Loading...
1/*
2 * INET An implementation of the TCP/IP protocol suite for the LINUX
3 * operating system. INET is implemented using the BSD Socket
4 * interface as the means of communication with the user level.
5 *
6 * Implementation of the Transmission Control Protocol(TCP).
7 *
8 * IPv4 specific functions
9 *
10 *
11 * code split from:
12 * linux/ipv4/tcp.c
13 * linux/ipv4/tcp_input.c
14 * linux/ipv4/tcp_output.c
15 *
16 * See tcp.c for author information
17 *
18 * This program is free software; you can redistribute it and/or
19 * modify it under the terms of the GNU General Public License
20 * as published by the Free Software Foundation; either version
21 * 2 of the License, or (at your option) any later version.
22 */
23
24/*
25 * Changes:
26 * David S. Miller : New socket lookup architecture.
27 * This code is dedicated to John Dyson.
28 * David S. Miller : Change semantics of established hash,
29 * half is devoted to TIME_WAIT sockets
30 * and the rest go in the other half.
31 * Andi Kleen : Add support for syncookies and fixed
32 * some bugs: ip options weren't passed to
33 * the TCP layer, missed a check for an
34 * ACK bit.
35 * Andi Kleen : Implemented fast path mtu discovery.
36 * Fixed many serious bugs in the
37 * request_sock handling and moved
38 * most of it into the af independent code.
39 * Added tail drop and some other bugfixes.
40 * Added new listen semantics.
41 * Mike McLagan : Routing by source
42 * Juan Jose Ciarlante: ip_dynaddr bits
43 * Andi Kleen: various fixes.
44 * Vitaly E. Lavrov : Transparent proxy revived after year
45 * coma.
46 * Andi Kleen : Fix new listen.
47 * Andi Kleen : Fix accept error reporting.
48 * YOSHIFUJI Hideaki @USAGI and: Support IPV6_V6ONLY socket option, which
49 * Alexey Kuznetsov allow both IPv4 and IPv6 sockets to bind
50 * a single port at the same time.
51 */
52
53#define pr_fmt(fmt) "TCP: " fmt
54
55#include <linux/bottom_half.h>
56#include <linux/types.h>
57#include <linux/fcntl.h>
58#include <linux/module.h>
59#include <linux/random.h>
60#include <linux/cache.h>
61#include <linux/jhash.h>
62#include <linux/init.h>
63#include <linux/times.h>
64#include <linux/slab.h>
65
66#include <net/net_namespace.h>
67#include <net/icmp.h>
68#include <net/inet_hashtables.h>
69#include <net/tcp.h>
70#include <net/transp_v6.h>
71#include <net/ipv6.h>
72#include <net/inet_common.h>
73#include <net/timewait_sock.h>
74#include <net/xfrm.h>
75#include <net/netdma.h>
76#include <net/secure_seq.h>
77#include <net/tcp_memcontrol.h>
78#include <net/busy_poll.h>
79
80#include <linux/inet.h>
81#include <linux/ipv6.h>
82#include <linux/stddef.h>
83#include <linux/proc_fs.h>
84#include <linux/seq_file.h>
85
86#include <linux/crypto.h>
87#include <linux/scatterlist.h>
88
89int sysctl_tcp_tw_reuse __read_mostly;
90int sysctl_tcp_low_latency __read_mostly;
91EXPORT_SYMBOL(sysctl_tcp_low_latency);
92
93
94#ifdef CONFIG_TCP_MD5SIG
95static int tcp_v4_md5_hash_hdr(char *md5_hash, const struct tcp_md5sig_key *key,
96 __be32 daddr, __be32 saddr, const struct tcphdr *th);
97#endif
98
99struct inet_hashinfo tcp_hashinfo;
100EXPORT_SYMBOL(tcp_hashinfo);
101
102static inline __u32 tcp_v4_init_sequence(const struct sk_buff *skb)
103{
104 return secure_tcp_sequence_number(ip_hdr(skb)->daddr,
105 ip_hdr(skb)->saddr,
106 tcp_hdr(skb)->dest,
107 tcp_hdr(skb)->source);
108}
109
110int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp)
111{
112 const struct tcp_timewait_sock *tcptw = tcp_twsk(sktw);
113 struct tcp_sock *tp = tcp_sk(sk);
114
115 /* With PAWS, it is safe from the viewpoint
116 of data integrity. Even without PAWS it is safe provided sequence
117 spaces do not overlap i.e. at data rates <= 80Mbit/sec.
118
119 Actually, the idea is close to VJ's one, only timestamp cache is
120 held not per host, but per port pair and TW bucket is used as state
121 holder.
122
123 If TW bucket has been already destroyed we fall back to VJ's scheme
124 and use initial timestamp retrieved from peer table.
125 */
126 if (tcptw->tw_ts_recent_stamp &&
127 (twp == NULL || (sysctl_tcp_tw_reuse &&
128 get_seconds() - tcptw->tw_ts_recent_stamp > 1))) {
129 tp->write_seq = tcptw->tw_snd_nxt + 65535 + 2;
130 if (tp->write_seq == 0)
131 tp->write_seq = 1;
132 tp->rx_opt.ts_recent = tcptw->tw_ts_recent;
133 tp->rx_opt.ts_recent_stamp = tcptw->tw_ts_recent_stamp;
134 sock_hold(sktw);
135 return 1;
136 }
137
138 return 0;
139}
140EXPORT_SYMBOL_GPL(tcp_twsk_unique);
141
142/* This will initiate an outgoing connection. */
143int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len)
144{
145 struct sockaddr_in *usin = (struct sockaddr_in *)uaddr;
146 struct inet_sock *inet = inet_sk(sk);
147 struct tcp_sock *tp = tcp_sk(sk);
148 __be16 orig_sport, orig_dport;
149 __be32 daddr, nexthop;
150 struct flowi4 *fl4;
151 struct rtable *rt;
152 int err;
153 struct ip_options_rcu *inet_opt;
154
155 if (addr_len < sizeof(struct sockaddr_in))
156 return -EINVAL;
157
158 if (usin->sin_family != AF_INET)
159 return -EAFNOSUPPORT;
160
161 nexthop = daddr = usin->sin_addr.s_addr;
162 inet_opt = rcu_dereference_protected(inet->inet_opt,
163 sock_owned_by_user(sk));
164 if (inet_opt && inet_opt->opt.srr) {
165 if (!daddr)
166 return -EINVAL;
167 nexthop = inet_opt->opt.faddr;
168 }
169
170 orig_sport = inet->inet_sport;
171 orig_dport = usin->sin_port;
172 fl4 = &inet->cork.fl.u.ip4;
173 rt = ip_route_connect(fl4, nexthop, inet->inet_saddr,
174 RT_CONN_FLAGS(sk), sk->sk_bound_dev_if,
175 IPPROTO_TCP,
176 orig_sport, orig_dport, sk);
177 if (IS_ERR(rt)) {
178 err = PTR_ERR(rt);
179 if (err == -ENETUNREACH)
180 IP_INC_STATS(sock_net(sk), IPSTATS_MIB_OUTNOROUTES);
181 return err;
182 }
183
184 if (rt->rt_flags & (RTCF_MULTICAST | RTCF_BROADCAST)) {
185 ip_rt_put(rt);
186 return -ENETUNREACH;
187 }
188
189 if (!inet_opt || !inet_opt->opt.srr)
190 daddr = fl4->daddr;
191
192 if (!inet->inet_saddr)
193 inet->inet_saddr = fl4->saddr;
194 inet->inet_rcv_saddr = inet->inet_saddr;
195
196 if (tp->rx_opt.ts_recent_stamp && inet->inet_daddr != daddr) {
197 /* Reset inherited state */
198 tp->rx_opt.ts_recent = 0;
199 tp->rx_opt.ts_recent_stamp = 0;
200 if (likely(!tp->repair))
201 tp->write_seq = 0;
202 }
203
204 if (tcp_death_row.sysctl_tw_recycle &&
205 !tp->rx_opt.ts_recent_stamp && fl4->daddr == daddr)
206 tcp_fetch_timewait_stamp(sk, &rt->dst);
207
208 inet->inet_dport = usin->sin_port;
209 inet->inet_daddr = daddr;
210
211 inet_csk(sk)->icsk_ext_hdr_len = 0;
212 if (inet_opt)
213 inet_csk(sk)->icsk_ext_hdr_len = inet_opt->opt.optlen;
214
215 tp->rx_opt.mss_clamp = TCP_MSS_DEFAULT;
216
217 /* Socket identity is still unknown (sport may be zero).
218 * However we set state to SYN-SENT and not releasing socket
219 * lock select source port, enter ourselves into the hash tables and
220 * complete initialization after this.
221 */
222 tcp_set_state(sk, TCP_SYN_SENT);
223 err = inet_hash_connect(&tcp_death_row, sk);
224 if (err)
225 goto failure;
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
254failure:
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}
265EXPORT_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 */
272static 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
304static 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 * This routine is called by the ICMP module when it gets some
314 * sort of error condition. If err < 0 then the socket should
315 * be closed and the error returned to the user. If err > 0
316 * it's just the icmp type << 8 | icmp code. After adjustment
317 * header points to the first 8 bytes of the tcp header. We need
318 * to find the appropriate port.
319 *
320 * The locking strategy used here is very "optimistic". When
321 * someone else accesses the socket the ICMP is just dropped
322 * and for some paths there is no check at all.
323 * A more general error queue to queue errors for later handling
324 * is probably better.
325 *
326 */
327
328void tcp_v4_err(struct sk_buff *icmp_skb, u32 info)
329{
330 const struct iphdr *iph = (const struct iphdr *)icmp_skb->data;
331 struct tcphdr *th = (struct tcphdr *)(icmp_skb->data + (iph->ihl << 2));
332 struct inet_connection_sock *icsk;
333 struct tcp_sock *tp;
334 struct inet_sock *inet;
335 const int type = icmp_hdr(icmp_skb)->type;
336 const int code = icmp_hdr(icmp_skb)->code;
337 struct sock *sk;
338 struct sk_buff *skb;
339 struct request_sock *req;
340 __u32 seq;
341 __u32 remaining;
342 int err;
343 struct net *net = dev_net(icmp_skb->dev);
344
345 if (icmp_skb->len < (iph->ihl << 2) + 8) {
346 ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS);
347 return;
348 }
349
350 sk = inet_lookup(net, &tcp_hashinfo, iph->daddr, th->dest,
351 iph->saddr, th->source, inet_iif(icmp_skb));
352 if (!sk) {
353 ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS);
354 return;
355 }
356 if (sk->sk_state == TCP_TIME_WAIT) {
357 inet_twsk_put(inet_twsk(sk));
358 return;
359 }
360
361 bh_lock_sock(sk);
362 /* If too many ICMPs get dropped on busy
363 * servers this needs to be solved differently.
364 * We do take care of PMTU discovery (RFC1191) special case :
365 * we can receive locally generated ICMP messages while socket is held.
366 */
367 if (sock_owned_by_user(sk)) {
368 if (!(type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED))
369 NET_INC_STATS_BH(net, LINUX_MIB_LOCKDROPPEDICMPS);
370 }
371 if (sk->sk_state == TCP_CLOSE)
372 goto out;
373
374 if (unlikely(iph->ttl < inet_sk(sk)->min_ttl)) {
375 NET_INC_STATS_BH(net, LINUX_MIB_TCPMINTTLDROP);
376 goto out;
377 }
378
379 icsk = inet_csk(sk);
380 tp = tcp_sk(sk);
381 req = tp->fastopen_rsk;
382 seq = ntohl(th->seq);
383 if (sk->sk_state != TCP_LISTEN &&
384 !between(seq, tp->snd_una, tp->snd_nxt) &&
385 (req == NULL || seq != tcp_rsk(req)->snt_isn)) {
386 /* For a Fast Open socket, allow seq to be snt_isn. */
387 NET_INC_STATS_BH(net, LINUX_MIB_OUTOFWINDOWICMPS);
388 goto out;
389 }
390
391 switch (type) {
392 case ICMP_REDIRECT:
393 do_redirect(icmp_skb, sk);
394 goto out;
395 case ICMP_SOURCE_QUENCH:
396 /* Just silently ignore these. */
397 goto out;
398 case ICMP_PARAMETERPROB:
399 err = EPROTO;
400 break;
401 case ICMP_DEST_UNREACH:
402 if (code > NR_ICMP_UNREACH)
403 goto out;
404
405 if (code == ICMP_FRAG_NEEDED) { /* PMTU discovery (RFC1191) */
406 /* We are not interested in TCP_LISTEN and open_requests
407 * (SYN-ACKs send out by Linux are always <576bytes so
408 * they should go through unfragmented).
409 */
410 if (sk->sk_state == TCP_LISTEN)
411 goto out;
412
413 tp->mtu_info = info;
414 if (!sock_owned_by_user(sk)) {
415 tcp_v4_mtu_reduced(sk);
416 } else {
417 if (!test_and_set_bit(TCP_MTU_REDUCED_DEFERRED, &tp->tsq_flags))
418 sock_hold(sk);
419 }
420 goto out;
421 }
422
423 err = icmp_err_convert[code].errno;
424 /* check if icmp_skb allows revert of backoff
425 * (see draft-zimmermann-tcp-lcd) */
426 if (code != ICMP_NET_UNREACH && code != ICMP_HOST_UNREACH)
427 break;
428 if (seq != tp->snd_una || !icsk->icsk_retransmits ||
429 !icsk->icsk_backoff)
430 break;
431
432 /* XXX (TFO) - revisit the following logic for TFO */
433
434 if (sock_owned_by_user(sk))
435 break;
436
437 icsk->icsk_backoff--;
438 inet_csk(sk)->icsk_rto = (tp->srtt_us ? __tcp_set_rto(tp) :
439 TCP_TIMEOUT_INIT) << icsk->icsk_backoff;
440 tcp_bound_rto(sk);
441
442 skb = tcp_write_queue_head(sk);
443 BUG_ON(!skb);
444
445 remaining = icsk->icsk_rto - min(icsk->icsk_rto,
446 tcp_time_stamp - TCP_SKB_CB(skb)->when);
447
448 if (remaining) {
449 inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
450 remaining, TCP_RTO_MAX);
451 } else {
452 /* RTO revert clocked out retransmission.
453 * Will retransmit now */
454 tcp_retransmit_timer(sk);
455 }
456
457 break;
458 case ICMP_TIME_EXCEEDED:
459 err = EHOSTUNREACH;
460 break;
461 default:
462 goto out;
463 }
464
465 /* XXX (TFO) - if it's a TFO socket and has been accepted, rather
466 * than following the TCP_SYN_RECV case and closing the socket,
467 * we ignore the ICMP error and keep trying like a fully established
468 * socket. Is this the right thing to do?
469 */
470 if (req && req->sk == NULL)
471 goto out;
472
473 switch (sk->sk_state) {
474 struct request_sock *req, **prev;
475 case TCP_LISTEN:
476 if (sock_owned_by_user(sk))
477 goto out;
478
479 req = inet_csk_search_req(sk, &prev, th->dest,
480 iph->daddr, iph->saddr);
481 if (!req)
482 goto out;
483
484 /* ICMPs are not backlogged, hence we cannot get
485 an established socket here.
486 */
487 WARN_ON(req->sk);
488
489 if (seq != tcp_rsk(req)->snt_isn) {
490 NET_INC_STATS_BH(net, LINUX_MIB_OUTOFWINDOWICMPS);
491 goto out;
492 }
493
494 /*
495 * Still in SYN_RECV, just remove it silently.
496 * There is no good way to pass the error to the newly
497 * created socket, and POSIX does not want network
498 * errors returned from accept().
499 */
500 inet_csk_reqsk_queue_drop(sk, req, prev);
501 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENDROPS);
502 goto out;
503
504 case TCP_SYN_SENT:
505 case TCP_SYN_RECV: /* Cannot happen.
506 It can f.e. if SYNs crossed,
507 or Fast Open.
508 */
509 if (!sock_owned_by_user(sk)) {
510 sk->sk_err = err;
511
512 sk->sk_error_report(sk);
513
514 tcp_done(sk);
515 } else {
516 sk->sk_err_soft = err;
517 }
518 goto out;
519 }
520
521 /* If we've already connected we will keep trying
522 * until we time out, or the user gives up.
523 *
524 * rfc1122 4.2.3.9 allows to consider as hard errors
525 * only PROTO_UNREACH and PORT_UNREACH (well, FRAG_FAILED too,
526 * but it is obsoleted by pmtu discovery).
527 *
528 * Note, that in modern internet, where routing is unreliable
529 * and in each dark corner broken firewalls sit, sending random
530 * errors ordered by their masters even this two messages finally lose
531 * their original sense (even Linux sends invalid PORT_UNREACHs)
532 *
533 * Now we are in compliance with RFCs.
534 * --ANK (980905)
535 */
536
537 inet = inet_sk(sk);
538 if (!sock_owned_by_user(sk) && inet->recverr) {
539 sk->sk_err = err;
540 sk->sk_error_report(sk);
541 } else { /* Only an error on timeout */
542 sk->sk_err_soft = err;
543 }
544
545out:
546 bh_unlock_sock(sk);
547 sock_put(sk);
548}
549
550void __tcp_v4_send_check(struct sk_buff *skb, __be32 saddr, __be32 daddr)
551{
552 struct tcphdr *th = tcp_hdr(skb);
553
554 if (skb->ip_summed == CHECKSUM_PARTIAL) {
555 th->check = ~tcp_v4_check(skb->len, saddr, daddr, 0);
556 skb->csum_start = skb_transport_header(skb) - skb->head;
557 skb->csum_offset = offsetof(struct tcphdr, check);
558 } else {
559 th->check = tcp_v4_check(skb->len, saddr, daddr,
560 csum_partial(th,
561 th->doff << 2,
562 skb->csum));
563 }
564}
565
566/* This routine computes an IPv4 TCP checksum. */
567void tcp_v4_send_check(struct sock *sk, struct sk_buff *skb)
568{
569 const struct inet_sock *inet = inet_sk(sk);
570
571 __tcp_v4_send_check(skb, inet->inet_saddr, inet->inet_daddr);
572}
573EXPORT_SYMBOL(tcp_v4_send_check);
574
575/*
576 * This routine will send an RST to the other tcp.
577 *
578 * Someone asks: why I NEVER use socket parameters (TOS, TTL etc.)
579 * for reset.
580 * Answer: if a packet caused RST, it is not for a socket
581 * existing in our system, if it is matched to a socket,
582 * it is just duplicate segment or bug in other side's TCP.
583 * So that we build reply only basing on parameters
584 * arrived with segment.
585 * Exception: precedence violation. We do not implement it in any case.
586 */
587
588static void tcp_v4_send_reset(struct sock *sk, struct sk_buff *skb)
589{
590 const struct tcphdr *th = tcp_hdr(skb);
591 struct {
592 struct tcphdr th;
593#ifdef CONFIG_TCP_MD5SIG
594 __be32 opt[(TCPOLEN_MD5SIG_ALIGNED >> 2)];
595#endif
596 } rep;
597 struct ip_reply_arg arg;
598#ifdef CONFIG_TCP_MD5SIG
599 struct tcp_md5sig_key *key;
600 const __u8 *hash_location = NULL;
601 unsigned char newhash[16];
602 int genhash;
603 struct sock *sk1 = NULL;
604#endif
605 struct net *net;
606
607 /* Never send a reset in response to a reset. */
608 if (th->rst)
609 return;
610
611 if (skb_rtable(skb)->rt_type != RTN_LOCAL)
612 return;
613
614 /* Swap the send and the receive. */
615 memset(&rep, 0, sizeof(rep));
616 rep.th.dest = th->source;
617 rep.th.source = th->dest;
618 rep.th.doff = sizeof(struct tcphdr) / 4;
619 rep.th.rst = 1;
620
621 if (th->ack) {
622 rep.th.seq = th->ack_seq;
623 } else {
624 rep.th.ack = 1;
625 rep.th.ack_seq = htonl(ntohl(th->seq) + th->syn + th->fin +
626 skb->len - (th->doff << 2));
627 }
628
629 memset(&arg, 0, sizeof(arg));
630 arg.iov[0].iov_base = (unsigned char *)&rep;
631 arg.iov[0].iov_len = sizeof(rep.th);
632
633#ifdef CONFIG_TCP_MD5SIG
634 hash_location = tcp_parse_md5sig_option(th);
635 if (!sk && hash_location) {
636 /*
637 * active side is lost. Try to find listening socket through
638 * source port, and then find md5 key through listening socket.
639 * we are not loose security here:
640 * Incoming packet is checked with md5 hash with finding key,
641 * no RST generated if md5 hash doesn't match.
642 */
643 sk1 = __inet_lookup_listener(dev_net(skb_dst(skb)->dev),
644 &tcp_hashinfo, ip_hdr(skb)->saddr,
645 th->source, ip_hdr(skb)->daddr,
646 ntohs(th->source), inet_iif(skb));
647 /* don't send rst if it can't find key */
648 if (!sk1)
649 return;
650 rcu_read_lock();
651 key = tcp_md5_do_lookup(sk1, (union tcp_md5_addr *)
652 &ip_hdr(skb)->saddr, AF_INET);
653 if (!key)
654 goto release_sk1;
655
656 genhash = tcp_v4_md5_hash_skb(newhash, key, NULL, NULL, skb);
657 if (genhash || memcmp(hash_location, newhash, 16) != 0)
658 goto release_sk1;
659 } else {
660 key = sk ? tcp_md5_do_lookup(sk, (union tcp_md5_addr *)
661 &ip_hdr(skb)->saddr,
662 AF_INET) : NULL;
663 }
664
665 if (key) {
666 rep.opt[0] = htonl((TCPOPT_NOP << 24) |
667 (TCPOPT_NOP << 16) |
668 (TCPOPT_MD5SIG << 8) |
669 TCPOLEN_MD5SIG);
670 /* Update length and the length the header thinks exists */
671 arg.iov[0].iov_len += TCPOLEN_MD5SIG_ALIGNED;
672 rep.th.doff = arg.iov[0].iov_len / 4;
673
674 tcp_v4_md5_hash_hdr((__u8 *) &rep.opt[1],
675 key, ip_hdr(skb)->saddr,
676 ip_hdr(skb)->daddr, &rep.th);
677 }
678#endif
679 arg.csum = csum_tcpudp_nofold(ip_hdr(skb)->daddr,
680 ip_hdr(skb)->saddr, /* XXX */
681 arg.iov[0].iov_len, IPPROTO_TCP, 0);
682 arg.csumoffset = offsetof(struct tcphdr, check) / 2;
683 arg.flags = (sk && inet_sk(sk)->transparent) ? IP_REPLY_ARG_NOSRCCHECK : 0;
684 /* When socket is gone, all binding information is lost.
685 * routing might fail in this case. No choice here, if we choose to force
686 * input interface, we will misroute in case of asymmetric route.
687 */
688 if (sk)
689 arg.bound_dev_if = sk->sk_bound_dev_if;
690
691 net = dev_net(skb_dst(skb)->dev);
692 arg.tos = ip_hdr(skb)->tos;
693 ip_send_unicast_reply(net, skb, ip_hdr(skb)->saddr,
694 ip_hdr(skb)->daddr, &arg, arg.iov[0].iov_len);
695
696 TCP_INC_STATS_BH(net, TCP_MIB_OUTSEGS);
697 TCP_INC_STATS_BH(net, TCP_MIB_OUTRSTS);
698
699#ifdef CONFIG_TCP_MD5SIG
700release_sk1:
701 if (sk1) {
702 rcu_read_unlock();
703 sock_put(sk1);
704 }
705#endif
706}
707
708/* The code following below sending ACKs in SYN-RECV and TIME-WAIT states
709 outside socket context is ugly, certainly. What can I do?
710 */
711
712static void tcp_v4_send_ack(struct sk_buff *skb, u32 seq, u32 ack,
713 u32 win, u32 tsval, u32 tsecr, int oif,
714 struct tcp_md5sig_key *key,
715 int reply_flags, u8 tos)
716{
717 const struct tcphdr *th = tcp_hdr(skb);
718 struct {
719 struct tcphdr th;
720 __be32 opt[(TCPOLEN_TSTAMP_ALIGNED >> 2)
721#ifdef CONFIG_TCP_MD5SIG
722 + (TCPOLEN_MD5SIG_ALIGNED >> 2)
723#endif
724 ];
725 } rep;
726 struct ip_reply_arg arg;
727 struct net *net = dev_net(skb_dst(skb)->dev);
728
729 memset(&rep.th, 0, sizeof(struct tcphdr));
730 memset(&arg, 0, sizeof(arg));
731
732 arg.iov[0].iov_base = (unsigned char *)&rep;
733 arg.iov[0].iov_len = sizeof(rep.th);
734 if (tsecr) {
735 rep.opt[0] = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
736 (TCPOPT_TIMESTAMP << 8) |
737 TCPOLEN_TIMESTAMP);
738 rep.opt[1] = htonl(tsval);
739 rep.opt[2] = htonl(tsecr);
740 arg.iov[0].iov_len += TCPOLEN_TSTAMP_ALIGNED;
741 }
742
743 /* Swap the send and the receive. */
744 rep.th.dest = th->source;
745 rep.th.source = th->dest;
746 rep.th.doff = arg.iov[0].iov_len / 4;
747 rep.th.seq = htonl(seq);
748 rep.th.ack_seq = htonl(ack);
749 rep.th.ack = 1;
750 rep.th.window = htons(win);
751
752#ifdef CONFIG_TCP_MD5SIG
753 if (key) {
754 int offset = (tsecr) ? 3 : 0;
755
756 rep.opt[offset++] = htonl((TCPOPT_NOP << 24) |
757 (TCPOPT_NOP << 16) |
758 (TCPOPT_MD5SIG << 8) |
759 TCPOLEN_MD5SIG);
760 arg.iov[0].iov_len += TCPOLEN_MD5SIG_ALIGNED;
761 rep.th.doff = arg.iov[0].iov_len/4;
762
763 tcp_v4_md5_hash_hdr((__u8 *) &rep.opt[offset],
764 key, ip_hdr(skb)->saddr,
765 ip_hdr(skb)->daddr, &rep.th);
766 }
767#endif
768 arg.flags = reply_flags;
769 arg.csum = csum_tcpudp_nofold(ip_hdr(skb)->daddr,
770 ip_hdr(skb)->saddr, /* XXX */
771 arg.iov[0].iov_len, IPPROTO_TCP, 0);
772 arg.csumoffset = offsetof(struct tcphdr, check) / 2;
773 if (oif)
774 arg.bound_dev_if = oif;
775 arg.tos = tos;
776 ip_send_unicast_reply(net, skb, ip_hdr(skb)->saddr,
777 ip_hdr(skb)->daddr, &arg, arg.iov[0].iov_len);
778
779 TCP_INC_STATS_BH(net, TCP_MIB_OUTSEGS);
780}
781
782static 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(skb, tcptw->tw_snd_nxt, tcptw->tw_rcv_nxt,
788 tcptw->tw_rcv_wnd >> tw->tw_rcv_wscale,
789 tcp_time_stamp + tcptw->tw_ts_offset,
790 tcptw->tw_ts_recent,
791 tw->tw_bound_dev_if,
792 tcp_twsk_md5_key(tcptw),
793 tw->tw_transparent ? IP_REPLY_ARG_NOSRCCHECK : 0,
794 tw->tw_tos
795 );
796
797 inet_twsk_put(tw);
798}
799
800static void tcp_v4_reqsk_send_ack(struct sock *sk, struct sk_buff *skb,
801 struct request_sock *req)
802{
803 /* sk->sk_state == TCP_LISTEN -> for regular TCP_SYN_RECV
804 * sk->sk_state == TCP_SYN_RECV -> for Fast Open.
