Loading...
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
54#include <linux/bottom_half.h>
55#include <linux/types.h>
56#include <linux/fcntl.h>
57#include <linux/module.h>
58#include <linux/random.h>
59#include <linux/cache.h>
60#include <linux/jhash.h>
61#include <linux/init.h>
62#include <linux/times.h>
63#include <linux/slab.h>
64
65#include <net/net_namespace.h>
66#include <net/icmp.h>
67#include <net/inet_hashtables.h>
68#include <net/tcp.h>
69#include <net/transp_v6.h>
70#include <net/ipv6.h>
71#include <net/inet_common.h>
72#include <net/timewait_sock.h>
73#include <net/xfrm.h>
74#include <net/netdma.h>
75#include <net/secure_seq.h>
76
77#include <linux/inet.h>
78#include <linux/ipv6.h>
79#include <linux/stddef.h>
80#include <linux/proc_fs.h>
81#include <linux/seq_file.h>
82
83#include <linux/crypto.h>
84#include <linux/scatterlist.h>
85
86int sysctl_tcp_tw_reuse __read_mostly;
87int sysctl_tcp_low_latency __read_mostly;
88EXPORT_SYMBOL(sysctl_tcp_low_latency);
89
90
91#ifdef CONFIG_TCP_MD5SIG
92static struct tcp_md5sig_key *tcp_v4_md5_do_lookup(struct sock *sk,
93 __be32 addr);
94static int tcp_v4_md5_hash_hdr(char *md5_hash, struct tcp_md5sig_key *key,
95 __be32 daddr, __be32 saddr, struct tcphdr *th);
96#else
97static inline
98struct tcp_md5sig_key *tcp_v4_md5_do_lookup(struct sock *sk, __be32 addr)
99{
100 return NULL;
101}
102#endif
103
104struct inet_hashinfo tcp_hashinfo;
105EXPORT_SYMBOL(tcp_hashinfo);
106
107static inline __u32 tcp_v4_init_sequence(struct sk_buff *skb)
108{
109 return secure_tcp_sequence_number(ip_hdr(skb)->daddr,
110 ip_hdr(skb)->saddr,
111 tcp_hdr(skb)->dest,
112 tcp_hdr(skb)->source);
113}
114
115int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp)
116{
117 const struct tcp_timewait_sock *tcptw = tcp_twsk(sktw);
118 struct tcp_sock *tp = tcp_sk(sk);
119
120 /* With PAWS, it is safe from the viewpoint
121 of data integrity. Even without PAWS it is safe provided sequence
122 spaces do not overlap i.e. at data rates <= 80Mbit/sec.
123
124 Actually, the idea is close to VJ's one, only timestamp cache is
125 held not per host, but per port pair and TW bucket is used as state
126 holder.
127
128 If TW bucket has been already destroyed we fall back to VJ's scheme
129 and use initial timestamp retrieved from peer table.
130 */
131 if (tcptw->tw_ts_recent_stamp &&
132 (twp == NULL || (sysctl_tcp_tw_reuse &&
133 get_seconds() - tcptw->tw_ts_recent_stamp > 1))) {
134 tp->write_seq = tcptw->tw_snd_nxt + 65535 + 2;
135 if (tp->write_seq == 0)
136 tp->write_seq = 1;
137 tp->rx_opt.ts_recent = tcptw->tw_ts_recent;
138 tp->rx_opt.ts_recent_stamp = tcptw->tw_ts_recent_stamp;
139 sock_hold(sktw);
140 return 1;
141 }
142
143 return 0;
144}
145EXPORT_SYMBOL_GPL(tcp_twsk_unique);
146
147/* This will initiate an outgoing connection. */
148int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len)
149{
150 struct sockaddr_in *usin = (struct sockaddr_in *)uaddr;
151 struct inet_sock *inet = inet_sk(sk);
152 struct tcp_sock *tp = tcp_sk(sk);
153 __be16 orig_sport, orig_dport;
154 __be32 daddr, nexthop;
155 struct flowi4 *fl4;
156 struct rtable *rt;
157 int err;
158 struct ip_options_rcu *inet_opt;
159
160 if (addr_len < sizeof(struct sockaddr_in))
161 return -EINVAL;
162
163 if (usin->sin_family != AF_INET)
164 return -EAFNOSUPPORT;
165
166 nexthop = daddr = usin->sin_addr.s_addr;
167 inet_opt = rcu_dereference_protected(inet->inet_opt,
168 sock_owned_by_user(sk));
169 if (inet_opt && inet_opt->opt.srr) {
170 if (!daddr)
171 return -EINVAL;
172 nexthop = inet_opt->opt.faddr;
173 }
174
175 orig_sport = inet->inet_sport;
176 orig_dport = usin->sin_port;
177 fl4 = &inet->cork.fl.u.ip4;
178 rt = ip_route_connect(fl4, nexthop, inet->inet_saddr,
179 RT_CONN_FLAGS(sk), sk->sk_bound_dev_if,
180 IPPROTO_TCP,
181 orig_sport, orig_dport, sk, true);
182 if (IS_ERR(rt)) {
183 err = PTR_ERR(rt);
184 if (err == -ENETUNREACH)
185 IP_INC_STATS_BH(sock_net(sk), IPSTATS_MIB_OUTNOROUTES);
186 return err;
187 }
188
189 if (rt->rt_flags & (RTCF_MULTICAST | RTCF_BROADCAST)) {
190 ip_rt_put(rt);
191 return -ENETUNREACH;
192 }
193
194 if (!inet_opt || !inet_opt->opt.srr)
195 daddr = fl4->daddr;
196
197 if (!inet->inet_saddr)
198 inet->inet_saddr = fl4->saddr;
199 inet->inet_rcv_saddr = inet->inet_saddr;
200
201 if (tp->rx_opt.ts_recent_stamp && inet->inet_daddr != daddr) {
202 /* Reset inherited state */
203 tp->rx_opt.ts_recent = 0;
204 tp->rx_opt.ts_recent_stamp = 0;
205 tp->write_seq = 0;
206 }
207
208 if (tcp_death_row.sysctl_tw_recycle &&
209 !tp->rx_opt.ts_recent_stamp && fl4->daddr == daddr) {
210 struct inet_peer *peer = rt_get_peer(rt, fl4->daddr);
211 /*
212 * VJ's idea. We save last timestamp seen from
213 * the destination in peer table, when entering state
214 * TIME-WAIT * and initialize rx_opt.ts_recent from it,
215 * when trying new connection.
216 */
217 if (peer) {
218 inet_peer_refcheck(peer);
219 if ((u32)get_seconds() - peer->tcp_ts_stamp <= TCP_PAWS_MSL) {
220 tp->rx_opt.ts_recent_stamp = peer->tcp_ts_stamp;
221 tp->rx_opt.ts_recent = peer->tcp_ts;
222 }
223 }
224 }
225
226 inet->inet_dport = usin->sin_port;
227 inet->inet_daddr = daddr;
228
229 inet_csk(sk)->icsk_ext_hdr_len = 0;
230 if (inet_opt)
231 inet_csk(sk)->icsk_ext_hdr_len = inet_opt->opt.optlen;
232
233 tp->rx_opt.mss_clamp = TCP_MSS_DEFAULT;
234
235 /* Socket identity is still unknown (sport may be zero).
236 * However we set state to SYN-SENT and not releasing socket
237 * lock select source port, enter ourselves into the hash tables and
238 * complete initialization after this.
239 */
240 tcp_set_state(sk, TCP_SYN_SENT);
241 err = inet_hash_connect(&tcp_death_row, sk);
242 if (err)
243 goto failure;
244
245 rt = ip_route_newports(fl4, rt, orig_sport, orig_dport,
246 inet->inet_sport, inet->inet_dport, sk);
247 if (IS_ERR(rt)) {
248 err = PTR_ERR(rt);
249 rt = NULL;
250 goto failure;
251 }
252 /* OK, now commit destination to socket. */
253 sk->sk_gso_type = SKB_GSO_TCPV4;
254 sk_setup_caps(sk, &rt->dst);
255
256 if (!tp->write_seq)
257 tp->write_seq = secure_tcp_sequence_number(inet->inet_saddr,
258 inet->inet_daddr,
259 inet->inet_sport,
260 usin->sin_port);
261
262 inet->inet_id = tp->write_seq ^ jiffies;
263
264 err = tcp_connect(sk);
265 rt = NULL;
266 if (err)
267 goto failure;
268
269 return 0;
270
271failure:
272 /*
273 * This unhashes the socket and releases the local port,
274 * if necessary.
275 */
276 tcp_set_state(sk, TCP_CLOSE);
277 ip_rt_put(rt);
278 sk->sk_route_caps = 0;
279 inet->inet_dport = 0;
280 return err;
281}
282EXPORT_SYMBOL(tcp_v4_connect);
283
284/*
285 * This routine does path mtu discovery as defined in RFC1191.
286 */
287static void do_pmtu_discovery(struct sock *sk, const struct iphdr *iph, u32 mtu)
288{
289 struct dst_entry *dst;
290 struct inet_sock *inet = inet_sk(sk);
291
292 /* We are not interested in TCP_LISTEN and open_requests (SYN-ACKs
293 * send out by Linux are always <576bytes so they should go through
294 * unfragmented).
295 */
296 if (sk->sk_state == TCP_LISTEN)
297 return;
298
299 /* We don't check in the destentry if pmtu discovery is forbidden
300 * on this route. We just assume that no packet_to_big packets
301 * are send back when pmtu discovery is not active.
302 * There is a small race when the user changes this flag in the
303 * route, but I think that's acceptable.
304 */
305 if ((dst = __sk_dst_check(sk, 0)) == NULL)
306 return;
307
308 dst->ops->update_pmtu(dst, mtu);
309
310 /* Something is about to be wrong... Remember soft error
311 * for the case, if this connection will not able to recover.
312 */
313 if (mtu < dst_mtu(dst) && ip_dont_fragment(sk, dst))
314 sk->sk_err_soft = EMSGSIZE;
315
316 mtu = dst_mtu(dst);
317
318 if (inet->pmtudisc != IP_PMTUDISC_DONT &&
319 inet_csk(sk)->icsk_pmtu_cookie > mtu) {
320 tcp_sync_mss(sk, mtu);
321
322 /* Resend the TCP packet because it's
323 * clear that the old packet has been
324 * dropped. This is the new "fast" path mtu
325 * discovery.
326 */
327 tcp_simple_retransmit(sk);
328 } /* else let the usual retransmit timer handle it */
329}
330
331/*
332 * This routine is called by the ICMP module when it gets some
333 * sort of error condition. If err < 0 then the socket should
334 * be closed and the error returned to the user. If err > 0
335 * it's just the icmp type << 8 | icmp code. After adjustment
336 * header points to the first 8 bytes of the tcp header. We need
337 * to find the appropriate port.
338 *
339 * The locking strategy used here is very "optimistic". When
340 * someone else accesses the socket the ICMP is just dropped
341 * and for some paths there is no check at all.
342 * A more general error queue to queue errors for later handling
343 * is probably better.
344 *
345 */
346
347void tcp_v4_err(struct sk_buff *icmp_skb, u32 info)
348{
349 const struct iphdr *iph = (const struct iphdr *)icmp_skb->data;
350 struct tcphdr *th = (struct tcphdr *)(icmp_skb->data + (iph->ihl << 2));
351 struct inet_connection_sock *icsk;
352 struct tcp_sock *tp;
353 struct inet_sock *inet;
354 const int type = icmp_hdr(icmp_skb)->type;
355 const int code = icmp_hdr(icmp_skb)->code;
356 struct sock *sk;
357 struct sk_buff *skb;
358 __u32 seq;
359 __u32 remaining;
360 int err;
361 struct net *net = dev_net(icmp_skb->dev);
362
363 if (icmp_skb->len < (iph->ihl << 2) + 8) {
364 ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS);
365 return;
366 }
367
368 sk = inet_lookup(net, &tcp_hashinfo, iph->daddr, th->dest,
369 iph->saddr, th->source, inet_iif(icmp_skb));
370 if (!sk) {
371 ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS);
372 return;
373 }
374 if (sk->sk_state == TCP_TIME_WAIT) {
375 inet_twsk_put(inet_twsk(sk));
376 return;
377 }
378
379 bh_lock_sock(sk);
380 /* If too many ICMPs get dropped on busy
381 * servers this needs to be solved differently.
382 */
383 if (sock_owned_by_user(sk))
384 NET_INC_STATS_BH(net, LINUX_MIB_LOCKDROPPEDICMPS);
385
386 if (sk->sk_state == TCP_CLOSE)
387 goto out;
388
389 if (unlikely(iph->ttl < inet_sk(sk)->min_ttl)) {
390 NET_INC_STATS_BH(net, LINUX_MIB_TCPMINTTLDROP);
391 goto out;
392 }
393
394 icsk = inet_csk(sk);
395 tp = tcp_sk(sk);
396 seq = ntohl(th->seq);
397 if (sk->sk_state != TCP_LISTEN &&
398 !between(seq, tp->snd_una, tp->snd_nxt)) {
399 NET_INC_STATS_BH(net, LINUX_MIB_OUTOFWINDOWICMPS);
400 goto out;
401 }
402
403 switch (type) {
404 case ICMP_SOURCE_QUENCH:
405 /* Just silently ignore these. */
406 goto out;
407 case ICMP_PARAMETERPROB:
408 err = EPROTO;
409 break;
410 case ICMP_DEST_UNREACH:
411 if (code > NR_ICMP_UNREACH)
412 goto out;
413
414 if (code == ICMP_FRAG_NEEDED) { /* PMTU discovery (RFC1191) */
415 if (!sock_owned_by_user(sk))
416 do_pmtu_discovery(sk, iph, info);
417 goto out;
418 }
419
420 err = icmp_err_convert[code].errno;
421 /* check if icmp_skb allows revert of backoff
422 * (see draft-zimmermann-tcp-lcd) */
423 if (code != ICMP_NET_UNREACH && code != ICMP_HOST_UNREACH)
424 break;
425 if (seq != tp->snd_una || !icsk->icsk_retransmits ||
426 !icsk->icsk_backoff)
427 break;
428
429 if (sock_owned_by_user(sk))
430 break;
431
432 icsk->icsk_backoff--;
433 inet_csk(sk)->icsk_rto = (tp->srtt ? __tcp_set_rto(tp) :
434 TCP_TIMEOUT_INIT) << icsk->icsk_backoff;
435 tcp_bound_rto(sk);
436
437 skb = tcp_write_queue_head(sk);
438 BUG_ON(!skb);
439
440 remaining = icsk->icsk_rto - min(icsk->icsk_rto,
441 tcp_time_stamp - TCP_SKB_CB(skb)->when);
442
443 if (remaining) {
444 inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
445 remaining, TCP_RTO_MAX);
446 } else {
447 /* RTO revert clocked out retransmission.
448 * Will retransmit now */
449 tcp_retransmit_timer(sk);
450 }
451
452 break;
453 case ICMP_TIME_EXCEEDED:
454 err = EHOSTUNREACH;
455 break;
456 default:
457 goto out;
458 }
459
460 switch (sk->sk_state) {
461 struct request_sock *req, **prev;
462 case TCP_LISTEN:
463 if (sock_owned_by_user(sk))
464 goto out;
465
466 req = inet_csk_search_req(sk, &prev, th->dest,
467 iph->daddr, iph->saddr);
468 if (!req)
469 goto out;
470
471 /* ICMPs are not backlogged, hence we cannot get
472 an established socket here.
473 */
474 WARN_ON(req->sk);
475
476 if (seq != tcp_rsk(req)->snt_isn) {
477 NET_INC_STATS_BH(net, LINUX_MIB_OUTOFWINDOWICMPS);
478 goto out;
479 }
480
481 /*
482 * Still in SYN_RECV, just remove it silently.
483 * There is no good way to pass the error to the newly
484 * created socket, and POSIX does not want network
485 * errors returned from accept().
486 */
487 inet_csk_reqsk_queue_drop(sk, req, prev);
488 goto out;
489
490 case TCP_SYN_SENT:
491 case TCP_SYN_RECV: /* Cannot happen.
492 It can f.e. if SYNs crossed.
493 */
494 if (!sock_owned_by_user(sk)) {
495 sk->sk_err = err;
496
497 sk->sk_error_report(sk);
498
499 tcp_done(sk);
500 } else {
501 sk->sk_err_soft = err;
502 }
503 goto out;
504 }
505
506 /* If we've already connected we will keep trying
507 * until we time out, or the user gives up.
508 *
509 * rfc1122 4.2.3.9 allows to consider as hard errors
510 * only PROTO_UNREACH and PORT_UNREACH (well, FRAG_FAILED too,
511 * but it is obsoleted by pmtu discovery).
512 *
513 * Note, that in modern internet, where routing is unreliable
514 * and in each dark corner broken firewalls sit, sending random
515 * errors ordered by their masters even this two messages finally lose
516 * their original sense (even Linux sends invalid PORT_UNREACHs)
517 *
518 * Now we are in compliance with RFCs.
519 * --ANK (980905)
520 */
521
522 inet = inet_sk(sk);
523 if (!sock_owned_by_user(sk) && inet->recverr) {
524 sk->sk_err = err;
525 sk->sk_error_report(sk);
526 } else { /* Only an error on timeout */
527 sk->sk_err_soft = err;
528 }
529
530out:
531 bh_unlock_sock(sk);
532 sock_put(sk);
533}
534
535static void __tcp_v4_send_check(struct sk_buff *skb,
536 __be32 saddr, __be32 daddr)
537{
538 struct tcphdr *th = tcp_hdr(skb);
539
540 if (skb->ip_summed == CHECKSUM_PARTIAL) {
541 th->check = ~tcp_v4_check(skb->len, saddr, daddr, 0);
542 skb->csum_start = skb_transport_header(skb) - skb->head;
543 skb->csum_offset = offsetof(struct tcphdr, check);
544 } else {
545 th->check = tcp_v4_check(skb->len, saddr, daddr,
546 csum_partial(th,
547 th->doff << 2,
548 skb->csum));
549 }
550}
551
552/* This routine computes an IPv4 TCP checksum. */
553void tcp_v4_send_check(struct sock *sk, struct sk_buff *skb)
554{
555 struct inet_sock *inet = inet_sk(sk);
556
557 __tcp_v4_send_check(skb, inet->inet_saddr, inet->inet_daddr);
558}
559EXPORT_SYMBOL(tcp_v4_send_check);
560
561int tcp_v4_gso_send_check(struct sk_buff *skb)
562{
563 const struct iphdr *iph;
564 struct tcphdr *th;
565
566 if (!pskb_may_pull(skb, sizeof(*th)))
567 return -EINVAL;
568
569 iph = ip_hdr(skb);
570 th = tcp_hdr(skb);
571
572 th->check = 0;
573 skb->ip_summed = CHECKSUM_PARTIAL;
574 __tcp_v4_send_check(skb, iph->saddr, iph->daddr);
575 return 0;
576}
577
578/*
579 * This routine will send an RST to the other tcp.
580 *
581 * Someone asks: why I NEVER use socket parameters (TOS, TTL etc.)
582 * for reset.
583 * Answer: if a packet caused RST, it is not for a socket
584 * existing in our system, if it is matched to a socket,
585 * it is just duplicate segment or bug in other side's TCP.
586 * So that we build reply only basing on parameters
587 * arrived with segment.
588 * Exception: precedence violation. We do not implement it in any case.
589 */
590
591static void tcp_v4_send_reset(struct sock *sk, struct sk_buff *skb)
592{
593 struct tcphdr *th = tcp_hdr(skb);
594 struct {
595 struct tcphdr th;
596#ifdef CONFIG_TCP_MD5SIG
597 __be32 opt[(TCPOLEN_MD5SIG_ALIGNED >> 2)];
598#endif
599 } rep;
600 struct ip_reply_arg arg;
601#ifdef CONFIG_TCP_MD5SIG
602 struct tcp_md5sig_key *key;
603#endif
604 struct net *net;
605
606 /* Never send a reset in response to a reset. */
607 if (th->rst)
608 return;
609
610 if (skb_rtable(skb)->rt_type != RTN_LOCAL)
611 return;
612
613 /* Swap the send and the receive. */
614 memset(&rep, 0, sizeof(rep));
615 rep.th.dest = th->source;
616 rep.th.source = th->dest;
617 rep.th.doff = sizeof(struct tcphdr) / 4;
618 rep.th.rst = 1;
619
620 if (th->ack) {
621 rep.th.seq = th->ack_seq;
622 } else {
623 rep.th.ack = 1;
624 rep.th.ack_seq = htonl(ntohl(th->seq) + th->syn + th->fin +
625 skb->len - (th->doff << 2));
626 }
627
628 memset(&arg, 0, sizeof(arg));
629 arg.iov[0].iov_base = (unsigned char *)&rep;
630 arg.iov[0].iov_len = sizeof(rep.th);
631
632#ifdef CONFIG_TCP_MD5SIG
633 key = sk ? tcp_v4_md5_do_lookup(sk, ip_hdr(skb)->daddr) : NULL;
634 if (key) {
635 rep.opt[0] = htonl((TCPOPT_NOP << 24) |
636 (TCPOPT_NOP << 16) |
637 (TCPOPT_MD5SIG << 8) |
638 TCPOLEN_MD5SIG);
639 /* Update length and the length the header thinks exists */
640 arg.iov[0].iov_len += TCPOLEN_MD5SIG_ALIGNED;
641 rep.th.doff = arg.iov[0].iov_len / 4;
642
643 tcp_v4_md5_hash_hdr((__u8 *) &rep.opt[1],
644 key, ip_hdr(skb)->saddr,
645 ip_hdr(skb)->daddr, &rep.th);
646 }
647#endif
648 arg.csum = csum_tcpudp_nofold(ip_hdr(skb)->daddr,
649 ip_hdr(skb)->saddr, /* XXX */
650 arg.iov[0].iov_len, IPPROTO_TCP, 0);
651 arg.csumoffset = offsetof(struct tcphdr, check) / 2;
652 arg.flags = (sk && inet_sk(sk)->transparent) ? IP_REPLY_ARG_NOSRCCHECK : 0;
653
654 net = dev_net(skb_dst(skb)->dev);
655 ip_send_reply(net->ipv4.tcp_sock, skb, ip_hdr(skb)->saddr,
656 &arg, arg.iov[0].iov_len);
657
658 TCP_INC_STATS_BH(net, TCP_MIB_OUTSEGS);
659 TCP_INC_STATS_BH(net, TCP_MIB_OUTRSTS);
660}
661
662/* The code following below sending ACKs in SYN-RECV and TIME-WAIT states
663 outside socket context is ugly, certainly. What can I do?
