current version compiles correctly,
1838	 * but you have been warned.
1839	 */
1840	return 0;
1841
1842csum_err:
1843	TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_CSUMERRORS);
1844	TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_INERRS);
1845	goto discard;
1846}
1847EXPORT_SYMBOL(tcp_v4_do_rcv);
1848
1849void tcp_v4_early_demux(struct sk_buff *skb)
1850{
1851	const struct iphdr *iph;
1852	const struct tcphdr *th;
1853	struct sock *sk;
1854
1855	if (skb->pkt_type != PACKET_HOST)
1856		return;
1857
1858	if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct tcphdr)))
1859		return;
1860
1861	iph = ip_hdr(skb);
1862	th = tcp_hdr(skb);
1863
1864	if (th->doff < sizeof(struct tcphdr) / 4)
1865		return;
1866
1867	sk = __inet_lookup_established(dev_net(skb->dev), &tcp_hashinfo,
1868				       iph->saddr, th->source,
1869				       iph->daddr, ntohs(th->dest),
1870				       skb->skb_iif);
1871	if (sk) {
1872		skb->sk = sk;
1873		skb->destructor = sock_edemux;
1874		if (sk->sk_state != TCP_TIME_WAIT) {
1875			struct dst_entry *dst = sk->sk_rx_dst;
1876
1877			if (dst)
1878				dst = dst_check(dst, 0);
1879			if (dst &&
1880			    inet_sk(sk)->rx_dst_ifindex == skb->skb_iif)
1881				skb_dst_set_noref(skb, dst);
1882		}
1883	}
 
1884}
1885
1886/* Packet is added to VJ-style prequeue for processing in process
1887 * context, if a reader task is waiting. Apparently, this exciting
1888 * idea (VJ's mail "Re: query about TCP header on tcp-ip" of 07 Sep 93)
1889 * failed somewhere. Latency? Burstiness? Well, at least now we will
1890 * see, why it failed. 8)8)				  --ANK
1891 *
1892 */
1893bool tcp_prequeue(struct sock *sk, struct sk_buff *skb)
1894{
1895	struct tcp_sock *tp = tcp_sk(sk);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1896
1897	if (sysctl_tcp_low_latency || !tp->ucopy.task)
1898		return false;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1899
1900	if (skb->len <= tcp_hdrlen(skb) &&
1901	    skb_queue_len(&tp->ucopy.prequeue) == 0)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1902		return false;
 
 
1903
1904	skb_dst_force(skb);
1905	__skb_queue_tail(&tp->ucopy.prequeue, skb);
1906	tp->ucopy.memory += skb->truesize;
1907	if (tp->ucopy.memory > sk->sk_rcvbuf) {
1908		struct sk_buff *skb1;
1909
1910		BUG_ON(sock_owned_by_user(sk));
1911
1912		while ((skb1 = __skb_dequeue(&tp->ucopy.prequeue)) != NULL) {
1913			sk_backlog_rcv(sk, skb1);
1914			NET_INC_STATS_BH(sock_net(sk),
1915					 LINUX_MIB_TCPPREQUEUEDROPPED);
1916		}
1917
1918		tp->ucopy.memory = 0;
1919	} else if (skb_queue_len(&tp->ucopy.prequeue) == 1) {
1920		wake_up_interruptible_sync_poll(sk_sleep(sk),
1921					   POLLIN | POLLRDNORM | POLLRDBAND);
1922		if (!inet_csk_ack_scheduled(sk))
1923			inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK,
1924						  (3 * tcp_rto_min(sk)) / 4,
1925						  TCP_RTO_MAX);
1926	}
1927	return true;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1928}
1929EXPORT_SYMBOL(tcp_prequeue);
1930
1931/*
1932 *	From tcp_input.c
1933 */
1934
1935int tcp_v4_rcv(struct sk_buff *skb)
1936{
 
 
 
 
1937	const struct iphdr *iph;
1938	const struct tcphdr *th;
 
1939	struct sock *sk;
1940	int ret;
1941	struct net *net = dev_net(skb->dev);
1942
1943	if (skb->pkt_type != PACKET_HOST)
1944		goto discard_it;
1945
1946	/* Count it even if it's bad */
1947	TCP_INC_STATS_BH(net, TCP_MIB_INSEGS);
1948
1949	if (!pskb_may_pull(skb, sizeof(struct tcphdr)))
1950		goto discard_it;
1951
1952	th = tcp_hdr(skb);
1953
1954	if (th->doff < sizeof(struct tcphdr) / 4)
1955		goto bad_packet;
1956	if (!pskb_may_pull(skb, th->doff * 4))
1957		goto discard_it;
1958
1959	/* An explanation is required here, I think.
1960	 * Packet length and doff are validated by header prediction,
1961	 * provided case of th->doff==0 is eliminated.
1962	 * So, we defer the checks. */
1963	if (!skb_csum_unnecessary(skb) && tcp_v4_checksum_init(skb))
 
