?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
ipv4: tcp: remove per net tcp_sock
[linux-2.6.git] / net / ipv4 / tcp_ipv4.c
blobd7d2fa50f07fedd9c0def65f5f49d208bf3195fa
1 /*
2 * INET An implementation of the TCP/IP protocol suite for the LINUX
3 * operating system. INET is implemented using the BSD Socket
4 * interface as the means of communication with the user level.
5 *
6 * Implementation of the Transmission Control Protocol(TCP).
7 *
8 * IPv4 specific functions
9 *
10 *
11 * code split from:
12 * linux/ipv4/tcp.c
13 * linux/ipv4/tcp_input.c
14 * linux/ipv4/tcp_output.c
15 *
16 * See tcp.c for author information
17 *
18 * This program is free software; you can redistribute it and/or
19 * modify it under the terms of the GNU General Public License
20 * as published by the Free Software Foundation; either version
21 * 2 of the License, or (at your option) any later version.
22 */
23
24 /*
25 * Changes:
26 * David S. Miller : New socket lookup architecture.
27 * This code is dedicated to John Dyson.
28 * David S. Miller : Change semantics of established hash,
29 * half is devoted to TIME_WAIT sockets
30 * and the rest go in the other half.
31 * Andi Kleen : Add support for syncookies and fixed
32 * some bugs: ip options weren't passed to
33 * the TCP layer, missed a check for an
34 * ACK bit.
35 * Andi Kleen : Implemented fast path mtu discovery.
36 * Fixed many serious bugs in the
37 * request_sock handling and moved
38 * most of it into the af independent code.
39 * Added tail drop and some other bugfixes.
40 * Added new listen semantics.
41 * Mike McLagan : Routing by source
42 * Juan Jose Ciarlante: ip_dynaddr bits
43 * Andi Kleen: various fixes.
44 * Vitaly E. Lavrov : Transparent proxy revived after year
45 * coma.
46 * Andi Kleen : Fix new listen.
47 * Andi Kleen : Fix accept error reporting.
48 * YOSHIFUJI Hideaki @USAGI and: Support IPV6_V6ONLY socket option, which
49 * Alexey Kuznetsov allow both IPv4 and IPv6 sockets to bind
50 * a single port at the same time.
51 */
52
53 #define pr_fmt(fmt) "TCP: " fmt
54
55 #include <linux/bottom_half.h>
56 #include <linux/types.h>
57 #include <linux/fcntl.h>
58 #include <linux/module.h>
59 #include <linux/random.h>
60 #include <linux/cache.h>
61 #include <linux/jhash.h>
62 #include <linux/init.h>
63 #include <linux/times.h>
64 #include <linux/slab.h>
65
66 #include <net/net_namespace.h>
67 #include <net/icmp.h>
68 #include <net/inet_hashtables.h>
69 #include <net/tcp.h>
70 #include <net/transp_v6.h>
71 #include <net/ipv6.h>
72 #include <net/inet_common.h>
73 #include <net/timewait_sock.h>
74 #include <net/xfrm.h>
75 #include <net/netdma.h>
76 #include <net/secure_seq.h>
77 #include <net/tcp_memcontrol.h>
78
79 #include <linux/inet.h>
80 #include <linux/ipv6.h>
81 #include <linux/stddef.h>
82 #include <linux/proc_fs.h>
83 #include <linux/seq_file.h>
84
85 #include <linux/crypto.h>
86 #include <linux/scatterlist.h>
87
88 int sysctl_tcp_tw_reuse __read_mostly;
89 int sysctl_tcp_low_latency __read_mostly;
90 EXPORT_SYMBOL(sysctl_tcp_low_latency);
91
92
93 #ifdef CONFIG_TCP_MD5SIG
94 static int tcp_v4_md5_hash_hdr(char *md5_hash, const struct tcp_md5sig_key *key,
95 __be32 daddr, __be32 saddr, const struct tcphdr *th);
96 #endif
97
98 struct inet_hashinfo tcp_hashinfo;
99 EXPORT_SYMBOL(tcp_hashinfo);
100
101 static inline __u32 tcp_v4_init_sequence(const struct sk_buff *skb)
102 {
103 return secure_tcp_sequence_number(ip_hdr(skb)->daddr,
104 ip_hdr(skb)->saddr,
105 tcp_hdr(skb)->dest,
106 tcp_hdr(skb)->source);
107 }
108
109 int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp)
110 {
111 const struct tcp_timewait_sock *tcptw = tcp_twsk(sktw);
112 struct tcp_sock *tp = tcp_sk(sk);
113
114 /* With PAWS, it is safe from the viewpoint
115 of data integrity. Even without PAWS it is safe provided sequence
116 spaces do not overlap i.e. at data rates <= 80Mbit/sec.
117
118 Actually, the idea is close to VJ's one, only timestamp cache is
119 held not per host, but per port pair and TW bucket is used as state
120 holder.
121
122 If TW bucket has been already destroyed we fall back to VJ's scheme
123 and use initial timestamp retrieved from peer table.
124 */
125 if (tcptw->tw_ts_recent_stamp &&
126 (twp == NULL || (sysctl_tcp_tw_reuse &&
127 get_seconds() - tcptw->tw_ts_recent_stamp > 1))) {
128 tp->write_seq = tcptw->tw_snd_nxt + 65535 + 2;
129 if (tp->write_seq == 0)
130 tp->write_seq = 1;
131 tp->rx_opt.ts_recent = tcptw->tw_ts_recent;
132 tp->rx_opt.ts_recent_stamp = tcptw->tw_ts_recent_stamp;
133 sock_hold(sktw);
134 return 1;
135 }
136
137 return 0;
138 }
139 EXPORT_SYMBOL_GPL(tcp_twsk_unique);
140
141 static int tcp_repair_connect(struct sock *sk)
142 {
143 tcp_connect_init(sk);
144 tcp_finish_connect(sk, NULL);
145
146 return 0;
147 }
148
149 /* This will initiate an outgoing connection. */
150 int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len)
151 {
152 struct sockaddr_in *usin = (struct sockaddr_in *)uaddr;
153 struct inet_sock *inet = inet_sk(sk);
154 struct tcp_sock *tp = tcp_sk(sk);
155 __be16 orig_sport, orig_dport;
156 __be32 daddr, nexthop;
157 struct flowi4 *fl4;
158 struct rtable *rt;
159 int err;
160 struct ip_options_rcu *inet_opt;
161
162 if (addr_len < sizeof(struct sockaddr_in))
163 return -EINVAL;
164
165 if (usin->sin_family != AF_INET)
166 return -EAFNOSUPPORT;
167
168 nexthop = daddr = usin->sin_addr.s_addr;
169 inet_opt = rcu_dereference_protected(inet->inet_opt,
170 sock_owned_by_user(sk));
171 if (inet_opt && inet_opt->opt.srr) {
172 if (!daddr)
173 return -EINVAL;
174 nexthop = inet_opt->opt.faddr;
175 }
176
177 orig_sport = inet->inet_sport;
178 orig_dport = usin->sin_port;
179 fl4 = &inet->cork.fl.u.ip4;
180 rt = ip_route_connect(fl4, nexthop, inet->inet_saddr,
181 RT_CONN_FLAGS(sk), sk->sk_bound_dev_if,
182 IPPROTO_TCP,
183 orig_sport, orig_dport, sk, true);
184 if (IS_ERR(rt)) {
185 err = PTR_ERR(rt);
186 if (err == -ENETUNREACH)
187 IP_INC_STATS_BH(sock_net(sk), IPSTATS_MIB_OUTNOROUTES);
188 return err;
189 }
190
191 if (rt->rt_flags & (RTCF_MULTICAST | RTCF_BROADCAST)) {
192 ip_rt_put(rt);
193 return -ENETUNREACH;
194 }
195
196 if (!inet_opt || !inet_opt->opt.srr)
197 daddr = fl4->daddr;
198
199 if (!inet->inet_saddr)
200 inet->inet_saddr = fl4->saddr;
201 inet->inet_rcv_saddr = inet->inet_saddr;
202
203 if (tp->rx_opt.ts_recent_stamp && inet->inet_daddr != daddr) {
204 /* Reset inherited state */
205 tp->rx_opt.ts_recent = 0;
206 tp->rx_opt.ts_recent_stamp = 0;
207 if (likely(!tp->repair))
208 tp->write_seq = 0;
209 }
210
211 if (tcp_death_row.sysctl_tw_recycle &&
212 !tp->rx_opt.ts_recent_stamp && fl4->daddr == daddr)
213 tcp_fetch_timewait_stamp(sk, &rt->dst);
214
215 inet->inet_dport = usin->sin_port;
216 inet->inet_daddr = daddr;
217
218 inet_csk(sk)->icsk_ext_hdr_len = 0;
219 if (inet_opt)
220 inet_csk(sk)->icsk_ext_hdr_len = inet_opt->opt.optlen;
221
222 tp->rx_opt.mss_clamp = TCP_MSS_DEFAULT;
223
224 /* Socket identity is still unknown (sport may be zero).
225 * However we set state to SYN-SENT and not releasing socket
226 * lock select source port, enter ourselves into the hash tables and
227 * complete initialization after this.
228 */
229 tcp_set_state(sk, TCP_SYN_SENT);
230 err = inet_hash_connect(&tcp_death_row, sk);
231 if (err)
232 goto failure;
233
234 rt = ip_route_newports(fl4, rt, orig_sport, orig_dport,
235 inet->inet_sport, inet->inet_dport, sk);
236 if (IS_ERR(rt)) {
237 err = PTR_ERR(rt);
238 rt = NULL;
239 goto failure;
240 }
241 /* OK, now commit destination to socket. */
242 sk->sk_gso_type = SKB_GSO_TCPV4;
243 sk_setup_caps(sk, &rt->dst);
244
245 if (!tp->write_seq && likely(!tp->repair))
246 tp->write_seq = secure_tcp_sequence_number(inet->inet_saddr,
247 inet->inet_daddr,
248 inet->inet_sport,
249 usin->sin_port);
250
251 inet->inet_id = tp->write_seq ^ jiffies;
252
253 if (likely(!tp->repair))
254 err = tcp_connect(sk);
255 else
256 err = tcp_repair_connect(sk);
257
258 rt = NULL;
259 if (err)
260 goto failure;
261
262 return 0;
263
264 failure:
265 /*
266 * This unhashes the socket and releases the local port,
267 * if necessary.
268 */
269 tcp_set_state(sk, TCP_CLOSE);
270 ip_rt_put(rt);
271 sk->sk_route_caps = 0;
272 inet->inet_dport = 0;
273 return err;
274 }
275 EXPORT_SYMBOL(tcp_v4_connect);
276
277 /*
278 * This routine does path mtu discovery as defined in RFC1191.
279 */
280 static void do_pmtu_discovery(struct sock *sk, const struct iphdr *iph, u32 mtu)
281 {
282 struct dst_entry *dst;
283 struct inet_sock *inet = inet_sk(sk);
284
285 /* We are not interested in TCP_LISTEN and open_requests (SYN-ACKs
286 * send out by Linux are always <576bytes so they should go through
287 * unfragmented).
288 */
289 if (sk->sk_state == TCP_LISTEN)
290 return;
291
292 dst = inet_csk_update_pmtu(sk, mtu);
293 if (!dst)
294 return;
295
296 /* Something is about to be wrong... Remember soft error
297 * for the case, if this connection will not able to recover.
298 */
299 if (mtu < dst_mtu(dst) && ip_dont_fragment(sk, dst))
300 sk->sk_err_soft = EMSGSIZE;
301
302 mtu = dst_mtu(dst);
303
304 if (inet->pmtudisc != IP_PMTUDISC_DONT &&
305 inet_csk(sk)->icsk_pmtu_cookie > mtu) {
306 tcp_sync_mss(sk, mtu);
307
308 /* Resend the TCP packet because it's
309 * clear that the old packet has been
310 * dropped. This is the new "fast" path mtu
311 * discovery.
312 */
313 tcp_simple_retransmit(sk);
314 } /* else let the usual retransmit timer handle it */
315 }
316
317 static void do_redirect(struct sk_buff *skb, struct sock *sk)
318 {
319 struct dst_entry *dst = __sk_dst_check(sk, 0);
320
321 if (dst)
322 dst->ops->redirect(dst, sk, skb);
323 }
324
325 /*
326 * This routine is called by the ICMP module when it gets some
327 * sort of error condition. If err < 0 then the socket should
328 * be closed and the error returned to the user. If err > 0
329 * it's just the icmp type << 8 | icmp code. After adjustment
330 * header points to the first 8 bytes of the tcp header. We need
331 * to find the appropriate port.
332 *
333 * The locking strategy used here is very "optimistic". When
334 * someone else accesses the socket the ICMP is just dropped
335 * and for some paths there is no check at all.
336 * A more general error queue to queue errors for later handling
337 * is probably better.
