?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Revert "[SCSI] extend the last_sector_bug flag to cover more sectors"
[linux-2.6/mini2440.git] / net / ipv4 / tcp_ipv4.c
blob91a8cfddf1c46f312d523a9e02d325e0cd322aad
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/types.h>
55 #include <linux/fcntl.h>
56 #include <linux/module.h>
57 #include <linux/random.h>
58 #include <linux/cache.h>
59 #include <linux/jhash.h>
60 #include <linux/init.h>
61 #include <linux/times.h>
62
63 #include <net/net_namespace.h>
64 #include <net/icmp.h>
65 #include <net/inet_hashtables.h>
66 #include <net/tcp.h>
67 #include <net/transp_v6.h>
68 #include <net/ipv6.h>
69 #include <net/inet_common.h>
70 #include <net/timewait_sock.h>
71 #include <net/xfrm.h>
72 #include <net/netdma.h>
73
74 #include <linux/inet.h>
75 #include <linux/ipv6.h>
76 #include <linux/stddef.h>
77 #include <linux/proc_fs.h>
78 #include <linux/seq_file.h>
79
80 #include <linux/crypto.h>
81 #include <linux/scatterlist.h>
82
83 int sysctl_tcp_tw_reuse __read_mostly;
84 int sysctl_tcp_low_latency __read_mostly;
85
86
87 #ifdef CONFIG_TCP_MD5SIG
88 static struct tcp_md5sig_key *tcp_v4_md5_do_lookup(struct sock *sk,
89 __be32 addr);
90 static int tcp_v4_md5_hash_hdr(char *md5_hash, struct tcp_md5sig_key *key,
91 __be32 daddr, __be32 saddr, struct tcphdr *th);
92 #else
93 static inline
94 struct tcp_md5sig_key *tcp_v4_md5_do_lookup(struct sock *sk, __be32 addr)
95 {
96 return NULL;
97 }
98 #endif
99
100 struct inet_hashinfo __cacheline_aligned tcp_hashinfo = {
101 .lhash_lock = __RW_LOCK_UNLOCKED(tcp_hashinfo.lhash_lock),
102 .lhash_users = ATOMIC_INIT(0),
103 .lhash_wait = __WAIT_QUEUE_HEAD_INITIALIZER(tcp_hashinfo.lhash_wait),
104 };
105
106 static inline __u32 tcp_v4_init_sequence(struct sk_buff *skb)
107 {
108 return secure_tcp_sequence_number(ip_hdr(skb)->daddr,
109 ip_hdr(skb)->saddr,
110 tcp_hdr(skb)->dest,
111 tcp_hdr(skb)->source);
112 }
113
114 int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp)
115 {
116 const struct tcp_timewait_sock *tcptw = tcp_twsk(sktw);
117 struct tcp_sock *tp = tcp_sk(sk);
118
119 /* With PAWS, it is safe from the viewpoint
120 of data integrity. Even without PAWS it is safe provided sequence
121 spaces do not overlap i.e. at data rates <= 80Mbit/sec.
122
123 Actually, the idea is close to VJ's one, only timestamp cache is
124 held not per host, but per port pair and TW bucket is used as state
125 holder.
126
127 If TW bucket has been already destroyed we fall back to VJ's scheme
128 and use initial timestamp retrieved from peer table.
129 */
130 if (tcptw->tw_ts_recent_stamp &&
131 (twp == NULL || (sysctl_tcp_tw_reuse &&
132 get_seconds() - tcptw->tw_ts_recent_stamp > 1))) {
133 tp->write_seq = tcptw->tw_snd_nxt + 65535 + 2;
134 if (tp->write_seq == 0)
135 tp->write_seq = 1;
136 tp->rx_opt.ts_recent = tcptw->tw_ts_recent;
137 tp->rx_opt.ts_recent_stamp = tcptw->tw_ts_recent_stamp;
138 sock_hold(sktw);
139 return 1;
140 }
141
142 return 0;
143 }
144
145 EXPORT_SYMBOL_GPL(tcp_twsk_unique);
146
147 /* This will initiate an outgoing connection. */
148 int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len)
149 {
150 struct inet_sock *inet = inet_sk(sk);
151 struct tcp_sock *tp = tcp_sk(sk);
152 struct sockaddr_in *usin = (struct sockaddr_in *)uaddr;
153 struct rtable *rt;
154 __be32 daddr, nexthop;
155 int tmp;
156 int err;
157
158 if (addr_len < sizeof(struct sockaddr_in))
159 return -EINVAL;
160
161 if (usin->sin_family != AF_INET)
162 return -EAFNOSUPPORT;
163
164 nexthop = daddr = usin->sin_addr.s_addr;
165 if (inet->opt && inet->opt->srr) {
166 if (!daddr)
167 return -EINVAL;
168 nexthop = inet->opt->faddr;
169 }
170
171 tmp = ip_route_connect(&rt, nexthop, inet->saddr,
172 RT_CONN_FLAGS(sk), sk->sk_bound_dev_if,
173 IPPROTO_TCP,
174 inet->sport, usin->sin_port, sk, 1);
175 if (tmp < 0) {
176 if (tmp == -ENETUNREACH)
177 IP_INC_STATS_BH(sock_net(sk), IPSTATS_MIB_OUTNOROUTES);
178 return tmp;
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->srr)
187 daddr = rt->rt_dst;
188
189 if (!inet->saddr)
190 inet->saddr = rt->rt_src;
191 inet->rcv_saddr = inet->saddr;
192
193 if (tp->rx_opt.ts_recent_stamp && inet->daddr != daddr) {
194 /* Reset inherited state */
195 tp->rx_opt.ts_recent = 0;
196 tp->rx_opt.ts_recent_stamp = 0;
197 tp->write_seq = 0;
198 }
199
200 if (tcp_death_row.sysctl_tw_recycle &&
201 !tp->rx_opt.ts_recent_stamp && rt->rt_dst == daddr) {
202 struct inet_peer *peer = rt_get_peer(rt);
203 /*
204 * VJ's idea. We save last timestamp seen from
205 * the destination in peer table, when entering state
206 * TIME-WAIT * and initialize rx_opt.ts_recent from it,
207 * when trying new connection.
208 */
209 if (peer != NULL &&
210 peer->tcp_ts_stamp + TCP_PAWS_MSL >= get_seconds()) {
211 tp->rx_opt.ts_recent_stamp = peer->tcp_ts_stamp;
212 tp->rx_opt.ts_recent = peer->tcp_ts;
213 }
214 }
215
216 inet->dport = usin->sin_port;
217 inet->daddr = daddr;
218
219 inet_csk(sk)->icsk_ext_hdr_len = 0;
220 if (inet->opt)
221 inet_csk(sk)->icsk_ext_hdr_len = inet->opt->optlen;
222
223 tp->rx_opt.mss_clamp = 536;
224
225 /* Socket identity is still unknown (sport may be zero).
226 * However we set state to SYN-SENT and not releasing socket
227 * lock select source port, enter ourselves into the hash tables and
228 * complete initialization after this.
229 */
230 tcp_set_state(sk, TCP_SYN_SENT);
231 err = inet_hash_connect(&tcp_death_row, sk);
232 if (err)
233 goto failure;
234
235 err = ip_route_newports(&rt, IPPROTO_TCP,
236 inet->sport, inet->dport, sk);
237 if (err)
238 goto failure;
239
240 /* OK, now commit destination to socket. */
241 sk->sk_gso_type = SKB_GSO_TCPV4;
242 sk_setup_caps(sk, &rt->u.dst);
243
244 if (!tp->write_seq)
245 tp->write_seq = secure_tcp_sequence_number(inet->saddr,
246 inet->daddr,
247 inet->sport,
248 usin->sin_port);
249
250 inet->id = tp->write_seq ^ jiffies;
251
252 err = tcp_connect(sk);
253 rt = NULL;
254 if (err)
255 goto failure;
256
257 return 0;
258
259 failure:
260 /*
261 * This unhashes the socket and releases the local port,
262 * if necessary.
263 */
264 tcp_set_state(sk, TCP_CLOSE);
265 ip_rt_put(rt);
266 sk->sk_route_caps = 0;
267 inet->dport = 0;
268 return err;
269 }
270
271 /*
272 * This routine does path mtu discovery as defined in RFC1191.
273 */
274 static void do_pmtu_discovery(struct sock *sk, struct iphdr *iph, u32 mtu)
275 {
276 struct dst_entry *dst;
277 struct inet_sock *inet = inet_sk(sk);
278
279 /* We are not interested in TCP_LISTEN and open_requests (SYN-ACKs
280 * send out by Linux are always <576bytes so they should go through
281 * unfragmented).
282 */
283 if (sk->sk_state == TCP_LISTEN)
284 return;
285
286 /* We don't check in the destentry if pmtu discovery is forbidden
287 * on this route. We just assume that no packet_to_big packets
288 * are send back when pmtu discovery is not active.
289 * There is a small race when the user changes this flag in the
290 * route, but I think that's acceptable.
291 */
292 if ((dst = __sk_dst_check(sk, 0)) == NULL)
293 return;
294
295 dst->ops->update_pmtu(dst, mtu);
296
297 /* Something is about to be wrong... Remember soft error
298 * for the case, if this connection will not able to recover.
299 */
300 if (mtu < dst_mtu(dst) && ip_dont_fragment(sk, dst))
301 sk->sk_err_soft = EMSGSIZE;
302
303 mtu = dst_mtu(dst);
304
305 if (inet->pmtudisc != IP_PMTUDISC_DONT &&
306 inet_csk(sk)->icsk_pmtu_cookie > mtu) {
307 tcp_sync_mss(sk, mtu);
308
309 /* Resend the TCP packet because it's
310 * clear that the old packet has been
311 * dropped. This is the new "fast" path mtu
312 * discovery.
313 */
314 tcp_simple_retransmit(sk);
315 } /* else let the usual retransmit timer handle it */
316 }
317
318 /*
319 * This routine is called by the ICMP module when it gets some
320 * sort of error condition. If err < 0 then the socket should
321 * be closed and the error returned to the user. If err > 0
322 * it's just the icmp type << 8 | icmp code. After adjustment
323 * header points to the first 8 bytes of the tcp header. We need
324 * to find the appropriate port.
325 *
326 * The locking strategy used here is very "optimistic". When
327 * someone else accesses the socket the ICMP is just dropped
328 * and for some paths there is no check at all.
329 * A more general error queue to queue errors for later handling
330 * is probably better.
331 *
332 */
333
334 void tcp_v4_err(struct sk_buff *skb, u32 info)
335 {
336 struct iphdr *iph = (struct iphdr *)skb->data;
337 struct tcphdr *th = (struct tcphdr *)(skb->data + (iph->ihl << 2));
338 struct tcp_sock *tp;
339 struct inet_sock *inet;
340 const int type = icmp_hdr(skb)->type;
341 const int code = icmp_hdr(skb)->code;
342 struct sock *sk;
343 __u32 seq;
344 int err;
345 struct net *net = dev_net(skb->dev);
346
347 if (skb->len < (iph->ihl << 2) + 8) {
348 ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS);
349 return;
350 }
351
352 sk = inet_lookup(net, &tcp_hashinfo, iph->daddr, th->dest,
353 iph->saddr, th->source, inet_iif(skb));
354 if (!sk) {
355 ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS);
356 return;
357 }
358 if (sk->sk_state == TCP_TIME_WAIT) {
359 inet_twsk_put(inet_twsk(sk));
360 return;
361 }
362
363 bh_lock_sock(sk);
364 /* If too many ICMPs get dropped on busy
365 * servers this needs to be solved differently.
