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
tcp_v6_md5_do_lookup(struct sock * sk,const struct in6_addr * addr)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
tcp_v6_hash(struct sock * sk)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
tcp_v6_check(int len,const struct in6_addr * saddr,const struct in6_addr * daddr,__wsum base)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
tcp_v6_init_sequence(const struct sk_buff * skb)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
tcp_v6_connect(struct sock * sk,struct sockaddr * uaddr,int addr_len)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
tcp_v6_err(struct sk_buff * skb,struct inet6_skb_parm * opt,u8 type,u8 code,int offset,__be32 info)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
tcp_v6_send_synack(struct sock * sk,struct request_sock * req,struct request_values * rvp)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
tcp_v6_rtx_synack(struct sock * sk,struct request_sock * req,struct request_values * rvp)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
tcp_v6_reqsk_destructor(struct request_sock * req)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
tcp_v6_md5_do_lookup(struct sock * sk,const struct in6_addr * addr)540 static struct tcp_md5sig_key *tcp_v6_md5_do_lookup(struct sock *sk,
541 const struct in6_addr *addr)
542 {
543 return tcp_md5_do_lookup(sk, (union tcp_md5_addr *)addr, AF_INET6);
544 }
545
tcp_v6_md5_lookup(struct sock * sk,struct sock * addr_sk)546 static struct tcp_md5sig_key *tcp_v6_md5_lookup(struct sock *sk,
547 struct sock *addr_sk)
548 {
549 return tcp_v6_md5_do_lookup(sk, &inet6_sk(addr_sk)->daddr);
550 }
551
tcp_v6_reqsk_md5_lookup(struct sock * sk,struct request_sock * req)552 static struct tcp_md5sig_key *tcp_v6_reqsk_md5_lookup(struct sock *sk,
553 struct request_sock *req)
554 {
555 return tcp_v6_md5_do_lookup(sk, &inet6_rsk(req)->rmt_addr);
556 }
557
tcp_v6_parse_md5_keys(struct sock * sk,char __user * optval,int optlen)558 static int tcp_v6_parse_md5_keys (struct sock *sk, char __user *optval,
559 int optlen)
560 {
561 struct tcp_md5sig cmd;
562 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)&cmd.tcpm_addr;
563
564 if (optlen < sizeof(cmd))
565 return -EINVAL;
566
567 if (copy_from_user(&cmd, optval, sizeof(cmd)))
568 return -EFAULT;
569
570 if (sin6->sin6_family != AF_INET6)
571 return -EINVAL;
572
573 if (!cmd.tcpm_keylen) {
574 if (ipv6_addr_v4mapped(&sin6->sin6_addr))
575 return tcp_md5_do_del(sk, (union tcp_md5_addr *)&sin6->sin6_addr.s6_addr32[3],
576 AF_INET);
577 return tcp_md5_do_del(sk, (union tcp_md5_addr *)&sin6->sin6_addr,
578 AF_INET6);
579 }
580
581 if (cmd.tcpm_keylen > TCP_MD5SIG_MAXKEYLEN)
582 return -EINVAL;
583
584 if (ipv6_addr_v4mapped(&sin6->sin6_addr))
585 return tcp_md5_do_add(sk, (union tcp_md5_addr *)&sin6->sin6_addr.s6_addr32[3],
586 AF_INET, cmd.tcpm_key, cmd.tcpm_keylen, GFP_KERNEL);
587
588 return tcp_md5_do_add(sk, (union tcp_md5_addr *)&sin6->sin6_addr,
589 AF_INET6, cmd.tcpm_key, cmd.tcpm_keylen, GFP_KERNEL);
590 }
591
tcp_v6_md5_hash_pseudoheader(struct tcp_md5sig_pool * hp,const struct in6_addr * daddr,const struct in6_addr * saddr,int nbytes)592 static int tcp_v6_md5_hash_pseudoheader(struct tcp_md5sig_pool *hp,
593 const struct in6_addr *daddr,
594 const struct in6_addr *saddr, int nbytes)
595 {
596 struct tcp6_pseudohdr *bp;
597 struct scatterlist sg;
598
599 bp = &hp->md5_blk.ip6;
600 /* 1. TCP pseudo-header (RFC2460) */
601 bp->saddr = *saddr;
602 bp->daddr = *daddr;
603 bp->protocol = cpu_to_be32(IPPROTO_TCP);
604 bp->len = cpu_to_be32(nbytes);
605
606 sg_init_one(&sg, bp, sizeof(*bp));
607 return crypto_hash_update(&hp->md5_desc, &sg, sizeof(*bp));
608 }
609
tcp_v6_md5_hash_hdr(char * md5_hash,struct tcp_md5sig_key * key,const struct in6_addr * daddr,struct in6_addr * saddr,const struct tcphdr * th)610 static int tcp_v6_md5_hash_hdr(char *md5_hash, struct tcp_md5sig_key *key,
611 const struct in6_addr *daddr, struct in6_addr *saddr,
612 const struct tcphdr *th)
613 {
614 struct tcp_md5sig_pool *hp;
615 struct hash_desc *desc;
616
617 hp = tcp_get_md5sig_pool();
618 if (!hp)
619 goto clear_hash_noput;
620 desc = &hp->md5_desc;
621
622 if (crypto_hash_init(desc))
623 goto clear_hash;
624 if (tcp_v6_md5_hash_pseudoheader(hp, daddr, saddr, th->doff << 2))
625 goto clear_hash;
626 if (tcp_md5_hash_header(hp, th))
627 goto clear_hash;
628 if (tcp_md5_hash_key(hp, key))
629 goto clear_hash;
630 if (crypto_hash_final(desc, md5_hash))
631 goto clear_hash;
632
633 tcp_put_md5sig_pool();
634 return 0;
635
636 clear_hash:
637 tcp_put_md5sig_pool();
638 clear_hash_noput:
639 memset(md5_hash, 0, 16);
640 return 1;
641 }
642
tcp_v6_md5_hash_skb(char * md5_hash,struct tcp_md5sig_key * key,const struct sock * sk,const struct request_sock * req,const struct sk_buff * skb)643 static int tcp_v6_md5_hash_skb(char *md5_hash, struct tcp_md5sig_key *key,
644 const struct sock *sk,
645 const struct request_sock *req,
646 const struct sk_buff *skb)
647 {
648 const struct in6_addr *saddr, *daddr;
649 struct tcp_md5sig_pool *hp;
650 struct hash_desc *desc;
651 const struct tcphdr *th = tcp_hdr(skb);
652
653 if (sk) {
654 saddr = &inet6_sk(sk)->saddr;
655 daddr = &inet6_sk(sk)->daddr;
656 } else if (req) {
657 saddr = &inet6_rsk(req)->loc_addr;
658 daddr = &inet6_rsk(req)->rmt_addr;
659 } else {
660 const struct ipv6hdr *ip6h = ipv6_hdr(skb);
661 saddr = &ip6h->saddr;
662 daddr = &ip6h->daddr;
663 }
664
665 hp = tcp_get_md5sig_pool();
666 if (!hp)
667 goto clear_hash_noput;
668 desc = &hp->md5_desc;
669
670 if (crypto_hash_init(desc))
671 goto clear_hash;
672
673 if (tcp_v6_md5_hash_pseudoheader(hp, daddr, saddr, skb->len))
674 goto clear_hash;
675 if (tcp_md5_hash_header(hp, th))
676 goto clear_hash;
677 if (tcp_md5_hash_skb_data(hp, skb, th->doff << 2))
678 goto clear_hash;
679 if (tcp_md5_hash_key(hp, key))
680 goto clear_hash;
681 if (crypto_hash_final(desc, md5_hash))
682 goto clear_hash;
683
684 tcp_put_md5sig_pool();
685 return 0;
686
687 clear_hash:
688 tcp_put_md5sig_pool();
689 clear_hash_noput:
690 memset(md5_hash, 0, 16);
691 return 1;
692 }
693
tcp_v6_inbound_md5_hash(struct sock * sk,const struct sk_buff * skb)694 static int tcp_v6_inbound_md5_hash(struct sock *sk, const struct sk_buff *skb)
695 {
696 const __u8 *hash_location = NULL;
697 struct tcp_md5sig_key *hash_expected;
698 const struct ipv6hdr *ip6h = ipv6_hdr(skb);
699 const struct tcphdr *th = tcp_hdr(skb);
700 int genhash;
701 u8 newhash[16];
702
703 hash_expected = tcp_v6_md5_do_lookup(sk, &ip6h->saddr);
704 hash_location = tcp_parse_md5sig_option(th);
705
706 /* We've parsed the options - do we have a hash? */
707 if (!hash_expected && !hash_location)
708 return 0;
709
710 if (hash_expected && !hash_location) {
711 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPMD5NOTFOUND);
712 return 1;
713 }
714
715 if (!hash_expected && hash_location) {