338 *
339 */
340
341 void tcp_v4_err(struct sk_buff *icmp_skb, u32 info)
342 {
343 const struct iphdr *iph = (const struct iphdr *)icmp_skb->data;
344 struct tcphdr *th = (struct tcphdr *)(icmp_skb->data + (iph->ihl << 2));
345 struct inet_connection_sock *icsk;
346 struct tcp_sock *tp;
347 struct inet_sock *inet;
348 const int type = icmp_hdr(icmp_skb)->type;
349 const int code = icmp_hdr(icmp_skb)->code;
350 struct sock *sk;
351 struct sk_buff *skb;
352 __u32 seq;
353 __u32 remaining;
354 int err;
355 struct net *net = dev_net(icmp_skb->dev);
356
357 if (icmp_skb->len < (iph->ihl << 2) + 8) {
358 ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS);
359 return;
360 }
361
362 sk = inet_lookup(net, &tcp_hashinfo, iph->daddr, th->dest,
363 iph->saddr, th->source, inet_iif(icmp_skb));
364 if (!sk) {
365 ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS);
366 return;
367 }
368 if (sk->sk_state == TCP_TIME_WAIT) {
369 inet_twsk_put(inet_twsk(sk));
370 return;
371 }
372
373 bh_lock_sock(sk);
374 /* If too many ICMPs get dropped on busy
375 * servers this needs to be solved differently.
376 */
377 if (sock_owned_by_user(sk))
378 NET_INC_STATS_BH(net, LINUX_MIB_LOCKDROPPEDICMPS);
379
380 if (sk->sk_state == TCP_CLOSE)
381 goto out;
382
383 if (unlikely(iph->ttl < inet_sk(sk)->min_ttl)) {
384 NET_INC_STATS_BH(net, LINUX_MIB_TCPMINTTLDROP);
385 goto out;
386 }
387
388 icsk = inet_csk(sk);
389 tp = tcp_sk(sk);
390 seq = ntohl(th->seq);
391 if (sk->sk_state != TCP_LISTEN &&
392 !between(seq, tp->snd_una, tp->snd_nxt)) {
393 NET_INC_STATS_BH(net, LINUX_MIB_OUTOFWINDOWICMPS);
394 goto out;
395 }
396
397 switch (type) {
398 case ICMP_REDIRECT:
399 do_redirect(icmp_skb, sk);
400 goto out;
401 case ICMP_SOURCE_QUENCH:
402 /* Just silently ignore these. */
403 goto out;
404 case ICMP_PARAMETERPROB:
405 err = EPROTO;
406 break;
407 case ICMP_DEST_UNREACH:
408 if (code > NR_ICMP_UNREACH)
409 goto out;
410
411 if (code == ICMP_FRAG_NEEDED) { /* PMTU discovery (RFC1191) */
412 if (!sock_owned_by_user(sk))
413 do_pmtu_discovery(sk, iph, info);
414 goto out;
415 }
416
417 err = icmp_err_convert[code].errno;
418 /* check if icmp_skb allows revert of backoff
419 * (see draft-zimmermann-tcp-lcd) */
420 if (code != ICMP_NET_UNREACH && code != ICMP_HOST_UNREACH)
421 break;
422 if (seq != tp->snd_una || !icsk->icsk_retransmits ||
423 !icsk->icsk_backoff)
424 break;
425
426 if (sock_owned_by_user(sk))
427 break;
428
429 icsk->icsk_backoff--;
430 inet_csk(sk)->icsk_rto = (tp->srtt ? __tcp_set_rto(tp) :
431 TCP_TIMEOUT_INIT) << icsk->icsk_backoff;
432 tcp_bound_rto(sk);
433
434 skb = tcp_write_queue_head(sk);
435 BUG_ON(!skb);
436
437 remaining = icsk->icsk_rto - min(icsk->icsk_rto,
438 tcp_time_stamp - TCP_SKB_CB(skb)->when);
439
440 if (remaining) {
441 inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
442 remaining, TCP_RTO_MAX);
443 } else {
444 /* RTO revert clocked out retransmission.
445 * Will retransmit now */
446 tcp_retransmit_timer(sk);
447 }
448
449 break;
450 case ICMP_TIME_EXCEEDED:
451 err = EHOSTUNREACH;
452 break;
453 default:
454 goto out;
455 }
456
457 switch (sk->sk_state) {
458 struct request_sock *req, **prev;
459 case TCP_LISTEN:
460 if (sock_owned_by_user(sk))
461 goto out;
462
463 req = inet_csk_search_req(sk, &prev, th->dest,
464 iph->daddr, iph->saddr);
465 if (!req)
466 goto out;
467
468 /* ICMPs are not backlogged, hence we cannot get
469 an established socket here.
470 */
471 WARN_ON(req->sk);
472
473 if (seq != tcp_rsk(req)->snt_isn) {
474 NET_INC_STATS_BH(net, LINUX_MIB_OUTOFWINDOWICMPS);
475 goto out;
476 }
477
478 /*
479 * Still in SYN_RECV, just remove it silently.
480 * There is no good way to pass the error to the newly
481 * created socket, and POSIX does not want network
482 * errors returned from accept().
483 */
484 inet_csk_reqsk_queue_drop(sk, req, prev);
485 goto out;
486
487 case TCP_SYN_SENT:
488 case TCP_SYN_RECV: /* Cannot happen.
489 It can f.e. if SYNs crossed.
490 */
491 if (!sock_owned_by_user(sk)) {
492 sk->sk_err = err;
493
494 sk->sk_error_report(sk);
495
496 tcp_done(sk);
497 } else {
498 sk->sk_err_soft = err;
499 }
500 goto out;
501 }
502
503 /* If we've already connected we will keep trying
504 * until we time out, or the user gives up.
505 *
506 * rfc1122 4.2.3.9 allows to consider as hard errors
507 * only PROTO_UNREACH and PORT_UNREACH (well, FRAG_FAILED too,
508 * but it is obsoleted by pmtu discovery).
509 *
510 * Note, that in modern internet, where routing is unreliable
511 * and in each dark corner broken firewalls sit, sending random
512 * errors ordered by their masters even this two messages finally lose
513 * their original sense (even Linux sends invalid PORT_UNREACHs)
514 *
515 * Now we are in compliance with RFCs.
516 * --ANK (980905)
517 */
518
519 inet = inet_sk(sk);
520 if (!sock_owned_by_user(sk) && inet->recverr) {
521 sk->sk_err = err;
522 sk->sk_error_report(sk);
523 } else { /* Only an error on timeout */
524 sk->sk_err_soft = err;
525 }
526
527 out:
528 bh_unlock_sock(sk);
529 sock_put(sk);
530 }
531
532 static void __tcp_v4_send_check(struct sk_buff *skb,
533 __be32 saddr, __be32 daddr)
534 {
535 struct tcphdr *th = tcp_hdr(skb);
536
537 if (skb->ip_summed == CHECKSUM_PARTIAL) {
538 th->check = ~tcp_v4_check(skb->len, saddr, daddr, 0);
539 skb->csum_start = skb_transport_header(skb) - skb->head;
540 skb->csum_offset = offsetof(struct tcphdr, check);
541 } else {
542 th->check = tcp_v4_check(skb->len, saddr, daddr,
543 csum_partial(th,
544 th->doff << 2,
545 skb->csum));
546 }
547 }
548
549 /* This routine computes an IPv4 TCP checksum. */
550 void tcp_v4_send_check(struct sock *sk, struct sk_buff *skb)
551 {
552 const struct inet_sock *inet = inet_sk(sk);
553
554 __tcp_v4_send_check(skb, inet->inet_saddr, inet->inet_daddr);
555 }
556 EXPORT_SYMBOL(tcp_v4_send_check);
557
558 int tcp_v4_gso_send_check(struct sk_buff *skb)
559 {
560 const struct iphdr *iph;
561 struct tcphdr *th;
562
563 if (!pskb_may_pull(skb, sizeof(*th)))
564 return -EINVAL;
565
566 iph = ip_hdr(skb);
567 th = tcp_hdr(skb);
568
569 th->check = 0;
570 skb->ip_summed = CHECKSUM_PARTIAL;
571 __tcp_v4_send_check(skb, iph->saddr, iph->daddr);
572 return 0;
573 }
574
575 /*
576 * This routine will send an RST to the other tcp.
577 *
578 * Someone asks: why I NEVER use socket parameters (TOS, TTL etc.)
579 * for reset.
580 * Answer: if a packet caused RST, it is not for a socket
581 * existing in our system, if it is matched to a socket,
582 * it is just duplicate segment or bug in other side's TCP.
583 * So that we build reply only basing on parameters
584 * arrived with segment.
585 * Exception: precedence violation. We do not implement it in any case.
586 */
587
588 static void tcp_v4_send_reset(struct sock *sk, struct sk_buff *skb)
589 {
590 const struct tcphdr *th = tcp_hdr(skb);
591 struct {
592 struct tcphdr th;
593 #ifdef CONFIG_TCP_MD5SIG
594 __be32 opt[(TCPOLEN_MD5SIG_ALIGNED >> 2)];
595 #endif
596 } rep;
597 struct ip_reply_arg arg;
598 #ifdef CONFIG_TCP_MD5SIG
599 struct tcp_md5sig_key *key;
600 const __u8 *hash_location = NULL;
601 unsigned char newhash[16];
602 int genhash;
603 struct sock *sk1 = NULL;
604 #endif
605 struct net *net;
606
607 /* Never send a reset in response to a reset. */
608 if (th->rst)
609 return;
610
611 if (skb_rtable(skb)->rt_type != RTN_LOCAL)
612 return;
613
614 /* Swap the send and the receive. */
615 memset(&rep, 0, sizeof(rep));
616 rep.th.dest = th->source;
617 rep.th.source = th->dest;
618 rep.th.doff = sizeof(struct tcphdr) / 4;
619 rep.th.rst = 1;
620
621 if (th->ack) {
622 rep.th.seq = th->ack_seq;
623 } else {
624 rep.th.ack = 1;
625 rep.th.ack_seq = htonl(ntohl(th->seq) + th->syn + th->fin +
626 skb->len - (th->doff << 2));
627 }
628
629 memset(&arg, 0, sizeof(arg));
630 arg.iov[0].iov_base = (unsigned char *)&rep;
631 arg.iov[0].iov_len = sizeof(rep.th);
632
633 #ifdef CONFIG_TCP_MD5SIG
634 hash_location = tcp_parse_md5sig_option(th);
635 if (!sk && hash_location) {
636 /*
637 * active side is lost. Try to find listening socket through
638 * source port, and then find md5 key through listening socket.
639 * we are not loose security here:
640 * Incoming packet is checked with md5 hash with finding key,
641 * no RST generated if md5 hash doesn't match.
642 */
643 sk1 = __inet_lookup_listener(dev_net(skb_dst(skb)->dev),
644 &tcp_hashinfo, ip_hdr(skb)->daddr,
645 ntohs(th->source), inet_iif(skb));
646 /* don't send rst if it can't find key */
647 if (!sk1)
648 return;
649 rcu_read_lock();
650 key = tcp_md5_do_lookup(sk1, (union tcp_md5_addr *)
651 &ip_hdr(skb)->saddr, AF_INET);
652 if (!key)
653 goto release_sk1;
654
655 genhash = tcp_v4_md5_hash_skb(newhash, key, NULL, NULL, skb);
656 if (genhash || memcmp(hash_location, newhash, 16) != 0)
657 goto release_sk1;
658 } else {
659 key = sk ? tcp_md5_do_lookup(sk, (union tcp_md5_addr *)
660 &ip_hdr(skb)->saddr,
661 AF_INET) : NULL;
662 }
663
664 if (key) {
665 rep.opt[0] = htonl((TCPOPT_NOP << 24) |
666 (TCPOPT_NOP << 16) |
667 (TCPOPT_MD5SIG << 8) |
668 TCPOLEN_MD5SIG);
669 /* Update length and the length the header thinks exists */
670 arg.iov[0].iov_len += TCPOLEN_MD5SIG_ALIGNED;
671 rep.th.doff = arg.iov[0].iov_len / 4;
672
673 tcp_v4_md5_hash_hdr((__u8 *) &rep.opt[1],
674 key, ip_hdr(skb)->saddr,
675 ip_hdr(skb)->daddr, &rep.th);
676 }
677 #endif
678 arg.csum = csum_tcpudp_nofold(ip_hdr(skb)->daddr,
679 ip_hdr(skb)->saddr, /* XXX */
680 arg.iov[0].iov_len, IPPROTO_TCP, 0);
681 arg.csumoffset = offsetof(struct tcphdr, check) / 2;
682 arg.flags = (sk && inet_sk(sk)->transparent) ? IP_REPLY_ARG_NOSRCCHECK : 0;
683 /* When socket is gone, all binding information is lost.
