?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
spi-topcliff-pch: Fix issue for transmitting over 4KByte
[zen-stable.git] / net / ipv6 / tcp_ipv6.c
bloba89ca8df38b90879afbc5f609abb7b3eacf58e72
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 #include <net/tcp_memcontrol.h>
66
67 #include <asm/uaccess.h>
68
69 #include <linux/proc_fs.h>
70 #include <linux/seq_file.h>
71
72 #include <linux/crypto.h>
73 #include <linux/scatterlist.h>
74
75 static void tcp_v6_send_reset(struct sock *sk, struct sk_buff *skb);
76 static void tcp_v6_reqsk_send_ack(struct sock *sk, struct sk_buff *skb,
77 struct request_sock *req);
78
79 static int tcp_v6_do_rcv(struct sock *sk, struct sk_buff *skb);
80 static void __tcp_v6_send_check(struct sk_buff *skb,
81 const struct in6_addr *saddr,
82 const struct in6_addr *daddr);
83
84 static const struct inet_connection_sock_af_ops ipv6_mapped;
85 static const struct inet_connection_sock_af_ops ipv6_specific;
86 #ifdef CONFIG_TCP_MD5SIG
87 static const struct tcp_sock_af_ops tcp_sock_ipv6_specific;
88 static const struct tcp_sock_af_ops tcp_sock_ipv6_mapped_specific;
89 #else
90 static struct tcp_md5sig_key *tcp_v6_md5_do_lookup(struct sock *sk,
91 const struct in6_addr *addr)
92 {
93 return NULL;
94 }
95 #endif
96
97 static void tcp_v6_hash(struct sock *sk)
98 {
99 if (sk->sk_state != TCP_CLOSE) {
100 if (inet_csk(sk)->icsk_af_ops == &ipv6_mapped) {
101 tcp_prot.hash(sk);
102 return;
103 }
104 local_bh_disable();
105 __inet6_hash(sk, NULL);
106 local_bh_enable();
107 }
108 }
109
110 static __inline__ __sum16 tcp_v6_check(int len,
111 const struct in6_addr *saddr,
112 const struct in6_addr *daddr,
113 __wsum base)
114 {
115 return csum_ipv6_magic(saddr, daddr, len, IPPROTO_TCP, base);
116 }
117
118 static __u32 tcp_v6_init_sequence(const struct sk_buff *skb)
119 {
120 return secure_tcpv6_sequence_number(ipv6_hdr(skb)->daddr.s6_addr32,
121 ipv6_hdr(skb)->saddr.s6_addr32,
122 tcp_hdr(skb)->dest,
123 tcp_hdr(skb)->source);
124 }
125
126 static int tcp_v6_connect(struct sock *sk, struct sockaddr *uaddr,
127 int addr_len)
128 {
129 struct sockaddr_in6 *usin = (struct sockaddr_in6 *) uaddr;
130 struct inet_sock *inet = inet_sk(sk);
131 struct inet_connection_sock *icsk = inet_csk(sk);
132 struct ipv6_pinfo *np = inet6_sk(sk);
133 struct tcp_sock *tp = tcp_sk(sk);
134 struct in6_addr *saddr = NULL, *final_p, final;
135 struct rt6_info *rt;
136 struct flowi6 fl6;
137 struct dst_entry *dst;
138 int addr_type;
139 int err;
140
141 if (addr_len < SIN6_LEN_RFC2133)
142 return -EINVAL;
143
144 if (usin->sin6_family != AF_INET6)
145 return -EAFNOSUPPORT;
146
147 memset(&fl6, 0, sizeof(fl6));
148
149 if (np->sndflow) {
150 fl6.flowlabel = usin->sin6_flowinfo&IPV6_FLOWINFO_MASK;
151 IP6_ECN_flow_init(fl6.flowlabel);
152 if (fl6.flowlabel&IPV6_FLOWLABEL_MASK) {
153 struct ip6_flowlabel *flowlabel;
154 flowlabel = fl6_sock_lookup(sk, fl6.flowlabel);
155 if (flowlabel == NULL)
156 return -EINVAL;
157 usin->sin6_addr = flowlabel->dst;
158 fl6_sock_release(flowlabel);
159 }
160 }
161
162 /*
163 * connect() to INADDR_ANY means loopback (BSD'ism).
164 */
165
166 if(ipv6_addr_any(&usin->sin6_addr))
167 usin->sin6_addr.s6_addr[15] = 0x1;
168
169 addr_type = ipv6_addr_type(&usin->sin6_addr);
170
171 if(addr_type & IPV6_ADDR_MULTICAST)
172 return -ENETUNREACH;
173
174 if (addr_type&IPV6_ADDR_LINKLOCAL) {
175 if (addr_len >= sizeof(struct sockaddr_in6) &&
176 usin->sin6_scope_id) {
177 /* If interface is set while binding, indices
178 * must coincide.
179 */
180 if (sk->sk_bound_dev_if &&
181 sk->sk_bound_dev_if != usin->sin6_scope_id)
182 return -EINVAL;
183
184 sk->sk_bound_dev_if = usin->sin6_scope_id;
185 }
186
187 /* Connect to link-local address requires an interface */
188 if (!sk->sk_bound_dev_if)
189 return -EINVAL;
190 }
191
192 if (tp->rx_opt.ts_recent_stamp &&
193 !ipv6_addr_equal(&np->daddr, &usin->sin6_addr)) {
194 tp->rx_opt.ts_recent = 0;
195 tp->rx_opt.ts_recent_stamp = 0;
196 tp->write_seq = 0;
197 }
198
199 np->daddr = usin->sin6_addr;
200 np->flow_label = fl6.flowlabel;
201
202 /*
203 * TCP over IPv4
204 */
205
206 if (addr_type == IPV6_ADDR_MAPPED) {
207 u32 exthdrlen = icsk->icsk_ext_hdr_len;
208 struct sockaddr_in sin;
209
210 SOCK_DEBUG(sk, "connect: ipv4 mapped\n");
211
212 if (__ipv6_only_sock(sk))
213 return -ENETUNREACH;
214
215 sin.sin_family = AF_INET;
216 sin.sin_port = usin->sin6_port;
217 sin.sin_addr.s_addr = usin->sin6_addr.s6_addr32[3];
218
219 icsk->icsk_af_ops = &ipv6_mapped;
220 sk->sk_backlog_rcv = tcp_v4_do_rcv;
221 #ifdef CONFIG_TCP_MD5SIG
222 tp->af_specific = &tcp_sock_ipv6_mapped_specific;
223 #endif
224
225 err = tcp_v4_connect(sk, (struct sockaddr *)&sin, sizeof(sin));
226
227 if (err) {
228 icsk->icsk_ext_hdr_len = exthdrlen;
229 icsk->icsk_af_ops = &ipv6_specific;
230 sk->sk_backlog_rcv = tcp_v6_do_rcv;
231 #ifdef CONFIG_TCP_MD5SIG
232 tp->af_specific = &tcp_sock_ipv6_specific;
233 #endif
234 goto failure;
235 } else {
236 ipv6_addr_set_v4mapped(inet->inet_saddr, &np->saddr);
237 ipv6_addr_set_v4mapped(inet->inet_rcv_saddr,
238 &np->rcv_saddr);
239 }
240
241 return err;
242 }
243
244 if (!ipv6_addr_any(&np->rcv_saddr))
245 saddr = &np->rcv_saddr;
246
247 fl6.flowi6_proto = IPPROTO_TCP;
248 fl6.daddr = np->daddr;
249 fl6.saddr = 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 np->rcv_saddr = *saddr;
268 }
269
270 /* set the source address */
271 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 fl6.daddr = np->daddr;
402 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 fl6.daddr = treq->rmt_addr;
493 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 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 tp->md5sig_info->keys6[tp->md5sig_info->entries6].addr = *peer;
621 tp->md5sig_info->keys6[tp->md5sig_info->entries6].base.key = newkey;
622 tp->md5sig_info->keys6[tp->md5sig_info->entries6].base.keylen = newkeylen;
623
624 tp->md5sig_info->entries6++;
625 }
626 return 0;
627 }
628
629 static int tcp_v6_md5_add_func(struct sock *sk, struct sock *addr_sk,
630 u8 *newkey, __u8 newkeylen)
631 {
632 return tcp_v6_md5_do_add(sk, &inet6_sk(addr_sk)->daddr,
633 newkey, newkeylen);
634 }
635
636 static int tcp_v6_md5_do_del(struct sock *sk, const struct in6_addr *peer)
637 {
638 struct tcp_sock *tp = tcp_sk(sk);
639 int i;
640
641 for (i = 0; i < tp->md5sig_info->entries6; i++) {