716 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPMD5UNEXPECTED);
717 return 1;
718 }
719
720 /* check the signature */
721 genhash = tcp_v6_md5_hash_skb(newhash,
722 hash_expected,
723 NULL, NULL, skb);
724
725 if (genhash || memcmp(hash_location, newhash, 16) != 0) {
726 if (net_ratelimit()) {
727 printk(KERN_INFO "MD5 Hash %s for [%pI6c]:%u->[%pI6c]:%u\n",
728 genhash ? "failed" : "mismatch",
729 &ip6h->saddr, ntohs(th->source),
730 &ip6h->daddr, ntohs(th->dest));
731 }
732 return 1;
733 }
734 return 0;
735 }
736 #endif
737
738 struct request_sock_ops tcp6_request_sock_ops __read_mostly = {
739 .family = AF_INET6,
740 .obj_size = sizeof(struct tcp6_request_sock),
741 .rtx_syn_ack = tcp_v6_rtx_synack,
742 .send_ack = tcp_v6_reqsk_send_ack,
743 .destructor = tcp_v6_reqsk_destructor,
744 .send_reset = tcp_v6_send_reset,
745 .syn_ack_timeout = tcp_syn_ack_timeout,
746 };
747
748 #ifdef CONFIG_TCP_MD5SIG
749 static const struct tcp_request_sock_ops tcp_request_sock_ipv6_ops = {
750 .md5_lookup = tcp_v6_reqsk_md5_lookup,
751 .calc_md5_hash = tcp_v6_md5_hash_skb,
752 };
753 #endif
754
__tcp_v6_send_check(struct sk_buff * skb,const struct in6_addr * saddr,const struct in6_addr * daddr)755 static void __tcp_v6_send_check(struct sk_buff *skb,
756 const struct in6_addr *saddr, const struct in6_addr *daddr)
757 {
758 struct tcphdr *th = tcp_hdr(skb);
759
760 if (skb->ip_summed == CHECKSUM_PARTIAL) {
761 th->check = ~tcp_v6_check(skb->len, saddr, daddr, 0);
762 skb->csum_start = skb_transport_header(skb) - skb->head;
763 skb->csum_offset = offsetof(struct tcphdr, check);
764 } else {
765 th->check = tcp_v6_check(skb->len, saddr, daddr,
766 csum_partial(th, th->doff << 2,
767 skb->csum));
768 }
769 }
770
tcp_v6_send_check(struct sock * sk,struct sk_buff * skb)771 static void tcp_v6_send_check(struct sock *sk, struct sk_buff *skb)
772 {
773 struct ipv6_pinfo *np = inet6_sk(sk);
774
775 __tcp_v6_send_check(skb, &np->saddr, &np->daddr);
776 }
777
tcp_v6_gso_send_check(struct sk_buff * skb)778 static int tcp_v6_gso_send_check(struct sk_buff *skb)
779 {
780 const struct ipv6hdr *ipv6h;
781 struct tcphdr *th;
782
783 if (!pskb_may_pull(skb, sizeof(*th)))
784 return -EINVAL;
785
786 ipv6h = ipv6_hdr(skb);
787 th = tcp_hdr(skb);
788
789 th->check = 0;
790 skb->ip_summed = CHECKSUM_PARTIAL;
791 __tcp_v6_send_check(skb, &ipv6h->saddr, &ipv6h->daddr);
792 return 0;
793 }
794
tcp6_gro_receive(struct sk_buff ** head,struct sk_buff * skb)795 static struct sk_buff **tcp6_gro_receive(struct sk_buff **head,
796 struct sk_buff *skb)
797 {
798 const struct ipv6hdr *iph = skb_gro_network_header(skb);
799
800 switch (skb->ip_summed) {
801 case CHECKSUM_COMPLETE:
802 if (!tcp_v6_check(skb_gro_len(skb), &iph->saddr, &iph->daddr,
803 skb->csum)) {
804 skb->ip_summed = CHECKSUM_UNNECESSARY;
805 break;
806 }
807
808 /* fall through */
809 case CHECKSUM_NONE:
810 NAPI_GRO_CB(skb)->flush = 1;
811 return NULL;
812 }
813
814 return tcp_gro_receive(head, skb);
815 }
816
tcp6_gro_complete(struct sk_buff * skb)817 static int tcp6_gro_complete(struct sk_buff *skb)
818 {
819 const struct ipv6hdr *iph = ipv6_hdr(skb);
820 struct tcphdr *th = tcp_hdr(skb);
821
822 th->check = ~tcp_v6_check(skb->len - skb_transport_offset(skb),
823 &iph->saddr, &iph->daddr, 0);
824 skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6;
825
826 return tcp_gro_complete(skb);
827 }
828
tcp_v6_send_response(struct sk_buff * skb,u32 seq,u32 ack,u32 win,u32 ts,struct tcp_md5sig_key * key,int rst,u8 tclass)829 static void tcp_v6_send_response(struct sk_buff *skb, u32 seq, u32 ack, u32 win,
830 u32 ts, struct tcp_md5sig_key *key, int rst, u8 tclass)
831 {
832 const struct tcphdr *th = tcp_hdr(skb);
833 struct tcphdr *t1;
834 struct sk_buff *buff;
835 struct flowi6 fl6;
836 struct net *net = dev_net(skb_dst(skb)->dev);
837 struct sock *ctl_sk = net->ipv6.tcp_sk;
838 unsigned int tot_len = sizeof(struct tcphdr);
839 struct dst_entry *dst;
840 __be32 *topt;
841
842 if (ts)
843 tot_len += TCPOLEN_TSTAMP_ALIGNED;
844 #ifdef CONFIG_TCP_MD5SIG
845 if (key)
846 tot_len += TCPOLEN_MD5SIG_ALIGNED;
847 #endif
848
849 buff = alloc_skb(MAX_HEADER + sizeof(struct ipv6hdr) + tot_len,
850 GFP_ATOMIC);
851 if (buff == NULL)
852 return;
853
854 skb_reserve(buff, MAX_HEADER + sizeof(struct ipv6hdr) + tot_len);
855
856 t1 = (struct tcphdr *) skb_push(buff, tot_len);
857 skb_reset_transport_header(buff);
858
859 /* Swap the send and the receive. */
860 memset(t1, 0, sizeof(*t1));
861 t1->dest = th->source;
862 t1->source = th->dest;
863 t1->doff = tot_len / 4;
864 t1->seq = htonl(seq);
865 t1->ack_seq = htonl(ack);
866 t1->ack = !rst || !th->ack;
867 t1->rst = rst;
868 t1->window = htons(win);
869
870 topt = (__be32 *)(t1 + 1);
871
872 if (ts) {
873 *topt++ = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
874 (TCPOPT_TIMESTAMP << 8) | TCPOLEN_TIMESTAMP);
875 *topt++ = htonl(tcp_time_stamp);
876 *topt++ = htonl(ts);
877 }
878
879 #ifdef CONFIG_TCP_MD5SIG
880 if (key) {
881 *topt++ = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
882 (TCPOPT_MD5SIG << 8) | TCPOLEN_MD5SIG);
883 tcp_v6_md5_hash_hdr((__u8 *)topt, key,
884 &ipv6_hdr(skb)->saddr,
885 &ipv6_hdr(skb)->daddr, t1);
886 }
887 #endif
888
889 memset(&fl6, 0, sizeof(fl6));
890 fl6.daddr = ipv6_hdr(skb)->saddr;
891 fl6.saddr = ipv6_hdr(skb)->daddr;
892
893 buff->ip_summed = CHECKSUM_PARTIAL;
894 buff->csum = 0;
895
896 __tcp_v6_send_check(buff, &fl6.saddr, &fl6.daddr);
897
898 fl6.flowi6_proto = IPPROTO_TCP;
899 if (ipv6_addr_type(&fl6.daddr) & IPV6_ADDR_LINKLOCAL)
900 fl6.flowi6_oif = inet6_iif(skb);
901 fl6.fl6_dport = t1->dest;
902 fl6.fl6_sport = t1->source;
903 security_skb_classify_flow(skb, flowi6_to_flowi(&fl6));
904
905 /* Pass a socket to ip6_dst_lookup either it is for RST
906 * Underlying function will use this to retrieve the network
907 * namespace
908 */
909 dst = ip6_dst_lookup_flow(ctl_sk, &fl6, NULL, false);
910 if (!IS_ERR(dst)) {
911 skb_dst_set(buff, dst);
912 ip6_xmit(ctl_sk, buff, &fl6, NULL, tclass);
913 TCP_INC_STATS_BH(net, TCP_MIB_OUTSEGS);
914 if (rst)
915 TCP_INC_STATS_BH(net, TCP_MIB_OUTRSTS);
916 return;
917 }
918
919 kfree_skb(buff);
920 }
921
tcp_v6_send_reset(struct sock * sk,struct sk_buff * skb)922 static void tcp_v6_send_reset(struct sock *sk, struct sk_buff *skb)
923 {
924 const struct tcphdr *th = tcp_hdr(skb);
925 u32 seq = 0, ack_seq = 0;
926 struct tcp_md5sig_key *key = NULL;
927 #ifdef CONFIG_TCP_MD5SIG
928 const __u8 *hash_location = NULL;
929 struct ipv6hdr *ipv6h = ipv6_hdr(skb);
930 unsigned char newhash[16];
931 int genhash;
932 struct sock *sk1 = NULL;
933 #endif
934
935 if (th->rst)
936 return;
937
938 if (!ipv6_unicast_destination(skb))
939 return;
940
941 #ifdef CONFIG_TCP_MD5SIG
942 hash_location = tcp_parse_md5sig_option(th);
943 if (!sk && hash_location) {
944 /*
945 * active side is lost. Try to find listening socket through
946 * source port, and then find md5 key through listening socket.