366 */
367 if (sock_owned_by_user(sk))
368 NET_INC_STATS_BH(net, LINUX_MIB_LOCKDROPPEDICMPS);
369
370 if (sk->sk_state == TCP_CLOSE)
371 goto out;
372
373 tp = tcp_sk(sk);
374 seq = ntohl(th->seq);
375 if (sk->sk_state != TCP_LISTEN &&
376 !between(seq, tp->snd_una, tp->snd_nxt)) {
377 NET_INC_STATS_BH(net, LINUX_MIB_OUTOFWINDOWICMPS);
378 goto out;
379 }
380
381 switch (type) {
382 case ICMP_SOURCE_QUENCH:
383 /* Just silently ignore these. */
384 goto out;
385 case ICMP_PARAMETERPROB:
386 err = EPROTO;
387 break;
388 case ICMP_DEST_UNREACH:
389 if (code > NR_ICMP_UNREACH)
390 goto out;
391
392 if (code == ICMP_FRAG_NEEDED) { /* PMTU discovery (RFC1191) */
393 if (!sock_owned_by_user(sk))
394 do_pmtu_discovery(sk, iph, info);
395 goto out;
396 }
397
398 err = icmp_err_convert[code].errno;
399 break;
400 case ICMP_TIME_EXCEEDED:
401 err = EHOSTUNREACH;
402 break;
403 default:
404 goto out;
405 }
406
407 switch (sk->sk_state) {
408 struct request_sock *req, **prev;
409 case TCP_LISTEN:
410 if (sock_owned_by_user(sk))
411 goto out;
412
413 req = inet_csk_search_req(sk, &prev, th->dest,
414 iph->daddr, iph->saddr);
415 if (!req)
416 goto out;
417
418 /* ICMPs are not backlogged, hence we cannot get
419 an established socket here.
420 */
421 WARN_ON(req->sk);
422
423 if (seq != tcp_rsk(req)->snt_isn) {
424 NET_INC_STATS_BH(net, LINUX_MIB_OUTOFWINDOWICMPS);
425 goto out;
426 }
427
428 /*
429 * Still in SYN_RECV, just remove it silently.
430 * There is no good way to pass the error to the newly
431 * created socket, and POSIX does not want network
432 * errors returned from accept().
433 */
434 inet_csk_reqsk_queue_drop(sk, req, prev);
435 goto out;
436
437 case TCP_SYN_SENT:
438 case TCP_SYN_RECV: /* Cannot happen.
439 It can f.e. if SYNs crossed.
440 */
441 if (!sock_owned_by_user(sk)) {
442 sk->sk_err = err;
443
444 sk->sk_error_report(sk);
445
446 tcp_done(sk);
447 } else {
448 sk->sk_err_soft = err;
449 }
450 goto out;
451 }
452
453 /* If we've already connected we will keep trying
454 * until we time out, or the user gives up.
455 *
456 * rfc1122 4.2.3.9 allows to consider as hard errors
457 * only PROTO_UNREACH and PORT_UNREACH (well, FRAG_FAILED too,
458 * but it is obsoleted by pmtu discovery).
459 *
460 * Note, that in modern internet, where routing is unreliable
461 * and in each dark corner broken firewalls sit, sending random
462 * errors ordered by their masters even this two messages finally lose
463 * their original sense (even Linux sends invalid PORT_UNREACHs)
464 *
465 * Now we are in compliance with RFCs.
466 * --ANK (980905)
467 */
468
469 inet = inet_sk(sk);
470 if (!sock_owned_by_user(sk) && inet->recverr) {
471 sk->sk_err = err;
472 sk->sk_error_report(sk);
473 } else { /* Only an error on timeout */
474 sk->sk_err_soft = err;
475 }
476
477 out:
478 bh_unlock_sock(sk);
479 sock_put(sk);
480 }
481
482 /* This routine computes an IPv4 TCP checksum. */
483 void tcp_v4_send_check(struct sock *sk, int len, struct sk_buff *skb)
484 {
485 struct inet_sock *inet = inet_sk(sk);
486 struct tcphdr *th = tcp_hdr(skb);
487
488 if (skb->ip_summed == CHECKSUM_PARTIAL) {
489 th->check = ~tcp_v4_check(len, inet->saddr,
490 inet->daddr, 0);
491 skb->csum_start = skb_transport_header(skb) - skb->head;
492 skb->csum_offset = offsetof(struct tcphdr, check);
493 } else {
494 th->check = tcp_v4_check(len, inet->saddr, inet->daddr,
495 csum_partial((char *)th,
496 th->doff << 2,
497 skb->csum));
498 }
499 }
500
501 int tcp_v4_gso_send_check(struct sk_buff *skb)
502 {
503 const struct iphdr *iph;
504 struct tcphdr *th;
505
506 if (!pskb_may_pull(skb, sizeof(*th)))
507 return -EINVAL;
508
509 iph = ip_hdr(skb);
510 th = tcp_hdr(skb);
511
512 th->check = 0;
513 th->check = ~tcp_v4_check(skb->len, iph->saddr, iph->daddr, 0);
514 skb->csum_start = skb_transport_header(skb) - skb->head;
515 skb->csum_offset = offsetof(struct tcphdr, check);
516 skb->ip_summed = CHECKSUM_PARTIAL;
517 return 0;
518 }
519
520 /*
521 * This routine will send an RST to the other tcp.
522 *
523 * Someone asks: why I NEVER use socket parameters (TOS, TTL etc.)
524 * for reset.
525 * Answer: if a packet caused RST, it is not for a socket
526 * existing in our system, if it is matched to a socket,
527 * it is just duplicate segment or bug in other side's TCP.
528 * So that we build reply only basing on parameters
529 * arrived with segment.
530 * Exception: precedence violation. We do not implement it in any case.
531 */
532
533 static void tcp_v4_send_reset(struct sock *sk, struct sk_buff *skb)
534 {
535 struct tcphdr *th = tcp_hdr(skb);
536 struct {
537 struct tcphdr th;
538 #ifdef CONFIG_TCP_MD5SIG
539 __be32 opt[(TCPOLEN_MD5SIG_ALIGNED >> 2)];
540 #endif
541 } rep;
542 struct ip_reply_arg arg;
543 #ifdef CONFIG_TCP_MD5SIG
544 struct tcp_md5sig_key *key;
545 #endif
546 struct net *net;
547
548 /* Never send a reset in response to a reset. */
549 if (th->rst)
550 return;
551
552 if (skb->rtable->rt_type != RTN_LOCAL)
553 return;
554
555 /* Swap the send and the receive. */
556 memset(&rep, 0, sizeof(rep));
557 rep.th.dest = th->source;
558 rep.th.source = th->dest;
559 rep.th.doff = sizeof(struct tcphdr) / 4;
560 rep.th.rst = 1;
561
562 if (th->ack) {
563 rep.th.seq = th->ack_seq;
564 } else {
565 rep.th.ack = 1;
566 rep.th.ack_seq = htonl(ntohl(th->seq) + th->syn + th->fin +
567 skb->len - (th->doff << 2));
568 }
569
570 memset(&arg, 0, sizeof(arg));
571 arg.iov[0].iov_base = (unsigned char *)&rep;
572 arg.iov[0].iov_len = sizeof(rep.th);
573
574 #ifdef CONFIG_TCP_MD5SIG
575 key = sk ? tcp_v4_md5_do_lookup(sk, ip_hdr(skb)->daddr) : NULL;
576 if (key) {
577 rep.opt[0] = htonl((TCPOPT_NOP << 24) |
578 (TCPOPT_NOP << 16) |
579 (TCPOPT_MD5SIG << 8) |
580 TCPOLEN_MD5SIG);
581 /* Update length and the length the header thinks exists */
582 arg.iov[0].iov_len += TCPOLEN_MD5SIG_ALIGNED;
583 rep.th.doff = arg.iov[0].iov_len / 4;
584
585 tcp_v4_md5_hash_hdr((__u8 *) &rep.opt[1],
586 key, ip_hdr(skb)->daddr,
587 ip_hdr(skb)->saddr, &rep.th);
588 }
589 #endif
590 arg.csum = csum_tcpudp_nofold(ip_hdr(skb)->daddr,
591 ip_hdr(skb)->saddr, /* XXX */
592 sizeof(struct tcphdr), IPPROTO_TCP, 0);
593 arg.csumoffset = offsetof(struct tcphdr, check) / 2;
594
595 net = dev_net(skb->dst->dev);
596 ip_send_reply(net->ipv4.tcp_sock, skb,
597 &arg, arg.iov[0].iov_len);
598
599 TCP_INC_STATS_BH(net, TCP_MIB_OUTSEGS);
600 TCP_INC_STATS_BH(net, TCP_MIB_OUTRSTS);
601 }
602
603 /* The code following below sending ACKs in SYN-RECV and TIME-WAIT states
604 outside socket context is ugly, certainly. What can I do?
605 */
606
607 static void tcp_v4_send_ack(struct sk_buff *skb, u32 seq, u32 ack,
608 u32 win, u32 ts, int oif,
609 struct tcp_md5sig_key *key)
610 {
611 struct tcphdr *th = tcp_hdr(skb);
612 struct {
613 struct tcphdr th;
614 __be32 opt[(TCPOLEN_TSTAMP_ALIGNED >> 2)
615 #ifdef CONFIG_TCP_MD5SIG
616 + (TCPOLEN_MD5SIG_ALIGNED >> 2)
617 #endif
618 ];
619 } rep;
620 struct ip_reply_arg arg;
621 struct net *net = dev_net(skb->dev);
622
623 memset(&rep.th, 0, sizeof(struct tcphdr));
624 memset(&arg, 0, sizeof(arg));
625
626 arg.iov[0].iov_base = (unsigned char *)&rep;
627 arg.iov[0].iov_len = sizeof(rep.th);
628 if (ts) {
629 rep.opt[0] = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
630 (TCPOPT_TIMESTAMP << 8) |
631 TCPOLEN_TIMESTAMP);
632 rep.opt[1] = htonl(tcp_time_stamp);
633 rep.opt[2] = htonl(ts);
634 arg.iov[0].iov_len += TCPOLEN_TSTAMP_ALIGNED;
635 }
636
637 /* Swap the send and the receive. */
638 rep.th.dest = th->source;
639 rep.th.source = th->dest;
640 rep.th.doff = arg.iov[0].iov_len / 4;
641 rep.th.seq = htonl(seq);
642 rep.th.ack_seq = htonl(ack);
643 rep.th.ack = 1;
644 rep.th.window = htons(win);
645
646 #ifdef CONFIG_TCP_MD5SIG
647 if (key) {
648 int offset = (ts) ? 3 : 0;
649
650 rep.opt[offset++] = htonl((TCPOPT_NOP << 24) |
651 (TCPOPT_NOP << 16) |
652 (TCPOPT_MD5SIG << 8) |
653 TCPOLEN_MD5SIG);
654 arg.iov[0].iov_len += TCPOLEN_MD5SIG_ALIGNED;
655 rep.th.doff = arg.iov[0].iov_len/4;
656
657 tcp_v4_md5_hash_hdr((__u8 *) &rep.opt[offset],
658 key, ip_hdr(skb)->saddr,
659 ip_hdr(skb)->daddr, &rep.th);
660 }
661 #endif
662 arg.csum = csum_tcpudp_nofold(ip_hdr(skb)->daddr,
663 ip_hdr(skb)->saddr, /* XXX */
664 arg.iov[0].iov_len, IPPROTO_TCP, 0);
665 arg.csumoffset = offsetof(struct tcphdr, check) / 2;
666 if (oif)
667 arg.bound_dev_if = oif;
668
669 ip_send_reply(net->ipv4.tcp_sock, skb,
670 &arg, arg.iov[0].iov_len);
671
672 TCP_INC_STATS_BH(net, TCP_MIB_OUTSEGS);
673 }
674
675 static void tcp_v4_timewait_ack(struct sock *sk, struct sk_buff *skb)
676 {
677 struct inet_timewait_sock *tw = inet_twsk(sk);
678 struct tcp_timewait_sock *tcptw = tcp_twsk(sk);
679
680 tcp_v4_send_ack(skb, tcptw->tw_snd_nxt, tcptw->tw_rcv_nxt,
681 tcptw->tw_rcv_wnd >> tw->tw_rcv_wscale,
682 tcptw->tw_ts_recent,
683 tw->tw_bound_dev_if,
684 tcp_twsk_md5_key(tcptw)
685 );
686
687 inet_twsk_put(tw);
688 }
689
690 static void tcp_v4_reqsk_send_ack(struct sk_buff *skb,
691 struct request_sock *req)
692 {
693 tcp_v4_send_ack(skb, tcp_rsk(req)->snt_isn + 1,
694 tcp_rsk(req)->rcv_isn + 1, req->rcv_wnd,
695 req->ts_recent,
696 0,
697 tcp_v4_md5_do_lookup(skb->sk, ip_hdr(skb)->daddr));
698 }
699
700 /*
701 * Send a SYN-ACK after having received a SYN.
