?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Merge branch 'fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/evalenti/linux...
[linux/fpc-iii.git] / net / ipv4 / tcp_ipv4.c
blobad450509029bceb74e324c096fbf0e1ab413b7cf
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
87 int sysctl_tcp_tw_reuse __read_mostly;
88 int sysctl_tcp_low_latency __read_mostly;
89 EXPORT_SYMBOL(sysctl_tcp_low_latency);
90
91 #ifdef CONFIG_TCP_MD5SIG
92 static int tcp_v4_md5_hash_hdr(char *md5_hash, const struct tcp_md5sig_key *key,
93 __be32 daddr, __be32 saddr, const struct tcphdr *th);
94 #endif
95
96 struct inet_hashinfo tcp_hashinfo;
97 EXPORT_SYMBOL(tcp_hashinfo);
98
99 static __u32 tcp_v4_init_sequence(const struct sk_buff *skb)
100 {
101 return secure_tcp_sequence_number(ip_hdr(skb)->daddr,
102 ip_hdr(skb)->saddr,
103 tcp_hdr(skb)->dest,
104 tcp_hdr(skb)->source);
105 }
106
107 int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp)
108 {
109 const struct tcp_timewait_sock *tcptw = tcp_twsk(sktw);
110 struct tcp_sock *tp = tcp_sk(sk);
111
112 /* With PAWS, it is safe from the viewpoint
113 of data integrity. Even without PAWS it is safe provided sequence
114 spaces do not overlap i.e. at data rates <= 80Mbit/sec.
115
116 Actually, the idea is close to VJ's one, only timestamp cache is
117 held not per host, but per port pair and TW bucket is used as state
118 holder.
119
120 If TW bucket has been already destroyed we fall back to VJ's scheme
121 and use initial timestamp retrieved from peer table.
122 */
123 if (tcptw->tw_ts_recent_stamp &&
124 (!twp || (sysctl_tcp_tw_reuse &&
125 get_seconds() - tcptw->tw_ts_recent_stamp > 1))) {
126 tp->write_seq = tcptw->tw_snd_nxt + 65535 + 2;
127 if (tp->write_seq == 0)
128 tp->write_seq = 1;
129 tp->rx_opt.ts_recent = tcptw->tw_ts_recent;
130 tp->rx_opt.ts_recent_stamp = tcptw->tw_ts_recent_stamp;
131 sock_hold(sktw);
132 return 1;
133 }
134
135 return 0;
136 }
137 EXPORT_SYMBOL_GPL(tcp_twsk_unique);
138
139 /* This will initiate an outgoing connection. */
140 int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len)
141 {
142 struct sockaddr_in *usin = (struct sockaddr_in *)uaddr;
143 struct inet_sock *inet = inet_sk(sk);
144 struct tcp_sock *tp = tcp_sk(sk);
145 __be16 orig_sport, orig_dport;
146 __be32 daddr, nexthop;
147 struct flowi4 *fl4;
148 struct rtable *rt;
149 int err;
150 struct ip_options_rcu *inet_opt;
151
152 if (addr_len < sizeof(struct sockaddr_in))
153 return -EINVAL;
154
155 if (usin->sin_family != AF_INET)
156 return -EAFNOSUPPORT;
157
158 nexthop = daddr = usin->sin_addr.s_addr;
159 inet_opt = rcu_dereference_protected(inet->inet_opt,
160 sock_owned_by_user(sk));
161 if (inet_opt && inet_opt->opt.srr) {
162 if (!daddr)
163 return -EINVAL;
164 nexthop = inet_opt->opt.faddr;
165 }
166
167 orig_sport = inet->inet_sport;
168 orig_dport = usin->sin_port;
169 fl4 = &inet->cork.fl.u.ip4;
170 rt = ip_route_connect(fl4, nexthop, inet->inet_saddr,
171 RT_CONN_FLAGS(sk), sk->sk_bound_dev_if,
172 IPPROTO_TCP,
173 orig_sport, orig_dport, sk);
174 if (IS_ERR(rt)) {
175 err = PTR_ERR(rt);
176 if (err == -ENETUNREACH)
177 IP_INC_STATS(sock_net(sk), IPSTATS_MIB_OUTNOROUTES);
178 return err;
179 }
180
181 if (rt->rt_flags & (RTCF_MULTICAST | RTCF_BROADCAST)) {
182 ip_rt_put(rt);
183 return -ENETUNREACH;
184 }
185
186 if (!inet_opt || !inet_opt->opt.srr)
187 daddr = fl4->daddr;
188
189 if (!inet->inet_saddr)
190 inet->inet_saddr = fl4->saddr;
191 sk_rcv_saddr_set(sk, inet->inet_saddr);
192
193 if (tp->rx_opt.ts_recent_stamp && inet->inet_daddr != daddr) {
194 /* Reset inherited state */
195 tp->rx_opt.ts_recent = 0;
196 tp->rx_opt.ts_recent_stamp = 0;
197 if (likely(!tp->repair))
198 tp->write_seq = 0;
199 }
200
201 if (tcp_death_row.sysctl_tw_recycle &&
202 !tp->rx_opt.ts_recent_stamp && fl4->daddr == daddr)
203 tcp_fetch_timewait_stamp(sk, &rt->dst);
204
205 inet->inet_dport = usin->sin_port;
206 sk_daddr_set(sk, daddr);
207
208 inet_csk(sk)->icsk_ext_hdr_len = 0;
209 if (inet_opt)
210 inet_csk(sk)->icsk_ext_hdr_len = inet_opt->opt.optlen;
211
212 tp->rx_opt.mss_clamp = TCP_MSS_DEFAULT;
213
214 /* Socket identity is still unknown (sport may be zero).
215 * However we set state to SYN-SENT and not releasing socket
216 * lock select source port, enter ourselves into the hash tables and
217 * complete initialization after this.
218 */
219 tcp_set_state(sk, TCP_SYN_SENT);
220 err = inet_hash_connect(&tcp_death_row, sk);
221 if (err)
222 goto failure;
223
224 sk_set_txhash(sk);
225
226 rt = ip_route_newports(fl4, rt, orig_sport, orig_dport,
227 inet->inet_sport, inet->inet_dport, sk);
228 if (IS_ERR(rt)) {
229 err = PTR_ERR(rt);
230 rt = NULL;
231 goto failure;
232 }
233 /* OK, now commit destination to socket. */
234 sk->sk_gso_type = SKB_GSO_TCPV4;
235 sk_setup_caps(sk, &rt->dst);
236
237 if (!tp->write_seq && likely(!tp->repair))
238 tp->write_seq = secure_tcp_sequence_number(inet->inet_saddr,
239 inet->inet_daddr,
240 inet->inet_sport,
241 usin->sin_port);
242
243 inet->inet_id = tp->write_seq ^ jiffies;
244
245 err = tcp_connect(sk);
246
247 rt = NULL;
248 if (err)
249 goto failure;
250
251 return 0;
252
253 failure:
254 /*
255 * This unhashes the socket and releases the local port,
256 * if necessary.
257 */
258 tcp_set_state(sk, TCP_CLOSE);
259 ip_rt_put(rt);
260 sk->sk_route_caps = 0;
261 inet->inet_dport = 0;
262 return err;
263 }
264 EXPORT_SYMBOL(tcp_v4_connect);
265
266 /*
267 * This routine reacts to ICMP_FRAG_NEEDED mtu indications as defined in RFC1191.
268 * It can be called through tcp_release_cb() if socket was owned by user
269 * at the time tcp_v4_err() was called to handle ICMP message.
270 */
271 void tcp_v4_mtu_reduced(struct sock *sk)
272 {
273 struct dst_entry *dst;
274 struct inet_sock *inet = inet_sk(sk);
275 u32 mtu = tcp_sk(sk)->mtu_info;
276
277 dst = inet_csk_update_pmtu(sk, mtu);
278 if (!dst)
279 return;
280
281 /* Something is about to be wrong... Remember soft error
282 * for the case, if this connection will not able to recover.
283 */
284 if (mtu < dst_mtu(dst) && ip_dont_fragment(sk, dst))
285 sk->sk_err_soft = EMSGSIZE;
286
287 mtu = dst_mtu(dst);
288
289 if (inet->pmtudisc != IP_PMTUDISC_DONT &&
290 ip_sk_accept_pmtu(sk) &&
291 inet_csk(sk)->icsk_pmtu_cookie > mtu) {
292 tcp_sync_mss(sk, mtu);
293
294 /* Resend the TCP packet because it's
295 * clear that the old packet has been
296 * dropped. This is the new "fast" path mtu
297 * discovery.
298 */
299 tcp_simple_retransmit(sk);
300 } /* else let the usual retransmit timer handle it */
301 }
302 EXPORT_SYMBOL(tcp_v4_mtu_reduced);
303
304 static void do_redirect(struct sk_buff *skb, struct sock *sk)
305 {
306 struct dst_entry *dst = __sk_dst_check(sk, 0);
307
308 if (dst)
309 dst->ops->redirect(dst, sk, skb);
310 }
311
312
313 /* handle ICMP messages on TCP_NEW_SYN_RECV request sockets */
314 void tcp_req_err(struct sock *sk, u32 seq, bool abort)
315 {
316 struct request_sock *req = inet_reqsk(sk);
317 struct net *net = sock_net(sk);
318
319 /* ICMPs are not backlogged, hence we cannot get
320 * an established socket here.
321 */
322 if (seq != tcp_rsk(req)->snt_isn) {
323 NET_INC_STATS_BH(net, LINUX_MIB_OUTOFWINDOWICMPS);
324 } else if (abort) {
325 /*
326 * Still in SYN_RECV, just remove it silently.
327 * There is no good way to pass the error to the newly
328 * created socket, and POSIX does not want network
329 * errors returned from accept().
330 */
331 inet_csk_reqsk_queue_drop(req->rsk_listener, req);
332 NET_INC_STATS_BH(net, LINUX_MIB_LISTENDROPS);
333 }
334 reqsk_put(req);
335 }
336 EXPORT_SYMBOL(tcp_req_err);
337
338 /*
339 * This routine is called by the ICMP module when it gets some
340 * sort of error condition. If err < 0 then the socket should
341 * be closed and the error returned to the user. If err > 0
342 * it's just the icmp type << 8 | icmp code. After adjustment
343 * header points to the first 8 bytes of the tcp header. We need
344 * to find the appropriate port.
345 *
346 * The locking strategy used here is very "optimistic". When
347 * someone else accesses the socket the ICMP is just dropped
348 * and for some paths there is no check at all.
349 * A more general error queue to queue errors for later handling
350 * is probably better.
351 *
352 */
353
354 void tcp_v4_err(struct sk_buff *icmp_skb, u32 info)
355 {
356 const struct iphdr *iph = (const struct iphdr *)icmp_skb->data;
357 struct tcphdr *th = (struct tcphdr *)(icmp_skb->data + (iph->ihl << 2));
358 struct inet_connection_sock *icsk;
359 struct tcp_sock *tp;
360 struct inet_sock *inet;
361 const int type = icmp_hdr(icmp_skb)->type;
362 const int code = icmp_hdr(icmp_skb)->code;
363 struct sock *sk;
364 struct sk_buff *skb;
365 struct request_sock *fastopen;
366 __u32 seq, snd_una;
367 __u32 remaining;
368 int err;
369 struct net *net = dev_net(icmp_skb->dev);
370
371 sk = __inet_lookup_established(net, &tcp_hashinfo, iph->daddr,
372 th->dest, iph->saddr, ntohs(th->source),
373 inet_iif(icmp_skb));
374 if (!sk) {
375 ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS);
376 return;
377 }
378 if (sk->sk_state == TCP_TIME_WAIT) {
379 inet_twsk_put(inet_twsk(sk));
380 return;
381 }
382 seq = ntohl(th->seq);
383 if (sk->sk_state == TCP_NEW_SYN_RECV)
384 return tcp_req_err(sk, seq,
385 type == ICMP_PARAMETERPROB ||
386 type == ICMP_TIME_EXCEEDED ||
387 (type == ICMP_DEST_UNREACH &&
388 (code == ICMP_NET_UNREACH ||
389 code == ICMP_HOST_UNREACH)));
390
391 bh_lock_sock(sk);
392 /* If too many ICMPs get dropped on busy
393 * servers this needs to be solved differently.