947 * we are not loose security here:
948 * Incoming packet is checked with md5 hash with finding key,
949 * no RST generated if md5 hash doesn't match.
950 */
951 sk1 = inet6_lookup_listener(dev_net(skb_dst(skb)->dev),
952 &tcp_hashinfo, &ipv6h->daddr,
953 ntohs(th->source), inet6_iif(skb));
954 if (!sk1)
955 return;
956
957 rcu_read_lock();
958 key = tcp_v6_md5_do_lookup(sk1, &ipv6h->saddr);
959 if (!key)
960 goto release_sk1;
961
962 genhash = tcp_v6_md5_hash_skb(newhash, key, NULL, NULL, skb);
963 if (genhash || memcmp(hash_location, newhash, 16) != 0)
964 goto release_sk1;
965 } else {
966 key = sk ? tcp_v6_md5_do_lookup(sk, &ipv6h->saddr) : NULL;
967 }
968 #endif
969
970 if (th->ack)
971 seq = ntohl(th->ack_seq);
972 else
973 ack_seq = ntohl(th->seq) + th->syn + th->fin + skb->len -
974 (th->doff << 2);
975
976 tcp_v6_send_response(skb, seq, ack_seq, 0, 0, key, 1, 0);
977
978 #ifdef CONFIG_TCP_MD5SIG
979 release_sk1:
980 if (sk1) {
981 rcu_read_unlock();
982 sock_put(sk1);
983 }
984 #endif
985 }
986
tcp_v6_send_ack(struct sk_buff * skb,u32 seq,u32 ack,u32 win,u32 ts,struct tcp_md5sig_key * key,u8 tclass)987 static void tcp_v6_send_ack(struct sk_buff *skb, u32 seq, u32 ack, u32 win, u32 ts,
988 struct tcp_md5sig_key *key, u8 tclass)
989 {
990 tcp_v6_send_response(skb, seq, ack, win, ts, key, 0, tclass);
991 }
992
tcp_v6_timewait_ack(struct sock * sk,struct sk_buff * skb)993 static void tcp_v6_timewait_ack(struct sock *sk, struct sk_buff *skb)
994 {
995 struct inet_timewait_sock *tw = inet_twsk(sk);
996 struct tcp_timewait_sock *tcptw = tcp_twsk(sk);
997
998 tcp_v6_send_ack(skb, tcptw->tw_snd_nxt, tcptw->tw_rcv_nxt,
999 tcptw->tw_rcv_wnd >> tw->tw_rcv_wscale,
1000 tcptw->tw_ts_recent, tcp_twsk_md5_key(tcptw),
1001 tw->tw_tclass);
1002
1003 inet_twsk_put(tw);
1004 }
1005
tcp_v6_reqsk_send_ack(struct sock * sk,struct sk_buff * skb,struct request_sock * req)1006 static void tcp_v6_reqsk_send_ack(struct sock *sk, struct sk_buff *skb,
1007 struct request_sock *req)
1008 {
1009 tcp_v6_send_ack(skb, tcp_rsk(req)->snt_isn + 1, tcp_rsk(req)->rcv_isn + 1, req->rcv_wnd, req->ts_recent,
1010 tcp_v6_md5_do_lookup(sk, &ipv6_hdr(skb)->daddr), 0);
1011 }
1012
1013
tcp_v6_hnd_req(struct sock * sk,struct sk_buff * skb)1014 static struct sock *tcp_v6_hnd_req(struct sock *sk,struct sk_buff *skb)
1015 {
1016 struct request_sock *req, **prev;
1017 const struct tcphdr *th = tcp_hdr(skb);
1018 struct sock *nsk;
1019
1020 /* Find possible connection requests. */
1021 req = inet6_csk_search_req(sk, &prev, th->source,
1022 &ipv6_hdr(skb)->saddr,
1023 &ipv6_hdr(skb)->daddr, inet6_iif(skb));
1024 if (req)
1025 return tcp_check_req(sk, skb, req, prev);
1026
1027 nsk = __inet6_lookup_established(sock_net(sk), &tcp_hashinfo,
1028 &ipv6_hdr(skb)->saddr, th->source,
1029 &ipv6_hdr(skb)->daddr, ntohs(th->dest), inet6_iif(skb));
1030
1031 if (nsk) {
1032 if (nsk->sk_state != TCP_TIME_WAIT) {
1033 bh_lock_sock(nsk);
1034 return nsk;
1035 }
1036 inet_twsk_put(inet_twsk(nsk));
1037 return NULL;
1038 }
1039
1040 #ifdef CONFIG_SYN_COOKIES
1041 if (!th->syn)
1042 sk = cookie_v6_check(sk, skb);
1043 #endif
1044 return sk;
1045 }
1046
1047 /* FIXME: this is substantially similar to the ipv4 code.
1048 * Can some kind of merge be done? -- erics
1049 */
tcp_v6_conn_request(struct sock * sk,struct sk_buff * skb)1050 static int tcp_v6_conn_request(struct sock *sk, struct sk_buff *skb)
1051 {
1052 struct tcp_extend_values tmp_ext;
1053 struct tcp_options_received tmp_opt;
1054 const u8 *hash_location;
1055 struct request_sock *req;
1056 struct inet6_request_sock *treq;
1057 struct ipv6_pinfo *np = inet6_sk(sk);
1058 struct tcp_sock *tp = tcp_sk(sk);
1059 __u32 isn = TCP_SKB_CB(skb)->when;
1060 struct dst_entry *dst = NULL;
1061 int want_cookie = 0;
1062
1063 if (skb->protocol == htons(ETH_P_IP))
1064 return tcp_v4_conn_request(sk, skb);
1065
1066 if (!ipv6_unicast_destination(skb))
1067 goto drop;
1068
1069 if (inet_csk_reqsk_queue_is_full(sk) && !isn) {
1070 want_cookie = tcp_syn_flood_action(sk, skb, "TCPv6");
1071 if (!want_cookie)
1072 goto drop;
1073 }
1074
1075 if (sk_acceptq_is_full(sk) && inet_csk_reqsk_queue_young(sk) > 1)
1076 goto drop;
1077
1078 req = inet6_reqsk_alloc(&tcp6_request_sock_ops);
1079 if (req == NULL)
1080 goto drop;
1081
1082 #ifdef CONFIG_TCP_MD5SIG
1083 tcp_rsk(req)->af_specific = &tcp_request_sock_ipv6_ops;
1084 #endif
1085
1086 tcp_clear_options(&tmp_opt);
1087 tmp_opt.mss_clamp = IPV6_MIN_MTU - sizeof(struct tcphdr) - sizeof(struct ipv6hdr);
1088 tmp_opt.user_mss = tp->rx_opt.user_mss;
1089 tcp_parse_options(skb, &tmp_opt, &hash_location, 0);
1090
1091 if (tmp_opt.cookie_plus > 0 &&
1092 tmp_opt.saw_tstamp &&
1093 !tp->rx_opt.cookie_out_never &&
1094 (sysctl_tcp_cookie_size > 0 ||
1095 (tp->cookie_values != NULL &&
1096 tp->cookie_values->cookie_desired > 0))) {
1097 u8 *c;
1098 u32 *d;
1099 u32 *mess = &tmp_ext.cookie_bakery[COOKIE_DIGEST_WORDS];
1100 int l = tmp_opt.cookie_plus - TCPOLEN_COOKIE_BASE;
1101
1102 if (tcp_cookie_generator(&tmp_ext.cookie_bakery[0]) != 0)
1103 goto drop_and_free;
1104
1105 /* Secret recipe starts with IP addresses */
1106 d = (__force u32 *)&ipv6_hdr(skb)->daddr.s6_addr32[0];
1107 *mess++ ^= *d++;
1108 *mess++ ^= *d++;
1109 *mess++ ^= *d++;
1110 *mess++ ^= *d++;
1111 d = (__force u32 *)&ipv6_hdr(skb)->saddr.s6_addr32[0];
1112 *mess++ ^= *d++;
1113 *mess++ ^= *d++;
1114 *mess++ ^= *d++;
1115 *mess++ ^= *d++;
1116
1117 /* plus variable length Initiator Cookie */
1118 c = (u8 *)mess;
1119 while (l-- > 0)
1120 *c++ ^= *hash_location++;
1121
1122 want_cookie = 0; /* not our kind of cookie */
1123 tmp_ext.cookie_out_never = 0; /* false */
1124 tmp_ext.cookie_plus = tmp_opt.cookie_plus;
1125 } else if (!tp->rx_opt.cookie_in_always) {
1126 /* redundant indications, but ensure initialization. */
1127 tmp_ext.cookie_out_never = 1; /* true */
1128 tmp_ext.cookie_plus = 0;
1129 } else {
1130 goto drop_and_free;
1131 }
1132 tmp_ext.cookie_in_always = tp->rx_opt.cookie_in_always;
1133
1134 if (want_cookie && !tmp_opt.saw_tstamp)
1135 tcp_clear_options(&tmp_opt);
1136
1137 tmp_opt.tstamp_ok = tmp_opt.saw_tstamp;
1138 tcp_openreq_init(req, &tmp_opt, skb);
1139
1140 treq = inet6_rsk(req);
1141 treq->rmt_addr = ipv6_hdr(skb)->saddr;
1142 treq->loc_addr = ipv6_hdr(skb)->daddr;
1143 if (!want_cookie || tmp_opt.tstamp_ok)
1144 TCP_ECN_create_request(req, tcp_hdr(skb));
1145
1146 treq->iif = sk->sk_bound_dev_if;
1147
1148 /* So that link locals have meaning */
1149 if (!sk->sk_bound_dev_if &&
1150 ipv6_addr_type(&treq->rmt_addr) & IPV6_ADDR_LINKLOCAL)
1151 treq->iif = inet6_iif(skb);
1152
1153 if (!isn) {
1154 struct inet_peer *peer = NULL;
1155
1156 if (ipv6_opt_accepted(sk, skb) ||
1157 np->rxopt.bits.rxinfo || np->rxopt.bits.rxoinfo ||
1158 np->rxopt.bits.rxhlim || np->rxopt.bits.rxohlim) {
1159 atomic_inc(&skb->users);
1160 treq->pktopts = skb;
1161 }
1162
1163 if (want_cookie) {
1164 isn = cookie_v6_init_sequence(sk, skb, &req->mss);
1165 req->cookie_ts = tmp_opt.tstamp_ok;
1166 goto have_isn;
1167 }
1168
1169 /* VJ's idea. We save last timestamp seen
1170 * from the destination in peer table, when entering
1171 * state TIME-WAIT, and check against it before
1172 * accepting new connection request.