684 * routing might fail in this case. using iif for oif to
685 * make sure we can deliver it
686 */
687 arg.bound_dev_if = sk ? sk->sk_bound_dev_if : inet_iif(skb);
688
689 net = dev_net(skb_dst(skb)->dev);
690 arg.tos = ip_hdr(skb)->tos;
691 ip_send_unicast_reply(net, skb, ip_hdr(skb)->saddr,
692 ip_hdr(skb)->daddr, &arg, arg.iov[0].iov_len);
693
694 TCP_INC_STATS_BH(net, TCP_MIB_OUTSEGS);
695 TCP_INC_STATS_BH(net, TCP_MIB_OUTRSTS);
696
697 #ifdef CONFIG_TCP_MD5SIG
698 release_sk1:
699 if (sk1) {
700 rcu_read_unlock();
701 sock_put(sk1);
702 }
703 #endif
704 }
705
706 /* The code following below sending ACKs in SYN-RECV and TIME-WAIT states
707 outside socket context is ugly, certainly. What can I do?
708 */
709
710 static void tcp_v4_send_ack(struct sk_buff *skb, u32 seq, u32 ack,
711 u32 win, u32 ts, int oif,
712 struct tcp_md5sig_key *key,
713 int reply_flags, u8 tos)
714 {
715 const struct tcphdr *th = tcp_hdr(skb);
716 struct {
717 struct tcphdr th;
718 __be32 opt[(TCPOLEN_TSTAMP_ALIGNED >> 2)
719 #ifdef CONFIG_TCP_MD5SIG
720 + (TCPOLEN_MD5SIG_ALIGNED >> 2)
721 #endif
722 ];
723 } rep;
724 struct ip_reply_arg arg;
725 struct net *net = dev_net(skb_dst(skb)->dev);
726
727 memset(&rep.th, 0, sizeof(struct tcphdr));
728 memset(&arg, 0, sizeof(arg));
729
730 arg.iov[0].iov_base = (unsigned char *)&rep;
731 arg.iov[0].iov_len = sizeof(rep.th);
732 if (ts) {
733 rep.opt[0] = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
734 (TCPOPT_TIMESTAMP << 8) |
735 TCPOLEN_TIMESTAMP);
736 rep.opt[1] = htonl(tcp_time_stamp);
737 rep.opt[2] = htonl(ts);
738 arg.iov[0].iov_len += TCPOLEN_TSTAMP_ALIGNED;
739 }
740
741 /* Swap the send and the receive. */
742 rep.th.dest = th->source;
743 rep.th.source = th->dest;
744 rep.th.doff = arg.iov[0].iov_len / 4;
745 rep.th.seq = htonl(seq);
746 rep.th.ack_seq = htonl(ack);
747 rep.th.ack = 1;
748 rep.th.window = htons(win);
749
750 #ifdef CONFIG_TCP_MD5SIG
751 if (key) {
752 int offset = (ts) ? 3 : 0;
753
754 rep.opt[offset++] = htonl((TCPOPT_NOP << 24) |
755 (TCPOPT_NOP << 16) |
756 (TCPOPT_MD5SIG << 8) |
757 TCPOLEN_MD5SIG);
758 arg.iov[0].iov_len += TCPOLEN_MD5SIG_ALIGNED;
759 rep.th.doff = arg.iov[0].iov_len/4;
760
761 tcp_v4_md5_hash_hdr((__u8 *) &rep.opt[offset],
762 key, ip_hdr(skb)->saddr,
763 ip_hdr(skb)->daddr, &rep.th);
764 }
765 #endif
766 arg.flags = reply_flags;
767 arg.csum = csum_tcpudp_nofold(ip_hdr(skb)->daddr,
768 ip_hdr(skb)->saddr, /* XXX */
769 arg.iov[0].iov_len, IPPROTO_TCP, 0);
770 arg.csumoffset = offsetof(struct tcphdr, check) / 2;
771 if (oif)
772 arg.bound_dev_if = oif;
773 arg.tos = tos;
774 ip_send_unicast_reply(net, skb, ip_hdr(skb)->saddr,
775 ip_hdr(skb)->daddr, &arg, arg.iov[0].iov_len);
776
777 TCP_INC_STATS_BH(net, TCP_MIB_OUTSEGS);
778 }
779
780 static void tcp_v4_timewait_ack(struct sock *sk, struct sk_buff *skb)
781 {
782 struct inet_timewait_sock *tw = inet_twsk(sk);
783 struct tcp_timewait_sock *tcptw = tcp_twsk(sk);
784
785 tcp_v4_send_ack(skb, tcptw->tw_snd_nxt, tcptw->tw_rcv_nxt,
786 tcptw->tw_rcv_wnd >> tw->tw_rcv_wscale,
787 tcptw->tw_ts_recent,
788 tw->tw_bound_dev_if,
789 tcp_twsk_md5_key(tcptw),
790 tw->tw_transparent ? IP_REPLY_ARG_NOSRCCHECK : 0,
791 tw->tw_tos
792 );
793
794 inet_twsk_put(tw);
795 }
796
797 static void tcp_v4_reqsk_send_ack(struct sock *sk, struct sk_buff *skb,
798 struct request_sock *req)
799 {
800 tcp_v4_send_ack(skb, tcp_rsk(req)->snt_isn + 1,
801 tcp_rsk(req)->rcv_isn + 1, req->rcv_wnd,
802 req->ts_recent,
803 0,
804 tcp_md5_do_lookup(sk, (union tcp_md5_addr *)&ip_hdr(skb)->daddr,
805 AF_INET),
806 inet_rsk(req)->no_srccheck ? IP_REPLY_ARG_NOSRCCHECK : 0,
807 ip_hdr(skb)->tos);
808 }
809
810 /*
811 * Send a SYN-ACK after having received a SYN.
812 * This still operates on a request_sock only, not on a big
813 * socket.
814 */
815 static int tcp_v4_send_synack(struct sock *sk, struct dst_entry *dst,
816 struct request_sock *req,
817 struct request_values *rvp,
818 u16 queue_mapping,
819 bool nocache)
820 {
821 const struct inet_request_sock *ireq = inet_rsk(req);
822 struct flowi4 fl4;
823 int err = -1;
824 struct sk_buff * skb;
825
826 /* First, grab a route. */
827 if (!dst && (dst = inet_csk_route_req(sk, &fl4, req, nocache)) == NULL)
828 return -1;
829
830 skb = tcp_make_synack(sk, dst, req, rvp);
831
832 if (skb) {
833 __tcp_v4_send_check(skb, ireq->loc_addr, ireq->rmt_addr);
834
835 skb_set_queue_mapping(skb, queue_mapping);
836 err = ip_build_and_send_pkt(skb, sk, ireq->loc_addr,
837 ireq->rmt_addr,
838 ireq->opt);
839 err = net_xmit_eval(err);
840 }
841
842 return err;
843 }
844
845 static int tcp_v4_rtx_synack(struct sock *sk, struct request_sock *req,
846 struct request_values *rvp)
847 {
848 TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_RETRANSSEGS);
849 return tcp_v4_send_synack(sk, NULL, req, rvp, 0, false);
850 }
851
852 /*
853 * IPv4 request_sock destructor.
854 */
855 static void tcp_v4_reqsk_destructor(struct request_sock *req)
856 {
857 kfree(inet_rsk(req)->opt);
858 }
859
860 /*
861 * Return true if a syncookie should be sent
862 */
863 bool tcp_syn_flood_action(struct sock *sk,
864 const struct sk_buff *skb,
865 const char *proto)
866 {
867 const char *msg = "Dropping request";
868 bool want_cookie = false;
869 struct listen_sock *lopt;
870
871
872
873 #ifdef CONFIG_SYN_COOKIES
874 if (sysctl_tcp_syncookies) {
875 msg = "Sending cookies";
876 want_cookie = true;
877 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPREQQFULLDOCOOKIES);
878 } else
879 #endif
880 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPREQQFULLDROP);
881
882 lopt = inet_csk(sk)->icsk_accept_queue.listen_opt;
883 if (!lopt->synflood_warned) {
884 lopt->synflood_warned = 1;
885 pr_info("%s: Possible SYN flooding on port %d. %s. Check SNMP counters.\n",
886 proto, ntohs(tcp_hdr(skb)->dest), msg);
887 }
888 return want_cookie;
889 }
890 EXPORT_SYMBOL(tcp_syn_flood_action);
891
892 /*
893 * Save and compile IPv4 options into the request_sock if needed.
894 */
895 static struct ip_options_rcu *tcp_v4_save_options(struct sock *sk,
896 struct sk_buff *skb)
897 {
898 const struct ip_options *opt = &(IPCB(skb)->opt);
899 struct ip_options_rcu *dopt = NULL;
900
901 if (opt && opt->optlen) {
902 int opt_size = sizeof(*dopt) + opt->optlen;
903
904 dopt = kmalloc(opt_size, GFP_ATOMIC);
905 if (dopt) {
906 if (ip_options_echo(&dopt->opt, skb)) {
907 kfree(dopt);
908 dopt = NULL;
909 }
910 }
911 }
912 return dopt;
913 }
914
915 #ifdef CONFIG_TCP_MD5SIG
916 /*
917 * RFC2385 MD5 checksumming requires a mapping of
918 * IP address->MD5 Key.
919 * We need to maintain these in the sk structure.
920 */
921
922 /* Find the Key structure for an address. */
923 struct tcp_md5sig_key *tcp_md5_do_lookup(struct sock *sk,
924 const union tcp_md5_addr *addr,
925 int family)
926 {
927 struct tcp_sock *tp = tcp_sk(sk);
928 struct tcp_md5sig_key *key;
929 struct hlist_node *pos;
930 unsigned int size = sizeof(struct in_addr);
931 struct tcp_md5sig_info *md5sig;
932
933 /* caller either holds rcu_read_lock() or socket lock */
934 md5sig = rcu_dereference_check(tp->md5sig_info,
935 sock_owned_by_user(sk) ||
936 lockdep_is_held(&sk->sk_lock.slock));
937 if (!md5sig)
938 return NULL;
939 #if IS_ENABLED(CONFIG_IPV6)
940 if (family == AF_INET6)
941 size = sizeof(struct in6_addr);
942 #endif
943 hlist_for_each_entry_rcu(key, pos, &md5sig->head, node) {
944 if (key->family != family)
945 continue;
946 if (!memcmp(&key->addr, addr, size))
947 return key;
948 }
949 return NULL;
950 }
951 EXPORT_SYMBOL(tcp_md5_do_lookup);
952
953 struct tcp_md5sig_key *tcp_v4_md5_lookup(struct sock *sk,
954 struct sock *addr_sk)
955 {
956 union tcp_md5_addr *addr;
957
958 addr = (union tcp_md5_addr *)&inet_sk(addr_sk)->inet_daddr;
959 return tcp_md5_do_lookup(sk, addr, AF_INET);
960 }
961 EXPORT_SYMBOL(tcp_v4_md5_lookup);
962
963 static struct tcp_md5sig_key *tcp_v4_reqsk_md5_lookup(struct sock *sk,
964 struct request_sock *req)
965 {
966 union tcp_md5_addr *addr;
967
968 addr = (union tcp_md5_addr *)&inet_rsk(req)->rmt_addr;
969 return tcp_md5_do_lookup(sk, addr, AF_INET);
970 }
971
972 /* This can be called on a newly created socket, from other files */
973 int tcp_md5_do_add(struct sock *sk, const union tcp_md5_addr *addr,
974 int family, const u8 *newkey, u8 newkeylen, gfp_t gfp)
975 {
976 /* Add Key to the list */
977 struct tcp_md5sig_key *key;
978 struct tcp_sock *tp = tcp_sk(sk);
979 struct tcp_md5sig_info *md5sig;
980
981 key = tcp_md5_do_lookup(sk, (union tcp_md5_addr *)&addr, AF_INET);
982 if (key) {
983 /* Pre-existing entry - just update that one. */
984 memcpy(key->key, newkey, newkeylen);
985 key->keylen = newkeylen;
986 return 0;
987 }
988
989 md5sig = rcu_dereference_protected(tp->md5sig_info,
990 sock_owned_by_user(sk));
991 if (!md5sig) {
992 md5sig = kmalloc(sizeof(*md5sig), gfp);
993 if (!md5sig)
994 return -ENOMEM;
995
996 sk_nocaps_add(sk, NETIF_F_GSO_MASK);
