?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Merge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/dtor/input
[linux-btrfs-devel.git] / net / ipv4 / tcp_ipv4.c
blob0ea10eefa60f005911b0849dd14218a1eb9ccf1d
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 sock_put(newsk);
1514 goto exit;
1515 }
1516 EXPORT_SYMBOL(tcp_v4_syn_recv_sock);
1517
1518 static struct sock *tcp_v4_hnd_req(struct sock *sk, struct sk_buff *skb)
1519 {
1520 struct tcphdr *th = tcp_hdr(skb);
1521 const struct iphdr *iph = ip_hdr(skb);
1522 struct sock *nsk;
1523 struct request_sock **prev;
1524 /* Find possible connection requests. */
1525 struct request_sock *req = inet_csk_search_req(sk, &prev, th->source,
1526 iph->saddr, iph->daddr);
1527 if (req)
1528 return tcp_check_req(sk, skb, req, prev);
1529
1530 nsk = inet_lookup_established(sock_net(sk), &tcp_hashinfo, iph->saddr,
1531 th->source, iph->daddr, th->dest, inet_iif(skb));
1532
1533 if (nsk) {
1534 if (nsk->sk_state != TCP_TIME_WAIT) {
1535 bh_lock_sock(nsk);
1536 return nsk;
1537 }
1538 inet_twsk_put(inet_twsk(nsk));
1539 return NULL;
1540 }
1541
1542 #ifdef CONFIG_SYN_COOKIES
1543 if (!th->syn)
1544 sk = cookie_v4_check(sk, skb, &(IPCB(skb)->opt));
1545 #endif
1546 return sk;
1547 }
1548
1549 static __sum16 tcp_v4_checksum_init(struct sk_buff *skb)
1550 {
1551 const struct iphdr *iph = ip_hdr(skb);
1552
1553 if (skb->ip_summed == CHECKSUM_COMPLETE) {
1554 if (!tcp_v4_check(skb->len, iph->saddr,
1555 iph->daddr, skb->csum)) {
1556 skb->ip_summed = CHECKSUM_UNNECESSARY;
1557 return 0;
1558 }
1559 }
1560
1561 skb->csum = csum_tcpudp_nofold(iph->saddr, iph->daddr,
1562 skb->len, IPPROTO_TCP, 0);
1563
1564 if (skb->len <= 76) {
1565 return __skb_checksum_complete(skb);
1566 }
1567 return 0;
1568 }
1569
1570
1571 /* The socket must have it's spinlock held when we get
1572 * here.
1573 *
1574 * We have a potential double-lock case here, so even when
1575 * doing backlog processing we use the BH locking scheme.
1576 * This is because we cannot sleep with the original spinlock
1577 * held.
1578 */
1579 int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb)
1580 {
1581 struct sock *rsk;
1582 #ifdef CONFIG_TCP_MD5SIG
1583 /*
1584 * We really want to reject the packet as early as possible
1585 * if:
1586 * o We're expecting an MD5'd packet and this is no MD5 tcp option
1587 * o There is an MD5 option and we're not expecting one
1588 */
1589 if (tcp_v4_inbound_md5_hash(sk, skb))
1590 goto discard;
1591 #endif
1592
1593 if (sk->sk_state == TCP_ESTABLISHED) { /* Fast path */
1594 sock_rps_save_rxhash(sk, skb);
1595 if (tcp_rcv_established(sk, skb, tcp_hdr(skb), skb->len)) {
1596 rsk = sk;
1597 goto reset;
1598 }
1599 return 0;
1600 }
1601
1602 if (skb->len < tcp_hdrlen(skb) || tcp_checksum_complete(skb))
1603 goto csum_err;
1604
1605 if (sk->sk_state == TCP_LISTEN) {
1606 struct sock *nsk = tcp_v4_hnd_req(sk, skb);
1607 if (!nsk)
1608 goto discard;
1609
1610 if (nsk != sk) {
1611 sock_rps_save_rxhash(nsk, skb);
1612 if (tcp_child_process(sk, nsk, skb)) {
1613 rsk = nsk;
1614 goto reset;
1615 }
1616 return 0;
1617 }
1618 } else
1619 sock_rps_save_rxhash(sk, skb);
1620
1621 if (tcp_rcv_state_process(sk, skb, tcp_hdr(skb), skb->len)) {
1622 rsk = sk;
1623 goto reset;
1624 }
1625 return 0;
1626
1627 reset:
1628 tcp_v4_send_reset(rsk, skb);
1629 discard:
1630 kfree_skb(skb);
1631 /* Be careful here. If this function gets more complicated and
1632 * gcc suffers from register pressure on the x86, sk (in %ebx)
1633 * might be destroyed here. This current version compiles correctly,
1634 * but you have been warned.
1635 */
1636 return 0;
1637
1638 csum_err:
1639 TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_INERRS);
1640 goto discard;
1641 }
1642 EXPORT_SYMBOL(tcp_v4_do_rcv);
1643
1644 /*
1645 * From tcp_input.c
1646 */
1647
1648 int tcp_v4_rcv(struct sk_buff *skb)
1649 {
1650 const struct iphdr *iph;
1651 const struct tcphdr *th;
1652 struct sock *sk;
1653 int ret;
1654 struct net *net = dev_net(skb->dev);
1655
1656 if (skb->pkt_type != PACKET_HOST)
1657 goto discard_it;
1658
1659 /* Count it even if it's bad */
1660 TCP_INC_STATS_BH(net, TCP_MIB_INSEGS);
1661
1662 if (!pskb_may_pull(skb, sizeof(struct tcphdr)))
1663 goto discard_it;
1664
1665 th = tcp_hdr(skb);
1666
1667 if (th->doff < sizeof(struct tcphdr) / 4)
1668 goto bad_packet;
1669 if (!pskb_may_pull(skb, th->doff * 4))
1670 goto discard_it;
1671
1672 /* An explanation is required here, I think.