805 */
806 tcp_v4_send_ack(skb, (sk->sk_state == TCP_LISTEN) ?
807 tcp_rsk(req)->snt_isn + 1 : tcp_sk(sk)->snd_nxt,
808 tcp_rsk(req)->rcv_nxt, req->rcv_wnd,
809 tcp_time_stamp,
810 req->ts_recent,
811 0,
812 tcp_md5_do_lookup(sk, (union tcp_md5_addr *)&ip_hdr(skb)->daddr,
813 AF_INET),
814 inet_rsk(req)->no_srccheck ? IP_REPLY_ARG_NOSRCCHECK : 0,
815 ip_hdr(skb)->tos);
816}
817
818/*
819 * Send a SYN-ACK after having received a SYN.
820 * This still operates on a request_sock only, not on a big
821 * socket.
822 */
823static int tcp_v4_send_synack(struct sock *sk, struct dst_entry *dst,
824 struct request_sock *req,
825 u16 queue_mapping)
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, NULL);
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 if (!tcp_rsk(req)->snt_synack && !err)
847 tcp_rsk(req)->snt_synack = tcp_time_stamp;
848 }
849
850 return err;
851}
852
853static int tcp_v4_rtx_synack(struct sock *sk, struct request_sock *req)
854{
855 int res = tcp_v4_send_synack(sk, NULL, req, 0);
856
857 if (!res) {
858 TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_RETRANSSEGS);
859 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPSYNRETRANS);
860 }
861 return res;
862}
863
864/*
865 * IPv4 request_sock destructor.
866 */
867static void tcp_v4_reqsk_destructor(struct request_sock *req)
868{
869 kfree(inet_rsk(req)->opt);
870}
871
872/*
873 * Return true if a syncookie should be sent
874 */
875bool tcp_syn_flood_action(struct sock *sk,
876 const struct sk_buff *skb,
877 const char *proto)
878{
879 const char *msg = "Dropping request";
880 bool want_cookie = false;
881 struct listen_sock *lopt;
882
883#ifdef CONFIG_SYN_COOKIES
884 if (sysctl_tcp_syncookies) {
885 msg = "Sending cookies";
886 want_cookie = true;
887 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPREQQFULLDOCOOKIES);
888 } else
889#endif
890 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPREQQFULLDROP);
891
892 lopt = inet_csk(sk)->icsk_accept_queue.listen_opt;
893 if (!lopt->synflood_warned && sysctl_tcp_syncookies != 2) {
894 lopt->synflood_warned = 1;
895 pr_info("%s: Possible SYN flooding on port %d. %s. Check SNMP counters.\n",
896 proto, ntohs(tcp_hdr(skb)->dest), msg);
897 }
898 return want_cookie;
899}
900EXPORT_SYMBOL(tcp_syn_flood_action);
901
902/*
903 * Save and compile IPv4 options into the request_sock if needed.
904 */
905static struct ip_options_rcu *tcp_v4_save_options(struct sk_buff *skb)
906{
907 const struct ip_options *opt = &(IPCB(skb)->opt);
908 struct ip_options_rcu *dopt = NULL;
909
910 if (opt && opt->optlen) {
911 int opt_size = sizeof(*dopt) + opt->optlen;
912
913 dopt = kmalloc(opt_size, GFP_ATOMIC);
914 if (dopt) {
915 if (ip_options_echo(&dopt->opt, skb)) {
916 kfree(dopt);
917 dopt = NULL;
918 }
919 }
920 }
921 return dopt;
922}
923
924#ifdef CONFIG_TCP_MD5SIG
925/*
926 * RFC2385 MD5 checksumming requires a mapping of
927 * IP address->MD5 Key.
928 * We need to maintain these in the sk structure.
929 */
930
931/* Find the Key structure for an address. */
932struct tcp_md5sig_key *tcp_md5_do_lookup(struct sock *sk,
933 const union tcp_md5_addr *addr,
934 int family)
935{
936 struct tcp_sock *tp = tcp_sk(sk);
937 struct tcp_md5sig_key *key;
938 unsigned int size = sizeof(struct in_addr);
939 struct tcp_md5sig_info *md5sig;
940
941 /* caller either holds rcu_read_lock() or socket lock */
942 md5sig = rcu_dereference_check(tp->md5sig_info,
943 sock_owned_by_user(sk) ||
944 lockdep_is_held(&sk->sk_lock.slock));
945 if (!md5sig)
946 return NULL;
947#if IS_ENABLED(CONFIG_IPV6)
948 if (family == AF_INET6)
949 size = sizeof(struct in6_addr);
950#endif
951 hlist_for_each_entry_rcu(key, &md5sig->head, node) {
952 if (key->family != family)
953 continue;
954 if (!memcmp(&key->addr, addr, size))
955 return key;
956 }
957 return NULL;
958}
959EXPORT_SYMBOL(tcp_md5_do_lookup);
960
961struct tcp_md5sig_key *tcp_v4_md5_lookup(struct sock *sk,
962 struct sock *addr_sk)
963{
964 union tcp_md5_addr *addr;
965
966 addr = (union tcp_md5_addr *)&inet_sk(addr_sk)->inet_daddr;
967 return tcp_md5_do_lookup(sk, addr, AF_INET);
968}
969EXPORT_SYMBOL(tcp_v4_md5_lookup);
970
971static struct tcp_md5sig_key *tcp_v4_reqsk_md5_lookup(struct sock *sk,
972 struct request_sock *req)
973{
974 union tcp_md5_addr *addr;
975
976 addr = (union tcp_md5_addr *)&inet_rsk(req)->ir_rmt_addr;
977 return tcp_md5_do_lookup(sk, addr, AF_INET);
978}
979
980/* This can be called on a newly created socket, from other files */
981int tcp_md5_do_add(struct sock *sk, const union tcp_md5_addr *addr,
982 int family, const u8 *newkey, u8 newkeylen, gfp_t gfp)
983{
984 /* Add Key to the list */
985 struct tcp_md5sig_key *key;
986 struct tcp_sock *tp = tcp_sk(sk);
987 struct tcp_md5sig_info *md5sig;
988
989 key = tcp_md5_do_lookup(sk, addr, family);
990 if (key) {
991 /* Pre-existing entry - just update that one. */
992 memcpy(key->key, newkey, newkeylen);
993 key->keylen = newkeylen;
994 return 0;
995 }
996
997 md5sig = rcu_dereference_protected(tp->md5sig_info,
998 sock_owned_by_user(sk));
999 if (!md5sig) {
1000 md5sig = kmalloc(sizeof(*md5sig), gfp);
1001 if (!md5sig)
1002 return -ENOMEM;
1003
1004 sk_nocaps_add(sk, NETIF_F_GSO_MASK);
1005 INIT_HLIST_HEAD(&md5sig->head);
1006 rcu_assign_pointer(tp->md5sig_info, md5sig);
1007 }
1008
1009 key = sock_kmalloc(sk, sizeof(*key), gfp);
1010 if (!key)
1011 return -ENOMEM;
1012 if (!tcp_alloc_md5sig_pool()) {
1013 sock_kfree_s(sk, key, sizeof(*key));
1014 return -ENOMEM;
1015 }
1016
1017 memcpy(key->key, newkey, newkeylen);
1018 key->keylen = newkeylen;
1019 key->family = family;
1020 memcpy(&key->addr, addr,
1021 (family == AF_INET6) ? sizeof(struct in6_addr) :
1022 sizeof(struct in_addr));
1023 hlist_add_head_rcu(&key->node, &md5sig->head);
1024 return 0;
1025}
1026EXPORT_SYMBOL(tcp_md5_do_add);
1027
1028int tcp_md5_do_del(struct sock *sk, const union tcp_md5_addr *addr, int family)
1029{
1030 struct tcp_md5sig_key *key;
1031
1032 key = tcp_md5_do_lookup(sk, addr, family);
1033 if (!key)
1034 return -ENOENT;
1035 hlist_del_rcu(&key->node);
1036 atomic_sub(sizeof(*key), &sk->sk_omem_alloc);
1037 kfree_rcu(key, rcu);
1038 return 0;
1039}
1040EXPORT_SYMBOL(tcp_md5_do_del);
1041
1042static void tcp_clear_md5_list(struct sock *sk)
1043{
1044 struct tcp_sock *tp = tcp_sk(sk);
1045 struct tcp_md5sig_key *key;
1046 struct hlist_node *n;
1047 struct tcp_md5sig_info *md5sig;
1048
1049 md5sig = rcu_dereference_protected(tp->md5sig_info, 1);
1050
1051 hlist_for_each_entry_safe(key, n, &md5sig->head, node) {
1052 hlist_del_rcu(&key->node);
1053 atomic_sub(sizeof(*key), &sk->sk_omem_alloc);
1054 kfree_rcu(key, rcu);
1055 }
1056}
1057
1058static int tcp_v4_parse_md5_keys(struct sock *sk, char __user *optval,
1059 int optlen)
1060{
1061 struct tcp_md5sig cmd;
1062 struct sockaddr_in *sin = (struct sockaddr_in *)&cmd.tcpm_addr;
1063
1064 if (optlen < sizeof(cmd))
1065 return -EINVAL;
1066
1067 if (copy_from_user(&cmd, optval, sizeof(cmd)))
1068 return -EFAULT;
1069
1070 if (sin->sin_family != AF_INET)
1071 return -EINVAL;
1072
1073 if (!cmd.tcpm_key || !cmd.tcpm_keylen)
1074 return tcp_md5_do_del(sk, (union tcp_md5_addr *)&sin->sin_addr.s_addr,
1075 AF_INET);
1076
1077 if (cmd.tcpm_keylen > TCP_MD5SIG_MAXKEYLEN)
1078 return -EINVAL;
1079
1080 return tcp_md5_do_add(sk, (union tcp_md5_addr *)&sin->sin_addr.s_addr,
1081 AF_INET, cmd.tcpm_key, cmd.tcpm_keylen,
1082 GFP_KERNEL);
1083}
1084
1085static int tcp_v4_md5_hash_pseudoheader(struct tcp_md5sig_pool *hp,
1086 __be32 daddr, __be32 saddr, int nbytes)
1087{
1088 struct tcp4_pseudohdr *bp;
1089 struct scatterlist sg;
1090
1091 bp = &hp->md5_blk.ip4;
1092
1093 /*
1094 * 1. the TCP pseudo-header (in the order: source IP address,
1095 * destination IP address, zero-padded protocol number, and
1096 * segment length)
1097 */
1098 bp->saddr = saddr;
1099 bp->daddr = daddr;
1100 bp->pad = 0;
1101 bp->protocol = IPPROTO_TCP;
1102 bp->len = cpu_to_be16(nbytes);
1103
1104 sg_init_one(&sg, bp, sizeof(*bp));
1105 return crypto_hash_update(&hp->md5_desc, &sg, sizeof(*bp));
1106}
1107
1108static int tcp_v4_md5_hash_hdr(char *md5_hash, const struct tcp_md5sig_key *key,
1109 __be32 daddr, __be32 saddr, const struct tcphdr *th)
1110{
1111 struct tcp_md5sig_pool *hp;
1112 struct hash_desc *desc;
1113
1114 hp = tcp_get_md5sig_pool();
1115 if (!hp)
1116 goto clear_hash_noput;
1117 desc = &hp->md5_desc;
1118
1119 if (crypto_hash_init(desc))
1120 goto clear_hash;
1121 if (tcp_v4_md5_hash_pseudoheader(hp, daddr, saddr, th->doff << 2))
1122 goto clear_hash;
1123 if (tcp_md5_hash_header(hp, th))
1124 goto clear_hash;
1125 if (tcp_md5_hash_key(hp, key))
1126 goto clear_hash;
1127 if (crypto_hash_final(desc, md5_hash))
1128 goto clear_hash;
1129
1130 tcp_put_md5sig_pool();
1131 return 0;
1132
1133clear_hash:
1134 tcp_put_md5sig_pool();
1135clear_hash_noput:
1136 memset(md5_hash, 0, 16);
1137 return 1;
1138}
1139
1140int tcp_v4_md5_hash_skb(char *md5_hash, struct tcp_md5sig_key *key,
1141 const struct sock *sk, const struct request_sock *req,
1142 const struct sk_buff *skb)
1143{
1144 struct tcp_md5sig_pool *hp;
1145 struct hash_desc *desc;
1146 const struct tcphdr *th = tcp_hdr(skb);
1147 __be32 saddr, daddr;
1148
1149 if (sk) {
1150 saddr = inet_sk(sk)->inet_saddr;
1151 daddr = inet_sk(sk)->inet_daddr;
1152 } else if (req) {
1153 saddr = inet_rsk(req)->ir_loc_addr;
1154 daddr = inet_rsk(req)->ir_rmt_addr;
1155 } else {
1156 const struct iphdr *iph = ip_hdr(skb);
1157 saddr = iph->saddr;
1158 daddr = iph->daddr;
1159 }
1160
1161 hp = tcp_get_md5sig_pool();
1162 if (!hp)
1163 goto clear_hash_noput;
1164 desc = &hp->md5_desc;
1165
1166 if (crypto_hash_init(desc))
1167 goto clear_hash;
1168
1169 if (tcp_v4_md5_hash_pseudoheader(hp, daddr, saddr, skb->len))
1170 goto clear_hash;
1171 if (tcp_md5_hash_header(hp, th))
1172 goto clear_hash;
1173 if (tcp_md5_hash_skb_data(hp, skb, th->doff << 2))
1174 goto clear_hash;
1175 if (tcp_md5_hash_key(hp, key))
1176 goto clear_hash;
1177 if (crypto_hash_final(desc, md5_hash))
1178 goto clear_hash;
1179
1180 tcp_put_md5sig_pool();
1181 return 0;
1182
1183clear_hash:
1184 tcp_put_md5sig_pool();
1185clear_hash_noput:
1186 memset(md5_hash, 0, 16);
1187 return 1;
1188}
1189EXPORT_SYMBOL(tcp_v4_md5_hash_skb);
1190
1191static bool tcp_v4_inbound_md5_hash(struct sock *sk, const struct sk_buff *skb)
1192{
1193 /*
1194 * This gets called for each TCP segment that arrives
1195 * so we want to be efficient.
1196 * We have 3 drop cases:
1197 * o No MD5 hash and one expected.
1198 * o MD5 hash and we're not expecting one.
1199 * o MD5 hash and its wrong.
1200 */
1201 const __u8 *hash_location = NULL;
1202 struct tcp_md5sig_key *hash_expected;
1203 const struct iphdr *iph = ip_hdr(skb);
1204 const struct tcphdr *th = tcp_hdr(skb);
1205 int genhash;
1206 unsigned char newhash[16];
1207
1208 hash_expected = tcp_md5_do_lookup(sk, (union tcp_md5_addr *)&iph->saddr,
1209 AF_INET);
1210 hash_location = tcp_parse_md5sig_option(th);
1211
1212 /* We've parsed the options - do we have a hash? */
1213 if (!hash_expected && !hash_location)
1214 return false;
1215
1216 if (hash_expected && !hash_location) {
1217 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPMD5NOTFOUND);
1218 return true;
1219 }
1220
1221 if (!hash_expected && hash_location) {
1222 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPMD5UNEXPECTED);
1223 return true;
1224 }
1225
1226 /* Okay, so this is hash_expected and hash_location -
1227 * so we need to calculate the checksum.
1228 */
1229 genhash = tcp_v4_md5_hash_skb(newhash,
1230 hash_expected,
1231 NULL, NULL, skb);
1232
1233 if (genhash || memcmp(hash_location, newhash, 16) != 0) {
1234 net_info_ratelimited("MD5 Hash failed for (%pI4, %d)->(%pI4, %d)%s\n",
1235 &iph->saddr, ntohs(th->source),
1236 &iph->daddr, ntohs(th->dest),
1237 genhash ? " tcp_v4_calc_md5_hash failed"
1238 : "");
1239 return true;
1240 }
1241 return false;
1242}
1243
1244#endif
1245
1246struct request_sock_ops tcp_request_sock_ops __read_mostly = {
1247 .family = PF_INET,
1248 .obj_size = sizeof(struct tcp_request_sock),
1249 .rtx_syn_ack = tcp_v4_rtx_synack,
1250 .send_ack = tcp_v4_reqsk_send_ack,
1251 .destructor = tcp_v4_reqsk_destructor,
1252 .send_reset = tcp_v4_send_reset,
1253 .syn_ack_timeout = tcp_syn_ack_timeout,
1254};
1255
1256#ifdef CONFIG_TCP_MD5SIG
1257static const struct tcp_request_sock_ops tcp_request_sock_ipv4_ops = {
1258 .md5_lookup = tcp_v4_reqsk_md5_lookup,
1259 .calc_md5_hash = tcp_v4_md5_hash_skb,
1260};
1261#endif
1262
1263static bool tcp_fastopen_check(struct sock *sk, struct sk_buff *skb,
1264 struct request_sock *req,
1265 struct tcp_fastopen_cookie *foc,
1266 struct tcp_fastopen_cookie *valid_foc)
1267{
1268 bool skip_cookie = false;
1269 struct fastopen_queue *fastopenq;
1270
1271 if (likely(!fastopen_cookie_present(foc))) {
1272 /* See include/net/tcp.h for the meaning of these knobs */
1273 if ((sysctl_tcp_fastopen & TFO_SERVER_ALWAYS) ||
1274 ((sysctl_tcp_fastopen & TFO_SERVER_COOKIE_NOT_REQD) &&
1275 (TCP_SKB_CB(skb)->end_seq != TCP_SKB_CB(skb)->seq + 1)))
1276 skip_cookie = true; /* no cookie to validate */
1277 else
1278 return false;
1279 }
1280 fastopenq = inet_csk(sk)->icsk_accept_queue.fastopenq;
1281 /* A FO option is present; bump the counter. */
1282 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPFASTOPENPASSIVE);
1283
1284 /* Make sure the listener has enabled fastopen, and we don't
1285 * exceed the max # of pending TFO requests allowed before trying
1286 * to validating the cookie in order to avoid burning CPU cycles
1287 * unnecessarily.
1288 *
1289 * XXX (TFO) - The implication of checking the max_qlen before
1290 * processing a cookie request is that clients can't differentiate
1291 * between qlen overflow causing Fast Open to be disabled
1292 * temporarily vs a server not supporting Fast Open at all.
1293 */
1294 if ((sysctl_tcp_fastopen & TFO_SERVER_ENABLE) == 0 ||
1295 fastopenq == NULL || fastopenq->max_qlen == 0)
1296 return false;
1297
1298 if (fastopenq->qlen >= fastopenq->max_qlen) {
1299 struct request_sock *req1;
1300 spin_lock(&fastopenq->lock);
1301 req1 = fastopenq->rskq_rst_head;
1302 if ((req1 == NULL) || time_after(req1->expires, jiffies)) {
1303 spin_unlock(&fastopenq->lock);
1304 NET_INC_STATS_BH(sock_net(sk),
1305 LINUX_MIB_TCPFASTOPENLISTENOVERFLOW);
1306 /* Avoid bumping LINUX_MIB_TCPFASTOPENPASSIVEFAIL*/
1307 foc->len = -1;
1308 return false;
1309 }
1310 fastopenq->rskq_rst_head = req1->dl_next;
1311 fastopenq->qlen--;
1312 spin_unlock(&fastopenq->lock);
1313 reqsk_free(req1);
1314 }
1315 if (skip_cookie) {
1316 tcp_rsk(req)->rcv_nxt = TCP_SKB_CB(skb)->end_seq;
1317 return true;
1318 }
1319
1320 if (foc->len == TCP_FASTOPEN_COOKIE_SIZE) {
1321 if ((sysctl_tcp_fastopen & TFO_SERVER_COOKIE_NOT_CHKED) == 0) {
1322 tcp_fastopen_cookie_gen(ip_hdr(skb)->saddr,
1323 ip_hdr(skb)->daddr, valid_foc);
1324 if ((valid_foc->len != TCP_FASTOPEN_COOKIE_SIZE) ||
1325 memcmp(&foc->val[0], &valid_foc->val[0],
1326 TCP_FASTOPEN_COOKIE_SIZE) != 0)
1327 return false;
1328 valid_foc->len = -1;
1329 }
1330 /* Acknowledge the data received from the peer. */
1331 tcp_rsk(req)->rcv_nxt = TCP_SKB_CB(skb)->end_seq;
1332 return true;
1333 } else if (foc->len == 0) { /* Client requesting a cookie */
1334 tcp_fastopen_cookie_gen(ip_hdr(skb)->saddr,
1335 ip_hdr(skb)->daddr, valid_foc);
1336 NET_INC_STATS_BH(sock_net(sk),
1337 LINUX_MIB_TCPFASTOPENCOOKIEREQD);
1338 } else {
1339 /* Client sent a cookie with wrong size. Treat it
1340 * the same as invalid and return a valid one.
1341 */
1342 tcp_fastopen_cookie_gen(ip_hdr(skb)->saddr,
1343 ip_hdr(skb)->daddr, valid_foc);
1344 }
1345 return false;
1346}
1347
1348static int tcp_v4_conn_req_fastopen(struct sock *sk,
1349 struct sk_buff *skb,
1350 struct sk_buff *skb_synack,
1351 struct request_sock *req)
1352{
1353 struct tcp_sock *tp = tcp_sk(sk);
1354 struct request_sock_queue *queue = &inet_csk(sk)->icsk_accept_queue;
1355 const struct inet_request_sock *ireq = inet_rsk(req);
1356 struct sock *child;
1357 int err;
1358
1359 req->num_retrans = 0;
1360 req->num_timeout = 0;
1361 req->sk = NULL;
1362
1363 child = inet_csk(sk)->icsk_af_ops->syn_recv_sock(sk, skb, req, NULL);
1364 if (child == NULL) {
1365 NET_INC_STATS_BH(sock_net(sk),
1366 LINUX_MIB_TCPFASTOPENPASSIVEFAIL);
1367 kfree_skb(skb_synack);
1368 return -1;
1369 }
1370 err = ip_build_and_send_pkt(skb_synack, sk, ireq->ir_loc_addr,
1371 ireq->ir_rmt_addr, ireq->opt);
1372 err = net_xmit_eval(err);
1373 if (!err)
1374 tcp_rsk(req)->snt_synack = tcp_time_stamp;
1375 /* XXX (TFO) - is it ok to ignore error and continue? */
1376
1377 spin_lock(&queue->fastopenq->lock);
1378 queue->fastopenq->qlen++;
1379 spin_unlock(&queue->fastopenq->lock);
1380
1381 /* Initialize the child socket. Have to fix some values to take
1382 * into account the child is a Fast Open socket and is created
1383 * only out of the bits carried in the SYN packet.
1384 */
1385 tp = tcp_sk(child);
1386
1387 tp->fastopen_rsk = req;
1388 /* Do a hold on the listner sk so that if the listener is being
1389 * closed, the child that has been accepted can live on and still
1390 * access listen_lock.
1391 */
1392 sock_hold(sk);
1393 tcp_rsk(req)->listener = sk;
1394
1395 /* RFC1323: The window in SYN & SYN/ACK segments is never
1396 * scaled. So correct it appropriately.
1397 */
1398 tp->snd_wnd = ntohs(tcp_hdr(skb)->window);
1399
1400 /* Activate the retrans timer so that SYNACK can be retransmitted.
1401 * The request socket is not added to the SYN table of the parent
1402 * because it's been added to the accept queue directly.
1403 */
1404 inet_csk_reset_xmit_timer(child, ICSK_TIME_RETRANS,
1405 TCP_TIMEOUT_INIT, TCP_RTO_MAX);
1406
1407 /* Add the child socket directly into the accept queue */
1408 inet_csk_reqsk_queue_add(sk, req, child);
1409
1410 /* Now finish processing the fastopen child socket. */
1411 inet_csk(child)->icsk_af_ops->rebuild_header(child);
1412 tcp_init_congestion_control(child);
1413 tcp_mtup_init(child);
1414 tcp_init_metrics(child);
1415 tcp_init_buffer_space(child);
1416
1417 /* Queue the data carried in the SYN packet. We need to first
1418 * bump skb's refcnt because the caller will attempt to free it.
1419 *
1420 * XXX (TFO) - we honor a zero-payload TFO request for now.
1421 * (Any reason not to?)
