?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
intel_menlo: extract return values using PTR_ERR
[linux-2.6/sactl.git] / net / ipv4 / tcp_ipv4.c
blob63414ea427c578d9a9b1cb527cd87e4cc53ef5f3
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 * Version: $Id: tcp_ipv4.c,v 1.240 2002/02/01 22:01:04 davem Exp $
9 *
10 * IPv4 specific functions
11 *
12 *
13 * code split from:
14 * linux/ipv4/tcp.c
15 * linux/ipv4/tcp_input.c
16 * linux/ipv4/tcp_output.c
17 *
18 * See tcp.c for author information
19 *
20 * This program is free software; you can redistribute it and/or
21 * modify it under the terms of the GNU General Public License
22 * as published by the Free Software Foundation; either version
23 * 2 of the License, or (at your option) any later version.
24 */
25
26 /*
27 * Changes:
28 * David S. Miller : New socket lookup architecture.
29 * This code is dedicated to John Dyson.
30 * David S. Miller : Change semantics of established hash,
31 * half is devoted to TIME_WAIT sockets
32 * and the rest go in the other half.
33 * Andi Kleen : Add support for syncookies and fixed
34 * some bugs: ip options weren't passed to
35 * the TCP layer, missed a check for an
36 * ACK bit.
37 * Andi Kleen : Implemented fast path mtu discovery.
38 * Fixed many serious bugs in the
39 * request_sock handling and moved
40 * most of it into the af independent code.
41 * Added tail drop and some other bugfixes.
42 * Added new listen semantics.
43 * Mike McLagan : Routing by source
44 * Juan Jose Ciarlante: ip_dynaddr bits
45 * Andi Kleen: various fixes.
46 * Vitaly E. Lavrov : Transparent proxy revived after year
47 * coma.
48 * Andi Kleen : Fix new listen.
49 * Andi Kleen : Fix accept error reporting.
50 * YOSHIFUJI Hideaki @USAGI and: Support IPV6_V6ONLY socket option, which
51 * Alexey Kuznetsov allow both IPv4 and IPv6 sockets to bind
52 * a single port at the same time.
53 */
54
55
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
65 #include <net/net_namespace.h>
66 #include <net/icmp.h>
67 #include <net/inet_hashtables.h>
68 #include <net/tcp.h>
69 #include <net/transp_v6.h>
70 #include <net/ipv6.h>
71 #include <net/inet_common.h>
72 #include <net/timewait_sock.h>
73 #include <net/xfrm.h>
74 #include <net/netdma.h>
75
76 #include <linux/inet.h>
77 #include <linux/ipv6.h>
78 #include <linux/stddef.h>
79 #include <linux/proc_fs.h>
80 #include <linux/seq_file.h>
81
82 #include <linux/crypto.h>
83 #include <linux/scatterlist.h>
84
85 int sysctl_tcp_tw_reuse __read_mostly;
86 int sysctl_tcp_low_latency __read_mostly;
87
88 /* Check TCP sequence numbers in ICMP packets. */
89 #define ICMP_MIN_LENGTH 8
90
91 /* Socket used for sending RSTs */
92 static struct socket *tcp_socket __read_mostly;
93
94 void tcp_v4_send_check(struct sock *sk, int len, struct sk_buff *skb);
95
96 #ifdef CONFIG_TCP_MD5SIG
97 static struct tcp_md5sig_key *tcp_v4_md5_do_lookup(struct sock *sk,
98 __be32 addr);
99 static int tcp_v4_do_calc_md5_hash(char *md5_hash, struct tcp_md5sig_key *key,
100 __be32 saddr, __be32 daddr,
101 struct tcphdr *th, int protocol,
102 unsigned int tcplen);
103 #endif
104
105 struct inet_hashinfo __cacheline_aligned tcp_hashinfo = {
106 .lhash_lock = __RW_LOCK_UNLOCKED(tcp_hashinfo.lhash_lock),
107 .lhash_users = ATOMIC_INIT(0),
108 .lhash_wait = __WAIT_QUEUE_HEAD_INITIALIZER(tcp_hashinfo.lhash_wait),
109 };
110
111 static inline __u32 tcp_v4_init_sequence(struct sk_buff *skb)
112 {
113 return secure_tcp_sequence_number(ip_hdr(skb)->daddr,
114 ip_hdr(skb)->saddr,
115 tcp_hdr(skb)->dest,
116 tcp_hdr(skb)->source);
117 }
118
119 int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp)
120 {
121 const struct tcp_timewait_sock *tcptw = tcp_twsk(sktw);
122 struct tcp_sock *tp = tcp_sk(sk);
123
124 /* With PAWS, it is safe from the viewpoint
125 of data integrity. Even without PAWS it is safe provided sequence
126 spaces do not overlap i.e. at data rates <= 80Mbit/sec.
127
128 Actually, the idea is close to VJ's one, only timestamp cache is
129 held not per host, but per port pair and TW bucket is used as state
130 holder.
131
132 If TW bucket has been already destroyed we fall back to VJ's scheme
133 and use initial timestamp retrieved from peer table.
134 */
135 if (tcptw->tw_ts_recent_stamp &&
136 (twp == NULL || (sysctl_tcp_tw_reuse &&
137 get_seconds() - tcptw->tw_ts_recent_stamp > 1))) {
138 tp->write_seq = tcptw->tw_snd_nxt + 65535 + 2;
139 if (tp->write_seq == 0)
140 tp->write_seq = 1;
141 tp->rx_opt.ts_recent = tcptw->tw_ts_recent;
142 tp->rx_opt.ts_recent_stamp = tcptw->tw_ts_recent_stamp;
143 sock_hold(sktw);
144 return 1;
145 }
146
147 return 0;
148 }
149
150 EXPORT_SYMBOL_GPL(tcp_twsk_unique);
151
152 /* This will initiate an outgoing connection. */
153 int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len)
154 {
155 struct inet_sock *inet = inet_sk(sk);
156 struct tcp_sock *tp = tcp_sk(sk);
157 struct sockaddr_in *usin = (struct sockaddr_in *)uaddr;
158 struct rtable *rt;
159 __be32 daddr, nexthop;
160 int tmp;
161 int err;
162
163 if (addr_len < sizeof(struct sockaddr_in))
164 return -EINVAL;
165
166 if (usin->sin_family != AF_INET)
167 return -EAFNOSUPPORT;
168
169 nexthop = daddr = usin->sin_addr.s_addr;
170 if (inet->opt && inet->opt->srr) {
171 if (!daddr)
172 return -EINVAL;
173 nexthop = inet->opt->faddr;
174 }
175
176 tmp = ip_route_connect(&rt, nexthop, inet->saddr,
177 RT_CONN_FLAGS(sk), sk->sk_bound_dev_if,
178 IPPROTO_TCP,
179 inet->sport, usin->sin_port, sk, 1);
180 if (tmp < 0) {
181 if (tmp == -ENETUNREACH)
182 IP_INC_STATS_BH(IPSTATS_MIB_OUTNOROUTES);
183 return tmp;
184 }
185
186 if (rt->rt_flags & (RTCF_MULTICAST | RTCF_BROADCAST)) {
187 ip_rt_put(rt);
188 return -ENETUNREACH;
189 }
190
191 if (!inet->opt || !inet->opt->srr)
192 daddr = rt->rt_dst;
193
194 if (!inet->saddr)
195 inet->saddr = rt->rt_src;
196 inet->rcv_saddr = inet->saddr;
197
198 if (tp->rx_opt.ts_recent_stamp && inet->daddr != daddr) {
199 /* Reset inherited state */
200 tp->rx_opt.ts_recent = 0;
201 tp->rx_opt.ts_recent_stamp = 0;
202 tp->write_seq = 0;
203 }
204
205 if (tcp_death_row.sysctl_tw_recycle &&
206 !tp->rx_opt.ts_recent_stamp && rt->rt_dst == daddr) {
207 struct inet_peer *peer = rt_get_peer(rt);
208 /*
209 * VJ's idea. We save last timestamp seen from
210 * the destination in peer table, when entering state
211 * TIME-WAIT * and initialize rx_opt.ts_recent from it,
212 * when trying new connection.
213 */
214 if (peer != NULL &&
215 peer->tcp_ts_stamp + TCP_PAWS_MSL >= get_seconds()) {
216 tp->rx_opt.ts_recent_stamp = peer->tcp_ts_stamp;
217 tp->rx_opt.ts_recent = peer->tcp_ts;
218 }
219 }
220
221 inet->dport = usin->sin_port;
222 inet->daddr = daddr;
223
224 inet_csk(sk)->icsk_ext_hdr_len = 0;
225 if (inet->opt)
226 inet_csk(sk)->icsk_ext_hdr_len = inet->opt->optlen;
227
228 tp->rx_opt.mss_clamp = 536;
229
230 /* Socket identity is still unknown (sport may be zero).
231 * However we set state to SYN-SENT and not releasing socket
232 * lock select source port, enter ourselves into the hash tables and
233 * complete initialization after this.
234 */
235 tcp_set_state(sk, TCP_SYN_SENT);
236 err = inet_hash_connect(&tcp_death_row, sk);
237 if (err)
238 goto failure;
239
240 err = ip_route_newports(&rt, IPPROTO_TCP,
241 inet->sport, inet->dport, sk);
242 if (err)
243 goto failure;
244
245 /* OK, now commit destination to socket. */
246 sk->sk_gso_type = SKB_GSO_TCPV4;
247 sk_setup_caps(sk, &rt->u.dst);
248
249 if (!tp->write_seq)
250 tp->write_seq = secure_tcp_sequence_number(inet->saddr,
251 inet->daddr,
252 inet->sport,
253 usin->sin_port);
254
255 inet->id = tp->write_seq ^ jiffies;
256
257 err = tcp_connect(sk);
258 rt = NULL;
259 if (err)
260 goto failure;
261
262 return 0;
263
264 failure:
265 /*
266 * This unhashes the socket and releases the local port,
267 * if necessary.
268 */
269 tcp_set_state(sk, TCP_CLOSE);
270 ip_rt_put(rt);
271 sk->sk_route_caps = 0;
272 inet->dport = 0;
273 return err;
274 }
275
276 /*
277 * This routine does path mtu discovery as defined in RFC1191.
278 */
279 static void do_pmtu_discovery(struct sock *sk, struct iphdr *iph, u32 mtu)
280 {
281 struct dst_entry *dst;
282 struct inet_sock *inet = inet_sk(sk);
283
284 /* We are not interested in TCP_LISTEN and open_requests (SYN-ACKs
285 * send out by Linux are always <576bytes so they should go through
286 * unfragmented).
287 */
288 if (sk->sk_state == TCP_LISTEN)
289 return;
290
291 /* We don't check in the destentry if pmtu discovery is forbidden
292 * on this route. We just assume that no packet_to_big packets
293 * are send back when pmtu discovery is not active.
294 * There is a small race when the user changes this flag in the
295 * route, but I think that's acceptable.
296 */
297 if ((dst = __sk_dst_check(sk, 0)) == NULL)
298 return;
299
300 dst->ops->update_pmtu(dst, mtu);
301
302 /* Something is about to be wrong... Remember soft error
303 * for the case, if this connection will not able to recover.
304 */
305 if (mtu < dst_mtu(dst) && ip_dont_fragment(sk, dst))
306 sk->sk_err_soft = EMSGSIZE;
307
308 mtu = dst_mtu(dst);
309
310 if (inet->pmtudisc != IP_PMTUDISC_DONT &&
311 inet_csk(sk)->icsk_pmtu_cookie > mtu) {
312 tcp_sync_mss(sk, mtu);
313
314 /* Resend the TCP packet because it's
315 * clear that the old packet has been
316 * dropped. This is the new "fast" path mtu
317 * discovery.