1673 * Packet length and doff are validated by header prediction,
1674 * provided case of th->doff==0 is eliminated.
1675 * So, we defer the checks. */
1676 if (!skb_csum_unnecessary(skb) && tcp_v4_checksum_init(skb))
1677 goto bad_packet;
1678
1679 th = tcp_hdr(skb);
1680 iph = ip_hdr(skb);
1681 TCP_SKB_CB(skb)->seq = ntohl(th->seq);
1682 TCP_SKB_CB(skb)->end_seq = (TCP_SKB_CB(skb)->seq + th->syn + th->fin +
1683 skb->len - th->doff * 4);
1684 TCP_SKB_CB(skb)->ack_seq = ntohl(th->ack_seq);
1685 TCP_SKB_CB(skb)->when = 0;
1686 TCP_SKB_CB(skb)->ip_dsfield = ipv4_get_dsfield(iph);
1687 TCP_SKB_CB(skb)->sacked = 0;
1688
1689 sk = __inet_lookup_skb(&tcp_hashinfo, skb, th->source, th->dest);
1690 if (!sk)
1691 goto no_tcp_socket;
1692
1693 process:
1694 if (sk->sk_state == TCP_TIME_WAIT)
1695 goto do_time_wait;
1696
1697 if (unlikely(iph->ttl < inet_sk(sk)->min_ttl)) {
1698 NET_INC_STATS_BH(net, LINUX_MIB_TCPMINTTLDROP);
1699 goto discard_and_relse;
1700 }
1701
1702 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
1703 goto discard_and_relse;
1704 nf_reset(skb);
1705
1706 if (sk_filter(sk, skb))
1707 goto discard_and_relse;
1708
1709 skb->dev = NULL;
1710
1711 bh_lock_sock_nested(sk);
1712 ret = 0;
1713 if (!sock_owned_by_user(sk)) {
1714 #ifdef CONFIG_NET_DMA
1715 struct tcp_sock *tp = tcp_sk(sk);
1716 if (!tp->ucopy.dma_chan && tp->ucopy.pinned_list)
1717 tp->ucopy.dma_chan = dma_find_channel(DMA_MEMCPY);
1718 if (tp->ucopy.dma_chan)
1719 ret = tcp_v4_do_rcv(sk, skb);
1720 else
1721 #endif
1722 {
1723 if (!tcp_prequeue(sk, skb))
1724 ret = tcp_v4_do_rcv(sk, skb);
1725 }
1726 } else if (unlikely(sk_add_backlog(sk, skb))) {
1727 bh_unlock_sock(sk);
1728 NET_INC_STATS_BH(net, LINUX_MIB_TCPBACKLOGDROP);
1729 goto discard_and_relse;
1730 }
1731 bh_unlock_sock(sk);
1732
1733 sock_put(sk);
1734
1735 return ret;
1736
1737 no_tcp_socket:
1738 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
1739 goto discard_it;
1740
1741 if (skb->len < (th->doff << 2) || tcp_checksum_complete(skb)) {
1742 bad_packet:
1743 TCP_INC_STATS_BH(net, TCP_MIB_INERRS);
1744 } else {
1745 tcp_v4_send_reset(NULL, skb);
1746 }
1747
1748 discard_it:
1749 /* Discard frame. */
1750 kfree_skb(skb);
1751 return 0;
1752
1753 discard_and_relse:
1754 sock_put(sk);
1755 goto discard_it;
1756
1757 do_time_wait:
1758 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) {
1759 inet_twsk_put(inet_twsk(sk));
1760 goto discard_it;
1761 }
1762
1763 if (skb->len < (th->doff << 2) || tcp_checksum_complete(skb)) {
1764 TCP_INC_STATS_BH(net, TCP_MIB_INERRS);
1765 inet_twsk_put(inet_twsk(sk));
1766 goto discard_it;
1767 }
1768 switch (tcp_timewait_state_process(inet_twsk(sk), skb, th)) {
1769 case TCP_TW_SYN: {
1770 struct sock *sk2 = inet_lookup_listener(dev_net(skb->dev),
1771 &tcp_hashinfo,
1772 iph->daddr, th->dest,
1773 inet_iif(skb));
1774 if (sk2) {
1775 inet_twsk_deschedule(inet_twsk(sk), &tcp_death_row);
1776 inet_twsk_put(inet_twsk(sk));
1777 sk = sk2;
1778 goto process;
1779 }
1780 /* Fall through to ACK */
1781 }
1782 case TCP_TW_ACK:
1783 tcp_v4_timewait_ack(sk, skb);
1784 break;
1785 case TCP_TW_RST:
1786 goto no_tcp_socket;
1787 case TCP_TW_SUCCESS:;
1788 }
1789 goto discard_it;
1790 }
1791
1792 struct inet_peer *tcp_v4_get_peer(struct sock *sk, bool *release_it)
1793 {
1794 struct rtable *rt = (struct rtable *) __sk_dst_get(sk);
1795 struct inet_sock *inet = inet_sk(sk);
1796 struct inet_peer *peer;
1797
1798 if (!rt ||
1799 inet->cork.fl.u.ip4.daddr != inet->inet_daddr) {
1800 peer = inet_getpeer_v4(inet->inet_daddr, 1);
1801 *release_it = true;
1802 } else {
1803 if (!rt->peer)
1804 rt_bind_peer(rt, inet->inet_daddr, 1);
1805 peer = rt->peer;
1806 *release_it = false;
1807 }
1808
1809 return peer;
1810 }
1811 EXPORT_SYMBOL(tcp_v4_get_peer);
1812
1813 void *tcp_v4_tw_get_peer(struct sock *sk)
1814 {
1815 const struct inet_timewait_sock *tw = inet_twsk(sk);
1816
1817 return inet_getpeer_v4(tw->tw_daddr, 1);
1818 }
1819 EXPORT_SYMBOL(tcp_v4_tw_get_peer);
1820
1821 static struct timewait_sock_ops tcp_timewait_sock_ops = {
1822 .twsk_obj_size = sizeof(struct tcp_timewait_sock),
1823 .twsk_unique = tcp_twsk_unique,