642 if (ipv6_addr_equal(&tp->md5sig_info->keys6[i].addr, peer)) {
643 /* Free the key */
644 kfree(tp->md5sig_info->keys6[i].base.key);
645 tp->md5sig_info->entries6--;
646
647 if (tp->md5sig_info->entries6 == 0) {
648 kfree(tp->md5sig_info->keys6);
649 tp->md5sig_info->keys6 = NULL;
650 tp->md5sig_info->alloced6 = 0;
651 tcp_free_md5sig_pool();
652 } else {
653 /* shrink the database */
654 if (tp->md5sig_info->entries6 != i)
655 memmove(&tp->md5sig_info->keys6[i],
656 &tp->md5sig_info->keys6[i+1],
657 (tp->md5sig_info->entries6 - i)
658 * sizeof (tp->md5sig_info->keys6[0]));
659 }
660 return 0;
661 }
662 }
663 return -ENOENT;
664 }
665
666 static void tcp_v6_clear_md5_list (struct sock *sk)
667 {
668 struct tcp_sock *tp = tcp_sk(sk);
669 int i;
670
671 if (tp->md5sig_info->entries6) {
672 for (i = 0; i < tp->md5sig_info->entries6; i++)
673 kfree(tp->md5sig_info->keys6[i].base.key);
674 tp->md5sig_info->entries6 = 0;
675 tcp_free_md5sig_pool();
676 }
677
678 kfree(tp->md5sig_info->keys6);
679 tp->md5sig_info->keys6 = NULL;
680 tp->md5sig_info->alloced6 = 0;
681
682 if (tp->md5sig_info->entries4) {
683 for (i = 0; i < tp->md5sig_info->entries4; i++)
684 kfree(tp->md5sig_info->keys4[i].base.key);
685 tp->md5sig_info->entries4 = 0;
686 tcp_free_md5sig_pool();
687 }
688
689 kfree(tp->md5sig_info->keys4);
690 tp->md5sig_info->keys4 = NULL;
691 tp->md5sig_info->alloced4 = 0;
692 }
693
694 static int tcp_v6_parse_md5_keys (struct sock *sk, char __user *optval,
695 int optlen)
696 {
697 struct tcp_md5sig cmd;
698 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)&cmd.tcpm_addr;
699 u8 *newkey;
700
701 if (optlen < sizeof(cmd))
702 return -EINVAL;
703
704 if (copy_from_user(&cmd, optval, sizeof(cmd)))
705 return -EFAULT;
706
707 if (sin6->sin6_family != AF_INET6)
708 return -EINVAL;
709
710 if (!cmd.tcpm_keylen) {
711 if (!tcp_sk(sk)->md5sig_info)
712 return -ENOENT;
713 if (ipv6_addr_v4mapped(&sin6->sin6_addr))
714 return tcp_v4_md5_do_del(sk, sin6->sin6_addr.s6_addr32[3]);
715 return tcp_v6_md5_do_del(sk, &sin6->sin6_addr);
716 }
717
718 if (cmd.tcpm_keylen > TCP_MD5SIG_MAXKEYLEN)
719 return -EINVAL;
720
721 if (!tcp_sk(sk)->md5sig_info) {
722 struct tcp_sock *tp = tcp_sk(sk);
723 struct tcp_md5sig_info *p;
724
725 p = kzalloc(sizeof(struct tcp_md5sig_info), GFP_KERNEL);
726 if (!p)
727 return -ENOMEM;
728
729 tp->md5sig_info = p;
730 sk_nocaps_add(sk, NETIF_F_GSO_MASK);
731 }
732
733 newkey = kmemdup(cmd.tcpm_key, cmd.tcpm_keylen, GFP_KERNEL);
734 if (!newkey)
735 return -ENOMEM;
736 if (ipv6_addr_v4mapped(&sin6->sin6_addr)) {
737 return tcp_v4_md5_do_add(sk, sin6->sin6_addr.s6_addr32[3],
738 newkey, cmd.tcpm_keylen);
739 }
740 return tcp_v6_md5_do_add(sk, &sin6->sin6_addr, newkey, cmd.tcpm_keylen);
741 }
742
743 static int tcp_v6_md5_hash_pseudoheader(struct tcp_md5sig_pool *hp,
744 const struct in6_addr *daddr,
745 const struct in6_addr *saddr, int nbytes)
746 {
747 struct tcp6_pseudohdr *bp;
748 struct scatterlist sg;
749
750 bp = &hp->md5_blk.ip6;
751 /* 1. TCP pseudo-header (RFC2460) */
752 bp->saddr = *saddr;
753 bp->daddr = *daddr;
754 bp->protocol = cpu_to_be32(IPPROTO_TCP);
755 bp->len = cpu_to_be32(nbytes);
756
757 sg_init_one(&sg, bp, sizeof(*bp));
758 return crypto_hash_update(&hp->md5_desc, &sg, sizeof(*bp));
759 }
760
761 static int tcp_v6_md5_hash_hdr(char *md5_hash, struct tcp_md5sig_key *key,
762 const struct in6_addr *daddr, struct in6_addr *saddr,
763 const struct tcphdr *th)
764 {
765 struct tcp_md5sig_pool *hp;
766 struct hash_desc *desc;
767
768 hp = tcp_get_md5sig_pool();
769 if (!hp)
770 goto clear_hash_noput;
771 desc = &hp->md5_desc;
772
773 if (crypto_hash_init(desc))
774 goto clear_hash;
775 if (tcp_v6_md5_hash_pseudoheader(hp, daddr, saddr, th->doff << 2))
776 goto clear_hash;
777 if (tcp_md5_hash_header(hp, th))
778 goto clear_hash;
779 if (tcp_md5_hash_key(hp, key))
780 goto clear_hash;
781 if (crypto_hash_final(desc, md5_hash))
782 goto clear_hash;
783
784 tcp_put_md5sig_pool();
785 return 0;
786
787 clear_hash:
788 tcp_put_md5sig_pool();
789 clear_hash_noput:
790 memset(md5_hash, 0, 16);
791 return 1;
792 }
793
794 static int tcp_v6_md5_hash_skb(char *md5_hash, struct tcp_md5sig_key *key,
795 const struct sock *sk,
796 const struct request_sock *req,
797 const struct sk_buff *skb)
798 {
799 const struct in6_addr *saddr, *daddr;
800 struct tcp_md5sig_pool *hp;
801 struct hash_desc *desc;
802 const struct tcphdr *th = tcp_hdr(skb);
803
804 if (sk) {
805 saddr = &inet6_sk(sk)->saddr;
806 daddr = &inet6_sk(sk)->daddr;
807 } else if (req) {
808 saddr = &inet6_rsk(req)->loc_addr;
809 daddr = &inet6_rsk(req)->rmt_addr;
810 } else {
811 const struct ipv6hdr *ip6h = ipv6_hdr(skb);
812 saddr = &ip6h->saddr;
813 daddr = &ip6h->daddr;
814 }
815
816 hp = tcp_get_md5sig_pool();
817 if (!hp)
818 goto clear_hash_noput;
819 desc = &hp->md5_desc;
820
821 if (crypto_hash_init(desc))
822 goto clear_hash;
823
824 if (tcp_v6_md5_hash_pseudoheader(hp, daddr, saddr, skb->len))
825 goto clear_hash;
826 if (tcp_md5_hash_header(hp, th))
827 goto clear_hash;
828 if (tcp_md5_hash_skb_data(hp, skb, th->doff << 2))
829 goto clear_hash;
830 if (tcp_md5_hash_key(hp, key))
831 goto clear_hash;
832 if (crypto_hash_final(desc, md5_hash))
833 goto clear_hash;
834
835 tcp_put_md5sig_pool();
836 return 0;
837
838 clear_hash:
839 tcp_put_md5sig_pool();
840 clear_hash_noput:
841 memset(md5_hash, 0, 16);
842 return 1;
843 }
844
845 static int tcp_v6_inbound_md5_hash(struct sock *sk, const struct sk_buff *skb)
846 {
847 const __u8 *hash_location = NULL;
848 struct tcp_md5sig_key *hash_expected;
849 const struct ipv6hdr *ip6h = ipv6_hdr(skb);
850 const struct tcphdr *th = tcp_hdr(skb);
851 int genhash;
852 u8 newhash[16];
853
854 hash_expected = tcp_v6_md5_do_lookup(sk, &ip6h->saddr);
855 hash_location = tcp_parse_md5sig_option(th);
856
857 /* We've parsed the options - do we have a hash? */
858 if (!hash_expected && !hash_location)
859 return 0;
860
861 if (hash_expected && !hash_location) {
862 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPMD5NOTFOUND);
863 return 1;
864 }
865
866 if (!hash_expected && hash_location) {
867 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPMD5UNEXPECTED);
868 return 1;
869 }
870
871 /* check the signature */
872 genhash = tcp_v6_md5_hash_skb(newhash,
873 hash_expected,
874 NULL, NULL, skb);
875