318 */
319 tcp_simple_retransmit(sk);
320 } /* else let the usual retransmit timer handle it */
321 }
322
323 /*
324 * This routine is called by the ICMP module when it gets some
325 * sort of error condition. If err < 0 then the socket should
326 * be closed and the error returned to the user. If err > 0
327 * it's just the icmp type << 8 | icmp code. After adjustment
328 * header points to the first 8 bytes of the tcp header. We need
329 * to find the appropriate port.
330 *
331 * The locking strategy used here is very "optimistic". When
332 * someone else accesses the socket the ICMP is just dropped
333 * and for some paths there is no check at all.
334 * A more general error queue to queue errors for later handling
335 * is probably better.
336 *
337 */
338
339 void tcp_v4_err(struct sk_buff *skb, u32 info)
340 {
341 struct iphdr *iph = (struct iphdr *)skb->data;
342 struct tcphdr *th = (struct tcphdr *)(skb->data + (iph->ihl << 2));
343 struct tcp_sock *tp;
344 struct inet_sock *inet;
345 const int type = icmp_hdr(skb)->type;
346 const int code = icmp_hdr(skb)->code;
347 struct sock *sk;
348 __u32 seq;
349 int err;
350
351 if (skb->len < (iph->ihl << 2) + 8) {
352 ICMP_INC_STATS_BH(ICMP_MIB_INERRORS);
353 return;
354 }
355
356 sk = inet_lookup(skb->dev->nd_net, &tcp_hashinfo, iph->daddr, th->dest,
357 iph->saddr, th->source, inet_iif(skb));
358 if (!sk) {
359 ICMP_INC_STATS_BH(ICMP_MIB_INERRORS);
360 return;
361 }
362 if (sk->sk_state == TCP_TIME_WAIT) {
363 inet_twsk_put(inet_twsk(sk));
364 return;
365 }
366
367 bh_lock_sock(sk);
368 /* If too many ICMPs get dropped on busy
369 * servers this needs to be solved differently.
370 */
371 if (sock_owned_by_user(sk))
372 NET_INC_STATS_BH(LINUX_MIB_LOCKDROPPEDICMPS);
373
374 if (sk->sk_state == TCP_CLOSE)
375 goto out;
376
377 tp = tcp_sk(sk);
378 seq = ntohl(th->seq);
379 if (sk->sk_state != TCP_LISTEN &&
380 !between(seq, tp->snd_una, tp->snd_nxt)) {
381 NET_INC_STATS_BH(LINUX_MIB_OUTOFWINDOWICMPS);
382 goto out;
383 }
384
385 switch (type) {
386 case ICMP_SOURCE_QUENCH:
387 /* Just silently ignore these. */
388 goto out;
389 case ICMP_PARAMETERPROB:
390 err = EPROTO;
391 break;
392 case ICMP_DEST_UNREACH:
393 if (code > NR_ICMP_UNREACH)
394 goto out;
395
396 if (code == ICMP_FRAG_NEEDED) { /* PMTU discovery (RFC1191) */
397 if (!sock_owned_by_user(sk))
398 do_pmtu_discovery(sk, iph, info);
399 goto out;
400 }
401
402 err = icmp_err_convert[code].errno;
403 break;
404 case ICMP_TIME_EXCEEDED:
405 err = EHOSTUNREACH;
406 break;
407 default:
408 goto out;
409 }
410
411 switch (sk->sk_state) {
412 struct request_sock *req, **prev;
413 case TCP_LISTEN:
414 if (sock_owned_by_user(sk))
415 goto out;
416
417 req = inet_csk_search_req(sk, &prev, th->dest,
418 iph->daddr, iph->saddr);
419 if (!req)
420 goto out;
421
422 /* ICMPs are not backlogged, hence we cannot get
423 an established socket here.
424 */
425 BUG_TRAP(!req->sk);
426
427 if (seq != tcp_rsk(req)->snt_isn) {
428 NET_INC_STATS_BH(LINUX_MIB_OUTOFWINDOWICMPS);
429 goto out;
430 }
431
432 /*
433 * Still in SYN_RECV, just remove it silently.
434 * There is no good way to pass the error to the newly
435 * created socket, and POSIX does not want network
436 * errors returned from accept().
437 */
438 inet_csk_reqsk_queue_drop(sk, req, prev);
439 goto out;
440
441 case TCP_SYN_SENT:
442 case TCP_SYN_RECV: /* Cannot happen.
443 It can f.e. if SYNs crossed.
444 */
445 if (!sock_owned_by_user(sk)) {
446 sk->sk_err = err;
447
448 sk->sk_error_report(sk);
449
450 tcp_done(sk);
451 } else {
452 sk->sk_err_soft = err;
453 }
454 goto out;
455 }
456
457 /* If we've already connected we will keep trying
458 * until we time out, or the user gives up.
459 *
460 * rfc1122 4.2.3.9 allows to consider as hard errors
461 * only PROTO_UNREACH and PORT_UNREACH (well, FRAG_FAILED too,
462 * but it is obsoleted by pmtu discovery).
463 *
464 * Note, that in modern internet, where routing is unreliable
465 * and in each dark corner broken firewalls sit, sending random
466 * errors ordered by their masters even this two messages finally lose
467 * their original sense (even Linux sends invalid PORT_UNREACHs)
468 *
469 * Now we are in compliance with RFCs.
470 * --ANK (980905)
471 */
472
473 inet = inet_sk(sk);
474 if (!sock_owned_by_user(sk) && inet->recverr) {
475 sk->sk_err = err;
476 sk->sk_error_report(sk);
477 } else { /* Only an error on timeout */
478 sk->sk_err_soft = err;
479 }
480
481 out:
482 bh_unlock_sock(sk);
483 sock_put(sk);
484 }
485
486 /* This routine computes an IPv4 TCP checksum. */
487 void tcp_v4_send_check(struct sock *sk, int len, struct sk_buff *skb)
488 {
489 struct inet_sock *inet = inet_sk(sk);
490 struct tcphdr *th = tcp_hdr(skb);
491
492 if (skb->ip_summed == CHECKSUM_PARTIAL) {
493 th->check = ~tcp_v4_check(len, inet->saddr,
494 inet->daddr, 0);
495 skb->csum_start = skb_transport_header(skb) - skb->head;
496 skb->csum_offset = offsetof(struct tcphdr, check);
497 } else {
498 th->check = tcp_v4_check(len, inet->saddr, inet->daddr,
499 csum_partial((char *)th,
500 th->doff << 2,
501 skb->csum));
502 }
503 }
504
505 int tcp_v4_gso_send_check(struct sk_buff *skb)
506 {
507 const struct iphdr *iph;
508 struct tcphdr *th;
509
510 if (!pskb_may_pull(skb, sizeof(*th)))
511 return -EINVAL;
512
513 iph = ip_hdr(skb);
514 th = tcp_hdr(skb);
515
516 th->check = 0;
517 th->check = ~tcp_v4_check(skb->len, iph->saddr, iph->daddr, 0);
518 skb->csum_start = skb_transport_header(skb) - skb->head;
519 skb->csum_offset = offsetof(struct tcphdr, check);
520 skb->ip_summed = CHECKSUM_PARTIAL;
521 return 0;
522 }
523
524 /*
525 * This routine will send an RST to the other tcp.
526 *
527 * Someone asks: why I NEVER use socket parameters (TOS, TTL etc.)
528 * for reset.
529 * Answer: if a packet caused RST, it is not for a socket
530 * existing in our system, if it is matched to a socket,
531 * it is just duplicate segment or bug in other side's TCP.
532 * So that we build reply only basing on parameters
533 * arrived with segment.
534 * Exception: precedence violation. We do not implement it in any case.
535 */
536
537 static void tcp_v4_send_reset(struct sock *sk, struct sk_buff *skb)
538 {
539 struct tcphdr *th = tcp_hdr(skb);
540 struct {
541 struct tcphdr th;
542 #ifdef CONFIG_TCP_MD5SIG
543 __be32 opt[(TCPOLEN_MD5SIG_ALIGNED >> 2)];
544 #endif
545 } rep;
546 struct ip_reply_arg arg;
547 #ifdef CONFIG_TCP_MD5SIG
548 struct tcp_md5sig_key *key;
549 #endif
550
551 /* Never send a reset in response to a reset. */
552 if (th->rst)
553 return;
554
555 if (((struct rtable *)skb->dst)->rt_type != RTN_LOCAL)
556 return;
557
558 /* Swap the send and the receive. */
559 memset(&rep, 0, sizeof(rep));
560 rep.th.dest = th->source;
561 rep.th.source = th->dest;
562 rep.th.doff = sizeof(struct tcphdr) / 4;
563 rep.th.rst = 1;
564
565 if (th->ack) {
566 rep.th.seq = th->ack_seq;
567 } else {
568 rep.th.ack = 1;
569 rep.th.ack_seq = htonl(ntohl(th->seq) + th->syn + th->fin +
570 skb->len - (th->doff << 2));
571 }
572
573 memset(&arg, 0, sizeof(arg));
574 arg.iov[0].iov_base = (unsigned char *)&rep;
575 arg.iov[0].iov_len = sizeof(rep.th);
576
577 #ifdef CONFIG_TCP_MD5SIG
578 key = sk ? tcp_v4_md5_do_lookup(sk, ip_hdr(skb)->daddr) : NULL;
579 if (key) {
580 rep.opt[0] = htonl((TCPOPT_NOP << 24) |
581 (TCPOPT_NOP << 16) |
582 (TCPOPT_MD5SIG << 8) |
583 TCPOLEN_MD5SIG);
584 /* Update length and the length the header thinks exists */
585 arg.iov[0].iov_len += TCPOLEN_MD5SIG_ALIGNED;
586 rep.th.doff = arg.iov[0].iov_len / 4;
587
588 tcp_v4_do_calc_md5_hash((__u8 *)&rep.opt[1],
589 key,
590 ip_hdr(skb)->daddr,
591 ip_hdr(skb)->saddr,
592 &rep.th, IPPROTO_TCP,
593 arg.iov[0].iov_len);
594 }
595 #endif
596 arg.csum = csum_tcpudp_nofold(ip_hdr(skb)->daddr,
597 ip_hdr(skb)->saddr, /* XXX */
598 sizeof(struct tcphdr), IPPROTO_TCP, 0);
599 arg.csumoffset = offsetof(struct tcphdr, check) / 2;
600
601 ip_send_reply(tcp_socket->sk, skb, &arg, arg.iov[0].iov_len);
602
603 TCP_INC_STATS_BH(TCP_MIB_OUTSEGS);
604 TCP_INC_STATS_BH(TCP_MIB_OUTRSTS);
605 }
606
607 /* The code following below sending ACKs in SYN-RECV and TIME-WAIT states
608 outside socket context is ugly, certainly. What can I do?
609 */
610
611 static void tcp_v4_send_ack(struct tcp_timewait_sock *twsk,
612 struct sk_buff *skb, u32 seq, u32 ack,
613 u32 win, u32 ts)
614 {
615 struct tcphdr *th = tcp_hdr(skb);
616 struct {
617 struct tcphdr th;
618 __be32 opt[(TCPOLEN_TSTAMP_ALIGNED >> 2)
619 #ifdef CONFIG_TCP_MD5SIG
620 + (TCPOLEN_MD5SIG_ALIGNED >> 2)
621 #endif
622 ];
623 } rep;
624 struct ip_reply_arg arg;
625 #ifdef CONFIG_TCP_MD5SIG
626 struct tcp_md5sig_key *key;
627 struct tcp_md5sig_key tw_key;
628 #endif
629
630 memset(&rep.th, 0, sizeof(struct tcphdr));
631 memset(&arg, 0, sizeof(arg));
632
633 arg.iov[0].iov_base = (unsigned char *)&rep;
634 arg.iov[0].iov_len = sizeof(rep.th);
635 if (ts) {
636 rep.opt[0] = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
637 (TCPOPT_TIMESTAMP << 8) |
638 TCPOLEN_TIMESTAMP);
639 rep.opt[1] = htonl(tcp_time_stamp);
640 rep.opt[2] = htonl(ts);
641 arg.iov[0].iov_len += TCPOLEN_TSTAMP_ALIGNED;
642 }
643
644 /* Swap the send and the receive. */
645 rep.th.dest = th->source;
646 rep.th.source = th->dest;
647 rep.th.doff = arg.iov[0].iov_len / 4;
648 rep.th.seq = htonl(seq);
649 rep.th.ack_seq = htonl(ack);
650 rep.th.ack = 1;
651 rep.th.window = htons(win);
652
653 #ifdef CONFIG_TCP_MD5SIG
654 /*
655 * The SKB holds an imcoming packet, but may not have a valid ->sk
656 * pointer. This is especially the case when we're dealing with a
657 * TIME_WAIT ack, because the sk structure is long gone, and only
658 * the tcp_timewait_sock remains. So the md5 key is stashed in that
659 * structure, and we use it in preference. I believe that (twsk ||
660 * skb->sk) holds true, but we program defensively.
