?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
ipv6: tcp: fix TCLASS value in ACK messages sent from TIME_WAIT
[linux-2.6.git] / net / ipv6 / tcp_ipv6.c
blob10b2b3165a1aff4ebc6cde947aae64a687b11a5b
1 /*
2 * TCP over IPv6
3 * Linux INET6 implementation
4 *
5 * Authors:
6 * Pedro Roque <roque@di.fc.ul.pt>
7 *
8 * Based on:
9 * linux/net/ipv4/tcp.c
10 * linux/net/ipv4/tcp_input.c
11 * linux/net/ipv4/tcp_output.c
12 *
13 * Fixes:
14 * Hideaki YOSHIFUJI : sin6_scope_id support
15 * YOSHIFUJI Hideaki @USAGI and: Support IPV6_V6ONLY socket option, which
16 * Alexey Kuznetsov allow both IPv4 and IPv6 sockets to bind
17 * a single port at the same time.
18 * YOSHIFUJI Hideaki @USAGI: convert /proc/net/tcp6 to seq_file.
19 *
20 * This program is free software; you can redistribute it and/or
21 * modify it under the terms of the GNU General Public License
22 * as published by the Free Software Foundation; either version
23 * 2 of the License, or (at your option) any later version.
24 */
25
26 #include <linux/bottom_half.h>
27 #include <linux/module.h>
28 #include <linux/errno.h>
29 #include <linux/types.h>
30 #include <linux/socket.h>
31 #include <linux/sockios.h>
32 #include <linux/net.h>
33 #include <linux/jiffies.h>
34 #include <linux/in.h>
35 #include <linux/in6.h>
36 #include <linux/netdevice.h>
37 #include <linux/init.h>
38 #include <linux/jhash.h>
39 #include <linux/ipsec.h>
40 #include <linux/times.h>
41 #include <linux/slab.h>
42
43 #include <linux/ipv6.h>
44 #include <linux/icmpv6.h>
45 #include <linux/random.h>
46
47 #include <net/tcp.h>
48 #include <net/ndisc.h>
49 #include <net/inet6_hashtables.h>
50 #include <net/inet6_connection_sock.h>
51 #include <net/ipv6.h>
52 #include <net/transp_v6.h>
53 #include <net/addrconf.h>
54 #include <net/ip6_route.h>
55 #include <net/ip6_checksum.h>
56 #include <net/inet_ecn.h>
57 #include <net/protocol.h>
58 #include <net/xfrm.h>
59 #include <net/snmp.h>
60 #include <net/dsfield.h>
61 #include <net/timewait_sock.h>
62 #include <net/netdma.h>
63 #include <net/inet_common.h>
64 #include <net/secure_seq.h>
65
66 #include <asm/uaccess.h>
67
68 #include <linux/proc_fs.h>
69 #include <linux/seq_file.h>
70
71 #include <linux/crypto.h>
72 #include <linux/scatterlist.h>
73
74 static void tcp_v6_send_reset(struct sock *sk, struct sk_buff *skb);
75 static void tcp_v6_reqsk_send_ack(struct sock *sk, struct sk_buff *skb,
76 struct request_sock *req);
77
78 static int tcp_v6_do_rcv(struct sock *sk, struct sk_buff *skb);
79 static void __tcp_v6_send_check(struct sk_buff *skb,
80 const struct in6_addr *saddr,
81 const struct in6_addr *daddr);
82
83 static const struct inet_connection_sock_af_ops ipv6_mapped;
84 static const struct inet_connection_sock_af_ops ipv6_specific;
85 #ifdef CONFIG_TCP_MD5SIG
86 static const struct tcp_sock_af_ops tcp_sock_ipv6_specific;
87 static const struct tcp_sock_af_ops tcp_sock_ipv6_mapped_specific;
88 #else
89 static struct tcp_md5sig_key *tcp_v6_md5_do_lookup(struct sock *sk,
90 const struct in6_addr *addr)
91 {
92 return NULL;
93 }
94 #endif
95
96 static void tcp_v6_hash(struct sock *sk)
97 {
98 if (sk->sk_state != TCP_CLOSE) {
99 if (inet_csk(sk)->icsk_af_ops == &ipv6_mapped) {
100 tcp_prot.hash(sk);
101 return;
102 }
103 local_bh_disable();
104 __inet6_hash(sk, NULL);
105 local_bh_enable();
106 }
107 }
108
109 static __inline__ __sum16 tcp_v6_check(int len,
110 const struct in6_addr *saddr,
111 const struct in6_addr *daddr,
112 __wsum base)
113 {
114 return csum_ipv6_magic(saddr, daddr, len, IPPROTO_TCP, base);
115 }
116
117 static __u32 tcp_v6_init_sequence(const struct sk_buff *skb)
118 {
119 return secure_tcpv6_sequence_number(ipv6_hdr(skb)->daddr.s6_addr32,
120 ipv6_hdr(skb)->saddr.s6_addr32,
121 tcp_hdr(skb)->dest,
122 tcp_hdr(skb)->source);
123 }
124
125 static int tcp_v6_connect(struct sock *sk, struct sockaddr *uaddr,
126 int addr_len)
127 {
128 struct sockaddr_in6 *usin = (struct sockaddr_in6 *) uaddr;
129 struct inet_sock *inet = inet_sk(sk);
130 struct inet_connection_sock *icsk = inet_csk(sk);
131 struct ipv6_pinfo *np = inet6_sk(sk);
132 struct tcp_sock *tp = tcp_sk(sk);
133 struct in6_addr *saddr = NULL, *final_p, final;
134 struct rt6_info *rt;
135 struct flowi6 fl6;
136 struct dst_entry *dst;
137 int addr_type;
138 int err;
139
140 if (addr_len < SIN6_LEN_RFC2133)
141 return -EINVAL;
142
143 if (usin->sin6_family != AF_INET6)
144 return -EAFNOSUPPORT;
145
146 memset(&fl6, 0, sizeof(fl6));
147
148 if (np->sndflow) {
149 fl6.flowlabel = usin->sin6_flowinfo&IPV6_FLOWINFO_MASK;
150 IP6_ECN_flow_init(fl6.flowlabel);
151 if (fl6.flowlabel&IPV6_FLOWLABEL_MASK) {
152 struct ip6_flowlabel *flowlabel;
153 flowlabel = fl6_sock_lookup(sk, fl6.flowlabel);
154 if (flowlabel == NULL)
155 return -EINVAL;
156 ipv6_addr_copy(&usin->sin6_addr, &flowlabel->dst);
157 fl6_sock_release(flowlabel);
158 }
159 }
160
161 /*
162 * connect() to INADDR_ANY means loopback (BSD'ism).
163 */
164
165 if(ipv6_addr_any(&usin->sin6_addr))
166 usin->sin6_addr.s6_addr[15] = 0x1;
167
168 addr_type = ipv6_addr_type(&usin->sin6_addr);
169
170 if(addr_type & IPV6_ADDR_MULTICAST)
171 return -ENETUNREACH;
172
173 if (addr_type&IPV6_ADDR_LINKLOCAL) {
174 if (addr_len >= sizeof(struct sockaddr_in6) &&
175 usin->sin6_scope_id) {
176 /* If interface is set while binding, indices
177 * must coincide.
178 */
179 if (sk->sk_bound_dev_if &&
180 sk->sk_bound_dev_if != usin->sin6_scope_id)
181 return -EINVAL;
182
183 sk->sk_bound_dev_if = usin->sin6_scope_id;
184 }
185
186 /* Connect to link-local address requires an interface */
187 if (!sk->sk_bound_dev_if)
188 return -EINVAL;
189 }
190
191 if (tp->rx_opt.ts_recent_stamp &&
192 !ipv6_addr_equal(&np->daddr, &usin->sin6_addr)) {
193 tp->rx_opt.ts_recent = 0;
194 tp->rx_opt.ts_recent_stamp = 0;
195 tp->write_seq = 0;
196 }
197
198 ipv6_addr_copy(&np->daddr, &usin->sin6_addr);
199 np->flow_label = fl6.flowlabel;
200
201 /*
202 * TCP over IPv4
203 */
204
205 if (addr_type == IPV6_ADDR_MAPPED) {
206 u32 exthdrlen = icsk->icsk_ext_hdr_len;
207 struct sockaddr_in sin;
208
209 SOCK_DEBUG(sk, "connect: ipv4 mapped\n");
210
211 if (__ipv6_only_sock(sk))
212 return -ENETUNREACH;
213
214 sin.sin_family = AF_INET;
215 sin.sin_port = usin->sin6_port;
216 sin.sin_addr.s_addr = usin->sin6_addr.s6_addr32[3];
217
218 icsk->icsk_af_ops = &ipv6_mapped;
219 sk->sk_backlog_rcv = tcp_v4_do_rcv;
220 #ifdef CONFIG_TCP_MD5SIG
221 tp->af_specific = &tcp_sock_ipv6_mapped_specific;
222 #endif
223
224 err = tcp_v4_connect(sk, (struct sockaddr *)&sin, sizeof(sin));
225
226 if (err) {
227 icsk->icsk_ext_hdr_len = exthdrlen;
228 icsk->icsk_af_ops = &ipv6_specific;
229 sk->sk_backlog_rcv = tcp_v6_do_rcv;
230 #ifdef CONFIG_TCP_MD5SIG
231 tp->af_specific = &tcp_sock_ipv6_specific;
232 #endif
233 goto failure;
234 } else {
235 ipv6_addr_set_v4mapped(inet->inet_saddr, &np->saddr);
236 ipv6_addr_set_v4mapped(inet->inet_rcv_saddr,
237 &np->rcv_saddr);
238 }
239
240 return err;
241 }
242
243 if (!ipv6_addr_any(&np->rcv_saddr))
244 saddr = &np->rcv_saddr;
245
246 fl6.flowi6_proto = IPPROTO_TCP;
247 ipv6_addr_copy(&fl6.daddr, &np->daddr);
248 ipv6_addr_copy(&fl6.saddr,
249 (saddr ? saddr : &np->saddr));
250 fl6.flowi6_oif = sk->sk_bound_dev_if;
251 fl6.flowi6_mark = sk->sk_mark;
252 fl6.fl6_dport = usin->sin6_port;
253 fl6.fl6_sport = inet->inet_sport;
254
255 final_p = fl6_update_dst(&fl6, np->opt, &final);
256
257 security_sk_classify_flow(sk, flowi6_to_flowi(&fl6));
258
259 dst = ip6_dst_lookup_flow(sk, &fl6, final_p, true);
260 if (IS_ERR(dst)) {
261 err = PTR_ERR(dst);
262 goto failure;
263 }
264
265 if (saddr == NULL) {
266 saddr = &fl6.saddr;
267 ipv6_addr_copy(&np->rcv_saddr, saddr);
268 }
269
270 /* set the source address */
271 ipv6_addr_copy(&np->saddr, saddr);
272 inet->inet_rcv_saddr = LOOPBACK4_IPV6;
273
274 sk->sk_gso_type = SKB_GSO_TCPV6;
275 __ip6_dst_store(sk, dst, NULL, NULL);
276
277 rt = (struct rt6_info *) dst;
278 if (tcp_death_row.sysctl_tw_recycle &&
279 !tp->rx_opt.ts_recent_stamp &&
280 ipv6_addr_equal(&rt->rt6i_dst.addr, &np->daddr)) {
281 struct inet_peer *peer = rt6_get_peer(rt);
282 /*
283 * VJ's idea. We save last timestamp seen from
284 * the destination in peer table, when entering state
285 * TIME-WAIT * and initialize rx_opt.ts_recent from it,
286 * when trying new connection.