1964		goto csum_error;
1965
1966	th = tcp_hdr(skb);
1967	iph = ip_hdr(skb);
1968	TCP_SKB_CB(skb)->seq = ntohl(th->seq);
1969	TCP_SKB_CB(skb)->end_seq = (TCP_SKB_CB(skb)->seq + th->syn + th->fin +
1970				    skb->len - th->doff * 4);
1971	TCP_SKB_CB(skb)->ack_seq = ntohl(th->ack_seq);
1972	TCP_SKB_CB(skb)->when	 = 0;
1973	TCP_SKB_CB(skb)->ip_dsfield = ipv4_get_dsfield(iph);
1974	TCP_SKB_CB(skb)->sacked	 = 0;
1975
1976	sk = __inet_lookup_skb(&tcp_hashinfo, skb, th->source, th->dest);
1977	if (!sk)
1978		goto no_tcp_socket;
1979
1980process:
1981	if (sk->sk_state == TCP_TIME_WAIT)
1982		goto do_time_wait;
1983
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1984	if (unlikely(iph->ttl < inet_sk(sk)->min_ttl)) {
1985		NET_INC_STATS_BH(net, LINUX_MIB_TCPMINTTLDROP);
1986		goto discard_and_relse;
1987	}
1988
1989	if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
1990		goto discard_and_relse;
1991	nf_reset(skb);
1992
1993	if (sk_filter(sk, skb))
1994		goto discard_and_relse;
1995
1996	sk_mark_napi_id(sk, skb);
 
 
 
 
 
 
 
1997	skb->dev = NULL;
1998
 
 
 
 
 
 
 
1999	bh_lock_sock_nested(sk);
 
2000	ret = 0;
2001	if (!sock_owned_by_user(sk)) {
2002#ifdef CONFIG_NET_DMA
2003		struct tcp_sock *tp = tcp_sk(sk);
2004		if (!tp->ucopy.dma_chan && tp->ucopy.pinned_list)
2005			tp->ucopy.dma_chan = net_dma_find_channel();
2006		if (tp->ucopy.dma_chan)
2007			ret = tcp_v4_do_rcv(sk, skb);
2008		else
2009#endif
2010		{
2011			if (!tcp_prequeue(sk, skb))
2012				ret = tcp_v4_do_rcv(sk, skb);
2013		}
2014	} else if (unlikely(sk_add_backlog(sk, skb,
2015					   sk->sk_rcvbuf + sk->sk_sndbuf))) {
2016		bh_unlock_sock(sk);
2017		NET_INC_STATS_BH(net, LINUX_MIB_TCPBACKLOGDROP);
2018		goto discard_and_relse;
2019	}
2020	bh_unlock_sock(sk);
 
 
2021
2022	sock_put(sk);
 
 
2023
2024	return ret;
2025
2026no_tcp_socket:
2027	if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
2028		goto discard_it;
2029
2030	if (skb->len < (th->doff << 2) || tcp_checksum_complete(skb)) {
 
 
2031csum_error:
2032		TCP_INC_STATS_BH(net, TCP_MIB_CSUMERRORS);
2033bad_packet:
2034		TCP_INC_STATS_BH(net, TCP_MIB_INERRS);
2035	} else {
2036		tcp_v4_send_reset(NULL, skb);
2037	}
2038
2039discard_it:
2040	/* Discard frame. */
2041	kfree_skb(skb);
2042	return 0;
2043
2044discard_and_relse:
2045	sock_put(sk);
 
 
2046	goto discard_it;
2047
2048do_time_wait:
2049	if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) {
2050		inet_twsk_put(inet_twsk(sk));
2051		goto discard_it;
2052	}
2053
2054	if (skb->len < (th->doff << 2)) {
2055		inet_twsk_put(inet_twsk(sk));
2056		goto bad_packet;
2057	}
2058	if (tcp_checksum_complete(skb)) {
2059		inet_twsk_put(inet_twsk(sk));
2060		goto csum_error;
2061	}
2062	switch (tcp_timewait_state_process(inet_twsk(sk), skb, th)) {
2063	case TCP_TW_SYN: {
2064		struct sock *sk2 = inet_lookup_listener(dev_net(skb->dev),
2065							&tcp_hashinfo,
 
2066							iph->saddr, th->source,
2067							iph->daddr, th->dest,
2068							inet_iif(skb));
 
2069		if (sk2) {
2070			inet_twsk_deschedule(inet_twsk(sk), &tcp_death_row);
2071			inet_twsk_put(inet_twsk(sk));
2072			sk = sk2;
 
 
2073			goto process;
2074		}
2075		/* Fall through to ACK */
2076	}
 
 
2077	case TCP_TW_ACK:
2078		tcp_v4_timewait_ack(sk, skb);
2079		break;
2080	case TCP_TW_RST:
2081		goto no_tcp_socket;
 