661 */
662 if (!twsk && skb->sk) {
663 key = tcp_v4_md5_do_lookup(skb->sk, ip_hdr(skb)->daddr);
664 } else if (twsk && twsk->tw_md5_keylen) {
665 tw_key.key = twsk->tw_md5_key;
666 tw_key.keylen = twsk->tw_md5_keylen;
667 key = &tw_key;
668 } else
669 key = NULL;
670
671 if (key) {
672 int offset = (ts) ? 3 : 0;
673
674 rep.opt[offset++] = htonl((TCPOPT_NOP << 24) |
675 (TCPOPT_NOP << 16) |
676 (TCPOPT_MD5SIG << 8) |
677 TCPOLEN_MD5SIG);
678 arg.iov[0].iov_len += TCPOLEN_MD5SIG_ALIGNED;
679 rep.th.doff = arg.iov[0].iov_len/4;
680
681 tcp_v4_do_calc_md5_hash((__u8 *)&rep.opt[offset],
682 key,
683 ip_hdr(skb)->daddr,
684 ip_hdr(skb)->saddr,
685 &rep.th, IPPROTO_TCP,
686 arg.iov[0].iov_len);
687 }
688 #endif
689 arg.csum = csum_tcpudp_nofold(ip_hdr(skb)->daddr,
690 ip_hdr(skb)->saddr, /* XXX */
691 arg.iov[0].iov_len, IPPROTO_TCP, 0);
692 arg.csumoffset = offsetof(struct tcphdr, check) / 2;
693 if (twsk)
694 arg.bound_dev_if = twsk->tw_sk.tw_bound_dev_if;
695
696 ip_send_reply(tcp_socket->sk, skb, &arg, arg.iov[0].iov_len);
697
698 TCP_INC_STATS_BH(TCP_MIB_OUTSEGS);
699 }
700
701 static void tcp_v4_timewait_ack(struct sock *sk, struct sk_buff *skb)
702 {
703 struct inet_timewait_sock *tw = inet_twsk(sk);
704 struct tcp_timewait_sock *tcptw = tcp_twsk(sk);
705
706 tcp_v4_send_ack(tcptw, skb, tcptw->tw_snd_nxt, tcptw->tw_rcv_nxt,
707 tcptw->tw_rcv_wnd >> tw->tw_rcv_wscale,
708 tcptw->tw_ts_recent);
709
710 inet_twsk_put(tw);
711 }
712
713 static void tcp_v4_reqsk_send_ack(struct sk_buff *skb,
714 struct request_sock *req)
715 {
716 tcp_v4_send_ack(NULL, skb, tcp_rsk(req)->snt_isn + 1,
717 tcp_rsk(req)->rcv_isn + 1, req->rcv_wnd,
718 req->ts_recent);
719 }
720
721 /*
722 * Send a SYN-ACK after having received an ACK.
723 * This still operates on a request_sock only, not on a big
724 * socket.
725 */
726 static int tcp_v4_send_synack(struct sock *sk, struct request_sock *req,
727 struct dst_entry *dst)
728 {
729 const struct inet_request_sock *ireq = inet_rsk(req);
730 int err = -1;
731 struct sk_buff * skb;
732
733 /* First, grab a route. */
734 if (!dst && (dst = inet_csk_route_req(sk, req)) == NULL)
735 goto out;
736
737 skb = tcp_make_synack(sk, dst, req);
738
739 if (skb) {
740 struct tcphdr *th = tcp_hdr(skb);
741
742 th->check = tcp_v4_check(skb->len,
743 ireq->loc_addr,
744 ireq->rmt_addr,
745 csum_partial((char *)th, skb->len,
746 skb->csum));
747
748 err = ip_build_and_send_pkt(skb, sk, ireq->loc_addr,
749 ireq->rmt_addr,
750 ireq->opt);
751 err = net_xmit_eval(err);
752 }
753
754 out:
755 dst_release(dst);
756 return err;
757 }
758
759 /*
760 * IPv4 request_sock destructor.
761 */
762 static void tcp_v4_reqsk_destructor(struct request_sock *req)
763 {
764 kfree(inet_rsk(req)->opt);
765 }
766
767 #ifdef CONFIG_SYN_COOKIES
768 static void syn_flood_warning(struct sk_buff *skb)
769 {
770 static unsigned long warntime;
771
772 if (time_after(jiffies, (warntime + HZ * 60))) {
773 warntime = jiffies;
774 printk(KERN_INFO
775 "possible SYN flooding on port %d. Sending cookies.\n",
776 ntohs(tcp_hdr(skb)->dest));
777 }
778 }
779 #endif
780
781 /*
782 * Save and compile IPv4 options into the request_sock if needed.
783 */
784 static struct ip_options *tcp_v4_save_options(struct sock *sk,
785 struct sk_buff *skb)
786 {
787 struct ip_options *opt = &(IPCB(skb)->opt);
788 struct ip_options *dopt = NULL;
789
790 if (opt && opt->optlen) {
791 int opt_size = optlength(opt);
792 dopt = kmalloc(opt_size, GFP_ATOMIC);
793 if (dopt) {
794 if (ip_options_echo(dopt, skb)) {
795 kfree(dopt);
796 dopt = NULL;
797 }
798 }
799 }
800 return dopt;
801 }
802
803 #ifdef CONFIG_TCP_MD5SIG
804 /*
805 * RFC2385 MD5 checksumming requires a mapping of
806 * IP address->MD5 Key.
807 * We need to maintain these in the sk structure.
808 */
809
810 /* Find the Key structure for an address. */
811 static struct tcp_md5sig_key *
812 tcp_v4_md5_do_lookup(struct sock *sk, __be32 addr)
813 {
814 struct tcp_sock *tp = tcp_sk(sk);
815 int i;
816
817 if (!tp->md5sig_info || !tp->md5sig_info->entries4)
818 return NULL;
819 for (i = 0; i < tp->md5sig_info->entries4; i++) {
820 if (tp->md5sig_info->keys4[i].addr == addr)
821 return &tp->md5sig_info->keys4[i].base;
822 }
823 return NULL;
824 }
825
826 struct tcp_md5sig_key *tcp_v4_md5_lookup(struct sock *sk,
827 struct sock *addr_sk)
828 {
829 return tcp_v4_md5_do_lookup(sk, inet_sk(addr_sk)->daddr);
830 }
831
832 EXPORT_SYMBOL(tcp_v4_md5_lookup);
833
834 static struct tcp_md5sig_key *tcp_v4_reqsk_md5_lookup(struct sock *sk,
835 struct request_sock *req)
836 {
837 return tcp_v4_md5_do_lookup(sk, inet_rsk(req)->rmt_addr);
838 }
839
840 /* This can be called on a newly created socket, from other files */
841 int tcp_v4_md5_do_add(struct sock *sk, __be32 addr,
842 u8 *newkey, u8 newkeylen)
843 {
844 /* Add Key to the list */
845 struct tcp_md5sig_key *key;
846 struct tcp_sock *tp = tcp_sk(sk);
847 struct tcp4_md5sig_key *keys;
848
849 key = tcp_v4_md5_do_lookup(sk, addr);
850 if (key) {
851 /* Pre-existing entry - just update that one. */
852 kfree(key->key);
853 key->key = newkey;
854 key->keylen = newkeylen;
855 } else {
856 struct tcp_md5sig_info *md5sig;
857
858 if (!tp->md5sig_info) {
859 tp->md5sig_info = kzalloc(sizeof(*tp->md5sig_info),
860 GFP_ATOMIC);
861 if (!tp->md5sig_info) {
862 kfree(newkey);
863 return -ENOMEM;
864 }
865 sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
866 }
867 if (tcp_alloc_md5sig_pool() == NULL) {
868 kfree(newkey);
869 return -ENOMEM;
870 }
871 md5sig = tp->md5sig_info;
872
873 if (md5sig->alloced4 == md5sig->entries4) {
874 keys = kmalloc((sizeof(*keys) *
875 (md5sig->entries4 + 1)), GFP_ATOMIC);
876 if (!keys) {
877 kfree(newkey);
878 tcp_free_md5sig_pool();
879 return -ENOMEM;
880 }
881
882 if (md5sig->entries4)
883 memcpy(keys, md5sig->keys4,
884 sizeof(*keys) * md5sig->entries4);
885
886 /* Free old key list, and reference new one */
887 kfree(md5sig->keys4);
888 md5sig->keys4 = keys;
889 md5sig->alloced4++;
890 }
891 md5sig->entries4++;
892 md5sig->keys4[md5sig->entries4 - 1].addr = addr;
893 md5sig->keys4[md5sig->entries4 - 1].base.key = newkey;
894 md5sig->keys4[md5sig->entries4 - 1].base.keylen = newkeylen;
895 }
896 return 0;
897 }
898
899 EXPORT_SYMBOL(tcp_v4_md5_do_add);
900
901 static int tcp_v4_md5_add_func(struct sock *sk, struct sock *addr_sk,
902 u8 *newkey, u8 newkeylen)
903 {
904 return tcp_v4_md5_do_add(sk, inet_sk(addr_sk)->daddr,
905 newkey, newkeylen);
906 }
907
908 int tcp_v4_md5_do_del(struct sock *sk, __be32 addr)
909 {
910 struct tcp_sock *tp = tcp_sk(sk);
911 int i;
912
913 for (i = 0; i < tp->md5sig_info->entries4; i++) {
914 if (tp->md5sig_info->keys4[i].addr == addr) {
915 /* Free the key */
916 kfree(tp->md5sig_info->keys4[i].base.key);
917 tp->md5sig_info->entries4--;
918
919 if (tp->md5sig_info->entries4 == 0) {
920 kfree(tp->md5sig_info->keys4);
921 tp->md5sig_info->keys4 = NULL;
922 tp->md5sig_info->alloced4 = 0;
923 } else if (tp->md5sig_info->entries4 != i) {
924 /* Need to do some manipulation */
925 memmove(&tp->md5sig_info->keys4[i],
926 &tp->md5sig_info->keys4[i+1],
927 (tp->md5sig_info->entries4 - i) *
928 sizeof(struct tcp4_md5sig_key));
929 }
930 tcp_free_md5sig_pool();
931 return 0;
932 }
933 }
934 return -ENOENT;
935 }
936
937 EXPORT_SYMBOL(tcp_v4_md5_do_del);
938
939 static void tcp_v4_clear_md5_list(struct sock *sk)
940 {
941 struct tcp_sock *tp = tcp_sk(sk);
942
943 /* Free each key, then the set of key keys,
944 * the crypto element, and then decrement our
945 * hold on the last resort crypto.