876 if (genhash || memcmp(hash_location, newhash, 16) != 0) {
877 if (net_ratelimit()) {
878 printk(KERN_INFO "MD5 Hash %s for [%pI6c]:%u->[%pI6c]:%u\n",
879 genhash ? "failed" : "mismatch",
880 &ip6h->saddr, ntohs(th->source),
881 &ip6h->daddr, ntohs(th->dest));
882 }
883 return 1;
884 }
885 return 0;
886 }
887 #endif
888
889 struct request_sock_ops tcp6_request_sock_ops __read_mostly = {
890 .family = AF_INET6,
891 .obj_size = sizeof(struct tcp6_request_sock),
892 .rtx_syn_ack = tcp_v6_rtx_synack,
893 .send_ack = tcp_v6_reqsk_send_ack,
894 .destructor = tcp_v6_reqsk_destructor,
895 .send_reset = tcp_v6_send_reset,
896 .syn_ack_timeout = tcp_syn_ack_timeout,
897 };
898
899 #ifdef CONFIG_TCP_MD5SIG
900 static const struct tcp_request_sock_ops tcp_request_sock_ipv6_ops = {
901 .md5_lookup = tcp_v6_reqsk_md5_lookup,
902 .calc_md5_hash = tcp_v6_md5_hash_skb,
903 };
904 #endif
905
906 static void __tcp_v6_send_check(struct sk_buff *skb,
907 const struct in6_addr *saddr, const struct in6_addr *daddr)
908 {
909 struct tcphdr *th = tcp_hdr(skb);
910
911 if (skb->ip_summed == CHECKSUM_PARTIAL) {
912 th->check = ~tcp_v6_check(skb->len, saddr, daddr, 0);
913 skb->csum_start = skb_transport_header(skb) - skb->head;
914 skb->csum_offset = offsetof(struct tcphdr, check);
915 } else {
916 th->check = tcp_v6_check(skb->len, saddr, daddr,
917 csum_partial(th, th->doff << 2,
918 skb->csum));
919 }
920 }
921
922 static void tcp_v6_send_check(struct sock *sk, struct sk_buff *skb)
923 {
924 struct ipv6_pinfo *np = inet6_sk(sk);
925
926 __tcp_v6_send_check(skb, &np->saddr, &np->daddr);
927 }
928
929 static int tcp_v6_gso_send_check(struct sk_buff *skb)
930 {
931 const struct ipv6hdr *ipv6h;
932 struct tcphdr *th;
933
934 if (!pskb_may_pull(skb, sizeof(*th)))
935 return -EINVAL;
936
937 ipv6h = ipv6_hdr(skb);
938 th = tcp_hdr(skb);
939
940 th->check = 0;
941 skb->ip_summed = CHECKSUM_PARTIAL;
942 __tcp_v6_send_check(skb, &ipv6h->saddr, &ipv6h->daddr);
943 return 0;
944 }
945
946 static struct sk_buff **tcp6_gro_receive(struct sk_buff **head,
947 struct sk_buff *skb)
948 {
949 const struct ipv6hdr *iph = skb_gro_network_header(skb);
950
951 switch (skb->ip_summed) {
952 case CHECKSUM_COMPLETE:
953 if (!tcp_v6_check(skb_gro_len(skb), &iph->saddr, &iph->daddr,
954 skb->csum)) {
955 skb->ip_summed = CHECKSUM_UNNECESSARY;
956 break;
957 }
958
959 /* fall through */
960 case CHECKSUM_NONE:
961 NAPI_GRO_CB(skb)->flush = 1;
962 return NULL;
963 }
964
965 return tcp_gro_receive(head, skb);
966 }
967
968 static int tcp6_gro_complete(struct sk_buff *skb)
969 {
970 const struct ipv6hdr *iph = ipv6_hdr(skb);
971 struct tcphdr *th = tcp_hdr(skb);
972
973 th->check = ~tcp_v6_check(skb->len - skb_transport_offset(skb),
974 &iph->saddr, &iph->daddr, 0);
975 skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6;
976
977 return tcp_gro_complete(skb);
978 }
979
980 static void tcp_v6_send_response(struct sk_buff *skb, u32 seq, u32 ack, u32 win,
981 u32 ts, struct tcp_md5sig_key *key, int rst, u8 tclass)
982 {
983 const struct tcphdr *th = tcp_hdr(skb);
984 struct tcphdr *t1;
985 struct sk_buff *buff;
986 struct flowi6 fl6;
987 struct net *net = dev_net(skb_dst(skb)->dev);
988 struct sock *ctl_sk = net->ipv6.tcp_sk;
989 unsigned int tot_len = sizeof(struct tcphdr);
990 struct dst_entry *dst;
991 __be32 *topt;
992
993 if (ts)
994 tot_len += TCPOLEN_TSTAMP_ALIGNED;
995 #ifdef CONFIG_TCP_MD5SIG
996 if (key)
997 tot_len += TCPOLEN_MD5SIG_ALIGNED;
998 #endif
999
1000 buff = alloc_skb(MAX_HEADER + sizeof(struct ipv6hdr) + tot_len,
1001 GFP_ATOMIC);
1002 if (buff == NULL)
1003 return;
1004
1005 skb_reserve(buff, MAX_HEADER + sizeof(struct ipv6hdr) + tot_len);
1006
1007 t1 = (struct tcphdr *) skb_push(buff, tot_len);
1008 skb_reset_transport_header(buff);
1009
1010 /* Swap the send and the receive. */
1011 memset(t1, 0, sizeof(*t1));
1012 t1->dest = th->source;
1013 t1->source = th->dest;
1014 t1->doff = tot_len / 4;
1015 t1->seq = htonl(seq);
1016 t1->ack_seq = htonl(ack);
1017 t1->ack = !rst || !th->ack;
1018 t1->rst = rst;
1019 t1->window = htons(win);
1020
1021 topt = (__be32 *)(t1 + 1);
1022
1023 if (ts) {
1024 *topt++ = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
1025 (TCPOPT_TIMESTAMP << 8) | TCPOLEN_TIMESTAMP);
1026 *topt++ = htonl(tcp_time_stamp);
1027 *topt++ = htonl(ts);
1028 }
1029
1030 #ifdef CONFIG_TCP_MD5SIG
1031 if (key) {
1032 *topt++ = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
1033 (TCPOPT_MD5SIG << 8) | TCPOLEN_MD5SIG);
1034 tcp_v6_md5_hash_hdr((__u8 *)topt, key,
1035 &ipv6_hdr(skb)->saddr,
1036 &ipv6_hdr(skb)->daddr, t1);
1037 }
1038 #endif
1039
1040 memset(&fl6, 0, sizeof(fl6));
1041 fl6.daddr = ipv6_hdr(skb)->saddr;
1042 fl6.saddr = ipv6_hdr(skb)->daddr;
1043
1044 buff->ip_summed = CHECKSUM_PARTIAL;
1045 buff->csum = 0;
1046
1047 __tcp_v6_send_check(buff, &fl6.saddr, &fl6.daddr);
1048
1049 fl6.flowi6_proto = IPPROTO_TCP;
1050 fl6.flowi6_oif = inet6_iif(skb);
1051 fl6.fl6_dport = t1->dest;
1052 fl6.fl6_sport = t1->source;
1053 security_skb_classify_flow(skb, flowi6_to_flowi(&fl6));
1054
1055 /* Pass a socket to ip6_dst_lookup either it is for RST
1056 * Underlying function will use this to retrieve the network
1057 * namespace
1058 */
1059 dst = ip6_dst_lookup_flow(ctl_sk, &fl6, NULL, false);
1060 if (!IS_ERR(dst)) {
1061 skb_dst_set(buff, dst);
1062 ip6_xmit(ctl_sk, buff, &fl6, NULL, tclass);
1063 TCP_INC_STATS_BH(net, TCP_MIB_OUTSEGS);
1064 if (rst)
1065 TCP_INC_STATS_BH(net, TCP_MIB_OUTRSTS);
1066 return;
1067 }
1068
1069 kfree_skb(buff);
1070 }
1071
1072 static void tcp_v6_send_reset(struct sock *sk, struct sk_buff *skb)
1073 {
1074 const struct tcphdr *th = tcp_hdr(skb);
1075 u32 seq = 0, ack_seq = 0;
1076 struct tcp_md5sig_key *key = NULL;
1077
1078 if (th->rst)
1079 return;
1080
1081 if (!ipv6_unicast_destination(skb))
1082 return;
1083
1084 #ifdef CONFIG_TCP_MD5SIG
1085 if (sk)
1086 key = tcp_v6_md5_do_lookup(sk, &ipv6_hdr(skb)->saddr);
1087 #endif
1088
1089 if (th->ack)
1090 seq = ntohl(th->ack_seq);
1091 else
1092 ack_seq = ntohl(th->seq) + th->syn + th->fin + skb->len -
1093 (th->doff << 2);
1094
1095 tcp_v6_send_response(skb, seq, ack_seq, 0, 0, key, 1, 0);
1096 }
1097
1098 static void tcp_v6_send_ack(struct sk_buff *skb, u32 seq, u32 ack, u32 win, u32 ts,
1099 struct tcp_md5sig_key *key, u8 tclass)
1100 {