287 */
288 if (peer) {
289 inet_peer_refcheck(peer);
290 if ((u32)get_seconds() - peer->tcp_ts_stamp <= TCP_PAWS_MSL) {
291 tp->rx_opt.ts_recent_stamp = peer->tcp_ts_stamp;
292 tp->rx_opt.ts_recent = peer->tcp_ts;
293 }
294 }
295 }
296
297 icsk->icsk_ext_hdr_len = 0;
298 if (np->opt)
299 icsk->icsk_ext_hdr_len = (np->opt->opt_flen +
300 np->opt->opt_nflen);
301
302 tp->rx_opt.mss_clamp = IPV6_MIN_MTU - sizeof(struct tcphdr) - sizeof(struct ipv6hdr);
303
304 inet->inet_dport = usin->sin6_port;
305
306 tcp_set_state(sk, TCP_SYN_SENT);
307 err = inet6_hash_connect(&tcp_death_row, sk);
308 if (err)
309 goto late_failure;
310
311 if (!tp->write_seq)
312 tp->write_seq = secure_tcpv6_sequence_number(np->saddr.s6_addr32,
313 np->daddr.s6_addr32,
314 inet->inet_sport,
315 inet->inet_dport);
316
317 err = tcp_connect(sk);
318 if (err)
319 goto late_failure;
320
321 return 0;
322
323 late_failure:
324 tcp_set_state(sk, TCP_CLOSE);
325 __sk_dst_reset(sk);
326 failure:
327 inet->inet_dport = 0;
328 sk->sk_route_caps = 0;
329 return err;
330 }
331
332 static void tcp_v6_err(struct sk_buff *skb, struct inet6_skb_parm *opt,
333 u8 type, u8 code, int offset, __be32 info)
334 {
335 const struct ipv6hdr *hdr = (const struct ipv6hdr*)skb->data;
336 const struct tcphdr *th = (struct tcphdr *)(skb->data+offset);
337 struct ipv6_pinfo *np;
338 struct sock *sk;
339 int err;
340 struct tcp_sock *tp;
341 __u32 seq;
342 struct net *net = dev_net(skb->dev);
343
344 sk = inet6_lookup(net, &tcp_hashinfo, &hdr->daddr,
345 th->dest, &hdr->saddr, th->source, skb->dev->ifindex);
346
347 if (sk == NULL) {
348 ICMP6_INC_STATS_BH(net, __in6_dev_get(skb->dev),
349 ICMP6_MIB_INERRORS);
350 return;
351 }
352
353 if (sk->sk_state == TCP_TIME_WAIT) {
354 inet_twsk_put(inet_twsk(sk));
355 return;
356 }
357
358 bh_lock_sock(sk);
359 if (sock_owned_by_user(sk))
360 NET_INC_STATS_BH(net, LINUX_MIB_LOCKDROPPEDICMPS);
361
362 if (sk->sk_state == TCP_CLOSE)
363 goto out;
364
365 if (ipv6_hdr(skb)->hop_limit < inet6_sk(sk)->min_hopcount) {
366 NET_INC_STATS_BH(net, LINUX_MIB_TCPMINTTLDROP);
367 goto out;
368 }
369
370 tp = tcp_sk(sk);
371 seq = ntohl(th->seq);
372 if (sk->sk_state != TCP_LISTEN &&
373 !between(seq, tp->snd_una, tp->snd_nxt)) {
374 NET_INC_STATS_BH(net, LINUX_MIB_OUTOFWINDOWICMPS);
375 goto out;
376 }
377
378 np = inet6_sk(sk);
379
380 if (type == ICMPV6_PKT_TOOBIG) {
381 struct dst_entry *dst;
382
383 if (sock_owned_by_user(sk))
384 goto out;
385 if ((1 << sk->sk_state) & (TCPF_LISTEN | TCPF_CLOSE))
386 goto out;
387
388 /* icmp should have updated the destination cache entry */
389 dst = __sk_dst_check(sk, np->dst_cookie);
390
391 if (dst == NULL) {
392 struct inet_sock *inet = inet_sk(sk);
393 struct flowi6 fl6;
394
395 /* BUGGG_FUTURE: Again, it is not clear how
396 to handle rthdr case. Ignore this complexity
397 for now.
398 */
399 memset(&fl6, 0, sizeof(fl6));
400 fl6.flowi6_proto = IPPROTO_TCP;
401 ipv6_addr_copy(&fl6.daddr, &np->daddr);
402 ipv6_addr_copy(&fl6.saddr, &np->saddr);
403 fl6.flowi6_oif = sk->sk_bound_dev_if;
404 fl6.flowi6_mark = sk->sk_mark;
405 fl6.fl6_dport = inet->inet_dport;
406 fl6.fl6_sport = inet->inet_sport;
407 security_skb_classify_flow(skb, flowi6_to_flowi(&fl6));
408
409 dst = ip6_dst_lookup_flow(sk, &fl6, NULL, false);
410 if (IS_ERR(dst)) {
411 sk->sk_err_soft = -PTR_ERR(dst);
412 goto out;
413 }
414
415 } else
416 dst_hold(dst);
417
418 if (inet_csk(sk)->icsk_pmtu_cookie > dst_mtu(dst)) {
419 tcp_sync_mss(sk, dst_mtu(dst));
420 tcp_simple_retransmit(sk);
421 } /* else let the usual retransmit timer handle it */
422 dst_release(dst);
423 goto out;
424 }
425
426 icmpv6_err_convert(type, code, &err);
427
428 /* Might be for an request_sock */
429 switch (sk->sk_state) {
430 struct request_sock *req, **prev;
431 case TCP_LISTEN:
432 if (sock_owned_by_user(sk))
433 goto out;
434
435 req = inet6_csk_search_req(sk, &prev, th->dest, &hdr->daddr,
436 &hdr->saddr, inet6_iif(skb));
437 if (!req)
438 goto out;
439
440 /* ICMPs are not backlogged, hence we cannot get
441 * an established socket here.
442 */
443 WARN_ON(req->sk != NULL);
444
445 if (seq != tcp_rsk(req)->snt_isn) {
446 NET_INC_STATS_BH(net, LINUX_MIB_OUTOFWINDOWICMPS);
447 goto out;
448 }
449
450 inet_csk_reqsk_queue_drop(sk, req, prev);
451 goto out;
452
453 case TCP_SYN_SENT:
454 case TCP_SYN_RECV: /* Cannot happen.
455 It can, it SYNs are crossed. --ANK */
456 if (!sock_owned_by_user(sk)) {
457 sk->sk_err = err;
458 sk->sk_error_report(sk); /* Wake people up to see the error (see connect in sock.c) */
459
460 tcp_done(sk);
461 } else
462 sk->sk_err_soft = err;
463 goto out;
464 }
465
466 if (!sock_owned_by_user(sk) && np->recverr) {
467 sk->sk_err = err;
468 sk->sk_error_report(sk);
469 } else
470 sk->sk_err_soft = err;
471
472 out:
473 bh_unlock_sock(sk);
474 sock_put(sk);
475 }
476
477
478 static int tcp_v6_send_synack(struct sock *sk, struct request_sock *req,
479 struct request_values *rvp)
480 {
481 struct inet6_request_sock *treq = inet6_rsk(req);
482 struct ipv6_pinfo *np = inet6_sk(sk);
483 struct sk_buff * skb;
484 struct ipv6_txoptions *opt = NULL;
485 struct in6_addr * final_p, final;
486 struct flowi6 fl6;
487 struct dst_entry *dst;
488 int err;
489
490 memset(&fl6, 0, sizeof(fl6));
491 fl6.flowi6_proto = IPPROTO_TCP;
492 ipv6_addr_copy(&fl6.daddr, &treq->rmt_addr);
493 ipv6_addr_copy(&fl6.saddr, &treq->loc_addr);
494 fl6.flowlabel = 0;
495 fl6.flowi6_oif = treq->iif;
496 fl6.flowi6_mark = sk->sk_mark;
497 fl6.fl6_dport = inet_rsk(req)->rmt_port;
498 fl6.fl6_sport = inet_rsk(req)->loc_port;
499 security_req_classify_flow(req, flowi6_to_flowi(&fl6));
500
501 opt = np->opt;
502 final_p = fl6_update_dst(&fl6, opt, &final);
503
504 dst = ip6_dst_lookup_flow(sk, &fl6, final_p, false);
505 if (IS_ERR(dst)) {
506 err = PTR_ERR(dst);
507 dst = NULL;
508 goto done;
509 }
510 skb = tcp_make_synack(sk, dst, req, rvp);
511 err = -ENOMEM;
512 if (skb) {
513 __tcp_v6_send_check(skb, &treq->loc_addr, &treq->rmt_addr);
514
515 ipv6_addr_copy(&fl6.daddr, &treq->rmt_addr);
516 err = ip6_xmit(sk, skb, &fl6, opt, np->tclass);
517 err = net_xmit_eval(err);
518 }
519
520 done:
521 if (opt && opt != np->opt)
522 sock_kfree_s(sk, opt, opt->tot_len);
523 dst_release(dst);
524 return err;
525 }
526
527 static int tcp_v6_rtx_synack(struct sock *sk, struct request_sock *req,
528 struct request_values *rvp)
529 {
530 TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_RETRANSSEGS);
531 return tcp_v6_send_synack(sk, req, rvp);
532 }
533
534 static void tcp_v6_reqsk_destructor(struct request_sock *req)
535 {
536 kfree_skb(inet6_rsk(req)->pktopts);
537 }
538
539 #ifdef CONFIG_TCP_MD5SIG
540 static struct tcp_md5sig_key *tcp_v6_md5_do_lookup(struct sock *sk,
541 const struct in6_addr *addr)
542 {
543 struct tcp_sock *tp = tcp_sk(sk);
544 int i;
545
546 BUG_ON(tp == NULL);
547
548 if (!tp->md5sig_info || !tp->md5sig_info->entries6)
549 return NULL;
550
551 for (i = 0; i < tp->md5sig_info->entries6; i++) {
552 if (ipv6_addr_equal(&tp->md5sig_info->keys6[i].addr, addr))
553 return &tp->md5sig_info->keys6[i].base;
554 }
555 return NULL;
556 }
557
558 static struct tcp_md5sig_key *tcp_v6_md5_lookup(struct sock *sk,
559 struct sock *addr_sk)
560 {
561 return tcp_v6_md5_do_lookup(sk, &inet6_sk(addr_sk)->daddr);
562 }
563
564 static struct tcp_md5sig_key *tcp_v6_reqsk_md5_lookup(struct sock *sk,
565 struct request_sock *req)
566 {
567 return tcp_v6_md5_do_lookup(sk, &inet6_rsk(req)->rmt_addr);
568 }
569
570 static int tcp_v6_md5_do_add(struct sock *sk, const struct in6_addr *peer,
571 char *newkey, u8 newkeylen)
572 {
573 /* Add key to the list */
574 struct tcp_md5sig_key *key;
575 struct tcp_sock *tp = tcp_sk(sk);
576 struct tcp6_md5sig_key *keys;
577
578 key = tcp_v6_md5_do_lookup(sk, peer);
579 if (key) {
580 /* modify existing entry - just update that one */
581 kfree(key->key);
582 key->key = newkey;
583 key->keylen = newkeylen;
584 } else {
585 /* reallocate new list if current one is full. */
586 if (!tp->md5sig_info) {