1422 */
1423 if (TCP_SKB_CB(skb)->end_seq == TCP_SKB_CB(skb)->seq + 1) {
1424 /* Don't queue the skb if there is no payload in SYN.
1425 * XXX (TFO) - How about SYN+FIN?
1426 */
1427 tp->rcv_nxt = TCP_SKB_CB(skb)->end_seq;
1428 } else {
1429 skb = skb_get(skb);
1430 skb_dst_drop(skb);
1431 __skb_pull(skb, tcp_hdr(skb)->doff * 4);
1432 skb_set_owner_r(skb, child);
1433 __skb_queue_tail(&child->sk_receive_queue, skb);
1434 tp->rcv_nxt = TCP_SKB_CB(skb)->end_seq;
1435 tp->syn_data_acked = 1;
1436 }
1437 sk->sk_data_ready(sk);
1438 bh_unlock_sock(child);
1439 sock_put(child);
1440 WARN_ON(req->sk == NULL);
1441 return 0;
1442}
1443
1444int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb)
1445{
1446 struct tcp_options_received tmp_opt;
1447 struct request_sock *req;
1448 struct inet_request_sock *ireq;
1449 struct tcp_sock *tp = tcp_sk(sk);
1450 struct dst_entry *dst = NULL;
1451 __be32 saddr = ip_hdr(skb)->saddr;
1452 __be32 daddr = ip_hdr(skb)->daddr;
1453 __u32 isn = TCP_SKB_CB(skb)->when;
1454 bool want_cookie = false;
1455 struct flowi4 fl4;
1456 struct tcp_fastopen_cookie foc = { .len = -1 };
1457 struct tcp_fastopen_cookie valid_foc = { .len = -1 };
1458 struct sk_buff *skb_synack;
1459 int do_fastopen;
1460
1461 /* Never answer to SYNs send to broadcast or multicast */
1462 if (skb_rtable(skb)->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))
1463 goto drop;
1464
1465 /* TW buckets are converted to open requests without
1466 * limitations, they conserve resources and peer is
1467 * evidently real one.
1468 */
1469 if ((sysctl_tcp_syncookies == 2 ||
1470 inet_csk_reqsk_queue_is_full(sk)) && !isn) {
1471 want_cookie = tcp_syn_flood_action(sk, skb, "TCP");
1472 if (!want_cookie)
1473 goto drop;
1474 }
1475
1476 /* Accept backlog is full. If we have already queued enough
1477 * of warm entries in syn queue, drop request. It is better than
1478 * clogging syn queue with openreqs with exponentially increasing
1479 * timeout.
1480 */
1481 if (sk_acceptq_is_full(sk) && inet_csk_reqsk_queue_young(sk) > 1) {
1482 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENOVERFLOWS);
1483 goto drop;
1484 }
1485
1486 req = inet_reqsk_alloc(&tcp_request_sock_ops);
1487 if (!req)
1488 goto drop;
1489
1490#ifdef CONFIG_TCP_MD5SIG
1491 tcp_rsk(req)->af_specific = &tcp_request_sock_ipv4_ops;
1492#endif
1493
1494 tcp_clear_options(&tmp_opt);
1495 tmp_opt.mss_clamp = TCP_MSS_DEFAULT;
1496 tmp_opt.user_mss = tp->rx_opt.user_mss;
1497 tcp_parse_options(skb, &tmp_opt, 0, want_cookie ? NULL : &foc);
1498
1499 if (want_cookie && !tmp_opt.saw_tstamp)
1500 tcp_clear_options(&tmp_opt);
1501
1502 tmp_opt.tstamp_ok = tmp_opt.saw_tstamp;
1503 tcp_openreq_init(req, &tmp_opt, skb);
1504
1505 ireq = inet_rsk(req);
1506 ireq->ir_loc_addr = daddr;
1507 ireq->ir_rmt_addr = saddr;
1508 ireq->no_srccheck = inet_sk(sk)->transparent;
1509 ireq->opt = tcp_v4_save_options(skb);
1510
1511 if (security_inet_conn_request(sk, skb, req))
1512 goto drop_and_free;
1513
1514 if (!want_cookie || tmp_opt.tstamp_ok)
1515 TCP_ECN_create_request(req, skb, sock_net(sk));
1516
1517 if (want_cookie) {
1518 isn = cookie_v4_init_sequence(sk, skb, &req->mss);
1519 req->cookie_ts = tmp_opt.tstamp_ok;
1520 } else if (!isn) {
1521 /* VJ's idea. We save last timestamp seen
1522 * from the destination in peer table, when entering
1523 * state TIME-WAIT, and check against it before
1524 * accepting new connection request.
1525 *
1526 * If "isn" is not zero, this request hit alive
1527 * timewait bucket, so that all the necessary checks
1528 * are made in the function processing timewait state.
1529 */
1530 if (tmp_opt.saw_tstamp &&
1531 tcp_death_row.sysctl_tw_recycle &&
1532 (dst = inet_csk_route_req(sk, &fl4, req)) != NULL &&
1533 fl4.daddr == saddr) {
1534 if (!tcp_peer_is_proven(req, dst, true)) {
1535 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_PAWSPASSIVEREJECTED);
1536 goto drop_and_release;
1537 }
1538 }
1539 /* Kill the following clause, if you dislike this way. */
1540 else if (!sysctl_tcp_syncookies &&
1541 (sysctl_max_syn_backlog - inet_csk_reqsk_queue_len(sk) <
1542 (sysctl_max_syn_backlog >> 2)) &&
1543 !tcp_peer_is_proven(req, dst, false)) {
1544 /* Without syncookies last quarter of
1545 * backlog is filled with destinations,
1546 * proven to be alive.
1547 * It means that we continue to communicate
1548 * to destinations, already remembered
1549 * to the moment of synflood.
1550 */
1551 LIMIT_NETDEBUG(KERN_DEBUG pr_fmt("drop open request from %pI4/%u\n"),
1552 &saddr, ntohs(tcp_hdr(skb)->source));
1553 goto drop_and_release;
1554 }
1555
1556 isn = tcp_v4_init_sequence(skb);
1557 }
1558 tcp_rsk(req)->snt_isn = isn;
1559
1560 if (dst == NULL) {
1561 dst = inet_csk_route_req(sk, &fl4, req);
1562 if (dst == NULL)
1563 goto drop_and_free;
1564 }
1565 do_fastopen = tcp_fastopen_check(sk, skb, req, &foc, &valid_foc);
1566
1567 /* We don't call tcp_v4_send_synack() directly because we need
1568 * to make sure a child socket can be created successfully before
1569 * sending back synack!
1570 *
1571 * XXX (TFO) - Ideally one would simply call tcp_v4_send_synack()
1572 * (or better yet, call tcp_send_synack() in the child context
1573 * directly, but will have to fix bunch of other code first)
1574 * after syn_recv_sock() except one will need to first fix the
1575 * latter to remove its dependency on the current implementation
1576 * of tcp_v4_send_synack()->tcp_select_initial_window().
1577 */
1578 skb_synack = tcp_make_synack(sk, dst, req,
1579 fastopen_cookie_present(&valid_foc) ? &valid_foc : NULL);
1580
1581 if (skb_synack) {
1582 __tcp_v4_send_check(skb_synack, ireq->ir_loc_addr, ireq->ir_rmt_addr);
1583 skb_set_queue_mapping(skb_synack, skb_get_queue_mapping(skb));
1584 } else
1585 goto drop_and_free;
1586
1587 if (likely(!do_fastopen)) {
1588 int err;
1589 err = ip_build_and_send_pkt(skb_synack, sk, ireq->ir_loc_addr,
1590 ireq->ir_rmt_addr, ireq->opt);
1591 err = net_xmit_eval(err);
1592 if (err || want_cookie)
1593 goto drop_and_free;
1594
1595 tcp_rsk(req)->snt_synack = tcp_time_stamp;
1596 tcp_rsk(req)->listener = NULL;
1597 /* Add the request_sock to the SYN table */
1598 inet_csk_reqsk_queue_hash_add(sk, req, TCP_TIMEOUT_INIT);
1599 if (fastopen_cookie_present(&foc) && foc.len != 0)
1600 NET_INC_STATS_BH(sock_net(sk),
1601 LINUX_MIB_TCPFASTOPENPASSIVEFAIL);
1602 } else if (tcp_v4_conn_req_fastopen(sk, skb, skb_synack, req))
1603 goto drop_and_free;
1604
1605 return 0;
1606
1607drop_and_release:
1608 dst_release(dst);
1609drop_and_free:
1610 reqsk_free(req);
1611drop:
1612 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENDROPS);
1613 return 0;
1614}
1615EXPORT_SYMBOL(tcp_v4_conn_request);
1616
1617
1618/*
1619 * The three way handshake has completed - we got a valid synack -
1620 * now create the new socket.
1621 */
1622struct sock *tcp_v4_syn_recv_sock(struct sock *sk, struct sk_buff *skb,
1623 struct request_sock *req,
1624 struct dst_entry *dst)
1625{
1626 struct inet_request_sock *ireq;
1627 struct inet_sock *newinet;
1628 struct tcp_sock *newtp;
1629 struct sock *newsk;
1630#ifdef CONFIG_TCP_MD5SIG
1631 struct tcp_md5sig_key *key;
1632#endif
1633 struct ip_options_rcu *inet_opt;
1634
1635 if (sk_acceptq_is_full(sk))
1636 goto exit_overflow;
1637
1638 newsk = tcp_create_openreq_child(sk, req, skb);
1639 if (!newsk)
1640 goto exit_nonewsk;
1641
1642 newsk->sk_gso_type = SKB_GSO_TCPV4;
1643 inet_sk_rx_dst_set(newsk, skb);
1644
1645 newtp = tcp_sk(newsk);
1646 newinet = inet_sk(newsk);
1647 ireq = inet_rsk(req);
1648 newinet->inet_daddr = ireq->ir_rmt_addr;
1649 newinet->inet_rcv_saddr = ireq->ir_loc_addr;
1650 newinet->inet_saddr = ireq->ir_loc_addr;
1651 inet_opt = ireq->opt;
1652 rcu_assign_pointer(newinet->inet_opt, inet_opt);
1653 ireq->opt = NULL;
1654 newinet->mc_index = inet_iif(skb);
1655 newinet->mc_ttl = ip_hdr(skb)->ttl;
1656 newinet->rcv_tos = ip_hdr(skb)->tos;
1657 inet_csk(newsk)->icsk_ext_hdr_len = 0;
1658 if (inet_opt)
1659 inet_csk(newsk)->icsk_ext_hdr_len = inet_opt->opt.optlen;
1660 newinet->inet_id = newtp->write_seq ^ jiffies;
1661
1662 if (!dst) {
1663 dst = inet_csk_route_child_sock(sk, newsk, req);
1664 if (!dst)
1665 goto put_and_exit;
1666 } else {
1667 /* syncookie case : see end of cookie_v4_check() */
1668 }
1669 sk_setup_caps(newsk, dst);
1670
1671 tcp_sync_mss(newsk, dst_mtu(dst));
1672 newtp->advmss = dst_metric_advmss(dst);
1673 if (tcp_sk(sk)->rx_opt.user_mss &&
1674 tcp_sk(sk)->rx_opt.user_mss < newtp->advmss)
1675 newtp->advmss = tcp_sk(sk)->rx_opt.user_mss;
1676
1677 tcp_initialize_rcv_mss(newsk);
1678
1679#ifdef CONFIG_TCP_MD5SIG
1680 /* Copy over the MD5 key from the original socket */
1681 key = tcp_md5_do_lookup(sk, (union tcp_md5_addr *)&newinet->inet_daddr,
1682 AF_INET);
1683 if (key != NULL) {
1684 /*
1685 * We're using one, so create a matching key
1686 * on the newsk structure. If we fail to get
1687 * memory, then we end up not copying the key
1688 * across. Shucks.
1689 */
1690 tcp_md5_do_add(newsk, (union tcp_md5_addr *)&newinet->inet_daddr,
1691 AF_INET, key->key, key->keylen, GFP_ATOMIC);
1692 sk_nocaps_add(newsk, NETIF_F_GSO_MASK);
1693 }
1694#endif
1695
1696 if (__inet_inherit_port(sk, newsk) < 0)
1697 goto put_and_exit;
1698 __inet_hash_nolisten(newsk, NULL);
1699
1700 return newsk;
1701
1702exit_overflow:
1703 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENOVERFLOWS);
1704exit_nonewsk:
1705 dst_release(dst);
1706exit:
1707 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENDROPS);
1708 return NULL;
1709put_and_exit:
1710 inet_csk_prepare_forced_close(newsk);
1711 tcp_done(newsk);
1712 goto exit;
1713}
1714EXPORT_SYMBOL(tcp_v4_syn_recv_sock);
1715
1716static struct sock *tcp_v4_hnd_req(struct sock *sk, struct sk_buff *skb)
1717{
1718 struct tcphdr *th = tcp_hdr(skb);
1719 const struct iphdr *iph = ip_hdr(skb);
1720 struct sock *nsk;
1721 struct request_sock **prev;
1722 /* Find possible connection requests. */
1723 struct request_sock *req = inet_csk_search_req(sk, &prev, th->source,
1724 iph->saddr, iph->daddr);
1725 if (req)
1726 return tcp_check_req(sk, skb, req, prev, false);
1727
1728 nsk = inet_lookup_established(sock_net(sk), &tcp_hashinfo, iph->saddr,
1729 th->source, iph->daddr, th->dest, inet_iif(skb));
1730
1731 if (nsk) {
1732 if (nsk->sk_state != TCP_TIME_WAIT) {
1733 bh_lock_sock(nsk);
1734 return nsk;
1735 }
1736 inet_twsk_put(inet_twsk(nsk));
1737 return NULL;
1738 }
1739
1740#ifdef CONFIG_SYN_COOKIES
1741 if (!th->syn)
1742 sk = cookie_v4_check(sk, skb, &(IPCB(skb)->opt));
1743#endif
1744 return sk;
1745}
1746
1747static __sum16 tcp_v4_checksum_init(struct sk_buff *skb)
1748{
1749 const struct iphdr *iph = ip_hdr(skb);
1750
1751 if (skb->ip_summed == CHECKSUM_COMPLETE) {
1752 if (!tcp_v4_check(skb->len, iph->saddr,
1753 iph->daddr, skb->csum)) {
1754 skb->ip_summed = CHECKSUM_UNNECESSARY;
1755 return 0;
1756 }
1757 }
1758
1759 skb->csum = csum_tcpudp_nofold(iph->saddr, iph->daddr,
1760 skb->len, IPPROTO_TCP, 0);
1761
1762 if (skb->len <= 76) {
1763 return __skb_checksum_complete(skb);
1764 }
1765 return 0;
1766}
1767
1768
1769/* The socket must have it's spinlock held when we get
1770 * here.
1771 *
1772 * We have a potential double-lock case here, so even when
1773 * doing backlog processing we use the BH locking scheme.
1774 * This is because we cannot sleep with the original spinlock
1775 * held.
1776 */
1777int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb)
1778{
1779 struct sock *rsk;
1780#ifdef CONFIG_TCP_MD5SIG
1781 /*
1782 * We really want to reject the packet as early as possible
1783 * if:
1784 * o We're expecting an MD5'd packet and this is no MD5 tcp option
1785 * o There is an MD5 option and we're not expecting one
1786 */
1787 if (tcp_v4_inbound_md5_hash(sk, skb))
1788 goto discard;
1789#endif
1790
1791 if (sk->sk_state == TCP_ESTABLISHED) { /* Fast path */
1792 struct dst_entry *dst = sk->sk_rx_dst;
1793
1794 sock_rps_save_rxhash(sk, skb);
1795 if (dst) {
1796 if (inet_sk(sk)->rx_dst_ifindex != skb->skb_iif ||
1797 dst->ops->check(dst, 0) == NULL) {
1798 dst_release(dst);
1799 sk->sk_rx_dst = NULL;
1800 }
1801 }
1802 tcp_rcv_established(sk, skb, tcp_hdr(skb), skb->len);
1803 return 0;
1804 }
1805
1806 if (skb->len < tcp_hdrlen(skb) || tcp_checksum_complete(skb))
1807 goto csum_err;
1808
1809 if (sk->sk_state == TCP_LISTEN) {
1810 struct sock *nsk = tcp_v4_hnd_req(sk, skb);
1811 if (!nsk)
1812 goto discard;
1813
1814 if (nsk != sk) {
1815 sock_rps_save_rxhash(nsk, skb);
1816 if (tcp_child_process(sk, nsk, skb)) {
1817 rsk = nsk;
1818 goto reset;
1819 }
1820 return 0;
1821 }
1822 } else
1823 sock_rps_save_rxhash(sk, skb);
1824
1825 if (tcp_rcv_state_process(sk, skb, tcp_hdr(skb), skb->len)) {
1826 rsk = sk;
1827 goto reset;
1828 }
1829 return 0;
1830
1831reset:
1832 tcp_v4_send_reset(rsk, skb);
1833discard:
1834 kfree_skb(skb);
1835 /* Be careful here. If this function gets more complicated and
1836 * gcc suffers from register pressure on the x86, sk (in %ebx)
1837 * might be destroyed here. This current version compiles correctly,
1838 * but you have been warned.
1839 */
1840 return 0;
1841
1842csum_err:
1843 TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_CSUMERRORS);
1844 TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_INERRS);
1845 goto discard;
1846}
1847EXPORT_SYMBOL(tcp_v4_do_rcv);
1848
1849void tcp_v4_early_demux(struct sk_buff *skb)
1850{
1851 const struct iphdr *iph;
1852 const struct tcphdr *th;
1853 struct sock *sk;
1854
1855 if (skb->pkt_type != PACKET_HOST)
1856 return;
1857
1858 if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct tcphdr)))
1859 return;
1860
1861 iph = ip_hdr(skb);
1862 th = tcp_hdr(skb);
1863
1864 if (th->doff < sizeof(struct tcphdr) / 4)
1865 return;
1866
1867 sk = __inet_lookup_established(dev_net(skb->dev), &tcp_hashinfo,
1868 iph->saddr, th->source,
1869 iph->daddr, ntohs(th->dest),
1870 skb->skb_iif);
1871 if (sk) {
1872 skb->sk = sk;
1873 skb->destructor = sock_edemux;
1874 if (sk->sk_state != TCP_TIME_WAIT) {
1875 struct dst_entry *dst = sk->sk_rx_dst;
1876
1877 if (dst)
1878 dst = dst_check(dst, 0);
1879 if (dst &&
1880 inet_sk(sk)->rx_dst_ifindex == skb->skb_iif)
1881 skb_dst_set_noref(skb, dst);
1882 }
1883 }
1884}
1885
1886/* Packet is added to VJ-style prequeue for processing in process
1887 * context, if a reader task is waiting. Apparently, this exciting
1888 * idea (VJ's mail "Re: query about TCP header on tcp-ip" of 07 Sep 93)
1889 * failed somewhere. Latency? Burstiness? Well, at least now we will
1890 * see, why it failed. 8)8) --ANK
1891 *
1892 */
1893bool tcp_prequeue(struct sock *sk, struct sk_buff *skb)
1894{
1895 struct tcp_sock *tp = tcp_sk(sk);
1896
1897 if (sysctl_tcp_low_latency || !tp->ucopy.task)
1898 return false;
1899
1900 if (skb->len <= tcp_hdrlen(skb) &&
1901 skb_queue_len(&tp->ucopy.prequeue) == 0)
1902 return false;
1903
1904 skb_dst_force(skb);
1905 __skb_queue_tail(&tp->ucopy.prequeue, skb);
1906 tp->ucopy.memory += skb->truesize;
1907 if (tp->ucopy.memory > sk->sk_rcvbuf) {
1908 struct sk_buff *skb1;
1909
1910 BUG_ON(sock_owned_by_user(sk));
1911
1912 while ((skb1 = __skb_dequeue(&tp->ucopy.prequeue)) != NULL) {
1913 sk_backlog_rcv(sk, skb1);
1914 NET_INC_STATS_BH(sock_net(sk),
1915 LINUX_MIB_TCPPREQUEUEDROPPED);
1916 }
1917
1918 tp->ucopy.memory = 0;
1919 } else if (skb_queue_len(&tp->ucopy.prequeue) == 1) {
1920 wake_up_interruptible_sync_poll(sk_sleep(sk),
1921 POLLIN | POLLRDNORM | POLLRDBAND);
1922 if (!inet_csk_ack_scheduled(sk))
1923 inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK,
1924 (3 * tcp_rto_min(sk)) / 4,
1925 TCP_RTO_MAX);
1926 }
1927 return true;
1928}
1929EXPORT_SYMBOL(tcp_prequeue);
1930
1931/*
1932 * From tcp_input.c
1933 */
1934
1935int tcp_v4_rcv(struct sk_buff *skb)
1936{
1937 const struct iphdr *iph;
1938 const struct tcphdr *th;
1939 struct sock *sk;
1940 int ret;
1941 struct net *net = dev_net(skb->dev);
1942
1943 if (skb->pkt_type != PACKET_HOST)
1944 goto discard_it;
1945
1946 /* Count it even if it's bad */
1947 TCP_INC_STATS_BH(net, TCP_MIB_INSEGS);
1948
1949 if (!pskb_may_pull(skb, sizeof(struct tcphdr)))
1950 goto discard_it;
1951
1952 th = tcp_hdr(skb);
1953
1954 if (th->doff < sizeof(struct tcphdr) / 4)
1955 goto bad_packet;
1956 if (!pskb_may_pull(skb, th->doff * 4))
1957 goto discard_it;
1958
1959 /* An explanation is required here, I think.
1960 * Packet length and doff are validated by header prediction,
1961 * provided case of th->doff==0 is eliminated.
1962 * So, we defer the checks. */
1963 if (!skb_csum_unnecessary(skb) && tcp_v4_checksum_init(skb))
1964 goto csum_error;
1965
1966 th = tcp_hdr(skb);
1967 iph = ip_hdr(skb);
1968 TCP_SKB_CB(skb)->seq = ntohl(th->seq);
1969 TCP_SKB_CB(skb)->end_seq = (TCP_SKB_CB(skb)->seq + th->syn + th->fin +
1970 skb->len - th->doff * 4);
1971 TCP_SKB_CB(skb)->ack_seq = ntohl(th->ack_seq);
1972 TCP_SKB_CB(skb)->when = 0;
1973 TCP_SKB_CB(skb)->ip_dsfield = ipv4_get_dsfield(iph);
1974 TCP_SKB_CB(skb)->sacked = 0;
1975
1976 sk = __inet_lookup_skb(&tcp_hashinfo, skb, th->source, th->dest);
1977 if (!sk)
1978 goto no_tcp_socket;
1979
1980process:
1981 if (sk->sk_state == TCP_TIME_WAIT)
1982 goto do_time_wait;
1983
1984 if (unlikely(iph->ttl < inet_sk(sk)->min_ttl)) {
1985 NET_INC_STATS_BH(net, LINUX_MIB_TCPMINTTLDROP);
1986 goto discard_and_relse;
1987 }
1988
1989 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
1990 goto discard_and_relse;
1991 nf_reset(skb);
1992
1993 if (sk_filter(sk, skb))
1994 goto discard_and_relse;
1995
1996 sk_mark_napi_id(sk, skb);
1997 skb->dev = NULL;
1998
1999 bh_lock_sock_nested(sk);
2000 ret = 0;
2001 if (!sock_owned_by_user(sk)) {
2002#ifdef CONFIG_NET_DMA
2003 struct tcp_sock *tp = tcp_sk(sk);
2004 if (!tp->ucopy.dma_chan && tp->ucopy.pinned_list)
2005 tp->ucopy.dma_chan = net_dma_find_channel();
2006 if (tp->ucopy.dma_chan)
2007 ret = tcp_v4_do_rcv(sk, skb);
2008 else
2009#endif
2010 {
2011 if (!tcp_prequeue(sk, skb))
2012 ret = tcp_v4_do_rcv(sk, skb);
2013 }
2014 } else if (unlikely(sk_add_backlog(sk, skb,
2015 sk->sk_rcvbuf + sk->sk_sndbuf))) {
2016 bh_unlock_sock(sk);
2017 NET_INC_STATS_BH(net, LINUX_MIB_TCPBACKLOGDROP);
2018 goto discard_and_relse;
2019 }
2020 bh_unlock_sock(sk);
2021
2022 sock_put(sk);
2023
2024 return ret;
2025
2026no_tcp_socket:
2027 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
2028 goto discard_it;
2029
2030 if (skb->len < (th->doff << 2) || tcp_checksum_complete(skb)) {
2031csum_error:
2032 TCP_INC_STATS_BH(net, TCP_MIB_CSUMERRORS);
2033bad_packet:
2034 TCP_INC_STATS_BH(net, TCP_MIB_INERRS);
2035 } else {
2036 tcp_v4_send_reset(NULL, skb);
2037 }
2038
2039discard_it:
2040 /* Discard frame. */
2041 kfree_skb(skb);
2042 return 0;
2043
2044discard_and_relse:
2045 sock_put(sk);
2046 goto discard_it;
2047
2048do_time_wait:
2049 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) {
2050 inet_twsk_put(inet_twsk(sk));
2051 goto discard_it;
2052 }
2053
2054 if (skb->len < (th->doff << 2)) {
2055 inet_twsk_put(inet_twsk(sk));
2056 goto bad_packet;
2057 }
2058 if (tcp_checksum_complete(skb)) {
2059 inet_twsk_put(inet_twsk(sk));
2060 goto csum_error;
2061 }
2062 switch (tcp_timewait_state_process(inet_twsk(sk), skb, th)) {
2063 case TCP_TW_SYN: {
2064 struct sock *sk2 = inet_lookup_listener(dev_net(skb->dev),
2065 &tcp_hashinfo,
2066 iph->saddr, th->source,
2067 iph->daddr, th->dest,
2068 inet_iif(skb));
2069 if (sk2) {
2070 inet_twsk_deschedule(inet_twsk(sk), &tcp_death_row);
2071 inet_twsk_put(inet_twsk(sk));
2072 sk = sk2;
2073 goto process;
2074 }
2075 /* Fall through to ACK */
2076 }
2077 case TCP_TW_ACK:
2078 tcp_v4_timewait_ack(sk, skb);
2079 break;
2080 case TCP_TW_RST:
2081 goto no_tcp_socket;
2082 case TCP_TW_SUCCESS:;
2083 }
2084 goto discard_it;
2085}
2086
2087static struct timewait_sock_ops tcp_timewait_sock_ops = {
2088 .twsk_obj_size = sizeof(struct tcp_timewait_sock),
2089 .twsk_unique = tcp_twsk_unique,
2090 .twsk_destructor= tcp_twsk_destructor,
2091};
2092
2093void inet_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb)
2094{
2095 struct dst_entry *dst = skb_dst(skb);
2096
2097 dst_hold(dst);
2098 sk->sk_rx_dst = dst;
2099 inet_sk(sk)->rx_dst_ifindex = skb->skb_iif;
2100}
2101EXPORT_SYMBOL(inet_sk_rx_dst_set);
2102
2103const struct inet_connection_sock_af_ops ipv4_specific = {
2104 .queue_xmit = ip_queue_xmit,
2105 .send_check = tcp_v4_send_check,
2106 .rebuild_header = inet_sk_rebuild_header,
2107 .sk_rx_dst_set = inet_sk_rx_dst_set,
2108 .conn_request = tcp_v4_conn_request,
2109 .syn_recv_sock = tcp_v4_syn_recv_sock,
2110 .net_header_len = sizeof(struct iphdr),
2111 .setsockopt = ip_setsockopt,
2112 .getsockopt = ip_getsockopt,
2113 .addr2sockaddr = inet_csk_addr2sockaddr,
2114 .sockaddr_len = sizeof(struct sockaddr_in),
2115 .bind_conflict = inet_csk_bind_conflict,
2116#ifdef CONFIG_COMPAT
2117 .compat_setsockopt = compat_ip_setsockopt,
2118 .compat_getsockopt = compat_ip_getsockopt,
2119#endif
2120};
2121EXPORT_SYMBOL(ipv4_specific);
2122
2123#ifdef CONFIG_TCP_MD5SIG
2124static const struct tcp_sock_af_ops tcp_sock_ipv4_specific = {
2125 .md5_lookup = tcp_v4_md5_lookup,
2126 .calc_md5_hash = tcp_v4_md5_hash_skb,
2127 .md5_parse = tcp_v4_parse_md5_keys,
2128};
2129#endif
2130
2131/* NOTE: A lot of things set to zero explicitly by call to
2132 * sk_alloc() so need not be done here.