702 * This still operates on a request_sock only, not on a big
703 * socket.
704 */
705 static int __tcp_v4_send_synack(struct sock *sk, struct request_sock *req,
706 struct dst_entry *dst)
707 {
708 const struct inet_request_sock *ireq = inet_rsk(req);
709 int err = -1;
710 struct sk_buff * skb;
711
712 /* First, grab a route. */
713 if (!dst && (dst = inet_csk_route_req(sk, req)) == NULL)
714 return -1;
715
716 skb = tcp_make_synack(sk, dst, req);
717
718 if (skb) {
719 struct tcphdr *th = tcp_hdr(skb);
720
721 th->check = tcp_v4_check(skb->len,
722 ireq->loc_addr,
723 ireq->rmt_addr,
724 csum_partial((char *)th, skb->len,
725 skb->csum));
726
727 err = ip_build_and_send_pkt(skb, sk, ireq->loc_addr,
728 ireq->rmt_addr,
729 ireq->opt);
730 err = net_xmit_eval(err);
731 }
732
733 dst_release(dst);
734 return err;
735 }
736
737 static int tcp_v4_send_synack(struct sock *sk, struct request_sock *req)
738 {
739 return __tcp_v4_send_synack(sk, req, NULL);
740 }
741
742 /*
743 * IPv4 request_sock destructor.
744 */
745 static void tcp_v4_reqsk_destructor(struct request_sock *req)
746 {
747 kfree(inet_rsk(req)->opt);
748 }
749
750 #ifdef CONFIG_SYN_COOKIES
751 static void syn_flood_warning(struct sk_buff *skb)
752 {
753 static unsigned long warntime;
754
755 if (time_after(jiffies, (warntime + HZ * 60))) {
756 warntime = jiffies;
757 printk(KERN_INFO
758 "possible SYN flooding on port %d. Sending cookies.\n",
759 ntohs(tcp_hdr(skb)->dest));
760 }
761 }
762 #endif
763
764 /*
765 * Save and compile IPv4 options into the request_sock if needed.
766 */
767 static struct ip_options *tcp_v4_save_options(struct sock *sk,
768 struct sk_buff *skb)
769 {
770 struct ip_options *opt = &(IPCB(skb)->opt);
771 struct ip_options *dopt = NULL;
772
773 if (opt && opt->optlen) {
774 int opt_size = optlength(opt);
775 dopt = kmalloc(opt_size, GFP_ATOMIC);
776 if (dopt) {
777 if (ip_options_echo(dopt, skb)) {
778 kfree(dopt);
779 dopt = NULL;
780 }
781 }
782 }
783 return dopt;
784 }
785
786 #ifdef CONFIG_TCP_MD5SIG
787 /*
788 * RFC2385 MD5 checksumming requires a mapping of
789 * IP address->MD5 Key.
790 * We need to maintain these in the sk structure.
791 */
792
793 /* Find the Key structure for an address. */
794 static struct tcp_md5sig_key *
795 tcp_v4_md5_do_lookup(struct sock *sk, __be32 addr)
796 {
797 struct tcp_sock *tp = tcp_sk(sk);
798 int i;
799
800 if (!tp->md5sig_info || !tp->md5sig_info->entries4)
801 return NULL;
802 for (i = 0; i < tp->md5sig_info->entries4; i++) {
803 if (tp->md5sig_info->keys4[i].addr == addr)
804 return &tp->md5sig_info->keys4[i].base;
805 }
806 return NULL;
807 }
808
809 struct tcp_md5sig_key *tcp_v4_md5_lookup(struct sock *sk,
810 struct sock *addr_sk)
811 {
812 return tcp_v4_md5_do_lookup(sk, inet_sk(addr_sk)->daddr);
813 }
814
815 EXPORT_SYMBOL(tcp_v4_md5_lookup);
816
817 static struct tcp_md5sig_key *tcp_v4_reqsk_md5_lookup(struct sock *sk,
818 struct request_sock *req)
819 {
820 return tcp_v4_md5_do_lookup(sk, inet_rsk(req)->rmt_addr);
821 }
822
823 /* This can be called on a newly created socket, from other files */
824 int tcp_v4_md5_do_add(struct sock *sk, __be32 addr,
825 u8 *newkey, u8 newkeylen)
826 {
827 /* Add Key to the list */
828 struct tcp_md5sig_key *key;
829 struct tcp_sock *tp = tcp_sk(sk);
830 struct tcp4_md5sig_key *keys;
831
832 key = tcp_v4_md5_do_lookup(sk, addr);
833 if (key) {
834 /* Pre-existing entry - just update that one. */
835 kfree(key->key);
836 key->key = newkey;
837 key->keylen = newkeylen;
838 } else {
839 struct tcp_md5sig_info *md5sig;
840
841 if (!tp->md5sig_info) {
842 tp->md5sig_info = kzalloc(sizeof(*tp->md5sig_info),
843 GFP_ATOMIC);
844 if (!tp->md5sig_info) {
845 kfree(newkey);
846 return -ENOMEM;
847 }
848 sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
849 }
850 if (tcp_alloc_md5sig_pool() == NULL) {
851 kfree(newkey);
852 return -ENOMEM;
853 }
854 md5sig = tp->md5sig_info;
855
856 if (md5sig->alloced4 == md5sig->entries4) {
857 keys = kmalloc((sizeof(*keys) *
858 (md5sig->entries4 + 1)), GFP_ATOMIC);
859 if (!keys) {
860 kfree(newkey);
861 tcp_free_md5sig_pool();
862 return -ENOMEM;
863 }
864
865 if (md5sig->entries4)
866 memcpy(keys, md5sig->keys4,
867 sizeof(*keys) * md5sig->entries4);
868
869 /* Free old key list, and reference new one */
870 kfree(md5sig->keys4);
871 md5sig->keys4 = keys;
872 md5sig->alloced4++;
873 }
874 md5sig->entries4++;
875 md5sig->keys4[md5sig->entries4 - 1].addr = addr;
876 md5sig->keys4[md5sig->entries4 - 1].base.key = newkey;
877 md5sig->keys4[md5sig->entries4 - 1].base.keylen = newkeylen;
878 }
879 return 0;
880 }
881
882 EXPORT_SYMBOL(tcp_v4_md5_do_add);
883
884 static int tcp_v4_md5_add_func(struct sock *sk, struct sock *addr_sk,
885 u8 *newkey, u8 newkeylen)
886 {
887 return tcp_v4_md5_do_add(sk, inet_sk(addr_sk)->daddr,
888 newkey, newkeylen);
889 }
890
891 int tcp_v4_md5_do_del(struct sock *sk, __be32 addr)
892 {
893 struct tcp_sock *tp = tcp_sk(sk);
894 int i;
895
896 for (i = 0; i < tp->md5sig_info->entries4; i++) {
897 if (tp->md5sig_info->keys4[i].addr == addr) {
898 /* Free the key */
899 kfree(tp->md5sig_info->keys4[i].base.key);
900 tp->md5sig_info->entries4--;
901
902 if (tp->md5sig_info->entries4 == 0) {
903 kfree(tp->md5sig_info->keys4);
904 tp->md5sig_info->keys4 = NULL;
905 tp->md5sig_info->alloced4 = 0;
906 } else if (tp->md5sig_info->entries4 != i) {
907 /* Need to do some manipulation */
908 memmove(&tp->md5sig_info->keys4[i],
909 &tp->md5sig_info->keys4[i+1],
910 (tp->md5sig_info->entries4 - i) *
911 sizeof(struct tcp4_md5sig_key));
912 }
913 tcp_free_md5sig_pool();
914 return 0;
915 }
916 }
917 return -ENOENT;
918 }
919
920 EXPORT_SYMBOL(tcp_v4_md5_do_del);
921
922 static void tcp_v4_clear_md5_list(struct sock *sk)
923 {
924 struct tcp_sock *tp = tcp_sk(sk);
925
926 /* Free each key, then the set of key keys,
927 * the crypto element, and then decrement our
928 * hold on the last resort crypto.