587 tp->md5sig_info = kzalloc(sizeof(*tp->md5sig_info), GFP_ATOMIC);
588 if (!tp->md5sig_info) {
589 kfree(newkey);
590 return -ENOMEM;
591 }
592 sk_nocaps_add(sk, NETIF_F_GSO_MASK);
593 }
594 if (tp->md5sig_info->entries6 == 0 &&
595 tcp_alloc_md5sig_pool(sk) == NULL) {
596 kfree(newkey);
597 return -ENOMEM;
598 }
599 if (tp->md5sig_info->alloced6 == tp->md5sig_info->entries6) {
600 keys = kmalloc((sizeof (tp->md5sig_info->keys6[0]) *
601 (tp->md5sig_info->entries6 + 1)), GFP_ATOMIC);
602
603 if (!keys) {
604 kfree(newkey);
605 if (tp->md5sig_info->entries6 == 0)
606 tcp_free_md5sig_pool();
607 return -ENOMEM;
608 }
609
610 if (tp->md5sig_info->entries6)
611 memmove(keys, tp->md5sig_info->keys6,
612 (sizeof (tp->md5sig_info->keys6[0]) *
613 tp->md5sig_info->entries6));
614
615 kfree(tp->md5sig_info->keys6);
616 tp->md5sig_info->keys6 = keys;
617 tp->md5sig_info->alloced6++;
618 }
619
620 ipv6_addr_copy(&tp->md5sig_info->keys6[tp->md5sig_info->entries6].addr,
621 peer);
622 tp->md5sig_info->keys6[tp->md5sig_info->entries6].base.key = newkey;
623 tp->md5sig_info->keys6[tp->md5sig_info->entries6].base.keylen = newkeylen;
624
625 tp->md5sig_info->entries6++;
626 }
627 return 0;
628 }
629
630 static int tcp_v6_md5_add_func(struct sock *sk, struct sock *addr_sk,
631 u8 *newkey, __u8 newkeylen)
632 {
633 return tcp_v6_md5_do_add(sk, &inet6_sk(addr_sk)->daddr,
634 newkey, newkeylen);
635 }
636
637 static int tcp_v6_md5_do_del(struct sock *sk, const struct in6_addr *peer)
638 {
639 struct tcp_sock *tp = tcp_sk(sk);
640 int i;
641
642 for (i = 0; i < tp->md5sig_info->entries6; i++) {
643 if (ipv6_addr_equal(&tp->md5sig_info->keys6[i].addr, peer)) {
644 /* Free the key */
645 kfree(tp->md5sig_info->keys6[i].base.key);
646 tp->md5sig_info->entries6--;
647
648 if (tp->md5sig_info->entries6 == 0) {
649 kfree(tp->md5sig_info->keys6);
650 tp->md5sig_info->keys6 = NULL;
651 tp->md5sig_info->alloced6 = 0;
652 tcp_free_md5sig_pool();
653 } else {
654 /* shrink the database */
655 if (tp->md5sig_info->entries6 != i)
656 memmove(&tp->md5sig_info->keys6[i],
657 &tp->md5sig_info->keys6[i+1],
658 (tp->md5sig_info->entries6 - i)
659 * sizeof (tp->md5sig_info->keys6[0]));
660 }
661 return 0;
662 }
663 }
664 return -ENOENT;
665 }
666
667 static void tcp_v6_clear_md5_list (struct sock *sk)
668 {
669 struct tcp_sock *tp = tcp_sk(sk);
670 int i;
671
672 if (tp->md5sig_info->entries6) {
673 for (i = 0; i < tp->md5sig_info->entries6; i++)
674 kfree(tp->md5sig_info->keys6[i].base.key);
675 tp->md5sig_info->entries6 = 0;
676 tcp_free_md5sig_pool();
677 }
678
679 kfree(tp->md5sig_info->keys6);
680 tp->md5sig_info->keys6 = NULL;
681 tp->md5sig_info->alloced6 = 0;
682
683 if (tp->md5sig_info->entries4) {
684 for (i = 0; i < tp->md5sig_info->entries4; i++)
685 kfree(tp->md5sig_info->keys4[i].base.key);
686 tp->md5sig_info->entries4 = 0;
687 tcp_free_md5sig_pool();
688 }
689
690 kfree(tp->md5sig_info->keys4);
691 tp->md5sig_info->keys4 = NULL;
692 tp->md5sig_info->alloced4 = 0;
693 }
694
695 static int tcp_v6_parse_md5_keys (struct sock *sk, char __user *optval,
696 int optlen)
697 {
698 struct tcp_md5sig cmd;
699 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)&cmd.tcpm_addr;
700 u8 *newkey;
701
702 if (optlen < sizeof(cmd))
703 return -EINVAL;
704
705 if (copy_from_user(&cmd, optval, sizeof(cmd)))
706 return -EFAULT;
707
708 if (sin6->sin6_family != AF_INET6)
709 return -EINVAL;
710
711 if (!cmd.tcpm_keylen) {
712 if (!tcp_sk(sk)->md5sig_info)
713 return -ENOENT;
714 if (ipv6_addr_v4mapped(&sin6->sin6_addr))
715 return tcp_v4_md5_do_del(sk, sin6->sin6_addr.s6_addr32[3]);
716 return tcp_v6_md5_do_del(sk, &sin6->sin6_addr);
717 }
718
719 if (cmd.tcpm_keylen > TCP_MD5SIG_MAXKEYLEN)
720 return -EINVAL;
721
722 if (!tcp_sk(sk)->md5sig_info) {
723 struct tcp_sock *tp = tcp_sk(sk);
724 struct tcp_md5sig_info *p;
725
726 p = kzalloc(sizeof(struct tcp_md5sig_info), GFP_KERNEL);
727 if (!p)
728 return -ENOMEM;
729
730 tp->md5sig_info = p;
731 sk_nocaps_add(sk, NETIF_F_GSO_MASK);
732 }
733
734 newkey = kmemdup(cmd.tcpm_key, cmd.tcpm_keylen, GFP_KERNEL);
735 if (!newkey)
736 return -ENOMEM;
737 if (ipv6_addr_v4mapped(&sin6->sin6_addr)) {
738 return tcp_v4_md5_do_add(sk, sin6->sin6_addr.s6_addr32[3],
739 newkey, cmd.tcpm_keylen);
740 }
741 return tcp_v6_md5_do_add(sk, &sin6->sin6_addr, newkey, cmd.tcpm_keylen);
742 }
743
744 static int tcp_v6_md5_hash_pseudoheader(struct tcp_md5sig_pool *hp,
745 const struct in6_addr *daddr,
746 const struct in6_addr *saddr, int nbytes)
747 {
748 struct tcp6_pseudohdr *bp;
749 struct scatterlist sg;
750
751 bp = &hp->md5_blk.ip6;
752 /* 1. TCP pseudo-header (RFC2460) */
753 ipv6_addr_copy(&bp->saddr, saddr);
754 ipv6_addr_copy(&bp->daddr, daddr);
755 bp->protocol = cpu_to_be32(IPPROTO_TCP);
756 bp->len = cpu_to_be32(nbytes);
757
758 sg_init_one(&sg, bp, sizeof(*bp));
759 return crypto_hash_update(&hp->md5_desc, &sg, sizeof(*bp));
760 }
761
762 static int tcp_v6_md5_hash_hdr(char *md5_hash, struct tcp_md5sig_key *key,
763 const struct in6_addr *daddr, struct in6_addr *saddr,
764 const struct tcphdr *th)
765 {
766 struct tcp_md5sig_pool *hp;
767 struct hash_desc *desc;
768
769 hp = tcp_get_md5sig_pool();
770 if (!hp)
771 goto clear_hash_noput;
772 desc = &hp->md5_desc;
773
774 if (crypto_hash_init(desc))
775 goto clear_hash;
776 if (tcp_v6_md5_hash_pseudoheader(hp, daddr, saddr, th->doff << 2))
777 goto clear_hash;
778 if (tcp_md5_hash_header(hp, th))
779 goto clear_hash;
780 if (tcp_md5_hash_key(hp, key))
781 goto clear_hash;
782 if (crypto_hash_final(desc, md5_hash))
783 goto clear_hash;
784
785 tcp_put_md5sig_pool();
786 return 0;
787
788 clear_hash:
789 tcp_put_md5sig_pool();
790 clear_hash_noput:
791 memset(md5_hash, 0, 16);
792 return 1;
793 }
794
795 static int tcp_v6_md5_hash_skb(char *md5_hash, struct tcp_md5sig_key *key,
796 const struct sock *sk,
797 const struct request_sock *req,
798 const struct sk_buff *skb)
799 {
800 const struct in6_addr *saddr, *daddr;
801 struct tcp_md5sig_pool *hp;
802 struct hash_desc *desc;
803 const struct tcphdr *th = tcp_hdr(skb);
804
805 if (sk) {
806 saddr = &inet6_sk(sk)->saddr;
807 daddr = &inet6_sk(sk)->daddr;
808 } else if (req) {
809 saddr = &inet6_rsk(req)->loc_addr;
810 daddr = &inet6_rsk(req)->rmt_addr;
811 } else {
812 const struct ipv6hdr *ip6h = ipv6_hdr(skb);
813 saddr = &ip6h->saddr;
814 daddr = &ip6h->daddr;
815 }
816
817 hp = tcp_get_md5sig_pool();
818 if (!hp)
819 goto clear_hash_noput;
820 desc = &hp->md5_desc;
821
822 if (crypto_hash_init(desc))
823 goto clear_hash;
824
825 if (tcp_v6_md5_hash_pseudoheader(hp, daddr, saddr, skb->len))
826 goto clear_hash;
827 if (tcp_md5_hash_header(hp, th))
828 goto clear_hash;
829 if (tcp_md5_hash_skb_data(hp, skb, th->doff << 2))
830 goto clear_hash;
831 if (tcp_md5_hash_key(hp, key))
832 goto clear_hash;
833 if (crypto_hash_final(desc, md5_hash))
834 goto clear_hash;
835
836 tcp_put_md5sig_pool();
837 return 0;
838
839 clear_hash:
840 tcp_put_md5sig_pool();
841 clear_hash_noput:
842 memset(md5_hash, 0, 16);
843 return 1;
844 }
845
846 static int tcp_v6_inbound_md5_hash(struct sock *sk, const struct sk_buff *skb)
847 {
848 const __u8 *hash_location = NULL;
849 struct tcp_md5sig_key *hash_expected;
850 const struct ipv6hdr *ip6h = ipv6_hdr(skb);
851 const struct tcphdr *th = tcp_hdr(skb);
852 int genhash;
853 u8 newhash[16];
854
855 hash_expected = tcp_v6_md5_do_lookup(sk, &ip6h->saddr);
856 hash_location = tcp_parse_md5sig_option(th);
857
858 /* We've parsed the options - do we have a hash? */
859 if (!hash_expected && !hash_location)
860 return 0;
861
862 if (hash_expected && !hash_location) {
863 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPMD5NOTFOUND);
864 return 1;
865 }
866
867 if (!hash_expected && hash_location) {
868 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPMD5UNEXPECTED);
869 return 1;
870 }
871
872 /* check the signature */
873 genhash = tcp_v6_md5_hash_skb(newhash,
874 hash_expected,
875 NULL, NULL, skb);
876
877 if (genhash || memcmp(hash_location, newhash, 16) != 0) {
878 if (net_ratelimit()) {