1824 .twsk_destructor= tcp_twsk_destructor,
1825 .twsk_getpeer = tcp_v4_tw_get_peer,
1826 };
1827
1828 const struct inet_connection_sock_af_ops ipv4_specific = {
1829 .queue_xmit = ip_queue_xmit,
1830 .send_check = tcp_v4_send_check,
1831 .rebuild_header = inet_sk_rebuild_header,
1832 .conn_request = tcp_v4_conn_request,
1833 .syn_recv_sock = tcp_v4_syn_recv_sock,
1834 .get_peer = tcp_v4_get_peer,
1835 .net_header_len = sizeof(struct iphdr),
1836 .setsockopt = ip_setsockopt,
1837 .getsockopt = ip_getsockopt,
1838 .addr2sockaddr = inet_csk_addr2sockaddr,
1839 .sockaddr_len = sizeof(struct sockaddr_in),
1840 .bind_conflict = inet_csk_bind_conflict,
1841 #ifdef CONFIG_COMPAT
1842 .compat_setsockopt = compat_ip_setsockopt,
1843 .compat_getsockopt = compat_ip_getsockopt,
1844 #endif
1845 };
1846 EXPORT_SYMBOL(ipv4_specific);
1847
1848 #ifdef CONFIG_TCP_MD5SIG
1849 static const struct tcp_sock_af_ops tcp_sock_ipv4_specific = {
1850 .md5_lookup = tcp_v4_md5_lookup,
1851 .calc_md5_hash = tcp_v4_md5_hash_skb,
1852 .md5_add = tcp_v4_md5_add_func,
1853 .md5_parse = tcp_v4_parse_md5_keys,
1854 };
1855 #endif
1856
1857 /* NOTE: A lot of things set to zero explicitly by call to
1858 * sk_alloc() so need not be done here.
1859 */
1860 static int tcp_v4_init_sock(struct sock *sk)
1861 {
1862 struct inet_connection_sock *icsk = inet_csk(sk);
1863 struct tcp_sock *tp = tcp_sk(sk);
1864
1865 skb_queue_head_init(&tp->out_of_order_queue);
1866 tcp_init_xmit_timers(sk);
1867 tcp_prequeue_init(tp);
1868
1869 icsk->icsk_rto = TCP_TIMEOUT_INIT;
1870 tp->mdev = TCP_TIMEOUT_INIT;
1871
1872 /* So many TCP implementations out there (incorrectly) count the
1873 * initial SYN frame in their delayed-ACK and congestion control
1874 * algorithms that we must have the following bandaid to talk
1875 * efficiently to them. -DaveM
1876 */
1877 tp->snd_cwnd = TCP_INIT_CWND;
1878
1879 /* See draft-stevens-tcpca-spec-01 for discussion of the
1880 * initialization of these values.
1881 */
1882 tp->snd_ssthresh = TCP_INFINITE_SSTHRESH;
1883 tp->snd_cwnd_clamp = ~0;
1884 tp->mss_cache = TCP_MSS_DEFAULT;
1885
1886 tp->reordering = sysctl_tcp_reordering;
1887 icsk->icsk_ca_ops = &tcp_init_congestion_ops;
1888
1889 sk->sk_state = TCP_CLOSE;
1890
1891 sk->sk_write_space = sk_stream_write_space;
1892 sock_set_flag(sk, SOCK_USE_WRITE_QUEUE);
1893
1894 icsk->icsk_af_ops = &ipv4_specific;
1895 icsk->icsk_sync_mss = tcp_sync_mss;
1896 #ifdef CONFIG_TCP_MD5SIG
1897 tp->af_specific = &tcp_sock_ipv4_specific;
1898 #endif
1899
1900 /* TCP Cookie Transactions */
1901 if (sysctl_tcp_cookie_size > 0) {
1902 /* Default, cookies without s_data_payload. */
1903 tp->cookie_values =
1904 kzalloc(sizeof(*tp->cookie_values),
1905 sk->sk_allocation);
1906 if (tp->cookie_values != NULL)
1907 kref_init(&tp->cookie_values->kref);
1908 }
1909 /* Presumed zeroed, in order of appearance:
1910 * cookie_in_always, cookie_out_never,
1911 * s_data_constant, s_data_in, s_data_out
1912 */
1913 sk->sk_sndbuf = sysctl_tcp_wmem[1];
1914 sk->sk_rcvbuf = sysctl_tcp_rmem[1];
1915
1916 local_bh_disable();
1917 percpu_counter_inc(&tcp_sockets_allocated);
1918 local_bh_enable();
1919
1920 return 0;
1921 }
1922
1923 void tcp_v4_destroy_sock(struct sock *sk)
1924 {
1925 struct tcp_sock *tp = tcp_sk(sk);
1926
1927 tcp_clear_xmit_timers(sk);
1928
1929 tcp_cleanup_congestion_control(sk);
1930
1931 /* Cleanup up the write buffer. */
1932 tcp_write_queue_purge(sk);
1933
1934 /* Cleans up our, hopefully empty, out_of_order_queue. */
1935 __skb_queue_purge(&tp->out_of_order_queue);
1936
1937 #ifdef CONFIG_TCP_MD5SIG
1938 /* Clean up the MD5 key list, if any */
1939 if (tp->md5sig_info) {
1940 tcp_v4_clear_md5_list(sk);
1941 kfree(tp->md5sig_info);
1942 tp->md5sig_info = NULL;
1943 }
1944 #endif
1945
1946 #ifdef CONFIG_NET_DMA
1947 /* Cleans up our sk_async_wait_queue */
1948 __skb_queue_purge(&sk->sk_async_wait_queue);
1949 #endif
1950
1951 /* Clean prequeue, it must be empty really */
1952 __skb_queue_purge(&tp->ucopy.prequeue);
1953
1954 /* Clean up a referenced TCP bind bucket. */
1955 if (inet_csk(sk)->icsk_bind_hash)
1956 inet_put_port(sk);
1957
1958 /*
1959 * If sendmsg cached page exists, toss it.