394 * We do take care of PMTU discovery (RFC1191) special case :
395 * we can receive locally generated ICMP messages while socket is held.
396 */
397 if (sock_owned_by_user(sk)) {
398 if (!(type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED))
399 NET_INC_STATS_BH(net, LINUX_MIB_LOCKDROPPEDICMPS);
400 }
401 if (sk->sk_state == TCP_CLOSE)
402 goto out;
403
404 if (unlikely(iph->ttl < inet_sk(sk)->min_ttl)) {
405 NET_INC_STATS_BH(net, LINUX_MIB_TCPMINTTLDROP);
406 goto out;
407 }
408
409 icsk = inet_csk(sk);
410 tp = tcp_sk(sk);
411 /* XXX (TFO) - tp->snd_una should be ISN (tcp_create_openreq_child() */
412 fastopen = tp->fastopen_rsk;
413 snd_una = fastopen ? tcp_rsk(fastopen)->snt_isn : tp->snd_una;
414 if (sk->sk_state != TCP_LISTEN &&
415 !between(seq, snd_una, tp->snd_nxt)) {
416 NET_INC_STATS_BH(net, LINUX_MIB_OUTOFWINDOWICMPS);
417 goto out;
418 }
419
420 switch (type) {
421 case ICMP_REDIRECT:
422 do_redirect(icmp_skb, sk);
423 goto out;
424 case ICMP_SOURCE_QUENCH:
425 /* Just silently ignore these. */
426 goto out;
427 case ICMP_PARAMETERPROB:
428 err = EPROTO;
429 break;
430 case ICMP_DEST_UNREACH:
431 if (code > NR_ICMP_UNREACH)
432 goto out;
433
434 if (code == ICMP_FRAG_NEEDED) { /* PMTU discovery (RFC1191) */
435 /* We are not interested in TCP_LISTEN and open_requests
436 * (SYN-ACKs send out by Linux are always <576bytes so
437 * they should go through unfragmented).
438 */
439 if (sk->sk_state == TCP_LISTEN)
440 goto out;
441
442 tp->mtu_info = info;
443 if (!sock_owned_by_user(sk)) {
444 tcp_v4_mtu_reduced(sk);
445 } else {
446 if (!test_and_set_bit(TCP_MTU_REDUCED_DEFERRED, &tp->tsq_flags))
447 sock_hold(sk);
448 }
449 goto out;
450 }
451
452 err = icmp_err_convert[code].errno;
453 /* check if icmp_skb allows revert of backoff
454 * (see draft-zimmermann-tcp-lcd) */
455 if (code != ICMP_NET_UNREACH && code != ICMP_HOST_UNREACH)
456 break;
457 if (seq != tp->snd_una || !icsk->icsk_retransmits ||
458 !icsk->icsk_backoff || fastopen)
459 break;
460
461 if (sock_owned_by_user(sk))
462 break;
463
464 icsk->icsk_backoff--;
465 icsk->icsk_rto = tp->srtt_us ? __tcp_set_rto(tp) :
466 TCP_TIMEOUT_INIT;
467 icsk->icsk_rto = inet_csk_rto_backoff(icsk, TCP_RTO_MAX);
468
469 skb = tcp_write_queue_head(sk);
470 BUG_ON(!skb);
471
472 remaining = icsk->icsk_rto -
473 min(icsk->icsk_rto,
474 tcp_time_stamp - tcp_skb_timestamp(skb));
475
476 if (remaining) {
477 inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
478 remaining, TCP_RTO_MAX);
479 } else {
480 /* RTO revert clocked out retransmission.
481 * Will retransmit now */
482 tcp_retransmit_timer(sk);
483 }
484
485 break;
486 case ICMP_TIME_EXCEEDED:
487 err = EHOSTUNREACH;
488 break;
489 default:
490 goto out;
491 }
492
493 switch (sk->sk_state) {
494 case TCP_SYN_SENT:
495 case TCP_SYN_RECV:
496 /* Only in fast or simultaneous open. If a fast open socket is
497 * is already accepted it is treated as a connected one below.
498 */
499 if (fastopen && !fastopen->sk)
500 break;
501
502 if (!sock_owned_by_user(sk)) {
503 sk->sk_err = err;
504
505 sk->sk_error_report(sk);
506
507 tcp_done(sk);
508 } else {
509 sk->sk_err_soft = err;
510 }
511 goto out;
512 }
513
514 /* If we've already connected we will keep trying
515 * until we time out, or the user gives up.
516 *
517 * rfc1122 4.2.3.9 allows to consider as hard errors
518 * only PROTO_UNREACH and PORT_UNREACH (well, FRAG_FAILED too,
519 * but it is obsoleted by pmtu discovery).
520 *
521 * Note, that in modern internet, where routing is unreliable
522 * and in each dark corner broken firewalls sit, sending random
523 * errors ordered by their masters even this two messages finally lose
524 * their original sense (even Linux sends invalid PORT_UNREACHs)
525 *
526 * Now we are in compliance with RFCs.
527 * --ANK (980905)
528 */
529
530 inet = inet_sk(sk);
531 if (!sock_owned_by_user(sk) && inet->recverr) {
532 sk->sk_err = err;
533 sk->sk_error_report(sk);
534 } else { /* Only an error on timeout */
535 sk->sk_err_soft = err;
536 }
537
538 out:
539 bh_unlock_sock(sk);
540 sock_put(sk);
541 }
542
543 void __tcp_v4_send_check(struct sk_buff *skb, __be32 saddr, __be32 daddr)
544 {
545 struct tcphdr *th = tcp_hdr(skb);
546
547 if (skb->ip_summed == CHECKSUM_PARTIAL) {
548 th->check = ~tcp_v4_check(skb->len, saddr, daddr, 0);
549 skb->csum_start = skb_transport_header(skb) - skb->head;
550 skb->csum_offset = offsetof(struct tcphdr, check);
551 } else {
552 th->check = tcp_v4_check(skb->len, saddr, daddr,
553 csum_partial(th,
554 th->doff << 2,
555 skb->csum));
556 }
557 }
558
559 /* This routine computes an IPv4 TCP checksum. */
560 void tcp_v4_send_check(struct sock *sk, struct sk_buff *skb)
561 {
562 const struct inet_sock *inet = inet_sk(sk);
563
564 __tcp_v4_send_check(skb, inet->inet_saddr, inet->inet_daddr);
565 }
566 EXPORT_SYMBOL(tcp_v4_send_check);
567
568 /*
569 * This routine will send an RST to the other tcp.
570 *
571 * Someone asks: why I NEVER use socket parameters (TOS, TTL etc.)
572 * for reset.
573 * Answer: if a packet caused RST, it is not for a socket
574 * existing in our system, if it is matched to a socket,
575 * it is just duplicate segment or bug in other side's TCP.
576 * So that we build reply only basing on parameters
577 * arrived with segment.
578 * Exception: precedence violation. We do not implement it in any case.
579 */
580
581 static void tcp_v4_send_reset(const struct sock *sk, struct sk_buff *skb)
582 {
583 const struct tcphdr *th = tcp_hdr(skb);
584 struct {
585 struct tcphdr th;
586 #ifdef CONFIG_TCP_MD5SIG
587 __be32 opt[(TCPOLEN_MD5SIG_ALIGNED >> 2)];
588 #endif
589 } rep;
590 struct ip_reply_arg arg;
591 #ifdef CONFIG_TCP_MD5SIG
592 struct tcp_md5sig_key *key = NULL;
593 const __u8 *hash_location = NULL;
594 unsigned char newhash[16];
595 int genhash;
596 struct sock *sk1 = NULL;
597 #endif
598 struct net *net;
599
600 /* Never send a reset in response to a reset. */
601 if (th->rst)
602 return;
603
604 /* If sk not NULL, it means we did a successful lookup and incoming
605 * route had to be correct. prequeue might have dropped our dst.
606 */
607 if (!sk && skb_rtable(skb)->rt_type != RTN_LOCAL)
608 return;
609
610 /* Swap the send and the receive. */
611 memset(&rep, 0, sizeof(rep));
612 rep.th.dest = th->source;
613 rep.th.source = th->dest;
614 rep.th.doff = sizeof(struct tcphdr) / 4;
615 rep.th.rst = 1;
616
617 if (th->ack) {
618 rep.th.seq = th->ack_seq;
619 } else {
620 rep.th.ack = 1;
621 rep.th.ack_seq = htonl(ntohl(th->seq) + th->syn + th->fin +
622 skb->len - (th->doff << 2));
623 }
624
625 memset(&arg, 0, sizeof(arg));
626 arg.iov[0].iov_base = (unsigned char *)&rep;
627 arg.iov[0].iov_len = sizeof(rep.th);
628
629 net = sk ? sock_net(sk) : dev_net(skb_dst(skb)->dev);
630 #ifdef CONFIG_TCP_MD5SIG
631 hash_location = tcp_parse_md5sig_option(th);
632 if (sk && sk_fullsock(sk)) {
633 key = tcp_md5_do_lookup(sk, (union tcp_md5_addr *)
634 &ip_hdr(skb)->saddr, AF_INET);
635 } else if (hash_location) {
636 /*
637 * active side is lost. Try to find listening socket through
638 * source port, and then find md5 key through listening socket.
639 * we are not loose security here:
640 * Incoming packet is checked with md5 hash with finding key,
641 * no RST generated if md5 hash doesn't match.
642 */
643 sk1 = __inet_lookup_listener(net, &tcp_hashinfo, NULL, 0,
644 ip_hdr(skb)->saddr,
645 th->source, ip_hdr(skb)->daddr,
646 ntohs(th->source), inet_iif(skb));
647 /* don't send rst if it can't find key */
648 if (!sk1)
649 return;
650 rcu_read_lock();
651 key = tcp_md5_do_lookup(sk1, (union tcp_md5_addr *)
652 &ip_hdr(skb)->saddr, AF_INET);
653 if (!key)
654 goto release_sk1;
655
656 genhash = tcp_v4_md5_hash_skb(newhash, key, NULL, skb);
657 if (genhash || memcmp(hash_location, newhash, 16) != 0)
658 goto release_sk1;
659 }
660
661 if (key) {
662 rep.opt[0] = htonl((TCPOPT_NOP << 24) |
663 (TCPOPT_NOP << 16) |
664 (TCPOPT_MD5SIG << 8) |
665 TCPOLEN_MD5SIG);
666 /* Update length and the length the header thinks exists */
667 arg.iov[0].iov_len += TCPOLEN_MD5SIG_ALIGNED;
668 rep.th.doff = arg.iov[0].iov_len / 4;
669
670 tcp_v4_md5_hash_hdr((__u8 *) &rep.opt[1],
671 key, ip_hdr(skb)->saddr,
672 ip_hdr(skb)->daddr, &rep.th);
673 }
674 #endif
675 arg.csum = csum_tcpudp_nofold(ip_hdr(skb)->daddr,
676 ip_hdr(skb)->saddr, /* XXX */
677 arg.iov[0].iov_len, IPPROTO_TCP, 0);
678 arg.csumoffset = offsetof(struct tcphdr, check) / 2;
679 arg.flags = (sk && inet_sk_transparent(sk)) ? IP_REPLY_ARG_NOSRCCHECK : 0;
680
681 /* When socket is gone, all binding information is lost.