1101 tcp_v6_send_response(skb, seq, ack, win, ts, key, 0, tclass);
1102 }
1103
1104 static void tcp_v6_timewait_ack(struct sock *sk, struct sk_buff *skb)
1105 {
1106 struct inet_timewait_sock *tw = inet_twsk(sk);
1107 struct tcp_timewait_sock *tcptw = tcp_twsk(sk);
1108
1109 tcp_v6_send_ack(skb, tcptw->tw_snd_nxt, tcptw->tw_rcv_nxt,
1110 tcptw->tw_rcv_wnd >> tw->tw_rcv_wscale,
1111 tcptw->tw_ts_recent, tcp_twsk_md5_key(tcptw),
1112 tw->tw_tclass);
1113
1114 inet_twsk_put(tw);
1115 }
1116
1117 static void tcp_v6_reqsk_send_ack(struct sock *sk, struct sk_buff *skb,
1118 struct request_sock *req)
1119 {
1120 tcp_v6_send_ack(skb, tcp_rsk(req)->snt_isn + 1, tcp_rsk(req)->rcv_isn + 1, req->rcv_wnd, req->ts_recent,
1121 tcp_v6_md5_do_lookup(sk, &ipv6_hdr(skb)->daddr), 0);
1122 }
1123
1124
1125 static struct sock *tcp_v6_hnd_req(struct sock *sk,struct sk_buff *skb)
1126 {
1127 struct request_sock *req, **prev;
1128 const struct tcphdr *th = tcp_hdr(skb);
1129 struct sock *nsk;
1130
1131 /* Find possible connection requests. */
1132 req = inet6_csk_search_req(sk, &prev, th->source,
1133 &ipv6_hdr(skb)->saddr,
1134 &ipv6_hdr(skb)->daddr, inet6_iif(skb));
1135 if (req)
1136 return tcp_check_req(sk, skb, req, prev);
1137
1138 nsk = __inet6_lookup_established(sock_net(sk), &tcp_hashinfo,
1139 &ipv6_hdr(skb)->saddr, th->source,
1140 &ipv6_hdr(skb)->daddr, ntohs(th->dest), inet6_iif(skb));
1141
1142 if (nsk) {
1143 if (nsk->sk_state != TCP_TIME_WAIT) {
1144 bh_lock_sock(nsk);
1145 return nsk;
1146 }
1147 inet_twsk_put(inet_twsk(nsk));
1148 return NULL;
1149 }
1150
1151 #ifdef CONFIG_SYN_COOKIES
1152 if (!th->syn)
1153 sk = cookie_v6_check(sk, skb);
1154 #endif
1155 return sk;
1156 }
1157
1158 /* FIXME: this is substantially similar to the ipv4 code.
1159 * Can some kind of merge be done? -- erics
1160 */
1161 static int tcp_v6_conn_request(struct sock *sk, struct sk_buff *skb)
1162 {
1163 struct tcp_extend_values tmp_ext;
1164 struct tcp_options_received tmp_opt;
1165 const u8 *hash_location;
1166 struct request_sock *req;
1167 struct inet6_request_sock *treq;
1168 struct ipv6_pinfo *np = inet6_sk(sk);
1169 struct tcp_sock *tp = tcp_sk(sk);
1170 __u32 isn = TCP_SKB_CB(skb)->when;
1171 struct dst_entry *dst = NULL;
1172 int want_cookie = 0;
1173
1174 if (skb->protocol == htons(ETH_P_IP))
1175 return tcp_v4_conn_request(sk, skb);
1176
1177 if (!ipv6_unicast_destination(skb))
1178 goto drop;
1179
1180 if (inet_csk_reqsk_queue_is_full(sk) && !isn) {
1181 want_cookie = tcp_syn_flood_action(sk, skb, "TCPv6");
1182 if (!want_cookie)
1183 goto drop;
1184 }
1185
1186 if (sk_acceptq_is_full(sk) && inet_csk_reqsk_queue_young(sk) > 1)
1187 goto drop;
1188
1189 req = inet6_reqsk_alloc(&tcp6_request_sock_ops);
1190 if (req == NULL)
1191 goto drop;
1192
1193 #ifdef CONFIG_TCP_MD5SIG
1194 tcp_rsk(req)->af_specific = &tcp_request_sock_ipv6_ops;
1195 #endif
1196
1197 tcp_clear_options(&tmp_opt);
1198 tmp_opt.mss_clamp = IPV6_MIN_MTU - sizeof(struct tcphdr) - sizeof(struct ipv6hdr);
1199 tmp_opt.user_mss = tp->rx_opt.user_mss;
1200 tcp_parse_options(skb, &tmp_opt, &hash_location, 0);
1201
1202 if (tmp_opt.cookie_plus > 0 &&
1203 tmp_opt.saw_tstamp &&
1204 !tp->rx_opt.cookie_out_never &&
1205 (sysctl_tcp_cookie_size > 0 ||
1206 (tp->cookie_values != NULL &&
1207 tp->cookie_values->cookie_desired > 0))) {
1208 u8 *c;
1209 u32 *d;
1210 u32 *mess = &tmp_ext.cookie_bakery[COOKIE_DIGEST_WORDS];
1211 int l = tmp_opt.cookie_plus - TCPOLEN_COOKIE_BASE;
1212
1213 if (tcp_cookie_generator(&tmp_ext.cookie_bakery[0]) != 0)
1214 goto drop_and_free;
1215
1216 /* Secret recipe starts with IP addresses */
1217 d = (__force u32 *)&ipv6_hdr(skb)->daddr.s6_addr32[0];
1218 *mess++ ^= *d++;
1219 *mess++ ^= *d++;
1220 *mess++ ^= *d++;
1221 *mess++ ^= *d++;
1222 d = (__force u32 *)&ipv6_hdr(skb)->saddr.s6_addr32[0];
1223 *mess++ ^= *d++;
1224 *mess++ ^= *d++;
1225 *mess++ ^= *d++;
1226 *mess++ ^= *d++;
1227
1228 /* plus variable length Initiator Cookie */
1229 c = (u8 *)mess;
1230 while (l-- > 0)
1231 *c++ ^= *hash_location++;
1232
1233 want_cookie = 0; /* not our kind of cookie */
1234 tmp_ext.cookie_out_never = 0; /* false */
1235 tmp_ext.cookie_plus = tmp_opt.cookie_plus;
1236 } else if (!tp->rx_opt.cookie_in_always) {
1237 /* redundant indications, but ensure initialization. */
1238 tmp_ext.cookie_out_never = 1; /* true */
1239 tmp_ext.cookie_plus = 0;
1240 } else {
1241 goto drop_and_free;
1242 }
1243 tmp_ext.cookie_in_always = tp->rx_opt.cookie_in_always;
1244
1245 if (want_cookie && !tmp_opt.saw_tstamp)
1246 tcp_clear_options(&tmp_opt);
1247
1248 tmp_opt.tstamp_ok = tmp_opt.saw_tstamp;
1249 tcp_openreq_init(req, &tmp_opt, skb);
1250
1251 treq = inet6_rsk(req);
1252 treq->rmt_addr = ipv6_hdr(skb)->saddr;
1253 treq->loc_addr = ipv6_hdr(skb)->daddr;
1254 if (!want_cookie || tmp_opt.tstamp_ok)
1255 TCP_ECN_create_request(req, tcp_hdr(skb));
1256
1257 treq->iif = sk->sk_bound_dev_if;
1258
1259 /* So that link locals have meaning */
1260 if (!sk->sk_bound_dev_if &&
1261 ipv6_addr_type(&treq->rmt_addr) & IPV6_ADDR_LINKLOCAL)
1262 treq->iif = inet6_iif(skb);
1263
1264 if (!isn) {
1265 struct inet_peer *peer = NULL;
1266
1267 if (ipv6_opt_accepted(sk, skb) ||
1268 np->rxopt.bits.rxinfo || np->rxopt.bits.rxoinfo ||
1269 np->rxopt.bits.rxhlim || np->rxopt.bits.rxohlim) {
1270 atomic_inc(&skb->users);
1271 treq->pktopts = skb;
1272 }
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 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 newnp->daddr = treq->rmt_addr;
1448 newnp->saddr = treq->loc_addr;
1449 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 if (tcp_sk(sk)->rx_opt.user_mss &&
1497 tcp_sk(sk)->rx_opt.user_mss < newtp->advmss)
1498 newtp->advmss = tcp_sk(sk)->rx_opt.user_mss;
1499
1500 tcp_initialize_rcv_mss(newsk);
1501 if (tcp_rsk(req)->snt_synack)
1502 tcp_valid_rtt_meas(newsk,
1503 tcp_time_stamp - tcp_rsk(req)->snt_synack);
1504 newtp->total_retrans = req->retrans;
1505
1506 newinet->inet_daddr = newinet->inet_saddr = LOOPBACK4_IPV6;
1507 newinet->inet_rcv_saddr = LOOPBACK4_IPV6;
1508
1509 #ifdef CONFIG_TCP_MD5SIG
1510 /* Copy over the MD5 key from the original socket */
1511 if ((key = tcp_v6_md5_do_lookup(sk, &newnp->daddr)) != NULL) {
1512 /* We're using one, so create a matching key
1513 * on the newsk structure. If we fail to get
1514 * memory, then we end up not copying the key
1515 * across. Shucks.