879 printk(KERN_INFO "MD5 Hash %s for [%pI6c]:%u->[%pI6c]:%u\n",
880 genhash ? "failed" : "mismatch",
881 &ip6h->saddr, ntohs(th->source),
882 &ip6h->daddr, ntohs(th->dest));
883 }
884 return 1;
885 }
886 return 0;
887 }
888 #endif
889
890 struct request_sock_ops tcp6_request_sock_ops __read_mostly = {
891 .family = AF_INET6,
892 .obj_size = sizeof(struct tcp6_request_sock),
893 .rtx_syn_ack = tcp_v6_rtx_synack,
894 .send_ack = tcp_v6_reqsk_send_ack,
895 .destructor = tcp_v6_reqsk_destructor,
896 .send_reset = tcp_v6_send_reset,
897 .syn_ack_timeout = tcp_syn_ack_timeout,
898 };
899
900 #ifdef CONFIG_TCP_MD5SIG
901 static const struct tcp_request_sock_ops tcp_request_sock_ipv6_ops = {
902 .md5_lookup = tcp_v6_reqsk_md5_lookup,
903 .calc_md5_hash = tcp_v6_md5_hash_skb,
904 };
905 #endif
906
907 static void __tcp_v6_send_check(struct sk_buff *skb,
908 const struct in6_addr *saddr, const struct in6_addr *daddr)
909 {
910 struct tcphdr *th = tcp_hdr(skb);
911
912 if (skb->ip_summed == CHECKSUM_PARTIAL) {
913 th->check = ~tcp_v6_check(skb->len, saddr, daddr, 0);
914 skb->csum_start = skb_transport_header(skb) - skb->head;
915 skb->csum_offset = offsetof(struct tcphdr, check);
916 } else {
917 th->check = tcp_v6_check(skb->len, saddr, daddr,
918 csum_partial(th, th->doff << 2,
919 skb->csum));
920 }
921 }
922
923 static void tcp_v6_send_check(struct sock *sk, struct sk_buff *skb)
924 {
925 struct ipv6_pinfo *np = inet6_sk(sk);
926
927 __tcp_v6_send_check(skb, &np->saddr, &np->daddr);
928 }
929
930 static int tcp_v6_gso_send_check(struct sk_buff *skb)
931 {
932 const struct ipv6hdr *ipv6h;
933 struct tcphdr *th;
934
935 if (!pskb_may_pull(skb, sizeof(*th)))
936 return -EINVAL;
937
938 ipv6h = ipv6_hdr(skb);
939 th = tcp_hdr(skb);
940
941 th->check = 0;
942 skb->ip_summed = CHECKSUM_PARTIAL;
943 __tcp_v6_send_check(skb, &ipv6h->saddr, &ipv6h->daddr);
944 return 0;
945 }
946
947 static struct sk_buff **tcp6_gro_receive(struct sk_buff **head,
948 struct sk_buff *skb)
949 {
950 const struct ipv6hdr *iph = skb_gro_network_header(skb);
951
952 switch (skb->ip_summed) {
953 case CHECKSUM_COMPLETE:
954 if (!tcp_v6_check(skb_gro_len(skb), &iph->saddr, &iph->daddr,
955 skb->csum)) {
956 skb->ip_summed = CHECKSUM_UNNECESSARY;
957 break;
958 }
959
960 /* fall through */
961 case CHECKSUM_NONE:
962 NAPI_GRO_CB(skb)->flush = 1;
963 return NULL;
964 }
965
966 return tcp_gro_receive(head, skb);
967 }
968
969 static int tcp6_gro_complete(struct sk_buff *skb)
970 {
971 const struct ipv6hdr *iph = ipv6_hdr(skb);
972 struct tcphdr *th = tcp_hdr(skb);
973
974 th->check = ~tcp_v6_check(skb->len - skb_transport_offset(skb),
975 &iph->saddr, &iph->daddr, 0);
976 skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6;
977
978 return tcp_gro_complete(skb);
979 }
980
981 static void tcp_v6_send_response(struct sk_buff *skb, u32 seq, u32 ack, u32 win,
982 u32 ts, struct tcp_md5sig_key *key, int rst, u8 tclass)
983 {
984 const struct tcphdr *th = tcp_hdr(skb);
985 struct tcphdr *t1;
986 struct sk_buff *buff;
987 struct flowi6 fl6;
988 struct net *net = dev_net(skb_dst(skb)->dev);
989 struct sock *ctl_sk = net->ipv6.tcp_sk;
990 unsigned int tot_len = sizeof(struct tcphdr);
991 struct dst_entry *dst;
992 __be32 *topt;
993
994 if (ts)
995 tot_len += TCPOLEN_TSTAMP_ALIGNED;
996 #ifdef CONFIG_TCP_MD5SIG
997 if (key)
998 tot_len += TCPOLEN_MD5SIG_ALIGNED;
999 #endif
1000
1001 buff = alloc_skb(MAX_HEADER + sizeof(struct ipv6hdr) + tot_len,
1002 GFP_ATOMIC);
1003 if (buff == NULL)
1004 return;
1005
1006 skb_reserve(buff, MAX_HEADER + sizeof(struct ipv6hdr) + tot_len);
1007
1008 t1 = (struct tcphdr *) skb_push(buff, tot_len);
1009 skb_reset_transport_header(buff);
1010
1011 /* Swap the send and the receive. */
1012 memset(t1, 0, sizeof(*t1));
1013 t1->dest = th->source;
1014 t1->source = th->dest;
1015 t1->doff = tot_len / 4;
1016 t1->seq = htonl(seq);
1017 t1->ack_seq = htonl(ack);
1018 t1->ack = !rst || !th->ack;
1019 t1->rst = rst;
1020 t1->window = htons(win);
1021
1022 topt = (__be32 *)(t1 + 1);
1023
1024 if (ts) {
1025 *topt++ = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
1026 (TCPOPT_TIMESTAMP << 8) | TCPOLEN_TIMESTAMP);
1027 *topt++ = htonl(tcp_time_stamp);
1028 *topt++ = htonl(ts);
1029 }
1030
1031 #ifdef CONFIG_TCP_MD5SIG
1032 if (key) {
1033 *topt++ = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
1034 (TCPOPT_MD5SIG << 8) | TCPOLEN_MD5SIG);
1035 tcp_v6_md5_hash_hdr((__u8 *)topt, key,
1036 &ipv6_hdr(skb)->saddr,
1037 &ipv6_hdr(skb)->daddr, t1);
1038 }
1039 #endif
1040
1041 memset(&fl6, 0, sizeof(fl6));
1042 ipv6_addr_copy(&fl6.daddr, &ipv6_hdr(skb)->saddr);
1043 ipv6_addr_copy(&fl6.saddr, &ipv6_hdr(skb)->daddr);
1044
1045 buff->ip_summed = CHECKSUM_PARTIAL;
1046 buff->csum = 0;
1047
1048 __tcp_v6_send_check(buff, &fl6.saddr, &fl6.daddr);
1049
1050 fl6.flowi6_proto = IPPROTO_TCP;
1051 fl6.flowi6_oif = inet6_iif(skb);
1052 fl6.fl6_dport = t1->dest;
1053 fl6.fl6_sport = t1->source;
1054 security_skb_classify_flow(skb, flowi6_to_flowi(&fl6));
1055
1056 /* Pass a socket to ip6_dst_lookup either it is for RST
1057 * Underlying function will use this to retrieve the network
1058 * namespace
1059 */
1060 dst = ip6_dst_lookup_flow(ctl_sk, &fl6, NULL, false);
1061 if (!IS_ERR(dst)) {
1062 skb_dst_set(buff, dst);
1063 ip6_xmit(ctl_sk, buff, &fl6, NULL, tclass);
1064 TCP_INC_STATS_BH(net, TCP_MIB_OUTSEGS);
1065 if (rst)
1066 TCP_INC_STATS_BH(net, TCP_MIB_OUTRSTS);
1067 return;
1068 }
1069
1070 kfree_skb(buff);
1071 }
1072
1073 static void tcp_v6_send_reset(struct sock *sk, struct sk_buff *skb)
1074 {
1075 const struct tcphdr *th = tcp_hdr(skb);
1076 u32 seq = 0, ack_seq = 0;
1077 struct tcp_md5sig_key *key = NULL;
1078
1079 if (th->rst)
1080 return;
1081
1082 if (!ipv6_unicast_destination(skb))
1083 return;
1084
1085 #ifdef CONFIG_TCP_MD5SIG
1086 if (sk)
1087 key = tcp_v6_md5_do_lookup(sk, &ipv6_hdr(skb)->daddr);
1088 #endif
1089
1090 if (th->ack)
1091 seq = ntohl(th->ack_seq);
1092 else
1093 ack_seq = ntohl(th->seq) + th->syn + th->fin + skb->len -
1094 (th->doff << 2);
1095
1096 tcp_v6_send_response(skb, seq, ack_seq, 0, 0, key, 1, 0);
1097 }
1098
1099 static void tcp_v6_send_ack(struct sk_buff *skb, u32 seq, u32 ack, u32 win, u32 ts,
1100 struct tcp_md5sig_key *key, u8 tclass)
1101 {
1102 tcp_v6_send_response(skb, seq, ack, win, ts, key, 0, tclass);
1103 }
1104
1105 static void tcp_v6_timewait_ack(struct sock *sk, struct sk_buff *skb)
1106 {
1107 struct inet_timewait_sock *tw = inet_twsk(sk);
1108 struct tcp_timewait_sock *tcptw = tcp_twsk(sk);
1109
1110 tcp_v6_send_ack(skb, tcptw->tw_snd_nxt, tcptw->tw_rcv_nxt,
1111 tcptw->tw_rcv_wnd >> tw->tw_rcv_wscale,
1112 tcptw->tw_ts_recent, tcp_twsk_md5_key(tcptw),
1113 tw->tw_tclass);
1114
1115 inet_twsk_put(tw);
1116 }
1117
1118 static void tcp_v6_reqsk_send_ack(struct sock *sk, struct sk_buff *skb,
1119 struct request_sock *req)
1120 {
1121 tcp_v6_send_ack(skb, tcp_rsk(req)->snt_isn + 1, tcp_rsk(req)->rcv_isn + 1, req->rcv_wnd, req->ts_recent,
1122 tcp_v6_md5_do_lookup(sk, &ipv6_hdr(skb)->daddr), 0);
1123 }
1124
1125
1126 static struct sock *tcp_v6_hnd_req(struct sock *sk,struct sk_buff *skb)
1127 {
1128 struct request_sock *req, **prev;
1129 const struct tcphdr *th = tcp_hdr(skb);
1130 struct sock *nsk;
1131
1132 /* Find possible connection requests. */
1133 req = inet6_csk_search_req(sk, &prev, th->source,
1134 &ipv6_hdr(skb)->saddr,
1135 &ipv6_hdr(skb)->daddr, inet6_iif(skb));
1136 if (req)
1137 return tcp_check_req(sk, skb, req, prev);
1138
1139 nsk = __inet6_lookup_established(sock_net(sk), &tcp_hashinfo,
1140 &ipv6_hdr(skb)->saddr, th->source,
1141 &ipv6_hdr(skb)->daddr, ntohs(th->dest), inet6_iif(skb));
1142
1143 if (nsk) {
1144 if (nsk->sk_state != TCP_TIME_WAIT) {
1145 bh_lock_sock(nsk);
1146 return nsk;
1147 }
1148 inet_twsk_put(inet_twsk(nsk));
1149 return NULL;
1150 }
1151
1152 #ifdef CONFIG_SYN_COOKIES
1153 if (!th->syn)
1154 sk = cookie_v6_check(sk, skb);
1155 #endif
1156 return sk;
1157 }
1158
1159 /* FIXME: this is substantially similar to the ipv4 code.
