projects / deliverable / linux.git / blob
? search:


69f9cf68474437511306d718ed870908b122bde8
[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/tcp_memcontrol.h>
77 #include <net/busy_poll.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 #ifdef CONFIG_TCP_MD5SIG
93 static int tcp_v4_md5_hash_hdr(char *md5_hash, const struct tcp_md5sig_key *key,
94 __be32 daddr, __be32 saddr, const struct tcphdr *th);
95 #endif
96
97 struct inet_hashinfo tcp_hashinfo;
98 EXPORT_SYMBOL(tcp_hashinfo);
99
100 static __u32 tcp_v4_init_sequence(const struct sk_buff *skb)
101 {
102 return secure_tcp_sequence_number(ip_hdr(skb)->daddr,
103 ip_hdr(skb)->saddr,
104 tcp_hdr(skb)->dest,
105 tcp_hdr(skb)->source);
106 }
107
108 int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp)
109 {
110 const struct tcp_timewait_sock *tcptw = tcp_twsk(sktw);
111 struct tcp_sock *tp = tcp_sk(sk);
112
113 /* With PAWS, it is safe from the viewpoint
114 of data integrity. Even without PAWS it is safe provided sequence
115 spaces do not overlap i.e. at data rates <= 80Mbit/sec.
116
117 Actually, the idea is close to VJ's one, only timestamp cache is
118 held not per host, but per port pair and TW bucket is used as state
119 holder.
120
121 If TW bucket has been already destroyed we fall back to VJ's scheme
122 and use initial timestamp retrieved from peer table.
123 */
124 if (tcptw->tw_ts_recent_stamp &&
125 (twp == NULL || (sysctl_tcp_tw_reuse &&
126 get_seconds() - tcptw->tw_ts_recent_stamp > 1))) {
127 tp->write_seq = tcptw->tw_snd_nxt + 65535 + 2;
128 if (tp->write_seq == 0)
129 tp->write_seq = 1;
130 tp->rx_opt.ts_recent = tcptw->tw_ts_recent;
131 tp->rx_opt.ts_recent_stamp = tcptw->tw_ts_recent_stamp;
132 sock_hold(sktw);
133 return 1;
134 }
135
136 return 0;
137 }
138 EXPORT_SYMBOL_GPL(tcp_twsk_unique);
139
140 /* This will initiate an outgoing connection. */
141 int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len)
142 {
143 struct sockaddr_in *usin = (struct sockaddr_in *)uaddr;
144 struct inet_sock *inet = inet_sk(sk);
145 struct tcp_sock *tp = tcp_sk(sk);
146 __be16 orig_sport, orig_dport;
147 __be32 daddr, nexthop;
148 struct flowi4 *fl4;
149 struct rtable *rt;
150 int err;
151 struct ip_options_rcu *inet_opt;
152
153 if (addr_len < sizeof(struct sockaddr_in))
154 return -EINVAL;
155
156 if (usin->sin_family != AF_INET)
157 return -EAFNOSUPPORT;
158
159 nexthop = daddr = usin->sin_addr.s_addr;
160 inet_opt = rcu_dereference_protected(inet->inet_opt,
161 sock_owned_by_user(sk));
162 if (inet_opt && inet_opt->opt.srr) {
163 if (!daddr)
164 return -EINVAL;
165 nexthop = inet_opt->opt.faddr;
166 }
167
168 orig_sport = inet->inet_sport;
169 orig_dport = usin->sin_port;
170 fl4 = &inet->cork.fl.u.ip4;
171 rt = ip_route_connect(fl4, nexthop, inet->inet_saddr,
172 RT_CONN_FLAGS(sk), sk->sk_bound_dev_if,
173 IPPROTO_TCP,
174 orig_sport, orig_dport, sk);
175 if (IS_ERR(rt)) {
176 err = PTR_ERR(rt);
177 if (err == -ENETUNREACH)
178 IP_INC_STATS(sock_net(sk), IPSTATS_MIB_OUTNOROUTES);
179 return err;
180 }
181
182 if (rt->rt_flags & (RTCF_MULTICAST | RTCF_BROADCAST)) {
183 ip_rt_put(rt);
184 return -ENETUNREACH;
185 }
186
187 if (!inet_opt || !inet_opt->opt.srr)
188 daddr = fl4->daddr;
189
190 if (!inet->inet_saddr)
191 inet->inet_saddr = fl4->saddr;
192 sk_rcv_saddr_set(sk, inet->inet_saddr);
193
194 if (tp->rx_opt.ts_recent_stamp && inet->inet_daddr != daddr) {
195 /* Reset inherited state */
196 tp->rx_opt.ts_recent = 0;
197 tp->rx_opt.ts_recent_stamp = 0;
198 if (likely(!tp->repair))
199 tp->write_seq = 0;
200 }
201
202 if (tcp_death_row.sysctl_tw_recycle &&
203 !tp->rx_opt.ts_recent_stamp && fl4->daddr == daddr)
204 tcp_fetch_timewait_stamp(sk, &rt->dst);
205
206 inet->inet_dport = usin->sin_port;
207 sk_daddr_set(sk, daddr);
208
209 inet_csk(sk)->icsk_ext_hdr_len = 0;
210 if (inet_opt)
211 inet_csk(sk)->icsk_ext_hdr_len = inet_opt->opt.optlen;
212
213 tp->rx_opt.mss_clamp = TCP_MSS_DEFAULT;
214
215 /* Socket identity is still unknown (sport may be zero).
216 * However we set state to SYN-SENT and not releasing socket
217 * lock select source port, enter ourselves into the hash tables and
218 * complete initialization after this.
219 */
220 tcp_set_state(sk, TCP_SYN_SENT);
221 err = inet_hash_connect(&tcp_death_row, sk);
222 if (err)
223 goto failure;
224
225 inet_set_txhash(sk);
226
227 rt = ip_route_newports(fl4, rt, orig_sport, orig_dport,
228 inet->inet_sport, inet->inet_dport, sk);
229 if (IS_ERR(rt)) {
230 err = PTR_ERR(rt);
231 rt = NULL;
232 goto failure;
233 }
234 /* OK, now commit destination to socket. */
235 sk->sk_gso_type = SKB_GSO_TCPV4;
236 sk_setup_caps(sk, &rt->dst);
237
238 if (!tp->write_seq && likely(!tp->repair))
239 tp->write_seq = secure_tcp_sequence_number(inet->inet_saddr,
240 inet->inet_daddr,
241 inet->inet_sport,
242 usin->sin_port);
243
244 inet->inet_id = tp->write_seq ^ jiffies;
245
246 err = tcp_connect(sk);
247
248 rt = NULL;
249 if (err)
250 goto failure;
251
252 return 0;
253
254 failure:
255 /*
256 * This unhashes the socket and releases the local port,
257 * if necessary.
258 */
259 tcp_set_state(sk, TCP_CLOSE);
260 ip_rt_put(rt);
261 sk->sk_route_caps = 0;
262 inet->inet_dport = 0;
263 return err;
264 }
265 EXPORT_SYMBOL(tcp_v4_connect);
266
267 /*
268 * This routine reacts to ICMP_FRAG_NEEDED mtu indications as defined in RFC1191.
269 * It can be called through tcp_release_cb() if socket was owned by user
270 * at the time tcp_v4_err() was called to handle ICMP message.
271 */
272 void tcp_v4_mtu_reduced(struct sock *sk)
273 {
274 struct dst_entry *dst;
275 struct inet_sock *inet = inet_sk(sk);
276 u32 mtu = tcp_sk(sk)->mtu_info;
277
278 dst = inet_csk_update_pmtu(sk, mtu);
279 if (!dst)
280 return;
281
282 /* Something is about to be wrong... Remember soft error
283 * for the case, if this connection will not able to recover.
284 */
285 if (mtu < dst_mtu(dst) && ip_dont_fragment(sk, dst))
286 sk->sk_err_soft = EMSGSIZE;
287
288 mtu = dst_mtu(dst);
289
290 if (inet->pmtudisc != IP_PMTUDISC_DONT &&
291 ip_sk_accept_pmtu(sk) &&
292 inet_csk(sk)->icsk_pmtu_cookie > mtu) {
293 tcp_sync_mss(sk, mtu);
294
295 /* Resend the TCP packet because it's
296 * clear that the old packet has been
297 * dropped. This is the new "fast" path mtu
298 * discovery.
299 */
300 tcp_simple_retransmit(sk);
301 } /* else let the usual retransmit timer handle it */
302 }
303 EXPORT_SYMBOL(tcp_v4_mtu_reduced);
304
305 static void do_redirect(struct sk_buff *skb, struct sock *sk)
306 {
307 struct dst_entry *dst = __sk_dst_check(sk, 0);
308
309 if (dst)
310 dst->ops->redirect(dst, sk, skb);
311 }
312
313
314 /* handle ICMP messages on TCP_NEW_SYN_RECV request sockets */
315 void tcp_req_err(struct sock *sk, u32 seq)
316 {
317 struct request_sock *req = inet_reqsk(sk);
318 struct net *net = sock_net(sk);
319
320 /* ICMPs are not backlogged, hence we cannot get
321 * an established socket here.
322 */
323 WARN_ON(req->sk);
324
325 if (seq != tcp_rsk(req)->snt_isn) {
326 NET_INC_STATS_BH(net, LINUX_MIB_OUTOFWINDOWICMPS);
327 reqsk_put(req);
328 } else {
329 /*
330 * Still in SYN_RECV, just remove it silently.
331 * There is no good way to pass the error to the newly
332 * created socket, and POSIX does not want network
333 * errors returned from accept().
334 */
335 NET_INC_STATS_BH(net, LINUX_MIB_LISTENDROPS);
336 inet_csk_reqsk_queue_drop(req->rsk_listener, req);
337 }
338 }
339 EXPORT_SYMBOL(tcp_req_err);
340
341 /*
342 * This routine is called by the ICMP module when it gets some
343 * sort of error condition. If err < 0 then the socket should
344 * be closed and the error returned to the user. If err > 0
345 * it's just the icmp type << 8 | icmp code. After adjustment
346 * header points to the first 8 bytes of the tcp header. We need
347 * to find the appropriate port.
348 *
349 * The locking strategy used here is very "optimistic". When
350 * someone else accesses the socket the ICMP is just dropped
351 * and for some paths there is no check at all.
352 * A more general error queue to queue errors for later handling
353 * is probably better.
354 *
355 */
356
357 void tcp_v4_err(struct sk_buff *icmp_skb, u32 info)
358 {
359 const struct iphdr *iph = (const struct iphdr *)icmp_skb->data;
360 struct tcphdr *th = (struct tcphdr *)(icmp_skb->data + (iph->ihl << 2));
361 struct inet_connection_sock *icsk;
362 struct tcp_sock *tp;
363 struct inet_sock *inet;
364 const int type = icmp_hdr(icmp_skb)->type;
365 const int code = icmp_hdr(icmp_skb)->code;
366 struct sock *sk;
367 struct sk_buff *skb;
368 struct request_sock *fastopen;
369 __u32 seq, snd_una;
370 __u32 remaining;
371 int err;
372 struct net *net = dev_net(icmp_skb->dev);
373
374 sk = __inet_lookup_established(net, &tcp_hashinfo, iph->daddr,
375 th->dest, iph->saddr, ntohs(th->source),
376 inet_iif(icmp_skb));
377 if (!sk) {
378 ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS);
379 return;
380 }
381 if (sk->sk_state == TCP_TIME_WAIT) {
382 inet_twsk_put(inet_twsk(sk));
383 return;
384 }
385 seq = ntohl(th->seq);
386 if (sk->sk_state == TCP_NEW_SYN_RECV)
387 return tcp_req_err(sk, seq);
388
389 bh_lock_sock(sk);
390 /* If too many ICMPs get dropped on busy
391 * servers this needs to be solved differently.