1516 */
1517 char *newkey = kmemdup(key->key, key->keylen, GFP_ATOMIC);
1518 if (newkey != NULL)
1519 tcp_v6_md5_do_add(newsk, &newnp->daddr,
1520 newkey, key->keylen);
1521 }
1522 #endif
1523
1524 if (__inet_inherit_port(sk, newsk) < 0) {
1525 sock_put(newsk);
1526 goto out;
1527 }
1528 __inet6_hash(newsk, NULL);
1529
1530 return newsk;
1531
1532 out_overflow:
1533 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENOVERFLOWS);
1534 out_nonewsk:
1535 if (opt && opt != np->opt)
1536 sock_kfree_s(sk, opt, opt->tot_len);
1537 dst_release(dst);
1538 out:
1539 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENDROPS);
1540 return NULL;
1541 }
1542
1543 static __sum16 tcp_v6_checksum_init(struct sk_buff *skb)
1544 {
1545 if (skb->ip_summed == CHECKSUM_COMPLETE) {
1546 if (!tcp_v6_check(skb->len, &ipv6_hdr(skb)->saddr,
1547 &ipv6_hdr(skb)->daddr, skb->csum)) {
1548 skb->ip_summed = CHECKSUM_UNNECESSARY;
1549 return 0;
1550 }
1551 }
1552
1553 skb->csum = ~csum_unfold(tcp_v6_check(skb->len,
1554 &ipv6_hdr(skb)->saddr,
1555 &ipv6_hdr(skb)->daddr, 0));
1556
1557 if (skb->len <= 76) {
1558 return __skb_checksum_complete(skb);
1559 }
1560 return 0;
1561 }
1562
1563 /* The socket must have it's spinlock held when we get
1564 * here.
1565 *
1566 * We have a potential double-lock case here, so even when
1567 * doing backlog processing we use the BH locking scheme.
1568 * This is because we cannot sleep with the original spinlock
1569 * held.
1570 */
1571 static int tcp_v6_do_rcv(struct sock *sk, struct sk_buff *skb)
1572 {
1573 struct ipv6_pinfo *np = inet6_sk(sk);
1574 struct tcp_sock *tp;
1575 struct sk_buff *opt_skb = NULL;
1576
1577 /* Imagine: socket is IPv6. IPv4 packet arrives,
1578 goes to IPv4 receive handler and backlogged.
1579 From backlog it always goes here. Kerboom...
1580 Fortunately, tcp_rcv_established and rcv_established
1581 handle them correctly, but it is not case with
1582 tcp_v6_hnd_req and tcp_v6_send_reset(). --ANK
1583 */
1584
1585 if (skb->protocol == htons(ETH_P_IP))
1586 return tcp_v4_do_rcv(sk, skb);
1587
1588 #ifdef CONFIG_TCP_MD5SIG
1589 if (tcp_v6_inbound_md5_hash (sk, skb))
1590 goto discard;
1591 #endif
1592
1593 if (sk_filter(sk, skb))
1594 goto discard;
1595
1596 /*
1597 * socket locking is here for SMP purposes as backlog rcv
1598 * is currently called with bh processing disabled.
1599 */
1600
1601 /* Do Stevens' IPV6_PKTOPTIONS.
1602
1603 Yes, guys, it is the only place in our code, where we
1604 may make it not affecting IPv4.
1605 The rest of code is protocol independent,
1606 and I do not like idea to uglify IPv4.
1607
1608 Actually, all the idea behind IPV6_PKTOPTIONS
1609 looks not very well thought. For now we latch
1610 options, received in the last packet, enqueued
1611 by tcp. Feel free to propose better solution.
1612 --ANK (980728)
1613 */
1614 if (np->rxopt.all)
1615 opt_skb = skb_clone(skb, GFP_ATOMIC);
1616
1617 if (sk->sk_state == TCP_ESTABLISHED) { /* Fast path */
1618 sock_rps_save_rxhash(sk, skb);
1619 if (tcp_rcv_established(sk, skb, tcp_hdr(skb), skb->len))
1620 goto reset;
1621 if (opt_skb)
1622 goto ipv6_pktoptions;
1623 return 0;
1624 }
1625
1626 if (skb->len < tcp_hdrlen(skb) || tcp_checksum_complete(skb))
1627 goto csum_err;
1628
1629 if (sk->sk_state == TCP_LISTEN) {
1630 struct sock *nsk = tcp_v6_hnd_req(sk, skb);
1631 if (!nsk)
1632 goto discard;
1633
1634 /*
1635 * Queue it on the new socket if the new socket is active,
1636 * otherwise we just shortcircuit this and continue with
1637 * the new socket..
1638 */
1639 if(nsk != sk) {
1640 sock_rps_save_rxhash(nsk, skb);
1641 if (tcp_child_process(sk, nsk, skb))
1642 goto reset;
1643 if (opt_skb)
1644 __kfree_skb(opt_skb);
1645 return 0;
1646 }
1647 } else
1648 sock_rps_save_rxhash(sk, skb);
1649
1650 if (tcp_rcv_state_process(sk, skb, tcp_hdr(skb), skb->len))
1651 goto reset;
1652 if (opt_skb)
1653 goto ipv6_pktoptions;
1654 return 0;
1655
1656 reset:
1657 tcp_v6_send_reset(sk, skb);
1658 discard:
1659 if (opt_skb)
1660 __kfree_skb(opt_skb);
1661 kfree_skb(skb);
1662 return 0;
1663 csum_err:
1664 TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_INERRS);
1665 goto discard;
1666
1667
1668 ipv6_pktoptions:
1669 /* Do you ask, what is it?
1670
1671 1. skb was enqueued by tcp.
1672 2. skb is added to tail of read queue, rather than out of order.
1673 3. socket is not in passive state.
1674 4. Finally, it really contains options, which user wants to receive.
1675 */
1676 tp = tcp_sk(sk);
1677 if (TCP_SKB_CB(opt_skb)->end_seq == tp->rcv_nxt &&
1678 !((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_LISTEN))) {
1679 if (np->rxopt.bits.rxinfo || np->rxopt.bits.rxoinfo)
1680 np->mcast_oif = inet6_iif(opt_skb);
1681 if (np->rxopt.bits.rxhlim || np->rxopt.bits.rxohlim)
1682 np->mcast_hops = ipv6_hdr(opt_skb)->hop_limit;
1683 if (ipv6_opt_accepted(sk, opt_skb)) {
1684 skb_set_owner_r(opt_skb, sk);
1685 opt_skb = xchg(&np->pktoptions, opt_skb);
1686 } else {
1687 __kfree_skb(opt_skb);
1688 opt_skb = xchg(&np->pktoptions, NULL);
1689 }
1690 }
1691
1692 kfree_skb(opt_skb);
1693 return 0;
1694 }
1695
1696 static int tcp_v6_rcv(struct sk_buff *skb)
1697 {
1698 const struct tcphdr *th;
1699 const struct ipv6hdr *hdr;
1700 struct sock *sk;
1701 int ret;
1702 struct net *net = dev_net(skb->dev);
1703
1704 if (skb->pkt_type != PACKET_HOST)
1705 goto discard_it;
1706
1707 /*
1708 * Count it even if it's bad.