2133 */
2134static int tcp_v4_init_sock(struct sock *sk)
2135{
2136 struct inet_connection_sock *icsk = inet_csk(sk);
2137
2138 tcp_init_sock(sk);
2139
2140 icsk->icsk_af_ops = &ipv4_specific;
2141
2142#ifdef CONFIG_TCP_MD5SIG
2143 tcp_sk(sk)->af_specific = &tcp_sock_ipv4_specific;
2144#endif
2145
2146 return 0;
2147}
2148
2149void tcp_v4_destroy_sock(struct sock *sk)
2150{
2151 struct tcp_sock *tp = tcp_sk(sk);
2152
2153 tcp_clear_xmit_timers(sk);
2154
2155 tcp_cleanup_congestion_control(sk);
2156
2157 /* Cleanup up the write buffer. */
2158 tcp_write_queue_purge(sk);
2159
2160 /* Cleans up our, hopefully empty, out_of_order_queue. */
2161 __skb_queue_purge(&tp->out_of_order_queue);
2162
2163#ifdef CONFIG_TCP_MD5SIG
2164 /* Clean up the MD5 key list, if any */
2165 if (tp->md5sig_info) {
2166 tcp_clear_md5_list(sk);
2167 kfree_rcu(tp->md5sig_info, rcu);
2168 tp->md5sig_info = NULL;
2169 }
2170#endif
2171
2172#ifdef CONFIG_NET_DMA
2173 /* Cleans up our sk_async_wait_queue */
2174 __skb_queue_purge(&sk->sk_async_wait_queue);
2175#endif
2176
2177 /* Clean prequeue, it must be empty really */
2178 __skb_queue_purge(&tp->ucopy.prequeue);
2179
2180 /* Clean up a referenced TCP bind bucket. */
2181 if (inet_csk(sk)->icsk_bind_hash)
2182 inet_put_port(sk);
2183
2184 BUG_ON(tp->fastopen_rsk != NULL);
2185
2186 /* If socket is aborted during connect operation */
2187 tcp_free_fastopen_req(tp);
2188
2189 sk_sockets_allocated_dec(sk);
2190 sock_release_memcg(sk);
2191}
2192EXPORT_SYMBOL(tcp_v4_destroy_sock);
2193
2194#ifdef CONFIG_PROC_FS
2195/* Proc filesystem TCP sock list dumping. */
2196
2197/*
2198 * Get next listener socket follow cur. If cur is NULL, get first socket
2199 * starting from bucket given in st->bucket; when st->bucket is zero the
2200 * very first socket in the hash table is returned.
2201 */
2202static void *listening_get_next(struct seq_file *seq, void *cur)
2203{
2204 struct inet_connection_sock *icsk;
2205 struct hlist_nulls_node *node;
2206 struct sock *sk = cur;
2207 struct inet_listen_hashbucket *ilb;
2208 struct tcp_iter_state *st = seq->private;
2209 struct net *net = seq_file_net(seq);
2210
2211 if (!sk) {
2212 ilb = &tcp_hashinfo.listening_hash[st->bucket];
2213 spin_lock_bh(&ilb->lock);
2214 sk = sk_nulls_head(&ilb->head);
2215 st->offset = 0;
2216 goto get_sk;
2217 }
2218 ilb = &tcp_hashinfo.listening_hash[st->bucket];
2219 ++st->num;
2220 ++st->offset;
2221
2222 if (st->state == TCP_SEQ_STATE_OPENREQ) {
2223 struct request_sock *req = cur;
2224
2225 icsk = inet_csk(st->syn_wait_sk);
2226 req = req->dl_next;
2227 while (1) {
2228 while (req) {
2229 if (req->rsk_ops->family == st->family) {
2230 cur = req;
2231 goto out;
2232 }
2233 req = req->dl_next;
2234 }
2235 if (++st->sbucket >= icsk->icsk_accept_queue.listen_opt->nr_table_entries)
2236 break;
2237get_req:
2238 req = icsk->icsk_accept_queue.listen_opt->syn_table[st->sbucket];
2239 }
2240 sk = sk_nulls_next(st->syn_wait_sk);
2241 st->state = TCP_SEQ_STATE_LISTENING;
2242 read_unlock_bh(&icsk->icsk_accept_queue.syn_wait_lock);
2243 } else {
2244 icsk = inet_csk(sk);
2245 read_lock_bh(&icsk->icsk_accept_queue.syn_wait_lock);
2246 if (reqsk_queue_len(&icsk->icsk_accept_queue))
2247 goto start_req;
2248 read_unlock_bh(&icsk->icsk_accept_queue.syn_wait_lock);
2249 sk = sk_nulls_next(sk);
2250 }
2251get_sk:
2252 sk_nulls_for_each_from(sk, node) {
2253 if (!net_eq(sock_net(sk), net))
2254 continue;
2255 if (sk->sk_family == st->family) {
2256 cur = sk;
2257 goto out;
2258 }
2259 icsk = inet_csk(sk);
2260 read_lock_bh(&icsk->icsk_accept_queue.syn_wait_lock);
2261 if (reqsk_queue_len(&icsk->icsk_accept_queue)) {
2262start_req:
2263 st->uid = sock_i_uid(sk);
2264 st->syn_wait_sk = sk;
2265 st->state = TCP_SEQ_STATE_OPENREQ;
2266 st->sbucket = 0;
2267 goto get_req;
2268 }
2269 read_unlock_bh(&icsk->icsk_accept_queue.syn_wait_lock);
2270 }
2271 spin_unlock_bh(&ilb->lock);
2272 st->offset = 0;
2273 if (++st->bucket < INET_LHTABLE_SIZE) {
2274 ilb = &tcp_hashinfo.listening_hash[st->bucket];
2275 spin_lock_bh(&ilb->lock);
2276 sk = sk_nulls_head(&ilb->head);
2277 goto get_sk;
2278 }
2279 cur = NULL;
2280out:
2281 return cur;
2282}
2283
2284static void *listening_get_idx(struct seq_file *seq, loff_t *pos)
2285{
2286 struct tcp_iter_state *st = seq->private;
2287 void *rc;
2288
2289 st->bucket = 0;
2290 st->offset = 0;
2291 rc = listening_get_next(seq, NULL);
2292
2293 while (rc && *pos) {
2294 rc = listening_get_next(seq, rc);
2295 --*pos;
2296 }
2297 return rc;
2298}
2299
2300static inline bool empty_bucket(const struct tcp_iter_state *st)
2301{
2302 return hlist_nulls_empty(&tcp_hashinfo.ehash[st->bucket].chain);
2303}
2304
2305/*
2306 * Get first established socket starting from bucket given in st->bucket.
2307 * If st->bucket is zero, the very first socket in the hash is returned.
2308 */
2309static void *established_get_first(struct seq_file *seq)
2310{
2311 struct tcp_iter_state *st = seq->private;
2312 struct net *net = seq_file_net(seq);
2313 void *rc = NULL;
2314
2315 st->offset = 0;
2316 for (; st->bucket <= tcp_hashinfo.ehash_mask; ++st->bucket) {
2317 struct sock *sk;
2318 struct hlist_nulls_node *node;
2319 spinlock_t *lock = inet_ehash_lockp(&tcp_hashinfo, st->bucket);
2320
2321 /* Lockless fast path for the common case of empty buckets */
2322 if (empty_bucket(st))
2323 continue;
2324
2325 spin_lock_bh(lock);
2326 sk_nulls_for_each(sk, node, &tcp_hashinfo.ehash[st->bucket].chain) {
2327 if (sk->sk_family != st->family ||
2328 !net_eq(sock_net(sk), net)) {
2329 continue;
2330 }
2331 rc = sk;
2332 goto out;
2333 }
2334 spin_unlock_bh(lock);
2335 }
2336out:
2337 return rc;
2338}
2339
2340static void *established_get_next(struct seq_file *seq, void *cur)
2341{
2342 struct sock *sk = cur;
2343 struct hlist_nulls_node *node;
2344 struct tcp_iter_state *st = seq->private;
2345 struct net *net = seq_file_net(seq);
2346
2347 ++st->num;
2348 ++st->offset;
2349
2350 sk = sk_nulls_next(sk);
2351
2352 sk_nulls_for_each_from(sk, node) {
2353 if (sk->sk_family == st->family && net_eq(sock_net(sk), net))
2354 return sk;
2355 }
2356
2357 spin_unlock_bh(inet_ehash_lockp(&tcp_hashinfo, st->bucket));
2358 ++st->bucket;
2359 return established_get_first(seq);
2360}
2361
2362static void *established_get_idx(struct seq_file *seq, loff_t pos)
2363{
2364 struct tcp_iter_state *st = seq->private;
2365 void *rc;
2366
2367 st->bucket = 0;
2368 rc = established_get_first(seq);
2369
2370 while (rc && pos) {
2371 rc = established_get_next(seq, rc);
2372 --pos;
2373 }
2374 return rc;
2375}
2376
2377static void *tcp_get_idx(struct seq_file *seq, loff_t pos)
2378{
2379 void *rc;
2380 struct tcp_iter_state *st = seq->private;
2381
2382 st->state = TCP_SEQ_STATE_LISTENING;
2383 rc = listening_get_idx(seq, &pos);
2384
2385 if (!rc) {
2386 st->state = TCP_SEQ_STATE_ESTABLISHED;
2387 rc = established_get_idx(seq, pos);
2388 }
2389
2390 return rc;
2391}
2392
2393static void *tcp_seek_last_pos(struct seq_file *seq)
2394{
2395 struct tcp_iter_state *st = seq->private;
2396 int offset = st->offset;
2397 int orig_num = st->num;
2398 void *rc = NULL;
2399
2400 switch (st->state) {
2401 case TCP_SEQ_STATE_OPENREQ:
2402 case TCP_SEQ_STATE_LISTENING:
2403 if (st->bucket >= INET_LHTABLE_SIZE)
2404 break;
2405 st->state = TCP_SEQ_STATE_LISTENING;
2406 rc = listening_get_next(seq, NULL);
2407 while (offset-- && rc)
2408 rc = listening_get_next(seq, rc);
2409 if (rc)
2410 break;
2411 st->bucket = 0;
2412 st->state = TCP_SEQ_STATE_ESTABLISHED;
2413 /* Fallthrough */
2414 case TCP_SEQ_STATE_ESTABLISHED:
2415 if (st->bucket > tcp_hashinfo.ehash_mask)
2416 break;
2417 rc = established_get_first(seq);
2418 while (offset-- && rc)
2419 rc = established_get_next(seq, rc);
2420 }
2421
2422 st->num = orig_num;
2423
2424 return rc;
2425}
2426
2427static void *tcp_seq_start(struct seq_file *seq, loff_t *pos)
2428{
2429 struct tcp_iter_state *st = seq->private;
2430 void *rc;
2431
2432 if (*pos && *pos == st->last_pos) {
2433 rc = tcp_seek_last_pos(seq);
2434 if (rc)
2435 goto out;
2436 }
2437
2438 st->state = TCP_SEQ_STATE_LISTENING;
2439 st->num = 0;
2440 st->bucket = 0;
2441 st->offset = 0;
2442 rc = *pos ? tcp_get_idx(seq, *pos - 1) : SEQ_START_TOKEN;
2443
2444out:
2445 st->last_pos = *pos;
2446 return rc;
2447}
2448
2449static void *tcp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2450{
2451 struct tcp_iter_state *st = seq->private;
2452 void *rc = NULL;
2453
2454 if (v == SEQ_START_TOKEN) {
2455 rc = tcp_get_idx(seq, 0);
2456 goto out;
2457 }
2458
2459 switch (st->state) {
2460 case TCP_SEQ_STATE_OPENREQ:
2461 case TCP_SEQ_STATE_LISTENING:
2462 rc = listening_get_next(seq, v);
2463 if (!rc) {
2464 st->state = TCP_SEQ_STATE_ESTABLISHED;
2465 st->bucket = 0;
2466 st->offset = 0;
2467 rc = established_get_first(seq);
2468 }
2469 break;
2470 case TCP_SEQ_STATE_ESTABLISHED:
2471 rc = established_get_next(seq, v);
2472 break;
2473 }
2474out:
2475 ++*pos;
2476 st->last_pos = *pos;
2477 return rc;
2478}
2479
2480static void tcp_seq_stop(struct seq_file *seq, void *v)
2481{
2482 struct tcp_iter_state *st = seq->private;
2483
2484 switch (st->state) {
2485 case TCP_SEQ_STATE_OPENREQ:
2486 if (v) {
2487 struct inet_connection_sock *icsk = inet_csk(st->syn_wait_sk);
2488 read_unlock_bh(&icsk->icsk_accept_queue.syn_wait_lock);
2489 }
2490 case TCP_SEQ_STATE_LISTENING:
2491 if (v != SEQ_START_TOKEN)
2492 spin_unlock_bh(&tcp_hashinfo.listening_hash[st->bucket].lock);
2493 break;
2494 case TCP_SEQ_STATE_ESTABLISHED:
2495 if (v)
2496 spin_unlock_bh(inet_ehash_lockp(&tcp_hashinfo, st->bucket));
2497 break;
2498 }
2499}
2500
2501int tcp_seq_open(struct inode *inode, struct file *file)
2502{
2503 struct tcp_seq_afinfo *afinfo = PDE_DATA(inode);
2504 struct tcp_iter_state *s;
2505 int err;
2506
2507 err = seq_open_net(inode, file, &afinfo->seq_ops,
2508 sizeof(struct tcp_iter_state));
2509 if (err < 0)
2510 return err;
2511
2512 s = ((struct seq_file *)file->private_data)->private;
2513 s->family = afinfo->family;
2514 s->last_pos = 0;
2515 return 0;
2516}
2517EXPORT_SYMBOL(tcp_seq_open);
2518
2519int tcp_proc_register(struct net *net, struct tcp_seq_afinfo *afinfo)
2520{
2521 int rc = 0;
2522 struct proc_dir_entry *p;
2523
2524 afinfo->seq_ops.start = tcp_seq_start;
2525 afinfo->seq_ops.next = tcp_seq_next;
2526 afinfo->seq_ops.stop = tcp_seq_stop;
2527
2528 p = proc_create_data(afinfo->name, S_IRUGO, net->proc_net,
2529 afinfo->seq_fops, afinfo);
2530 if (!p)
2531 rc = -ENOMEM;
2532 return rc;
2533}
2534EXPORT_SYMBOL(tcp_proc_register);
2535
2536void tcp_proc_unregister(struct net *net, struct tcp_seq_afinfo *afinfo)
2537{
2538 remove_proc_entry(afinfo->name, net->proc_net);
2539}
2540EXPORT_SYMBOL(tcp_proc_unregister);
2541
2542static void get_openreq4(const struct sock *sk, const struct request_sock *req,
2543 struct seq_file *f, int i, kuid_t uid)
2544{
2545 const struct inet_request_sock *ireq = inet_rsk(req);
2546 long delta = req->expires - jiffies;
2547
2548 seq_printf(f, "%4d: %08X:%04X %08X:%04X"
2549 " %02X %08X:%08X %02X:%08lX %08X %5u %8d %u %d %pK",
2550 i,
2551 ireq->ir_loc_addr,
2552 ntohs(inet_sk(sk)->inet_sport),
2553 ireq->ir_rmt_addr,
2554 ntohs(ireq->ir_rmt_port),
2555 TCP_SYN_RECV,
2556 0, 0, /* could print option size, but that is af dependent. */
2557 1, /* timers active (only the expire timer) */
2558 jiffies_delta_to_clock_t(delta),
2559 req->num_timeout,
2560 from_kuid_munged(seq_user_ns(f), uid),
2561 0, /* non standard timer */
2562 0, /* open_requests have no inode */
2563 atomic_read(&sk->sk_refcnt),
2564 req);
2565}
2566
2567static void get_tcp4_sock(struct sock *sk, struct seq_file *f, int i)
2568{
2569 int timer_active;
2570 unsigned long timer_expires;
2571 const struct tcp_sock *tp = tcp_sk(sk);
2572 const struct inet_connection_sock *icsk = inet_csk(sk);
2573 const struct inet_sock *inet = inet_sk(sk);
2574 struct fastopen_queue *fastopenq = icsk->icsk_accept_queue.fastopenq;
2575 __be32 dest = inet->inet_daddr;
2576 __be32 src = inet->inet_rcv_saddr;
2577 __u16 destp = ntohs(inet->inet_dport);
2578 __u16 srcp = ntohs(inet->inet_sport);
2579 int rx_queue;
2580
2581 if (icsk->icsk_pending == ICSK_TIME_RETRANS ||
2582 icsk->icsk_pending == ICSK_TIME_EARLY_RETRANS ||
2583 icsk->icsk_pending == ICSK_TIME_LOSS_PROBE) {
2584 timer_active = 1;
2585 timer_expires = icsk->icsk_timeout;
2586 } else if (icsk->icsk_pending == ICSK_TIME_PROBE0) {
2587 timer_active = 4;
2588 timer_expires = icsk->icsk_timeout;
2589 } else if (timer_pending(&sk->sk_timer)) {
2590 timer_active = 2;
2591 timer_expires = sk->sk_timer.expires;
2592 } else {
2593 timer_active = 0;
2594 timer_expires = jiffies;
2595 }
2596
2597 if (sk->sk_state == TCP_LISTEN)
2598 rx_queue = sk->sk_ack_backlog;
2599 else
2600 /*
2601 * because we dont lock socket, we might find a transient negative value
2602 */
2603 rx_queue = max_t(int, tp->rcv_nxt - tp->copied_seq, 0);
2604
2605 seq_printf(f, "%4d: %08X:%04X %08X:%04X %02X %08X:%08X %02X:%08lX "
2606 "%08X %5u %8d %lu %d %pK %lu %lu %u %u %d",
2607 i, src, srcp, dest, destp, sk->sk_state,
2608 tp->write_seq - tp->snd_una,
2609 rx_queue,
2610 timer_active,
2611 jiffies_delta_to_clock_t(timer_expires - jiffies),
2612 icsk->icsk_retransmits,
2613 from_kuid_munged(seq_user_ns(f), sock_i_uid(sk)),
2614 icsk->icsk_probes_out,
2615 sock_i_ino(sk),
2616 atomic_read(&sk->sk_refcnt), sk,
2617 jiffies_to_clock_t(icsk->icsk_rto),
2618 jiffies_to_clock_t(icsk->icsk_ack.ato),
2619 (icsk->icsk_ack.quick << 1) | icsk->icsk_ack.pingpong,
2620 tp->snd_cwnd,
2621 sk->sk_state == TCP_LISTEN ?
2622 (fastopenq ? fastopenq->max_qlen : 0) :
2623 (tcp_in_initial_slowstart(tp) ? -1 : tp->snd_ssthresh));
2624}
2625
2626static void get_timewait4_sock(const struct inet_timewait_sock *tw,
2627 struct seq_file *f, int i)
2628{
2629 __be32 dest, src;
2630 __u16 destp, srcp;
2631 s32 delta = tw->tw_ttd - inet_tw_time_stamp();
2632
2633 dest = tw->tw_daddr;
2634 src = tw->tw_rcv_saddr;
2635 destp = ntohs(tw->tw_dport);
2636 srcp = ntohs(tw->tw_sport);
2637
2638 seq_printf(f, "%4d: %08X:%04X %08X:%04X"
2639 " %02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %pK",
2640 i, src, srcp, dest, destp, tw->tw_substate, 0, 0,
2641 3, jiffies_delta_to_clock_t(delta), 0, 0, 0, 0,
2642 atomic_read(&tw->tw_refcnt), tw);
2643}
2644
2645#define TMPSZ 150
2646
2647static int tcp4_seq_show(struct seq_file *seq, void *v)
2648{
2649 struct tcp_iter_state *st;
2650 struct sock *sk = v;
2651
2652 seq_setwidth(seq, TMPSZ - 1);
2653 if (v == SEQ_START_TOKEN) {
2654 seq_puts(seq, " sl local_address rem_address st tx_queue "
2655 "rx_queue tr tm->when retrnsmt uid timeout "
2656 "inode");
2657 goto out;
2658 }
2659 st = seq->private;
2660
2661 switch (st->state) {
2662 case TCP_SEQ_STATE_LISTENING:
2663 case TCP_SEQ_STATE_ESTABLISHED:
2664 if (sk->sk_state == TCP_TIME_WAIT)
2665 get_timewait4_sock(v, seq, st->num);
2666 else
2667 get_tcp4_sock(v, seq, st->num);
2668 break;
2669 case TCP_SEQ_STATE_OPENREQ:
2670 get_openreq4(st->syn_wait_sk, v, seq, st->num, st->uid);
2671 break;
2672 }
2673out:
2674 seq_pad(seq, '\n');
2675 return 0;
2676}
2677
2678static const struct file_operations tcp_afinfo_seq_fops = {
2679 .owner = THIS_MODULE,
2680 .open = tcp_seq_open,
2681 .read = seq_read,
2682 .llseek = seq_lseek,
2683 .release = seq_release_net
2684};
2685
2686static struct tcp_seq_afinfo tcp4_seq_afinfo = {
2687 .name = "tcp",
2688 .family = AF_INET,
2689 .seq_fops = &tcp_afinfo_seq_fops,
2690 .seq_ops = {
2691 .show = tcp4_seq_show,
2692 },
2693};
2694
2695static int __net_init tcp4_proc_init_net(struct net *net)
2696{
2697 return tcp_proc_register(net, &tcp4_seq_afinfo);
2698}
2699
2700static void __net_exit tcp4_proc_exit_net(struct net *net)
2701{
2702 tcp_proc_unregister(net, &tcp4_seq_afinfo);
2703}
2704
2705static struct pernet_operations tcp4_net_ops = {
2706 .init = tcp4_proc_init_net,
2707 .exit = tcp4_proc_exit_net,
2708};
2709
2710int __init tcp4_proc_init(void)
2711{
2712 return register_pernet_subsys(&tcp4_net_ops);
2713}
2714
2715void tcp4_proc_exit(void)
2716{
2717 unregister_pernet_subsys(&tcp4_net_ops);
2718}
2719#endif /* CONFIG_PROC_FS */
2720
2721struct proto tcp_prot = {
2722 .name = "TCP",
2723 .owner = THIS_MODULE,
2724 .close = tcp_close,
2725 .connect = tcp_v4_connect,
2726 .disconnect = tcp_disconnect,
2727 .accept = inet_csk_accept,
2728 .ioctl = tcp_ioctl,
2729 .init = tcp_v4_init_sock,
2730 .destroy = tcp_v4_destroy_sock,
2731 .shutdown = tcp_shutdown,
2732 .setsockopt = tcp_setsockopt,
2733 .getsockopt = tcp_getsockopt,
2734 .recvmsg = tcp_recvmsg,
2735 .sendmsg = tcp_sendmsg,
2736 .sendpage = tcp_sendpage,
2737 .backlog_rcv = tcp_v4_do_rcv,
2738 .release_cb = tcp_release_cb,
2739 .mtu_reduced = tcp_v4_mtu_reduced,
2740 .hash = inet_hash,
2741 .unhash = inet_unhash,
2742 .get_port = inet_csk_get_port,
2743 .enter_memory_pressure = tcp_enter_memory_pressure,
2744 .stream_memory_free = tcp_stream_memory_free,
2745 .sockets_allocated = &tcp_sockets_allocated,
2746 .orphan_count = &tcp_orphan_count,
2747 .memory_allocated = &tcp_memory_allocated,
2748 .memory_pressure = &tcp_memory_pressure,
2749 .sysctl_mem = sysctl_tcp_mem,
2750 .sysctl_wmem = sysctl_tcp_wmem,
2751 .sysctl_rmem = sysctl_tcp_rmem,
2752 .max_header = MAX_TCP_HEADER,
2753 .obj_size = sizeof(struct tcp_sock),
2754 .slab_flags = SLAB_DESTROY_BY_RCU,
2755 .twsk_prot = &tcp_timewait_sock_ops,
2756 .rsk_prot = &tcp_request_sock_ops,
2757 .h.hashinfo = &tcp_hashinfo,
2758 .no_autobind = true,
2759#ifdef CONFIG_COMPAT
2760 .compat_setsockopt = compat_tcp_setsockopt,
2761 .compat_getsockopt = compat_tcp_getsockopt,
2762#endif
2763#ifdef CONFIG_MEMCG_KMEM
2764 .init_cgroup = tcp_init_cgroup,
2765 .destroy_cgroup = tcp_destroy_cgroup,
2766 .proto_cgroup = tcp_proto_cgroup,
2767#endif
2768};
2769EXPORT_SYMBOL(tcp_prot);
2770
2771static int __net_init tcp_sk_init(struct net *net)
2772{
2773 net->ipv4.sysctl_tcp_ecn = 2;
2774 return 0;
2775}
2776
2777static void __net_exit tcp_sk_exit(struct net *net)
2778{
2779}
2780
2781static void __net_exit tcp_sk_exit_batch(struct list_head *net_exit_list)
2782{
2783 inet_twsk_purge(&tcp_hashinfo, &tcp_death_row, AF_INET);
2784}
2785
2786static struct pernet_operations __net_initdata tcp_sk_ops = {
2787 .init = tcp_sk_init,
2788 .exit = tcp_sk_exit,
2789 .exit_batch = tcp_sk_exit_batch,
2790};
2791
2792void __init tcp_v4_init(void)
2793{
2794 inet_hashinfo_init(&tcp_hashinfo);
2795 if (register_pernet_subsys(&tcp_sk_ops))
2796 panic("Failed to create the TCP control socket.\n");
2797}
1/*
2 * INET An implementation of the TCP/IP protocol suite for the LINUX
3 * operating system. INET is implemented using the BSD Socket
4 * interface as the means of communication with the user level.