929 */
930 if (tp->md5sig_info->entries4) {
931 int i;
932 for (i = 0; i < tp->md5sig_info->entries4; i++)
933 kfree(tp->md5sig_info->keys4[i].base.key);
934 tp->md5sig_info->entries4 = 0;
935 tcp_free_md5sig_pool();
936 }
937 if (tp->md5sig_info->keys4) {
938 kfree(tp->md5sig_info->keys4);
939 tp->md5sig_info->keys4 = NULL;
940 tp->md5sig_info->alloced4 = 0;
941 }
942 }
943
944 static int tcp_v4_parse_md5_keys(struct sock *sk, char __user *optval,
945 int optlen)
946 {
947 struct tcp_md5sig cmd;
948 struct sockaddr_in *sin = (struct sockaddr_in *)&cmd.tcpm_addr;
949 u8 *newkey;
950
951 if (optlen < sizeof(cmd))
952 return -EINVAL;
953
954 if (copy_from_user(&cmd, optval, sizeof(cmd)))
955 return -EFAULT;
956
957 if (sin->sin_family != AF_INET)
958 return -EINVAL;
959
960 if (!cmd.tcpm_key || !cmd.tcpm_keylen) {
961 if (!tcp_sk(sk)->md5sig_info)
962 return -ENOENT;
963 return tcp_v4_md5_do_del(sk, sin->sin_addr.s_addr);
964 }
965
966 if (cmd.tcpm_keylen > TCP_MD5SIG_MAXKEYLEN)
967 return -EINVAL;
968
969 if (!tcp_sk(sk)->md5sig_info) {
970 struct tcp_sock *tp = tcp_sk(sk);
971 struct tcp_md5sig_info *p = kzalloc(sizeof(*p), GFP_KERNEL);
972
973 if (!p)
974 return -EINVAL;
975
976 tp->md5sig_info = p;
977 sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
978 }
979
980 newkey = kmemdup(cmd.tcpm_key, cmd.tcpm_keylen, GFP_KERNEL);
981 if (!newkey)
982 return -ENOMEM;
983 return tcp_v4_md5_do_add(sk, sin->sin_addr.s_addr,
984 newkey, cmd.tcpm_keylen);
985 }
986
987 static int tcp_v4_md5_hash_pseudoheader(struct tcp_md5sig_pool *hp,
988 __be32 daddr, __be32 saddr, int nbytes)
989 {
990 struct tcp4_pseudohdr *bp;
991 struct scatterlist sg;
992
993 bp = &hp->md5_blk.ip4;
994
995 /*
996 * 1. the TCP pseudo-header (in the order: source IP address,
997 * destination IP address, zero-padded protocol number, and
998 * segment length)
999 */
1000 bp->saddr = saddr;
1001 bp->daddr = daddr;
1002 bp->pad = 0;
1003 bp->protocol = IPPROTO_TCP;
1004 bp->len = cpu_to_be16(nbytes);
1005
1006 sg_init_one(&sg, bp, sizeof(*bp));
1007 return crypto_hash_update(&hp->md5_desc, &sg, sizeof(*bp));
1008 }
1009
1010 static int tcp_v4_md5_hash_hdr(char *md5_hash, struct tcp_md5sig_key *key,
1011 __be32 daddr, __be32 saddr, struct tcphdr *th)
1012 {
1013 struct tcp_md5sig_pool *hp;
1014 struct hash_desc *desc;
1015
1016 hp = tcp_get_md5sig_pool();
1017 if (!hp)
1018 goto clear_hash_noput;
1019 desc = &hp->md5_desc;
1020
1021 if (crypto_hash_init(desc))
1022 goto clear_hash;
1023 if (tcp_v4_md5_hash_pseudoheader(hp, daddr, saddr, th->doff << 2))
1024 goto clear_hash;
1025 if (tcp_md5_hash_header(hp, th))
1026 goto clear_hash;
1027 if (tcp_md5_hash_key(hp, key))
1028 goto clear_hash;
1029 if (crypto_hash_final(desc, md5_hash))
1030 goto clear_hash;
1031
1032 tcp_put_md5sig_pool();
1033 return 0;
1034
1035 clear_hash:
1036 tcp_put_md5sig_pool();
1037 clear_hash_noput:
1038 memset(md5_hash, 0, 16);
1039 return 1;
1040 }
1041
1042 int tcp_v4_md5_hash_skb(char *md5_hash, struct tcp_md5sig_key *key,
1043 struct sock *sk, struct request_sock *req,
1044 struct sk_buff *skb)
1045 {
1046 struct tcp_md5sig_pool *hp;
1047 struct hash_desc *desc;
1048 struct tcphdr *th = tcp_hdr(skb);
1049 __be32 saddr, daddr;
1050
1051 if (sk) {
1052 saddr = inet_sk(sk)->saddr;
1053 daddr = inet_sk(sk)->daddr;
1054 } else if (req) {
1055 saddr = inet_rsk(req)->loc_addr;
1056 daddr = inet_rsk(req)->rmt_addr;
1057 } else {
1058 const struct iphdr *iph = ip_hdr(skb);
1059 saddr = iph->saddr;
1060 daddr = iph->daddr;
1061 }
1062
1063 hp = tcp_get_md5sig_pool();
1064 if (!hp)
1065 goto clear_hash_noput;
1066 desc = &hp->md5_desc;
1067
1068 if (crypto_hash_init(desc))
1069 goto clear_hash;
1070
1071 if (tcp_v4_md5_hash_pseudoheader(hp, daddr, saddr, skb->len))
1072 goto clear_hash;
1073 if (tcp_md5_hash_header(hp, th))
1074 goto clear_hash;
1075 if (tcp_md5_hash_skb_data(hp, skb, th->doff << 2))
1076 goto clear_hash;
1077 if (tcp_md5_hash_key(hp, key))
1078 goto clear_hash;
1079 if (crypto_hash_final(desc, md5_hash))
1080 goto clear_hash;
1081
1082 tcp_put_md5sig_pool();
1083 return 0;
1084
1085 clear_hash:
1086 tcp_put_md5sig_pool();
1087 clear_hash_noput:
1088 memset(md5_hash, 0, 16);
1089 return 1;
1090 }
1091
1092 EXPORT_SYMBOL(tcp_v4_md5_hash_skb);
1093
1094 static int tcp_v4_inbound_md5_hash(struct sock *sk, struct sk_buff *skb)
1095 {
1096 /*
1097 * This gets called for each TCP segment that arrives
1098 * so we want to be efficient.
1099 * We have 3 drop cases:
1100 * o No MD5 hash and one expected.
1101 * o MD5 hash and we're not expecting one.
1102 * o MD5 hash and its wrong.
1103 */
1104 __u8 *hash_location = NULL;
1105 struct tcp_md5sig_key *hash_expected;
1106 const struct iphdr *iph = ip_hdr(skb);
1107 struct tcphdr *th = tcp_hdr(skb);
1108 int genhash;
1109 unsigned char newhash[16];
1110
1111 hash_expected = tcp_v4_md5_do_lookup(sk, iph->saddr);
1112 hash_location = tcp_parse_md5sig_option(th);
1113
1114 /* We've parsed the options - do we have a hash? */
1115 if (!hash_expected && !hash_location)
1116 return 0;
1117
1118 if (hash_expected && !hash_location) {
1119 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPMD5NOTFOUND);
1120 return 1;
1121 }
1122
1123 if (!hash_expected && hash_location) {
1124 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPMD5UNEXPECTED);
1125 return 1;
1126 }
1127
1128 /* Okay, so this is hash_expected and hash_location -
1129 * so we need to calculate the checksum.
1130 */
1131 genhash = tcp_v4_md5_hash_skb(newhash,
1132 hash_expected,
1133 NULL, NULL, skb);
1134
1135 if (genhash || memcmp(hash_location, newhash, 16) != 0) {
1136 if (net_ratelimit()) {
1137 printk(KERN_INFO "MD5 Hash failed for "
1138 "(" NIPQUAD_FMT ", %d)->(" NIPQUAD_FMT ", %d)%s\n",
1139 NIPQUAD(iph->saddr), ntohs(th->source),
1140 NIPQUAD(iph->daddr), ntohs(th->dest),
1141 genhash ? " tcp_v4_calc_md5_hash failed" : "");
1142 }
1143 return 1;
1144 }
1145 return 0;
1146 }
1147
1148 #endif
1149
1150 struct request_sock_ops tcp_request_sock_ops __read_mostly = {
1151 .family = PF_INET,
1152 .obj_size = sizeof(struct tcp_request_sock),
1153 .rtx_syn_ack = tcp_v4_send_synack,
1154 .send_ack = tcp_v4_reqsk_send_ack,
1155 .destructor = tcp_v4_reqsk_destructor,
1156 .send_reset = tcp_v4_send_reset,
1157 };
1158
1159 #ifdef CONFIG_TCP_MD5SIG
1160 static struct tcp_request_sock_ops tcp_request_sock_ipv4_ops = {
1161 .md5_lookup = tcp_v4_reqsk_md5_lookup,
1162 };
1163 #endif
1164
1165 static struct timewait_sock_ops tcp_timewait_sock_ops = {
1166 .twsk_obj_size = sizeof(struct tcp_timewait_sock),
1167 .twsk_unique = tcp_twsk_unique,
1168 .twsk_destructor= tcp_twsk_destructor,
1169 };
1170
1171 int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb)
1172 {
1173 struct inet_request_sock *ireq;
1174 struct tcp_options_received tmp_opt;
1175 struct request_sock *req;
1176 __be32 saddr = ip_hdr(skb)->saddr;
1177 __be32 daddr = ip_hdr(skb)->daddr;
1178 __u32 isn = TCP_SKB_CB(skb)->when;
1179 struct dst_entry *dst = NULL;
1180 #ifdef CONFIG_SYN_COOKIES
1181 int want_cookie = 0;
1182 #else
1183 #define want_cookie 0 /* Argh, why doesn't gcc optimize this :( */
1184 #endif
1185
1186 /* Never answer to SYNs send to broadcast or multicast */
1187 if (skb->rtable->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))
1188 goto drop;
1189
1190 /* TW buckets are converted to open requests without
1191 * limitations, they conserve resources and peer is
1192 * evidently real one.
1193 */
1194 if (inet_csk_reqsk_queue_is_full(sk) && !isn) {
1195 #ifdef CONFIG_SYN_COOKIES
1196 if (sysctl_tcp_syncookies) {
1197 want_cookie = 1;
1198 } else
1199 #endif
1200 goto drop;
1201 }
1202
1203 /* Accept backlog is full. If we have already queued enough
1204 * of warm entries in syn queue, drop request. It is better than
1205 * clogging syn queue with openreqs with exponentially increasing
1206 * timeout.
1207 */
1208 if (sk_acceptq_is_full(sk) && inet_csk_reqsk_queue_young(sk) > 1)
1209 goto drop;
1210
1211 req = inet_reqsk_alloc(&tcp_request_sock_ops);
1212 if (!req)
1213 goto drop;
1214
1215 #ifdef CONFIG_TCP_MD5SIG
1216 tcp_rsk(req)->af_specific = &tcp_request_sock_ipv4_ops;
1217 #endif
1218
1219 tcp_clear_options(&tmp_opt);
1220 tmp_opt.mss_clamp = 536;
1221 tmp_opt.user_mss = tcp_sk(sk)->rx_opt.user_mss;
1222
1223 tcp_parse_options(skb, &tmp_opt, 0);
1224
1225 if (want_cookie && !tmp_opt.saw_tstamp)
1226 tcp_clear_options(&tmp_opt);
1227
1228 if (tmp_opt.saw_tstamp && !tmp_opt.rcv_tsval) {
1229 /* Some OSes (unknown ones, but I see them on web server, which
1230 * contains information interesting only for windows'
1231 * users) do not send their stamp in SYN. It is easy case.
1232 * We simply do not advertise TS support.
1233 */
1234 tmp_opt.saw_tstamp = 0;
1235 tmp_opt.tstamp_ok = 0;
1236 }
1237 tmp_opt.tstamp_ok = tmp_opt.saw_tstamp;
1238
1239 tcp_openreq_init(req, &tmp_opt, skb);
1240
1241 if (security_inet_conn_request(sk, skb, req))
1242 goto drop_and_free;
1243
1244 ireq = inet_rsk(req);
1245 ireq->loc_addr = daddr;
1246 ireq->rmt_addr = saddr;
1247 ireq->opt = tcp_v4_save_options(sk, skb);
1248 if (!want_cookie)
1249 TCP_ECN_create_request(req, tcp_hdr(skb));
1250
1251 if (want_cookie) {
1252 #ifdef CONFIG_SYN_COOKIES
1253 syn_flood_warning(skb);
1254 req->cookie_ts = tmp_opt.tstamp_ok;
1255 #endif
1256 isn = cookie_v4_init_sequence(sk, skb, &req->mss);
1257 } else if (!isn) {
1258 struct inet_peer *peer = NULL;
1259
1260 /* VJ's idea. We save last timestamp seen
1261 * from the destination in peer table, when entering
1262 * state TIME-WAIT, and check against it before
1263 * accepting new connection request.