1173 *
1174 * If "isn" is not zero, this request hit alive
1175 * timewait bucket, so that all the necessary checks
1176 * are made in the function processing timewait state.
1177 */
1178 if (tmp_opt.saw_tstamp &&
1179 tcp_death_row.sysctl_tw_recycle &&
1180 (dst = inet6_csk_route_req(sk, req)) != NULL &&
1181 (peer = rt6_get_peer((struct rt6_info *)dst)) != NULL &&
1182 ipv6_addr_equal((struct in6_addr *)peer->daddr.addr.a6,
1183 &treq->rmt_addr)) {
1184 inet_peer_refcheck(peer);
1185 if ((u32)get_seconds() - peer->tcp_ts_stamp < TCP_PAWS_MSL &&
1186 (s32)(peer->tcp_ts - req->ts_recent) >
1187 TCP_PAWS_WINDOW) {
1188 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_PAWSPASSIVEREJECTED);
1189 goto drop_and_release;
1190 }
1191 }
1192 /* Kill the following clause, if you dislike this way. */
1193 else if (!sysctl_tcp_syncookies &&
1194 (sysctl_max_syn_backlog - inet_csk_reqsk_queue_len(sk) <
1195 (sysctl_max_syn_backlog >> 2)) &&
1196 (!peer || !peer->tcp_ts_stamp) &&
1197 (!dst || !dst_metric(dst, RTAX_RTT))) {
1198 /* Without syncookies last quarter of
1199 * backlog is filled with destinations,
1200 * proven to be alive.
1201 * It means that we continue to communicate
1202 * to destinations, already remembered
1203 * to the moment of synflood.
1204 */
1205 LIMIT_NETDEBUG(KERN_DEBUG "TCP: drop open request from %pI6/%u\n",
1206 &treq->rmt_addr, ntohs(tcp_hdr(skb)->source));
1207 goto drop_and_release;
1208 }
1209
1210 isn = tcp_v6_init_sequence(skb);
1211 }
1212 have_isn:
1213 tcp_rsk(req)->snt_isn = isn;
1214 tcp_rsk(req)->snt_synack = tcp_time_stamp;
1215
1216 security_inet_conn_request(sk, skb, req);
1217
1218 if (tcp_v6_send_synack(sk, req,
1219 (struct request_values *)&tmp_ext) ||
1220 want_cookie)
1221 goto drop_and_free;
1222
1223 inet6_csk_reqsk_queue_hash_add(sk, req, TCP_TIMEOUT_INIT);
1224 return 0;
1225
1226 drop_and_release:
1227 dst_release(dst);
1228 drop_and_free:
1229 reqsk_free(req);
1230 drop:
1231 return 0; /* don't send reset */
1232 }
1233
tcp_v6_syn_recv_sock(struct sock * sk,struct sk_buff * skb,struct request_sock * req,struct dst_entry * dst)1234 static struct sock * tcp_v6_syn_recv_sock(struct sock *sk, struct sk_buff *skb,
1235 struct request_sock *req,
1236 struct dst_entry *dst)
1237 {
1238 struct inet6_request_sock *treq;
1239 struct ipv6_pinfo *newnp, *np = inet6_sk(sk);
1240 struct tcp6_sock *newtcp6sk;
1241 struct inet_sock *newinet;
1242 struct tcp_sock *newtp;
1243 struct sock *newsk;
1244 struct ipv6_txoptions *opt;
1245 #ifdef CONFIG_TCP_MD5SIG
1246 struct tcp_md5sig_key *key;
1247 #endif
1248
1249 if (skb->protocol == htons(ETH_P_IP)) {
1250 /*
1251 * v6 mapped
1252 */
1253
1254 newsk = tcp_v4_syn_recv_sock(sk, skb, req, dst);
1255
1256 if (newsk == NULL)
1257 return NULL;
1258
1259 newtcp6sk = (struct tcp6_sock *)newsk;
1260 inet_sk(newsk)->pinet6 = &newtcp6sk->inet6;
1261
1262 newinet = inet_sk(newsk);
1263 newnp = inet6_sk(newsk);
1264 newtp = tcp_sk(newsk);
1265
1266 memcpy(newnp, np, sizeof(struct ipv6_pinfo));
1267
1268 ipv6_addr_set_v4mapped(newinet->inet_daddr, &newnp->daddr);
1269
1270 ipv6_addr_set_v4mapped(newinet->inet_saddr, &newnp->saddr);
1271
1272 newnp->rcv_saddr = newnp->saddr;
1273
1274 inet_csk(newsk)->icsk_af_ops = &ipv6_mapped;
1275 newsk->sk_backlog_rcv = tcp_v4_do_rcv;
1276 #ifdef CONFIG_TCP_MD5SIG
1277 newtp->af_specific = &tcp_sock_ipv6_mapped_specific;
1278 #endif
1279
1280 newnp->ipv6_ac_list = NULL;
1281 newnp->ipv6_fl_list = NULL;
1282 newnp->pktoptions = NULL;
1283 newnp->opt = NULL;
1284 newnp->mcast_oif = inet6_iif(skb);
1285 newnp->mcast_hops = ipv6_hdr(skb)->hop_limit;
1286 newnp->rcv_tclass = ipv6_tclass(ipv6_hdr(skb));
1287
1288 /*
1289 * No need to charge this sock to the relevant IPv6 refcnt debug socks count
1290 * here, tcp_create_openreq_child now does this for us, see the comment in
1291 * that function for the gory details. -acme
1292 */
1293
1294 /* It is tricky place. Until this moment IPv4 tcp
1295 worked with IPv6 icsk.icsk_af_ops.
1296 Sync it now.
1297 */
1298 tcp_sync_mss(newsk, inet_csk(newsk)->icsk_pmtu_cookie);
1299
1300 return newsk;
1301 }
1302
1303 treq = inet6_rsk(req);
1304 opt = np->opt;
1305
1306 if (sk_acceptq_is_full(sk))
1307 goto out_overflow;
1308
1309 if (!dst) {
1310 dst = inet6_csk_route_req(sk, req);
1311 if (!dst)
1312 goto out;
1313 }
1314
1315 newsk = tcp_create_openreq_child(sk, req, skb);
1316 if (newsk == NULL)
1317 goto out_nonewsk;
1318
1319 /*
1320 * No need to charge this sock to the relevant IPv6 refcnt debug socks
1321 * count here, tcp_create_openreq_child now does this for us, see the
1322 * comment in that function for the gory details. -acme
1323 */
1324
1325 newsk->sk_gso_type = SKB_GSO_TCPV6;
1326 __ip6_dst_store(newsk, dst, NULL, NULL);
1327
1328 newtcp6sk = (struct tcp6_sock *)newsk;
1329 inet_sk(newsk)->pinet6 = &newtcp6sk->inet6;
1330
1331 newtp = tcp_sk(newsk);
1332 newinet = inet_sk(newsk);
1333 newnp = inet6_sk(newsk);
1334
1335 memcpy(newnp, np, sizeof(struct ipv6_pinfo));
1336
1337 newnp->daddr = treq->rmt_addr;
1338 newnp->saddr = treq->loc_addr;
1339 newnp->rcv_saddr = treq->loc_addr;
1340 newsk->sk_bound_dev_if = treq->iif;
1341
1342 /* Now IPv6 options...
