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/secure_seq.h>
76#include <net/busy_poll.h>
77
78#include <linux/inet.h>
79#include <linux/ipv6.h>
80#include <linux/stddef.h>
81#include <linux/proc_fs.h>
82#include <linux/seq_file.h>
83
84#include <crypto/hash.h>
85#include <linux/scatterlist.h>
86
87int sysctl_tcp_low_latency __read_mostly;
88
89#ifdef CONFIG_TCP_MD5SIG
90static int tcp_v4_md5_hash_hdr(char *md5_hash, const struct tcp_md5sig_key *key,
91 __be32 daddr, __be32 saddr, const struct tcphdr *th);
92#endif
93
94struct inet_hashinfo tcp_hashinfo;
95EXPORT_SYMBOL(tcp_hashinfo);
96
97static u32 tcp_v4_init_sequence(const struct sk_buff *skb, u32 *tsoff)
98{
99 return secure_tcp_sequence_number(ip_hdr(skb)->daddr,
100 ip_hdr(skb)->saddr,
101 tcp_hdr(skb)->dest,
102 tcp_hdr(skb)->source, tsoff);
103}
104
105int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp)
106{
107 const struct tcp_timewait_sock *tcptw = tcp_twsk(sktw);
108 struct tcp_sock *tp = tcp_sk(sk);
109
110 /* With PAWS, it is safe from the viewpoint
111 of data integrity. Even without PAWS it is safe provided sequence
112 spaces do not overlap i.e. at data rates <= 80Mbit/sec.
113
114 Actually, the idea is close to VJ's one, only timestamp cache is
115 held not per host, but per port pair and TW bucket is used as state
116 holder.
117
118 If TW bucket has been already destroyed we fall back to VJ's scheme
119 and use initial timestamp retrieved from peer table.
120 */
121 if (tcptw->tw_ts_recent_stamp &&
122 (!twp || (sock_net(sk)->ipv4.sysctl_tcp_tw_reuse &&
123 get_seconds() - tcptw->tw_ts_recent_stamp > 1))) {
124 tp->write_seq = tcptw->tw_snd_nxt + 65535 + 2;
125 if (tp->write_seq == 0)
126 tp->write_seq = 1;
127 tp->rx_opt.ts_recent = tcptw->tw_ts_recent;
128 tp->rx_opt.ts_recent_stamp = tcptw->tw_ts_recent_stamp;
129 sock_hold(sktw);
130 return 1;
131 }
132
133 return 0;
134}
135EXPORT_SYMBOL_GPL(tcp_twsk_unique);
136
137/* This will initiate an outgoing connection. */
138int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len)
139{
140 struct sockaddr_in *usin = (struct sockaddr_in *)uaddr;
141 struct inet_sock *inet = inet_sk(sk);
142 struct tcp_sock *tp = tcp_sk(sk);
143 __be16 orig_sport, orig_dport;
144 __be32 daddr, nexthop;
145 struct flowi4 *fl4;
146 struct rtable *rt;
147 int err;
148 u32 seq;
149 struct ip_options_rcu *inet_opt;
150 struct inet_timewait_death_row *tcp_death_row = &sock_net(sk)->ipv4.tcp_death_row;
151
152 if (addr_len < sizeof(struct sockaddr_in))
153 return -EINVAL;
154
155 if (usin->sin_family != AF_INET)
156 return -EAFNOSUPPORT;
157
158 nexthop = daddr = usin->sin_addr.s_addr;
159 inet_opt = rcu_dereference_protected(inet->inet_opt,
160 lockdep_sock_is_held(sk));
161 if (inet_opt && inet_opt->opt.srr) {
162 if (!daddr)
163 return -EINVAL;
164 nexthop = inet_opt->opt.faddr;
165 }
166
167 orig_sport = inet->inet_sport;
168 orig_dport = usin->sin_port;
169 fl4 = &inet->cork.fl.u.ip4;
170 rt = ip_route_connect(fl4, nexthop, inet->inet_saddr,
171 RT_CONN_FLAGS(sk), sk->sk_bound_dev_if,
172 IPPROTO_TCP,
173 orig_sport, orig_dport, sk);
174 if (IS_ERR(rt)) {
175 err = PTR_ERR(rt);
176 if (err == -ENETUNREACH)
177 IP_INC_STATS(sock_net(sk), IPSTATS_MIB_OUTNOROUTES);
178 return err;
179 }
180
181 if (rt->rt_flags & (RTCF_MULTICAST | RTCF_BROADCAST)) {
182 ip_rt_put(rt);
183 return -ENETUNREACH;
184 }
185
186 if (!inet_opt || !inet_opt->opt.srr)
187 daddr = fl4->daddr;
188
189 if (!inet->inet_saddr)
190 inet->inet_saddr = fl4->saddr;
191 sk_rcv_saddr_set(sk, inet->inet_saddr);
192
193 if (tp->rx_opt.ts_recent_stamp && inet->inet_daddr != daddr) {
194 /* Reset inherited state */
195 tp->rx_opt.ts_recent = 0;
196 tp->rx_opt.ts_recent_stamp = 0;
197 if (likely(!tp->repair))
198 tp->write_seq = 0;
199 }
200
201 if (tcp_death_row->sysctl_tw_recycle &&
202 !tp->rx_opt.ts_recent_stamp && fl4->daddr == daddr)
203 tcp_fetch_timewait_stamp(sk, &rt->dst);
204
205 inet->inet_dport = usin->sin_port;
206 sk_daddr_set(sk, daddr);
207
208 inet_csk(sk)->icsk_ext_hdr_len = 0;
209 if (inet_opt)
210 inet_csk(sk)->icsk_ext_hdr_len = inet_opt->opt.optlen;
211
212 tp->rx_opt.mss_clamp = TCP_MSS_DEFAULT;
213
214 /* Socket identity is still unknown (sport may be zero).
215 * However we set state to SYN-SENT and not releasing socket
216 * lock select source port, enter ourselves into the hash tables and
217 * complete initialization after this.
218 */
219 tcp_set_state(sk, TCP_SYN_SENT);
220 err = inet_hash_connect(tcp_death_row, sk);
221 if (err)
222 goto failure;
223
224 sk_set_txhash(sk);
225
226 rt = ip_route_newports(fl4, rt, orig_sport, orig_dport,
227 inet->inet_sport, inet->inet_dport, sk);
228 if (IS_ERR(rt)) {
229 err = PTR_ERR(rt);
230 rt = NULL;
231 goto failure;
232 }
233 /* OK, now commit destination to socket. */
234 sk->sk_gso_type = SKB_GSO_TCPV4;
235 sk_setup_caps(sk, &rt->dst);
236 rt = NULL;
237
238 if (likely(!tp->repair)) {
239 seq = secure_tcp_sequence_number(inet->inet_saddr,
240 inet->inet_daddr,
241 inet->inet_sport,
242 usin->sin_port,
243 &tp->tsoffset);
244 if (!tp->write_seq)
245 tp->write_seq = seq;
246 }
247
248 inet->inet_id = tp->write_seq ^ jiffies;
249
250 if (tcp_fastopen_defer_connect(sk, &err))
251 return err;
252 if (err)
253 goto failure;
254
255 err = tcp_connect(sk);
256
257 if (err)
258 goto failure;
259
260 return 0;
261
262failure:
263 /*
264 * This unhashes the socket and releases the local port,
265 * if necessary.
266 */
267 tcp_set_state(sk, TCP_CLOSE);
268 ip_rt_put(rt);
269 sk->sk_route_caps = 0;
270 inet->inet_dport = 0;
271 return err;
272}
273EXPORT_SYMBOL(tcp_v4_connect);
274
275/*
276 * This routine reacts to ICMP_FRAG_NEEDED mtu indications as defined in RFC1191.
277 * It can be called through tcp_release_cb() if socket was owned by user
278 * at the time tcp_v4_err() was called to handle ICMP message.
279 */
280void tcp_v4_mtu_reduced(struct sock *sk)
281{
282 struct inet_sock *inet = inet_sk(sk);
283 struct dst_entry *dst;
284 u32 mtu;
285
286 if ((1 << sk->sk_state) & (TCPF_LISTEN | TCPF_CLOSE))
287 return;
288 mtu = tcp_sk(sk)->mtu_info;
289 dst = inet_csk_update_pmtu(sk, mtu);
290 if (!dst)
291 return;
292
293 /* Something is about to be wrong... Remember soft error
294 * for the case, if this connection will not able to recover.
295 */
296 if (mtu < dst_mtu(dst) && ip_dont_fragment(sk, dst))
297 sk->sk_err_soft = EMSGSIZE;
298
299 mtu = dst_mtu(dst);
300
301 if (inet->pmtudisc != IP_PMTUDISC_DONT &&
302 ip_sk_accept_pmtu(sk) &&
303 inet_csk(sk)->icsk_pmtu_cookie > mtu) {
304 tcp_sync_mss(sk, mtu);
305
306 /* Resend the TCP packet because it's
307 * clear that the old packet has been
308 * dropped. This is the new "fast" path mtu
309 * discovery.
310 */
311 tcp_simple_retransmit(sk);
312 } /* else let the usual retransmit timer handle it */
313}
314EXPORT_SYMBOL(tcp_v4_mtu_reduced);
315
316static void do_redirect(struct sk_buff *skb, struct sock *sk)
317{
318 struct dst_entry *dst = __sk_dst_check(sk, 0);
319
320 if (dst)
321 dst->ops->redirect(dst, sk, skb);
322}
323
324
325/* handle ICMP messages on TCP_NEW_SYN_RECV request sockets */
326void tcp_req_err(struct sock *sk, u32 seq, bool abort)
327{
328 struct request_sock *req = inet_reqsk(sk);
329 struct net *net = sock_net(sk);
330
331 /* ICMPs are not backlogged, hence we cannot get
332 * an established socket here.
333 */
334 if (seq != tcp_rsk(req)->snt_isn) {
335 __NET_INC_STATS(net, LINUX_MIB_OUTOFWINDOWICMPS);
336 } else if (abort) {
337 /*
338 * Still in SYN_RECV, just remove it silently.
339 * There is no good way to pass the error to the newly
340 * created socket, and POSIX does not want network
341 * errors returned from accept().
342 */
343 inet_csk_reqsk_queue_drop(req->rsk_listener, req);
344 tcp_listendrop(req->rsk_listener);
345 }
346 reqsk_put(req);
347}
348EXPORT_SYMBOL(tcp_req_err);
349
350/*
351 * This routine is called by the ICMP module when it gets some
352 * sort of error condition. If err < 0 then the socket should
353 * be closed and the error returned to the user. If err > 0
354 * it's just the icmp type << 8 | icmp code. After adjustment
355 * header points to the first 8 bytes of the tcp header. We need
356 * to find the appropriate port.
357 *
358 * The locking strategy used here is very "optimistic". When
359 * someone else accesses the socket the ICMP is just dropped
360 * and for some paths there is no check at all.
361 * A more general error queue to queue errors for later handling
362 * is probably better.
363 *
364 */
365
366void tcp_v4_err(struct sk_buff *icmp_skb, u32 info)
367{
368 const struct iphdr *iph = (const struct iphdr *)icmp_skb->data;
369 struct tcphdr *th = (struct tcphdr *)(icmp_skb->data + (iph->ihl << 2));
370 struct inet_connection_sock *icsk;
371 struct tcp_sock *tp;
372 struct inet_sock *inet;
373 const int type = icmp_hdr(icmp_skb)->type;
374 const int code = icmp_hdr(icmp_skb)->code;
375 struct sock *sk;
376 struct sk_buff *skb;
377 struct request_sock *fastopen;
378 __u32 seq, snd_una;
379 __u32 remaining;
380 int err;
381 struct net *net = dev_net(icmp_skb->dev);
382
383 sk = __inet_lookup_established(net, &tcp_hashinfo, iph->daddr,
384 th->dest, iph->saddr, ntohs(th->source),
385 inet_iif(icmp_skb));
386 if (!sk) {
387 __ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
388 return;
389 }
390 if (sk->sk_state == TCP_TIME_WAIT) {
391 inet_twsk_put(inet_twsk(sk));
392 return;
393 }
394 seq = ntohl(th->seq);
395 if (sk->sk_state == TCP_NEW_SYN_RECV)
396 return tcp_req_err(sk, seq,
397 type == ICMP_PARAMETERPROB ||
398 type == ICMP_TIME_EXCEEDED ||
399 (type == ICMP_DEST_UNREACH &&
400 (code == ICMP_NET_UNREACH ||
401 code == ICMP_HOST_UNREACH)));
402
403 bh_lock_sock(sk);
404 /* If too many ICMPs get dropped on busy
405 * servers this needs to be solved differently.