1264 *
1265 * If "isn" is not zero, this request hit alive
1266 * timewait bucket, so that all the necessary checks
1267 * are made in the function processing timewait state.
1268 */
1269 if (tmp_opt.saw_tstamp &&
1270 tcp_death_row.sysctl_tw_recycle &&
1271 (dst = inet_csk_route_req(sk, req)) != NULL &&
1272 (peer = rt_get_peer((struct rtable *)dst)) != NULL &&
1273 peer->v4daddr == saddr) {
1274 if (get_seconds() < peer->tcp_ts_stamp + TCP_PAWS_MSL &&
1275 (s32)(peer->tcp_ts - req->ts_recent) >
1276 TCP_PAWS_WINDOW) {
1277 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_PAWSPASSIVEREJECTED);
1278 goto drop_and_release;
1279 }
1280 }
1281 /* Kill the following clause, if you dislike this way. */
1282 else if (!sysctl_tcp_syncookies &&
1283 (sysctl_max_syn_backlog - inet_csk_reqsk_queue_len(sk) <
1284 (sysctl_max_syn_backlog >> 2)) &&
1285 (!peer || !peer->tcp_ts_stamp) &&
1286 (!dst || !dst_metric(dst, RTAX_RTT))) {
1287 /* Without syncookies last quarter of
1288 * backlog is filled with destinations,
1289 * proven to be alive.
1290 * It means that we continue to communicate
1291 * to destinations, already remembered
1292 * to the moment of synflood.
1293 */
1294 LIMIT_NETDEBUG(KERN_DEBUG "TCP: drop open "
1295 "request from " NIPQUAD_FMT "/%u\n",
1296 NIPQUAD(saddr),
1297 ntohs(tcp_hdr(skb)->source));
1298 goto drop_and_release;
1299 }
1300
1301 isn = tcp_v4_init_sequence(skb);
1302 }
1303 tcp_rsk(req)->snt_isn = isn;
1304
1305 if (__tcp_v4_send_synack(sk, req, dst) || want_cookie)
1306 goto drop_and_free;
1307
1308 inet_csk_reqsk_queue_hash_add(sk, req, TCP_TIMEOUT_INIT);
1309 return 0;
1310
1311 drop_and_release:
1312 dst_release(dst);
1313 drop_and_free:
1314 reqsk_free(req);
1315 drop:
1316 return 0;
1317 }
1318
1319
1320 /*
1321 * The three way handshake has completed - we got a valid synack -
1322 * now create the new socket.
1323 */
1324 struct sock *tcp_v4_syn_recv_sock(struct sock *sk, struct sk_buff *skb,
1325 struct request_sock *req,
1326 struct dst_entry *dst)
1327 {
1328 struct inet_request_sock *ireq;
1329 struct inet_sock *newinet;
1330 struct tcp_sock *newtp;
1331 struct sock *newsk;
1332 #ifdef CONFIG_TCP_MD5SIG
1333 struct tcp_md5sig_key *key;
1334 #endif
1335
1336 if (sk_acceptq_is_full(sk))
1337 goto exit_overflow;
1338
1339 if (!dst && (dst = inet_csk_route_req(sk, req)) == NULL)
1340 goto exit;
1341
1342 newsk = tcp_create_openreq_child(sk, req, skb);
1343 if (!newsk)
1344 goto exit;
1345
1346 newsk->sk_gso_type = SKB_GSO_TCPV4;
1347 sk_setup_caps(newsk, dst);
1348
1349 newtp = tcp_sk(newsk);
1350 newinet = inet_sk(newsk);
1351 ireq = inet_rsk(req);
1352 newinet->daddr = ireq->rmt_addr;
1353 newinet->rcv_saddr = ireq->loc_addr;
1354 newinet->saddr = ireq->loc_addr;
1355 newinet->opt = ireq->opt;
1356 ireq->opt = NULL;
1357 newinet->mc_index = inet_iif(skb);
1358 newinet->mc_ttl = ip_hdr(skb)->ttl;
1359 inet_csk(newsk)->icsk_ext_hdr_len = 0;
1360 if (newinet->opt)
1361 inet_csk(newsk)->icsk_ext_hdr_len = newinet->opt->optlen;
1362 newinet->id = newtp->write_seq ^ jiffies;
1363
1364 tcp_mtup_init(newsk);
1365 tcp_sync_mss(newsk, dst_mtu(dst));
1366 newtp->advmss = dst_metric(dst, RTAX_ADVMSS);
1367 tcp_initialize_rcv_mss(newsk);
1368
1369 #ifdef CONFIG_TCP_MD5SIG
1370 /* Copy over the MD5 key from the original socket */
1371 if ((key = tcp_v4_md5_do_lookup(sk, newinet->daddr)) != NULL) {
1372 /*
1373 * We're using one, so create a matching key
1374 * on the newsk structure. If we fail to get
1375 * memory, then we end up not copying the key
1376 * across. Shucks.
1377 */
1378 char *newkey = kmemdup(key->key, key->keylen, GFP_ATOMIC);
1379 if (newkey != NULL)
1380 tcp_v4_md5_do_add(newsk, inet_sk(sk)->daddr,
1381 newkey, key->keylen);
1382 newsk->sk_route_caps &= ~NETIF_F_GSO_MASK;
1383 }
1384 #endif
1385
1386 __inet_hash_nolisten(newsk);
1387 __inet_inherit_port(sk, newsk);
1388
1389 return newsk;
1390
1391 exit_overflow:
1392 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENOVERFLOWS);
1393 exit:
1394 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENDROPS);
1395 dst_release(dst);
1396 return NULL;
1397 }
1398
1399 static struct sock *tcp_v4_hnd_req(struct sock *sk, struct sk_buff *skb)
1400 {
1401 struct tcphdr *th = tcp_hdr(skb);
1402 const struct iphdr *iph = ip_hdr(skb);
1403 struct sock *nsk;
1404 struct request_sock **prev;
1405 /* Find possible connection requests. */
1406 struct request_sock *req = inet_csk_search_req(sk, &prev, th->source,
1407 iph->saddr, iph->daddr);
1408 if (req)
1409 return tcp_check_req(sk, skb, req, prev);
1410
1411 nsk = inet_lookup_established(sock_net(sk), &tcp_hashinfo, iph->saddr,
1412 th->source, iph->daddr, th->dest, inet_iif(skb));
1413
1414 if (nsk) {
1415 if (nsk->sk_state != TCP_TIME_WAIT) {
1416 bh_lock_sock(nsk);
1417 return nsk;
1418 }
1419 inet_twsk_put(inet_twsk(nsk));
1420 return NULL;
1421 }
1422
1423 #ifdef CONFIG_SYN_COOKIES
1424 if (!th->rst && !th->syn && th->ack)
1425 sk = cookie_v4_check(sk, skb, &(IPCB(skb)->opt));
1426 #endif
1427 return sk;
1428 }
1429
1430 static __sum16 tcp_v4_checksum_init(struct sk_buff *skb)
1431 {
1432 const struct iphdr *iph = ip_hdr(skb);
1433
1434 if (skb->ip_summed == CHECKSUM_COMPLETE) {
1435 if (!tcp_v4_check(skb->len, iph->saddr,
1436 iph->daddr, skb->csum)) {
1437 skb->ip_summed = CHECKSUM_UNNECESSARY;
1438 return 0;
1439 }
1440 }
1441
1442 skb->csum = csum_tcpudp_nofold(iph->saddr, iph->daddr,
1443 skb->len, IPPROTO_TCP, 0);
1444
1445 if (skb->len <= 76) {
1446 return __skb_checksum_complete(skb);
1447 }
1448 return 0;
1449 }
1450
1451
1452 /* The socket must have it's spinlock held when we get
1453 * here.
1454 *
1455 * We have a potential double-lock case here, so even when
1456 * doing backlog processing we use the BH locking scheme.
1457 * This is because we cannot sleep with the original spinlock
1458 * held.
1459 */
1460 int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb)
1461 {
1462 struct sock *rsk;
1463 #ifdef CONFIG_TCP_MD5SIG
1464 /*
1465 * We really want to reject the packet as early as possible
1466 * if:
1467 * o We're expecting an MD5'd packet and this is no MD5 tcp option
1468 * o There is an MD5 option and we're not expecting one
1469 */
1470 if (tcp_v4_inbound_md5_hash(sk, skb))
1471 goto discard;
1472 #endif
1473
1474 if (sk->sk_state == TCP_ESTABLISHED) { /* Fast path */
1475 TCP_CHECK_TIMER(sk);
1476 if (tcp_rcv_established(sk, skb, tcp_hdr(skb), skb->len)) {
1477 rsk = sk;
1478 goto reset;
1479 }
1480 TCP_CHECK_TIMER(sk);
1481 return 0;
1482 }
1483
1484 if (skb->len < tcp_hdrlen(skb) || tcp_checksum_complete(skb))
1485 goto csum_err;
1486
1487 if (sk->sk_state == TCP_LISTEN) {
1488 struct sock *nsk = tcp_v4_hnd_req(sk, skb);
1489 if (!nsk)
1490 goto discard;
1491
1492 if (nsk != sk) {
1493 if (tcp_child_process(sk, nsk, skb)) {
1494 rsk = nsk;
1495 goto reset;
1496 }
1497 return 0;
1498 }
1499 }
1500
1501 TCP_CHECK_TIMER(sk);
1502 if (tcp_rcv_state_process(sk, skb, tcp_hdr(skb), skb->len)) {
1503 rsk = sk;
1504 goto reset;
1505 }
1506 TCP_CHECK_TIMER(sk);
1507 return 0;
1508
1509 reset:
1510 tcp_v4_send_reset(rsk, skb);
1511 discard:
1512 kfree_skb(skb);
1513 /* Be careful here. If this function gets more complicated and
1514 * gcc suffers from register pressure on the x86, sk (in %ebx)
1515 * might be destroyed here. This current version compiles correctly,
1516 * but you have been warned.
1517 */
1518 return 0;
1519
1520 csum_err:
1521 TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_INERRS);
1522 goto discard;
1523 }
1524
1525 /*
1526 * From tcp_input.c
1527 */
1528
1529 int tcp_v4_rcv(struct sk_buff *skb)
1530 {
1531 const struct iphdr *iph;
1532 struct tcphdr *th;
1533 struct sock *sk;
1534 int ret;
1535 struct net *net = dev_net(skb->dev);
1536
1537 if (skb->pkt_type != PACKET_HOST)
1538 goto discard_it;
1539
1540 /* Count it even if it's bad */
1541 TCP_INC_STATS_BH(net, TCP_MIB_INSEGS);
1542
1543 if (!pskb_may_pull(skb, sizeof(struct tcphdr)))
1544 goto discard_it;
1545
1546 th = tcp_hdr(skb);
1547
1548 if (th->doff < sizeof(struct tcphdr) / 4)
1549 goto bad_packet;
1550 if (!pskb_may_pull(skb, th->doff * 4))
1551 goto discard_it;
1552
1553 /* An explanation is required here, I think.
1554 * Packet length and doff are validated by header prediction,
1555 * provided case of th->doff==0 is eliminated.