682 * routing might fail in this case. No choice here, if we choose to force
683 * input interface, we will misroute in case of asymmetric route.
684 */
685 if (sk)
686 arg.bound_dev_if = sk->sk_bound_dev_if;
687
688 BUILD_BUG_ON(offsetof(struct sock, sk_bound_dev_if) !=
689 offsetof(struct inet_timewait_sock, tw_bound_dev_if));
690
691 arg.tos = ip_hdr(skb)->tos;
692 ip_send_unicast_reply(*this_cpu_ptr(net->ipv4.tcp_sk),
693 skb, &TCP_SKB_CB(skb)->header.h4.opt,
694 ip_hdr(skb)->saddr, ip_hdr(skb)->daddr,
695 &arg, arg.iov[0].iov_len);
696
697 TCP_INC_STATS_BH(net, TCP_MIB_OUTSEGS);
698 TCP_INC_STATS_BH(net, TCP_MIB_OUTRSTS);
699
700 #ifdef CONFIG_TCP_MD5SIG
701 release_sk1:
702 if (sk1) {
703 rcu_read_unlock();
704 sock_put(sk1);
705 }
706 #endif
707 }
708
709 /* The code following below sending ACKs in SYN-RECV and TIME-WAIT states
710 outside socket context is ugly, certainly. What can I do?
711 */
712
713 static void tcp_v4_send_ack(struct net *net,
714 struct sk_buff *skb, u32 seq, u32 ack,
715 u32 win, u32 tsval, u32 tsecr, int oif,
716 struct tcp_md5sig_key *key,
717 int reply_flags, u8 tos)
718 {
719 const struct tcphdr *th = tcp_hdr(skb);
720 struct {
721 struct tcphdr th;
722 __be32 opt[(TCPOLEN_TSTAMP_ALIGNED >> 2)
723 #ifdef CONFIG_TCP_MD5SIG
724 + (TCPOLEN_MD5SIG_ALIGNED >> 2)
725 #endif
726 ];
727 } rep;
728 struct ip_reply_arg arg;
729
730 memset(&rep.th, 0, sizeof(struct tcphdr));
731 memset(&arg, 0, sizeof(arg));
732
733 arg.iov[0].iov_base = (unsigned char *)&rep;
734 arg.iov[0].iov_len = sizeof(rep.th);
735 if (tsecr) {
736 rep.opt[0] = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
737 (TCPOPT_TIMESTAMP << 8) |
738 TCPOLEN_TIMESTAMP);
739 rep.opt[1] = htonl(tsval);
740 rep.opt[2] = htonl(tsecr);
741 arg.iov[0].iov_len += TCPOLEN_TSTAMP_ALIGNED;
742 }
743
744 /* Swap the send and the receive. */
745 rep.th.dest = th->source;
746 rep.th.source = th->dest;
747 rep.th.doff = arg.iov[0].iov_len / 4;
748 rep.th.seq = htonl(seq);
749 rep.th.ack_seq = htonl(ack);
750 rep.th.ack = 1;
751 rep.th.window = htons(win);
752
753 #ifdef CONFIG_TCP_MD5SIG
754 if (key) {
755 int offset = (tsecr) ? 3 : 0;
756
757 rep.opt[offset++] = htonl((TCPOPT_NOP << 24) |
758 (TCPOPT_NOP << 16) |
759 (TCPOPT_MD5SIG << 8) |
760 TCPOLEN_MD5SIG);
761 arg.iov[0].iov_len += TCPOLEN_MD5SIG_ALIGNED;
762 rep.th.doff = arg.iov[0].iov_len/4;
763
764 tcp_v4_md5_hash_hdr((__u8 *) &rep.opt[offset],
765 key, ip_hdr(skb)->saddr,
766 ip_hdr(skb)->daddr, &rep.th);
767 }
768 #endif
769 arg.flags = reply_flags;
770 arg.csum = csum_tcpudp_nofold(ip_hdr(skb)->daddr,
771 ip_hdr(skb)->saddr, /* XXX */
772 arg.iov[0].iov_len, IPPROTO_TCP, 0);
773 arg.csumoffset = offsetof(struct tcphdr, check) / 2;
774 if (oif)
775 arg.bound_dev_if = oif;
776 arg.tos = tos;
777 ip_send_unicast_reply(*this_cpu_ptr(net->ipv4.tcp_sk),
778 skb, &TCP_SKB_CB(skb)->header.h4.opt,
779 ip_hdr(skb)->saddr, ip_hdr(skb)->daddr,
780 &arg, arg.iov[0].iov_len);
781
782 TCP_INC_STATS_BH(net, TCP_MIB_OUTSEGS);
783 }
784
785 static void tcp_v4_timewait_ack(struct sock *sk, struct sk_buff *skb)
786 {
787 struct inet_timewait_sock *tw = inet_twsk(sk);
788 struct tcp_timewait_sock *tcptw = tcp_twsk(sk);
789
790 tcp_v4_send_ack(sock_net(sk), skb,
791 tcptw->tw_snd_nxt, tcptw->tw_rcv_nxt,
792 tcptw->tw_rcv_wnd >> tw->tw_rcv_wscale,
793 tcp_time_stamp + tcptw->tw_ts_offset,
794 tcptw->tw_ts_recent,
795 tw->tw_bound_dev_if,
796 tcp_twsk_md5_key(tcptw),
797 tw->tw_transparent ? IP_REPLY_ARG_NOSRCCHECK : 0,
798 tw->tw_tos
799 );
800
801 inet_twsk_put(tw);
802 }
803
804 static void tcp_v4_reqsk_send_ack(const struct sock *sk, struct sk_buff *skb,
805 struct request_sock *req)
806 {
807 /* sk->sk_state == TCP_LISTEN -> for regular TCP_SYN_RECV
808 * sk->sk_state == TCP_SYN_RECV -> for Fast Open.
809 */
810 u32 seq = (sk->sk_state == TCP_LISTEN) ? tcp_rsk(req)->snt_isn + 1 :
811 tcp_sk(sk)->snd_nxt;
812
813 tcp_v4_send_ack(sock_net(sk), skb, seq,
814 tcp_rsk(req)->rcv_nxt, req->rsk_rcv_wnd,
815 tcp_time_stamp,
816 req->ts_recent,
817 0,
818 tcp_md5_do_lookup(sk, (union tcp_md5_addr *)&ip_hdr(skb)->daddr,
819 AF_INET),
820 inet_rsk(req)->no_srccheck ? IP_REPLY_ARG_NOSRCCHECK : 0,
821 ip_hdr(skb)->tos);
822 }
823
824 /*
825 * Send a SYN-ACK after having received a SYN.
826 * This still operates on a request_sock only, not on a big
827 * socket.
828 */
829 static int tcp_v4_send_synack(const struct sock *sk, struct dst_entry *dst,
830 struct flowi *fl,
831 struct request_sock *req,
832 struct tcp_fastopen_cookie *foc,
833 bool attach_req)
834 {
835 const struct inet_request_sock *ireq = inet_rsk(req);
836 struct flowi4 fl4;
837 int err = -1;
838 struct sk_buff *skb;
839
840 /* First, grab a route. */
841 if (!dst && (dst = inet_csk_route_req(sk, &fl4, req)) == NULL)
842 return -1;
843
844 skb = tcp_make_synack(sk, dst, req, foc, attach_req);
845
846 if (skb) {
847 __tcp_v4_send_check(skb, ireq->ir_loc_addr, ireq->ir_rmt_addr);
848
849 err = ip_build_and_send_pkt(skb, sk, ireq->ir_loc_addr,
850 ireq->ir_rmt_addr,
851 ireq->opt);
852 err = net_xmit_eval(err);
853 }
854
855 return err;
856 }
857
858 /*
859 * IPv4 request_sock destructor.
860 */
861 static void tcp_v4_reqsk_destructor(struct request_sock *req)
862 {
863 kfree(inet_rsk(req)->opt);
864 }
865
866 #ifdef CONFIG_TCP_MD5SIG
867 /*
868 * RFC2385 MD5 checksumming requires a mapping of
869 * IP address->MD5 Key.
870 * We need to maintain these in the sk structure.