 
2082	case TCP_TW_SUCCESS:;
2083	}
2084	goto discard_it;
2085}
2086
2087static struct timewait_sock_ops tcp_timewait_sock_ops = {
2088	.twsk_obj_size	= sizeof(struct tcp_timewait_sock),
2089	.twsk_unique	= tcp_twsk_unique,
2090	.twsk_destructor= tcp_twsk_destructor,
2091};
2092
2093void inet_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb)
2094{
2095	struct dst_entry *dst = skb_dst(skb);
2096
2097	dst_hold(dst);
2098	sk->sk_rx_dst = dst;
2099	inet_sk(sk)->rx_dst_ifindex = skb->skb_iif;
 
2100}
2101EXPORT_SYMBOL(inet_sk_rx_dst_set);
2102
2103const struct inet_connection_sock_af_ops ipv4_specific = {
2104	.queue_xmit	   = ip_queue_xmit,
2105	.send_check	   = tcp_v4_send_check,
2106	.rebuild_header	   = inet_sk_rebuild_header,
2107	.sk_rx_dst_set	   = inet_sk_rx_dst_set,
2108	.conn_request	   = tcp_v4_conn_request,
2109	.syn_recv_sock	   = tcp_v4_syn_recv_sock,
2110	.net_header_len	   = sizeof(struct iphdr),
2111	.setsockopt	   = ip_setsockopt,
2112	.getsockopt	   = ip_getsockopt,
2113	.addr2sockaddr	   = inet_csk_addr2sockaddr,
2114	.sockaddr_len	   = sizeof(struct sockaddr_in),
2115	.bind_conflict	   = inet_csk_bind_conflict,
2116#ifdef CONFIG_COMPAT
2117	.compat_setsockopt = compat_ip_setsockopt,
2118	.compat_getsockopt = compat_ip_getsockopt,
2119#endif
2120};
2121EXPORT_SYMBOL(ipv4_specific);
2122
2123#ifdef CONFIG_TCP_MD5SIG
2124static const struct tcp_sock_af_ops tcp_sock_ipv4_specific = {
2125	.md5_lookup		= tcp_v4_md5_lookup,
2126	.calc_md5_hash		= tcp_v4_md5_hash_skb,
2127	.md5_parse		= tcp_v4_parse_md5_keys,
2128};
2129#endif
2130
2131/* NOTE: A lot of things set to zero explicitly by call to
2132 *       sk_alloc() so need not be done here.
2133 */
2134static int tcp_v4_init_sock(struct sock *sk)
2135{
2136	struct inet_connection_sock *icsk = inet_csk(sk);
2137
2138	tcp_init_sock(sk);
2139
2140	icsk->icsk_af_ops = &ipv4_specific;
2141
2142#ifdef CONFIG_TCP_MD5SIG
2143	tcp_sk(sk)->af_specific = &tcp_sock_ipv4_specific;
2144#endif
2145
2146	return 0;
2147}
2148
2149void tcp_v4_destroy_sock(struct sock *sk)
2150{
2151	struct tcp_sock *tp = tcp_sk(sk);
2152
 
 
2153	tcp_clear_xmit_timers(sk);
2154
2155	tcp_cleanup_congestion_control(sk);
2156
 
 
2157	/* Cleanup up the write buffer. */
2158	tcp_write_queue_purge(sk);
2159
 
 
 
2160	/* Cleans up our, hopefully empty, out_of_order_queue. */
2161	__skb_queue_purge(&tp->out_of_order_queue);
2162
2163#ifdef CONFIG_TCP_MD5SIG
2164	/* Clean up the MD5 key list, if any */
2165	if (tp->md5sig_info) {
2166		tcp_clear_md5_list(sk);
2167		kfree_rcu(tp->md5sig_info, rcu);
2168		tp->md5sig_info = NULL;
2169	}
2170#endif
2171
2172#ifdef CONFIG_NET_DMA
2173	/* Cleans up our sk_async_wait_queue */
2174	__skb_queue_purge(&sk->sk_async_wait_queue);
2175#endif
2176
2177	/* Clean prequeue, it must be empty really */
2178	__skb_queue_purge(&tp->ucopy.prequeue);
2179
2180	/* Clean up a referenced TCP bind bucket. */
2181	if (inet_csk(sk)->icsk_bind_hash)
2182		inet_put_port(sk);
2183
2184	BUG_ON(tp->fastopen_rsk != NULL);
2185
2186	/* If socket is aborted during connect operation */
2187	tcp_free_fastopen_req(tp);
 
 
2188
2189	sk_sockets_allocated_dec(sk);
2190	sock_release_memcg(sk);
2191}
2192EXPORT_SYMBOL(tcp_v4_destroy_sock);
2193
2194#ifdef CONFIG_PROC_FS
2195/* Proc filesystem TCP sock list dumping. */
2196
2197/*
2198 * Get next listener socket follow cur.  If cur is NULL, get first socket
2199 * starting from bucket given in st->bucket; when st->bucket is zero the
2200 * very first socket in the hash table is returned.
2201 */
2202static void *listening_get_next(struct seq_file *seq, void *cur)
2203{
2204	struct inet_connection_sock *icsk;
 