406 * We do take care of PMTU discovery (RFC1191) special case :
407 * we can receive locally generated ICMP messages while socket is held.
408 */
409 if (sock_owned_by_user(sk)) {
410 if (!(type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED))
411 __NET_INC_STATS(net, LINUX_MIB_LOCKDROPPEDICMPS);
412 }
413 if (sk->sk_state == TCP_CLOSE)
414 goto out;
415
416 if (unlikely(iph->ttl < inet_sk(sk)->min_ttl)) {
417 __NET_INC_STATS(net, LINUX_MIB_TCPMINTTLDROP);
418 goto out;
419 }
420
421 icsk = inet_csk(sk);
422 tp = tcp_sk(sk);
423 /* XXX (TFO) - tp->snd_una should be ISN (tcp_create_openreq_child() */
424 fastopen = tp->fastopen_rsk;
425 snd_una = fastopen ? tcp_rsk(fastopen)->snt_isn : tp->snd_una;
426 if (sk->sk_state != TCP_LISTEN &&
427 !between(seq, snd_una, tp->snd_nxt)) {
428 __NET_INC_STATS(net, LINUX_MIB_OUTOFWINDOWICMPS);
429 goto out;
430 }
431
432 switch (type) {
433 case ICMP_REDIRECT:
434 if (!sock_owned_by_user(sk))
435 do_redirect(icmp_skb, sk);
436 goto out;
437 case ICMP_SOURCE_QUENCH:
438 /* Just silently ignore these. */
439 goto out;
440 case ICMP_PARAMETERPROB:
441 err = EPROTO;
442 break;
443 case ICMP_DEST_UNREACH:
444 if (code > NR_ICMP_UNREACH)
445 goto out;
446
447 if (code == ICMP_FRAG_NEEDED) { /* PMTU discovery (RFC1191) */
448 /* We are not interested in TCP_LISTEN and open_requests
449 * (SYN-ACKs send out by Linux are always <576bytes so
450 * they should go through unfragmented).
451 */
452 if (sk->sk_state == TCP_LISTEN)
453 goto out;
454
455 tp->mtu_info = info;
456 if (!sock_owned_by_user(sk)) {
457 tcp_v4_mtu_reduced(sk);
458 } else {
459 if (!test_and_set_bit(TCP_MTU_REDUCED_DEFERRED, &sk->sk_tsq_flags))
460 sock_hold(sk);
461 }
462 goto out;
463 }
464
465 err = icmp_err_convert[code].errno;
466 /* check if icmp_skb allows revert of backoff
467 * (see draft-zimmermann-tcp-lcd) */
468 if (code != ICMP_NET_UNREACH && code != ICMP_HOST_UNREACH)
469 break;
470 if (seq != tp->snd_una || !icsk->icsk_retransmits ||
471 !icsk->icsk_backoff || fastopen)
472 break;
473
474 if (sock_owned_by_user(sk))
475 break;
476
477 icsk->icsk_backoff--;
478 icsk->icsk_rto = tp->srtt_us ? __tcp_set_rto(tp) :
479 TCP_TIMEOUT_INIT;
480 icsk->icsk_rto = inet_csk_rto_backoff(icsk, TCP_RTO_MAX);
481
482 skb = tcp_write_queue_head(sk);
483 BUG_ON(!skb);
484
485 remaining = icsk->icsk_rto -
486 min(icsk->icsk_rto,
487 tcp_time_stamp - tcp_skb_timestamp(skb));
488
489 if (remaining) {
490 inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
491 remaining, TCP_RTO_MAX);
492 } else {
493 /* RTO revert clocked out retransmission.
494 * Will retransmit now */
495 tcp_retransmit_timer(sk);
496 }
497
498 break;
499 case ICMP_TIME_EXCEEDED:
500 err = EHOSTUNREACH;
501 break;
502 default:
503 goto out;
504 }
505
506 switch (sk->sk_state) {
507 case TCP_SYN_SENT:
508 case TCP_SYN_RECV:
509 /* Only in fast or simultaneous open. If a fast open socket is
510 * is already accepted it is treated as a connected one below.
511 */
512 if (fastopen && !fastopen->sk)
513 break;
514
515 if (!sock_owned_by_user(sk)) {
516 sk->sk_err = err;
517
518 sk->sk_error_report(sk);
519
520 tcp_done(sk);
521 } else {
522 sk->sk_err_soft = err;
523 }
524 goto out;
525 }
526
527 /* If we've already connected we will keep trying
528 * until we time out, or the user gives up.
529 *
530 * rfc1122 4.2.3.9 allows to consider as hard errors
531 * only PROTO_UNREACH and PORT_UNREACH (well, FRAG_FAILED too,
532 * but it is obsoleted by pmtu discovery).
533 *
534 * Note, that in modern internet, where routing is unreliable
535 * and in each dark corner broken firewalls sit, sending random
536 * errors ordered by their masters even this two messages finally lose
537 * their original sense (even Linux sends invalid PORT_UNREACHs)
538 *
539 * Now we are in compliance with RFCs.
540 * --ANK (980905)
541 */
542
543 inet = inet_sk(sk);
544 if (!sock_owned_by_user(sk) && inet->recverr) {
545 sk->sk_err = err;
546 sk->sk_error_report(sk);
547 } else { /* Only an error on timeout */
548 sk->sk_err_soft = err;
549 }
550
551out:
552 bh_unlock_sock(sk);
553 sock_put(sk);
554}
555
556void __tcp_v4_send_check(struct sk_buff *skb, __be32 saddr, __be32 daddr)
557{
558 struct tcphdr *th = tcp_hdr(skb);
559
560 if (skb->ip_summed == CHECKSUM_PARTIAL) {
561 th->check = ~tcp_v4_check(skb->len, saddr, daddr, 0);
562 skb->csum_start = skb_transport_header(skb) - skb->head;
563 skb->csum_offset = offsetof(struct tcphdr, check);
564 } else {
565 th->check = tcp_v4_check(skb->len, saddr, daddr,
566 csum_partial(th,
567 th->doff << 2,
568 skb->csum));
569 }
570}
571
572/* This routine computes an IPv4 TCP checksum. */
573void tcp_v4_send_check(struct sock *sk, struct sk_buff *skb)
574{
575 const struct inet_sock *inet = inet_sk(sk);
576
577 __tcp_v4_send_check(skb, inet->inet_saddr, inet->inet_daddr);
578}
579EXPORT_SYMBOL(tcp_v4_send_check);
580
581/*
582 * This routine will send an RST to the other tcp.
583 *
584 * Someone asks: why I NEVER use socket parameters (TOS, TTL etc.)
585 * for reset.
586 * Answer: if a packet caused RST, it is not for a socket
587 * existing in our system, if it is matched to a socket,
588 * it is just duplicate segment or bug in other side's TCP.
589 * So that we build reply only basing on parameters
590 * arrived with segment.
591 * Exception: precedence violation. We do not implement it in any case.
592 */
593
594static void tcp_v4_send_reset(const struct sock *sk, struct sk_buff *skb)
595{
596 const struct tcphdr *th = tcp_hdr(skb);
597 struct {
598 struct tcphdr th;
599#ifdef CONFIG_TCP_MD5SIG
600 __be32 opt[(TCPOLEN_MD5SIG_ALIGNED >> 2)];
601#endif
602 } rep;
603 struct ip_reply_arg arg;
604#ifdef CONFIG_TCP_MD5SIG
605 struct tcp_md5sig_key *key = NULL;
606 const __u8 *hash_location = NULL;
607 unsigned char newhash[16];
608 int genhash;
609 struct sock *sk1 = NULL;
610#endif
611 struct net *net;
612
613 /* Never send a reset in response to a reset. */
614 if (th->rst)
615 return;
616
617 /* If sk not NULL, it means we did a successful lookup and incoming
618 * route had to be correct. prequeue might have dropped our dst.
619 */
620 if (!sk && skb_rtable(skb)->rt_type != RTN_LOCAL)
621 return;
622
623 /* Swap the send and the receive. */
624 memset(&rep, 0, sizeof(rep));
625 rep.th.dest = th->source;
626 rep.th.source = th->dest;
627 rep.th.doff = sizeof(struct tcphdr) / 4;
628 rep.th.rst = 1;
629
630 if (th->ack) {
631 rep.th.seq = th->ack_seq;
632 } else {
633 rep.th.ack = 1;
634 rep.th.ack_seq = htonl(ntohl(th->seq) + th->syn + th->fin +
635 skb->len - (th->doff << 2));
636 }
637
638 memset(&arg, 0, sizeof(arg));
639 arg.iov[0].iov_base = (unsigned char *)&rep;
640 arg.iov[0].iov_len = sizeof(rep.th);
641
642 net = sk ? sock_net(sk) : dev_net(skb_dst(skb)->dev);
643#ifdef CONFIG_TCP_MD5SIG
644 rcu_read_lock();
645 hash_location = tcp_parse_md5sig_option(th);
646 if (sk && sk_fullsock(sk)) {
647 key = tcp_md5_do_lookup(sk, (union tcp_md5_addr *)
648 &ip_hdr(skb)->saddr, AF_INET);
649 } else if (hash_location) {
650 /*
651 * active side is lost. Try to find listening socket through
652 * source port, and then find md5 key through listening socket.
653 * we are not loose security here:
654 * Incoming packet is checked with md5 hash with finding key,
655 * no RST generated if md5 hash doesn't match.
656 */
657 sk1 = __inet_lookup_listener(net, &tcp_hashinfo, NULL, 0,
658 ip_hdr(skb)->saddr,
659 th->source, ip_hdr(skb)->daddr,
660 ntohs(th->source), inet_iif(skb));
661 /* don't send rst if it can't find key */
662 if (!sk1)
663 goto out;
664
665 key = tcp_md5_do_lookup(sk1, (union tcp_md5_addr *)
666 &ip_hdr(skb)->saddr, AF_INET);
667 if (!key)
668 goto out;
669
670
671 genhash = tcp_v4_md5_hash_skb(newhash, key, NULL, skb);
672 if (genhash || memcmp(hash_location, newhash, 16) != 0)
673 goto out;
674
675 }
676
677 if (key) {
678 rep.opt[0] = htonl((TCPOPT_NOP << 24) |
679 (TCPOPT_NOP << 16) |
680 (TCPOPT_MD5SIG << 8) |
681 TCPOLEN_MD5SIG);
682 /* Update length and the length the header thinks exists */
683 arg.iov[0].iov_len += TCPOLEN_MD5SIG_ALIGNED;
684 rep.th.doff = arg.iov[0].iov_len / 4;
685
686 tcp_v4_md5_hash_hdr((__u8 *) &rep.opt[1],
687 key, ip_hdr(skb)->saddr,
688 ip_hdr(skb)->daddr, &rep.th);
689 }
690#endif
691 arg.csum = csum_tcpudp_nofold(ip_hdr(skb)->daddr,
692 ip_hdr(skb)->saddr, /* XXX */
693 arg.iov[0].iov_len, IPPROTO_TCP, 0);
694 arg.csumoffset = offsetof(struct tcphdr, check) / 2;
695 arg.flags = (sk && inet_sk_transparent(sk)) ? IP_REPLY_ARG_NOSRCCHECK : 0;
696
697 /* When socket is gone, all binding information is lost.