5 *
6 * Implementation of the Transmission Control Protocol(TCP).
7 *
8 * IPv4 specific functions
9 *
10 *
11 * code split from:
12 * linux/ipv4/tcp.c
13 * linux/ipv4/tcp_input.c
14 * linux/ipv4/tcp_output.c
15 *
16 * See tcp.c for author information
17 *
18 * This program is free software; you can redistribute it and/or
19 * modify it under the terms of the GNU General Public License
20 * as published by the Free Software Foundation; either version
21 * 2 of the License, or (at your option) any later version.
22 */
23
24/*
25 * Changes:
26 * David S. Miller : New socket lookup architecture.
27 * This code is dedicated to John Dyson.
28 * David S. Miller : Change semantics of established hash,
29 * half is devoted to TIME_WAIT sockets
30 * and the rest go in the other half.
31 * Andi Kleen : Add support for syncookies and fixed
32 * some bugs: ip options weren't passed to
33 * the TCP layer, missed a check for an
34 * ACK bit.
35 * Andi Kleen : Implemented fast path mtu discovery.
36 * Fixed many serious bugs in the
37 * request_sock handling and moved
38 * most of it into the af independent code.
39 * Added tail drop and some other bugfixes.
40 * Added new listen semantics.
41 * Mike McLagan : Routing by source
42 * Juan Jose Ciarlante: ip_dynaddr bits
43 * Andi Kleen: various fixes.
44 * Vitaly E. Lavrov : Transparent proxy revived after year
45 * coma.
46 * Andi Kleen : Fix new listen.
47 * Andi Kleen : Fix accept error reporting.
48 * YOSHIFUJI Hideaki @USAGI and: Support IPV6_V6ONLY socket option, which
49 * Alexey Kuznetsov allow both IPv4 and IPv6 sockets to bind
50 * a single port at the same time.
51 */
52
53#define pr_fmt(fmt) "TCP: " fmt
54
55#include <linux/bottom_half.h>
56#include <linux/types.h>
57#include <linux/fcntl.h>
58#include <linux/module.h>
59#include <linux/random.h>
60#include <linux/cache.h>
61#include <linux/jhash.h>
62#include <linux/init.h>
63#include <linux/times.h>
64#include <linux/slab.h>
65
66#include <net/net_namespace.h>
67#include <net/icmp.h>
68#include <net/inet_hashtables.h>
69#include <net/tcp.h>
70#include <net/transp_v6.h>
71#include <net/ipv6.h>
72#include <net/inet_common.h>
73#include <net/timewait_sock.h>
74#include <net/xfrm.h>
75#include <net/netdma.h>
76#include <net/secure_seq.h>
77#include <net/tcp_memcontrol.h>
78
79#include <linux/inet.h>
80#include <linux/ipv6.h>
81#include <linux/stddef.h>
82#include <linux/proc_fs.h>
83#include <linux/seq_file.h>
84
85#include <linux/crypto.h>
86#include <linux/scatterlist.h>
87
88int sysctl_tcp_tw_reuse __read_mostly;
89int sysctl_tcp_low_latency __read_mostly;
90EXPORT_SYMBOL(sysctl_tcp_low_latency);
91
92
93#ifdef CONFIG_TCP_MD5SIG
94static int tcp_v4_md5_hash_hdr(char *md5_hash, const struct tcp_md5sig_key *key,
95 __be32 daddr, __be32 saddr, const struct tcphdr *th);
96#endif
97
98struct inet_hashinfo tcp_hashinfo;
99EXPORT_SYMBOL(tcp_hashinfo);
100
101static inline __u32 tcp_v4_init_sequence(const struct sk_buff *skb)
102{
103 return secure_tcp_sequence_number(ip_hdr(skb)->daddr,
104 ip_hdr(skb)->saddr,
105 tcp_hdr(skb)->dest,
106 tcp_hdr(skb)->source);
107}
108
109int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp)
110{
111 const struct tcp_timewait_sock *tcptw = tcp_twsk(sktw);
112 struct tcp_sock *tp = tcp_sk(sk);
113
114 /* With PAWS, it is safe from the viewpoint
115 of data integrity. Even without PAWS it is safe provided sequence
116 spaces do not overlap i.e. at data rates <= 80Mbit/sec.
117
118 Actually, the idea is close to VJ's one, only timestamp cache is
119 held not per host, but per port pair and TW bucket is used as state
120 holder.
121
122 If TW bucket has been already destroyed we fall back to VJ's scheme
123 and use initial timestamp retrieved from peer table.
124 */
125 if (tcptw->tw_ts_recent_stamp &&
126 (twp == NULL || (sysctl_tcp_tw_reuse &&
127 get_seconds() - tcptw->tw_ts_recent_stamp > 1))) {
128 tp->write_seq = tcptw->tw_snd_nxt + 65535 + 2;
129 if (tp->write_seq == 0)
130 tp->write_seq = 1;
131 tp->rx_opt.ts_recent = tcptw->tw_ts_recent;
132 tp->rx_opt.ts_recent_stamp = tcptw->tw_ts_recent_stamp;
133 sock_hold(sktw);
134 return 1;
135 }
136
137 return 0;
138}
139EXPORT_SYMBOL_GPL(tcp_twsk_unique);
140
141static int tcp_repair_connect(struct sock *sk)
142{
143 tcp_connect_init(sk);
144 tcp_finish_connect(sk, NULL);
145
146 return 0;
147}
148
149/* This will initiate an outgoing connection. */
150int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len)
151{
152 struct sockaddr_in *usin = (struct sockaddr_in *)uaddr;
153 struct inet_sock *inet = inet_sk(sk);
154 struct tcp_sock *tp = tcp_sk(sk);
155 __be16 orig_sport, orig_dport;
156 __be32 daddr, nexthop;
157 struct flowi4 *fl4;
158 struct rtable *rt;
159 int err;
160 struct ip_options_rcu *inet_opt;
161
162 if (addr_len < sizeof(struct sockaddr_in))
163 return -EINVAL;
164
165 if (usin->sin_family != AF_INET)
166 return -EAFNOSUPPORT;
167
168 nexthop = daddr = usin->sin_addr.s_addr;
169 inet_opt = rcu_dereference_protected(inet->inet_opt,
170 sock_owned_by_user(sk));
171 if (inet_opt && inet_opt->opt.srr) {
172 if (!daddr)
173 return -EINVAL;
174 nexthop = inet_opt->opt.faddr;
175 }
176
177 orig_sport = inet->inet_sport;
178 orig_dport = usin->sin_port;
179 fl4 = &inet->cork.fl.u.ip4;
180 rt = ip_route_connect(fl4, nexthop, inet->inet_saddr,
181 RT_CONN_FLAGS(sk), sk->sk_bound_dev_if,
182 IPPROTO_TCP,
183 orig_sport, orig_dport, sk, true);
184 if (IS_ERR(rt)) {
185 err = PTR_ERR(rt);
186 if (err == -ENETUNREACH)
187 IP_INC_STATS_BH(sock_net(sk), IPSTATS_MIB_OUTNOROUTES);
188 return err;
189 }
190
191 if (rt->rt_flags & (RTCF_MULTICAST | RTCF_BROADCAST)) {
192 ip_rt_put(rt);
193 return -ENETUNREACH;
194 }
195
196 if (!inet_opt || !inet_opt->opt.srr)
197 daddr = fl4->daddr;
198
199 if (!inet->inet_saddr)
200 inet->inet_saddr = fl4->saddr;
201 inet->inet_rcv_saddr = inet->inet_saddr;
202
203 if (tp->rx_opt.ts_recent_stamp && inet->inet_daddr != daddr) {
204 /* Reset inherited state */
205 tp->rx_opt.ts_recent = 0;
206 tp->rx_opt.ts_recent_stamp = 0;
207 if (likely(!tp->repair))
208 tp->write_seq = 0;
209 }
210
211 if (tcp_death_row.sysctl_tw_recycle &&
212 !tp->rx_opt.ts_recent_stamp && fl4->daddr == daddr) {
213 struct inet_peer *peer = rt_get_peer(rt, fl4->daddr);
214 /*
215 * VJ's idea. We save last timestamp seen from
216 * the destination in peer table, when entering state
217 * TIME-WAIT * and initialize rx_opt.ts_recent from it,
218 * when trying new connection.
219 */
220 if (peer) {
221 inet_peer_refcheck(peer);
222 if ((u32)get_seconds() - peer->tcp_ts_stamp <= TCP_PAWS_MSL) {
223 tp->rx_opt.ts_recent_stamp = peer->tcp_ts_stamp;
224 tp->rx_opt.ts_recent = peer->tcp_ts;
225 }
226 }
227 }
228
229 inet->inet_dport = usin->sin_port;
230 inet->inet_daddr = daddr;
231
232 inet_csk(sk)->icsk_ext_hdr_len = 0;
233 if (inet_opt)
234 inet_csk(sk)->icsk_ext_hdr_len = inet_opt->opt.optlen;
235
236 tp->rx_opt.mss_clamp = TCP_MSS_DEFAULT;
237
238 /* Socket identity is still unknown (sport may be zero).
239 * However we set state to SYN-SENT and not releasing socket
240 * lock select source port, enter ourselves into the hash tables and
241 * complete initialization after this.
242 */
243 tcp_set_state(sk, TCP_SYN_SENT);
244 err = inet_hash_connect(&tcp_death_row, sk);
245 if (err)
246 goto failure;
247
248 rt = ip_route_newports(fl4, rt, orig_sport, orig_dport,
249 inet->inet_sport, inet->inet_dport, sk);
250 if (IS_ERR(rt)) {
251 err = PTR_ERR(rt);
252 rt = NULL;
253 goto failure;
254 }
255 /* OK, now commit destination to socket. */
256 sk->sk_gso_type = SKB_GSO_TCPV4;
257 sk_setup_caps(sk, &rt->dst);
258
259 if (!tp->write_seq && likely(!tp->repair))
260 tp->write_seq = secure_tcp_sequence_number(inet->inet_saddr,
261 inet->inet_daddr,
262 inet->inet_sport,
263 usin->sin_port);
264
265 inet->inet_id = tp->write_seq ^ jiffies;
266
267 if (likely(!tp->repair))
268 err = tcp_connect(sk);
269 else
270 err = tcp_repair_connect(sk);
271
272 rt = NULL;
273 if (err)
274 goto failure;
275
276 return 0;
277
278failure:
279 /*
280 * This unhashes the socket and releases the local port,
281 * if necessary.
282 */
283 tcp_set_state(sk, TCP_CLOSE);
284 ip_rt_put(rt);
285 sk->sk_route_caps = 0;
286 inet->inet_dport = 0;
287 return err;
288}
289EXPORT_SYMBOL(tcp_v4_connect);
290
291/*
292 * This routine does path mtu discovery as defined in RFC1191.
293 */
294static void do_pmtu_discovery(struct sock *sk, const struct iphdr *iph, u32 mtu)
295{
296 struct dst_entry *dst;
297 struct inet_sock *inet = inet_sk(sk);
298
299 /* We are not interested in TCP_LISTEN and open_requests (SYN-ACKs
300 * send out by Linux are always <576bytes so they should go through
301 * unfragmented).
302 */
303 if (sk->sk_state == TCP_LISTEN)
304 return;
305
306 /* We don't check in the destentry if pmtu discovery is forbidden
307 * on this route. We just assume that no packet_to_big packets
308 * are send back when pmtu discovery is not active.
309 * There is a small race when the user changes this flag in the
310 * route, but I think that's acceptable.
311 */
312 if ((dst = __sk_dst_check(sk, 0)) == NULL)
313 return;
314
315 dst->ops->update_pmtu(dst, mtu);
316
317 /* Something is about to be wrong... Remember soft error
318 * for the case, if this connection will not able to recover.
319 */
320 if (mtu < dst_mtu(dst) && ip_dont_fragment(sk, dst))
321 sk->sk_err_soft = EMSGSIZE;
322
323 mtu = dst_mtu(dst);
324
325 if (inet->pmtudisc != IP_PMTUDISC_DONT &&
326 inet_csk(sk)->icsk_pmtu_cookie > mtu) {
327 tcp_sync_mss(sk, mtu);
328
329 /* Resend the TCP packet because it's
330 * clear that the old packet has been
331 * dropped. This is the new "fast" path mtu
332 * discovery.
333 */
334 tcp_simple_retransmit(sk);
335 } /* else let the usual retransmit timer handle it */
336}
337
338/*
339 * This routine is called by the ICMP module when it gets some
340 * sort of error condition. If err < 0 then the socket should
341 * be closed and the error returned to the user. If err > 0
342 * it's just the icmp type << 8 | icmp code. After adjustment
343 * header points to the first 8 bytes of the tcp header. We need
344 * to find the appropriate port.
345 *
346 * The locking strategy used here is very "optimistic". When
347 * someone else accesses the socket the ICMP is just dropped
348 * and for some paths there is no check at all.
349 * A more general error queue to queue errors for later handling
350 * is probably better.
351 *
352 */
353
354void tcp_v4_err(struct sk_buff *icmp_skb, u32 info)
355{
356 const struct iphdr *iph = (const struct iphdr *)icmp_skb->data;
357 struct tcphdr *th = (struct tcphdr *)(icmp_skb->data + (iph->ihl << 2));
358 struct inet_connection_sock *icsk;
359 struct tcp_sock *tp;
360 struct inet_sock *inet;
361 const int type = icmp_hdr(icmp_skb)->type;
362 const int code = icmp_hdr(icmp_skb)->code;
363 struct sock *sk;
364 struct sk_buff *skb;
365 __u32 seq;
366 __u32 remaining;
367 int err;
368 struct net *net = dev_net(icmp_skb->dev);
369
370 if (icmp_skb->len < (iph->ihl << 2) + 8) {
371 ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS);
372 return;
373 }
374
375 sk = inet_lookup(net, &tcp_hashinfo, iph->daddr, th->dest,
376 iph->saddr, th->source, 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
386 bh_lock_sock(sk);
387 /* If too many ICMPs get dropped on busy
388 * servers this needs to be solved differently.
389 */
390 if (sock_owned_by_user(sk))
391 NET_INC_STATS_BH(net, LINUX_MIB_LOCKDROPPEDICMPS);
392
393 if (sk->sk_state == TCP_CLOSE)
394 goto out;
395
396 if (unlikely(iph->ttl < inet_sk(sk)->min_ttl)) {
397 NET_INC_STATS_BH(net, LINUX_MIB_TCPMINTTLDROP);
398 goto out;
399 }
400
401 icsk = inet_csk(sk);
402 tp = tcp_sk(sk);
403 seq = ntohl(th->seq);
404 if (sk->sk_state != TCP_LISTEN &&
405 !between(seq, tp->snd_una, tp->snd_nxt)) {
406 NET_INC_STATS_BH(net, LINUX_MIB_OUTOFWINDOWICMPS);
407 goto out;
408 }
409
410 switch (type) {
411 case ICMP_SOURCE_QUENCH:
412 /* Just silently ignore these. */
413 goto out;
414 case ICMP_PARAMETERPROB:
415 err = EPROTO;
416 break;
417 case ICMP_DEST_UNREACH:
418 if (code > NR_ICMP_UNREACH)
419 goto out;
420
421 if (code == ICMP_FRAG_NEEDED) { /* PMTU discovery (RFC1191) */
422 if (!sock_owned_by_user(sk))
423 do_pmtu_discovery(sk, iph, info);
424 goto out;
425 }
426
427 err = icmp_err_convert[code].errno;
428 /* check if icmp_skb allows revert of backoff
429 * (see draft-zimmermann-tcp-lcd) */
430 if (code != ICMP_NET_UNREACH && code != ICMP_HOST_UNREACH)
431 break;
432 if (seq != tp->snd_una || !icsk->icsk_retransmits ||
433 !icsk->icsk_backoff)
434 break;
435
436 if (sock_owned_by_user(sk))
437 break;
438
439 icsk->icsk_backoff--;
440 inet_csk(sk)->icsk_rto = (tp->srtt ? __tcp_set_rto(tp) :
441 TCP_TIMEOUT_INIT) << icsk->icsk_backoff;
442 tcp_bound_rto(sk);
443
444 skb = tcp_write_queue_head(sk);
445 BUG_ON(!skb);
446
447 remaining = icsk->icsk_rto - min(icsk->icsk_rto,
448 tcp_time_stamp - TCP_SKB_CB(skb)->when);
449
450 if (remaining) {
451 inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
452 remaining, TCP_RTO_MAX);
453 } else {
454 /* RTO revert clocked out retransmission.
455 * Will retransmit now */
456 tcp_retransmit_timer(sk);
457 }
458
459 break;
460 case ICMP_TIME_EXCEEDED:
461 err = EHOSTUNREACH;
462 break;
463 default:
464 goto out;
465 }
466
467 switch (sk->sk_state) {
468 struct request_sock *req, **prev;
469 case TCP_LISTEN:
470 if (sock_owned_by_user(sk))
471 goto out;
472
473 req = inet_csk_search_req(sk, &prev, th->dest,
474 iph->daddr, iph->saddr);
475 if (!req)
476 goto out;
477
478 /* ICMPs are not backlogged, hence we cannot get
479 an established socket here.
480 */
481 WARN_ON(req->sk);
482
483 if (seq != tcp_rsk(req)->snt_isn) {
484 NET_INC_STATS_BH(net, LINUX_MIB_OUTOFWINDOWICMPS);
485 goto out;
486 }
487
488 /*
489 * Still in SYN_RECV, just remove it silently.
490 * There is no good way to pass the error to the newly
491 * created socket, and POSIX does not want network
492 * errors returned from accept().
493 */
494 inet_csk_reqsk_queue_drop(sk, req, prev);
495 goto out;
496
497 case TCP_SYN_SENT:
498 case TCP_SYN_RECV: /* Cannot happen.
499 It can f.e. if SYNs crossed.
500 */
501 if (!sock_owned_by_user(sk)) {
502 sk->sk_err = err;
503
504 sk->sk_error_report(sk);
505
506 tcp_done(sk);
507 } else {
508 sk->sk_err_soft = err;
509 }
510 goto out;
511 }
512
513 /* If we've already connected we will keep trying
514 * until we time out, or the user gives up.
515 *
516 * rfc1122 4.2.3.9 allows to consider as hard errors
517 * only PROTO_UNREACH and PORT_UNREACH (well, FRAG_FAILED too,
518 * but it is obsoleted by pmtu discovery).
519 *
520 * Note, that in modern internet, where routing is unreliable
521 * and in each dark corner broken firewalls sit, sending random
522 * errors ordered by their masters even this two messages finally lose
523 * their original sense (even Linux sends invalid PORT_UNREACHs)
524 *
525 * Now we are in compliance with RFCs.
526 * --ANK (980905)
527 */
528
529 inet = inet_sk(sk);
530 if (!sock_owned_by_user(sk) && inet->recverr) {
531 sk->sk_err = err;
532 sk->sk_error_report(sk);
533 } else { /* Only an error on timeout */
534 sk->sk_err_soft = err;
535 }
536
537out:
538 bh_unlock_sock(sk);
539 sock_put(sk);
540}
541
542static void __tcp_v4_send_check(struct sk_buff *skb,
543 __be32 saddr, __be32 daddr)
544{
545 struct tcphdr *th = tcp_hdr(skb);
546
547 if (skb->ip_summed == CHECKSUM_PARTIAL) {
548 th->check = ~tcp_v4_check(skb->len, saddr, daddr, 0);
549 skb->csum_start = skb_transport_header(skb) - skb->head;
550 skb->csum_offset = offsetof(struct tcphdr, check);
551 } else {
552 th->check = tcp_v4_check(skb->len, saddr, daddr,
553 csum_partial(th,
554 th->doff << 2,
555 skb->csum));
556 }
557}
558
559/* This routine computes an IPv4 TCP checksum. */
560void tcp_v4_send_check(struct sock *sk, struct sk_buff *skb)
561{
562 const struct inet_sock *inet = inet_sk(sk);
563
564 __tcp_v4_send_check(skb, inet->inet_saddr, inet->inet_daddr);
565}
566EXPORT_SYMBOL(tcp_v4_send_check);
567
568int tcp_v4_gso_send_check(struct sk_buff *skb)
569{
570 const struct iphdr *iph;
571 struct tcphdr *th;
572
573 if (!pskb_may_pull(skb, sizeof(*th)))
574 return -EINVAL;
575
576 iph = ip_hdr(skb);
577 th = tcp_hdr(skb);
578
579 th->check = 0;
580 skb->ip_summed = CHECKSUM_PARTIAL;
581 __tcp_v4_send_check(skb, iph->saddr, iph->daddr);
582 return 0;
583}
584
585/*
586 * This routine will send an RST to the other tcp.
587 *
588 * Someone asks: why I NEVER use socket parameters (TOS, TTL etc.)
589 * for reset.
590 * Answer: if a packet caused RST, it is not for a socket
591 * existing in our system, if it is matched to a socket,
592 * it is just duplicate segment or bug in other side's TCP.
593 * So that we build reply only basing on parameters
594 * arrived with segment.
595 * Exception: precedence violation. We do not implement it in any case.
596 */
597
598static void tcp_v4_send_reset(struct sock *sk, struct sk_buff *skb)
599{
600 const struct tcphdr *th = tcp_hdr(skb);
601 struct {
602 struct tcphdr th;
603#ifdef CONFIG_TCP_MD5SIG
604 __be32 opt[(TCPOLEN_MD5SIG_ALIGNED >> 2)];
605#endif
606 } rep;
607 struct ip_reply_arg arg;
608#ifdef CONFIG_TCP_MD5SIG
609 struct tcp_md5sig_key *key;
610 const __u8 *hash_location = NULL;
611 unsigned char newhash[16];
612 int genhash;
613 struct sock *sk1 = NULL;
614#endif
615 struct net *net;
616
617 /* Never send a reset in response to a reset. */
618 if (th->rst)
619 return;
620
621 if (skb_rtable(skb)->rt_type != RTN_LOCAL)
622 return;
623
624 /* Swap the send and the receive. */
625 memset(&rep, 0, sizeof(rep));
626 rep.th.dest = th->source;
627 rep.th.source = th->dest;
628 rep.th.doff = sizeof(struct tcphdr) / 4;
629 rep.th.rst = 1;
630
631 if (th->ack) {
632 rep.th.seq = th->ack_seq;
633 } else {
634 rep.th.ack = 1;
635 rep.th.ack_seq = htonl(ntohl(th->seq) + th->syn + th->fin +
636 skb->len - (th->doff << 2));
637 }
638
639 memset(&arg, 0, sizeof(arg));
640 arg.iov[0].iov_base = (unsigned char *)&rep;
641 arg.iov[0].iov_len = sizeof(rep.th);
642
643#ifdef CONFIG_TCP_MD5SIG
644 hash_location = tcp_parse_md5sig_option(th);
645 if (!sk && hash_location) {
646 /*
647 * active side is lost. Try to find listening socket through
648 * source port, and then find md5 key through listening socket.
649 * we are not loose security here:
650 * Incoming packet is checked with md5 hash with finding key,
651 * no RST generated if md5 hash doesn't match.