 
 
2205	struct hlist_nulls_node *node;
2206	struct sock *sk = cur;
2207	struct inet_listen_hashbucket *ilb;
2208	struct tcp_iter_state *st = seq->private;
2209	struct net *net = seq_file_net(seq);
 
 
2210
2211	if (!sk) {
 
2212		ilb = &tcp_hashinfo.listening_hash[st->bucket];
2213		spin_lock_bh(&ilb->lock);
2214		sk = sk_nulls_head(&ilb->head);
2215		st->offset = 0;
2216		goto get_sk;
2217	}
2218	ilb = &tcp_hashinfo.listening_hash[st->bucket];
2219	++st->num;
2220	++st->offset;
2221
2222	if (st->state == TCP_SEQ_STATE_OPENREQ) {
2223		struct request_sock *req = cur;
2224
2225		icsk = inet_csk(st->syn_wait_sk);
2226		req = req->dl_next;
2227		while (1) {
2228			while (req) {
2229				if (req->rsk_ops->family == st->family) {
2230					cur = req;
2231					goto out;
2232				}
2233				req = req->dl_next;
2234			}
2235			if (++st->sbucket >= icsk->icsk_accept_queue.listen_opt->nr_table_entries)
2236				break;
2237get_req:
2238			req = icsk->icsk_accept_queue.listen_opt->syn_table[st->sbucket];
2239		}
2240		sk	  = sk_nulls_next(st->syn_wait_sk);
2241		st->state = TCP_SEQ_STATE_LISTENING;
2242		read_unlock_bh(&icsk->icsk_accept_queue.syn_wait_lock);
2243	} else {
2244		icsk = inet_csk(sk);
2245		read_lock_bh(&icsk->icsk_accept_queue.syn_wait_lock);
2246		if (reqsk_queue_len(&icsk->icsk_accept_queue))
2247			goto start_req;
2248		read_unlock_bh(&icsk->icsk_accept_queue.syn_wait_lock);
2249		sk = sk_nulls_next(sk);
2250	}
2251get_sk:
2252	sk_nulls_for_each_from(sk, node) {
2253		if (!net_eq(sock_net(sk), net))
2254			continue;
2255		if (sk->sk_family == st->family) {
2256			cur = sk;
2257			goto out;
2258		}
2259		icsk = inet_csk(sk);
2260		read_lock_bh(&icsk->icsk_accept_queue.syn_wait_lock);
2261		if (reqsk_queue_len(&icsk->icsk_accept_queue)) {
2262start_req:
2263			st->uid		= sock_i_uid(sk);
2264			st->syn_wait_sk = sk;
2265			st->state	= TCP_SEQ_STATE_OPENREQ;
2266			st->sbucket	= 0;
2267			goto get_req;
2268		}
2269		read_unlock_bh(&icsk->icsk_accept_queue.syn_wait_lock);
2270	}
2271	spin_unlock_bh(&ilb->lock);
2272	st->offset = 0;
2273	if (++st->bucket < INET_LHTABLE_SIZE) {
2274		ilb = &tcp_hashinfo.listening_hash[st->bucket];
2275		spin_lock_bh(&ilb->lock);
2276		sk = sk_nulls_head(&ilb->head);
2277		goto get_sk;
2278	}
2279	cur = NULL;
2280out:
2281	return cur;
2282}
2283
2284static void *listening_get_idx(struct seq_file *seq, loff_t *pos)
2285{
2286	struct tcp_iter_state *st = seq->private;
2287	void *rc;
2288
2289	st->bucket = 0;
2290	st->offset = 0;
2291	rc = listening_get_next(seq, NULL);
2292
2293	while (rc && *pos) {
2294		rc = listening_get_next(seq, rc);
2295		--*pos;
2296	}
2297	return rc;
2298}
2299
2300static inline bool empty_bucket(const struct tcp_iter_state *st)
2301{
2302	return hlist_nulls_empty(&tcp_hashinfo.ehash[st->bucket].chain);
2303}
2304
2305/*
2306 * Get first established socket starting from bucket given in st->bucket.
2307 * If st->bucket is zero, the very first socket in the hash is returned.
2308 */
2309static void *established_get_first(struct seq_file *seq)
2310{
 
2311	struct tcp_iter_state *st = seq->private;
2312	struct net *net = seq_file_net(seq);
2313	void *rc = NULL;
2314
 
 
 
 
 