1160 * Can some kind of merge be done? -- erics
1161 */
1162 static int tcp_v6_conn_request(struct sock *sk, struct sk_buff *skb)
1163 {
1164 struct tcp_extend_values tmp_ext;
1165 struct tcp_options_received tmp_opt;
1166 const u8 *hash_location;
1167 struct request_sock *req;
1168 struct inet6_request_sock *treq;
1169 struct ipv6_pinfo *np = inet6_sk(sk);
1170 struct tcp_sock *tp = tcp_sk(sk);
1171 __u32 isn = TCP_SKB_CB(skb)->when;
1172 struct dst_entry *dst = NULL;
1173 int want_cookie = 0;
1174
1175 if (skb->protocol == htons(ETH_P_IP))
1176 return tcp_v4_conn_request(sk, skb);
1177
1178 if (!ipv6_unicast_destination(skb))
1179 goto drop;
1180
1181 if (inet_csk_reqsk_queue_is_full(sk) && !isn) {
1182 want_cookie = tcp_syn_flood_action(sk, skb, "TCPv6");
1183 if (!want_cookie)
1184 goto drop;
1185 }
1186
1187 if (sk_acceptq_is_full(sk) && inet_csk_reqsk_queue_young(sk) > 1)
1188 goto drop;
1189
1190 req = inet6_reqsk_alloc(&tcp6_request_sock_ops);
1191 if (req == NULL)
1192 goto drop;
1193
1194 #ifdef CONFIG_TCP_MD5SIG
1195 tcp_rsk(req)->af_specific = &tcp_request_sock_ipv6_ops;
1196 #endif
1197
1198 tcp_clear_options(&tmp_opt);
1199 tmp_opt.mss_clamp = IPV6_MIN_MTU - sizeof(struct tcphdr) - sizeof(struct ipv6hdr);
1200 tmp_opt.user_mss = tp->rx_opt.user_mss;
1201 tcp_parse_options(skb, &tmp_opt, &hash_location, 0);
1202
1203 if (tmp_opt.cookie_plus > 0 &&
1204 tmp_opt.saw_tstamp &&
1205 !tp->rx_opt.cookie_out_never &&
1206 (sysctl_tcp_cookie_size > 0 ||
1207 (tp->cookie_values != NULL &&
1208 tp->cookie_values->cookie_desired > 0))) {
1209 u8 *c;
1210 u32 *d;
1211 u32 *mess = &tmp_ext.cookie_bakery[COOKIE_DIGEST_WORDS];
1212 int l = tmp_opt.cookie_plus - TCPOLEN_COOKIE_BASE;
1213
1214 if (tcp_cookie_generator(&tmp_ext.cookie_bakery[0]) != 0)
1215 goto drop_and_free;
1216
1217 /* Secret recipe starts with IP addresses */
1218 d = (__force u32 *)&ipv6_hdr(skb)->daddr.s6_addr32[0];
1219 *mess++ ^= *d++;
1220 *mess++ ^= *d++;
1221 *mess++ ^= *d++;
1222 *mess++ ^= *d++;
1223 d = (__force u32 *)&ipv6_hdr(skb)->saddr.s6_addr32[0];
1224 *mess++ ^= *d++;
1225 *mess++ ^= *d++;
1226 *mess++ ^= *d++;
1227 *mess++ ^= *d++;
1228
1229 /* plus variable length Initiator Cookie */
1230 c = (u8 *)mess;
1231 while (l-- > 0)
1232 *c++ ^= *hash_location++;
1233
1234 want_cookie = 0; /* not our kind of cookie */
1235 tmp_ext.cookie_out_never = 0; /* false */
1236 tmp_ext.cookie_plus = tmp_opt.cookie_plus;
1237 } else if (!tp->rx_opt.cookie_in_always) {
1238 /* redundant indications, but ensure initialization. */
1239 tmp_ext.cookie_out_never = 1; /* true */
1240 tmp_ext.cookie_plus = 0;
1241 } else {
1242 goto drop_and_free;
1243 }
1244 tmp_ext.cookie_in_always = tp->rx_opt.cookie_in_always;
1245
1246 if (want_cookie && !tmp_opt.saw_tstamp)
1247 tcp_clear_options(&tmp_opt);
1248
1249 tmp_opt.tstamp_ok = tmp_opt.saw_tstamp;
1250 tcp_openreq_init(req, &tmp_opt, skb);
1251
1252 treq = inet6_rsk(req);
1253 ipv6_addr_copy(&treq->rmt_addr, &ipv6_hdr(skb)->saddr);
1254 ipv6_addr_copy(&treq->loc_addr, &ipv6_hdr(skb)->daddr);
1255 if (!want_cookie || tmp_opt.tstamp_ok)
1256 TCP_ECN_create_request(req, tcp_hdr(skb));
1257
1258 if (!isn) {
1259 struct inet_peer *peer = NULL;
1260
1261 if (ipv6_opt_accepted(sk, skb) ||
1262 np->rxopt.bits.rxinfo || np->rxopt.bits.rxoinfo ||
1263 np->rxopt.bits.rxhlim || np->rxopt.bits.rxohlim) {
1264 atomic_inc(&skb->users);
1265 treq->pktopts = skb;
1266 }
1267 treq->iif = sk->sk_bound_dev_if;
1268
1269 /* So that link locals have meaning */
1270 if (!sk->sk_bound_dev_if &&
1271 ipv6_addr_type(&treq->rmt_addr) & IPV6_ADDR_LINKLOCAL)
1272 treq->iif = inet6_iif(skb);
1273
1274 if (want_cookie) {
1275 isn = cookie_v6_init_sequence(sk, skb, &req->mss);
1276 req->cookie_ts = tmp_opt.tstamp_ok;
1277 goto have_isn;
1278 }
1279
1280 /* VJ's idea. We save last timestamp seen
1281 * from the destination in peer table, when entering
1282 * state TIME-WAIT, and check against it before
1283 * accepting new connection request.
1284 *
1285 * If "isn" is not zero, this request hit alive
1286 * timewait bucket, so that all the necessary checks
1287 * are made in the function processing timewait state.
1288 */
1289 if (tmp_opt.saw_tstamp &&
1290 tcp_death_row.sysctl_tw_recycle &&
1291 (dst = inet6_csk_route_req(sk, req)) != NULL &&
1292 (peer = rt6_get_peer((struct rt6_info *)dst)) != NULL &&
1293 ipv6_addr_equal((struct in6_addr *)peer->daddr.addr.a6,
1294 &treq->rmt_addr)) {
1295 inet_peer_refcheck(peer);
1296 if ((u32)get_seconds() - peer->tcp_ts_stamp < TCP_PAWS_MSL &&
1297 (s32)(peer->tcp_ts - req->ts_recent) >
1298 TCP_PAWS_WINDOW) {
1299 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_PAWSPASSIVEREJECTED);
1300 goto drop_and_release;
1301 }
1302 }
1303 /* Kill the following clause, if you dislike this way. */
1304 else if (!sysctl_tcp_syncookies &&
1305 (sysctl_max_syn_backlog - inet_csk_reqsk_queue_len(sk) <
1306 (sysctl_max_syn_backlog >> 2)) &&
1307 (!peer || !peer->tcp_ts_stamp) &&
1308 (!dst || !dst_metric(dst, RTAX_RTT))) {
1309 /* Without syncookies last quarter of
1310 * backlog is filled with destinations,
1311 * proven to be alive.
1312 * It means that we continue to communicate
1313 * to destinations, already remembered
1314 * to the moment of synflood.
1315 */
1316 LIMIT_NETDEBUG(KERN_DEBUG "TCP: drop open request from %pI6/%u\n",
1317 &treq->rmt_addr, ntohs(tcp_hdr(skb)->source));
1318 goto drop_and_release;
1319 }
1320
1321 isn = tcp_v6_init_sequence(skb);
1322 }
1323 have_isn:
1324 tcp_rsk(req)->snt_isn = isn;
1325 tcp_rsk(req)->snt_synack = tcp_time_stamp;
1326
1327 security_inet_conn_request(sk, skb, req);
1328
1329 if (tcp_v6_send_synack(sk, req,
1330 (struct request_values *)&tmp_ext) ||
1331 want_cookie)
1332 goto drop_and_free;
1333
1334 inet6_csk_reqsk_queue_hash_add(sk, req, TCP_TIMEOUT_INIT);
1335 return 0;
1336
1337 drop_and_release:
1338 dst_release(dst);
1339 drop_and_free:
1340 reqsk_free(req);
1341 drop:
1342 return 0; /* don't send reset */
1343 }
1344
1345 static struct sock * tcp_v6_syn_recv_sock(struct sock *sk, struct sk_buff *skb,
1346 struct request_sock *req,
1347 struct dst_entry *dst)
1348 {
1349 struct inet6_request_sock *treq;
1350 struct ipv6_pinfo *newnp, *np = inet6_sk(sk);
1351 struct tcp6_sock *newtcp6sk;
1352 struct inet_sock *newinet;
1353 struct tcp_sock *newtp;
1354 struct sock *newsk;
1355 struct ipv6_txoptions *opt;
1356 #ifdef CONFIG_TCP_MD5SIG
1357 struct tcp_md5sig_key *key;
1358 #endif
1359
1360 if (skb->protocol == htons(ETH_P_IP)) {
1361 /*
1362 * v6 mapped
1363 */
1364
1365 newsk = tcp_v4_syn_recv_sock(sk, skb, req, dst);
1366
1367 if (newsk == NULL)
1368 return NULL;
1369
1370 newtcp6sk = (struct tcp6_sock *)newsk;
1371 inet_sk(newsk)->pinet6 = &newtcp6sk->inet6;
1372
1373 newinet = inet_sk(newsk);
1374 newnp = inet6_sk(newsk);
1375 newtp = tcp_sk(newsk);
1376
1377 memcpy(newnp, np, sizeof(struct ipv6_pinfo));
1378
1379 ipv6_addr_set_v4mapped(newinet->inet_daddr, &newnp->daddr);
1380
1381 ipv6_addr_set_v4mapped(newinet->inet_saddr, &newnp->saddr);
1382
1383 ipv6_addr_copy(&newnp->rcv_saddr, &newnp->saddr);
1384
1385 inet_csk(newsk)->icsk_af_ops = &ipv6_mapped;
1386 newsk->sk_backlog_rcv = tcp_v4_do_rcv;
1387 #ifdef CONFIG_TCP_MD5SIG
1388 newtp->af_specific = &tcp_sock_ipv6_mapped_specific;
1389 #endif
1390
1391 newnp->ipv6_ac_list = NULL;
1392 newnp->ipv6_fl_list = NULL;
1393 newnp->pktoptions = NULL;
1394 newnp->opt = NULL;
1395 newnp->mcast_oif = inet6_iif(skb);
1396 newnp->mcast_hops = ipv6_hdr(skb)->hop_limit;
1397
1398 /*
1399 * No need to charge this sock to the relevant IPv6 refcnt debug socks count
1400 * here, tcp_create_openreq_child now does this for us, see the comment in
1401 * that function for the gory details. -acme
1402 */
1403
1404 /* It is tricky place. Until this moment IPv4 tcp
1405 worked with IPv6 icsk.icsk_af_ops.
1406 Sync it now.
1407 */
1408 tcp_sync_mss(newsk, inet_csk(newsk)->icsk_pmtu_cookie);
1409
1410 return newsk;
1411 }
1412
1413 treq = inet6_rsk(req);
1414 opt = np->opt;
1415
1416 if (sk_acceptq_is_full(sk))
1417 goto out_overflow;
1418
1419 if (!dst) {
1420 dst = inet6_csk_route_req(sk, req);
1421 if (!dst)
1422 goto out;
1423 }
1424
1425 newsk = tcp_create_openreq_child(sk, req, skb);
1426 if (newsk == NULL)
1427 goto out_nonewsk;
1428
1429 /*
1430 * No need to charge this sock to the relevant IPv6 refcnt debug socks
1431 * count here, tcp_create_openreq_child now does this for us, see the
1432 * comment in that function for the gory details. -acme
1433 */
1434
1435 newsk->sk_gso_type = SKB_GSO_TCPV6;
1436 __ip6_dst_store(newsk, dst, NULL, NULL);
1437
1438 newtcp6sk = (struct tcp6_sock *)newsk;
1439 inet_sk(newsk)->pinet6 = &newtcp6sk->inet6;
1440
1441 newtp = tcp_sk(newsk);
1442 newinet = inet_sk(newsk);
1443 newnp = inet6_sk(newsk);
1444
1445 memcpy(newnp, np, sizeof(struct ipv6_pinfo));
1446
1447 ipv6_addr_copy(&newnp->daddr, &treq->rmt_addr);
1448 ipv6_addr_copy(&newnp->saddr, &treq->loc_addr);
1449 ipv6_addr_copy(&newnp->rcv_saddr, &treq->loc_addr);
1450 newsk->sk_bound_dev_if = treq->iif;
1451
1452 /* Now IPv6 options...