871 */
872
873 /* Find the Key structure for an address. */
874 struct tcp_md5sig_key *tcp_md5_do_lookup(const struct sock *sk,
875 const union tcp_md5_addr *addr,
876 int family)
877 {
878 const struct tcp_sock *tp = tcp_sk(sk);
879 struct tcp_md5sig_key *key;
880 unsigned int size = sizeof(struct in_addr);
881 const struct tcp_md5sig_info *md5sig;
882
883 /* caller either holds rcu_read_lock() or socket lock */
884 md5sig = rcu_dereference_check(tp->md5sig_info,
885 sock_owned_by_user(sk) ||
886 lockdep_is_held((spinlock_t *)&sk->sk_lock.slock));
887 if (!md5sig)
888 return NULL;
889 #if IS_ENABLED(CONFIG_IPV6)
890 if (family == AF_INET6)
891 size = sizeof(struct in6_addr);
892 #endif
893 hlist_for_each_entry_rcu(key, &md5sig->head, node) {
894 if (key->family != family)
895 continue;
896 if (!memcmp(&key->addr, addr, size))
897 return key;
898 }
899 return NULL;
900 }
901 EXPORT_SYMBOL(tcp_md5_do_lookup);
902
903 struct tcp_md5sig_key *tcp_v4_md5_lookup(const struct sock *sk,
904 const struct sock *addr_sk)
905 {
906 const union tcp_md5_addr *addr;
907
908 addr = (const union tcp_md5_addr *)&addr_sk->sk_daddr;
909 return tcp_md5_do_lookup(sk, addr, AF_INET);
910 }
911 EXPORT_SYMBOL(tcp_v4_md5_lookup);
912
913 /* This can be called on a newly created socket, from other files */
914 int tcp_md5_do_add(struct sock *sk, const union tcp_md5_addr *addr,
915 int family, const u8 *newkey, u8 newkeylen, gfp_t gfp)
916 {
917 /* Add Key to the list */
918 struct tcp_md5sig_key *key;
919 struct tcp_sock *tp = tcp_sk(sk);
920 struct tcp_md5sig_info *md5sig;
921
922 key = tcp_md5_do_lookup(sk, addr, family);
923 if (key) {
924 /* Pre-existing entry - just update that one. */
925 memcpy(key->key, newkey, newkeylen);
926 key->keylen = newkeylen;
927 return 0;
928 }
929
930 md5sig = rcu_dereference_protected(tp->md5sig_info,
931 sock_owned_by_user(sk) ||
932 lockdep_is_held(&sk->sk_lock.slock));
933 if (!md5sig) {
934 md5sig = kmalloc(sizeof(*md5sig), gfp);
935 if (!md5sig)
936 return -ENOMEM;
937
938 sk_nocaps_add(sk, NETIF_F_GSO_MASK);
939 INIT_HLIST_HEAD(&md5sig->head);
940 rcu_assign_pointer(tp->md5sig_info, md5sig);
941 }
942
943 key = sock_kmalloc(sk, sizeof(*key), gfp);
944 if (!key)
945 return -ENOMEM;
946 if (!tcp_alloc_md5sig_pool()) {
947 sock_kfree_s(sk, key, sizeof(*key));
948 return -ENOMEM;
949 }
950
951 memcpy(key->key, newkey, newkeylen);
952 key->keylen = newkeylen;
953 key->family = family;
954 memcpy(&key->addr, addr,
955 (family == AF_INET6) ? sizeof(struct in6_addr) :
956 sizeof(struct in_addr));
957 hlist_add_head_rcu(&key->node, &md5sig->head);
958 return 0;
959 }
960 EXPORT_SYMBOL(tcp_md5_do_add);
961
962 int tcp_md5_do_del(struct sock *sk, const union tcp_md5_addr *addr, int family)
963 {
964 struct tcp_md5sig_key *key;
965
966 key = tcp_md5_do_lookup(sk, addr, family);
967 if (!key)
968 return -ENOENT;
969 hlist_del_rcu(&key->node);
970 atomic_sub(sizeof(*key), &sk->sk_omem_alloc);
971 kfree_rcu(key, rcu);
972 return 0;
973 }
974 EXPORT_SYMBOL(tcp_md5_do_del);
975
976 static void tcp_clear_md5_list(struct sock *sk)
977 {
978 struct tcp_sock *tp = tcp_sk(sk);
979 struct tcp_md5sig_key *key;
980 struct hlist_node *n;
981 struct tcp_md5sig_info *md5sig;
982
983 md5sig = rcu_dereference_protected(tp->md5sig_info, 1);
984
985 hlist_for_each_entry_safe(key, n, &md5sig->head, node) {
986 hlist_del_rcu(&key->node);
987 atomic_sub(sizeof(*key), &sk->sk_omem_alloc);
988 kfree_rcu(key, rcu);
989 }
990 }
991
992 static int tcp_v4_parse_md5_keys(struct sock *sk, char __user *optval,
993 int optlen)
994 {
995 struct tcp_md5sig cmd;
996 struct sockaddr_in *sin = (struct sockaddr_in *)&cmd.tcpm_addr;
997
998 if (optlen < sizeof(cmd))
999 return -EINVAL;
1000
1001 if (copy_from_user(&cmd, optval, sizeof(cmd)))
1002 return -EFAULT;
1003
1004 if (sin->sin_family != AF_INET)
1005 return -EINVAL;
1006
1007 if (!cmd.tcpm_keylen)
1008 return tcp_md5_do_del(sk, (union tcp_md5_addr *)&sin->sin_addr.s_addr,
1009 AF_INET);
1010
1011 if (cmd.tcpm_keylen > TCP_MD5SIG_MAXKEYLEN)
1012 return -EINVAL;
1013
1014 return tcp_md5_do_add(sk, (union tcp_md5_addr *)&sin->sin_addr.s_addr,
1015 AF_INET, cmd.tcpm_key, cmd.tcpm_keylen,
1016 GFP_KERNEL);
1017 }
1018
1019 static int tcp_v4_md5_hash_pseudoheader(struct tcp_md5sig_pool *hp,
1020 __be32 daddr, __be32 saddr, int nbytes)
1021 {
1022 struct tcp4_pseudohdr *bp;
1023 struct scatterlist sg;
1024
1025 bp = &hp->md5_blk.ip4;
1026
1027 /*
1028 * 1. the TCP pseudo-header (in the order: source IP address,
1029 * destination IP address, zero-padded protocol number, and
1030 * segment length)
1031 */
1032 bp->saddr = saddr;
1033 bp->daddr = daddr;
1034 bp->pad = 0;
1035 bp->protocol = IPPROTO_TCP;
1036 bp->len = cpu_to_be16(nbytes);
1037
1038 sg_init_one(&sg, bp, sizeof(*bp));
1039 ahash_request_set_crypt(hp->md5_req, &sg, NULL, sizeof(*bp));
1040 return crypto_ahash_update(hp->md5_req);
1041 }
1042
1043 static int tcp_v4_md5_hash_hdr(char *md5_hash, const struct tcp_md5sig_key *key,
1044 __be32 daddr, __be32 saddr, const struct tcphdr *th)
1045 {
1046 struct tcp_md5sig_pool *hp;
1047 struct ahash_request *req;
1048
1049 hp = tcp_get_md5sig_pool();
1050 if (!hp)
1051 goto clear_hash_noput;
1052 req = hp->md5_req;
1053
1054 if (crypto_ahash_init(req))
1055 goto clear_hash;
1056 if (tcp_v4_md5_hash_pseudoheader(hp, daddr, saddr, th->doff << 2))
1057 goto clear_hash;
1058 if (tcp_md5_hash_header(hp, th))
1059 goto clear_hash;
1060 if (tcp_md5_hash_key(hp, key))
1061 goto clear_hash;
1062 ahash_request_set_crypt(req, NULL, md5_hash, 0);
1063 if (crypto_ahash_final(req))
1064 goto clear_hash;
1065
1066 tcp_put_md5sig_pool();
1067 return 0;
1068
1069 clear_hash:
1070 tcp_put_md5sig_pool();
1071 clear_hash_noput:
1072 memset(md5_hash, 0, 16);
1073 return 1;
1074 }
1075
1076 int tcp_v4_md5_hash_skb(char *md5_hash, const struct tcp_md5sig_key *key,
1077 const struct sock *sk,
1078 const struct sk_buff *skb)
1079 {
1080 struct tcp_md5sig_pool *hp;
1081 struct ahash_request *req;
1082 const struct tcphdr *th = tcp_hdr(skb);
1083 __be32 saddr, daddr;
1084
1085 if (sk) { /* valid for establish/request sockets */
1086 saddr = sk->sk_rcv_saddr;
1087 daddr = sk->sk_daddr;
1088 } else {
1089 const struct iphdr *iph = ip_hdr(skb);
1090 saddr = iph->saddr;
1091 daddr = iph->daddr;
1092 }
1093
1094 hp = tcp_get_md5sig_pool();
1095 if (!hp)
1096 goto clear_hash_noput;
1097 req = hp->md5_req;
1098
1099 if (crypto_ahash_init(req))
1100 goto clear_hash;
1101
1102 if (tcp_v4_md5_hash_pseudoheader(hp, daddr, saddr, skb->len))
1103 goto clear_hash;
1104 if (tcp_md5_hash_header(hp, th))
1105 goto clear_hash;
1106 if (tcp_md5_hash_skb_data(hp, skb, th->doff << 2))
1107 goto clear_hash;
1108 if (tcp_md5_hash_key(hp, key))
1109 goto clear_hash;
1110 ahash_request_set_crypt(req, NULL, md5_hash, 0);
1111 if (crypto_ahash_final(req))
1112 goto clear_hash;
1113
1114 tcp_put_md5sig_pool();
1115 return 0;
1116
1117 clear_hash:
1118 tcp_put_md5sig_pool();
1119 clear_hash_noput:
1120 memset(md5_hash, 0, 16);
1121 return 1;
1122 }
1123 EXPORT_SYMBOL(tcp_v4_md5_hash_skb);
1124
1125 #endif
1126
1127 /* Called with rcu_read_lock() */
1128 static bool tcp_v4_inbound_md5_hash(const struct sock *sk,
1129 const struct sk_buff *skb)
1130 {
1131 #ifdef CONFIG_TCP_MD5SIG
1132 /*
1133 * This gets called for each TCP segment that arrives
1134 * so we want to be efficient.
1135 * We have 3 drop cases:
1136 * o No MD5 hash and one expected.
1137 * o MD5 hash and we're not expecting one.
1138 * o MD5 hash and its wrong.
1139 */
1140 const __u8 *hash_location = NULL;
1141 struct tcp_md5sig_key *hash_expected;
1142 const struct iphdr *iph = ip_hdr(skb);
1143 const struct tcphdr *th = tcp_hdr(skb);
1144 int genhash;
1145 unsigned char newhash[16];
1146
1147 hash_expected = tcp_md5_do_lookup(sk, (union tcp_md5_addr *)&iph->saddr,
1148 AF_INET);
1149 hash_location = tcp_parse_md5sig_option(th);
1150
1151 /* We've parsed the options - do we have a hash? */
1152 if (!hash_expected && !hash_location)
1153 return false;
1154
1155 if (hash_expected && !hash_location) {
1156 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPMD5NOTFOUND);
1157 return true;
1158 }
1159
1160 if (!hash_expected && hash_location) {
1161 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPMD5UNEXPECTED);
1162 return true;
1163 }
1164
1165 /* Okay, so this is hash_expected and hash_location -
1166 * so we need to calculate the checksum.
1167 */
1168 genhash = tcp_v4_md5_hash_skb(newhash,
1169 hash_expected,
1170 NULL, skb);
1171
1172 if (genhash || memcmp(hash_location, newhash, 16) != 0) {
1173 net_info_ratelimited("MD5 Hash failed for (%pI4, %d)->(%pI4, %d)%s\n",
1174 &iph->saddr, ntohs(th->source),
1175 &iph->daddr, ntohs(th->dest),
1176 genhash ? " tcp_v4_calc_md5_hash failed"
1177 : "");
1178 return true;
1179 }
1180 return false;
1181 #endif
1182 return false;
1183 }
1184
1185 static void tcp_v4_init_req(struct request_sock *req,
1186 const struct sock *sk_listener,
1187 struct sk_buff *skb)
1188 {
1189 struct inet_request_sock *ireq = inet_rsk(req);
1190
1191 sk_rcv_saddr_set(req_to_sk(req), ip_hdr(skb)->daddr);
1192 sk_daddr_set(req_to_sk(req), ip_hdr(skb)->saddr);
1193 ireq->no_srccheck = inet_sk(sk_listener)->transparent;
1194 ireq->opt = tcp_v4_save_options(skb);
1195 }
1196
1197 static struct dst_entry *tcp_v4_route_req(const struct sock *sk,
1198 struct flowi *fl,
1199 const struct request_sock *req,
1200 bool *strict)
1201 {
1202 struct dst_entry *dst = inet_csk_route_req(sk, &fl->u.ip4, req);
1203
1204 if (strict) {
1205 if (fl->u.ip4.daddr == inet_rsk(req)->ir_rmt_addr)
1206 *strict = true;
1207 else
1208 *strict = false;
1209 }
1210
1211 return dst;
1212 }
1213
1214 struct request_sock_ops tcp_request_sock_ops __read_mostly = {
1215 .family = PF_INET,
1216 .obj_size = sizeof(struct tcp_request_sock),
1217 .rtx_syn_ack = tcp_rtx_synack,
1218 .send_ack = tcp_v4_reqsk_send_ack,
1219 .destructor = tcp_v4_reqsk_destructor,
1220 .send_reset = tcp_v4_send_reset,
1221 .syn_ack_timeout = tcp_syn_ack_timeout,
1222 };
1223
1224 static const struct tcp_request_sock_ops tcp_request_sock_ipv4_ops = {
1225 .mss_clamp = TCP_MSS_DEFAULT,
1226 #ifdef CONFIG_TCP_MD5SIG
1227 .req_md5_lookup = tcp_v4_md5_lookup,
1228 .calc_md5_hash = tcp_v4_md5_hash_skb,
1229 #endif
1230 .init_req = tcp_v4_init_req,
1231 #ifdef CONFIG_SYN_COOKIES
1232 .cookie_init_seq = cookie_v4_init_sequence,
1233 #endif
1234 .route_req = tcp_v4_route_req,
1235 .init_seq = tcp_v4_init_sequence,
1236 .send_synack = tcp_v4_send_synack,
1237 };
1238
1239 int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb)
1240 {
1241 /* Never answer to SYNs send to broadcast or multicast */
1242 if (skb_rtable(skb)->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))
1243 goto drop;
1244
1245 return tcp_conn_request(&tcp_request_sock_ops,
1246 &tcp_request_sock_ipv4_ops, sk, skb);
1247
1248 drop:
1249 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENDROPS);
1250 return 0;
1251 }
1252 EXPORT_SYMBOL(tcp_v4_conn_request);
1253
1254
1255 /*
1256 * The three way handshake has completed - we got a valid synack -
1257 * now create the new socket.