1960 */
1961 if (sk->sk_sndmsg_page) {
1962 __free_page(sk->sk_sndmsg_page);
1963 sk->sk_sndmsg_page = NULL;
1964 }
1965
1966 /* TCP Cookie Transactions */
1967 if (tp->cookie_values != NULL) {
1968 kref_put(&tp->cookie_values->kref,
1969 tcp_cookie_values_release);
1970 tp->cookie_values = NULL;
1971 }
1972
1973 percpu_counter_dec(&tcp_sockets_allocated);
1974 }
1975 EXPORT_SYMBOL(tcp_v4_destroy_sock);
1976
1977 #ifdef CONFIG_PROC_FS
1978 /* Proc filesystem TCP sock list dumping. */
1979
1980 static inline struct inet_timewait_sock *tw_head(struct hlist_nulls_head *head)
1981 {
1982 return hlist_nulls_empty(head) ? NULL :
1983 list_entry(head->first, struct inet_timewait_sock, tw_node);
1984 }
1985
1986 static inline struct inet_timewait_sock *tw_next(struct inet_timewait_sock *tw)
1987 {
1988 return !is_a_nulls(tw->tw_node.next) ?
1989 hlist_nulls_entry(tw->tw_node.next, typeof(*tw), tw_node) : NULL;
1990 }
1991
1992 /*
1993 * Get next listener socket follow cur. If cur is NULL, get first socket
1994 * starting from bucket given in st->bucket; when st->bucket is zero the
1995 * very first socket in the hash table is returned.
1996 */
1997 static void *listening_get_next(struct seq_file *seq, void *cur)
1998 {
1999 struct inet_connection_sock *icsk;
2000 struct hlist_nulls_node *node;
2001 struct sock *sk = cur;
2002 struct inet_listen_hashbucket *ilb;
2003 struct tcp_iter_state *st = seq->private;
2004 struct net *net = seq_file_net(seq);
2005
2006 if (!sk) {
2007 ilb = &tcp_hashinfo.listening_hash[st->bucket];
2008 spin_lock_bh(&ilb->lock);
2009 sk = sk_nulls_head(&ilb->head);
2010 st->offset = 0;
2011 goto get_sk;
2012 }
2013 ilb = &tcp_hashinfo.listening_hash[st->bucket];
2014 ++st->num;
2015 ++st->offset;
2016
2017 if (st->state == TCP_SEQ_STATE_OPENREQ) {
2018 struct request_sock *req = cur;
2019
2020 icsk = inet_csk(st->syn_wait_sk);
2021 req = req->dl_next;
2022 while (1) {
2023 while (req) {
2024 if (req->rsk_ops->family == st->family) {
2025 cur = req;
2026 goto out;
2027 }
2028 req = req->dl_next;
2029 }
2030 if (++st->sbucket >= icsk->icsk_accept_queue.listen_opt->nr_table_entries)
2031 break;
2032 get_req:
2033 req = icsk->icsk_accept_queue.listen_opt->syn_table[st->sbucket];
2034 }
2035 sk = sk_nulls_next(st->syn_wait_sk);
2036 st->state = TCP_SEQ_STATE_LISTENING;
2037 read_unlock_bh(&icsk->icsk_accept_queue.syn_wait_lock);
2038 } else {
2039 icsk = inet_csk(sk);
2040 read_lock_bh(&icsk->icsk_accept_queue.syn_wait_lock);
2041 if (reqsk_queue_len(&icsk->icsk_accept_queue))
2042 goto start_req;
2043 read_unlock_bh(&icsk->icsk_accept_queue.syn_wait_lock);
2044 sk = sk_nulls_next(sk);
2045 }
2046 get_sk:
2047 sk_nulls_for_each_from(sk, node) {
2048 if (!net_eq(sock_net(sk), net))
2049 continue;
2050 if (sk->sk_family == st->family) {
2051 cur = sk;
2052 goto out;
2053 }
2054 icsk = inet_csk(sk);
2055 read_lock_bh(&icsk->icsk_accept_queue.syn_wait_lock);
2056 if (reqsk_queue_len(&icsk->icsk_accept_queue)) {
2057 start_req:
2058 st->uid = sock_i_uid(sk);
2059 st->syn_wait_sk = sk;
2060 st->state = TCP_SEQ_STATE_OPENREQ;
2061 st->sbucket = 0;
2062 goto get_req;
2063 }
2064 read_unlock_bh(&icsk->icsk_accept_queue.syn_wait_lock);
2065 }
2066 spin_unlock_bh(&ilb->lock);
2067 st->offset = 0;
2068 if (++st->bucket < INET_LHTABLE_SIZE) {
2069 ilb = &tcp_hashinfo.listening_hash[st->bucket];
2070 spin_lock_bh(&ilb->lock);
2071 sk = sk_nulls_head(&ilb->head);
2072 goto get_sk;
2073 }
2074 cur = NULL;
2075 out:
2076 return cur;
2077 }
2078
2079 static void *listening_get_idx(struct seq_file *seq, loff_t *pos)
2080 {
2081 struct tcp_iter_state *st = seq->private;
2082 void *rc;
2083
2084 st->bucket = 0;
2085 st->offset = 0;
2086 rc = listening_get_next(seq, NULL);
2087
2088 while (rc && *pos) {
2089 rc = listening_get_next(seq, rc);
2090 --*pos;
2091 }
2092 return rc;
2093 }
2094
2095 static inline int empty_bucket(struct tcp_iter_state *st)
2096 {
2097 return hlist_nulls_empty(&tcp_hashinfo.ehash[st->bucket].chain) &&
2098 hlist_nulls_empty(&tcp_hashinfo.ehash[st->bucket].twchain);
2099 }
2100
2101 /*
2102 * Get first established socket starting from bucket given in st->bucket.