1453
1454 First: no IPv4 options.
1455 */
1456 newinet->inet_opt = NULL;
1457 newnp->ipv6_ac_list = NULL;
1458 newnp->ipv6_fl_list = NULL;
1459
1460 /* Clone RX bits */
1461 newnp->rxopt.all = np->rxopt.all;
1462
1463 /* Clone pktoptions received with SYN */
1464 newnp->pktoptions = NULL;
1465 if (treq->pktopts != NULL) {
1466 newnp->pktoptions = skb_clone(treq->pktopts, GFP_ATOMIC);
1467 kfree_skb(treq->pktopts);
1468 treq->pktopts = NULL;
1469 if (newnp->pktoptions)
1470 skb_set_owner_r(newnp->pktoptions, newsk);
1471 }
1472 newnp->opt = NULL;
1473 newnp->mcast_oif = inet6_iif(skb);
1474 newnp->mcast_hops = ipv6_hdr(skb)->hop_limit;
1475
1476 /* Clone native IPv6 options from listening socket (if any)
1477
1478 Yes, keeping reference count would be much more clever,
1479 but we make one more one thing there: reattach optmem
1480 to newsk.
1481 */
1482 if (opt) {
1483 newnp->opt = ipv6_dup_options(newsk, opt);
1484 if (opt != np->opt)
1485 sock_kfree_s(sk, opt, opt->tot_len);
1486 }
1487
1488 inet_csk(newsk)->icsk_ext_hdr_len = 0;
1489 if (newnp->opt)
1490 inet_csk(newsk)->icsk_ext_hdr_len = (newnp->opt->opt_nflen +
1491 newnp->opt->opt_flen);
1492
1493 tcp_mtup_init(newsk);
1494 tcp_sync_mss(newsk, dst_mtu(dst));
1495 newtp->advmss = dst_metric_advmss(dst);
1496 tcp_initialize_rcv_mss(newsk);
1497 if (tcp_rsk(req)->snt_synack)
1498 tcp_valid_rtt_meas(newsk,
1499 tcp_time_stamp - tcp_rsk(req)->snt_synack);
1500 newtp->total_retrans = req->retrans;
1501
1502 newinet->inet_daddr = newinet->inet_saddr = LOOPBACK4_IPV6;
1503 newinet->inet_rcv_saddr = LOOPBACK4_IPV6;
1504
1505 #ifdef CONFIG_TCP_MD5SIG
1506 /* Copy over the MD5 key from the original socket */
1507 if ((key = tcp_v6_md5_do_lookup(sk, &newnp->daddr)) != NULL) {
1508 /* We're using one, so create a matching key
1509 * on the newsk structure. If we fail to get
1510 * memory, then we end up not copying the key
1511 * across. Shucks.
1512 */
1513 char *newkey = kmemdup(key->key, key->keylen, GFP_ATOMIC);
1514 if (newkey != NULL)
1515 tcp_v6_md5_do_add(newsk, &newnp->daddr,
1516 newkey, key->keylen);
1517 }
1518 #endif
1519
1520 if (__inet_inherit_port(sk, newsk) < 0) {
1521 sock_put(newsk);
1522 goto out;
1523 }
1524 __inet6_hash(newsk, NULL);
1525
1526 return newsk;
1527
1528 out_overflow:
1529 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENOVERFLOWS);
1530 out_nonewsk:
1531 if (opt && opt != np->opt)
1532 sock_kfree_s(sk, opt, opt->tot_len);
1533 dst_release(dst);
1534 out:
1535 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENDROPS);
1536 return NULL;
1537 }
1538
1539 static __sum16 tcp_v6_checksum_init(struct sk_buff *skb)
1540 {
1541 if (skb->ip_summed == CHECKSUM_COMPLETE) {
1542 if (!tcp_v6_check(skb->len, &ipv6_hdr(skb)->saddr,
1543 &ipv6_hdr(skb)->daddr, skb->csum)) {
1544 skb->ip_summed = CHECKSUM_UNNECESSARY;
1545 return 0;
1546 }
1547 }
1548
1549 skb->csum = ~csum_unfold(tcp_v6_check(skb->len,
1550 &ipv6_hdr(skb)->saddr,
1551 &ipv6_hdr(skb)->daddr, 0));
1552
1553 if (skb->len <= 76) {
1554 return __skb_checksum_complete(skb);
1555 }
1556 return 0;
1557 }
1558
1559 /* The socket must have it's spinlock held when we get
1560 * here.
1561 *
1562 * We have a potential double-lock case here, so even when
1563 * doing backlog processing we use the BH locking scheme.
1564 * This is because we cannot sleep with the original spinlock
1565 * held.
1566 */
1567 static int tcp_v6_do_rcv(struct sock *sk, struct sk_buff *skb)
1568 {
1569 struct ipv6_pinfo *np = inet6_sk(sk);
1570 struct tcp_sock *tp;
1571 struct sk_buff *opt_skb = NULL;
1572
1573 /* Imagine: socket is IPv6. IPv4 packet arrives,
1574 goes to IPv4 receive handler and backlogged.
1575 From backlog it always goes here. Kerboom...
1576 Fortunately, tcp_rcv_established and rcv_established
1577 handle them correctly, but it is not case with
1578 tcp_v6_hnd_req and tcp_v6_send_reset(). --ANK
1579 */
1580
1581 if (skb->protocol == htons(ETH_P_IP))
1582 return tcp_v4_do_rcv(sk, skb);
1583
1584 #ifdef CONFIG_TCP_MD5SIG
1585 if (tcp_v6_inbound_md5_hash (sk, skb))
1586 goto discard;
1587 #endif
1588
1589 if (sk_filter(sk, skb))
1590 goto discard;
1591
1592 /*
1593 * socket locking is here for SMP purposes as backlog rcv
1594 * is currently called with bh processing disabled.
1595 */
1596
1597 /* Do Stevens' IPV6_PKTOPTIONS.
1598
1599 Yes, guys, it is the only place in our code, where we
1600 may make it not affecting IPv4.
1601 The rest of code is protocol independent,
1602 and I do not like idea to uglify IPv4.
1603
1604 Actually, all the idea behind IPV6_PKTOPTIONS
1605 looks not very well thought. For now we latch
1606 options, received in the last packet, enqueued
1607 by tcp. Feel free to propose better solution.
1608 --ANK (980728)
1609 */
1610 if (np->rxopt.all)
1611 opt_skb = skb_clone(skb, GFP_ATOMIC);
1612
1613 if (sk->sk_state == TCP_ESTABLISHED) { /* Fast path */
1614 sock_rps_save_rxhash(sk, skb);
1615 if (tcp_rcv_established(sk, skb, tcp_hdr(skb), skb->len))
1616 goto reset;
1617 if (opt_skb)
1618 goto ipv6_pktoptions;
1619 return 0;
1620 }
1621
1622 if (skb->len < tcp_hdrlen(skb) || tcp_checksum_complete(skb))
1623 goto csum_err;
1624
1625 if (sk->sk_state == TCP_LISTEN) {
1626 struct sock *nsk = tcp_v6_hnd_req(sk, skb);
1627 if (!nsk)
1628 goto discard;
1629
1630 /*
1631 * Queue it on the new socket if the new socket is active,
1632 * otherwise we just shortcircuit this and continue with
1633 * the new socket..
1634 */
1635 if(nsk != sk) {
1636 sock_rps_save_rxhash(nsk, skb);
1637 if (tcp_child_process(sk, nsk, skb))
1638 goto reset;
1639 if (opt_skb)
1640 __kfree_skb(opt_skb);
1641 return 0;
1642 }
1643 } else
1644 sock_rps_save_rxhash(sk, skb);
1645
1646 if (tcp_rcv_state_process(sk, skb, tcp_hdr(skb), skb->len))
1647 goto reset;
1648 if (opt_skb)
1649 goto ipv6_pktoptions;
1650 return 0;
1651
1652 reset:
1653 tcp_v6_send_reset(sk, skb);
1654 discard:
1655 if (opt_skb)
1656 __kfree_skb(opt_skb);
1657 kfree_skb(skb);
1658 return 0;
1659 csum_err:
1660 TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_INERRS);
1661 goto discard;
1662
1663
1664 ipv6_pktoptions:
1665 /* Do you ask, what is it?
1666
1667 1. skb was enqueued by tcp.
1668 2. skb is added to tail of read queue, rather than out of order.
1669 3. socket is not in passive state.
1670 4. Finally, it really contains options, which user wants to receive.
1671 */
1672 tp = tcp_sk(sk);
1673 if (TCP_SKB_CB(opt_skb)->end_seq == tp->rcv_nxt &&
1674 !((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_LISTEN))) {
1675 if (np->rxopt.bits.rxinfo || np->rxopt.bits.rxoinfo)
1676 np->mcast_oif = inet6_iif(opt_skb);
1677 if (np->rxopt.bits.rxhlim || np->rxopt.bits.rxohlim)
1678 np->mcast_hops = ipv6_hdr(opt_skb)->hop_limit;
1679 if (ipv6_opt_accepted(sk, opt_skb)) {
1680 skb_set_owner_r(opt_skb, sk);
1681 opt_skb = xchg(&np->pktoptions, opt_skb);
1682 } else {
1683 __kfree_skb(opt_skb);
1684 opt_skb = xchg(&np->pktoptions, NULL);
1685 }
1686 }
1687
1688 kfree_skb(opt_skb);
1689 return 0;
1690 }
1691
1692 static int tcp_v6_rcv(struct sk_buff *skb)
1693 {
1694 const struct tcphdr *th;
1695 const struct ipv6hdr *hdr;
1696 struct sock *sk;
1697 int ret;
1698 struct net *net = dev_net(skb->dev);
1699
1700 if (skb->pkt_type != PACKET_HOST)
1701 goto discard_it;
1702
1703 /*
1704 * Count it even if it's bad.