1343
1344 First: no IPv4 options.
1345 */
1346 newinet->inet_opt = NULL;
1347 newnp->ipv6_ac_list = NULL;
1348 newnp->ipv6_fl_list = NULL;
1349
1350 /* Clone RX bits */
1351 newnp->rxopt.all = np->rxopt.all;
1352
1353 /* Clone pktoptions received with SYN */
1354 newnp->pktoptions = NULL;
1355 if (treq->pktopts != NULL) {
1356 newnp->pktoptions = skb_clone(treq->pktopts, GFP_ATOMIC);
1357 kfree_skb(treq->pktopts);
1358 treq->pktopts = NULL;
1359 if (newnp->pktoptions)
1360 skb_set_owner_r(newnp->pktoptions, newsk);
1361 }
1362 newnp->opt = NULL;
1363 newnp->mcast_oif = inet6_iif(skb);
1364 newnp->mcast_hops = ipv6_hdr(skb)->hop_limit;
1365 newnp->rcv_tclass = ipv6_tclass(ipv6_hdr(skb));
1366
1367 /* Clone native IPv6 options from listening socket (if any)
1368
1369 Yes, keeping reference count would be much more clever,
1370 but we make one more one thing there: reattach optmem
1371 to newsk.
1372 */
1373 if (opt) {
1374 newnp->opt = ipv6_dup_options(newsk, opt);
1375 if (opt != np->opt)
1376 sock_kfree_s(sk, opt, opt->tot_len);
1377 }
1378
1379 inet_csk(newsk)->icsk_ext_hdr_len = 0;
1380 if (newnp->opt)
1381 inet_csk(newsk)->icsk_ext_hdr_len = (newnp->opt->opt_nflen +
1382 newnp->opt->opt_flen);
1383
1384 tcp_mtup_init(newsk);
1385 tcp_sync_mss(newsk, dst_mtu(dst));
1386 newtp->advmss = dst_metric_advmss(dst);
1387 if (tcp_sk(sk)->rx_opt.user_mss &&
1388 tcp_sk(sk)->rx_opt.user_mss < newtp->advmss)
1389 newtp->advmss = tcp_sk(sk)->rx_opt.user_mss;
1390
1391 tcp_initialize_rcv_mss(newsk);
1392 if (tcp_rsk(req)->snt_synack)
1393 tcp_valid_rtt_meas(newsk,
1394 tcp_time_stamp - tcp_rsk(req)->snt_synack);
1395 newtp->total_retrans = req->retrans;
1396
1397 newinet->inet_daddr = newinet->inet_saddr = LOOPBACK4_IPV6;
1398 newinet->inet_rcv_saddr = LOOPBACK4_IPV6;
1399
1400 #ifdef CONFIG_TCP_MD5SIG
1401 /* Copy over the MD5 key from the original socket */
1402 if ((key = tcp_v6_md5_do_lookup(sk, &newnp->daddr)) != NULL) {
1403 /* We're using one, so create a matching key
1404 * on the newsk structure. If we fail to get
1405 * memory, then we end up not copying the key
1406 * across. Shucks.
1407 */
1408 tcp_md5_do_add(newsk, (union tcp_md5_addr *)&newnp->daddr,
1409 AF_INET6, key->key, key->keylen, GFP_ATOMIC);
1410 }
1411 #endif
1412
1413 if (__inet_inherit_port(sk, newsk) < 0) {
1414 inet_csk_prepare_forced_close(newsk);
1415 tcp_done(newsk);
1416 goto out;
1417 }
1418 __inet6_hash(newsk, NULL);
1419
1420 return newsk;
1421
1422 out_overflow:
1423 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENOVERFLOWS);
1424 out_nonewsk:
1425 if (opt && opt != np->opt)
1426 sock_kfree_s(sk, opt, opt->tot_len);
1427 dst_release(dst);
1428 out:
1429 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENDROPS);
1430 return NULL;
1431 }
1432
tcp_v6_checksum_init(struct sk_buff * skb)1433 static __sum16 tcp_v6_checksum_init(struct sk_buff *skb)
1434 {
1435 if (skb->ip_summed == CHECKSUM_COMPLETE) {
1436 if (!tcp_v6_check(skb->len, &ipv6_hdr(skb)->saddr,
1437 &ipv6_hdr(skb)->daddr, skb->csum)) {
1438 skb->ip_summed = CHECKSUM_UNNECESSARY;
1439 return 0;
1440 }
1441 }
1442
1443 skb->csum = ~csum_unfold(tcp_v6_check(skb->len,
1444 &ipv6_hdr(skb)->saddr,
1445 &ipv6_hdr(skb)->daddr, 0));
1446
1447 if (skb->len <= 76) {
1448 return __skb_checksum_complete(skb);
1449 }
1450 return 0;
1451 }
1452
1453 /* The socket must have it's spinlock held when we get
1454 * here.
1455 *
1456 * We have a potential double-lock case here, so even when
1457 * doing backlog processing we use the BH locking scheme.
1458 * This is because we cannot sleep with the original spinlock
1459 * held.
1460 */
tcp_v6_do_rcv(struct sock * sk,struct sk_buff * skb)1461 static int tcp_v6_do_rcv(struct sock *sk, struct sk_buff *skb)
1462 {
1463 struct ipv6_pinfo *np = inet6_sk(sk);
1464 struct tcp_sock *tp;
1465 struct sk_buff *opt_skb = NULL;
1466
1467 /* Imagine: socket is IPv6. IPv4 packet arrives,
1468 goes to IPv4 receive handler and backlogged.
1469 From backlog it always goes here. Kerboom...
1470 Fortunately, tcp_rcv_established and rcv_established
1471 handle them correctly, but it is not case with
1472 tcp_v6_hnd_req and tcp_v6_send_reset(). --ANK
1473 */
1474
1475 if (skb->protocol == htons(ETH_P_IP))
1476 return tcp_v4_do_rcv(sk, skb);
1477
1478 #ifdef CONFIG_TCP_MD5SIG
1479 if (tcp_v6_inbound_md5_hash (sk, skb))
1480 goto discard;
1481 #endif
1482
1483 if (sk_filter(sk, skb))
1484 goto discard;
1485
1486 /*
1487 * socket locking is here for SMP purposes as backlog rcv
1488 * is currently called with bh processing disabled.
1489 */
1490
1491 /* Do Stevens' IPV6_PKTOPTIONS.
1492
1493 Yes, guys, it is the only place in our code, where we
1494 may make it not affecting IPv4.
1495 The rest of code is protocol independent,
1496 and I do not like idea to uglify IPv4.
1497
1498 Actually, all the idea behind IPV6_PKTOPTIONS
1499 looks not very well thought. For now we latch
1500 options, received in the last packet, enqueued
1501 by tcp. Feel free to propose better solution.
1502 --ANK (980728)
1503 */
1504 if (np->rxopt.all)
1505 opt_skb = skb_clone(skb, GFP_ATOMIC);
1506
1507 if (sk->sk_state == TCP_ESTABLISHED) { /* Fast path */
1508 sock_rps_save_rxhash(sk, skb);
1509 if (tcp_rcv_established(sk, skb, tcp_hdr(skb), skb->len))
1510 goto reset;
1511 if (opt_skb)
1512 goto ipv6_pktoptions;
1513 return 0;
1514 }
1515
1516 if (skb->len < tcp_hdrlen(skb) || tcp_checksum_complete(skb))
1517 goto csum_err;
1518
1519 if (sk->sk_state == TCP_LISTEN) {
1520 struct sock *nsk = tcp_v6_hnd_req(sk, skb);
1521 if (!nsk)
1522 goto discard;
1523
1524 /*
1525 * Queue it on the new socket if the new socket is active,
1526 * otherwise we just shortcircuit this and continue with
1527 * the new socket..
1528 */
1529 if(nsk != sk) {
1530 sock_rps_save_rxhash(nsk, skb);
1531 if (tcp_child_process(sk, nsk, skb))
1532 goto reset;
1533 if (opt_skb)
1534 __kfree_skb(opt_skb);
1535 return 0;
1536 }
1537 } else
1538 sock_rps_save_rxhash(sk, skb);
1539
1540 if (tcp_rcv_state_process(sk, skb, tcp_hdr(skb), skb->len))
1541 goto reset;
1542 if (opt_skb)
1543 goto ipv6_pktoptions;
1544 return 0;
1545
1546 reset:
1547 tcp_v6_send_reset(sk, skb);
1548 discard:
1549 if (opt_skb)
1550 __kfree_skb(opt_skb);
1551 kfree_skb(skb);
1552 return 0;
1553 csum_err:
1554 TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_INERRS);
1555 goto discard;
1556
1557
1558 ipv6_pktoptions:
1559 /* Do you ask, what is it?
1560
1561 1. skb was enqueued by tcp.
1562 2. skb is added to tail of read queue, rather than out of order.
1563 3. socket is not in passive state.
1564 4. Finally, it really contains options, which user wants to receive.