2315	st->offset = 0;
2316	for (; st->bucket <= tcp_hashinfo.ehash_mask; ++st->bucket) {
2317		struct sock *sk;
2318		struct hlist_nulls_node *node;
2319		spinlock_t *lock = inet_ehash_lockp(&tcp_hashinfo, st->bucket);
2320
2321		/* Lockless fast path for the common case of empty buckets */
2322		if (empty_bucket(st))
2323			continue;
2324
2325		spin_lock_bh(lock);
2326		sk_nulls_for_each(sk, node, &tcp_hashinfo.ehash[st->bucket].chain) {
2327			if (sk->sk_family != st->family ||
 
2328			    !net_eq(sock_net(sk), net)) {
2329				continue;
2330			}
2331			rc = sk;
2332			goto out;
2333		}
2334		spin_unlock_bh(lock);
2335	}
2336out:
2337	return rc;
2338}
2339
2340static void *established_get_next(struct seq_file *seq, void *cur)
2341{
 
2342	struct sock *sk = cur;
2343	struct hlist_nulls_node *node;
2344	struct tcp_iter_state *st = seq->private;
2345	struct net *net = seq_file_net(seq);
2346
 
 
 
 
 
2347	++st->num;
2348	++st->offset;
2349
2350	sk = sk_nulls_next(sk);
2351
2352	sk_nulls_for_each_from(sk, node) {
2353		if (sk->sk_family == st->family && net_eq(sock_net(sk), net))
 
 
2354			return sk;
2355	}
2356
2357	spin_unlock_bh(inet_ehash_lockp(&tcp_hashinfo, st->bucket));
2358	++st->bucket;
2359	return established_get_first(seq);
2360}
2361
2362static void *established_get_idx(struct seq_file *seq, loff_t pos)
2363{
2364	struct tcp_iter_state *st = seq->private;
2365	void *rc;
2366
2367	st->bucket = 0;
2368	rc = established_get_first(seq);
2369
2370	while (rc && pos) {
2371		rc = established_get_next(seq, rc);
2372		--pos;
2373	}
2374	return rc;
2375}
2376
2377static void *tcp_get_idx(struct seq_file *seq, loff_t pos)
2378{
2379	void *rc;
2380	struct tcp_iter_state *st = seq->private;
2381
2382	st->state = TCP_SEQ_STATE_LISTENING;
2383	rc	  = listening_get_idx(seq, &pos);
2384
2385	if (!rc) {
2386		st->state = TCP_SEQ_STATE_ESTABLISHED;
2387		rc	  = established_get_idx(seq, pos);
2388	}
2389
2390	return rc;
2391}
2392
2393static void *tcp_seek_last_pos(struct seq_file *seq)
2394{
2395	struct tcp_iter_state *st = seq->private;
2396	int offset = st->offset;
2397	int orig_num = st->num;
2398	void *rc = NULL;
2399
2400	switch (st->state) {
2401	case TCP_SEQ_STATE_OPENREQ:
2402	case TCP_SEQ_STATE_LISTENING:
2403		if (st->bucket >= INET_LHTABLE_SIZE)
2404			break;
2405		st->state = TCP_SEQ_STATE_LISTENING;
2406		rc = listening_get_next(seq, NULL);
2407		while (offset-- && rc)
2408			rc = listening_get_next(seq, rc);
2409		if (rc)
2410			break;
2411		st->bucket = 0;
2412		st->state = TCP_SEQ_STATE_ESTABLISHED;
2413		/* Fallthrough */
2414	case TCP_SEQ_STATE_ESTABLISHED:
2415		if (st->bucket > tcp_hashinfo.ehash_mask)
2416			break;
2417		rc = established_get_first(seq);
2418		while (offset-- && rc)
2419			rc = established_get_next(seq, rc);
2420	}
2421
2422	st->num = orig_num;
2423
2424	return rc;
2425}
2426
2427static void *tcp_seq_start(struct seq_file *seq, loff_t *pos)
2428{
2429	struct tcp_iter_state *st = seq->private;
2430	void *rc;
2431
2432	if (*pos && *pos == st->last_pos) {
2433		rc = tcp_seek_last_pos(seq);
2434		if (rc)
2435			goto out;
2436	}
2437
2438	st->state = TCP_SEQ_STATE_LISTENING;
2439	st->num = 0;
2440	st->bucket = 0;
2441	st->offset = 0;
2442	rc = *pos ? tcp_get_idx(seq, *pos - 1) : SEQ_START_TOKEN;
2443
2444out:
2445	st->last_pos = *pos;
2446	return rc;
2447}
 
2448
2449static void *tcp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2450{
2451	struct tcp_iter_state *st = seq->private;
2452	void *rc = NULL;
2453
2454	if (v == SEQ_START_TOKEN) {
2455		rc = tcp_get_idx(seq, 0);
2456		goto out;
2457	}
2458
2459	switch (st->state) {
2460	case TCP_SEQ_STATE_OPENREQ:
2461	case TCP_SEQ_STATE_LISTENING:
2462		rc = listening_get_next(seq, v);
2463		if (!rc) {
2464			st->state = TCP_SEQ_STATE_ESTABLISHED;
2465			st->bucket = 0;
2466			st->offset = 0;
2467			rc	  = established_get_first(seq);
2468		}
2469		break;
2470	case TCP_SEQ_STATE_ESTABLISHED:
2471		rc = established_get_next(seq, v);
2472		break;
2473	}
2474out:
2475	++*pos;
2476	st->last_pos = *pos;
2477	return rc;
2478}
 