1709 */
1710 TCP_INC_STATS_BH(net, TCP_MIB_INSEGS);
1711
1712 if (!pskb_may_pull(skb, sizeof(struct tcphdr)))
1713 goto discard_it;
1714
1715 th = tcp_hdr(skb);
1716
1717 if (th->doff < sizeof(struct tcphdr)/4)
1718 goto bad_packet;
1719 if (!pskb_may_pull(skb, th->doff*4))
1720 goto discard_it;
1721
1722 if (!skb_csum_unnecessary(skb) && tcp_v6_checksum_init(skb))
1723 goto bad_packet;
1724
1725 th = tcp_hdr(skb);
1726 hdr = ipv6_hdr(skb);
1727 TCP_SKB_CB(skb)->seq = ntohl(th->seq);
1728 TCP_SKB_CB(skb)->end_seq = (TCP_SKB_CB(skb)->seq + th->syn + th->fin +
1729 skb->len - th->doff*4);
1730 TCP_SKB_CB(skb)->ack_seq = ntohl(th->ack_seq);
1731 TCP_SKB_CB(skb)->when = 0;
1732 TCP_SKB_CB(skb)->ip_dsfield = ipv6_get_dsfield(hdr);
1733 TCP_SKB_CB(skb)->sacked = 0;
1734
1735 sk = __inet6_lookup_skb(&tcp_hashinfo, skb, th->source, th->dest);
1736 if (!sk)
1737 goto no_tcp_socket;
1738
1739 process:
1740 if (sk->sk_state == TCP_TIME_WAIT)
1741 goto do_time_wait;
1742
1743 if (hdr->hop_limit < inet6_sk(sk)->min_hopcount) {
1744 NET_INC_STATS_BH(net, LINUX_MIB_TCPMINTTLDROP);
1745 goto discard_and_relse;
1746 }
1747
1748 if (!xfrm6_policy_check(sk, XFRM_POLICY_IN, skb))
1749 goto discard_and_relse;
1750
1751 if (sk_filter(sk, skb))
1752 goto discard_and_relse;
1753
1754 skb->dev = NULL;
1755
1756 bh_lock_sock_nested(sk);
1757 ret = 0;
1758 if (!sock_owned_by_user(sk)) {
1759 #ifdef CONFIG_NET_DMA
1760 struct tcp_sock *tp = tcp_sk(sk);
1761 if (!tp->ucopy.dma_chan && tp->ucopy.pinned_list)
1762 tp->ucopy.dma_chan = dma_find_channel(DMA_MEMCPY);
1763 if (tp->ucopy.dma_chan)
1764 ret = tcp_v6_do_rcv(sk, skb);
1765 else
1766 #endif
1767 {
1768 if (!tcp_prequeue(sk, skb))
1769 ret = tcp_v6_do_rcv(sk, skb);
1770 }
1771 } else if (unlikely(sk_add_backlog(sk, skb))) {
1772 bh_unlock_sock(sk);
1773 NET_INC_STATS_BH(net, LINUX_MIB_TCPBACKLOGDROP);
1774 goto discard_and_relse;
1775 }
1776 bh_unlock_sock(sk);
1777
1778 sock_put(sk);
1779 return ret ? -1 : 0;
1780
1781 no_tcp_socket:
1782 if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb))
1783 goto discard_it;
1784
1785 if (skb->len < (th->doff<<2) || tcp_checksum_complete(skb)) {
1786 bad_packet:
1787 TCP_INC_STATS_BH(net, TCP_MIB_INERRS);
1788 } else {
1789 tcp_v6_send_reset(NULL, skb);
1790 }
1791
1792 discard_it:
1793
1794 /*
1795 * Discard frame
1796 */
1797
1798 kfree_skb(skb);
1799 return 0;
1800
1801 discard_and_relse:
1802 sock_put(sk);
1803 goto discard_it;
1804
1805 do_time_wait:
1806 if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb)) {
1807 inet_twsk_put(inet_twsk(sk));
1808 goto discard_it;
1809 }
1810
1811 if (skb->len < (th->doff<<2) || tcp_checksum_complete(skb)) {
1812 TCP_INC_STATS_BH(net, TCP_MIB_INERRS);
1813 inet_twsk_put(inet_twsk(sk));
1814 goto discard_it;
1815 }
1816
1817 switch (tcp_timewait_state_process(inet_twsk(sk), skb, th)) {
1818 case TCP_TW_SYN:
1819 {
1820 struct sock *sk2;
1821
1822 sk2 = inet6_lookup_listener(dev_net(skb->dev), &tcp_hashinfo,
1823 &ipv6_hdr(skb)->daddr,
1824 ntohs(th->dest), inet6_iif(skb));
1825 if (sk2 != NULL) {
1826 struct inet_timewait_sock *tw = inet_twsk(sk);
1827 inet_twsk_deschedule(tw, &tcp_death_row);
1828 inet_twsk_put(tw);
1829 sk = sk2;
1830 goto process;
1831 }
1832 /* Fall through to ACK */
1833 }
1834 case TCP_TW_ACK:
1835 tcp_v6_timewait_ack(sk, skb);
1836 break;
1837 case TCP_TW_RST:
1838 goto no_tcp_socket;
1839 case TCP_TW_SUCCESS:;
1840 }
1841 goto discard_it;
1842 }
1843
1844 static struct inet_peer *tcp_v6_get_peer(struct sock *sk, bool *release_it)
1845 {
1846 struct rt6_info *rt = (struct rt6_info *) __sk_dst_get(sk);
1847 struct ipv6_pinfo *np = inet6_sk(sk);
1848 struct inet_peer *peer;
1849
1850 if (!rt ||
1851 !ipv6_addr_equal(&np->daddr, &rt->rt6i_dst.addr)) {
1852 peer = inet_getpeer_v6(&np->daddr, 1);
1853 *release_it = true;
1854 } else {
1855 if (!rt->rt6i_peer)
1856 rt6_bind_peer(rt, 1);
1857 peer = rt->rt6i_peer;
1858 *release_it = false;
1859 }
1860
1861 return peer;
1862 }
1863
1864 static void *tcp_v6_tw_get_peer(struct sock *sk)
1865 {
1866 const struct inet6_timewait_sock *tw6 = inet6_twsk(sk);
1867 const struct inet_timewait_sock *tw = inet_twsk(sk);
1868
1869 if (tw->tw_family == AF_INET)
1870 return tcp_v4_tw_get_peer(sk);
1871
1872 return inet_getpeer_v6(&tw6->tw_v6_daddr, 1);
1873 }
1874
1875 static struct timewait_sock_ops tcp6_timewait_sock_ops = {
1876 .twsk_obj_size = sizeof(struct tcp6_timewait_sock),
1877 .twsk_unique = tcp_twsk_unique,
1878 .twsk_destructor= tcp_twsk_destructor,
1879 .twsk_getpeer = tcp_v6_tw_get_peer,
1880 };
1881
1882 static const struct inet_connection_sock_af_ops ipv6_specific = {
1883 .queue_xmit = inet6_csk_xmit,
1884 .send_check = tcp_v6_send_check,
1885 .rebuild_header = inet6_sk_rebuild_header,
1886 .conn_request = tcp_v6_conn_request,
1887 .syn_recv_sock = tcp_v6_syn_recv_sock,
1888 .get_peer = tcp_v6_get_peer,
1889 .net_header_len = sizeof(struct ipv6hdr),
1890 .setsockopt = ipv6_setsockopt,
1891 .getsockopt = ipv6_getsockopt,
1892 .addr2sockaddr = inet6_csk_addr2sockaddr,
1893 .sockaddr_len = sizeof(struct sockaddr_in6),
1894 .bind_conflict = inet6_csk_bind_conflict,
1895 #ifdef CONFIG_COMPAT
1896 .compat_setsockopt = compat_ipv6_setsockopt,
1897 .compat_getsockopt = compat_ipv6_getsockopt,
1898 #endif
1899 };
1900
1901 #ifdef CONFIG_TCP_MD5SIG
1902 static const struct tcp_sock_af_ops tcp_sock_ipv6_specific = {
1903 .md5_lookup = tcp_v6_md5_lookup,
1904 .calc_md5_hash = tcp_v6_md5_hash_skb,
1905 .md5_add = tcp_v6_md5_add_func,
1906 .md5_parse = tcp_v6_parse_md5_keys,
1907 };
1908 #endif
1909
1910 /*
1911 * TCP over IPv4 via INET6 API
1912 */
1913
1914 static const struct inet_connection_sock_af_ops ipv6_mapped = {
1915 .queue_xmit = ip_queue_xmit,
1916 .send_check = tcp_v4_send_check,
1917 .rebuild_header = inet_sk_rebuild_header,
1918 .conn_request = tcp_v6_conn_request,
1919 .syn_recv_sock = tcp_v6_syn_recv_sock,
1920 .get_peer = tcp_v4_get_peer,
1921 .net_header_len = sizeof(struct iphdr),
1922 .setsockopt = ipv6_setsockopt,
1923 .getsockopt = ipv6_getsockopt,
1924 .addr2sockaddr = inet6_csk_addr2sockaddr,
1925 .sockaddr_len = sizeof(struct sockaddr_in6),
1926 .bind_conflict = inet6_csk_bind_conflict,
1927 #ifdef CONFIG_COMPAT
1928 .compat_setsockopt = compat_ipv6_setsockopt,
1929 .compat_getsockopt = compat_ipv6_getsockopt,
1930 #endif
1931 };
1932
1933 #ifdef CONFIG_TCP_MD5SIG
1934 static const struct tcp_sock_af_ops tcp_sock_ipv6_mapped_specific = {
1935 .md5_lookup = tcp_v4_md5_lookup,
1936 .calc_md5_hash = tcp_v4_md5_hash_skb,
1937 .md5_add = tcp_v6_md5_add_func,
1938 .md5_parse = tcp_v6_parse_md5_keys,
1939 };
1940 #endif
1941
1942 /* NOTE: A lot of things set to zero explicitly by call to
1943 * sk_alloc() so need not be done here.