664 */
665
666static void tcp_v4_send_ack(struct sk_buff *skb, u32 seq, u32 ack,
667 u32 win, u32 ts, int oif,
668 struct tcp_md5sig_key *key,
669 int reply_flags)
670{
671 struct tcphdr *th = tcp_hdr(skb);
672 struct {
673 struct tcphdr th;
674 __be32 opt[(TCPOLEN_TSTAMP_ALIGNED >> 2)
675#ifdef CONFIG_TCP_MD5SIG
676 + (TCPOLEN_MD5SIG_ALIGNED >> 2)
677#endif
678 ];
679 } rep;
680 struct ip_reply_arg arg;
681 struct net *net = dev_net(skb_dst(skb)->dev);
682
683 memset(&rep.th, 0, sizeof(struct tcphdr));
684 memset(&arg, 0, sizeof(arg));
685
686 arg.iov[0].iov_base = (unsigned char *)&rep;
687 arg.iov[0].iov_len = sizeof(rep.th);
688 if (ts) {
689 rep.opt[0] = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
690 (TCPOPT_TIMESTAMP << 8) |
691 TCPOLEN_TIMESTAMP);
692 rep.opt[1] = htonl(tcp_time_stamp);
693 rep.opt[2] = htonl(ts);
694 arg.iov[0].iov_len += TCPOLEN_TSTAMP_ALIGNED;
695 }
696
697 /* Swap the send and the receive. */
698 rep.th.dest = th->source;
699 rep.th.source = th->dest;
700 rep.th.doff = arg.iov[0].iov_len / 4;
701 rep.th.seq = htonl(seq);
702 rep.th.ack_seq = htonl(ack);
703 rep.th.ack = 1;
704 rep.th.window = htons(win);
705
706#ifdef CONFIG_TCP_MD5SIG
707 if (key) {
708 int offset = (ts) ? 3 : 0;
709
710 rep.opt[offset++] = htonl((TCPOPT_NOP << 24) |
711 (TCPOPT_NOP << 16) |
712 (TCPOPT_MD5SIG << 8) |
713 TCPOLEN_MD5SIG);
714 arg.iov[0].iov_len += TCPOLEN_MD5SIG_ALIGNED;
715 rep.th.doff = arg.iov[0].iov_len/4;
716
717 tcp_v4_md5_hash_hdr((__u8 *) &rep.opt[offset],
718 key, ip_hdr(skb)->saddr,
719 ip_hdr(skb)->daddr, &rep.th);
720 }
721#endif
722 arg.flags = reply_flags;
723 arg.csum = csum_tcpudp_nofold(ip_hdr(skb)->daddr,
724 ip_hdr(skb)->saddr, /* XXX */
725 arg.iov[0].iov_len, IPPROTO_TCP, 0);
726 arg.csumoffset = offsetof(struct tcphdr, check) / 2;
727 if (oif)
728 arg.bound_dev_if = oif;
729
730 ip_send_reply(net->ipv4.tcp_sock, skb, ip_hdr(skb)->saddr,
731 &arg, arg.iov[0].iov_len);
732
733 TCP_INC_STATS_BH(net, TCP_MIB_OUTSEGS);
734}
735
736static void tcp_v4_timewait_ack(struct sock *sk, struct sk_buff *skb)
737{
738 struct inet_timewait_sock *tw = inet_twsk(sk);
739 struct tcp_timewait_sock *tcptw = tcp_twsk(sk);
740
741 tcp_v4_send_ack(skb, tcptw->tw_snd_nxt, tcptw->tw_rcv_nxt,
742 tcptw->tw_rcv_wnd >> tw->tw_rcv_wscale,
743 tcptw->tw_ts_recent,
744 tw->tw_bound_dev_if,
745 tcp_twsk_md5_key(tcptw),
746 tw->tw_transparent ? IP_REPLY_ARG_NOSRCCHECK : 0
747 );
748
749 inet_twsk_put(tw);
750}
751
752static void tcp_v4_reqsk_send_ack(struct sock *sk, struct sk_buff *skb,
753 struct request_sock *req)
754{
755 tcp_v4_send_ack(skb, tcp_rsk(req)->snt_isn + 1,
756 tcp_rsk(req)->rcv_isn + 1, req->rcv_wnd,
757 req->ts_recent,
758 0,
759 tcp_v4_md5_do_lookup(sk, ip_hdr(skb)->daddr),
760 inet_rsk(req)->no_srccheck ? IP_REPLY_ARG_NOSRCCHECK : 0);
761}
762
763/*
764 * Send a SYN-ACK after having received a SYN.
765 * This still operates on a request_sock only, not on a big
766 * socket.
767 */
768static int tcp_v4_send_synack(struct sock *sk, struct dst_entry *dst,
769 struct request_sock *req,
770 struct request_values *rvp)
771{
772 const struct inet_request_sock *ireq = inet_rsk(req);
773 struct flowi4 fl4;
774 int err = -1;
775 struct sk_buff * skb;
776
777 /* First, grab a route. */
778 if (!dst && (dst = inet_csk_route_req(sk, &fl4, req)) == NULL)
779 return -1;
780
781 skb = tcp_make_synack(sk, dst, req, rvp);
782
783 if (skb) {
784 __tcp_v4_send_check(skb, ireq->loc_addr, ireq->rmt_addr);
785
786 err = ip_build_and_send_pkt(skb, sk, ireq->loc_addr,
787 ireq->rmt_addr,
788 ireq->opt);
789 err = net_xmit_eval(err);
790 }
791
792 dst_release(dst);
793 return err;
794}
795
796static int tcp_v4_rtx_synack(struct sock *sk, struct request_sock *req,
797 struct request_values *rvp)
798{
799 TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_RETRANSSEGS);
800 return tcp_v4_send_synack(sk, NULL, req, rvp);
801}
802
803/*
804 * IPv4 request_sock destructor.
805 */
806static void tcp_v4_reqsk_destructor(struct request_sock *req)
807{
808 kfree(inet_rsk(req)->opt);
809}
810
811/*
812 * Return 1 if a syncookie should be sent
813 */
814int tcp_syn_flood_action(struct sock *sk,
815 const struct sk_buff *skb,
816 const char *proto)
817{
818 const char *msg = "Dropping request";
819 int want_cookie = 0;
820 struct listen_sock *lopt;
821
822
823
824#ifdef CONFIG_SYN_COOKIES
825 if (sysctl_tcp_syncookies) {
826 msg = "Sending cookies";
827 want_cookie = 1;
828 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPREQQFULLDOCOOKIES);
829 } else
830#endif
831 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPREQQFULLDROP);
832
833 lopt = inet_csk(sk)->icsk_accept_queue.listen_opt;
834 if (!lopt->synflood_warned) {
835 lopt->synflood_warned = 1;
836 pr_info("%s: Possible SYN flooding on port %d. %s. "
837 " Check SNMP counters.\n",
838 proto, ntohs(tcp_hdr(skb)->dest), msg);
839 }
840 return want_cookie;
841}
842EXPORT_SYMBOL(tcp_syn_flood_action);
843
844/*
845 * Save and compile IPv4 options into the request_sock if needed.
846 */
847static struct ip_options_rcu *tcp_v4_save_options(struct sock *sk,
848 struct sk_buff *skb)
849{
850 const struct ip_options *opt = &(IPCB(skb)->opt);
851 struct ip_options_rcu *dopt = NULL;
852
853 if (opt && opt->optlen) {
854 int opt_size = sizeof(*dopt) + opt->optlen;
855
856 dopt = kmalloc(opt_size, GFP_ATOMIC);
857 if (dopt) {
858 if (ip_options_echo(&dopt->opt, skb)) {
859 kfree(dopt);
860 dopt = NULL;
861 }
862 }
863 }
864 return dopt;
865}
866
867#ifdef CONFIG_TCP_MD5SIG
868/*
869 * RFC2385 MD5 checksumming requires a mapping of
870 * IP address->MD5 Key.
871 * We need to maintain these in the sk structure.
872 */
873
874/* Find the Key structure for an address. */
875static struct tcp_md5sig_key *
876 tcp_v4_md5_do_lookup(struct sock *sk, __be32 addr)
877{
878 struct tcp_sock *tp = tcp_sk(sk);
879 int i;
880
881 if (!tp->md5sig_info || !tp->md5sig_info->entries4)
882 return NULL;
883 for (i = 0; i < tp->md5sig_info->entries4; i++) {
884 if (tp->md5sig_info->keys4[i].addr == addr)
885 return &tp->md5sig_info->keys4[i].base;
886 }
887 return NULL;
888}
889
890struct tcp_md5sig_key *tcp_v4_md5_lookup(struct sock *sk,
891 struct sock *addr_sk)
892{
893 return tcp_v4_md5_do_lookup(sk, inet_sk(addr_sk)->inet_daddr);
894}
895EXPORT_SYMBOL(tcp_v4_md5_lookup);
896
897static struct tcp_md5sig_key *tcp_v4_reqsk_md5_lookup(struct sock *sk,
898 struct request_sock *req)
899{
900 return tcp_v4_md5_do_lookup(sk, inet_rsk(req)->rmt_addr);
901}
902
903/* This can be called on a newly created socket, from other files */
904int tcp_v4_md5_do_add(struct sock *sk, __be32 addr,
905 u8 *newkey, u8 newkeylen)
906{
907 /* Add Key to the list */
908 struct tcp_md5sig_key *key;
909 struct tcp_sock *tp = tcp_sk(sk);
910 struct tcp4_md5sig_key *keys;
911
912 key = tcp_v4_md5_do_lookup(sk, addr);
913 if (key) {
914 /* Pre-existing entry - just update that one. */
915 kfree(key->key);
916 key->key = newkey;
917 key->keylen = newkeylen;
918 } else {
919 struct tcp_md5sig_info *md5sig;
920
921 if (!tp->md5sig_info) {
922 tp->md5sig_info = kzalloc(sizeof(*tp->md5sig_info),
923 GFP_ATOMIC);
924 if (!tp->md5sig_info) {
925 kfree(newkey);
926 return -ENOMEM;
927 }
928 sk_nocaps_add(sk, NETIF_F_GSO_MASK);
929 }
930
931 md5sig = tp->md5sig_info;
932 if (md5sig->entries4 == 0 &&
933 tcp_alloc_md5sig_pool(sk) == NULL) {
934 kfree(newkey);
935 return -ENOMEM;
936 }
937
938 if (md5sig->alloced4 == md5sig->entries4) {
939 keys = kmalloc((sizeof(*keys) *
940 (md5sig->entries4 + 1)), GFP_ATOMIC);
941 if (!keys) {
942 kfree(newkey);
943 if (md5sig->entries4 == 0)
944 tcp_free_md5sig_pool();
945 return -ENOMEM;
946 }
947
948 if (md5sig->entries4)
949 memcpy(keys, md5sig->keys4,
950 sizeof(*keys) * md5sig->entries4);
951
952 /* Free old key list, and reference new one */
953 kfree(md5sig->keys4);
954 md5sig->keys4 = keys;
955 md5sig->alloced4++;
956 }
957 md5sig->entries4++;
958 md5sig->keys4[md5sig->entries4 - 1].addr = addr;
959 md5sig->keys4[md5sig->entries4 - 1].base.key = newkey;
960 md5sig->keys4[md5sig->entries4 - 1].base.keylen = newkeylen;
961 }
962 return 0;
963}
964EXPORT_SYMBOL(tcp_v4_md5_do_add);
965
966static int tcp_v4_md5_add_func(struct sock *sk, struct sock *addr_sk,
967 u8 *newkey, u8 newkeylen)
968{
969 return tcp_v4_md5_do_add(sk, inet_sk(addr_sk)->inet_daddr,
970 newkey, newkeylen);
971}
972
973int tcp_v4_md5_do_del(struct sock *sk, __be32 addr)
974{
975 struct tcp_sock *tp = tcp_sk(sk);
976 int i;
977
978 for (i = 0; i < tp->md5sig_info->entries4; i++) {
979 if (tp->md5sig_info->keys4[i].addr == addr) {
980 /* Free the key */
981 kfree(tp->md5sig_info->keys4[i].base.key);
982 tp->md5sig_info->entries4--;
983
984 if (tp->md5sig_info->entries4 == 0) {
985 kfree(tp->md5sig_info->keys4);
986 tp->md5sig_info->keys4 = NULL;
987 tp->md5sig_info->alloced4 = 0;
988 tcp_free_md5sig_pool();
989 } else if (tp->md5sig_info->entries4 != i) {
990 /* Need to do some manipulation */
991 memmove(&tp->md5sig_info->keys4[i],
992 &tp->md5sig_info->keys4[i+1],
993 (tp->md5sig_info->entries4 - i) *
994 sizeof(struct tcp4_md5sig_key));
995 }
996 return 0;
997 }
998 }
999 return -ENOENT;
1000}
1001EXPORT_SYMBOL(tcp_v4_md5_do_del);
1002
1003static void tcp_v4_clear_md5_list(struct sock *sk)
1004{
1005 struct tcp_sock *tp = tcp_sk(sk);
1006
1007 /* Free each key, then the set of key keys,
1008 * the crypto element, and then decrement our
1009 * hold on the last resort crypto.
1010 */
1011 if (tp->md5sig_info->entries4) {
1012 int i;
1013 for (i = 0; i < tp->md5sig_info->entries4; i++)
1014 kfree(tp->md5sig_info->keys4[i].base.key);
1015 tp->md5sig_info->entries4 = 0;
1016 tcp_free_md5sig_pool();
1017 }
1018 if (tp->md5sig_info->keys4) {
1019 kfree(tp->md5sig_info->keys4);
1020 tp->md5sig_info->keys4 = NULL;
1021 tp->md5sig_info->alloced4 = 0;
1022 }
1023}
1024
1025static int tcp_v4_parse_md5_keys(struct sock *sk, char __user *optval,
1026 int optlen)
1027{
1028 struct tcp_md5sig cmd;
1029 struct sockaddr_in *sin = (struct sockaddr_in *)&cmd.tcpm_addr;
1030 u8 *newkey;
1031
1032 if (optlen < sizeof(cmd))
1033 return -EINVAL;
1034
1035 if (copy_from_user(&cmd, optval, sizeof(cmd)))
1036 return -EFAULT;
1037
1038 if (sin->sin_family != AF_INET)
1039 return -EINVAL;
1040
1041 if (!cmd.tcpm_key || !cmd.tcpm_keylen) {
1042 if (!tcp_sk(sk)->md5sig_info)
1043 return -ENOENT;
1044 return tcp_v4_md5_do_del(sk, sin->sin_addr.s_addr);
1045 }
1046
1047 if (cmd.tcpm_keylen > TCP_MD5SIG_MAXKEYLEN)
1048 return -EINVAL;
1049
1050 if (!tcp_sk(sk)->md5sig_info) {
1051 struct tcp_sock *tp = tcp_sk(sk);
1052 struct tcp_md5sig_info *p;
1053
1054 p = kzalloc(sizeof(*p), sk->sk_allocation);
1055 if (!p)
1056 return -EINVAL;
1057
1058 tp->md5sig_info = p;
1059 sk_nocaps_add(sk, NETIF_F_GSO_MASK);
1060 }
1061
1062 newkey = kmemdup(cmd.tcpm_key, cmd.tcpm_keylen, sk->sk_allocation);
1063 if (!newkey)
1064 return -ENOMEM;
1065 return tcp_v4_md5_do_add(sk, sin->sin_addr.s_addr,
1066 newkey, cmd.tcpm_keylen);
1067}
1068
1069static int tcp_v4_md5_hash_pseudoheader(struct tcp_md5sig_pool *hp,
1070 __be32 daddr, __be32 saddr, int nbytes)
1071{
1072 struct tcp4_pseudohdr *bp;
1073 struct scatterlist sg;
1074
1075 bp = &hp->md5_blk.ip4;
1076
1077 /*
1078 * 1. the TCP pseudo-header (in the order: source IP address,
1079 * destination IP address, zero-padded protocol number, and
1080 * segment length)
1081 */
1082 bp->saddr = saddr;
1083 bp->daddr = daddr;
1084 bp->pad = 0;
1085 bp->protocol = IPPROTO_TCP;
1086 bp->len = cpu_to_be16(nbytes);
1087
1088 sg_init_one(&sg, bp, sizeof(*bp));
1089 return crypto_hash_update(&hp->md5_desc, &sg, sizeof(*bp));
1090}
1091
1092static int tcp_v4_md5_hash_hdr(char *md5_hash, struct tcp_md5sig_key *key,
1093 __be32 daddr, __be32 saddr, struct tcphdr *th)
1094{
1095 struct tcp_md5sig_pool *hp;
1096 struct hash_desc *desc;
1097
1098 hp = tcp_get_md5sig_pool();
1099 if (!hp)
1100 goto clear_hash_noput;
1101 desc = &hp->md5_desc;
1102
1103 if (crypto_hash_init(desc))
1104 goto clear_hash;
1105 if (tcp_v4_md5_hash_pseudoheader(hp, daddr, saddr, th->doff << 2))
1106 goto clear_hash;
1107 if (tcp_md5_hash_header(hp, th))
1108 goto clear_hash;
1109 if (tcp_md5_hash_key(hp, key))
1110 goto clear_hash;
1111 if (crypto_hash_final(desc, md5_hash))
1112 goto clear_hash;
1113
1114 tcp_put_md5sig_pool();
1115 return 0;
1116
1117clear_hash:
1118 tcp_put_md5sig_pool();
1119clear_hash_noput:
1120 memset(md5_hash, 0, 16);
1121 return 1;
1122}
1123
1124int tcp_v4_md5_hash_skb(char *md5_hash, struct tcp_md5sig_key *key,
1125 struct sock *sk, struct request_sock *req,
1126 struct sk_buff *skb)
1127{
1128 struct tcp_md5sig_pool *hp;
1129 struct hash_desc *desc;
1130 struct tcphdr *th = tcp_hdr(skb);
1131 __be32 saddr, daddr;
1132
1133 if (sk) {
1134 saddr = inet_sk(sk)->inet_saddr;
1135 daddr = inet_sk(sk)->inet_daddr;
1136 } else if (req) {
1137 saddr = inet_rsk(req)->loc_addr;
1138 daddr = inet_rsk(req)->rmt_addr;
1139 } else {
1140 const struct iphdr *iph = ip_hdr(skb);
1141 saddr = iph->saddr;
1142 daddr = iph->daddr;
1143 }
1144
1145 hp = tcp_get_md5sig_pool();
1146 if (!hp)
1147 goto clear_hash_noput;
1148 desc = &hp->md5_desc;
1149
1150 if (crypto_hash_init(desc))
1151 goto clear_hash;
1152
1153 if (tcp_v4_md5_hash_pseudoheader(hp, daddr, saddr, skb->len))
1154 goto clear_hash;
1155 if (tcp_md5_hash_header(hp, th))
1156 goto clear_hash;
1157 if (tcp_md5_hash_skb_data(hp, skb, th->doff << 2))
1158 goto clear_hash;
1159 if (tcp_md5_hash_key(hp, key))
1160 goto clear_hash;
1161 if (crypto_hash_final(desc, md5_hash))
1162 goto clear_hash;
1163
1164 tcp_put_md5sig_pool();
1165 return 0;
1166
1167clear_hash:
1168 tcp_put_md5sig_pool();
1169clear_hash_noput:
1170 memset(md5_hash, 0, 16);
1171 return 1;
1172}
1173EXPORT_SYMBOL(tcp_v4_md5_hash_skb);
1174
1175static int tcp_v4_inbound_md5_hash(struct sock *sk, struct sk_buff *skb)
1176{
1177 /*
1178 * This gets called for each TCP segment that arrives
1179 * so we want to be efficient.
1180 * We have 3 drop cases:
1181 * o No MD5 hash and one expected.
1182 * o MD5 hash and we're not expecting one.
1183 * o MD5 hash and its wrong.
1184 */
1185 __u8 *hash_location = NULL;
1186 struct tcp_md5sig_key *hash_expected;
1187 const struct iphdr *iph = ip_hdr(skb);
1188 struct tcphdr *th = tcp_hdr(skb);
1189 int genhash;
1190 unsigned char newhash[16];
1191
1192 hash_expected = tcp_v4_md5_do_lookup(sk, iph->saddr);
1193 hash_location = tcp_parse_md5sig_option(th);
1194
1195 /* We've parsed the options - do we have a hash? */
1196 if (!hash_expected && !hash_location)
1197 return 0;
1198
1199 if (hash_expected && !hash_location) {
1200 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPMD5NOTFOUND);
1201 return 1;
1202 }
1203
1204 if (!hash_expected && hash_location) {
1205 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPMD5UNEXPECTED);
1206 return 1;
1207 }
1208
1209 /* Okay, so this is hash_expected and hash_location -
1210 * so we need to calculate the checksum.
1211 */
1212 genhash = tcp_v4_md5_hash_skb(newhash,
1213 hash_expected,
1214 NULL, NULL, skb);
1215
1216 if (genhash || memcmp(hash_location, newhash, 16) != 0) {
1217 if (net_ratelimit()) {
1218 printk(KERN_INFO "MD5 Hash failed for (%pI4, %d)->(%pI4, %d)%s\n",
1219 &iph->saddr, ntohs(th->source),
1220 &iph->daddr, ntohs(th->dest),
1221 genhash ? " tcp_v4_calc_md5_hash failed" : "");
1222 }
1223 return 1;
1224 }
1225 return 0;
1226}
1227
1228#endif
1229
1230struct request_sock_ops tcp_request_sock_ops __read_mostly = {
1231 .family = PF_INET,
1232 .obj_size = sizeof(struct tcp_request_sock),
1233 .rtx_syn_ack = tcp_v4_rtx_synack,
1234 .send_ack = tcp_v4_reqsk_send_ack,
1235 .destructor = tcp_v4_reqsk_destructor,
1236 .send_reset = tcp_v4_send_reset,
1237 .syn_ack_timeout = tcp_syn_ack_timeout,
1238};
1239
1240#ifdef CONFIG_TCP_MD5SIG
1241static const struct tcp_request_sock_ops tcp_request_sock_ipv4_ops = {
1242 .md5_lookup = tcp_v4_reqsk_md5_lookup,
1243 .calc_md5_hash = tcp_v4_md5_hash_skb,
1244};
1245#endif
1246
1247int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb)
1248{
1249 struct tcp_extend_values tmp_ext;
1250 struct tcp_options_received tmp_opt;
1251 u8 *hash_location;
1252 struct request_sock *req;
1253 struct inet_request_sock *ireq;
1254 struct tcp_sock *tp = tcp_sk(sk);
1255 struct dst_entry *dst = NULL;
1256 __be32 saddr = ip_hdr(skb)->saddr;
1257 __be32 daddr = ip_hdr(skb)->daddr;
1258 __u32 isn = TCP_SKB_CB(skb)->when;
1259 int want_cookie = 0;
1260
1261 /* Never answer to SYNs send to broadcast or multicast */
1262 if (skb_rtable(skb)->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))
1263 goto drop;
1264
1265 /* TW buckets are converted to open requests without
1266 * limitations, they conserve resources and peer is
1267 * evidently real one.
1268 */
1269 if (inet_csk_reqsk_queue_is_full(sk) && !isn) {
1270 want_cookie = tcp_syn_flood_action(sk, skb, "TCP");
1271 if (!want_cookie)
1272 goto drop;
1273 }
1274
1275 /* Accept backlog is full. If we have already queued enough
1276 * of warm entries in syn queue, drop request. It is better than
1277 * clogging syn queue with openreqs with exponentially increasing
1278 * timeout.