2479
2480static void tcp_seq_stop(struct seq_file *seq, void *v)
2481{
2482	struct tcp_iter_state *st = seq->private;
2483
2484	switch (st->state) {
2485	case TCP_SEQ_STATE_OPENREQ:
2486		if (v) {
2487			struct inet_connection_sock *icsk = inet_csk(st->syn_wait_sk);
2488			read_unlock_bh(&icsk->icsk_accept_queue.syn_wait_lock);
2489		}
2490	case TCP_SEQ_STATE_LISTENING:
2491		if (v != SEQ_START_TOKEN)
2492			spin_unlock_bh(&tcp_hashinfo.listening_hash[st->bucket].lock);
2493		break;
2494	case TCP_SEQ_STATE_ESTABLISHED:
2495		if (v)
2496			spin_unlock_bh(inet_ehash_lockp(&tcp_hashinfo, st->bucket));
2497		break;
2498	}
2499}
 
2500
2501int tcp_seq_open(struct inode *inode, struct file *file)
2502{
2503	struct tcp_seq_afinfo *afinfo = PDE_DATA(inode);
2504	struct tcp_iter_state *s;
2505	int err;
2506
2507	err = seq_open_net(inode, file, &afinfo->seq_ops,
2508			  sizeof(struct tcp_iter_state));
2509	if (err < 0)
2510		return err;
2511
2512	s = ((struct seq_file *)file->private_data)->private;
2513	s->family		= afinfo->family;
2514	s->last_pos 		= 0;
2515	return 0;
2516}
2517EXPORT_SYMBOL(tcp_seq_open);
2518
2519int tcp_proc_register(struct net *net, struct tcp_seq_afinfo *afinfo)
2520{
2521	int rc = 0;
2522	struct proc_dir_entry *p;
2523
2524	afinfo->seq_ops.start		= tcp_seq_start;
2525	afinfo->seq_ops.next		= tcp_seq_next;
2526	afinfo->seq_ops.stop		= tcp_seq_stop;
2527
2528	p = proc_create_data(afinfo->name, S_IRUGO, net->proc_net,
2529			     afinfo->seq_fops, afinfo);
2530	if (!p)
2531		rc = -ENOMEM;
2532	return rc;
2533}
2534EXPORT_SYMBOL(tcp_proc_register);
2535
2536void tcp_proc_unregister(struct net *net, struct tcp_seq_afinfo *afinfo)
2537{
2538	remove_proc_entry(afinfo->name, net->proc_net);
2539}
2540EXPORT_SYMBOL(tcp_proc_unregister);
2541
2542static void get_openreq4(const struct sock *sk, const struct request_sock *req,
2543			 struct seq_file *f, int i, kuid_t uid)
2544{
2545	const struct inet_request_sock *ireq = inet_rsk(req);
2546	long delta = req->expires - jiffies;
2547
2548	seq_printf(f, "%4d: %08X:%04X %08X:%04X"
2549		" %02X %08X:%08X %02X:%08lX %08X %5u %8d %u %d %pK",
2550		i,
2551		ireq->ir_loc_addr,
2552		ntohs(inet_sk(sk)->inet_sport),
2553		ireq->ir_rmt_addr,
2554		ntohs(ireq->ir_rmt_port),
2555		TCP_SYN_RECV,
2556		0, 0, /* could print option size, but that is af dependent. */
2557		1,    /* timers active (only the expire timer) */
2558		jiffies_delta_to_clock_t(delta),
2559		req->num_timeout,
2560		from_kuid_munged(seq_user_ns(f), uid),
 
2561		0,  /* non standard timer */
2562		0, /* open_requests have no inode */
2563		atomic_read(&sk->sk_refcnt),
2564		req);
2565}
2566
2567static void get_tcp4_sock(struct sock *sk, struct seq_file *f, int i)
2568{
2569	int timer_active;
2570	unsigned long timer_expires;
2571	const struct tcp_sock *tp = tcp_sk(sk);
2572	const struct inet_connection_sock *icsk = inet_csk(sk);
2573	const struct inet_sock *inet = inet_sk(sk);
2574	struct fastopen_queue *fastopenq = icsk->icsk_accept_queue.fastopenq;
2575	__be32 dest = inet->inet_daddr;
2576	__be32 src = inet->inet_rcv_saddr;
2577	__u16 destp = ntohs(inet->inet_dport);
2578	__u16 srcp = ntohs(inet->inet_sport);
2579	int rx_queue;
 