1944 */
1945 static int tcp_v6_init_sock(struct sock *sk)
1946 {
1947 struct inet_connection_sock *icsk = inet_csk(sk);
1948 struct tcp_sock *tp = tcp_sk(sk);
1949
1950 skb_queue_head_init(&tp->out_of_order_queue);
1951 tcp_init_xmit_timers(sk);
1952 tcp_prequeue_init(tp);
1953
1954 icsk->icsk_rto = TCP_TIMEOUT_INIT;
1955 tp->mdev = TCP_TIMEOUT_INIT;
1956
1957 /* So many TCP implementations out there (incorrectly) count the
1958 * initial SYN frame in their delayed-ACK and congestion control
1959 * algorithms that we must have the following bandaid to talk
1960 * efficiently to them. -DaveM
1961 */
1962 tp->snd_cwnd = 2;
1963
1964 /* See draft-stevens-tcpca-spec-01 for discussion of the
1965 * initialization of these values.
1966 */
1967 tp->snd_ssthresh = TCP_INFINITE_SSTHRESH;
1968 tp->snd_cwnd_clamp = ~0;
1969 tp->mss_cache = TCP_MSS_DEFAULT;
1970
1971 tp->reordering = sysctl_tcp_reordering;
1972
1973 sk->sk_state = TCP_CLOSE;
1974
1975 icsk->icsk_af_ops = &ipv6_specific;
1976 icsk->icsk_ca_ops = &tcp_init_congestion_ops;
1977 icsk->icsk_sync_mss = tcp_sync_mss;
1978 sk->sk_write_space = sk_stream_write_space;
1979 sock_set_flag(sk, SOCK_USE_WRITE_QUEUE);
1980
1981 #ifdef CONFIG_TCP_MD5SIG
1982 tp->af_specific = &tcp_sock_ipv6_specific;
1983 #endif
1984
1985 /* TCP Cookie Transactions */
1986 if (sysctl_tcp_cookie_size > 0) {
1987 /* Default, cookies without s_data_payload. */
1988 tp->cookie_values =
1989 kzalloc(sizeof(*tp->cookie_values),
1990 sk->sk_allocation);
1991 if (tp->cookie_values != NULL)
1992 kref_init(&tp->cookie_values->kref);
1993 }
1994 /* Presumed zeroed, in order of appearance:
1995 * cookie_in_always, cookie_out_never,
1996 * s_data_constant, s_data_in, s_data_out
1997 */
1998 sk->sk_sndbuf = sysctl_tcp_wmem[1];
1999 sk->sk_rcvbuf = sysctl_tcp_rmem[1];
2000
2001 local_bh_disable();
2002 sock_update_memcg(sk);
2003 sk_sockets_allocated_inc(sk);
2004 local_bh_enable();
2005
2006 return 0;
2007 }
2008
2009 static void tcp_v6_destroy_sock(struct sock *sk)
2010 {
2011 #ifdef CONFIG_TCP_MD5SIG
2012 /* Clean up the MD5 key list */
2013 if (tcp_sk(sk)->md5sig_info)
2014 tcp_v6_clear_md5_list(sk);
2015 #endif
2016 tcp_v4_destroy_sock(sk);
2017 inet6_destroy_sock(sk);
2018 }
2019
2020 #ifdef CONFIG_PROC_FS
2021 /* Proc filesystem TCPv6 sock list dumping. */
2022 static void get_openreq6(struct seq_file *seq,
2023 const struct sock *sk, struct request_sock *req, int i, int uid)
2024 {
2025 int ttd = req->expires - jiffies;
2026 const struct in6_addr *src = &inet6_rsk(req)->loc_addr;
2027 const struct in6_addr *dest = &inet6_rsk(req)->rmt_addr;
2028
2029 if (ttd < 0)
2030 ttd = 0;
2031
2032 seq_printf(seq,
2033 "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
2034 "%02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %pK\n",
2035 i,
2036 src->s6_addr32[0], src->s6_addr32[1],
2037 src->s6_addr32[2], src->s6_addr32[3],
2038 ntohs(inet_rsk(req)->loc_port),
2039 dest->s6_addr32[0], dest->s6_addr32[1],
2040 dest->s6_addr32[2], dest->s6_addr32[3],
2041 ntohs(inet_rsk(req)->rmt_port),
2042 TCP_SYN_RECV,
2043 0,0, /* could print option size, but that is af dependent. */
2044 1, /* timers active (only the expire timer) */
2045 jiffies_to_clock_t(ttd),
2046 req->retrans,
2047 uid,
2048 0, /* non standard timer */
2049 0, /* open_requests have no inode */
2050 0, req);
2051 }
2052
2053 static void get_tcp6_sock(struct seq_file *seq, struct sock *sp, int i)
2054 {
2055 const struct in6_addr *dest, *src;
2056 __u16 destp, srcp;
2057 int timer_active;
2058 unsigned long timer_expires;
2059 const struct inet_sock *inet = inet_sk(sp);
2060 const struct tcp_sock *tp = tcp_sk(sp);
2061 const struct inet_connection_sock *icsk = inet_csk(sp);
2062 const struct ipv6_pinfo *np = inet6_sk(sp);
2063
2064 dest = &np->daddr;
2065 src = &np->rcv_saddr;
2066 destp = ntohs(inet->inet_dport);
2067 srcp = ntohs(inet->inet_sport);
2068
2069 if (icsk->icsk_pending == ICSK_TIME_RETRANS) {
2070 timer_active = 1;
2071 timer_expires = icsk->icsk_timeout;
2072 } else if (icsk->icsk_pending == ICSK_TIME_PROBE0) {
2073 timer_active = 4;
2074 timer_expires = icsk->icsk_timeout;
2075 } else if (timer_pending(&sp->sk_timer)) {
2076 timer_active = 2;
2077 timer_expires = sp->sk_timer.expires;
2078 } else {
2079 timer_active = 0;
2080 timer_expires = jiffies;
2081 }
2082
2083 seq_printf(seq,
2084 "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
2085 "%02X %08X:%08X %02X:%08lX %08X %5d %8d %lu %d %pK %lu %lu %u %u %d\n",
2086 i,
2087 src->s6_addr32[0], src->s6_addr32[1],
2088 src->s6_addr32[2], src->s6_addr32[3], srcp,
2089 dest->s6_addr32[0], dest->s6_addr32[1],
2090 dest->s6_addr32[2], dest->s6_addr32[3], destp,
2091 sp->sk_state,
2092 tp->write_seq-tp->snd_una,
2093 (sp->sk_state == TCP_LISTEN) ? sp->sk_ack_backlog : (tp->rcv_nxt - tp->copied_seq),
2094 timer_active,
2095 jiffies_to_clock_t(timer_expires - jiffies),
2096 icsk->icsk_retransmits,
2097 sock_i_uid(sp),
2098 icsk->icsk_probes_out,
2099 sock_i_ino(sp),
2100 atomic_read(&sp->sk_refcnt), sp,
2101 jiffies_to_clock_t(icsk->icsk_rto),
2102 jiffies_to_clock_t(icsk->icsk_ack.ato),
2103 (icsk->icsk_ack.quick << 1 ) | icsk->icsk_ack.pingpong,
2104 tp->snd_cwnd,
2105 tcp_in_initial_slowstart(tp) ? -1 : tp->snd_ssthresh
2106 );
2107 }
2108
2109 static void get_timewait6_sock(struct seq_file *seq,
2110 struct inet_timewait_sock *tw, int i)
2111 {
2112 const struct in6_addr *dest, *src;
2113 __u16 destp, srcp;
2114 const struct inet6_timewait_sock *tw6 = inet6_twsk((struct sock *)tw);
2115 int ttd = tw->tw_ttd - jiffies;