2103 * If st->bucket is zero, the very first socket in the hash is returned.
2104 */
2105 static void *established_get_first(struct seq_file *seq)
2106 {
2107 struct tcp_iter_state *st = seq->private;
2108 struct net *net = seq_file_net(seq);
2109 void *rc = NULL;
2110
2111 st->offset = 0;
2112 for (; st->bucket <= tcp_hashinfo.ehash_mask; ++st->bucket) {
2113 struct sock *sk;
2114 struct hlist_nulls_node *node;
2115 struct inet_timewait_sock *tw;
2116 spinlock_t *lock = inet_ehash_lockp(&tcp_hashinfo, st->bucket);
2117
2118 /* Lockless fast path for the common case of empty buckets */
2119 if (empty_bucket(st))
2120 continue;
2121
2122 spin_lock_bh(lock);
2123 sk_nulls_for_each(sk, node, &tcp_hashinfo.ehash[st->bucket].chain) {
2124 if (sk->sk_family != st->family ||
2125 !net_eq(sock_net(sk), net)) {
2126 continue;
2127 }
2128 rc = sk;
2129 goto out;
2130 }
2131 st->state = TCP_SEQ_STATE_TIME_WAIT;
2132 inet_twsk_for_each(tw, node,
2133 &tcp_hashinfo.ehash[st->bucket].twchain) {
2134 if (tw->tw_family != st->family ||
2135 !net_eq(twsk_net(tw), net)) {
2136 continue;
2137 }
2138 rc = tw;
2139 goto out;
2140 }
2141 spin_unlock_bh(lock);
2142 st->state = TCP_SEQ_STATE_ESTABLISHED;
2143 }
2144 out:
2145 return rc;
2146 }
2147
2148 static void *established_get_next(struct seq_file *seq, void *cur)
2149 {
2150 struct sock *sk = cur;
2151 struct inet_timewait_sock *tw;
2152 struct hlist_nulls_node *node;
2153 struct tcp_iter_state *st = seq->private;
2154 struct net *net = seq_file_net(seq);
2155
2156 ++st->num;
2157 ++st->offset;
2158
2159 if (st->state == TCP_SEQ_STATE_TIME_WAIT) {
2160 tw = cur;
2161 tw = tw_next(tw);
2162 get_tw:
2163 while (tw && (tw->tw_family != st->family || !net_eq(twsk_net(tw), net))) {
2164 tw = tw_next(tw);
2165 }
2166 if (tw) {
2167 cur = tw;
2168 goto out;
2169 }
2170 spin_unlock_bh(inet_ehash_lockp(&tcp_hashinfo, st->bucket));
2171 st->state = TCP_SEQ_STATE_ESTABLISHED;
2172
2173 /* Look for next non empty bucket */
2174 st->offset = 0;
2175 while (++st->bucket <= tcp_hashinfo.ehash_mask &&
2176 empty_bucket(st))
2177 ;
2178 if (st->bucket > tcp_hashinfo.ehash_mask)
2179 return NULL;
2180
2181 spin_lock_bh(inet_ehash_lockp(&tcp_hashinfo, st->bucket));
2182 sk = sk_nulls_head(&tcp_hashinfo.ehash[st->bucket].chain);
2183 } else
2184 sk = sk_nulls_next(sk);
2185
2186 sk_nulls_for_each_from(sk, node) {
2187 if (sk->sk_family == st->family && net_eq(sock_net(sk), net))
2188 goto found;
2189 }
2190
2191 st->state = TCP_SEQ_STATE_TIME_WAIT;
2192 tw = tw_head(&tcp_hashinfo.ehash[st->bucket].twchain);
2193 goto get_tw;
2194 found:
2195 cur = sk;
2196 out:
2197 return cur;
2198 }
2199
2200 static void *established_get_idx(struct seq_file *seq, loff_t pos)
2201 {
2202 struct tcp_iter_state *st = seq->private;
2203 void *rc;
2204
2205 st->bucket = 0;
2206 rc = established_get_first(seq);
2207
2208 while (rc && pos) {
2209 rc = established_get_next(seq, rc);
2210 --pos;
2211 }
2212 return rc;
2213 }
2214
2215 static void *tcp_get_idx(struct seq_file *seq, loff_t pos)
2216 {
2217 void *rc;
2218 struct tcp_iter_state *st = seq->private;
2219
2220 st->state = TCP_SEQ_STATE_LISTENING;
2221 rc = listening_get_idx(seq, &pos);
2222
2223 if (!rc) {
2224 st->state = TCP_SEQ_STATE_ESTABLISHED;
2225 rc = established_get_idx(seq, pos);
2226 }
2227
2228 return rc;
2229 }
2230
2231 static void *tcp_seek_last_pos(struct seq_file *seq)
2232 {
2233 struct tcp_iter_state *st = seq->private;
2234 int offset = st->offset;
2235 int orig_num = st->num;
2236 void *rc = NULL;
2237
2238 switch (st->state) {
2239 case TCP_SEQ_STATE_OPENREQ:
2240 case TCP_SEQ_STATE_LISTENING:
2241 if (st->bucket >= INET_LHTABLE_SIZE)
2242 break;
2243 st->state = TCP_SEQ_STATE_LISTENING;
2244 rc = listening_get_next(seq, NULL);
2245 while (offset-- && rc)
2246 rc = listening_get_next(seq, rc);
2247 if (rc)
2248 break;
2249 st->bucket = 0;