392 * We do take care of PMTU discovery (RFC1191) special case :
393 * we can receive locally generated ICMP messages while socket is held.
394 */
395 if (sock_owned_by_user(sk)) {
396 if (!(type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED))
397 NET_INC_STATS_BH(net, LINUX_MIB_LOCKDROPPEDICMPS);
398 }
399 if (sk->sk_state == TCP_CLOSE)
400 goto out;
401
402 if (unlikely(iph->ttl < inet_sk(sk)->min_ttl)) {
403 NET_INC_STATS_BH(net, LINUX_MIB_TCPMINTTLDROP);
404 goto out;
405 }
406
407 icsk = inet_csk(sk);
408 tp = tcp_sk(sk);
409 /* XXX (TFO) - tp->snd_una should be ISN (tcp_create_openreq_child() */
410 fastopen = tp->fastopen_rsk;
411 snd_una = fastopen ? tcp_rsk(fastopen)->snt_isn : tp->snd_una;
412 if (sk->sk_state != TCP_LISTEN &&
413 !between(seq, snd_una, tp->snd_nxt)) {
414 NET_INC_STATS_BH(net, LINUX_MIB_OUTOFWINDOWICMPS);
415 goto out;
416 }
417
418 switch (type) {
419 case ICMP_REDIRECT:
420 do_redirect(icmp_skb, sk);
421 goto out;
422 case ICMP_SOURCE_QUENCH:
423 /* Just silently ignore these. */
424 goto out;
425 case ICMP_PARAMETERPROB:
426 err = EPROTO;
427 break;
428 case ICMP_DEST_UNREACH:
429 if (code > NR_ICMP_UNREACH)
430 goto out;
431
432 if (code == ICMP_FRAG_NEEDED) { /* PMTU discovery (RFC1191) */
433 /* We are not interested in TCP_LISTEN and open_requests
434 * (SYN-ACKs send out by Linux are always <576bytes so
435 * they should go through unfragmented).
436 */
437 if (sk->sk_state == TCP_LISTEN)
438 goto out;
439
440 tp->mtu_info = info;
441 if (!sock_owned_by_user(sk)) {
442 tcp_v4_mtu_reduced(sk);
443 } else {
444 if (!test_and_set_bit(TCP_MTU_REDUCED_DEFERRED, &tp->tsq_flags))
445 sock_hold(sk);
446 }
447 goto out;
448 }
449
450 err = icmp_err_convert[code].errno;
451 /* check if icmp_skb allows revert of backoff
452 * (see draft-zimmermann-tcp-lcd) */
453 if (code != ICMP_NET_UNREACH && code != ICMP_HOST_UNREACH)
454 break;
455 if (seq != tp->snd_una || !icsk->icsk_retransmits ||
456 !icsk->icsk_backoff || fastopen)
457 break;
458
459 if (sock_owned_by_user(sk))
460 break;
461
462 icsk->icsk_backoff--;
463 icsk->icsk_rto = tp->srtt_us ? __tcp_set_rto(tp) :
464 TCP_TIMEOUT_INIT;
465 icsk->icsk_rto = inet_csk_rto_backoff(icsk, TCP_RTO_MAX);
466
467 skb = tcp_write_queue_head(sk);
468 BUG_ON(!skb);
469
470 remaining = icsk->icsk_rto -
471 min(icsk->icsk_rto,
472 tcp_time_stamp - tcp_skb_timestamp(skb));
473
474 if (remaining) {
475 inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
476 remaining, TCP_RTO_MAX);
477 } else {
478 /* RTO revert clocked out retransmission.
479 * Will retransmit now */
480 tcp_retransmit_timer(sk);
481 }
482
483 break;
484 case ICMP_TIME_EXCEEDED:
485 err = EHOSTUNREACH;
486 break;
487 default:
488 goto out;
489 }
490
491 switch (sk->sk_state) {
492 case TCP_SYN_SENT:
493 case TCP_SYN_RECV:
494 /* Only in fast or simultaneous open. If a fast open socket is
495 * is already accepted it is treated as a connected one below.
496 */
497 if (fastopen && fastopen->sk == NULL)
498 break;
499
500 if (!sock_owned_by_user(sk)) {
501 sk->sk_err = err;
502
503 sk->sk_error_report(sk);
504
505 tcp_done(sk);
506 } else {
507 sk->sk_err_soft = err;
508 }
509 goto out;
510 }
511
512 /* If we've already connected we will keep trying
513 * until we time out, or the user gives up.
514 *
515 * rfc1122 4.2.3.9 allows to consider as hard errors
516 * only PROTO_UNREACH and PORT_UNREACH (well, FRAG_FAILED too,
517 * but it is obsoleted by pmtu discovery).
518 *
519 * Note, that in modern internet, where routing is unreliable
520 * and in each dark corner broken firewalls sit, sending random
521 * errors ordered by their masters even this two messages finally lose
522 * their original sense (even Linux sends invalid PORT_UNREACHs)
523 *
524 * Now we are in compliance with RFCs.
525 * --ANK (980905)
526 */
527
528 inet = inet_sk(sk);
529 if (!sock_owned_by_user(sk) && inet->recverr) {
530 sk->sk_err = err;
531 sk->sk_error_report(sk);
532 } else { /* Only an error on timeout */
533 sk->sk_err_soft = err;
534 }
535
536 out:
537 bh_unlock_sock(sk);
538 sock_put(sk);
539 }
540
541 void __tcp_v4_send_check(struct sk_buff *skb, __be32 saddr, __be32 daddr)
542 {
543 struct tcphdr *th = tcp_hdr(skb);
544
545 if (skb->ip_summed == CHECKSUM_PARTIAL) {
546 th->check = ~tcp_v4_check(skb->len, saddr, daddr, 0);
547 skb->csum_start = skb_transport_header(skb) - skb->head;
548 skb->csum_offset = offsetof(struct tcphdr, check);
549 } else {
550 th->check = tcp_v4_check(skb->len, saddr, daddr,
551 csum_partial(th,
552 th->doff << 2,
553 skb->csum));
554 }
555 }
556
557 /* This routine computes an IPv4 TCP checksum. */
558 void tcp_v4_send_check(struct sock *sk, struct sk_buff *skb)
559 {
560 const struct inet_sock *inet = inet_sk(sk);
561
562 __tcp_v4_send_check(skb, inet->inet_saddr, inet->inet_daddr);
563 }
564 EXPORT_SYMBOL(tcp_v4_send_check);
565
566 /*
567 * This routine will send an RST to the other tcp.
568 *
569 * Someone asks: why I NEVER use socket parameters (TOS, TTL etc.)
570 * for reset.
571 * Answer: if a packet caused RST, it is not for a socket
572 * existing in our system, if it is matched to a socket,
573 * it is just duplicate segment or bug in other side's TCP.
574 * So that we build reply only basing on parameters
575 * arrived with segment.
576 * Exception: precedence violation. We do not implement it in any case.
577 */
578
579 static void tcp_v4_send_reset(struct sock *sk, struct sk_buff *skb)
580 {
581 const struct tcphdr *th = tcp_hdr(skb);
582 struct {
583 struct tcphdr th;
584 #ifdef CONFIG_TCP_MD5SIG
585 __be32 opt[(TCPOLEN_MD5SIG_ALIGNED >> 2)];
586 #endif
587 } rep;
588 struct ip_reply_arg arg;
589 #ifdef CONFIG_TCP_MD5SIG
590 struct tcp_md5sig_key *key;
591 const __u8 *hash_location = NULL;
592 unsigned char newhash[16];
593 int genhash;
594 struct sock *sk1 = NULL;
595 #endif
596 struct net *net;
597
598 /* Never send a reset in response to a reset. */
599 if (th->rst)
600 return;
601
602 /* If sk not NULL, it means we did a successful lookup and incoming
603 * route had to be correct. prequeue might have dropped our dst.
604 */
605 if (!sk && skb_rtable(skb)->rt_type != RTN_LOCAL)
606 return;
607
608 /* Swap the send and the receive. */
609 memset(&rep, 0, sizeof(rep));
610 rep.th.dest = th->source;
611 rep.th.source = th->dest;
612 rep.th.doff = sizeof(struct tcphdr) / 4;
613 rep.th.rst = 1;
614
615 if (th->ack) {
616 rep.th.seq = th->ack_seq;
617 } else {
618 rep.th.ack = 1;
619 rep.th.ack_seq = htonl(ntohl(th->seq) + th->syn + th->fin +
620 skb->len - (th->doff << 2));
621 }
622
623 memset(&arg, 0, sizeof(arg));
624 arg.iov[0].iov_base = (unsigned char *)&rep;
625 arg.iov[0].iov_len = sizeof(rep.th);
626
627 net = sk ? sock_net(sk) : dev_net(skb_dst(skb)->dev);
628 #ifdef CONFIG_TCP_MD5SIG
629 hash_location = tcp_parse_md5sig_option(th);
630 if (!sk && hash_location) {
631 /*
632 * active side is lost. Try to find listening socket through
633 * source port, and then find md5 key through listening socket.
634 * we are not loose security here:
635 * Incoming packet is checked with md5 hash with finding key,
636 * no RST generated if md5 hash doesn't match.
637 */
638 sk1 = __inet_lookup_listener(net,
639 &tcp_hashinfo, ip_hdr(skb)->saddr,
640 th->source, ip_hdr(skb)->daddr,
641 ntohs(th->source), inet_iif(skb));
642 /* don't send rst if it can't find key */
643 if (!sk1)
644 return;
645 rcu_read_lock();
646 key = tcp_md5_do_lookup(sk1, (union tcp_md5_addr *)
647 &ip_hdr(skb)->saddr, AF_INET);
648 if (!key)
649 goto release_sk1;
650
651 genhash = tcp_v4_md5_hash_skb(newhash, key, NULL, skb);
652 if (genhash || memcmp(hash_location, newhash, 16) != 0)
653 goto release_sk1;
654 } else {
655 key = sk ? tcp_md5_do_lookup(sk, (union tcp_md5_addr *)
656 &ip_hdr(skb)->saddr,
657 AF_INET) : NULL;
658 }
659
660 if (key) {
661 rep.opt[0] = htonl((TCPOPT_NOP << 24) |
662 (TCPOPT_NOP << 16) |
663 (TCPOPT_MD5SIG << 8) |
664 TCPOLEN_MD5SIG);
665 /* Update length and the length the header thinks exists */
666 arg.iov[0].iov_len += TCPOLEN_MD5SIG_ALIGNED;
667 rep.th.doff = arg.iov[0].iov_len / 4;
668
669 tcp_v4_md5_hash_hdr((__u8 *) &rep.opt[1],
670 key, ip_hdr(skb)->saddr,
671 ip_hdr(skb)->daddr, &rep.th);
672 }
673 #endif
674 arg.csum = csum_tcpudp_nofold(ip_hdr(skb)->daddr,
675 ip_hdr(skb)->saddr, /* XXX */
676 arg.iov[0].iov_len, IPPROTO_TCP, 0);
677 arg.csumoffset = offsetof(struct tcphdr, check) / 2;
678 arg.flags = (sk && inet_sk(sk)->transparent) ? IP_REPLY_ARG_NOSRCCHECK : 0;
679 /* When socket is gone, all binding information is lost.
680 * routing might fail in this case. No choice here, if we choose to force
681 * input interface, we will misroute in case of asymmetric route.