1258 */
1259 struct sock *tcp_v4_syn_recv_sock(const struct sock *sk, struct sk_buff *skb,
1260 struct request_sock *req,
1261 struct dst_entry *dst,
1262 struct request_sock *req_unhash,
1263 bool *own_req)
1264 {
1265 struct inet_request_sock *ireq;
1266 struct inet_sock *newinet;
1267 struct tcp_sock *newtp;
1268 struct sock *newsk;
1269 #ifdef CONFIG_TCP_MD5SIG
1270 struct tcp_md5sig_key *key;
1271 #endif
1272 struct ip_options_rcu *inet_opt;
1273
1274 if (sk_acceptq_is_full(sk))
1275 goto exit_overflow;
1276
1277 newsk = tcp_create_openreq_child(sk, req, skb);
1278 if (!newsk)
1279 goto exit_nonewsk;
1280
1281 newsk->sk_gso_type = SKB_GSO_TCPV4;
1282 inet_sk_rx_dst_set(newsk, skb);
1283
1284 newtp = tcp_sk(newsk);
1285 newinet = inet_sk(newsk);
1286 ireq = inet_rsk(req);
1287 sk_daddr_set(newsk, ireq->ir_rmt_addr);
1288 sk_rcv_saddr_set(newsk, ireq->ir_loc_addr);
1289 newsk->sk_bound_dev_if = ireq->ir_iif;
1290 newinet->inet_saddr = ireq->ir_loc_addr;
1291 inet_opt = ireq->opt;
1292 rcu_assign_pointer(newinet->inet_opt, inet_opt);
1293 ireq->opt = NULL;
1294 newinet->mc_index = inet_iif(skb);
1295 newinet->mc_ttl = ip_hdr(skb)->ttl;
1296 newinet->rcv_tos = ip_hdr(skb)->tos;
1297 inet_csk(newsk)->icsk_ext_hdr_len = 0;
1298 if (inet_opt)
1299 inet_csk(newsk)->icsk_ext_hdr_len = inet_opt->opt.optlen;
1300 newinet->inet_id = newtp->write_seq ^ jiffies;
1301
1302 if (!dst) {
1303 dst = inet_csk_route_child_sock(sk, newsk, req);
1304 if (!dst)
1305 goto put_and_exit;
1306 } else {
1307 /* syncookie case : see end of cookie_v4_check() */
1308 }
1309 sk_setup_caps(newsk, dst);
1310
1311 tcp_ca_openreq_child(newsk, dst);
1312
1313 tcp_sync_mss(newsk, dst_mtu(dst));
1314 newtp->advmss = dst_metric_advmss(dst);
1315 if (tcp_sk(sk)->rx_opt.user_mss &&
1316 tcp_sk(sk)->rx_opt.user_mss < newtp->advmss)
1317 newtp->advmss = tcp_sk(sk)->rx_opt.user_mss;
1318
1319 tcp_initialize_rcv_mss(newsk);
1320
1321 #ifdef CONFIG_TCP_MD5SIG
1322 /* Copy over the MD5 key from the original socket */
1323 key = tcp_md5_do_lookup(sk, (union tcp_md5_addr *)&newinet->inet_daddr,
1324 AF_INET);
1325 if (key) {
1326 /*
1327 * We're using one, so create a matching key
1328 * on the newsk structure. If we fail to get
1329 * memory, then we end up not copying the key
1330 * across. Shucks.
1331 */
1332 tcp_md5_do_add(newsk, (union tcp_md5_addr *)&newinet->inet_daddr,
1333 AF_INET, key->key, key->keylen, GFP_ATOMIC);
1334 sk_nocaps_add(newsk, NETIF_F_GSO_MASK);
1335 }
1336 #endif
1337
1338 if (__inet_inherit_port(sk, newsk) < 0)
1339 goto put_and_exit;
1340 *own_req = inet_ehash_nolisten(newsk, req_to_sk(req_unhash));
1341 if (*own_req)
1342 tcp_move_syn(newtp, req);
1343
1344 return newsk;
1345
1346 exit_overflow:
1347 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENOVERFLOWS);
1348 exit_nonewsk:
1349 dst_release(dst);
1350 exit:
1351 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENDROPS);
1352 return NULL;
1353 put_and_exit:
1354 inet_csk_prepare_forced_close(newsk);
1355 tcp_done(newsk);
1356 goto exit;
1357 }
1358 EXPORT_SYMBOL(tcp_v4_syn_recv_sock);
1359
1360 static struct sock *tcp_v4_cookie_check(struct sock *sk, struct sk_buff *skb)
1361 {
1362 #ifdef CONFIG_SYN_COOKIES
1363 const struct tcphdr *th = tcp_hdr(skb);
1364
1365 if (!th->syn)
1366 sk = cookie_v4_check(sk, skb);
1367 #endif
1368 return sk;
1369 }
1370
1371 /* The socket must have it's spinlock held when we get
1372 * here, unless it is a TCP_LISTEN socket.
1373 *
1374 * We have a potential double-lock case here, so even when
1375 * doing backlog processing we use the BH locking scheme.
1376 * This is because we cannot sleep with the original spinlock
1377 * held.
1378 */
1379 int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb)
1380 {
1381 struct sock *rsk;
1382
1383 if (sk->sk_state == TCP_ESTABLISHED) { /* Fast path */
1384 struct dst_entry *dst = sk->sk_rx_dst;
1385
1386 sock_rps_save_rxhash(sk, skb);
1387 sk_mark_napi_id(sk, skb);
1388 if (dst) {
1389 if (inet_sk(sk)->rx_dst_ifindex != skb->skb_iif ||
1390 !dst->ops->check(dst, 0)) {
1391 dst_release(dst);
1392 sk->sk_rx_dst = NULL;
1393 }
1394 }
1395 tcp_rcv_established(sk, skb, tcp_hdr(skb), skb->len);
1396 return 0;
1397 }
1398
1399 if (tcp_checksum_complete(skb))
1400 goto csum_err;
1401
1402 if (sk->sk_state == TCP_LISTEN) {
1403 struct sock *nsk = tcp_v4_cookie_check(sk, skb);
1404
1405 if (!nsk)
1406 goto discard;
1407 if (nsk != sk) {
1408 sock_rps_save_rxhash(nsk, skb);
1409 sk_mark_napi_id(nsk, skb);
1410 if (tcp_child_process(sk, nsk, skb)) {
1411 rsk = nsk;
1412 goto reset;
1413 }
1414 return 0;
1415 }
1416 } else
1417 sock_rps_save_rxhash(sk, skb);
1418
1419 if (tcp_rcv_state_process(sk, skb)) {
1420 rsk = sk;
1421 goto reset;
1422 }
1423 return 0;
1424
1425 reset:
1426 tcp_v4_send_reset(rsk, skb);
1427 discard:
1428 kfree_skb(skb);
1429 /* Be careful here. If this function gets more complicated and
1430 * gcc suffers from register pressure on the x86, sk (in %ebx)
1431 * might be destroyed here. This current version compiles correctly,
1432 * but you have been warned.
1433 */
1434 return 0;
1435
1436 csum_err:
1437 TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_CSUMERRORS);
1438 TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_INERRS);
1439 goto discard;
1440 }
1441 EXPORT_SYMBOL(tcp_v4_do_rcv);
1442
1443 void tcp_v4_early_demux(struct sk_buff *skb)
1444 {
1445 const struct iphdr *iph;
1446 const struct tcphdr *th;
1447 struct sock *sk;
1448
1449 if (skb->pkt_type != PACKET_HOST)
1450 return;
1451
1452 if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct tcphdr)))
1453 return;
1454
1455 iph = ip_hdr(skb);
1456 th = tcp_hdr(skb);
1457
1458 if (th->doff < sizeof(struct tcphdr) / 4)
1459 return;
1460
1461 sk = __inet_lookup_established(dev_net(skb->dev), &tcp_hashinfo,
1462 iph->saddr, th->source,
1463 iph->daddr, ntohs(th->dest),
1464 skb->skb_iif);
1465 if (sk) {
1466 skb->sk = sk;
1467 skb->destructor = sock_edemux;
1468 if (sk_fullsock(sk)) {
1469 struct dst_entry *dst = READ_ONCE(sk->sk_rx_dst);
1470
1471 if (dst)
1472 dst = dst_check(dst, 0);
1473 if (dst &&
1474 inet_sk(sk)->rx_dst_ifindex == skb->skb_iif)
1475 skb_dst_set_noref(skb, dst);
1476 }
1477 }
1478 }
1479
1480 /* Packet is added to VJ-style prequeue for processing in process
1481 * context, if a reader task is waiting. Apparently, this exciting
1482 * idea (VJ's mail "Re: query about TCP header on tcp-ip" of 07 Sep 93)
1483 * failed somewhere. Latency? Burstiness? Well, at least now we will
1484 * see, why it failed. 8)8) --ANK
1485 *
1486 */
1487 bool tcp_prequeue(struct sock *sk, struct sk_buff *skb)
1488 {
1489 struct tcp_sock *tp = tcp_sk(sk);
1490
1491 if (sysctl_tcp_low_latency || !tp->ucopy.task)
1492 return false;
1493
1494 if (skb->len <= tcp_hdrlen(skb) &&
1495 skb_queue_len(&tp->ucopy.prequeue) == 0)
1496 return false;
1497
1498 /* Before escaping RCU protected region, we need to take care of skb
1499 * dst. Prequeue is only enabled for established sockets.
1500 * For such sockets, we might need the skb dst only to set sk->sk_rx_dst
1501 * Instead of doing full sk_rx_dst validity here, let's perform
1502 * an optimistic check.
1503 */
1504 if (likely(sk->sk_rx_dst))
1505 skb_dst_drop(skb);
1506 else
1507 skb_dst_force_safe(skb);
1508
1509 __skb_queue_tail(&tp->ucopy.prequeue, skb);
1510 tp->ucopy.memory += skb->truesize;
1511 if (tp->ucopy.memory > sk->sk_rcvbuf) {
1512 struct sk_buff *skb1;
1513
1514 BUG_ON(sock_owned_by_user(sk));
1515
1516 while ((skb1 = __skb_dequeue(&tp->ucopy.prequeue)) != NULL) {
1517 sk_backlog_rcv(sk, skb1);
1518 NET_INC_STATS_BH(sock_net(sk),
1519 LINUX_MIB_TCPPREQUEUEDROPPED);
1520 }
1521
1522 tp->ucopy.memory = 0;
1523 } else if (skb_queue_len(&tp->ucopy.prequeue) == 1) {
1524 wake_up_interruptible_sync_poll(sk_sleep(sk),
1525 POLLIN | POLLRDNORM | POLLRDBAND);
1526 if (!inet_csk_ack_scheduled(sk))
1527 inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK,
1528 (3 * tcp_rto_min(sk)) / 4,
1529 TCP_RTO_MAX);
1530 }
1531 return true;
1532 }
1533 EXPORT_SYMBOL(tcp_prequeue);
1534
1535 /*
1536 * From tcp_input.c
1537 */
1538
1539 int tcp_v4_rcv(struct sk_buff *skb)
1540 {
1541 const struct iphdr *iph;
1542 const struct tcphdr *th;
1543 struct sock *sk;
1544 int ret;
1545 struct net *net = dev_net(skb->dev);
1546
1547 if (skb->pkt_type != PACKET_HOST)
1548 goto discard_it;
1549
1550 /* Count it even if it's bad */
1551 TCP_INC_STATS_BH(net, TCP_MIB_INSEGS);
1552
1553 if (!pskb_may_pull(skb, sizeof(struct tcphdr)))
1554 goto discard_it;
1555
1556 th = tcp_hdr(skb);
1557
1558 if (th->doff < sizeof(struct tcphdr) / 4)
1559 goto bad_packet;
1560 if (!pskb_may_pull(skb, th->doff * 4))
1561 goto discard_it;
1562
1563 /* An explanation is required here, I think.