2250 /* Fallthrough */
2251 case TCP_SEQ_STATE_ESTABLISHED:
2252 case TCP_SEQ_STATE_TIME_WAIT:
2253 st->state = TCP_SEQ_STATE_ESTABLISHED;
2254 if (st->bucket > tcp_hashinfo.ehash_mask)
2255 break;
2256 rc = established_get_first(seq);
2257 while (offset-- && rc)
2258 rc = established_get_next(seq, rc);
2259 }
2260
2261 st->num = orig_num;
2262
2263 return rc;
2264 }
2265
2266 static void *tcp_seq_start(struct seq_file *seq, loff_t *pos)
2267 {
2268 struct tcp_iter_state *st = seq->private;
2269 void *rc;
2270
2271 if (*pos && *pos == st->last_pos) {
2272 rc = tcp_seek_last_pos(seq);
2273 if (rc)
2274 goto out;
2275 }
2276
2277 st->state = TCP_SEQ_STATE_LISTENING;
2278 st->num = 0;
2279 st->bucket = 0;
2280 st->offset = 0;
2281 rc = *pos ? tcp_get_idx(seq, *pos - 1) : SEQ_START_TOKEN;
2282
2283 out:
2284 st->last_pos = *pos;
2285 return rc;
2286 }
2287
2288 static void *tcp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2289 {
2290 struct tcp_iter_state *st = seq->private;
2291 void *rc = NULL;
2292
2293 if (v == SEQ_START_TOKEN) {
2294 rc = tcp_get_idx(seq, 0);
2295 goto out;
2296 }
2297
2298 switch (st->state) {
2299 case TCP_SEQ_STATE_OPENREQ:
2300 case TCP_SEQ_STATE_LISTENING:
2301 rc = listening_get_next(seq, v);
2302 if (!rc) {
2303 st->state = TCP_SEQ_STATE_ESTABLISHED;
2304 st->bucket = 0;
2305 st->offset = 0;
2306 rc = established_get_first(seq);
2307 }
2308 break;
2309 case TCP_SEQ_STATE_ESTABLISHED:
2310 case TCP_SEQ_STATE_TIME_WAIT:
2311 rc = established_get_next(seq, v);
2312 break;
2313 }
2314 out:
2315 ++*pos;
2316 st->last_pos = *pos;
2317 return rc;
2318 }
2319
2320 static void tcp_seq_stop(struct seq_file *seq, void *v)
2321 {
2322 struct tcp_iter_state *st = seq->private;
2323
2324 switch (st->state) {
2325 case TCP_SEQ_STATE_OPENREQ:
2326 if (v) {
2327 struct inet_connection_sock *icsk = inet_csk(st->syn_wait_sk);
2328 read_unlock_bh(&icsk->icsk_accept_queue.syn_wait_lock);
2329 }
2330 case TCP_SEQ_STATE_LISTENING:
2331 if (v != SEQ_START_TOKEN)
2332 spin_unlock_bh(&tcp_hashinfo.listening_hash[st->bucket].lock);
2333 break;
2334 case TCP_SEQ_STATE_TIME_WAIT:
2335 case TCP_SEQ_STATE_ESTABLISHED:
2336 if (v)
2337 spin_unlock_bh(inet_ehash_lockp(&tcp_hashinfo, st->bucket));
2338 break;
2339 }
2340 }
2341
2342 static int tcp_seq_open(struct inode *inode, struct file *file)
2343 {
2344 struct tcp_seq_afinfo *afinfo = PDE(inode)->data;
2345 struct tcp_iter_state *s;
2346 int err;
2347
2348 err = seq_open_net(inode, file, &afinfo->seq_ops,
2349 sizeof(struct tcp_iter_state));
2350 if (err < 0)
2351 return err;
2352
2353 s = ((struct seq_file *)file->private_data)->private;
2354 s->family = afinfo->family;
2355 s->last_pos = 0;
2356 return 0;
2357 }
2358
2359 int tcp_proc_register(struct net *net, struct tcp_seq_afinfo *afinfo)
2360 {
2361 int rc = 0;
2362 struct proc_dir_entry *p;
2363
2364 afinfo->seq_fops.open = tcp_seq_open;
2365 afinfo->seq_fops.read = seq_read;
2366 afinfo->seq_fops.llseek = seq_lseek;
2367 afinfo->seq_fops.release = seq_release_net;
2368
2369 afinfo->seq_ops.start = tcp_seq_start;
2370 afinfo->seq_ops.next = tcp_seq_next;
2371 afinfo->seq_ops.stop = tcp_seq_stop;
2372
2373 p = proc_create_data(afinfo->name, S_IRUGO, net->proc_net,
2374 &afinfo->seq_fops, afinfo);
2375 if (!p)
2376 rc = -ENOMEM;
2377 return rc;
2378 }
2379 EXPORT_SYMBOL(tcp_proc_register);
2380
2381 void tcp_proc_unregister(struct net *net, struct tcp_seq_afinfo *afinfo)
2382 {
2383 proc_net_remove(net, afinfo->name);
2384 }
2385 EXPORT_SYMBOL(tcp_proc_unregister);
2386
2387 static void get_openreq4(const struct sock *sk, const struct request_sock *req,
2388 struct seq_file *f, int i, int uid, int *len)
2389 {
2390 const struct inet_request_sock *ireq = inet_rsk(req);
2391 int ttd = req->expires - jiffies;
2392
2393 seq_printf(f, "%4d: %08X:%04X %08X:%04X"
2394 " %02X %08X:%08X %02X:%08lX %08X %5d %8d %u %d %pK%n",