698 * routing might fail in this case. No choice here, if we choose to force
699 * input interface, we will misroute in case of asymmetric route.
700 */
701 if (sk)
702 arg.bound_dev_if = sk->sk_bound_dev_if;
703
704 BUILD_BUG_ON(offsetof(struct sock, sk_bound_dev_if) !=
705 offsetof(struct inet_timewait_sock, tw_bound_dev_if));
706
707 arg.tos = ip_hdr(skb)->tos;
708 arg.uid = sock_net_uid(net, sk && sk_fullsock(sk) ? sk : NULL);
709 local_bh_disable();
710 ip_send_unicast_reply(*this_cpu_ptr(net->ipv4.tcp_sk),
711 skb, &TCP_SKB_CB(skb)->header.h4.opt,
712 ip_hdr(skb)->saddr, ip_hdr(skb)->daddr,
713 &arg, arg.iov[0].iov_len);
714
715 __TCP_INC_STATS(net, TCP_MIB_OUTSEGS);
716 __TCP_INC_STATS(net, TCP_MIB_OUTRSTS);
717 local_bh_enable();
718
719#ifdef CONFIG_TCP_MD5SIG
720out:
721 rcu_read_unlock();
722#endif
723}
724
725/* The code following below sending ACKs in SYN-RECV and TIME-WAIT states
726 outside socket context is ugly, certainly. What can I do?
727 */
728
729static void tcp_v4_send_ack(const struct sock *sk,
730 struct sk_buff *skb, u32 seq, u32 ack,
731 u32 win, u32 tsval, u32 tsecr, int oif,
732 struct tcp_md5sig_key *key,
733 int reply_flags, u8 tos)
734{
735 const struct tcphdr *th = tcp_hdr(skb);
736 struct {
737 struct tcphdr th;
738 __be32 opt[(TCPOLEN_TSTAMP_ALIGNED >> 2)
739#ifdef CONFIG_TCP_MD5SIG
740 + (TCPOLEN_MD5SIG_ALIGNED >> 2)
741#endif
742 ];
743 } rep;
744 struct net *net = sock_net(sk);
745 struct ip_reply_arg arg;
746
747 memset(&rep.th, 0, sizeof(struct tcphdr));
748 memset(&arg, 0, sizeof(arg));
749
750 arg.iov[0].iov_base = (unsigned char *)&rep;
751 arg.iov[0].iov_len = sizeof(rep.th);
752 if (tsecr) {
753 rep.opt[0] = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
754 (TCPOPT_TIMESTAMP << 8) |
755 TCPOLEN_TIMESTAMP);
756 rep.opt[1] = htonl(tsval);
757 rep.opt[2] = htonl(tsecr);
758 arg.iov[0].iov_len += TCPOLEN_TSTAMP_ALIGNED;
759 }
760
761 /* Swap the send and the receive. */
762 rep.th.dest = th->source;
763 rep.th.source = th->dest;
764 rep.th.doff = arg.iov[0].iov_len / 4;
765 rep.th.seq = htonl(seq);
766 rep.th.ack_seq = htonl(ack);
767 rep.th.ack = 1;
768 rep.th.window = htons(win);
769
770#ifdef CONFIG_TCP_MD5SIG
771 if (key) {
772 int offset = (tsecr) ? 3 : 0;
773
774 rep.opt[offset++] = htonl((TCPOPT_NOP << 24) |
775 (TCPOPT_NOP << 16) |
776 (TCPOPT_MD5SIG << 8) |
777 TCPOLEN_MD5SIG);
778 arg.iov[0].iov_len += TCPOLEN_MD5SIG_ALIGNED;
779 rep.th.doff = arg.iov[0].iov_len/4;
780
781 tcp_v4_md5_hash_hdr((__u8 *) &rep.opt[offset],
782 key, ip_hdr(skb)->saddr,
783 ip_hdr(skb)->daddr, &rep.th);
784 }
785#endif
786 arg.flags = reply_flags;
787 arg.csum = csum_tcpudp_nofold(ip_hdr(skb)->daddr,
788 ip_hdr(skb)->saddr, /* XXX */
789 arg.iov[0].iov_len, IPPROTO_TCP, 0);
790 arg.csumoffset = offsetof(struct tcphdr, check) / 2;
791 if (oif)
792 arg.bound_dev_if = oif;
793 arg.tos = tos;
794 arg.uid = sock_net_uid(net, sk_fullsock(sk) ? sk : NULL);
795 local_bh_disable();
796 ip_send_unicast_reply(*this_cpu_ptr(net->ipv4.tcp_sk),
797 skb, &TCP_SKB_CB(skb)->header.h4.opt,
798 ip_hdr(skb)->saddr, ip_hdr(skb)->daddr,
799 &arg, arg.iov[0].iov_len);
800
801 __TCP_INC_STATS(net, TCP_MIB_OUTSEGS);
802 local_bh_enable();
803}
804
805static void tcp_v4_timewait_ack(struct sock *sk, struct sk_buff *skb)
806{
807 struct inet_timewait_sock *tw = inet_twsk(sk);
808 struct tcp_timewait_sock *tcptw = tcp_twsk(sk);
809
810 tcp_v4_send_ack(sk, skb,
811 tcptw->tw_snd_nxt, tcptw->tw_rcv_nxt,
812 tcptw->tw_rcv_wnd >> tw->tw_rcv_wscale,
813 tcp_time_stamp + tcptw->tw_ts_offset,
814 tcptw->tw_ts_recent,
815 tw->tw_bound_dev_if,
816 tcp_twsk_md5_key(tcptw),
817 tw->tw_transparent ? IP_REPLY_ARG_NOSRCCHECK : 0,
818 tw->tw_tos
819 );
820
821 inet_twsk_put(tw);
822}
823
824static void tcp_v4_reqsk_send_ack(const struct sock *sk, struct sk_buff *skb,
825 struct request_sock *req)
826{
827 /* sk->sk_state == TCP_LISTEN -> for regular TCP_SYN_RECV
828 * sk->sk_state == TCP_SYN_RECV -> for Fast Open.
829 */
830 u32 seq = (sk->sk_state == TCP_LISTEN) ? tcp_rsk(req)->snt_isn + 1 :
831 tcp_sk(sk)->snd_nxt;
832
833 /* RFC 7323 2.3
834 * The window field (SEG.WND) of every outgoing segment, with the
835 * exception of <SYN> segments, MUST be right-shifted by
836 * Rcv.Wind.Shift bits:
837 */
838 tcp_v4_send_ack(sk, skb, seq,
839 tcp_rsk(req)->rcv_nxt,
840 req->rsk_rcv_wnd >> inet_rsk(req)->rcv_wscale,
841 tcp_time_stamp + tcp_rsk(req)->ts_off,
842 req->ts_recent,
843 0,
844 tcp_md5_do_lookup(sk, (union tcp_md5_addr *)&ip_hdr(skb)->daddr,
845 AF_INET),
846 inet_rsk(req)->no_srccheck ? IP_REPLY_ARG_NOSRCCHECK : 0,
847 ip_hdr(skb)->tos);
848}
849
850/*
851 * Send a SYN-ACK after having received a SYN.
852 * This still operates on a request_sock only, not on a big
853 * socket.
854 */
855static int tcp_v4_send_synack(const struct sock *sk, struct dst_entry *dst,
856 struct flowi *fl,
857 struct request_sock *req,
858 struct tcp_fastopen_cookie *foc,
859 enum tcp_synack_type synack_type)
860{
861 const struct inet_request_sock *ireq = inet_rsk(req);
862 struct flowi4 fl4;
863 int err = -1;
864 struct sk_buff *skb;
865
866 /* First, grab a route. */
867 if (!dst && (dst = inet_csk_route_req(sk, &fl4, req)) == NULL)
868 return -1;
869
870 skb = tcp_make_synack(sk, dst, req, foc, synack_type);
871
872 if (skb) {
873 __tcp_v4_send_check(skb, ireq->ir_loc_addr, ireq->ir_rmt_addr);
874
875 err = ip_build_and_send_pkt(skb, sk, ireq->ir_loc_addr,
876 ireq->ir_rmt_addr,
877 ireq->opt);
878 err = net_xmit_eval(err);
879 }
880
881 return err;
882}
883
884/*
885 * IPv4 request_sock destructor.
886 */
887static void tcp_v4_reqsk_destructor(struct request_sock *req)
888{
889 kfree(inet_rsk(req)->opt);
890}
891
892#ifdef CONFIG_TCP_MD5SIG
893/*
894 * RFC2385 MD5 checksumming requires a mapping of
895 * IP address->MD5 Key.
896 * We need to maintain these in the sk structure.
