projects / deliverable / linux.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
ipv4: Namespaceify tcp synack retries sysctl knob
[deliverable/linux.git] / net / ipv4 / tcp_ipv4.c
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/secure_seq.h>
76 #include <net/busy_poll.h>
77
78 #include <linux/inet.h>
79 #include <linux/ipv6.h>
80 #include <linux/stddef.h>
81 #include <linux/proc_fs.h>
82 #include <linux/seq_file.h>
83
84 #include <linux/crypto.h>
85 #include <linux/scatterlist.h>
86
87 int sysctl_tcp_tw_reuse __read_mostly;
88 int sysctl_tcp_low_latency __read_mostly;
89 EXPORT_SYMBOL(sysctl_tcp_low_latency);
90
91 #ifdef CONFIG_TCP_MD5SIG
92 static int tcp_v4_md5_hash_hdr(char *md5_hash, const struct tcp_md5sig_key *key,
93 __be32 daddr, __be32 saddr, const struct tcphdr *th);
94 #endif
95
96 struct inet_hashinfo tcp_hashinfo;
97 EXPORT_SYMBOL(tcp_hashinfo);
98
99 static __u32 tcp_v4_init_sequence(const struct sk_buff *skb)
100 {
101 return secure_tcp_sequence_number(ip_hdr(skb)->daddr,
102 ip_hdr(skb)->saddr,
103 tcp_hdr(skb)->dest,
104 tcp_hdr(skb)->source);
105 }
106
107 int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp)
108 {
109 const struct tcp_timewait_sock *tcptw = tcp_twsk(sktw);
110 struct tcp_sock *tp = tcp_sk(sk);
111
112 /* With PAWS, it is safe from the viewpoint
113 of data integrity. Even without PAWS it is safe provided sequence
114 spaces do not overlap i.e. at data rates <= 80Mbit/sec.
115
116 Actually, the idea is close to VJ's one, only timestamp cache is
117 held not per host, but per port pair and TW bucket is used as state
118 holder.
119
120 If TW bucket has been already destroyed we fall back to VJ's scheme
121 and use initial timestamp retrieved from peer table.
122 */
123 if (tcptw->tw_ts_recent_stamp &&
124 (!twp || (sysctl_tcp_tw_reuse &&
125 get_seconds() - tcptw->tw_ts_recent_stamp > 1))) {
126 tp->write_seq = tcptw->tw_snd_nxt + 65535 + 2;
127 if (tp->write_seq == 0)
128 tp->write_seq = 1;
129 tp->rx_opt.ts_recent = tcptw->tw_ts_recent;
130 tp->rx_opt.ts_recent_stamp = tcptw->tw_ts_recent_stamp;
131 sock_hold(sktw);
132 return 1;
133 }
134
135 return 0;
136 }
137 EXPORT_SYMBOL_GPL(tcp_twsk_unique);
138
139 /* This will initiate an outgoing connection. */
140 int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len)
141 {
142 struct sockaddr_in *usin = (struct sockaddr_in *)uaddr;
143 struct inet_sock *inet = inet_sk(sk);
144 struct tcp_sock *tp = tcp_sk(sk);
145 __be16 orig_sport, orig_dport;
146 __be32 daddr, nexthop;
147 struct flowi4 *fl4;
148 struct rtable *rt;
149 int err;
150 struct ip_options_rcu *inet_opt;
151
152 if (addr_len < sizeof(struct sockaddr_in))
153 return -EINVAL;
154
155 if (usin->sin_family != AF_INET)
156 return -EAFNOSUPPORT;
157
158 nexthop = daddr = usin->sin_addr.s_addr;
159 inet_opt = rcu_dereference_protected(inet->inet_opt,
160 sock_owned_by_user(sk));
161 if (inet_opt && inet_opt->opt.srr) {
162 if (!daddr)
163 return -EINVAL;
164 nexthop = inet_opt->opt.faddr;
165 }
166
167 orig_sport = inet->inet_sport;
168 orig_dport = usin->sin_port;
169 fl4 = &inet->cork.fl.u.ip4;
170 rt = ip_route_connect(fl4, nexthop, inet->inet_saddr,
171 RT_CONN_FLAGS(sk), sk->sk_bound_dev_if,
172 IPPROTO_TCP,
173 orig_sport, orig_dport, sk);
174 if (IS_ERR(rt)) {
175 err = PTR_ERR(rt);
176 if (err == -ENETUNREACH)
177 IP_INC_STATS(sock_net(sk), IPSTATS_MIB_OUTNOROUTES);
178 return err;
179 }
180
181 if (rt->rt_flags & (RTCF_MULTICAST | RTCF_BROADCAST)) {
182 ip_rt_put(rt);
183 return -ENETUNREACH;
184 }
185
186 if (!inet_opt || !inet_opt->opt.srr)
187 daddr = fl4->daddr;
188
189 if (!inet->inet_saddr)
190 inet->inet_saddr = fl4->saddr;
191 sk_rcv_saddr_set(sk, inet->inet_saddr);
192
193 if (tp->rx_opt.ts_recent_stamp && inet->inet_daddr != daddr) {
194 /* Reset inherited state */
195 tp->rx_opt.ts_recent = 0;
196 tp->rx_opt.ts_recent_stamp = 0;
197 if (likely(!tp->repair))
198 tp->write_seq = 0;
199 }
200
201 if (tcp_death_row.sysctl_tw_recycle &&
202 !tp->rx_opt.ts_recent_stamp && fl4->daddr == daddr)
203 tcp_fetch_timewait_stamp(sk, &rt->dst);
204
205 inet->inet_dport = usin->sin_port;
206 sk_daddr_set(sk, daddr);
207
208 inet_csk(sk)->icsk_ext_hdr_len = 0;
209 if (inet_opt)
210 inet_csk(sk)->icsk_ext_hdr_len = inet_opt->opt.optlen;
211
212 tp->rx_opt.mss_clamp = TCP_MSS_DEFAULT;
213
214 /* Socket identity is still unknown (sport may be zero).
215 * However we set state to SYN-SENT and not releasing socket
216 * lock select source port, enter ourselves into the hash tables and
217 * complete initialization after this.
218 */
219 tcp_set_state(sk, TCP_SYN_SENT);
220 err = inet_hash_connect(&tcp_death_row, sk);
221 if (err)
222 goto failure;
223
224 sk_set_txhash(sk);
225
226 rt = ip_route_newports(fl4, rt, orig_sport, orig_dport,
227 inet->inet_sport, inet->inet_dport, sk);
228 if (IS_ERR(rt)) {
229 err = PTR_ERR(rt);
230 rt = NULL;
231 goto failure;
232 }
233 /* OK, now commit destination to socket. */
234 sk->sk_gso_type = SKB_GSO_TCPV4;
235 sk_setup_caps(sk, &rt->dst);
236
237 if (!tp->write_seq && likely(!tp->repair))
238 tp->write_seq = secure_tcp_sequence_number(inet->inet_saddr,
239 inet->inet_daddr,
240 inet->inet_sport,
241 usin->sin_port);
242
243 inet->inet_id = tp->write_seq ^ jiffies;
244
245 err = tcp_connect(sk);
246
247 rt = NULL;
248 if (err)
249 goto failure;
250
251 return 0;
252
253 failure:
254 /*
255 * This unhashes the socket and releases the local port,
256 * if necessary.
257 */
258 tcp_set_state(sk, TCP_CLOSE);
259 ip_rt_put(rt);
260 sk->sk_route_caps = 0;
261 inet->inet_dport = 0;
262 return err;
263 }
264 EXPORT_SYMBOL(tcp_v4_connect);
265
266 /*
267 * This routine reacts to ICMP_FRAG_NEEDED mtu indications as defined in RFC1191.
268 * It can be called through tcp_release_cb() if socket was owned by user
269 * at the time tcp_v4_err() was called to handle ICMP message.
270 */
271 void tcp_v4_mtu_reduced(struct sock *sk)
272 {
273 struct dst_entry *dst;
274 struct inet_sock *inet = inet_sk(sk);
275 u32 mtu = tcp_sk(sk)->mtu_info;
276
277 dst = inet_csk_update_pmtu(sk, mtu);
278 if (!dst)
279 return;
280
281 /* Something is about to be wrong... Remember soft error
282 * for the case, if this connection will not able to recover.
283 */
284 if (mtu < dst_mtu(dst) && ip_dont_fragment(sk, dst))
285 sk->sk_err_soft = EMSGSIZE;
286
287 mtu = dst_mtu(dst);
288
289 if (inet->pmtudisc != IP_PMTUDISC_DONT &&
290 ip_sk_accept_pmtu(sk) &&
291 inet_csk(sk)->icsk_pmtu_cookie > mtu) {
292 tcp_sync_mss(sk, mtu);
293
294 /* Resend the TCP packet because it's
295 * clear that the old packet has been
296 * dropped. This is the new "fast" path mtu
297 * discovery.
298 */
299 tcp_simple_retransmit(sk);
300 } /* else let the usual retransmit timer handle it */
301 }
302 EXPORT_SYMBOL(tcp_v4_mtu_reduced);
303
304 static void do_redirect(struct sk_buff *skb, struct sock *sk)
305 {
306 struct dst_entry *dst = __sk_dst_check(sk, 0);
307
308 if (dst)
309 dst->ops->redirect(dst, sk, skb);
310 }
311
312
313 /* handle ICMP messages on TCP_NEW_SYN_RECV request sockets */
314 void tcp_req_err(struct sock *sk, u32 seq)
315 {
316 struct request_sock *req = inet_reqsk(sk);
317 struct net *net = sock_net(sk);
318
319 /* ICMPs are not backlogged, hence we cannot get
320 * an established socket here.
321 */
322 WARN_ON(req->sk);
323
324 if (seq != tcp_rsk(req)->snt_isn) {
325 NET_INC_STATS_BH(net, LINUX_MIB_OUTOFWINDOWICMPS);
326 } else {
327 /*
328 * Still in SYN_RECV, just remove it silently.
329 * There is no good way to pass the error to the newly
330 * created socket, and POSIX does not want network
331 * errors returned from accept().
332 */
333 inet_csk_reqsk_queue_drop(req->rsk_listener, req);
334 NET_INC_STATS_BH(net, LINUX_MIB_LISTENDROPS);
335 }
336 reqsk_put(req);
337 }
338 EXPORT_SYMBOL(tcp_req_err);
339
340 /*
341 * This routine is called by the ICMP module when it gets some
342 * sort of error condition. If err < 0 then the socket should
343 * be closed and the error returned to the user. If err > 0
344 * it's just the icmp type << 8 | icmp code. After adjustment
345 * header points to the first 8 bytes of the tcp header. We need
346 * to find the appropriate port.
347 *
348 * The locking strategy used here is very "optimistic". When
349 * someone else accesses the socket the ICMP is just dropped
350 * and for some paths there is no check at all.
351 * A more general error queue to queue errors for later handling
352 * is probably better.