1565 */
1566 tp = tcp_sk(sk);
1567 if (TCP_SKB_CB(opt_skb)->end_seq == tp->rcv_nxt &&
1568 !((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_LISTEN))) {
1569 if (np->rxopt.bits.rxinfo || np->rxopt.bits.rxoinfo)
1570 np->mcast_oif = inet6_iif(opt_skb);
1571 if (np->rxopt.bits.rxhlim || np->rxopt.bits.rxohlim)
1572 np->mcast_hops = ipv6_hdr(opt_skb)->hop_limit;
1573 if (np->rxopt.bits.rxtclass)
1574 np->rcv_tclass = ipv6_tclass(ipv6_hdr(opt_skb));
1575 if (ipv6_opt_accepted(sk, opt_skb)) {
1576 skb_set_owner_r(opt_skb, sk);
1577 opt_skb = xchg(&np->pktoptions, opt_skb);
1578 } else {
1579 __kfree_skb(opt_skb);
1580 opt_skb = xchg(&np->pktoptions, NULL);
1581 }
1582 }
1583
1584 kfree_skb(opt_skb);
1585 return 0;
1586 }
1587
tcp_v6_rcv(struct sk_buff * skb)1588 static int tcp_v6_rcv(struct sk_buff *skb)
1589 {
1590 const struct tcphdr *th;
1591 const struct ipv6hdr *hdr;
1592 struct sock *sk;
1593 int ret;
1594 struct net *net = dev_net(skb->dev);
1595
1596 if (skb->pkt_type != PACKET_HOST)
1597 goto discard_it;
1598
1599 /*
1600 * Count it even if it's bad.
1601 */
1602 TCP_INC_STATS_BH(net, TCP_MIB_INSEGS);
1603
1604 if (!pskb_may_pull(skb, sizeof(struct tcphdr)))
1605 goto discard_it;
1606
1607 th = tcp_hdr(skb);
1608
1609 if (th->doff < sizeof(struct tcphdr)/4)
1610 goto bad_packet;
1611 if (!pskb_may_pull(skb, th->doff*4))
1612 goto discard_it;
1613
1614 if (!skb_csum_unnecessary(skb) && tcp_v6_checksum_init(skb))
1615 goto bad_packet;
1616
1617 th = tcp_hdr(skb);
1618 hdr = ipv6_hdr(skb);
1619 TCP_SKB_CB(skb)->seq = ntohl(th->seq);
1620 TCP_SKB_CB(skb)->end_seq = (TCP_SKB_CB(skb)->seq + th->syn + th->fin +
1621 skb->len - th->doff*4);
1622 TCP_SKB_CB(skb)->ack_seq = ntohl(th->ack_seq);
1623 TCP_SKB_CB(skb)->when = 0;
1624 TCP_SKB_CB(skb)->ip_dsfield = ipv6_get_dsfield(hdr);
1625 TCP_SKB_CB(skb)->sacked = 0;
1626
1627 sk = __inet6_lookup_skb(&tcp_hashinfo, skb, th->source, th->dest);
1628 if (!sk)
1629 goto no_tcp_socket;
1630
1631 process:
1632 if (sk->sk_state == TCP_TIME_WAIT)
1633 goto do_time_wait;
1634
1635 if (hdr->hop_limit < inet6_sk(sk)->min_hopcount) {
1636 NET_INC_STATS_BH(net, LINUX_MIB_TCPMINTTLDROP);
1637 goto discard_and_relse;
1638 }
1639
1640 if (!xfrm6_policy_check(sk, XFRM_POLICY_IN, skb))
1641 goto discard_and_relse;
1642
1643 if (sk_filter(sk, skb))
1644 goto discard_and_relse;
1645
1646 skb->dev = NULL;
1647
1648 bh_lock_sock_nested(sk);
1649 ret = 0;
1650 if (!sock_owned_by_user(sk)) {
1651 #ifdef CONFIG_NET_DMA
1652 struct tcp_sock *tp = tcp_sk(sk);
1653 if (!tp->ucopy.dma_chan && tp->ucopy.pinned_list)
1654 tp->ucopy.dma_chan = net_dma_find_channel();
1655 if (tp->ucopy.dma_chan)
1656 ret = tcp_v6_do_rcv(sk, skb);
1657 else
1658 #endif
1659 {
1660 if (!tcp_prequeue(sk, skb))
1661 ret = tcp_v6_do_rcv(sk, skb);
1662 }
1663 } else if (unlikely(sk_add_backlog(sk, skb))) {
1664 bh_unlock_sock(sk);
1665 NET_INC_STATS_BH(net, LINUX_MIB_TCPBACKLOGDROP);
1666 goto discard_and_relse;
1667 }
1668 bh_unlock_sock(sk);
1669
1670 sock_put(sk);
1671 return ret ? -1 : 0;
1672
1673 no_tcp_socket:
1674 if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb))
1675 goto discard_it;
1676
1677 if (skb->len < (th->doff<<2) || tcp_checksum_complete(skb)) {
1678 bad_packet:
1679 TCP_INC_STATS_BH(net, TCP_MIB_INERRS);
1680 } else {
1681 tcp_v6_send_reset(NULL, skb);
1682 }
1683
1684 discard_it:
1685
1686 /*
1687 * Discard frame
1688 */
1689
1690 kfree_skb(skb);
1691 return 0;
1692
1693 discard_and_relse:
1694 sock_put(sk);
1695 goto discard_it;
1696
1697 do_time_wait:
1698 if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb)) {
1699 inet_twsk_put(inet_twsk(sk));
1700 goto discard_it;
1701 }
1702
1703 if (skb->len < (th->doff<<2) || tcp_checksum_complete(skb)) {
1704 TCP_INC_STATS_BH(net, TCP_MIB_INERRS);
1705 inet_twsk_put(inet_twsk(sk));
1706 goto discard_it;
1707 }
1708
1709 switch (tcp_timewait_state_process(inet_twsk(sk), skb, th)) {
1710 case TCP_TW_SYN:
1711 {
1712 struct sock *sk2;
1713
1714 sk2 = inet6_lookup_listener(dev_net(skb->dev), &tcp_hashinfo,
1715 &ipv6_hdr(skb)->daddr,
1716 ntohs(th->dest), inet6_iif(skb));
1717 if (sk2 != NULL) {
1718 struct inet_timewait_sock *tw = inet_twsk(sk);
1719 inet_twsk_deschedule(tw, &tcp_death_row);
1720 inet_twsk_put(tw);
1721 sk = sk2;
1722 goto process;
1723 }
1724 /* Fall through to ACK */
1725 }
1726 case TCP_TW_ACK:
1727 tcp_v6_timewait_ack(sk, skb);
1728 break;
1729 case TCP_TW_RST:
1730 goto no_tcp_socket;
1731 case TCP_TW_SUCCESS:;
1732 }
1733 goto discard_it;
1734 }
1735
tcp_v6_get_peer(struct sock * sk,bool * release_it)1736 static struct inet_peer *tcp_v6_get_peer(struct sock *sk, bool *release_it)
1737 {
1738 struct rt6_info *rt = (struct rt6_info *) __sk_dst_get(sk);
1739 struct ipv6_pinfo *np = inet6_sk(sk);
1740 struct inet_peer *peer;
1741
1742 if (!rt ||
1743 !ipv6_addr_equal(&np->daddr, &rt->rt6i_dst.addr)) {
1744 peer = inet_getpeer_v6(&np->daddr, 1);
1745 *release_it = true;
1746 } else {
1747 if (!rt->rt6i_peer)
1748 rt6_bind_peer(rt, 1);
1749 peer = rt->rt6i_peer;
1750 *release_it = false;
1751 }
1752
1753 return peer;
1754 }
1755
tcp_v6_tw_get_peer(struct sock * sk)1756 static void *tcp_v6_tw_get_peer(struct sock *sk)
1757 {
1758 const struct inet6_timewait_sock *tw6 = inet6_twsk(sk);
1759 const struct inet_timewait_sock *tw = inet_twsk(sk);
1760
1761 if (tw->tw_family == AF_INET)
1762 return tcp_v4_tw_get_peer(sk);
1763
1764 return inet_getpeer_v6(&tw6->tw_v6_daddr, 1);
1765 }
1766
1767 static struct timewait_sock_ops tcp6_timewait_sock_ops = {
1768 .twsk_obj_size = sizeof(struct tcp6_timewait_sock),
1769 .twsk_unique = tcp_twsk_unique,
1770 .twsk_destructor= tcp_twsk_destructor,
1771 .twsk_getpeer = tcp_v6_tw_get_peer,
1772 };
1773
1774 static const struct inet_connection_sock_af_ops ipv6_specific = {
1775 .queue_xmit = inet6_csk_xmit,
1776 .send_check = tcp_v6_send_check,
1777 .rebuild_header = inet6_sk_rebuild_header,
1778 .conn_request = tcp_v6_conn_request,
1779 .syn_recv_sock = tcp_v6_syn_recv_sock,
1780 .get_peer = tcp_v6_get_peer,
1781 .net_header_len = sizeof(struct ipv6hdr),
1782 .setsockopt = ipv6_setsockopt,
1783 .getsockopt = ipv6_getsockopt,
1784 .addr2sockaddr = inet6_csk_addr2sockaddr,
1785 .sockaddr_len = sizeof(struct sockaddr_in6),
1786 .bind_conflict = inet6_csk_bind_conflict,
1787 #ifdef CONFIG_COMPAT
1788 .compat_setsockopt = compat_ipv6_setsockopt,
1789 .compat_getsockopt = compat_ipv6_getsockopt,
1790 #endif
1791 };
1792
1793 #ifdef CONFIG_TCP_MD5SIG
1794 static const struct tcp_sock_af_ops tcp_sock_ipv6_specific = {
1795 .md5_lookup = tcp_v6_md5_lookup,
1796 .calc_md5_hash = tcp_v6_md5_hash_skb,
1797 .md5_parse = tcp_v6_parse_md5_keys,
1798 };
1799 #endif
1800
1801 /*
1802 * TCP over IPv4 via INET6 API
1803 */
1804
1805 static const struct inet_connection_sock_af_ops ipv6_mapped = {
1806 .queue_xmit = ip_queue_xmit,
1807 .send_check = tcp_v4_send_check,
1808 .rebuild_header = inet_sk_rebuild_header,
1809 .conn_request = tcp_v6_conn_request,
1810 .syn_recv_sock = tcp_v6_syn_recv_sock,
1811 .get_peer = tcp_v4_get_peer,
1812 .net_header_len = sizeof(struct iphdr),
1813 .setsockopt = ipv6_setsockopt,
1814 .getsockopt = ipv6_getsockopt,
1815 .addr2sockaddr = inet6_csk_addr2sockaddr,
1816 .sockaddr_len = sizeof(struct sockaddr_in6),
1817 .bind_conflict = inet6_csk_bind_conflict,
1818 #ifdef CONFIG_COMPAT
1819 .compat_setsockopt = compat_ipv6_setsockopt,
1820 .compat_getsockopt = compat_ipv6_getsockopt,
1821 #endif
1822 };
1823
1824 #ifdef CONFIG_TCP_MD5SIG
1825 static const struct tcp_sock_af_ops tcp_sock_ipv6_mapped_specific = {
1826 .md5_lookup = tcp_v4_md5_lookup,
1827 .calc_md5_hash = tcp_v4_md5_hash_skb,
1828 .md5_parse = tcp_v6_parse_md5_keys,
1829 };
1830 #endif
1831
1832 /* NOTE: A lot of things set to zero explicitly by call to
1833 * sk_alloc() so need not be done here.