1279 */
1280 if (sk_acceptq_is_full(sk) && inet_csk_reqsk_queue_young(sk) > 1)
1281 goto drop;
1282
1283 req = inet_reqsk_alloc(&tcp_request_sock_ops);
1284 if (!req)
1285 goto drop;
1286
1287#ifdef CONFIG_TCP_MD5SIG
1288 tcp_rsk(req)->af_specific = &tcp_request_sock_ipv4_ops;
1289#endif
1290
1291 tcp_clear_options(&tmp_opt);
1292 tmp_opt.mss_clamp = TCP_MSS_DEFAULT;
1293 tmp_opt.user_mss = tp->rx_opt.user_mss;
1294 tcp_parse_options(skb, &tmp_opt, &hash_location, 0);
1295
1296 if (tmp_opt.cookie_plus > 0 &&
1297 tmp_opt.saw_tstamp &&
1298 !tp->rx_opt.cookie_out_never &&
1299 (sysctl_tcp_cookie_size > 0 ||
1300 (tp->cookie_values != NULL &&
1301 tp->cookie_values->cookie_desired > 0))) {
1302 u8 *c;
1303 u32 *mess = &tmp_ext.cookie_bakery[COOKIE_DIGEST_WORDS];
1304 int l = tmp_opt.cookie_plus - TCPOLEN_COOKIE_BASE;
1305
1306 if (tcp_cookie_generator(&tmp_ext.cookie_bakery[0]) != 0)
1307 goto drop_and_release;
1308
1309 /* Secret recipe starts with IP addresses */
1310 *mess++ ^= (__force u32)daddr;
1311 *mess++ ^= (__force u32)saddr;
1312
1313 /* plus variable length Initiator Cookie */
1314 c = (u8 *)mess;
1315 while (l-- > 0)
1316 *c++ ^= *hash_location++;
1317
1318 want_cookie = 0; /* not our kind of cookie */
1319 tmp_ext.cookie_out_never = 0; /* false */
1320 tmp_ext.cookie_plus = tmp_opt.cookie_plus;
1321 } else if (!tp->rx_opt.cookie_in_always) {
1322 /* redundant indications, but ensure initialization. */
1323 tmp_ext.cookie_out_never = 1; /* true */
1324 tmp_ext.cookie_plus = 0;
1325 } else {
1326 goto drop_and_release;
1327 }
1328 tmp_ext.cookie_in_always = tp->rx_opt.cookie_in_always;
1329
1330 if (want_cookie && !tmp_opt.saw_tstamp)
1331 tcp_clear_options(&tmp_opt);
1332
1333 tmp_opt.tstamp_ok = tmp_opt.saw_tstamp;
1334 tcp_openreq_init(req, &tmp_opt, skb);
1335
1336 ireq = inet_rsk(req);
1337 ireq->loc_addr = daddr;
1338 ireq->rmt_addr = saddr;
1339 ireq->no_srccheck = inet_sk(sk)->transparent;
1340 ireq->opt = tcp_v4_save_options(sk, skb);
1341
1342 if (security_inet_conn_request(sk, skb, req))
1343 goto drop_and_free;
1344
1345 if (!want_cookie || tmp_opt.tstamp_ok)
1346 TCP_ECN_create_request(req, tcp_hdr(skb));
1347
1348 if (want_cookie) {
1349 isn = cookie_v4_init_sequence(sk, skb, &req->mss);
1350 req->cookie_ts = tmp_opt.tstamp_ok;
1351 } else if (!isn) {
1352 struct inet_peer *peer = NULL;
1353 struct flowi4 fl4;
1354
1355 /* VJ's idea. We save last timestamp seen
1356 * from the destination in peer table, when entering
1357 * state TIME-WAIT, and check against it before
1358 * accepting new connection request.
1359 *
1360 * If "isn" is not zero, this request hit alive
1361 * timewait bucket, so that all the necessary checks
1362 * are made in the function processing timewait state.
1363 */
1364 if (tmp_opt.saw_tstamp &&
1365 tcp_death_row.sysctl_tw_recycle &&
1366 (dst = inet_csk_route_req(sk, &fl4, req)) != NULL &&
1367 fl4.daddr == saddr &&
1368 (peer = rt_get_peer((struct rtable *)dst, fl4.daddr)) != NULL) {
1369 inet_peer_refcheck(peer);
1370 if ((u32)get_seconds() - peer->tcp_ts_stamp < TCP_PAWS_MSL &&
1371 (s32)(peer->tcp_ts - req->ts_recent) >
1372 TCP_PAWS_WINDOW) {
1373 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_PAWSPASSIVEREJECTED);
1374 goto drop_and_release;
1375 }
1376 }
1377 /* Kill the following clause, if you dislike this way. */
1378 else if (!sysctl_tcp_syncookies &&
1379 (sysctl_max_syn_backlog - inet_csk_reqsk_queue_len(sk) <
1380 (sysctl_max_syn_backlog >> 2)) &&
1381 (!peer || !peer->tcp_ts_stamp) &&
1382 (!dst || !dst_metric(dst, RTAX_RTT))) {
1383 /* Without syncookies last quarter of
1384 * backlog is filled with destinations,
1385 * proven to be alive.
1386 * It means that we continue to communicate
1387 * to destinations, already remembered
1388 * to the moment of synflood.
1389 */
1390 LIMIT_NETDEBUG(KERN_DEBUG "TCP: drop open request from %pI4/%u\n",
1391 &saddr, ntohs(tcp_hdr(skb)->source));
1392 goto drop_and_release;
1393 }
1394
1395 isn = tcp_v4_init_sequence(skb);
1396 }
1397 tcp_rsk(req)->snt_isn = isn;
1398 tcp_rsk(req)->snt_synack = tcp_time_stamp;
1399
1400 if (tcp_v4_send_synack(sk, dst, req,
1401 (struct request_values *)&tmp_ext) ||
1402 want_cookie)
1403 goto drop_and_free;
1404
1405 inet_csk_reqsk_queue_hash_add(sk, req, TCP_TIMEOUT_INIT);
1406 return 0;
1407
1408drop_and_release:
1409 dst_release(dst);
1410drop_and_free:
1411 reqsk_free(req);
1412drop:
1413 return 0;
1414}
1415EXPORT_SYMBOL(tcp_v4_conn_request);
1416
1417
1418/*
1419 * The three way handshake has completed - we got a valid synack -
1420 * now create the new socket.
1421 */
1422struct sock *tcp_v4_syn_recv_sock(struct sock *sk, struct sk_buff *skb,
1423 struct request_sock *req,
1424 struct dst_entry *dst)
1425{
1426 struct inet_request_sock *ireq;
1427 struct inet_sock *newinet;
1428 struct tcp_sock *newtp;
1429 struct sock *newsk;
1430#ifdef CONFIG_TCP_MD5SIG
1431 struct tcp_md5sig_key *key;
1432#endif
1433 struct ip_options_rcu *inet_opt;
1434
1435 if (sk_acceptq_is_full(sk))
1436 goto exit_overflow;
1437
1438 newsk = tcp_create_openreq_child(sk, req, skb);
1439 if (!newsk)
1440 goto exit_nonewsk;
1441
1442 newsk->sk_gso_type = SKB_GSO_TCPV4;
1443
1444 newtp = tcp_sk(newsk);
1445 newinet = inet_sk(newsk);
1446 ireq = inet_rsk(req);
1447 newinet->inet_daddr = ireq->rmt_addr;
1448 newinet->inet_rcv_saddr = ireq->loc_addr;
1449 newinet->inet_saddr = ireq->loc_addr;
1450 inet_opt = ireq->opt;
1451 rcu_assign_pointer(newinet->inet_opt, inet_opt);
1452 ireq->opt = NULL;
1453 newinet->mc_index = inet_iif(skb);
1454 newinet->mc_ttl = ip_hdr(skb)->ttl;
1455 inet_csk(newsk)->icsk_ext_hdr_len = 0;
1456 if (inet_opt)
1457 inet_csk(newsk)->icsk_ext_hdr_len = inet_opt->opt.optlen;
1458 newinet->inet_id = newtp->write_seq ^ jiffies;
1459
1460 if (!dst && (dst = inet_csk_route_child_sock(sk, newsk, req)) == NULL)
1461 goto put_and_exit;
1462
1463 sk_setup_caps(newsk, dst);
1464
1465 tcp_mtup_init(newsk);
1466 tcp_sync_mss(newsk, dst_mtu(dst));
1467 newtp->advmss = dst_metric_advmss(dst);
1468 if (tcp_sk(sk)->rx_opt.user_mss &&
1469 tcp_sk(sk)->rx_opt.user_mss < newtp->advmss)
1470 newtp->advmss = tcp_sk(sk)->rx_opt.user_mss;
1471
1472 tcp_initialize_rcv_mss(newsk);
1473 if (tcp_rsk(req)->snt_synack)
1474 tcp_valid_rtt_meas(newsk,
1475 tcp_time_stamp - tcp_rsk(req)->snt_synack);
1476 newtp->total_retrans = req->retrans;
1477
1478#ifdef CONFIG_TCP_MD5SIG
1479 /* Copy over the MD5 key from the original socket */
1480 key = tcp_v4_md5_do_lookup(sk, newinet->inet_daddr);
1481 if (key != NULL) {
1482 /*
1483 * We're using one, so create a matching key
1484 * on the newsk structure. If we fail to get
1485 * memory, then we end up not copying the key
1486 * across. Shucks.
1487 */
1488 char *newkey = kmemdup(key->key, key->keylen, GFP_ATOMIC);
1489 if (newkey != NULL)
1490 tcp_v4_md5_do_add(newsk, newinet->inet_daddr,
1491 newkey, key->keylen);
1492 sk_nocaps_add(newsk, NETIF_F_GSO_MASK);
1493 }
1494#endif
1495
1496 if (__inet_inherit_port(sk, newsk) < 0)
1497 goto put_and_exit;
1498 __inet_hash_nolisten(newsk, NULL);
1499
1500 return newsk;
1501
1502exit_overflow:
1503 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENOVERFLOWS);
1504exit_nonewsk:
1505 dst_release(dst);
1506exit:
1507 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENDROPS);
1508 return NULL;
1509put_and_exit:
1510 sock_put(newsk);
1511 goto exit;
1512}
1513EXPORT_SYMBOL(tcp_v4_syn_recv_sock);
1514
1515static struct sock *tcp_v4_hnd_req(struct sock *sk, struct sk_buff *skb)
1516{
1517 struct tcphdr *th = tcp_hdr(skb);
1518 const struct iphdr *iph = ip_hdr(skb);
1519 struct sock *nsk;
1520 struct request_sock **prev;
1521 /* Find possible connection requests. */
1522 struct request_sock *req = inet_csk_search_req(sk, &prev, th->source,
1523 iph->saddr, iph->daddr);
1524 if (req)
1525 return tcp_check_req(sk, skb, req, prev);
1526
1527 nsk = inet_lookup_established(sock_net(sk), &tcp_hashinfo, iph->saddr,
1528 th->source, iph->daddr, th->dest, inet_iif(skb));
1529
1530 if (nsk) {
1531 if (nsk->sk_state != TCP_TIME_WAIT) {
1532 bh_lock_sock(nsk);
1533 return nsk;
1534 }
1535 inet_twsk_put(inet_twsk(nsk));
1536 return NULL;
1537 }
1538
1539#ifdef CONFIG_SYN_COOKIES
1540 if (!th->syn)
1541 sk = cookie_v4_check(sk, skb, &(IPCB(skb)->opt));
1542#endif
1543 return sk;
1544}
1545
1546static __sum16 tcp_v4_checksum_init(struct sk_buff *skb)
1547{
1548 const struct iphdr *iph = ip_hdr(skb);
1549
1550 if (skb->ip_summed == CHECKSUM_COMPLETE) {
1551 if (!tcp_v4_check(skb->len, iph->saddr,
1552 iph->daddr, skb->csum)) {
1553 skb->ip_summed = CHECKSUM_UNNECESSARY;
1554 return 0;
1555 }
1556 }
1557
1558 skb->csum = csum_tcpudp_nofold(iph->saddr, iph->daddr,
1559 skb->len, IPPROTO_TCP, 0);
1560
1561 if (skb->len <= 76) {
1562 return __skb_checksum_complete(skb);
1563 }
1564 return 0;
1565}
1566
1567
1568/* The socket must have it's spinlock held when we get
1569 * here.
1570 *
1571 * We have a potential double-lock case here, so even when
1572 * doing backlog processing we use the BH locking scheme.
1573 * This is because we cannot sleep with the original spinlock
1574 * held.
1575 */
1576int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb)
1577{
1578 struct sock *rsk;
1579#ifdef CONFIG_TCP_MD5SIG
1580 /*
1581 * We really want to reject the packet as early as possible
1582 * if:
1583 * o We're expecting an MD5'd packet and this is no MD5 tcp option
1584 * o There is an MD5 option and we're not expecting one
1585 */
1586 if (tcp_v4_inbound_md5_hash(sk, skb))
1587 goto discard;
1588#endif
1589
1590 if (sk->sk_state == TCP_ESTABLISHED) { /* Fast path */
1591 sock_rps_save_rxhash(sk, skb->rxhash);
1592 if (tcp_rcv_established(sk, skb, tcp_hdr(skb), skb->len)) {
1593 rsk = sk;
1594 goto reset;
1595 }
1596 return 0;
1597 }
1598
1599 if (skb->len < tcp_hdrlen(skb) || tcp_checksum_complete(skb))
1600 goto csum_err;
1601
1602 if (sk->sk_state == TCP_LISTEN) {
1603 struct sock *nsk = tcp_v4_hnd_req(sk, skb);
1604 if (!nsk)
1605 goto discard;
1606
1607 if (nsk != sk) {
1608 sock_rps_save_rxhash(nsk, skb->rxhash);
1609 if (tcp_child_process(sk, nsk, skb)) {
1610 rsk = nsk;
1611 goto reset;
1612 }
1613 return 0;
1614 }
1615 } else
1616 sock_rps_save_rxhash(sk, skb->rxhash);
1617
1618 if (tcp_rcv_state_process(sk, skb, tcp_hdr(skb), skb->len)) {
1619 rsk = sk;
1620 goto reset;
1621 }
1622 return 0;
1623
1624reset:
1625 tcp_v4_send_reset(rsk, skb);
1626discard:
1627 kfree_skb(skb);
1628 /* Be careful here. If this function gets more complicated and
1629 * gcc suffers from register pressure on the x86, sk (in %ebx)
1630 * might be destroyed here. This current version compiles correctly,
1631 * but you have been warned.
1632 */
1633 return 0;
1634
1635csum_err:
1636 TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_INERRS);
1637 goto discard;
1638}
1639EXPORT_SYMBOL(tcp_v4_do_rcv);
1640
1641/*
1642 * From tcp_input.c
1643 */
1644
1645int tcp_v4_rcv(struct sk_buff *skb)
1646{
1647 const struct iphdr *iph;
1648 struct tcphdr *th;
1649 struct sock *sk;
1650 int ret;
1651 struct net *net = dev_net(skb->dev);
1652
1653 if (skb->pkt_type != PACKET_HOST)
1654 goto discard_it;
1655
1656 /* Count it even if it's bad */
1657 TCP_INC_STATS_BH(net, TCP_MIB_INSEGS);
1658
1659 if (!pskb_may_pull(skb, sizeof(struct tcphdr)))
1660 goto discard_it;
1661
1662 th = tcp_hdr(skb);
1663
1664 if (th->doff < sizeof(struct tcphdr) / 4)
1665 goto bad_packet;
1666 if (!pskb_may_pull(skb, th->doff * 4))
1667 goto discard_it;
1668
1669 /* An explanation is required here, I think.
1670 * Packet length and doff are validated by header prediction,
1671 * provided case of th->doff==0 is eliminated.
1672 * So, we defer the checks. */
1673 if (!skb_csum_unnecessary(skb) && tcp_v4_checksum_init(skb))
1674 goto bad_packet;
1675
1676 th = tcp_hdr(skb);
1677 iph = ip_hdr(skb);
1678 TCP_SKB_CB(skb)->seq = ntohl(th->seq);
1679 TCP_SKB_CB(skb)->end_seq = (TCP_SKB_CB(skb)->seq + th->syn + th->fin +
1680 skb->len - th->doff * 4);
1681 TCP_SKB_CB(skb)->ack_seq = ntohl(th->ack_seq);
1682 TCP_SKB_CB(skb)->when = 0;
1683 TCP_SKB_CB(skb)->flags = iph->tos;
1684 TCP_SKB_CB(skb)->sacked = 0;
1685
1686 sk = __inet_lookup_skb(&tcp_hashinfo, skb, th->source, th->dest);
1687 if (!sk)
1688 goto no_tcp_socket;
1689
1690process:
1691 if (sk->sk_state == TCP_TIME_WAIT)
1692 goto do_time_wait;
1693
1694 if (unlikely(iph->ttl < inet_sk(sk)->min_ttl)) {
1695 NET_INC_STATS_BH(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 nf_reset(skb);
1702
1703 if (sk_filter(sk, skb))
1704 goto discard_and_relse;
1705
1706 skb->dev = NULL;
1707
1708 bh_lock_sock_nested(sk);
1709 ret = 0;
1710 if (!sock_owned_by_user(sk)) {
1711#ifdef CONFIG_NET_DMA
1712 struct tcp_sock *tp = tcp_sk(sk);
1713 if (!tp->ucopy.dma_chan && tp->ucopy.pinned_list)
1714 tp->ucopy.dma_chan = dma_find_channel(DMA_MEMCPY);
1715 if (tp->ucopy.dma_chan)
1716 ret = tcp_v4_do_rcv(sk, skb);
1717 else
1718#endif
1719 {
1720 if (!tcp_prequeue(sk, skb))
1721 ret = tcp_v4_do_rcv(sk, skb);
1722 }
1723 } else if (unlikely(sk_add_backlog(sk, skb))) {
1724 bh_unlock_sock(sk);
1725 NET_INC_STATS_BH(net, LINUX_MIB_TCPBACKLOGDROP);
1726 goto discard_and_relse;
1727 }
1728 bh_unlock_sock(sk);
1729
1730 sock_put(sk);
1731
1732 return ret;
1733
1734no_tcp_socket:
1735 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
1736 goto discard_it;
1737
1738 if (skb->len < (th->doff << 2) || tcp_checksum_complete(skb)) {
1739bad_packet:
1740 TCP_INC_STATS_BH(net, TCP_MIB_INERRS);
1741 } else {
1742 tcp_v4_send_reset(NULL, skb);
1743 }
1744
1745discard_it:
1746 /* Discard frame. */
1747 kfree_skb(skb);
1748 return 0;
1749
1750discard_and_relse:
1751 sock_put(sk);
1752 goto discard_it;
1753
1754do_time_wait:
1755 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) {
1756 inet_twsk_put(inet_twsk(sk));
1757 goto discard_it;
1758 }
1759
1760 if (skb->len < (th->doff << 2) || tcp_checksum_complete(skb)) {
1761 TCP_INC_STATS_BH(net, TCP_MIB_INERRS);
1762 inet_twsk_put(inet_twsk(sk));
1763 goto discard_it;
1764 }
1765 switch (tcp_timewait_state_process(inet_twsk(sk), skb, th)) {
1766 case TCP_TW_SYN: {
1767 struct sock *sk2 = inet_lookup_listener(dev_net(skb->dev),
1768 &tcp_hashinfo,
1769 iph->daddr, th->dest,
1770 inet_iif(skb));
1771 if (sk2) {
1772 inet_twsk_deschedule(inet_twsk(sk), &tcp_death_row);
1773 inet_twsk_put(inet_twsk(sk));
1774 sk = sk2;
1775 goto process;
1776 }
1777 /* Fall through to ACK */
1778 }
1779 case TCP_TW_ACK:
1780 tcp_v4_timewait_ack(sk, skb);
1781 break;
1782 case TCP_TW_RST:
1783 goto no_tcp_socket;
1784 case TCP_TW_SUCCESS:;
1785 }
1786 goto discard_it;
1787}
1788
1789struct inet_peer *tcp_v4_get_peer(struct sock *sk, bool *release_it)
1790{
1791 struct rtable *rt = (struct rtable *) __sk_dst_get(sk);
1792 struct inet_sock *inet = inet_sk(sk);
1793 struct inet_peer *peer;
1794
1795 if (!rt ||
1796 inet->cork.fl.u.ip4.daddr != inet->inet_daddr) {
1797 peer = inet_getpeer_v4(inet->inet_daddr, 1);
1798 *release_it = true;
1799 } else {
1800 if (!rt->peer)
1801 rt_bind_peer(rt, inet->inet_daddr, 1);
1802 peer = rt->peer;
1803 *release_it = false;
1804 }
1805
1806 return peer;
1807}
1808EXPORT_SYMBOL(tcp_v4_get_peer);
1809
1810void *tcp_v4_tw_get_peer(struct sock *sk)
1811{
1812 struct inet_timewait_sock *tw = inet_twsk(sk);
1813
1814 return inet_getpeer_v4(tw->tw_daddr, 1);
1815}
1816EXPORT_SYMBOL(tcp_v4_tw_get_peer);
1817
1818static struct timewait_sock_ops tcp_timewait_sock_ops = {
1819 .twsk_obj_size = sizeof(struct tcp_timewait_sock),
1820 .twsk_unique = tcp_twsk_unique,
1821 .twsk_destructor= tcp_twsk_destructor,
1822 .twsk_getpeer = tcp_v4_tw_get_peer,
1823};
1824
1825const struct inet_connection_sock_af_ops ipv4_specific = {
1826 .queue_xmit = ip_queue_xmit,
1827 .send_check = tcp_v4_send_check,
1828 .rebuild_header = inet_sk_rebuild_header,
1829 .conn_request = tcp_v4_conn_request,
1830 .syn_recv_sock = tcp_v4_syn_recv_sock,
1831 .get_peer = tcp_v4_get_peer,
1832 .net_header_len = sizeof(struct iphdr),
1833 .setsockopt = ip_setsockopt,
1834 .getsockopt = ip_getsockopt,
1835 .addr2sockaddr = inet_csk_addr2sockaddr,
1836 .sockaddr_len = sizeof(struct sockaddr_in),
1837 .bind_conflict = inet_csk_bind_conflict,
1838#ifdef CONFIG_COMPAT
1839 .compat_setsockopt = compat_ip_setsockopt,
1840 .compat_getsockopt = compat_ip_getsockopt,
1841#endif
1842};
1843EXPORT_SYMBOL(ipv4_specific);
1844
1845#ifdef CONFIG_TCP_MD5SIG
1846static const struct tcp_sock_af_ops tcp_sock_ipv4_specific = {
1847 .md5_lookup = tcp_v4_md5_lookup,
1848 .calc_md5_hash = tcp_v4_md5_hash_skb,
1849 .md5_add = tcp_v4_md5_add_func,
1850 .md5_parse = tcp_v4_parse_md5_keys,
1851};
1852#endif
1853
1854/* NOTE: A lot of things set to zero explicitly by call to
1855 * sk_alloc() so need not be done here.
1856 */
1857static int tcp_v4_init_sock(struct sock *sk)
1858{
1859 struct inet_connection_sock *icsk = inet_csk(sk);
1860 struct tcp_sock *tp = tcp_sk(sk);
1861
1862 skb_queue_head_init(&tp->out_of_order_queue);
1863 tcp_init_xmit_timers(sk);
1864 tcp_prequeue_init(tp);
1865
1866 icsk->icsk_rto = TCP_TIMEOUT_INIT;
1867 tp->mdev = TCP_TIMEOUT_INIT;
1868
1869 /* So many TCP implementations out there (incorrectly) count the
1870 * initial SYN frame in their delayed-ACK and congestion control
1871 * algorithms that we must have the following bandaid to talk
1872 * efficiently to them. -DaveM
1873 */
1874 tp->snd_cwnd = TCP_INIT_CWND;
1875
1876 /* See draft-stevens-tcpca-spec-01 for discussion of the
1877 * initialization of these values.