353 *
354 */
355
356 void tcp_v4_err(struct sk_buff *icmp_skb, u32 info)
357 {
358 const struct iphdr *iph = (const struct iphdr *)icmp_skb->data;
359 struct tcphdr *th = (struct tcphdr *)(icmp_skb->data + (iph->ihl << 2));
360 struct inet_connection_sock *icsk;
361 struct tcp_sock *tp;
362 struct inet_sock *inet;
363 const int type = icmp_hdr(icmp_skb)->type;
364 const int code = icmp_hdr(icmp_skb)->code;
365 struct sock *sk;
366 struct sk_buff *skb;
367 struct request_sock *fastopen;
368 __u32 seq, snd_una;
369 __u32 remaining;
370 int err;
371 struct net *net = dev_net(icmp_skb->dev);
372
373 sk = __inet_lookup_established(net, &tcp_hashinfo, iph->daddr,
374 th->dest, iph->saddr, ntohs(th->source),
375 inet_iif(icmp_skb));
376 if (!sk) {
377 ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS);
378 return;
379 }
380 if (sk->sk_state == TCP_TIME_WAIT) {
381 inet_twsk_put(inet_twsk(sk));
382 return;
383 }
384 seq = ntohl(th->seq);
385 if (sk->sk_state == TCP_NEW_SYN_RECV)
386 return tcp_req_err(sk, seq);
387
388 bh_lock_sock(sk);
389 /* If too many ICMPs get dropped on busy
390 * servers this needs to be solved differently.
391 * We do take care of PMTU discovery (RFC1191) special case :
392 * we can receive locally generated ICMP messages while socket is held.
393 */
394 if (sock_owned_by_user(sk)) {
395 if (!(type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED))
396 NET_INC_STATS_BH(net, LINUX_MIB_LOCKDROPPEDICMPS);
397 }
398 if (sk->sk_state == TCP_CLOSE)
399 goto out;
400
401 if (unlikely(iph->ttl < inet_sk(sk)->min_ttl)) {
402 NET_INC_STATS_BH(net, LINUX_MIB_TCPMINTTLDROP);
403 goto out;
404 }
405
406 icsk = inet_csk(sk);
407 tp = tcp_sk(sk);
408 /* XXX (TFO) - tp->snd_una should be ISN (tcp_create_openreq_child() */
409 fastopen = tp->fastopen_rsk;
410 snd_una = fastopen ? tcp_rsk(fastopen)->snt_isn : tp->snd_una;
411 if (sk->sk_state != TCP_LISTEN &&
412 !between(seq, snd_una, tp->snd_nxt)) {
413 NET_INC_STATS_BH(net, LINUX_MIB_OUTOFWINDOWICMPS);
414 goto out;
415 }
416
417 switch (type) {
418 case ICMP_REDIRECT:
419 do_redirect(icmp_skb, sk);
420 goto out;
421 case ICMP_SOURCE_QUENCH:
422 /* Just silently ignore these. */
423 goto out;
424 case ICMP_PARAMETERPROB:
425 err = EPROTO;
426 break;
427 case ICMP_DEST_UNREACH:
428 if (code > NR_ICMP_UNREACH)
429 goto out;
430
431 if (code == ICMP_FRAG_NEEDED) { /* PMTU discovery (RFC1191) */
432 /* We are not interested in TCP_LISTEN and open_requests
433 * (SYN-ACKs send out by Linux are always <576bytes so
434 * they should go through unfragmented).
435 */
436 if (sk->sk_state == TCP_LISTEN)
437 goto out;
438
439 tp->mtu_info = info;
440 if (!sock_owned_by_user(sk)) {
441 tcp_v4_mtu_reduced(sk);
442 } else {
443 if (!test_and_set_bit(TCP_MTU_REDUCED_DEFERRED, &tp->tsq_flags))
444 sock_hold(sk);
445 }
446 goto out;
447 }
448
449 err = icmp_err_convert[code].errno;
450 /* check if icmp_skb allows revert of backoff
451 * (see draft-zimmermann-tcp-lcd) */
452 if (code != ICMP_NET_UNREACH && code != ICMP_HOST_UNREACH)
453 break;
454 if (seq != tp->snd_una || !icsk->icsk_retransmits ||
455 !icsk->icsk_backoff || fastopen)
456 break;
457
458 if (sock_owned_by_user(sk))
459 break;
460
461 icsk->icsk_backoff--;
462 icsk->icsk_rto = tp->srtt_us ? __tcp_set_rto(tp) :
463 TCP_TIMEOUT_INIT;
464 icsk->icsk_rto = inet_csk_rto_backoff(icsk, TCP_RTO_MAX);
465
466 skb = tcp_write_queue_head(sk);
467 BUG_ON(!skb);
468
469 remaining = icsk->icsk_rto -
470 min(icsk->icsk_rto,
471 tcp_time_stamp - tcp_skb_timestamp(skb));
472
473 if (remaining) {
474 inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
475 remaining, TCP_RTO_MAX);
476 } else {
477 /* RTO revert clocked out retransmission.
478 * Will retransmit now */
479 tcp_retransmit_timer(sk);
480 }
481
482 break;
483 case ICMP_TIME_EXCEEDED:
484 err = EHOSTUNREACH;
485 break;
486 default:
487 goto out;
488 }
489
490 switch (sk->sk_state) {
491 case TCP_SYN_SENT:
492 case TCP_SYN_RECV:
493 /* Only in fast or simultaneous open. If a fast open socket is
494 * is already accepted it is treated as a connected one below.
495 */
496 if (fastopen && !fastopen->sk)
497 break;
498
499 if (!sock_owned_by_user(sk)) {
500 sk->sk_err = err;
501
502 sk->sk_error_report(sk);
503
504 tcp_done(sk);
505 } else {
506 sk->sk_err_soft = err;
507 }
508 goto out;
509 }
510
511 /* If we've already connected we will keep trying
512 * until we time out, or the user gives up.
513 *
514 * rfc1122 4.2.3.9 allows to consider as hard errors
515 * only PROTO_UNREACH and PORT_UNREACH (well, FRAG_FAILED too,
516 * but it is obsoleted by pmtu discovery).
517 *
518 * Note, that in modern internet, where routing is unreliable
519 * and in each dark corner broken firewalls sit, sending random
520 * errors ordered by their masters even this two messages finally lose
521 * their original sense (even Linux sends invalid PORT_UNREACHs)
522 *
523 * Now we are in compliance with RFCs.
524 * --ANK (980905)
525 */
526
527 inet = inet_sk(sk);
528 if (!sock_owned_by_user(sk) && inet->recverr) {
529 sk->sk_err = err;
530 sk->sk_error_report(sk);
531 } else { /* Only an error on timeout */
532 sk->sk_err_soft = err;
533 }
534
535 out:
536 bh_unlock_sock(sk);
537 sock_put(sk);
538 }
539
540 void __tcp_v4_send_check(struct sk_buff *skb, __be32 saddr, __be32 daddr)
541 {
542 struct tcphdr *th = tcp_hdr(skb);
543
544 if (skb->ip_summed == CHECKSUM_PARTIAL) {
545 th->check = ~tcp_v4_check(skb->len, saddr, daddr, 0);
546 skb->csum_start = skb_transport_header(skb) - skb->head;
547 skb->csum_offset = offsetof(struct tcphdr, check);
548 } else {
549 th->check = tcp_v4_check(skb->len, saddr, daddr,
550 csum_partial(th,
551 th->doff << 2,
552 skb->csum));
553 }
554 }
555
556 /* This routine computes an IPv4 TCP checksum. */
557 void tcp_v4_send_check(struct sock *sk, struct sk_buff *skb)
558 {
559 const struct inet_sock *inet = inet_sk(sk);
560
561 __tcp_v4_send_check(skb, inet->inet_saddr, inet->inet_daddr);
562 }
563 EXPORT_SYMBOL(tcp_v4_send_check);
564
565 /*
566 * This routine will send an RST to the other tcp.
567 *
568 * Someone asks: why I NEVER use socket parameters (TOS, TTL etc.)
569 * for reset.
570 * Answer: if a packet caused RST, it is not for a socket
571 * existing in our system, if it is matched to a socket,
572 * it is just duplicate segment or bug in other side's TCP.
573 * So that we build reply only basing on parameters
574 * arrived with segment.
575 * Exception: precedence violation. We do not implement it in any case.
576 */
577
578 static void tcp_v4_send_reset(const struct sock *sk, struct sk_buff *skb)
579 {
580 const struct tcphdr *th = tcp_hdr(skb);
581 struct {
582 struct tcphdr th;
583 #ifdef CONFIG_TCP_MD5SIG
584 __be32 opt[(TCPOLEN_MD5SIG_ALIGNED >> 2)];
585 #endif
586 } rep;
587 struct ip_reply_arg arg;
588 #ifdef CONFIG_TCP_MD5SIG
589 struct tcp_md5sig_key *key = NULL;
590 const __u8 *hash_location = NULL;
591 unsigned char newhash[16];
592 int genhash;
593 struct sock *sk1 = NULL;
594 #endif
595 struct net *net;
596
597 /* Never send a reset in response to a reset. */
598 if (th->rst)
599 return;
600
601 /* If sk not NULL, it means we did a successful lookup and incoming
602 * route had to be correct. prequeue might have dropped our dst.
603 */
604 if (!sk && skb_rtable(skb)->rt_type != RTN_LOCAL)
605 return;
606
607 /* Swap the send and the receive. */
608 memset(&rep, 0, sizeof(rep));
609 rep.th.dest = th->source;
610 rep.th.source = th->dest;
611 rep.th.doff = sizeof(struct tcphdr) / 4;
612 rep.th.rst = 1;
613
614 if (th->ack) {
615 rep.th.seq = th->ack_seq;
616 } else {
617 rep.th.ack = 1;
618 rep.th.ack_seq = htonl(ntohl(th->seq) + th->syn + th->fin +
619 skb->len - (th->doff << 2));
620 }
621
622 memset(&arg, 0, sizeof(arg));
623 arg.iov[0].iov_base = (unsigned char *)&rep;
624 arg.iov[0].iov_len = sizeof(rep.th);
625
626 net = sk ? sock_net(sk) : dev_net(skb_dst(skb)->dev);
627 #ifdef CONFIG_TCP_MD5SIG
628 hash_location = tcp_parse_md5sig_option(th);
629 if (sk && sk_fullsock(sk)) {
630 key = tcp_md5_do_lookup(sk, (union tcp_md5_addr *)
631 &ip_hdr(skb)->saddr, AF_INET);
632 } else if (hash_location) {
633 /*
634 * active side is lost. Try to find listening socket through
635 * source port, and then find md5 key through listening socket.
636 * we are not loose security here:
637 * Incoming packet is checked with md5 hash with finding key,
638 * no RST generated if md5 hash doesn't match.