1705 */
1706 TCP_INC_STATS_BH(net, TCP_MIB_INSEGS);
1707
1708 if (!pskb_may_pull(skb, sizeof(struct tcphdr)))
1709 goto discard_it;
1710
1711 th = tcp_hdr(skb);
1712
1713 if (th->doff < sizeof(struct tcphdr)/4)
1714 goto bad_packet;
1715 if (!pskb_may_pull(skb, th->doff*4))
1716 goto discard_it;
1717
1718 if (!skb_csum_unnecessary(skb) && tcp_v6_checksum_init(skb))
1719 goto bad_packet;
1720
1721 th = tcp_hdr(skb);
1722 hdr = ipv6_hdr(skb);
1723 TCP_SKB_CB(skb)->seq = ntohl(th->seq);
1724 TCP_SKB_CB(skb)->end_seq = (TCP_SKB_CB(skb)->seq + th->syn + th->fin +
1725 skb->len - th->doff*4);
1726 TCP_SKB_CB(skb)->ack_seq = ntohl(th->ack_seq);
1727 TCP_SKB_CB(skb)->when = 0;
1728 TCP_SKB_CB(skb)->ip_dsfield = ipv6_get_dsfield(hdr);
1729 TCP_SKB_CB(skb)->sacked = 0;
1730
1731 sk = __inet6_lookup_skb(&tcp_hashinfo, skb, th->source, th->dest);
1732 if (!sk)
1733 goto no_tcp_socket;
1734
1735 process:
1736 if (sk->sk_state == TCP_TIME_WAIT)
1737 goto do_time_wait;
1738
1739 if (hdr->hop_limit < inet6_sk(sk)->min_hopcount) {
1740 NET_INC_STATS_BH(net, LINUX_MIB_TCPMINTTLDROP);
1741 goto discard_and_relse;
1742 }
1743
1744 if (!xfrm6_policy_check(sk, XFRM_POLICY_IN, skb))
1745 goto discard_and_relse;
1746
1747 if (sk_filter(sk, skb))
1748 goto discard_and_relse;
1749
1750 skb->dev = NULL;
1751
1752 bh_lock_sock_nested(sk);
1753 ret = 0;
1754 if (!sock_owned_by_user(sk)) {
1755 #ifdef CONFIG_NET_DMA
1756 struct tcp_sock *tp = tcp_sk(sk);
1757 if (!tp->ucopy.dma_chan && tp->ucopy.pinned_list)
1758 tp->ucopy.dma_chan = dma_find_channel(DMA_MEMCPY);
1759 if (tp->ucopy.dma_chan)
1760 ret = tcp_v6_do_rcv(sk, skb);
1761 else
1762 #endif
1763 {
1764 if (!tcp_prequeue(sk, skb))
1765 ret = tcp_v6_do_rcv(sk, skb);
1766 }
1767 } else if (unlikely(sk_add_backlog(sk, skb))) {
1768 bh_unlock_sock(sk);
1769 NET_INC_STATS_BH(net, LINUX_MIB_TCPBACKLOGDROP);
1770 goto discard_and_relse;
1771 }
1772 bh_unlock_sock(sk);
1773
1774 sock_put(sk);
1775 return ret ? -1 : 0;
1776
1777 no_tcp_socket:
1778 if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb))
1779 goto discard_it;
1780
1781 if (skb->len < (th->doff<<2) || tcp_checksum_complete(skb)) {
1782 bad_packet:
1783 TCP_INC_STATS_BH(net, TCP_MIB_INERRS);
1784 } else {
1785 tcp_v6_send_reset(NULL, skb);
1786 }
1787
1788 discard_it:
1789
1790 /*
1791 * Discard frame
1792 */
1793
1794 kfree_skb(skb);
1795 return 0;
1796
1797 discard_and_relse:
1798 sock_put(sk);
1799 goto discard_it;
1800
1801 do_time_wait:
1802 if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb)) {
1803 inet_twsk_put(inet_twsk(sk));
1804 goto discard_it;
1805 }
1806
1807 if (skb->len < (th->doff<<2) || tcp_checksum_complete(skb)) {
1808 TCP_INC_STATS_BH(net, TCP_MIB_INERRS);
1809 inet_twsk_put(inet_twsk(sk));
1810 goto discard_it;
1811 }
1812
1813 switch (tcp_timewait_state_process(inet_twsk(sk), skb, th)) {
1814 case TCP_TW_SYN:
1815 {
1816 struct sock *sk2;
1817
1818 sk2 = inet6_lookup_listener(dev_net(skb->dev), &tcp_hashinfo,
1819 &ipv6_hdr(skb)->daddr,
1820 ntohs(th->dest), inet6_iif(skb));
1821 if (sk2 != NULL) {
1822 struct inet_timewait_sock *tw = inet_twsk(sk);
1823 inet_twsk_deschedule(tw, &tcp_death_row);
1824 inet_twsk_put(tw);
1825 sk = sk2;
1826 goto process;
1827 }
1828 /* Fall through to ACK */
1829 }
1830 case TCP_TW_ACK:
1831 tcp_v6_timewait_ack(sk, skb);
1832 break;
1833 case TCP_TW_RST:
1834 goto no_tcp_socket;
1835 case TCP_TW_SUCCESS:;
1836 }
1837 goto discard_it;
1838 }
1839
1840 static struct inet_peer *tcp_v6_get_peer(struct sock *sk, bool *release_it)
1841 {
1842 struct rt6_info *rt = (struct rt6_info *) __sk_dst_get(sk);
1843 struct ipv6_pinfo *np = inet6_sk(sk);
1844 struct inet_peer *peer;
1845
1846 if (!rt ||
1847 !ipv6_addr_equal(&np->daddr, &rt->rt6i_dst.addr)) {
1848 peer = inet_getpeer_v6(&np->daddr, 1);
1849 *release_it = true;
1850 } else {
1851 if (!rt->rt6i_peer)
1852 rt6_bind_peer(rt, 1);
1853 peer = rt->rt6i_peer;
1854 *release_it = false;
1855 }
1856
1857 return peer;
1858 }
1859
1860 static void *tcp_v6_tw_get_peer(struct sock *sk)
1861 {
1862 const struct inet6_timewait_sock *tw6 = inet6_twsk(sk);
1863 const struct inet_timewait_sock *tw = inet_twsk(sk);
1864
1865 if (tw->tw_family == AF_INET)
1866 return tcp_v4_tw_get_peer(sk);
1867
1868 return inet_getpeer_v6(&tw6->tw_v6_daddr, 1);
1869 }
1870
1871 static struct timewait_sock_ops tcp6_timewait_sock_ops = {
1872 .twsk_obj_size = sizeof(struct tcp6_timewait_sock),
1873 .twsk_unique = tcp_twsk_unique,
1874 .twsk_destructor= tcp_twsk_destructor,
1875 .twsk_getpeer = tcp_v6_tw_get_peer,
1876 };
1877
1878 static const struct inet_connection_sock_af_ops ipv6_specific = {
1879 .queue_xmit = inet6_csk_xmit,
1880 .send_check = tcp_v6_send_check,
1881 .rebuild_header = inet6_sk_rebuild_header,
1882 .conn_request = tcp_v6_conn_request,
1883 .syn_recv_sock = tcp_v6_syn_recv_sock,
1884 .get_peer = tcp_v6_get_peer,
1885 .net_header_len = sizeof(struct ipv6hdr),
1886 .setsockopt = ipv6_setsockopt,
1887 .getsockopt = ipv6_getsockopt,
1888 .addr2sockaddr = inet6_csk_addr2sockaddr,
1889 .sockaddr_len = sizeof(struct sockaddr_in6),
1890 .bind_conflict = inet6_csk_bind_conflict,
1891 #ifdef CONFIG_COMPAT
1892 .compat_setsockopt = compat_ipv6_setsockopt,
1893 .compat_getsockopt = compat_ipv6_getsockopt,
1894 #endif
1895 };
1896
1897 #ifdef CONFIG_TCP_MD5SIG
1898 static const struct tcp_sock_af_ops tcp_sock_ipv6_specific = {
1899 .md5_lookup = tcp_v6_md5_lookup,
1900 .calc_md5_hash = tcp_v6_md5_hash_skb,
1901 .md5_add = tcp_v6_md5_add_func,
1902 .md5_parse = tcp_v6_parse_md5_keys,
1903 };
1904 #endif
1905
1906 /*
1907 * TCP over IPv4 via INET6 API
1908 */
1909
1910 static const struct inet_connection_sock_af_ops ipv6_mapped = {
1911 .queue_xmit = ip_queue_xmit,
1912 .send_check = tcp_v4_send_check,
1913 .rebuild_header = inet_sk_rebuild_header,
1914 .conn_request = tcp_v6_conn_request,
1915 .syn_recv_sock = tcp_v6_syn_recv_sock,
1916 .get_peer = tcp_v4_get_peer,
1917 .net_header_len = sizeof(struct iphdr),
1918 .setsockopt = ipv6_setsockopt,
1919 .getsockopt = ipv6_getsockopt,
1920 .addr2sockaddr = inet6_csk_addr2sockaddr,
1921 .sockaddr_len = sizeof(struct sockaddr_in6),
1922 .bind_conflict = inet6_csk_bind_conflict,
1923 #ifdef CONFIG_COMPAT
1924 .compat_setsockopt = compat_ipv6_setsockopt,
1925 .compat_getsockopt = compat_ipv6_getsockopt,
1926 #endif
1927 };
1928
1929 #ifdef CONFIG_TCP_MD5SIG
1930 static const struct tcp_sock_af_ops tcp_sock_ipv6_mapped_specific = {
1931 .md5_lookup = tcp_v4_md5_lookup,
1932 .calc_md5_hash = tcp_v4_md5_hash_skb,
1933 .md5_add = tcp_v6_md5_add_func,
1934 .md5_parse = tcp_v6_parse_md5_keys,
1935 };
1936 #endif
1937
1938 /* NOTE: A lot of things set to zero explicitly by call to
1939 * sk_alloc() so need not be done here.
1940 */
1941 static int tcp_v6_init_sock(struct sock *sk)
1942 {
1943 struct inet_connection_sock *icsk = inet_csk(sk);
1944 struct tcp_sock *tp = tcp_sk(sk);
1945
1946 skb_queue_head_init(&tp->out_of_order_queue);
1947 tcp_init_xmit_timers(sk);
1948 tcp_prequeue_init(tp);
1949
1950 icsk->icsk_rto = TCP_TIMEOUT_INIT;
1951 tp->mdev = TCP_TIMEOUT_INIT;
1952
1953 /* So many TCP implementations out there (incorrectly) count the
1954 * initial SYN frame in their delayed-ACK and congestion control
1955 * algorithms that we must have the following bandaid to talk
1956 * efficiently to them. -DaveM
1957 */
1958 tp->snd_cwnd = 2;
1959
1960 /* See draft-stevens-tcpca-spec-01 for discussion of the
1961 * initialization of these values.