652 */
653 sk1 = __inet_lookup_listener(dev_net(skb_dst(skb)->dev),
654 &tcp_hashinfo, ip_hdr(skb)->daddr,
655 ntohs(th->source), inet_iif(skb));
656 /* don't send rst if it can't find key */
657 if (!sk1)
658 return;
659 rcu_read_lock();
660 key = tcp_md5_do_lookup(sk1, (union tcp_md5_addr *)
661 &ip_hdr(skb)->saddr, AF_INET);
662 if (!key)
663 goto release_sk1;
664
665 genhash = tcp_v4_md5_hash_skb(newhash, key, NULL, NULL, skb);
666 if (genhash || memcmp(hash_location, newhash, 16) != 0)
667 goto release_sk1;
668 } else {
669 key = sk ? tcp_md5_do_lookup(sk, (union tcp_md5_addr *)
670 &ip_hdr(skb)->saddr,
671 AF_INET) : NULL;
672 }
673
674 if (key) {
675 rep.opt[0] = htonl((TCPOPT_NOP << 24) |
676 (TCPOPT_NOP << 16) |
677 (TCPOPT_MD5SIG << 8) |
678 TCPOLEN_MD5SIG);
679 /* Update length and the length the header thinks exists */
680 arg.iov[0].iov_len += TCPOLEN_MD5SIG_ALIGNED;
681 rep.th.doff = arg.iov[0].iov_len / 4;
682
683 tcp_v4_md5_hash_hdr((__u8 *) &rep.opt[1],
684 key, ip_hdr(skb)->saddr,
685 ip_hdr(skb)->daddr, &rep.th);
686 }
687#endif
688 arg.csum = csum_tcpudp_nofold(ip_hdr(skb)->daddr,
689 ip_hdr(skb)->saddr, /* XXX */
690 arg.iov[0].iov_len, IPPROTO_TCP, 0);
691 arg.csumoffset = offsetof(struct tcphdr, check) / 2;
692 arg.flags = (sk && inet_sk(sk)->transparent) ? IP_REPLY_ARG_NOSRCCHECK : 0;
693 /* When socket is gone, all binding information is lost.
694 * routing might fail in this case. using iif for oif to
695 * make sure we can deliver it
696 */
697 arg.bound_dev_if = sk ? sk->sk_bound_dev_if : inet_iif(skb);
698
699 net = dev_net(skb_dst(skb)->dev);
700 arg.tos = ip_hdr(skb)->tos;
701 ip_send_reply(net->ipv4.tcp_sock, skb, ip_hdr(skb)->saddr,
702 &arg, arg.iov[0].iov_len);
703
704 TCP_INC_STATS_BH(net, TCP_MIB_OUTSEGS);
705 TCP_INC_STATS_BH(net, TCP_MIB_OUTRSTS);
706
707#ifdef CONFIG_TCP_MD5SIG
708release_sk1:
709 if (sk1) {
710 rcu_read_unlock();
711 sock_put(sk1);
712 }
713#endif
714}
715
716/* The code following below sending ACKs in SYN-RECV and TIME-WAIT states
717 outside socket context is ugly, certainly. What can I do?
718 */
719
720static void tcp_v4_send_ack(struct sk_buff *skb, u32 seq, u32 ack,
721 u32 win, u32 ts, int oif,
722 struct tcp_md5sig_key *key,
723 int reply_flags, u8 tos)
724{
725 const struct tcphdr *th = tcp_hdr(skb);
726 struct {
727 struct tcphdr th;
728 __be32 opt[(TCPOLEN_TSTAMP_ALIGNED >> 2)
729#ifdef CONFIG_TCP_MD5SIG
730 + (TCPOLEN_MD5SIG_ALIGNED >> 2)
731#endif
732 ];
733 } rep;
734 struct ip_reply_arg arg;
735 struct net *net = dev_net(skb_dst(skb)->dev);
736
737 memset(&rep.th, 0, sizeof(struct tcphdr));
738 memset(&arg, 0, sizeof(arg));
739
740 arg.iov[0].iov_base = (unsigned char *)&rep;
741 arg.iov[0].iov_len = sizeof(rep.th);
742 if (ts) {
743 rep.opt[0] = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
744 (TCPOPT_TIMESTAMP << 8) |
745 TCPOLEN_TIMESTAMP);
746 rep.opt[1] = htonl(tcp_time_stamp);
747 rep.opt[2] = htonl(ts);
748 arg.iov[0].iov_len += TCPOLEN_TSTAMP_ALIGNED;
749 }
750
751 /* Swap the send and the receive. */
752 rep.th.dest = th->source;
753 rep.th.source = th->dest;
754 rep.th.doff = arg.iov[0].iov_len / 4;
755 rep.th.seq = htonl(seq);
756 rep.th.ack_seq = htonl(ack);
757 rep.th.ack = 1;
758 rep.th.window = htons(win);
759
760#ifdef CONFIG_TCP_MD5SIG
761 if (key) {
762 int offset = (ts) ? 3 : 0;
763
764 rep.opt[offset++] = htonl((TCPOPT_NOP << 24) |
765 (TCPOPT_NOP << 16) |
766 (TCPOPT_MD5SIG << 8) |
767 TCPOLEN_MD5SIG);
768 arg.iov[0].iov_len += TCPOLEN_MD5SIG_ALIGNED;
769 rep.th.doff = arg.iov[0].iov_len/4;
770
771 tcp_v4_md5_hash_hdr((__u8 *) &rep.opt[offset],
772 key, ip_hdr(skb)->saddr,
773 ip_hdr(skb)->daddr, &rep.th);
774 }
775#endif
776 arg.flags = reply_flags;
777 arg.csum = csum_tcpudp_nofold(ip_hdr(skb)->daddr,
778 ip_hdr(skb)->saddr, /* XXX */
779 arg.iov[0].iov_len, IPPROTO_TCP, 0);
780 arg.csumoffset = offsetof(struct tcphdr, check) / 2;
781 if (oif)
782 arg.bound_dev_if = oif;
783 arg.tos = tos;
784 ip_send_reply(net->ipv4.tcp_sock, skb, ip_hdr(skb)->saddr,
785 &arg, arg.iov[0].iov_len);
786
787 TCP_INC_STATS_BH(net, TCP_MIB_OUTSEGS);
788}
789
790static void tcp_v4_timewait_ack(struct sock *sk, struct sk_buff *skb)
791{
792 struct inet_timewait_sock *tw = inet_twsk(sk);
793 struct tcp_timewait_sock *tcptw = tcp_twsk(sk);
794
795 tcp_v4_send_ack(skb, tcptw->tw_snd_nxt, tcptw->tw_rcv_nxt,
796 tcptw->tw_rcv_wnd >> tw->tw_rcv_wscale,
797 tcptw->tw_ts_recent,
798 tw->tw_bound_dev_if,
799 tcp_twsk_md5_key(tcptw),
800 tw->tw_transparent ? IP_REPLY_ARG_NOSRCCHECK : 0,
801 tw->tw_tos
802 );
803
804 inet_twsk_put(tw);
805}
806
807static void tcp_v4_reqsk_send_ack(struct sock *sk, struct sk_buff *skb,
808 struct request_sock *req)
809{
810 tcp_v4_send_ack(skb, tcp_rsk(req)->snt_isn + 1,
811 tcp_rsk(req)->rcv_isn + 1, req->rcv_wnd,
812 req->ts_recent,
813 0,
814 tcp_md5_do_lookup(sk, (union tcp_md5_addr *)&ip_hdr(skb)->daddr,
815 AF_INET),
816 inet_rsk(req)->no_srccheck ? IP_REPLY_ARG_NOSRCCHECK : 0,
817 ip_hdr(skb)->tos);
818}
819
820/*
821 * Send a SYN-ACK after having received a SYN.
822 * This still operates on a request_sock only, not on a big
823 * socket.
824 */
825static int tcp_v4_send_synack(struct sock *sk, struct dst_entry *dst,
826 struct request_sock *req,
827 struct request_values *rvp,
828 u16 queue_mapping)
829{
830 const struct inet_request_sock *ireq = inet_rsk(req);
831 struct flowi4 fl4;
832 int err = -1;
833 struct sk_buff * skb;
834
835 /* First, grab a route. */
836 if (!dst && (dst = inet_csk_route_req(sk, &fl4, req)) == NULL)
837 return -1;
838
839 skb = tcp_make_synack(sk, dst, req, rvp);
840
841 if (skb) {
842 __tcp_v4_send_check(skb, ireq->loc_addr, ireq->rmt_addr);
843
844 skb_set_queue_mapping(skb, queue_mapping);
845 err = ip_build_and_send_pkt(skb, sk, ireq->loc_addr,
846 ireq->rmt_addr,
847 ireq->opt);
848 err = net_xmit_eval(err);
849 }
850
851 dst_release(dst);
852 return err;
853}
854
855static int tcp_v4_rtx_synack(struct sock *sk, struct request_sock *req,
856 struct request_values *rvp)
857{
858 TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_RETRANSSEGS);
859 return tcp_v4_send_synack(sk, NULL, req, rvp, 0);
860}
861
862/*
863 * IPv4 request_sock destructor.
864 */
865static void tcp_v4_reqsk_destructor(struct request_sock *req)
866{
867 kfree(inet_rsk(req)->opt);
868}
869
870/*
871 * Return true if a syncookie should be sent
872 */
873bool tcp_syn_flood_action(struct sock *sk,
874 const struct sk_buff *skb,
875 const char *proto)
876{
877 const char *msg = "Dropping request";
878 bool want_cookie = false;
879 struct listen_sock *lopt;
880
881
882
883#ifdef CONFIG_SYN_COOKIES
884 if (sysctl_tcp_syncookies) {
885 msg = "Sending cookies";
886 want_cookie = true;
887 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPREQQFULLDOCOOKIES);
888 } else
889#endif
890 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPREQQFULLDROP);
891
892 lopt = inet_csk(sk)->icsk_accept_queue.listen_opt;
893 if (!lopt->synflood_warned) {
894 lopt->synflood_warned = 1;
895 pr_info("%s: Possible SYN flooding on port %d. %s. Check SNMP counters.\n",
896 proto, ntohs(tcp_hdr(skb)->dest), msg);
897 }
898 return want_cookie;
899}
900EXPORT_SYMBOL(tcp_syn_flood_action);
901
902/*
903 * Save and compile IPv4 options into the request_sock if needed.
904 */
905static struct ip_options_rcu *tcp_v4_save_options(struct sock *sk,
906 struct sk_buff *skb)
907{
908 const struct ip_options *opt = &(IPCB(skb)->opt);
909 struct ip_options_rcu *dopt = NULL;
910
911 if (opt && opt->optlen) {
912 int opt_size = sizeof(*dopt) + opt->optlen;
913
914 dopt = kmalloc(opt_size, GFP_ATOMIC);
915 if (dopt) {
916 if (ip_options_echo(&dopt->opt, skb)) {
917 kfree(dopt);
918 dopt = NULL;
919 }
920 }
921 }
922 return dopt;
923}
924
925#ifdef CONFIG_TCP_MD5SIG
926/*
927 * RFC2385 MD5 checksumming requires a mapping of
928 * IP address->MD5 Key.
929 * We need to maintain these in the sk structure.
930 */
931
932/* Find the Key structure for an address. */
933struct tcp_md5sig_key *tcp_md5_do_lookup(struct sock *sk,
934 const union tcp_md5_addr *addr,
935 int family)
936{
937 struct tcp_sock *tp = tcp_sk(sk);
938 struct tcp_md5sig_key *key;
939 struct hlist_node *pos;
940 unsigned int size = sizeof(struct in_addr);
941 struct tcp_md5sig_info *md5sig;
942
943 /* caller either holds rcu_read_lock() or socket lock */
944 md5sig = rcu_dereference_check(tp->md5sig_info,
945 sock_owned_by_user(sk) ||
946 lockdep_is_held(&sk->sk_lock.slock));
947 if (!md5sig)
948 return NULL;
949#if IS_ENABLED(CONFIG_IPV6)
950 if (family == AF_INET6)
951 size = sizeof(struct in6_addr);
952#endif
953 hlist_for_each_entry_rcu(key, pos, &md5sig->head, node) {
954 if (key->family != family)
955 continue;
956 if (!memcmp(&key->addr, addr, size))
957 return key;
958 }
959 return NULL;
960}
961EXPORT_SYMBOL(tcp_md5_do_lookup);
962
963struct tcp_md5sig_key *tcp_v4_md5_lookup(struct sock *sk,
964 struct sock *addr_sk)
965{
966 union tcp_md5_addr *addr;
967
968 addr = (union tcp_md5_addr *)&inet_sk(addr_sk)->inet_daddr;
969 return tcp_md5_do_lookup(sk, addr, AF_INET);
970}
971EXPORT_SYMBOL(tcp_v4_md5_lookup);
972
973static struct tcp_md5sig_key *tcp_v4_reqsk_md5_lookup(struct sock *sk,
974 struct request_sock *req)
975{
976 union tcp_md5_addr *addr;
977
978 addr = (union tcp_md5_addr *)&inet_rsk(req)->rmt_addr;
979 return tcp_md5_do_lookup(sk, addr, AF_INET);
980}
981
982/* This can be called on a newly created socket, from other files */
983int tcp_md5_do_add(struct sock *sk, const union tcp_md5_addr *addr,
984 int family, const u8 *newkey, u8 newkeylen, gfp_t gfp)
985{
986 /* Add Key to the list */
987 struct tcp_md5sig_key *key;
988 struct tcp_sock *tp = tcp_sk(sk);
989 struct tcp_md5sig_info *md5sig;
990
991 key = tcp_md5_do_lookup(sk, (union tcp_md5_addr *)&addr, AF_INET);
992 if (key) {
993 /* Pre-existing entry - just update that one. */
994 memcpy(key->key, newkey, newkeylen);
995 key->keylen = newkeylen;
996 return 0;
997 }
998
999 md5sig = rcu_dereference_protected(tp->md5sig_info,
1000 sock_owned_by_user(sk));
1001 if (!md5sig) {
1002 md5sig = kmalloc(sizeof(*md5sig), gfp);
1003 if (!md5sig)
1004 return -ENOMEM;
1005
1006 sk_nocaps_add(sk, NETIF_F_GSO_MASK);
1007 INIT_HLIST_HEAD(&md5sig->head);
1008 rcu_assign_pointer(tp->md5sig_info, md5sig);
1009 }
1010
1011 key = sock_kmalloc(sk, sizeof(*key), gfp);
1012 if (!key)
1013 return -ENOMEM;
1014 if (hlist_empty(&md5sig->head) && !tcp_alloc_md5sig_pool(sk)) {
1015 sock_kfree_s(sk, key, sizeof(*key));
1016 return -ENOMEM;
1017 }
1018
1019 memcpy(key->key, newkey, newkeylen);
1020 key->keylen = newkeylen;
1021 key->family = family;
1022 memcpy(&key->addr, addr,
1023 (family == AF_INET6) ? sizeof(struct in6_addr) :
1024 sizeof(struct in_addr));
1025 hlist_add_head_rcu(&key->node, &md5sig->head);
1026 return 0;
1027}
1028EXPORT_SYMBOL(tcp_md5_do_add);
1029
1030int tcp_md5_do_del(struct sock *sk, const union tcp_md5_addr *addr, int family)
1031{
1032 struct tcp_sock *tp = tcp_sk(sk);
1033 struct tcp_md5sig_key *key;
1034 struct tcp_md5sig_info *md5sig;
1035
1036 key = tcp_md5_do_lookup(sk, (union tcp_md5_addr *)&addr, AF_INET);
1037 if (!key)
1038 return -ENOENT;
1039 hlist_del_rcu(&key->node);
1040 atomic_sub(sizeof(*key), &sk->sk_omem_alloc);
1041 kfree_rcu(key, rcu);
1042 md5sig = rcu_dereference_protected(tp->md5sig_info,
1043 sock_owned_by_user(sk));
1044 if (hlist_empty(&md5sig->head))
1045 tcp_free_md5sig_pool();
1046 return 0;
1047}
1048EXPORT_SYMBOL(tcp_md5_do_del);
1049
1050void tcp_clear_md5_list(struct sock *sk)
1051{
1052 struct tcp_sock *tp = tcp_sk(sk);
1053 struct tcp_md5sig_key *key;
1054 struct hlist_node *pos, *n;
1055 struct tcp_md5sig_info *md5sig;
1056
1057 md5sig = rcu_dereference_protected(tp->md5sig_info, 1);
1058
1059 if (!hlist_empty(&md5sig->head))
1060 tcp_free_md5sig_pool();
1061 hlist_for_each_entry_safe(key, pos, n, &md5sig->head, node) {
1062 hlist_del_rcu(&key->node);
1063 atomic_sub(sizeof(*key), &sk->sk_omem_alloc);
1064 kfree_rcu(key, rcu);
1065 }
1066}
1067
1068static int tcp_v4_parse_md5_keys(struct sock *sk, char __user *optval,
1069 int optlen)
1070{
1071 struct tcp_md5sig cmd;
1072 struct sockaddr_in *sin = (struct sockaddr_in *)&cmd.tcpm_addr;
1073
1074 if (optlen < sizeof(cmd))
1075 return -EINVAL;
1076
1077 if (copy_from_user(&cmd, optval, sizeof(cmd)))
1078 return -EFAULT;
1079
1080 if (sin->sin_family != AF_INET)
1081 return -EINVAL;
1082
1083 if (!cmd.tcpm_key || !cmd.tcpm_keylen)
1084 return tcp_md5_do_del(sk, (union tcp_md5_addr *)&sin->sin_addr.s_addr,
1085 AF_INET);
1086
1087 if (cmd.tcpm_keylen > TCP_MD5SIG_MAXKEYLEN)
1088 return -EINVAL;
1089
1090 return tcp_md5_do_add(sk, (union tcp_md5_addr *)&sin->sin_addr.s_addr,
1091 AF_INET, cmd.tcpm_key, cmd.tcpm_keylen,
1092 GFP_KERNEL);
1093}
1094
1095static int tcp_v4_md5_hash_pseudoheader(struct tcp_md5sig_pool *hp,
1096 __be32 daddr, __be32 saddr, int nbytes)
1097{
1098 struct tcp4_pseudohdr *bp;
1099 struct scatterlist sg;
1100
1101 bp = &hp->md5_blk.ip4;
1102
1103 /*
1104 * 1. the TCP pseudo-header (in the order: source IP address,
1105 * destination IP address, zero-padded protocol number, and
1106 * segment length)
1107 */
1108 bp->saddr = saddr;
1109 bp->daddr = daddr;
1110 bp->pad = 0;
1111 bp->protocol = IPPROTO_TCP;
1112 bp->len = cpu_to_be16(nbytes);
1113
1114 sg_init_one(&sg, bp, sizeof(*bp));
1115 return crypto_hash_update(&hp->md5_desc, &sg, sizeof(*bp));
1116}
1117
1118static int tcp_v4_md5_hash_hdr(char *md5_hash, const struct tcp_md5sig_key *key,
1119 __be32 daddr, __be32 saddr, const struct tcphdr *th)
1120{
1121 struct tcp_md5sig_pool *hp;
1122 struct hash_desc *desc;
1123
1124 hp = tcp_get_md5sig_pool();
1125 if (!hp)
1126 goto clear_hash_noput;
1127 desc = &hp->md5_desc;
1128
1129 if (crypto_hash_init(desc))
1130 goto clear_hash;
1131 if (tcp_v4_md5_hash_pseudoheader(hp, daddr, saddr, th->doff << 2))
1132 goto clear_hash;
1133 if (tcp_md5_hash_header(hp, th))
1134 goto clear_hash;
1135 if (tcp_md5_hash_key(hp, key))
1136 goto clear_hash;
1137 if (crypto_hash_final(desc, md5_hash))
1138 goto clear_hash;
1139
1140 tcp_put_md5sig_pool();
1141 return 0;
1142
1143clear_hash:
1144 tcp_put_md5sig_pool();
1145clear_hash_noput:
1146 memset(md5_hash, 0, 16);
1147 return 1;
1148}
1149
1150int tcp_v4_md5_hash_skb(char *md5_hash, struct tcp_md5sig_key *key,
1151 const struct sock *sk, const struct request_sock *req,
1152 const struct sk_buff *skb)
1153{
1154 struct tcp_md5sig_pool *hp;
1155 struct hash_desc *desc;
1156 const struct tcphdr *th = tcp_hdr(skb);
1157 __be32 saddr, daddr;
1158
1159 if (sk) {
1160 saddr = inet_sk(sk)->inet_saddr;
1161 daddr = inet_sk(sk)->inet_daddr;
1162 } else if (req) {
1163 saddr = inet_rsk(req)->loc_addr;
1164 daddr = inet_rsk(req)->rmt_addr;
1165 } else {
1166 const struct iphdr *iph = ip_hdr(skb);
1167 saddr = iph->saddr;
1168 daddr = iph->daddr;
1169 }
1170
1171 hp = tcp_get_md5sig_pool();
1172 if (!hp)
1173 goto clear_hash_noput;
1174 desc = &hp->md5_desc;
1175
1176 if (crypto_hash_init(desc))
1177 goto clear_hash;
1178
1179 if (tcp_v4_md5_hash_pseudoheader(hp, daddr, saddr, skb->len))
1180 goto clear_hash;
1181 if (tcp_md5_hash_header(hp, th))
1182 goto clear_hash;
1183 if (tcp_md5_hash_skb_data(hp, skb, th->doff << 2))
1184 goto clear_hash;
1185 if (tcp_md5_hash_key(hp, key))
1186 goto clear_hash;
1187 if (crypto_hash_final(desc, md5_hash))
1188 goto clear_hash;
1189
1190 tcp_put_md5sig_pool();
1191 return 0;
1192
1193clear_hash:
1194 tcp_put_md5sig_pool();
1195clear_hash_noput:
1196 memset(md5_hash, 0, 16);
1197 return 1;
1198}
1199EXPORT_SYMBOL(tcp_v4_md5_hash_skb);
1200
1201static bool tcp_v4_inbound_md5_hash(struct sock *sk, const struct sk_buff *skb)
1202{
1203 /*
1204 * This gets called for each TCP segment that arrives
1205 * so we want to be efficient.
1206 * We have 3 drop cases:
1207 * o No MD5 hash and one expected.
1208 * o MD5 hash and we're not expecting one.
1209 * o MD5 hash and its wrong.
1210 */
1211 const __u8 *hash_location = NULL;
1212 struct tcp_md5sig_key *hash_expected;
1213 const struct iphdr *iph = ip_hdr(skb);
1214 const struct tcphdr *th = tcp_hdr(skb);
1215 int genhash;
1216 unsigned char newhash[16];
1217
1218 hash_expected = tcp_md5_do_lookup(sk, (union tcp_md5_addr *)&iph->saddr,
1219 AF_INET);
1220 hash_location = tcp_parse_md5sig_option(th);
1221
1222 /* We've parsed the options - do we have a hash? */
1223 if (!hash_expected && !hash_location)
1224 return false;
1225
1226 if (hash_expected && !hash_location) {
1227 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPMD5NOTFOUND);
1228 return true;
1229 }
1230
1231 if (!hash_expected && hash_location) {
1232 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPMD5UNEXPECTED);
1233 return true;
1234 }
1235
1236 /* Okay, so this is hash_expected and hash_location -
1237 * so we need to calculate the checksum.
1238 */
1239 genhash = tcp_v4_md5_hash_skb(newhash,
1240 hash_expected,
1241 NULL, NULL, skb);
1242
1243 if (genhash || memcmp(hash_location, newhash, 16) != 0) {
1244 net_info_ratelimited("MD5 Hash failed for (%pI4, %d)->(%pI4, %d)%s\n",
1245 &iph->saddr, ntohs(th->source),
1246 &iph->daddr, ntohs(th->dest),
1247 genhash ? " tcp_v4_calc_md5_hash failed"
1248 : "");
1249 return true;
1250 }
1251 return false;
1252}
1253
1254#endif
1255
1256struct request_sock_ops tcp_request_sock_ops __read_mostly = {
1257 .family = PF_INET,
1258 .obj_size = sizeof(struct tcp_request_sock),
1259 .rtx_syn_ack = tcp_v4_rtx_synack,
1260 .send_ack = tcp_v4_reqsk_send_ack,
1261 .destructor = tcp_v4_reqsk_destructor,
1262 .send_reset = tcp_v4_send_reset,
1263 .syn_ack_timeout = tcp_syn_ack_timeout,
1264};
1265
1266#ifdef CONFIG_TCP_MD5SIG
1267static const struct tcp_request_sock_ops tcp_request_sock_ipv4_ops = {
1268 .md5_lookup = tcp_v4_reqsk_md5_lookup,
1269 .calc_md5_hash = tcp_v4_md5_hash_skb,
1270};
1271#endif
1272
1273int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb)
1274{
1275 struct tcp_extend_values tmp_ext;
1276 struct tcp_options_received tmp_opt;
1277 const u8 *hash_location;
1278 struct request_sock *req;
1279 struct inet_request_sock *ireq;
1280 struct tcp_sock *tp = tcp_sk(sk);
1281 struct dst_entry *dst = NULL;
1282 __be32 saddr = ip_hdr(skb)->saddr;
1283 __be32 daddr = ip_hdr(skb)->daddr;
1284 __u32 isn = TCP_SKB_CB(skb)->when;
1285 bool want_cookie = false;
1286
1287 /* Never answer to SYNs send to broadcast or multicast */
1288 if (skb_rtable(skb)->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))
1289 goto drop;
1290
1291 /* TW buckets are converted to open requests without
1292 * limitations, they conserve resources and peer is
1293 * evidently real one.
1294 */
1295 if (inet_csk_reqsk_queue_is_full(sk) && !isn) {
1296 want_cookie = tcp_syn_flood_action(sk, skb, "TCP");
1297 if (!want_cookie)
1298 goto drop;
1299 }
1300
1301 /* Accept backlog is full. If we have already queued enough
1302 * of warm entries in syn queue, drop request. It is better than
1303 * clogging syn queue with openreqs with exponentially increasing
1304 * timeout.