639 */
640 sk1 = __inet_lookup_listener(net,
641 &tcp_hashinfo, ip_hdr(skb)->saddr,
642 th->source, ip_hdr(skb)->daddr,
643 ntohs(th->source), inet_iif(skb));
644 /* don't send rst if it can't find key */
645 if (!sk1)
646 return;
647 rcu_read_lock();
648 key = tcp_md5_do_lookup(sk1, (union tcp_md5_addr *)
649 &ip_hdr(skb)->saddr, AF_INET);
650 if (!key)
651 goto release_sk1;
652
653 genhash = tcp_v4_md5_hash_skb(newhash, key, NULL, skb);
654 if (genhash || memcmp(hash_location, newhash, 16) != 0)
655 goto release_sk1;
656 }
657
658 if (key) {
659 rep.opt[0] = htonl((TCPOPT_NOP << 24) |
660 (TCPOPT_NOP << 16) |
661 (TCPOPT_MD5SIG << 8) |
662 TCPOLEN_MD5SIG);
663 /* Update length and the length the header thinks exists */
664 arg.iov[0].iov_len += TCPOLEN_MD5SIG_ALIGNED;
665 rep.th.doff = arg.iov[0].iov_len / 4;
666
667 tcp_v4_md5_hash_hdr((__u8 *) &rep.opt[1],
668 key, ip_hdr(skb)->saddr,
669 ip_hdr(skb)->daddr, &rep.th);
670 }
671 #endif
672 arg.csum = csum_tcpudp_nofold(ip_hdr(skb)->daddr,
673 ip_hdr(skb)->saddr, /* XXX */
674 arg.iov[0].iov_len, IPPROTO_TCP, 0);
675 arg.csumoffset = offsetof(struct tcphdr, check) / 2;
676 arg.flags = (sk && inet_sk_transparent(sk)) ? IP_REPLY_ARG_NOSRCCHECK : 0;
677
678 /* When socket is gone, all binding information is lost.
679 * routing might fail in this case. No choice here, if we choose to force
680 * input interface, we will misroute in case of asymmetric route.
681 */
682 if (sk)
683 arg.bound_dev_if = sk->sk_bound_dev_if;
684
685 BUILD_BUG_ON(offsetof(struct sock, sk_bound_dev_if) !=
686 offsetof(struct inet_timewait_sock, tw_bound_dev_if));
687
688 arg.tos = ip_hdr(skb)->tos;
689 ip_send_unicast_reply(*this_cpu_ptr(net->ipv4.tcp_sk),
690 skb, &TCP_SKB_CB(skb)->header.h4.opt,
691 ip_hdr(skb)->saddr, ip_hdr(skb)->daddr,
692 &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 net *net,
711 struct sk_buff *skb, u32 seq, u32 ack,
712 u32 win, u32 tsval, u32 tsecr, int oif,
713 struct tcp_md5sig_key *key,
714 int reply_flags, u8 tos)
715 {
716 const struct tcphdr *th = tcp_hdr(skb);
717 struct {
718 struct tcphdr th;
719 __be32 opt[(TCPOLEN_TSTAMP_ALIGNED >> 2)
720 #ifdef CONFIG_TCP_MD5SIG
721 + (TCPOLEN_MD5SIG_ALIGNED >> 2)
722 #endif
723 ];
724 } rep;
725 struct ip_reply_arg arg;
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 (tsecr) {
733 rep.opt[0] = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
734 (TCPOPT_TIMESTAMP << 8) |
735 TCPOLEN_TIMESTAMP);
736 rep.opt[1] = htonl(tsval);
737 rep.opt[2] = htonl(tsecr);
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 = (tsecr) ? 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(*this_cpu_ptr(net->ipv4.tcp_sk),
775 skb, &TCP_SKB_CB(skb)->header.h4.opt,
776 ip_hdr(skb)->saddr, ip_hdr(skb)->daddr,
777 &arg, arg.iov[0].iov_len);
778
779 TCP_INC_STATS_BH(net, TCP_MIB_OUTSEGS);
780 }
781
782 static void tcp_v4_timewait_ack(struct sock *sk, struct sk_buff *skb)
783 {
784 struct inet_timewait_sock *tw = inet_twsk(sk);
785 struct tcp_timewait_sock *tcptw = tcp_twsk(sk);
786
787 tcp_v4_send_ack(sock_net(sk), skb,
788 tcptw->tw_snd_nxt, tcptw->tw_rcv_nxt,
789 tcptw->tw_rcv_wnd >> tw->tw_rcv_wscale,
790 tcp_time_stamp + tcptw->tw_ts_offset,
791 tcptw->tw_ts_recent,
792 tw->tw_bound_dev_if,
793 tcp_twsk_md5_key(tcptw),
794 tw->tw_transparent ? IP_REPLY_ARG_NOSRCCHECK : 0,
795 tw->tw_tos
796 );
797
798 inet_twsk_put(tw);
799 }
800
801 static void tcp_v4_reqsk_send_ack(const struct sock *sk, struct sk_buff *skb,
802 struct request_sock *req)
803 {
804 /* sk->sk_state == TCP_LISTEN -> for regular TCP_SYN_RECV
805 * sk->sk_state == TCP_SYN_RECV -> for Fast Open.
806 */
807 u32 seq = (sk->sk_state == TCP_LISTEN) ? tcp_rsk(req)->snt_isn + 1 :
808 tcp_sk(sk)->snd_nxt;
809
810 tcp_v4_send_ack(sock_net(sk), skb, seq,
811 tcp_rsk(req)->rcv_nxt, req->rsk_rcv_wnd,
812 tcp_time_stamp,
813 req->ts_recent,
814 0,
815 tcp_md5_do_lookup(sk, (union tcp_md5_addr *)&ip_hdr(skb)->daddr,
816 AF_INET),
817 inet_rsk(req)->no_srccheck ? IP_REPLY_ARG_NOSRCCHECK : 0,
818 ip_hdr(skb)->tos);
819 }
820
821 /*
822 * Send a SYN-ACK after having received a SYN.
823 * This still operates on a request_sock only, not on a big
824 * socket.
825 */
826 static int tcp_v4_send_synack(const struct sock *sk, struct dst_entry *dst,
827 struct flowi *fl,
828 struct request_sock *req,
829 struct tcp_fastopen_cookie *foc,
830 bool attach_req)
831 {
832 const struct inet_request_sock *ireq = inet_rsk(req);
833 struct flowi4 fl4;
834 int err = -1;
835 struct sk_buff *skb;
836
837 /* First, grab a route. */
838 if (!dst && (dst = inet_csk_route_req(sk, &fl4, req)) == NULL)
839 return -1;
840
841 skb = tcp_make_synack(sk, dst, req, foc, attach_req);
842
843 if (skb) {
844 __tcp_v4_send_check(skb, ireq->ir_loc_addr, ireq->ir_rmt_addr);
845
846 err = ip_build_and_send_pkt(skb, sk, ireq->ir_loc_addr,
847 ireq->ir_rmt_addr,
848 ireq->opt);
849 err = net_xmit_eval(err);
850 }
851
852 return err;
853 }
854
855 /*
856 * IPv4 request_sock destructor.
857 */
858 static void tcp_v4_reqsk_destructor(struct request_sock *req)
859 {
860 kfree(inet_rsk(req)->opt);
861 }
862
863
864 #ifdef CONFIG_TCP_MD5SIG
865 /*
866 * RFC2385 MD5 checksumming requires a mapping of
867 * IP address->MD5 Key.
868 * We need to maintain these in the sk structure.
869 */
870
871 /* Find the Key structure for an address. */
872 struct tcp_md5sig_key *tcp_md5_do_lookup(const struct sock *sk,
873 const union tcp_md5_addr *addr,
874 int family)
875 {
876 const struct tcp_sock *tp = tcp_sk(sk);
877 struct tcp_md5sig_key *key;
878 unsigned int size = sizeof(struct in_addr);
879 const struct tcp_md5sig_info *md5sig;
880
881 /* caller either holds rcu_read_lock() or socket lock */
882 md5sig = rcu_dereference_check(tp->md5sig_info,
883 sock_owned_by_user(sk) ||
884 lockdep_is_held((spinlock_t *)&sk->sk_lock.slock));
885 if (!md5sig)
886 return NULL;
887 #if IS_ENABLED(CONFIG_IPV6)
888 if (family == AF_INET6)
889 size = sizeof(struct in6_addr);
890 #endif
891 hlist_for_each_entry_rcu(key, &md5sig->head, node) {
892 if (key->family != family)
893 continue;
894 if (!memcmp(&key->addr, addr, size))
895 return key;
896 }
897 return NULL;
898 }
899 EXPORT_SYMBOL(tcp_md5_do_lookup);
900
901 struct tcp_md5sig_key *tcp_v4_md5_lookup(const struct sock *sk,
902 const struct sock *addr_sk)
903 {
904 const union tcp_md5_addr *addr;
905
906 addr = (const union tcp_md5_addr *)&addr_sk->sk_daddr;
907 return tcp_md5_do_lookup(sk, addr, AF_INET);
908 }
909 EXPORT_SYMBOL(tcp_v4_md5_lookup);
910
911 /* This can be called on a newly created socket, from other files */
912 int tcp_md5_do_add(struct sock *sk, const union tcp_md5_addr *addr,
913 int family, const u8 *newkey, u8 newkeylen, gfp_t gfp)
914 {
915 /* Add Key to the list */
916 struct tcp_md5sig_key *key;
917 struct tcp_sock *tp = tcp_sk(sk);
918 struct tcp_md5sig_info *md5sig;
919
920 key = tcp_md5_do_lookup(sk, addr, family);
921 if (key) {
922 /* Pre-existing entry - just update that one. */
923 memcpy(key->key, newkey, newkeylen);
924 key->keylen = newkeylen;
925 return 0;
926 }
927
928 md5sig = rcu_dereference_protected(tp->md5sig_info,
929 sock_owned_by_user(sk) ||
930 lockdep_is_held(&sk->sk_lock.slock));
931 if (!md5sig) {
932 md5sig = kmalloc(sizeof(*md5sig), gfp);
933 if (!md5sig)
934 return -ENOMEM;
935
936 sk_nocaps_add(sk, NETIF_F_GSO_MASK);
937 INIT_HLIST_HEAD(&md5sig->head);
938 rcu_assign_pointer(tp->md5sig_info, md5sig);
939 }
940
941 key = sock_kmalloc(sk, sizeof(*key), gfp);
942 if (!key)
943 return -ENOMEM;
944 if (!tcp_alloc_md5sig_pool()) {
945 sock_kfree_s(sk, key, sizeof(*key));
946 return -ENOMEM;
947 }
948
949 memcpy(key->key, newkey, newkeylen);
950 key->keylen = newkeylen;
951 key->family = family;
952 memcpy(&key->addr, addr,
953 (family == AF_INET6) ? sizeof(struct in6_addr) :
954 sizeof(struct in_addr));
955 hlist_add_head_rcu(&key->node, &md5sig->head);
956 return 0;
957 }
958 EXPORT_SYMBOL(tcp_md5_do_add);
959
960 int tcp_md5_do_del(struct sock *sk, const union tcp_md5_addr *addr, int family)
961 {
962 struct tcp_md5sig_key *key;
963
964 key = tcp_md5_do_lookup(sk, addr, family);
965 if (!key)
966 return -ENOENT;