997 INIT_HLIST_HEAD(&md5sig->head);
998 rcu_assign_pointer(tp->md5sig_info, md5sig);
999 }
1000
1001 key = sock_kmalloc(sk, sizeof(*key), gfp);
1002 if (!key)
1003 return -ENOMEM;
1004 if (hlist_empty(&md5sig->head) && !tcp_alloc_md5sig_pool(sk)) {
1005 sock_kfree_s(sk, key, sizeof(*key));
1006 return -ENOMEM;
1007 }
1008
1009 memcpy(key->key, newkey, newkeylen);
1010 key->keylen = newkeylen;
1011 key->family = family;
1012 memcpy(&key->addr, addr,
1013 (family == AF_INET6) ? sizeof(struct in6_addr) :
1014 sizeof(struct in_addr));
1015 hlist_add_head_rcu(&key->node, &md5sig->head);
1016 return 0;
1017 }
1018 EXPORT_SYMBOL(tcp_md5_do_add);
1019
1020 int tcp_md5_do_del(struct sock *sk, const union tcp_md5_addr *addr, int family)
1021 {
1022 struct tcp_sock *tp = tcp_sk(sk);
1023 struct tcp_md5sig_key *key;
1024 struct tcp_md5sig_info *md5sig;
1025
1026 key = tcp_md5_do_lookup(sk, (union tcp_md5_addr *)&addr, AF_INET);
1027 if (!key)
1028 return -ENOENT;
1029 hlist_del_rcu(&key->node);
1030 atomic_sub(sizeof(*key), &sk->sk_omem_alloc);
1031 kfree_rcu(key, rcu);
1032 md5sig = rcu_dereference_protected(tp->md5sig_info,
1033 sock_owned_by_user(sk));
1034 if (hlist_empty(&md5sig->head))
1035 tcp_free_md5sig_pool();
1036 return 0;
1037 }
1038 EXPORT_SYMBOL(tcp_md5_do_del);
1039
1040 void tcp_clear_md5_list(struct sock *sk)
1041 {
1042 struct tcp_sock *tp = tcp_sk(sk);
1043 struct tcp_md5sig_key *key;
1044 struct hlist_node *pos, *n;
1045 struct tcp_md5sig_info *md5sig;
1046
1047 md5sig = rcu_dereference_protected(tp->md5sig_info, 1);
1048
1049 if (!hlist_empty(&md5sig->head))
1050 tcp_free_md5sig_pool();
1051 hlist_for_each_entry_safe(key, pos, n, &md5sig->head, node) {
1052 hlist_del_rcu(&key->node);
1053 atomic_sub(sizeof(*key), &sk->sk_omem_alloc);
1054 kfree_rcu(key, rcu);
1055 }
1056 }
1057
1058 static int tcp_v4_parse_md5_keys(struct sock *sk, char __user *optval,
1059 int optlen)
1060 {
1061 struct tcp_md5sig cmd;
1062 struct sockaddr_in *sin = (struct sockaddr_in *)&cmd.tcpm_addr;
1063
1064 if (optlen < sizeof(cmd))
1065 return -EINVAL;
1066
1067 if (copy_from_user(&cmd, optval, sizeof(cmd)))
1068 return -EFAULT;
1069
1070 if (sin->sin_family != AF_INET)
1071 return -EINVAL;
1072
1073 if (!cmd.tcpm_key || !cmd.tcpm_keylen)
1074 return tcp_md5_do_del(sk, (union tcp_md5_addr *)&sin->sin_addr.s_addr,
1075 AF_INET);
1076
1077 if (cmd.tcpm_keylen > TCP_MD5SIG_MAXKEYLEN)
1078 return -EINVAL;
1079
1080 return tcp_md5_do_add(sk, (union tcp_md5_addr *)&sin->sin_addr.s_addr,
1081 AF_INET, cmd.tcpm_key, cmd.tcpm_keylen,
1082 GFP_KERNEL);
1083 }
1084
1085 static int tcp_v4_md5_hash_pseudoheader(struct tcp_md5sig_pool *hp,
1086 __be32 daddr, __be32 saddr, int nbytes)
1087 {
1088 struct tcp4_pseudohdr *bp;
1089 struct scatterlist sg;
1090
1091 bp = &hp->md5_blk.ip4;
1092
1093 /*
1094 * 1. the TCP pseudo-header (in the order: source IP address,
1095 * destination IP address, zero-padded protocol number, and
1096 * segment length)
1097 */
1098 bp->saddr = saddr;
1099 bp->daddr = daddr;
1100 bp->pad = 0;
1101 bp->protocol = IPPROTO_TCP;
1102 bp->len = cpu_to_be16(nbytes);
1103
1104 sg_init_one(&sg, bp, sizeof(*bp));
1105 return crypto_hash_update(&hp->md5_desc, &sg, sizeof(*bp));
1106 }
1107
1108 static int tcp_v4_md5_hash_hdr(char *md5_hash, const struct tcp_md5sig_key *key,
1109 __be32 daddr, __be32 saddr, const struct tcphdr *th)
1110 {
1111 struct tcp_md5sig_pool *hp;
1112 struct hash_desc *desc;
1113
1114 hp = tcp_get_md5sig_pool();
1115 if (!hp)
1116 goto clear_hash_noput;
1117 desc = &hp->md5_desc;
1118
1119 if (crypto_hash_init(desc))
1120 goto clear_hash;
1121 if (tcp_v4_md5_hash_pseudoheader(hp, daddr, saddr, th->doff << 2))
1122 goto clear_hash;
1123 if (tcp_md5_hash_header(hp, th))
1124 goto clear_hash;
1125 if (tcp_md5_hash_key(hp, key))
1126 goto clear_hash;
1127 if (crypto_hash_final(desc, md5_hash))
1128 goto clear_hash;
1129
1130 tcp_put_md5sig_pool();
1131 return 0;
1132
1133 clear_hash:
1134 tcp_put_md5sig_pool();
1135 clear_hash_noput:
1136 memset(md5_hash, 0, 16);
1137 return 1;
1138 }
1139
1140 int tcp_v4_md5_hash_skb(char *md5_hash, struct tcp_md5sig_key *key,
1141 const struct sock *sk, const struct request_sock *req,
1142 const struct sk_buff *skb)
1143 {
1144 struct tcp_md5sig_pool *hp;
1145 struct hash_desc *desc;
1146 const struct tcphdr *th = tcp_hdr(skb);
1147 __be32 saddr, daddr;
1148
1149 if (sk) {
1150 saddr = inet_sk(sk)->inet_saddr;
1151 daddr = inet_sk(sk)->inet_daddr;
1152 } else if (req) {
1153 saddr = inet_rsk(req)->loc_addr;
1154 daddr = inet_rsk(req)->rmt_addr;
1155 } else {
1156 const struct iphdr *iph = ip_hdr(skb);
1157 saddr = iph->saddr;
1158 daddr = iph->daddr;
1159 }
1160
1161 hp = tcp_get_md5sig_pool();
1162 if (!hp)
1163 goto clear_hash_noput;
1164 desc = &hp->md5_desc;
1165
1166 if (crypto_hash_init(desc))
1167 goto clear_hash;
1168
1169 if (tcp_v4_md5_hash_pseudoheader(hp, daddr, saddr, skb->len))
1170 goto clear_hash;
1171 if (tcp_md5_hash_header(hp, th))
1172 goto clear_hash;
1173 if (tcp_md5_hash_skb_data(hp, skb, th->doff << 2))
1174 goto clear_hash;
1175 if (tcp_md5_hash_key(hp, key))
1176 goto clear_hash;
1177 if (crypto_hash_final(desc, md5_hash))
1178 goto clear_hash;
1179
1180 tcp_put_md5sig_pool();
1181 return 0;
1182
1183 clear_hash:
1184 tcp_put_md5sig_pool();
1185 clear_hash_noput:
1186 memset(md5_hash, 0, 16);
1187 return 1;
1188 }
1189 EXPORT_SYMBOL(tcp_v4_md5_hash_skb);
1190
1191 static bool tcp_v4_inbound_md5_hash(struct sock *sk, const struct sk_buff *skb)
1192 {
1193 /*
1194 * This gets called for each TCP segment that arrives
1195 * so we want to be efficient.
1196 * We have 3 drop cases:
1197 * o No MD5 hash and one expected.
1198 * o MD5 hash and we're not expecting one.
1199 * o MD5 hash and its wrong.
1200 */
1201 const __u8 *hash_location = NULL;
1202 struct tcp_md5sig_key *hash_expected;
1203 const struct iphdr *iph = ip_hdr(skb);
1204 const struct tcphdr *th = tcp_hdr(skb);
1205 int genhash;
1206 unsigned char newhash[16];
1207
1208 hash_expected = tcp_md5_do_lookup(sk, (union tcp_md5_addr *)&iph->saddr,
1209 AF_INET);
1210 hash_location = tcp_parse_md5sig_option(th);
1211
1212 /* We've parsed the options - do we have a hash? */
1213 if (!hash_expected && !hash_location)
1214 return false;
1215
1216 if (hash_expected && !hash_location) {
1217 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPMD5NOTFOUND);
1218 return true;
1219 }
1220
1221 if (!hash_expected && hash_location) {
1222 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPMD5UNEXPECTED);
1223 return true;
1224 }
1225
1226 /* Okay, so this is hash_expected and hash_location -
1227 * so we need to calculate the checksum.
1228 */
1229 genhash = tcp_v4_md5_hash_skb(newhash,
1230 hash_expected,
1231 NULL, NULL, skb);
1232
1233 if (genhash || memcmp(hash_location, newhash, 16) != 0) {
1234 net_info_ratelimited("MD5 Hash failed for (%pI4, %d)->(%pI4, %d)%s\n",
1235 &iph->saddr, ntohs(th->source),
1236 &iph->daddr, ntohs(th->dest),
1237 genhash ? " tcp_v4_calc_md5_hash failed"
1238 : "");
1239 return true;
1240 }
1241 return false;
1242 }
1243
1244 #endif
1245
1246 struct request_sock_ops tcp_request_sock_ops __read_mostly = {
1247 .family = PF_INET,
1248 .obj_size = sizeof(struct tcp_request_sock),
1249 .rtx_syn_ack = tcp_v4_rtx_synack,
1250 .send_ack = tcp_v4_reqsk_send_ack,
1251 .destructor = tcp_v4_reqsk_destructor,
1252 .send_reset = tcp_v4_send_reset,
1253 .syn_ack_timeout = tcp_syn_ack_timeout,
1254 };
1255
1256 #ifdef CONFIG_TCP_MD5SIG
1257 static const struct tcp_request_sock_ops tcp_request_sock_ipv4_ops = {
1258 .md5_lookup = tcp_v4_reqsk_md5_lookup,
1259 .calc_md5_hash = tcp_v4_md5_hash_skb,
1260 };
1261 #endif
1262
1263 int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb)
1264 {
1265 struct tcp_extend_values tmp_ext;
1266 struct tcp_options_received tmp_opt;
1267 const u8 *hash_location;
1268 struct request_sock *req;
1269 struct inet_request_sock *ireq;
1270 struct tcp_sock *tp = tcp_sk(sk);
1271 struct dst_entry *dst = NULL;
1272 __be32 saddr = ip_hdr(skb)->saddr;
1273 __be32 daddr = ip_hdr(skb)->daddr;
1274 __u32 isn = TCP_SKB_CB(skb)->when;
1275 bool want_cookie = false;
1276
1277 /* Never answer to SYNs send to broadcast or multicast */
1278 if (skb_rtable(skb)->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))
1279 goto drop;
1280
1281 /* TW buckets are converted to open requests without
1282 * limitations, they conserve resources and peer is
1283 * evidently real one.
1284 */
1285 if (inet_csk_reqsk_queue_is_full(sk) && !isn) {
1286 want_cookie = tcp_syn_flood_action(sk, skb, "TCP");
1287 if (!want_cookie)
1288 goto drop;
1289 }
1290
1291 /* Accept backlog is full. If we have already queued enough
1292 * of warm entries in syn queue, drop request. It is better than
1293 * clogging syn queue with openreqs with exponentially increasing
1294 * timeout.
