?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
foundations of per-cgroup memory pressure controlling.
[linux-2.6/libata-dev.git] / net / ipv4 / tcp_ipv4.c
blobf48bf312cfe82efb9b37750a878acd09786a2d7d
1 /*
2 * INET An implementation of the TCP/IP protocol suite for the LINUX
3 * operating system. INET is implemented using the BSD Socket
4 * interface as the means of communication with the user level.
5 *
6 * Implementation of the Transmission Control Protocol(TCP).
7 *
8 * IPv4 specific functions
9 *
10 *
11 * code split from:
12 * linux/ipv4/tcp.c
13 * linux/ipv4/tcp_input.c
14 * linux/ipv4/tcp_output.c
15 *
16 * See tcp.c for author information
17 *
18 * This program is free software; you can redistribute it and/or
19 * modify it under the terms of the GNU General Public License
20 * as published by the Free Software Foundation; either version
21 * 2 of the License, or (at your option) any later version.
22 */
23
24 /*
25 * Changes:
26 * David S. Miller : New socket lookup architecture.
27 * This code is dedicated to John Dyson.
28 * David S. Miller : Change semantics of established hash,
29 * half is devoted to TIME_WAIT sockets
30 * and the rest go in the other half.
31 * Andi Kleen : Add support for syncookies and fixed
32 * some bugs: ip options weren't passed to
33 * the TCP layer, missed a check for an
34 * ACK bit.
35 * Andi Kleen : Implemented fast path mtu discovery.
36 * Fixed many serious bugs in the
37 * request_sock handling and moved
38 * most of it into the af independent code.
39 * Added tail drop and some other bugfixes.
40 * Added new listen semantics.
41 * Mike McLagan : Routing by source
42 * Juan Jose Ciarlante: ip_dynaddr bits
43 * Andi Kleen: various fixes.
44 * Vitaly E. Lavrov : Transparent proxy revived after year
45 * coma.
46 * Andi Kleen : Fix new listen.
47 * Andi Kleen : Fix accept error reporting.
48 * YOSHIFUJI Hideaki @USAGI and: Support IPV6_V6ONLY socket option, which
49 * Alexey Kuznetsov allow both IPv4 and IPv6 sockets to bind
50 * a single port at the same time.
51 */
52
53
54 #include <linux/bottom_half.h>
55 #include <linux/types.h>
56 #include <linux/fcntl.h>
57 #include <linux/module.h>
58 #include <linux/random.h>
59 #include <linux/cache.h>
60 #include <linux/jhash.h>
61 #include <linux/init.h>
62 #include <linux/times.h>
63 #include <linux/slab.h>
64
65 #include <net/net_namespace.h>
66 #include <net/icmp.h>
67 #include <net/inet_hashtables.h>
68 #include <net/tcp.h>
69 #include <net/transp_v6.h>
70 #include <net/ipv6.h>
71 #include <net/inet_common.h>
72 #include <net/timewait_sock.h>
73 #include <net/xfrm.h>
74 #include <net/netdma.h>
75 #include <net/secure_seq.h>
76
77 #include <linux/inet.h>
78 #include <linux/ipv6.h>
79 #include <linux/stddef.h>
80 #include <linux/proc_fs.h>
81 #include <linux/seq_file.h>
82
83 #include <linux/crypto.h>
84 #include <linux/scatterlist.h>
85
86 int sysctl_tcp_tw_reuse __read_mostly;
87 int sysctl_tcp_low_latency __read_mostly;
88 EXPORT_SYMBOL(sysctl_tcp_low_latency);
89
90
91 #ifdef CONFIG_TCP_MD5SIG
92 static struct tcp_md5sig_key *tcp_v4_md5_do_lookup(struct sock *sk,
93 __be32 addr);
94 static int tcp_v4_md5_hash_hdr(char *md5_hash, struct tcp_md5sig_key *key,
95 __be32 daddr, __be32 saddr, const struct tcphdr *th);
96 #else
97 static inline
98 struct tcp_md5sig_key *tcp_v4_md5_do_lookup(struct sock *sk, __be32 addr)
99 {
100 return NULL;
101 }
102 #endif
103
104 struct inet_hashinfo tcp_hashinfo;
105 EXPORT_SYMBOL(tcp_hashinfo);
106
107 static inline __u32 tcp_v4_init_sequence(const struct sk_buff *skb)
108 {
109 return secure_tcp_sequence_number(ip_hdr(skb)->daddr,
110 ip_hdr(skb)->saddr,
111 tcp_hdr(skb)->dest,
112 tcp_hdr(skb)->source);
113 }
114
115 int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp)
116 {
117 const struct tcp_timewait_sock *tcptw = tcp_twsk(sktw);
118 struct tcp_sock *tp = tcp_sk(sk);
119
120 /* With PAWS, it is safe from the viewpoint
121 of data integrity. Even without PAWS it is safe provided sequence
122 spaces do not overlap i.e. at data rates <= 80Mbit/sec.
123
124 Actually, the idea is close to VJ's one, only timestamp cache is
125 held not per host, but per port pair and TW bucket is used as state
126 holder.
127
128 If TW bucket has been already destroyed we fall back to VJ's scheme
129 and use initial timestamp retrieved from peer table.
130 */
131 if (tcptw->tw_ts_recent_stamp &&
132 (twp == NULL || (sysctl_tcp_tw_reuse &&
133 get_seconds() - tcptw->tw_ts_recent_stamp > 1))) {
134 tp->write_seq = tcptw->tw_snd_nxt + 65535 + 2;
135 if (tp->write_seq == 0)
136 tp->write_seq = 1;
137 tp->rx_opt.ts_recent = tcptw->tw_ts_recent;
138 tp->rx_opt.ts_recent_stamp = tcptw->tw_ts_recent_stamp;
139 sock_hold(sktw);
140 return 1;
141 }
142
143 return 0;
144 }
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 sockaddr_in *usin = (struct sockaddr_in *)uaddr;
151 struct inet_sock *inet = inet_sk(sk);
152 struct tcp_sock *tp = tcp_sk(sk);
153 __be16 orig_sport, orig_dport;
154 __be32 daddr, nexthop;
155 struct flowi4 *fl4;
156 struct rtable *rt;
157 int err;
158 struct ip_options_rcu *inet_opt;
159
160 if (addr_len < sizeof(struct sockaddr_in))
161 return -EINVAL;
162
163 if (usin->sin_family != AF_INET)
164 return -EAFNOSUPPORT;
165
166 nexthop = daddr = usin->sin_addr.s_addr;
167 inet_opt = rcu_dereference_protected(inet->inet_opt,
168 sock_owned_by_user(sk));
169 if (inet_opt && inet_opt->opt.srr) {
170 if (!daddr)
171 return -EINVAL;
172 nexthop = inet_opt->opt.faddr;
173 }
174
175 orig_sport = inet->inet_sport;
176 orig_dport = usin->sin_port;
177 fl4 = &inet->cork.fl.u.ip4;
178 rt = ip_route_connect(fl4, nexthop, inet->inet_saddr,
179 RT_CONN_FLAGS(sk), sk->sk_bound_dev_if,
180 IPPROTO_TCP,
181 orig_sport, orig_dport, sk, true);
182 if (IS_ERR(rt)) {
183 err = PTR_ERR(rt);
184 if (err == -ENETUNREACH)
185 IP_INC_STATS_BH(sock_net(sk), IPSTATS_MIB_OUTNOROUTES);
186 return err;
187 }
188
189 if (rt->rt_flags & (RTCF_MULTICAST | RTCF_BROADCAST)) {
190 ip_rt_put(rt);
191 return -ENETUNREACH;
192 }
193
194 if (!inet_opt || !inet_opt->opt.srr)
195 daddr = fl4->daddr;
196
197 if (!inet->inet_saddr)
198 inet->inet_saddr = fl4->saddr;
199 inet->inet_rcv_saddr = inet->inet_saddr;
200
201 if (tp->rx_opt.ts_recent_stamp && inet->inet_daddr != daddr) {
202 /* Reset inherited state */
203 tp->rx_opt.ts_recent = 0;
204 tp->rx_opt.ts_recent_stamp = 0;
205 tp->write_seq = 0;
206 }
207
208 if (tcp_death_row.sysctl_tw_recycle &&
209 !tp->rx_opt.ts_recent_stamp && fl4->daddr == daddr) {
210 struct inet_peer *peer = rt_get_peer(rt, fl4->daddr);
211 /*
212 * VJ's idea. We save last timestamp seen from
213 * the destination in peer table, when entering state
214 * TIME-WAIT * and initialize rx_opt.ts_recent from it,
215 * when trying new connection.
216 */
217 if (peer) {
218 inet_peer_refcheck(peer);
219 if ((u32)get_seconds() - peer->tcp_ts_stamp <= TCP_PAWS_MSL) {
220 tp->rx_opt.ts_recent_stamp = peer->tcp_ts_stamp;
221 tp->rx_opt.ts_recent = peer->tcp_ts;
222 }
223 }
224 }
225
226 inet->inet_dport = usin->sin_port;
227 inet->inet_daddr = daddr;
228
229 inet_csk(sk)->icsk_ext_hdr_len = 0;
230 if (inet_opt)
231 inet_csk(sk)->icsk_ext_hdr_len = inet_opt->opt.optlen;
232
233 tp->rx_opt.mss_clamp = TCP_MSS_DEFAULT;
234
235 /* Socket identity is still unknown (sport may be zero).
236 * However we set state to SYN-SENT and not releasing socket
237 * lock select source port, enter ourselves into the hash tables and
238 * complete initialization after this.
239 */
240 tcp_set_state(sk, TCP_SYN_SENT);
241 err = inet_hash_connect(&tcp_death_row, sk);
242 if (err)
243 goto failure;
244
245 rt = ip_route_newports(fl4, rt, orig_sport, orig_dport,
246 inet->inet_sport, inet->inet_dport, sk);
247 if (IS_ERR(rt)) {
248 err = PTR_ERR(rt);
249 rt = NULL;
250 goto failure;
251 }
252 /* OK, now commit destination to socket. */
253 sk->sk_gso_type = SKB_GSO_TCPV4;
254 sk_setup_caps(sk, &rt->dst);
255
256 if (!tp->write_seq)
257 tp->write_seq = secure_tcp_sequence_number(inet->inet_saddr,
258 inet->inet_daddr,
259 inet->inet_sport,
260 usin->sin_port);
261
262 inet->inet_id = tp->write_seq ^ jiffies;
263
264 err = tcp_connect(sk);
265 rt = NULL;
266 if (err)
267 goto failure;
268
269 return 0;
270
271 failure:
272 /*
273 * This unhashes the socket and releases the local port,
274 * if necessary.
275 */
276 tcp_set_state(sk, TCP_CLOSE);
277 ip_rt_put(rt);
278 sk->sk_route_caps = 0;
279 inet->inet_dport = 0;
280 return err;
281 }
282 EXPORT_SYMBOL(tcp_v4_connect);
283
284 /*
285 * This routine does path mtu discovery as defined in RFC1191.
286 */
287 static void do_pmtu_discovery(struct sock *sk, const struct iphdr *iph, u32 mtu)
288 {
289 struct dst_entry *dst;
290 struct inet_sock *inet = inet_sk(sk);
291
292 /* We are not interested in TCP_LISTEN and open_requests (SYN-ACKs
293 * send out by Linux are always <576bytes so they should go through
294 * unfragmented).
295 */
296 if (sk->sk_state == TCP_LISTEN)
297 return;
298
299 /* We don't check in the destentry if pmtu discovery is forbidden
300 * on this route. We just assume that no packet_to_big packets
301 * are send back when pmtu discovery is not active.
302 * There is a small race when the user changes this flag in the
303 * route, but I think that's acceptable.
304 */
305 if ((dst = __sk_dst_check(sk, 0)) == NULL)
306 return;
307
308 dst->ops->update_pmtu(dst, mtu);
309
310 /* Something is about to be wrong... Remember soft error
311 * for the case, if this connection will not able to recover.
312 */
313 if (mtu < dst_mtu(dst) && ip_dont_fragment(sk, dst))
314 sk->sk_err_soft = EMSGSIZE;
315
316 mtu = dst_mtu(dst);
317
318 if (inet->pmtudisc != IP_PMTUDISC_DONT &&
319 inet_csk(sk)->icsk_pmtu_cookie > mtu) {
320 tcp_sync_mss(sk, mtu);
321
322 /* Resend the TCP packet because it's
323 * clear that the old packet has been
324 * dropped. This is the new "fast" path mtu
325 * discovery.
326 */
327 tcp_simple_retransmit(sk);
328 } /* else let the usual retransmit timer handle it */
329 }
330
331 /*
332 * This routine is called by the ICMP module when it gets some
333 * sort of error condition. If err < 0 then the socket should
334 * be closed and the error returned to the user. If err > 0
335 * it's just the icmp type << 8 | icmp code. After adjustment
336 * header points to the first 8 bytes of the tcp header. We need
337 * to find the appropriate port.