967 hlist_del_rcu(&key->node);
968 atomic_sub(sizeof(*key), &sk->sk_omem_alloc);
969 kfree_rcu(key, rcu);
970 return 0;
971 }
972 EXPORT_SYMBOL(tcp_md5_do_del);
973
974 static void tcp_clear_md5_list(struct sock *sk)
975 {
976 struct tcp_sock *tp = tcp_sk(sk);
977 struct tcp_md5sig_key *key;
978 struct hlist_node *n;
979 struct tcp_md5sig_info *md5sig;
980
981 md5sig = rcu_dereference_protected(tp->md5sig_info, 1);
982
983 hlist_for_each_entry_safe(key, n, &md5sig->head, node) {
984 hlist_del_rcu(&key->node);
985 atomic_sub(sizeof(*key), &sk->sk_omem_alloc);
986 kfree_rcu(key, rcu);
987 }
988 }
989
990 static int tcp_v4_parse_md5_keys(struct sock *sk, char __user *optval,
991 int optlen)
992 {
993 struct tcp_md5sig cmd;
994 struct sockaddr_in *sin = (struct sockaddr_in *)&cmd.tcpm_addr;
995
996 if (optlen < sizeof(cmd))
997 return -EINVAL;
998
999 if (copy_from_user(&cmd, optval, sizeof(cmd)))
1000 return -EFAULT;
1001
1002 if (sin->sin_family != AF_INET)
1003 return -EINVAL;
1004
1005 if (!cmd.tcpm_keylen)
1006 return tcp_md5_do_del(sk, (union tcp_md5_addr *)&sin->sin_addr.s_addr,
1007 AF_INET);
1008
1009 if (cmd.tcpm_keylen > TCP_MD5SIG_MAXKEYLEN)
1010 return -EINVAL;
1011
1012 return tcp_md5_do_add(sk, (union tcp_md5_addr *)&sin->sin_addr.s_addr,
1013 AF_INET, cmd.tcpm_key, cmd.tcpm_keylen,
1014 GFP_KERNEL);
1015 }
1016
1017 static int tcp_v4_md5_hash_pseudoheader(struct tcp_md5sig_pool *hp,
1018 __be32 daddr, __be32 saddr, int nbytes)
1019 {
1020 struct tcp4_pseudohdr *bp;
1021 struct scatterlist sg;
1022
1023 bp = &hp->md5_blk.ip4;
1024
1025 /*
1026 * 1. the TCP pseudo-header (in the order: source IP address,
1027 * destination IP address, zero-padded protocol number, and
1028 * segment length)
1029 */
1030 bp->saddr = saddr;
1031 bp->daddr = daddr;
1032 bp->pad = 0;
1033 bp->protocol = IPPROTO_TCP;
1034 bp->len = cpu_to_be16(nbytes);
1035
1036 sg_init_one(&sg, bp, sizeof(*bp));
1037 return crypto_hash_update(&hp->md5_desc, &sg, sizeof(*bp));
1038 }
1039
1040 static int tcp_v4_md5_hash_hdr(char *md5_hash, const struct tcp_md5sig_key *key,
1041 __be32 daddr, __be32 saddr, const struct tcphdr *th)
1042 {
1043 struct tcp_md5sig_pool *hp;
1044 struct hash_desc *desc;
1045
1046 hp = tcp_get_md5sig_pool();
1047 if (!hp)
1048 goto clear_hash_noput;
1049 desc = &hp->md5_desc;
1050
1051 if (crypto_hash_init(desc))
1052 goto clear_hash;
1053 if (tcp_v4_md5_hash_pseudoheader(hp, daddr, saddr, th->doff << 2))
1054 goto clear_hash;
1055 if (tcp_md5_hash_header(hp, th))
1056 goto clear_hash;
1057 if (tcp_md5_hash_key(hp, key))
1058 goto clear_hash;
1059 if (crypto_hash_final(desc, md5_hash))
1060 goto clear_hash;
1061
1062 tcp_put_md5sig_pool();
1063 return 0;
1064
1065 clear_hash:
1066 tcp_put_md5sig_pool();
1067 clear_hash_noput:
1068 memset(md5_hash, 0, 16);
1069 return 1;
1070 }
1071
1072 int tcp_v4_md5_hash_skb(char *md5_hash, const struct tcp_md5sig_key *key,
1073 const struct sock *sk,
1074 const struct sk_buff *skb)
1075 {
1076 struct tcp_md5sig_pool *hp;
1077 struct hash_desc *desc;
1078 const struct tcphdr *th = tcp_hdr(skb);
1079 __be32 saddr, daddr;
1080
1081 if (sk) { /* valid for establish/request sockets */
1082 saddr = sk->sk_rcv_saddr;
1083 daddr = sk->sk_daddr;
1084 } else {
1085 const struct iphdr *iph = ip_hdr(skb);
1086 saddr = iph->saddr;
1087 daddr = iph->daddr;
1088 }
1089
1090 hp = tcp_get_md5sig_pool();
1091 if (!hp)
1092 goto clear_hash_noput;
1093 desc = &hp->md5_desc;
1094
1095 if (crypto_hash_init(desc))
1096 goto clear_hash;
1097
1098 if (tcp_v4_md5_hash_pseudoheader(hp, daddr, saddr, skb->len))
1099 goto clear_hash;
1100 if (tcp_md5_hash_header(hp, th))
1101 goto clear_hash;
1102 if (tcp_md5_hash_skb_data(hp, skb, th->doff << 2))
1103 goto clear_hash;
1104 if (tcp_md5_hash_key(hp, key))
1105 goto clear_hash;
1106 if (crypto_hash_final(desc, md5_hash))
1107 goto clear_hash;
1108
1109 tcp_put_md5sig_pool();
1110 return 0;
1111
1112 clear_hash:
1113 tcp_put_md5sig_pool();
1114 clear_hash_noput:
1115 memset(md5_hash, 0, 16);
1116 return 1;
1117 }
1118 EXPORT_SYMBOL(tcp_v4_md5_hash_skb);
1119
1120 #endif
1121
1122 /* Called with rcu_read_lock() */
1123 static bool tcp_v4_inbound_md5_hash(const struct sock *sk,
1124 const struct sk_buff *skb)
1125 {
1126 #ifdef CONFIG_TCP_MD5SIG
1127 /*
1128 * This gets called for each TCP segment that arrives
1129 * so we want to be efficient.
1130 * We have 3 drop cases:
1131 * o No MD5 hash and one expected.
1132 * o MD5 hash and we're not expecting one.
1133 * o MD5 hash and its wrong.
1134 */
1135 const __u8 *hash_location = NULL;
1136 struct tcp_md5sig_key *hash_expected;
1137 const struct iphdr *iph = ip_hdr(skb);
1138 const struct tcphdr *th = tcp_hdr(skb);
1139 int genhash;
1140 unsigned char newhash[16];
1141
1142 hash_expected = tcp_md5_do_lookup(sk, (union tcp_md5_addr *)&iph->saddr,
1143 AF_INET);
1144 hash_location = tcp_parse_md5sig_option(th);
1145
1146 /* We've parsed the options - do we have a hash? */
1147 if (!hash_expected && !hash_location)
1148 return false;
1149
1150 if (hash_expected && !hash_location) {
1151 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPMD5NOTFOUND);
1152 return true;
1153 }
1154
1155 if (!hash_expected && hash_location) {
1156 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPMD5UNEXPECTED);
1157 return true;
1158 }
1159
1160 /* Okay, so this is hash_expected and hash_location -
1161 * so we need to calculate the checksum.
1162 */
1163 genhash = tcp_v4_md5_hash_skb(newhash,
1164 hash_expected,
1165 NULL, skb);
1166
1167 if (genhash || memcmp(hash_location, newhash, 16) != 0) {
1168 net_info_ratelimited("MD5 Hash failed for (%pI4, %d)->(%pI4, %d)%s\n",
1169 &iph->saddr, ntohs(th->source),
1170 &iph->daddr, ntohs(th->dest),
1171 genhash ? " tcp_v4_calc_md5_hash failed"
1172 : "");
1173 return true;
1174 }
1175 return false;
1176 #endif
1177 return false;
1178 }
1179
1180 static void tcp_v4_init_req(struct request_sock *req,
1181 const struct sock *sk_listener,
1182 struct sk_buff *skb)
1183 {
1184 struct inet_request_sock *ireq = inet_rsk(req);
1185
1186 sk_rcv_saddr_set(req_to_sk(req), ip_hdr(skb)->daddr);
1187 sk_daddr_set(req_to_sk(req), ip_hdr(skb)->saddr);
1188 ireq->no_srccheck = inet_sk(sk_listener)->transparent;
1189 ireq->opt = tcp_v4_save_options(skb);
1190 }
1191
1192 static struct dst_entry *tcp_v4_route_req(const struct sock *sk,
1193 struct flowi *fl,
1194 const struct request_sock *req,
1195 bool *strict)
1196 {
1197 struct dst_entry *dst = inet_csk_route_req(sk, &fl->u.ip4, req);
1198
1199 if (strict) {
1200 if (fl->u.ip4.daddr == inet_rsk(req)->ir_rmt_addr)
1201 *strict = true;
1202 else
1203 *strict = false;
1204 }
1205
1206 return dst;
1207 }
1208
1209 struct request_sock_ops tcp_request_sock_ops __read_mostly = {
1210 .family = PF_INET,
1211 .obj_size = sizeof(struct tcp_request_sock),
1212 .rtx_syn_ack = tcp_rtx_synack,
1213 .send_ack = tcp_v4_reqsk_send_ack,
1214 .destructor = tcp_v4_reqsk_destructor,
1215 .send_reset = tcp_v4_send_reset,
1216 .syn_ack_timeout = tcp_syn_ack_timeout,
1217 };
1218
1219 static const struct tcp_request_sock_ops tcp_request_sock_ipv4_ops = {
1220 .mss_clamp = TCP_MSS_DEFAULT,
1221 #ifdef CONFIG_TCP_MD5SIG
1222 .req_md5_lookup = tcp_v4_md5_lookup,
1223 .calc_md5_hash = tcp_v4_md5_hash_skb,
1224 #endif
1225 .init_req = tcp_v4_init_req,
1226 #ifdef CONFIG_SYN_COOKIES
1227 .cookie_init_seq = cookie_v4_init_sequence,
1228 #endif
1229 .route_req = tcp_v4_route_req,
1230 .init_seq = tcp_v4_init_sequence,
1231 .send_synack = tcp_v4_send_synack,
1232 };
1233
1234 int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb)
1235 {
1236 /* Never answer to SYNs send to broadcast or multicast */
1237 if (skb_rtable(skb)->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))
1238 goto drop;
1239
1240 return tcp_conn_request(&tcp_request_sock_ops,
1241 &tcp_request_sock_ipv4_ops, sk, skb);
1242
1243 drop:
1244 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENDROPS);
1245 return 0;
1246 }
1247 EXPORT_SYMBOL(tcp_v4_conn_request);
1248
1249
1250 /*
1251 * The three way handshake has completed - we got a valid synack -
1252 * now create the new socket.