946 */
947 if (tp->md5sig_info->entries4) {
948 int i;
949 for (i = 0; i < tp->md5sig_info->entries4; i++)
950 kfree(tp->md5sig_info->keys4[i].base.key);
951 tp->md5sig_info->entries4 = 0;
952 tcp_free_md5sig_pool();
953 }
954 if (tp->md5sig_info->keys4) {
955 kfree(tp->md5sig_info->keys4);
956 tp->md5sig_info->keys4 = NULL;
957 tp->md5sig_info->alloced4 = 0;
958 }
959 }
960
961 static int tcp_v4_parse_md5_keys(struct sock *sk, char __user *optval,
962 int optlen)
963 {
964 struct tcp_md5sig cmd;
965 struct sockaddr_in *sin = (struct sockaddr_in *)&cmd.tcpm_addr;
966 u8 *newkey;
967
968 if (optlen < sizeof(cmd))
969 return -EINVAL;
970
971 if (copy_from_user(&cmd, optval, sizeof(cmd)))
972 return -EFAULT;
973
974 if (sin->sin_family != AF_INET)
975 return -EINVAL;
976
977 if (!cmd.tcpm_key || !cmd.tcpm_keylen) {
978 if (!tcp_sk(sk)->md5sig_info)
979 return -ENOENT;
980 return tcp_v4_md5_do_del(sk, sin->sin_addr.s_addr);
981 }
982
983 if (cmd.tcpm_keylen > TCP_MD5SIG_MAXKEYLEN)
984 return -EINVAL;
985
986 if (!tcp_sk(sk)->md5sig_info) {
987 struct tcp_sock *tp = tcp_sk(sk);
988 struct tcp_md5sig_info *p = kzalloc(sizeof(*p), GFP_KERNEL);
989
990 if (!p)
991 return -EINVAL;
992
993 tp->md5sig_info = p;
994 sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
995 }
996
997 newkey = kmemdup(cmd.tcpm_key, cmd.tcpm_keylen, GFP_KERNEL);
998 if (!newkey)
999 return -ENOMEM;
1000 return tcp_v4_md5_do_add(sk, sin->sin_addr.s_addr,
1001 newkey, cmd.tcpm_keylen);
1002 }
1003
1004 static int tcp_v4_do_calc_md5_hash(char *md5_hash, struct tcp_md5sig_key *key,
1005 __be32 saddr, __be32 daddr,
1006 struct tcphdr *th, int protocol,
1007 unsigned int tcplen)
1008 {
1009 struct scatterlist sg[4];
1010 __u16 data_len;
1011 int block = 0;
1012 __sum16 old_checksum;
1013 struct tcp_md5sig_pool *hp;
1014 struct tcp4_pseudohdr *bp;
1015 struct hash_desc *desc;
1016 int err;
1017 unsigned int nbytes = 0;
1018
1019 /*
1020 * Okay, so RFC2385 is turned on for this connection,
1021 * so we need to generate the MD5 hash for the packet now.
1022 */
1023
1024 hp = tcp_get_md5sig_pool();
1025 if (!hp)
1026 goto clear_hash_noput;
1027
1028 bp = &hp->md5_blk.ip4;
1029 desc = &hp->md5_desc;
1030
1031 /*
1032 * 1. the TCP pseudo-header (in the order: source IP address,
1033 * destination IP address, zero-padded protocol number, and
1034 * segment length)
1035 */
1036 bp->saddr = saddr;
1037 bp->daddr = daddr;
1038 bp->pad = 0;
1039 bp->protocol = protocol;
1040 bp->len = htons(tcplen);
1041
1042 sg_init_table(sg, 4);
1043
1044 sg_set_buf(&sg[block++], bp, sizeof(*bp));
1045 nbytes += sizeof(*bp);
1046
1047 /* 2. the TCP header, excluding options, and assuming a
1048 * checksum of zero/
1049 */
1050 old_checksum = th->check;
1051 th->check = 0;
1052 sg_set_buf(&sg[block++], th, sizeof(struct tcphdr));
1053 nbytes += sizeof(struct tcphdr);
1054
1055 /* 3. the TCP segment data (if any) */
1056 data_len = tcplen - (th->doff << 2);
1057 if (data_len > 0) {
1058 unsigned char *data = (unsigned char *)th + (th->doff << 2);
1059 sg_set_buf(&sg[block++], data, data_len);
1060 nbytes += data_len;
1061 }
1062
1063 /* 4. an independently-specified key or password, known to both
1064 * TCPs and presumably connection-specific
1065 */
1066 sg_set_buf(&sg[block++], key->key, key->keylen);
1067 nbytes += key->keylen;
1068
1069 sg_mark_end(&sg[block - 1]);
1070
1071 /* Now store the Hash into the packet */
1072 err = crypto_hash_init(desc);
1073 if (err)
1074 goto clear_hash;
1075 err = crypto_hash_update(desc, sg, nbytes);
1076 if (err)
1077 goto clear_hash;
1078 err = crypto_hash_final(desc, md5_hash);
1079 if (err)
1080 goto clear_hash;
1081
1082 /* Reset header, and free up the crypto */
1083 tcp_put_md5sig_pool();
1084 th->check = old_checksum;
1085
1086 out:
1087 return 0;
1088 clear_hash:
1089 tcp_put_md5sig_pool();
1090 clear_hash_noput:
1091 memset(md5_hash, 0, 16);
1092 goto out;
1093 }
1094
1095 int tcp_v4_calc_md5_hash(char *md5_hash, struct tcp_md5sig_key *key,
1096 struct sock *sk,
1097 struct dst_entry *dst,
1098 struct request_sock *req,
1099 struct tcphdr *th, int protocol,
1100 unsigned int tcplen)
1101 {
1102 __be32 saddr, daddr;
1103
1104 if (sk) {
1105 saddr = inet_sk(sk)->saddr;
1106 daddr = inet_sk(sk)->daddr;
1107 } else {
1108 struct rtable *rt = (struct rtable *)dst;
1109 BUG_ON(!rt);
1110 saddr = rt->rt_src;
1111 daddr = rt->rt_dst;
1112 }
1113 return tcp_v4_do_calc_md5_hash(md5_hash, key,
1114 saddr, daddr,
1115 th, protocol, tcplen);
1116 }
1117
1118 EXPORT_SYMBOL(tcp_v4_calc_md5_hash);
1119
1120 static int tcp_v4_inbound_md5_hash(struct sock *sk, struct sk_buff *skb)
1121 {
1122 /*
1123 * This gets called for each TCP segment that arrives
1124 * so we want to be efficient.
1125 * We have 3 drop cases:
1126 * o No MD5 hash and one expected.
1127 * o MD5 hash and we're not expecting one.
1128 * o MD5 hash and its wrong.
1129 */
1130 __u8 *hash_location = NULL;
1131 struct tcp_md5sig_key *hash_expected;
1132 const struct iphdr *iph = ip_hdr(skb);
1133 struct tcphdr *th = tcp_hdr(skb);
1134 int length = (th->doff << 2) - sizeof(struct tcphdr);
1135 int genhash;
1136 unsigned char *ptr;
1137 unsigned char newhash[16];
1138
1139 hash_expected = tcp_v4_md5_do_lookup(sk, iph->saddr);
1140
1141 /*
1142 * If the TCP option length is less than the TCP_MD5SIG
1143 * option length, then we can shortcut
1144 */
1145 if (length < TCPOLEN_MD5SIG) {
1146 if (hash_expected)
1147 return 1;
1148 else
1149 return 0;
1150 }
1151
1152 /* Okay, we can't shortcut - we have to grub through the options */
1153 ptr = (unsigned char *)(th + 1);
1154 while (length > 0) {
1155 int opcode = *ptr++;
1156 int opsize;
1157
1158 switch (opcode) {
1159 case TCPOPT_EOL:
1160 goto done_opts;
1161 case TCPOPT_NOP:
1162 length--;
1163 continue;
1164 default:
1165 opsize = *ptr++;
1166 if (opsize < 2)
1167 goto done_opts;
1168 if (opsize > length)
1169 goto done_opts;
1170
1171 if (opcode == TCPOPT_MD5SIG) {
1172 hash_location = ptr;
1173 goto done_opts;
1174 }
1175 }
1176 ptr += opsize-2;
1177 length -= opsize;
1178 }
1179 done_opts:
1180 /* We've parsed the options - do we have a hash? */
1181 if (!hash_expected && !hash_location)
1182 return 0;
1183
1184 if (hash_expected && !hash_location) {
1185 LIMIT_NETDEBUG(KERN_INFO "MD5 Hash expected but NOT found "
1186 "(" NIPQUAD_FMT ", %d)->(" NIPQUAD_FMT ", %d)\n",
1187 NIPQUAD(iph->saddr), ntohs(th->source),
1188 NIPQUAD(iph->daddr), ntohs(th->dest));
1189 return 1;
1190 }
1191
1192 if (!hash_expected && hash_location) {
1193 LIMIT_NETDEBUG(KERN_INFO "MD5 Hash NOT expected but found "
1194 "(" NIPQUAD_FMT ", %d)->(" NIPQUAD_FMT ", %d)\n",
1195 NIPQUAD(iph->saddr), ntohs(th->source),
1196 NIPQUAD(iph->daddr), ntohs(th->dest));
1197 return 1;
1198 }
1199
1200 /* Okay, so this is hash_expected and hash_location -
1201 * so we need to calculate the checksum.
1202 */
1203 genhash = tcp_v4_do_calc_md5_hash(newhash,
1204 hash_expected,
1205 iph->saddr, iph->daddr,
1206 th, sk->sk_protocol,
1207 skb->len);
1208
1209 if (genhash || memcmp(hash_location, newhash, 16) != 0) {
1210 if (net_ratelimit()) {
1211 printk(KERN_INFO "MD5 Hash failed for "
1212 "(" NIPQUAD_FMT ", %d)->(" NIPQUAD_FMT ", %d)%s\n",
1213 NIPQUAD(iph->saddr), ntohs(th->source),
1214 NIPQUAD(iph->daddr), ntohs(th->dest),
1215 genhash ? " tcp_v4_calc_md5_hash failed" : "");
1216 }
1217 return 1;
1218 }
1219 return 0;
1220 }
1221
1222 #endif
1223
1224 struct request_sock_ops tcp_request_sock_ops __read_mostly = {
1225 .family = PF_INET,
1226 .obj_size = sizeof(struct tcp_request_sock),
1227 .rtx_syn_ack = tcp_v4_send_synack,
1228 .send_ack = tcp_v4_reqsk_send_ack,
1229 .destructor = tcp_v4_reqsk_destructor,
1230 .send_reset = tcp_v4_send_reset,
1231 };
1232
1233 #ifdef CONFIG_TCP_MD5SIG
1234 static struct tcp_request_sock_ops tcp_request_sock_ipv4_ops = {
1235 .md5_lookup = tcp_v4_reqsk_md5_lookup,
1236 };
1237 #endif
1238
1239 static struct timewait_sock_ops tcp_timewait_sock_ops = {
1240 .twsk_obj_size = sizeof(struct tcp_timewait_sock),
1241 .twsk_unique = tcp_twsk_unique,
1242 .twsk_destructor= tcp_twsk_destructor,
1243 };
1244
1245 int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb)
1246 {
1247 struct inet_request_sock *ireq;
1248 struct tcp_options_received tmp_opt;
1249 struct request_sock *req;
1250 __be32 saddr = ip_hdr(skb)->saddr;
1251 __be32 daddr = ip_hdr(skb)->daddr;
1252 __u32 isn = TCP_SKB_CB(skb)->when;
1253 struct dst_entry *dst = NULL;
1254 #ifdef CONFIG_SYN_COOKIES
1255 int want_cookie = 0;
1256 #else
1257 #define want_cookie 0 /* Argh, why doesn't gcc optimize this :( */
1258 #endif
1259
1260 /* Never answer to SYNs send to broadcast or multicast */
1261 if (((struct rtable *)skb->dst)->rt_flags &
1262 (RTCF_BROADCAST | RTCF_MULTICAST))
1263 goto drop;
1264
1265 /* TW buckets are converted to open requests without
1266 * limitations, they conserve resources and peer is
1267 * evidently real one.