682 */
683 if (sk)
684 arg.bound_dev_if = sk->sk_bound_dev_if;
685
686 arg.tos = ip_hdr(skb)->tos;
687 ip_send_unicast_reply(*this_cpu_ptr(net->ipv4.tcp_sk),
688 skb, &TCP_SKB_CB(skb)->header.h4.opt,
689 ip_hdr(skb)->saddr, ip_hdr(skb)->daddr,
690 &arg, arg.iov[0].iov_len);
691
692 TCP_INC_STATS_BH(net, TCP_MIB_OUTSEGS);
693 TCP_INC_STATS_BH(net, TCP_MIB_OUTRSTS);
694
695 #ifdef CONFIG_TCP_MD5SIG
696 release_sk1:
697 if (sk1) {
698 rcu_read_unlock();
699 sock_put(sk1);
700 }
701 #endif
702 }
703
704 /* The code following below sending ACKs in SYN-RECV and TIME-WAIT states
705 outside socket context is ugly, certainly. What can I do?
706 */
707
708 static void tcp_v4_send_ack(struct sk_buff *skb, u32 seq, u32 ack,
709 u32 win, u32 tsval, u32 tsecr, int oif,
710 struct tcp_md5sig_key *key,
711 int reply_flags, u8 tos)
712 {
713 const struct tcphdr *th = tcp_hdr(skb);
714 struct {
715 struct tcphdr th;
716 __be32 opt[(TCPOLEN_TSTAMP_ALIGNED >> 2)
717 #ifdef CONFIG_TCP_MD5SIG
718 + (TCPOLEN_MD5SIG_ALIGNED >> 2)
719 #endif
720 ];
721 } rep;
722 struct ip_reply_arg arg;
723 struct net *net = dev_net(skb_dst(skb)->dev);
724
725 memset(&rep.th, 0, sizeof(struct tcphdr));
726 memset(&arg, 0, sizeof(arg));
727
728 arg.iov[0].iov_base = (unsigned char *)&rep;
729 arg.iov[0].iov_len = sizeof(rep.th);
730 if (tsecr) {
731 rep.opt[0] = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
732 (TCPOPT_TIMESTAMP << 8) |
733 TCPOLEN_TIMESTAMP);
734 rep.opt[1] = htonl(tsval);
735 rep.opt[2] = htonl(tsecr);
736 arg.iov[0].iov_len += TCPOLEN_TSTAMP_ALIGNED;
737 }
738
739 /* Swap the send and the receive. */
740 rep.th.dest = th->source;
741 rep.th.source = th->dest;
742 rep.th.doff = arg.iov[0].iov_len / 4;
743 rep.th.seq = htonl(seq);
744 rep.th.ack_seq = htonl(ack);
745 rep.th.ack = 1;
746 rep.th.window = htons(win);
747
748 #ifdef CONFIG_TCP_MD5SIG
749 if (key) {
750 int offset = (tsecr) ? 3 : 0;
751
752 rep.opt[offset++] = htonl((TCPOPT_NOP << 24) |
753 (TCPOPT_NOP << 16) |
754 (TCPOPT_MD5SIG << 8) |
755 TCPOLEN_MD5SIG);
756 arg.iov[0].iov_len += TCPOLEN_MD5SIG_ALIGNED;
757 rep.th.doff = arg.iov[0].iov_len/4;
758
759 tcp_v4_md5_hash_hdr((__u8 *) &rep.opt[offset],
760 key, ip_hdr(skb)->saddr,
761 ip_hdr(skb)->daddr, &rep.th);
762 }
763 #endif
764 arg.flags = reply_flags;
765 arg.csum = csum_tcpudp_nofold(ip_hdr(skb)->daddr,
766 ip_hdr(skb)->saddr, /* XXX */
767 arg.iov[0].iov_len, IPPROTO_TCP, 0);
768 arg.csumoffset = offsetof(struct tcphdr, check) / 2;
769 if (oif)
770 arg.bound_dev_if = oif;
771 arg.tos = tos;
772 ip_send_unicast_reply(*this_cpu_ptr(net->ipv4.tcp_sk),
773 skb, &TCP_SKB_CB(skb)->header.h4.opt,
774 ip_hdr(skb)->saddr, ip_hdr(skb)->daddr,
775 &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 tcp_time_stamp + tcptw->tw_ts_offset,
788 tcptw->tw_ts_recent,
789 tw->tw_bound_dev_if,
790 tcp_twsk_md5_key(tcptw),
791 tw->tw_transparent ? IP_REPLY_ARG_NOSRCCHECK : 0,
792 tw->tw_tos
793 );
794
795 inet_twsk_put(tw);
796 }
797
798 static void tcp_v4_reqsk_send_ack(struct sock *sk, struct sk_buff *skb,
799 struct request_sock *req)
800 {
801 /* sk->sk_state == TCP_LISTEN -> for regular TCP_SYN_RECV
802 * sk->sk_state == TCP_SYN_RECV -> for Fast Open.
803 */
804 tcp_v4_send_ack(skb, (sk->sk_state == TCP_LISTEN) ?
805 tcp_rsk(req)->snt_isn + 1 : tcp_sk(sk)->snd_nxt,
806 tcp_rsk(req)->rcv_nxt, req->rcv_wnd,
807 tcp_time_stamp,
808 req->ts_recent,
809 0,
810 tcp_md5_do_lookup(sk, (union tcp_md5_addr *)&ip_hdr(skb)->daddr,
811 AF_INET),
812 inet_rsk(req)->no_srccheck ? IP_REPLY_ARG_NOSRCCHECK : 0,
813 ip_hdr(skb)->tos);
814 }
815
816 /*
817 * Send a SYN-ACK after having received a SYN.
818 * This still operates on a request_sock only, not on a big
819 * socket.
820 */
821 static int tcp_v4_send_synack(struct sock *sk, struct dst_entry *dst,
822 struct flowi *fl,
823 struct request_sock *req,
824 u16 queue_mapping,
825 struct tcp_fastopen_cookie *foc)
826 {
827 const struct inet_request_sock *ireq = inet_rsk(req);
828 struct flowi4 fl4;
829 int err = -1;
830 struct sk_buff *skb;
831
832 /* First, grab a route. */
833 if (!dst && (dst = inet_csk_route_req(sk, &fl4, req)) == NULL)
834 return -1;
835
836 skb = tcp_make_synack(sk, dst, req, foc);
837
838 if (skb) {
839 __tcp_v4_send_check(skb, ireq->ir_loc_addr, ireq->ir_rmt_addr);
840
841 skb_set_queue_mapping(skb, queue_mapping);
842 err = ip_build_and_send_pkt(skb, sk, ireq->ir_loc_addr,
843 ireq->ir_rmt_addr,
844 ireq->opt);
845 err = net_xmit_eval(err);
846 }
847
848 return err;
849 }
850
851 /*
852 * IPv4 request_sock destructor.
853 */
854 static void tcp_v4_reqsk_destructor(struct request_sock *req)
855 {
856 kfree(inet_rsk(req)->opt);
857 }
858
859
860 #ifdef CONFIG_TCP_MD5SIG
861 /*
862 * RFC2385 MD5 checksumming requires a mapping of
863 * IP address->MD5 Key.
864 * We need to maintain these in the sk structure.
865 */
866
867 /* Find the Key structure for an address. */
868 struct tcp_md5sig_key *tcp_md5_do_lookup(struct sock *sk,
869 const union tcp_md5_addr *addr,
870 int family)
871 {
872 const struct tcp_sock *tp = tcp_sk(sk);
873 struct tcp_md5sig_key *key;
874 unsigned int size = sizeof(struct in_addr);
875 const struct tcp_md5sig_info *md5sig;
876
877 /* caller either holds rcu_read_lock() or socket lock */
878 md5sig = rcu_dereference_check(tp->md5sig_info,
879 sock_owned_by_user(sk) ||
880 lockdep_is_held(&sk->sk_lock.slock));
881 if (!md5sig)
882 return NULL;
883 #if IS_ENABLED(CONFIG_IPV6)
884 if (family == AF_INET6)
885 size = sizeof(struct in6_addr);
886 #endif
887 hlist_for_each_entry_rcu(key, &md5sig->head, node) {
888 if (key->family != family)
889 continue;
890 if (!memcmp(&key->addr, addr, size))
891 return key;
892 }
893 return NULL;
894 }
895 EXPORT_SYMBOL(tcp_md5_do_lookup);
896
897 struct tcp_md5sig_key *tcp_v4_md5_lookup(struct sock *sk,
898 const struct sock *addr_sk)
899 {
900 union tcp_md5_addr *addr;
901
902 addr = (union tcp_md5_addr *)&sk->sk_daddr;
903 return tcp_md5_do_lookup(sk, addr, AF_INET);
904 }
905 EXPORT_SYMBOL(tcp_v4_md5_lookup);
906
907 /* This can be called on a newly created socket, from other files */
908 int tcp_md5_do_add(struct sock *sk, const union tcp_md5_addr *addr,
909 int family, const u8 *newkey, u8 newkeylen, gfp_t gfp)
910 {
911 /* Add Key to the list */
912 struct tcp_md5sig_key *key;
913 struct tcp_sock *tp = tcp_sk(sk);
914 struct tcp_md5sig_info *md5sig;
915
916 key = tcp_md5_do_lookup(sk, addr, family);
917 if (key) {
918 /* Pre-existing entry - just update that one. */
919 memcpy(key->key, newkey, newkeylen);
920 key->keylen = newkeylen;
921 return 0;
922 }
923
924 md5sig = rcu_dereference_protected(tp->md5sig_info,
925 sock_owned_by_user(sk));
926 if (!md5sig) {
927 md5sig = kmalloc(sizeof(*md5sig), gfp);
928 if (!md5sig)
929 return -ENOMEM;
930
931 sk_nocaps_add(sk, NETIF_F_GSO_MASK);
932 INIT_HLIST_HEAD(&md5sig->head);
933 rcu_assign_pointer(tp->md5sig_info, md5sig);
934 }
935
936 key = sock_kmalloc(sk, sizeof(*key), gfp);
937 if (!key)
938 return -ENOMEM;
939 if (!tcp_alloc_md5sig_pool()) {
940 sock_kfree_s(sk, key, sizeof(*key));
941 return -ENOMEM;
942 }
943
944 memcpy(key->key, newkey, newkeylen);
945 key->keylen = newkeylen;
946 key->family = family;
947 memcpy(&key->addr, addr,
948 (family == AF_INET6) ? sizeof(struct in6_addr) :
949 sizeof(struct in_addr));
950 hlist_add_head_rcu(&key->node, &md5sig->head);
951 return 0;
952 }
953 EXPORT_SYMBOL(tcp_md5_do_add);
954
955 int tcp_md5_do_del(struct sock *sk, const union tcp_md5_addr *addr, int family)
956 {
957 struct tcp_md5sig_key *key;
958
959 key = tcp_md5_do_lookup(sk, addr, family);
960 if (!key)
961 return -ENOENT;
962 hlist_del_rcu(&key->node);
963 atomic_sub(sizeof(*key), &sk->sk_omem_alloc);
964 kfree_rcu(key, rcu);
965 return 0;
966 }
967 EXPORT_SYMBOL(tcp_md5_do_del);
968
969 static void tcp_clear_md5_list(struct sock *sk)
970 {