1253 */
1254 struct sock *tcp_v4_syn_recv_sock(const struct sock *sk, struct sk_buff *skb,
1255 struct request_sock *req,
1256 struct dst_entry *dst,
1257 struct request_sock *req_unhash,
1258 bool *own_req)
1259 {
1260 struct inet_request_sock *ireq;
1261 struct inet_sock *newinet;
1262 struct tcp_sock *newtp;
1263 struct sock *newsk;
1264 #ifdef CONFIG_TCP_MD5SIG
1265 struct tcp_md5sig_key *key;
1266 #endif
1267 struct ip_options_rcu *inet_opt;
1268
1269 if (sk_acceptq_is_full(sk))
1270 goto exit_overflow;
1271
1272 newsk = tcp_create_openreq_child(sk, req, skb);
1273 if (!newsk)
1274 goto exit_nonewsk;
1275
1276 newsk->sk_gso_type = SKB_GSO_TCPV4;
1277 inet_sk_rx_dst_set(newsk, skb);
1278
1279 newtp = tcp_sk(newsk);
1280 newinet = inet_sk(newsk);
1281 ireq = inet_rsk(req);
1282 sk_daddr_set(newsk, ireq->ir_rmt_addr);
1283 sk_rcv_saddr_set(newsk, ireq->ir_loc_addr);
1284 newsk->sk_bound_dev_if = ireq->ir_iif;
1285 newinet->inet_saddr = ireq->ir_loc_addr;
1286 inet_opt = ireq->opt;
1287 rcu_assign_pointer(newinet->inet_opt, inet_opt);
1288 ireq->opt = NULL;
1289 newinet->mc_index = inet_iif(skb);
1290 newinet->mc_ttl = ip_hdr(skb)->ttl;
1291 newinet->rcv_tos = ip_hdr(skb)->tos;
1292 inet_csk(newsk)->icsk_ext_hdr_len = 0;
1293 if (inet_opt)
1294 inet_csk(newsk)->icsk_ext_hdr_len = inet_opt->opt.optlen;
1295 newinet->inet_id = newtp->write_seq ^ jiffies;
1296
1297 if (!dst) {
1298 dst = inet_csk_route_child_sock(sk, newsk, req);
1299 if (!dst)
1300 goto put_and_exit;
1301 } else {
1302 /* syncookie case : see end of cookie_v4_check() */
1303 }
1304 sk_setup_caps(newsk, dst);
1305
1306 tcp_ca_openreq_child(newsk, dst);
1307
1308 tcp_sync_mss(newsk, dst_mtu(dst));
1309 newtp->advmss = dst_metric_advmss(dst);
1310 if (tcp_sk(sk)->rx_opt.user_mss &&
1311 tcp_sk(sk)->rx_opt.user_mss < newtp->advmss)
1312 newtp->advmss = tcp_sk(sk)->rx_opt.user_mss;
1313
1314 tcp_initialize_rcv_mss(newsk);
1315
1316 #ifdef CONFIG_TCP_MD5SIG
1317 /* Copy over the MD5 key from the original socket */
1318 key = tcp_md5_do_lookup(sk, (union tcp_md5_addr *)&newinet->inet_daddr,
1319 AF_INET);
1320 if (key) {
1321 /*
1322 * We're using one, so create a matching key
1323 * on the newsk structure. If we fail to get
1324 * memory, then we end up not copying the key
1325 * across. Shucks.
1326 */
1327 tcp_md5_do_add(newsk, (union tcp_md5_addr *)&newinet->inet_daddr,
1328 AF_INET, key->key, key->keylen, GFP_ATOMIC);
1329 sk_nocaps_add(newsk, NETIF_F_GSO_MASK);
1330 }
1331 #endif
1332
1333 if (__inet_inherit_port(sk, newsk) < 0)
1334 goto put_and_exit;
1335 *own_req = inet_ehash_nolisten(newsk, req_to_sk(req_unhash));
1336 if (*own_req)
1337 tcp_move_syn(newtp, req);
1338
1339 return newsk;
1340
1341 exit_overflow:
1342 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENOVERFLOWS);
1343 exit_nonewsk:
1344 dst_release(dst);
1345 exit:
1346 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENDROPS);
1347 return NULL;
1348 put_and_exit:
1349 inet_csk_prepare_forced_close(newsk);
1350 tcp_done(newsk);
1351 goto exit;
1352 }
1353 EXPORT_SYMBOL(tcp_v4_syn_recv_sock);
1354
1355 static struct sock *tcp_v4_cookie_check(struct sock *sk, struct sk_buff *skb)
1356 {
1357 #ifdef CONFIG_SYN_COOKIES
1358 const struct tcphdr *th = tcp_hdr(skb);
1359
1360 if (!th->syn)
1361 sk = cookie_v4_check(sk, skb);
1362 #endif
1363 return sk;
1364 }
1365
1366 /* The socket must have it's spinlock held when we get
1367 * here, unless it is a TCP_LISTEN socket.
1368 *
1369 * We have a potential double-lock case here, so even when
1370 * doing backlog processing we use the BH locking scheme.
1371 * This is because we cannot sleep with the original spinlock
1372 * held.
1373 */
1374 int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb)
1375 {
1376 struct sock *rsk;
1377
1378 if (sk->sk_state == TCP_ESTABLISHED) { /* Fast path */
1379 struct dst_entry *dst = sk->sk_rx_dst;
1380
1381 sock_rps_save_rxhash(sk, skb);
1382 sk_mark_napi_id(sk, skb);
1383 if (dst) {
1384 if (inet_sk(sk)->rx_dst_ifindex != skb->skb_iif ||
1385 !dst->ops->check(dst, 0)) {
1386 dst_release(dst);
1387 sk->sk_rx_dst = NULL;
1388 }
1389 }
1390 tcp_rcv_established(sk, skb, tcp_hdr(skb), skb->len);
1391 return 0;
1392 }
1393
1394 if (tcp_checksum_complete(skb))
1395 goto csum_err;
1396
1397 if (sk->sk_state == TCP_LISTEN) {
1398 struct sock *nsk = tcp_v4_cookie_check(sk, skb);
1399
1400 if (!nsk)
1401 goto discard;
1402 if (nsk != sk) {
1403 sock_rps_save_rxhash(nsk, skb);
1404 sk_mark_napi_id(nsk, skb);
1405 if (tcp_child_process(sk, nsk, skb)) {
1406 rsk = nsk;
1407 goto reset;
1408 }
1409 return 0;
1410 }
1411 } else
1412 sock_rps_save_rxhash(sk, skb);
1413
1414 if (tcp_rcv_state_process(sk, skb)) {
1415 rsk = sk;
1416 goto reset;
1417 }
1418 return 0;
1419
1420 reset:
1421 tcp_v4_send_reset(rsk, skb);
1422 discard:
1423 kfree_skb(skb);
1424 /* Be careful here. If this function gets more complicated and
1425 * gcc suffers from register pressure on the x86, sk (in %ebx)
1426 * might be destroyed here. This current version compiles correctly,
1427 * but you have been warned.
1428 */
1429 return 0;
1430
1431 csum_err:
1432 TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_CSUMERRORS);
1433 TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_INERRS);
1434 goto discard;
1435 }
1436 EXPORT_SYMBOL(tcp_v4_do_rcv);
1437
1438 void tcp_v4_early_demux(struct sk_buff *skb)
1439 {
1440 const struct iphdr *iph;
1441 const struct tcphdr *th;
1442 struct sock *sk;
1443
1444 if (skb->pkt_type != PACKET_HOST)
1445 return;
1446
1447 if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct tcphdr)))
1448 return;
1449
1450 iph = ip_hdr(skb);
1451 th = tcp_hdr(skb);
1452
1453 if (th->doff < sizeof(struct tcphdr) / 4)
1454 return;
1455
1456 sk = __inet_lookup_established(dev_net(skb->dev), &tcp_hashinfo,
1457 iph->saddr, th->source,
1458 iph->daddr, ntohs(th->dest),
1459 skb->skb_iif);
1460 if (sk) {
1461 skb->sk = sk;
1462 skb->destructor = sock_edemux;
1463 if (sk_fullsock(sk)) {
1464 struct dst_entry *dst = READ_ONCE(sk->sk_rx_dst);
1465
1466 if (dst)
1467 dst = dst_check(dst, 0);
1468 if (dst &&
1469 inet_sk(sk)->rx_dst_ifindex == skb->skb_iif)
1470 skb_dst_set_noref(skb, dst);
1471 }
1472 }
1473 }
1474
1475 /* Packet is added to VJ-style prequeue for processing in process
1476 * context, if a reader task is waiting. Apparently, this exciting
1477 * idea (VJ's mail "Re: query about TCP header on tcp-ip" of 07 Sep 93)
1478 * failed somewhere. Latency? Burstiness? Well, at least now we will
1479 * see, why it failed. 8)8) --ANK
1480 *
1481 */
1482 bool tcp_prequeue(struct sock *sk, struct sk_buff *skb)
1483 {
1484 struct tcp_sock *tp = tcp_sk(sk);
1485
1486 if (sysctl_tcp_low_latency || !tp->ucopy.task)
1487 return false;
1488
1489 if (skb->len <= tcp_hdrlen(skb) &&
1490 skb_queue_len(&tp->ucopy.prequeue) == 0)
1491 return false;
1492
1493 /* Before escaping RCU protected region, we need to take care of skb
1494 * dst. Prequeue is only enabled for established sockets.
1495 * For such sockets, we might need the skb dst only to set sk->sk_rx_dst
1496 * Instead of doing full sk_rx_dst validity here, let's perform
1497 * an optimistic check.
1498 */
1499 if (likely(sk->sk_rx_dst))
1500 skb_dst_drop(skb);
1501 else
1502 skb_dst_force_safe(skb);
1503
1504 __skb_queue_tail(&tp->ucopy.prequeue, skb);
1505 tp->ucopy.memory += skb->truesize;
1506 if (tp->ucopy.memory > sk->sk_rcvbuf) {
1507 struct sk_buff *skb1;
1508
1509 BUG_ON(sock_owned_by_user(sk));
1510
1511 while ((skb1 = __skb_dequeue(&tp->ucopy.prequeue)) != NULL) {
1512 sk_backlog_rcv(sk, skb1);
1513 NET_INC_STATS_BH(sock_net(sk),
1514 LINUX_MIB_TCPPREQUEUEDROPPED);
1515 }
1516
1517 tp->ucopy.memory = 0;
1518 } else if (skb_queue_len(&tp->ucopy.prequeue) == 1) {
1519 wake_up_interruptible_sync_poll(sk_sleep(sk),
1520 POLLIN | POLLRDNORM | POLLRDBAND);
1521 if (!inet_csk_ack_scheduled(sk))
1522 inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK,
1523 (3 * tcp_rto_min(sk)) / 4,
1524 TCP_RTO_MAX);
1525 }
1526 return true;
1527 }
1528 EXPORT_SYMBOL(tcp_prequeue);
1529
1530 /*
1531 * From tcp_input.c
1532 */
1533
1534 int tcp_v4_rcv(struct sk_buff *skb)
1535 {
1536 const struct iphdr *iph;
1537 const struct tcphdr *th;
1538 struct sock *sk;
1539 int ret;
1540 struct net *net = dev_net(skb->dev);
1541
1542 if (skb->pkt_type != PACKET_HOST)
1543 goto discard_it;
1544
1545 /* Count it even if it's bad */
1546 TCP_INC_STATS_BH(net, TCP_MIB_INSEGS);
1547
1548 if (!pskb_may_pull(skb, sizeof(struct tcphdr)))
1549 goto discard_it;
1550
1551 th = tcp_hdr(skb);
1552
1553 if (th->doff < sizeof(struct tcphdr) / 4)
1554 goto bad_packet;
1555 if (!pskb_may_pull(skb, th->doff * 4))
1556 goto discard_it;
1557
1558 /* An explanation is required here, I think.
1559 * Packet length and doff are validated by header prediction,
1560 * provided case of th->doff==0 is eliminated.