338 *
339 * The locking strategy used here is very "optimistic". When
340 * someone else accesses the socket the ICMP is just dropped
341 * and for some paths there is no check at all.
342 * A more general error queue to queue errors for later handling
343 * is probably better.
344 *
345 */
346
347 void tcp_v4_err(struct sk_buff *icmp_skb, u32 info)
348 {
349 const struct iphdr *iph = (const struct iphdr *)icmp_skb->data;
350 struct tcphdr *th = (struct tcphdr *)(icmp_skb->data + (iph->ihl << 2));
351 struct inet_connection_sock *icsk;
352 struct tcp_sock *tp;
353 struct inet_sock *inet;
354 const int type = icmp_hdr(icmp_skb)->type;
355 const int code = icmp_hdr(icmp_skb)->code;
356 struct sock *sk;
357 struct sk_buff *skb;
358 __u32 seq;
359 __u32 remaining;
360 int err;
361 struct net *net = dev_net(icmp_skb->dev);
362
363 if (icmp_skb->len < (iph->ihl << 2) + 8) {
364 ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS);
365 return;
366 }
367
368 sk = inet_lookup(net, &tcp_hashinfo, iph->daddr, th->dest,
369 iph->saddr, th->source, inet_iif(icmp_skb));
370 if (!sk) {
371 ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS);
372 return;
373 }
374 if (sk->sk_state == TCP_TIME_WAIT) {
375 inet_twsk_put(inet_twsk(sk));
376 return;
377 }
378
379 bh_lock_sock(sk);
380 /* If too many ICMPs get dropped on busy
381 * servers this needs to be solved differently.
382 */
383 if (sock_owned_by_user(sk))
384 NET_INC_STATS_BH(net, LINUX_MIB_LOCKDROPPEDICMPS);
385
386 if (sk->sk_state == TCP_CLOSE)
387 goto out;
388
389 if (unlikely(iph->ttl < inet_sk(sk)->min_ttl)) {
390 NET_INC_STATS_BH(net, LINUX_MIB_TCPMINTTLDROP);
391 goto out;
392 }
393
394 icsk = inet_csk(sk);
395 tp = tcp_sk(sk);
396 seq = ntohl(th->seq);
397 if (sk->sk_state != TCP_LISTEN &&
398 !between(seq, tp->snd_una, tp->snd_nxt)) {
399 NET_INC_STATS_BH(net, LINUX_MIB_OUTOFWINDOWICMPS);
400 goto out;
401 }
402
403 switch (type) {
404 case ICMP_SOURCE_QUENCH:
405 /* Just silently ignore these. */
406 goto out;
407 case ICMP_PARAMETERPROB:
408 err = EPROTO;
409 break;
410 case ICMP_DEST_UNREACH:
411 if (code > NR_ICMP_UNREACH)
412 goto out;
413
414 if (code == ICMP_FRAG_NEEDED) { /* PMTU discovery (RFC1191) */
415 if (!sock_owned_by_user(sk))
416 do_pmtu_discovery(sk, iph, info);
417 goto out;
418 }
419
420 err = icmp_err_convert[code].errno;
421 /* check if icmp_skb allows revert of backoff
422 * (see draft-zimmermann-tcp-lcd) */
423 if (code != ICMP_NET_UNREACH && code != ICMP_HOST_UNREACH)
424 break;
425 if (seq != tp->snd_una || !icsk->icsk_retransmits ||
426 !icsk->icsk_backoff)
427 break;
428
429 if (sock_owned_by_user(sk))
430 break;
431
432 icsk->icsk_backoff--;
433 inet_csk(sk)->icsk_rto = (tp->srtt ? __tcp_set_rto(tp) :
434 TCP_TIMEOUT_INIT) << icsk->icsk_backoff;
435 tcp_bound_rto(sk);
436
437 skb = tcp_write_queue_head(sk);
438 BUG_ON(!skb);
439
440 remaining = icsk->icsk_rto - min(icsk->icsk_rto,
441 tcp_time_stamp - TCP_SKB_CB(skb)->when);
442
443 if (remaining) {
444 inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
445 remaining, TCP_RTO_MAX);
446 } else {
447 /* RTO revert clocked out retransmission.
448 * Will retransmit now */
449 tcp_retransmit_timer(sk);
450 }
451
452 break;
453 case ICMP_TIME_EXCEEDED:
454 err = EHOSTUNREACH;
455 break;
456 default:
457 goto out;
458 }
459
460 switch (sk->sk_state) {
461 struct request_sock *req, **prev;
462 case TCP_LISTEN:
463 if (sock_owned_by_user(sk))
464 goto out;
465
466 req = inet_csk_search_req(sk, &prev, th->dest,
467 iph->daddr, iph->saddr);
468 if (!req)
469 goto out;
470
471 /* ICMPs are not backlogged, hence we cannot get
472 an established socket here.
473 */
474 WARN_ON(req->sk);
475
476 if (seq != tcp_rsk(req)->snt_isn) {
477 NET_INC_STATS_BH(net, LINUX_MIB_OUTOFWINDOWICMPS);
478 goto out;
479 }
480
481 /*
482 * Still in SYN_RECV, just remove it silently.
483 * There is no good way to pass the error to the newly
484 * created socket, and POSIX does not want network
485 * errors returned from accept().
486 */
487 inet_csk_reqsk_queue_drop(sk, req, prev);
488 goto out;
489
490 case TCP_SYN_SENT:
491 case TCP_SYN_RECV: /* Cannot happen.
492 It can f.e. if SYNs crossed.
493 */
494 if (!sock_owned_by_user(sk)) {
495 sk->sk_err = err;
496
497 sk->sk_error_report(sk);
498
499 tcp_done(sk);
500 } else {
501 sk->sk_err_soft = err;
502 }
503 goto out;
504 }
505
506 /* If we've already connected we will keep trying
507 * until we time out, or the user gives up.
508 *
509 * rfc1122 4.2.3.9 allows to consider as hard errors
510 * only PROTO_UNREACH and PORT_UNREACH (well, FRAG_FAILED too,
511 * but it is obsoleted by pmtu discovery).
512 *
513 * Note, that in modern internet, where routing is unreliable
514 * and in each dark corner broken firewalls sit, sending random
515 * errors ordered by their masters even this two messages finally lose
516 * their original sense (even Linux sends invalid PORT_UNREACHs)
517 *
518 * Now we are in compliance with RFCs.
519 * --ANK (980905)
520 */
521
522 inet = inet_sk(sk);
523 if (!sock_owned_by_user(sk) && inet->recverr) {
524 sk->sk_err = err;
525 sk->sk_error_report(sk);
526 } else { /* Only an error on timeout */
527 sk->sk_err_soft = err;
528 }
529
530 out:
531 bh_unlock_sock(sk);
532 sock_put(sk);
533 }
534
535 static void __tcp_v4_send_check(struct sk_buff *skb,
536 __be32 saddr, __be32 daddr)
537 {
538 struct tcphdr *th = tcp_hdr(skb);
539
540 if (skb->ip_summed == CHECKSUM_PARTIAL) {
541 th->check = ~tcp_v4_check(skb->len, saddr, daddr, 0);
542 skb->csum_start = skb_transport_header(skb) - skb->head;
543 skb->csum_offset = offsetof(struct tcphdr, check);
544 } else {
545 th->check = tcp_v4_check(skb->len, saddr, daddr,
546 csum_partial(th,
547 th->doff << 2,
548 skb->csum));
549 }
550 }
551
552 /* This routine computes an IPv4 TCP checksum. */
553 void tcp_v4_send_check(struct sock *sk, struct sk_buff *skb)
554 {
555 const struct inet_sock *inet = inet_sk(sk);
556
557 __tcp_v4_send_check(skb, inet->inet_saddr, inet->inet_daddr);
558 }
559 EXPORT_SYMBOL(tcp_v4_send_check);
560
561 int tcp_v4_gso_send_check(struct sk_buff *skb)
562 {
563 const struct iphdr *iph;
564 struct tcphdr *th;
565
566 if (!pskb_may_pull(skb, sizeof(*th)))
567 return -EINVAL;
568
569 iph = ip_hdr(skb);
570 th = tcp_hdr(skb);
571
572 th->check = 0;
573 skb->ip_summed = CHECKSUM_PARTIAL;
574 __tcp_v4_send_check(skb, iph->saddr, iph->daddr);
575 return 0;
576 }
577
578 /*
579 * This routine will send an RST to the other tcp.
580 *
581 * Someone asks: why I NEVER use socket parameters (TOS, TTL etc.)
582 * for reset.
583 * Answer: if a packet caused RST, it is not for a socket
584 * existing in our system, if it is matched to a socket,
585 * it is just duplicate segment or bug in other side's TCP.
586 * So that we build reply only basing on parameters
587 * arrived with segment.
588 * Exception: precedence violation. We do not implement it in any case.
589 */
590
591 static void tcp_v4_send_reset(struct sock *sk, struct sk_buff *skb)
592 {
593 const struct tcphdr *th = tcp_hdr(skb);
594 struct {
595 struct tcphdr th;
596 #ifdef CONFIG_TCP_MD5SIG
597 __be32 opt[(TCPOLEN_MD5SIG_ALIGNED >> 2)];
598 #endif
599 } rep;
600 struct ip_reply_arg arg;
601 #ifdef CONFIG_TCP_MD5SIG
602 struct tcp_md5sig_key *key;
603 #endif
604 struct net *net;
605
606 /* Never send a reset in response to a reset. */
607 if (th->rst)
608 return;
609
610 if (skb_rtable(skb)->rt_type != RTN_LOCAL)
611 return;
612
613 /* Swap the send and the receive. */
614 memset(&rep, 0, sizeof(rep));
615 rep.th.dest = th->source;
616 rep.th.source = th->dest;
617 rep.th.doff = sizeof(struct tcphdr) / 4;
618 rep.th.rst = 1;
619
620 if (th->ack) {
621 rep.th.seq = th->ack_seq;
622 } else {
623 rep.th.ack = 1;
624 rep.th.ack_seq = htonl(ntohl(th->seq) + th->syn + th->fin +
625 skb->len - (th->doff << 2));
626 }
627
628 memset(&arg, 0, sizeof(arg));
629 arg.iov[0].iov_base = (unsigned char *)&rep;
630 arg.iov[0].iov_len = sizeof(rep.th);
631
632 #ifdef CONFIG_TCP_MD5SIG
633 key = sk ? tcp_v4_md5_do_lookup(sk, ip_hdr(skb)->daddr) : NULL;
634 if (key) {
635 rep.opt[0] = htonl((TCPOPT_NOP << 24) |
636 (TCPOPT_NOP << 16) |
637 (TCPOPT_MD5SIG << 8) |
638 TCPOLEN_MD5SIG);
639 /* Update length and the length the header thinks exists */
640 arg.iov[0].iov_len += TCPOLEN_MD5SIG_ALIGNED;
641 rep.th.doff = arg.iov[0].iov_len / 4;
642
643 tcp_v4_md5_hash_hdr((__u8 *) &rep.opt[1],
644 key, ip_hdr(skb)->saddr,
645 ip_hdr(skb)->daddr, &rep.th);
646 }
647 #endif
648 arg.csum = csum_tcpudp_nofold(ip_hdr(skb)->daddr,
649 ip_hdr(skb)->saddr, /* XXX */
650 arg.iov[0].iov_len, IPPROTO_TCP, 0);
651 arg.csumoffset = offsetof(struct tcphdr, check) / 2;
652 arg.flags = (sk && inet_sk(sk)->transparent) ? IP_REPLY_ARG_NOSRCCHECK : 0;
653
654 net = dev_net(skb_dst(skb)->dev);
655 arg.tos = ip_hdr(skb)->tos;
656 ip_send_reply(net->ipv4.tcp_sock, skb, ip_hdr(skb)->saddr,
657 &arg, arg.iov[0].iov_len);
658
659 TCP_INC_STATS_BH(net, TCP_MIB_OUTSEGS);
660 TCP_INC_STATS_BH(net, TCP_MIB_OUTRSTS);
661 }
662
663 /* The code following below sending ACKs in SYN-RECV and TIME-WAIT states
664 outside socket context is ugly, certainly. What can I do?