1295 */
1296 if (sk_acceptq_is_full(sk) && inet_csk_reqsk_queue_young(sk) > 1)
1297 goto drop;
1298
1299 req = inet_reqsk_alloc(&tcp_request_sock_ops);
1300 if (!req)
1301 goto drop;
1302
1303 #ifdef CONFIG_TCP_MD5SIG
1304 tcp_rsk(req)->af_specific = &tcp_request_sock_ipv4_ops;
1305 #endif
1306
1307 tcp_clear_options(&tmp_opt);
1308 tmp_opt.mss_clamp = TCP_MSS_DEFAULT;
1309 tmp_opt.user_mss = tp->rx_opt.user_mss;
1310 tcp_parse_options(skb, &tmp_opt, &hash_location, 0);
1311
1312 if (tmp_opt.cookie_plus > 0 &&
1313 tmp_opt.saw_tstamp &&
1314 !tp->rx_opt.cookie_out_never &&
1315 (sysctl_tcp_cookie_size > 0 ||
1316 (tp->cookie_values != NULL &&
1317 tp->cookie_values->cookie_desired > 0))) {
1318 u8 *c;
1319 u32 *mess = &tmp_ext.cookie_bakery[COOKIE_DIGEST_WORDS];
1320 int l = tmp_opt.cookie_plus - TCPOLEN_COOKIE_BASE;
1321
1322 if (tcp_cookie_generator(&tmp_ext.cookie_bakery[0]) != 0)
1323 goto drop_and_release;
1324
1325 /* Secret recipe starts with IP addresses */
1326 *mess++ ^= (__force u32)daddr;
1327 *mess++ ^= (__force u32)saddr;
1328
1329 /* plus variable length Initiator Cookie */
1330 c = (u8 *)mess;
1331 while (l-- > 0)
1332 *c++ ^= *hash_location++;
1333
1334 want_cookie = false; /* not our kind of cookie */
1335 tmp_ext.cookie_out_never = 0; /* false */
1336 tmp_ext.cookie_plus = tmp_opt.cookie_plus;
1337 } else if (!tp->rx_opt.cookie_in_always) {
1338 /* redundant indications, but ensure initialization. */
1339 tmp_ext.cookie_out_never = 1; /* true */
1340 tmp_ext.cookie_plus = 0;
1341 } else {
1342 goto drop_and_release;
1343 }
1344 tmp_ext.cookie_in_always = tp->rx_opt.cookie_in_always;
1345
1346 if (want_cookie && !tmp_opt.saw_tstamp)
1347 tcp_clear_options(&tmp_opt);
1348
1349 tmp_opt.tstamp_ok = tmp_opt.saw_tstamp;
1350 tcp_openreq_init(req, &tmp_opt, skb);
1351
1352 ireq = inet_rsk(req);
1353 ireq->loc_addr = daddr;
1354 ireq->rmt_addr = saddr;
1355 ireq->no_srccheck = inet_sk(sk)->transparent;
1356 ireq->opt = tcp_v4_save_options(sk, skb);
1357
1358 if (security_inet_conn_request(sk, skb, req))
1359 goto drop_and_free;
1360
1361 if (!want_cookie || tmp_opt.tstamp_ok)
1362 TCP_ECN_create_request(req, skb);
1363
1364 if (want_cookie) {
1365 isn = cookie_v4_init_sequence(sk, skb, &req->mss);
1366 req->cookie_ts = tmp_opt.tstamp_ok;
1367 } else if (!isn) {
1368 struct flowi4 fl4;
1369
1370 /* VJ's idea. We save last timestamp seen
1371 * from the destination in peer table, when entering
1372 * state TIME-WAIT, and check against it before
1373 * accepting new connection request.
1374 *
1375 * If "isn" is not zero, this request hit alive
1376 * timewait bucket, so that all the necessary checks
1377 * are made in the function processing timewait state.
1378 */
1379 if (tmp_opt.saw_tstamp &&
1380 tcp_death_row.sysctl_tw_recycle &&
1381 (dst = inet_csk_route_req(sk, &fl4, req, want_cookie)) != NULL &&
1382 fl4.daddr == saddr) {
1383 if (!tcp_peer_is_proven(req, dst, true)) {
1384 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_PAWSPASSIVEREJECTED);
1385 goto drop_and_release;
1386 }
1387 }
1388 /* Kill the following clause, if you dislike this way. */
1389 else if (!sysctl_tcp_syncookies &&
1390 (sysctl_max_syn_backlog - inet_csk_reqsk_queue_len(sk) <
1391 (sysctl_max_syn_backlog >> 2)) &&
1392 !tcp_peer_is_proven(req, dst, false)) {
1393 /* Without syncookies last quarter of
1394 * backlog is filled with destinations,
1395 * proven to be alive.
1396 * It means that we continue to communicate
1397 * to destinations, already remembered
1398 * to the moment of synflood.
1399 */
1400 LIMIT_NETDEBUG(KERN_DEBUG pr_fmt("drop open request from %pI4/%u\n"),
1401 &saddr, ntohs(tcp_hdr(skb)->source));
1402 goto drop_and_release;
1403 }
1404
1405 isn = tcp_v4_init_sequence(skb);
1406 }
1407 tcp_rsk(req)->snt_isn = isn;
1408 tcp_rsk(req)->snt_synack = tcp_time_stamp;
1409
1410 if (tcp_v4_send_synack(sk, dst, req,
1411 (struct request_values *)&tmp_ext,
1412 skb_get_queue_mapping(skb),
1413 want_cookie) ||
1414 want_cookie)
1415 goto drop_and_free;
1416
1417 inet_csk_reqsk_queue_hash_add(sk, req, TCP_TIMEOUT_INIT);
1418 return 0;
1419
1420 drop_and_release:
1421 dst_release(dst);
1422 drop_and_free:
1423 reqsk_free(req);
1424 drop:
1425 return 0;
1426 }
1427 EXPORT_SYMBOL(tcp_v4_conn_request);
1428
1429
1430 /*
1431 * The three way handshake has completed - we got a valid synack -
1432 * now create the new socket.
1433 */
1434 struct sock *tcp_v4_syn_recv_sock(struct sock *sk, struct sk_buff *skb,
1435 struct request_sock *req,
1436 struct dst_entry *dst)
1437 {
1438 struct inet_request_sock *ireq;
1439 struct inet_sock *newinet;
1440 struct tcp_sock *newtp;
1441 struct sock *newsk;
1442 #ifdef CONFIG_TCP_MD5SIG
1443 struct tcp_md5sig_key *key;
1444 #endif
1445 struct ip_options_rcu *inet_opt;
1446
1447 if (sk_acceptq_is_full(sk))
1448 goto exit_overflow;
1449
1450 newsk = tcp_create_openreq_child(sk, req, skb);
1451 if (!newsk)
1452 goto exit_nonewsk;
1453
1454 newsk->sk_gso_type = SKB_GSO_TCPV4;
1455
1456 newtp = tcp_sk(newsk);
1457 newinet = inet_sk(newsk);
1458 ireq = inet_rsk(req);
1459 newinet->inet_daddr = ireq->rmt_addr;
1460 newinet->inet_rcv_saddr = ireq->loc_addr;
1461 newinet->inet_saddr = ireq->loc_addr;
1462 inet_opt = ireq->opt;
1463 rcu_assign_pointer(newinet->inet_opt, inet_opt);
1464 ireq->opt = NULL;
1465 newinet->mc_index = inet_iif(skb);
1466 newinet->mc_ttl = ip_hdr(skb)->ttl;
1467 newinet->rcv_tos = ip_hdr(skb)->tos;
1468 inet_csk(newsk)->icsk_ext_hdr_len = 0;
1469 if (inet_opt)
1470 inet_csk(newsk)->icsk_ext_hdr_len = inet_opt->opt.optlen;
1471 newinet->inet_id = newtp->write_seq ^ jiffies;
1472
1473 if (!dst) {
1474 dst = inet_csk_route_child_sock(sk, newsk, req);
1475 if (!dst)
1476 goto put_and_exit;
1477 } else {
1478 /* syncookie case : see end of cookie_v4_check() */
1479 }
1480 sk_setup_caps(newsk, dst);
1481
1482 tcp_mtup_init(newsk);
1483 tcp_sync_mss(newsk, dst_mtu(dst));
1484 newtp->advmss = dst_metric_advmss(dst);
1485 if (tcp_sk(sk)->rx_opt.user_mss &&
1486 tcp_sk(sk)->rx_opt.user_mss < newtp->advmss)
1487 newtp->advmss = tcp_sk(sk)->rx_opt.user_mss;
1488
1489 tcp_initialize_rcv_mss(newsk);
1490 if (tcp_rsk(req)->snt_synack)
1491 tcp_valid_rtt_meas(newsk,
1492 tcp_time_stamp - tcp_rsk(req)->snt_synack);
1493 newtp->total_retrans = req->retrans;
1494
1495 #ifdef CONFIG_TCP_MD5SIG
1496 /* Copy over the MD5 key from the original socket */
1497 key = tcp_md5_do_lookup(sk, (union tcp_md5_addr *)&newinet->inet_daddr,
1498 AF_INET);
1499 if (key != NULL) {
1500 /*
1501 * We're using one, so create a matching key
1502 * on the newsk structure. If we fail to get
1503 * memory, then we end up not copying the key
1504 * across. Shucks.
1505 */
1506 tcp_md5_do_add(newsk, (union tcp_md5_addr *)&newinet->inet_daddr,
1507 AF_INET, key->key, key->keylen, GFP_ATOMIC);
1508 sk_nocaps_add(newsk, NETIF_F_GSO_MASK);
1509 }
1510 #endif
1511
1512 if (__inet_inherit_port(sk, newsk) < 0)
1513 goto put_and_exit;
1514 __inet_hash_nolisten(newsk, NULL);
1515
1516 return newsk;
1517
1518 exit_overflow:
1519 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENOVERFLOWS);
1520 exit_nonewsk:
1521 dst_release(dst);
1522 exit:
1523 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENDROPS);
1524 return NULL;
1525 put_and_exit:
1526 tcp_clear_xmit_timers(newsk);
1527 tcp_cleanup_congestion_control(newsk);
1528 bh_unlock_sock(newsk);
1529 sock_put(newsk);
1530 goto exit;
1531 }
1532 EXPORT_SYMBOL(tcp_v4_syn_recv_sock);
1533
1534 static struct sock *tcp_v4_hnd_req(struct sock *sk, struct sk_buff *skb)
1535 {
1536 struct tcphdr *th = tcp_hdr(skb);
1537 const struct iphdr *iph = ip_hdr(skb);
1538 struct sock *nsk;
1539 struct request_sock **prev;
1540 /* Find possible connection requests. */
1541 struct request_sock *req = inet_csk_search_req(sk, &prev, th->source,
1542 iph->saddr, iph->daddr);
1543 if (req)
1544 return tcp_check_req(sk, skb, req, prev);
1545
1546 nsk = inet_lookup_established(sock_net(sk), &tcp_hashinfo, iph->saddr,
1547 th->source, iph->daddr, th->dest, inet_iif(skb));
1548
1549 if (nsk) {
1550 if (nsk->sk_state != TCP_TIME_WAIT) {
1551 bh_lock_sock(nsk);
1552 return nsk;
1553 }
1554 inet_twsk_put(inet_twsk(nsk));
1555 return NULL;
1556 }
1557
1558 #ifdef CONFIG_SYN_COOKIES
1559 if (!th->syn)
1560 sk = cookie_v4_check(sk, skb, &(IPCB(skb)->opt));
1561 #endif
1562 return sk;
1563 }
1564
1565 static __sum16 tcp_v4_checksum_init(struct sk_buff *skb)
1566 {
1567 const struct iphdr *iph = ip_hdr(skb);
1568
1569 if (skb->ip_summed == CHECKSUM_COMPLETE) {
1570 if (!tcp_v4_check(skb->len, iph->saddr,
1571 iph->daddr, skb->csum)) {
1572 skb->ip_summed = CHECKSUM_UNNECESSARY;
1573 return 0;
1574 }
1575 }
1576
1577 skb->csum = csum_tcpudp_nofold(iph->saddr, iph->daddr,
1578 skb->len, IPPROTO_TCP, 0);
1579
1580 if (skb->len <= 76) {
1581 return __skb_checksum_complete(skb);
1582 }
1583 return 0;
1584 }
1585
1586
1587 /* The socket must have it's spinlock held when we get
1588 * here.
1589 *
1590 * We have a potential double-lock case here, so even when
1591 * doing backlog processing we use the BH locking scheme.
1592 * This is because we cannot sleep with the original spinlock
1593 * held.
1594 */
1595 int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb)
1596 {
1597 struct sock *rsk;
1598 #ifdef CONFIG_TCP_MD5SIG
1599 /*
1600 * We really want to reject the packet as early as possible
1601 * if:
1602 * o We're expecting an MD5'd packet and this is no MD5 tcp option
1603 * o There is an MD5 option and we're not expecting one
1604 */
1605 if (tcp_v4_inbound_md5_hash(sk, skb))
1606 goto discard;
1607 #endif
1608
1609 if (sk->sk_state == TCP_ESTABLISHED) { /* Fast path */
1610 sock_rps_save_rxhash(sk, skb);
1611 if (tcp_rcv_established(sk, skb, tcp_hdr(skb), skb->len)) {
1612 rsk = sk;
1613 goto reset;
1614 }
1615 return 0;
1616 }
1617
1618 if (skb->len < tcp_hdrlen(skb) || tcp_checksum_complete(skb))
1619 goto csum_err;
1620
1621 if (sk->sk_state == TCP_LISTEN) {
1622 struct sock *nsk = tcp_v4_hnd_req(sk, skb);
1623 if (!nsk)
1624 goto discard;
1625
1626 if (nsk != sk) {
1627 sock_rps_save_rxhash(nsk, skb);
1628 if (tcp_child_process(sk, nsk, skb)) {
1629 rsk = nsk;
1630 goto reset;
1631 }
1632 return 0;
1633 }
1634 } else
1635 sock_rps_save_rxhash(sk, skb);
1636
1637 if (tcp_rcv_state_process(sk, skb, tcp_hdr(skb), skb->len)) {
1638 rsk = sk;
1639 goto reset;
1640 }
1641 return 0;
1642
1643 reset:
1644 tcp_v4_send_reset(rsk, skb);
1645 discard:
1646 kfree_skb(skb);
1647 /* Be careful here. If this function gets more complicated and
1648 * gcc suffers from register pressure on the x86, sk (in %ebx)
1649 * might be destroyed here. This current version compiles correctly,
1650 * but you have been warned.