1564 * Packet length and doff are validated by header prediction,
1565 * provided case of th->doff==0 is eliminated.
1566 * So, we defer the checks. */
1567
1568 if (skb_checksum_init(skb, IPPROTO_TCP, inet_compute_pseudo))
1569 goto csum_error;
1570
1571 th = tcp_hdr(skb);
1572 iph = ip_hdr(skb);
1573 /* This is tricky : We move IPCB at its correct location into TCP_SKB_CB()
1574 * barrier() makes sure compiler wont play fool^Waliasing games.
1575 */
1576 memmove(&TCP_SKB_CB(skb)->header.h4, IPCB(skb),
1577 sizeof(struct inet_skb_parm));
1578 barrier();
1579
1580 TCP_SKB_CB(skb)->seq = ntohl(th->seq);
1581 TCP_SKB_CB(skb)->end_seq = (TCP_SKB_CB(skb)->seq + th->syn + th->fin +
1582 skb->len - th->doff * 4);
1583 TCP_SKB_CB(skb)->ack_seq = ntohl(th->ack_seq);
1584 TCP_SKB_CB(skb)->tcp_flags = tcp_flag_byte(th);
1585 TCP_SKB_CB(skb)->tcp_tw_isn = 0;
1586 TCP_SKB_CB(skb)->ip_dsfield = ipv4_get_dsfield(iph);
1587 TCP_SKB_CB(skb)->sacked = 0;
1588
1589 lookup:
1590 sk = __inet_lookup_skb(&tcp_hashinfo, skb, __tcp_hdrlen(th), th->source,
1591 th->dest);
1592 if (!sk)
1593 goto no_tcp_socket;
1594
1595 process:
1596 if (sk->sk_state == TCP_TIME_WAIT)
1597 goto do_time_wait;
1598
1599 if (sk->sk_state == TCP_NEW_SYN_RECV) {
1600 struct request_sock *req = inet_reqsk(sk);
1601 struct sock *nsk;
1602
1603 sk = req->rsk_listener;
1604 if (unlikely(tcp_v4_inbound_md5_hash(sk, skb))) {
1605 reqsk_put(req);
1606 goto discard_it;
1607 }
1608 if (unlikely(sk->sk_state != TCP_LISTEN)) {
1609 inet_csk_reqsk_queue_drop_and_put(sk, req);
1610 goto lookup;
1611 }
1612 sock_hold(sk);
1613 nsk = tcp_check_req(sk, skb, req, false);
1614 if (!nsk) {
1615 reqsk_put(req);
1616 goto discard_and_relse;
1617 }
1618 if (nsk == sk) {
1619 reqsk_put(req);
1620 } else if (tcp_child_process(sk, nsk, skb)) {
1621 tcp_v4_send_reset(nsk, skb);
1622 goto discard_and_relse;
1623 } else {
1624 sock_put(sk);
1625 return 0;
1626 }
1627 }
1628 if (unlikely(iph->ttl < inet_sk(sk)->min_ttl)) {
1629 NET_INC_STATS_BH(net, LINUX_MIB_TCPMINTTLDROP);
1630 goto discard_and_relse;
1631 }
1632
1633 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
1634 goto discard_and_relse;
1635
1636 if (tcp_v4_inbound_md5_hash(sk, skb))
1637 goto discard_and_relse;
1638
1639 nf_reset(skb);
1640
1641 if (sk_filter(sk, skb))
1642 goto discard_and_relse;
1643
1644 skb->dev = NULL;
1645
1646 if (sk->sk_state == TCP_LISTEN) {
1647 ret = tcp_v4_do_rcv(sk, skb);
1648 goto put_and_return;
1649 }
1650
1651 sk_incoming_cpu_update(sk);
1652
1653 bh_lock_sock_nested(sk);
1654 tcp_segs_in(tcp_sk(sk), skb);
1655 ret = 0;
1656 if (!sock_owned_by_user(sk)) {
1657 if (!tcp_prequeue(sk, skb))
1658 ret = tcp_v4_do_rcv(sk, skb);
1659 } else if (unlikely(sk_add_backlog(sk, skb,
1660 sk->sk_rcvbuf + sk->sk_sndbuf))) {
1661 bh_unlock_sock(sk);
1662 NET_INC_STATS_BH(net, LINUX_MIB_TCPBACKLOGDROP);
1663 goto discard_and_relse;
1664 }
1665 bh_unlock_sock(sk);
1666
1667 put_and_return:
1668 sock_put(sk);
1669
1670 return ret;
1671
1672 no_tcp_socket:
1673 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
1674 goto discard_it;
1675
1676 if (tcp_checksum_complete(skb)) {
1677 csum_error:
1678 TCP_INC_STATS_BH(net, TCP_MIB_CSUMERRORS);
1679 bad_packet:
1680 TCP_INC_STATS_BH(net, TCP_MIB_INERRS);
1681 } else {
1682 tcp_v4_send_reset(NULL, skb);
1683 }
1684
1685 discard_it:
1686 /* Discard frame. */
1687 kfree_skb(skb);
1688 return 0;
1689
1690 discard_and_relse:
1691 sock_put(sk);
1692 goto discard_it;
1693
1694 do_time_wait:
1695 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) {
1696 inet_twsk_put(inet_twsk(sk));
1697 goto discard_it;
1698 }
1699
1700 if (tcp_checksum_complete(skb)) {
1701 inet_twsk_put(inet_twsk(sk));
1702 goto csum_error;
1703 }
1704 switch (tcp_timewait_state_process(inet_twsk(sk), skb, th)) {
1705 case TCP_TW_SYN: {
1706 struct sock *sk2 = inet_lookup_listener(dev_net(skb->dev),
1707 &tcp_hashinfo, skb,
1708 __tcp_hdrlen(th),
1709 iph->saddr, th->source,
1710 iph->daddr, th->dest,
1711 inet_iif(skb));
1712 if (sk2) {
1713 inet_twsk_deschedule_put(inet_twsk(sk));
1714 sk = sk2;
1715 goto process;
1716 }
1717 /* Fall through to ACK */
1718 }
1719 case TCP_TW_ACK:
1720 tcp_v4_timewait_ack(sk, skb);
1721 break;
1722 case TCP_TW_RST:
1723 tcp_v4_send_reset(sk, skb);
1724 inet_twsk_deschedule_put(inet_twsk(sk));
1725 goto discard_it;
1726 case TCP_TW_SUCCESS:;
1727 }
1728 goto discard_it;
1729 }
1730
1731 static struct timewait_sock_ops tcp_timewait_sock_ops = {
1732 .twsk_obj_size = sizeof(struct tcp_timewait_sock),
1733 .twsk_unique = tcp_twsk_unique,
1734 .twsk_destructor= tcp_twsk_destructor,
1735 };
1736
1737 void inet_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb)
1738 {
1739 struct dst_entry *dst = skb_dst(skb);
1740
1741 if (dst && dst_hold_safe(dst)) {
1742 sk->sk_rx_dst = dst;
1743 inet_sk(sk)->rx_dst_ifindex = skb->skb_iif;
1744 }
1745 }
1746 EXPORT_SYMBOL(inet_sk_rx_dst_set);
1747
1748 const struct inet_connection_sock_af_ops ipv4_specific = {
1749 .queue_xmit = ip_queue_xmit,
1750 .send_check = tcp_v4_send_check,
1751 .rebuild_header = inet_sk_rebuild_header,
1752 .sk_rx_dst_set = inet_sk_rx_dst_set,
1753 .conn_request = tcp_v4_conn_request,
1754 .syn_recv_sock = tcp_v4_syn_recv_sock,
1755 .net_header_len = sizeof(struct iphdr),
1756 .setsockopt = ip_setsockopt,
1757 .getsockopt = ip_getsockopt,
1758 .addr2sockaddr = inet_csk_addr2sockaddr,
1759 .sockaddr_len = sizeof(struct sockaddr_in),
1760 .bind_conflict = inet_csk_bind_conflict,
1761 #ifdef CONFIG_COMPAT
1762 .compat_setsockopt = compat_ip_setsockopt,
1763 .compat_getsockopt = compat_ip_getsockopt,
1764 #endif
1765 .mtu_reduced = tcp_v4_mtu_reduced,
1766 };
1767 EXPORT_SYMBOL(ipv4_specific);
1768
1769 #ifdef CONFIG_TCP_MD5SIG
1770 static const struct tcp_sock_af_ops tcp_sock_ipv4_specific = {
1771 .md5_lookup = tcp_v4_md5_lookup,
1772 .calc_md5_hash = tcp_v4_md5_hash_skb,
1773 .md5_parse = tcp_v4_parse_md5_keys,
1774 };
1775 #endif
1776
1777 /* NOTE: A lot of things set to zero explicitly by call to
1778 * sk_alloc() so need not be done here.
1779 */
1780 static int tcp_v4_init_sock(struct sock *sk)
1781 {
1782 struct inet_connection_sock *icsk = inet_csk(sk);
1783
1784 tcp_init_sock(sk);
1785
1786 icsk->icsk_af_ops = &ipv4_specific;
1787
1788 #ifdef CONFIG_TCP_MD5SIG
1789 tcp_sk(sk)->af_specific = &tcp_sock_ipv4_specific;
1790 #endif
1791
1792 return 0;
1793 }
1794
1795 void tcp_v4_destroy_sock(struct sock *sk)
1796 {
1797 struct tcp_sock *tp = tcp_sk(sk);
1798
1799 tcp_clear_xmit_timers(sk);
1800
1801 tcp_cleanup_congestion_control(sk);
1802
1803 /* Cleanup up the write buffer. */
1804 tcp_write_queue_purge(sk);
1805
1806 /* Cleans up our, hopefully empty, out_of_order_queue. */
1807 __skb_queue_purge(&tp->out_of_order_queue);
1808
1809 #ifdef CONFIG_TCP_MD5SIG
1810 /* Clean up the MD5 key list, if any */
1811 if (tp->md5sig_info) {
1812 tcp_clear_md5_list(sk);
1813 kfree_rcu(tp->md5sig_info, rcu);
1814 tp->md5sig_info = NULL;
1815 }
1816 #endif
1817
1818 /* Clean prequeue, it must be empty really */
1819 __skb_queue_purge(&tp->ucopy.prequeue);
1820
1821 /* Clean up a referenced TCP bind bucket. */
1822 if (inet_csk(sk)->icsk_bind_hash)
1823 inet_put_port(sk);
1824
1825 BUG_ON(tp->fastopen_rsk);
1826
1827 /* If socket is aborted during connect operation */
1828 tcp_free_fastopen_req(tp);
1829 tcp_saved_syn_free(tp);
1830
1831 sk_sockets_allocated_dec(sk);
1832
1833 if (mem_cgroup_sockets_enabled && sk->sk_memcg)
1834 sock_release_memcg(sk);
1835 }
1836 EXPORT_SYMBOL(tcp_v4_destroy_sock);
1837
1838 #ifdef CONFIG_PROC_FS
1839 /* Proc filesystem TCP sock list dumping. */
1840
1841 /*
1842 * Get next listener socket follow cur. If cur is NULL, get first socket
1843 * starting from bucket given in st->bucket; when st->bucket is zero the
1844 * very first socket in the hash table is returned.