1561 * So, we defer the checks. */
1562
1563 if (skb_checksum_init(skb, IPPROTO_TCP, inet_compute_pseudo))
1564 goto csum_error;
1565
1566 th = tcp_hdr(skb);
1567 iph = ip_hdr(skb);
1568 /* This is tricky : We move IPCB at its correct location into TCP_SKB_CB()
1569 * barrier() makes sure compiler wont play fool^Waliasing games.
1570 */
1571 memmove(&TCP_SKB_CB(skb)->header.h4, IPCB(skb),
1572 sizeof(struct inet_skb_parm));
1573 barrier();
1574
1575 TCP_SKB_CB(skb)->seq = ntohl(th->seq);
1576 TCP_SKB_CB(skb)->end_seq = (TCP_SKB_CB(skb)->seq + th->syn + th->fin +
1577 skb->len - th->doff * 4);
1578 TCP_SKB_CB(skb)->ack_seq = ntohl(th->ack_seq);
1579 TCP_SKB_CB(skb)->tcp_flags = tcp_flag_byte(th);
1580 TCP_SKB_CB(skb)->tcp_tw_isn = 0;
1581 TCP_SKB_CB(skb)->ip_dsfield = ipv4_get_dsfield(iph);
1582 TCP_SKB_CB(skb)->sacked = 0;
1583
1584 lookup:
1585 sk = __inet_lookup_skb(&tcp_hashinfo, skb, th->source, th->dest);
1586 if (!sk)
1587 goto no_tcp_socket;
1588
1589 process:
1590 if (sk->sk_state == TCP_TIME_WAIT)
1591 goto do_time_wait;
1592
1593 if (sk->sk_state == TCP_NEW_SYN_RECV) {
1594 struct request_sock *req = inet_reqsk(sk);
1595 struct sock *nsk = NULL;
1596
1597 sk = req->rsk_listener;
1598 if (tcp_v4_inbound_md5_hash(sk, skb))
1599 goto discard_and_relse;
1600 if (likely(sk->sk_state == TCP_LISTEN)) {
1601 nsk = tcp_check_req(sk, skb, req, false);
1602 } else {
1603 inet_csk_reqsk_queue_drop_and_put(sk, req);
1604 goto lookup;
1605 }
1606 if (!nsk) {
1607 reqsk_put(req);
1608 goto discard_it;
1609 }
1610 if (nsk == sk) {
1611 sock_hold(sk);
1612 reqsk_put(req);
1613 } else if (tcp_child_process(sk, nsk, skb)) {
1614 tcp_v4_send_reset(nsk, skb);
1615 goto discard_it;
1616 } else {
1617 return 0;
1618 }
1619 }
1620 if (unlikely(iph->ttl < inet_sk(sk)->min_ttl)) {
1621 NET_INC_STATS_BH(net, LINUX_MIB_TCPMINTTLDROP);
1622 goto discard_and_relse;
1623 }
1624
1625 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
1626 goto discard_and_relse;
1627
1628 if (tcp_v4_inbound_md5_hash(sk, skb))
1629 goto discard_and_relse;
1630
1631 nf_reset(skb);
1632
1633 if (sk_filter(sk, skb))
1634 goto discard_and_relse;
1635
1636 skb->dev = NULL;
1637
1638 if (sk->sk_state == TCP_LISTEN) {
1639 ret = tcp_v4_do_rcv(sk, skb);
1640 goto put_and_return;
1641 }
1642
1643 sk_incoming_cpu_update(sk);
1644
1645 bh_lock_sock_nested(sk);
1646 tcp_sk(sk)->segs_in += max_t(u16, 1, skb_shinfo(skb)->gso_segs);
1647 ret = 0;
1648 if (!sock_owned_by_user(sk)) {
1649 if (!tcp_prequeue(sk, skb))
1650 ret = tcp_v4_do_rcv(sk, skb);
1651 } else if (unlikely(sk_add_backlog(sk, skb,
1652 sk->sk_rcvbuf + sk->sk_sndbuf))) {
1653 bh_unlock_sock(sk);
1654 NET_INC_STATS_BH(net, LINUX_MIB_TCPBACKLOGDROP);
1655 goto discard_and_relse;
1656 }
1657 bh_unlock_sock(sk);
1658
1659 put_and_return:
1660 sock_put(sk);
1661
1662 return ret;
1663
1664 no_tcp_socket:
1665 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
1666 goto discard_it;
1667
1668 if (tcp_checksum_complete(skb)) {
1669 csum_error:
1670 TCP_INC_STATS_BH(net, TCP_MIB_CSUMERRORS);
1671 bad_packet:
1672 TCP_INC_STATS_BH(net, TCP_MIB_INERRS);
1673 } else {
1674 tcp_v4_send_reset(NULL, skb);
1675 }
1676
1677 discard_it:
1678 /* Discard frame. */
1679 kfree_skb(skb);
1680 return 0;
1681
1682 discard_and_relse:
1683 sock_put(sk);
1684 goto discard_it;
1685
1686 do_time_wait:
1687 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) {
1688 inet_twsk_put(inet_twsk(sk));
1689 goto discard_it;
1690 }
1691
1692 if (tcp_checksum_complete(skb)) {
1693 inet_twsk_put(inet_twsk(sk));
1694 goto csum_error;
1695 }
1696 switch (tcp_timewait_state_process(inet_twsk(sk), skb, th)) {
1697 case TCP_TW_SYN: {
1698 struct sock *sk2 = inet_lookup_listener(dev_net(skb->dev),
1699 &tcp_hashinfo,
1700 iph->saddr, th->source,
1701 iph->daddr, th->dest,
1702 inet_iif(skb));
1703 if (sk2) {
1704 inet_twsk_deschedule_put(inet_twsk(sk));
1705 sk = sk2;
1706 goto process;
1707 }
1708 /* Fall through to ACK */
1709 }
1710 case TCP_TW_ACK:
1711 tcp_v4_timewait_ack(sk, skb);
1712 break;
1713 case TCP_TW_RST:
1714 tcp_v4_send_reset(sk, skb);
1715 inet_twsk_deschedule_put(inet_twsk(sk));
1716 goto discard_it;
1717 case TCP_TW_SUCCESS:;
1718 }
1719 goto discard_it;
1720 }
1721
1722 static struct timewait_sock_ops tcp_timewait_sock_ops = {
1723 .twsk_obj_size = sizeof(struct tcp_timewait_sock),
1724 .twsk_unique = tcp_twsk_unique,
1725 .twsk_destructor= tcp_twsk_destructor,
1726 };
1727
1728 void inet_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb)
1729 {
1730 struct dst_entry *dst = skb_dst(skb);
1731
1732 if (dst && dst_hold_safe(dst)) {
1733 sk->sk_rx_dst = dst;
1734 inet_sk(sk)->rx_dst_ifindex = skb->skb_iif;
1735 }
1736 }
1737 EXPORT_SYMBOL(inet_sk_rx_dst_set);
1738
1739 const struct inet_connection_sock_af_ops ipv4_specific = {
1740 .queue_xmit = ip_queue_xmit,
1741 .send_check = tcp_v4_send_check,
1742 .rebuild_header = inet_sk_rebuild_header,
1743 .sk_rx_dst_set = inet_sk_rx_dst_set,
1744 .conn_request = tcp_v4_conn_request,
1745 .syn_recv_sock = tcp_v4_syn_recv_sock,
1746 .net_header_len = sizeof(struct iphdr),
1747 .setsockopt = ip_setsockopt,
1748 .getsockopt = ip_getsockopt,
1749 .addr2sockaddr = inet_csk_addr2sockaddr,
1750 .sockaddr_len = sizeof(struct sockaddr_in),
1751 .bind_conflict = inet_csk_bind_conflict,
1752 #ifdef CONFIG_COMPAT
1753 .compat_setsockopt = compat_ip_setsockopt,
1754 .compat_getsockopt = compat_ip_getsockopt,
1755 #endif
1756 .mtu_reduced = tcp_v4_mtu_reduced,
1757 };
1758 EXPORT_SYMBOL(ipv4_specific);
1759
1760 #ifdef CONFIG_TCP_MD5SIG
1761 static const struct tcp_sock_af_ops tcp_sock_ipv4_specific = {
1762 .md5_lookup = tcp_v4_md5_lookup,
1763 .calc_md5_hash = tcp_v4_md5_hash_skb,
1764 .md5_parse = tcp_v4_parse_md5_keys,
1765 };
1766 #endif
1767
1768 /* NOTE: A lot of things set to zero explicitly by call to
1769 * sk_alloc() so need not be done here.
1770 */
1771 static int tcp_v4_init_sock(struct sock *sk)
1772 {
1773 struct inet_connection_sock *icsk = inet_csk(sk);
1774
1775 tcp_init_sock(sk);
1776
1777 icsk->icsk_af_ops = &ipv4_specific;
1778
1779 #ifdef CONFIG_TCP_MD5SIG
1780 tcp_sk(sk)->af_specific = &tcp_sock_ipv4_specific;
1781 #endif
1782
1783 return 0;
1784 }
1785
1786 void tcp_v4_destroy_sock(struct sock *sk)
1787 {
1788 struct tcp_sock *tp = tcp_sk(sk);
1789
1790 tcp_clear_xmit_timers(sk);
1791
1792 tcp_cleanup_congestion_control(sk);
1793
1794 /* Cleanup up the write buffer. */
1795 tcp_write_queue_purge(sk);
1796
1797 /* Cleans up our, hopefully empty, out_of_order_queue. */
1798 __skb_queue_purge(&tp->out_of_order_queue);
1799
1800 #ifdef CONFIG_TCP_MD5SIG
1801 /* Clean up the MD5 key list, if any */
1802 if (tp->md5sig_info) {
1803 tcp_clear_md5_list(sk);
1804 kfree_rcu(tp->md5sig_info, rcu);
1805 tp->md5sig_info = NULL;
1806 }
1807 #endif
1808
1809 /* Clean prequeue, it must be empty really */
1810 __skb_queue_purge(&tp->ucopy.prequeue);
1811
1812 /* Clean up a referenced TCP bind bucket. */
1813 if (inet_csk(sk)->icsk_bind_hash)
1814 inet_put_port(sk);
1815
1816 BUG_ON(tp->fastopen_rsk);
1817
1818 /* If socket is aborted during connect operation */
1819 tcp_free_fastopen_req(tp);
1820 tcp_saved_syn_free(tp);
1821
1822 sk_sockets_allocated_dec(sk);
1823
1824 if (mem_cgroup_sockets_enabled && sk->sk_memcg)
1825 sock_release_memcg(sk);
1826 }
1827 EXPORT_SYMBOL(tcp_v4_destroy_sock);
1828
1829 #ifdef CONFIG_PROC_FS
1830 /* Proc filesystem TCP sock list dumping. */
1831
1832 /*
1833 * Get next listener socket follow cur. If cur is NULL, get first socket
1834 * starting from bucket given in st->bucket; when st->bucket is zero the
1835 * very first socket in the hash table is returned.