897 */
898
899/* Find the Key structure for an address. */
900struct tcp_md5sig_key *tcp_md5_do_lookup(const struct sock *sk,
901 const union tcp_md5_addr *addr,
902 int family)
903{
904 const struct tcp_sock *tp = tcp_sk(sk);
905 struct tcp_md5sig_key *key;
906 unsigned int size = sizeof(struct in_addr);
907 const struct tcp_md5sig_info *md5sig;
908
909 /* caller either holds rcu_read_lock() or socket lock */
910 md5sig = rcu_dereference_check(tp->md5sig_info,
911 lockdep_sock_is_held(sk));
912 if (!md5sig)
913 return NULL;
914#if IS_ENABLED(CONFIG_IPV6)
915 if (family == AF_INET6)
916 size = sizeof(struct in6_addr);
917#endif
918 hlist_for_each_entry_rcu(key, &md5sig->head, node) {
919 if (key->family != family)
920 continue;
921 if (!memcmp(&key->addr, addr, size))
922 return key;
923 }
924 return NULL;
925}
926EXPORT_SYMBOL(tcp_md5_do_lookup);
927
928struct tcp_md5sig_key *tcp_v4_md5_lookup(const struct sock *sk,
929 const struct sock *addr_sk)
930{
931 const union tcp_md5_addr *addr;
932
933 addr = (const union tcp_md5_addr *)&addr_sk->sk_daddr;
934 return tcp_md5_do_lookup(sk, addr, AF_INET);
935}
936EXPORT_SYMBOL(tcp_v4_md5_lookup);
937
938/* This can be called on a newly created socket, from other files */
939int tcp_md5_do_add(struct sock *sk, const union tcp_md5_addr *addr,
940 int family, const u8 *newkey, u8 newkeylen, gfp_t gfp)
941{
942 /* Add Key to the list */
943 struct tcp_md5sig_key *key;
944 struct tcp_sock *tp = tcp_sk(sk);
945 struct tcp_md5sig_info *md5sig;
946
947 key = tcp_md5_do_lookup(sk, addr, family);
948 if (key) {
949 /* Pre-existing entry - just update that one. */
950 memcpy(key->key, newkey, newkeylen);
951 key->keylen = newkeylen;
952 return 0;
953 }
954
955 md5sig = rcu_dereference_protected(tp->md5sig_info,
956 lockdep_sock_is_held(sk));
957 if (!md5sig) {
958 md5sig = kmalloc(sizeof(*md5sig), gfp);
959 if (!md5sig)
960 return -ENOMEM;
961
962 sk_nocaps_add(sk, NETIF_F_GSO_MASK);
963 INIT_HLIST_HEAD(&md5sig->head);
964 rcu_assign_pointer(tp->md5sig_info, md5sig);
965 }
966
967 key = sock_kmalloc(sk, sizeof(*key), gfp);
968 if (!key)
969 return -ENOMEM;
970 if (!tcp_alloc_md5sig_pool()) {
971 sock_kfree_s(sk, key, sizeof(*key));
972 return -ENOMEM;
973 }
974
975 memcpy(key->key, newkey, newkeylen);
976 key->keylen = newkeylen;
977 key->family = family;
978 memcpy(&key->addr, addr,
979 (family == AF_INET6) ? sizeof(struct in6_addr) :
980 sizeof(struct in_addr));
981 hlist_add_head_rcu(&key->node, &md5sig->head);
982 return 0;
983}
984EXPORT_SYMBOL(tcp_md5_do_add);
985
986int tcp_md5_do_del(struct sock *sk, const union tcp_md5_addr *addr, int family)
987{
988 struct tcp_md5sig_key *key;
989
990 key = tcp_md5_do_lookup(sk, addr, family);
991 if (!key)
992 return -ENOENT;
993 hlist_del_rcu(&key->node);
994 atomic_sub(sizeof(*key), &sk->sk_omem_alloc);
995 kfree_rcu(key, rcu);
996 return 0;
997}
998EXPORT_SYMBOL(tcp_md5_do_del);
999
1000static void tcp_clear_md5_list(struct sock *sk)
1001{
1002 struct tcp_sock *tp = tcp_sk(sk);
1003 struct tcp_md5sig_key *key;
1004 struct hlist_node *n;
1005 struct tcp_md5sig_info *md5sig;
1006
1007 md5sig = rcu_dereference_protected(tp->md5sig_info, 1);
1008
1009 hlist_for_each_entry_safe(key, n, &md5sig->head, node) {
1010 hlist_del_rcu(&key->node);
1011 atomic_sub(sizeof(*key), &sk->sk_omem_alloc);
1012 kfree_rcu(key, rcu);
1013 }
1014}
1015
1016static int tcp_v4_parse_md5_keys(struct sock *sk, char __user *optval,
1017 int optlen)
1018{
1019 struct tcp_md5sig cmd;
1020 struct sockaddr_in *sin = (struct sockaddr_in *)&cmd.tcpm_addr;
1021
1022 if (optlen < sizeof(cmd))
1023 return -EINVAL;
1024
1025 if (copy_from_user(&cmd, optval, sizeof(cmd)))
1026 return -EFAULT;
1027
1028 if (sin->sin_family != AF_INET)
1029 return -EINVAL;
1030
1031 if (!cmd.tcpm_keylen)
1032 return tcp_md5_do_del(sk, (union tcp_md5_addr *)&sin->sin_addr.s_addr,
1033 AF_INET);
1034
1035 if (cmd.tcpm_keylen > TCP_MD5SIG_MAXKEYLEN)
1036 return -EINVAL;
1037
1038 return tcp_md5_do_add(sk, (union tcp_md5_addr *)&sin->sin_addr.s_addr,
1039 AF_INET, cmd.tcpm_key, cmd.tcpm_keylen,
1040 GFP_KERNEL);
1041}
1042
1043static int tcp_v4_md5_hash_headers(struct tcp_md5sig_pool *hp,
1044 __be32 daddr, __be32 saddr,
1045 const struct tcphdr *th, int nbytes)
1046{
1047 struct tcp4_pseudohdr *bp;
1048 struct scatterlist sg;
1049 struct tcphdr *_th;
1050
1051 bp = hp->scratch;
1052 bp->saddr = saddr;
1053 bp->daddr = daddr;
1054 bp->pad = 0;
1055 bp->protocol = IPPROTO_TCP;
1056 bp->len = cpu_to_be16(nbytes);
1057
1058 _th = (struct tcphdr *)(bp + 1);
1059 memcpy(_th, th, sizeof(*th));
1060 _th->check = 0;
1061
1062 sg_init_one(&sg, bp, sizeof(*bp) + sizeof(*th));
1063 ahash_request_set_crypt(hp->md5_req, &sg, NULL,
1064 sizeof(*bp) + sizeof(*th));
1065 return crypto_ahash_update(hp->md5_req);
1066}
1067
1068static int tcp_v4_md5_hash_hdr(char *md5_hash, const struct tcp_md5sig_key *key,
1069 __be32 daddr, __be32 saddr, const struct tcphdr *th)
1070{
1071 struct tcp_md5sig_pool *hp;
1072 struct ahash_request *req;
1073
1074 hp = tcp_get_md5sig_pool();
1075 if (!hp)
1076 goto clear_hash_noput;
1077 req = hp->md5_req;
1078
1079 if (crypto_ahash_init(req))
1080 goto clear_hash;
1081 if (tcp_v4_md5_hash_headers(hp, daddr, saddr, th, th->doff << 2))
1082 goto clear_hash;
1083 if (tcp_md5_hash_key(hp, key))
1084 goto clear_hash;
1085 ahash_request_set_crypt(req, NULL, md5_hash, 0);
1086 if (crypto_ahash_final(req))
1087 goto clear_hash;
1088
1089 tcp_put_md5sig_pool();
1090 return 0;
1091
1092clear_hash:
1093 tcp_put_md5sig_pool();
1094clear_hash_noput:
1095 memset(md5_hash, 0, 16);
1096 return 1;
1097}
1098
1099int tcp_v4_md5_hash_skb(char *md5_hash, const struct tcp_md5sig_key *key,
1100 const struct sock *sk,
1101 const struct sk_buff *skb)
1102{
1103 struct tcp_md5sig_pool *hp;
1104 struct ahash_request *req;
1105 const struct tcphdr *th = tcp_hdr(skb);
1106 __be32 saddr, daddr;
1107
1108 if (sk) { /* valid for establish/request sockets */
1109 saddr = sk->sk_rcv_saddr;
1110 daddr = sk->sk_daddr;
1111 } else {
1112 const struct iphdr *iph = ip_hdr(skb);
1113 saddr = iph->saddr;
1114 daddr = iph->daddr;
1115 }
1116
1117 hp = tcp_get_md5sig_pool();
1118 if (!hp)
1119 goto clear_hash_noput;
1120 req = hp->md5_req;
1121
1122 if (crypto_ahash_init(req))
1123 goto clear_hash;
1124
1125 if (tcp_v4_md5_hash_headers(hp, daddr, saddr, th, skb->len))
1126 goto clear_hash;
1127 if (tcp_md5_hash_skb_data(hp, skb, th->doff << 2))
1128 goto clear_hash;
1129 if (tcp_md5_hash_key(hp, key))
1130 goto clear_hash;
1131 ahash_request_set_crypt(req, NULL, md5_hash, 0);
1132 if (crypto_ahash_final(req))
1133 goto clear_hash;
1134
1135 tcp_put_md5sig_pool();
1136 return 0;
1137
1138clear_hash:
1139 tcp_put_md5sig_pool();
1140clear_hash_noput:
1141 memset(md5_hash, 0, 16);
1142 return 1;
1143}
1144EXPORT_SYMBOL(tcp_v4_md5_hash_skb);
1145
1146#endif
1147
1148/* Called with rcu_read_lock() */
1149static bool tcp_v4_inbound_md5_hash(const struct sock *sk,
1150 const struct sk_buff *skb)
1151{
1152#ifdef CONFIG_TCP_MD5SIG
1153 /*
1154 * This gets called for each TCP segment that arrives
1155 * so we want to be efficient.
1156 * We have 3 drop cases:
1157 * o No MD5 hash and one expected.
1158 * o MD5 hash and we're not expecting one.
1159 * o MD5 hash and its wrong.
1160 */
1161 const __u8 *hash_location = NULL;
1162 struct tcp_md5sig_key *hash_expected;
1163 const struct iphdr *iph = ip_hdr(skb);
1164 const struct tcphdr *th = tcp_hdr(skb);
1165 int genhash;
1166 unsigned char newhash[16];
1167
1168 hash_expected = tcp_md5_do_lookup(sk, (union tcp_md5_addr *)&iph->saddr,
1169 AF_INET);
1170 hash_location = tcp_parse_md5sig_option(th);
1171
1172 /* We've parsed the options - do we have a hash? */
1173 if (!hash_expected && !hash_location)
1174 return false;
1175
1176 if (hash_expected && !hash_location) {
1177 NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPMD5NOTFOUND);
1178 return true;
1179 }
1180
1181 if (!hash_expected && hash_location) {
1182 NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPMD5UNEXPECTED);
1183 return true;
1184 }
1185
1186 /* Okay, so this is hash_expected and hash_location -
1187 * so we need to calculate the checksum.
1188 */
1189 genhash = tcp_v4_md5_hash_skb(newhash,
1190 hash_expected,
1191 NULL, skb);
1192
1193 if (genhash || memcmp(hash_location, newhash, 16) != 0) {
1194 NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPMD5FAILURE);
1195 net_info_ratelimited("MD5 Hash failed for (%pI4, %d)->(%pI4, %d)%s\n",
1196 &iph->saddr, ntohs(th->source),
1197 &iph->daddr, ntohs(th->dest),
1198 genhash ? " tcp_v4_calc_md5_hash failed"
1199 : "");
1200 return true;
1201 }
1202 return false;
1203#endif
1204 return false;
1205}
1206
1207static void tcp_v4_init_req(struct request_sock *req,
1208 const struct sock *sk_listener,
1209 struct sk_buff *skb)
1210{
1211 struct inet_request_sock *ireq = inet_rsk(req);
1212
1213 sk_rcv_saddr_set(req_to_sk(req), ip_hdr(skb)->daddr);
1214 sk_daddr_set(req_to_sk(req), ip_hdr(skb)->saddr);
1215 ireq->opt = tcp_v4_save_options(skb);
1216}
1217
1218static struct dst_entry *tcp_v4_route_req(const struct sock *sk,
1219 struct flowi *fl,
1220 const struct request_sock *req,
1221 bool *strict)
1222{
1223 struct dst_entry *dst = inet_csk_route_req(sk, &fl->u.ip4, req);
1224
1225 if (strict) {
1226 if (fl->u.ip4.daddr == inet_rsk(req)->ir_rmt_addr)
1227 *strict = true;
1228 else
1229 *strict = false;
1230 }
1231
1232 return dst;
1233}
1234
1235struct request_sock_ops tcp_request_sock_ops __read_mostly = {
1236 .family = PF_INET,
1237 .obj_size = sizeof(struct tcp_request_sock),
1238 .rtx_syn_ack = tcp_rtx_synack,
1239 .send_ack = tcp_v4_reqsk_send_ack,
1240 .destructor = tcp_v4_reqsk_destructor,
1241 .send_reset = tcp_v4_send_reset,
1242 .syn_ack_timeout = tcp_syn_ack_timeout,
1243};
1244
1245static const struct tcp_request_sock_ops tcp_request_sock_ipv4_ops = {
1246 .mss_clamp = TCP_MSS_DEFAULT,
1247#ifdef CONFIG_TCP_MD5SIG
1248 .req_md5_lookup = tcp_v4_md5_lookup,
1249 .calc_md5_hash = tcp_v4_md5_hash_skb,
1250#endif
1251 .init_req = tcp_v4_init_req,
1252#ifdef CONFIG_SYN_COOKIES
1253 .cookie_init_seq = cookie_v4_init_sequence,
1254#endif
1255 .route_req = tcp_v4_route_req,
1256 .init_seq = tcp_v4_init_sequence,
1257 .send_synack = tcp_v4_send_synack,
1258};
1259
1260int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb)
1261{
1262 /* Never answer to SYNs send to broadcast or multicast */
1263 if (skb_rtable(skb)->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))
1264 goto drop;
1265
1266 return tcp_conn_request(&tcp_request_sock_ops,
1267 &tcp_request_sock_ipv4_ops, sk, skb);
1268
1269drop:
1270 tcp_listendrop(sk);
1271 return 0;
1272}
1273EXPORT_SYMBOL(tcp_v4_conn_request);
1274
1275
1276/*
1277 * The three way handshake has completed - we got a valid synack -
1278 * now create the new socket.