1268 */
1269 if (inet_csk_reqsk_queue_is_full(sk) && !isn) {
1270 #ifdef CONFIG_SYN_COOKIES
1271 if (sysctl_tcp_syncookies) {
1272 want_cookie = 1;
1273 } else
1274 #endif
1275 goto drop;
1276 }
1277
1278 /* Accept backlog is full. If we have already queued enough
1279 * of warm entries in syn queue, drop request. It is better than
1280 * clogging syn queue with openreqs with exponentially increasing
1281 * timeout.
1282 */
1283 if (sk_acceptq_is_full(sk) && inet_csk_reqsk_queue_young(sk) > 1)
1284 goto drop;
1285
1286 req = reqsk_alloc(&tcp_request_sock_ops);
1287 if (!req)
1288 goto drop;
1289
1290 #ifdef CONFIG_TCP_MD5SIG
1291 tcp_rsk(req)->af_specific = &tcp_request_sock_ipv4_ops;
1292 #endif
1293
1294 tcp_clear_options(&tmp_opt);
1295 tmp_opt.mss_clamp = 536;
1296 tmp_opt.user_mss = tcp_sk(sk)->rx_opt.user_mss;
1297
1298 tcp_parse_options(skb, &tmp_opt, 0);
1299
1300 if (want_cookie) {
1301 tcp_clear_options(&tmp_opt);
1302 tmp_opt.saw_tstamp = 0;
1303 }
1304
1305 if (tmp_opt.saw_tstamp && !tmp_opt.rcv_tsval) {
1306 /* Some OSes (unknown ones, but I see them on web server, which
1307 * contains information interesting only for windows'
1308 * users) do not send their stamp in SYN. It is easy case.
1309 * We simply do not advertise TS support.
1310 */
1311 tmp_opt.saw_tstamp = 0;
1312 tmp_opt.tstamp_ok = 0;
1313 }
1314 tmp_opt.tstamp_ok = tmp_opt.saw_tstamp;
1315
1316 tcp_openreq_init(req, &tmp_opt, skb);
1317
1318 if (security_inet_conn_request(sk, skb, req))
1319 goto drop_and_free;
1320
1321 ireq = inet_rsk(req);
1322 ireq->loc_addr = daddr;
1323 ireq->rmt_addr = saddr;
1324 ireq->opt = tcp_v4_save_options(sk, skb);
1325 if (!want_cookie)
1326 TCP_ECN_create_request(req, tcp_hdr(skb));
1327
1328 if (want_cookie) {
1329 #ifdef CONFIG_SYN_COOKIES
1330 syn_flood_warning(skb);
1331 #endif
1332 isn = cookie_v4_init_sequence(sk, skb, &req->mss);
1333 } else if (!isn) {
1334 struct inet_peer *peer = NULL;
1335
1336 /* VJ's idea. We save last timestamp seen
1337 * from the destination in peer table, when entering
1338 * state TIME-WAIT, and check against it before
1339 * accepting new connection request.
1340 *
1341 * If "isn" is not zero, this request hit alive
1342 * timewait bucket, so that all the necessary checks
1343 * are made in the function processing timewait state.
1344 */
1345 if (tmp_opt.saw_tstamp &&
1346 tcp_death_row.sysctl_tw_recycle &&
1347 (dst = inet_csk_route_req(sk, req)) != NULL &&
1348 (peer = rt_get_peer((struct rtable *)dst)) != NULL &&
1349 peer->v4daddr == saddr) {
1350 if (get_seconds() < peer->tcp_ts_stamp + TCP_PAWS_MSL &&
1351 (s32)(peer->tcp_ts - req->ts_recent) >
1352 TCP_PAWS_WINDOW) {
1353 NET_INC_STATS_BH(LINUX_MIB_PAWSPASSIVEREJECTED);
1354 dst_release(dst);
1355 goto drop_and_free;
1356 }
1357 }
1358 /* Kill the following clause, if you dislike this way. */
1359 else if (!sysctl_tcp_syncookies &&
1360 (sysctl_max_syn_backlog - inet_csk_reqsk_queue_len(sk) <
1361 (sysctl_max_syn_backlog >> 2)) &&
1362 (!peer || !peer->tcp_ts_stamp) &&
1363 (!dst || !dst_metric(dst, RTAX_RTT))) {
1364 /* Without syncookies last quarter of
1365 * backlog is filled with destinations,
1366 * proven to be alive.
1367 * It means that we continue to communicate
1368 * to destinations, already remembered
1369 * to the moment of synflood.
1370 */
1371 LIMIT_NETDEBUG(KERN_DEBUG "TCP: drop open "
1372 "request from %u.%u.%u.%u/%u\n",
1373 NIPQUAD(saddr),
1374 ntohs(tcp_hdr(skb)->source));
1375 dst_release(dst);
1376 goto drop_and_free;
1377 }
1378
1379 isn = tcp_v4_init_sequence(skb);
1380 }
1381 tcp_rsk(req)->snt_isn = isn;
1382
1383 if (tcp_v4_send_synack(sk, req, dst))
1384 goto drop_and_free;
1385
1386 if (want_cookie) {
1387 reqsk_free(req);
1388 } else {
1389 inet_csk_reqsk_queue_hash_add(sk, req, TCP_TIMEOUT_INIT);
1390 }
1391 return 0;
1392
1393 drop_and_free:
1394 reqsk_free(req);
1395 drop:
1396 return 0;
1397 }
1398
1399
1400 /*
1401 * The three way handshake has completed - we got a valid synack -
1402 * now create the new socket.
1403 */
1404 struct sock *tcp_v4_syn_recv_sock(struct sock *sk, struct sk_buff *skb,
1405 struct request_sock *req,
1406 struct dst_entry *dst)
1407 {
1408 struct inet_request_sock *ireq;
1409 struct inet_sock *newinet;
1410 struct tcp_sock *newtp;
1411 struct sock *newsk;
1412 #ifdef CONFIG_TCP_MD5SIG
1413 struct tcp_md5sig_key *key;
1414 #endif
1415
1416 if (sk_acceptq_is_full(sk))
1417 goto exit_overflow;
1418
1419 if (!dst && (dst = inet_csk_route_req(sk, req)) == NULL)
1420 goto exit;
1421
1422 newsk = tcp_create_openreq_child(sk, req, skb);
1423 if (!newsk)
1424 goto exit;
1425
1426 newsk->sk_gso_type = SKB_GSO_TCPV4;
1427 sk_setup_caps(newsk, dst);
1428
1429 newtp = tcp_sk(newsk);
1430 newinet = inet_sk(newsk);
1431 ireq = inet_rsk(req);
1432 newinet->daddr = ireq->rmt_addr;
1433 newinet->rcv_saddr = ireq->loc_addr;
1434 newinet->saddr = ireq->loc_addr;
1435 newinet->opt = ireq->opt;
1436 ireq->opt = NULL;
1437 newinet->mc_index = inet_iif(skb);
1438 newinet->mc_ttl = ip_hdr(skb)->ttl;
1439 inet_csk(newsk)->icsk_ext_hdr_len = 0;
1440 if (newinet->opt)
1441 inet_csk(newsk)->icsk_ext_hdr_len = newinet->opt->optlen;
1442 newinet->id = newtp->write_seq ^ jiffies;
1443
1444 tcp_mtup_init(newsk);
1445 tcp_sync_mss(newsk, dst_mtu(dst));
1446 newtp->advmss = dst_metric(dst, RTAX_ADVMSS);
1447 tcp_initialize_rcv_mss(newsk);
1448
1449 #ifdef CONFIG_TCP_MD5SIG
1450 /* Copy over the MD5 key from the original socket */
1451 if ((key = tcp_v4_md5_do_lookup(sk, newinet->daddr)) != NULL) {
1452 /*
1453 * We're using one, so create a matching key
1454 * on the newsk structure. If we fail to get
1455 * memory, then we end up not copying the key
1456 * across. Shucks.
1457 */
1458 char *newkey = kmemdup(key->key, key->keylen, GFP_ATOMIC);
1459 if (newkey != NULL)
1460 tcp_v4_md5_do_add(newsk, inet_sk(sk)->daddr,
1461 newkey, key->keylen);
1462 }
1463 #endif
1464
1465 __inet_hash_nolisten(newsk);
1466 __inet_inherit_port(sk, newsk);
1467
1468 return newsk;
1469
1470 exit_overflow:
1471 NET_INC_STATS_BH(LINUX_MIB_LISTENOVERFLOWS);
1472 exit:
1473 NET_INC_STATS_BH(LINUX_MIB_LISTENDROPS);
1474 dst_release(dst);
1475 return NULL;
1476 }
1477
1478 static struct sock *tcp_v4_hnd_req(struct sock *sk, struct sk_buff *skb)
1479 {
1480 struct tcphdr *th = tcp_hdr(skb);
1481 const struct iphdr *iph = ip_hdr(skb);
1482 struct sock *nsk;
1483 struct request_sock **prev;
1484 /* Find possible connection requests. */
1485 struct request_sock *req = inet_csk_search_req(sk, &prev, th->source,
1486 iph->saddr, iph->daddr);
1487 if (req)
1488 return tcp_check_req(sk, skb, req, prev);
1489
1490 nsk = inet_lookup_established(sk->sk_net, &tcp_hashinfo, iph->saddr,
1491 th->source, iph->daddr, th->dest, inet_iif(skb));
1492
1493 if (nsk) {
1494 if (nsk->sk_state != TCP_TIME_WAIT) {
1495 bh_lock_sock(nsk);
1496 return nsk;
1497 }
1498 inet_twsk_put(inet_twsk(nsk));
1499 return NULL;
1500 }
1501
1502 #ifdef CONFIG_SYN_COOKIES
1503 if (!th->rst && !th->syn && th->ack)
1504 sk = cookie_v4_check(sk, skb, &(IPCB(skb)->opt));
1505 #endif
1506 return sk;
1507 }
1508
1509 static __sum16 tcp_v4_checksum_init(struct sk_buff *skb)
1510 {
1511 const struct iphdr *iph = ip_hdr(skb);
1512
1513 if (skb->ip_summed == CHECKSUM_COMPLETE) {
1514 if (!tcp_v4_check(skb->len, iph->saddr,
1515 iph->daddr, skb->csum)) {
1516 skb->ip_summed = CHECKSUM_UNNECESSARY;
1517 return 0;
1518 }
1519 }
1520
1521 skb->csum = csum_tcpudp_nofold(iph->saddr, iph->daddr,
1522 skb->len, IPPROTO_TCP, 0);
1523
1524 if (skb->len <= 76) {
1525 return __skb_checksum_complete(skb);
1526 }
1527 return 0;
1528 }
1529
1530
1531 /* The socket must have it's spinlock held when we get
1532 * here.
1533 *
1534 * We have a potential double-lock case here, so even when
1535 * doing backlog processing we use the BH locking scheme.
1536 * This is because we cannot sleep with the original spinlock
1537 * held.