1834 */
tcp_v6_init_sock(struct sock * sk)1835 static int tcp_v6_init_sock(struct sock *sk)
1836 {
1837 struct inet_connection_sock *icsk = inet_csk(sk);
1838 struct tcp_sock *tp = tcp_sk(sk);
1839
1840 skb_queue_head_init(&tp->out_of_order_queue);
1841 tcp_init_xmit_timers(sk);
1842 tcp_prequeue_init(tp);
1843
1844 icsk->icsk_rto = TCP_TIMEOUT_INIT;
1845 tp->mdev = TCP_TIMEOUT_INIT;
1846
1847 /* So many TCP implementations out there (incorrectly) count the
1848 * initial SYN frame in their delayed-ACK and congestion control
1849 * algorithms that we must have the following bandaid to talk
1850 * efficiently to them. -DaveM
1851 */
1852 tp->snd_cwnd = 2;
1853
1854 /* See draft-stevens-tcpca-spec-01 for discussion of the
1855 * initialization of these values.
1856 */
1857 tp->snd_ssthresh = TCP_INFINITE_SSTHRESH;
1858 tp->snd_cwnd_clamp = ~0;
1859 tp->mss_cache = TCP_MSS_DEFAULT;
1860
1861 tp->reordering = sysctl_tcp_reordering;
1862
1863 sk->sk_state = TCP_CLOSE;
1864
1865 icsk->icsk_af_ops = &ipv6_specific;
1866 icsk->icsk_ca_ops = &tcp_init_congestion_ops;
1867 icsk->icsk_sync_mss = tcp_sync_mss;
1868 sk->sk_write_space = sk_stream_write_space;
1869 sock_set_flag(sk, SOCK_USE_WRITE_QUEUE);
1870
1871 #ifdef CONFIG_TCP_MD5SIG
1872 tp->af_specific = &tcp_sock_ipv6_specific;
1873 #endif
1874
1875 /* TCP Cookie Transactions */
1876 if (sysctl_tcp_cookie_size > 0) {
1877 /* Default, cookies without s_data_payload. */
1878 tp->cookie_values =
1879 kzalloc(sizeof(*tp->cookie_values),
1880 sk->sk_allocation);
1881 if (tp->cookie_values != NULL)
1882 kref_init(&tp->cookie_values->kref);
1883 }
1884 /* Presumed zeroed, in order of appearance:
1885 * cookie_in_always, cookie_out_never,
1886 * s_data_constant, s_data_in, s_data_out
1887 */
1888 sk->sk_sndbuf = sysctl_tcp_wmem[1];
1889 sk->sk_rcvbuf = sysctl_tcp_rmem[1];
1890
1891 local_bh_disable();
1892 sock_update_memcg(sk);
1893 sk_sockets_allocated_inc(sk);
1894 local_bh_enable();
1895
1896 return 0;
1897 }
1898
tcp_v6_destroy_sock(struct sock * sk)1899 static void tcp_v6_destroy_sock(struct sock *sk)
1900 {
1901 tcp_v4_destroy_sock(sk);
1902 inet6_destroy_sock(sk);
1903 }
1904
1905 #ifdef CONFIG_PROC_FS
1906 /* Proc filesystem TCPv6 sock list dumping. */
get_openreq6(struct seq_file * seq,const struct sock * sk,struct request_sock * req,int i,int uid)1907 static void get_openreq6(struct seq_file *seq,
1908 const struct sock *sk, struct request_sock *req, int i, int uid)
1909 {
1910 int ttd = req->expires - jiffies;
1911 const struct in6_addr *src = &inet6_rsk(req)->loc_addr;
1912 const struct in6_addr *dest = &inet6_rsk(req)->rmt_addr;
1913
1914 if (ttd < 0)
1915 ttd = 0;
1916
1917 seq_printf(seq,
1918 "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
1919 "%02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %pK\n",
1920 i,
1921 src->s6_addr32[0], src->s6_addr32[1],
1922 src->s6_addr32[2], src->s6_addr32[3],
1923 ntohs(inet_rsk(req)->loc_port),
1924 dest->s6_addr32[0], dest->s6_addr32[1],
1925 dest->s6_addr32[2], dest->s6_addr32[3],
1926 ntohs(inet_rsk(req)->rmt_port),
1927 TCP_SYN_RECV,
1928 0,0, /* could print option size, but that is af dependent. */
1929 1, /* timers active (only the expire timer) */
1930 jiffies_to_clock_t(ttd),
1931 req->retrans,
1932 uid,
1933 0, /* non standard timer */
1934 0, /* open_requests have no inode */
1935 0, req);
1936 }
1937
get_tcp6_sock(struct seq_file * seq,struct sock * sp,int i)1938 static void get_tcp6_sock(struct seq_file *seq, struct sock *sp, int i)
1939 {
1940 const struct in6_addr *dest, *src;
1941 __u16 destp, srcp;
1942 int timer_active;
1943 unsigned long timer_expires;
1944 const struct inet_sock *inet = inet_sk(sp);
1945 const struct tcp_sock *tp = tcp_sk(sp);
1946 const struct inet_connection_sock *icsk = inet_csk(sp);
1947 const struct ipv6_pinfo *np = inet6_sk(sp);
1948
1949 dest = &np->daddr;
1950 src = &np->rcv_saddr;
1951 destp = ntohs(inet->inet_dport);
1952 srcp = ntohs(inet->inet_sport);
1953
1954 if (icsk->icsk_pending == ICSK_TIME_RETRANS) {
1955 timer_active = 1;
1956 timer_expires = icsk->icsk_timeout;
1957 } else if (icsk->icsk_pending == ICSK_TIME_PROBE0) {
1958 timer_active = 4;
1959 timer_expires = icsk->icsk_timeout;
1960 } else if (timer_pending(&sp->sk_timer)) {
1961 timer_active = 2;
1962 timer_expires = sp->sk_timer.expires;
1963 } else {
1964 timer_active = 0;
1965 timer_expires = jiffies;
1966 }
1967
1968 seq_printf(seq,
1969 "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
1970 "%02X %08X:%08X %02X:%08lX %08X %5d %8d %lu %d %pK %lu %lu %u %u %d\n",
1971 i,
1972 src->s6_addr32[0], src->s6_addr32[1],
1973 src->s6_addr32[2], src->s6_addr32[3], srcp,
1974 dest->s6_addr32[0], dest->s6_addr32[1],
1975 dest->s6_addr32[2], dest->s6_addr32[3], destp,
1976 sp->sk_state,
1977 tp->write_seq-tp->snd_una,
1978 (sp->sk_state == TCP_LISTEN) ? sp->sk_ack_backlog : (tp->rcv_nxt - tp->copied_seq),
1979 timer_active,
1980 jiffies_to_clock_t(timer_expires - jiffies),
1981 icsk->icsk_retransmits,
1982 sock_i_uid(sp),
1983 icsk->icsk_probes_out,
1984 sock_i_ino(sp),
1985 atomic_read(&sp->sk_refcnt), sp,
1986 jiffies_to_clock_t(icsk->icsk_rto),
1987 jiffies_to_clock_t(icsk->icsk_ack.ato),
1988 (icsk->icsk_ack.quick << 1 ) | icsk->icsk_ack.pingpong,
1989 tp->snd_cwnd,
1990 tcp_in_initial_slowstart(tp) ? -1 : tp->snd_ssthresh
1991 );
1992 }
1993
get_timewait6_sock(struct seq_file * seq,struct inet_timewait_sock * tw,int i)1994 static void get_timewait6_sock(struct seq_file *seq,
1995 struct inet_timewait_sock *tw, int i)
1996 {
1997 const struct in6_addr *dest, *src;
1998 __u16 destp, srcp;
1999 const struct inet6_timewait_sock *tw6 = inet6_twsk((struct sock *)tw);
2000 int ttd = tw->tw_ttd - jiffies;
2001
2002 if (ttd < 0)
2003 ttd = 0;
2004
2005 dest = &tw6->tw_v6_daddr;
2006 src = &tw6->tw_v6_rcv_saddr;
2007 destp = ntohs(tw->tw_dport);
2008 srcp = ntohs(tw->tw_sport);
2009
2010 seq_printf(seq,
2011 "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