665 */
666
667 static void tcp_v4_send_ack(struct sk_buff *skb, u32 seq, u32 ack,
668 u32 win, u32 ts, int oif,
669 struct tcp_md5sig_key *key,
670 int reply_flags, u8 tos)
671 {
672 const struct tcphdr *th = tcp_hdr(skb);
673 struct {
674 struct tcphdr th;
675 __be32 opt[(TCPOLEN_TSTAMP_ALIGNED >> 2)
676 #ifdef CONFIG_TCP_MD5SIG
677 + (TCPOLEN_MD5SIG_ALIGNED >> 2)
678 #endif
679 ];
680 } rep;
681 struct ip_reply_arg arg;
682 struct net *net = dev_net(skb_dst(skb)->dev);
683
684 memset(&rep.th, 0, sizeof(struct tcphdr));
685 memset(&arg, 0, sizeof(arg));
686
687 arg.iov[0].iov_base = (unsigned char *)&rep;
688 arg.iov[0].iov_len = sizeof(rep.th);
689 if (ts) {
690 rep.opt[0] = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
691 (TCPOPT_TIMESTAMP << 8) |
692 TCPOLEN_TIMESTAMP);
693 rep.opt[1] = htonl(tcp_time_stamp);
694 rep.opt[2] = htonl(ts);
695 arg.iov[0].iov_len += TCPOLEN_TSTAMP_ALIGNED;
696 }
697
698 /* Swap the send and the receive. */
699 rep.th.dest = th->source;
700 rep.th.source = th->dest;
701 rep.th.doff = arg.iov[0].iov_len / 4;
702 rep.th.seq = htonl(seq);
703 rep.th.ack_seq = htonl(ack);
704 rep.th.ack = 1;
705 rep.th.window = htons(win);
706
707 #ifdef CONFIG_TCP_MD5SIG
708 if (key) {
709 int offset = (ts) ? 3 : 0;
710
711 rep.opt[offset++] = htonl((TCPOPT_NOP << 24) |
712 (TCPOPT_NOP << 16) |
713 (TCPOPT_MD5SIG << 8) |
714 TCPOLEN_MD5SIG);
715 arg.iov[0].iov_len += TCPOLEN_MD5SIG_ALIGNED;
716 rep.th.doff = arg.iov[0].iov_len/4;
717
718 tcp_v4_md5_hash_hdr((__u8 *) &rep.opt[offset],
719 key, ip_hdr(skb)->saddr,
720 ip_hdr(skb)->daddr, &rep.th);
721 }
722 #endif
723 arg.flags = reply_flags;
724 arg.csum = csum_tcpudp_nofold(ip_hdr(skb)->daddr,
725 ip_hdr(skb)->saddr, /* XXX */
726 arg.iov[0].iov_len, IPPROTO_TCP, 0);
727 arg.csumoffset = offsetof(struct tcphdr, check) / 2;
728 if (oif)
729 arg.bound_dev_if = oif;
730 arg.tos = tos;
731 ip_send_reply(net->ipv4.tcp_sock, skb, ip_hdr(skb)->saddr,
732 &arg, arg.iov[0].iov_len);
733
734 TCP_INC_STATS_BH(net, TCP_MIB_OUTSEGS);
735 }
736
737 static void tcp_v4_timewait_ack(struct sock *sk, struct sk_buff *skb)
738 {
739 struct inet_timewait_sock *tw = inet_twsk(sk);
740 struct tcp_timewait_sock *tcptw = tcp_twsk(sk);
741
742 tcp_v4_send_ack(skb, tcptw->tw_snd_nxt, tcptw->tw_rcv_nxt,
743 tcptw->tw_rcv_wnd >> tw->tw_rcv_wscale,
744 tcptw->tw_ts_recent,
745 tw->tw_bound_dev_if,
746 tcp_twsk_md5_key(tcptw),
747 tw->tw_transparent ? IP_REPLY_ARG_NOSRCCHECK : 0,
748 tw->tw_tos
749 );
750
751 inet_twsk_put(tw);
752 }
753
754 static void tcp_v4_reqsk_send_ack(struct sock *sk, struct sk_buff *skb,
755 struct request_sock *req)
756 {
757 tcp_v4_send_ack(skb, tcp_rsk(req)->snt_isn + 1,
758 tcp_rsk(req)->rcv_isn + 1, req->rcv_wnd,
759 req->ts_recent,
760 0,
761 tcp_v4_md5_do_lookup(sk, ip_hdr(skb)->daddr),
762 inet_rsk(req)->no_srccheck ? IP_REPLY_ARG_NOSRCCHECK : 0,
763 ip_hdr(skb)->tos);
764 }
765
766 /*
767 * Send a SYN-ACK after having received a SYN.
768 * This still operates on a request_sock only, not on a big
769 * socket.
770 */
771 static int tcp_v4_send_synack(struct sock *sk, struct dst_entry *dst,
772 struct request_sock *req,
773 struct request_values *rvp)
774 {
775 const struct inet_request_sock *ireq = inet_rsk(req);
776 struct flowi4 fl4;
777 int err = -1;
778 struct sk_buff * skb;
779
780 /* First, grab a route. */
781 if (!dst && (dst = inet_csk_route_req(sk, &fl4, req)) == NULL)
782 return -1;
783
784 skb = tcp_make_synack(sk, dst, req, rvp);
785
786 if (skb) {
787 __tcp_v4_send_check(skb, ireq->loc_addr, ireq->rmt_addr);
788
789 err = ip_build_and_send_pkt(skb, sk, ireq->loc_addr,
790 ireq->rmt_addr,
791 ireq->opt);
792 err = net_xmit_eval(err);
793 }
794
795 dst_release(dst);
796 return err;
797 }
798
799 static int tcp_v4_rtx_synack(struct sock *sk, struct request_sock *req,
800 struct request_values *rvp)
801 {
802 TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_RETRANSSEGS);
803 return tcp_v4_send_synack(sk, NULL, req, rvp);
804 }
805
806 /*
807 * IPv4 request_sock destructor.
808 */
809 static void tcp_v4_reqsk_destructor(struct request_sock *req)
810 {
811 kfree(inet_rsk(req)->opt);
812 }
813
814 /*
815 * Return 1 if a syncookie should be sent
816 */
817 int tcp_syn_flood_action(struct sock *sk,
818 const struct sk_buff *skb,
819 const char *proto)
820 {
821 const char *msg = "Dropping request";
822 int want_cookie = 0;
823 struct listen_sock *lopt;
824
825
826
827 #ifdef CONFIG_SYN_COOKIES
828 if (sysctl_tcp_syncookies) {
829 msg = "Sending cookies";
830 want_cookie = 1;
831 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPREQQFULLDOCOOKIES);
832 } else
833 #endif
834 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPREQQFULLDROP);
835
836 lopt = inet_csk(sk)->icsk_accept_queue.listen_opt;
837 if (!lopt->synflood_warned) {
838 lopt->synflood_warned = 1;
839 pr_info("%s: Possible SYN flooding on port %d. %s. "
840 " Check SNMP counters.\n",
841 proto, ntohs(tcp_hdr(skb)->dest), msg);
842 }
843 return want_cookie;
844 }
845 EXPORT_SYMBOL(tcp_syn_flood_action);
846
847 /*
848 * Save and compile IPv4 options into the request_sock if needed.
849 */
850 static struct ip_options_rcu *tcp_v4_save_options(struct sock *sk,
851 struct sk_buff *skb)
852 {
853 const struct ip_options *opt = &(IPCB(skb)->opt);
854 struct ip_options_rcu *dopt = NULL;
855
856 if (opt && opt->optlen) {
857 int opt_size = sizeof(*dopt) + opt->optlen;
858
859 dopt = kmalloc(opt_size, GFP_ATOMIC);
860 if (dopt) {
861 if (ip_options_echo(&dopt->opt, skb)) {
862 kfree(dopt);
863 dopt = NULL;
864 }
865 }
866 }
867 return dopt;
868 }
869
870 #ifdef CONFIG_TCP_MD5SIG
871 /*
872 * RFC2385 MD5 checksumming requires a mapping of
873 * IP address->MD5 Key.
874 * We need to maintain these in the sk structure.
875 */
876
877 /* Find the Key structure for an address. */
878 static struct tcp_md5sig_key *
879 tcp_v4_md5_do_lookup(struct sock *sk, __be32 addr)
880 {
881 struct tcp_sock *tp = tcp_sk(sk);
882 int i;
883
884 if (!tp->md5sig_info || !tp->md5sig_info->entries4)
885 return NULL;
886 for (i = 0; i < tp->md5sig_info->entries4; i++) {
887 if (tp->md5sig_info->keys4[i].addr == addr)
888 return &tp->md5sig_info->keys4[i].base;
889 }
890 return NULL;
891 }
892
893 struct tcp_md5sig_key *tcp_v4_md5_lookup(struct sock *sk,
894 struct sock *addr_sk)
895 {
896 return tcp_v4_md5_do_lookup(sk, inet_sk(addr_sk)->inet_daddr);
897 }
898 EXPORT_SYMBOL(tcp_v4_md5_lookup);
899
900 static struct tcp_md5sig_key *tcp_v4_reqsk_md5_lookup(struct sock *sk,
901 struct request_sock *req)
902 {
903 return tcp_v4_md5_do_lookup(sk, inet_rsk(req)->rmt_addr);
904 }
905
906 /* This can be called on a newly created socket, from other files */
907 int tcp_v4_md5_do_add(struct sock *sk, __be32 addr,
908 u8 *newkey, u8 newkeylen)
909 {
910 /* Add Key to the list */
911 struct tcp_md5sig_key *key;
912 struct tcp_sock *tp = tcp_sk(sk);
913 struct tcp4_md5sig_key *keys;
914
915 key = tcp_v4_md5_do_lookup(sk, addr);
916 if (key) {
917 /* Pre-existing entry - just update that one. */
918 kfree(key->key);
919 key->key = newkey;
920 key->keylen = newkeylen;
921 } else {
922 struct tcp_md5sig_info *md5sig;
923
924 if (!tp->md5sig_info) {
925 tp->md5sig_info = kzalloc(sizeof(*tp->md5sig_info),
926 GFP_ATOMIC);
927 if (!tp->md5sig_info) {
928 kfree(newkey);
929 return -ENOMEM;
930 }
931 sk_nocaps_add(sk, NETIF_F_GSO_MASK);
932 }
933
934 md5sig = tp->md5sig_info;
935 if (md5sig->entries4 == 0 &&
936 tcp_alloc_md5sig_pool(sk) == NULL) {
937 kfree(newkey);
938 return -ENOMEM;
939 }
940
941 if (md5sig->alloced4 == md5sig->entries4) {
942 keys = kmalloc((sizeof(*keys) *
943 (md5sig->entries4 + 1)), GFP_ATOMIC);
944 if (!keys) {
945 kfree(newkey);
946 if (md5sig->entries4 == 0)
947 tcp_free_md5sig_pool();
948 return -ENOMEM;
949 }
950
951 if (md5sig->entries4)
952 memcpy(keys, md5sig->keys4,
953 sizeof(*keys) * md5sig->entries4);
954
955 /* Free old key list, and reference new one */
956 kfree(md5sig->keys4);
957 md5sig->keys4 = keys;
958 md5sig->alloced4++;
959 }
960 md5sig->entries4++;
961 md5sig->keys4[md5sig->entries4 - 1].addr = addr;
962 md5sig->keys4[md5sig->entries4 - 1].base.key = newkey;
963 md5sig->keys4[md5sig->entries4 - 1].base.keylen = newkeylen;
964 }
965 return 0;
966 }
967 EXPORT_SYMBOL(tcp_v4_md5_do_add);
968
969 static int tcp_v4_md5_add_func(struct sock *sk, struct sock *addr_sk,
970 u8 *newkey, u8 newkeylen)
971 {
972 return tcp_v4_md5_do_add(sk, inet_sk(addr_sk)->inet_daddr,
973 newkey, newkeylen);
974 }
975
976 int tcp_v4_md5_do_del(struct sock *sk, __be32 addr)
977 {
978 struct tcp_sock *tp = tcp_sk(sk);
979 int i;
980
981 for (i = 0; i < tp->md5sig_info->entries4; i++) {
982 if (tp->md5sig_info->keys4[i].addr == addr) {
983 /* Free the key */
984 kfree(tp->md5sig_info->keys4[i].base.key);
985 tp->md5sig_info->entries4--;
986
987 if (tp->md5sig_info->entries4 == 0) {
988 kfree(tp->md5sig_info->keys4);
989 tp->md5sig_info->keys4 = NULL;
990 tp->md5sig_info->alloced4 = 0;
991 tcp_free_md5sig_pool();
992 } else if (tp->md5sig_info->entries4 != i) {
993 /* Need to do some manipulation */
994 memmove(&tp->md5sig_info->keys4[i],
995 &tp->md5sig_info->keys4[i+1],
996 (tp->md5sig_info->entries4 - i) *
997 sizeof(struct tcp4_md5sig_key));
998 }
999 return 0;
1000 }
1001 }
1002 return -ENOENT;
1003 }
1004 EXPORT_SYMBOL(tcp_v4_md5_do_del);
1005
1006 static void tcp_v4_clear_md5_list(struct sock *sk)
1007 {
1008 struct tcp_sock *tp = tcp_sk(sk);
1009
1010 /* Free each key, then the set of key keys,
1011 * the crypto element, and then decrement our
1012 * hold on the last resort crypto.