971 struct tcp_sock *tp = tcp_sk(sk);
972 struct tcp_md5sig_key *key;
973 struct hlist_node *n;
974 struct tcp_md5sig_info *md5sig;
975
976 md5sig = rcu_dereference_protected(tp->md5sig_info, 1);
977
978 hlist_for_each_entry_safe(key, n, &md5sig->head, node) {
979 hlist_del_rcu(&key->node);
980 atomic_sub(sizeof(*key), &sk->sk_omem_alloc);
981 kfree_rcu(key, rcu);
982 }
983 }
984
985 static int tcp_v4_parse_md5_keys(struct sock *sk, char __user *optval,
986 int optlen)
987 {
988 struct tcp_md5sig cmd;
989 struct sockaddr_in *sin = (struct sockaddr_in *)&cmd.tcpm_addr;
990
991 if (optlen < sizeof(cmd))
992 return -EINVAL;
993
994 if (copy_from_user(&cmd, optval, sizeof(cmd)))
995 return -EFAULT;
996
997 if (sin->sin_family != AF_INET)
998 return -EINVAL;
999
1000 if (!cmd.tcpm_keylen)
1001 return tcp_md5_do_del(sk, (union tcp_md5_addr *)&sin->sin_addr.s_addr,
1002 AF_INET);
1003
1004 if (cmd.tcpm_keylen > TCP_MD5SIG_MAXKEYLEN)
1005 return -EINVAL;
1006
1007 return tcp_md5_do_add(sk, (union tcp_md5_addr *)&sin->sin_addr.s_addr,
1008 AF_INET, cmd.tcpm_key, cmd.tcpm_keylen,
1009 GFP_KERNEL);
1010 }
1011
1012 static int tcp_v4_md5_hash_pseudoheader(struct tcp_md5sig_pool *hp,
1013 __be32 daddr, __be32 saddr, int nbytes)
1014 {
1015 struct tcp4_pseudohdr *bp;
1016 struct scatterlist sg;
1017
1018 bp = &hp->md5_blk.ip4;
1019
1020 /*
1021 * 1. the TCP pseudo-header (in the order: source IP address,
1022 * destination IP address, zero-padded protocol number, and
1023 * segment length)
1024 */
1025 bp->saddr = saddr;
1026 bp->daddr = daddr;
1027 bp->pad = 0;
1028 bp->protocol = IPPROTO_TCP;
1029 bp->len = cpu_to_be16(nbytes);
1030
1031 sg_init_one(&sg, bp, sizeof(*bp));
1032 return crypto_hash_update(&hp->md5_desc, &sg, sizeof(*bp));
1033 }
1034
1035 static int tcp_v4_md5_hash_hdr(char *md5_hash, const struct tcp_md5sig_key *key,
1036 __be32 daddr, __be32 saddr, const struct tcphdr *th)
1037 {
1038 struct tcp_md5sig_pool *hp;
1039 struct hash_desc *desc;
1040
1041 hp = tcp_get_md5sig_pool();
1042 if (!hp)
1043 goto clear_hash_noput;
1044 desc = &hp->md5_desc;
1045
1046 if (crypto_hash_init(desc))
1047 goto clear_hash;
1048 if (tcp_v4_md5_hash_pseudoheader(hp, daddr, saddr, th->doff << 2))
1049 goto clear_hash;
1050 if (tcp_md5_hash_header(hp, th))
1051 goto clear_hash;
1052 if (tcp_md5_hash_key(hp, key))
1053 goto clear_hash;
1054 if (crypto_hash_final(desc, md5_hash))
1055 goto clear_hash;
1056
1057 tcp_put_md5sig_pool();
1058 return 0;
1059
1060 clear_hash:
1061 tcp_put_md5sig_pool();
1062 clear_hash_noput:
1063 memset(md5_hash, 0, 16);
1064 return 1;
1065 }
1066
1067 int tcp_v4_md5_hash_skb(char *md5_hash, const struct tcp_md5sig_key *key,
1068 const struct sock *sk,
1069 const struct sk_buff *skb)
1070 {
1071 struct tcp_md5sig_pool *hp;
1072 struct hash_desc *desc;
1073 const struct tcphdr *th = tcp_hdr(skb);
1074 __be32 saddr, daddr;
1075
1076 if (sk) { /* valid for establish/request sockets */
1077 saddr = sk->sk_rcv_saddr;
1078 daddr = sk->sk_daddr;
1079 } else {
1080 const struct iphdr *iph = ip_hdr(skb);
1081 saddr = iph->saddr;
1082 daddr = iph->daddr;
1083 }
1084
1085 hp = tcp_get_md5sig_pool();
1086 if (!hp)
1087 goto clear_hash_noput;
1088 desc = &hp->md5_desc;
1089
1090 if (crypto_hash_init(desc))
1091 goto clear_hash;
1092
1093 if (tcp_v4_md5_hash_pseudoheader(hp, daddr, saddr, skb->len))
1094 goto clear_hash;
1095 if (tcp_md5_hash_header(hp, th))
1096 goto clear_hash;
1097 if (tcp_md5_hash_skb_data(hp, skb, th->doff << 2))
1098 goto clear_hash;
1099 if (tcp_md5_hash_key(hp, key))
1100 goto clear_hash;
1101 if (crypto_hash_final(desc, md5_hash))
1102 goto clear_hash;
1103
1104 tcp_put_md5sig_pool();
1105 return 0;
1106
1107 clear_hash:
1108 tcp_put_md5sig_pool();
1109 clear_hash_noput:
1110 memset(md5_hash, 0, 16);
1111 return 1;
1112 }
1113 EXPORT_SYMBOL(tcp_v4_md5_hash_skb);
1114
1115 /* Called with rcu_read_lock() */
1116 static bool tcp_v4_inbound_md5_hash(struct sock *sk,
1117 const struct sk_buff *skb)
1118 {
1119 /*
1120 * This gets called for each TCP segment that arrives
1121 * so we want to be efficient.
1122 * We have 3 drop cases:
1123 * o No MD5 hash and one expected.
1124 * o MD5 hash and we're not expecting one.
1125 * o MD5 hash and its wrong.
1126 */
1127 const __u8 *hash_location = NULL;
1128 struct tcp_md5sig_key *hash_expected;
1129 const struct iphdr *iph = ip_hdr(skb);
1130 const struct tcphdr *th = tcp_hdr(skb);
1131 int genhash;
1132 unsigned char newhash[16];
1133
1134 hash_expected = tcp_md5_do_lookup(sk, (union tcp_md5_addr *)&iph->saddr,
1135 AF_INET);
1136 hash_location = tcp_parse_md5sig_option(th);
1137
1138 /* We've parsed the options - do we have a hash? */
1139 if (!hash_expected && !hash_location)
1140 return false;
1141
1142 if (hash_expected && !hash_location) {
1143 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPMD5NOTFOUND);
1144 return true;
1145 }
1146
1147 if (!hash_expected && hash_location) {
1148 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPMD5UNEXPECTED);
1149 return true;
1150 }
1151
1152 /* Okay, so this is hash_expected and hash_location -
1153 * so we need to calculate the checksum.
1154 */
1155 genhash = tcp_v4_md5_hash_skb(newhash,
1156 hash_expected,
1157 NULL, skb);
1158
1159 if (genhash || memcmp(hash_location, newhash, 16) != 0) {
1160 net_info_ratelimited("MD5 Hash failed for (%pI4, %d)->(%pI4, %d)%s\n",
1161 &iph->saddr, ntohs(th->source),
1162 &iph->daddr, ntohs(th->dest),
1163 genhash ? " tcp_v4_calc_md5_hash failed"
1164 : "");
1165 return true;
1166 }
1167 return false;
1168 }
1169 #endif
1170
1171 static void tcp_v4_init_req(struct request_sock *req, struct sock *sk_listener,
1172 struct sk_buff *skb)
1173 {
1174 struct inet_request_sock *ireq = inet_rsk(req);
1175
1176 sk_rcv_saddr_set(req_to_sk(req), ip_hdr(skb)->daddr);
1177 sk_daddr_set(req_to_sk(req), ip_hdr(skb)->saddr);
1178 ireq->no_srccheck = inet_sk(sk_listener)->transparent;
1179 ireq->opt = tcp_v4_save_options(skb);
1180 }
1181
1182 static struct dst_entry *tcp_v4_route_req(struct sock *sk, struct flowi *fl,
1183 const struct request_sock *req,
1184 bool *strict)
1185 {
1186 struct dst_entry *dst = inet_csk_route_req(sk, &fl->u.ip4, req);
1187
1188 if (strict) {
1189 if (fl->u.ip4.daddr == inet_rsk(req)->ir_rmt_addr)
1190 *strict = true;
1191 else
1192 *strict = false;
1193 }
1194
1195 return dst;
1196 }
1197
1198 struct request_sock_ops tcp_request_sock_ops __read_mostly = {
1199 .family = PF_INET,
1200 .obj_size = sizeof(struct tcp_request_sock),
1201 .rtx_syn_ack = tcp_rtx_synack,
1202 .send_ack = tcp_v4_reqsk_send_ack,
1203 .destructor = tcp_v4_reqsk_destructor,
1204 .send_reset = tcp_v4_send_reset,
1205 .syn_ack_timeout = tcp_syn_ack_timeout,
1206 };
1207
1208 static const struct tcp_request_sock_ops tcp_request_sock_ipv4_ops = {
1209 .mss_clamp = TCP_MSS_DEFAULT,
1210 #ifdef CONFIG_TCP_MD5SIG
1211 .req_md5_lookup = tcp_v4_md5_lookup,
1212 .calc_md5_hash = tcp_v4_md5_hash_skb,
1213 #endif
1214 .init_req = tcp_v4_init_req,
1215 #ifdef CONFIG_SYN_COOKIES
1216 .cookie_init_seq = cookie_v4_init_sequence,
1217 #endif
1218 .route_req = tcp_v4_route_req,
1219 .init_seq = tcp_v4_init_sequence,
1220 .send_synack = tcp_v4_send_synack,
1221 .queue_hash_add = inet_csk_reqsk_queue_hash_add,
1222 };
1223
1224 int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb)
1225 {
1226 /* Never answer to SYNs send to broadcast or multicast */
1227 if (skb_rtable(skb)->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))
1228 goto drop;
1229
1230 return tcp_conn_request(&tcp_request_sock_ops,
1231 &tcp_request_sock_ipv4_ops, sk, skb);
1232
1233 drop:
1234 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENDROPS);
1235 return 0;
1236 }
1237 EXPORT_SYMBOL(tcp_v4_conn_request);
1238
1239
1240 /*
1241 * The three way handshake has completed - we got a valid synack -
1242 * now create the new socket.