1836 */
1837 static void *listening_get_next(struct seq_file *seq, void *cur)
1838 {
1839 struct inet_connection_sock *icsk;
1840 struct hlist_nulls_node *node;
1841 struct sock *sk = cur;
1842 struct inet_listen_hashbucket *ilb;
1843 struct tcp_iter_state *st = seq->private;
1844 struct net *net = seq_file_net(seq);
1845
1846 if (!sk) {
1847 ilb = &tcp_hashinfo.listening_hash[st->bucket];
1848 spin_lock_bh(&ilb->lock);
1849 sk = sk_nulls_head(&ilb->head);
1850 st->offset = 0;
1851 goto get_sk;
1852 }
1853 ilb = &tcp_hashinfo.listening_hash[st->bucket];
1854 ++st->num;
1855 ++st->offset;
1856
1857 sk = sk_nulls_next(sk);
1858 get_sk:
1859 sk_nulls_for_each_from(sk, node) {
1860 if (!net_eq(sock_net(sk), net))
1861 continue;
1862 if (sk->sk_family == st->family) {
1863 cur = sk;
1864 goto out;
1865 }
1866 icsk = inet_csk(sk);
1867 }
1868 spin_unlock_bh(&ilb->lock);
1869 st->offset = 0;
1870 if (++st->bucket < INET_LHTABLE_SIZE) {
1871 ilb = &tcp_hashinfo.listening_hash[st->bucket];
1872 spin_lock_bh(&ilb->lock);
1873 sk = sk_nulls_head(&ilb->head);
1874 goto get_sk;
1875 }
1876 cur = NULL;
1877 out:
1878 return cur;
1879 }
1880
1881 static void *listening_get_idx(struct seq_file *seq, loff_t *pos)
1882 {
1883 struct tcp_iter_state *st = seq->private;
1884 void *rc;
1885
1886 st->bucket = 0;
1887 st->offset = 0;
1888 rc = listening_get_next(seq, NULL);
1889
1890 while (rc && *pos) {
1891 rc = listening_get_next(seq, rc);
1892 --*pos;
1893 }
1894 return rc;
1895 }
1896
1897 static inline bool empty_bucket(const struct tcp_iter_state *st)
1898 {
1899 return hlist_nulls_empty(&tcp_hashinfo.ehash[st->bucket].chain);
1900 }
1901
1902 /*
1903 * Get first established socket starting from bucket given in st->bucket.
1904 * If st->bucket is zero, the very first socket in the hash is returned.
1905 */
1906 static void *established_get_first(struct seq_file *seq)
1907 {
1908 struct tcp_iter_state *st = seq->private;
1909 struct net *net = seq_file_net(seq);
1910 void *rc = NULL;
1911
1912 st->offset = 0;
1913 for (; st->bucket <= tcp_hashinfo.ehash_mask; ++st->bucket) {
1914 struct sock *sk;
1915 struct hlist_nulls_node *node;
1916 spinlock_t *lock = inet_ehash_lockp(&tcp_hashinfo, st->bucket);
1917
1918 /* Lockless fast path for the common case of empty buckets */
1919 if (empty_bucket(st))
1920 continue;
1921
1922 spin_lock_bh(lock);
1923 sk_nulls_for_each(sk, node, &tcp_hashinfo.ehash[st->bucket].chain) {
1924 if (sk->sk_family != st->family ||
1925 !net_eq(sock_net(sk), net)) {
1926 continue;
1927 }
1928 rc = sk;
1929 goto out;
1930 }
1931 spin_unlock_bh(lock);
1932 }
1933 out:
1934 return rc;
1935 }
1936
1937 static void *established_get_next(struct seq_file *seq, void *cur)
1938 {
1939 struct sock *sk = cur;
1940 struct hlist_nulls_node *node;
1941 struct tcp_iter_state *st = seq->private;
1942 struct net *net = seq_file_net(seq);
1943
1944 ++st->num;
1945 ++st->offset;
1946
1947 sk = sk_nulls_next(sk);
1948
1949 sk_nulls_for_each_from(sk, node) {
1950 if (sk->sk_family == st->family && net_eq(sock_net(sk), net))
1951 return sk;
1952 }
1953
1954 spin_unlock_bh(inet_ehash_lockp(&tcp_hashinfo, st->bucket));
1955 ++st->bucket;
1956 return established_get_first(seq);
1957 }
1958
1959 static void *established_get_idx(struct seq_file *seq, loff_t pos)
1960 {
1961 struct tcp_iter_state *st = seq->private;
1962 void *rc;
1963
1964 st->bucket = 0;
1965 rc = established_get_first(seq);
1966
1967 while (rc && pos) {
1968 rc = established_get_next(seq, rc);
1969 --pos;
1970 }
1971 return rc;
1972 }
1973
1974 static void *tcp_get_idx(struct seq_file *seq, loff_t pos)
1975 {
1976 void *rc;
1977 struct tcp_iter_state *st = seq->private;
1978
1979 st->state = TCP_SEQ_STATE_LISTENING;
1980 rc = listening_get_idx(seq, &pos);
1981
1982 if (!rc) {
1983 st->state = TCP_SEQ_STATE_ESTABLISHED;
1984 rc = established_get_idx(seq, pos);
1985 }
1986
1987 return rc;
1988 }
1989
1990 static void *tcp_seek_last_pos(struct seq_file *seq)
1991 {
1992 struct tcp_iter_state *st = seq->private;
1993 int offset = st->offset;
1994 int orig_num = st->num;
1995 void *rc = NULL;
1996
1997 switch (st->state) {
1998 case TCP_SEQ_STATE_LISTENING:
1999 if (st->bucket >= INET_LHTABLE_SIZE)
2000 break;
2001 st->state = TCP_SEQ_STATE_LISTENING;
2002 rc = listening_get_next(seq, NULL);
2003 while (offset-- && rc)
2004 rc = listening_get_next(seq, rc);
2005 if (rc)
2006 break;
2007 st->bucket = 0;
2008 st->state = TCP_SEQ_STATE_ESTABLISHED;
2009 /* Fallthrough */
2010 case TCP_SEQ_STATE_ESTABLISHED:
2011 if (st->bucket > tcp_hashinfo.ehash_mask)
2012 break;
2013 rc = established_get_first(seq);
2014 while (offset-- && rc)
2015 rc = established_get_next(seq, rc);
2016 }
2017
2018 st->num = orig_num;
2019
2020 return rc;
2021 }
2022
2023 static void *tcp_seq_start(struct seq_file *seq, loff_t *pos)
2024 {
2025 struct tcp_iter_state *st = seq->private;
2026 void *rc;
2027
2028 if (*pos && *pos == st->last_pos) {
2029 rc = tcp_seek_last_pos(seq);
2030 if (rc)
2031 goto out;
2032 }
2033
2034 st->state = TCP_SEQ_STATE_LISTENING;
2035 st->num = 0;
2036 st->bucket = 0;
2037 st->offset = 0;
2038 rc = *pos ? tcp_get_idx(seq, *pos - 1) : SEQ_START_TOKEN;
2039
2040 out:
2041 st->last_pos = *pos;
2042 return rc;
2043 }
2044
2045 static void *tcp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2046 {
2047 struct tcp_iter_state *st = seq->private;
2048 void *rc = NULL;
2049
2050 if (v == SEQ_START_TOKEN) {
2051 rc = tcp_get_idx(seq, 0);
2052 goto out;
2053 }
2054
2055 switch (st->state) {
2056 case TCP_SEQ_STATE_LISTENING:
2057 rc = listening_get_next(seq, v);
2058 if (!rc) {
2059 st->state = TCP_SEQ_STATE_ESTABLISHED;
2060 st->bucket = 0;
2061 st->offset = 0;
2062 rc = established_get_first(seq);
2063 }
2064 break;
2065 case TCP_SEQ_STATE_ESTABLISHED:
2066 rc = established_get_next(seq, v);
2067 break;
2068 }
2069 out:
2070 ++*pos;
2071 st->last_pos = *pos;
2072 return rc;
2073 }
2074
2075 static void tcp_seq_stop(struct seq_file *seq, void *v)
2076 {
2077 struct tcp_iter_state *st = seq->private;
2078
2079 switch (st->state) {
2080 case TCP_SEQ_STATE_LISTENING:
2081 if (v != SEQ_START_TOKEN)
2082 spin_unlock_bh(&tcp_hashinfo.listening_hash[st->bucket].lock);
2083 break;
2084 case TCP_SEQ_STATE_ESTABLISHED:
2085 if (v)
2086 spin_unlock_bh(inet_ehash_lockp(&tcp_hashinfo, st->bucket));
2087 break;
2088 }
2089 }
2090
2091 int tcp_seq_open(struct inode *inode, struct file *file)
2092 {
2093 struct tcp_seq_afinfo *afinfo = PDE_DATA(inode);
2094 struct tcp_iter_state *s;
2095 int err;
2096
2097 err = seq_open_net(inode, file, &afinfo->seq_ops,
2098 sizeof(struct tcp_iter_state));
2099 if (err < 0)
2100 return err;
2101
2102 s = ((struct seq_file *)file->private_data)->private;
2103 s->family = afinfo->family;
2104 s->last_pos = 0;
2105 return 0;
2106 }
2107 EXPORT_SYMBOL(tcp_seq_open);
2108
2109 int tcp_proc_register(struct net *net, struct tcp_seq_afinfo *afinfo)
2110 {
2111 int rc = 0;
2112 struct proc_dir_entry *p;
2113
2114 afinfo->seq_ops.start = tcp_seq_start;
2115 afinfo->seq_ops.next = tcp_seq_next;
2116 afinfo->seq_ops.stop = tcp_seq_stop;
2117
2118 p = proc_create_data(afinfo->name, S_IRUGO, net->proc_net,
2119 afinfo->seq_fops, afinfo);
2120 if (!p)
2121 rc = -ENOMEM;
2122 return rc;
2123 }
2124 EXPORT_SYMBOL(tcp_proc_register);
2125
2126 void tcp_proc_unregister(struct net *net, struct tcp_seq_afinfo *afinfo)
2127 {
2128 remove_proc_entry(afinfo->name, net->proc_net);
2129 }
2130 EXPORT_SYMBOL(tcp_proc_unregister);
2131
2132 static void get_openreq4(const struct request_sock *req,
2133 struct seq_file *f, int i)
2134 {
2135 const struct inet_request_sock *ireq = inet_rsk(req);
2136 long delta = req->rsk_timer.expires - jiffies;
2137
2138 seq_printf(f, "%4d: %08X:%04X %08X:%04X"
2139 " %02X %08X:%08X %02X:%08lX %08X %5u %8d %u %d %pK",
2140 i,
2141 ireq->ir_loc_addr,
2142 ireq->ir_num,
2143 ireq->ir_rmt_addr,
2144 ntohs(ireq->ir_rmt_port),
2145 TCP_SYN_RECV,
2146 0, 0, /* could print option size, but that is af dependent. */
2147 1, /* timers active (only the expire timer) */
2148 jiffies_delta_to_clock_t(delta),
2149 req->num_timeout,
2150 from_kuid_munged(seq_user_ns(f),
2151 sock_i_uid(req->rsk_listener)),
2152 0, /* non standard timer */
2153 0, /* open_requests have no inode */
2154 0,
2155 req);
2156 }
2157
2158 static void get_tcp4_sock(struct sock *sk, struct seq_file *f, int i)
2159 {
2160 int timer_active;
2161 unsigned long timer_expires;
2162 const struct tcp_sock *tp = tcp_sk(sk);
2163 const struct inet_connection_sock *icsk = inet_csk(sk);
2164 const struct inet_sock *inet = inet_sk(sk);
2165 const struct fastopen_queue *fastopenq = &icsk->icsk_accept_queue.fastopenq;
2166 __be32 dest = inet->inet_daddr;
2167 __be32 src = inet->inet_rcv_saddr;
2168 __u16 destp = ntohs(inet->inet_dport);
2169 __u16 srcp = ntohs(inet->inet_sport);
2170 int rx_queue;
2171 int state;
2172
2173 if (icsk->icsk_pending == ICSK_TIME_RETRANS ||
2174 icsk->icsk_pending == ICSK_TIME_EARLY_RETRANS ||
2175 icsk->icsk_pending == ICSK_TIME_LOSS_PROBE) {
2176 timer_active = 1;
2177 timer_expires = icsk->icsk_timeout;
2178 } else if (icsk->icsk_pending == ICSK_TIME_PROBE0) {
2179 timer_active = 4;
2180 timer_expires = icsk->icsk_timeout;
2181 } else if (timer_pending(&sk->sk_timer)) {
2182 timer_active = 2;
2183 timer_expires = sk->sk_timer.expires;
2184 } else {
2185 timer_active = 0;
2186 timer_expires = jiffies;
2187 }
2188
2189 state = sk_state_load(sk);
2190 if (state == TCP_LISTEN)
2191 rx_queue = sk->sk_ack_backlog;
2192 else
2193 /* Because we don't lock the socket,
2194 * we might find a transient negative value.