2580
2581	if (icsk->icsk_pending == ICSK_TIME_RETRANS ||
2582	    icsk->icsk_pending == ICSK_TIME_EARLY_RETRANS ||
2583	    icsk->icsk_pending == ICSK_TIME_LOSS_PROBE) {
2584		timer_active	= 1;
2585		timer_expires	= icsk->icsk_timeout;
2586	} else if (icsk->icsk_pending == ICSK_TIME_PROBE0) {
2587		timer_active	= 4;
2588		timer_expires	= icsk->icsk_timeout;
2589	} else if (timer_pending(&sk->sk_timer)) {
2590		timer_active	= 2;
2591		timer_expires	= sk->sk_timer.expires;
2592	} else {
2593		timer_active	= 0;
2594		timer_expires = jiffies;
2595	}
2596
2597	if (sk->sk_state == TCP_LISTEN)
2598		rx_queue = sk->sk_ack_backlog;
 
2599	else
2600		/*
2601		 * because we dont lock socket, we might find a transient negative value
2602		 */
2603		rx_queue = max_t(int, tp->rcv_nxt - tp->copied_seq, 0);
 
2604
2605	seq_printf(f, "%4d: %08X:%04X %08X:%04X %02X %08X:%08X %02X:%08lX "
2606			"%08X %5u %8d %lu %d %pK %lu %lu %u %u %d",
2607		i, src, srcp, dest, destp, sk->sk_state,
2608		tp->write_seq - tp->snd_una,
2609		rx_queue,
2610		timer_active,
2611		jiffies_delta_to_clock_t(timer_expires - jiffies),
2612		icsk->icsk_retransmits,
2613		from_kuid_munged(seq_user_ns(f), sock_i_uid(sk)),
2614		icsk->icsk_probes_out,
2615		sock_i_ino(sk),
2616		atomic_read(&sk->sk_refcnt), sk,
2617		jiffies_to_clock_t(icsk->icsk_rto),
2618		jiffies_to_clock_t(icsk->icsk_ack.ato),
2619		(icsk->icsk_ack.quick << 1) | icsk->icsk_ack.pingpong,
2620		tp->snd_cwnd,
2621		sk->sk_state == TCP_LISTEN ?
2622		    (fastopenq ? fastopenq->max_qlen : 0) :
2623		    (tcp_in_initial_slowstart(tp) ? -1 : tp->snd_ssthresh));
2624}
2625
2626static void get_timewait4_sock(const struct inet_timewait_sock *tw,
2627			       struct seq_file *f, int i)
2628{
 
2629	__be32 dest, src;
2630	__u16 destp, srcp;
2631	s32 delta = tw->tw_ttd - inet_tw_time_stamp();
2632
2633	dest  = tw->tw_daddr;
2634	src   = tw->tw_rcv_saddr;
2635	destp = ntohs(tw->tw_dport);
2636	srcp  = ntohs(tw->tw_sport);
2637
2638	seq_printf(f, "%4d: %08X:%04X %08X:%04X"
2639		" %02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %pK",
2640		i, src, srcp, dest, destp, tw->tw_substate, 0, 0,
2641		3, jiffies_delta_to_clock_t(delta), 0, 0, 0, 0,
2642		atomic_read(&tw->tw_refcnt), tw);
2643}
2644
2645#define TMPSZ 150
2646
2647static int tcp4_seq_show(struct seq_file *seq, void *v)
2648{
2649	struct tcp_iter_state *st;
2650	struct sock *sk = v;
2651
2652	seq_setwidth(seq, TMPSZ - 1);
2653	if (v == SEQ_START_TOKEN) {
2654		seq_puts(seq, "  sl  local_address rem_address   st tx_queue "
2655			   "rx_queue tr tm->when retrnsmt   uid  timeout "
2656			   "inode");
2657		goto out;
2658	}
2659	st = seq->private;
2660
2661	switch (st->state) {
2662	case TCP_SEQ_STATE_LISTENING:
2663	case TCP_SEQ_STATE_ESTABLISHED:
2664		if (sk->sk_state == TCP_TIME_WAIT)
2665			get_timewait4_sock(v, seq, st->num);
2666		else
2667			get_tcp4_sock(v, seq, st->num);
2668		break;
2669	case TCP_SEQ_STATE_OPENREQ:
2670		get_openreq4(st->syn_wait_sk, v, seq, st->num, st->uid);
2671		break;
2672	}
2673out:
2674	seq_pad(seq, '\n');
2675	return 0;
2676}
2677
2678static const struct file_operations tcp_afinfo_seq_fops = {
2679	.owner   = THIS_MODULE,
2680	.open    = tcp_seq_open,
2681	.read    = seq_read,
2682	.llseek  = seq_lseek,
2683	.release = seq_release_net
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2684};
2685
2686static struct tcp_seq_afinfo tcp4_seq_afinfo = {
2687	.name		= "tcp",
2688	.family		= AF_INET,
2689	.seq_fops	= &tcp_afinfo_seq_fops,
2690	.seq_ops	= {
2691		.show		= tcp4_seq_show,
2692	},
2693};
2694
2695static int __net_init tcp4_proc_init_net(struct net *net)
2696{
2697	return tcp_proc_register(net, &tcp4_seq_afinfo);
 