1878 */
1879 tp->snd_ssthresh = TCP_INFINITE_SSTHRESH;
1880 tp->snd_cwnd_clamp = ~0;
1881 tp->mss_cache = TCP_MSS_DEFAULT;
1882
1883 tp->reordering = sysctl_tcp_reordering;
1884 icsk->icsk_ca_ops = &tcp_init_congestion_ops;
1885
1886 sk->sk_state = TCP_CLOSE;
1887
1888 sk->sk_write_space = sk_stream_write_space;
1889 sock_set_flag(sk, SOCK_USE_WRITE_QUEUE);
1890
1891 icsk->icsk_af_ops = &ipv4_specific;
1892 icsk->icsk_sync_mss = tcp_sync_mss;
1893#ifdef CONFIG_TCP_MD5SIG
1894 tp->af_specific = &tcp_sock_ipv4_specific;
1895#endif
1896
1897 /* TCP Cookie Transactions */
1898 if (sysctl_tcp_cookie_size > 0) {
1899 /* Default, cookies without s_data_payload. */
1900 tp->cookie_values =
1901 kzalloc(sizeof(*tp->cookie_values),
1902 sk->sk_allocation);
1903 if (tp->cookie_values != NULL)
1904 kref_init(&tp->cookie_values->kref);
1905 }
1906 /* Presumed zeroed, in order of appearance:
1907 * cookie_in_always, cookie_out_never,
1908 * s_data_constant, s_data_in, s_data_out
1909 */
1910 sk->sk_sndbuf = sysctl_tcp_wmem[1];
1911 sk->sk_rcvbuf = sysctl_tcp_rmem[1];
1912
1913 local_bh_disable();
1914 percpu_counter_inc(&tcp_sockets_allocated);
1915 local_bh_enable();
1916
1917 return 0;
1918}
1919
1920void tcp_v4_destroy_sock(struct sock *sk)
1921{
1922 struct tcp_sock *tp = tcp_sk(sk);
1923
1924 tcp_clear_xmit_timers(sk);
1925
1926 tcp_cleanup_congestion_control(sk);
1927
1928 /* Cleanup up the write buffer. */
1929 tcp_write_queue_purge(sk);
1930
1931 /* Cleans up our, hopefully empty, out_of_order_queue. */
1932 __skb_queue_purge(&tp->out_of_order_queue);
1933
1934#ifdef CONFIG_TCP_MD5SIG
1935 /* Clean up the MD5 key list, if any */
1936 if (tp->md5sig_info) {
1937 tcp_v4_clear_md5_list(sk);
1938 kfree(tp->md5sig_info);
1939 tp->md5sig_info = NULL;
1940 }
1941#endif
1942
1943#ifdef CONFIG_NET_DMA
1944 /* Cleans up our sk_async_wait_queue */
1945 __skb_queue_purge(&sk->sk_async_wait_queue);
1946#endif
1947
1948 /* Clean prequeue, it must be empty really */
1949 __skb_queue_purge(&tp->ucopy.prequeue);
1950
1951 /* Clean up a referenced TCP bind bucket. */
1952 if (inet_csk(sk)->icsk_bind_hash)
1953 inet_put_port(sk);
1954
1955 /*
1956 * If sendmsg cached page exists, toss it.
1957 */
1958 if (sk->sk_sndmsg_page) {
1959 __free_page(sk->sk_sndmsg_page);
1960 sk->sk_sndmsg_page = NULL;
1961 }
1962
1963 /* TCP Cookie Transactions */
1964 if (tp->cookie_values != NULL) {
1965 kref_put(&tp->cookie_values->kref,
1966 tcp_cookie_values_release);
1967 tp->cookie_values = NULL;
1968 }
1969
1970 percpu_counter_dec(&tcp_sockets_allocated);
1971}
1972EXPORT_SYMBOL(tcp_v4_destroy_sock);
1973
1974#ifdef CONFIG_PROC_FS
1975/* Proc filesystem TCP sock list dumping. */
1976
1977static inline struct inet_timewait_sock *tw_head(struct hlist_nulls_head *head)
1978{
1979 return hlist_nulls_empty(head) ? NULL :
1980 list_entry(head->first, struct inet_timewait_sock, tw_node);
1981}
1982
1983static inline struct inet_timewait_sock *tw_next(struct inet_timewait_sock *tw)
1984{
1985 return !is_a_nulls(tw->tw_node.next) ?
1986 hlist_nulls_entry(tw->tw_node.next, typeof(*tw), tw_node) : NULL;
1987}
1988
1989/*
1990 * Get next listener socket follow cur. If cur is NULL, get first socket
1991 * starting from bucket given in st->bucket; when st->bucket is zero the
1992 * very first socket in the hash table is returned.
1993 */
1994static void *listening_get_next(struct seq_file *seq, void *cur)
1995{
1996 struct inet_connection_sock *icsk;
1997 struct hlist_nulls_node *node;
1998 struct sock *sk = cur;
1999 struct inet_listen_hashbucket *ilb;
2000 struct tcp_iter_state *st = seq->private;
2001 struct net *net = seq_file_net(seq);
2002
2003 if (!sk) {
2004 ilb = &tcp_hashinfo.listening_hash[st->bucket];
2005 spin_lock_bh(&ilb->lock);
2006 sk = sk_nulls_head(&ilb->head);
2007 st->offset = 0;
2008 goto get_sk;
2009 }
2010 ilb = &tcp_hashinfo.listening_hash[st->bucket];
2011 ++st->num;
2012 ++st->offset;
2013
2014 if (st->state == TCP_SEQ_STATE_OPENREQ) {
2015 struct request_sock *req = cur;
2016
2017 icsk = inet_csk(st->syn_wait_sk);
2018 req = req->dl_next;
2019 while (1) {
2020 while (req) {
2021 if (req->rsk_ops->family == st->family) {
2022 cur = req;
2023 goto out;
2024 }
2025 req = req->dl_next;
2026 }
2027 if (++st->sbucket >= icsk->icsk_accept_queue.listen_opt->nr_table_entries)
2028 break;
2029get_req:
2030 req = icsk->icsk_accept_queue.listen_opt->syn_table[st->sbucket];
2031 }
2032 sk = sk_nulls_next(st->syn_wait_sk);
2033 st->state = TCP_SEQ_STATE_LISTENING;
2034 read_unlock_bh(&icsk->icsk_accept_queue.syn_wait_lock);
2035 } else {
2036 icsk = inet_csk(sk);
2037 read_lock_bh(&icsk->icsk_accept_queue.syn_wait_lock);
2038 if (reqsk_queue_len(&icsk->icsk_accept_queue))
2039 goto start_req;
2040 read_unlock_bh(&icsk->icsk_accept_queue.syn_wait_lock);
2041 sk = sk_nulls_next(sk);
2042 }
2043get_sk:
2044 sk_nulls_for_each_from(sk, node) {
2045 if (!net_eq(sock_net(sk), net))
2046 continue;
2047 if (sk->sk_family == st->family) {
2048 cur = sk;
2049 goto out;
2050 }
2051 icsk = inet_csk(sk);
2052 read_lock_bh(&icsk->icsk_accept_queue.syn_wait_lock);
2053 if (reqsk_queue_len(&icsk->icsk_accept_queue)) {
2054start_req:
2055 st->uid = sock_i_uid(sk);
2056 st->syn_wait_sk = sk;
2057 st->state = TCP_SEQ_STATE_OPENREQ;
2058 st->sbucket = 0;
2059 goto get_req;
2060 }
2061 read_unlock_bh(&icsk->icsk_accept_queue.syn_wait_lock);
2062 }
2063 spin_unlock_bh(&ilb->lock);
2064 st->offset = 0;
2065 if (++st->bucket < INET_LHTABLE_SIZE) {
2066 ilb = &tcp_hashinfo.listening_hash[st->bucket];
2067 spin_lock_bh(&ilb->lock);
2068 sk = sk_nulls_head(&ilb->head);
2069 goto get_sk;
2070 }
2071 cur = NULL;
2072out:
2073 return cur;
2074}
2075
2076static void *listening_get_idx(struct seq_file *seq, loff_t *pos)
2077{
2078 struct tcp_iter_state *st = seq->private;
2079 void *rc;
2080
2081 st->bucket = 0;
2082 st->offset = 0;
2083 rc = listening_get_next(seq, NULL);
2084
2085 while (rc && *pos) {
2086 rc = listening_get_next(seq, rc);
2087 --*pos;
2088 }
2089 return rc;
2090}
2091
2092static inline int empty_bucket(struct tcp_iter_state *st)
2093{
2094 return hlist_nulls_empty(&tcp_hashinfo.ehash[st->bucket].chain) &&
2095 hlist_nulls_empty(&tcp_hashinfo.ehash[st->bucket].twchain);
2096}
2097
2098/*
2099 * Get first established socket starting from bucket given in st->bucket.
2100 * If st->bucket is zero, the very first socket in the hash is returned.
2101 */
2102static void *established_get_first(struct seq_file *seq)
2103{
2104 struct tcp_iter_state *st = seq->private;
2105 struct net *net = seq_file_net(seq);
2106 void *rc = NULL;
2107
2108 st->offset = 0;
2109 for (; st->bucket <= tcp_hashinfo.ehash_mask; ++st->bucket) {
2110 struct sock *sk;
2111 struct hlist_nulls_node *node;
2112 struct inet_timewait_sock *tw;
2113 spinlock_t *lock = inet_ehash_lockp(&tcp_hashinfo, st->bucket);
2114
2115 /* Lockless fast path for the common case of empty buckets */
2116 if (empty_bucket(st))
2117 continue;
2118
2119 spin_lock_bh(lock);
2120 sk_nulls_for_each(sk, node, &tcp_hashinfo.ehash[st->bucket].chain) {
2121 if (sk->sk_family != st->family ||
2122 !net_eq(sock_net(sk), net)) {
2123 continue;
2124 }
2125 rc = sk;
2126 goto out;
2127 }
2128 st->state = TCP_SEQ_STATE_TIME_WAIT;
2129 inet_twsk_for_each(tw, node,
2130 &tcp_hashinfo.ehash[st->bucket].twchain) {
2131 if (tw->tw_family != st->family ||
2132 !net_eq(twsk_net(tw), net)) {
2133 continue;
2134 }
2135 rc = tw;
2136 goto out;
2137 }
2138 spin_unlock_bh(lock);
2139 st->state = TCP_SEQ_STATE_ESTABLISHED;
2140 }
2141out:
2142 return rc;
2143}
2144
2145static void *established_get_next(struct seq_file *seq, void *cur)
2146{
2147 struct sock *sk = cur;
2148 struct inet_timewait_sock *tw;
2149 struct hlist_nulls_node *node;
2150 struct tcp_iter_state *st = seq->private;
2151 struct net *net = seq_file_net(seq);
2152
2153 ++st->num;
2154 ++st->offset;
2155
2156 if (st->state == TCP_SEQ_STATE_TIME_WAIT) {
2157 tw = cur;
2158 tw = tw_next(tw);
2159get_tw:
2160 while (tw && (tw->tw_family != st->family || !net_eq(twsk_net(tw), net))) {
2161 tw = tw_next(tw);
2162 }
2163 if (tw) {
2164 cur = tw;
2165 goto out;
2166 }
2167 spin_unlock_bh(inet_ehash_lockp(&tcp_hashinfo, st->bucket));
2168 st->state = TCP_SEQ_STATE_ESTABLISHED;
2169
2170 /* Look for next non empty bucket */
2171 st->offset = 0;
2172 while (++st->bucket <= tcp_hashinfo.ehash_mask &&
2173 empty_bucket(st))
2174 ;
2175 if (st->bucket > tcp_hashinfo.ehash_mask)
2176 return NULL;
2177
2178 spin_lock_bh(inet_ehash_lockp(&tcp_hashinfo, st->bucket));
2179 sk = sk_nulls_head(&tcp_hashinfo.ehash[st->bucket].chain);
2180 } else
2181 sk = sk_nulls_next(sk);
2182
2183 sk_nulls_for_each_from(sk, node) {
2184 if (sk->sk_family == st->family && net_eq(sock_net(sk), net))
2185 goto found;
2186 }
2187
2188 st->state = TCP_SEQ_STATE_TIME_WAIT;
2189 tw = tw_head(&tcp_hashinfo.ehash[st->bucket].twchain);
2190 goto get_tw;
2191found:
2192 cur = sk;
2193out:
2194 return cur;
2195}
2196
2197static void *established_get_idx(struct seq_file *seq, loff_t pos)
2198{
2199 struct tcp_iter_state *st = seq->private;
2200 void *rc;
2201
2202 st->bucket = 0;
2203 rc = established_get_first(seq);
2204
2205 while (rc && pos) {
2206 rc = established_get_next(seq, rc);
2207 --pos;
2208 }
2209 return rc;
2210}
2211
2212static void *tcp_get_idx(struct seq_file *seq, loff_t pos)
2213{
2214 void *rc;
2215 struct tcp_iter_state *st = seq->private;
2216
2217 st->state = TCP_SEQ_STATE_LISTENING;
2218 rc = listening_get_idx(seq, &pos);
2219
2220 if (!rc) {
2221 st->state = TCP_SEQ_STATE_ESTABLISHED;
2222 rc = established_get_idx(seq, pos);
2223 }
2224
2225 return rc;
2226}
2227
2228static void *tcp_seek_last_pos(struct seq_file *seq)
2229{
2230 struct tcp_iter_state *st = seq->private;
2231 int offset = st->offset;
2232 int orig_num = st->num;
2233 void *rc = NULL;
2234
2235 switch (st->state) {
2236 case TCP_SEQ_STATE_OPENREQ:
2237 case TCP_SEQ_STATE_LISTENING:
2238 if (st->bucket >= INET_LHTABLE_SIZE)
2239 break;
2240 st->state = TCP_SEQ_STATE_LISTENING;
2241 rc = listening_get_next(seq, NULL);
2242 while (offset-- && rc)
2243 rc = listening_get_next(seq, rc);
2244 if (rc)
2245 break;
2246 st->bucket = 0;
2247 /* Fallthrough */
2248 case TCP_SEQ_STATE_ESTABLISHED:
2249 case TCP_SEQ_STATE_TIME_WAIT:
2250 st->state = TCP_SEQ_STATE_ESTABLISHED;
2251 if (st->bucket > tcp_hashinfo.ehash_mask)
2252 break;
2253 rc = established_get_first(seq);
2254 while (offset-- && rc)
2255 rc = established_get_next(seq, rc);
2256 }
2257
2258 st->num = orig_num;
2259
2260 return rc;
2261}
2262
2263static void *tcp_seq_start(struct seq_file *seq, loff_t *pos)
2264{
2265 struct tcp_iter_state *st = seq->private;
2266 void *rc;
2267
2268 if (*pos && *pos == st->last_pos) {
2269 rc = tcp_seek_last_pos(seq);
2270 if (rc)
2271 goto out;
2272 }
2273
2274 st->state = TCP_SEQ_STATE_LISTENING;
2275 st->num = 0;
2276 st->bucket = 0;
2277 st->offset = 0;
2278 rc = *pos ? tcp_get_idx(seq, *pos - 1) : SEQ_START_TOKEN;
2279
2280out:
2281 st->last_pos = *pos;
2282 return rc;
2283}
2284
2285static void *tcp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2286{
2287 struct tcp_iter_state *st = seq->private;
2288 void *rc = NULL;
2289
2290 if (v == SEQ_START_TOKEN) {
2291 rc = tcp_get_idx(seq, 0);
2292 goto out;
2293 }
2294
2295 switch (st->state) {
2296 case TCP_SEQ_STATE_OPENREQ:
2297 case TCP_SEQ_STATE_LISTENING:
2298 rc = listening_get_next(seq, v);
2299 if (!rc) {
2300 st->state = TCP_SEQ_STATE_ESTABLISHED;
2301 st->bucket = 0;
2302 st->offset = 0;
2303 rc = established_get_first(seq);
2304 }
2305 break;
2306 case TCP_SEQ_STATE_ESTABLISHED:
2307 case TCP_SEQ_STATE_TIME_WAIT:
2308 rc = established_get_next(seq, v);
2309 break;
2310 }
2311out:
2312 ++*pos;
2313 st->last_pos = *pos;
2314 return rc;
2315}
2316
2317static void tcp_seq_stop(struct seq_file *seq, void *v)
2318{
2319 struct tcp_iter_state *st = seq->private;
2320
2321 switch (st->state) {
2322 case TCP_SEQ_STATE_OPENREQ:
2323 if (v) {
2324 struct inet_connection_sock *icsk = inet_csk(st->syn_wait_sk);
2325 read_unlock_bh(&icsk->icsk_accept_queue.syn_wait_lock);
2326 }
2327 case TCP_SEQ_STATE_LISTENING:
2328 if (v != SEQ_START_TOKEN)
2329 spin_unlock_bh(&tcp_hashinfo.listening_hash[st->bucket].lock);
2330 break;
2331 case TCP_SEQ_STATE_TIME_WAIT:
2332 case TCP_SEQ_STATE_ESTABLISHED:
2333 if (v)
2334 spin_unlock_bh(inet_ehash_lockp(&tcp_hashinfo, st->bucket));
2335 break;
2336 }
2337}
2338
2339static int tcp_seq_open(struct inode *inode, struct file *file)
2340{
2341 struct tcp_seq_afinfo *afinfo = PDE(inode)->data;
2342 struct tcp_iter_state *s;
2343 int err;
2344
2345 err = seq_open_net(inode, file, &afinfo->seq_ops,
2346 sizeof(struct tcp_iter_state));
2347 if (err < 0)
2348 return err;
2349
2350 s = ((struct seq_file *)file->private_data)->private;
2351 s->family = afinfo->family;
2352 s->last_pos = 0;
2353 return 0;
2354}
2355
2356int tcp_proc_register(struct net *net, struct tcp_seq_afinfo *afinfo)
2357{
2358 int rc = 0;
2359 struct proc_dir_entry *p;
2360
2361 afinfo->seq_fops.open = tcp_seq_open;
2362 afinfo->seq_fops.read = seq_read;
2363 afinfo->seq_fops.llseek = seq_lseek;
2364 afinfo->seq_fops.release = seq_release_net;
2365
2366 afinfo->seq_ops.start = tcp_seq_start;
2367 afinfo->seq_ops.next = tcp_seq_next;
2368 afinfo->seq_ops.stop = tcp_seq_stop;
2369
2370 p = proc_create_data(afinfo->name, S_IRUGO, net->proc_net,
2371 &afinfo->seq_fops, afinfo);
2372 if (!p)
2373 rc = -ENOMEM;
2374 return rc;
2375}
2376EXPORT_SYMBOL(tcp_proc_register);
2377
2378void tcp_proc_unregister(struct net *net, struct tcp_seq_afinfo *afinfo)
2379{
2380 proc_net_remove(net, afinfo->name);
2381}
2382EXPORT_SYMBOL(tcp_proc_unregister);
2383
2384static void get_openreq4(struct sock *sk, struct request_sock *req,
2385 struct seq_file *f, int i, int uid, int *len)
2386{
2387 const struct inet_request_sock *ireq = inet_rsk(req);
2388 int ttd = req->expires - jiffies;
2389
2390 seq_printf(f, "%4d: %08X:%04X %08X:%04X"
2391 " %02X %08X:%08X %02X:%08lX %08X %5d %8d %u %d %pK%n",
2392 i,
2393 ireq->loc_addr,
2394 ntohs(inet_sk(sk)->inet_sport),
2395 ireq->rmt_addr,
2396 ntohs(ireq->rmt_port),
2397 TCP_SYN_RECV,
2398 0, 0, /* could print option size, but that is af dependent. */
2399 1, /* timers active (only the expire timer) */
2400 jiffies_to_clock_t(ttd),
2401 req->retrans,
2402 uid,
2403 0, /* non standard timer */
2404 0, /* open_requests have no inode */
2405 atomic_read(&sk->sk_refcnt),
2406 req,
2407 len);
2408}
2409
2410static void get_tcp4_sock(struct sock *sk, struct seq_file *f, int i, int *len)
2411{
2412 int timer_active;
2413 unsigned long timer_expires;
2414 struct tcp_sock *tp = tcp_sk(sk);
2415 const struct inet_connection_sock *icsk = inet_csk(sk);
2416 struct inet_sock *inet = inet_sk(sk);
2417 __be32 dest = inet->inet_daddr;
2418 __be32 src = inet->inet_rcv_saddr;
2419 __u16 destp = ntohs(inet->inet_dport);
2420 __u16 srcp = ntohs(inet->inet_sport);
2421 int rx_queue;
2422
2423 if (icsk->icsk_pending == ICSK_TIME_RETRANS) {
2424 timer_active = 1;
2425 timer_expires = icsk->icsk_timeout;
2426 } else if (icsk->icsk_pending == ICSK_TIME_PROBE0) {
2427 timer_active = 4;
2428 timer_expires = icsk->icsk_timeout;
2429 } else if (timer_pending(&sk->sk_timer)) {
2430 timer_active = 2;
2431 timer_expires = sk->sk_timer.expires;
2432 } else {
2433 timer_active = 0;
2434 timer_expires = jiffies;
2435 }
2436
2437 if (sk->sk_state == TCP_LISTEN)
2438 rx_queue = sk->sk_ack_backlog;
2439 else
2440 /*
2441 * because we dont lock socket, we might find a transient negative value
2442 */
2443 rx_queue = max_t(int, tp->rcv_nxt - tp->copied_seq, 0);
2444
2445 seq_printf(f, "%4d: %08X:%04X %08X:%04X %02X %08X:%08X %02X:%08lX "
2446 "%08X %5d %8d %lu %d %pK %lu %lu %u %u %d%n",
2447 i, src, srcp, dest, destp, sk->sk_state,
2448 tp->write_seq - tp->snd_una,
2449 rx_queue,
2450 timer_active,
2451 jiffies_to_clock_t(timer_expires - jiffies),
2452 icsk->icsk_retransmits,
2453 sock_i_uid(sk),
2454 icsk->icsk_probes_out,
2455 sock_i_ino(sk),
2456 atomic_read(&sk->sk_refcnt), sk,
2457 jiffies_to_clock_t(icsk->icsk_rto),
2458 jiffies_to_clock_t(icsk->icsk_ack.ato),
2459 (icsk->icsk_ack.quick << 1) | icsk->icsk_ack.pingpong,
2460 tp->snd_cwnd,
2461 tcp_in_initial_slowstart(tp) ? -1 : tp->snd_ssthresh,
2462 len);
2463}
2464
2465static void get_timewait4_sock(struct inet_timewait_sock *tw,
2466 struct seq_file *f, int i, int *len)
2467{
2468 __be32 dest, src;
2469 __u16 destp, srcp;
2470 int ttd = tw->tw_ttd - jiffies;
2471
2472 if (ttd < 0)
2473 ttd = 0;
2474
2475 dest = tw->tw_daddr;
2476 src = tw->tw_rcv_saddr;
2477 destp = ntohs(tw->tw_dport);
2478 srcp = ntohs(tw->tw_sport);
2479
2480 seq_printf(f, "%4d: %08X:%04X %08X:%04X"
2481 " %02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %pK%n",
2482 i, src, srcp, dest, destp, tw->tw_substate, 0, 0,
2483 3, jiffies_to_clock_t(ttd), 0, 0, 0, 0,
2484 atomic_read(&tw->tw_refcnt), tw, len);
2485}
2486
2487#define TMPSZ 150
2488
2489static int tcp4_seq_show(struct seq_file *seq, void *v)
2490{
2491 struct tcp_iter_state *st;
2492 int len;
2493
2494 if (v == SEQ_START_TOKEN) {
2495 seq_printf(seq, "%-*s\n", TMPSZ - 1,
2496 " sl local_address rem_address st tx_queue "
2497 "rx_queue tr tm->when retrnsmt uid timeout "
2498 "inode");
2499 goto out;
2500 }
2501 st = seq->private;
2502
2503 switch (st->state) {
2504 case TCP_SEQ_STATE_LISTENING:
2505 case TCP_SEQ_STATE_ESTABLISHED:
2506 get_tcp4_sock(v, seq, st->num, &len);
2507 break;
2508 case TCP_SEQ_STATE_OPENREQ:
2509 get_openreq4(st->syn_wait_sk, v, seq, st->num, st->uid, &len);
2510 break;
2511 case TCP_SEQ_STATE_TIME_WAIT:
2512 get_timewait4_sock(v, seq, st->num, &len);
2513 break;
2514 }
2515 seq_printf(seq, "%*s\n", TMPSZ - 1 - len, "");
2516out:
2517 return 0;
2518}
2519
2520static struct tcp_seq_afinfo tcp4_seq_afinfo = {
2521 .name = "tcp",
2522 .family = AF_INET,
2523 .seq_fops = {
2524 .owner = THIS_MODULE,
2525 },
2526 .seq_ops = {
2527 .show = tcp4_seq_show,
2528 },
2529};
2530
2531static int __net_init tcp4_proc_init_net(struct net *net)
2532{
2533 return tcp_proc_register(net, &tcp4_seq_afinfo);
2534}
2535