1243 */
1244 struct sock *tcp_v4_syn_recv_sock(struct sock *sk, struct sk_buff *skb,
1245 struct request_sock *req,
1246 struct dst_entry *dst)
1247 {
1248 struct inet_request_sock *ireq;
1249 struct inet_sock *newinet;
1250 struct tcp_sock *newtp;
1251 struct sock *newsk;
1252 #ifdef CONFIG_TCP_MD5SIG
1253 struct tcp_md5sig_key *key;
1254 #endif
1255 struct ip_options_rcu *inet_opt;
1256
1257 if (sk_acceptq_is_full(sk))
1258 goto exit_overflow;
1259
1260 newsk = tcp_create_openreq_child(sk, req, skb);
1261 if (!newsk)
1262 goto exit_nonewsk;
1263
1264 newsk->sk_gso_type = SKB_GSO_TCPV4;
1265 inet_sk_rx_dst_set(newsk, skb);
1266
1267 newtp = tcp_sk(newsk);
1268 newinet = inet_sk(newsk);
1269 ireq = inet_rsk(req);
1270 sk_daddr_set(newsk, ireq->ir_rmt_addr);
1271 sk_rcv_saddr_set(newsk, ireq->ir_loc_addr);
1272 newinet->inet_saddr = ireq->ir_loc_addr;
1273 inet_opt = ireq->opt;
1274 rcu_assign_pointer(newinet->inet_opt, inet_opt);
1275 ireq->opt = NULL;
1276 newinet->mc_index = inet_iif(skb);
1277 newinet->mc_ttl = ip_hdr(skb)->ttl;
1278 newinet->rcv_tos = ip_hdr(skb)->tos;
1279 inet_csk(newsk)->icsk_ext_hdr_len = 0;
1280 inet_set_txhash(newsk);
1281 if (inet_opt)
1282 inet_csk(newsk)->icsk_ext_hdr_len = inet_opt->opt.optlen;
1283 newinet->inet_id = newtp->write_seq ^ jiffies;
1284
1285 if (!dst) {
1286 dst = inet_csk_route_child_sock(sk, newsk, req);
1287 if (!dst)
1288 goto put_and_exit;
1289 } else {
1290 /* syncookie case : see end of cookie_v4_check() */
1291 }
1292 sk_setup_caps(newsk, dst);
1293
1294 tcp_ca_openreq_child(newsk, dst);
1295
1296 tcp_sync_mss(newsk, dst_mtu(dst));
1297 newtp->advmss = dst_metric_advmss(dst);
1298 if (tcp_sk(sk)->rx_opt.user_mss &&
1299 tcp_sk(sk)->rx_opt.user_mss < newtp->advmss)
1300 newtp->advmss = tcp_sk(sk)->rx_opt.user_mss;
1301
1302 tcp_initialize_rcv_mss(newsk);
1303
1304 #ifdef CONFIG_TCP_MD5SIG
1305 /* Copy over the MD5 key from the original socket */
1306 key = tcp_md5_do_lookup(sk, (union tcp_md5_addr *)&newinet->inet_daddr,
1307 AF_INET);
1308 if (key != NULL) {
1309 /*
1310 * We're using one, so create a matching key
1311 * on the newsk structure. If we fail to get
1312 * memory, then we end up not copying the key
1313 * across. Shucks.
1314 */
1315 tcp_md5_do_add(newsk, (union tcp_md5_addr *)&newinet->inet_daddr,
1316 AF_INET, key->key, key->keylen, GFP_ATOMIC);
1317 sk_nocaps_add(newsk, NETIF_F_GSO_MASK);
1318 }
1319 #endif
1320
1321 if (__inet_inherit_port(sk, newsk) < 0)
1322 goto put_and_exit;
1323 __inet_hash_nolisten(newsk, NULL);
1324
1325 return newsk;
1326
1327 exit_overflow:
1328 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENOVERFLOWS);
1329 exit_nonewsk:
1330 dst_release(dst);
1331 exit:
1332 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENDROPS);
1333 return NULL;
1334 put_and_exit:
1335 inet_csk_prepare_forced_close(newsk);
1336 tcp_done(newsk);
1337 goto exit;
1338 }
1339 EXPORT_SYMBOL(tcp_v4_syn_recv_sock);
1340
1341 static struct sock *tcp_v4_hnd_req(struct sock *sk, struct sk_buff *skb)
1342 {
1343 const struct tcphdr *th = tcp_hdr(skb);
1344 const struct iphdr *iph = ip_hdr(skb);
1345 struct request_sock *req;
1346 struct sock *nsk;
1347
1348 req = inet_csk_search_req(sk, th->source, iph->saddr, iph->daddr);
1349 if (req) {
1350 nsk = tcp_check_req(sk, skb, req, false);
1351 reqsk_put(req);
1352 return nsk;
1353 }
1354
1355 nsk = inet_lookup_established(sock_net(sk), &tcp_hashinfo, iph->saddr,
1356 th->source, iph->daddr, th->dest, inet_iif(skb));
1357
1358 if (nsk) {
1359 if (nsk->sk_state != TCP_TIME_WAIT) {
1360 bh_lock_sock(nsk);
1361 return nsk;
1362 }
1363 inet_twsk_put(inet_twsk(nsk));
1364 return NULL;
1365 }
1366
1367 #ifdef CONFIG_SYN_COOKIES
1368 if (!th->syn)
1369 sk = cookie_v4_check(sk, skb);
1370 #endif
1371 return sk;
1372 }
1373
1374 /* The socket must have it's spinlock held when we get
1375 * here.
1376 *
1377 * We have a potential double-lock case here, so even when
1378 * doing backlog processing we use the BH locking scheme.
1379 * This is because we cannot sleep with the original spinlock
1380 * held.
1381 */
1382 int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb)
1383 {
1384 struct sock *rsk;
1385
1386 if (sk->sk_state == TCP_ESTABLISHED) { /* Fast path */
1387 struct dst_entry *dst = sk->sk_rx_dst;
1388
1389 sock_rps_save_rxhash(sk, skb);
1390 sk_mark_napi_id(sk, skb);
1391 if (dst) {
1392 if (inet_sk(sk)->rx_dst_ifindex != skb->skb_iif ||
1393 dst->ops->check(dst, 0) == NULL) {
1394 dst_release(dst);
1395 sk->sk_rx_dst = NULL;
1396 }
1397 }
1398 tcp_rcv_established(sk, skb, tcp_hdr(skb), skb->len);
1399 return 0;
1400 }
1401
1402 if (skb->len < tcp_hdrlen(skb) || tcp_checksum_complete(skb))
1403 goto csum_err;
1404
1405 if (sk->sk_state == TCP_LISTEN) {
1406 struct sock *nsk = tcp_v4_hnd_req(sk, skb);
1407 if (!nsk)
1408 goto discard;
1409
1410 if (nsk != sk) {
1411 sock_rps_save_rxhash(nsk, skb);
1412 sk_mark_napi_id(sk, skb);
1413 if (tcp_child_process(sk, nsk, skb)) {
1414 rsk = nsk;
1415 goto reset;
1416 }
1417 return 0;
1418 }
1419 } else
1420 sock_rps_save_rxhash(sk, skb);
1421
1422 if (tcp_rcv_state_process(sk, skb, tcp_hdr(skb), skb->len)) {
1423 rsk = sk;
1424 goto reset;
1425 }
1426 return 0;
1427
1428 reset:
1429 tcp_v4_send_reset(rsk, skb);
1430 discard:
1431 kfree_skb(skb);
1432 /* Be careful here. If this function gets more complicated and
1433 * gcc suffers from register pressure on the x86, sk (in %ebx)
1434 * might be destroyed here. This current version compiles correctly,
1435 * but you have been warned.
1436 */
1437 return 0;
1438
1439 csum_err:
1440 TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_CSUMERRORS);
1441 TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_INERRS);
1442 goto discard;
1443 }
1444 EXPORT_SYMBOL(tcp_v4_do_rcv);
1445
1446 void tcp_v4_early_demux(struct sk_buff *skb)
1447 {
1448 const struct iphdr *iph;
1449 const struct tcphdr *th;
1450 struct sock *sk;
1451
1452 if (skb->pkt_type != PACKET_HOST)
1453 return;
1454
1455 if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct tcphdr)))
1456 return;
1457
1458 iph = ip_hdr(skb);
1459 th = tcp_hdr(skb);
1460
1461 if (th->doff < sizeof(struct tcphdr) / 4)
1462 return;
1463
1464 sk = __inet_lookup_established(dev_net(skb->dev), &tcp_hashinfo,
1465 iph->saddr, th->source,
1466 iph->daddr, ntohs(th->dest),
1467 skb->skb_iif);
1468 if (sk) {
1469 skb->sk = sk;
1470 skb->destructor = sock_edemux;
1471 if (sk_fullsock(sk)) {
1472 struct dst_entry *dst = READ_ONCE(sk->sk_rx_dst);
1473
1474 if (dst)
1475 dst = dst_check(dst, 0);
1476 if (dst &&
1477 inet_sk(sk)->rx_dst_ifindex == skb->skb_iif)
1478 skb_dst_set_noref(skb, dst);
1479 }
1480 }
1481 }
1482
1483 /* Packet is added to VJ-style prequeue for processing in process
1484 * context, if a reader task is waiting. Apparently, this exciting
1485 * idea (VJ's mail "Re: query about TCP header on tcp-ip" of 07 Sep 93)
1486 * failed somewhere. Latency? Burstiness? Well, at least now we will
1487 * see, why it failed. 8)8) --ANK
1488 *
1489 */
1490 bool tcp_prequeue(struct sock *sk, struct sk_buff *skb)
1491 {
1492 struct tcp_sock *tp = tcp_sk(sk);
1493
1494 if (sysctl_tcp_low_latency || !tp->ucopy.task)
1495 return false;
1496
1497 if (skb->len <= tcp_hdrlen(skb) &&
1498 skb_queue_len(&tp->ucopy.prequeue) == 0)
1499 return false;
1500
1501 /* Before escaping RCU protected region, we need to take care of skb
1502 * dst. Prequeue is only enabled for established sockets.
1503 * For such sockets, we might need the skb dst only to set sk->sk_rx_dst
1504 * Instead of doing full sk_rx_dst validity here, let's perform
1505 * an optimistic check.
1506 */
1507 if (likely(sk->sk_rx_dst))
1508 skb_dst_drop(skb);
1509 else
1510 skb_dst_force(skb);
1511
1512 __skb_queue_tail(&tp->ucopy.prequeue, skb);
1513 tp->ucopy.memory += skb->truesize;
1514 if (tp->ucopy.memory > sk->sk_rcvbuf) {
1515 struct sk_buff *skb1;
1516
1517 BUG_ON(sock_owned_by_user(sk));
1518
1519 while ((skb1 = __skb_dequeue(&tp->ucopy.prequeue)) != NULL) {
1520 sk_backlog_rcv(sk, skb1);
1521 NET_INC_STATS_BH(sock_net(sk),
1522 LINUX_MIB_TCPPREQUEUEDROPPED);
1523 }
1524
1525 tp->ucopy.memory = 0;
1526 } else if (skb_queue_len(&tp->ucopy.prequeue) == 1) {
1527 wake_up_interruptible_sync_poll(sk_sleep(sk),
1528 POLLIN | POLLRDNORM | POLLRDBAND);
1529 if (!inet_csk_ack_scheduled(sk))
1530 inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK,
1531 (3 * tcp_rto_min(sk)) / 4,
1532 TCP_RTO_MAX);
1533 }
1534 return true;
1535 }
1536 EXPORT_SYMBOL(tcp_prequeue);
1537
1538 /*
1539 * From tcp_input.c
1540 */
1541
1542 int tcp_v4_rcv(struct sk_buff *skb)
1543 {
1544 const struct iphdr *iph;
1545 const struct tcphdr *th;
1546 struct sock *sk;
1547 int ret;
1548 struct net *net = dev_net(skb->dev);
1549
1550 if (skb->pkt_type != PACKET_HOST)
1551 goto discard_it;
1552
1553 /* Count it even if it's bad */
1554 TCP_INC_STATS_BH(net, TCP_MIB_INSEGS);
1555
1556 if (!pskb_may_pull(skb, sizeof(struct tcphdr)))
1557 goto discard_it;
1558
1559 th = tcp_hdr(skb);
1560
1561 if (th->doff < sizeof(struct tcphdr) / 4)
1562 goto bad_packet;
1563 if (!pskb_may_pull(skb, th->doff * 4))
1564 goto discard_it;
1565
1566 /* An explanation is required here, I think.
1567 * Packet length and doff are validated by header prediction,
1568 * provided case of th->doff==0 is eliminated.