1845 */
1846 static void *listening_get_next(struct seq_file *seq, void *cur)
1847 {
1848 struct inet_connection_sock *icsk;
1849 struct hlist_nulls_node *node;
1850 struct sock *sk = cur;
1851 struct inet_listen_hashbucket *ilb;
1852 struct tcp_iter_state *st = seq->private;
1853 struct net *net = seq_file_net(seq);
1854
1855 if (!sk) {
1856 ilb = &tcp_hashinfo.listening_hash[st->bucket];
1857 spin_lock_bh(&ilb->lock);
1858 sk = sk_nulls_head(&ilb->head);
1859 st->offset = 0;
1860 goto get_sk;
1861 }
1862 ilb = &tcp_hashinfo.listening_hash[st->bucket];
1863 ++st->num;
1864 ++st->offset;
1865
1866 sk = sk_nulls_next(sk);
1867 get_sk:
1868 sk_nulls_for_each_from(sk, node) {
1869 if (!net_eq(sock_net(sk), net))
1870 continue;
1871 if (sk->sk_family == st->family) {
1872 cur = sk;
1873 goto out;
1874 }
1875 icsk = inet_csk(sk);
1876 }
1877 spin_unlock_bh(&ilb->lock);
1878 st->offset = 0;
1879 if (++st->bucket < INET_LHTABLE_SIZE) {
1880 ilb = &tcp_hashinfo.listening_hash[st->bucket];
1881 spin_lock_bh(&ilb->lock);
1882 sk = sk_nulls_head(&ilb->head);
1883 goto get_sk;
1884 }
1885 cur = NULL;
1886 out:
1887 return cur;
1888 }
1889
1890 static void *listening_get_idx(struct seq_file *seq, loff_t *pos)
1891 {
1892 struct tcp_iter_state *st = seq->private;
1893 void *rc;
1894
1895 st->bucket = 0;
1896 st->offset = 0;
1897 rc = listening_get_next(seq, NULL);
1898
1899 while (rc && *pos) {
1900 rc = listening_get_next(seq, rc);
1901 --*pos;
1902 }
1903 return rc;
1904 }
1905
1906 static inline bool empty_bucket(const struct tcp_iter_state *st)
1907 {
1908 return hlist_nulls_empty(&tcp_hashinfo.ehash[st->bucket].chain);
1909 }
1910
1911 /*
1912 * Get first established socket starting from bucket given in st->bucket.
1913 * If st->bucket is zero, the very first socket in the hash is returned.
1914 */
1915 static void *established_get_first(struct seq_file *seq)
1916 {
1917 struct tcp_iter_state *st = seq->private;
1918 struct net *net = seq_file_net(seq);
1919 void *rc = NULL;
1920
1921 st->offset = 0;
1922 for (; st->bucket <= tcp_hashinfo.ehash_mask; ++st->bucket) {
1923 struct sock *sk;
1924 struct hlist_nulls_node *node;
1925 spinlock_t *lock = inet_ehash_lockp(&tcp_hashinfo, st->bucket);
1926
1927 /* Lockless fast path for the common case of empty buckets */
1928 if (empty_bucket(st))
1929 continue;
1930
1931 spin_lock_bh(lock);
1932 sk_nulls_for_each(sk, node, &tcp_hashinfo.ehash[st->bucket].chain) {
1933 if (sk->sk_family != st->family ||
1934 !net_eq(sock_net(sk), net)) {
1935 continue;
1936 }
1937 rc = sk;
1938 goto out;
1939 }
1940 spin_unlock_bh(lock);
1941 }
1942 out:
1943 return rc;
1944 }
1945
1946 static void *established_get_next(struct seq_file *seq, void *cur)
1947 {
1948 struct sock *sk = cur;
1949 struct hlist_nulls_node *node;
1950 struct tcp_iter_state *st = seq->private;
1951 struct net *net = seq_file_net(seq);
1952
1953 ++st->num;
1954 ++st->offset;
1955
1956 sk = sk_nulls_next(sk);
1957
1958 sk_nulls_for_each_from(sk, node) {
1959 if (sk->sk_family == st->family && net_eq(sock_net(sk), net))
1960 return sk;
1961 }
1962
1963 spin_unlock_bh(inet_ehash_lockp(&tcp_hashinfo, st->bucket));
1964 ++st->bucket;
1965 return established_get_first(seq);
1966 }
1967
1968 static void *established_get_idx(struct seq_file *seq, loff_t pos)
1969 {
1970 struct tcp_iter_state *st = seq->private;
1971 void *rc;
1972
1973 st->bucket = 0;
1974 rc = established_get_first(seq);
1975
1976 while (rc && pos) {
1977 rc = established_get_next(seq, rc);
1978 --pos;
1979 }
1980 return rc;
1981 }
1982
1983 static void *tcp_get_idx(struct seq_file *seq, loff_t pos)
1984 {
1985 void *rc;
1986 struct tcp_iter_state *st = seq->private;
1987
1988 st->state = TCP_SEQ_STATE_LISTENING;
1989 rc = listening_get_idx(seq, &pos);
1990
1991 if (!rc) {
1992 st->state = TCP_SEQ_STATE_ESTABLISHED;
1993 rc = established_get_idx(seq, pos);
1994 }
1995
1996 return rc;
1997 }
1998
1999 static void *tcp_seek_last_pos(struct seq_file *seq)
2000 {
2001 struct tcp_iter_state *st = seq->private;
2002 int offset = st->offset;
2003 int orig_num = st->num;
2004 void *rc = NULL;
2005
2006 switch (st->state) {
2007 case TCP_SEQ_STATE_LISTENING:
2008 if (st->bucket >= INET_LHTABLE_SIZE)
2009 break;
2010 st->state = TCP_SEQ_STATE_LISTENING;
2011 rc = listening_get_next(seq, NULL);
2012 while (offset-- && rc)
2013 rc = listening_get_next(seq, rc);
2014 if (rc)
2015 break;
2016 st->bucket = 0;
2017 st->state = TCP_SEQ_STATE_ESTABLISHED;
2018 /* Fallthrough */
2019 case TCP_SEQ_STATE_ESTABLISHED:
2020 if (st->bucket > tcp_hashinfo.ehash_mask)
2021 break;
2022 rc = established_get_first(seq);
2023 while (offset-- && rc)
2024 rc = established_get_next(seq, rc);
2025 }
2026
2027 st->num = orig_num;
2028
2029 return rc;
2030 }
2031
2032 static void *tcp_seq_start(struct seq_file *seq, loff_t *pos)
2033 {
2034 struct tcp_iter_state *st = seq->private;
2035 void *rc;
2036
2037 if (*pos && *pos == st->last_pos) {
2038 rc = tcp_seek_last_pos(seq);
2039 if (rc)
2040 goto out;
2041 }
2042
2043 st->state = TCP_SEQ_STATE_LISTENING;
2044 st->num = 0;
2045 st->bucket = 0;
2046 st->offset = 0;
2047 rc = *pos ? tcp_get_idx(seq, *pos - 1) : SEQ_START_TOKEN;
2048
2049 out:
2050 st->last_pos = *pos;
2051 return rc;
2052 }
2053
2054 static void *tcp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2055 {
2056 struct tcp_iter_state *st = seq->private;
2057 void *rc = NULL;
2058
2059 if (v == SEQ_START_TOKEN) {
2060 rc = tcp_get_idx(seq, 0);
2061 goto out;
2062 }
2063
2064 switch (st->state) {
2065 case TCP_SEQ_STATE_LISTENING:
2066 rc = listening_get_next(seq, v);
2067 if (!rc) {
2068 st->state = TCP_SEQ_STATE_ESTABLISHED;
2069 st->bucket = 0;
2070 st->offset = 0;
2071 rc = established_get_first(seq);
2072 }
2073 break;
2074 case TCP_SEQ_STATE_ESTABLISHED:
2075 rc = established_get_next(seq, v);
2076 break;
2077 }
2078 out:
2079 ++*pos;
2080 st->last_pos = *pos;
2081 return rc;
2082 }
2083
2084 static void tcp_seq_stop(struct seq_file *seq, void *v)
2085 {
2086 struct tcp_iter_state *st = seq->private;
2087
2088 switch (st->state) {
2089 case TCP_SEQ_STATE_LISTENING:
2090 if (v != SEQ_START_TOKEN)
2091 spin_unlock_bh(&tcp_hashinfo.listening_hash[st->bucket].lock);
2092 break;
2093 case TCP_SEQ_STATE_ESTABLISHED:
2094 if (v)
2095 spin_unlock_bh(inet_ehash_lockp(&tcp_hashinfo, st->bucket));
2096 break;
2097 }
2098 }
2099
2100 int tcp_seq_open(struct inode *inode, struct file *file)
2101 {
2102 struct tcp_seq_afinfo *afinfo = PDE_DATA(inode);
2103 struct tcp_iter_state *s;
2104 int err;
2105
2106 err = seq_open_net(inode, file, &afinfo->seq_ops,
2107 sizeof(struct tcp_iter_state));
2108 if (err < 0)
2109 return err;
2110
2111 s = ((struct seq_file *)file->private_data)->private;
2112 s->family = afinfo->family;
2113 s->last_pos = 0;
2114 return 0;
2115 }
2116 EXPORT_SYMBOL(tcp_seq_open);
2117
2118 int tcp_proc_register(struct net *net, struct tcp_seq_afinfo *afinfo)
2119 {
2120 int rc = 0;
2121 struct proc_dir_entry *p;
2122
2123 afinfo->seq_ops.start = tcp_seq_start;
2124 afinfo->seq_ops.next = tcp_seq_next;
2125 afinfo->seq_ops.stop = tcp_seq_stop;
2126
2127 p = proc_create_data(afinfo->name, S_IRUGO, net->proc_net,
2128 afinfo->seq_fops, afinfo);
2129 if (!p)
2130 rc = -ENOMEM;
2131 return rc;
2132 }
2133 EXPORT_SYMBOL(tcp_proc_register);
2134
2135 void tcp_proc_unregister(struct net *net, struct tcp_seq_afinfo *afinfo)
2136 {
2137 remove_proc_entry(afinfo->name, net->proc_net);
2138 }
2139 EXPORT_SYMBOL(tcp_proc_unregister);
2140
2141 static void get_openreq4(const struct request_sock *req,
2142 struct seq_file *f, int i)
2143 {
2144 const struct inet_request_sock *ireq = inet_rsk(req);
2145 long delta = req->rsk_timer.expires - jiffies;
2146
2147 seq_printf(f, "%4d: %08X:%04X %08X:%04X"
2148 " %02X %08X:%08X %02X:%08lX %08X %5u %8d %u %d %pK",
2149 i,
2150 ireq->ir_loc_addr,
2151 ireq->ir_num,
2152 ireq->ir_rmt_addr,
2153 ntohs(ireq->ir_rmt_port),
2154 TCP_SYN_RECV,
2155 0, 0, /* could print option size, but that is af dependent. */
2156 1, /* timers active (only the expire timer) */
2157 jiffies_delta_to_clock_t(delta),
2158 req->num_timeout,
2159 from_kuid_munged(seq_user_ns(f),
2160 sock_i_uid(req->rsk_listener)),
2161 0, /* non standard timer */
2162 0, /* open_requests have no inode */
2163 0,
2164 req);
2165 }
2166
2167 static void get_tcp4_sock(struct sock *sk, struct seq_file *f, int i)
2168 {
2169 int timer_active;
2170 unsigned long timer_expires;
2171 const struct tcp_sock *tp = tcp_sk(sk);
2172 const struct inet_connection_sock *icsk = inet_csk(sk);
2173 const struct inet_sock *inet = inet_sk(sk);
2174 const struct fastopen_queue *fastopenq = &icsk->icsk_accept_queue.fastopenq;
2175 __be32 dest = inet->inet_daddr;
2176 __be32 src = inet->inet_rcv_saddr;
2177 __u16 destp = ntohs(inet->inet_dport);
2178 __u16 srcp = ntohs(inet->inet_sport);
2179 int rx_queue;
2180 int state;
2181
2182 if (icsk->icsk_pending == ICSK_TIME_RETRANS ||
2183 icsk->icsk_pending == ICSK_TIME_EARLY_RETRANS ||
2184 icsk->icsk_pending == ICSK_TIME_LOSS_PROBE) {
2185 timer_active = 1;
2186 timer_expires = icsk->icsk_timeout;
2187 } else if (icsk->icsk_pending == ICSK_TIME_PROBE0) {
2188 timer_active = 4;
2189 timer_expires = icsk->icsk_timeout;
2190 } else if (timer_pending(&sk->sk_timer)) {
2191 timer_active = 2;
2192 timer_expires = sk->sk_timer.expires;
2193 } else {
2194 timer_active = 0;
2195 timer_expires = jiffies;
2196 }
2197
2198 state = sk_state_load(sk);
2199 if (state == TCP_LISTEN)
2200 rx_queue = sk->sk_ack_backlog;
2201 else
2202 /* Because we don't lock the socket,
2203 * we might find a transient negative value.