 
 
2698}
2699
2700static void __net_exit tcp4_proc_exit_net(struct net *net)
2701{
2702	tcp_proc_unregister(net, &tcp4_seq_afinfo);
2703}
2704
2705static struct pernet_operations tcp4_net_ops = {
2706	.init = tcp4_proc_init_net,
2707	.exit = tcp4_proc_exit_net,
2708};
2709
2710int __init tcp4_proc_init(void)
2711{
2712	return register_pernet_subsys(&tcp4_net_ops);
2713}
2714
2715void tcp4_proc_exit(void)
2716{
2717	unregister_pernet_subsys(&tcp4_net_ops);
2718}
2719#endif /* CONFIG_PROC_FS */
2720
2721struct proto tcp_prot = {
2722	.name			= "TCP",
2723	.owner			= THIS_MODULE,
2724	.close			= tcp_close,
 
2725	.connect		= tcp_v4_connect,
2726	.disconnect		= tcp_disconnect,
2727	.accept			= inet_csk_accept,
2728	.ioctl			= tcp_ioctl,
2729	.init			= tcp_v4_init_sock,
2730	.destroy		= tcp_v4_destroy_sock,
2731	.shutdown		= tcp_shutdown,
2732	.setsockopt		= tcp_setsockopt,
2733	.getsockopt		= tcp_getsockopt,
 
2734	.recvmsg		= tcp_recvmsg,
2735	.sendmsg		= tcp_sendmsg,
2736	.sendpage		= tcp_sendpage,
2737	.backlog_rcv		= tcp_v4_do_rcv,
2738	.release_cb		= tcp_release_cb,
2739	.mtu_reduced		= tcp_v4_mtu_reduced,
2740	.hash			= inet_hash,
2741	.unhash			= inet_unhash,
2742	.get_port		= inet_csk_get_port,
2743	.enter_memory_pressure	= tcp_enter_memory_pressure,
 
2744	.stream_memory_free	= tcp_stream_memory_free,
2745	.sockets_allocated	= &tcp_sockets_allocated,
2746	.orphan_count		= &tcp_orphan_count,
2747	.memory_allocated	= &tcp_memory_allocated,
2748	.memory_pressure	= &tcp_memory_pressure,
2749	.sysctl_mem		= sysctl_tcp_mem,
2750	.sysctl_wmem		= sysctl_tcp_wmem,
2751	.sysctl_rmem		= sysctl_tcp_rmem,
2752	.max_header		= MAX_TCP_HEADER,
2753	.obj_size		= sizeof(struct tcp_sock),
2754	.slab_flags		= SLAB_DESTROY_BY_RCU,
2755	.twsk_prot		= &tcp_timewait_sock_ops,
2756	.rsk_prot		= &tcp_request_sock_ops,
2757	.h.hashinfo		= &tcp_hashinfo,
2758	.no_autobind		= true,
2759#ifdef CONFIG_COMPAT
2760	.compat_setsockopt	= compat_tcp_setsockopt,
2761	.compat_getsockopt	= compat_tcp_getsockopt,
2762#endif
2763#ifdef CONFIG_MEMCG_KMEM
2764	.init_cgroup		= tcp_init_cgroup,
2765	.destroy_cgroup		= tcp_destroy_cgroup,
2766	.proto_cgroup		= tcp_proto_cgroup,
2767#endif
2768};
2769EXPORT_SYMBOL(tcp_prot);
2770
 
 
 
 
 
 
 
 
 
 
 
 
 
2771static int __net_init tcp_sk_init(struct net *net)
2772{
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2773	net->ipv4.sysctl_tcp_ecn = 2;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2774	return 0;
2775}
 
2776
2777static void __net_exit tcp_sk_exit(struct net *net)
2778{
2779}
2780
2781static void __net_exit tcp_sk_exit_batch(struct list_head *net_exit_list)
2782{
2783	inet_twsk_purge(&tcp_hashinfo, &tcp_death_row, AF_INET);
 
 
 
 
 
2784}
2785
2786static struct pernet_operations __net_initdata tcp_sk_ops = {
2787       .init	   = tcp_sk_init,
2788       .exit	   = tcp_sk_exit,
2789       .exit_batch = tcp_sk_exit_batch,
2790};
2791
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2792void __init tcp_v4_init(void)
2793{
2794	inet_hashinfo_init(&tcp_hashinfo);
2795	if (register_pernet_subsys(&tcp_sk_ops))
2796		panic("Failed to create the TCP control socket.\n");
 
 
 
 
2797}