2536static void __net_exit tcp4_proc_exit_net(struct net *net)
2537{
2538 tcp_proc_unregister(net, &tcp4_seq_afinfo);
2539}
2540
2541static struct pernet_operations tcp4_net_ops = {
2542 .init = tcp4_proc_init_net,
2543 .exit = tcp4_proc_exit_net,
2544};
2545
2546int __init tcp4_proc_init(void)
2547{
2548 return register_pernet_subsys(&tcp4_net_ops);
2549}
2550
2551void tcp4_proc_exit(void)
2552{
2553 unregister_pernet_subsys(&tcp4_net_ops);
2554}
2555#endif /* CONFIG_PROC_FS */
2556
2557struct sk_buff **tcp4_gro_receive(struct sk_buff **head, struct sk_buff *skb)
2558{
2559 const struct iphdr *iph = skb_gro_network_header(skb);
2560
2561 switch (skb->ip_summed) {
2562 case CHECKSUM_COMPLETE:
2563 if (!tcp_v4_check(skb_gro_len(skb), iph->saddr, iph->daddr,
2564 skb->csum)) {
2565 skb->ip_summed = CHECKSUM_UNNECESSARY;
2566 break;
2567 }
2568
2569 /* fall through */
2570 case CHECKSUM_NONE:
2571 NAPI_GRO_CB(skb)->flush = 1;
2572 return NULL;
2573 }
2574
2575 return tcp_gro_receive(head, skb);
2576}
2577
2578int tcp4_gro_complete(struct sk_buff *skb)
2579{
2580 const struct iphdr *iph = ip_hdr(skb);
2581 struct tcphdr *th = tcp_hdr(skb);
2582
2583 th->check = ~tcp_v4_check(skb->len - skb_transport_offset(skb),
2584 iph->saddr, iph->daddr, 0);
2585 skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
2586
2587 return tcp_gro_complete(skb);
2588}
2589
2590struct proto tcp_prot = {
2591 .name = "TCP",
2592 .owner = THIS_MODULE,
2593 .close = tcp_close,
2594 .connect = tcp_v4_connect,
2595 .disconnect = tcp_disconnect,
2596 .accept = inet_csk_accept,
2597 .ioctl = tcp_ioctl,
2598 .init = tcp_v4_init_sock,
2599 .destroy = tcp_v4_destroy_sock,
2600 .shutdown = tcp_shutdown,
2601 .setsockopt = tcp_setsockopt,
2602 .getsockopt = tcp_getsockopt,
2603 .recvmsg = tcp_recvmsg,
2604 .sendmsg = tcp_sendmsg,
2605 .sendpage = tcp_sendpage,
2606 .backlog_rcv = tcp_v4_do_rcv,
2607 .hash = inet_hash,
2608 .unhash = inet_unhash,
2609 .get_port = inet_csk_get_port,
2610 .enter_memory_pressure = tcp_enter_memory_pressure,
2611 .sockets_allocated = &tcp_sockets_allocated,
2612 .orphan_count = &tcp_orphan_count,
2613 .memory_allocated = &tcp_memory_allocated,
2614 .memory_pressure = &tcp_memory_pressure,
2615 .sysctl_mem = sysctl_tcp_mem,
2616 .sysctl_wmem = sysctl_tcp_wmem,
2617 .sysctl_rmem = sysctl_tcp_rmem,
2618 .max_header = MAX_TCP_HEADER,
2619 .obj_size = sizeof(struct tcp_sock),
2620 .slab_flags = SLAB_DESTROY_BY_RCU,
2621 .twsk_prot = &tcp_timewait_sock_ops,
2622 .rsk_prot = &tcp_request_sock_ops,
2623 .h.hashinfo = &tcp_hashinfo,
2624 .no_autobind = true,
2625#ifdef CONFIG_COMPAT
2626 .compat_setsockopt = compat_tcp_setsockopt,
2627 .compat_getsockopt = compat_tcp_getsockopt,
2628#endif
2629};
2630EXPORT_SYMBOL(tcp_prot);
2631
2632
2633static int __net_init tcp_sk_init(struct net *net)
2634{
2635 return inet_ctl_sock_create(&net->ipv4.tcp_sock,
2636 PF_INET, SOCK_RAW, IPPROTO_TCP, net);
2637}
2638
2639static void __net_exit tcp_sk_exit(struct net *net)
2640{
2641 inet_ctl_sock_destroy(net->ipv4.tcp_sock);
2642}
2643
2644static void __net_exit tcp_sk_exit_batch(struct list_head *net_exit_list)
2645{
2646 inet_twsk_purge(&tcp_hashinfo, &tcp_death_row, AF_INET);
2647}
2648
2649static struct pernet_operations __net_initdata tcp_sk_ops = {
2650 .init = tcp_sk_init,
2651 .exit = tcp_sk_exit,
2652 .exit_batch = tcp_sk_exit_batch,
2653};
2654
2655void __init tcp_v4_init(void)
2656{
2657 inet_hashinfo_init(&tcp_hashinfo);
2658 if (register_pernet_subsys(&tcp_sk_ops))
2659 panic("Failed to create the TCP control socket.\n");
2660}
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 struct ip_options_rcu *inet_opt;
149
150 if (addr_len < sizeof(struct sockaddr_in))
151 return -EINVAL;
152
153 if (usin->sin_family != AF_INET)
154 return -EAFNOSUPPORT;
155
156 nexthop = daddr = usin->sin_addr.s_addr;
157 inet_opt = rcu_dereference_protected(inet->inet_opt,
158 lockdep_sock_is_held(sk));
159 if (inet_opt && inet_opt->opt.srr) {
160 if (!daddr)
161 return -EINVAL;
162 nexthop = inet_opt->opt.faddr;
163 }
164
165 orig_sport = inet->inet_sport;
166 orig_dport = usin->sin_port;
167 fl4 = &inet->cork.fl.u.ip4;
168 rt = ip_route_connect(fl4, nexthop, inet->inet_saddr,
169 RT_CONN_FLAGS(sk), sk->sk_bound_dev_if,
170 IPPROTO_TCP,
171 orig_sport, orig_dport, sk);
172 if (IS_ERR(rt)) {
173 err = PTR_ERR(rt);
174 if (err == -ENETUNREACH)
175 IP_INC_STATS(sock_net(sk), IPSTATS_MIB_OUTNOROUTES);
176 return err;
177 }
178
179 if (rt->rt_flags & (RTCF_MULTICAST | RTCF_BROADCAST)) {
180 ip_rt_put(rt);
181 return -ENETUNREACH;
182 }
183
184 if (!inet_opt || !inet_opt->opt.srr)
185 daddr = fl4->daddr;
186
187 if (!inet->inet_saddr)
188 inet->inet_saddr = fl4->saddr;
189 sk_rcv_saddr_set(sk, inet->inet_saddr);
190
191 if (tp->rx_opt.ts_recent_stamp && inet->inet_daddr != daddr) {
192 /* Reset inherited state */
193 tp->rx_opt.ts_recent = 0;
194 tp->rx_opt.ts_recent_stamp = 0;
195 if (likely(!tp->repair))
196 tp->write_seq = 0;
197 }
198
199 if (tcp_death_row.sysctl_tw_recycle &&
200 !tp->rx_opt.ts_recent_stamp && fl4->daddr == daddr)
201 tcp_fetch_timewait_stamp(sk, &rt->dst);
202
203 inet->inet_dport = usin->sin_port;
204 sk_daddr_set(sk, daddr);
205
206 inet_csk(sk)->icsk_ext_hdr_len = 0;
207 if (inet_opt)
208 inet_csk(sk)->icsk_ext_hdr_len = inet_opt->opt.optlen;
209
210 tp->rx_opt.mss_clamp = TCP_MSS_DEFAULT;
211
212 /* Socket identity is still unknown (sport may be zero).
213 * However we set state to SYN-SENT and not releasing socket
214 * lock select source port, enter ourselves into the hash tables and
215 * complete initialization after this.
216 */
217 tcp_set_state(sk, TCP_SYN_SENT);
218 err = inet_hash_connect(&tcp_death_row, sk);
219 if (err)
220 goto failure;
221
222 sk_set_txhash(sk);
223
224 rt = ip_route_newports(fl4, rt, orig_sport, orig_dport,
225 inet->inet_sport, inet->inet_dport, sk);
226 if (IS_ERR(rt)) {
227 err = PTR_ERR(rt);
228 rt = NULL;
229 goto failure;
230 }
231 /* OK, now commit destination to socket. */
232 sk->sk_gso_type = SKB_GSO_TCPV4;
233 sk_setup_caps(sk, &rt->dst);
234
235 if (!tp->write_seq && likely(!tp->repair))
236 tp->write_seq = secure_tcp_sequence_number(inet->inet_saddr,
237 inet->inet_daddr,
238 inet->inet_sport,
239 usin->sin_port,
240 &tp->tsoffset);
241
242 inet->inet_id = tp->write_seq ^ jiffies;
243
244 err = tcp_connect(sk);
245
246 rt = NULL;
247 if (err)
248 goto failure;
249
250 return 0;
251
252failure:
253 /*
254 * This unhashes the socket and releases the local port,
255 * if necessary.
256 */
257 tcp_set_state(sk, TCP_CLOSE);
258 ip_rt_put(rt);
259 sk->sk_route_caps = 0;
260 inet->inet_dport = 0;
261 return err;
262}
263EXPORT_SYMBOL(tcp_v4_connect);
264
265/*
266 * This routine reacts to ICMP_FRAG_NEEDED mtu indications as defined in RFC1191.
267 * It can be called through tcp_release_cb() if socket was owned by user
268 * at the time tcp_v4_err() was called to handle ICMP message.
269 */
270void tcp_v4_mtu_reduced(struct sock *sk)
271{
272 struct inet_sock *inet = inet_sk(sk);
273 struct dst_entry *dst;
274 u32 mtu;
275
276 if ((1 << sk->sk_state) & (TCPF_LISTEN | TCPF_CLOSE))
277 return;
278 mtu = tcp_sk(sk)->mtu_info;
279 dst = inet_csk_update_pmtu(sk, mtu);
280 if (!dst)
281 return;
282
283 /* Something is about to be wrong... Remember soft error
284 * for the case, if this connection will not able to recover.
285 */
286 if (mtu < dst_mtu(dst) && ip_dont_fragment(sk, dst))
287 sk->sk_err_soft = EMSGSIZE;
288
289 mtu = dst_mtu(dst);
290
291 if (inet->pmtudisc != IP_PMTUDISC_DONT &&
292 ip_sk_accept_pmtu(sk) &&
293 inet_csk(sk)->icsk_pmtu_cookie > mtu) {
294 tcp_sync_mss(sk, mtu);
295
296 /* Resend the TCP packet because it's
297 * clear that the old packet has been
298 * dropped. This is the new "fast" path mtu
299 * discovery.
300 */
301 tcp_simple_retransmit(sk);
302 } /* else let the usual retransmit timer handle it */
303}
304EXPORT_SYMBOL(tcp_v4_mtu_reduced);
305
306static void do_redirect(struct sk_buff *skb, struct sock *sk)
307{
308 struct dst_entry *dst = __sk_dst_check(sk, 0);
309
310 if (dst)
311 dst->ops->redirect(dst, sk, skb);
312}
313
314
315/* handle ICMP messages on TCP_NEW_SYN_RECV request sockets */
316void tcp_req_err(struct sock *sk, u32 seq, bool abort)
317{
318 struct request_sock *req = inet_reqsk(sk);
319 struct net *net = sock_net(sk);
320
321 /* ICMPs are not backlogged, hence we cannot get
322 * an established socket here.
323 */
324 if (seq != tcp_rsk(req)->snt_isn) {
325 __NET_INC_STATS(net, LINUX_MIB_OUTOFWINDOWICMPS);
326 } else if (abort) {
327 /*
328 * Still in SYN_RECV, just remove it silently.
329 * There is no good way to pass the error to the newly
330 * created socket, and POSIX does not want network
331 * errors returned from accept().
332 */
333 inet_csk_reqsk_queue_drop(req->rsk_listener, req);
334 tcp_listendrop(req->rsk_listener);
335 }
336 reqsk_put(req);
337}
338EXPORT_SYMBOL(tcp_req_err);
339
340/*
341 * This routine is called by the ICMP module when it gets some
342 * sort of error condition. If err < 0 then the socket should
343 * be closed and the error returned to the user. If err > 0
344 * it's just the icmp type << 8 | icmp code. After adjustment
345 * header points to the first 8 bytes of the tcp header. We need
346 * to find the appropriate port.
347 *
348 * The locking strategy used here is very "optimistic". When
349 * someone else accesses the socket the ICMP is just dropped
350 * and for some paths there is no check at all.
351 * A more general error queue to queue errors for later handling
352 * is probably better.
353 *
354 */
355
356void tcp_v4_err(struct sk_buff *icmp_skb, u32 info)
357{
358 const struct iphdr *iph = (const struct iphdr *)icmp_skb->data;
359 struct tcphdr *th = (struct tcphdr *)(icmp_skb->data + (iph->ihl << 2));
360 struct inet_connection_sock *icsk;
361 struct tcp_sock *tp;
362 struct inet_sock *inet;
363 const int type = icmp_hdr(icmp_skb)->type;
364 const int code = icmp_hdr(icmp_skb)->code;
365 struct sock *sk;
366 struct sk_buff *skb;
367 struct request_sock *fastopen;
368 __u32 seq, snd_una;
369 __u32 remaining;
370 int err;
371 struct net *net = dev_net(icmp_skb->dev);
372
373 sk = __inet_lookup_established(net, &tcp_hashinfo, iph->daddr,
374 th->dest, iph->saddr, ntohs(th->source),
375 inet_iif(icmp_skb));
376 if (!sk) {
377 __ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
378 return;
379 }
380 if (sk->sk_state == TCP_TIME_WAIT) {
381 inet_twsk_put(inet_twsk(sk));
382 return;
383 }
384 seq = ntohl(th->seq);
385 if (sk->sk_state == TCP_NEW_SYN_RECV)
386 return tcp_req_err(sk, seq,
387 type == ICMP_PARAMETERPROB ||
388 type == ICMP_TIME_EXCEEDED ||
389 (type == ICMP_DEST_UNREACH &&
390 (code == ICMP_NET_UNREACH ||
391 code == ICMP_HOST_UNREACH)));
392
393 bh_lock_sock(sk);
394 /* If too many ICMPs get dropped on busy
395 * servers this needs to be solved differently.
396 * We do take care of PMTU discovery (RFC1191) special case :
397 * we can receive locally generated ICMP messages while socket is held.
398 */
399 if (sock_owned_by_user(sk)) {
400 if (!(type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED))
401 __NET_INC_STATS(net, LINUX_MIB_LOCKDROPPEDICMPS);
402 }
403 if (sk->sk_state == TCP_CLOSE)
404 goto out;
405
406 if (unlikely(iph->ttl < inet_sk(sk)->min_ttl)) {
407 __NET_INC_STATS(net, LINUX_MIB_TCPMINTTLDROP);
408 goto out;
409 }
410
411 icsk = inet_csk(sk);
412 tp = tcp_sk(sk);
413 /* XXX (TFO) - tp->snd_una should be ISN (tcp_create_openreq_child() */
414 fastopen = tp->fastopen_rsk;
415 snd_una = fastopen ? tcp_rsk(fastopen)->snt_isn : tp->snd_una;
416 if (sk->sk_state != TCP_LISTEN &&
417 !between(seq, snd_una, tp->snd_nxt)) {
418 __NET_INC_STATS(net, LINUX_MIB_OUTOFWINDOWICMPS);
419 goto out;
420 }
421
422 switch (type) {
423 case ICMP_REDIRECT:
424 if (!sock_owned_by_user(sk))
425 do_redirect(icmp_skb, sk);
426 goto out;
427 case ICMP_SOURCE_QUENCH:
428 /* Just silently ignore these. */
429 goto out;
430 case ICMP_PARAMETERPROB:
431 err = EPROTO;
432 break;
433 case ICMP_DEST_UNREACH:
434 if (code > NR_ICMP_UNREACH)
435 goto out;
436
437 if (code == ICMP_FRAG_NEEDED) { /* PMTU discovery (RFC1191) */
438 /* We are not interested in TCP_LISTEN and open_requests
439 * (SYN-ACKs send out by Linux are always <576bytes so
440 * they should go through unfragmented).
441 */
442 if (sk->sk_state == TCP_LISTEN)
443 goto out;
444
445 tp->mtu_info = info;
446 if (!sock_owned_by_user(sk)) {
447 tcp_v4_mtu_reduced(sk);
448 } else {
449 if (!test_and_set_bit(TCP_MTU_REDUCED_DEFERRED, &sk->sk_tsq_flags))
450 sock_hold(sk);
451 }
452 goto out;
453 }
454
455 err = icmp_err_convert[code].errno;
456 /* check if icmp_skb allows revert of backoff
457 * (see draft-zimmermann-tcp-lcd) */
458 if (code != ICMP_NET_UNREACH && code != ICMP_HOST_UNREACH)
459 break;
460 if (seq != tp->snd_una || !icsk->icsk_retransmits ||
461 !icsk->icsk_backoff || fastopen)
462 break;
463
464 if (sock_owned_by_user(sk))
465 break;
466
467 icsk->icsk_backoff--;
468 icsk->icsk_rto = tp->srtt_us ? __tcp_set_rto(tp) :
469 TCP_TIMEOUT_INIT;
470 icsk->icsk_rto = inet_csk_rto_backoff(icsk, TCP_RTO_MAX);
471
472 skb = tcp_write_queue_head(sk);
473 BUG_ON(!skb);
474
475 remaining = icsk->icsk_rto -
476 min(icsk->icsk_rto,
477 tcp_time_stamp - tcp_skb_timestamp(skb));
478
479 if (remaining) {
480 inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
481 remaining, TCP_RTO_MAX);
482 } else {
483 /* RTO revert clocked out retransmission.
484 * Will retransmit now */
485 tcp_retransmit_timer(sk);
486 }
487
488 break;
489 case ICMP_TIME_EXCEEDED:
490 err = EHOSTUNREACH;
491 break;
492 default:
493 goto out;
494 }
495
496 switch (sk->sk_state) {
497 case TCP_SYN_SENT:
498 case TCP_SYN_RECV:
499 /* Only in fast or simultaneous open. If a fast open socket is
500 * is already accepted it is treated as a connected one below.
501 */
502 if (fastopen && !fastopen->sk)
503 break;
504
505 if (!sock_owned_by_user(sk)) {
506 sk->sk_err = err;
507
508 sk->sk_error_report(sk);
509
510 tcp_done(sk);
511 } else {
512 sk->sk_err_soft = err;
513 }
514 goto out;
515 }
516
517 /* If we've already connected we will keep trying
518 * until we time out, or the user gives up.
519 *
520 * rfc1122 4.2.3.9 allows to consider as hard errors
521 * only PROTO_UNREACH and PORT_UNREACH (well, FRAG_FAILED too,
522 * but it is obsoleted by pmtu discovery).
523 *
524 * Note, that in modern internet, where routing is unreliable
525 * and in each dark corner broken firewalls sit, sending random
526 * errors ordered by their masters even this two messages finally lose
527 * their original sense (even Linux sends invalid PORT_UNREACHs)
528 *
529 * Now we are in compliance with RFCs.
530 * --ANK (980905)
531 */
532
533 inet = inet_sk(sk);
534 if (!sock_owned_by_user(sk) && inet->recverr) {
535 sk->sk_err = err;
536 sk->sk_error_report(sk);
537 } else { /* Only an error on timeout */
538 sk->sk_err_soft = err;
539 }
540
541out:
542 bh_unlock_sock(sk);
543 sock_put(sk);
544}
545
546void __tcp_v4_send_check(struct sk_buff *skb, __be32 saddr, __be32 daddr)
547{
548 struct tcphdr *th = tcp_hdr(skb);
549
550 if (skb->ip_summed == CHECKSUM_PARTIAL) {
551 th->check = ~tcp_v4_check(skb->len, saddr, daddr, 0);
552 skb->csum_start = skb_transport_header(skb) - skb->head;
553 skb->csum_offset = offsetof(struct tcphdr, check);
554 } else {
555 th->check = tcp_v4_check(skb->len, saddr, daddr,
556 csum_partial(th,
557 th->doff << 2,
558 skb->csum));
559 }
560}
561
562/* This routine computes an IPv4 TCP checksum. */
563void tcp_v4_send_check(struct sock *sk, struct sk_buff *skb)
564{
565 const struct inet_sock *inet = inet_sk(sk);
566
567 __tcp_v4_send_check(skb, inet->inet_saddr, inet->inet_daddr);
568}
569EXPORT_SYMBOL(tcp_v4_send_check);
570
571/*
572 * This routine will send an RST to the other tcp.
573 *
574 * Someone asks: why I NEVER use socket parameters (TOS, TTL etc.)
575 * for reset.
576 * Answer: if a packet caused RST, it is not for a socket
577 * existing in our system, if it is matched to a socket,
578 * it is just duplicate segment or bug in other side's TCP.
579 * So that we build reply only basing on parameters
580 * arrived with segment.
581 * Exception: precedence violation. We do not implement it in any case.
582 */
583
584static void tcp_v4_send_reset(const struct sock *sk, struct sk_buff *skb)
585{
586 const struct tcphdr *th = tcp_hdr(skb);
587 struct {
588 struct tcphdr th;
589#ifdef CONFIG_TCP_MD5SIG
590 __be32 opt[(TCPOLEN_MD5SIG_ALIGNED >> 2)];
591#endif
592 } rep;
593 struct ip_reply_arg arg;
594#ifdef CONFIG_TCP_MD5SIG
595 struct tcp_md5sig_key *key = NULL;
596 const __u8 *hash_location = NULL;
597 unsigned char newhash[16];
598 int genhash;
599 struct sock *sk1 = NULL;
600#endif
601 struct net *net;
602
603 /* Never send a reset in response to a reset. */
604 if (th->rst)
605 return;
606
607 /* If sk not NULL, it means we did a successful lookup and incoming
608 * route had to be correct. prequeue might have dropped our dst.
609 */
610 if (!sk && skb_rtable(skb)->rt_type != RTN_LOCAL)
611 return;
612
613 /* Swap the send and the receive. */
614 memset(&rep, 0, sizeof(rep));
615 rep.th.dest = th->source;
616 rep.th.source = th->dest;
617 rep.th.doff = sizeof(struct tcphdr) / 4;
618 rep.th.rst = 1;
619
620 if (th->ack) {
621 rep.th.seq = th->ack_seq;
622 } else {
623 rep.th.ack = 1;
624 rep.th.ack_seq = htonl(ntohl(th->seq) + th->syn + th->fin +
625 skb->len - (th->doff << 2));
626 }
627
628 memset(&arg, 0, sizeof(arg));
629 arg.iov[0].iov_base = (unsigned char *)&rep;
630 arg.iov[0].iov_len = sizeof(rep.th);
631
632 net = sk ? sock_net(sk) : dev_net(skb_dst(skb)->dev);
633#ifdef CONFIG_TCP_MD5SIG
634 rcu_read_lock();
635 hash_location = tcp_parse_md5sig_option(th);
636 if (sk && sk_fullsock(sk)) {
637 key = tcp_md5_do_lookup(sk, (union tcp_md5_addr *)
638 &ip_hdr(skb)->saddr, AF_INET);
639 } else if (hash_location) {
640 /*
641 * active side is lost. Try to find listening socket through
642 * source port, and then find md5 key through listening socket.
643 * we are not loose security here:
644 * Incoming packet is checked with md5 hash with finding key,
645 * no RST generated if md5 hash doesn't match.