1013 */
1014 if (tp->md5sig_info->entries4) {
1015 int i;
1016 for (i = 0; i < tp->md5sig_info->entries4; i++)
1017 kfree(tp->md5sig_info->keys4[i].base.key);
1018 tp->md5sig_info->entries4 = 0;
1019 tcp_free_md5sig_pool();
1020 }
1021 if (tp->md5sig_info->keys4) {
1022 kfree(tp->md5sig_info->keys4);
1023 tp->md5sig_info->keys4 = NULL;
1024 tp->md5sig_info->alloced4 = 0;
1025 }
1026 }
1027
1028 static int tcp_v4_parse_md5_keys(struct sock *sk, char __user *optval,
1029 int optlen)
1030 {
1031 struct tcp_md5sig cmd;
1032 struct sockaddr_in *sin = (struct sockaddr_in *)&cmd.tcpm_addr;
1033 u8 *newkey;
1034
1035 if (optlen < sizeof(cmd))
1036 return -EINVAL;
1037
1038 if (copy_from_user(&cmd, optval, sizeof(cmd)))
1039 return -EFAULT;
1040
1041 if (sin->sin_family != AF_INET)
1042 return -EINVAL;
1043
1044 if (!cmd.tcpm_key || !cmd.tcpm_keylen) {
1045 if (!tcp_sk(sk)->md5sig_info)
1046 return -ENOENT;
1047 return tcp_v4_md5_do_del(sk, sin->sin_addr.s_addr);
1048 }
1049
1050 if (cmd.tcpm_keylen > TCP_MD5SIG_MAXKEYLEN)
1051 return -EINVAL;
1052
1053 if (!tcp_sk(sk)->md5sig_info) {
1054 struct tcp_sock *tp = tcp_sk(sk);
1055 struct tcp_md5sig_info *p;
1056
1057 p = kzalloc(sizeof(*p), sk->sk_allocation);
1058 if (!p)
1059 return -EINVAL;
1060
1061 tp->md5sig_info = p;
1062 sk_nocaps_add(sk, NETIF_F_GSO_MASK);
1063 }
1064
1065 newkey = kmemdup(cmd.tcpm_key, cmd.tcpm_keylen, sk->sk_allocation);
1066 if (!newkey)
1067 return -ENOMEM;
1068 return tcp_v4_md5_do_add(sk, sin->sin_addr.s_addr,
1069 newkey, cmd.tcpm_keylen);
1070 }
1071
1072 static int tcp_v4_md5_hash_pseudoheader(struct tcp_md5sig_pool *hp,
1073 __be32 daddr, __be32 saddr, int nbytes)
1074 {
1075 struct tcp4_pseudohdr *bp;
1076 struct scatterlist sg;
1077
1078 bp = &hp->md5_blk.ip4;
1079
1080 /*
1081 * 1. the TCP pseudo-header (in the order: source IP address,
1082 * destination IP address, zero-padded protocol number, and
1083 * segment length)
1084 */
1085 bp->saddr = saddr;
1086 bp->daddr = daddr;
1087 bp->pad = 0;
1088 bp->protocol = IPPROTO_TCP;
1089 bp->len = cpu_to_be16(nbytes);
1090
1091 sg_init_one(&sg, bp, sizeof(*bp));
1092 return crypto_hash_update(&hp->md5_desc, &sg, sizeof(*bp));
1093 }
1094
1095 static int tcp_v4_md5_hash_hdr(char *md5_hash, struct tcp_md5sig_key *key,
1096 __be32 daddr, __be32 saddr, const struct tcphdr *th)
1097 {
1098 struct tcp_md5sig_pool *hp;
1099 struct hash_desc *desc;
1100
1101 hp = tcp_get_md5sig_pool();
1102 if (!hp)
1103 goto clear_hash_noput;
1104 desc = &hp->md5_desc;
1105
1106 if (crypto_hash_init(desc))
1107 goto clear_hash;
1108 if (tcp_v4_md5_hash_pseudoheader(hp, daddr, saddr, th->doff << 2))
1109 goto clear_hash;
1110 if (tcp_md5_hash_header(hp, th))
1111 goto clear_hash;
1112 if (tcp_md5_hash_key(hp, key))
1113 goto clear_hash;
1114 if (crypto_hash_final(desc, md5_hash))
1115 goto clear_hash;
1116
1117 tcp_put_md5sig_pool();
1118 return 0;
1119
1120 clear_hash:
1121 tcp_put_md5sig_pool();
1122 clear_hash_noput:
1123 memset(md5_hash, 0, 16);
1124 return 1;
1125 }
1126
1127 int tcp_v4_md5_hash_skb(char *md5_hash, struct tcp_md5sig_key *key,
1128 const struct sock *sk, const struct request_sock *req,
1129 const struct sk_buff *skb)
1130 {
1131 struct tcp_md5sig_pool *hp;
1132 struct hash_desc *desc;
1133 const struct tcphdr *th = tcp_hdr(skb);
1134 __be32 saddr, daddr;
1135
1136 if (sk) {
1137 saddr = inet_sk(sk)->inet_saddr;
1138 daddr = inet_sk(sk)->inet_daddr;
1139 } else if (req) {
1140 saddr = inet_rsk(req)->loc_addr;
1141 daddr = inet_rsk(req)->rmt_addr;
1142 } else {
1143 const struct iphdr *iph = ip_hdr(skb);
1144 saddr = iph->saddr;
1145 daddr = iph->daddr;
1146 }
1147
1148 hp = tcp_get_md5sig_pool();
1149 if (!hp)
1150 goto clear_hash_noput;
1151 desc = &hp->md5_desc;
1152
1153 if (crypto_hash_init(desc))
1154 goto clear_hash;
1155
1156 if (tcp_v4_md5_hash_pseudoheader(hp, daddr, saddr, skb->len))
1157 goto clear_hash;
1158 if (tcp_md5_hash_header(hp, th))
1159 goto clear_hash;
1160 if (tcp_md5_hash_skb_data(hp, skb, th->doff << 2))
1161 goto clear_hash;
1162 if (tcp_md5_hash_key(hp, key))
1163 goto clear_hash;
1164 if (crypto_hash_final(desc, md5_hash))
1165 goto clear_hash;
1166
1167 tcp_put_md5sig_pool();
1168 return 0;
1169
1170 clear_hash:
1171 tcp_put_md5sig_pool();
1172 clear_hash_noput:
1173 memset(md5_hash, 0, 16);
1174 return 1;
1175 }
1176 EXPORT_SYMBOL(tcp_v4_md5_hash_skb);
1177
1178 static int tcp_v4_inbound_md5_hash(struct sock *sk, const struct sk_buff *skb)
1179 {
1180 /*
1181 * This gets called for each TCP segment that arrives
1182 * so we want to be efficient.
1183 * We have 3 drop cases:
1184 * o No MD5 hash and one expected.
1185 * o MD5 hash and we're not expecting one.
1186 * o MD5 hash and its wrong.
1187 */
1188 const __u8 *hash_location = NULL;
1189 struct tcp_md5sig_key *hash_expected;
1190 const struct iphdr *iph = ip_hdr(skb);
1191 const struct tcphdr *th = tcp_hdr(skb);
1192 int genhash;
1193 unsigned char newhash[16];
1194
1195 hash_expected = tcp_v4_md5_do_lookup(sk, iph->saddr);
1196 hash_location = tcp_parse_md5sig_option(th);
1197
1198 /* We've parsed the options - do we have a hash? */
1199 if (!hash_expected && !hash_location)
1200 return 0;
1201
1202 if (hash_expected && !hash_location) {
1203 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPMD5NOTFOUND);
1204 return 1;
1205 }
1206
1207 if (!hash_expected && hash_location) {
1208 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPMD5UNEXPECTED);
1209 return 1;
1210 }
1211
1212 /* Okay, so this is hash_expected and hash_location -
1213 * so we need to calculate the checksum.
1214 */
1215 genhash = tcp_v4_md5_hash_skb(newhash,
1216 hash_expected,
1217 NULL, NULL, skb);
1218
1219 if (genhash || memcmp(hash_location, newhash, 16) != 0) {
1220 if (net_ratelimit()) {
1221 printk(KERN_INFO "MD5 Hash failed for (%pI4, %d)->(%pI4, %d)%s\n",
1222 &iph->saddr, ntohs(th->source),
1223 &iph->daddr, ntohs(th->dest),
1224 genhash ? " tcp_v4_calc_md5_hash failed" : "");
1225 }
1226 return 1;
1227 }
1228 return 0;
1229 }
1230
1231 #endif
1232
1233 struct request_sock_ops tcp_request_sock_ops __read_mostly = {
1234 .family = PF_INET,
1235 .obj_size = sizeof(struct tcp_request_sock),
1236 .rtx_syn_ack = tcp_v4_rtx_synack,
1237 .send_ack = tcp_v4_reqsk_send_ack,
1238 .destructor = tcp_v4_reqsk_destructor,
1239 .send_reset = tcp_v4_send_reset,
1240 .syn_ack_timeout = tcp_syn_ack_timeout,
1241 };
1242
1243 #ifdef CONFIG_TCP_MD5SIG
1244 static const struct tcp_request_sock_ops tcp_request_sock_ipv4_ops = {
1245 .md5_lookup = tcp_v4_reqsk_md5_lookup,
1246 .calc_md5_hash = tcp_v4_md5_hash_skb,
1247 };
1248 #endif
1249
1250 int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb)
1251 {
1252 struct tcp_extend_values tmp_ext;
1253 struct tcp_options_received tmp_opt;
1254 const u8 *hash_location;
1255 struct request_sock *req;
1256 struct inet_request_sock *ireq;
1257 struct tcp_sock *tp = tcp_sk(sk);
1258 struct dst_entry *dst = NULL;
1259 __be32 saddr = ip_hdr(skb)->saddr;
1260 __be32 daddr = ip_hdr(skb)->daddr;
1261 __u32 isn = TCP_SKB_CB(skb)->when;
1262 int want_cookie = 0;
1263
1264 /* Never answer to SYNs send to broadcast or multicast */
1265 if (skb_rtable(skb)->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))
1266 goto drop;
1267
1268 /* TW buckets are converted to open requests without
1269 * limitations, they conserve resources and peer is
1270 * evidently real one.
1271 */
1272 if (inet_csk_reqsk_queue_is_full(sk) && !isn) {
1273 want_cookie = tcp_syn_flood_action(sk, skb, "TCP");
1274 if (!want_cookie)
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 = inet_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 = TCP_MSS_DEFAULT;
1296 tmp_opt.user_mss = tp->rx_opt.user_mss;
1297 tcp_parse_options(skb, &tmp_opt, &hash_location, 0);
1298
1299 if (tmp_opt.cookie_plus > 0 &&
1300 tmp_opt.saw_tstamp &&
1301 !tp->rx_opt.cookie_out_never &&
1302 (sysctl_tcp_cookie_size > 0 ||
1303 (tp->cookie_values != NULL &&
1304 tp->cookie_values->cookie_desired > 0))) {
1305 u8 *c;
1306 u32 *mess = &tmp_ext.cookie_bakery[COOKIE_DIGEST_WORDS];
1307 int l = tmp_opt.cookie_plus - TCPOLEN_COOKIE_BASE;
1308
1309 if (tcp_cookie_generator(&tmp_ext.cookie_bakery[0]) != 0)
1310 goto drop_and_release;
1311
1312 /* Secret recipe starts with IP addresses */
1313 *mess++ ^= (__force u32)daddr;
1314 *mess++ ^= (__force u32)saddr;
1315
1316 /* plus variable length Initiator Cookie */
1317 c = (u8 *)mess;
1318 while (l-- > 0)
1319 *c++ ^= *hash_location++;
1320
1321 want_cookie = 0; /* not our kind of cookie */
1322 tmp_ext.cookie_out_never = 0; /* false */
1323 tmp_ext.cookie_plus = tmp_opt.cookie_plus;
1324 } else if (!tp->rx_opt.cookie_in_always) {
1325 /* redundant indications, but ensure initialization. */
1326 tmp_ext.cookie_out_never = 1; /* true */
1327 tmp_ext.cookie_plus = 0;
1328 } else {
1329 goto drop_and_release;
1330 }
1331 tmp_ext.cookie_in_always = tp->rx_opt.cookie_in_always;
1332
1333 if (want_cookie && !tmp_opt.saw_tstamp)
1334 tcp_clear_options(&tmp_opt);
1335
1336 tmp_opt.tstamp_ok = tmp_opt.saw_tstamp;
1337 tcp_openreq_init(req, &tmp_opt, skb);
1338
1339 ireq = inet_rsk(req);
1340 ireq->loc_addr = daddr;
1341 ireq->rmt_addr = saddr;
1342 ireq->no_srccheck = inet_sk(sk)->transparent;
1343 ireq->opt = tcp_v4_save_options(sk, skb);
1344
1345 if (security_inet_conn_request(sk, skb, req))
1346 goto drop_and_free;
1347
1348 if (!want_cookie || tmp_opt.tstamp_ok)
1349 TCP_ECN_create_request(req, tcp_hdr(skb));
1350
1351 if (want_cookie) {
1352 isn = cookie_v4_init_sequence(sk, skb, &req->mss);
1353 req->cookie_ts = tmp_opt.tstamp_ok;
1354 } else if (!isn) {
1355 struct inet_peer *peer = NULL;
1356 struct flowi4 fl4;
1357
1358 /* VJ's idea. We save last timestamp seen
1359 * from the destination in peer table, when entering
1360 * state TIME-WAIT, and check against it before
1361 * accepting new connection request.