1651 */
1652 return 0;
1653
1654 csum_err:
1655 TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_INERRS);
1656 goto discard;
1657 }
1658 EXPORT_SYMBOL(tcp_v4_do_rcv);
1659
1660 void tcp_v4_early_demux(struct sk_buff *skb)
1661 {
1662 struct net *net = dev_net(skb->dev);
1663 const struct iphdr *iph;
1664 const struct tcphdr *th;
1665 struct net_device *dev;
1666 struct sock *sk;
1667
1668 if (skb->pkt_type != PACKET_HOST)
1669 return;
1670
1671 if (!pskb_may_pull(skb, ip_hdrlen(skb) + sizeof(struct tcphdr)))
1672 return;
1673
1674 iph = ip_hdr(skb);
1675 th = (struct tcphdr *) ((char *)iph + ip_hdrlen(skb));
1676
1677 if (th->doff < sizeof(struct tcphdr) / 4)
1678 return;
1679
1680 if (!pskb_may_pull(skb, ip_hdrlen(skb) + th->doff * 4))
1681 return;
1682
1683 dev = skb->dev;
1684 sk = __inet_lookup_established(net, &tcp_hashinfo,
1685 iph->saddr, th->source,
1686 iph->daddr, ntohs(th->dest),
1687 dev->ifindex);
1688 if (sk) {
1689 skb->sk = sk;
1690 skb->destructor = sock_edemux;
1691 if (sk->sk_state != TCP_TIME_WAIT) {
1692 struct dst_entry *dst = sk->sk_rx_dst;
1693 if (dst)
1694 dst = dst_check(dst, 0);
1695 if (dst) {
1696 struct rtable *rt = (struct rtable *) dst;
1697
1698 if (rt->rt_iif == dev->ifindex)
1699 skb_dst_set_noref(skb, dst);
1700 }
1701 }
1702 }
1703 }
1704
1705 /*
1706 * From tcp_input.c
1707 */
1708
1709 int tcp_v4_rcv(struct sk_buff *skb)
1710 {
1711 const struct iphdr *iph;
1712 const struct tcphdr *th;
1713 struct sock *sk;
1714 int ret;
1715 struct net *net = dev_net(skb->dev);
1716
1717 if (skb->pkt_type != PACKET_HOST)
1718 goto discard_it;
1719
1720 /* Count it even if it's bad */
1721 TCP_INC_STATS_BH(net, TCP_MIB_INSEGS);
1722
1723 if (!pskb_may_pull(skb, sizeof(struct tcphdr)))
1724 goto discard_it;
1725
1726 th = tcp_hdr(skb);
1727
1728 if (th->doff < sizeof(struct tcphdr) / 4)
1729 goto bad_packet;
1730 if (!pskb_may_pull(skb, th->doff * 4))
1731 goto discard_it;
1732
1733 /* An explanation is required here, I think.
1734 * Packet length and doff are validated by header prediction,
1735 * provided case of th->doff==0 is eliminated.
1736 * So, we defer the checks. */
1737 if (!skb_csum_unnecessary(skb) && tcp_v4_checksum_init(skb))
1738 goto bad_packet;
1739
1740 th = tcp_hdr(skb);
1741 iph = ip_hdr(skb);
1742 TCP_SKB_CB(skb)->seq = ntohl(th->seq);
1743 TCP_SKB_CB(skb)->end_seq = (TCP_SKB_CB(skb)->seq + th->syn + th->fin +
1744 skb->len - th->doff * 4);
1745 TCP_SKB_CB(skb)->ack_seq = ntohl(th->ack_seq);
1746 TCP_SKB_CB(skb)->when = 0;
1747 TCP_SKB_CB(skb)->ip_dsfield = ipv4_get_dsfield(iph);
1748 TCP_SKB_CB(skb)->sacked = 0;
1749
1750 sk = __inet_lookup_skb(&tcp_hashinfo, skb, th->source, th->dest);
1751 if (!sk)
1752 goto no_tcp_socket;
1753
1754 process:
1755 if (sk->sk_state == TCP_TIME_WAIT)
1756 goto do_time_wait;
1757
1758 if (unlikely(iph->ttl < inet_sk(sk)->min_ttl)) {
1759 NET_INC_STATS_BH(net, LINUX_MIB_TCPMINTTLDROP);
1760 goto discard_and_relse;
1761 }
1762
1763 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
1764 goto discard_and_relse;
1765 nf_reset(skb);
1766
1767 if (sk_filter(sk, skb))
1768 goto discard_and_relse;
1769
1770 skb->dev = NULL;
1771
1772 bh_lock_sock_nested(sk);
1773 ret = 0;
1774 if (!sock_owned_by_user(sk)) {
1775 #ifdef CONFIG_NET_DMA
1776 struct tcp_sock *tp = tcp_sk(sk);
1777 if (!tp->ucopy.dma_chan && tp->ucopy.pinned_list)
1778 tp->ucopy.dma_chan = net_dma_find_channel();
1779 if (tp->ucopy.dma_chan)
1780 ret = tcp_v4_do_rcv(sk, skb);
1781 else
1782 #endif
1783 {
1784 if (!tcp_prequeue(sk, skb))
1785 ret = tcp_v4_do_rcv(sk, skb);
1786 }
1787 } else if (unlikely(sk_add_backlog(sk, skb,
1788 sk->sk_rcvbuf + sk->sk_sndbuf))) {
1789 bh_unlock_sock(sk);
1790 NET_INC_STATS_BH(net, LINUX_MIB_TCPBACKLOGDROP);
1791 goto discard_and_relse;
1792 }
1793 bh_unlock_sock(sk);
1794
1795 sock_put(sk);
1796
1797 return ret;
1798
1799 no_tcp_socket:
1800 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
1801 goto discard_it;
1802
1803 if (skb->len < (th->doff << 2) || tcp_checksum_complete(skb)) {
1804 bad_packet:
1805 TCP_INC_STATS_BH(net, TCP_MIB_INERRS);
1806 } else {
1807 tcp_v4_send_reset(NULL, skb);
1808 }
1809
1810 discard_it:
1811 /* Discard frame. */
1812 kfree_skb(skb);
1813 return 0;
1814
1815 discard_and_relse:
1816 sock_put(sk);
1817 goto discard_it;
1818
1819 do_time_wait:
1820 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) {
1821 inet_twsk_put(inet_twsk(sk));
1822 goto discard_it;
1823 }
1824
1825 if (skb->len < (th->doff << 2) || tcp_checksum_complete(skb)) {
1826 TCP_INC_STATS_BH(net, TCP_MIB_INERRS);
1827 inet_twsk_put(inet_twsk(sk));
1828 goto discard_it;
1829 }
1830 switch (tcp_timewait_state_process(inet_twsk(sk), skb, th)) {
1831 case TCP_TW_SYN: {
1832 struct sock *sk2 = inet_lookup_listener(dev_net(skb->dev),
1833 &tcp_hashinfo,
1834 iph->daddr, th->dest,
1835 inet_iif(skb));
1836 if (sk2) {
1837 inet_twsk_deschedule(inet_twsk(sk), &tcp_death_row);
1838 inet_twsk_put(inet_twsk(sk));
1839 sk = sk2;
1840 goto process;
1841 }
1842 /* Fall through to ACK */
1843 }
1844 case TCP_TW_ACK:
1845 tcp_v4_timewait_ack(sk, skb);
1846 break;
1847 case TCP_TW_RST:
1848 goto no_tcp_socket;
1849 case TCP_TW_SUCCESS:;
1850 }
1851 goto discard_it;
1852 }
1853
1854 static struct timewait_sock_ops tcp_timewait_sock_ops = {
1855 .twsk_obj_size = sizeof(struct tcp_timewait_sock),
1856 .twsk_unique = tcp_twsk_unique,
1857 .twsk_destructor= tcp_twsk_destructor,
1858 };
1859
1860 const struct inet_connection_sock_af_ops ipv4_specific = {
1861 .queue_xmit = ip_queue_xmit,
1862 .send_check = tcp_v4_send_check,
1863 .rebuild_header = inet_sk_rebuild_header,
1864 .conn_request = tcp_v4_conn_request,
1865 .syn_recv_sock = tcp_v4_syn_recv_sock,
1866 .net_header_len = sizeof(struct iphdr),
1867 .setsockopt = ip_setsockopt,
1868 .getsockopt = ip_getsockopt,
1869 .addr2sockaddr = inet_csk_addr2sockaddr,
1870 .sockaddr_len = sizeof(struct sockaddr_in),
1871 .bind_conflict = inet_csk_bind_conflict,
1872 #ifdef CONFIG_COMPAT
1873 .compat_setsockopt = compat_ip_setsockopt,
1874 .compat_getsockopt = compat_ip_getsockopt,
1875 #endif
1876 };
1877 EXPORT_SYMBOL(ipv4_specific);
1878
1879 #ifdef CONFIG_TCP_MD5SIG
1880 static const struct tcp_sock_af_ops tcp_sock_ipv4_specific = {
1881 .md5_lookup = tcp_v4_md5_lookup,
1882 .calc_md5_hash = tcp_v4_md5_hash_skb,
1883 .md5_parse = tcp_v4_parse_md5_keys,
1884 };
1885 #endif
1886
1887 /* NOTE: A lot of things set to zero explicitly by call to
1888 * sk_alloc() so need not be done here.
1889 */
1890 static int tcp_v4_init_sock(struct sock *sk)
1891 {
1892 struct inet_connection_sock *icsk = inet_csk(sk);
1893
1894 tcp_init_sock(sk);
1895
1896 icsk->icsk_af_ops = &ipv4_specific;
1897
1898 #ifdef CONFIG_TCP_MD5SIG
1899 tcp_sk(sk)->af_specific = &tcp_sock_ipv4_specific;
1900 #endif
1901
1902 return 0;
1903 }
1904
1905 void tcp_v4_destroy_sock(struct sock *sk)
1906 {
1907 struct tcp_sock *tp = tcp_sk(sk);
1908
1909 tcp_clear_xmit_timers(sk);
1910
1911 tcp_cleanup_congestion_control(sk);
1912
1913 /* Cleanup up the write buffer. */
1914 tcp_write_queue_purge(sk);
1915
1916 /* Cleans up our, hopefully empty, out_of_order_queue. */
1917 __skb_queue_purge(&tp->out_of_order_queue);
1918
1919 #ifdef CONFIG_TCP_MD5SIG
1920 /* Clean up the MD5 key list, if any */
1921 if (tp->md5sig_info) {
1922 tcp_clear_md5_list(sk);
1923 kfree_rcu(tp->md5sig_info, rcu);
1924 tp->md5sig_info = NULL;
1925 }
1926 #endif
1927
1928 #ifdef CONFIG_NET_DMA
1929 /* Cleans up our sk_async_wait_queue */
1930 __skb_queue_purge(&sk->sk_async_wait_queue);
1931 #endif
1932
1933 /* Clean prequeue, it must be empty really */
1934 __skb_queue_purge(&tp->ucopy.prequeue);
1935
1936 /* Clean up a referenced TCP bind bucket. */
1937 if (inet_csk(sk)->icsk_bind_hash)
1938 inet_put_port(sk);
1939
1940 /*
1941 * If sendmsg cached page exists, toss it.
1942 */
1943 if (sk->sk_sndmsg_page) {
1944 __free_page(sk->sk_sndmsg_page);
1945 sk->sk_sndmsg_page = NULL;
1946 }
1947
1948 /* TCP Cookie Transactions */
1949 if (tp->cookie_values != NULL) {
1950 kref_put(&tp->cookie_values->kref,
1951 tcp_cookie_values_release);
1952 tp->cookie_values = NULL;
1953 }
1954
1955 sk_sockets_allocated_dec(sk);
1956 sock_release_memcg(sk);
1957 }
1958 EXPORT_SYMBOL(tcp_v4_destroy_sock);
1959
1960 #ifdef CONFIG_PROC_FS
1961 /* Proc filesystem TCP sock list dumping. */
1962
1963 static inline struct inet_timewait_sock *tw_head(struct hlist_nulls_head *head)
1964 {
1965 return hlist_nulls_empty(head) ? NULL :
1966 list_entry(head->first, struct inet_timewait_sock, tw_node);
1967 }
1968
1969 static inline struct inet_timewait_sock *tw_next(struct inet_timewait_sock *tw)
1970 {
1971 return !is_a_nulls(tw->tw_node.next) ?
1972 hlist_nulls_entry(tw->tw_node.next, typeof(*tw), tw_node) : NULL;
1973 }
1974
1975 /*
1976 * Get next listener socket follow cur. If cur is NULL, get first socket
1977 * starting from bucket given in st->bucket; when st->bucket is zero the
1978 * very first socket in the hash table is returned.