1569 * So, we defer the checks. */
1570
1571 if (skb_checksum_init(skb, IPPROTO_TCP, inet_compute_pseudo))
1572 goto csum_error;
1573
1574 th = tcp_hdr(skb);
1575 iph = ip_hdr(skb);
1576 /* This is tricky : We move IPCB at its correct location into TCP_SKB_CB()
1577 * barrier() makes sure compiler wont play fool^Waliasing games.
1578 */
1579 memmove(&TCP_SKB_CB(skb)->header.h4, IPCB(skb),
1580 sizeof(struct inet_skb_parm));
1581 barrier();
1582
1583 TCP_SKB_CB(skb)->seq = ntohl(th->seq);
1584 TCP_SKB_CB(skb)->end_seq = (TCP_SKB_CB(skb)->seq + th->syn + th->fin +
1585 skb->len - th->doff * 4);
1586 TCP_SKB_CB(skb)->ack_seq = ntohl(th->ack_seq);
1587 TCP_SKB_CB(skb)->tcp_flags = tcp_flag_byte(th);
1588 TCP_SKB_CB(skb)->tcp_tw_isn = 0;
1589 TCP_SKB_CB(skb)->ip_dsfield = ipv4_get_dsfield(iph);
1590 TCP_SKB_CB(skb)->sacked = 0;
1591
1592 sk = __inet_lookup_skb(&tcp_hashinfo, skb, th->source, th->dest);
1593 if (!sk)
1594 goto no_tcp_socket;
1595
1596 process:
1597 if (sk->sk_state == TCP_TIME_WAIT)
1598 goto do_time_wait;
1599
1600 if (unlikely(iph->ttl < inet_sk(sk)->min_ttl)) {
1601 NET_INC_STATS_BH(net, LINUX_MIB_TCPMINTTLDROP);
1602 goto discard_and_relse;
1603 }
1604
1605 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
1606 goto discard_and_relse;
1607
1608 #ifdef CONFIG_TCP_MD5SIG
1609 /*
1610 * We really want to reject the packet as early as possible
1611 * if:
1612 * o We're expecting an MD5'd packet and this is no MD5 tcp option
1613 * o There is an MD5 option and we're not expecting one
1614 */
1615 if (tcp_v4_inbound_md5_hash(sk, skb))
1616 goto discard_and_relse;
1617 #endif
1618
1619 nf_reset(skb);
1620
1621 if (sk_filter(sk, skb))
1622 goto discard_and_relse;
1623
1624 sk_incoming_cpu_update(sk);
1625 skb->dev = NULL;
1626
1627 bh_lock_sock_nested(sk);
1628 ret = 0;
1629 if (!sock_owned_by_user(sk)) {
1630 if (!tcp_prequeue(sk, skb))
1631 ret = tcp_v4_do_rcv(sk, skb);
1632 } else if (unlikely(sk_add_backlog(sk, skb,
1633 sk->sk_rcvbuf + sk->sk_sndbuf))) {
1634 bh_unlock_sock(sk);
1635 NET_INC_STATS_BH(net, LINUX_MIB_TCPBACKLOGDROP);
1636 goto discard_and_relse;
1637 }
1638 bh_unlock_sock(sk);
1639
1640 sock_put(sk);
1641
1642 return ret;
1643
1644 no_tcp_socket:
1645 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
1646 goto discard_it;
1647
1648 if (skb->len < (th->doff << 2) || tcp_checksum_complete(skb)) {
1649 csum_error:
1650 TCP_INC_STATS_BH(net, TCP_MIB_CSUMERRORS);
1651 bad_packet:
1652 TCP_INC_STATS_BH(net, TCP_MIB_INERRS);
1653 } else {
1654 tcp_v4_send_reset(NULL, skb);
1655 }
1656
1657 discard_it:
1658 /* Discard frame. */
1659 kfree_skb(skb);
1660 return 0;
1661
1662 discard_and_relse:
1663 sock_put(sk);
1664 goto discard_it;
1665
1666 do_time_wait:
1667 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) {
1668 inet_twsk_put(inet_twsk(sk));
1669 goto discard_it;
1670 }
1671
1672 if (skb->len < (th->doff << 2)) {
1673 inet_twsk_put(inet_twsk(sk));
1674 goto bad_packet;
1675 }
1676 if (tcp_checksum_complete(skb)) {
1677 inet_twsk_put(inet_twsk(sk));
1678 goto csum_error;
1679 }
1680 switch (tcp_timewait_state_process(inet_twsk(sk), skb, th)) {
1681 case TCP_TW_SYN: {
1682 struct sock *sk2 = inet_lookup_listener(dev_net(skb->dev),
1683 &tcp_hashinfo,
1684 iph->saddr, th->source,
1685 iph->daddr, th->dest,
1686 inet_iif(skb));
1687 if (sk2) {
1688 inet_twsk_deschedule(inet_twsk(sk), &tcp_death_row);
1689 inet_twsk_put(inet_twsk(sk));
1690 sk = sk2;
1691 goto process;
1692 }
1693 /* Fall through to ACK */
1694 }
1695 case TCP_TW_ACK:
1696 tcp_v4_timewait_ack(sk, skb);
1697 break;
1698 case TCP_TW_RST:
1699 goto no_tcp_socket;
1700 case TCP_TW_SUCCESS:;
1701 }
1702 goto discard_it;
1703 }
1704
1705 static struct timewait_sock_ops tcp_timewait_sock_ops = {
1706 .twsk_obj_size = sizeof(struct tcp_timewait_sock),
1707 .twsk_unique = tcp_twsk_unique,
1708 .twsk_destructor= tcp_twsk_destructor,
1709 };
1710
1711 void inet_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb)
1712 {
1713 struct dst_entry *dst = skb_dst(skb);
1714
1715 if (dst) {
1716 dst_hold(dst);
1717 sk->sk_rx_dst = dst;
1718 inet_sk(sk)->rx_dst_ifindex = skb->skb_iif;
1719 }
1720 }
1721 EXPORT_SYMBOL(inet_sk_rx_dst_set);
1722
1723 const struct inet_connection_sock_af_ops ipv4_specific = {
1724 .queue_xmit = ip_queue_xmit,
1725 .send_check = tcp_v4_send_check,
1726 .rebuild_header = inet_sk_rebuild_header,
1727 .sk_rx_dst_set = inet_sk_rx_dst_set,
1728 .conn_request = tcp_v4_conn_request,
1729 .syn_recv_sock = tcp_v4_syn_recv_sock,
1730 .net_header_len = sizeof(struct iphdr),
1731 .setsockopt = ip_setsockopt,
1732 .getsockopt = ip_getsockopt,
1733 .addr2sockaddr = inet_csk_addr2sockaddr,
1734 .sockaddr_len = sizeof(struct sockaddr_in),
1735 .bind_conflict = inet_csk_bind_conflict,
1736 #ifdef CONFIG_COMPAT
1737 .compat_setsockopt = compat_ip_setsockopt,
1738 .compat_getsockopt = compat_ip_getsockopt,
1739 #endif
1740 .mtu_reduced = tcp_v4_mtu_reduced,
1741 };
1742 EXPORT_SYMBOL(ipv4_specific);
1743
1744 #ifdef CONFIG_TCP_MD5SIG
1745 static const struct tcp_sock_af_ops tcp_sock_ipv4_specific = {
1746 .md5_lookup = tcp_v4_md5_lookup,
1747 .calc_md5_hash = tcp_v4_md5_hash_skb,
1748 .md5_parse = tcp_v4_parse_md5_keys,
1749 };
1750 #endif
1751
1752 /* NOTE: A lot of things set to zero explicitly by call to
1753 * sk_alloc() so need not be done here.
1754 */
1755 static int tcp_v4_init_sock(struct sock *sk)
1756 {
1757 struct inet_connection_sock *icsk = inet_csk(sk);
1758
1759 tcp_init_sock(sk);
1760
1761 icsk->icsk_af_ops = &ipv4_specific;
1762
1763 #ifdef CONFIG_TCP_MD5SIG
1764 tcp_sk(sk)->af_specific = &tcp_sock_ipv4_specific;
1765 #endif
1766
1767 return 0;
1768 }
1769
1770 void tcp_v4_destroy_sock(struct sock *sk)
1771 {
1772 struct tcp_sock *tp = tcp_sk(sk);
1773
1774 tcp_clear_xmit_timers(sk);
1775
1776 tcp_cleanup_congestion_control(sk);
1777
1778 /* Cleanup up the write buffer. */
1779 tcp_write_queue_purge(sk);
1780
1781 /* Cleans up our, hopefully empty, out_of_order_queue. */
1782 __skb_queue_purge(&tp->out_of_order_queue);
1783
1784 #ifdef CONFIG_TCP_MD5SIG
1785 /* Clean up the MD5 key list, if any */
1786 if (tp->md5sig_info) {
1787 tcp_clear_md5_list(sk);
1788 kfree_rcu(tp->md5sig_info, rcu);
1789 tp->md5sig_info = NULL;
1790 }
1791 #endif
1792
1793 /* Clean prequeue, it must be empty really */
1794 __skb_queue_purge(&tp->ucopy.prequeue);
1795
1796 /* Clean up a referenced TCP bind bucket. */
1797 if (inet_csk(sk)->icsk_bind_hash)
1798 inet_put_port(sk);
1799
1800 BUG_ON(tp->fastopen_rsk != NULL);
1801
1802 /* If socket is aborted during connect operation */
1803 tcp_free_fastopen_req(tp);
1804
1805 sk_sockets_allocated_dec(sk);
1806 sock_release_memcg(sk);
1807 }
1808 EXPORT_SYMBOL(tcp_v4_destroy_sock);
1809
1810 #ifdef CONFIG_PROC_FS
1811 /* Proc filesystem TCP sock list dumping. */
1812
1813 /*
1814 * Get next listener socket follow cur. If cur is NULL, get first socket
1815 * starting from bucket given in st->bucket; when st->bucket is zero the
1816 * very first socket in the hash table is returned.