646 */
647 sk1 = __inet_lookup_listener(net, &tcp_hashinfo, NULL, 0,
648 ip_hdr(skb)->saddr,
649 th->source, ip_hdr(skb)->daddr,
650 ntohs(th->source), inet_iif(skb));
651 /* don't send rst if it can't find key */
652 if (!sk1)
653 goto out;
654
655 key = tcp_md5_do_lookup(sk1, (union tcp_md5_addr *)
656 &ip_hdr(skb)->saddr, AF_INET);
657 if (!key)
658 goto out;
659
660
661 genhash = tcp_v4_md5_hash_skb(newhash, key, NULL, skb);
662 if (genhash || memcmp(hash_location, newhash, 16) != 0)
663 goto out;
664
665 }
666
667 if (key) {
668 rep.opt[0] = htonl((TCPOPT_NOP << 24) |
669 (TCPOPT_NOP << 16) |
670 (TCPOPT_MD5SIG << 8) |
671 TCPOLEN_MD5SIG);
672 /* Update length and the length the header thinks exists */
673 arg.iov[0].iov_len += TCPOLEN_MD5SIG_ALIGNED;
674 rep.th.doff = arg.iov[0].iov_len / 4;
675
676 tcp_v4_md5_hash_hdr((__u8 *) &rep.opt[1],
677 key, ip_hdr(skb)->saddr,
678 ip_hdr(skb)->daddr, &rep.th);
679 }
680#endif
681 arg.csum = csum_tcpudp_nofold(ip_hdr(skb)->daddr,
682 ip_hdr(skb)->saddr, /* XXX */
683 arg.iov[0].iov_len, IPPROTO_TCP, 0);
684 arg.csumoffset = offsetof(struct tcphdr, check) / 2;
685 arg.flags = (sk && inet_sk_transparent(sk)) ? IP_REPLY_ARG_NOSRCCHECK : 0;
686
687 /* When socket is gone, all binding information is lost.
688 * routing might fail in this case. No choice here, if we choose to force
689 * input interface, we will misroute in case of asymmetric route.
690 */
691 if (sk)
692 arg.bound_dev_if = sk->sk_bound_dev_if;
693
694 BUILD_BUG_ON(offsetof(struct sock, sk_bound_dev_if) !=
695 offsetof(struct inet_timewait_sock, tw_bound_dev_if));
696
697 arg.tos = ip_hdr(skb)->tos;
698 arg.uid = sock_net_uid(net, sk && sk_fullsock(sk) ? sk : NULL);
699 local_bh_disable();
700 ip_send_unicast_reply(*this_cpu_ptr(net->ipv4.tcp_sk),
701 skb, &TCP_SKB_CB(skb)->header.h4.opt,
702 ip_hdr(skb)->saddr, ip_hdr(skb)->daddr,
703 &arg, arg.iov[0].iov_len);
704
705 __TCP_INC_STATS(net, TCP_MIB_OUTSEGS);
706 __TCP_INC_STATS(net, TCP_MIB_OUTRSTS);
707 local_bh_enable();
708
709#ifdef CONFIG_TCP_MD5SIG
710out:
711 rcu_read_unlock();
712#endif
713}
714
715/* The code following below sending ACKs in SYN-RECV and TIME-WAIT states
716 outside socket context is ugly, certainly. What can I do?
717 */
718
719static void tcp_v4_send_ack(const struct sock *sk,
720 struct sk_buff *skb, u32 seq, u32 ack,
721 u32 win, u32 tsval, u32 tsecr, int oif,
722 struct tcp_md5sig_key *key,
723 int reply_flags, u8 tos)
724{
725 const struct tcphdr *th = tcp_hdr(skb);
726 struct {
727 struct tcphdr th;
728 __be32 opt[(TCPOLEN_TSTAMP_ALIGNED >> 2)
729#ifdef CONFIG_TCP_MD5SIG
730 + (TCPOLEN_MD5SIG_ALIGNED >> 2)
731#endif
732 ];
733 } rep;
734 struct net *net = sock_net(sk);
735 struct ip_reply_arg arg;
736
737 memset(&rep.th, 0, sizeof(struct tcphdr));
738 memset(&arg, 0, sizeof(arg));
739
740 arg.iov[0].iov_base = (unsigned char *)&rep;
741 arg.iov[0].iov_len = sizeof(rep.th);
742 if (tsecr) {
743 rep.opt[0] = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
744 (TCPOPT_TIMESTAMP << 8) |
745 TCPOLEN_TIMESTAMP);
746 rep.opt[1] = htonl(tsval);
747 rep.opt[2] = htonl(tsecr);
748 arg.iov[0].iov_len += TCPOLEN_TSTAMP_ALIGNED;
749 }
750
751 /* Swap the send and the receive. */
752 rep.th.dest = th->source;
753 rep.th.source = th->dest;
754 rep.th.doff = arg.iov[0].iov_len / 4;
755 rep.th.seq = htonl(seq);
756 rep.th.ack_seq = htonl(ack);
757 rep.th.ack = 1;
758 rep.th.window = htons(win);
759
760#ifdef CONFIG_TCP_MD5SIG
761 if (key) {
762 int offset = (tsecr) ? 3 : 0;
763
764 rep.opt[offset++] = htonl((TCPOPT_NOP << 24) |
765 (TCPOPT_NOP << 16) |
766 (TCPOPT_MD5SIG << 8) |
767 TCPOLEN_MD5SIG);
768 arg.iov[0].iov_len += TCPOLEN_MD5SIG_ALIGNED;
769 rep.th.doff = arg.iov[0].iov_len/4;
770
771 tcp_v4_md5_hash_hdr((__u8 *) &rep.opt[offset],
772 key, ip_hdr(skb)->saddr,
773 ip_hdr(skb)->daddr, &rep.th);
774 }
775#endif
776 arg.flags = reply_flags;
777 arg.csum = csum_tcpudp_nofold(ip_hdr(skb)->daddr,
778 ip_hdr(skb)->saddr, /* XXX */
779 arg.iov[0].iov_len, IPPROTO_TCP, 0);
780 arg.csumoffset = offsetof(struct tcphdr, check) / 2;
781 if (oif)
782 arg.bound_dev_if = oif;
783 arg.tos = tos;
784 arg.uid = sock_net_uid(net, sk_fullsock(sk) ? sk : NULL);
785 local_bh_disable();
786 ip_send_unicast_reply(*this_cpu_ptr(net->ipv4.tcp_sk),
787 skb, &TCP_SKB_CB(skb)->header.h4.opt,
788 ip_hdr(skb)->saddr, ip_hdr(skb)->daddr,
789 &arg, arg.iov[0].iov_len);
790
791 __TCP_INC_STATS(net, TCP_MIB_OUTSEGS);
792 local_bh_enable();
793}
794
795static void tcp_v4_timewait_ack(struct sock *sk, struct sk_buff *skb)
796{
797 struct inet_timewait_sock *tw = inet_twsk(sk);
798 struct tcp_timewait_sock *tcptw = tcp_twsk(sk);
799
800 tcp_v4_send_ack(sk, skb,
801 tcptw->tw_snd_nxt, tcptw->tw_rcv_nxt,
802 tcptw->tw_rcv_wnd >> tw->tw_rcv_wscale,
803 tcp_time_stamp + tcptw->tw_ts_offset,
804 tcptw->tw_ts_recent,
805 tw->tw_bound_dev_if,
806 tcp_twsk_md5_key(tcptw),
807 tw->tw_transparent ? IP_REPLY_ARG_NOSRCCHECK : 0,
808 tw->tw_tos
809 );
810
811 inet_twsk_put(tw);
812}
813
814static void tcp_v4_reqsk_send_ack(const struct sock *sk, struct sk_buff *skb,
815 struct request_sock *req)
816{
817 /* sk->sk_state == TCP_LISTEN -> for regular TCP_SYN_RECV
818 * sk->sk_state == TCP_SYN_RECV -> for Fast Open.
819 */
820 u32 seq = (sk->sk_state == TCP_LISTEN) ? tcp_rsk(req)->snt_isn + 1 :
821 tcp_sk(sk)->snd_nxt;
822
823 /* RFC 7323 2.3
824 * The window field (SEG.WND) of every outgoing segment, with the
825 * exception of <SYN> segments, MUST be right-shifted by
826 * Rcv.Wind.Shift bits:
827 */
828 tcp_v4_send_ack(sk, skb, seq,
829 tcp_rsk(req)->rcv_nxt,
830 req->rsk_rcv_wnd >> inet_rsk(req)->rcv_wscale,
831 tcp_time_stamp + tcp_rsk(req)->ts_off,
832 req->ts_recent,
833 0,
834 tcp_md5_do_lookup(sk, (union tcp_md5_addr *)&ip_hdr(skb)->daddr,
835 AF_INET),
836 inet_rsk(req)->no_srccheck ? IP_REPLY_ARG_NOSRCCHECK : 0,
837 ip_hdr(skb)->tos);
838}
839
840/*
841 * Send a SYN-ACK after having received a SYN.
842 * This still operates on a request_sock only, not on a big
843 * socket.
844 */
845static int tcp_v4_send_synack(const struct sock *sk, struct dst_entry *dst,
846 struct flowi *fl,
847 struct request_sock *req,
848 struct tcp_fastopen_cookie *foc,
849 enum tcp_synack_type synack_type)
850{
851 const struct inet_request_sock *ireq = inet_rsk(req);
852 struct flowi4 fl4;
853 int err = -1;
854 struct sk_buff *skb;
855
856 /* First, grab a route. */
857 if (!dst && (dst = inet_csk_route_req(sk, &fl4, req)) == NULL)
858 return -1;
859
860 skb = tcp_make_synack(sk, dst, req, foc, synack_type);
861
862 if (skb) {
863 __tcp_v4_send_check(skb, ireq->ir_loc_addr, ireq->ir_rmt_addr);
864
865 err = ip_build_and_send_pkt(skb, sk, ireq->ir_loc_addr,
866 ireq->ir_rmt_addr,
867 ireq->opt);
868 err = net_xmit_eval(err);
869 }
870
871 return err;
872}
873
874/*
875 * IPv4 request_sock destructor.
876 */
877static void tcp_v4_reqsk_destructor(struct request_sock *req)
878{
879 kfree(inet_rsk(req)->opt);
880}
881
882#ifdef CONFIG_TCP_MD5SIG
883/*
884 * RFC2385 MD5 checksumming requires a mapping of
885 * IP address->MD5 Key.
886 * We need to maintain these in the sk structure.
887 */
888
889/* Find the Key structure for an address. */
890struct tcp_md5sig_key *tcp_md5_do_lookup(const struct sock *sk,
891 const union tcp_md5_addr *addr,
892 int family)
893{
894 const struct tcp_sock *tp = tcp_sk(sk);
895 struct tcp_md5sig_key *key;
896 unsigned int size = sizeof(struct in_addr);
897 const struct tcp_md5sig_info *md5sig;
898
899 /* caller either holds rcu_read_lock() or socket lock */
900 md5sig = rcu_dereference_check(tp->md5sig_info,
901 lockdep_sock_is_held(sk));
902 if (!md5sig)
903 return NULL;
904#if IS_ENABLED(CONFIG_IPV6)
905 if (family == AF_INET6)
906 size = sizeof(struct in6_addr);
907#endif
908 hlist_for_each_entry_rcu(key, &md5sig->head, node) {
909 if (key->family != family)
910 continue;
911 if (!memcmp(&key->addr, addr, size))
912 return key;
913 }
914 return NULL;
915}
916EXPORT_SYMBOL(tcp_md5_do_lookup);
917
918struct tcp_md5sig_key *tcp_v4_md5_lookup(const struct sock *sk,
919 const struct sock *addr_sk)
920{
921 const union tcp_md5_addr *addr;
922
923 addr = (const union tcp_md5_addr *)&addr_sk->sk_daddr;
924 return tcp_md5_do_lookup(sk, addr, AF_INET);
925}
926EXPORT_SYMBOL(tcp_v4_md5_lookup);
927
928/* This can be called on a newly created socket, from other files */
929int tcp_md5_do_add(struct sock *sk, const union tcp_md5_addr *addr,
930 int family, const u8 *newkey, u8 newkeylen, gfp_t gfp)
931{
932 /* Add Key to the list */
933 struct tcp_md5sig_key *key;
934 struct tcp_sock *tp = tcp_sk(sk);
935 struct tcp_md5sig_info *md5sig;
936
937 key = tcp_md5_do_lookup(sk, addr, family);
938 if (key) {
939 /* Pre-existing entry - just update that one. */
940 memcpy(key->key, newkey, newkeylen);
941 key->keylen = newkeylen;
942 return 0;
943 }
944
945 md5sig = rcu_dereference_protected(tp->md5sig_info,
946 lockdep_sock_is_held(sk));
947 if (!md5sig) {
948 md5sig = kmalloc(sizeof(*md5sig), gfp);
949 if (!md5sig)
950 return -ENOMEM;
951
952 sk_nocaps_add(sk, NETIF_F_GSO_MASK);
953 INIT_HLIST_HEAD(&md5sig->head);
954 rcu_assign_pointer(tp->md5sig_info, md5sig);
955 }
956
957 key = sock_kmalloc(sk, sizeof(*key), gfp);
958 if (!key)
959 return -ENOMEM;
960 if (!tcp_alloc_md5sig_pool()) {
961 sock_kfree_s(sk, key, sizeof(*key));
962 return -ENOMEM;
963 }
964
965 memcpy(key->key, newkey, newkeylen);
966 key->keylen = newkeylen;
967 key->family = family;
968 memcpy(&key->addr, addr,
969 (family == AF_INET6) ? sizeof(struct in6_addr) :
970 sizeof(struct in_addr));
971 hlist_add_head_rcu(&key->node, &md5sig->head);
972 return 0;
973}
974EXPORT_SYMBOL(tcp_md5_do_add);
975
976int tcp_md5_do_del(struct sock *sk, const union tcp_md5_addr *addr, int family)
977{
978 struct tcp_md5sig_key *key;
979
980 key = tcp_md5_do_lookup(sk, addr, family);
981 if (!key)
982 return -ENOENT;
983 hlist_del_rcu(&key->node);
984 atomic_sub(sizeof(*key), &sk->sk_omem_alloc);
985 kfree_rcu(key, rcu);
986 return 0;
987}
988EXPORT_SYMBOL(tcp_md5_do_del);
989
990static void tcp_clear_md5_list(struct sock *sk)
991{
992 struct tcp_sock *tp = tcp_sk(sk);
993 struct tcp_md5sig_key *key;
994 struct hlist_node *n;
995 struct tcp_md5sig_info *md5sig;
996
997 md5sig = rcu_dereference_protected(tp->md5sig_info, 1);
998
999 hlist_for_each_entry_safe(key, n, &md5sig->head, node) {
1000 hlist_del_rcu(&key->node);
1001 atomic_sub(sizeof(*key), &sk->sk_omem_alloc);
1002 kfree_rcu(key, rcu);
1003 }
1004}
1005
1006static int tcp_v4_parse_md5_keys(struct sock *sk, char __user *optval,
1007 int optlen)
1008{
1009 struct tcp_md5sig cmd;
1010 struct sockaddr_in *sin = (struct sockaddr_in *)&cmd.tcpm_addr;
1011
1012 if (optlen < sizeof(cmd))
1013 return -EINVAL;
1014
1015 if (copy_from_user(&cmd, optval, sizeof(cmd)))
1016 return -EFAULT;
1017
1018 if (sin->sin_family != AF_INET)
1019 return -EINVAL;
1020
1021 if (!cmd.tcpm_keylen)
1022 return tcp_md5_do_del(sk, (union tcp_md5_addr *)&sin->sin_addr.s_addr,
1023 AF_INET);
1024
1025 if (cmd.tcpm_keylen > TCP_MD5SIG_MAXKEYLEN)
1026 return -EINVAL;
1027
1028 return tcp_md5_do_add(sk, (union tcp_md5_addr *)&sin->sin_addr.s_addr,
1029 AF_INET, cmd.tcpm_key, cmd.tcpm_keylen,
1030 GFP_KERNEL);
1031}
1032
1033static int tcp_v4_md5_hash_headers(struct tcp_md5sig_pool *hp,
1034 __be32 daddr, __be32 saddr,
1035 const struct tcphdr *th, int nbytes)
1036{
1037 struct tcp4_pseudohdr *bp;
1038 struct scatterlist sg;
1039 struct tcphdr *_th;
1040
1041 bp = hp->scratch;
1042 bp->saddr = saddr;
1043 bp->daddr = daddr;
1044 bp->pad = 0;
1045 bp->protocol = IPPROTO_TCP;
1046 bp->len = cpu_to_be16(nbytes);
1047
1048 _th = (struct tcphdr *)(bp + 1);
1049 memcpy(_th, th, sizeof(*th));
1050 _th->check = 0;
1051
1052 sg_init_one(&sg, bp, sizeof(*bp) + sizeof(*th));
1053 ahash_request_set_crypt(hp->md5_req, &sg, NULL,
1054 sizeof(*bp) + sizeof(*th));
1055 return crypto_ahash_update(hp->md5_req);
1056}
1057
1058static int tcp_v4_md5_hash_hdr(char *md5_hash, const struct tcp_md5sig_key *key,
1059 __be32 daddr, __be32 saddr, const struct tcphdr *th)
1060{
1061 struct tcp_md5sig_pool *hp;
1062 struct ahash_request *req;
1063
1064 hp = tcp_get_md5sig_pool();
1065 if (!hp)
1066 goto clear_hash_noput;
1067 req = hp->md5_req;
1068
1069 if (crypto_ahash_init(req))
1070 goto clear_hash;
1071 if (tcp_v4_md5_hash_headers(hp, daddr, saddr, th, th->doff << 2))
1072 goto clear_hash;
1073 if (tcp_md5_hash_key(hp, key))
1074 goto clear_hash;
1075 ahash_request_set_crypt(req, NULL, md5_hash, 0);
1076 if (crypto_ahash_final(req))
1077 goto clear_hash;
1078
1079 tcp_put_md5sig_pool();
1080 return 0;
1081
1082clear_hash:
1083 tcp_put_md5sig_pool();
1084clear_hash_noput:
1085 memset(md5_hash, 0, 16);
1086 return 1;
1087}
1088
1089int tcp_v4_md5_hash_skb(char *md5_hash, const struct tcp_md5sig_key *key,
1090 const struct sock *sk,
1091 const struct sk_buff *skb)
1092{
1093 struct tcp_md5sig_pool *hp;
1094 struct ahash_request *req;
1095 const struct tcphdr *th = tcp_hdr(skb);
1096 __be32 saddr, daddr;
1097
1098 if (sk) { /* valid for establish/request sockets */
1099 saddr = sk->sk_rcv_saddr;
1100 daddr = sk->sk_daddr;
1101 } else {
1102 const struct iphdr *iph = ip_hdr(skb);
1103 saddr = iph->saddr;
1104 daddr = iph->daddr;
1105 }
1106
1107 hp = tcp_get_md5sig_pool();
1108 if (!hp)
1109 goto clear_hash_noput;
1110 req = hp->md5_req;
1111
1112 if (crypto_ahash_init(req))
1113 goto clear_hash;
1114
1115 if (tcp_v4_md5_hash_headers(hp, daddr, saddr, th, skb->len))
1116 goto clear_hash;
1117 if (tcp_md5_hash_skb_data(hp, skb, th->doff << 2))
1118 goto clear_hash;
1119 if (tcp_md5_hash_key(hp, key))
1120 goto clear_hash;
1121 ahash_request_set_crypt(req, NULL, md5_hash, 0);
1122 if (crypto_ahash_final(req))
1123 goto clear_hash;
1124
1125 tcp_put_md5sig_pool();
1126 return 0;
1127
1128clear_hash:
1129 tcp_put_md5sig_pool();
1130clear_hash_noput:
1131 memset(md5_hash, 0, 16);
1132 return 1;
1133}
1134EXPORT_SYMBOL(tcp_v4_md5_hash_skb);
1135
1136#endif
1137
1138/* Called with rcu_read_lock() */
1139static bool tcp_v4_inbound_md5_hash(const struct sock *sk,
1140 const struct sk_buff *skb)
1141{
1142#ifdef CONFIG_TCP_MD5SIG
1143 /*
1144 * This gets called for each TCP segment that arrives
1145 * so we want to be efficient.
1146 * We have 3 drop cases:
1147 * o No MD5 hash and one expected.
1148 * o MD5 hash and we're not expecting one.
1149 * o MD5 hash and its wrong.
1150 */
1151 const __u8 *hash_location = NULL;
1152 struct tcp_md5sig_key *hash_expected;
1153 const struct iphdr *iph = ip_hdr(skb);
1154 const struct tcphdr *th = tcp_hdr(skb);
1155 int genhash;
1156 unsigned char newhash[16];
1157
1158 hash_expected = tcp_md5_do_lookup(sk, (union tcp_md5_addr *)&iph->saddr,
1159 AF_INET);
1160 hash_location = tcp_parse_md5sig_option(th);
1161
1162 /* We've parsed the options - do we have a hash? */
1163 if (!hash_expected && !hash_location)
1164 return false;
1165
1166 if (hash_expected && !hash_location) {
1167 NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPMD5NOTFOUND);
1168 return true;
1169 }
1170
1171 if (!hash_expected && hash_location) {
1172 NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPMD5UNEXPECTED);
1173 return true;
1174 }
1175
1176 /* Okay, so this is hash_expected and hash_location -
1177 * so we need to calculate the checksum.
1178 */
1179 genhash = tcp_v4_md5_hash_skb(newhash,
1180 hash_expected,
1181 NULL, skb);
1182
1183 if (genhash || memcmp(hash_location, newhash, 16) != 0) {
1184 NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPMD5FAILURE);
1185 net_info_ratelimited("MD5 Hash failed for (%pI4, %d)->(%pI4, %d)%s\n",
1186 &iph->saddr, ntohs(th->source),
1187 &iph->daddr, ntohs(th->dest),
1188 genhash ? " tcp_v4_calc_md5_hash failed"
1189 : "");
1190 return true;
1191 }
1192 return false;
1193#endif
1194 return false;
1195}
1196
1197static void tcp_v4_init_req(struct request_sock *req,
1198 const struct sock *sk_listener,
1199 struct sk_buff *skb)
1200{
1201 struct inet_request_sock *ireq = inet_rsk(req);
1202
1203 sk_rcv_saddr_set(req_to_sk(req), ip_hdr(skb)->daddr);
1204 sk_daddr_set(req_to_sk(req), ip_hdr(skb)->saddr);
1205 ireq->opt = tcp_v4_save_options(skb);
1206}
1207
1208static struct dst_entry *tcp_v4_route_req(const struct sock *sk,
1209 struct flowi *fl,
1210 const struct request_sock *req,
1211 bool *strict)
1212{
1213 struct dst_entry *dst = inet_csk_route_req(sk, &fl->u.ip4, req);
1214
1215 if (strict) {
1216 if (fl->u.ip4.daddr == inet_rsk(req)->ir_rmt_addr)
1217 *strict = true;
1218 else
1219 *strict = false;
1220 }
1221
1222 return dst;
1223}
1224
1225struct request_sock_ops tcp_request_sock_ops __read_mostly = {
1226 .family = PF_INET,
1227 .obj_size = sizeof(struct tcp_request_sock),
1228 .rtx_syn_ack = tcp_rtx_synack,
1229 .send_ack = tcp_v4_reqsk_send_ack,
1230 .destructor = tcp_v4_reqsk_destructor,
1231 .send_reset = tcp_v4_send_reset,
1232 .syn_ack_timeout = tcp_syn_ack_timeout,
1233};
1234
1235static const struct tcp_request_sock_ops tcp_request_sock_ipv4_ops = {
1236 .mss_clamp = TCP_MSS_DEFAULT,
1237#ifdef CONFIG_TCP_MD5SIG
1238 .req_md5_lookup = tcp_v4_md5_lookup,
1239 .calc_md5_hash = tcp_v4_md5_hash_skb,
1240#endif
1241 .init_req = tcp_v4_init_req,
1242#ifdef CONFIG_SYN_COOKIES
1243 .cookie_init_seq = cookie_v4_init_sequence,
1244#endif
1245 .route_req = tcp_v4_route_req,
1246 .init_seq = tcp_v4_init_sequence,
1247 .send_synack = tcp_v4_send_synack,
1248};
1249
1250int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb)
1251{
1252 /* Never answer to SYNs send to broadcast or multicast */
1253 if (skb_rtable(skb)->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))
1254 goto drop;
1255
1256 return tcp_conn_request(&tcp_request_sock_ops,
1257 &tcp_request_sock_ipv4_ops, sk, skb);
1258
1259drop:
1260 tcp_listendrop(sk);
1261 return 0;
1262}
1263EXPORT_SYMBOL(tcp_v4_conn_request);
1264
1265
1266/*
1267 * The three way handshake has completed - we got a valid synack -
1268 * now create the new socket.