1962 */
1963 tp->snd_ssthresh = TCP_INFINITE_SSTHRESH;
1964 tp->snd_cwnd_clamp = ~0;
1965 tp->mss_cache = TCP_MSS_DEFAULT;
1966
1967 tp->reordering = sysctl_tcp_reordering;
1968
1969 sk->sk_state = TCP_CLOSE;
1970
1971 icsk->icsk_af_ops = &ipv6_specific;
1972 icsk->icsk_ca_ops = &tcp_init_congestion_ops;
1973 icsk->icsk_sync_mss = tcp_sync_mss;
1974 sk->sk_write_space = sk_stream_write_space;
1975 sock_set_flag(sk, SOCK_USE_WRITE_QUEUE);
1976
1977 #ifdef CONFIG_TCP_MD5SIG
1978 tp->af_specific = &tcp_sock_ipv6_specific;
1979 #endif
1980
1981 /* TCP Cookie Transactions */
1982 if (sysctl_tcp_cookie_size > 0) {
1983 /* Default, cookies without s_data_payload. */
1984 tp->cookie_values =
1985 kzalloc(sizeof(*tp->cookie_values),
1986 sk->sk_allocation);
1987 if (tp->cookie_values != NULL)
1988 kref_init(&tp->cookie_values->kref);
1989 }
1990 /* Presumed zeroed, in order of appearance:
1991 * cookie_in_always, cookie_out_never,
1992 * s_data_constant, s_data_in, s_data_out
1993 */
1994 sk->sk_sndbuf = sysctl_tcp_wmem[1];
1995 sk->sk_rcvbuf = sysctl_tcp_rmem[1];
1996
1997 local_bh_disable();
1998 percpu_counter_inc(&tcp_sockets_allocated);
1999 local_bh_enable();
2000
2001 return 0;
2002 }
2003
2004 static void tcp_v6_destroy_sock(struct sock *sk)
2005 {
2006 #ifdef CONFIG_TCP_MD5SIG
2007 /* Clean up the MD5 key list */
2008 if (tcp_sk(sk)->md5sig_info)
2009 tcp_v6_clear_md5_list(sk);
2010 #endif
2011 tcp_v4_destroy_sock(sk);
2012 inet6_destroy_sock(sk);
2013 }
2014
2015 #ifdef CONFIG_PROC_FS
2016 /* Proc filesystem TCPv6 sock list dumping. */
2017 static void get_openreq6(struct seq_file *seq,
2018 const struct sock *sk, struct request_sock *req, int i, int uid)
2019 {
2020 int ttd = req->expires - jiffies;
2021 const struct in6_addr *src = &inet6_rsk(req)->loc_addr;
2022 const struct in6_addr *dest = &inet6_rsk(req)->rmt_addr;
2023
2024 if (ttd < 0)
2025 ttd = 0;
2026
2027 seq_printf(seq,
2028 "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
2029 "%02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %pK\n",
2030 i,
2031 src->s6_addr32[0], src->s6_addr32[1],
2032 src->s6_addr32[2], src->s6_addr32[3],
2033 ntohs(inet_rsk(req)->loc_port),
2034 dest->s6_addr32[0], dest->s6_addr32[1],
2035 dest->s6_addr32[2], dest->s6_addr32[3],
2036 ntohs(inet_rsk(req)->rmt_port),
2037 TCP_SYN_RECV,
2038 0,0, /* could print option size, but that is af dependent. */
2039 1, /* timers active (only the expire timer) */
2040 jiffies_to_clock_t(ttd),
2041 req->retrans,
2042 uid,
2043 0, /* non standard timer */
2044 0, /* open_requests have no inode */
2045 0, req);
2046 }
2047
2048 static void get_tcp6_sock(struct seq_file *seq, struct sock *sp, int i)
2049 {
2050 const struct in6_addr *dest, *src;
2051 __u16 destp, srcp;
2052 int timer_active;
2053 unsigned long timer_expires;
2054 const struct inet_sock *inet = inet_sk(sp);
2055 const struct tcp_sock *tp = tcp_sk(sp);
2056 const struct inet_connection_sock *icsk = inet_csk(sp);
2057 const struct ipv6_pinfo *np = inet6_sk(sp);
2058
2059 dest = &np->daddr;
2060 src = &np->rcv_saddr;
2061 destp = ntohs(inet->inet_dport);
2062 srcp = ntohs(inet->inet_sport);
2063
2064 if (icsk->icsk_pending == ICSK_TIME_RETRANS) {
2065 timer_active = 1;
2066 timer_expires = icsk->icsk_timeout;
2067 } else if (icsk->icsk_pending == ICSK_TIME_PROBE0) {
2068 timer_active = 4;
2069 timer_expires = icsk->icsk_timeout;
2070 } else if (timer_pending(&sp->sk_timer)) {
2071 timer_active = 2;
2072 timer_expires = sp->sk_timer.expires;
2073 } else {
2074 timer_active = 0;
2075 timer_expires = jiffies;
2076 }
2077
2078 seq_printf(seq,
2079 "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
2080 "%02X %08X:%08X %02X:%08lX %08X %5d %8d %lu %d %pK %lu %lu %u %u %d\n",
2081 i,
2082 src->s6_addr32[0], src->s6_addr32[1],
2083 src->s6_addr32[2], src->s6_addr32[3], srcp,
2084 dest->s6_addr32[0], dest->s6_addr32[1],
2085 dest->s6_addr32[2], dest->s6_addr32[3], destp,
2086 sp->sk_state,
2087 tp->write_seq-tp->snd_una,
2088 (sp->sk_state == TCP_LISTEN) ? sp->sk_ack_backlog : (tp->rcv_nxt - tp->copied_seq),
2089 timer_active,
2090 jiffies_to_clock_t(timer_expires - jiffies),
2091 icsk->icsk_retransmits,
2092 sock_i_uid(sp),
2093 icsk->icsk_probes_out,
2094 sock_i_ino(sp),
2095 atomic_read(&sp->sk_refcnt), sp,
2096 jiffies_to_clock_t(icsk->icsk_rto),
2097 jiffies_to_clock_t(icsk->icsk_ack.ato),
2098 (icsk->icsk_ack.quick << 1 ) | icsk->icsk_ack.pingpong,
2099 tp->snd_cwnd,
2100 tcp_in_initial_slowstart(tp) ? -1 : tp->snd_ssthresh
2101 );
2102 }
2103
2104 static void get_timewait6_sock(struct seq_file *seq,
2105 struct inet_timewait_sock *tw, int i)
2106 {
2107 const struct in6_addr *dest, *src;
2108 __u16 destp, srcp;
2109 const struct inet6_timewait_sock *tw6 = inet6_twsk((struct sock *)tw);
2110 int ttd = tw->tw_ttd - jiffies;
2111
2112 if (ttd < 0)
2113 ttd = 0;
2114
2115 dest = &tw6->tw_v6_daddr;
2116 src = &tw6->tw_v6_rcv_saddr;
2117 destp = ntohs(tw->tw_dport);
2118 srcp = ntohs(tw->tw_sport);
2119
2120 seq_printf(seq,
2121 "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
2122 "%02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %pK\n",
2123 i,
2124 src->s6_addr32[0], src->s6_addr32[1],
2125 src->s6_addr32[2], src->s6_addr32[3], srcp,
2126 dest->s6_addr32[0], dest->s6_addr32[1],
2127 dest->s6_addr32[2], dest->s6_addr32[3], destp,
2128 tw->tw_substate, 0, 0,
2129 3, jiffies_to_clock_t(ttd), 0, 0, 0, 0,
2130 atomic_read(&tw->tw_refcnt), tw);
2131 }
2132
2133 static int tcp6_seq_show(struct seq_file *seq, void *v)
2134 {
2135 struct tcp_iter_state *st;
2136
2137 if (v == SEQ_START_TOKEN) {
2138 seq_puts(seq,
2139 " sl "
2140 "local_address "
2141 "remote_address "
2142 "st tx_queue rx_queue tr tm->when retrnsmt"
2143 " uid timeout inode\n");
2144 goto out;
2145 }
2146 st = seq->private;
2147
2148 switch (st->state) {
2149 case TCP_SEQ_STATE_LISTENING:
2150 case TCP_SEQ_STATE_ESTABLISHED:
2151 get_tcp6_sock(seq, v, st->num);
2152 break;
2153 case TCP_SEQ_STATE_OPENREQ:
2154 get_openreq6(seq, st->syn_wait_sk, v, st->num, st->uid);
2155 break;
2156 case TCP_SEQ_STATE_TIME_WAIT:
2157 get_timewait6_sock(seq, v, st->num);
2158 break;
2159 }
2160 out:
2161 return 0;
2162 }
2163
2164 static struct tcp_seq_afinfo tcp6_seq_afinfo = {
2165 .name = "tcp6",
2166 .family = AF_INET6,
2167 .seq_fops = {
2168 .owner = THIS_MODULE,
2169 },
2170 .seq_ops = {
2171 .show = tcp6_seq_show,
2172 },
2173 };
2174
2175 int __net_init tcp6_proc_init(struct net *net)
2176 {
2177 return tcp_proc_register(net, &tcp6_seq_afinfo);
2178 }
2179
2180 void tcp6_proc_exit(struct net *net)
2181 {
2182 tcp_proc_unregister(net, &tcp6_seq_afinfo);
2183 }
2184 #endif
2185
2186 struct proto tcpv6_prot = {
2187 .name = "TCPv6",
2188 .owner = THIS_MODULE,
2189 .close = tcp_close,
2190 .connect = tcp_v6_connect,
2191 .disconnect = tcp_disconnect,
2192 .accept = inet_csk_accept,
2193 .ioctl = tcp_ioctl,
2194 .init = tcp_v6_init_sock,
2195 .destroy = tcp_v6_destroy_sock,
2196 .shutdown = tcp_shutdown,
2197 .setsockopt = tcp_setsockopt,
2198 .getsockopt = tcp_getsockopt,
2199 .recvmsg = tcp_recvmsg,
2200 .sendmsg = tcp_sendmsg,
2201 .sendpage = tcp_sendpage,
2202 .backlog_rcv = tcp_v6_do_rcv,
2203 .hash = tcp_v6_hash,
2204 .unhash = inet_unhash,
2205 .get_port = inet_csk_get_port,
2206 .enter_memory_pressure = tcp_enter_memory_pressure,
2207 .sockets_allocated = &tcp_sockets_allocated,
2208 .memory_allocated = &tcp_memory_allocated,
2209 .memory_pressure = &tcp_memory_pressure,
2210 .orphan_count = &tcp_orphan_count,
2211 .sysctl_mem = sysctl_tcp_mem,
2212 .sysctl_wmem = sysctl_tcp_wmem,
2213 .sysctl_rmem = sysctl_tcp_rmem,
2214 .max_header = MAX_TCP_HEADER,
2215 .obj_size = sizeof(struct tcp6_sock),
2216 .slab_flags = SLAB_DESTROY_BY_RCU,
2217 .twsk_prot = &tcp6_timewait_sock_ops,
2218 .rsk_prot = &tcp6_request_sock_ops,
2219 .h.hashinfo = &tcp_hashinfo,
2220 .no_autobind = true,
2221 #ifdef CONFIG_COMPAT
2222 .compat_setsockopt = compat_tcp_setsockopt,
2223 .compat_getsockopt = compat_tcp_getsockopt,
2224 #endif
2225 };
2226
2227 static const struct inet6_protocol tcpv6_protocol = {
2228 .handler = tcp_v6_rcv,
2229 .err_handler = tcp_v6_err,
2230 .gso_send_check = tcp_v6_gso_send_check,
2231 .gso_segment = tcp_tso_segment,
2232 .gro_receive = tcp6_gro_receive,
2233 .gro_complete = tcp6_gro_complete,
2234 .flags = INET6_PROTO_NOPOLICY|INET6_PROTO_FINAL,
2235 };
2236
2237 static struct inet_protosw tcpv6_protosw = {
2238 .type = SOCK_STREAM,
2239 .protocol = IPPROTO_TCP,
2240 .prot = &tcpv6_prot,
2241 .ops = &inet6_stream_ops,
2242 .no_check = 0,
2243 .flags = INET_PROTOSW_PERMANENT |
2244 INET_PROTOSW_ICSK,
2245 };
2246
2247 static int __net_init tcpv6_net_init(struct net *net)
2248 {
2249 return inet_ctl_sock_create(&net->ipv6.tcp_sk, PF_INET6,
2250 SOCK_RAW, IPPROTO_TCP, net);
2251 }
2252
2253 static void __net_exit tcpv6_net_exit(struct net *net)
2254 {
2255 inet_ctl_sock_destroy(net->ipv6.tcp_sk);
2256 }
2257
2258 static void __net_exit tcpv6_net_exit_batch(struct list_head *net_exit_list)
2259 {
2260 inet_twsk_purge(&tcp_hashinfo, &tcp_death_row, AF_INET6);
2261 }
2262
2263 static struct pernet_operations tcpv6_net_ops = {
2264 .init = tcpv6_net_init,
2265 .exit = tcpv6_net_exit,
2266 .exit_batch = tcpv6_net_exit_batch,
2267 };
2268
2269 int __init tcpv6_init(void)
2270 {
2271 int ret;
2272
2273 ret = inet6_add_protocol(&tcpv6_protocol, IPPROTO_TCP);
2274 if (ret)
2275 goto out;
2276
2277 /* register inet6 protocol */
2278 ret = inet6_register_protosw(&tcpv6_protosw);
2279 if (ret)
2280 goto out_tcpv6_protocol;
2281
2282 ret = register_pernet_subsys(&tcpv6_net_ops);
2283 if (ret)
2284 goto out_tcpv6_protosw;
2285 out:
2286 return ret;
2287
2288 out_tcpv6_protocol:
2289 inet6_del_protocol(&tcpv6_protocol, IPPROTO_TCP);
2290 out_tcpv6_protosw:
2291 inet6_unregister_protosw(&tcpv6_protosw);
2292 goto out;
2293 }
2294
2295 void tcpv6_exit(void)
2296 {
2297 unregister_pernet_subsys(&tcpv6_net_ops);
2298 inet6_unregister_protosw(&tcpv6_protosw);
2299 inet6_del_protocol(&tcpv6_protocol, IPPROTO_TCP);
2300 }
Linus' kernel tree