1305 */
1306 if (sk_acceptq_is_full(sk) && inet_csk_reqsk_queue_young(sk) > 1)
1307 goto drop;
1308
1309 req = inet_reqsk_alloc(&tcp_request_sock_ops);
1310 if (!req)
1311 goto drop;
1312
1313#ifdef CONFIG_TCP_MD5SIG
1314 tcp_rsk(req)->af_specific = &tcp_request_sock_ipv4_ops;
1315#endif
1316
1317 tcp_clear_options(&tmp_opt);
1318 tmp_opt.mss_clamp = TCP_MSS_DEFAULT;
1319 tmp_opt.user_mss = tp->rx_opt.user_mss;
1320 tcp_parse_options(skb, &tmp_opt, &hash_location, 0);
1321
1322 if (tmp_opt.cookie_plus > 0 &&
1323 tmp_opt.saw_tstamp &&
1324 !tp->rx_opt.cookie_out_never &&
1325 (sysctl_tcp_cookie_size > 0 ||
1326 (tp->cookie_values != NULL &&
1327 tp->cookie_values->cookie_desired > 0))) {
1328 u8 *c;
1329 u32 *mess = &tmp_ext.cookie_bakery[COOKIE_DIGEST_WORDS];
1330 int l = tmp_opt.cookie_plus - TCPOLEN_COOKIE_BASE;
1331
1332 if (tcp_cookie_generator(&tmp_ext.cookie_bakery[0]) != 0)
1333 goto drop_and_release;
1334
1335 /* Secret recipe starts with IP addresses */
1336 *mess++ ^= (__force u32)daddr;
1337 *mess++ ^= (__force u32)saddr;
1338
1339 /* plus variable length Initiator Cookie */
1340 c = (u8 *)mess;
1341 while (l-- > 0)
1342 *c++ ^= *hash_location++;
1343
1344 want_cookie = false; /* not our kind of cookie */
1345 tmp_ext.cookie_out_never = 0; /* false */
1346 tmp_ext.cookie_plus = tmp_opt.cookie_plus;
1347 } else if (!tp->rx_opt.cookie_in_always) {
1348 /* redundant indications, but ensure initialization. */
1349 tmp_ext.cookie_out_never = 1; /* true */
1350 tmp_ext.cookie_plus = 0;
1351 } else {
1352 goto drop_and_release;
1353 }
1354 tmp_ext.cookie_in_always = tp->rx_opt.cookie_in_always;
1355
1356 if (want_cookie && !tmp_opt.saw_tstamp)
1357 tcp_clear_options(&tmp_opt);
1358
1359 tmp_opt.tstamp_ok = tmp_opt.saw_tstamp;
1360 tcp_openreq_init(req, &tmp_opt, skb);
1361
1362 ireq = inet_rsk(req);
1363 ireq->loc_addr = daddr;
1364 ireq->rmt_addr = saddr;
1365 ireq->no_srccheck = inet_sk(sk)->transparent;
1366 ireq->opt = tcp_v4_save_options(sk, skb);
1367
1368 if (security_inet_conn_request(sk, skb, req))
1369 goto drop_and_free;
1370
1371 if (!want_cookie || tmp_opt.tstamp_ok)
1372 TCP_ECN_create_request(req, skb);
1373
1374 if (want_cookie) {
1375 isn = cookie_v4_init_sequence(sk, skb, &req->mss);
1376 req->cookie_ts = tmp_opt.tstamp_ok;
1377 } else if (!isn) {
1378 struct inet_peer *peer = NULL;
1379 struct flowi4 fl4;
1380
1381 /* VJ's idea. We save last timestamp seen
1382 * from the destination in peer table, when entering
1383 * state TIME-WAIT, and check against it before
1384 * accepting new connection request.
1385 *
1386 * If "isn" is not zero, this request hit alive
1387 * timewait bucket, so that all the necessary checks
1388 * are made in the function processing timewait state.
1389 */
1390 if (tmp_opt.saw_tstamp &&
1391 tcp_death_row.sysctl_tw_recycle &&
1392 (dst = inet_csk_route_req(sk, &fl4, req)) != NULL &&
1393 fl4.daddr == saddr &&
1394 (peer = rt_get_peer((struct rtable *)dst, fl4.daddr)) != NULL) {
1395 inet_peer_refcheck(peer);
1396 if ((u32)get_seconds() - peer->tcp_ts_stamp < TCP_PAWS_MSL &&
1397 (s32)(peer->tcp_ts - req->ts_recent) >
1398 TCP_PAWS_WINDOW) {
1399 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_PAWSPASSIVEREJECTED);
1400 goto drop_and_release;
1401 }
1402 }
1403 /* Kill the following clause, if you dislike this way. */
1404 else if (!sysctl_tcp_syncookies &&
1405 (sysctl_max_syn_backlog - inet_csk_reqsk_queue_len(sk) <
1406 (sysctl_max_syn_backlog >> 2)) &&
1407 (!peer || !peer->tcp_ts_stamp) &&
1408 (!dst || !dst_metric(dst, RTAX_RTT))) {
1409 /* Without syncookies last quarter of
1410 * backlog is filled with destinations,
1411 * proven to be alive.
1412 * It means that we continue to communicate
1413 * to destinations, already remembered
1414 * to the moment of synflood.
1415 */
1416 LIMIT_NETDEBUG(KERN_DEBUG pr_fmt("drop open request from %pI4/%u\n"),
1417 &saddr, ntohs(tcp_hdr(skb)->source));
1418 goto drop_and_release;
1419 }
1420
1421 isn = tcp_v4_init_sequence(skb);
1422 }
1423 tcp_rsk(req)->snt_isn = isn;
1424 tcp_rsk(req)->snt_synack = tcp_time_stamp;
1425
1426 if (tcp_v4_send_synack(sk, dst, req,
1427 (struct request_values *)&tmp_ext,
1428 skb_get_queue_mapping(skb)) ||
1429 want_cookie)
1430 goto drop_and_free;
1431
1432 inet_csk_reqsk_queue_hash_add(sk, req, TCP_TIMEOUT_INIT);
1433 return 0;
1434
1435drop_and_release:
1436 dst_release(dst);
1437drop_and_free:
1438 reqsk_free(req);
1439drop:
1440 return 0;
1441}
1442EXPORT_SYMBOL(tcp_v4_conn_request);
1443
1444
1445/*
1446 * The three way handshake has completed - we got a valid synack -
1447 * now create the new socket.
1448 */
1449struct sock *tcp_v4_syn_recv_sock(struct sock *sk, struct sk_buff *skb,
1450 struct request_sock *req,
1451 struct dst_entry *dst)
1452{
1453 struct inet_request_sock *ireq;
1454 struct inet_sock *newinet;
1455 struct tcp_sock *newtp;
1456 struct sock *newsk;
1457#ifdef CONFIG_TCP_MD5SIG
1458 struct tcp_md5sig_key *key;
1459#endif
1460 struct ip_options_rcu *inet_opt;
1461
1462 if (sk_acceptq_is_full(sk))
1463 goto exit_overflow;
1464
1465 newsk = tcp_create_openreq_child(sk, req, skb);
1466 if (!newsk)
1467 goto exit_nonewsk;
1468
1469 newsk->sk_gso_type = SKB_GSO_TCPV4;
1470
1471 newtp = tcp_sk(newsk);
1472 newinet = inet_sk(newsk);
1473 ireq = inet_rsk(req);
1474 newinet->inet_daddr = ireq->rmt_addr;
1475 newinet->inet_rcv_saddr = ireq->loc_addr;
1476 newinet->inet_saddr = ireq->loc_addr;
1477 inet_opt = ireq->opt;
1478 rcu_assign_pointer(newinet->inet_opt, inet_opt);
1479 ireq->opt = NULL;
1480 newinet->mc_index = inet_iif(skb);
1481 newinet->mc_ttl = ip_hdr(skb)->ttl;
1482 newinet->rcv_tos = ip_hdr(skb)->tos;
1483 inet_csk(newsk)->icsk_ext_hdr_len = 0;
1484 if (inet_opt)
1485 inet_csk(newsk)->icsk_ext_hdr_len = inet_opt->opt.optlen;
1486 newinet->inet_id = newtp->write_seq ^ jiffies;
1487
1488 if (!dst) {
1489 dst = inet_csk_route_child_sock(sk, newsk, req);
1490 if (!dst)
1491 goto put_and_exit;
1492 } else {
1493 /* syncookie case : see end of cookie_v4_check() */
1494 }
1495 sk_setup_caps(newsk, dst);
1496
1497 tcp_mtup_init(newsk);
1498 tcp_sync_mss(newsk, dst_mtu(dst));
1499 newtp->advmss = dst_metric_advmss(dst);
1500 if (tcp_sk(sk)->rx_opt.user_mss &&
1501 tcp_sk(sk)->rx_opt.user_mss < newtp->advmss)
1502 newtp->advmss = tcp_sk(sk)->rx_opt.user_mss;
1503
1504 tcp_initialize_rcv_mss(newsk);
1505 if (tcp_rsk(req)->snt_synack)
1506 tcp_valid_rtt_meas(newsk,
1507 tcp_time_stamp - tcp_rsk(req)->snt_synack);
1508 newtp->total_retrans = req->retrans;
1509
1510#ifdef CONFIG_TCP_MD5SIG
1511 /* Copy over the MD5 key from the original socket */
1512 key = tcp_md5_do_lookup(sk, (union tcp_md5_addr *)&newinet->inet_daddr,
1513 AF_INET);
1514 if (key != NULL) {
1515 /*
1516 * We're using one, so create a matching key
1517 * on the newsk structure. If we fail to get
1518 * memory, then we end up not copying the key
1519 * across. Shucks.
1520 */
1521 tcp_md5_do_add(newsk, (union tcp_md5_addr *)&newinet->inet_daddr,
1522 AF_INET, key->key, key->keylen, GFP_ATOMIC);
1523 sk_nocaps_add(newsk, NETIF_F_GSO_MASK);
1524 }
1525#endif
1526
1527 if (__inet_inherit_port(sk, newsk) < 0)
1528 goto put_and_exit;
1529 __inet_hash_nolisten(newsk, NULL);
1530
1531 return newsk;
1532
1533exit_overflow:
1534 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENOVERFLOWS);
1535exit_nonewsk:
1536 dst_release(dst);
1537exit:
1538 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENDROPS);
1539 return NULL;
1540put_and_exit:
1541 tcp_clear_xmit_timers(newsk);
1542 tcp_cleanup_congestion_control(newsk);
1543 bh_unlock_sock(newsk);
1544 sock_put(newsk);
1545 goto exit;
1546}
1547EXPORT_SYMBOL(tcp_v4_syn_recv_sock);
1548
1549static struct sock *tcp_v4_hnd_req(struct sock *sk, struct sk_buff *skb)
1550{
1551 struct tcphdr *th = tcp_hdr(skb);
1552 const struct iphdr *iph = ip_hdr(skb);
1553 struct sock *nsk;
1554 struct request_sock **prev;
1555 /* Find possible connection requests. */
1556 struct request_sock *req = inet_csk_search_req(sk, &prev, th->source,
1557 iph->saddr, iph->daddr);
1558 if (req)
1559 return tcp_check_req(sk, skb, req, prev);
1560
1561 nsk = inet_lookup_established(sock_net(sk), &tcp_hashinfo, iph->saddr,
1562 th->source, iph->daddr, th->dest, inet_iif(skb));
1563
1564 if (nsk) {
1565 if (nsk->sk_state != TCP_TIME_WAIT) {
1566 bh_lock_sock(nsk);
1567 return nsk;
1568 }
1569 inet_twsk_put(inet_twsk(nsk));
1570 return NULL;
1571 }
1572
1573#ifdef CONFIG_SYN_COOKIES
1574 if (!th->syn)
1575 sk = cookie_v4_check(sk, skb, &(IPCB(skb)->opt));
1576#endif
1577 return sk;
1578}
1579
1580static __sum16 tcp_v4_checksum_init(struct sk_buff *skb)
1581{
1582 const struct iphdr *iph = ip_hdr(skb);
1583
1584 if (skb->ip_summed == CHECKSUM_COMPLETE) {
1585 if (!tcp_v4_check(skb->len, iph->saddr,
1586 iph->daddr, skb->csum)) {
1587 skb->ip_summed = CHECKSUM_UNNECESSARY;
1588 return 0;
1589 }
1590 }
1591
1592 skb->csum = csum_tcpudp_nofold(iph->saddr, iph->daddr,
1593 skb->len, IPPROTO_TCP, 0);
1594
1595 if (skb->len <= 76) {
1596 return __skb_checksum_complete(skb);
1597 }
1598 return 0;
1599}
1600
1601
1602/* The socket must have it's spinlock held when we get
1603 * here.
1604 *
1605 * We have a potential double-lock case here, so even when
1606 * doing backlog processing we use the BH locking scheme.
1607 * This is because we cannot sleep with the original spinlock
1608 * held.
1609 */
1610int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb)
1611{
1612 struct sock *rsk;
1613#ifdef CONFIG_TCP_MD5SIG
1614 /*
1615 * We really want to reject the packet as early as possible
1616 * if:
1617 * o We're expecting an MD5'd packet and this is no MD5 tcp option
1618 * o There is an MD5 option and we're not expecting one
1619 */
1620 if (tcp_v4_inbound_md5_hash(sk, skb))
1621 goto discard;
1622#endif
1623
1624 if (sk->sk_state == TCP_ESTABLISHED) { /* Fast path */
1625 sock_rps_save_rxhash(sk, skb);
1626 if (tcp_rcv_established(sk, skb, tcp_hdr(skb), skb->len)) {
1627 rsk = sk;
1628 goto reset;
1629 }
1630 return 0;
1631 }
1632
1633 if (skb->len < tcp_hdrlen(skb) || tcp_checksum_complete(skb))
1634 goto csum_err;
1635
1636 if (sk->sk_state == TCP_LISTEN) {
1637 struct sock *nsk = tcp_v4_hnd_req(sk, skb);
1638 if (!nsk)
1639 goto discard;
1640
1641 if (nsk != sk) {
1642 sock_rps_save_rxhash(nsk, skb);
1643 if (tcp_child_process(sk, nsk, skb)) {
1644 rsk = nsk;
1645 goto reset;
1646 }
1647 return 0;
1648 }
1649 } else
1650 sock_rps_save_rxhash(sk, skb);
1651
1652 if (tcp_rcv_state_process(sk, skb, tcp_hdr(skb), skb->len)) {
1653 rsk = sk;
1654 goto reset;
1655 }
1656 return 0;
1657
1658reset:
1659 tcp_v4_send_reset(rsk, skb);
1660discard:
1661 kfree_skb(skb);
1662 /* Be careful here. If this function gets more complicated and
1663 * gcc suffers from register pressure on the x86, sk (in %ebx)
1664 * might be destroyed here. This current version compiles correctly,
1665 * but you have been warned.
1666 */
1667 return 0;
1668
1669csum_err:
1670 TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_INERRS);
1671 goto discard;
1672}
1673EXPORT_SYMBOL(tcp_v4_do_rcv);
1674
1675/*
1676 * From tcp_input.c
1677 */
1678
1679int tcp_v4_rcv(struct sk_buff *skb)
1680{
1681 const struct iphdr *iph;
1682 const struct tcphdr *th;
1683 struct sock *sk;
1684 int ret;
1685 struct net *net = dev_net(skb->dev);
1686
1687 if (skb->pkt_type != PACKET_HOST)
1688 goto discard_it;
1689
1690 /* Count it even if it's bad */
1691 TCP_INC_STATS_BH(net, TCP_MIB_INSEGS);
1692
1693 if (!pskb_may_pull(skb, sizeof(struct tcphdr)))
1694 goto discard_it;
1695
1696 th = tcp_hdr(skb);
1697
1698 if (th->doff < sizeof(struct tcphdr) / 4)
1699 goto bad_packet;
1700 if (!pskb_may_pull(skb, th->doff * 4))
1701 goto discard_it;
1702
1703 /* An explanation is required here, I think.
1704 * Packet length and doff are validated by header prediction,
1705 * provided case of th->doff==0 is eliminated.
1706 * So, we defer the checks. */
1707 if (!skb_csum_unnecessary(skb) && tcp_v4_checksum_init(skb))
1708 goto bad_packet;
1709
1710 th = tcp_hdr(skb);
1711 iph = ip_hdr(skb);
1712 TCP_SKB_CB(skb)->seq = ntohl(th->seq);
1713 TCP_SKB_CB(skb)->end_seq = (TCP_SKB_CB(skb)->seq + th->syn + th->fin +
1714 skb->len - th->doff * 4);
1715 TCP_SKB_CB(skb)->ack_seq = ntohl(th->ack_seq);
1716 TCP_SKB_CB(skb)->when = 0;
1717 TCP_SKB_CB(skb)->ip_dsfield = ipv4_get_dsfield(iph);
1718 TCP_SKB_CB(skb)->sacked = 0;
1719
1720 sk = __inet_lookup_skb(&tcp_hashinfo, skb, th->source, th->dest);
1721 if (!sk)
1722 goto no_tcp_socket;
1723
1724process:
1725 if (sk->sk_state == TCP_TIME_WAIT)
1726 goto do_time_wait;
1727
1728 if (unlikely(iph->ttl < inet_sk(sk)->min_ttl)) {
1729 NET_INC_STATS_BH(net, LINUX_MIB_TCPMINTTLDROP);
1730 goto discard_and_relse;
1731 }
1732
1733 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
1734 goto discard_and_relse;
1735 nf_reset(skb);
1736
1737 if (sk_filter(sk, skb))
1738 goto discard_and_relse;
1739
1740 skb->dev = NULL;
1741
1742 bh_lock_sock_nested(sk);
1743 ret = 0;
1744 if (!sock_owned_by_user(sk)) {
1745#ifdef CONFIG_NET_DMA
1746 struct tcp_sock *tp = tcp_sk(sk);
1747 if (!tp->ucopy.dma_chan && tp->ucopy.pinned_list)
1748 tp->ucopy.dma_chan = net_dma_find_channel();
1749 if (tp->ucopy.dma_chan)
1750 ret = tcp_v4_do_rcv(sk, skb);
1751 else
1752#endif
1753 {
1754 if (!tcp_prequeue(sk, skb))
1755 ret = tcp_v4_do_rcv(sk, skb);
1756 }
1757 } else if (unlikely(sk_add_backlog(sk, skb,
1758 sk->sk_rcvbuf + sk->sk_sndbuf))) {
1759 bh_unlock_sock(sk);
1760 NET_INC_STATS_BH(net, LINUX_MIB_TCPBACKLOGDROP);
1761 goto discard_and_relse;
1762 }
1763 bh_unlock_sock(sk);
1764
1765 sock_put(sk);
1766
1767 return ret;
1768
1769no_tcp_socket:
1770 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
1771 goto discard_it;
1772
1773 if (skb->len < (th->doff << 2) || tcp_checksum_complete(skb)) {
1774bad_packet:
1775 TCP_INC_STATS_BH(net, TCP_MIB_INERRS);
1776 } else {
1777 tcp_v4_send_reset(NULL, skb);
1778 }
1779
1780discard_it:
1781 /* Discard frame. */
1782 kfree_skb(skb);
1783 return 0;
1784
1785discard_and_relse:
1786 sock_put(sk);
1787 goto discard_it;
1788
1789do_time_wait:
1790 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) {
1791 inet_twsk_put(inet_twsk(sk));
1792 goto discard_it;
1793 }
1794
1795 if (skb->len < (th->doff << 2) || tcp_checksum_complete(skb)) {
1796 TCP_INC_STATS_BH(net, TCP_MIB_INERRS);
1797 inet_twsk_put(inet_twsk(sk));
1798 goto discard_it;
1799 }
1800 switch (tcp_timewait_state_process(inet_twsk(sk), skb, th)) {
1801 case TCP_TW_SYN: {
1802 struct sock *sk2 = inet_lookup_listener(dev_net(skb->dev),
1803 &tcp_hashinfo,
1804 iph->daddr, th->dest,
1805 inet_iif(skb));
1806 if (sk2) {
1807 inet_twsk_deschedule(inet_twsk(sk), &tcp_death_row);
1808 inet_twsk_put(inet_twsk(sk));
1809 sk = sk2;
1810 goto process;
1811 }
1812 /* Fall through to ACK */
1813 }
1814 case TCP_TW_ACK:
1815 tcp_v4_timewait_ack(sk, skb);
1816 break;
1817 case TCP_TW_RST:
1818 goto no_tcp_socket;
1819 case TCP_TW_SUCCESS:;
1820 }
1821 goto discard_it;
1822}
1823
1824struct inet_peer *tcp_v4_get_peer(struct sock *sk, bool *release_it)
1825{
1826 struct rtable *rt = (struct rtable *) __sk_dst_get(sk);
1827 struct inet_sock *inet = inet_sk(sk);
1828 struct inet_peer *peer;
1829
1830 if (!rt ||
1831 inet->cork.fl.u.ip4.daddr != inet->inet_daddr) {
1832 peer = inet_getpeer_v4(inet->inet_daddr, 1);
1833 *release_it = true;
1834 } else {
1835 if (!rt->peer)
1836 rt_bind_peer(rt, inet->inet_daddr, 1);
1837 peer = rt->peer;
1838 *release_it = false;
1839 }
1840
1841 return peer;
1842}
1843EXPORT_SYMBOL(tcp_v4_get_peer);
1844
1845void *tcp_v4_tw_get_peer(struct sock *sk)
1846{
1847 const struct inet_timewait_sock *tw = inet_twsk(sk);
1848
1849 return inet_getpeer_v4(tw->tw_daddr, 1);
1850}
1851EXPORT_SYMBOL(tcp_v4_tw_get_peer);
1852
1853static struct timewait_sock_ops tcp_timewait_sock_ops = {
1854 .twsk_obj_size = sizeof(struct tcp_timewait_sock),
1855 .twsk_unique = tcp_twsk_unique,
1856 .twsk_destructor= tcp_twsk_destructor,
1857 .twsk_getpeer = tcp_v4_tw_get_peer,
1858};
1859
1860const struct inet_connection_sock_af_ops ipv4_specific = {
1861 .queue_xmit = ip_queue_xmit,
1862 .send_check = tcp_v4_send_check,
1863 .rebuild_header = inet_sk_rebuild_header,
1864 .conn_request = tcp_v4_conn_request,
1865 .syn_recv_sock = tcp_v4_syn_recv_sock,
1866 .get_peer = tcp_v4_get_peer,
1867 .net_header_len = sizeof(struct iphdr),
1868 .setsockopt = ip_setsockopt,
1869 .getsockopt = ip_getsockopt,
1870 .addr2sockaddr = inet_csk_addr2sockaddr,
1871 .sockaddr_len = sizeof(struct sockaddr_in),
1872 .bind_conflict = inet_csk_bind_conflict,
1873#ifdef CONFIG_COMPAT
1874 .compat_setsockopt = compat_ip_setsockopt,
1875 .compat_getsockopt = compat_ip_getsockopt,
1876#endif
1877};
1878EXPORT_SYMBOL(ipv4_specific);
1879
1880#ifdef CONFIG_TCP_MD5SIG
1881static const struct tcp_sock_af_ops tcp_sock_ipv4_specific = {
1882 .md5_lookup = tcp_v4_md5_lookup,
1883 .calc_md5_hash = tcp_v4_md5_hash_skb,
1884 .md5_parse = tcp_v4_parse_md5_keys,
1885};
1886#endif
1887
1888/* NOTE: A lot of things set to zero explicitly by call to
1889 * sk_alloc() so need not be done here.
1890 */
1891static int tcp_v4_init_sock(struct sock *sk)
1892{
1893 struct inet_connection_sock *icsk = inet_csk(sk);
1894
1895 tcp_init_sock(sk);
1896
1897 icsk->icsk_af_ops = &ipv4_specific;
1898
1899#ifdef CONFIG_TCP_MD5SIG
1900 tcp_sk(sk)->af_specific = &tcp_sock_ipv4_specific;
1901#endif
1902
1903 return 0;
1904}
1905
1906void tcp_v4_destroy_sock(struct sock *sk)
1907{
1908 struct tcp_sock *tp = tcp_sk(sk);
1909
1910 tcp_clear_xmit_timers(sk);
1911
1912 tcp_cleanup_congestion_control(sk);
1913
1914 /* Cleanup up the write buffer. */
1915 tcp_write_queue_purge(sk);
1916
1917 /* Cleans up our, hopefully empty, out_of_order_queue. */
1918 __skb_queue_purge(&tp->out_of_order_queue);
1919
1920#ifdef CONFIG_TCP_MD5SIG
1921 /* Clean up the MD5 key list, if any */
1922 if (tp->md5sig_info) {
1923 tcp_clear_md5_list(sk);
1924 kfree_rcu(tp->md5sig_info, rcu);
1925 tp->md5sig_info = NULL;
1926 }
1927#endif
1928
1929#ifdef CONFIG_NET_DMA
1930 /* Cleans up our sk_async_wait_queue */
1931 __skb_queue_purge(&sk->sk_async_wait_queue);
1932#endif
1933
1934 /* Clean prequeue, it must be empty really */
1935 __skb_queue_purge(&tp->ucopy.prequeue);
1936
1937 /* Clean up a referenced TCP bind bucket. */
1938 if (inet_csk(sk)->icsk_bind_hash)
1939 inet_put_port(sk);
1940
1941 /*
1942 * If sendmsg cached page exists, toss it.
1943 */
1944 if (sk->sk_sndmsg_page) {
1945 __free_page(sk->sk_sndmsg_page);
1946 sk->sk_sndmsg_page = NULL;
1947 }
1948
1949 /* TCP Cookie Transactions */
1950 if (tp->cookie_values != NULL) {
1951 kref_put(&tp->cookie_values->kref,
1952 tcp_cookie_values_release);
1953 tp->cookie_values = NULL;
1954 }
1955
1956 sk_sockets_allocated_dec(sk);
1957 sock_release_memcg(sk);
1958}
1959EXPORT_SYMBOL(tcp_v4_destroy_sock);
1960
1961#ifdef CONFIG_PROC_FS
1962/* Proc filesystem TCP sock list dumping. */
1963
1964static inline struct inet_timewait_sock *tw_head(struct hlist_nulls_head *head)
1965{
1966 return hlist_nulls_empty(head) ? NULL :
1967 list_entry(head->first, struct inet_timewait_sock, tw_node);
1968}
1969
1970static inline struct inet_timewait_sock *tw_next(struct inet_timewait_sock *tw)
1971{
1972 return !is_a_nulls(tw->tw_node.next) ?