2204 */
2205 rx_queue = max_t(int, tp->rcv_nxt - tp->copied_seq, 0);
2206
2207 seq_printf(f, "%4d: %08X:%04X %08X:%04X %02X %08X:%08X %02X:%08lX "
2208 "%08X %5u %8d %lu %d %pK %lu %lu %u %u %d",
2209 i, src, srcp, dest, destp, state,
2210 tp->write_seq - tp->snd_una,
2211 rx_queue,
2212 timer_active,
2213 jiffies_delta_to_clock_t(timer_expires - jiffies),
2214 icsk->icsk_retransmits,
2215 from_kuid_munged(seq_user_ns(f), sock_i_uid(sk)),
2216 icsk->icsk_probes_out,
2217 sock_i_ino(sk),
2218 atomic_read(&sk->sk_refcnt), sk,
2219 jiffies_to_clock_t(icsk->icsk_rto),
2220 jiffies_to_clock_t(icsk->icsk_ack.ato),
2221 (icsk->icsk_ack.quick << 1) | icsk->icsk_ack.pingpong,
2222 tp->snd_cwnd,
2223 state == TCP_LISTEN ?
2224 fastopenq->max_qlen :
2225 (tcp_in_initial_slowstart(tp) ? -1 : tp->snd_ssthresh));
2226 }
2227
2228 static void get_timewait4_sock(const struct inet_timewait_sock *tw,
2229 struct seq_file *f, int i)
2230 {
2231 long delta = tw->tw_timer.expires - jiffies;
2232 __be32 dest, src;
2233 __u16 destp, srcp;
2234
2235 dest = tw->tw_daddr;
2236 src = tw->tw_rcv_saddr;
2237 destp = ntohs(tw->tw_dport);
2238 srcp = ntohs(tw->tw_sport);
2239
2240 seq_printf(f, "%4d: %08X:%04X %08X:%04X"
2241 " %02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %pK",
2242 i, src, srcp, dest, destp, tw->tw_substate, 0, 0,
2243 3, jiffies_delta_to_clock_t(delta), 0, 0, 0, 0,
2244 atomic_read(&tw->tw_refcnt), tw);
2245 }
2246
2247 #define TMPSZ 150
2248
2249 static int tcp4_seq_show(struct seq_file *seq, void *v)
2250 {
2251 struct tcp_iter_state *st;
2252 struct sock *sk = v;
2253
2254 seq_setwidth(seq, TMPSZ - 1);
2255 if (v == SEQ_START_TOKEN) {
2256 seq_puts(seq, " sl local_address rem_address st tx_queue "
2257 "rx_queue tr tm->when retrnsmt uid timeout "
2258 "inode");
2259 goto out;
2260 }
2261 st = seq->private;
2262
2263 if (sk->sk_state == TCP_TIME_WAIT)
2264 get_timewait4_sock(v, seq, st->num);
2265 else if (sk->sk_state == TCP_NEW_SYN_RECV)
2266 get_openreq4(v, seq, st->num);
2267 else
2268 get_tcp4_sock(v, seq, st->num);
2269 out:
2270 seq_pad(seq, '\n');
2271 return 0;
2272 }
2273
2274 static const struct file_operations tcp_afinfo_seq_fops = {
2275 .owner = THIS_MODULE,
2276 .open = tcp_seq_open,
2277 .read = seq_read,
2278 .llseek = seq_lseek,
2279 .release = seq_release_net
2280 };
2281
2282 static struct tcp_seq_afinfo tcp4_seq_afinfo = {
2283 .name = "tcp",
2284 .family = AF_INET,
2285 .seq_fops = &tcp_afinfo_seq_fops,
2286 .seq_ops = {
2287 .show = tcp4_seq_show,
2288 },
2289 };
2290
2291 static int __net_init tcp4_proc_init_net(struct net *net)
2292 {
2293 return tcp_proc_register(net, &tcp4_seq_afinfo);
2294 }
2295
2296 static void __net_exit tcp4_proc_exit_net(struct net *net)
2297 {
2298 tcp_proc_unregister(net, &tcp4_seq_afinfo);
2299 }
2300
2301 static struct pernet_operations tcp4_net_ops = {
2302 .init = tcp4_proc_init_net,
2303 .exit = tcp4_proc_exit_net,
2304 };
2305
2306 int __init tcp4_proc_init(void)
2307 {
2308 return register_pernet_subsys(&tcp4_net_ops);
2309 }
2310
2311 void tcp4_proc_exit(void)
2312 {
2313 unregister_pernet_subsys(&tcp4_net_ops);
2314 }
2315 #endif /* CONFIG_PROC_FS */
2316
2317 struct proto tcp_prot = {
2318 .name = "TCP",
2319 .owner = THIS_MODULE,
2320 .close = tcp_close,
2321 .connect = tcp_v4_connect,
2322 .disconnect = tcp_disconnect,
2323 .accept = inet_csk_accept,
2324 .ioctl = tcp_ioctl,
2325 .init = tcp_v4_init_sock,
2326 .destroy = tcp_v4_destroy_sock,
2327 .shutdown = tcp_shutdown,
2328 .setsockopt = tcp_setsockopt,
2329 .getsockopt = tcp_getsockopt,
2330 .recvmsg = tcp_recvmsg,
2331 .sendmsg = tcp_sendmsg,
2332 .sendpage = tcp_sendpage,
2333 .backlog_rcv = tcp_v4_do_rcv,
2334 .release_cb = tcp_release_cb,
2335 .hash = inet_hash,
2336 .unhash = inet_unhash,
2337 .get_port = inet_csk_get_port,
2338 .enter_memory_pressure = tcp_enter_memory_pressure,
2339 .stream_memory_free = tcp_stream_memory_free,
2340 .sockets_allocated = &tcp_sockets_allocated,
2341 .orphan_count = &tcp_orphan_count,
2342 .memory_allocated = &tcp_memory_allocated,
2343 .memory_pressure = &tcp_memory_pressure,
2344 .sysctl_mem = sysctl_tcp_mem,
2345 .sysctl_wmem = sysctl_tcp_wmem,
2346 .sysctl_rmem = sysctl_tcp_rmem,
2347 .max_header = MAX_TCP_HEADER,
2348 .obj_size = sizeof(struct tcp_sock),
2349 .slab_flags = SLAB_DESTROY_BY_RCU,
2350 .twsk_prot = &tcp_timewait_sock_ops,
2351 .rsk_prot = &tcp_request_sock_ops,
2352 .h.hashinfo = &tcp_hashinfo,
2353 .no_autobind = true,
2354 #ifdef CONFIG_COMPAT
2355 .compat_setsockopt = compat_tcp_setsockopt,
2356 .compat_getsockopt = compat_tcp_getsockopt,
2357 #endif
2358 .diag_destroy = tcp_abort,
2359 };
2360 EXPORT_SYMBOL(tcp_prot);
2361
2362 static void __net_exit tcp_sk_exit(struct net *net)
2363 {
2364 int cpu;
2365
2366 for_each_possible_cpu(cpu)
2367 inet_ctl_sock_destroy(*per_cpu_ptr(net->ipv4.tcp_sk, cpu));
2368 free_percpu(net->ipv4.tcp_sk);
2369 }
2370
2371 static int __net_init tcp_sk_init(struct net *net)
2372 {
2373 int res, cpu;
2374
2375 net->ipv4.tcp_sk = alloc_percpu(struct sock *);
2376 if (!net->ipv4.tcp_sk)
2377 return -ENOMEM;
2378
2379 for_each_possible_cpu(cpu) {
2380 struct sock *sk;
2381
2382 res = inet_ctl_sock_create(&sk, PF_INET, SOCK_RAW,
2383 IPPROTO_TCP, net);
2384 if (res)
2385 goto fail;
2386 *per_cpu_ptr(net->ipv4.tcp_sk, cpu) = sk;
2387 }
2388
2389 net->ipv4.sysctl_tcp_ecn = 2;
2390 net->ipv4.sysctl_tcp_ecn_fallback = 1;
2391
2392 net->ipv4.sysctl_tcp_base_mss = TCP_BASE_MSS;
2393 net->ipv4.sysctl_tcp_probe_threshold = TCP_PROBE_THRESHOLD;
2394 net->ipv4.sysctl_tcp_probe_interval = TCP_PROBE_INTERVAL;
2395
2396 net->ipv4.sysctl_tcp_keepalive_time = TCP_KEEPALIVE_TIME;
2397 net->ipv4.sysctl_tcp_keepalive_probes = TCP_KEEPALIVE_PROBES;
2398 net->ipv4.sysctl_tcp_keepalive_intvl = TCP_KEEPALIVE_INTVL;
2399
2400 net->ipv4.sysctl_tcp_syn_retries = TCP_SYN_RETRIES;
2401 net->ipv4.sysctl_tcp_synack_retries = TCP_SYNACK_RETRIES;
2402 net->ipv4.sysctl_tcp_syncookies = 1;
2403 net->ipv4.sysctl_tcp_reordering = TCP_FASTRETRANS_THRESH;
2404 net->ipv4.sysctl_tcp_retries1 = TCP_RETR1;
2405 net->ipv4.sysctl_tcp_retries2 = TCP_RETR2;
2406 net->ipv4.sysctl_tcp_orphan_retries = 0;
2407 net->ipv4.sysctl_tcp_fin_timeout = TCP_FIN_TIMEOUT;
2408 net->ipv4.sysctl_tcp_notsent_lowat = UINT_MAX;
2409
2410 return 0;
2411 fail:
2412 tcp_sk_exit(net);
2413
2414 return res;
2415 }
2416
2417 static void __net_exit tcp_sk_exit_batch(struct list_head *net_exit_list)
2418 {
2419 inet_twsk_purge(&tcp_hashinfo, &tcp_death_row, AF_INET);
2420 }
2421
2422 static struct pernet_operations __net_initdata tcp_sk_ops = {
2423 .init = tcp_sk_init,
2424 .exit = tcp_sk_exit,
2425 .exit_batch = tcp_sk_exit_batch,
2426 };
2427
2428 void __init tcp_v4_init(void)
2429 {
2430 inet_hashinfo_init(&tcp_hashinfo);
2431 if (register_pernet_subsys(&tcp_sk_ops))
2432 panic("Failed to create the TCP control socket.\n");
2433 }
Linux with added FPC-III support