1279 */
1280struct sock *tcp_v4_syn_recv_sock(const struct sock *sk, struct sk_buff *skb,
1281 struct request_sock *req,
1282 struct dst_entry *dst,
1283 struct request_sock *req_unhash,
1284 bool *own_req)
1285{
1286 struct inet_request_sock *ireq;
1287 struct inet_sock *newinet;
1288 struct tcp_sock *newtp;
1289 struct sock *newsk;
1290#ifdef CONFIG_TCP_MD5SIG
1291 struct tcp_md5sig_key *key;
1292#endif
1293 struct ip_options_rcu *inet_opt;
1294
1295 if (sk_acceptq_is_full(sk))
1296 goto exit_overflow;
1297
1298 newsk = tcp_create_openreq_child(sk, req, skb);
1299 if (!newsk)
1300 goto exit_nonewsk;
1301
1302 newsk->sk_gso_type = SKB_GSO_TCPV4;
1303 inet_sk_rx_dst_set(newsk, skb);
1304
1305 newtp = tcp_sk(newsk);
1306 newinet = inet_sk(newsk);
1307 ireq = inet_rsk(req);
1308 sk_daddr_set(newsk, ireq->ir_rmt_addr);
1309 sk_rcv_saddr_set(newsk, ireq->ir_loc_addr);
1310 newsk->sk_bound_dev_if = ireq->ir_iif;
1311 newinet->inet_saddr = ireq->ir_loc_addr;
1312 inet_opt = ireq->opt;
1313 rcu_assign_pointer(newinet->inet_opt, inet_opt);
1314 ireq->opt = NULL;
1315 newinet->mc_index = inet_iif(skb);
1316 newinet->mc_ttl = ip_hdr(skb)->ttl;
1317 newinet->rcv_tos = ip_hdr(skb)->tos;
1318 inet_csk(newsk)->icsk_ext_hdr_len = 0;
1319 if (inet_opt)
1320 inet_csk(newsk)->icsk_ext_hdr_len = inet_opt->opt.optlen;
1321 newinet->inet_id = newtp->write_seq ^ jiffies;
1322
1323 if (!dst) {
1324 dst = inet_csk_route_child_sock(sk, newsk, req);
1325 if (!dst)
1326 goto put_and_exit;
1327 } else {
1328 /* syncookie case : see end of cookie_v4_check() */
1329 }
1330 sk_setup_caps(newsk, dst);
1331
1332 tcp_ca_openreq_child(newsk, dst);
1333
1334 tcp_sync_mss(newsk, dst_mtu(dst));
1335 newtp->advmss = tcp_mss_clamp(tcp_sk(sk), dst_metric_advmss(dst));
1336
1337 tcp_initialize_rcv_mss(newsk);
1338
1339#ifdef CONFIG_TCP_MD5SIG
1340 /* Copy over the MD5 key from the original socket */
1341 key = tcp_md5_do_lookup(sk, (union tcp_md5_addr *)&newinet->inet_daddr,
1342 AF_INET);
1343 if (key) {
1344 /*
1345 * We're using one, so create a matching key
1346 * on the newsk structure. If we fail to get
1347 * memory, then we end up not copying the key
1348 * across. Shucks.
1349 */
1350 tcp_md5_do_add(newsk, (union tcp_md5_addr *)&newinet->inet_daddr,
1351 AF_INET, key->key, key->keylen, GFP_ATOMIC);
1352 sk_nocaps_add(newsk, NETIF_F_GSO_MASK);
1353 }
1354#endif
1355
1356 if (__inet_inherit_port(sk, newsk) < 0)
1357 goto put_and_exit;
1358 *own_req = inet_ehash_nolisten(newsk, req_to_sk(req_unhash));
1359 if (*own_req)
1360 tcp_move_syn(newtp, req);
1361
1362 return newsk;
1363
1364exit_overflow:
1365 NET_INC_STATS(sock_net(sk), LINUX_MIB_LISTENOVERFLOWS);
1366exit_nonewsk:
1367 dst_release(dst);
1368exit:
1369 tcp_listendrop(sk);
1370 return NULL;
1371put_and_exit:
1372 inet_csk_prepare_forced_close(newsk);
1373 tcp_done(newsk);
1374 goto exit;
1375}
1376EXPORT_SYMBOL(tcp_v4_syn_recv_sock);
1377
1378static struct sock *tcp_v4_cookie_check(struct sock *sk, struct sk_buff *skb)
1379{
1380#ifdef CONFIG_SYN_COOKIES
1381 const struct tcphdr *th = tcp_hdr(skb);
1382
1383 if (!th->syn)
1384 sk = cookie_v4_check(sk, skb);
1385#endif
1386 return sk;
1387}
1388
1389/* The socket must have it's spinlock held when we get
1390 * here, unless it is a TCP_LISTEN socket.
1391 *
1392 * We have a potential double-lock case here, so even when
1393 * doing backlog processing we use the BH locking scheme.
1394 * This is because we cannot sleep with the original spinlock
1395 * held.
1396 */
1397int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb)
1398{
1399 struct sock *rsk;
1400
1401 if (sk->sk_state == TCP_ESTABLISHED) { /* Fast path */
1402 struct dst_entry *dst = sk->sk_rx_dst;
1403
1404 sock_rps_save_rxhash(sk, skb);
1405 sk_mark_napi_id(sk, skb);
1406 if (dst) {
1407 if (inet_sk(sk)->rx_dst_ifindex != skb->skb_iif ||
1408 !dst->ops->check(dst, 0)) {
1409 dst_release(dst);
1410 sk->sk_rx_dst = NULL;
1411 }
1412 }
1413 tcp_rcv_established(sk, skb, tcp_hdr(skb), skb->len);
1414 return 0;
1415 }
1416
1417 if (tcp_checksum_complete(skb))
1418 goto csum_err;
1419
1420 if (sk->sk_state == TCP_LISTEN) {
1421 struct sock *nsk = tcp_v4_cookie_check(sk, skb);
1422
1423 if (!nsk)
1424 goto discard;
1425 if (nsk != sk) {
1426 sock_rps_save_rxhash(nsk, skb);
1427 sk_mark_napi_id(nsk, skb);
1428 if (tcp_child_process(sk, nsk, skb)) {
1429 rsk = nsk;
1430 goto reset;
1431 }
1432 return 0;
1433 }
1434 } else
1435 sock_rps_save_rxhash(sk, skb);
1436
1437 if (tcp_rcv_state_process(sk, skb)) {
1438 rsk = sk;
1439 goto reset;
1440 }
1441 return 0;
1442
1443reset:
1444 tcp_v4_send_reset(rsk, skb);
1445discard:
1446 kfree_skb(skb);
1447 /* Be careful here. If this function gets more complicated and
1448 * gcc suffers from register pressure on the x86, sk (in %ebx)
1449 * might be destroyed here. This current version compiles correctly,
1450 * but you have been warned.
1451 */
1452 return 0;
1453
1454csum_err:
1455 TCP_INC_STATS(sock_net(sk), TCP_MIB_CSUMERRORS);
1456 TCP_INC_STATS(sock_net(sk), TCP_MIB_INERRS);
1457 goto discard;
1458}
1459EXPORT_SYMBOL(tcp_v4_do_rcv);
1460
1461void tcp_v4_early_demux(struct sk_buff *skb)
1462{
1463 const struct iphdr *iph;
1464 const struct tcphdr *th;
1465 struct sock *sk;
1466
1467 if (skb->pkt_type != PACKET_HOST)
1468 return;
1469
1470 if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct tcphdr)))
1471 return;
1472
1473 iph = ip_hdr(skb);
1474 th = tcp_hdr(skb);
1475
1476 if (th->doff < sizeof(struct tcphdr) / 4)
1477 return;
1478
1479 sk = __inet_lookup_established(dev_net(skb->dev), &tcp_hashinfo,
1480 iph->saddr, th->source,
1481 iph->daddr, ntohs(th->dest),
1482 skb->skb_iif);
1483 if (sk) {
1484 skb->sk = sk;
1485 skb->destructor = sock_edemux;
1486 if (sk_fullsock(sk)) {
1487 struct dst_entry *dst = READ_ONCE(sk->sk_rx_dst);
1488
1489 if (dst)
1490 dst = dst_check(dst, 0);
1491 if (dst &&
1492 inet_sk(sk)->rx_dst_ifindex == skb->skb_iif)
1493 skb_dst_set_noref(skb, dst);
1494 }
1495 }
1496}
1497
1498/* Packet is added to VJ-style prequeue for processing in process
1499 * context, if a reader task is waiting. Apparently, this exciting
1500 * idea (VJ's mail "Re: query about TCP header on tcp-ip" of 07 Sep 93)
1501 * failed somewhere. Latency? Burstiness? Well, at least now we will
1502 * see, why it failed. 8)8) --ANK
1503 *
1504 */
1505bool tcp_prequeue(struct sock *sk, struct sk_buff *skb)
1506{
1507 struct tcp_sock *tp = tcp_sk(sk);
1508
1509 if (sysctl_tcp_low_latency || !tp->ucopy.task)
1510 return false;
1511
1512 if (skb->len <= tcp_hdrlen(skb) &&
1513 skb_queue_len(&tp->ucopy.prequeue) == 0)
1514 return false;
1515
1516 /* Before escaping RCU protected region, we need to take care of skb
1517 * dst. Prequeue is only enabled for established sockets.
1518 * For such sockets, we might need the skb dst only to set sk->sk_rx_dst
1519 * Instead of doing full sk_rx_dst validity here, let's perform
1520 * an optimistic check.
1521 */
1522 if (likely(sk->sk_rx_dst))
1523 skb_dst_drop(skb);
1524 else
1525 skb_dst_force_safe(skb);
1526
1527 __skb_queue_tail(&tp->ucopy.prequeue, skb);
1528 tp->ucopy.memory += skb->truesize;
1529 if (skb_queue_len(&tp->ucopy.prequeue) >= 32 ||
1530 tp->ucopy.memory + atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf) {
1531 struct sk_buff *skb1;
1532
1533 BUG_ON(sock_owned_by_user(sk));
1534 __NET_ADD_STATS(sock_net(sk), LINUX_MIB_TCPPREQUEUEDROPPED,
1535 skb_queue_len(&tp->ucopy.prequeue));
1536
1537 while ((skb1 = __skb_dequeue(&tp->ucopy.prequeue)) != NULL)
1538 sk_backlog_rcv(sk, skb1);
1539
1540 tp->ucopy.memory = 0;
1541 } else if (skb_queue_len(&tp->ucopy.prequeue) == 1) {
1542 wake_up_interruptible_sync_poll(sk_sleep(sk),
1543 POLLIN | POLLRDNORM | POLLRDBAND);
1544 if (!inet_csk_ack_scheduled(sk))
1545 inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK,
1546 (3 * tcp_rto_min(sk)) / 4,
1547 TCP_RTO_MAX);
1548 }
1549 return true;
1550}
1551EXPORT_SYMBOL(tcp_prequeue);
1552
1553bool tcp_add_backlog(struct sock *sk, struct sk_buff *skb)
1554{
1555 u32 limit = sk->sk_rcvbuf + sk->sk_sndbuf;
1556
1557 /* Only socket owner can try to collapse/prune rx queues
1558 * to reduce memory overhead, so add a little headroom here.
1559 * Few sockets backlog are possibly concurrently non empty.
1560 */
1561 limit += 64*1024;
1562
1563 /* In case all data was pulled from skb frags (in __pskb_pull_tail()),
1564 * we can fix skb->truesize to its real value to avoid future drops.
1565 * This is valid because skb is not yet charged to the socket.
1566 * It has been noticed pure SACK packets were sometimes dropped
1567 * (if cooked by drivers without copybreak feature).