1973 hlist_nulls_entry(tw->tw_node.next, typeof(*tw), tw_node) : NULL;
1974}
1975
1976/*
1977 * Get next listener socket follow cur. If cur is NULL, get first socket
1978 * starting from bucket given in st->bucket; when st->bucket is zero the
1979 * very first socket in the hash table is returned.
1980 */
1981static void *listening_get_next(struct seq_file *seq, void *cur)
1982{
1983 struct inet_connection_sock *icsk;
1984 struct hlist_nulls_node *node;
1985 struct sock *sk = cur;
1986 struct inet_listen_hashbucket *ilb;
1987 struct tcp_iter_state *st = seq->private;
1988 struct net *net = seq_file_net(seq);
1989
1990 if (!sk) {
1991 ilb = &tcp_hashinfo.listening_hash[st->bucket];
1992 spin_lock_bh(&ilb->lock);
1993 sk = sk_nulls_head(&ilb->head);
1994 st->offset = 0;
1995 goto get_sk;
1996 }
1997 ilb = &tcp_hashinfo.listening_hash[st->bucket];
1998 ++st->num;
1999 ++st->offset;
2000
2001 if (st->state == TCP_SEQ_STATE_OPENREQ) {
2002 struct request_sock *req = cur;
2003
2004 icsk = inet_csk(st->syn_wait_sk);
2005 req = req->dl_next;
2006 while (1) {
2007 while (req) {
2008 if (req->rsk_ops->family == st->family) {
2009 cur = req;
2010 goto out;
2011 }
2012 req = req->dl_next;
2013 }
2014 if (++st->sbucket >= icsk->icsk_accept_queue.listen_opt->nr_table_entries)
2015 break;
2016get_req:
2017 req = icsk->icsk_accept_queue.listen_opt->syn_table[st->sbucket];
2018 }
2019 sk = sk_nulls_next(st->syn_wait_sk);
2020 st->state = TCP_SEQ_STATE_LISTENING;
2021 read_unlock_bh(&icsk->icsk_accept_queue.syn_wait_lock);
2022 } else {
2023 icsk = inet_csk(sk);
2024 read_lock_bh(&icsk->icsk_accept_queue.syn_wait_lock);
2025 if (reqsk_queue_len(&icsk->icsk_accept_queue))
2026 goto start_req;
2027 read_unlock_bh(&icsk->icsk_accept_queue.syn_wait_lock);
2028 sk = sk_nulls_next(sk);
2029 }
2030get_sk:
2031 sk_nulls_for_each_from(sk, node) {
2032 if (!net_eq(sock_net(sk), net))
2033 continue;
2034 if (sk->sk_family == st->family) {
2035 cur = sk;
2036 goto out;
2037 }
2038 icsk = inet_csk(sk);
2039 read_lock_bh(&icsk->icsk_accept_queue.syn_wait_lock);
2040 if (reqsk_queue_len(&icsk->icsk_accept_queue)) {
2041start_req:
2042 st->uid = sock_i_uid(sk);
2043 st->syn_wait_sk = sk;
2044 st->state = TCP_SEQ_STATE_OPENREQ;
2045 st->sbucket = 0;
2046 goto get_req;
2047 }
2048 read_unlock_bh(&icsk->icsk_accept_queue.syn_wait_lock);
2049 }
2050 spin_unlock_bh(&ilb->lock);
2051 st->offset = 0;
2052 if (++st->bucket < INET_LHTABLE_SIZE) {
2053 ilb = &tcp_hashinfo.listening_hash[st->bucket];
2054 spin_lock_bh(&ilb->lock);
2055 sk = sk_nulls_head(&ilb->head);
2056 goto get_sk;
2057 }
2058 cur = NULL;
2059out:
2060 return cur;
2061}
2062
2063static void *listening_get_idx(struct seq_file *seq, loff_t *pos)
2064{
2065 struct tcp_iter_state *st = seq->private;
2066 void *rc;
2067
2068 st->bucket = 0;
2069 st->offset = 0;
2070 rc = listening_get_next(seq, NULL);
2071
2072 while (rc && *pos) {
2073 rc = listening_get_next(seq, rc);
2074 --*pos;
2075 }
2076 return rc;
2077}
2078
2079static inline bool empty_bucket(struct tcp_iter_state *st)
2080{
2081 return hlist_nulls_empty(&tcp_hashinfo.ehash[st->bucket].chain) &&
2082 hlist_nulls_empty(&tcp_hashinfo.ehash[st->bucket].twchain);
2083}
2084
2085/*
2086 * Get first established socket starting from bucket given in st->bucket.
2087 * If st->bucket is zero, the very first socket in the hash is returned.
2088 */
2089static void *established_get_first(struct seq_file *seq)
2090{
2091 struct tcp_iter_state *st = seq->private;
2092 struct net *net = seq_file_net(seq);
2093 void *rc = NULL;
2094
2095 st->offset = 0;
2096 for (; st->bucket <= tcp_hashinfo.ehash_mask; ++st->bucket) {
2097 struct sock *sk;
2098 struct hlist_nulls_node *node;
2099 struct inet_timewait_sock *tw;
2100 spinlock_t *lock = inet_ehash_lockp(&tcp_hashinfo, st->bucket);
2101
2102 /* Lockless fast path for the common case of empty buckets */
2103 if (empty_bucket(st))
2104 continue;
2105
2106 spin_lock_bh(lock);
2107 sk_nulls_for_each(sk, node, &tcp_hashinfo.ehash[st->bucket].chain) {
2108 if (sk->sk_family != st->family ||
2109 !net_eq(sock_net(sk), net)) {
2110 continue;
2111 }
2112 rc = sk;
2113 goto out;
2114 }
2115 st->state = TCP_SEQ_STATE_TIME_WAIT;
2116 inet_twsk_for_each(tw, node,
2117 &tcp_hashinfo.ehash[st->bucket].twchain) {
2118 if (tw->tw_family != st->family ||
2119 !net_eq(twsk_net(tw), net)) {
2120 continue;
2121 }
2122 rc = tw;
2123 goto out;
2124 }
2125 spin_unlock_bh(lock);
2126 st->state = TCP_SEQ_STATE_ESTABLISHED;
2127 }
2128out:
2129 return rc;
2130}
2131
2132static void *established_get_next(struct seq_file *seq, void *cur)
2133{
2134 struct sock *sk = cur;
2135 struct inet_timewait_sock *tw;
2136 struct hlist_nulls_node *node;
2137 struct tcp_iter_state *st = seq->private;
2138 struct net *net = seq_file_net(seq);
2139
2140 ++st->num;
2141 ++st->offset;
2142
2143 if (st->state == TCP_SEQ_STATE_TIME_WAIT) {
2144 tw = cur;
2145 tw = tw_next(tw);
2146get_tw:
2147 while (tw && (tw->tw_family != st->family || !net_eq(twsk_net(tw), net))) {
2148 tw = tw_next(tw);
2149 }
2150 if (tw) {
2151 cur = tw;
2152 goto out;
2153 }
2154 spin_unlock_bh(inet_ehash_lockp(&tcp_hashinfo, st->bucket));
2155 st->state = TCP_SEQ_STATE_ESTABLISHED;
2156
2157 /* Look for next non empty bucket */
2158 st->offset = 0;
2159 while (++st->bucket <= tcp_hashinfo.ehash_mask &&
2160 empty_bucket(st))
2161 ;
2162 if (st->bucket > tcp_hashinfo.ehash_mask)
2163 return NULL;
2164
2165 spin_lock_bh(inet_ehash_lockp(&tcp_hashinfo, st->bucket));
2166 sk = sk_nulls_head(&tcp_hashinfo.ehash[st->bucket].chain);
2167 } else
2168 sk = sk_nulls_next(sk);
2169
2170 sk_nulls_for_each_from(sk, node) {
2171 if (sk->sk_family == st->family && net_eq(sock_net(sk), net))
2172 goto found;
2173 }
2174
2175 st->state = TCP_SEQ_STATE_TIME_WAIT;
2176 tw = tw_head(&tcp_hashinfo.ehash[st->bucket].twchain);
2177 goto get_tw;
2178found:
2179 cur = sk;
2180out:
2181 return cur;
2182}
2183
2184static void *established_get_idx(struct seq_file *seq, loff_t pos)
2185{
2186 struct tcp_iter_state *st = seq->private;
2187 void *rc;
2188
2189 st->bucket = 0;
2190 rc = established_get_first(seq);
2191
2192 while (rc && pos) {
2193 rc = established_get_next(seq, rc);
2194 --pos;
2195 }
2196 return rc;
2197}
2198
2199static void *tcp_get_idx(struct seq_file *seq, loff_t pos)
2200{
2201 void *rc;
2202 struct tcp_iter_state *st = seq->private;
2203
2204 st->state = TCP_SEQ_STATE_LISTENING;
2205 rc = listening_get_idx(seq, &pos);
2206
2207 if (!rc) {
2208 st->state = TCP_SEQ_STATE_ESTABLISHED;
2209 rc = established_get_idx(seq, pos);
2210 }
2211
2212 return rc;
2213}
2214
2215static void *tcp_seek_last_pos(struct seq_file *seq)
2216{
2217 struct tcp_iter_state *st = seq->private;
2218 int offset = st->offset;
2219 int orig_num = st->num;
2220 void *rc = NULL;
2221
2222 switch (st->state) {
2223 case TCP_SEQ_STATE_OPENREQ:
2224 case TCP_SEQ_STATE_LISTENING:
2225 if (st->bucket >= INET_LHTABLE_SIZE)
2226 break;
2227 st->state = TCP_SEQ_STATE_LISTENING;
2228 rc = listening_get_next(seq, NULL);
2229 while (offset-- && rc)
2230 rc = listening_get_next(seq, rc);
2231 if (rc)
2232 break;
2233 st->bucket = 0;
2234 /* Fallthrough */
2235 case TCP_SEQ_STATE_ESTABLISHED:
2236 case TCP_SEQ_STATE_TIME_WAIT:
2237 st->state = TCP_SEQ_STATE_ESTABLISHED;
2238 if (st->bucket > tcp_hashinfo.ehash_mask)
2239 break;
2240 rc = established_get_first(seq);
2241 while (offset-- && rc)
2242 rc = established_get_next(seq, rc);
2243 }
2244
2245 st->num = orig_num;
2246
2247 return rc;
2248}
2249
2250static void *tcp_seq_start(struct seq_file *seq, loff_t *pos)
2251{
2252 struct tcp_iter_state *st = seq->private;
2253 void *rc;
2254
2255 if (*pos && *pos == st->last_pos) {
2256 rc = tcp_seek_last_pos(seq);
2257 if (rc)
2258 goto out;
2259 }
2260
2261 st->state = TCP_SEQ_STATE_LISTENING;
2262 st->num = 0;
2263 st->bucket = 0;
2264 st->offset = 0;
2265 rc = *pos ? tcp_get_idx(seq, *pos - 1) : SEQ_START_TOKEN;
2266
2267out:
2268 st->last_pos = *pos;
2269 return rc;
2270}
2271
2272static void *tcp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2273{
2274 struct tcp_iter_state *st = seq->private;
2275 void *rc = NULL;
2276
2277 if (v == SEQ_START_TOKEN) {
2278 rc = tcp_get_idx(seq, 0);
2279 goto out;
2280 }
2281
2282 switch (st->state) {
2283 case TCP_SEQ_STATE_OPENREQ:
2284 case TCP_SEQ_STATE_LISTENING:
2285 rc = listening_get_next(seq, v);
2286 if (!rc) {
2287 st->state = TCP_SEQ_STATE_ESTABLISHED;
2288 st->bucket = 0;
2289 st->offset = 0;
2290 rc = established_get_first(seq);
2291 }
2292 break;
2293 case TCP_SEQ_STATE_ESTABLISHED:
2294 case TCP_SEQ_STATE_TIME_WAIT:
2295 rc = established_get_next(seq, v);
2296 break;
2297 }
2298out:
2299 ++*pos;
2300 st->last_pos = *pos;
2301 return rc;
2302}
2303
2304static void tcp_seq_stop(struct seq_file *seq, void *v)
2305{
2306 struct tcp_iter_state *st = seq->private;
2307
2308 switch (st->state) {
2309 case TCP_SEQ_STATE_OPENREQ:
2310 if (v) {
2311 struct inet_connection_sock *icsk = inet_csk(st->syn_wait_sk);
2312 read_unlock_bh(&icsk->icsk_accept_queue.syn_wait_lock);
2313 }
2314 case TCP_SEQ_STATE_LISTENING:
2315 if (v != SEQ_START_TOKEN)
2316 spin_unlock_bh(&tcp_hashinfo.listening_hash[st->bucket].lock);
2317 break;
2318 case TCP_SEQ_STATE_TIME_WAIT:
2319 case TCP_SEQ_STATE_ESTABLISHED:
2320 if (v)
2321 spin_unlock_bh(inet_ehash_lockp(&tcp_hashinfo, st->bucket));
2322 break;
2323 }
2324}
2325
2326int tcp_seq_open(struct inode *inode, struct file *file)
2327{
2328 struct tcp_seq_afinfo *afinfo = PDE(inode)->data;
2329 struct tcp_iter_state *s;
2330 int err;
2331
2332 err = seq_open_net(inode, file, &afinfo->seq_ops,
2333 sizeof(struct tcp_iter_state));
2334 if (err < 0)
2335 return err;
2336
2337 s = ((struct seq_file *)file->private_data)->private;
2338 s->family = afinfo->family;
2339 s->last_pos = 0;
2340 return 0;
2341}
2342EXPORT_SYMBOL(tcp_seq_open);
2343
2344int tcp_proc_register(struct net *net, struct tcp_seq_afinfo *afinfo)
2345{
2346 int rc = 0;
2347 struct proc_dir_entry *p;
2348
2349 afinfo->seq_ops.start = tcp_seq_start;
2350 afinfo->seq_ops.next = tcp_seq_next;
2351 afinfo->seq_ops.stop = tcp_seq_stop;
2352
2353 p = proc_create_data(afinfo->name, S_IRUGO, net->proc_net,
2354 afinfo->seq_fops, afinfo);
2355 if (!p)
2356 rc = -ENOMEM;
2357 return rc;
2358}
2359EXPORT_SYMBOL(tcp_proc_register);
2360
2361void tcp_proc_unregister(struct net *net, struct tcp_seq_afinfo *afinfo)
2362{
2363 proc_net_remove(net, afinfo->name);
2364}
2365EXPORT_SYMBOL(tcp_proc_unregister);
2366
2367static void get_openreq4(const struct sock *sk, const struct request_sock *req,
2368 struct seq_file *f, int i, int uid, int *len)
2369{
2370 const struct inet_request_sock *ireq = inet_rsk(req);
2371 int ttd = req->expires - jiffies;
2372
2373 seq_printf(f, "%4d: %08X:%04X %08X:%04X"
2374 " %02X %08X:%08X %02X:%08lX %08X %5d %8d %u %d %pK%n",
2375 i,
2376 ireq->loc_addr,
2377 ntohs(inet_sk(sk)->inet_sport),
2378 ireq->rmt_addr,
2379 ntohs(ireq->rmt_port),
2380 TCP_SYN_RECV,
2381 0, 0, /* could print option size, but that is af dependent. */
2382 1, /* timers active (only the expire timer) */
2383 jiffies_to_clock_t(ttd),
2384 req->retrans,
2385 uid,
2386 0, /* non standard timer */
2387 0, /* open_requests have no inode */
2388 atomic_read(&sk->sk_refcnt),
2389 req,
2390 len);
2391}
2392
2393static void get_tcp4_sock(struct sock *sk, struct seq_file *f, int i, int *len)
2394{
2395 int timer_active;
2396 unsigned long timer_expires;
2397 const struct tcp_sock *tp = tcp_sk(sk);
2398 const struct inet_connection_sock *icsk = inet_csk(sk);
2399 const struct inet_sock *inet = inet_sk(sk);
2400 __be32 dest = inet->inet_daddr;
2401 __be32 src = inet->inet_rcv_saddr;
2402 __u16 destp = ntohs(inet->inet_dport);
2403 __u16 srcp = ntohs(inet->inet_sport);
2404 int rx_queue;
2405
2406 if (icsk->icsk_pending == ICSK_TIME_RETRANS) {
2407 timer_active = 1;
2408 timer_expires = icsk->icsk_timeout;
2409 } else if (icsk->icsk_pending == ICSK_TIME_PROBE0) {
2410 timer_active = 4;
2411 timer_expires = icsk->icsk_timeout;
2412 } else if (timer_pending(&sk->sk_timer)) {
2413 timer_active = 2;
2414 timer_expires = sk->sk_timer.expires;
2415 } else {
2416 timer_active = 0;
2417 timer_expires = jiffies;
2418 }
2419
2420 if (sk->sk_state == TCP_LISTEN)
2421 rx_queue = sk->sk_ack_backlog;
2422 else
2423 /*
2424 * because we dont lock socket, we might find a transient negative value
2425 */
2426 rx_queue = max_t(int, tp->rcv_nxt - tp->copied_seq, 0);
2427
2428 seq_printf(f, "%4d: %08X:%04X %08X:%04X %02X %08X:%08X %02X:%08lX "
2429 "%08X %5d %8d %lu %d %pK %lu %lu %u %u %d%n",
2430 i, src, srcp, dest, destp, sk->sk_state,
2431 tp->write_seq - tp->snd_una,
2432 rx_queue,
2433 timer_active,
2434 jiffies_to_clock_t(timer_expires - jiffies),
2435 icsk->icsk_retransmits,
2436 sock_i_uid(sk),
2437 icsk->icsk_probes_out,
2438 sock_i_ino(sk),
2439 atomic_read(&sk->sk_refcnt), sk,
2440 jiffies_to_clock_t(icsk->icsk_rto),
2441 jiffies_to_clock_t(icsk->icsk_ack.ato),
2442 (icsk->icsk_ack.quick << 1) | icsk->icsk_ack.pingpong,
2443 tp->snd_cwnd,
2444 tcp_in_initial_slowstart(tp) ? -1 : tp->snd_ssthresh,
2445 len);
2446}
2447
2448static void get_timewait4_sock(const struct inet_timewait_sock *tw,
2449 struct seq_file *f, int i, int *len)
2450{
2451 __be32 dest, src;
2452 __u16 destp, srcp;
2453 int ttd = tw->tw_ttd - jiffies;
2454
2455 if (ttd < 0)
2456 ttd = 0;
2457
2458 dest = tw->tw_daddr;
2459 src = tw->tw_rcv_saddr;
2460 destp = ntohs(tw->tw_dport);
2461 srcp = ntohs(tw->tw_sport);
2462
2463 seq_printf(f, "%4d: %08X:%04X %08X:%04X"
2464 " %02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %pK%n",
2465 i, src, srcp, dest, destp, tw->tw_substate, 0, 0,
2466 3, jiffies_to_clock_t(ttd), 0, 0, 0, 0,
2467 atomic_read(&tw->tw_refcnt), tw, len);
2468}
2469
2470#define TMPSZ 150
2471
2472static int tcp4_seq_show(struct seq_file *seq, void *v)
2473{
2474 struct tcp_iter_state *st;
2475 int len;
2476
2477 if (v == SEQ_START_TOKEN) {
2478 seq_printf(seq, "%-*s\n", TMPSZ - 1,
2479 " sl local_address rem_address st tx_queue "
2480 "rx_queue tr tm->when retrnsmt uid timeout "
2481 "inode");
2482 goto out;
2483 }
2484 st = seq->private;
2485
2486 switch (st->state) {
2487 case TCP_SEQ_STATE_LISTENING:
2488 case TCP_SEQ_STATE_ESTABLISHED:
2489 get_tcp4_sock(v, seq, st->num, &len);
2490 break;
2491 case TCP_SEQ_STATE_OPENREQ:
2492 get_openreq4(st->syn_wait_sk, v, seq, st->num, st->uid, &len);
2493 break;
2494 case TCP_SEQ_STATE_TIME_WAIT:
2495 get_timewait4_sock(v, seq, st->num, &len);
2496 break;
2497 }
2498 seq_printf(seq, "%*s\n", TMPSZ - 1 - len, "");
2499out:
2500 return 0;
2501}
2502
2503static const struct file_operations tcp_afinfo_seq_fops = {
2504 .owner = THIS_MODULE,
2505 .open = tcp_seq_open,
2506 .read = seq_read,
2507 .llseek = seq_lseek,
2508 .release = seq_release_net
2509};
2510
2511static struct tcp_seq_afinfo tcp4_seq_afinfo = {
2512 .name = "tcp",
2513 .family = AF_INET,
2514 .seq_fops = &tcp_afinfo_seq_fops,
2515 .seq_ops = {
2516 .show = tcp4_seq_show,
2517 },
2518};
2519
2520static int __net_init tcp4_proc_init_net(struct net *net)
2521{
2522 return tcp_proc_register(net, &tcp4_seq_afinfo);
2523}
2524
2525static void __net_exit tcp4_proc_exit_net(struct net *net)
2526{
2527 tcp_proc_unregister(net, &tcp4_seq_afinfo);
2528}
2529
2530static struct pernet_operations tcp4_net_ops = {
2531 .init = tcp4_proc_init_net,
2532 .exit = tcp4_proc_exit_net,
2533};
2534
2535int __init tcp4_proc_init(void)
2536{
2537 return register_pernet_subsys(&tcp4_net_ops);
2538}
2539
2540void tcp4_proc_exit(void)
2541{
2542 unregister_pernet_subsys(&tcp4_net_ops);
2543}
2544#endif /* CONFIG_PROC_FS */
2545
2546struct sk_buff **tcp4_gro_receive(struct sk_buff **head, struct sk_buff *skb)
2547{
2548 const struct iphdr *iph = skb_gro_network_header(skb);
2549
2550 switch (skb->ip_summed) {
2551 case CHECKSUM_COMPLETE:
2552 if (!tcp_v4_check(skb_gro_len(skb), iph->saddr, iph->daddr,
2553 skb->csum)) {
2554 skb->ip_summed = CHECKSUM_UNNECESSARY;
2555 break;
2556 }
2557
2558 /* fall through */
2559 case CHECKSUM_NONE:
2560 NAPI_GRO_CB(skb)->flush = 1;
2561 return NULL;
2562 }
2563
2564 return tcp_gro_receive(head, skb);
2565}
2566
2567int tcp4_gro_complete(struct sk_buff *skb)
2568{
2569 const struct iphdr *iph = ip_hdr(skb);
2570 struct tcphdr *th = tcp_hdr(skb);
2571
2572 th->check = ~tcp_v4_check(skb->len - skb_transport_offset(skb),
2573 iph->saddr, iph->daddr, 0);
2574 skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
2575
2576 return tcp_gro_complete(skb);
2577}
2578
2579struct proto tcp_prot = {
2580 .name = "TCP",
2581 .owner = THIS_MODULE,
2582 .close = tcp_close,
2583 .connect = tcp_v4_connect,
2584 .disconnect = tcp_disconnect,
2585 .accept = inet_csk_accept,
2586 .ioctl = tcp_ioctl,
2587 .init = tcp_v4_init_sock,
2588 .destroy = tcp_v4_destroy_sock,
2589 .shutdown = tcp_shutdown,
2590 .setsockopt = tcp_setsockopt,
2591 .getsockopt = tcp_getsockopt,
2592 .recvmsg = tcp_recvmsg,
2593 .sendmsg = tcp_sendmsg,
2594 .sendpage = tcp_sendpage,
2595 .backlog_rcv = tcp_v4_do_rcv,
2596 .hash = inet_hash,
2597 .unhash = inet_unhash,
2598 .get_port = inet_csk_get_port,
2599 .enter_memory_pressure = tcp_enter_memory_pressure,
2600 .sockets_allocated = &tcp_sockets_allocated,
2601 .orphan_count = &tcp_orphan_count,
2602 .memory_allocated = &tcp_memory_allocated,
2603 .memory_pressure = &tcp_memory_pressure,
2604 .sysctl_wmem = sysctl_tcp_wmem,
2605 .sysctl_rmem = sysctl_tcp_rmem,
2606 .max_header = MAX_TCP_HEADER,
2607 .obj_size = sizeof(struct tcp_sock),
2608 .slab_flags = SLAB_DESTROY_BY_RCU,
2609 .twsk_prot = &tcp_timewait_sock_ops,
2610 .rsk_prot = &tcp_request_sock_ops,
2611 .h.hashinfo = &tcp_hashinfo,
2612 .no_autobind = true,
2613#ifdef CONFIG_COMPAT
2614 .compat_setsockopt = compat_tcp_setsockopt,
2615 .compat_getsockopt = compat_tcp_getsockopt,
2616#endif
2617#ifdef CONFIG_CGROUP_MEM_RES_CTLR_KMEM
2618 .init_cgroup = tcp_init_cgroup,
2619 .destroy_cgroup = tcp_destroy_cgroup,
2620 .proto_cgroup = tcp_proto_cgroup,
2621#endif
2622};
2623EXPORT_SYMBOL(tcp_prot);
2624
2625static int __net_init tcp_sk_init(struct net *net)
2626{
2627 return inet_ctl_sock_create(&net->ipv4.tcp_sock,
2628 PF_INET, SOCK_RAW, IPPROTO_TCP, net);
2629}
2630
2631static void __net_exit tcp_sk_exit(struct net *net)
2632{
2633 inet_ctl_sock_destroy(net->ipv4.tcp_sock);
2634}
2635
2636static void __net_exit tcp_sk_exit_batch(struct list_head *net_exit_list)
2637{
2638 inet_twsk_purge(&tcp_hashinfo, &tcp_death_row, AF_INET);
2639}
2640
2641static struct pernet_operations __net_initdata tcp_sk_ops = {
2642 .init = tcp_sk_init,
2643 .exit = tcp_sk_exit,
2644 .exit_batch = tcp_sk_exit_batch,
2645};
2646
2647void __init tcp_v4_init(void)
2648{
2649 inet_hashinfo_init(&tcp_hashinfo);
2650 if (register_pernet_subsys(&tcp_sk_ops))
2651 panic("Failed to create the TCP control socket.\n");
2652}