1362 *
1363 * If "isn" is not zero, this request hit alive
1364 * timewait bucket, so that all the necessary checks
1365 * are made in the function processing timewait state.
1366 */
1367 if (tmp_opt.saw_tstamp &&
1368 tcp_death_row.sysctl_tw_recycle &&
1369 (dst = inet_csk_route_req(sk, &fl4, req)) != NULL &&
1370 fl4.daddr == saddr &&
1371 (peer = rt_get_peer((struct rtable *)dst, fl4.daddr)) != NULL) {
1372 inet_peer_refcheck(peer);
1373 if ((u32)get_seconds() - peer->tcp_ts_stamp < TCP_PAWS_MSL &&
1374 (s32)(peer->tcp_ts - req->ts_recent) >
1375 TCP_PAWS_WINDOW) {
1376 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_PAWSPASSIVEREJECTED);
1377 goto drop_and_release;
1378 }
1379 }
1380 /* Kill the following clause, if you dislike this way. */
1381 else if (!sysctl_tcp_syncookies &&
1382 (sysctl_max_syn_backlog - inet_csk_reqsk_queue_len(sk) <
1383 (sysctl_max_syn_backlog >> 2)) &&
1384 (!peer || !peer->tcp_ts_stamp) &&
1385 (!dst || !dst_metric(dst, RTAX_RTT))) {
1386 /* Without syncookies last quarter of
1387 * backlog is filled with destinations,
1388 * proven to be alive.
1389 * It means that we continue to communicate
1390 * to destinations, already remembered
1391 * to the moment of synflood.
1392 */
1393 LIMIT_NETDEBUG(KERN_DEBUG "TCP: drop open request from %pI4/%u\n",
1394 &saddr, ntohs(tcp_hdr(skb)->source));
1395 goto drop_and_release;
1396 }
1397
1398 isn = tcp_v4_init_sequence(skb);
1399 }
1400 tcp_rsk(req)->snt_isn = isn;
1401 tcp_rsk(req)->snt_synack = tcp_time_stamp;
1402
1403 if (tcp_v4_send_synack(sk, dst, req,
1404 (struct request_values *)&tmp_ext) ||
1405 want_cookie)
1406 goto drop_and_free;
1407
1408 inet_csk_reqsk_queue_hash_add(sk, req, TCP_TIMEOUT_INIT);
1409 return 0;
1410
1411 drop_and_release:
1412 dst_release(dst);
1413 drop_and_free:
1414 reqsk_free(req);
1415 drop:
1416 return 0;
1417 }
1418 EXPORT_SYMBOL(tcp_v4_conn_request);
1419
1420
1421 /*
1422 * The three way handshake has completed - we got a valid synack -
1423 * now create the new socket.
1424 */
1425 struct sock *tcp_v4_syn_recv_sock(struct sock *sk, struct sk_buff *skb,
1426 struct request_sock *req,
1427 struct dst_entry *dst)
1428 {
1429 struct inet_request_sock *ireq;
1430 struct inet_sock *newinet;
1431 struct tcp_sock *newtp;
1432 struct sock *newsk;
1433 #ifdef CONFIG_TCP_MD5SIG
1434 struct tcp_md5sig_key *key;
1435 #endif
1436 struct ip_options_rcu *inet_opt;
1437
1438 if (sk_acceptq_is_full(sk))
1439 goto exit_overflow;
1440
1441 newsk = tcp_create_openreq_child(sk, req, skb);
1442 if (!newsk)
1443 goto exit_nonewsk;
1444
1445 newsk->sk_gso_type = SKB_GSO_TCPV4;
1446
1447 newtp = tcp_sk(newsk);
1448 newinet = inet_sk(newsk);
1449 ireq = inet_rsk(req);
1450 newinet->inet_daddr = ireq->rmt_addr;
1451 newinet->inet_rcv_saddr = ireq->loc_addr;
1452 newinet->inet_saddr = ireq->loc_addr;
1453 inet_opt = ireq->opt;
1454 rcu_assign_pointer(newinet->inet_opt, inet_opt);
1455 ireq->opt = NULL;
1456 newinet->mc_index = inet_iif(skb);
1457 newinet->mc_ttl = ip_hdr(skb)->ttl;
1458 inet_csk(newsk)->icsk_ext_hdr_len = 0;
1459 if (inet_opt)
1460 inet_csk(newsk)->icsk_ext_hdr_len = inet_opt->opt.optlen;
1461 newinet->inet_id = newtp->write_seq ^ jiffies;
1462
1463 if (!dst && (dst = inet_csk_route_child_sock(sk, newsk, req)) == NULL)
1464 goto put_and_exit;
1465
1466 sk_setup_caps(newsk, dst);
1467
1468 tcp_mtup_init(newsk);
1469 tcp_sync_mss(newsk, dst_mtu(dst));
1470 newtp->advmss = dst_metric_advmss(dst);
1471 if (tcp_sk(sk)->rx_opt.user_mss &&
1472 tcp_sk(sk)->rx_opt.user_mss < newtp->advmss)
1473 newtp->advmss = tcp_sk(sk)->rx_opt.user_mss;
1474
1475 tcp_initialize_rcv_mss(newsk);
1476 if (tcp_rsk(req)->snt_synack)
1477 tcp_valid_rtt_meas(newsk,
1478 tcp_time_stamp - tcp_rsk(req)->snt_synack);
1479 newtp->total_retrans = req->retrans;
1480
1481 #ifdef CONFIG_TCP_MD5SIG
1482 /* Copy over the MD5 key from the original socket */
1483 key = tcp_v4_md5_do_lookup(sk, newinet->inet_daddr);
1484 if (key != NULL) {
1485 /*
1486 * We're using one, so create a matching key
1487 * on the newsk structure. If we fail to get
1488 * memory, then we end up not copying the key
1489 * across. Shucks.
1490 */
1491 char *newkey = kmemdup(key->key, key->keylen, GFP_ATOMIC);
1492 if (newkey != NULL)
1493 tcp_v4_md5_do_add(newsk, newinet->inet_daddr,
1494 newkey, key->keylen);
1495 sk_nocaps_add(newsk, NETIF_F_GSO_MASK);
1496 }
1497 #endif
1498
1499 if (__inet_inherit_port(sk, newsk) < 0)
1500 goto put_and_exit;
1501 __inet_hash_nolisten(newsk, NULL);
1502
1503 return newsk;
1504
1505 exit_overflow:
1506 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENOVERFLOWS);
1507 exit_nonewsk:
1508 dst_release(dst);
1509 exit:
1510 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENDROPS);
1511 return NULL;
1512 put_and_exit:
1513 tcp_clear_xmit_timers(newsk);
1514 tcp_cleanup_congestion_control(newsk);
1515 bh_unlock_sock(newsk);
1516 sock_put(newsk);
1517 goto exit;
1518 }
1519 EXPORT_SYMBOL(tcp_v4_syn_recv_sock);
1520
1521 static struct sock *tcp_v4_hnd_req(struct sock *sk, struct sk_buff *skb)
1522 {
1523 struct tcphdr *th = tcp_hdr(skb);
1524 const struct iphdr *iph = ip_hdr(skb);
1525 struct sock *nsk;
1526 struct request_sock **prev;
1527 /* Find possible connection requests. */
1528 struct request_sock *req = inet_csk_search_req(sk, &prev, th->source,
1529 iph->saddr, iph->daddr);
1530 if (req)
1531 return tcp_check_req(sk, skb, req, prev);
1532
1533 nsk = inet_lookup_established(sock_net(sk), &tcp_hashinfo, iph->saddr,
1534 th->source, iph->daddr, th->dest, inet_iif(skb));
1535
1536 if (nsk) {
1537 if (nsk->sk_state != TCP_TIME_WAIT) {
1538 bh_lock_sock(nsk);
1539 return nsk;
1540 }
1541 inet_twsk_put(inet_twsk(nsk));
1542 return NULL;
1543 }
1544
1545 #ifdef CONFIG_SYN_COOKIES
1546 if (!th->syn)
1547 sk = cookie_v4_check(sk, skb, &(IPCB(skb)->opt));
1548 #endif
1549 return sk;
1550 }
1551
1552 static __sum16 tcp_v4_checksum_init(struct sk_buff *skb)
1553 {
1554 const struct iphdr *iph = ip_hdr(skb);
1555
1556 if (skb->ip_summed == CHECKSUM_COMPLETE) {
1557 if (!tcp_v4_check(skb->len, iph->saddr,
1558 iph->daddr, skb->csum)) {
1559 skb->ip_summed = CHECKSUM_UNNECESSARY;
1560 return 0;
1561 }
1562 }
1563
1564 skb->csum = csum_tcpudp_nofold(iph->saddr, iph->daddr,
1565 skb->len, IPPROTO_TCP, 0);
1566
1567 if (skb->len <= 76) {
1568 return __skb_checksum_complete(skb);
1569 }
1570 return 0;
1571 }
1572
1573
1574 /* The socket must have it's spinlock held when we get
1575 * here.
1576 *
1577 * We have a potential double-lock case here, so even when
1578 * doing backlog processing we use the BH locking scheme.
1579 * This is because we cannot sleep with the original spinlock
1580 * held.
1581 */
1582 int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb)
1583 {
1584 struct sock *rsk;
1585 #ifdef CONFIG_TCP_MD5SIG
1586 /*
1587 * We really want to reject the packet as early as possible
1588 * if:
1589 * o We're expecting an MD5'd packet and this is no MD5 tcp option
1590 * o There is an MD5 option and we're not expecting one
1591 */
1592 if (tcp_v4_inbound_md5_hash(sk, skb))
1593 goto discard;
1594 #endif
1595
1596 if (sk->sk_state == TCP_ESTABLISHED) { /* Fast path */
1597 sock_rps_save_rxhash(sk, skb);
1598 if (tcp_rcv_established(sk, skb, tcp_hdr(skb), skb->len)) {
1599 rsk = sk;
1600 goto reset;
1601 }
1602 return 0;
1603 }
1604
1605 if (skb->len < tcp_hdrlen(skb) || tcp_checksum_complete(skb))
1606 goto csum_err;
1607
1608 if (sk->sk_state == TCP_LISTEN) {
1609 struct sock *nsk = tcp_v4_hnd_req(sk, skb);
1610 if (!nsk)
1611 goto discard;
1612
1613 if (nsk != sk) {
1614 sock_rps_save_rxhash(nsk, skb);
1615 if (tcp_child_process(sk, nsk, skb)) {
1616 rsk = nsk;
1617 goto reset;
1618 }
1619 return 0;
1620 }
1621 } else
1622 sock_rps_save_rxhash(sk, skb);
1623
1624 if (tcp_rcv_state_process(sk, skb, tcp_hdr(skb), skb->len)) {
1625 rsk = sk;
1626 goto reset;
1627 }
1628 return 0;
1629
1630 reset:
1631 tcp_v4_send_reset(rsk, skb);
1632 discard:
1633 kfree_skb(skb);
1634 /* Be careful here. If this function gets more complicated and
1635 * gcc suffers from register pressure on the x86, sk (in %ebx)
1636 * might be destroyed here. This current version compiles correctly,
1637 * but you have been warned.
1638 */
1639 return 0;
1640
1641 csum_err:
1642 TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_INERRS);
1643 goto discard;
1644 }
1645 EXPORT_SYMBOL(tcp_v4_do_rcv);
1646
1647 /*
1648 * From tcp_input.c
1649 */
1650
1651 int tcp_v4_rcv(struct sk_buff *skb)
1652 {
1653 const struct iphdr *iph;
1654 const struct tcphdr *th;
1655 struct sock *sk;
1656 int ret;
1657 struct net *net = dev_net(skb->dev);
1658
1659 if (skb->pkt_type != PACKET_HOST)
1660 goto discard_it;
1661
1662 /* Count it even if it's bad */
1663 TCP_INC_STATS_BH(net, TCP_MIB_INSEGS);
1664
1665 if (!pskb_may_pull(skb, sizeof(struct tcphdr)))
1666 goto discard_it;
1667
1668 th = tcp_hdr(skb);
1669
1670 if (th->doff < sizeof(struct tcphdr) / 4)
1671 goto bad_packet;
1672 if (!pskb_may_pull(skb, th->doff * 4))
1673 goto discard_it;
1674
1675 /* An explanation is required here, I think.
1676 * Packet length and doff are validated by header prediction,
1677 * provided case of th->doff==0 is eliminated.