1556 * So, we defer the checks. */
1557 if (!skb_csum_unnecessary(skb) && tcp_v4_checksum_init(skb))
1558 goto bad_packet;
1559
1560 th = tcp_hdr(skb);
1561 iph = ip_hdr(skb);
1562 TCP_SKB_CB(skb)->seq = ntohl(th->seq);
1563 TCP_SKB_CB(skb)->end_seq = (TCP_SKB_CB(skb)->seq + th->syn + th->fin +
1564 skb->len - th->doff * 4);
1565 TCP_SKB_CB(skb)->ack_seq = ntohl(th->ack_seq);
1566 TCP_SKB_CB(skb)->when = 0;
1567 TCP_SKB_CB(skb)->flags = iph->tos;
1568 TCP_SKB_CB(skb)->sacked = 0;
1569
1570 sk = __inet_lookup(net, &tcp_hashinfo, iph->saddr,
1571 th->source, iph->daddr, th->dest, inet_iif(skb));
1572 if (!sk)
1573 goto no_tcp_socket;
1574
1575 process:
1576 if (sk->sk_state == TCP_TIME_WAIT)
1577 goto do_time_wait;
1578
1579 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
1580 goto discard_and_relse;
1581 nf_reset(skb);
1582
1583 if (sk_filter(sk, skb))
1584 goto discard_and_relse;
1585
1586 skb->dev = NULL;
1587
1588 bh_lock_sock_nested(sk);
1589 ret = 0;
1590 if (!sock_owned_by_user(sk)) {
1591 #ifdef CONFIG_NET_DMA
1592 struct tcp_sock *tp = tcp_sk(sk);
1593 if (!tp->ucopy.dma_chan && tp->ucopy.pinned_list)
1594 tp->ucopy.dma_chan = get_softnet_dma();
1595 if (tp->ucopy.dma_chan)
1596 ret = tcp_v4_do_rcv(sk, skb);
1597 else
1598 #endif
1599 {
1600 if (!tcp_prequeue(sk, skb))
1601 ret = tcp_v4_do_rcv(sk, skb);
1602 }
1603 } else
1604 sk_add_backlog(sk, skb);
1605 bh_unlock_sock(sk);
1606
1607 sock_put(sk);
1608
1609 return ret;
1610
1611 no_tcp_socket:
1612 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
1613 goto discard_it;
1614
1615 if (skb->len < (th->doff << 2) || tcp_checksum_complete(skb)) {
1616 bad_packet:
1617 TCP_INC_STATS_BH(net, TCP_MIB_INERRS);
1618 } else {
1619 tcp_v4_send_reset(NULL, skb);
1620 }
1621
1622 discard_it:
1623 /* Discard frame. */
1624 kfree_skb(skb);
1625 return 0;
1626
1627 discard_and_relse:
1628 sock_put(sk);
1629 goto discard_it;
1630
1631 do_time_wait:
1632 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) {
1633 inet_twsk_put(inet_twsk(sk));
1634 goto discard_it;
1635 }
1636
1637 if (skb->len < (th->doff << 2) || tcp_checksum_complete(skb)) {
1638 TCP_INC_STATS_BH(net, TCP_MIB_INERRS);
1639 inet_twsk_put(inet_twsk(sk));
1640 goto discard_it;
1641 }
1642 switch (tcp_timewait_state_process(inet_twsk(sk), skb, th)) {
1643 case TCP_TW_SYN: {
1644 struct sock *sk2 = inet_lookup_listener(dev_net(skb->dev),
1645 &tcp_hashinfo,
1646 iph->daddr, th->dest,
1647 inet_iif(skb));
1648 if (sk2) {
1649 inet_twsk_deschedule(inet_twsk(sk), &tcp_death_row);
1650 inet_twsk_put(inet_twsk(sk));
1651 sk = sk2;
1652 goto process;
1653 }
1654 /* Fall through to ACK */
1655 }
1656 case TCP_TW_ACK:
1657 tcp_v4_timewait_ack(sk, skb);
1658 break;
1659 case TCP_TW_RST:
1660 goto no_tcp_socket;
1661 case TCP_TW_SUCCESS:;
1662 }
1663 goto discard_it;
1664 }
1665
1666 /* VJ's idea. Save last timestamp seen from this destination
1667 * and hold it at least for normal timewait interval to use for duplicate
1668 * segment detection in subsequent connections, before they enter synchronized
1669 * state.
1670 */
1671
1672 int tcp_v4_remember_stamp(struct sock *sk)
1673 {
1674 struct inet_sock *inet = inet_sk(sk);
1675 struct tcp_sock *tp = tcp_sk(sk);
1676 struct rtable *rt = (struct rtable *)__sk_dst_get(sk);
1677 struct inet_peer *peer = NULL;
1678 int release_it = 0;
1679
1680 if (!rt || rt->rt_dst != inet->daddr) {
1681 peer = inet_getpeer(inet->daddr, 1);
1682 release_it = 1;
1683 } else {
1684 if (!rt->peer)
1685 rt_bind_peer(rt, 1);
1686 peer = rt->peer;
1687 }
1688
1689 if (peer) {
1690 if ((s32)(peer->tcp_ts - tp->rx_opt.ts_recent) <= 0 ||
1691 (peer->tcp_ts_stamp + TCP_PAWS_MSL < get_seconds() &&
1692 peer->tcp_ts_stamp <= tp->rx_opt.ts_recent_stamp)) {
1693 peer->tcp_ts_stamp = tp->rx_opt.ts_recent_stamp;
1694 peer->tcp_ts = tp->rx_opt.ts_recent;
1695 }
1696 if (release_it)
1697 inet_putpeer(peer);
1698 return 1;
1699 }
1700
1701 return 0;
1702 }
1703
1704 int tcp_v4_tw_remember_stamp(struct inet_timewait_sock *tw)
1705 {
1706 struct inet_peer *peer = inet_getpeer(tw->tw_daddr, 1);
1707
1708 if (peer) {
1709 const struct tcp_timewait_sock *tcptw = tcp_twsk((struct sock *)tw);
1710
1711 if ((s32)(peer->tcp_ts - tcptw->tw_ts_recent) <= 0 ||
1712 (peer->tcp_ts_stamp + TCP_PAWS_MSL < get_seconds() &&
1713 peer->tcp_ts_stamp <= tcptw->tw_ts_recent_stamp)) {
1714 peer->tcp_ts_stamp = tcptw->tw_ts_recent_stamp;
1715 peer->tcp_ts = tcptw->tw_ts_recent;
1716 }
1717 inet_putpeer(peer);
1718 return 1;
1719 }
1720
1721 return 0;
1722 }
1723
1724 struct inet_connection_sock_af_ops ipv4_specific = {
1725 .queue_xmit = ip_queue_xmit,
1726 .send_check = tcp_v4_send_check,
1727 .rebuild_header = inet_sk_rebuild_header,
1728 .conn_request = tcp_v4_conn_request,
1729 .syn_recv_sock = tcp_v4_syn_recv_sock,
1730 .remember_stamp = tcp_v4_remember_stamp,
1731 .net_header_len = sizeof(struct iphdr),
1732 .setsockopt = ip_setsockopt,
1733 .getsockopt = ip_getsockopt,
1734 .addr2sockaddr = inet_csk_addr2sockaddr,
1735 .sockaddr_len = sizeof(struct sockaddr_in),
1736 .bind_conflict = inet_csk_bind_conflict,
1737 #ifdef CONFIG_COMPAT
1738 .compat_setsockopt = compat_ip_setsockopt,
1739 .compat_getsockopt = compat_ip_getsockopt,
1740 #endif
1741 };
1742
1743 #ifdef CONFIG_TCP_MD5SIG
1744 static struct tcp_sock_af_ops tcp_sock_ipv4_specific = {
1745 .md5_lookup = tcp_v4_md5_lookup,
1746 .calc_md5_hash = tcp_v4_md5_hash_skb,
1747 .md5_add = tcp_v4_md5_add_func,
1748 .md5_parse = tcp_v4_parse_md5_keys,
1749 };
1750 #endif
1751
1752 /* NOTE: A lot of things set to zero explicitly by call to
1753 * sk_alloc() so need not be done here.
1754 */
1755 static int tcp_v4_init_sock(struct sock *sk)
1756 {
1757 struct inet_connection_sock *icsk = inet_csk(sk);
1758 struct tcp_sock *tp = tcp_sk(sk);
1759
1760 skb_queue_head_init(&tp->out_of_order_queue);
1761 tcp_init_xmit_timers(sk);
1762 tcp_prequeue_init(tp);
1763
1764 icsk->icsk_rto = TCP_TIMEOUT_INIT;
1765 tp->mdev = TCP_TIMEOUT_INIT;
1766
1767 /* So many TCP implementations out there (incorrectly) count the
1768 * initial SYN frame in their delayed-ACK and congestion control
1769 * algorithms that we must have the following bandaid to talk
1770 * efficiently to them. -DaveM
1771 */
1772 tp->snd_cwnd = 2;
1773
1774 /* See draft-stevens-tcpca-spec-01 for discussion of the
1775 * initialization of these values.
1776 */
1777 tp->snd_ssthresh = 0x7fffffff; /* Infinity */
1778 tp->snd_cwnd_clamp = ~0;
1779 tp->mss_cache = 536;
1780
1781 tp->reordering = sysctl_tcp_reordering;
1782 icsk->icsk_ca_ops = &tcp_init_congestion_ops;
1783
1784 sk->sk_state = TCP_CLOSE;
1785
1786 sk->sk_write_space = sk_stream_write_space;
1787 sock_set_flag(sk, SOCK_USE_WRITE_QUEUE);
1788
1789 icsk->icsk_af_ops = &ipv4_specific;
1790 icsk->icsk_sync_mss = tcp_sync_mss;
1791 #ifdef CONFIG_TCP_MD5SIG
1792 tp->af_specific = &tcp_sock_ipv4_specific;
1793 #endif
1794
1795 sk->sk_sndbuf = sysctl_tcp_wmem[1];
1796 sk->sk_rcvbuf = sysctl_tcp_rmem[1];
1797
1798 atomic_inc(&tcp_sockets_allocated);
1799
1800 return 0;
1801 }
1802
1803 void tcp_v4_destroy_sock(struct sock *sk)
1804 {
1805 struct tcp_sock *tp = tcp_sk(sk);
1806
1807 tcp_clear_xmit_timers(sk);
1808
1809 tcp_cleanup_congestion_control(sk);
1810
1811 /* Cleanup up the write buffer. */
1812 tcp_write_queue_purge(sk);
1813
1814 /* Cleans up our, hopefully empty, out_of_order_queue. */
1815 __skb_queue_purge(&tp->out_of_order_queue);
1816
1817 #ifdef CONFIG_TCP_MD5SIG
1818 /* Clean up the MD5 key list, if any */
1819 if (tp->md5sig_info) {
1820 tcp_v4_clear_md5_list(sk);
1821 kfree(tp->md5sig_info);
1822 tp->md5sig_info = NULL;
1823 }
1824 #endif
1825
1826 #ifdef CONFIG_NET_DMA
1827 /* Cleans up our sk_async_wait_queue */
1828 __skb_queue_purge(&sk->sk_async_wait_queue);
1829 #endif
1830
1831 /* Clean prequeue, it must be empty really */
1832 __skb_queue_purge(&tp->ucopy.prequeue);
1833
1834 /* Clean up a referenced TCP bind bucket. */
1835 if (inet_csk(sk)->icsk_bind_hash)
1836 inet_put_port(sk);
1837
1838 /*
1839 * If sendmsg cached page exists, toss it.