1979 */
1980 static void *listening_get_next(struct seq_file *seq, void *cur)
1981 {
1982 struct inet_connection_sock *icsk;
1983 struct hlist_nulls_node *node;
1984 struct sock *sk = cur;
1985 struct inet_listen_hashbucket *ilb;
1986 struct tcp_iter_state *st = seq->private;
1987 struct net *net = seq_file_net(seq);
1988
1989 if (!sk) {
1990 ilb = &tcp_hashinfo.listening_hash[st->bucket];
1991 spin_lock_bh(&ilb->lock);
1992 sk = sk_nulls_head(&ilb->head);
1993 st->offset = 0;
1994 goto get_sk;
1995 }
1996 ilb = &tcp_hashinfo.listening_hash[st->bucket];
1997 ++st->num;
1998 ++st->offset;
1999
2000 if (st->state == TCP_SEQ_STATE_OPENREQ) {
2001 struct request_sock *req = cur;
2002
2003 icsk = inet_csk(st->syn_wait_sk);
2004 req = req->dl_next;
2005 while (1) {
2006 while (req) {
2007 if (req->rsk_ops->family == st->family) {
2008 cur = req;
2009 goto out;
2010 }
2011 req = req->dl_next;
2012 }
2013 if (++st->sbucket >= icsk->icsk_accept_queue.listen_opt->nr_table_entries)
2014 break;
2015 get_req:
2016 req = icsk->icsk_accept_queue.listen_opt->syn_table[st->sbucket];
2017 }
2018 sk = sk_nulls_next(st->syn_wait_sk);
2019 st->state = TCP_SEQ_STATE_LISTENING;
2020 read_unlock_bh(&icsk->icsk_accept_queue.syn_wait_lock);
2021 } else {
2022 icsk = inet_csk(sk);
2023 read_lock_bh(&icsk->icsk_accept_queue.syn_wait_lock);
2024 if (reqsk_queue_len(&icsk->icsk_accept_queue))
2025 goto start_req;
2026 read_unlock_bh(&icsk->icsk_accept_queue.syn_wait_lock);
2027 sk = sk_nulls_next(sk);
2028 }
2029 get_sk:
2030 sk_nulls_for_each_from(sk, node) {
2031 if (!net_eq(sock_net(sk), net))
2032 continue;
2033 if (sk->sk_family == st->family) {
2034 cur = sk;
2035 goto out;
2036 }
2037 icsk = inet_csk(sk);
2038 read_lock_bh(&icsk->icsk_accept_queue.syn_wait_lock);
2039 if (reqsk_queue_len(&icsk->icsk_accept_queue)) {
2040 start_req:
2041 st->uid = sock_i_uid(sk);
2042 st->syn_wait_sk = sk;
2043 st->state = TCP_SEQ_STATE_OPENREQ;
2044 st->sbucket = 0;
2045 goto get_req;
2046 }
2047 read_unlock_bh(&icsk->icsk_accept_queue.syn_wait_lock);
2048 }
2049 spin_unlock_bh(&ilb->lock);
2050 st->offset = 0;
2051 if (++st->bucket < INET_LHTABLE_SIZE) {
2052 ilb = &tcp_hashinfo.listening_hash[st->bucket];
2053 spin_lock_bh(&ilb->lock);
2054 sk = sk_nulls_head(&ilb->head);
2055 goto get_sk;
2056 }
2057 cur = NULL;
2058 out:
2059 return cur;
2060 }
2061
2062 static void *listening_get_idx(struct seq_file *seq, loff_t *pos)
2063 {
2064 struct tcp_iter_state *st = seq->private;
2065 void *rc;
2066
2067 st->bucket = 0;
2068 st->offset = 0;
2069 rc = listening_get_next(seq, NULL);
2070
2071 while (rc && *pos) {
2072 rc = listening_get_next(seq, rc);
2073 --*pos;
2074 }
2075 return rc;
2076 }
2077
2078 static inline bool empty_bucket(struct tcp_iter_state *st)
2079 {
2080 return hlist_nulls_empty(&tcp_hashinfo.ehash[st->bucket].chain) &&
2081 hlist_nulls_empty(&tcp_hashinfo.ehash[st->bucket].twchain);
2082 }
2083
2084 /*
2085 * Get first established socket starting from bucket given in st->bucket.
2086 * If st->bucket is zero, the very first socket in the hash is returned.
2087 */
2088 static void *established_get_first(struct seq_file *seq)
2089 {
2090 struct tcp_iter_state *st = seq->private;
2091 struct net *net = seq_file_net(seq);
2092 void *rc = NULL;
2093
2094 st->offset = 0;
2095 for (; st->bucket <= tcp_hashinfo.ehash_mask; ++st->bucket) {
2096 struct sock *sk;
2097 struct hlist_nulls_node *node;
2098 struct inet_timewait_sock *tw;
2099 spinlock_t *lock = inet_ehash_lockp(&tcp_hashinfo, st->bucket);
2100
2101 /* Lockless fast path for the common case of empty buckets */
2102 if (empty_bucket(st))
2103 continue;
2104
2105 spin_lock_bh(lock);
2106 sk_nulls_for_each(sk, node, &tcp_hashinfo.ehash[st->bucket].chain) {
2107 if (sk->sk_family != st->family ||
2108 !net_eq(sock_net(sk), net)) {
2109 continue;
2110 }
2111 rc = sk;
2112 goto out;
2113 }
2114 st->state = TCP_SEQ_STATE_TIME_WAIT;
2115 inet_twsk_for_each(tw, node,
2116 &tcp_hashinfo.ehash[st->bucket].twchain) {
2117 if (tw->tw_family != st->family ||
2118 !net_eq(twsk_net(tw), net)) {
2119 continue;
2120 }
2121 rc = tw;
2122 goto out;
2123 }
2124 spin_unlock_bh(lock);
2125 st->state = TCP_SEQ_STATE_ESTABLISHED;
2126 }
2127 out:
2128 return rc;
2129 }
2130
2131 static void *established_get_next(struct seq_file *seq, void *cur)
2132 {
2133 struct sock *sk = cur;
2134 struct inet_timewait_sock *tw;
2135 struct hlist_nulls_node *node;
2136 struct tcp_iter_state *st = seq->private;
2137 struct net *net = seq_file_net(seq);
2138
2139 ++st->num;
2140 ++st->offset;
2141
2142 if (st->state == TCP_SEQ_STATE_TIME_WAIT) {
2143 tw = cur;
2144 tw = tw_next(tw);
2145 get_tw:
2146 while (tw && (tw->tw_family != st->family || !net_eq(twsk_net(tw), net))) {
2147 tw = tw_next(tw);
2148 }
2149 if (tw) {
2150 cur = tw;
2151 goto out;
2152 }
2153 spin_unlock_bh(inet_ehash_lockp(&tcp_hashinfo, st->bucket));
2154 st->state = TCP_SEQ_STATE_ESTABLISHED;
2155
2156 /* Look for next non empty bucket */
2157 st->offset = 0;
2158 while (++st->bucket <= tcp_hashinfo.ehash_mask &&
2159 empty_bucket(st))
2160 ;
2161 if (st->bucket > tcp_hashinfo.ehash_mask)
2162 return NULL;
2163
2164 spin_lock_bh(inet_ehash_lockp(&tcp_hashinfo, st->bucket));
2165 sk = sk_nulls_head(&tcp_hashinfo.ehash[st->bucket].chain);
2166 } else
2167 sk = sk_nulls_next(sk);
2168
2169 sk_nulls_for_each_from(sk, node) {
2170 if (sk->sk_family == st->family && net_eq(sock_net(sk), net))
2171 goto found;
2172 }
2173
2174 st->state = TCP_SEQ_STATE_TIME_WAIT;
2175 tw = tw_head(&tcp_hashinfo.ehash[st->bucket].twchain);
2176 goto get_tw;
2177 found:
2178 cur = sk;
2179 out:
2180 return cur;
2181 }
2182
2183 static void *established_get_idx(struct seq_file *seq, loff_t pos)
2184 {
2185 struct tcp_iter_state *st = seq->private;
2186 void *rc;
2187
2188 st->bucket = 0;
2189 rc = established_get_first(seq);
2190
2191 while (rc && pos) {
2192 rc = established_get_next(seq, rc);
2193 --pos;
2194 }
2195 return rc;
2196 }
2197
2198 static void *tcp_get_idx(struct seq_file *seq, loff_t pos)
2199 {
2200 void *rc;
2201 struct tcp_iter_state *st = seq->private;
2202
2203 st->state = TCP_SEQ_STATE_LISTENING;
2204 rc = listening_get_idx(seq, &pos);
2205
2206 if (!rc) {
2207 st->state = TCP_SEQ_STATE_ESTABLISHED;
2208 rc = established_get_idx(seq, pos);
2209 }
2210
2211 return rc;
2212 }
2213
2214 static void *tcp_seek_last_pos(struct seq_file *seq)
2215 {
2216 struct tcp_iter_state *st = seq->private;
2217 int offset = st->offset;
2218 int orig_num = st->num;
2219 void *rc = NULL;
2220
2221 switch (st->state) {
2222 case TCP_SEQ_STATE_OPENREQ:
2223 case TCP_SEQ_STATE_LISTENING:
2224 if (st->bucket >= INET_LHTABLE_SIZE)
2225 break;
2226 st->state = TCP_SEQ_STATE_LISTENING;
2227 rc = listening_get_next(seq, NULL);
2228 while (offset-- && rc)
2229 rc = listening_get_next(seq, rc);
2230 if (rc)
2231 break;
2232 st->bucket = 0;
2233 /* Fallthrough */
2234 case TCP_SEQ_STATE_ESTABLISHED:
2235 case TCP_SEQ_STATE_TIME_WAIT:
2236 st->state = TCP_SEQ_STATE_ESTABLISHED;
2237 if (st->bucket > tcp_hashinfo.ehash_mask)
2238 break;
2239 rc = established_get_first(seq);
2240 while (offset-- && rc)
2241 rc = established_get_next(seq, rc);
2242 }
2243
2244 st->num = orig_num;
2245
2246 return rc;
2247 }
2248
2249 static void *tcp_seq_start(struct seq_file *seq, loff_t *pos)
2250 {
2251 struct tcp_iter_state *st = seq->private;
2252 void *rc;
2253
2254 if (*pos && *pos == st->last_pos) {
2255 rc = tcp_seek_last_pos(seq);
2256 if (rc)
2257 goto out;
2258 }
2259
2260 st->state = TCP_SEQ_STATE_LISTENING;
2261 st->num = 0;
2262 st->bucket = 0;
2263 st->offset = 0;
2264 rc = *pos ? tcp_get_idx(seq, *pos - 1) : SEQ_START_TOKEN;
2265
2266 out:
2267 st->last_pos = *pos;
2268 return rc;
2269 }
2270
2271 static void *tcp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2272 {
2273 struct tcp_iter_state *st = seq->private;
2274 void *rc = NULL;
2275
2276 if (v == SEQ_START_TOKEN) {
2277 rc = tcp_get_idx(seq, 0);
2278 goto out;
2279 }
2280
2281 switch (st->state) {
2282 case TCP_SEQ_STATE_OPENREQ:
2283 case TCP_SEQ_STATE_LISTENING:
2284 rc = listening_get_next(seq, v);
2285 if (!rc) {
2286 st->state = TCP_SEQ_STATE_ESTABLISHED;
2287 st->bucket = 0;
2288 st->offset = 0;
2289 rc = established_get_first(seq);
2290 }
2291 break;
2292 case TCP_SEQ_STATE_ESTABLISHED:
2293 case TCP_SEQ_STATE_TIME_WAIT:
2294 rc = established_get_next(seq, v);
2295 break;
2296 }
2297 out:
2298 ++*pos;
2299 st->last_pos = *pos;
2300 return rc;
2301 }
2302
2303 static void tcp_seq_stop(struct seq_file *seq, void *v)
2304 {
2305 struct tcp_iter_state *st = seq->private;
2306
2307 switch (st->state) {
2308 case TCP_SEQ_STATE_OPENREQ:
2309 if (v) {
2310 struct inet_connection_sock *icsk = inet_csk(st->syn_wait_sk);
2311 read_unlock_bh(&icsk->icsk_accept_queue.syn_wait_lock);
2312 }
2313 case TCP_SEQ_STATE_LISTENING:
2314 if (v != SEQ_START_TOKEN)
2315 spin_unlock_bh(&tcp_hashinfo.listening_hash[st->bucket].lock);
2316 break;
2317 case TCP_SEQ_STATE_TIME_WAIT:
2318 case TCP_SEQ_STATE_ESTABLISHED:
2319 if (v)
2320 spin_unlock_bh(inet_ehash_lockp(&tcp_hashinfo, st->bucket));
2321 break;
2322 }
2323 }
2324
2325 int tcp_seq_open(struct inode *inode, struct file *file)
2326 {
2327 struct tcp_seq_afinfo *afinfo = PDE(inode)->data;
2328 struct tcp_iter_state *s;
2329 int err;
2330