1678 * So, we defer the checks. */
1679 if (!skb_csum_unnecessary(skb) && tcp_v4_checksum_init(skb))
1680 goto bad_packet;
1681
1682 th = tcp_hdr(skb);
1683 iph = ip_hdr(skb);
1684 TCP_SKB_CB(skb)->seq = ntohl(th->seq);
1685 TCP_SKB_CB(skb)->end_seq = (TCP_SKB_CB(skb)->seq + th->syn + th->fin +
1686 skb->len - th->doff * 4);
1687 TCP_SKB_CB(skb)->ack_seq = ntohl(th->ack_seq);
1688 TCP_SKB_CB(skb)->when = 0;
1689 TCP_SKB_CB(skb)->ip_dsfield = ipv4_get_dsfield(iph);
1690 TCP_SKB_CB(skb)->sacked = 0;
1691
1692 sk = __inet_lookup_skb(&tcp_hashinfo, skb, th->source, th->dest);
1693 if (!sk)
1694 goto no_tcp_socket;
1695
1696 process:
1697 if (sk->sk_state == TCP_TIME_WAIT)
1698 goto do_time_wait;
1699
1700 if (unlikely(iph->ttl < inet_sk(sk)->min_ttl)) {
1701 NET_INC_STATS_BH(net, LINUX_MIB_TCPMINTTLDROP);
1702 goto discard_and_relse;
1703 }
1704
1705 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
1706 goto discard_and_relse;
1707 nf_reset(skb);
1708
1709 if (sk_filter(sk, skb))
1710 goto discard_and_relse;
1711
1712 skb->dev = NULL;
1713
1714 bh_lock_sock_nested(sk);
1715 ret = 0;
1716 if (!sock_owned_by_user(sk)) {
1717 #ifdef CONFIG_NET_DMA
1718 struct tcp_sock *tp = tcp_sk(sk);
1719 if (!tp->ucopy.dma_chan && tp->ucopy.pinned_list)
1720 tp->ucopy.dma_chan = dma_find_channel(DMA_MEMCPY);
1721 if (tp->ucopy.dma_chan)
1722 ret = tcp_v4_do_rcv(sk, skb);
1723 else
1724 #endif
1725 {
1726 if (!tcp_prequeue(sk, skb))
1727 ret = tcp_v4_do_rcv(sk, skb);
1728 }
1729 } else if (unlikely(sk_add_backlog(sk, skb))) {
1730 bh_unlock_sock(sk);
1731 NET_INC_STATS_BH(net, LINUX_MIB_TCPBACKLOGDROP);
1732 goto discard_and_relse;
1733 }
1734 bh_unlock_sock(sk);
1735
1736 sock_put(sk);
1737
1738 return ret;
1739
1740 no_tcp_socket:
1741 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
1742 goto discard_it;
1743
1744 if (skb->len < (th->doff << 2) || tcp_checksum_complete(skb)) {
1745 bad_packet:
1746 TCP_INC_STATS_BH(net, TCP_MIB_INERRS);
1747 } else {
1748 tcp_v4_send_reset(NULL, skb);
1749 }
1750
1751 discard_it:
1752 /* Discard frame. */
1753 kfree_skb(skb);
1754 return 0;
1755
1756 discard_and_relse:
1757 sock_put(sk);
1758 goto discard_it;
1759
1760 do_time_wait:
1761 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) {
1762 inet_twsk_put(inet_twsk(sk));
1763 goto discard_it;
1764 }
1765
1766 if (skb->len < (th->doff << 2) || tcp_checksum_complete(skb)) {
1767 TCP_INC_STATS_BH(net, TCP_MIB_INERRS);
1768 inet_twsk_put(inet_twsk(sk));
1769 goto discard_it;
1770 }
1771 switch (tcp_timewait_state_process(inet_twsk(sk), skb, th)) {
1772 case TCP_TW_SYN: {
1773 struct sock *sk2 = inet_lookup_listener(dev_net(skb->dev),
1774 &tcp_hashinfo,
1775 iph->daddr, th->dest,
1776 inet_iif(skb));
1777 if (sk2) {
1778 inet_twsk_deschedule(inet_twsk(sk), &tcp_death_row);
1779 inet_twsk_put(inet_twsk(sk));
1780 sk = sk2;
1781 goto process;
1782 }
1783 /* Fall through to ACK */
1784 }
1785 case TCP_TW_ACK:
1786 tcp_v4_timewait_ack(sk, skb);
1787 break;
1788 case TCP_TW_RST:
1789 goto no_tcp_socket;
1790 case TCP_TW_SUCCESS:;
1791 }
1792 goto discard_it;
1793 }
1794
1795 struct inet_peer *tcp_v4_get_peer(struct sock *sk, bool *release_it)
1796 {
1797 struct rtable *rt = (struct rtable *) __sk_dst_get(sk);
1798 struct inet_sock *inet = inet_sk(sk);
1799 struct inet_peer *peer;
1800
1801 if (!rt ||
1802 inet->cork.fl.u.ip4.daddr != inet->inet_daddr) {
1803 peer = inet_getpeer_v4(inet->inet_daddr, 1);
1804 *release_it = true;
1805 } else {
1806 if (!rt->peer)
1807 rt_bind_peer(rt, inet->inet_daddr, 1);
1808 peer = rt->peer;
1809 *release_it = false;
1810 }
1811
1812 return peer;
1813 }
1814 EXPORT_SYMBOL(tcp_v4_get_peer);
1815
1816 void *tcp_v4_tw_get_peer(struct sock *sk)
1817 {
1818 const struct inet_timewait_sock *tw = inet_twsk(sk);
1819
1820 return inet_getpeer_v4(tw->tw_daddr, 1);
1821 }
1822 EXPORT_SYMBOL(tcp_v4_tw_get_peer);
1823
1824 static struct timewait_sock_ops tcp_timewait_sock_ops = {
1825 .twsk_obj_size = sizeof(struct tcp_timewait_sock),
1826 .twsk_unique = tcp_twsk_unique,
1827 .twsk_destructor= tcp_twsk_destructor,
1828 .twsk_getpeer = tcp_v4_tw_get_peer,
1829 };
1830
1831 const struct inet_connection_sock_af_ops ipv4_specific = {
1832 .queue_xmit = ip_queue_xmit,
1833 .send_check = tcp_v4_send_check,
1834 .rebuild_header = inet_sk_rebuild_header,
1835 .conn_request = tcp_v4_conn_request,
1836 .syn_recv_sock = tcp_v4_syn_recv_sock,
1837 .get_peer = tcp_v4_get_peer,
1838 .net_header_len = sizeof(struct iphdr),
1839 .setsockopt = ip_setsockopt,
1840 .getsockopt = ip_getsockopt,
1841 .addr2sockaddr = inet_csk_addr2sockaddr,
1842 .sockaddr_len = sizeof(struct sockaddr_in),
1843 .bind_conflict = inet_csk_bind_conflict,
1844 #ifdef CONFIG_COMPAT
1845 .compat_setsockopt = compat_ip_setsockopt,
1846 .compat_getsockopt = compat_ip_getsockopt,
1847 #endif
1848 };
1849 EXPORT_SYMBOL(ipv4_specific);
1850
1851 #ifdef CONFIG_TCP_MD5SIG
1852 static const struct tcp_sock_af_ops tcp_sock_ipv4_specific = {
1853 .md5_lookup = tcp_v4_md5_lookup,
1854 .calc_md5_hash = tcp_v4_md5_hash_skb,
1855 .md5_add = tcp_v4_md5_add_func,
1856 .md5_parse = tcp_v4_parse_md5_keys,
1857 };
1858 #endif
1859
1860 /* NOTE: A lot of things set to zero explicitly by call to
1861 * sk_alloc() so need not be done here.
1862 */
1863 static int tcp_v4_init_sock(struct sock *sk)
1864 {
1865 struct inet_connection_sock *icsk = inet_csk(sk);
1866 struct tcp_sock *tp = tcp_sk(sk);
1867
1868 skb_queue_head_init(&tp->out_of_order_queue);
1869 tcp_init_xmit_timers(sk);
1870 tcp_prequeue_init(tp);
1871
1872 icsk->icsk_rto = TCP_TIMEOUT_INIT;
1873 tp->mdev = TCP_TIMEOUT_INIT;
1874
1875 /* So many TCP implementations out there (incorrectly) count the
1876 * initial SYN frame in their delayed-ACK and congestion control
1877 * algorithms that we must have the following bandaid to talk
1878 * efficiently to them. -DaveM
1879 */
1880 tp->snd_cwnd = TCP_INIT_CWND;
1881
1882 /* See draft-stevens-tcpca-spec-01 for discussion of the
1883 * initialization of these values.
1884 */
1885 tp->snd_ssthresh = TCP_INFINITE_SSTHRESH;
1886 tp->snd_cwnd_clamp = ~0;
1887 tp->mss_cache = TCP_MSS_DEFAULT;
1888
1889 tp->reordering = sysctl_tcp_reordering;
1890 icsk->icsk_ca_ops = &tcp_init_congestion_ops;
1891
1892 sk->sk_state = TCP_CLOSE;
1893
1894 sk->sk_write_space = sk_stream_write_space;
1895 sock_set_flag(sk, SOCK_USE_WRITE_QUEUE);
1896
1897 icsk->icsk_af_ops = &ipv4_specific;
1898 icsk->icsk_sync_mss = tcp_sync_mss;
1899 #ifdef CONFIG_TCP_MD5SIG
1900 tp->af_specific = &tcp_sock_ipv4_specific;
1901 #endif
1902
1903 /* TCP Cookie Transactions */
1904 if (sysctl_tcp_cookie_size > 0) {
1905 /* Default, cookies without s_data_payload. */
1906 tp->cookie_values =
1907 kzalloc(sizeof(*tp->cookie_values),
1908 sk->sk_allocation);
1909 if (tp->cookie_values != NULL)
1910 kref_init(&tp->cookie_values->kref);
1911 }
1912 /* Presumed zeroed, in order of appearance:
1913 * cookie_in_always, cookie_out_never,
1914 * s_data_constant, s_data_in, s_data_out
1915 */
1916 sk->sk_sndbuf = sysctl_tcp_wmem[1];
1917 sk->sk_rcvbuf = sysctl_tcp_rmem[1];
1918
1919 local_bh_disable();
1920 sk_sockets_allocated_inc(sk);
1921 local_bh_enable();
1922
1923 return 0;
1924 }
1925
1926 void tcp_v4_destroy_sock(struct sock *sk)
1927 {
1928 struct tcp_sock *tp = tcp_sk(sk);
1929
1930 tcp_clear_xmit_timers(sk);
1931
1932 tcp_cleanup_congestion_control(sk);
1933
1934 /* Cleanup up the write buffer. */
1935 tcp_write_queue_purge(sk);
1936
1937 /* Cleans up our, hopefully empty, out_of_order_queue. */
1938 __skb_queue_purge(&tp->out_of_order_queue);
1939
1940 #ifdef CONFIG_TCP_MD5SIG
1941 /* Clean up the MD5 key list, if any */
1942 if (tp->md5sig_info) {
1943 tcp_v4_clear_md5_list(sk);
1944 kfree(tp->md5sig_info);
1945 tp->md5sig_info = NULL;
1946 }
1947 #endif
1948
1949 #ifdef CONFIG_NET_DMA
1950 /* Cleans up our sk_async_wait_queue */
1951 __skb_queue_purge(&sk->sk_async_wait_queue);
1952 #endif
1953
1954 /* Clean prequeue, it must be empty really */
1955 __skb_queue_purge(&tp->ucopy.prequeue);
1956
1957 /* Clean up a referenced TCP bind bucket. */
1958 if (inet_csk(sk)->icsk_bind_hash)
1959 inet_put_port(sk);
1960
1961 /*
1962 * If sendmsg cached page exists, toss it.
1963 */
1964 if (sk->sk_sndmsg_page) {
1965 __free_page(sk->sk_sndmsg_page);
1966 sk->sk_sndmsg_page = NULL;
1967 }
1968
1969 /* TCP Cookie Transactions */
1970 if (tp->cookie_values != NULL) {
1971 kref_put(&tp->cookie_values->kref,
1972 tcp_cookie_values_release);
1973 tp->cookie_values = NULL;
1974 }
1975
1976 sk_sockets_allocated_dec(sk);
1977 }
1978 EXPORT_SYMBOL(tcp_v4_destroy_sock);
1979
1980 #ifdef CONFIG_PROC_FS
1981 /* Proc filesystem TCP sock list dumping. */
1982
1983 static inline struct inet_timewait_sock *tw_head(struct hlist_nulls_head *head)
1984 {
1985 return hlist_nulls_empty(head) ? NULL :
1986 list_entry(head->first, struct inet_timewait_sock, tw_node);
1987 }
1988
1989 static inline struct inet_timewait_sock *tw_next(struct inet_timewait_sock *tw)
1990 {
1991 return !is_a_nulls(tw->tw_node.next) ?
1992 hlist_nulls_entry(tw->tw_node.next, typeof(*tw), tw_node) : NULL;
1993 }
1994
1995 /*
1996 * Get next listener socket follow cur. If cur is NULL, get first socket
1997 * starting from bucket given in st->bucket; when st->bucket is zero the
1998 * very first socket in the hash table is returned.