1840 */
1841 if (sk->sk_sndmsg_page) {
1842 __free_page(sk->sk_sndmsg_page);
1843 sk->sk_sndmsg_page = NULL;
1844 }
1845
1846 atomic_dec(&tcp_sockets_allocated);
1847 }
1848
1849 EXPORT_SYMBOL(tcp_v4_destroy_sock);
1850
1851 #ifdef CONFIG_PROC_FS
1852 /* Proc filesystem TCP sock list dumping. */
1853
1854 static inline struct inet_timewait_sock *tw_head(struct hlist_head *head)
1855 {
1856 return hlist_empty(head) ? NULL :
1857 list_entry(head->first, struct inet_timewait_sock, tw_node);
1858 }
1859
1860 static inline struct inet_timewait_sock *tw_next(struct inet_timewait_sock *tw)
1861 {
1862 return tw->tw_node.next ?
1863 hlist_entry(tw->tw_node.next, typeof(*tw), tw_node) : NULL;
1864 }
1865
1866 static void *listening_get_next(struct seq_file *seq, void *cur)
1867 {
1868 struct inet_connection_sock *icsk;
1869 struct hlist_node *node;
1870 struct sock *sk = cur;
1871 struct tcp_iter_state* st = seq->private;
1872 struct net *net = seq_file_net(seq);
1873
1874 if (!sk) {
1875 st->bucket = 0;
1876 sk = sk_head(&tcp_hashinfo.listening_hash[0]);
1877 goto get_sk;
1878 }
1879
1880 ++st->num;
1881
1882 if (st->state == TCP_SEQ_STATE_OPENREQ) {
1883 struct request_sock *req = cur;
1884
1885 icsk = inet_csk(st->syn_wait_sk);
1886 req = req->dl_next;
1887 while (1) {
1888 while (req) {
1889 if (req->rsk_ops->family == st->family) {
1890 cur = req;
1891 goto out;
1892 }
1893 req = req->dl_next;
1894 }
1895 if (++st->sbucket >= icsk->icsk_accept_queue.listen_opt->nr_table_entries)
1896 break;
1897 get_req:
1898 req = icsk->icsk_accept_queue.listen_opt->syn_table[st->sbucket];
1899 }
1900 sk = sk_next(st->syn_wait_sk);
1901 st->state = TCP_SEQ_STATE_LISTENING;
1902 read_unlock_bh(&icsk->icsk_accept_queue.syn_wait_lock);
1903 } else {
1904 icsk = inet_csk(sk);
1905 read_lock_bh(&icsk->icsk_accept_queue.syn_wait_lock);
1906 if (reqsk_queue_len(&icsk->icsk_accept_queue))
1907 goto start_req;
1908 read_unlock_bh(&icsk->icsk_accept_queue.syn_wait_lock);
1909 sk = sk_next(sk);
1910 }
1911 get_sk:
1912 sk_for_each_from(sk, node) {
1913 if (sk->sk_family == st->family && net_eq(sock_net(sk), net)) {
1914 cur = sk;
1915 goto out;
1916 }
1917 icsk = inet_csk(sk);
1918 read_lock_bh(&icsk->icsk_accept_queue.syn_wait_lock);
1919 if (reqsk_queue_len(&icsk->icsk_accept_queue)) {
1920 start_req:
1921 st->uid = sock_i_uid(sk);
1922 st->syn_wait_sk = sk;
1923 st->state = TCP_SEQ_STATE_OPENREQ;
1924 st->sbucket = 0;
1925 goto get_req;
1926 }
1927 read_unlock_bh(&icsk->icsk_accept_queue.syn_wait_lock);
1928 }
1929 if (++st->bucket < INET_LHTABLE_SIZE) {
1930 sk = sk_head(&tcp_hashinfo.listening_hash[st->bucket]);
1931 goto get_sk;
1932 }
1933 cur = NULL;
1934 out:
1935 return cur;
1936 }
1937
1938 static void *listening_get_idx(struct seq_file *seq, loff_t *pos)
1939 {
1940 void *rc = listening_get_next(seq, NULL);
1941
1942 while (rc && *pos) {
1943 rc = listening_get_next(seq, rc);
1944 --*pos;
1945 }
1946 return rc;
1947 }
1948
1949 static void *established_get_first(struct seq_file *seq)
1950 {
1951 struct tcp_iter_state* st = seq->private;
1952 struct net *net = seq_file_net(seq);
1953 void *rc = NULL;
1954
1955 for (st->bucket = 0; st->bucket < tcp_hashinfo.ehash_size; ++st->bucket) {
1956 struct sock *sk;
1957 struct hlist_node *node;
1958 struct inet_timewait_sock *tw;
1959 rwlock_t *lock = inet_ehash_lockp(&tcp_hashinfo, st->bucket);
1960
1961 read_lock_bh(lock);
1962 sk_for_each(sk, node, &tcp_hashinfo.ehash[st->bucket].chain) {
1963 if (sk->sk_family != st->family ||
1964 !net_eq(sock_net(sk), net)) {
1965 continue;
1966 }
1967 rc = sk;
1968 goto out;
1969 }
1970 st->state = TCP_SEQ_STATE_TIME_WAIT;
1971 inet_twsk_for_each(tw, node,
1972 &tcp_hashinfo.ehash[st->bucket].twchain) {
1973 if (tw->tw_family != st->family ||
1974 !net_eq(twsk_net(tw), net)) {
1975 continue;
1976 }
1977 rc = tw;
1978 goto out;
1979 }
1980 read_unlock_bh(lock);
1981 st->state = TCP_SEQ_STATE_ESTABLISHED;
1982 }
1983 out:
1984 return rc;
1985 }
1986
1987 static void *established_get_next(struct seq_file *seq, void *cur)
1988 {
1989 struct sock *sk = cur;
1990 struct inet_timewait_sock *tw;
1991 struct hlist_node *node;
1992 struct tcp_iter_state* st = seq->private;
1993 struct net *net = seq_file_net(seq);
1994
1995 ++st->num;
1996
1997 if (st->state == TCP_SEQ_STATE_TIME_WAIT) {
1998 tw = cur;
1999 tw = tw_next(tw);
2000 get_tw:
2001 while (tw && (tw->tw_family != st->family || !net_eq(twsk_net(tw), net))) {
2002 tw = tw_next(tw);
2003 }
2004 if (tw) {
2005 cur = tw;
2006 goto out;
2007 }
2008 read_unlock_bh(inet_ehash_lockp(&tcp_hashinfo, st->bucket));
2009 st->state = TCP_SEQ_STATE_ESTABLISHED;
2010
2011 if (++st->bucket < tcp_hashinfo.ehash_size) {
2012 read_lock_bh(inet_ehash_lockp(&tcp_hashinfo, st->bucket));
2013 sk = sk_head(&tcp_hashinfo.ehash[st->bucket].chain);
2014 } else {
2015 cur = NULL;
2016 goto out;
2017 }
2018 } else
2019 sk = sk_next(sk);
2020
2021 sk_for_each_from(sk, node) {
2022 if (sk->sk_family == st->family && net_eq(sock_net(sk), net))
2023 goto found;
2024 }
2025
2026 st->state = TCP_SEQ_STATE_TIME_WAIT;
2027 tw = tw_head(&tcp_hashinfo.ehash[st->bucket].twchain);
2028 goto get_tw;
2029 found:
2030 cur = sk;
2031 out:
2032 return cur;
2033 }
2034
2035 static void *established_get_idx(struct seq_file *seq, loff_t pos)
2036 {
2037 void *rc = established_get_first(seq);
2038
2039 while (rc && pos) {
2040 rc = established_get_next(seq, rc);
2041 --pos;
2042 }
2043 return rc;
2044 }
2045
2046 static void *tcp_get_idx(struct seq_file *seq, loff_t pos)
2047 {
2048 void *rc;
2049 struct tcp_iter_state* st = seq->private;
2050
2051 inet_listen_lock(&tcp_hashinfo);
2052 st->state = TCP_SEQ_STATE_LISTENING;
2053 rc = listening_get_idx(seq, &pos);
2054
2055 if (!rc) {
2056 inet_listen_unlock(&tcp_hashinfo);
2057 st->state = TCP_SEQ_STATE_ESTABLISHED;
2058 rc = established_get_idx(seq, pos);
2059 }
2060
2061 return rc;
2062 }
2063
2064 static void *tcp_seq_start(struct seq_file *seq, loff_t *pos)
2065 {
2066 struct tcp_iter_state* st = seq->private;
2067 st->state = TCP_SEQ_STATE_LISTENING;
2068 st->num = 0;
2069 return *pos ? tcp_get_idx(seq, *pos - 1) : SEQ_START_TOKEN;
2070 }
2071
2072 static void *tcp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2073 {
2074 void *rc = NULL;
2075 struct tcp_iter_state* st;
2076
2077 if (v == SEQ_START_TOKEN) {
2078 rc = tcp_get_idx(seq, 0);
2079 goto out;
2080 }
2081 st = seq->private;
2082
2083 switch (st->state) {
2084 case TCP_SEQ_STATE_OPENREQ:
2085 case TCP_SEQ_STATE_LISTENING:
2086 rc = listening_get_next(seq, v);
2087 if (!rc) {
2088 inet_listen_unlock(&tcp_hashinfo);
2089 st->state = TCP_SEQ_STATE_ESTABLISHED;
2090 rc = established_get_first(seq);
2091 }
2092 break;
2093 case TCP_SEQ_STATE_ESTABLISHED:
2094 case TCP_SEQ_STATE_TIME_WAIT:
2095 rc = established_get_next(seq, v);
2096 break;
2097 }
2098 out:
2099 ++*pos;
2100 return rc;
2101 }
2102
2103 static void tcp_seq_stop(struct seq_file *seq, void *v)
2104 {
2105 struct tcp_iter_state* st = seq->private;
2106
2107 switch (st->state) {
2108 case TCP_SEQ_STATE_OPENREQ:
2109 if (v) {
2110 struct inet_connection_sock *icsk = inet_csk(st->syn_wait_sk);
2111 read_unlock_bh(&icsk->icsk_accept_queue.syn_wait_lock);
2112 }
2113 case TCP_SEQ_STATE_LISTENING:
2114 if (v != SEQ_START_TOKEN)
2115 inet_listen_unlock(&tcp_hashinfo);
2116 break;
2117 case TCP_SEQ_STATE_TIME_WAIT:
2118 case TCP_SEQ_STATE_ESTABLISHED:
2119 if (v)
2120 read_unlock_bh(inet_ehash_lockp(&tcp_hashinfo, st->bucket));
2121 break;
2122 }
2123 }
2124
2125 static int tcp_seq_open(struct inode *inode, struct file *file)
2126 {
2127 struct tcp_seq_afinfo *afinfo = PDE(inode)->data;
2128 struct tcp_iter_state *s;
2129 int err;
2130
2131 err = seq_open_net(inode, file, &afinfo->seq_ops,
2132 sizeof(struct tcp_iter_state));
2133 if (err < 0)
2134 return err;
2135
2136 s = ((struct seq_file *)file->private_data)->private;