1817 */
1818 static void *listening_get_next(struct seq_file *seq, void *cur)
1819 {
1820 struct inet_connection_sock *icsk;
1821 struct hlist_nulls_node *node;
1822 struct sock *sk = cur;
1823 struct inet_listen_hashbucket *ilb;
1824 struct tcp_iter_state *st = seq->private;
1825 struct net *net = seq_file_net(seq);
1826
1827 if (!sk) {
1828 ilb = &tcp_hashinfo.listening_hash[st->bucket];
1829 spin_lock_bh(&ilb->lock);
1830 sk = sk_nulls_head(&ilb->head);
1831 st->offset = 0;
1832 goto get_sk;
1833 }
1834 ilb = &tcp_hashinfo.listening_hash[st->bucket];
1835 ++st->num;
1836 ++st->offset;
1837
1838 if (st->state == TCP_SEQ_STATE_OPENREQ) {
1839 struct request_sock *req = cur;
1840
1841 icsk = inet_csk(st->syn_wait_sk);
1842 req = req->dl_next;
1843 while (1) {
1844 while (req) {
1845 if (req->rsk_ops->family == st->family) {
1846 cur = req;
1847 goto out;
1848 }
1849 req = req->dl_next;
1850 }
1851 if (++st->sbucket >= icsk->icsk_accept_queue.listen_opt->nr_table_entries)
1852 break;
1853 get_req:
1854 req = icsk->icsk_accept_queue.listen_opt->syn_table[st->sbucket];
1855 }
1856 sk = sk_nulls_next(st->syn_wait_sk);
1857 st->state = TCP_SEQ_STATE_LISTENING;
1858 spin_unlock_bh(&icsk->icsk_accept_queue.syn_wait_lock);
1859 } else {
1860 icsk = inet_csk(sk);
1861 spin_lock_bh(&icsk->icsk_accept_queue.syn_wait_lock);
1862 if (reqsk_queue_len(&icsk->icsk_accept_queue))
1863 goto start_req;
1864 spin_unlock_bh(&icsk->icsk_accept_queue.syn_wait_lock);
1865 sk = sk_nulls_next(sk);
1866 }
1867 get_sk:
1868 sk_nulls_for_each_from(sk, node) {
1869 if (!net_eq(sock_net(sk), net))
1870 continue;
1871 if (sk->sk_family == st->family) {
1872 cur = sk;
1873 goto out;
1874 }
1875 icsk = inet_csk(sk);
1876 spin_lock_bh(&icsk->icsk_accept_queue.syn_wait_lock);
1877 if (reqsk_queue_len(&icsk->icsk_accept_queue)) {
1878 start_req:
1879 st->uid = sock_i_uid(sk);
1880 st->syn_wait_sk = sk;
1881 st->state = TCP_SEQ_STATE_OPENREQ;
1882 st->sbucket = 0;
1883 goto get_req;
1884 }
1885 spin_unlock_bh(&icsk->icsk_accept_queue.syn_wait_lock);
1886 }
1887 spin_unlock_bh(&ilb->lock);
1888 st->offset = 0;
1889 if (++st->bucket < INET_LHTABLE_SIZE) {
1890 ilb = &tcp_hashinfo.listening_hash[st->bucket];
1891 spin_lock_bh(&ilb->lock);
1892 sk = sk_nulls_head(&ilb->head);
1893 goto get_sk;
1894 }
1895 cur = NULL;
1896 out:
1897 return cur;
1898 }
1899
1900 static void *listening_get_idx(struct seq_file *seq, loff_t *pos)
1901 {
1902 struct tcp_iter_state *st = seq->private;
1903 void *rc;
1904
1905 st->bucket = 0;
1906 st->offset = 0;
1907 rc = listening_get_next(seq, NULL);
1908
1909 while (rc && *pos) {
1910 rc = listening_get_next(seq, rc);
1911 --*pos;
1912 }
1913 return rc;
1914 }
1915
1916 static inline bool empty_bucket(const struct tcp_iter_state *st)
1917 {
1918 return hlist_nulls_empty(&tcp_hashinfo.ehash[st->bucket].chain);
1919 }
1920
1921 /*
1922 * Get first established socket starting from bucket given in st->bucket.
1923 * If st->bucket is zero, the very first socket in the hash is returned.
1924 */
1925 static void *established_get_first(struct seq_file *seq)
1926 {
1927 struct tcp_iter_state *st = seq->private;
1928 struct net *net = seq_file_net(seq);
1929 void *rc = NULL;
1930
1931 st->offset = 0;
1932 for (; st->bucket <= tcp_hashinfo.ehash_mask; ++st->bucket) {
1933 struct sock *sk;
1934 struct hlist_nulls_node *node;
1935 spinlock_t *lock = inet_ehash_lockp(&tcp_hashinfo, st->bucket);
1936
1937 /* Lockless fast path for the common case of empty buckets */
1938 if (empty_bucket(st))
1939 continue;
1940
1941 spin_lock_bh(lock);
1942 sk_nulls_for_each(sk, node, &tcp_hashinfo.ehash[st->bucket].chain) {
1943 if (sk->sk_family != st->family ||
1944 !net_eq(sock_net(sk), net)) {
1945 continue;
1946 }
1947 rc = sk;
1948 goto out;
1949 }
1950 spin_unlock_bh(lock);
1951 }
1952 out:
1953 return rc;
1954 }
1955
1956 static void *established_get_next(struct seq_file *seq, void *cur)
1957 {
1958 struct sock *sk = cur;
1959 struct hlist_nulls_node *node;
1960 struct tcp_iter_state *st = seq->private;
1961 struct net *net = seq_file_net(seq);
1962
1963 ++st->num;
1964 ++st->offset;
1965
1966 sk = sk_nulls_next(sk);
1967
1968 sk_nulls_for_each_from(sk, node) {
1969 if (sk->sk_family == st->family && net_eq(sock_net(sk), net))
1970 return sk;
1971 }
1972
1973 spin_unlock_bh(inet_ehash_lockp(&tcp_hashinfo, st->bucket));
1974 ++st->bucket;
1975 return established_get_first(seq);
1976 }
1977
1978 static void *established_get_idx(struct seq_file *seq, loff_t pos)
1979 {
1980 struct tcp_iter_state *st = seq->private;
1981 void *rc;
1982
1983 st->bucket = 0;
1984 rc = established_get_first(seq);
1985
1986 while (rc && pos) {
1987 rc = established_get_next(seq, rc);
1988 --pos;
1989 }
1990 return rc;
1991 }
1992
1993 static void *tcp_get_idx(struct seq_file *seq, loff_t pos)
1994 {
1995 void *rc;
1996 struct tcp_iter_state *st = seq->private;
1997
1998 st->state = TCP_SEQ_STATE_LISTENING;
1999 rc = listening_get_idx(seq, &pos);
2000
2001 if (!rc) {
2002 st->state = TCP_SEQ_STATE_ESTABLISHED;
2003 rc = established_get_idx(seq, pos);
2004 }
2005
2006 return rc;
2007 }
2008
2009 static void *tcp_seek_last_pos(struct seq_file *seq)
2010 {
2011 struct tcp_iter_state *st = seq->private;
2012 int offset = st->offset;
2013 int orig_num = st->num;
2014 void *rc = NULL;
2015
2016 switch (st->state) {
2017 case TCP_SEQ_STATE_OPENREQ:
2018 case TCP_SEQ_STATE_LISTENING:
2019 if (st->bucket >= INET_LHTABLE_SIZE)
2020 break;
2021 st->state = TCP_SEQ_STATE_LISTENING;
2022 rc = listening_get_next(seq, NULL);
2023 while (offset-- && rc)
2024 rc = listening_get_next(seq, rc);
2025 if (rc)
2026 break;
2027 st->bucket = 0;
2028 st->state = TCP_SEQ_STATE_ESTABLISHED;
2029 /* Fallthrough */
2030 case TCP_SEQ_STATE_ESTABLISHED:
2031 if (st->bucket > tcp_hashinfo.ehash_mask)
2032 break;
2033 rc = established_get_first(seq);
2034 while (offset-- && rc)
2035 rc = established_get_next(seq, rc);
2036 }
2037
2038 st->num = orig_num;
2039
2040 return rc;
2041 }
2042
2043 static void *tcp_seq_start(struct seq_file *seq, loff_t *pos)
2044 {
2045 struct tcp_iter_state *st = seq->private;
2046 void *rc;
2047
2048 if (*pos && *pos == st->last_pos) {
2049 rc = tcp_seek_last_pos(seq);
2050 if (rc)
2051 goto out;
2052 }
2053
2054 st->state = TCP_SEQ_STATE_LISTENING;
2055 st->num = 0;
2056 st->bucket = 0;
2057 st->offset = 0;
2058 rc = *pos ? tcp_get_idx(seq, *pos - 1) : SEQ_START_TOKEN;
2059
2060 out:
2061 st->last_pos = *pos;
2062 return rc;
2063 }
2064
2065 static void *tcp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2066 {
2067 struct tcp_iter_state *st = seq->private;
2068 void *rc = NULL;
2069
2070 if (v == SEQ_START_TOKEN) {
2071 rc = tcp_get_idx(seq, 0);
2072 goto out;
2073 }
2074
2075 switch (st->state) {
2076 case TCP_SEQ_STATE_OPENREQ:
2077 case TCP_SEQ_STATE_LISTENING:
2078 rc = listening_get_next(seq, v);
2079 if (!rc) {
2080 st->state = TCP_SEQ_STATE_ESTABLISHED;
2081 st->bucket = 0;
2082 st->offset = 0;
2083 rc = established_get_first(seq);
2084 }
2085 break;
2086 case TCP_SEQ_STATE_ESTABLISHED:
2087 rc = established_get_next(seq, v);
2088 break;
2089 }
2090 out:
2091 ++*pos;
2092 st->last_pos = *pos;
2093 return rc;
2094 }
2095
2096 static void tcp_seq_stop(struct seq_file *seq, void *v)
2097 {
2098 struct tcp_iter_state *st = seq->private;
2099
2100 switch (st->state) {
2101 case TCP_SEQ_STATE_OPENREQ:
2102 if (v) {
2103 struct inet_connection_sock *icsk = inet_csk(st->syn_wait_sk);
2104 spin_unlock_bh(&icsk->icsk_accept_queue.syn_wait_lock);
2105 }
2106 case TCP_SEQ_STATE_LISTENING:
2107 if (v != SEQ_START_TOKEN)
2108 spin_unlock_bh(&tcp_hashinfo.listening_hash[st->bucket].lock);
2109 break;
2110 case TCP_SEQ_STATE_ESTABLISHED:
2111 if (v)
2112 spin_unlock_bh(inet_ehash_lockp(&tcp_hashinfo, st->bucket));
2113 break;
2114 }
2115 }
2116
2117 int tcp_seq_open(struct inode *inode, struct file *file)
2118 {
2119 struct tcp_seq_afinfo *afinfo = PDE_DATA(inode);
2120 struct tcp_iter_state *s;
2121 int err;
2122
2123 err = seq_open_net(inode, file, &afinfo->seq_ops,
2124 sizeof(struct tcp_iter_state));
2125 if (err < 0)
2126 return err;
2127
2128 s = ((struct seq_file *)file->private_data)->private;
2129 s->family = afinfo->family;
2130 s->last_pos = 0;
2131 return 0;
2132 }
2133 EXPORT_SYMBOL(tcp_seq_open);
2134
2135 int tcp_proc_register(struct net *net, struct tcp_seq_afinfo *afinfo)
2136 {
2137 int rc = 0;
2138 struct proc_dir_entry *p;
2139
2140 afinfo->seq_ops.start = tcp_seq_start;
2141 afinfo->seq_ops.next = tcp_seq_next;
2142 afinfo->seq_ops.stop = tcp_seq_stop;
2143
2144 p = proc_create_data(afinfo->name, S_IRUGO, net->proc_net,
2145 afinfo->seq_fops, afinfo);
2146 if (!p)
2147 rc = -ENOMEM;
2148 return rc;
2149 }
2150 EXPORT_SYMBOL(tcp_proc_register);
2151
2152 void tcp_proc_unregister(struct net *net, struct tcp_seq_afinfo *afinfo)
2153 {
2154 remove_proc_entry(afinfo->name, net->proc_net);
2155 }
2156 EXPORT_SYMBOL(tcp_proc_unregister);
2157
2158 static void get_openreq4(const struct request_sock *req,
2159 struct seq_file *f, int i, kuid_t uid)
2160 {
2161 const struct inet_request_sock *ireq = inet_rsk(req);
2162 long delta = req->rsk_timer.expires - jiffies;
2163
2164 seq_printf(f, "%4d: %08X:%04X %08X:%04X"
2165 " %02X %08X:%08X %02X:%08lX %08X %5u %8d %u %d %pK",
2166 i,
2167 ireq->ir_loc_addr,
2168 ireq->ir_num,
2169 ireq->ir_rmt_addr,
2170 ntohs(ireq->ir_rmt_port),
2171 TCP_SYN_RECV,
2172 0, 0, /* could print option size, but that is af dependent. */
2173 1, /* timers active (only the expire timer) */
2174 jiffies_delta_to_clock_t(delta),
2175 req->num_timeout,
2176 from_kuid_munged(seq_user_ns(f), uid),
2177 0, /* non standard timer */
2178 0, /* open_requests have no inode */
2179 0,
2180 req);
2181 }
2182
2183 static void get_tcp4_sock(struct sock *sk, struct seq_file *f, int i)
2184 {
2185 int timer_active;
2186 unsigned long timer_expires;
2187 const struct tcp_sock *tp = tcp_sk(sk);
2188 const struct inet_connection_sock *icsk = inet_csk(sk);
2189 const struct inet_sock *inet = inet_sk(sk);
2190 struct fastopen_queue *fastopenq = icsk->icsk_accept_queue.fastopenq;
2191 __be32 dest = inet->inet_daddr;
2192 __be32 src = inet->inet_rcv_saddr;
2193 __u16 destp = ntohs(inet->inet_dport);
2194 __u16 srcp = ntohs(inet->inet_sport);
2195 int rx_queue;
2196
2197 if (icsk->icsk_pending == ICSK_TIME_RETRANS ||
2198 icsk->icsk_pending == ICSK_TIME_EARLY_RETRANS ||
2199 icsk->icsk_pending == ICSK_TIME_LOSS_PROBE) {
2200 timer_active = 1;
2201 timer_expires = icsk->icsk_timeout;
2202 } else if (icsk->icsk_pending == ICSK_TIME_PROBE0) {
2203 timer_active = 4;
2204 timer_expires = icsk->icsk_timeout;
2205 } else if (timer_pending(&sk->sk_timer)) {
2206 timer_active = 2;
2207 timer_expires = sk->sk_timer.expires;
2208 } else {
2209 timer_active = 0;
2210 timer_expires = jiffies;
2211 }
2212
2213 if (sk->sk_state == TCP_LISTEN)
2214 rx_queue = sk->sk_ack_backlog;
2215 else
2216 /*
2217 * because we dont lock socket, we might find a transient negative value
2218 */
2219 rx_queue = max_t(int, tp->rcv_nxt - tp->copied_seq, 0);
2220
2221 seq_printf(f, "%4d: %08X:%04X %08X:%04X %02X %08X:%08X %02X:%08lX "
2222 "%08X %5u %8d %lu %d %pK %lu %lu %u %u %d",
2223 i, src, srcp, dest, destp, sk->sk_state,
2224 tp->write_seq - tp->snd_una,
2225 rx_queue,
2226 timer_active,
2227 jiffies_delta_to_clock_t(timer_expires - jiffies),
2228 icsk->icsk_retransmits,
2229 from_kuid_munged(seq_user_ns(f), sock_i_uid(sk)),
2230 icsk->icsk_probes_out,
2231 sock_i_ino(sk),
2232 atomic_read(&sk->sk_refcnt), sk,
2233 jiffies_to_clock_t(icsk->icsk_rto),
2234 jiffies_to_clock_t(icsk->icsk_ack.ato),
2235 (icsk->icsk_ack.quick << 1) | icsk->icsk_ack.pingpong,
2236 tp->snd_cwnd,
2237 sk->sk_state == TCP_LISTEN ?