1999 */
2000 static void *listening_get_next(struct seq_file *seq, void *cur)
2001 {
2002 struct inet_connection_sock *icsk;
2003 struct hlist_nulls_node *node;
2004 struct sock *sk = cur;
2005 struct inet_listen_hashbucket *ilb;
2006 struct tcp_iter_state *st = seq->private;
2007 struct net *net = seq_file_net(seq);
2008
2009 if (!sk) {
2010 ilb = &tcp_hashinfo.listening_hash[st->bucket];
2011 spin_lock_bh(&ilb->lock);
2012 sk = sk_nulls_head(&ilb->head);
2013 st->offset = 0;
2014 goto get_sk;
2015 }
2016 ilb = &tcp_hashinfo.listening_hash[st->bucket];
2017 ++st->num;
2018 ++st->offset;
2019
2020 if (st->state == TCP_SEQ_STATE_OPENREQ) {
2021 struct request_sock *req = cur;
2022
2023 icsk = inet_csk(st->syn_wait_sk);
2024 req = req->dl_next;
2025 while (1) {
2026 while (req) {
2027 if (req->rsk_ops->family == st->family) {
2028 cur = req;
2029 goto out;
2030 }
2031 req = req->dl_next;
2032 }
2033 if (++st->sbucket >= icsk->icsk_accept_queue.listen_opt->nr_table_entries)
2034 break;
2035 get_req:
2036 req = icsk->icsk_accept_queue.listen_opt->syn_table[st->sbucket];
2037 }
2038 sk = sk_nulls_next(st->syn_wait_sk);
2039 st->state = TCP_SEQ_STATE_LISTENING;
2040 read_unlock_bh(&icsk->icsk_accept_queue.syn_wait_lock);
2041 } else {
2042 icsk = inet_csk(sk);
2043 read_lock_bh(&icsk->icsk_accept_queue.syn_wait_lock);
2044 if (reqsk_queue_len(&icsk->icsk_accept_queue))
2045 goto start_req;
2046 read_unlock_bh(&icsk->icsk_accept_queue.syn_wait_lock);
2047 sk = sk_nulls_next(sk);
2048 }
2049 get_sk:
2050 sk_nulls_for_each_from(sk, node) {
2051 if (!net_eq(sock_net(sk), net))
2052 continue;
2053 if (sk->sk_family == st->family) {
2054 cur = sk;
2055 goto out;
2056 }
2057 icsk = inet_csk(sk);
2058 read_lock_bh(&icsk->icsk_accept_queue.syn_wait_lock);
2059 if (reqsk_queue_len(&icsk->icsk_accept_queue)) {
2060 start_req:
2061 st->uid = sock_i_uid(sk);
2062 st->syn_wait_sk = sk;
2063 st->state = TCP_SEQ_STATE_OPENREQ;
2064 st->sbucket = 0;
2065 goto get_req;
2066 }
2067 read_unlock_bh(&icsk->icsk_accept_queue.syn_wait_lock);
2068 }
2069 spin_unlock_bh(&ilb->lock);
2070 st->offset = 0;
2071 if (++st->bucket < INET_LHTABLE_SIZE) {
2072 ilb = &tcp_hashinfo.listening_hash[st->bucket];
2073 spin_lock_bh(&ilb->lock);
2074 sk = sk_nulls_head(&ilb->head);
2075 goto get_sk;
2076 }
2077 cur = NULL;
2078 out:
2079 return cur;
2080 }
2081
2082 static void *listening_get_idx(struct seq_file *seq, loff_t *pos)
2083 {
2084 struct tcp_iter_state *st = seq->private;
2085 void *rc;
2086
2087 st->bucket = 0;
2088 st->offset = 0;
2089 rc = listening_get_next(seq, NULL);
2090
2091 while (rc && *pos) {
2092 rc = listening_get_next(seq, rc);
2093 --*pos;
2094 }
2095 return rc;
2096 }
2097
2098 static inline int empty_bucket(struct tcp_iter_state *st)
2099 {
2100 return hlist_nulls_empty(&tcp_hashinfo.ehash[st->bucket].chain) &&
2101 hlist_nulls_empty(&tcp_hashinfo.ehash[st->bucket].twchain);
2102 }
2103
2104 /*
2105 * Get first established socket starting from bucket given in st->bucket.
2106 * If st->bucket is zero, the very first socket in the hash is returned.
2107 */
2108 static void *established_get_first(struct seq_file *seq)
2109 {
2110 struct tcp_iter_state *st = seq->private;
2111 struct net *net = seq_file_net(seq);
2112 void *rc = NULL;
2113
2114 st->offset = 0;
2115 for (; st->bucket <= tcp_hashinfo.ehash_mask; ++st->bucket) {
2116 struct sock *sk;
2117 struct hlist_nulls_node *node;
2118 struct inet_timewait_sock *tw;
2119 spinlock_t *lock = inet_ehash_lockp(&tcp_hashinfo, st->bucket);
2120
2121 /* Lockless fast path for the common case of empty buckets */
2122 if (empty_bucket(st))
2123 continue;
2124
2125 spin_lock_bh(lock);
2126 sk_nulls_for_each(sk, node, &tcp_hashinfo.ehash[st->bucket].chain) {
2127 if (sk->sk_family != st->family ||
2128 !net_eq(sock_net(sk), net)) {
2129 continue;
2130 }
2131 rc = sk;
2132 goto out;
2133 }
2134 st->state = TCP_SEQ_STATE_TIME_WAIT;
2135 inet_twsk_for_each(tw, node,
2136 &tcp_hashinfo.ehash[st->bucket].twchain) {
2137 if (tw->tw_family != st->family ||
2138 !net_eq(twsk_net(tw), net)) {
2139 continue;
2140 }
2141 rc = tw;
2142 goto out;
2143 }
2144 spin_unlock_bh(lock);
2145 st->state = TCP_SEQ_STATE_ESTABLISHED;
2146 }
2147 out:
2148 return rc;
2149 }
2150
2151 static void *established_get_next(struct seq_file *seq, void *cur)
2152 {
2153 struct sock *sk = cur;
2154 struct inet_timewait_sock *tw;
2155 struct hlist_nulls_node *node;
2156 struct tcp_iter_state *st = seq->private;
2157 struct net *net = seq_file_net(seq);
2158
2159 ++st->num;
2160 ++st->offset;
2161
2162 if (st->state == TCP_SEQ_STATE_TIME_WAIT) {
2163 tw = cur;
2164 tw = tw_next(tw);
2165 get_tw:
2166 while (tw && (tw->tw_family != st->family || !net_eq(twsk_net(tw), net))) {
2167 tw = tw_next(tw);
2168 }
2169 if (tw) {
2170 cur = tw;
2171 goto out;
2172 }
2173 spin_unlock_bh(inet_ehash_lockp(&tcp_hashinfo, st->bucket));
2174 st->state = TCP_SEQ_STATE_ESTABLISHED;
2175
2176 /* Look for next non empty bucket */
2177 st->offset = 0;
2178 while (++st->bucket <= tcp_hashinfo.ehash_mask &&
2179 empty_bucket(st))
2180 ;
2181 if (st->bucket > tcp_hashinfo.ehash_mask)
2182 return NULL;
2183
2184 spin_lock_bh(inet_ehash_lockp(&tcp_hashinfo, st->bucket));
2185 sk = sk_nulls_head(&tcp_hashinfo.ehash[st->bucket].chain);
2186 } else
2187 sk = sk_nulls_next(sk);
2188
2189 sk_nulls_for_each_from(sk, node) {
2190 if (sk->sk_family == st->family && net_eq(sock_net(sk), net))
2191 goto found;
2192 }
2193
2194 st->state = TCP_SEQ_STATE_TIME_WAIT;
2195 tw = tw_head(&tcp_hashinfo.ehash[st->bucket].twchain);
2196 goto get_tw;
2197 found:
2198 cur = sk;
2199 out:
2200 return cur;
2201 }
2202
2203 static void *established_get_idx(struct seq_file *seq, loff_t pos)
2204 {
2205 struct tcp_iter_state *st = seq->private;
2206 void *rc;
2207
2208 st->bucket = 0;
2209 rc = established_get_first(seq);
2210
2211 while (rc && pos) {
2212 rc = established_get_next(seq, rc);
2213 --pos;
2214 }
2215 return rc;
2216 }
2217
2218 static void *tcp_get_idx(struct seq_file *seq, loff_t pos)
2219 {
2220 void *rc;
2221 struct tcp_iter_state *st = seq->private;
2222
2223 st->state = TCP_SEQ_STATE_LISTENING;
2224 rc = listening_get_idx(seq, &pos);
2225
2226 if (!rc) {
2227 st->state = TCP_SEQ_STATE_ESTABLISHED;
2228 rc = established_get_idx(seq, pos);
2229 }
2230
2231 return rc;
2232 }
2233
2234 static void *tcp_seek_last_pos(struct seq_file *seq)
2235 {
2236 struct tcp_iter_state *st = seq->private;
2237 int offset = st->offset;
2238 int orig_num = st->num;
2239 void *rc = NULL;
2240
2241 switch (st->state) {
2242 case TCP_SEQ_STATE_OPENREQ:
2243 case TCP_SEQ_STATE_LISTENING:
2244 if (st->bucket >= INET_LHTABLE_SIZE)
2245 break;
2246 st->state = TCP_SEQ_STATE_LISTENING;
2247 rc = listening_get_next(seq, NULL);
2248 while (offset-- && rc)
2249 rc = listening_get_next(seq, rc);
2250 if (rc)
2251 break;
2252 st->bucket = 0;
2253 /* Fallthrough */
2254 case TCP_SEQ_STATE_ESTABLISHED:
2255 case TCP_SEQ_STATE_TIME_WAIT:
2256 st->state = TCP_SEQ_STATE_ESTABLISHED;
2257 if (st->bucket > tcp_hashinfo.ehash_mask)
2258 break;
2259 rc = established_get_first(seq);
2260 while (offset-- && rc)
2261 rc = established_get_next(seq, rc);
2262 }
2263
2264 st->num = orig_num;
2265
2266 return rc;
2267 }
2268
2269 static void *tcp_seq_start(struct seq_file *seq, loff_t *pos)
2270 {
2271 struct tcp_iter_state *st = seq->private;
2272 void *rc;
2273
2274 if (*pos && *pos == st->last_pos) {
2275 rc = tcp_seek_last_pos(seq);
2276 if (rc)
2277 goto out;
2278 }
2279
2280 st->state = TCP_SEQ_STATE_LISTENING;
2281 st->num = 0;
2282 st->bucket = 0;
2283 st->offset = 0;
2284 rc = *pos ? tcp_get_idx(seq, *pos - 1) : SEQ_START_TOKEN;
2285
2286 out:
2287 st->last_pos = *pos;
2288 return rc;
2289 }
2290
2291 static void *tcp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2292 {
2293 struct tcp_iter_state *st = seq->private;
2294 void *rc = NULL;
2295
2296 if (v == SEQ_START_TOKEN) {
2297 rc = tcp_get_idx(seq, 0);
2298 goto out;
2299 }
2300
2301 switch (st->state) {
2302 case TCP_SEQ_STATE_OPENREQ:
2303 case TCP_SEQ_STATE_LISTENING:
2304 rc = listening_get_next(seq, v);
2305 if (!rc) {
2306 st->state = TCP_SEQ_STATE_ESTABLISHED;
2307 st->bucket = 0;
2308 st->offset = 0;
2309 rc = established_get_first(seq);
2310 }
2311 break;
2312 case TCP_SEQ_STATE_ESTABLISHED:
2313 case TCP_SEQ_STATE_TIME_WAIT:
2314 rc = established_get_next(seq, v);
2315 break;
2316 }
2317 out:
2318 ++*pos;
2319 st->last_pos = *pos;
2320 return rc;
2321 }
2322
2323 static void tcp_seq_stop(struct seq_file *seq, void *v)
2324 {
2325 struct tcp_iter_state *st = seq->private;
2326
2327 switch (st->state) {
2328 case TCP_SEQ_STATE_OPENREQ:
2329 if (v) {
2330 struct inet_connection_sock *icsk = inet_csk(st->syn_wait_sk);
2331 read_unlock_bh(&icsk->icsk_accept_queue.syn_wait_lock);
2332 }
2333 case TCP_SEQ_STATE_LISTENING:
2334 if (v != SEQ_START_TOKEN)
2335 spin_unlock_bh(&tcp_hashinfo.listening_hash[st->bucket].lock);
2336 break;
2337 case TCP_SEQ_STATE_TIME_WAIT:
2338 case TCP_SEQ_STATE_ESTABLISHED:
2339 if (v)
2340 spin_unlock_bh(inet_ehash_lockp(&tcp_hashinfo, st->bucket));
2341 break;
2342 }
2343 }
2344
2345 int tcp_seq_open(struct inode *inode, struct file *file)
2346 {
2347 struct tcp_seq_afinfo *afinfo = PDE(inode)->data;
2348 struct tcp_iter_state *s;