1269 */
1270struct sock *tcp_v4_syn_recv_sock(const struct sock *sk, struct sk_buff *skb,
1271 struct request_sock *req,
1272 struct dst_entry *dst,
1273 struct request_sock *req_unhash,
1274 bool *own_req)
1275{
1276 struct inet_request_sock *ireq;
1277 struct inet_sock *newinet;
1278 struct tcp_sock *newtp;
1279 struct sock *newsk;
1280#ifdef CONFIG_TCP_MD5SIG
1281 struct tcp_md5sig_key *key;
1282#endif
1283 struct ip_options_rcu *inet_opt;
1284
1285 if (sk_acceptq_is_full(sk))
1286 goto exit_overflow;
1287
1288 newsk = tcp_create_openreq_child(sk, req, skb);
1289 if (!newsk)
1290 goto exit_nonewsk;
1291
1292 newsk->sk_gso_type = SKB_GSO_TCPV4;
1293 inet_sk_rx_dst_set(newsk, skb);
1294
1295 newtp = tcp_sk(newsk);
1296 newinet = inet_sk(newsk);
1297 ireq = inet_rsk(req);
1298 sk_daddr_set(newsk, ireq->ir_rmt_addr);
1299 sk_rcv_saddr_set(newsk, ireq->ir_loc_addr);
1300 newsk->sk_bound_dev_if = ireq->ir_iif;
1301 newinet->inet_saddr = ireq->ir_loc_addr;
1302 inet_opt = ireq->opt;
1303 rcu_assign_pointer(newinet->inet_opt, inet_opt);
1304 ireq->opt = NULL;
1305 newinet->mc_index = inet_iif(skb);
1306 newinet->mc_ttl = ip_hdr(skb)->ttl;
1307 newinet->rcv_tos = ip_hdr(skb)->tos;
1308 inet_csk(newsk)->icsk_ext_hdr_len = 0;
1309 if (inet_opt)
1310 inet_csk(newsk)->icsk_ext_hdr_len = inet_opt->opt.optlen;
1311 newinet->inet_id = newtp->write_seq ^ jiffies;
1312
1313 if (!dst) {
1314 dst = inet_csk_route_child_sock(sk, newsk, req);
1315 if (!dst)
1316 goto put_and_exit;
1317 } else {
1318 /* syncookie case : see end of cookie_v4_check() */
1319 }
1320 sk_setup_caps(newsk, dst);
1321
1322 tcp_ca_openreq_child(newsk, dst);
1323
1324 tcp_sync_mss(newsk, dst_mtu(dst));
1325 newtp->advmss = dst_metric_advmss(dst);
1326 if (tcp_sk(sk)->rx_opt.user_mss &&
1327 tcp_sk(sk)->rx_opt.user_mss < newtp->advmss)
1328 newtp->advmss = tcp_sk(sk)->rx_opt.user_mss;
1329
1330 tcp_initialize_rcv_mss(newsk);
1331
1332#ifdef CONFIG_TCP_MD5SIG
1333 /* Copy over the MD5 key from the original socket */
1334 key = tcp_md5_do_lookup(sk, (union tcp_md5_addr *)&newinet->inet_daddr,
1335 AF_INET);
1336 if (key) {
1337 /*
1338 * We're using one, so create a matching key
1339 * on the newsk structure. If we fail to get
1340 * memory, then we end up not copying the key
1341 * across. Shucks.
1342 */
1343 tcp_md5_do_add(newsk, (union tcp_md5_addr *)&newinet->inet_daddr,
1344 AF_INET, key->key, key->keylen, GFP_ATOMIC);
1345 sk_nocaps_add(newsk, NETIF_F_GSO_MASK);
1346 }
1347#endif
1348
1349 if (__inet_inherit_port(sk, newsk) < 0)
1350 goto put_and_exit;
1351 *own_req = inet_ehash_nolisten(newsk, req_to_sk(req_unhash));
1352 if (*own_req)
1353 tcp_move_syn(newtp, req);
1354
1355 return newsk;
1356
1357exit_overflow:
1358 NET_INC_STATS(sock_net(sk), LINUX_MIB_LISTENOVERFLOWS);
1359exit_nonewsk:
1360 dst_release(dst);
1361exit:
1362 tcp_listendrop(sk);
1363 return NULL;
1364put_and_exit:
1365 inet_csk_prepare_forced_close(newsk);
1366 tcp_done(newsk);
1367 goto exit;
1368}
1369EXPORT_SYMBOL(tcp_v4_syn_recv_sock);
1370
1371static struct sock *tcp_v4_cookie_check(struct sock *sk, struct sk_buff *skb)
1372{
1373#ifdef CONFIG_SYN_COOKIES
1374 const struct tcphdr *th = tcp_hdr(skb);
1375
1376 if (!th->syn)
1377 sk = cookie_v4_check(sk, skb);
1378#endif
1379 return sk;
1380}
1381
1382/* The socket must have it's spinlock held when we get
1383 * here, unless it is a TCP_LISTEN socket.
1384 *
1385 * We have a potential double-lock case here, so even when
1386 * doing backlog processing we use the BH locking scheme.
1387 * This is because we cannot sleep with the original spinlock
1388 * held.
1389 */
1390int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb)
1391{
1392 struct sock *rsk;
1393
1394 if (sk->sk_state == TCP_ESTABLISHED) { /* Fast path */
1395 struct dst_entry *dst = sk->sk_rx_dst;
1396
1397 sock_rps_save_rxhash(sk, skb);
1398 sk_mark_napi_id(sk, skb);
1399 if (dst) {
1400 if (inet_sk(sk)->rx_dst_ifindex != skb->skb_iif ||
1401 !dst->ops->check(dst, 0)) {
1402 dst_release(dst);
1403 sk->sk_rx_dst = NULL;
1404 }
1405 }
1406 tcp_rcv_established(sk, skb, tcp_hdr(skb), skb->len);
1407 return 0;
1408 }
1409
1410 if (tcp_checksum_complete(skb))
1411 goto csum_err;
1412
1413 if (sk->sk_state == TCP_LISTEN) {
1414 struct sock *nsk = tcp_v4_cookie_check(sk, skb);
1415
1416 if (!nsk)
1417 goto discard;
1418 if (nsk != sk) {
1419 sock_rps_save_rxhash(nsk, skb);
1420 sk_mark_napi_id(nsk, skb);
1421 if (tcp_child_process(sk, nsk, skb)) {
1422 rsk = nsk;
1423 goto reset;
1424 }
1425 return 0;
1426 }
1427 } else
1428 sock_rps_save_rxhash(sk, skb);
1429
1430 if (tcp_rcv_state_process(sk, skb)) {
1431 rsk = sk;
1432 goto reset;
1433 }
1434 return 0;
1435
1436reset:
1437 tcp_v4_send_reset(rsk, skb);
1438discard:
1439 kfree_skb(skb);
1440 /* Be careful here. If this function gets more complicated and
1441 * gcc suffers from register pressure on the x86, sk (in %ebx)
1442 * might be destroyed here. This current version compiles correctly,
1443 * but you have been warned.
1444 */
1445 return 0;
1446
1447csum_err:
1448 TCP_INC_STATS(sock_net(sk), TCP_MIB_CSUMERRORS);
1449 TCP_INC_STATS(sock_net(sk), TCP_MIB_INERRS);
1450 goto discard;
1451}
1452EXPORT_SYMBOL(tcp_v4_do_rcv);
1453
1454void tcp_v4_early_demux(struct sk_buff *skb)
1455{
1456 const struct iphdr *iph;
1457 const struct tcphdr *th;
1458 struct sock *sk;
1459
1460 if (skb->pkt_type != PACKET_HOST)
1461 return;
1462
1463 if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct tcphdr)))
1464 return;
1465
1466 iph = ip_hdr(skb);
1467 th = tcp_hdr(skb);
1468
1469 if (th->doff < sizeof(struct tcphdr) / 4)
1470 return;
1471
1472 sk = __inet_lookup_established(dev_net(skb->dev), &tcp_hashinfo,
1473 iph->saddr, th->source,
1474 iph->daddr, ntohs(th->dest),
1475 skb->skb_iif);
1476 if (sk) {
1477 skb->sk = sk;
1478 skb->destructor = sock_edemux;
1479 if (sk_fullsock(sk)) {
1480 struct dst_entry *dst = READ_ONCE(sk->sk_rx_dst);
1481
1482 if (dst)
1483 dst = dst_check(dst, 0);
1484 if (dst &&
1485 inet_sk(sk)->rx_dst_ifindex == skb->skb_iif)
1486 skb_dst_set_noref(skb, dst);
1487 }
1488 }
1489}
1490
1491/* Packet is added to VJ-style prequeue for processing in process
1492 * context, if a reader task is waiting. Apparently, this exciting
1493 * idea (VJ's mail "Re: query about TCP header on tcp-ip" of 07 Sep 93)
1494 * failed somewhere. Latency? Burstiness? Well, at least now we will
1495 * see, why it failed. 8)8) --ANK
1496 *
1497 */
1498bool tcp_prequeue(struct sock *sk, struct sk_buff *skb)
1499{
1500 struct tcp_sock *tp = tcp_sk(sk);
1501
1502 if (sysctl_tcp_low_latency || !tp->ucopy.task)
1503 return false;
1504
1505 if (skb->len <= tcp_hdrlen(skb) &&
1506 skb_queue_len(&tp->ucopy.prequeue) == 0)
1507 return false;
1508
1509 /* Before escaping RCU protected region, we need to take care of skb
1510 * dst. Prequeue is only enabled for established sockets.
1511 * For such sockets, we might need the skb dst only to set sk->sk_rx_dst
1512 * Instead of doing full sk_rx_dst validity here, let's perform
1513 * an optimistic check.
1514 */
1515 if (likely(sk->sk_rx_dst))
1516 skb_dst_drop(skb);
1517 else
1518 skb_dst_force_safe(skb);
1519
1520 __skb_queue_tail(&tp->ucopy.prequeue, skb);
1521 tp->ucopy.memory += skb->truesize;
1522 if (skb_queue_len(&tp->ucopy.prequeue) >= 32 ||
1523 tp->ucopy.memory + atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf) {
1524 struct sk_buff *skb1;
1525
1526 BUG_ON(sock_owned_by_user(sk));
1527 __NET_ADD_STATS(sock_net(sk), LINUX_MIB_TCPPREQUEUEDROPPED,
1528 skb_queue_len(&tp->ucopy.prequeue));
1529
1530 while ((skb1 = __skb_dequeue(&tp->ucopy.prequeue)) != NULL)
1531 sk_backlog_rcv(sk, skb1);
1532
1533 tp->ucopy.memory = 0;
1534 } else if (skb_queue_len(&tp->ucopy.prequeue) == 1) {
1535 wake_up_interruptible_sync_poll(sk_sleep(sk),
1536 POLLIN | POLLRDNORM | POLLRDBAND);
1537 if (!inet_csk_ack_scheduled(sk))
1538 inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK,
1539 (3 * tcp_rto_min(sk)) / 4,
1540 TCP_RTO_MAX);
1541 }
1542 return true;
1543}
1544EXPORT_SYMBOL(tcp_prequeue);
1545
1546bool tcp_add_backlog(struct sock *sk, struct sk_buff *skb)
1547{
1548 u32 limit = sk->sk_rcvbuf + sk->sk_sndbuf;
1549
1550 /* Only socket owner can try to collapse/prune rx queues
1551 * to reduce memory overhead, so add a little headroom here.
1552 * Few sockets backlog are possibly concurrently non empty.
1553 */
1554 limit += 64*1024;
1555
1556 /* In case all data was pulled from skb frags (in __pskb_pull_tail()),
1557 * we can fix skb->truesize to its real value to avoid future drops.
1558 * This is valid because skb is not yet charged to the socket.
1559 * It has been noticed pure SACK packets were sometimes dropped
1560 * (if cooked by drivers without copybreak feature).
1561 */
1562 if (!skb->data_len)
1563 skb->truesize = SKB_TRUESIZE(skb_end_offset(skb));
1564
1565 if (unlikely(sk_add_backlog(sk, skb, limit))) {
1566 bh_unlock_sock(sk);
1567 __NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPBACKLOGDROP);
1568 return true;
1569 }
1570 return false;
1571}
1572EXPORT_SYMBOL(tcp_add_backlog);
1573
1574int tcp_filter(struct sock *sk, struct sk_buff *skb)
1575{
1576 struct tcphdr *th = (struct tcphdr *)skb->data;
1577 unsigned int eaten = skb->len;
1578 int err;
1579
1580 err = sk_filter_trim_cap(sk, skb, th->doff * 4);
1581 if (!err) {
1582 eaten -= skb->len;
1583 TCP_SKB_CB(skb)->end_seq -= eaten;
1584 }
1585 return err;
1586}
1587EXPORT_SYMBOL(tcp_filter);
1588
1589/*
1590 * From tcp_input.c
1591 */
1592
1593int tcp_v4_rcv(struct sk_buff *skb)
1594{
1595 struct net *net = dev_net(skb->dev);
1596 const struct iphdr *iph;
1597 const struct tcphdr *th;
1598 bool refcounted;
1599 struct sock *sk;
1600 int ret;
1601
1602 if (skb->pkt_type != PACKET_HOST)
1603 goto discard_it;
1604
1605 /* Count it even if it's bad */
1606 __TCP_INC_STATS(net, TCP_MIB_INSEGS);
1607
1608 if (!pskb_may_pull(skb, sizeof(struct tcphdr)))
1609 goto discard_it;
1610
1611 th = (const struct tcphdr *)skb->data;
1612
1613 if (unlikely(th->doff < sizeof(struct tcphdr) / 4))
1614 goto bad_packet;
1615 if (!pskb_may_pull(skb, th->doff * 4))
1616 goto discard_it;
1617
1618 /* An explanation is required here, I think.
1619 * Packet length and doff are validated by header prediction,
1620 * provided case of th->doff==0 is eliminated.
1621 * So, we defer the checks. */
1622
1623 if (skb_checksum_init(skb, IPPROTO_TCP, inet_compute_pseudo))
1624 goto csum_error;
1625
1626 th = (const struct tcphdr *)skb->data;
1627 iph = ip_hdr(skb);
1628 /* This is tricky : We move IPCB at its correct location into TCP_SKB_CB()
1629 * barrier() makes sure compiler wont play fool^Waliasing games.
1630 */
1631 memmove(&TCP_SKB_CB(skb)->header.h4, IPCB(skb),
1632 sizeof(struct inet_skb_parm));
1633 barrier();
1634
1635 TCP_SKB_CB(skb)->seq = ntohl(th->seq);
1636 TCP_SKB_CB(skb)->end_seq = (TCP_SKB_CB(skb)->seq + th->syn + th->fin +
1637 skb->len - th->doff * 4);
1638 TCP_SKB_CB(skb)->ack_seq = ntohl(th->ack_seq);
1639 TCP_SKB_CB(skb)->tcp_flags = tcp_flag_byte(th);
1640 TCP_SKB_CB(skb)->tcp_tw_isn = 0;
1641 TCP_SKB_CB(skb)->ip_dsfield = ipv4_get_dsfield(iph);
1642 TCP_SKB_CB(skb)->sacked = 0;
1643
1644lookup:
1645 sk = __inet_lookup_skb(&tcp_hashinfo, skb, __tcp_hdrlen(th), th->source,
1646 th->dest, &refcounted);
1647 if (!sk)
1648 goto no_tcp_socket;
1649
1650process:
1651 if (sk->sk_state == TCP_TIME_WAIT)
1652 goto do_time_wait;
1653
1654 if (sk->sk_state == TCP_NEW_SYN_RECV) {
1655 struct request_sock *req = inet_reqsk(sk);
1656 struct sock *nsk;
1657
1658 sk = req->rsk_listener;
1659 if (unlikely(tcp_v4_inbound_md5_hash(sk, skb))) {
1660 sk_drops_add(sk, skb);
1661 reqsk_put(req);
1662 goto discard_it;
1663 }
1664 if (unlikely(sk->sk_state != TCP_LISTEN)) {
1665 inet_csk_reqsk_queue_drop_and_put(sk, req);
1666 goto lookup;
1667 }
1668 /* We own a reference on the listener, increase it again
1669 * as we might lose it too soon.
1670 */
1671 sock_hold(sk);
1672 refcounted = true;
1673 nsk = tcp_check_req(sk, skb, req, false);
1674 if (!nsk) {
1675 reqsk_put(req);
1676 goto discard_and_relse;
1677 }
1678 if (nsk == sk) {
1679 reqsk_put(req);
1680 } else if (tcp_child_process(sk, nsk, skb)) {
1681 tcp_v4_send_reset(nsk, skb);
1682 goto discard_and_relse;
1683 } else {
1684 sock_put(sk);
1685 return 0;
1686 }
1687 }
1688 if (unlikely(iph->ttl < inet_sk(sk)->min_ttl)) {
1689 __NET_INC_STATS(net, LINUX_MIB_TCPMINTTLDROP);
1690 goto discard_and_relse;
1691 }
1692
1693 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
1694 goto discard_and_relse;
1695
1696 if (tcp_v4_inbound_md5_hash(sk, skb))
1697 goto discard_and_relse;
1698
1699 nf_reset(skb);
1700
1701 if (tcp_filter(sk, skb))
1702 goto discard_and_relse;
1703 th = (const struct tcphdr *)skb->data;
1704 iph = ip_hdr(skb);
1705
1706 skb->dev = NULL;
1707
1708 if (sk->sk_state == TCP_LISTEN) {
1709 ret = tcp_v4_do_rcv(sk, skb);
1710 goto put_and_return;
1711 }
1712
1713 sk_incoming_cpu_update(sk);
1714
1715 bh_lock_sock_nested(sk);
1716 tcp_segs_in(tcp_sk(sk), skb);
1717 ret = 0;
1718 if (!sock_owned_by_user(sk)) {
1719 if (!tcp_prequeue(sk, skb))
1720 ret = tcp_v4_do_rcv(sk, skb);
1721 } else if (tcp_add_backlog(sk, skb)) {
1722 goto discard_and_relse;
1723 }
1724 bh_unlock_sock(sk);
1725
1726put_and_return:
1727 if (refcounted)
1728 sock_put(sk);
1729
1730 return ret;
1731
1732no_tcp_socket:
1733 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
1734 goto discard_it;
1735
1736 if (tcp_checksum_complete(skb)) {
1737csum_error:
1738 __TCP_INC_STATS(net, TCP_MIB_CSUMERRORS);
1739bad_packet:
1740 __TCP_INC_STATS(net, TCP_MIB_INERRS);
1741 } else {
1742 tcp_v4_send_reset(NULL, skb);
1743 }
1744
1745discard_it:
1746 /* Discard frame. */
1747 kfree_skb(skb);
1748 return 0;
1749
1750discard_and_relse:
1751 sk_drops_add(sk, skb);
1752 if (refcounted)
1753 sock_put(sk);
1754 goto discard_it;
1755
1756do_time_wait:
1757 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) {
1758 inet_twsk_put(inet_twsk(sk));
1759 goto discard_it;
1760 }
1761
1762 if (tcp_checksum_complete(skb)) {
1763 inet_twsk_put(inet_twsk(sk));
1764 goto csum_error;
1765 }
1766 switch (tcp_timewait_state_process(inet_twsk(sk), skb, th)) {
1767 case TCP_TW_SYN: {
1768 struct sock *sk2 = inet_lookup_listener(dev_net(skb->dev),
1769 &tcp_hashinfo, skb,
1770 __tcp_hdrlen(th),
1771 iph->saddr, th->source,
1772 iph->daddr, th->dest,
1773 inet_iif(skb));
1774 if (sk2) {
1775 inet_twsk_deschedule_put(inet_twsk(sk));
1776 sk = sk2;
1777 refcounted = false;
1778 goto process;
1779 }
1780 /* Fall through to ACK */
1781 }
1782 case TCP_TW_ACK:
1783 tcp_v4_timewait_ack(sk, skb);
1784 break;
1785 case TCP_TW_RST:
1786 tcp_v4_send_reset(sk, skb);
1787 inet_twsk_deschedule_put(inet_twsk(sk));
1788 goto discard_it;
1789 case TCP_TW_SUCCESS:;
1790 }
1791 goto discard_it;
1792}
1793
1794static struct timewait_sock_ops tcp_timewait_sock_ops = {
1795 .twsk_obj_size = sizeof(struct tcp_timewait_sock),
1796 .twsk_unique = tcp_twsk_unique,
1797 .twsk_destructor= tcp_twsk_destructor,
1798};
1799
1800void inet_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb)
1801{
1802 struct dst_entry *dst = skb_dst(skb);
1803
1804 if (dst && dst_hold_safe(dst)) {
1805 sk->sk_rx_dst = dst;
1806 inet_sk(sk)->rx_dst_ifindex = skb->skb_iif;
1807 }
1808}
1809EXPORT_SYMBOL(inet_sk_rx_dst_set);
1810
1811const struct inet_connection_sock_af_ops ipv4_specific = {
1812 .queue_xmit = ip_queue_xmit,
1813 .send_check = tcp_v4_send_check,
1814 .rebuild_header = inet_sk_rebuild_header,
1815 .sk_rx_dst_set = inet_sk_rx_dst_set,
1816 .conn_request = tcp_v4_conn_request,
1817 .syn_recv_sock = tcp_v4_syn_recv_sock,
1818 .net_header_len = sizeof(struct iphdr),
1819 .setsockopt = ip_setsockopt,
1820 .getsockopt = ip_getsockopt,
1821 .addr2sockaddr = inet_csk_addr2sockaddr,
1822 .sockaddr_len = sizeof(struct sockaddr_in),
1823 .bind_conflict = inet_csk_bind_conflict,
1824#ifdef CONFIG_COMPAT
1825 .compat_setsockopt = compat_ip_setsockopt,
1826 .compat_getsockopt = compat_ip_getsockopt,
1827#endif
1828 .mtu_reduced = tcp_v4_mtu_reduced,
1829};
1830EXPORT_SYMBOL(ipv4_specific);
1831
1832#ifdef CONFIG_TCP_MD5SIG
1833static const struct tcp_sock_af_ops tcp_sock_ipv4_specific = {
1834 .md5_lookup = tcp_v4_md5_lookup,
1835 .calc_md5_hash = tcp_v4_md5_hash_skb,
1836 .md5_parse = tcp_v4_parse_md5_keys,
1837};
1838#endif
1839
1840/* NOTE: A lot of things set to zero explicitly by call to
1841 * sk_alloc() so need not be done here.