2238 (fastopenq ? fastopenq->max_qlen : 0) :
2239 (tcp_in_initial_slowstart(tp) ? -1 : tp->snd_ssthresh));
2240 }
2241
2242 static void get_timewait4_sock(const struct inet_timewait_sock *tw,
2243 struct seq_file *f, int i)
2244 {
2245 __be32 dest, src;
2246 __u16 destp, srcp;
2247 s32 delta = tw->tw_ttd - inet_tw_time_stamp();
2248
2249 dest = tw->tw_daddr;
2250 src = tw->tw_rcv_saddr;
2251 destp = ntohs(tw->tw_dport);
2252 srcp = ntohs(tw->tw_sport);
2253
2254 seq_printf(f, "%4d: %08X:%04X %08X:%04X"
2255 " %02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %pK",
2256 i, src, srcp, dest, destp, tw->tw_substate, 0, 0,
2257 3, jiffies_delta_to_clock_t(delta), 0, 0, 0, 0,
2258 atomic_read(&tw->tw_refcnt), tw);
2259 }
2260
2261 #define TMPSZ 150
2262
2263 static int tcp4_seq_show(struct seq_file *seq, void *v)
2264 {
2265 struct tcp_iter_state *st;
2266 struct sock *sk = v;
2267
2268 seq_setwidth(seq, TMPSZ - 1);
2269 if (v == SEQ_START_TOKEN) {
2270 seq_puts(seq, " sl local_address rem_address st tx_queue "
2271 "rx_queue tr tm->when retrnsmt uid timeout "
2272 "inode");
2273 goto out;
2274 }
2275 st = seq->private;
2276
2277 switch (st->state) {
2278 case TCP_SEQ_STATE_LISTENING:
2279 case TCP_SEQ_STATE_ESTABLISHED:
2280 if (sk->sk_state == TCP_TIME_WAIT)
2281 get_timewait4_sock(v, seq, st->num);
2282 else
2283 get_tcp4_sock(v, seq, st->num);
2284 break;
2285 case TCP_SEQ_STATE_OPENREQ:
2286 get_openreq4(v, seq, st->num, st->uid);
2287 break;
2288 }
2289 out:
2290 seq_pad(seq, '\n');
2291 return 0;
2292 }
2293
2294 static const struct file_operations tcp_afinfo_seq_fops = {
2295 .owner = THIS_MODULE,
2296 .open = tcp_seq_open,
2297 .read = seq_read,
2298 .llseek = seq_lseek,
2299 .release = seq_release_net
2300 };
2301
2302 static struct tcp_seq_afinfo tcp4_seq_afinfo = {
2303 .name = "tcp",
2304 .family = AF_INET,
2305 .seq_fops = &tcp_afinfo_seq_fops,
2306 .seq_ops = {
2307 .show = tcp4_seq_show,
2308 },
2309 };
2310
2311 static int __net_init tcp4_proc_init_net(struct net *net)
2312 {
2313 return tcp_proc_register(net, &tcp4_seq_afinfo);
2314 }
2315
2316 static void __net_exit tcp4_proc_exit_net(struct net *net)
2317 {
2318 tcp_proc_unregister(net, &tcp4_seq_afinfo);
2319 }
2320
2321 static struct pernet_operations tcp4_net_ops = {
2322 .init = tcp4_proc_init_net,
2323 .exit = tcp4_proc_exit_net,
2324 };
2325
2326 int __init tcp4_proc_init(void)
2327 {
2328 return register_pernet_subsys(&tcp4_net_ops);
2329 }
2330
2331 void tcp4_proc_exit(void)
2332 {
2333 unregister_pernet_subsys(&tcp4_net_ops);
2334 }
2335 #endif /* CONFIG_PROC_FS */
2336
2337 struct proto tcp_prot = {
2338 .name = "TCP",
2339 .owner = THIS_MODULE,
2340 .close = tcp_close,
2341 .connect = tcp_v4_connect,
2342 .disconnect = tcp_disconnect,
2343 .accept = inet_csk_accept,
2344 .ioctl = tcp_ioctl,
2345 .init = tcp_v4_init_sock,
2346 .destroy = tcp_v4_destroy_sock,
2347 .shutdown = tcp_shutdown,
2348 .setsockopt = tcp_setsockopt,
2349 .getsockopt = tcp_getsockopt,
2350 .recvmsg = tcp_recvmsg,
2351 .sendmsg = tcp_sendmsg,
2352 .sendpage = tcp_sendpage,
2353 .backlog_rcv = tcp_v4_do_rcv,
2354 .release_cb = tcp_release_cb,
2355 .hash = inet_hash,
2356 .unhash = inet_unhash,
2357 .get_port = inet_csk_get_port,
2358 .enter_memory_pressure = tcp_enter_memory_pressure,
2359 .stream_memory_free = tcp_stream_memory_free,
2360 .sockets_allocated = &tcp_sockets_allocated,
2361 .orphan_count = &tcp_orphan_count,
2362 .memory_allocated = &tcp_memory_allocated,
2363 .memory_pressure = &tcp_memory_pressure,
2364 .sysctl_mem = sysctl_tcp_mem,
2365 .sysctl_wmem = sysctl_tcp_wmem,
2366 .sysctl_rmem = sysctl_tcp_rmem,
2367 .max_header = MAX_TCP_HEADER,
2368 .obj_size = sizeof(struct tcp_sock),
2369 .slab_flags = SLAB_DESTROY_BY_RCU,
2370 .twsk_prot = &tcp_timewait_sock_ops,
2371 .rsk_prot = &tcp_request_sock_ops,
2372 .h.hashinfo = &tcp_hashinfo,
2373 .no_autobind = true,
2374 #ifdef CONFIG_COMPAT
2375 .compat_setsockopt = compat_tcp_setsockopt,
2376 .compat_getsockopt = compat_tcp_getsockopt,
2377 #endif
2378 #ifdef CONFIG_MEMCG_KMEM
2379 .init_cgroup = tcp_init_cgroup,
2380 .destroy_cgroup = tcp_destroy_cgroup,
2381 .proto_cgroup = tcp_proto_cgroup,
2382 #endif
2383 };
2384 EXPORT_SYMBOL(tcp_prot);
2385
2386 static void __net_exit tcp_sk_exit(struct net *net)
2387 {
2388 int cpu;
2389
2390 for_each_possible_cpu(cpu)
2391 inet_ctl_sock_destroy(*per_cpu_ptr(net->ipv4.tcp_sk, cpu));
2392 free_percpu(net->ipv4.tcp_sk);
2393 }
2394
2395 static int __net_init tcp_sk_init(struct net *net)
2396 {
2397 int res, cpu;
2398
2399 net->ipv4.tcp_sk = alloc_percpu(struct sock *);
2400 if (!net->ipv4.tcp_sk)
2401 return -ENOMEM;
2402
2403 for_each_possible_cpu(cpu) {
2404 struct sock *sk;
2405
2406 res = inet_ctl_sock_create(&sk, PF_INET, SOCK_RAW,
2407 IPPROTO_TCP, net);
2408 if (res)
2409 goto fail;
2410 *per_cpu_ptr(net->ipv4.tcp_sk, cpu) = sk;
2411 }
2412 net->ipv4.sysctl_tcp_ecn = 2;
2413 net->ipv4.sysctl_tcp_base_mss = TCP_BASE_MSS;
2414 net->ipv4.sysctl_tcp_probe_threshold = TCP_PROBE_THRESHOLD;
2415 net->ipv4.sysctl_tcp_probe_interval = TCP_PROBE_INTERVAL;
2416 return 0;
2417
2418 fail:
2419 tcp_sk_exit(net);
2420
2421 return res;
2422 }
2423
2424 static void __net_exit tcp_sk_exit_batch(struct list_head *net_exit_list)
2425 {
2426 inet_twsk_purge(&tcp_hashinfo, &tcp_death_row, AF_INET);
2427 }
2428
2429 static struct pernet_operations __net_initdata tcp_sk_ops = {
2430 .init = tcp_sk_init,
2431 .exit = tcp_sk_exit,
2432 .exit_batch = tcp_sk_exit_batch,
2433 };
2434
2435 void __init tcp_v4_init(void)
2436 {
2437 inet_hashinfo_init(&tcp_hashinfo);
2438 if (register_pernet_subsys(&tcp_sk_ops))
2439 panic("Failed to create the TCP control socket.\n");
2440 }
Linux kernel - Deliverables
This page took 0.077823 seconds and 4 git commands to generate.