1538 */
1539 int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb)
1540 {
1541 struct sock *rsk;
1542 #ifdef CONFIG_TCP_MD5SIG
1543 /*
1544 * We really want to reject the packet as early as possible
1545 * if:
1546 * o We're expecting an MD5'd packet and this is no MD5 tcp option
1547 * o There is an MD5 option and we're not expecting one
1548 */
1549 if (tcp_v4_inbound_md5_hash(sk, skb))
1550 goto discard;
1551 #endif
1552
1553 if (sk->sk_state == TCP_ESTABLISHED) { /* Fast path */
1554 TCP_CHECK_TIMER(sk);
1555 if (tcp_rcv_established(sk, skb, tcp_hdr(skb), skb->len)) {
1556 rsk = sk;
1557 goto reset;
1558 }
1559 TCP_CHECK_TIMER(sk);
1560 return 0;
1561 }
1562
1563 if (skb->len < tcp_hdrlen(skb) || tcp_checksum_complete(skb))
1564 goto csum_err;
1565
1566 if (sk->sk_state == TCP_LISTEN) {
1567 struct sock *nsk = tcp_v4_hnd_req(sk, skb);
1568 if (!nsk)
1569 goto discard;
1570
1571 if (nsk != sk) {
1572 if (tcp_child_process(sk, nsk, skb)) {
1573 rsk = nsk;
1574 goto reset;
1575 }
1576 return 0;
1577 }
1578 }
1579
1580 TCP_CHECK_TIMER(sk);
1581 if (tcp_rcv_state_process(sk, skb, tcp_hdr(skb), skb->len)) {
1582 rsk = sk;
1583 goto reset;
1584 }
1585 TCP_CHECK_TIMER(sk);
1586 return 0;
1587
1588 reset:
1589 tcp_v4_send_reset(rsk, skb);
1590 discard:
1591 kfree_skb(skb);
1592 /* Be careful here. If this function gets more complicated and
1593 * gcc suffers from register pressure on the x86, sk (in %ebx)
1594 * might be destroyed here. This current version compiles correctly,
1595 * but you have been warned.
1596 */
1597 return 0;
1598
1599 csum_err:
1600 TCP_INC_STATS_BH(TCP_MIB_INERRS);
1601 goto discard;
1602 }
1603
1604 /*
1605 * From tcp_input.c
1606 */
1607
1608 int tcp_v4_rcv(struct sk_buff *skb)
1609 {
1610 const struct iphdr *iph;
1611 struct tcphdr *th;
1612 struct sock *sk;
1613 int ret;
1614
1615 if (skb->pkt_type != PACKET_HOST)
1616 goto discard_it;
1617
1618 /* Count it even if it's bad */
1619 TCP_INC_STATS_BH(TCP_MIB_INSEGS);
1620
1621 if (!pskb_may_pull(skb, sizeof(struct tcphdr)))
1622 goto discard_it;
1623
1624 th = tcp_hdr(skb);
1625
1626 if (th->doff < sizeof(struct tcphdr) / 4)
1627 goto bad_packet;
1628 if (!pskb_may_pull(skb, th->doff * 4))
1629 goto discard_it;
1630
1631 /* An explanation is required here, I think.
1632 * Packet length and doff are validated by header prediction,
1633 * provided case of th->doff==0 is eliminated.
1634 * So, we defer the checks. */
1635 if (!skb_csum_unnecessary(skb) && tcp_v4_checksum_init(skb))
1636 goto bad_packet;
1637
1638 th = tcp_hdr(skb);
1639 iph = ip_hdr(skb);
1640 TCP_SKB_CB(skb)->seq = ntohl(th->seq);
1641 TCP_SKB_CB(skb)->end_seq = (TCP_SKB_CB(skb)->seq + th->syn + th->fin +
1642 skb->len - th->doff * 4);
1643 TCP_SKB_CB(skb)->ack_seq = ntohl(th->ack_seq);
1644 TCP_SKB_CB(skb)->when = 0;
1645 TCP_SKB_CB(skb)->flags = iph->tos;
1646 TCP_SKB_CB(skb)->sacked = 0;
1647
1648 sk = __inet_lookup(skb->dev->nd_net, &tcp_hashinfo, iph->saddr,
1649 th->source, iph->daddr, th->dest, inet_iif(skb));
1650 if (!sk)
1651 goto no_tcp_socket;
1652
1653 process:
1654 if (sk->sk_state == TCP_TIME_WAIT)
1655 goto do_time_wait;
1656
1657 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
1658 goto discard_and_relse;
1659 nf_reset(skb);
1660
1661 if (sk_filter(sk, skb))
1662 goto discard_and_relse;
1663
1664 skb->dev = NULL;
1665
1666 bh_lock_sock_nested(sk);
1667 ret = 0;
1668 if (!sock_owned_by_user(sk)) {
1669 #ifdef CONFIG_NET_DMA
1670 struct tcp_sock *tp = tcp_sk(sk);
1671 if (!tp->ucopy.dma_chan && tp->ucopy.pinned_list)
1672 tp->ucopy.dma_chan = get_softnet_dma();
1673 if (tp->ucopy.dma_chan)
1674 ret = tcp_v4_do_rcv(sk, skb);
1675 else
1676 #endif
1677 {
1678 if (!tcp_prequeue(sk, skb))
1679 ret = tcp_v4_do_rcv(sk, skb);
1680 }
1681 } else
1682 sk_add_backlog(sk, skb);
1683 bh_unlock_sock(sk);
1684
1685 sock_put(sk);
1686
1687 return ret;
1688
1689 no_tcp_socket:
1690 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
1691 goto discard_it;
1692
1693 if (skb->len < (th->doff << 2) || tcp_checksum_complete(skb)) {
1694 bad_packet:
1695 TCP_INC_STATS_BH(TCP_MIB_INERRS);
1696 } else {
1697 tcp_v4_send_reset(NULL, skb);
1698 }
1699
1700 discard_it:
1701 /* Discard frame. */
1702 kfree_skb(skb);
1703 return 0;
1704
1705 discard_and_relse:
1706 sock_put(sk);
1707 goto discard_it;
1708
1709 do_time_wait:
1710 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) {
1711 inet_twsk_put(inet_twsk(sk));
1712 goto discard_it;
1713 }
1714
1715 if (skb->len < (th->doff << 2) || tcp_checksum_complete(skb)) {
1716 TCP_INC_STATS_BH(TCP_MIB_INERRS);
1717 inet_twsk_put(inet_twsk(sk));
1718 goto discard_it;
1719 }
1720 switch (tcp_timewait_state_process(inet_twsk(sk), skb, th)) {
1721 case TCP_TW_SYN: {
1722 struct sock *sk2 = inet_lookup_listener(skb->dev->nd_net,
1723 &tcp_hashinfo,
1724 iph->daddr, th->dest,
1725 inet_iif(skb));
1726 if (sk2) {
1727 inet_twsk_deschedule(inet_twsk(sk), &tcp_death_row);
1728 inet_twsk_put(inet_twsk(sk));
1729 sk = sk2;
1730 goto process;
1731 }
1732 /* Fall through to ACK */
1733 }
1734 case TCP_TW_ACK:
1735 tcp_v4_timewait_ack(sk, skb);
1736 break;
1737 case TCP_TW_RST:
1738 goto no_tcp_socket;
1739 case TCP_TW_SUCCESS:;
1740 }
1741 goto discard_it;
1742 }
1743
1744 /* VJ's idea. Save last timestamp seen from this destination
1745 * and hold it at least for normal timewait interval to use for duplicate
1746 * segment detection in subsequent connections, before they enter synchronized
1747 * state.
1748 */
1749
1750 int tcp_v4_remember_stamp(struct sock *sk)
1751 {
1752 struct inet_sock *inet = inet_sk(sk);
1753 struct tcp_sock *tp = tcp_sk(sk);
1754 struct rtable *rt = (struct rtable *)__sk_dst_get(sk);
1755 struct inet_peer *peer = NULL;
1756 int release_it = 0;
1757
1758 if (!rt || rt->rt_dst != inet->daddr) {
1759 peer = inet_getpeer(inet->daddr, 1);
1760 release_it = 1;
1761 } else {
1762 if (!rt->peer)
1763 rt_bind_peer(rt, 1);
1764 peer = rt->peer;
1765 }
1766
1767 if (peer) {
1768 if ((s32)(peer->tcp_ts - tp->rx_opt.ts_recent) <= 0 ||
1769 (peer->tcp_ts_stamp + TCP_PAWS_MSL < get_seconds() &&
1770 peer->tcp_ts_stamp <= tp->rx_opt.ts_recent_stamp)) {
1771 peer->tcp_ts_stamp = tp->rx_opt.ts_recent_stamp;
1772 peer->tcp_ts = tp->rx_opt.ts_recent;
1773 }
1774 if (release_it)
1775 inet_putpeer(peer);
1776 return 1;
1777 }
1778
1779 return 0;
1780 }
1781
1782 int tcp_v4_tw_remember_stamp(struct inet_timewait_sock *tw)
1783 {
1784 struct inet_peer *peer = inet_getpeer(tw->tw_daddr, 1);
1785
1786 if (peer) {
1787 const struct tcp_timewait_sock *tcptw = tcp_twsk((struct sock *)tw);
1788
1789 if ((s32)(peer->tcp_ts - tcptw->tw_ts_recent) <= 0 ||
1790 (peer->tcp_ts_stamp + TCP_PAWS_MSL < get_seconds() &&
1791 peer->tcp_ts_stamp <= tcptw->tw_ts_recent_stamp)) {
1792 peer->tcp_ts_stamp = tcptw->tw_ts_recent_stamp;
1793 peer->tcp_ts = tcptw->tw_ts_recent;
1794 }
1795 inet_putpeer(peer);
1796 return 1;
1797 }
1798
1799 return 0;
1800 }
1801
1802 struct inet_connection_sock_af_ops ipv4_specific = {
1803 .queue_xmit = ip_queue_xmit,
1804 .send_check = tcp_v4_send_check,
1805 .rebuild_header = inet_sk_rebuild_header,
1806 .conn_request = tcp_v4_conn_request,
1807 .syn_recv_sock = tcp_v4_syn_recv_sock,
1808 .remember_stamp = tcp_v4_remember_stamp,
1809 .net_header_len = sizeof(struct iphdr),
1810 .setsockopt = ip_setsockopt,
1811 .getsockopt = ip_getsockopt,
1812 .addr2sockaddr = inet_csk_addr2sockaddr,
1813 .sockaddr_len = sizeof(struct sockaddr_in),
1814 .bind_conflict = inet_csk_bind_conflict,
1815 #ifdef CONFIG_COMPAT
1816 .compat_setsockopt = compat_ip_setsockopt,
1817 .compat_getsockopt = compat_ip_getsockopt,
1818 #endif
1819 };
1820
1821 #ifdef CONFIG_TCP_MD5SIG
1822 static struct tcp_sock_af_ops tcp_sock_ipv4_specific = {
1823 .md5_lookup = tcp_v4_md5_lookup,
1824 .calc_md5_hash = tcp_v4_calc_md5_hash,
1825 .md5_add = tcp_v4_md5_add_func,
1826 .md5_parse = tcp_v4_parse_md5_keys,
1827 };
1828 #endif
1829
1830 /* NOTE: A lot of things set to zero explicitly by call to
1831 * sk_alloc() so need not be done here.
1832 */
1833 static int tcp_v4_init_sock(struct sock *sk)
1834 {
1835 struct inet_connection_sock *icsk = inet_csk(sk);
1836 struct tcp_sock *tp = tcp_sk(sk);
1837
1838 skb_queue_head_init(&tp->out_of_order_queue);
1839 tcp_init_xmit_timers(sk);
1840 tcp_prequeue_init(tp);
1841
1842 icsk->icsk_rto = TCP_TIMEOUT_INIT;
1843 tp->mdev = TCP_TIMEOUT_INIT;
1844
1845 /* So many TCP implementations out there (incorrectly) count the
1846 * initial SYN frame in their delayed-ACK and congestion control
1847 * algorithms that we must have the following bandaid to talk
1848 * efficiently to them. -DaveM
1849 */
1850 tp->snd_cwnd = 2;
1851
1852 /* See draft-stevens-tcpca-spec-01 for discussion of the
1853 * initialization of these values.