2395 i,
2396 ireq->loc_addr,
2397 ntohs(inet_sk(sk)->inet_sport),
2398 ireq->rmt_addr,
2399 ntohs(ireq->rmt_port),
2400 TCP_SYN_RECV,
2401 0, 0, /* could print option size, but that is af dependent. */
2402 1, /* timers active (only the expire timer) */
2403 jiffies_to_clock_t(ttd),
2404 req->retrans,
2405 uid,
2406 0, /* non standard timer */
2407 0, /* open_requests have no inode */
2408 atomic_read(&sk->sk_refcnt),
2409 req,
2410 len);
2411 }
2412
2413 static void get_tcp4_sock(struct sock *sk, struct seq_file *f, int i, int *len)
2414 {
2415 int timer_active;
2416 unsigned long timer_expires;
2417 const struct tcp_sock *tp = tcp_sk(sk);
2418 const struct inet_connection_sock *icsk = inet_csk(sk);
2419 const struct inet_sock *inet = inet_sk(sk);
2420 __be32 dest = inet->inet_daddr;
2421 __be32 src = inet->inet_rcv_saddr;
2422 __u16 destp = ntohs(inet->inet_dport);
2423 __u16 srcp = ntohs(inet->inet_sport);
2424 int rx_queue;
2425
2426 if (icsk->icsk_pending == ICSK_TIME_RETRANS) {
2427 timer_active = 1;
2428 timer_expires = icsk->icsk_timeout;
2429 } else if (icsk->icsk_pending == ICSK_TIME_PROBE0) {
2430 timer_active = 4;
2431 timer_expires = icsk->icsk_timeout;
2432 } else if (timer_pending(&sk->sk_timer)) {
2433 timer_active = 2;
2434 timer_expires = sk->sk_timer.expires;
2435 } else {
2436 timer_active = 0;
2437 timer_expires = jiffies;
2438 }
2439
2440 if (sk->sk_state == TCP_LISTEN)
2441 rx_queue = sk->sk_ack_backlog;
2442 else
2443 /*
2444 * because we dont lock socket, we might find a transient negative value
2445 */
2446 rx_queue = max_t(int, tp->rcv_nxt - tp->copied_seq, 0);
2447
2448 seq_printf(f, "%4d: %08X:%04X %08X:%04X %02X %08X:%08X %02X:%08lX "
2449 "%08X %5d %8d %lu %d %pK %lu %lu %u %u %d%n",
2450 i, src, srcp, dest, destp, sk->sk_state,
2451 tp->write_seq - tp->snd_una,
2452 rx_queue,
2453 timer_active,
2454 jiffies_to_clock_t(timer_expires - jiffies),
2455 icsk->icsk_retransmits,
2456 sock_i_uid(sk),
2457 icsk->icsk_probes_out,
2458 sock_i_ino(sk),
2459 atomic_read(&sk->sk_refcnt), sk,
2460 jiffies_to_clock_t(icsk->icsk_rto),
2461 jiffies_to_clock_t(icsk->icsk_ack.ato),
2462 (icsk->icsk_ack.quick << 1) | icsk->icsk_ack.pingpong,
2463 tp->snd_cwnd,
2464 tcp_in_initial_slowstart(tp) ? -1 : tp->snd_ssthresh,
2465 len);
2466 }
2467
2468 static void get_timewait4_sock(const struct inet_timewait_sock *tw,
2469 struct seq_file *f, int i, int *len)
2470 {
2471 __be32 dest, src;
2472 __u16 destp, srcp;
2473 int ttd = tw->tw_ttd - jiffies;
2474
2475 if (ttd < 0)
2476 ttd = 0;
2477
2478 dest = tw->tw_daddr;
2479 src = tw->tw_rcv_saddr;
2480 destp = ntohs(tw->tw_dport);
2481 srcp = ntohs(tw->tw_sport);
2482
2483 seq_printf(f, "%4d: %08X:%04X %08X:%04X"
2484 " %02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %pK%n",
2485 i, src, srcp, dest, destp, tw->tw_substate, 0, 0,
2486 3, jiffies_to_clock_t(ttd), 0, 0, 0, 0,
2487 atomic_read(&tw->tw_refcnt), tw, len);
2488 }
2489
2490 #define TMPSZ 150
2491
2492 static int tcp4_seq_show(struct seq_file *seq, void *v)
2493 {
2494 struct tcp_iter_state *st;
2495 int len;
2496
2497 if (v == SEQ_START_TOKEN) {
2498 seq_printf(seq, "%-*s\n", TMPSZ - 1,
2499 " sl local_address rem_address st tx_queue "
2500 "rx_queue tr tm->when retrnsmt uid timeout "
2501 "inode");
2502 goto out;
2503 }
2504 st = seq->private;
2505
2506 switch (st->state) {
2507 case TCP_SEQ_STATE_LISTENING:
2508 case TCP_SEQ_STATE_ESTABLISHED:
2509 get_tcp4_sock(v, seq, st->num, &len);
2510 break;
2511 case TCP_SEQ_STATE_OPENREQ:
2512 get_openreq4(st->syn_wait_sk, v, seq, st->num, st->uid, &len);
2513 break;
2514 case TCP_SEQ_STATE_TIME_WAIT:
2515 get_timewait4_sock(v, seq, st->num, &len);
2516 break;
2517 }
2518 seq_printf(seq, "%*s\n", TMPSZ - 1 - len, "");
2519 out:
2520 return 0;
2521 }
2522
2523 static struct tcp_seq_afinfo tcp4_seq_afinfo = {
2524 .name = "tcp",
2525 .family = AF_INET,