1568 */
1569 skb_condense(skb);
1570
1571 if (unlikely(sk_add_backlog(sk, skb, limit))) {
1572 bh_unlock_sock(sk);
1573 __NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPBACKLOGDROP);
1574 return true;
1575 }
1576 return false;
1577}
1578EXPORT_SYMBOL(tcp_add_backlog);
1579
1580int tcp_filter(struct sock *sk, struct sk_buff *skb)
1581{
1582 struct tcphdr *th = (struct tcphdr *)skb->data;
1583 unsigned int eaten = skb->len;
1584 int err;
1585
1586 err = sk_filter_trim_cap(sk, skb, th->doff * 4);
1587 if (!err) {
1588 eaten -= skb->len;
1589 TCP_SKB_CB(skb)->end_seq -= eaten;
1590 }
1591 return err;
1592}
1593EXPORT_SYMBOL(tcp_filter);
1594
1595/*
1596 * From tcp_input.c
1597 */
1598
1599int tcp_v4_rcv(struct sk_buff *skb)
1600{
1601 struct net *net = dev_net(skb->dev);
1602 const struct iphdr *iph;
1603 const struct tcphdr *th;
1604 bool refcounted;
1605 struct sock *sk;
1606 int ret;
1607
1608 if (skb->pkt_type != PACKET_HOST)
1609 goto discard_it;
1610
1611 /* Count it even if it's bad */
1612 __TCP_INC_STATS(net, TCP_MIB_INSEGS);
1613
1614 if (!pskb_may_pull(skb, sizeof(struct tcphdr)))
1615 goto discard_it;
1616
1617 th = (const struct tcphdr *)skb->data;
1618
1619 if (unlikely(th->doff < sizeof(struct tcphdr) / 4))
1620 goto bad_packet;
1621 if (!pskb_may_pull(skb, th->doff * 4))
1622 goto discard_it;
1623
1624 /* An explanation is required here, I think.
1625 * Packet length and doff are validated by header prediction,
1626 * provided case of th->doff==0 is eliminated.
1627 * So, we defer the checks. */
1628
1629 if (skb_checksum_init(skb, IPPROTO_TCP, inet_compute_pseudo))
1630 goto csum_error;
1631
1632 th = (const struct tcphdr *)skb->data;
1633 iph = ip_hdr(skb);
1634 /* This is tricky : We move IPCB at its correct location into TCP_SKB_CB()
1635 * barrier() makes sure compiler wont play fool^Waliasing games.
1636 */
1637 memmove(&TCP_SKB_CB(skb)->header.h4, IPCB(skb),
1638 sizeof(struct inet_skb_parm));
1639 barrier();
1640
1641 TCP_SKB_CB(skb)->seq = ntohl(th->seq);
1642 TCP_SKB_CB(skb)->end_seq = (TCP_SKB_CB(skb)->seq + th->syn + th->fin +
1643 skb->len - th->doff * 4);
1644 TCP_SKB_CB(skb)->ack_seq = ntohl(th->ack_seq);
1645 TCP_SKB_CB(skb)->tcp_flags = tcp_flag_byte(th);
1646 TCP_SKB_CB(skb)->tcp_tw_isn = 0;
1647 TCP_SKB_CB(skb)->ip_dsfield = ipv4_get_dsfield(iph);
1648 TCP_SKB_CB(skb)->sacked = 0;
1649
1650lookup:
1651 sk = __inet_lookup_skb(&tcp_hashinfo, skb, __tcp_hdrlen(th), th->source,
1652 th->dest, &refcounted);
1653 if (!sk)
1654 goto no_tcp_socket;
1655
1656process:
1657 if (sk->sk_state == TCP_TIME_WAIT)
1658 goto do_time_wait;
1659
1660 if (sk->sk_state == TCP_NEW_SYN_RECV) {
1661 struct request_sock *req = inet_reqsk(sk);
1662 struct sock *nsk;
1663
1664 sk = req->rsk_listener;
1665 if (unlikely(tcp_v4_inbound_md5_hash(sk, skb))) {
1666 sk_drops_add(sk, skb);
1667 reqsk_put(req);
1668 goto discard_it;
1669 }
1670 if (unlikely(sk->sk_state != TCP_LISTEN)) {
1671 inet_csk_reqsk_queue_drop_and_put(sk, req);
1672 goto lookup;
1673 }
1674 /* We own a reference on the listener, increase it again
1675 * as we might lose it too soon.
1676 */
1677 sock_hold(sk);
1678 refcounted = true;
1679 nsk = tcp_check_req(sk, skb, req, false);
1680 if (!nsk) {
1681 reqsk_put(req);
1682 goto discard_and_relse;
1683 }
1684 if (nsk == sk) {
1685 reqsk_put(req);
1686 } else if (tcp_child_process(sk, nsk, skb)) {
1687 tcp_v4_send_reset(nsk, skb);
1688 goto discard_and_relse;
1689 } else {
1690 sock_put(sk);
1691 return 0;
1692 }
1693 }
1694 if (unlikely(iph->ttl < inet_sk(sk)->min_ttl)) {
1695 __NET_INC_STATS(net, LINUX_MIB_TCPMINTTLDROP);
1696 goto discard_and_relse;
1697 }
1698
1699 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
1700 goto discard_and_relse;
1701
1702 if (tcp_v4_inbound_md5_hash(sk, skb))
1703 goto discard_and_relse;
1704
1705 nf_reset(skb);
1706
1707 if (tcp_filter(sk, skb))
1708 goto discard_and_relse;
1709 th = (const struct tcphdr *)skb->data;
1710 iph = ip_hdr(skb);
1711
1712 skb->dev = NULL;
1713
1714 if (sk->sk_state == TCP_LISTEN) {
1715 ret = tcp_v4_do_rcv(sk, skb);
1716 goto put_and_return;
1717 }
1718
1719 sk_incoming_cpu_update(sk);
1720
1721 bh_lock_sock_nested(sk);
1722 tcp_segs_in(tcp_sk(sk), skb);
1723 ret = 0;
1724 if (!sock_owned_by_user(sk)) {
1725 if (!tcp_prequeue(sk, skb))
1726 ret = tcp_v4_do_rcv(sk, skb);
1727 } else if (tcp_add_backlog(sk, skb)) {
1728 goto discard_and_relse;
1729 }
1730 bh_unlock_sock(sk);
1731
1732put_and_return:
1733 if (refcounted)
1734 sock_put(sk);
1735
1736 return ret;
1737
1738no_tcp_socket:
1739 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
1740 goto discard_it;
1741
1742 if (tcp_checksum_complete(skb)) {
1743csum_error:
1744 __TCP_INC_STATS(net, TCP_MIB_CSUMERRORS);
1745bad_packet:
1746 __TCP_INC_STATS(net, TCP_MIB_INERRS);
1747 } else {
1748 tcp_v4_send_reset(NULL, skb);
1749 }
1750
1751discard_it:
1752 /* Discard frame. */
1753 kfree_skb(skb);
1754 return 0;
1755
1756discard_and_relse:
1757 sk_drops_add(sk, skb);
1758 if (refcounted)
1759 sock_put(sk);
1760 goto discard_it;
1761
1762do_time_wait:
1763 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) {
1764 inet_twsk_put(inet_twsk(sk));
1765 goto discard_it;
1766 }
1767
1768 if (tcp_checksum_complete(skb)) {
1769 inet_twsk_put(inet_twsk(sk));
1770 goto csum_error;
1771 }
1772 switch (tcp_timewait_state_process(inet_twsk(sk), skb, th)) {
1773 case TCP_TW_SYN: {
1774 struct sock *sk2 = inet_lookup_listener(dev_net(skb->dev),
1775 &tcp_hashinfo, skb,
1776 __tcp_hdrlen(th),
1777 iph->saddr, th->source,
1778 iph->daddr, th->dest,
1779 inet_iif(skb));
1780 if (sk2) {
1781 inet_twsk_deschedule_put(inet_twsk(sk));
1782 sk = sk2;
1783 refcounted = false;
1784 goto process;
1785 }
1786 /* Fall through to ACK */
1787 }
1788 case TCP_TW_ACK:
1789 tcp_v4_timewait_ack(sk, skb);
1790 break;
1791 case TCP_TW_RST:
1792 tcp_v4_send_reset(sk, skb);
1793 inet_twsk_deschedule_put(inet_twsk(sk));
1794 goto discard_it;
1795 case TCP_TW_SUCCESS:;
1796 }
1797 goto discard_it;
1798}
1799
1800static struct timewait_sock_ops tcp_timewait_sock_ops = {
1801 .twsk_obj_size = sizeof(struct tcp_timewait_sock),
1802 .twsk_unique = tcp_twsk_unique,
1803 .twsk_destructor= tcp_twsk_destructor,
1804};
1805
1806void inet_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb)
1807{
1808 struct dst_entry *dst = skb_dst(skb);
1809
1810 if (dst && dst_hold_safe(dst)) {
1811 sk->sk_rx_dst = dst;
1812 inet_sk(sk)->rx_dst_ifindex = skb->skb_iif;
1813 }
1814}
1815EXPORT_SYMBOL(inet_sk_rx_dst_set);
1816
1817const struct inet_connection_sock_af_ops ipv4_specific = {
1818 .queue_xmit = ip_queue_xmit,
1819 .send_check = tcp_v4_send_check,
1820 .rebuild_header = inet_sk_rebuild_header,
1821 .sk_rx_dst_set = inet_sk_rx_dst_set,
1822 .conn_request = tcp_v4_conn_request,
1823 .syn_recv_sock = tcp_v4_syn_recv_sock,
1824 .net_header_len = sizeof(struct iphdr),
1825 .setsockopt = ip_setsockopt,
1826 .getsockopt = ip_getsockopt,
1827 .addr2sockaddr = inet_csk_addr2sockaddr,
1828 .sockaddr_len = sizeof(struct sockaddr_in),
1829#ifdef CONFIG_COMPAT
1830 .compat_setsockopt = compat_ip_setsockopt,
1831 .compat_getsockopt = compat_ip_getsockopt,
1832#endif
1833 .mtu_reduced = tcp_v4_mtu_reduced,
1834};
1835EXPORT_SYMBOL(ipv4_specific);
1836
1837#ifdef CONFIG_TCP_MD5SIG
1838static const struct tcp_sock_af_ops tcp_sock_ipv4_specific = {
1839 .md5_lookup = tcp_v4_md5_lookup,
1840 .calc_md5_hash = tcp_v4_md5_hash_skb,
1841 .md5_parse = tcp_v4_parse_md5_keys,
1842};
1843#endif
1844
1845/* NOTE: A lot of things set to zero explicitly by call to
1846 * sk_alloc() so need not be done here.
1847 */
1848static int tcp_v4_init_sock(struct sock *sk)
1849{
1850 struct inet_connection_sock *icsk = inet_csk(sk);
1851
1852 tcp_init_sock(sk);
1853
1854 icsk->icsk_af_ops = &ipv4_specific;
1855
1856#ifdef CONFIG_TCP_MD5SIG
1857 tcp_sk(sk)->af_specific = &tcp_sock_ipv4_specific;
1858#endif
1859
1860 return 0;
1861}
1862
1863void tcp_v4_destroy_sock(struct sock *sk)
1864{
1865 struct tcp_sock *tp = tcp_sk(sk);
1866
1867 tcp_clear_xmit_timers(sk);
1868
1869 tcp_cleanup_congestion_control(sk);
1870
1871 /* Cleanup up the write buffer. */
1872 tcp_write_queue_purge(sk);
1873
1874 /* Cleans up our, hopefully empty, out_of_order_queue. */
1875 skb_rbtree_purge(&tp->out_of_order_queue);
1876
1877#ifdef CONFIG_TCP_MD5SIG
1878 /* Clean up the MD5 key list, if any */
1879 if (tp->md5sig_info) {
1880 tcp_clear_md5_list(sk);
1881 kfree_rcu(tp->md5sig_info, rcu);
1882 tp->md5sig_info = NULL;
1883 }
1884#endif
1885
1886 /* Clean prequeue, it must be empty really */
1887 __skb_queue_purge(&tp->ucopy.prequeue);
1888
1889 /* Clean up a referenced TCP bind bucket. */
1890 if (inet_csk(sk)->icsk_bind_hash)
1891 inet_put_port(sk);
1892
1893 BUG_ON(tp->fastopen_rsk);
1894
1895 /* If socket is aborted during connect operation */
1896 tcp_free_fastopen_req(tp);
1897 tcp_saved_syn_free(tp);
1898
1899 sk_sockets_allocated_dec(sk);
1900}
1901EXPORT_SYMBOL(tcp_v4_destroy_sock);
1902
1903#ifdef CONFIG_PROC_FS
1904/* Proc filesystem TCP sock list dumping. */
1905
1906/*
1907 * Get next listener socket follow cur. If cur is NULL, get first socket
1908 * starting from bucket given in st->bucket; when st->bucket is zero the
1909 * very first socket in the hash table is returned.