1854 */
1855 tp->snd_ssthresh = 0x7fffffff; /* Infinity */
1856 tp->snd_cwnd_clamp = ~0;
1857 tp->mss_cache = 536;
1858
1859 tp->reordering = sysctl_tcp_reordering;
1860 icsk->icsk_ca_ops = &tcp_init_congestion_ops;
1861
1862 sk->sk_state = TCP_CLOSE;
1863
1864 sk->sk_write_space = sk_stream_write_space;
1865 sock_set_flag(sk, SOCK_USE_WRITE_QUEUE);
1866
1867 icsk->icsk_af_ops = &ipv4_specific;
1868 icsk->icsk_sync_mss = tcp_sync_mss;
1869 #ifdef CONFIG_TCP_MD5SIG
1870 tp->af_specific = &tcp_sock_ipv4_specific;
1871 #endif
1872
1873 sk->sk_sndbuf = sysctl_tcp_wmem[1];
1874 sk->sk_rcvbuf = sysctl_tcp_rmem[1];
1875
1876 atomic_inc(&tcp_sockets_allocated);
1877
1878 return 0;
1879 }
1880
1881 int tcp_v4_destroy_sock(struct sock *sk)
1882 {
1883 struct tcp_sock *tp = tcp_sk(sk);
1884
1885 tcp_clear_xmit_timers(sk);
1886
1887 tcp_cleanup_congestion_control(sk);
1888
1889 /* Cleanup up the write buffer. */
1890 tcp_write_queue_purge(sk);
1891
1892 /* Cleans up our, hopefully empty, out_of_order_queue. */
1893 __skb_queue_purge(&tp->out_of_order_queue);
1894
1895 #ifdef CONFIG_TCP_MD5SIG
1896 /* Clean up the MD5 key list, if any */
1897 if (tp->md5sig_info) {
1898 tcp_v4_clear_md5_list(sk);
1899 kfree(tp->md5sig_info);
1900 tp->md5sig_info = NULL;
1901 }
1902 #endif
1903
1904 #ifdef CONFIG_NET_DMA
1905 /* Cleans up our sk_async_wait_queue */
1906 __skb_queue_purge(&sk->sk_async_wait_queue);
1907 #endif
1908
1909 /* Clean prequeue, it must be empty really */
1910 __skb_queue_purge(&tp->ucopy.prequeue);
1911
1912 /* Clean up a referenced TCP bind bucket. */
1913 if (inet_csk(sk)->icsk_bind_hash)
1914 inet_put_port(sk);
1915
1916 /*
1917 * If sendmsg cached page exists, toss it.
1918 */
1919 if (sk->sk_sndmsg_page) {
1920 __free_page(sk->sk_sndmsg_page);
1921 sk->sk_sndmsg_page = NULL;
1922 }
1923
1924 atomic_dec(&tcp_sockets_allocated);
1925
1926 return 0;
1927 }
1928
1929 EXPORT_SYMBOL(tcp_v4_destroy_sock);
1930
1931 #ifdef CONFIG_PROC_FS
1932 /* Proc filesystem TCP sock list dumping. */
1933
1934 static inline struct inet_timewait_sock *tw_head(struct hlist_head *head)
1935 {
1936 return hlist_empty(head) ? NULL :
1937 list_entry(head->first, struct inet_timewait_sock, tw_node);
1938 }
1939
1940 static inline struct inet_timewait_sock *tw_next(struct inet_timewait_sock *tw)
1941 {
1942 return tw->tw_node.next ?
1943 hlist_entry(tw->tw_node.next, typeof(*tw), tw_node) : NULL;
1944 }
1945
1946 static void *listening_get_next(struct seq_file *seq, void *cur)
1947 {
1948 struct inet_connection_sock *icsk;
1949 struct hlist_node *node;
1950 struct sock *sk = cur;
1951 struct tcp_iter_state* st = seq->private;
1952
1953 if (!sk) {
1954 st->bucket = 0;
1955 sk = sk_head(&tcp_hashinfo.listening_hash[0]);
1956 goto get_sk;
1957 }
1958
1959 ++st->num;
1960
1961 if (st->state == TCP_SEQ_STATE_OPENREQ) {
1962 struct request_sock *req = cur;
1963
1964 icsk = inet_csk(st->syn_wait_sk);
1965 req = req->dl_next;
1966 while (1) {
1967 while (req) {
1968 if (req->rsk_ops->family == st->family) {
1969 cur = req;
1970 goto out;
1971 }
1972 req = req->dl_next;
1973 }
1974 if (++st->sbucket >= icsk->icsk_accept_queue.listen_opt->nr_table_entries)
1975 break;
1976 get_req:
1977 req = icsk->icsk_accept_queue.listen_opt->syn_table[st->sbucket];
1978 }
1979 sk = sk_next(st->syn_wait_sk);
1980 st->state = TCP_SEQ_STATE_LISTENING;
1981 read_unlock_bh(&icsk->icsk_accept_queue.syn_wait_lock);
1982 } else {
1983 icsk = inet_csk(sk);
1984 read_lock_bh(&icsk->icsk_accept_queue.syn_wait_lock);
1985 if (reqsk_queue_len(&icsk->icsk_accept_queue))
1986 goto start_req;
1987 read_unlock_bh(&icsk->icsk_accept_queue.syn_wait_lock);
1988 sk = sk_next(sk);
1989 }
1990 get_sk:
1991 sk_for_each_from(sk, node) {
1992 if (sk->sk_family == st->family) {
1993 cur = sk;
1994 goto out;
1995 }
1996 icsk = inet_csk(sk);
1997 read_lock_bh(&icsk->icsk_accept_queue.syn_wait_lock);
1998 if (reqsk_queue_len(&icsk->icsk_accept_queue)) {
1999 start_req:
2000 st->uid = sock_i_uid(sk);
2001 st->syn_wait_sk = sk;
2002 st->state = TCP_SEQ_STATE_OPENREQ;
2003 st->sbucket = 0;
2004 goto get_req;
2005 }
2006 read_unlock_bh(&icsk->icsk_accept_queue.syn_wait_lock);
2007 }
2008 if (++st->bucket < INET_LHTABLE_SIZE) {
2009 sk = sk_head(&tcp_hashinfo.listening_hash[st->bucket]);
2010 goto get_sk;
2011 }
2012 cur = NULL;
2013 out:
2014 return cur;
2015 }
2016
2017 static void *listening_get_idx(struct seq_file *seq, loff_t *pos)
2018 {
2019 void *rc = listening_get_next(seq, NULL);
2020
2021 while (rc && *pos) {
2022 rc = listening_get_next(seq, rc);
2023 --*pos;
2024 }
2025 return rc;
2026 }
2027
2028 static void *established_get_first(struct seq_file *seq)
2029 {
2030 struct tcp_iter_state* st = seq->private;
2031 void *rc = NULL;
2032
2033 for (st->bucket = 0; st->bucket < tcp_hashinfo.ehash_size; ++st->bucket) {
2034 struct sock *sk;
2035 struct hlist_node *node;
2036 struct inet_timewait_sock *tw;
2037 rwlock_t *lock = inet_ehash_lockp(&tcp_hashinfo, st->bucket);
2038
2039 read_lock_bh(lock);
2040 sk_for_each(sk, node, &tcp_hashinfo.ehash[st->bucket].chain) {
2041 if (sk->sk_family != st->family) {
2042 continue;
2043 }
2044 rc = sk;
2045 goto out;
2046 }
2047 st->state = TCP_SEQ_STATE_TIME_WAIT;
2048 inet_twsk_for_each(tw, node,
2049 &tcp_hashinfo.ehash[st->bucket].twchain) {
2050 if (tw->tw_family != st->family) {
2051 continue;
2052 }
2053 rc = tw;
2054 goto out;
2055 }
2056 read_unlock_bh(lock);
2057 st->state = TCP_SEQ_STATE_ESTABLISHED;
2058 }
2059 out:
2060 return rc;
2061 }
2062
2063 static void *established_get_next(struct seq_file *seq, void *cur)
2064 {
2065 struct sock *sk = cur;
2066 struct inet_timewait_sock *tw;
2067 struct hlist_node *node;
2068 struct tcp_iter_state* st = seq->private;
2069
2070 ++st->num;
2071
2072 if (st->state == TCP_SEQ_STATE_TIME_WAIT) {
2073 tw = cur;
2074 tw = tw_next(tw);
2075 get_tw:
2076 while (tw && tw->tw_family != st->family) {
2077 tw = tw_next(tw);
2078 }
2079 if (tw) {
2080 cur = tw;
2081 goto out;
2082 }
2083 read_unlock_bh(inet_ehash_lockp(&tcp_hashinfo, st->bucket));
2084 st->state = TCP_SEQ_STATE_ESTABLISHED;
2085
2086 if (++st->bucket < tcp_hashinfo.ehash_size) {
2087 read_lock_bh(inet_ehash_lockp(&tcp_hashinfo, st->bucket));
2088 sk = sk_head(&tcp_hashinfo.ehash[st->bucket].chain);
2089 } else {
2090 cur = NULL;
2091 goto out;
2092 }
2093 } else
2094 sk = sk_next(sk);
2095
2096 sk_for_each_from(sk, node) {
2097 if (sk->sk_family == st->family)
2098 goto found;
2099 }
2100
2101 st->state = TCP_SEQ_STATE_TIME_WAIT;
2102 tw = tw_head(&tcp_hashinfo.ehash[st->bucket].twchain);
2103 goto get_tw;
2104 found:
2105 cur = sk;
2106 out:
2107 return cur;
2108 }
2109
2110 static void *established_get_idx(struct seq_file *seq, loff_t pos)
2111 {
2112 void *rc = established_get_first(seq);
2113
2114 while (rc && pos) {
2115 rc = established_get_next(seq, rc);
2116 --pos;
2117 }
2118 return rc;
2119 }
2120
2121 static void *tcp_get_idx(struct seq_file *seq, loff_t pos)
2122 {
2123 void *rc;
2124 struct tcp_iter_state* st = seq->private;
2125
2126 inet_listen_lock(&tcp_hashinfo);
2127 st->state = TCP_SEQ_STATE_LISTENING;
2128 rc = listening_get_idx(seq, &pos);
2129
2130 if (!rc) {
2131 inet_listen_unlock(&tcp_hashinfo);
2132 st->state = TCP_SEQ_STATE_ESTABLISHED;
2133 rc = established_get_idx(seq, pos);
2134 }
2135
2136 return rc;
2137 }
2138
2139 static void *tcp_seq_start(struct seq_file *seq, loff_t *pos)
2140 {
2141 struct tcp_iter_state* st = seq->private;
2142 st->state = TCP_SEQ_STATE_LISTENING;
2143 st->num = 0;
2144 return *pos ? tcp_get_idx(seq, *pos - 1) : SEQ_START_TOKEN;
2145 }
2146
2147 static void *tcp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2148 {
2149 void *rc = NULL;
2150 struct tcp_iter_state* st;
2151
2152 if (v == SEQ_START_TOKEN) {
2153 rc = tcp_get_idx(seq, 0);
2154 goto out;
2155 }
2156 st = seq->private;
2157
2158 switch (st->state) {
2159 case TCP_SEQ_STATE_OPENREQ:
2160 case TCP_SEQ_STATE_LISTENING:
2161 rc = listening_get_next(seq, v);
2162 if (!rc) {
2163 inet_listen_unlock(&tcp_hashinfo);
2164 st->state = TCP_SEQ_STATE_ESTABLISHED;
2165 rc = established_get_first(seq);
2166 }
2167 break;
2168 case TCP_SEQ_STATE_ESTABLISHED:
2169 case TCP_SEQ_STATE_TIME_WAIT:
2170 rc = established_get_next(seq, v);
2171 break;
2172 }
2173 out:
2174 ++*pos;
2175 return rc;
2176 }
2177
2178 static void tcp_seq_stop(struct seq_file *seq, void *v)
2179 {