2526 .seq_fops = {
2527 .owner = THIS_MODULE,
2528 },
2529 .seq_ops = {
2530 .show = tcp4_seq_show,
2531 },
2532 };
2533
2534 static int __net_init tcp4_proc_init_net(struct net *net)
2535 {
2536 return tcp_proc_register(net, &tcp4_seq_afinfo);
2537 }
2538
2539 static void __net_exit tcp4_proc_exit_net(struct net *net)
2540 {
2541 tcp_proc_unregister(net, &tcp4_seq_afinfo);
2542 }
2543
2544 static struct pernet_operations tcp4_net_ops = {
2545 .init = tcp4_proc_init_net,
2546 .exit = tcp4_proc_exit_net,
2547 };
2548
2549 int __init tcp4_proc_init(void)
2550 {
2551 return register_pernet_subsys(&tcp4_net_ops);
2552 }
2553
2554 void tcp4_proc_exit(void)
2555 {
2556 unregister_pernet_subsys(&tcp4_net_ops);
2557 }
2558 #endif /* CONFIG_PROC_FS */
2559
2560 struct sk_buff **tcp4_gro_receive(struct sk_buff **head, struct sk_buff *skb)
2561 {
2562 const struct iphdr *iph = skb_gro_network_header(skb);
2563
2564 switch (skb->ip_summed) {
2565 case CHECKSUM_COMPLETE:
2566 if (!tcp_v4_check(skb_gro_len(skb), iph->saddr, iph->daddr,
2567 skb->csum)) {
2568 skb->ip_summed = CHECKSUM_UNNECESSARY;
2569 break;
2570 }
2571
2572 /* fall through */
2573 case CHECKSUM_NONE:
2574 NAPI_GRO_CB(skb)->flush = 1;
2575 return NULL;
2576 }
2577
2578 return tcp_gro_receive(head, skb);
2579 }
2580
2581 int tcp4_gro_complete(struct sk_buff *skb)
2582 {
2583 const struct iphdr *iph = ip_hdr(skb);
2584 struct tcphdr *th = tcp_hdr(skb);
2585
2586 th->check = ~tcp_v4_check(skb->len - skb_transport_offset(skb),
2587 iph->saddr, iph->daddr, 0);
2588 skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
2589
2590 return tcp_gro_complete(skb);
2591 }
2592
2593 struct proto tcp_prot = {
2594 .name = "TCP",
2595 .owner = THIS_MODULE,
2596 .close = tcp_close,
2597 .connect = tcp_v4_connect,
2598 .disconnect = tcp_disconnect,
2599 .accept = inet_csk_accept,
2600 .ioctl = tcp_ioctl,
2601 .init = tcp_v4_init_sock,
2602 .destroy = tcp_v4_destroy_sock,
2603 .shutdown = tcp_shutdown,
2604 .setsockopt = tcp_setsockopt,
2605 .getsockopt = tcp_getsockopt,
2606 .recvmsg = tcp_recvmsg,
2607 .sendmsg = tcp_sendmsg,
2608 .sendpage = tcp_sendpage,
2609 .backlog_rcv = tcp_v4_do_rcv,
2610 .hash = inet_hash,
2611 .unhash = inet_unhash,
2612 .get_port = inet_csk_get_port,
2613 .enter_memory_pressure = tcp_enter_memory_pressure,
2614 .sockets_allocated = &tcp_sockets_allocated,
2615 .orphan_count = &tcp_orphan_count,
2616 .memory_allocated = &tcp_memory_allocated,
2617 .memory_pressure = &tcp_memory_pressure,
2618 .sysctl_mem = sysctl_tcp_mem,
2619 .sysctl_wmem = sysctl_tcp_wmem,
2620 .sysctl_rmem = sysctl_tcp_rmem,
2621 .max_header = MAX_TCP_HEADER,
2622 .obj_size = sizeof(struct tcp_sock),
2623 .slab_flags = SLAB_DESTROY_BY_RCU,
2624 .twsk_prot = &tcp_timewait_sock_ops,
2625 .rsk_prot = &tcp_request_sock_ops,
2626 .h.hashinfo = &tcp_hashinfo,
2627 .no_autobind = true,
2628 #ifdef CONFIG_COMPAT
2629 .compat_setsockopt = compat_tcp_setsockopt,
2630 .compat_getsockopt = compat_tcp_getsockopt,
2631 #endif
2632 };
2633 EXPORT_SYMBOL(tcp_prot);
2634
2635
2636 static int __net_init tcp_sk_init(struct net *net)
2637 {
2638 return inet_ctl_sock_create(&net->ipv4.tcp_sock,
2639 PF_INET, SOCK_RAW, IPPROTO_TCP, net);
2640 }
2641
2642 static void __net_exit tcp_sk_exit(struct net *net)
2643 {
2644 inet_ctl_sock_destroy(net->ipv4.tcp_sock);
2645 }
2646
2647 static void __net_exit tcp_sk_exit_batch(struct list_head *net_exit_list)
2648 {
2649 inet_twsk_purge(&tcp_hashinfo, &tcp_death_row, AF_INET);
2650 }
2651
2652 static struct pernet_operations __net_initdata tcp_sk_ops = {
2653 .init = tcp_sk_init,
2654 .exit = tcp_sk_exit,
2655 .exit_batch = tcp_sk_exit_batch,
2656 };
2657
2658 void __init tcp_v4_init(void)
2659 {
2660 inet_hashinfo_init(&tcp_hashinfo);
2661 if (register_pernet_subsys(&tcp_sk_ops))
2662 panic("Failed to create the TCP control socket.\n");
2663 }
Linux Btrfs Development