2195 */
2196 rx_queue = max_t(int, tp->rcv_nxt - tp->copied_seq, 0);
2197
2198 seq_printf(f, "%4d: %08X:%04X %08X:%04X %02X %08X:%08X %02X:%08lX "
2199 "%08X %5u %8d %lu %d %pK %lu %lu %u %u %d",
2200 i, src, srcp, dest, destp, state,
2201 tp->write_seq - tp->snd_una,
2202 rx_queue,
2203 timer_active,
2204 jiffies_delta_to_clock_t(timer_expires - jiffies),
2205 icsk->icsk_retransmits,
2206 from_kuid_munged(seq_user_ns(f), sock_i_uid(sk)),
2207 icsk->icsk_probes_out,
2208 sock_i_ino(sk),
2209 atomic_read(&sk->sk_refcnt), sk,
2210 jiffies_to_clock_t(icsk->icsk_rto),
2211 jiffies_to_clock_t(icsk->icsk_ack.ato),
2212 (icsk->icsk_ack.quick << 1) | icsk->icsk_ack.pingpong,
2213 tp->snd_cwnd,
2214 state == TCP_LISTEN ?
2215 fastopenq->max_qlen :
2216 (tcp_in_initial_slowstart(tp) ? -1 : tp->snd_ssthresh));
2217 }
2218
2219 static void get_timewait4_sock(const struct inet_timewait_sock *tw,
2220 struct seq_file *f, int i)
2221 {
2222 long delta = tw->tw_timer.expires - jiffies;
2223 __be32 dest, src;
2224 __u16 destp, srcp;
2225
2226 dest = tw->tw_daddr;
2227 src = tw->tw_rcv_saddr;
2228 destp = ntohs(tw->tw_dport);
2229 srcp = ntohs(tw->tw_sport);
2230
2231 seq_printf(f, "%4d: %08X:%04X %08X:%04X"
2232 " %02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %pK",
2233 i, src, srcp, dest, destp, tw->tw_substate, 0, 0,
2234 3, jiffies_delta_to_clock_t(delta), 0, 0, 0, 0,
2235 atomic_read(&tw->tw_refcnt), tw);
2236 }
2237
2238 #define TMPSZ 150
2239
2240 static int tcp4_seq_show(struct seq_file *seq, void *v)
2241 {
2242 struct tcp_iter_state *st;
2243 struct sock *sk = v;
2244
2245 seq_setwidth(seq, TMPSZ - 1);
2246 if (v == SEQ_START_TOKEN) {
2247 seq_puts(seq, " sl local_address rem_address st tx_queue "
2248 "rx_queue tr tm->when retrnsmt uid timeout "
2249 "inode");
2250 goto out;
2251 }
2252 st = seq->private;
2253
2254 if (sk->sk_state == TCP_TIME_WAIT)
2255 get_timewait4_sock(v, seq, st->num);
2256 else if (sk->sk_state == TCP_NEW_SYN_RECV)
2257 get_openreq4(v, seq, st->num);
2258 else
2259 get_tcp4_sock(v, seq, st->num);
2260 out:
2261 seq_pad(seq, '\n');
2262 return 0;
2263 }
2264
2265 static const struct file_operations tcp_afinfo_seq_fops = {
2266 .owner = THIS_MODULE,
2267 .open = tcp_seq_open,
2268 .read = seq_read,
2269 .llseek = seq_lseek,
2270 .release = seq_release_net
2271 };
2272
2273 static struct tcp_seq_afinfo tcp4_seq_afinfo = {
2274 .name = "tcp",
2275 .family = AF_INET,
2276 .seq_fops = &tcp_afinfo_seq_fops,
2277 .seq_ops = {
2278 .show = tcp4_seq_show,
2279 },
2280 };
2281
2282 static int __net_init tcp4_proc_init_net(struct net *net)
2283 {
2284 return tcp_proc_register(net, &tcp4_seq_afinfo);
2285 }
2286
2287 static void __net_exit tcp4_proc_exit_net(struct net *net)
2288 {
2289 tcp_proc_unregister(net, &tcp4_seq_afinfo);
2290 }
2291
2292 static struct pernet_operations tcp4_net_ops = {
2293 .init = tcp4_proc_init_net,
2294 .exit = tcp4_proc_exit_net,
2295 };
2296
2297 int __init tcp4_proc_init(void)
2298 {
2299 return register_pernet_subsys(&tcp4_net_ops);
2300 }
2301
2302 void tcp4_proc_exit(void)
2303 {
2304 unregister_pernet_subsys(&tcp4_net_ops);
2305 }
2306 #endif /* CONFIG_PROC_FS */
2307
2308 struct proto tcp_prot = {
2309 .name = "TCP",
2310 .owner = THIS_MODULE,
2311 .close = tcp_close,
2312 .connect = tcp_v4_connect,
2313 .disconnect = tcp_disconnect,
2314 .accept = inet_csk_accept,
2315 .ioctl = tcp_ioctl,
2316 .init = tcp_v4_init_sock,
2317 .destroy = tcp_v4_destroy_sock,
2318 .shutdown = tcp_shutdown,
2319 .setsockopt = tcp_setsockopt,
2320 .getsockopt = tcp_getsockopt,
2321 .recvmsg = tcp_recvmsg,
2322 .sendmsg = tcp_sendmsg,
2323 .sendpage = tcp_sendpage,
2324 .backlog_rcv = tcp_v4_do_rcv,
2325 .release_cb = tcp_release_cb,
2326 .hash = inet_hash,
2327 .unhash = inet_unhash,
2328 .get_port = inet_csk_get_port,
2329 .enter_memory_pressure = tcp_enter_memory_pressure,
2330 .stream_memory_free = tcp_stream_memory_free,
2331 .sockets_allocated = &tcp_sockets_allocated,
2332 .orphan_count = &tcp_orphan_count,
2333 .memory_allocated = &tcp_memory_allocated,
2334 .memory_pressure = &tcp_memory_pressure,
2335 .sysctl_mem = sysctl_tcp_mem,
2336 .sysctl_wmem = sysctl_tcp_wmem,
2337 .sysctl_rmem = sysctl_tcp_rmem,
2338 .max_header = MAX_TCP_HEADER,
2339 .obj_size = sizeof(struct tcp_sock),
2340 .slab_flags = SLAB_DESTROY_BY_RCU,
2341 .twsk_prot = &tcp_timewait_sock_ops,
2342 .rsk_prot = &tcp_request_sock_ops,
2343 .h.hashinfo = &tcp_hashinfo,
2344 .no_autobind = true,
2345 #ifdef CONFIG_COMPAT
2346 .compat_setsockopt = compat_tcp_setsockopt,
2347 .compat_getsockopt = compat_tcp_getsockopt,
2348 #endif
2349 .diag_destroy = tcp_abort,
2350 };
2351 EXPORT_SYMBOL(tcp_prot);
2352
2353 static void __net_exit tcp_sk_exit(struct net *net)
2354 {
2355 int cpu;
2356
2357 for_each_possible_cpu(cpu)
2358 inet_ctl_sock_destroy(*per_cpu_ptr(net->ipv4.tcp_sk, cpu));
2359 free_percpu(net->ipv4.tcp_sk);
2360 }
2361
2362 static int __net_init tcp_sk_init(struct net *net)
2363 {
2364 int res, cpu;
2365
2366 net->ipv4.tcp_sk = alloc_percpu(struct sock *);
2367 if (!net->ipv4.tcp_sk)
2368 return -ENOMEM;
2369
2370 for_each_possible_cpu(cpu) {
2371 struct sock *sk;
2372
2373 res = inet_ctl_sock_create(&sk, PF_INET, SOCK_RAW,
2374 IPPROTO_TCP, net);
2375 if (res)
2376 goto fail;
2377 *per_cpu_ptr(net->ipv4.tcp_sk, cpu) = sk;
2378 }
2379
2380 net->ipv4.sysctl_tcp_ecn = 2;
2381 net->ipv4.sysctl_tcp_ecn_fallback = 1;
2382
2383 net->ipv4.sysctl_tcp_base_mss = TCP_BASE_MSS;
2384 net->ipv4.sysctl_tcp_probe_threshold = TCP_PROBE_THRESHOLD;
2385 net->ipv4.sysctl_tcp_probe_interval = TCP_PROBE_INTERVAL;
2386
2387 net->ipv4.sysctl_tcp_keepalive_time = TCP_KEEPALIVE_TIME;
2388 net->ipv4.sysctl_tcp_keepalive_probes = TCP_KEEPALIVE_PROBES;
2389 net->ipv4.sysctl_tcp_keepalive_intvl = TCP_KEEPALIVE_INTVL;
2390
2391 net->ipv4.sysctl_tcp_syn_retries = TCP_SYN_RETRIES;
2392 net->ipv4.sysctl_tcp_synack_retries = TCP_SYNACK_RETRIES;
2393
2394 return 0;
2395 fail:
2396 tcp_sk_exit(net);
2397
2398 return res;
2399 }
2400
2401 static void __net_exit tcp_sk_exit_batch(struct list_head *net_exit_list)
2402 {
2403 inet_twsk_purge(&tcp_hashinfo, &tcp_death_row, AF_INET);
2404 }
2405
2406 static struct pernet_operations __net_initdata tcp_sk_ops = {
2407 .init = tcp_sk_init,
2408 .exit = tcp_sk_exit,
2409 .exit_batch = tcp_sk_exit_batch,
2410 };
2411
2412 void __init tcp_v4_init(void)
2413 {
2414 inet_hashinfo_init(&tcp_hashinfo);
2415 if (register_pernet_subsys(&tcp_sk_ops))
2416 panic("Failed to create the TCP control socket.\n");
2417 }
Linux kernel - Deliverables
This page took 0.097669 seconds and 5 git commands to generate.