2180 struct tcp_iter_state* st = seq->private;
2181
2182 switch (st->state) {
2183 case TCP_SEQ_STATE_OPENREQ:
2184 if (v) {
2185 struct inet_connection_sock *icsk = inet_csk(st->syn_wait_sk);
2186 read_unlock_bh(&icsk->icsk_accept_queue.syn_wait_lock);
2187 }
2188 case TCP_SEQ_STATE_LISTENING:
2189 if (v != SEQ_START_TOKEN)
2190 inet_listen_unlock(&tcp_hashinfo);
2191 break;
2192 case TCP_SEQ_STATE_TIME_WAIT:
2193 case TCP_SEQ_STATE_ESTABLISHED:
2194 if (v)
2195 read_unlock_bh(inet_ehash_lockp(&tcp_hashinfo, st->bucket));
2196 break;
2197 }
2198 }
2199
2200 static int tcp_seq_open(struct inode *inode, struct file *file)
2201 {
2202 struct tcp_seq_afinfo *afinfo = PDE(inode)->data;
2203 struct seq_file *seq;
2204 struct tcp_iter_state *s;
2205 int rc;
2206
2207 if (unlikely(afinfo == NULL))
2208 return -EINVAL;
2209
2210 s = kzalloc(sizeof(*s), GFP_KERNEL);
2211 if (!s)
2212 return -ENOMEM;
2213 s->family = afinfo->family;
2214 s->seq_ops.start = tcp_seq_start;
2215 s->seq_ops.next = tcp_seq_next;
2216 s->seq_ops.show = afinfo->seq_show;
2217 s->seq_ops.stop = tcp_seq_stop;
2218
2219 rc = seq_open(file, &s->seq_ops);
2220 if (rc)
2221 goto out_kfree;
2222 seq = file->private_data;
2223 seq->private = s;
2224 out:
2225 return rc;
2226 out_kfree:
2227 kfree(s);
2228 goto out;
2229 }
2230
2231 int tcp_proc_register(struct tcp_seq_afinfo *afinfo)
2232 {
2233 int rc = 0;
2234 struct proc_dir_entry *p;
2235
2236 if (!afinfo)
2237 return -EINVAL;
2238 afinfo->seq_fops->owner = afinfo->owner;
2239 afinfo->seq_fops->open = tcp_seq_open;
2240 afinfo->seq_fops->read = seq_read;
2241 afinfo->seq_fops->llseek = seq_lseek;
2242 afinfo->seq_fops->release = seq_release_private;
2243
2244 p = proc_net_fops_create(&init_net, afinfo->name, S_IRUGO, afinfo->seq_fops);
2245 if (p)
2246 p->data = afinfo;
2247 else
2248 rc = -ENOMEM;
2249 return rc;
2250 }
2251
2252 void tcp_proc_unregister(struct tcp_seq_afinfo *afinfo)
2253 {
2254 if (!afinfo)
2255 return;
2256 proc_net_remove(&init_net, afinfo->name);
2257 memset(afinfo->seq_fops, 0, sizeof(*afinfo->seq_fops));
2258 }
2259
2260 static void get_openreq4(struct sock *sk, struct request_sock *req,
2261 char *tmpbuf, int i, int uid)
2262 {
2263 const struct inet_request_sock *ireq = inet_rsk(req);
2264 int ttd = req->expires - jiffies;
2265
2266 sprintf(tmpbuf, "%4d: %08X:%04X %08X:%04X"
2267 " %02X %08X:%08X %02X:%08lX %08X %5d %8d %u %d %p",
2268 i,
2269 ireq->loc_addr,
2270 ntohs(inet_sk(sk)->sport),
2271 ireq->rmt_addr,
2272 ntohs(ireq->rmt_port),
2273 TCP_SYN_RECV,
2274 0, 0, /* could print option size, but that is af dependent. */
2275 1, /* timers active (only the expire timer) */
2276 jiffies_to_clock_t(ttd),
2277 req->retrans,
2278 uid,
2279 0, /* non standard timer */
2280 0, /* open_requests have no inode */
2281 atomic_read(&sk->sk_refcnt),
2282 req);
2283 }
2284
2285 static void get_tcp4_sock(struct sock *sk, char *tmpbuf, int i)
2286 {
2287 int timer_active;
2288 unsigned long timer_expires;
2289 struct tcp_sock *tp = tcp_sk(sk);
2290 const struct inet_connection_sock *icsk = inet_csk(sk);
2291 struct inet_sock *inet = inet_sk(sk);
2292 __be32 dest = inet->daddr;
2293 __be32 src = inet->rcv_saddr;
2294 __u16 destp = ntohs(inet->dport);
2295 __u16 srcp = ntohs(inet->sport);
2296
2297 if (icsk->icsk_pending == ICSK_TIME_RETRANS) {
2298 timer_active = 1;
2299 timer_expires = icsk->icsk_timeout;
2300 } else if (icsk->icsk_pending == ICSK_TIME_PROBE0) {
2301 timer_active = 4;
2302 timer_expires = icsk->icsk_timeout;
2303 } else if (timer_pending(&sk->sk_timer)) {
2304 timer_active = 2;
2305 timer_expires = sk->sk_timer.expires;
2306 } else {
2307 timer_active = 0;
2308 timer_expires = jiffies;
2309 }
2310
2311 sprintf(tmpbuf, "%4d: %08X:%04X %08X:%04X %02X %08X:%08X %02X:%08lX "
2312 "%08X %5d %8d %lu %d %p %u %u %u %u %d",
2313 i, src, srcp, dest, destp, sk->sk_state,
2314 tp->write_seq - tp->snd_una,
2315 sk->sk_state == TCP_LISTEN ? sk->sk_ack_backlog :
2316 (tp->rcv_nxt - tp->copied_seq),
2317 timer_active,
2318 jiffies_to_clock_t(timer_expires - jiffies),
2319 icsk->icsk_retransmits,
2320 sock_i_uid(sk),
2321 icsk->icsk_probes_out,
2322 sock_i_ino(sk),
2323 atomic_read(&sk->sk_refcnt), sk,
2324 icsk->icsk_rto,
2325 icsk->icsk_ack.ato,
2326 (icsk->icsk_ack.quick << 1) | icsk->icsk_ack.pingpong,
2327 tp->snd_cwnd,
2328 tp->snd_ssthresh >= 0xFFFF ? -1 : tp->snd_ssthresh);
2329 }
2330
2331 static void get_timewait4_sock(struct inet_timewait_sock *tw,
2332 char *tmpbuf, int i)
2333 {
2334 __be32 dest, src;
2335 __u16 destp, srcp;
2336 int ttd = tw->tw_ttd - jiffies;
2337
2338 if (ttd < 0)
2339 ttd = 0;
2340
2341 dest = tw->tw_daddr;
2342 src = tw->tw_rcv_saddr;
2343 destp = ntohs(tw->tw_dport);
2344 srcp = ntohs(tw->tw_sport);
2345
2346 sprintf(tmpbuf, "%4d: %08X:%04X %08X:%04X"
2347 " %02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %p",
2348 i, src, srcp, dest, destp, tw->tw_substate, 0, 0,
2349 3, jiffies_to_clock_t(ttd), 0, 0, 0, 0,
2350 atomic_read(&tw->tw_refcnt), tw);
2351 }
2352
2353 #define TMPSZ 150
2354
2355 static int tcp4_seq_show(struct seq_file *seq, void *v)
2356 {
2357 struct tcp_iter_state* st;
2358 char tmpbuf[TMPSZ + 1];
2359
2360 if (v == SEQ_START_TOKEN) {
2361 seq_printf(seq, "%-*s\n", TMPSZ - 1,
2362 " sl local_address rem_address st tx_queue "
2363 "rx_queue tr tm->when retrnsmt uid timeout "
2364 "inode");
2365 goto out;
2366 }
2367 st = seq->private;
2368
2369 switch (st->state) {
2370 case TCP_SEQ_STATE_LISTENING:
2371 case TCP_SEQ_STATE_ESTABLISHED:
2372 get_tcp4_sock(v, tmpbuf, st->num);
2373 break;
2374 case TCP_SEQ_STATE_OPENREQ:
2375 get_openreq4(st->syn_wait_sk, v, tmpbuf, st->num, st->uid);
2376 break;
2377 case TCP_SEQ_STATE_TIME_WAIT:
2378 get_timewait4_sock(v, tmpbuf, st->num);
2379 break;
2380 }
2381 seq_printf(seq, "%-*s\n", TMPSZ - 1, tmpbuf);
2382 out:
2383 return 0;
2384 }
2385
2386 static struct file_operations tcp4_seq_fops;
2387 static struct tcp_seq_afinfo tcp4_seq_afinfo = {
2388 .owner = THIS_MODULE,
2389 .name = "tcp",
2390 .family = AF_INET,
2391 .seq_show = tcp4_seq_show,
2392 .seq_fops = &tcp4_seq_fops,
2393 };
2394
2395 int __init tcp4_proc_init(void)
2396 {
2397 return tcp_proc_register(&tcp4_seq_afinfo);
2398 }
2399
2400 void tcp4_proc_exit(void)
2401 {
2402 tcp_proc_unregister(&tcp4_seq_afinfo);
2403 }
2404 #endif /* CONFIG_PROC_FS */
2405
2406 DEFINE_PROTO_INUSE(tcp)
2407
2408 struct proto tcp_prot = {
2409 .name = "TCP",
2410 .owner = THIS_MODULE,
2411 .close = tcp_close,
2412 .connect = tcp_v4_connect,
2413 .disconnect = tcp_disconnect,
2414 .accept = inet_csk_accept,
2415 .ioctl = tcp_ioctl,
2416 .init = tcp_v4_init_sock,
2417 .destroy = tcp_v4_destroy_sock,
2418 .shutdown = tcp_shutdown,
2419 .setsockopt = tcp_setsockopt,
2420 .getsockopt = tcp_getsockopt,
2421 .recvmsg = tcp_recvmsg,
2422 .backlog_rcv = tcp_v4_do_rcv,
2423 .hash = inet_hash,
2424 .unhash = inet_unhash,
2425 .get_port = inet_csk_get_port,
2426 .enter_memory_pressure = tcp_enter_memory_pressure,
2427 .sockets_allocated = &tcp_sockets_allocated,
2428 .orphan_count = &tcp_orphan_count,
2429 .memory_allocated = &tcp_memory_allocated,
2430 .memory_pressure = &tcp_memory_pressure,
2431 .sysctl_mem = sysctl_tcp_mem,
2432 .sysctl_wmem = sysctl_tcp_wmem,
2433 .sysctl_rmem = sysctl_tcp_rmem,
2434 .max_header = MAX_TCP_HEADER,
2435 .obj_size = sizeof(struct tcp_sock),
2436 .twsk_prot = &tcp_timewait_sock_ops,
2437 .rsk_prot = &tcp_request_sock_ops,
2438 .hashinfo = &tcp_hashinfo,
2439 #ifdef CONFIG_COMPAT
2440 .compat_setsockopt = compat_tcp_setsockopt,
2441 .compat_getsockopt = compat_tcp_getsockopt,
2442 #endif
2443 REF_PROTO_INUSE(tcp)
2444 };
2445
2446 void __init tcp_v4_init(struct net_proto_family *ops)
2447 {
2448 if (inet_csk_ctl_sock_create(&tcp_socket, PF_INET, SOCK_RAW,
2449 IPPROTO_TCP) < 0)
2450 panic("Failed to create the TCP control socket.\n");
2451 }
2452
2453 EXPORT_SYMBOL(ipv4_specific);
2454 EXPORT_SYMBOL(tcp_hashinfo);
2455 EXPORT_SYMBOL(tcp_prot);
2456 EXPORT_SYMBOL(tcp_v4_conn_request);
2457 EXPORT_SYMBOL(tcp_v4_connect);
2458 EXPORT_SYMBOL(tcp_v4_do_rcv);
2459 EXPORT_SYMBOL(tcp_v4_remember_stamp);
2460 EXPORT_SYMBOL(tcp_v4_send_check);
2461 EXPORT_SYMBOL(tcp_v4_syn_recv_sock);
2462
2463 #ifdef CONFIG_PROC_FS
2464 EXPORT_SYMBOL(tcp_proc_register);
2465 EXPORT_SYMBOL(tcp_proc_unregister);
2466 #endif
2467 EXPORT_SYMBOL(sysctl_tcp_low_latency);
2468
SocketAccessConTroL development