2349 int err;
2350
2351 err = seq_open_net(inode, file, &afinfo->seq_ops,
2352 sizeof(struct tcp_iter_state));
2353 if (err < 0)
2354 return err;
2355
2356 s = ((struct seq_file *)file->private_data)->private;
2357 s->family = afinfo->family;
2358 s->last_pos = 0;
2359 return 0;
2360 }
2361 EXPORT_SYMBOL(tcp_seq_open);
2362
2363 int tcp_proc_register(struct net *net, struct tcp_seq_afinfo *afinfo)
2364 {
2365 int rc = 0;
2366 struct proc_dir_entry *p;
2367
2368 afinfo->seq_ops.start = tcp_seq_start;
2369 afinfo->seq_ops.next = tcp_seq_next;
2370 afinfo->seq_ops.stop = tcp_seq_stop;
2371
2372 p = proc_create_data(afinfo->name, S_IRUGO, net->proc_net,
2373 afinfo->seq_fops, afinfo);
2374 if (!p)
2375 rc = -ENOMEM;
2376 return rc;
2377 }
2378 EXPORT_SYMBOL(tcp_proc_register);
2379
2380 void tcp_proc_unregister(struct net *net, struct tcp_seq_afinfo *afinfo)
2381 {
2382 proc_net_remove(net, afinfo->name);
2383 }
2384 EXPORT_SYMBOL(tcp_proc_unregister);
2385
2386 static void get_openreq4(const struct sock *sk, const struct request_sock *req,
2387 struct seq_file *f, int i, int uid, int *len)
2388 {
2389 const struct inet_request_sock *ireq = inet_rsk(req);
2390 int ttd = req->expires - jiffies;
2391
2392 seq_printf(f, "%4d: %08X:%04X %08X:%04X"
2393 " %02X %08X:%08X %02X:%08lX %08X %5d %8d %u %d %pK%n",
2394 i,
2395 ireq->loc_addr,
2396 ntohs(inet_sk(sk)->inet_sport),
2397 ireq->rmt_addr,
2398 ntohs(ireq->rmt_port),
2399 TCP_SYN_RECV,
2400 0, 0, /* could print option size, but that is af dependent. */
2401 1, /* timers active (only the expire timer) */
2402 jiffies_to_clock_t(ttd),
2403 req->retrans,
2404 uid,
2405 0, /* non standard timer */
2406 0, /* open_requests have no inode */
2407 atomic_read(&sk->sk_refcnt),
2408 req,
2409 len);
2410 }
2411
2412 static void get_tcp4_sock(struct sock *sk, struct seq_file *f, int i, int *len)
2413 {
2414 int timer_active;
2415 unsigned long timer_expires;
2416 const struct tcp_sock *tp = tcp_sk(sk);
2417 const struct inet_connection_sock *icsk = inet_csk(sk);
2418 const struct inet_sock *inet = inet_sk(sk);
2419 __be32 dest = inet->inet_daddr;
2420 __be32 src = inet->inet_rcv_saddr;
2421 __u16 destp = ntohs(inet->inet_dport);
2422 __u16 srcp = ntohs(inet->inet_sport);
2423 int rx_queue;
2424
2425 if (icsk->icsk_pending == ICSK_TIME_RETRANS) {
2426 timer_active = 1;
2427 timer_expires = icsk->icsk_timeout;
2428 } else if (icsk->icsk_pending == ICSK_TIME_PROBE0) {
2429 timer_active = 4;
2430 timer_expires = icsk->icsk_timeout;
2431 } else if (timer_pending(&sk->sk_timer)) {
2432 timer_active = 2;
2433 timer_expires = sk->sk_timer.expires;
2434 } else {
2435 timer_active = 0;
2436 timer_expires = jiffies;
2437 }
2438
2439 if (sk->sk_state == TCP_LISTEN)
2440 rx_queue = sk->sk_ack_backlog;
2441 else
2442 /*
2443 * because we dont lock socket, we might find a transient negative value
2444 */
2445 rx_queue = max_t(int, tp->rcv_nxt - tp->copied_seq, 0);
2446
2447 seq_printf(f, "%4d: %08X:%04X %08X:%04X %02X %08X:%08X %02X:%08lX "
2448 "%08X %5d %8d %lu %d %pK %lu %lu %u %u %d%n",
2449 i, src, srcp, dest, destp, sk->sk_state,
2450 tp->write_seq - tp->snd_una,
2451 rx_queue,
2452 timer_active,
2453 jiffies_to_clock_t(timer_expires - jiffies),
2454 icsk->icsk_retransmits,
2455 sock_i_uid(sk),
2456 icsk->icsk_probes_out,
2457 sock_i_ino(sk),
2458 atomic_read(&sk->sk_refcnt), sk,
2459 jiffies_to_clock_t(icsk->icsk_rto),
2460 jiffies_to_clock_t(icsk->icsk_ack.ato),
2461 (icsk->icsk_ack.quick << 1) | icsk->icsk_ack.pingpong,
2462 tp->snd_cwnd,
2463 tcp_in_initial_slowstart(tp) ? -1 : tp->snd_ssthresh,
2464 len);
2465 }
2466
2467 static void get_timewait4_sock(const struct inet_timewait_sock *tw,
2468 struct seq_file *f, int i, int *len)
2469 {
2470 __be32 dest, src;
2471 __u16 destp, srcp;
2472 int ttd = tw->tw_ttd - jiffies;
2473
2474 if (ttd < 0)
2475 ttd = 0;
2476
2477 dest = tw->tw_daddr;
2478 src = tw->tw_rcv_saddr;
2479 destp = ntohs(tw->tw_dport);
2480 srcp = ntohs(tw->tw_sport);
2481
2482 seq_printf(f, "%4d: %08X:%04X %08X:%04X"
2483 " %02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %pK%n",
2484 i, src, srcp, dest, destp, tw->tw_substate, 0, 0,
2485 3, jiffies_to_clock_t(ttd), 0, 0, 0, 0,
2486 atomic_read(&tw->tw_refcnt), tw, len);
2487 }
2488
2489 #define TMPSZ 150
2490
2491 static int tcp4_seq_show(struct seq_file *seq, void *v)
2492 {
2493 struct tcp_iter_state *st;
2494 int len;
2495
2496 if (v == SEQ_START_TOKEN) {
2497 seq_printf(seq, "%-*s\n", TMPSZ - 1,
2498 " sl local_address rem_address st tx_queue "
2499 "rx_queue tr tm->when retrnsmt uid timeout "
2500 "inode");
2501 goto out;
2502 }
2503 st = seq->private;
2504
2505 switch (st->state) {
2506 case TCP_SEQ_STATE_LISTENING:
2507 case TCP_SEQ_STATE_ESTABLISHED:
2508 get_tcp4_sock(v, seq, st->num, &len);
2509 break;
2510 case TCP_SEQ_STATE_OPENREQ:
2511 get_openreq4(st->syn_wait_sk, v, seq, st->num, st->uid, &len);
2512 break;
2513 case TCP_SEQ_STATE_TIME_WAIT:
2514 get_timewait4_sock(v, seq, st->num, &len);
2515 break;
2516 }
2517 seq_printf(seq, "%*s\n", TMPSZ - 1 - len, "");
2518 out:
2519 return 0;
2520 }
2521
2522 static const struct file_operations tcp_afinfo_seq_fops = {
2523 .owner = THIS_MODULE,
2524 .open = tcp_seq_open,
2525 .read = seq_read,
2526 .llseek = seq_lseek,
2527 .release = seq_release_net
2528 };
2529
2530 static struct tcp_seq_afinfo tcp4_seq_afinfo = {
2531 .name = "tcp",
2532 .family = AF_INET,
2533 .seq_fops = &tcp_afinfo_seq_fops,
2534 .seq_ops = {
2535 .show = tcp4_seq_show,
2536 },
2537 };
2538
2539 static int __net_init tcp4_proc_init_net(struct net *net)
2540 {
2541 return tcp_proc_register(net, &tcp4_seq_afinfo);
2542 }
2543
2544 static void __net_exit tcp4_proc_exit_net(struct net *net)
2545 {
2546 tcp_proc_unregister(net, &tcp4_seq_afinfo);
2547 }
2548
2549 static struct pernet_operations tcp4_net_ops = {
2550 .init = tcp4_proc_init_net,
2551 .exit = tcp4_proc_exit_net,
2552 };
2553
2554 int __init tcp4_proc_init(void)
2555 {
2556 return register_pernet_subsys(&tcp4_net_ops);
2557 }
2558
2559 void tcp4_proc_exit(void)
2560 {
2561 unregister_pernet_subsys(&tcp4_net_ops);
2562 }
2563 #endif /* CONFIG_PROC_FS */
2564
2565 struct sk_buff **tcp4_gro_receive(struct sk_buff **head, struct sk_buff *skb)
2566 {
2567 const struct iphdr *iph = skb_gro_network_header(skb);
2568
2569 switch (skb->ip_summed) {
2570 case CHECKSUM_COMPLETE:
2571 if (!tcp_v4_check(skb_gro_len(skb), iph->saddr, iph->daddr,
2572 skb->csum)) {
2573 skb->ip_summed = CHECKSUM_UNNECESSARY;
2574 break;
2575 }
2576
2577 /* fall through */
2578 case CHECKSUM_NONE:
2579 NAPI_GRO_CB(skb)->flush = 1;
2580 return NULL;
2581 }
2582
2583 return tcp_gro_receive(head, skb);
2584 }
2585
2586 int tcp4_gro_complete(struct sk_buff *skb)
2587 {
2588 const struct iphdr *iph = ip_hdr(skb);
2589 struct tcphdr *th = tcp_hdr(skb);
2590
2591 th->check = ~tcp_v4_check(skb->len - skb_transport_offset(skb),
2592 iph->saddr, iph->daddr, 0);
2593 skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
2594
2595 return tcp_gro_complete(skb);
2596 }
2597
2598 struct proto tcp_prot = {
2599 .name = "TCP",
2600 .owner = THIS_MODULE,
2601 .close = tcp_close,
2602 .connect = tcp_v4_connect,
2603 .disconnect = tcp_disconnect,
2604 .accept = inet_csk_accept,
2605 .ioctl = tcp_ioctl,
2606 .init = tcp_v4_init_sock,
2607 .destroy = tcp_v4_destroy_sock,
2608 .shutdown = tcp_shutdown,
2609 .setsockopt = tcp_setsockopt,
2610 .getsockopt = tcp_getsockopt,
2611 .recvmsg = tcp_recvmsg,
2612 .sendmsg = tcp_sendmsg,
2613 .sendpage = tcp_sendpage,
2614 .backlog_rcv = tcp_v4_do_rcv,
2615 .hash = inet_hash,
2616 .unhash = inet_unhash,
2617 .get_port = inet_csk_get_port,
2618 .enter_memory_pressure = tcp_enter_memory_pressure,
2619 .sockets_allocated = &tcp_sockets_allocated,
2620 .orphan_count = &tcp_orphan_count,
2621 .memory_allocated = &tcp_memory_allocated,
2622 .memory_pressure = &tcp_memory_pressure,
2623 .sysctl_mem = sysctl_tcp_mem,
2624 .sysctl_wmem = sysctl_tcp_wmem,
2625 .sysctl_rmem = sysctl_tcp_rmem,
2626 .max_header = MAX_TCP_HEADER,
2627 .obj_size = sizeof(struct tcp_sock),
2628 .slab_flags = SLAB_DESTROY_BY_RCU,
2629 .twsk_prot = &tcp_timewait_sock_ops,
2630 .rsk_prot = &tcp_request_sock_ops,
2631 .h.hashinfo = &tcp_hashinfo,
2632 .no_autobind = true,
2633 #ifdef CONFIG_COMPAT
2634 .compat_setsockopt = compat_tcp_setsockopt,
2635 .compat_getsockopt = compat_tcp_getsockopt,
2636 #endif
2637 };
2638 EXPORT_SYMBOL(tcp_prot);
2639
2640
2641 static int __net_init tcp_sk_init(struct net *net)
2642 {
2643 return inet_ctl_sock_create(&net->ipv4.tcp_sock,
2644 PF_INET, SOCK_RAW, IPPROTO_TCP, net);
2645 }
2646
2647 static void __net_exit tcp_sk_exit(struct net *net)
2648 {
2649 inet_ctl_sock_destroy(net->ipv4.tcp_sock);
2650 }
2651
2652 static void __net_exit tcp_sk_exit_batch(struct list_head *net_exit_list)
2653 {
2654 inet_twsk_purge(&tcp_hashinfo, &tcp_death_row, AF_INET);
2655 }
2656
2657 static struct pernet_operations __net_initdata tcp_sk_ops = {
2658 .init = tcp_sk_init,
2659 .exit = tcp_sk_exit,
2660 .exit_batch = tcp_sk_exit_batch,
2661 };
2662
2663 void __init tcp_v4_init(void)
2664 {
2665 inet_hashinfo_init(&tcp_hashinfo);
2666 if (register_pernet_subsys(&tcp_sk_ops))
2667 panic("Failed to create the TCP control socket.\n");
2668 }
Linux 2.6 libata-dev (mirror)