1910 */
1911static void *listening_get_next(struct seq_file *seq, void *cur)
1912{
1913 struct tcp_iter_state *st = seq->private;
1914 struct net *net = seq_file_net(seq);
1915 struct inet_listen_hashbucket *ilb;
1916 struct sock *sk = cur;
1917
1918 if (!sk) {
1919get_head:
1920 ilb = &tcp_hashinfo.listening_hash[st->bucket];
1921 spin_lock(&ilb->lock);
1922 sk = sk_head(&ilb->head);
1923 st->offset = 0;
1924 goto get_sk;
1925 }
1926 ilb = &tcp_hashinfo.listening_hash[st->bucket];
1927 ++st->num;
1928 ++st->offset;
1929
1930 sk = sk_next(sk);
1931get_sk:
1932 sk_for_each_from(sk) {
1933 if (!net_eq(sock_net(sk), net))
1934 continue;
1935 if (sk->sk_family == st->family)
1936 return sk;
1937 }
1938 spin_unlock(&ilb->lock);
1939 st->offset = 0;
1940 if (++st->bucket < INET_LHTABLE_SIZE)
1941 goto get_head;
1942 return NULL;
1943}
1944
1945static void *listening_get_idx(struct seq_file *seq, loff_t *pos)
1946{
1947 struct tcp_iter_state *st = seq->private;
1948 void *rc;
1949
1950 st->bucket = 0;
1951 st->offset = 0;
1952 rc = listening_get_next(seq, NULL);
1953
1954 while (rc && *pos) {
1955 rc = listening_get_next(seq, rc);
1956 --*pos;
1957 }
1958 return rc;
1959}
1960
1961static inline bool empty_bucket(const struct tcp_iter_state *st)
1962{
1963 return hlist_nulls_empty(&tcp_hashinfo.ehash[st->bucket].chain);
1964}
1965
1966/*
1967 * Get first established socket starting from bucket given in st->bucket.
1968 * If st->bucket is zero, the very first socket in the hash is returned.
1969 */
1970static void *established_get_first(struct seq_file *seq)
1971{
1972 struct tcp_iter_state *st = seq->private;
1973 struct net *net = seq_file_net(seq);
1974 void *rc = NULL;
1975
1976 st->offset = 0;
1977 for (; st->bucket <= tcp_hashinfo.ehash_mask; ++st->bucket) {
1978 struct sock *sk;
1979 struct hlist_nulls_node *node;
1980 spinlock_t *lock = inet_ehash_lockp(&tcp_hashinfo, st->bucket);
1981
1982 /* Lockless fast path for the common case of empty buckets */
1983 if (empty_bucket(st))
1984 continue;
1985
1986 spin_lock_bh(lock);
1987 sk_nulls_for_each(sk, node, &tcp_hashinfo.ehash[st->bucket].chain) {
1988 if (sk->sk_family != st->family ||
1989 !net_eq(sock_net(sk), net)) {
1990 continue;
1991 }
1992 rc = sk;
1993 goto out;
1994 }
1995 spin_unlock_bh(lock);
1996 }
1997out:
1998 return rc;
1999}
2000
2001static void *established_get_next(struct seq_file *seq, void *cur)
2002{
2003 struct sock *sk = cur;
2004 struct hlist_nulls_node *node;
2005 struct tcp_iter_state *st = seq->private;
2006 struct net *net = seq_file_net(seq);
2007
2008 ++st->num;
2009 ++st->offset;
2010
2011 sk = sk_nulls_next(sk);
2012
2013 sk_nulls_for_each_from(sk, node) {
2014 if (sk->sk_family == st->family && net_eq(sock_net(sk), net))
2015 return sk;
2016 }
2017
2018 spin_unlock_bh(inet_ehash_lockp(&tcp_hashinfo, st->bucket));
2019 ++st->bucket;
2020 return established_get_first(seq);
2021}
2022
2023static void *established_get_idx(struct seq_file *seq, loff_t pos)
2024{
2025 struct tcp_iter_state *st = seq->private;
2026 void *rc;
2027
2028 st->bucket = 0;
2029 rc = established_get_first(seq);
2030
2031 while (rc && pos) {
2032 rc = established_get_next(seq, rc);
2033 --pos;
2034 }
2035 return rc;
2036}
2037
2038static void *tcp_get_idx(struct seq_file *seq, loff_t pos)
2039{
2040 void *rc;
2041 struct tcp_iter_state *st = seq->private;
2042
2043 st->state = TCP_SEQ_STATE_LISTENING;
2044 rc = listening_get_idx(seq, &pos);
2045
2046 if (!rc) {
2047 st->state = TCP_SEQ_STATE_ESTABLISHED;
2048 rc = established_get_idx(seq, pos);
2049 }
2050
2051 return rc;
2052}
2053
2054static void *tcp_seek_last_pos(struct seq_file *seq)
2055{
2056 struct tcp_iter_state *st = seq->private;
2057 int offset = st->offset;
2058 int orig_num = st->num;
2059 void *rc = NULL;
2060
2061 switch (st->state) {
2062 case TCP_SEQ_STATE_LISTENING:
2063 if (st->bucket >= INET_LHTABLE_SIZE)
2064 break;
2065 st->state = TCP_SEQ_STATE_LISTENING;
2066 rc = listening_get_next(seq, NULL);
2067 while (offset-- && rc)
2068 rc = listening_get_next(seq, rc);
2069 if (rc)
2070 break;
2071 st->bucket = 0;
2072 st->state = TCP_SEQ_STATE_ESTABLISHED;
2073 /* Fallthrough */
2074 case TCP_SEQ_STATE_ESTABLISHED:
2075 if (st->bucket > tcp_hashinfo.ehash_mask)
2076 break;
2077 rc = established_get_first(seq);
2078 while (offset-- && rc)
2079 rc = established_get_next(seq, rc);
2080 }
2081
2082 st->num = orig_num;
2083
2084 return rc;
2085}
2086
2087static void *tcp_seq_start(struct seq_file *seq, loff_t *pos)
2088{
2089 struct tcp_iter_state *st = seq->private;
2090 void *rc;
2091
2092 if (*pos && *pos == st->last_pos) {
2093 rc = tcp_seek_last_pos(seq);
2094 if (rc)
2095 goto out;
2096 }
2097
2098 st->state = TCP_SEQ_STATE_LISTENING;
2099 st->num = 0;
2100 st->bucket = 0;
2101 st->offset = 0;
2102 rc = *pos ? tcp_get_idx(seq, *pos - 1) : SEQ_START_TOKEN;
2103
2104out:
2105 st->last_pos = *pos;
2106 return rc;
2107}
2108
2109static void *tcp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2110{
2111 struct tcp_iter_state *st = seq->private;
2112 void *rc = NULL;
2113
2114 if (v == SEQ_START_TOKEN) {
2115 rc = tcp_get_idx(seq, 0);
2116 goto out;
2117 }
2118
2119 switch (st->state) {
2120 case TCP_SEQ_STATE_LISTENING:
2121 rc = listening_get_next(seq, v);
2122 if (!rc) {
2123 st->state = TCP_SEQ_STATE_ESTABLISHED;
2124 st->bucket = 0;
2125 st->offset = 0;
2126 rc = established_get_first(seq);
2127 }
2128 break;
2129 case TCP_SEQ_STATE_ESTABLISHED:
2130 rc = established_get_next(seq, v);
2131 break;
2132 }
2133out:
2134 ++*pos;
2135 st->last_pos = *pos;
2136 return rc;
2137}
2138
2139static void tcp_seq_stop(struct seq_file *seq, void *v)
2140{
2141 struct tcp_iter_state *st = seq->private;
2142
2143 switch (st->state) {
2144 case TCP_SEQ_STATE_LISTENING:
2145 if (v != SEQ_START_TOKEN)
2146 spin_unlock(&tcp_hashinfo.listening_hash[st->bucket].lock);
2147 break;
2148 case TCP_SEQ_STATE_ESTABLISHED:
2149 if (v)
2150 spin_unlock_bh(inet_ehash_lockp(&tcp_hashinfo, st->bucket));
2151 break;
2152 }
2153}
2154
2155int tcp_seq_open(struct inode *inode, struct file *file)
2156{
2157 struct tcp_seq_afinfo *afinfo = PDE_DATA(inode);
2158 struct tcp_iter_state *s;
2159 int err;
2160
2161 err = seq_open_net(inode, file, &afinfo->seq_ops,
2162 sizeof(struct tcp_iter_state));
2163 if (err < 0)
2164 return err;
2165
2166 s = ((struct seq_file *)file->private_data)->private;
2167 s->family = afinfo->family;
2168 s->last_pos = 0;
2169 return 0;
2170}
2171EXPORT_SYMBOL(tcp_seq_open);
2172
2173int tcp_proc_register(struct net *net, struct tcp_seq_afinfo *afinfo)
2174{
2175 int rc = 0;
2176 struct proc_dir_entry *p;
2177
2178 afinfo->seq_ops.start = tcp_seq_start;
2179 afinfo->seq_ops.next = tcp_seq_next;
2180 afinfo->seq_ops.stop = tcp_seq_stop;
2181
2182 p = proc_create_data(afinfo->name, S_IRUGO, net->proc_net,
2183 afinfo->seq_fops, afinfo);
2184 if (!p)
2185 rc = -ENOMEM;
2186 return rc;
2187}
2188EXPORT_SYMBOL(tcp_proc_register);
2189
2190void tcp_proc_unregister(struct net *net, struct tcp_seq_afinfo *afinfo)
2191{
2192 remove_proc_entry(afinfo->name, net->proc_net);
2193}
2194EXPORT_SYMBOL(tcp_proc_unregister);
2195
2196static void get_openreq4(const struct request_sock *req,
2197 struct seq_file *f, int i)
2198{
2199 const struct inet_request_sock *ireq = inet_rsk(req);
2200 long delta = req->rsk_timer.expires - jiffies;
2201
2202 seq_printf(f, "%4d: %08X:%04X %08X:%04X"
2203 " %02X %08X:%08X %02X:%08lX %08X %5u %8d %u %d %pK",
2204 i,
2205 ireq->ir_loc_addr,
2206 ireq->ir_num,
2207 ireq->ir_rmt_addr,
2208 ntohs(ireq->ir_rmt_port),
2209 TCP_SYN_RECV,
2210 0, 0, /* could print option size, but that is af dependent. */
2211 1, /* timers active (only the expire timer) */
2212 jiffies_delta_to_clock_t(delta),
2213 req->num_timeout,
2214 from_kuid_munged(seq_user_ns(f),
2215 sock_i_uid(req->rsk_listener)),
2216 0, /* non standard timer */
2217 0, /* open_requests have no inode */
2218 0,
2219 req);
2220}
2221
2222static void get_tcp4_sock(struct sock *sk, struct seq_file *f, int i)
2223{
2224 int timer_active;
2225 unsigned long timer_expires;
2226 const struct tcp_sock *tp = tcp_sk(sk);
2227 const struct inet_connection_sock *icsk = inet_csk(sk);
2228 const struct inet_sock *inet = inet_sk(sk);
2229 const struct fastopen_queue *fastopenq = &icsk->icsk_accept_queue.fastopenq;
2230 __be32 dest = inet->inet_daddr;
2231 __be32 src = inet->inet_rcv_saddr;
2232 __u16 destp = ntohs(inet->inet_dport);
2233 __u16 srcp = ntohs(inet->inet_sport);
2234 int rx_queue;
2235 int state;
2236
2237 if (icsk->icsk_pending == ICSK_TIME_RETRANS ||
2238 icsk->icsk_pending == ICSK_TIME_REO_TIMEOUT ||
2239 icsk->icsk_pending == ICSK_TIME_LOSS_PROBE) {
2240 timer_active = 1;
2241 timer_expires = icsk->icsk_timeout;
2242 } else if (icsk->icsk_pending == ICSK_TIME_PROBE0) {
2243 timer_active = 4;
2244 timer_expires = icsk->icsk_timeout;
2245 } else if (timer_pending(&sk->sk_timer)) {
2246 timer_active = 2;
2247 timer_expires = sk->sk_timer.expires;
2248 } else {
2249 timer_active = 0;
2250 timer_expires = jiffies;
2251 }
2252
2253 state = sk_state_load(sk);
2254 if (state == TCP_LISTEN)
2255 rx_queue = sk->sk_ack_backlog;
2256 else
2257 /* Because we don't lock the socket,
2258 * we might find a transient negative value.