2137 s->family = afinfo->family;
2138 return 0;
2139 }
2140
2141 int tcp_proc_register(struct net *net, struct tcp_seq_afinfo *afinfo)
2142 {
2143 int rc = 0;
2144 struct proc_dir_entry *p;
2145
2146 afinfo->seq_fops.open = tcp_seq_open;
2147 afinfo->seq_fops.read = seq_read;
2148 afinfo->seq_fops.llseek = seq_lseek;
2149 afinfo->seq_fops.release = seq_release_net;
2150
2151 afinfo->seq_ops.start = tcp_seq_start;
2152 afinfo->seq_ops.next = tcp_seq_next;
2153 afinfo->seq_ops.stop = tcp_seq_stop;
2154
2155 p = proc_create_data(afinfo->name, S_IRUGO, net->proc_net,
2156 &afinfo->seq_fops, afinfo);
2157 if (!p)
2158 rc = -ENOMEM;
2159 return rc;
2160 }
2161
2162 void tcp_proc_unregister(struct net *net, struct tcp_seq_afinfo *afinfo)
2163 {
2164 proc_net_remove(net, afinfo->name);
2165 }
2166
2167 static void get_openreq4(struct sock *sk, struct request_sock *req,
2168 struct seq_file *f, int i, int uid, int *len)
2169 {
2170 const struct inet_request_sock *ireq = inet_rsk(req);
2171 int ttd = req->expires - jiffies;
2172
2173 seq_printf(f, "%4d: %08X:%04X %08X:%04X"
2174 " %02X %08X:%08X %02X:%08lX %08X %5d %8d %u %d %p%n",
2175 i,
2176 ireq->loc_addr,
2177 ntohs(inet_sk(sk)->sport),
2178 ireq->rmt_addr,
2179 ntohs(ireq->rmt_port),
2180 TCP_SYN_RECV,
2181 0, 0, /* could print option size, but that is af dependent. */
2182 1, /* timers active (only the expire timer) */
2183 jiffies_to_clock_t(ttd),
2184 req->retrans,
2185 uid,
2186 0, /* non standard timer */
2187 0, /* open_requests have no inode */
2188 atomic_read(&sk->sk_refcnt),
2189 req,
2190 len);
2191 }
2192
2193 static void get_tcp4_sock(struct sock *sk, struct seq_file *f, int i, int *len)
2194 {
2195 int timer_active;
2196 unsigned long timer_expires;
2197 struct tcp_sock *tp = tcp_sk(sk);
2198 const struct inet_connection_sock *icsk = inet_csk(sk);
2199 struct inet_sock *inet = inet_sk(sk);
2200 __be32 dest = inet->daddr;
2201 __be32 src = inet->rcv_saddr;
2202 __u16 destp = ntohs(inet->dport);
2203 __u16 srcp = ntohs(inet->sport);
2204
2205 if (icsk->icsk_pending == ICSK_TIME_RETRANS) {
2206 timer_active = 1;
2207 timer_expires = icsk->icsk_timeout;
2208 } else if (icsk->icsk_pending == ICSK_TIME_PROBE0) {
2209 timer_active = 4;
2210 timer_expires = icsk->icsk_timeout;
2211 } else if (timer_pending(&sk->sk_timer)) {
2212 timer_active = 2;
2213 timer_expires = sk->sk_timer.expires;
2214 } else {
2215 timer_active = 0;
2216 timer_expires = jiffies;
2217 }
2218
2219 seq_printf(f, "%4d: %08X:%04X %08X:%04X %02X %08X:%08X %02X:%08lX "
2220 "%08X %5d %8d %lu %d %p %lu %lu %u %u %d%n",
2221 i, src, srcp, dest, destp, sk->sk_state,
2222 tp->write_seq - tp->snd_una,
2223 sk->sk_state == TCP_LISTEN ? sk->sk_ack_backlog :
2224 (tp->rcv_nxt - tp->copied_seq),
2225 timer_active,
2226 jiffies_to_clock_t(timer_expires - jiffies),
2227 icsk->icsk_retransmits,
2228 sock_i_uid(sk),
2229 icsk->icsk_probes_out,
2230 sock_i_ino(sk),
2231 atomic_read(&sk->sk_refcnt), sk,
2232 jiffies_to_clock_t(icsk->icsk_rto),
2233 jiffies_to_clock_t(icsk->icsk_ack.ato),
2234 (icsk->icsk_ack.quick << 1) | icsk->icsk_ack.pingpong,
2235 tp->snd_cwnd,
2236 tp->snd_ssthresh >= 0xFFFF ? -1 : tp->snd_ssthresh,
2237 len);
2238 }
2239
2240 static void get_timewait4_sock(struct inet_timewait_sock *tw,
2241 struct seq_file *f, int i, int *len)
2242 {
2243 __be32 dest, src;
2244 __u16 destp, srcp;
2245 int ttd = tw->tw_ttd - jiffies;
2246
2247 if (ttd < 0)
2248 ttd = 0;
2249
2250 dest = tw->tw_daddr;
2251 src = tw->tw_rcv_saddr;
2252 destp = ntohs(tw->tw_dport);
2253 srcp = ntohs(tw->tw_sport);
2254
2255 seq_printf(f, "%4d: %08X:%04X %08X:%04X"
2256 " %02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %p%n",
2257 i, src, srcp, dest, destp, tw->tw_substate, 0, 0,
2258 3, jiffies_to_clock_t(ttd), 0, 0, 0, 0,
2259 atomic_read(&tw->tw_refcnt), tw, len);
2260 }
2261
2262 #define TMPSZ 150
2263
2264 static int tcp4_seq_show(struct seq_file *seq, void *v)
2265 {
2266 struct tcp_iter_state* st;
2267 int len;
2268
2269 if (v == SEQ_START_TOKEN) {
2270 seq_printf(seq, "%-*s\n", TMPSZ - 1,
2271 " sl local_address rem_address st tx_queue "
2272 "rx_queue tr tm->when retrnsmt uid timeout "
2273 "inode");
2274 goto out;
2275 }
2276 st = seq->private;
2277
2278 switch (st->state) {
2279 case TCP_SEQ_STATE_LISTENING:
2280 case TCP_SEQ_STATE_ESTABLISHED:
2281 get_tcp4_sock(v, seq, st->num, &len);
2282 break;
2283 case TCP_SEQ_STATE_OPENREQ:
2284 get_openreq4(st->syn_wait_sk, v, seq, st->num, st->uid, &len);
2285 break;
2286 case TCP_SEQ_STATE_TIME_WAIT:
2287 get_timewait4_sock(v, seq, st->num, &len);
2288 break;
2289 }
2290 seq_printf(seq, "%*s\n", TMPSZ - 1 - len, "");
2291 out:
2292 return 0;
2293 }
2294
2295 static struct tcp_seq_afinfo tcp4_seq_afinfo = {
2296 .name = "tcp",
2297 .family = AF_INET,
2298 .seq_fops = {
2299 .owner = THIS_MODULE,
2300 },
2301 .seq_ops = {
2302 .show = tcp4_seq_show,
2303 },
2304 };
2305
2306 static int tcp4_proc_init_net(struct net *net)
2307 {
2308 return tcp_proc_register(net, &tcp4_seq_afinfo);
2309 }
2310
2311 static void tcp4_proc_exit_net(struct net *net)
2312 {
2313 tcp_proc_unregister(net, &tcp4_seq_afinfo);
2314 }
2315
2316 static struct pernet_operations tcp4_net_ops = {
2317 .init = tcp4_proc_init_net,
2318 .exit = tcp4_proc_exit_net,
2319 };
2320
2321 int __init tcp4_proc_init(void)
2322 {
2323 return register_pernet_subsys(&tcp4_net_ops);
2324 }
2325
2326 void tcp4_proc_exit(void)
2327 {
2328 unregister_pernet_subsys(&tcp4_net_ops);
2329 }
2330 #endif /* CONFIG_PROC_FS */
2331
2332 struct proto tcp_prot = {
2333 .name = "TCP",
2334 .owner = THIS_MODULE,
2335 .close = tcp_close,
2336 .connect = tcp_v4_connect,
2337 .disconnect = tcp_disconnect,
2338 .accept = inet_csk_accept,
2339 .ioctl = tcp_ioctl,
2340 .init = tcp_v4_init_sock,
2341 .destroy = tcp_v4_destroy_sock,
2342 .shutdown = tcp_shutdown,
2343 .setsockopt = tcp_setsockopt,
2344 .getsockopt = tcp_getsockopt,
2345 .recvmsg = tcp_recvmsg,
2346 .backlog_rcv = tcp_v4_do_rcv,
2347 .hash = inet_hash,
2348 .unhash = inet_unhash,
2349 .get_port = inet_csk_get_port,
2350 .enter_memory_pressure = tcp_enter_memory_pressure,
2351 .sockets_allocated = &tcp_sockets_allocated,
2352 .orphan_count = &tcp_orphan_count,
2353 .memory_allocated = &tcp_memory_allocated,
2354 .memory_pressure = &tcp_memory_pressure,
2355 .sysctl_mem = sysctl_tcp_mem,
2356 .sysctl_wmem = sysctl_tcp_wmem,
2357 .sysctl_rmem = sysctl_tcp_rmem,
2358 .max_header = MAX_TCP_HEADER,
2359 .obj_size = sizeof(struct tcp_sock),
2360 .twsk_prot = &tcp_timewait_sock_ops,
2361 .rsk_prot = &tcp_request_sock_ops,
2362 .h.hashinfo = &tcp_hashinfo,
2363 #ifdef CONFIG_COMPAT
2364 .compat_setsockopt = compat_tcp_setsockopt,
2365 .compat_getsockopt = compat_tcp_getsockopt,
2366 #endif
2367 };
2368
2369
2370 static int __net_init tcp_sk_init(struct net *net)
2371 {
2372 return inet_ctl_sock_create(&net->ipv4.tcp_sock,
2373 PF_INET, SOCK_RAW, IPPROTO_TCP, net);
2374 }
2375
2376 static void __net_exit tcp_sk_exit(struct net *net)
2377 {
2378 inet_ctl_sock_destroy(net->ipv4.tcp_sock);
2379 }
2380
2381 static struct pernet_operations __net_initdata tcp_sk_ops = {
2382 .init = tcp_sk_init,
2383 .exit = tcp_sk_exit,
2384 };
2385
2386 void __init tcp_v4_init(void)
2387 {
2388 if (register_pernet_device(&tcp_sk_ops))
2389 panic("Failed to create the TCP control socket.\n");
2390 }
2391
2392 EXPORT_SYMBOL(ipv4_specific);
2393 EXPORT_SYMBOL(tcp_hashinfo);
2394 EXPORT_SYMBOL(tcp_prot);
2395 EXPORT_SYMBOL(tcp_v4_conn_request);
2396 EXPORT_SYMBOL(tcp_v4_connect);
2397 EXPORT_SYMBOL(tcp_v4_do_rcv);
2398 EXPORT_SYMBOL(tcp_v4_remember_stamp);
2399 EXPORT_SYMBOL(tcp_v4_send_check);
2400 EXPORT_SYMBOL(tcp_v4_syn_recv_sock);
2401
2402 #ifdef CONFIG_PROC_FS
2403 EXPORT_SYMBOL(tcp_proc_register);
2404 EXPORT_SYMBOL(tcp_proc_unregister);
2405 #endif
2406 EXPORT_SYMBOL(sysctl_tcp_low_latency);
2407
Support for the Chinese Samsung S3C2440 based development boards