2331 err = seq_open_net(inode, file, &afinfo->seq_ops,
2332 sizeof(struct tcp_iter_state));
2333 if (err < 0)
2334 return err;
2335
2336 s = ((struct seq_file *)file->private_data)->private;
2337 s->family = afinfo->family;
2338 s->last_pos = 0;
2339 return 0;
2340 }
2341 EXPORT_SYMBOL(tcp_seq_open);
2342
2343 int tcp_proc_register(struct net *net, struct tcp_seq_afinfo *afinfo)
2344 {
2345 int rc = 0;
2346 struct proc_dir_entry *p;
2347
2348 afinfo->seq_ops.start = tcp_seq_start;
2349 afinfo->seq_ops.next = tcp_seq_next;
2350 afinfo->seq_ops.stop = tcp_seq_stop;
2351
2352 p = proc_create_data(afinfo->name, S_IRUGO, net->proc_net,
2353 afinfo->seq_fops, afinfo);
2354 if (!p)
2355 rc = -ENOMEM;
2356 return rc;
2357 }
2358 EXPORT_SYMBOL(tcp_proc_register);
2359
2360 void tcp_proc_unregister(struct net *net, struct tcp_seq_afinfo *afinfo)
2361 {
2362 proc_net_remove(net, afinfo->name);
2363 }
2364 EXPORT_SYMBOL(tcp_proc_unregister);
2365
2366 static void get_openreq4(const struct sock *sk, const struct request_sock *req,
2367 struct seq_file *f, int i, int uid, int *len)
2368 {
2369 const struct inet_request_sock *ireq = inet_rsk(req);
2370 int ttd = req->expires - jiffies;
2371
2372 seq_printf(f, "%4d: %08X:%04X %08X:%04X"
2373 " %02X %08X:%08X %02X:%08lX %08X %5d %8d %u %d %pK%n",
2374 i,
2375 ireq->loc_addr,
2376 ntohs(inet_sk(sk)->inet_sport),
2377 ireq->rmt_addr,
2378 ntohs(ireq->rmt_port),
2379 TCP_SYN_RECV,
2380 0, 0, /* could print option size, but that is af dependent. */
2381 1, /* timers active (only the expire timer) */
2382 jiffies_to_clock_t(ttd),
2383 req->retrans,
2384 uid,
2385 0, /* non standard timer */
2386 0, /* open_requests have no inode */
2387 atomic_read(&sk->sk_refcnt),
2388 req,
2389 len);
2390 }
2391
2392 static void get_tcp4_sock(struct sock *sk, struct seq_file *f, int i, int *len)
2393 {
2394 int timer_active;
2395 unsigned long timer_expires;
2396 const struct tcp_sock *tp = tcp_sk(sk);
2397 const struct inet_connection_sock *icsk = inet_csk(sk);
2398 const struct inet_sock *inet = inet_sk(sk);
2399 __be32 dest = inet->inet_daddr;
2400 __be32 src = inet->inet_rcv_saddr;
2401 __u16 destp = ntohs(inet->inet_dport);
2402 __u16 srcp = ntohs(inet->inet_sport);
2403 int rx_queue;
2404
2405 if (icsk->icsk_pending == ICSK_TIME_RETRANS) {
2406 timer_active = 1;
2407 timer_expires = icsk->icsk_timeout;
2408 } else if (icsk->icsk_pending == ICSK_TIME_PROBE0) {
2409 timer_active = 4;
2410 timer_expires = icsk->icsk_timeout;
2411 } else if (timer_pending(&sk->sk_timer)) {
2412 timer_active = 2;
2413 timer_expires = sk->sk_timer.expires;
2414 } else {
2415 timer_active = 0;
2416 timer_expires = jiffies;
2417 }
2418
2419 if (sk->sk_state == TCP_LISTEN)
2420 rx_queue = sk->sk_ack_backlog;
2421 else
2422 /*
2423 * because we dont lock socket, we might find a transient negative value
2424 */
2425 rx_queue = max_t(int, tp->rcv_nxt - tp->copied_seq, 0);
2426
2427 seq_printf(f, "%4d: %08X:%04X %08X:%04X %02X %08X:%08X %02X:%08lX "
2428 "%08X %5d %8d %lu %d %pK %lu %lu %u %u %d%n",
2429 i, src, srcp, dest, destp, sk->sk_state,
2430 tp->write_seq - tp->snd_una,
2431 rx_queue,
2432 timer_active,
2433 jiffies_to_clock_t(timer_expires - jiffies),
2434 icsk->icsk_retransmits,
2435 sock_i_uid(sk),
2436 icsk->icsk_probes_out,
2437 sock_i_ino(sk),
2438 atomic_read(&sk->sk_refcnt), sk,
2439 jiffies_to_clock_t(icsk->icsk_rto),
2440 jiffies_to_clock_t(icsk->icsk_ack.ato),
2441 (icsk->icsk_ack.quick << 1) | icsk->icsk_ack.pingpong,
2442 tp->snd_cwnd,
2443 tcp_in_initial_slowstart(tp) ? -1 : tp->snd_ssthresh,
2444 len);
2445 }
2446
2447 static void get_timewait4_sock(const struct inet_timewait_sock *tw,
2448 struct seq_file *f, int i, int *len)
2449 {
2450 __be32 dest, src;
2451 __u16 destp, srcp;
2452 int ttd = tw->tw_ttd - jiffies;
2453
2454 if (ttd < 0)
2455 ttd = 0;
2456
2457 dest = tw->tw_daddr;
2458 src = tw->tw_rcv_saddr;
2459 destp = ntohs(tw->tw_dport);
2460 srcp = ntohs(tw->tw_sport);
2461
2462 seq_printf(f, "%4d: %08X:%04X %08X:%04X"
2463 " %02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %pK%n",
2464 i, src, srcp, dest, destp, tw->tw_substate, 0, 0,
2465 3, jiffies_to_clock_t(ttd), 0, 0, 0, 0,
2466 atomic_read(&tw->tw_refcnt), tw, len);
2467 }
2468
2469 #define TMPSZ 150
2470
2471 static int tcp4_seq_show(struct seq_file *seq, void *v)
2472 {
2473 struct tcp_iter_state *st;
2474 int len;
2475
2476 if (v == SEQ_START_TOKEN) {
2477 seq_printf(seq, "%-*s\n", TMPSZ - 1,
2478 " sl local_address rem_address st tx_queue "
2479 "rx_queue tr tm->when retrnsmt uid timeout "
2480 "inode");
2481 goto out;
2482 }
2483 st = seq->private;
2484
2485 switch (st->state) {
2486 case TCP_SEQ_STATE_LISTENING:
2487 case TCP_SEQ_STATE_ESTABLISHED:
2488 get_tcp4_sock(v, seq, st->num, &len);
2489 break;
2490 case TCP_SEQ_STATE_OPENREQ:
2491 get_openreq4(st->syn_wait_sk, v, seq, st->num, st->uid, &len);
2492 break;
2493 case TCP_SEQ_STATE_TIME_WAIT:
2494 get_timewait4_sock(v, seq, st->num, &len);
2495 break;
2496 }
2497 seq_printf(seq, "%*s\n", TMPSZ - 1 - len, "");
2498 out:
2499 return 0;
2500 }
2501
2502 static const struct file_operations tcp_afinfo_seq_fops = {
2503 .owner = THIS_MODULE,
2504 .open = tcp_seq_open,
2505 .read = seq_read,
2506 .llseek = seq_lseek,
2507 .release = seq_release_net
2508 };
2509
2510 static struct tcp_seq_afinfo tcp4_seq_afinfo = {
2511 .name = "tcp",
2512 .family = AF_INET,
2513 .seq_fops = &tcp_afinfo_seq_fops,
2514 .seq_ops = {
2515 .show = tcp4_seq_show,
2516 },
2517 };
2518
2519 static int __net_init tcp4_proc_init_net(struct net *net)
2520 {
2521 return tcp_proc_register(net, &tcp4_seq_afinfo);
2522 }
2523
2524 static void __net_exit tcp4_proc_exit_net(struct net *net)
2525 {
2526 tcp_proc_unregister(net, &tcp4_seq_afinfo);
2527 }
2528
2529 static struct pernet_operations tcp4_net_ops = {
2530 .init = tcp4_proc_init_net,
2531 .exit = tcp4_proc_exit_net,
2532 };
2533
2534 int __init tcp4_proc_init(void)
2535 {
2536 return register_pernet_subsys(&tcp4_net_ops);
2537 }
2538
2539 void tcp4_proc_exit(void)
2540 {
2541 unregister_pernet_subsys(&tcp4_net_ops);
2542 }
2543 #endif /* CONFIG_PROC_FS */
2544
2545 struct sk_buff **tcp4_gro_receive(struct sk_buff **head, struct sk_buff *skb)
2546 {
2547 const struct iphdr *iph = skb_gro_network_header(skb);
2548
2549 switch (skb->ip_summed) {
2550 case CHECKSUM_COMPLETE:
2551 if (!tcp_v4_check(skb_gro_len(skb), iph->saddr, iph->daddr,
2552 skb->csum)) {
2553 skb->ip_summed = CHECKSUM_UNNECESSARY;
2554 break;
2555 }
2556
2557 /* fall through */
2558 case CHECKSUM_NONE:
2559 NAPI_GRO_CB(skb)->flush = 1;
2560 return NULL;
2561 }
2562
2563 return tcp_gro_receive(head, skb);
2564 }
2565
2566 int tcp4_gro_complete(struct sk_buff *skb)
2567 {
2568 const struct iphdr *iph = ip_hdr(skb);
2569 struct tcphdr *th = tcp_hdr(skb);
2570
2571 th->check = ~tcp_v4_check(skb->len - skb_transport_offset(skb),
2572 iph->saddr, iph->daddr, 0);
2573 skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
2574
2575 return tcp_gro_complete(skb);
2576 }
2577
2578 struct proto tcp_prot = {
2579 .name = "TCP",
2580 .owner = THIS_MODULE,
2581 .close = tcp_close,
2582 .connect = tcp_v4_connect,
2583 .disconnect = tcp_disconnect,
2584 .accept = inet_csk_accept,
2585 .ioctl = tcp_ioctl,
2586 .init = tcp_v4_init_sock,
2587 .destroy = tcp_v4_destroy_sock,
2588 .shutdown = tcp_shutdown,
2589 .setsockopt = tcp_setsockopt,
2590 .getsockopt = tcp_getsockopt,
2591 .recvmsg = tcp_recvmsg,
2592 .sendmsg = tcp_sendmsg,
2593 .sendpage = tcp_sendpage,
2594 .backlog_rcv = tcp_v4_do_rcv,
2595 .release_cb = tcp_release_cb,
2596 .hash = inet_hash,
2597 .unhash = inet_unhash,
2598 .get_port = inet_csk_get_port,
2599 .enter_memory_pressure = tcp_enter_memory_pressure,
2600 .sockets_allocated = &tcp_sockets_allocated,
2601 .orphan_count = &tcp_orphan_count,
2602 .memory_allocated = &tcp_memory_allocated,
2603 .memory_pressure = &tcp_memory_pressure,
2604 .sysctl_wmem = sysctl_tcp_wmem,
2605 .sysctl_rmem = sysctl_tcp_rmem,
2606 .max_header = MAX_TCP_HEADER,
2607 .obj_size = sizeof(struct tcp_sock),
2608 .slab_flags = SLAB_DESTROY_BY_RCU,
2609 .twsk_prot = &tcp_timewait_sock_ops,
2610 .rsk_prot = &tcp_request_sock_ops,
2611 .h.hashinfo = &tcp_hashinfo,
2612 .no_autobind = true,
2613 #ifdef CONFIG_COMPAT
2614 .compat_setsockopt = compat_tcp_setsockopt,
2615 .compat_getsockopt = compat_tcp_getsockopt,
2616 #endif
2617 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_KMEM
2618 .init_cgroup = tcp_init_cgroup,
2619 .destroy_cgroup = tcp_destroy_cgroup,
2620 .proto_cgroup = tcp_proto_cgroup,
2621 #endif
2622 };
2623 EXPORT_SYMBOL(tcp_prot);
2624
2625 static int __net_init tcp_sk_init(struct net *net)
2626 {
2627 return 0;
2628 }
2629
2630 static void __net_exit tcp_sk_exit(struct net *net)
2631 {
2632 }
2633
2634 static void __net_exit tcp_sk_exit_batch(struct list_head *net_exit_list)
2635 {
2636 inet_twsk_purge(&tcp_hashinfo, &tcp_death_row, AF_INET);
2637 }
2638
2639 static struct pernet_operations __net_initdata tcp_sk_ops = {
2640 .init = tcp_sk_init,
2641 .exit = tcp_sk_exit,
2642 .exit_batch = tcp_sk_exit_batch,
2643 };
2644
2645 void __init tcp_v4_init(void)
2646 {
2647 inet_hashinfo_init(&tcp_hashinfo);
2648 if (register_pernet_subsys(&tcp_sk_ops))
2649 panic("Failed to create the TCP control socket.\n");
2650 }
Linus' kernel tree