2259 */
2260 rx_queue = max_t(int, tp->rcv_nxt - tp->copied_seq, 0);
2261
2262 seq_printf(f, "%4d: %08X:%04X %08X:%04X %02X %08X:%08X %02X:%08lX "
2263 "%08X %5u %8d %lu %d %pK %lu %lu %u %u %d",
2264 i, src, srcp, dest, destp, state,
2265 tp->write_seq - tp->snd_una,
2266 rx_queue,
2267 timer_active,
2268 jiffies_delta_to_clock_t(timer_expires - jiffies),
2269 icsk->icsk_retransmits,
2270 from_kuid_munged(seq_user_ns(f), sock_i_uid(sk)),
2271 icsk->icsk_probes_out,
2272 sock_i_ino(sk),
2273 atomic_read(&sk->sk_refcnt), sk,
2274 jiffies_to_clock_t(icsk->icsk_rto),
2275 jiffies_to_clock_t(icsk->icsk_ack.ato),
2276 (icsk->icsk_ack.quick << 1) | icsk->icsk_ack.pingpong,
2277 tp->snd_cwnd,
2278 state == TCP_LISTEN ?
2279 fastopenq->max_qlen :
2280 (tcp_in_initial_slowstart(tp) ? -1 : tp->snd_ssthresh));
2281}
2282
2283static void get_timewait4_sock(const struct inet_timewait_sock *tw,
2284 struct seq_file *f, int i)
2285{
2286 long delta = tw->tw_timer.expires - jiffies;
2287 __be32 dest, src;
2288 __u16 destp, srcp;
2289
2290 dest = tw->tw_daddr;
2291 src = tw->tw_rcv_saddr;
2292 destp = ntohs(tw->tw_dport);
2293 srcp = ntohs(tw->tw_sport);
2294
2295 seq_printf(f, "%4d: %08X:%04X %08X:%04X"
2296 " %02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %pK",
2297 i, src, srcp, dest, destp, tw->tw_substate, 0, 0,
2298 3, jiffies_delta_to_clock_t(delta), 0, 0, 0, 0,
2299 atomic_read(&tw->tw_refcnt), tw);
2300}
2301
2302#define TMPSZ 150
2303
2304static int tcp4_seq_show(struct seq_file *seq, void *v)
2305{
2306 struct tcp_iter_state *st;
2307 struct sock *sk = v;
2308
2309 seq_setwidth(seq, TMPSZ - 1);
2310 if (v == SEQ_START_TOKEN) {
2311 seq_puts(seq, " sl local_address rem_address st tx_queue "
2312 "rx_queue tr tm->when retrnsmt uid timeout "
2313 "inode");
2314 goto out;
2315 }
2316 st = seq->private;
2317
2318 if (sk->sk_state == TCP_TIME_WAIT)
2319 get_timewait4_sock(v, seq, st->num);
2320 else if (sk->sk_state == TCP_NEW_SYN_RECV)
2321 get_openreq4(v, seq, st->num);
2322 else
2323 get_tcp4_sock(v, seq, st->num);
2324out:
2325 seq_pad(seq, '\n');
2326 return 0;
2327}
2328
2329static const struct file_operations tcp_afinfo_seq_fops = {
2330 .owner = THIS_MODULE,
2331 .open = tcp_seq_open,
2332 .read = seq_read,
2333 .llseek = seq_lseek,
2334 .release = seq_release_net
2335};
2336
2337static struct tcp_seq_afinfo tcp4_seq_afinfo = {
2338 .name = "tcp",
2339 .family = AF_INET,
2340 .seq_fops = &tcp_afinfo_seq_fops,
2341 .seq_ops = {
2342 .show = tcp4_seq_show,
2343 },
2344};
2345
2346static int __net_init tcp4_proc_init_net(struct net *net)
2347{
2348 return tcp_proc_register(net, &tcp4_seq_afinfo);
2349}
2350
2351static void __net_exit tcp4_proc_exit_net(struct net *net)
2352{
2353 tcp_proc_unregister(net, &tcp4_seq_afinfo);
2354}
2355
2356static struct pernet_operations tcp4_net_ops = {
2357 .init = tcp4_proc_init_net,
2358 .exit = tcp4_proc_exit_net,
2359};
2360
2361int __init tcp4_proc_init(void)
2362{
2363 return register_pernet_subsys(&tcp4_net_ops);
2364}
2365
2366void tcp4_proc_exit(void)
2367{
2368 unregister_pernet_subsys(&tcp4_net_ops);
2369}
2370#endif /* CONFIG_PROC_FS */
2371
2372struct proto tcp_prot = {
2373 .name = "TCP",
2374 .owner = THIS_MODULE,
2375 .close = tcp_close,
2376 .connect = tcp_v4_connect,
2377 .disconnect = tcp_disconnect,
2378 .accept = inet_csk_accept,
2379 .ioctl = tcp_ioctl,
2380 .init = tcp_v4_init_sock,
2381 .destroy = tcp_v4_destroy_sock,
2382 .shutdown = tcp_shutdown,
2383 .setsockopt = tcp_setsockopt,
2384 .getsockopt = tcp_getsockopt,
2385 .keepalive = tcp_set_keepalive,
2386 .recvmsg = tcp_recvmsg,
2387 .sendmsg = tcp_sendmsg,
2388 .sendpage = tcp_sendpage,
2389 .backlog_rcv = tcp_v4_do_rcv,
2390 .release_cb = tcp_release_cb,
2391 .hash = inet_hash,
2392 .unhash = inet_unhash,
2393 .get_port = inet_csk_get_port,
2394 .enter_memory_pressure = tcp_enter_memory_pressure,
2395 .stream_memory_free = tcp_stream_memory_free,
2396 .sockets_allocated = &tcp_sockets_allocated,
2397 .orphan_count = &tcp_orphan_count,
2398 .memory_allocated = &tcp_memory_allocated,
2399 .memory_pressure = &tcp_memory_pressure,
2400 .sysctl_mem = sysctl_tcp_mem,
2401 .sysctl_wmem = sysctl_tcp_wmem,
2402 .sysctl_rmem = sysctl_tcp_rmem,
2403 .max_header = MAX_TCP_HEADER,
2404 .obj_size = sizeof(struct tcp_sock),
2405 .slab_flags = SLAB_DESTROY_BY_RCU,
2406 .twsk_prot = &tcp_timewait_sock_ops,
2407 .rsk_prot = &tcp_request_sock_ops,
2408 .h.hashinfo = &tcp_hashinfo,
2409 .no_autobind = true,
2410#ifdef CONFIG_COMPAT
2411 .compat_setsockopt = compat_tcp_setsockopt,
2412 .compat_getsockopt = compat_tcp_getsockopt,
2413#endif
2414 .diag_destroy = tcp_abort,
2415};
2416EXPORT_SYMBOL(tcp_prot);
2417
2418static void __net_exit tcp_sk_exit(struct net *net)
2419{
2420 int cpu;
2421
2422 for_each_possible_cpu(cpu)
2423 inet_ctl_sock_destroy(*per_cpu_ptr(net->ipv4.tcp_sk, cpu));
2424 free_percpu(net->ipv4.tcp_sk);
2425}
2426
2427static int __net_init tcp_sk_init(struct net *net)
2428{
2429 int res, cpu, cnt;
2430
2431 net->ipv4.tcp_sk = alloc_percpu(struct sock *);
2432 if (!net->ipv4.tcp_sk)
2433 return -ENOMEM;
2434
2435 for_each_possible_cpu(cpu) {
2436 struct sock *sk;
2437
2438 res = inet_ctl_sock_create(&sk, PF_INET, SOCK_RAW,
2439 IPPROTO_TCP, net);
2440 if (res)
2441 goto fail;
2442 sock_set_flag(sk, SOCK_USE_WRITE_QUEUE);
2443 *per_cpu_ptr(net->ipv4.tcp_sk, cpu) = sk;
2444 }
2445
2446 net->ipv4.sysctl_tcp_ecn = 2;
2447 net->ipv4.sysctl_tcp_ecn_fallback = 1;
2448
2449 net->ipv4.sysctl_tcp_base_mss = TCP_BASE_MSS;
2450 net->ipv4.sysctl_tcp_probe_threshold = TCP_PROBE_THRESHOLD;
2451 net->ipv4.sysctl_tcp_probe_interval = TCP_PROBE_INTERVAL;
2452
2453 net->ipv4.sysctl_tcp_keepalive_time = TCP_KEEPALIVE_TIME;
2454 net->ipv4.sysctl_tcp_keepalive_probes = TCP_KEEPALIVE_PROBES;
2455 net->ipv4.sysctl_tcp_keepalive_intvl = TCP_KEEPALIVE_INTVL;
2456
2457 net->ipv4.sysctl_tcp_syn_retries = TCP_SYN_RETRIES;
2458 net->ipv4.sysctl_tcp_synack_retries = TCP_SYNACK_RETRIES;
2459 net->ipv4.sysctl_tcp_syncookies = 1;
2460 net->ipv4.sysctl_tcp_reordering = TCP_FASTRETRANS_THRESH;
2461 net->ipv4.sysctl_tcp_retries1 = TCP_RETR1;
2462 net->ipv4.sysctl_tcp_retries2 = TCP_RETR2;
2463 net->ipv4.sysctl_tcp_orphan_retries = 0;
2464 net->ipv4.sysctl_tcp_fin_timeout = TCP_FIN_TIMEOUT;
2465 net->ipv4.sysctl_tcp_notsent_lowat = UINT_MAX;
2466 net->ipv4.sysctl_tcp_tw_reuse = 0;
2467
2468 cnt = tcp_hashinfo.ehash_mask + 1;
2469 net->ipv4.tcp_death_row.sysctl_tw_recycle = 0;
2470 net->ipv4.tcp_death_row.sysctl_max_tw_buckets = (cnt + 1) / 2;
2471 net->ipv4.tcp_death_row.hashinfo = &tcp_hashinfo;
2472
2473 net->ipv4.sysctl_max_syn_backlog = max(128, cnt / 256);
2474
2475 return 0;
2476fail:
2477 tcp_sk_exit(net);
2478
2479 return res;
2480}
2481
2482static void __net_exit tcp_sk_exit_batch(struct list_head *net_exit_list)
2483{
2484 inet_twsk_purge(&tcp_hashinfo, AF_INET);
2485}
2486
2487static struct pernet_operations __net_initdata tcp_sk_ops = {
2488 .init = tcp_sk_init,
2489 .exit = tcp_sk_exit,
2490 .exit_batch = tcp_sk_exit_batch,
2491};
2492
2493void __init tcp_v4_init(void)
2494{
2495 if (register_pernet_subsys(&tcp_sk_ops))
2496 panic("Failed to create the TCP control socket.\n");
2497}
2498