2012 "%02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %pK\n",
2013 i,
2014 src->s6_addr32[0], src->s6_addr32[1],
2015 src->s6_addr32[2], src->s6_addr32[3], srcp,
2016 dest->s6_addr32[0], dest->s6_addr32[1],
2017 dest->s6_addr32[2], dest->s6_addr32[3], destp,
2018 tw->tw_substate, 0, 0,
2019 3, jiffies_to_clock_t(ttd), 0, 0, 0, 0,
2020 atomic_read(&tw->tw_refcnt), tw);
2021 }
2022
tcp6_seq_show(struct seq_file * seq,void * v)2023 static int tcp6_seq_show(struct seq_file *seq, void *v)
2024 {
2025 struct tcp_iter_state *st;
2026
2027 if (v == SEQ_START_TOKEN) {
2028 seq_puts(seq,
2029 " sl "
2030 "local_address "
2031 "remote_address "
2032 "st tx_queue rx_queue tr tm->when retrnsmt"
2033 " uid timeout inode\n");
2034 goto out;
2035 }
2036 st = seq->private;
2037
2038 switch (st->state) {
2039 case TCP_SEQ_STATE_LISTENING:
2040 case TCP_SEQ_STATE_ESTABLISHED:
2041 get_tcp6_sock(seq, v, st->num);
2042 break;
2043 case TCP_SEQ_STATE_OPENREQ:
2044 get_openreq6(seq, st->syn_wait_sk, v, st->num, st->uid);
2045 break;
2046 case TCP_SEQ_STATE_TIME_WAIT:
2047 get_timewait6_sock(seq, v, st->num);
2048 break;
2049 }
2050 out:
2051 return 0;
2052 }
2053
2054 static const struct file_operations tcp6_afinfo_seq_fops = {
2055 .owner = THIS_MODULE,
2056 .open = tcp_seq_open,
2057 .read = seq_read,
2058 .llseek = seq_lseek,
2059 .release = seq_release_net
2060 };
2061
2062 static struct tcp_seq_afinfo tcp6_seq_afinfo = {
2063 .name = "tcp6",
2064 .family = AF_INET6,
2065 .seq_fops = &tcp6_afinfo_seq_fops,
2066 .seq_ops = {
2067 .show = tcp6_seq_show,
2068 },
2069 };
2070
tcp6_proc_init(struct net * net)2071 int __net_init tcp6_proc_init(struct net *net)
2072 {
2073 return tcp_proc_register(net, &tcp6_seq_afinfo);
2074 }
2075
tcp6_proc_exit(struct net * net)2076 void tcp6_proc_exit(struct net *net)
2077 {
2078 tcp_proc_unregister(net, &tcp6_seq_afinfo);
2079 }
2080 #endif
2081
tcp_v6_clear_sk(struct sock * sk,int size)2082 static void tcp_v6_clear_sk(struct sock *sk, int size)
2083 {
2084 struct inet_sock *inet = inet_sk(sk);
2085
2086 /* we do not want to clear pinet6 field, because of RCU lookups */
2087 sk_prot_clear_nulls(sk, offsetof(struct inet_sock, pinet6));
2088
2089 size -= offsetof(struct inet_sock, pinet6) + sizeof(inet->pinet6);
2090 memset(&inet->pinet6 + 1, 0, size);
2091 }
2092
2093 struct proto tcpv6_prot = {
2094 .name = "TCPv6",
2095 .owner = THIS_MODULE,
2096 .close = tcp_close,
2097 .connect = tcp_v6_connect,
2098 .disconnect = tcp_disconnect,
2099 .accept = inet_csk_accept,
2100 .ioctl = tcp_ioctl,
2101 .init = tcp_v6_init_sock,
2102 .destroy = tcp_v6_destroy_sock,
2103 .shutdown = tcp_shutdown,
2104 .setsockopt = tcp_setsockopt,
2105 .getsockopt = tcp_getsockopt,
2106 .recvmsg = tcp_recvmsg,
2107 .sendmsg = tcp_sendmsg,
2108 .sendpage = tcp_sendpage,
2109 .backlog_rcv = tcp_v6_do_rcv,
2110 .hash = tcp_v6_hash,
2111 .unhash = inet_unhash,
2112 .get_port = inet_csk_get_port,
2113 .enter_memory_pressure = tcp_enter_memory_pressure,
2114 .sockets_allocated = &tcp_sockets_allocated,
2115 .memory_allocated = &tcp_memory_allocated,
2116 .memory_pressure = &tcp_memory_pressure,
2117 .orphan_count = &tcp_orphan_count,
2118 .sysctl_wmem = sysctl_tcp_wmem,
2119 .sysctl_rmem = sysctl_tcp_rmem,
2120 .max_header = MAX_TCP_HEADER,
2121 .obj_size = sizeof(struct tcp6_sock),
2122 .slab_flags = SLAB_DESTROY_BY_RCU,
2123 .twsk_prot = &tcp6_timewait_sock_ops,
2124 .rsk_prot = &tcp6_request_sock_ops,
2125 .h.hashinfo = &tcp_hashinfo,
2126 .no_autobind = true,
2127 #ifdef CONFIG_COMPAT
2128 .compat_setsockopt = compat_tcp_setsockopt,
2129 .compat_getsockopt = compat_tcp_getsockopt,
2130 #endif
2131 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_KMEM
2132 .proto_cgroup = tcp_proto_cgroup,
2133 #endif
2134 .clear_sk = tcp_v6_clear_sk,
2135 };
2136
2137 static const struct inet6_protocol tcpv6_protocol = {
2138 .handler = tcp_v6_rcv,
2139 .err_handler = tcp_v6_err,
2140 .gso_send_check = tcp_v6_gso_send_check,
2141 .gso_segment = tcp_tso_segment,
2142 .gro_receive = tcp6_gro_receive,
2143 .gro_complete = tcp6_gro_complete,
2144 .flags = INET6_PROTO_NOPOLICY|INET6_PROTO_FINAL,
2145 };
2146
2147 static struct inet_protosw tcpv6_protosw = {
2148 .type = SOCK_STREAM,
2149 .protocol = IPPROTO_TCP,
2150 .prot = &tcpv6_prot,
2151 .ops = &inet6_stream_ops,
2152 .no_check = 0,
2153 .flags = INET_PROTOSW_PERMANENT |
2154 INET_PROTOSW_ICSK,
2155 };
2156
tcpv6_net_init(struct net * net)2157 static int __net_init tcpv6_net_init(struct net *net)
2158 {
2159 return inet_ctl_sock_create(&net->ipv6.tcp_sk, PF_INET6,
2160 SOCK_RAW, IPPROTO_TCP, net);
2161 }
2162
tcpv6_net_exit(struct net * net)2163 static void __net_exit tcpv6_net_exit(struct net *net)
2164 {
2165 inet_ctl_sock_destroy(net->ipv6.tcp_sk);
2166 }
2167
tcpv6_net_exit_batch(struct list_head * net_exit_list)2168 static void __net_exit tcpv6_net_exit_batch(struct list_head *net_exit_list)
2169 {
2170 inet_twsk_purge(&tcp_hashinfo, &tcp_death_row, AF_INET6);
2171 }
2172
2173 static struct pernet_operations tcpv6_net_ops = {
2174 .init = tcpv6_net_init,
2175 .exit = tcpv6_net_exit,
2176 .exit_batch = tcpv6_net_exit_batch,
2177 };
2178
tcpv6_init(void)2179 int __init tcpv6_init(void)
2180 {
2181 int ret;
2182
2183 ret = inet6_add_protocol(&tcpv6_protocol, IPPROTO_TCP);
2184 if (ret)
2185 goto out;
2186
2187 /* register inet6 protocol */
2188 ret = inet6_register_protosw(&tcpv6_protosw);
2189 if (ret)
2190 goto out_tcpv6_protocol;
2191
2192 ret = register_pernet_subsys(&tcpv6_net_ops);
2193 if (ret)
2194 goto out_tcpv6_protosw;
2195 out:
2196 return ret;
2197
2198 out_tcpv6_protocol:
2199 inet6_del_protocol(&tcpv6_protocol, IPPROTO_TCP);
2200 out_tcpv6_protosw:
2201 inet6_unregister_protosw(&tcpv6_protosw);
2202 goto out;
2203 }
2204
tcpv6_exit(void)2205 void tcpv6_exit(void)
2206 {
2207 unregister_pernet_subsys(&tcpv6_net_ops);
2208 inet6_unregister_protosw(&tcpv6_protosw);
2209 inet6_del_protocol(&tcpv6_protocol, IPPROTO_TCP);
2210 }
2211