1842 */
1843static int tcp_v4_init_sock(struct sock *sk)
1844{
1845 struct inet_connection_sock *icsk = inet_csk(sk);
1846
1847 tcp_init_sock(sk);
1848
1849 icsk->icsk_af_ops = &ipv4_specific;
1850
1851#ifdef CONFIG_TCP_MD5SIG
1852 tcp_sk(sk)->af_specific = &tcp_sock_ipv4_specific;
1853#endif
1854
1855 return 0;
1856}
1857
1858void tcp_v4_destroy_sock(struct sock *sk)
1859{
1860 struct tcp_sock *tp = tcp_sk(sk);
1861
1862 tcp_clear_xmit_timers(sk);
1863
1864 tcp_cleanup_congestion_control(sk);
1865
1866 /* Cleanup up the write buffer. */
1867 tcp_write_queue_purge(sk);
1868
1869 /* Cleans up our, hopefully empty, out_of_order_queue. */
1870 skb_rbtree_purge(&tp->out_of_order_queue);
1871
1872#ifdef CONFIG_TCP_MD5SIG
1873 /* Clean up the MD5 key list, if any */
1874 if (tp->md5sig_info) {
1875 tcp_clear_md5_list(sk);
1876 kfree_rcu(tp->md5sig_info, rcu);
1877 tp->md5sig_info = NULL;
1878 }
1879#endif
1880
1881 /* Clean prequeue, it must be empty really */
1882 __skb_queue_purge(&tp->ucopy.prequeue);
1883
1884 /* Clean up a referenced TCP bind bucket. */
1885 if (inet_csk(sk)->icsk_bind_hash)
1886 inet_put_port(sk);
1887
1888 BUG_ON(tp->fastopen_rsk);
1889
1890 /* If socket is aborted during connect operation */
1891 tcp_free_fastopen_req(tp);
1892 tcp_saved_syn_free(tp);
1893
1894 local_bh_disable();
1895 sk_sockets_allocated_dec(sk);
1896 local_bh_enable();
1897}
1898EXPORT_SYMBOL(tcp_v4_destroy_sock);
1899
1900#ifdef CONFIG_PROC_FS
1901/* Proc filesystem TCP sock list dumping. */
1902
1903/*
1904 * Get next listener socket follow cur. If cur is NULL, get first socket
1905 * starting from bucket given in st->bucket; when st->bucket is zero the
1906 * very first socket in the hash table is returned.
1907 */
1908static void *listening_get_next(struct seq_file *seq, void *cur)
1909{
1910 struct tcp_iter_state *st = seq->private;
1911 struct net *net = seq_file_net(seq);
1912 struct inet_listen_hashbucket *ilb;
1913 struct sock *sk = cur;
1914
1915 if (!sk) {
1916get_head:
1917 ilb = &tcp_hashinfo.listening_hash[st->bucket];
1918 spin_lock(&ilb->lock);
1919 sk = sk_head(&ilb->head);
1920 st->offset = 0;
1921 goto get_sk;
1922 }
1923 ilb = &tcp_hashinfo.listening_hash[st->bucket];
1924 ++st->num;
1925 ++st->offset;
1926
1927 sk = sk_next(sk);
1928get_sk:
1929 sk_for_each_from(sk) {
1930 if (!net_eq(sock_net(sk), net))
1931 continue;
1932 if (sk->sk_family == st->family)
1933 return sk;
1934 }
1935 spin_unlock(&ilb->lock);
1936 st->offset = 0;
1937 if (++st->bucket < INET_LHTABLE_SIZE)
1938 goto get_head;
1939 return NULL;
1940}
1941
1942static void *listening_get_idx(struct seq_file *seq, loff_t *pos)
1943{
1944 struct tcp_iter_state *st = seq->private;
1945 void *rc;
1946
1947 st->bucket = 0;
1948 st->offset = 0;
1949 rc = listening_get_next(seq, NULL);
1950
1951 while (rc && *pos) {
1952 rc = listening_get_next(seq, rc);
1953 --*pos;
1954 }
1955 return rc;
1956}
1957
1958static inline bool empty_bucket(const struct tcp_iter_state *st)
1959{
1960 return hlist_nulls_empty(&tcp_hashinfo.ehash[st->bucket].chain);
1961}
1962
1963/*
1964 * Get first established socket starting from bucket given in st->bucket.
1965 * If st->bucket is zero, the very first socket in the hash is returned.
1966 */
1967static void *established_get_first(struct seq_file *seq)
1968{
1969 struct tcp_iter_state *st = seq->private;
1970 struct net *net = seq_file_net(seq);
1971 void *rc = NULL;
1972
1973 st->offset = 0;
1974 for (; st->bucket <= tcp_hashinfo.ehash_mask; ++st->bucket) {
1975 struct sock *sk;
1976 struct hlist_nulls_node *node;
1977 spinlock_t *lock = inet_ehash_lockp(&tcp_hashinfo, st->bucket);
1978
1979 /* Lockless fast path for the common case of empty buckets */
1980 if (empty_bucket(st))
1981 continue;
1982
1983 spin_lock_bh(lock);
1984 sk_nulls_for_each(sk, node, &tcp_hashinfo.ehash[st->bucket].chain) {
1985 if (sk->sk_family != st->family ||
1986 !net_eq(sock_net(sk), net)) {
1987 continue;
1988 }
1989 rc = sk;
1990 goto out;
1991 }
1992 spin_unlock_bh(lock);
1993 }
1994out:
1995 return rc;
1996}
1997
1998static void *established_get_next(struct seq_file *seq, void *cur)
1999{
2000 struct sock *sk = cur;
2001 struct hlist_nulls_node *node;
2002 struct tcp_iter_state *st = seq->private;
2003 struct net *net = seq_file_net(seq);
2004
2005 ++st->num;
2006 ++st->offset;
2007
2008 sk = sk_nulls_next(sk);
2009
2010 sk_nulls_for_each_from(sk, node) {
2011 if (sk->sk_family == st->family && net_eq(sock_net(sk), net))
2012 return sk;
2013 }
2014
2015 spin_unlock_bh(inet_ehash_lockp(&tcp_hashinfo, st->bucket));
2016 ++st->bucket;
2017 return established_get_first(seq);
2018}
2019
2020static void *established_get_idx(struct seq_file *seq, loff_t pos)
2021{
2022 struct tcp_iter_state *st = seq->private;
2023 void *rc;
2024
2025 st->bucket = 0;
2026 rc = established_get_first(seq);
2027
2028 while (rc && pos) {
2029 rc = established_get_next(seq, rc);
2030 --pos;
2031 }
2032 return rc;
2033}
2034
2035static void *tcp_get_idx(struct seq_file *seq, loff_t pos)
2036{
2037 void *rc;
2038 struct tcp_iter_state *st = seq->private;
2039
2040 st->state = TCP_SEQ_STATE_LISTENING;
2041 rc = listening_get_idx(seq, &pos);
2042
2043 if (!rc) {
2044 st->state = TCP_SEQ_STATE_ESTABLISHED;
2045 rc = established_get_idx(seq, pos);
2046 }
2047
2048 return rc;
2049}
2050
2051static void *tcp_seek_last_pos(struct seq_file *seq)
2052{
2053 struct tcp_iter_state *st = seq->private;
2054 int offset = st->offset;
2055 int orig_num = st->num;
2056 void *rc = NULL;
2057
2058 switch (st->state) {
2059 case TCP_SEQ_STATE_LISTENING:
2060 if (st->bucket >= INET_LHTABLE_SIZE)
2061 break;
2062 st->state = TCP_SEQ_STATE_LISTENING;
2063 rc = listening_get_next(seq, NULL);
2064 while (offset-- && rc)
2065 rc = listening_get_next(seq, rc);
2066 if (rc)
2067 break;
2068 st->bucket = 0;
2069 st->state = TCP_SEQ_STATE_ESTABLISHED;
2070 /* Fallthrough */
2071 case TCP_SEQ_STATE_ESTABLISHED:
2072 if (st->bucket > tcp_hashinfo.ehash_mask)
2073 break;
2074 rc = established_get_first(seq);
2075 while (offset-- && rc)
2076 rc = established_get_next(seq, rc);
2077 }
2078
2079 st->num = orig_num;
2080
2081 return rc;
2082}
2083
2084static void *tcp_seq_start(struct seq_file *seq, loff_t *pos)
2085{
2086 struct tcp_iter_state *st = seq->private;
2087 void *rc;
2088
2089 if (*pos && *pos == st->last_pos) {
2090 rc = tcp_seek_last_pos(seq);
2091 if (rc)
2092 goto out;
2093 }
2094
2095 st->state = TCP_SEQ_STATE_LISTENING;
2096 st->num = 0;
2097 st->bucket = 0;
2098 st->offset = 0;
2099 rc = *pos ? tcp_get_idx(seq, *pos - 1) : SEQ_START_TOKEN;
2100
2101out:
2102 st->last_pos = *pos;
2103 return rc;
2104}
2105
2106static void *tcp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2107{
2108 struct tcp_iter_state *st = seq->private;
2109 void *rc = NULL;
2110
2111 if (v == SEQ_START_TOKEN) {
2112 rc = tcp_get_idx(seq, 0);
2113 goto out;
2114 }
2115
2116 switch (st->state) {
2117 case TCP_SEQ_STATE_LISTENING:
2118 rc = listening_get_next(seq, v);
2119 if (!rc) {
2120 st->state = TCP_SEQ_STATE_ESTABLISHED;
2121 st->bucket = 0;
2122 st->offset = 0;
2123 rc = established_get_first(seq);
2124 }
2125 break;
2126 case TCP_SEQ_STATE_ESTABLISHED:
2127 rc = established_get_next(seq, v);
2128 break;
2129 }
2130out:
2131 ++*pos;
2132 st->last_pos = *pos;
2133 return rc;
2134}
2135
2136static void tcp_seq_stop(struct seq_file *seq, void *v)
2137{
2138 struct tcp_iter_state *st = seq->private;
2139
2140 switch (st->state) {
2141 case TCP_SEQ_STATE_LISTENING:
2142 if (v != SEQ_START_TOKEN)
2143 spin_unlock(&tcp_hashinfo.listening_hash[st->bucket].lock);
2144 break;
2145 case TCP_SEQ_STATE_ESTABLISHED:
2146 if (v)
2147 spin_unlock_bh(inet_ehash_lockp(&tcp_hashinfo, st->bucket));
2148 break;
2149 }
2150}
2151
2152int tcp_seq_open(struct inode *inode, struct file *file)
2153{
2154 struct tcp_seq_afinfo *afinfo = PDE_DATA(inode);
2155 struct tcp_iter_state *s;
2156 int err;
2157
2158 err = seq_open_net(inode, file, &afinfo->seq_ops,
2159 sizeof(struct tcp_iter_state));
2160 if (err < 0)
2161 return err;
2162
2163 s = ((struct seq_file *)file->private_data)->private;
2164 s->family = afinfo->family;
2165 s->last_pos = 0;
2166 return 0;
2167}
2168EXPORT_SYMBOL(tcp_seq_open);
2169
2170int tcp_proc_register(struct net *net, struct tcp_seq_afinfo *afinfo)
2171{
2172 int rc = 0;
2173 struct proc_dir_entry *p;
2174
2175 afinfo->seq_ops.start = tcp_seq_start;
2176 afinfo->seq_ops.next = tcp_seq_next;
2177 afinfo->seq_ops.stop = tcp_seq_stop;
2178
2179 p = proc_create_data(afinfo->name, S_IRUGO, net->proc_net,
2180 afinfo->seq_fops, afinfo);
2181 if (!p)
2182 rc = -ENOMEM;
2183 return rc;
2184}
2185EXPORT_SYMBOL(tcp_proc_register);
2186
2187void tcp_proc_unregister(struct net *net, struct tcp_seq_afinfo *afinfo)
2188{
2189 remove_proc_entry(afinfo->name, net->proc_net);
2190}
2191EXPORT_SYMBOL(tcp_proc_unregister);
2192
2193static void get_openreq4(const struct request_sock *req,
2194 struct seq_file *f, int i)
2195{
2196 const struct inet_request_sock *ireq = inet_rsk(req);
2197 long delta = req->rsk_timer.expires - jiffies;
2198
2199 seq_printf(f, "%4d: %08X:%04X %08X:%04X"
2200 " %02X %08X:%08X %02X:%08lX %08X %5u %8d %u %d %pK",
2201 i,
2202 ireq->ir_loc_addr,
2203 ireq->ir_num,
2204 ireq->ir_rmt_addr,
2205 ntohs(ireq->ir_rmt_port),
2206 TCP_SYN_RECV,
2207 0, 0, /* could print option size, but that is af dependent. */
2208 1, /* timers active (only the expire timer) */
2209 jiffies_delta_to_clock_t(delta),
2210 req->num_timeout,
2211 from_kuid_munged(seq_user_ns(f),
2212 sock_i_uid(req->rsk_listener)),
2213 0, /* non standard timer */
2214 0, /* open_requests have no inode */
2215 0,
2216 req);
2217}
2218
2219static void get_tcp4_sock(struct sock *sk, struct seq_file *f, int i)
2220{
2221 int timer_active;
2222 unsigned long timer_expires;
2223 const struct tcp_sock *tp = tcp_sk(sk);
2224 const struct inet_connection_sock *icsk = inet_csk(sk);
2225 const struct inet_sock *inet = inet_sk(sk);
2226 const struct fastopen_queue *fastopenq = &icsk->icsk_accept_queue.fastopenq;
2227 __be32 dest = inet->inet_daddr;
2228 __be32 src = inet->inet_rcv_saddr;
2229 __u16 destp = ntohs(inet->inet_dport);
2230 __u16 srcp = ntohs(inet->inet_sport);
2231 int rx_queue;
2232 int state;
2233
2234 if (icsk->icsk_pending == ICSK_TIME_RETRANS ||
2235 icsk->icsk_pending == ICSK_TIME_EARLY_RETRANS ||
2236 icsk->icsk_pending == ICSK_TIME_LOSS_PROBE) {
2237 timer_active = 1;
2238 timer_expires = icsk->icsk_timeout;
2239 } else if (icsk->icsk_pending == ICSK_TIME_PROBE0) {
2240 timer_active = 4;
2241 timer_expires = icsk->icsk_timeout;
2242 } else if (timer_pending(&sk->sk_timer)) {
2243 timer_active = 2;
2244 timer_expires = sk->sk_timer.expires;
2245 } else {
2246 timer_active = 0;
2247 timer_expires = jiffies;
2248 }
2249
2250 state = sk_state_load(sk);
2251 if (state == TCP_LISTEN)
2252 rx_queue = sk->sk_ack_backlog;
2253 else
2254 /* Because we don't lock the socket,
2255 * we might find a transient negative value.
2256 */
2257 rx_queue = max_t(int, tp->rcv_nxt - tp->copied_seq, 0);
2258
2259 seq_printf(f, "%4d: %08X:%04X %08X:%04X %02X %08X:%08X %02X:%08lX "
2260 "%08X %5u %8d %lu %d %pK %lu %lu %u %u %d",
2261 i, src, srcp, dest, destp, state,
2262 tp->write_seq - tp->snd_una,
2263 rx_queue,
2264 timer_active,
2265 jiffies_delta_to_clock_t(timer_expires - jiffies),
2266 icsk->icsk_retransmits,
2267 from_kuid_munged(seq_user_ns(f), sock_i_uid(sk)),
2268 icsk->icsk_probes_out,
2269 sock_i_ino(sk),
2270 atomic_read(&sk->sk_refcnt), sk,
2271 jiffies_to_clock_t(icsk->icsk_rto),
2272 jiffies_to_clock_t(icsk->icsk_ack.ato),
2273 (icsk->icsk_ack.quick << 1) | icsk->icsk_ack.pingpong,
2274 tp->snd_cwnd,
2275 state == TCP_LISTEN ?
2276 fastopenq->max_qlen :
2277 (tcp_in_initial_slowstart(tp) ? -1 : tp->snd_ssthresh));
2278}
2279
2280static void get_timewait4_sock(const struct inet_timewait_sock *tw,
2281 struct seq_file *f, int i)
2282{
2283 long delta = tw->tw_timer.expires - jiffies;
2284 __be32 dest, src;
2285 __u16 destp, srcp;
2286
2287 dest = tw->tw_daddr;
2288 src = tw->tw_rcv_saddr;
2289 destp = ntohs(tw->tw_dport);
2290 srcp = ntohs(tw->tw_sport);
2291
2292 seq_printf(f, "%4d: %08X:%04X %08X:%04X"
2293 " %02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %pK",
2294 i, src, srcp, dest, destp, tw->tw_substate, 0, 0,
2295 3, jiffies_delta_to_clock_t(delta), 0, 0, 0, 0,
2296 atomic_read(&tw->tw_refcnt), tw);
2297}
2298
2299#define TMPSZ 150
2300
2301static int tcp4_seq_show(struct seq_file *seq, void *v)
2302{
2303 struct tcp_iter_state *st;
2304 struct sock *sk = v;
2305
2306 seq_setwidth(seq, TMPSZ - 1);
2307 if (v == SEQ_START_TOKEN) {
2308 seq_puts(seq, " sl local_address rem_address st tx_queue "
2309 "rx_queue tr tm->when retrnsmt uid timeout "
2310 "inode");
2311 goto out;
2312 }
2313 st = seq->private;
2314
2315 if (sk->sk_state == TCP_TIME_WAIT)
2316 get_timewait4_sock(v, seq, st->num);
2317 else if (sk->sk_state == TCP_NEW_SYN_RECV)
2318 get_openreq4(v, seq, st->num);
2319 else
2320 get_tcp4_sock(v, seq, st->num);
2321out:
2322 seq_pad(seq, '\n');
2323 return 0;
2324}
2325
2326static const struct file_operations tcp_afinfo_seq_fops = {
2327 .owner = THIS_MODULE,
2328 .open = tcp_seq_open,
2329 .read = seq_read,
2330 .llseek = seq_lseek,
2331 .release = seq_release_net
2332};
2333
2334static struct tcp_seq_afinfo tcp4_seq_afinfo = {
2335 .name = "tcp",
2336 .family = AF_INET,
2337 .seq_fops = &tcp_afinfo_seq_fops,
2338 .seq_ops = {
2339 .show = tcp4_seq_show,
2340 },
2341};
2342
2343static int __net_init tcp4_proc_init_net(struct net *net)
2344{
2345 return tcp_proc_register(net, &tcp4_seq_afinfo);
2346}
2347
2348static void __net_exit tcp4_proc_exit_net(struct net *net)
2349{
2350 tcp_proc_unregister(net, &tcp4_seq_afinfo);
2351}
2352
2353static struct pernet_operations tcp4_net_ops = {
2354 .init = tcp4_proc_init_net,
2355 .exit = tcp4_proc_exit_net,
2356};
2357
2358int __init tcp4_proc_init(void)
2359{
2360 return register_pernet_subsys(&tcp4_net_ops);
2361}
2362
2363void tcp4_proc_exit(void)
2364{
2365 unregister_pernet_subsys(&tcp4_net_ops);
2366}
2367#endif /* CONFIG_PROC_FS */
2368
2369struct proto tcp_prot = {
2370 .name = "TCP",
2371 .owner = THIS_MODULE,
2372 .close = tcp_close,
2373 .connect = tcp_v4_connect,
2374 .disconnect = tcp_disconnect,
2375 .accept = inet_csk_accept,
2376 .ioctl = tcp_ioctl,
2377 .init = tcp_v4_init_sock,
2378 .destroy = tcp_v4_destroy_sock,
2379 .shutdown = tcp_shutdown,
2380 .setsockopt = tcp_setsockopt,
2381 .getsockopt = tcp_getsockopt,
2382 .recvmsg = tcp_recvmsg,
2383 .sendmsg = tcp_sendmsg,
2384 .sendpage = tcp_sendpage,
2385 .backlog_rcv = tcp_v4_do_rcv,
2386 .release_cb = tcp_release_cb,
2387 .hash = inet_hash,
2388 .unhash = inet_unhash,
2389 .get_port = inet_csk_get_port,
2390 .enter_memory_pressure = tcp_enter_memory_pressure,
2391 .stream_memory_free = tcp_stream_memory_free,
2392 .sockets_allocated = &tcp_sockets_allocated,
2393 .orphan_count = &tcp_orphan_count,
2394 .memory_allocated = &tcp_memory_allocated,
2395 .memory_pressure = &tcp_memory_pressure,
2396 .sysctl_mem = sysctl_tcp_mem,
2397 .sysctl_wmem = sysctl_tcp_wmem,
2398 .sysctl_rmem = sysctl_tcp_rmem,
2399 .max_header = MAX_TCP_HEADER,
2400 .obj_size = sizeof(struct tcp_sock),
2401 .slab_flags = SLAB_DESTROY_BY_RCU,
2402 .twsk_prot = &tcp_timewait_sock_ops,
2403 .rsk_prot = &tcp_request_sock_ops,
2404 .h.hashinfo = &tcp_hashinfo,
2405 .no_autobind = true,
2406#ifdef CONFIG_COMPAT
2407 .compat_setsockopt = compat_tcp_setsockopt,
2408 .compat_getsockopt = compat_tcp_getsockopt,
2409#endif
2410 .diag_destroy = tcp_abort,
2411};
2412EXPORT_SYMBOL(tcp_prot);
2413
2414static void __net_exit tcp_sk_exit(struct net *net)
2415{
2416 int cpu;
2417
2418 for_each_possible_cpu(cpu)
2419 inet_ctl_sock_destroy(*per_cpu_ptr(net->ipv4.tcp_sk, cpu));
2420 free_percpu(net->ipv4.tcp_sk);
2421}
2422
2423static int __net_init tcp_sk_init(struct net *net)
2424{
2425 int res, cpu;
2426
2427 net->ipv4.tcp_sk = alloc_percpu(struct sock *);
2428 if (!net->ipv4.tcp_sk)
2429 return -ENOMEM;
2430
2431 for_each_possible_cpu(cpu) {
2432 struct sock *sk;
2433
2434 res = inet_ctl_sock_create(&sk, PF_INET, SOCK_RAW,
2435 IPPROTO_TCP, net);
2436 if (res)
2437 goto fail;
2438 sock_set_flag(sk, SOCK_USE_WRITE_QUEUE);
2439 *per_cpu_ptr(net->ipv4.tcp_sk, cpu) = sk;
2440 }
2441
2442 net->ipv4.sysctl_tcp_ecn = 2;
2443 net->ipv4.sysctl_tcp_ecn_fallback = 1;
2444
2445 net->ipv4.sysctl_tcp_base_mss = TCP_BASE_MSS;
2446 net->ipv4.sysctl_tcp_probe_threshold = TCP_PROBE_THRESHOLD;
2447 net->ipv4.sysctl_tcp_probe_interval = TCP_PROBE_INTERVAL;
2448
2449 net->ipv4.sysctl_tcp_keepalive_time = TCP_KEEPALIVE_TIME;
2450 net->ipv4.sysctl_tcp_keepalive_probes = TCP_KEEPALIVE_PROBES;
2451 net->ipv4.sysctl_tcp_keepalive_intvl = TCP_KEEPALIVE_INTVL;
2452
2453 net->ipv4.sysctl_tcp_syn_retries = TCP_SYN_RETRIES;
2454 net->ipv4.sysctl_tcp_synack_retries = TCP_SYNACK_RETRIES;
2455 net->ipv4.sysctl_tcp_syncookies = 1;
2456 net->ipv4.sysctl_tcp_reordering = TCP_FASTRETRANS_THRESH;
2457 net->ipv4.sysctl_tcp_retries1 = TCP_RETR1;
2458 net->ipv4.sysctl_tcp_retries2 = TCP_RETR2;
2459 net->ipv4.sysctl_tcp_orphan_retries = 0;
2460 net->ipv4.sysctl_tcp_fin_timeout = TCP_FIN_TIMEOUT;
2461 net->ipv4.sysctl_tcp_notsent_lowat = UINT_MAX;
2462 net->ipv4.sysctl_tcp_tw_reuse = 0;
2463
2464 return 0;
2465fail:
2466 tcp_sk_exit(net);
2467
2468 return res;
2469}
2470
2471static void __net_exit tcp_sk_exit_batch(struct list_head *net_exit_list)
2472{
2473 inet_twsk_purge(&tcp_hashinfo, &tcp_death_row, AF_INET);
2474}
2475
2476static struct pernet_operations __net_initdata tcp_sk_ops = {
2477 .init = tcp_sk_init,
2478 .exit = tcp_sk_exit,
2479 .exit_batch = tcp_sk_exit_batch,
2480};
2481
2482void __init tcp_v4_init(void)
2483{
2484 inet_hashinfo_init(&tcp_hashinfo);
2485 if (register_pernet_subsys(&tcp_sk_ops))
2486 panic("Failed to create the TCP control socket.\n");
2487}