2116
2117 if (ttd < 0)
2118 ttd = 0;
2119
2120 dest = &tw6->tw_v6_daddr;
2121 src = &tw6->tw_v6_rcv_saddr;
2122 destp = ntohs(tw->tw_dport);
2123 srcp = ntohs(tw->tw_sport);
2124
2125 seq_printf(seq,
2126 "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
2127 "%02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %pK\n",
2128 i,
2129 src->s6_addr32[0], src->s6_addr32[1],
2130 src->s6_addr32[2], src->s6_addr32[3], srcp,
2131 dest->s6_addr32[0], dest->s6_addr32[1],
2132 dest->s6_addr32[2], dest->s6_addr32[3], destp,
2133 tw->tw_substate, 0, 0,
2134 3, jiffies_to_clock_t(ttd), 0, 0, 0, 0,
2135 atomic_read(&tw->tw_refcnt), tw);
2136 }
2137
2138 static int tcp6_seq_show(struct seq_file *seq, void *v)
2139 {
2140 struct tcp_iter_state *st;
2141
2142 if (v == SEQ_START_TOKEN) {
2143 seq_puts(seq,
2144 " sl "
2145 "local_address "
2146 "remote_address "
2147 "st tx_queue rx_queue tr tm->when retrnsmt"
2148 " uid timeout inode\n");
2149 goto out;
2150 }
2151 st = seq->private;
2152
2153 switch (st->state) {
2154 case TCP_SEQ_STATE_LISTENING:
2155 case TCP_SEQ_STATE_ESTABLISHED:
2156 get_tcp6_sock(seq, v, st->num);
2157 break;
2158 case TCP_SEQ_STATE_OPENREQ:
2159 get_openreq6(seq, st->syn_wait_sk, v, st->num, st->uid);
2160 break;
2161 case TCP_SEQ_STATE_TIME_WAIT:
2162 get_timewait6_sock(seq, v, st->num);
2163 break;
2164 }
2165 out:
2166 return 0;
2167 }
2168
2169 static const struct file_operations tcp6_afinfo_seq_fops = {
2170 .owner = THIS_MODULE,
2171 .open = tcp_seq_open,
2172 .read = seq_read,
2173 .llseek = seq_lseek,
2174 .release = seq_release_net
2175 };
2176
2177 static struct tcp_seq_afinfo tcp6_seq_afinfo = {
2178 .name = "tcp6",
2179 .family = AF_INET6,
2180 .seq_fops = &tcp6_afinfo_seq_fops,
2181 .seq_ops = {
2182 .show = tcp6_seq_show,
2183 },
2184 };
2185
2186 int __net_init tcp6_proc_init(struct net *net)
2187 {
2188 return tcp_proc_register(net, &tcp6_seq_afinfo);
2189 }
2190
2191 void tcp6_proc_exit(struct net *net)
2192 {
2193 tcp_proc_unregister(net, &tcp6_seq_afinfo);
2194 }
2195 #endif
2196
2197 struct proto tcpv6_prot = {
2198 .name = "TCPv6",
2199 .owner = THIS_MODULE,
2200 .close = tcp_close,
2201 .connect = tcp_v6_connect,
2202 .disconnect = tcp_disconnect,
2203 .accept = inet_csk_accept,
2204 .ioctl = tcp_ioctl,
2205 .init = tcp_v6_init_sock,
2206 .destroy = tcp_v6_destroy_sock,
2207 .shutdown = tcp_shutdown,
2208 .setsockopt = tcp_setsockopt,
2209 .getsockopt = tcp_getsockopt,
2210 .recvmsg = tcp_recvmsg,
2211 .sendmsg = tcp_sendmsg,
2212 .sendpage = tcp_sendpage,
2213 .backlog_rcv = tcp_v6_do_rcv,
2214 .hash = tcp_v6_hash,
2215 .unhash = inet_unhash,
2216 .get_port = inet_csk_get_port,
2217 .enter_memory_pressure = tcp_enter_memory_pressure,
2218 .sockets_allocated = &tcp_sockets_allocated,
2219 .memory_allocated = &tcp_memory_allocated,
2220 .memory_pressure = &tcp_memory_pressure,
2221 .orphan_count = &tcp_orphan_count,
2222 .sysctl_wmem = sysctl_tcp_wmem,
2223 .sysctl_rmem = sysctl_tcp_rmem,
2224 .max_header = MAX_TCP_HEADER,
2225 .obj_size = sizeof(struct tcp6_sock),
2226 .slab_flags = SLAB_DESTROY_BY_RCU,
2227 .twsk_prot = &tcp6_timewait_sock_ops,
2228 .rsk_prot = &tcp6_request_sock_ops,
2229 .h.hashinfo = &tcp_hashinfo,
2230 .no_autobind = true,
2231 #ifdef CONFIG_COMPAT
2232 .compat_setsockopt = compat_tcp_setsockopt,
2233 .compat_getsockopt = compat_tcp_getsockopt,
2234 #endif
2235 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_KMEM
2236 .proto_cgroup = tcp_proto_cgroup,
2237 #endif
2238 };
2239
2240 static const struct inet6_protocol tcpv6_protocol = {
2241 .handler = tcp_v6_rcv,
2242 .err_handler = tcp_v6_err,
2243 .gso_send_check = tcp_v6_gso_send_check,
2244 .gso_segment = tcp_tso_segment,
2245 .gro_receive = tcp6_gro_receive,
2246 .gro_complete = tcp6_gro_complete,
2247 .flags = INET6_PROTO_NOPOLICY|INET6_PROTO_FINAL,
2248 };
2249
2250 static struct inet_protosw tcpv6_protosw = {
2251 .type = SOCK_STREAM,
2252 .protocol = IPPROTO_TCP,
2253 .prot = &tcpv6_prot,
2254 .ops = &inet6_stream_ops,
2255 .no_check = 0,
2256 .flags = INET_PROTOSW_PERMANENT |
2257 INET_PROTOSW_ICSK,
2258 };
2259
2260 static int __net_init tcpv6_net_init(struct net *net)
2261 {
2262 return inet_ctl_sock_create(&net->ipv6.tcp_sk, PF_INET6,
2263 SOCK_RAW, IPPROTO_TCP, net);
2264 }
2265
2266 static void __net_exit tcpv6_net_exit(struct net *net)
2267 {
2268 inet_ctl_sock_destroy(net->ipv6.tcp_sk);
2269 }
2270
2271 static void __net_exit tcpv6_net_exit_batch(struct list_head *net_exit_list)
2272 {
2273 inet_twsk_purge(&tcp_hashinfo, &tcp_death_row, AF_INET6);
2274 }
2275
2276 static struct pernet_operations tcpv6_net_ops = {
2277 .init = tcpv6_net_init,
2278 .exit = tcpv6_net_exit,
2279 .exit_batch = tcpv6_net_exit_batch,
2280 };
2281
2282 int __init tcpv6_init(void)
2283 {
2284 int ret;
2285
2286 ret = inet6_add_protocol(&tcpv6_protocol, IPPROTO_TCP);
2287 if (ret)
2288 goto out;
2289
2290 /* register inet6 protocol */
2291 ret = inet6_register_protosw(&tcpv6_protosw);
2292 if (ret)
2293 goto out_tcpv6_protocol;
2294
2295 ret = register_pernet_subsys(&tcpv6_net_ops);
2296 if (ret)
2297 goto out_tcpv6_protosw;
2298 out:
2299 return ret;
2300
2301 out_tcpv6_protocol:
2302 inet6_del_protocol(&tcpv6_protocol, IPPROTO_TCP);
2303 out_tcpv6_protosw:
2304 inet6_unregister_protosw(&tcpv6_protosw);
2305 goto out;
2306 }
2307
2308 void tcpv6_exit(void)
2309 {
2310 unregister_pernet_subsys(&tcpv6_net_ops);
2311 inet6_unregister_protosw(&tcpv6_protosw);
2312 inet6_del_protocol(&tcpv6_protocol, IPPROTO_TCP);
2313 }