1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3 * TCP over IPv6
4 * Linux INET6 implementation
5 *
6 * Authors:
7 * Pedro Roque <roque@di.fc.ul.pt>
8 *
9 * Based on:
10 * linux/net/ipv4/tcp.c
11 * linux/net/ipv4/tcp_input.c
12 * linux/net/ipv4/tcp_output.c
13 *
14 * Fixes:
15 * Hideaki YOSHIFUJI : sin6_scope_id support
16 * YOSHIFUJI Hideaki @USAGI and: Support IPV6_V6ONLY socket option, which
17 * Alexey Kuznetsov allow both IPv4 and IPv6 sockets to bind
18 * a single port at the same time.
19 * YOSHIFUJI Hideaki @USAGI: convert /proc/net/tcp6 to seq_file.
20 */
21
22 #include <linux/bottom_half.h>
23 #include <linux/module.h>
24 #include <linux/errno.h>
25 #include <linux/types.h>
26 #include <linux/socket.h>
27 #include <linux/sockios.h>
28 #include <linux/net.h>
29 #include <linux/jiffies.h>
30 #include <linux/in.h>
31 #include <linux/in6.h>
32 #include <linux/netdevice.h>
33 #include <linux/init.h>
34 #include <linux/jhash.h>
35 #include <linux/ipsec.h>
36 #include <linux/times.h>
37 #include <linux/slab.h>
38 #include <linux/uaccess.h>
39 #include <linux/ipv6.h>
40 #include <linux/icmpv6.h>
41 #include <linux/random.h>
42 #include <linux/indirect_call_wrapper.h>
43
44 #include <net/tcp.h>
45 #include <net/ndisc.h>
46 #include <net/inet6_hashtables.h>
47 #include <net/inet6_connection_sock.h>
48 #include <net/ipv6.h>
49 #include <net/transp_v6.h>
50 #include <net/addrconf.h>
51 #include <net/ip6_route.h>
52 #include <net/ip6_checksum.h>
53 #include <net/inet_ecn.h>
54 #include <net/protocol.h>
55 #include <net/xfrm.h>
56 #include <net/snmp.h>
57 #include <net/dsfield.h>
58 #include <net/timewait_sock.h>
59 #include <net/inet_common.h>
60 #include <net/secure_seq.h>
61 #include <net/busy_poll.h>
62
63 #include <linux/proc_fs.h>
64 #include <linux/seq_file.h>
65
66 #include <crypto/hash.h>
67 #include <linux/scatterlist.h>
68
69 #include <trace/events/tcp.h>
70
71 static void tcp_v6_send_reset(const struct sock *sk, struct sk_buff *skb);
72 static void tcp_v6_reqsk_send_ack(const struct sock *sk, struct sk_buff *skb,
73 struct request_sock *req);
74
75 static int tcp_v6_do_rcv(struct sock *sk, struct sk_buff *skb);
76
77 static const struct inet_connection_sock_af_ops ipv6_mapped;
78 const struct inet_connection_sock_af_ops ipv6_specific;
79 #ifdef CONFIG_TCP_MD5SIG
80 static const struct tcp_sock_af_ops tcp_sock_ipv6_specific;
81 static const struct tcp_sock_af_ops tcp_sock_ipv6_mapped_specific;
82 #else
tcp_v6_md5_do_lookup(const struct sock * sk,const struct in6_addr * addr,int l3index)83 static struct tcp_md5sig_key *tcp_v6_md5_do_lookup(const struct sock *sk,
84 const struct in6_addr *addr,
85 int l3index)
86 {
87 return NULL;
88 }
89 #endif
90
91 /* Helper returning the inet6 address from a given tcp socket.
92 * It can be used in TCP stack instead of inet6_sk(sk).
93 * This avoids a dereference and allow compiler optimizations.
94 * It is a specialized version of inet6_sk_generic().
95 */
tcp_inet6_sk(const struct sock * sk)96 static struct ipv6_pinfo *tcp_inet6_sk(const struct sock *sk)
97 {
98 unsigned int offset = sizeof(struct tcp6_sock) - sizeof(struct ipv6_pinfo);
99
100 return (struct ipv6_pinfo *)(((u8 *)sk) + offset);
101 }
102
inet6_sk_rx_dst_set(struct sock * sk,const struct sk_buff * skb)103 static void inet6_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb)
104 {
105 struct dst_entry *dst = skb_dst(skb);
106
107 if (dst && dst_hold_safe(dst)) {
108 const struct rt6_info *rt = (const struct rt6_info *)dst;
109
110 sk->sk_rx_dst = dst;
111 inet_sk(sk)->rx_dst_ifindex = skb->skb_iif;
112 tcp_inet6_sk(sk)->rx_dst_cookie = rt6_get_cookie(rt);
113 }
114 }
115
tcp_v6_init_seq(const struct sk_buff * skb)116 static u32 tcp_v6_init_seq(const struct sk_buff *skb)
117 {
118 return secure_tcpv6_seq(ipv6_hdr(skb)->daddr.s6_addr32,
119 ipv6_hdr(skb)->saddr.s6_addr32,
120 tcp_hdr(skb)->dest,
121 tcp_hdr(skb)->source);
122 }
123
tcp_v6_init_ts_off(const struct net * net,const struct sk_buff * skb)124 static u32 tcp_v6_init_ts_off(const struct net *net, const struct sk_buff *skb)
125 {
126 return secure_tcpv6_ts_off(net, ipv6_hdr(skb)->daddr.s6_addr32,
127 ipv6_hdr(skb)->saddr.s6_addr32);
128 }
129
tcp_v6_pre_connect(struct sock * sk,struct sockaddr * uaddr,int addr_len)130 static int tcp_v6_pre_connect(struct sock *sk, struct sockaddr *uaddr,
131 int addr_len)
132 {
133 /* This check is replicated from tcp_v6_connect() and intended to
134 * prevent BPF program called below from accessing bytes that are out
135 * of the bound specified by user in addr_len.
136 */
137 if (addr_len < SIN6_LEN_RFC2133)
138 return -EINVAL;
139
140 sock_owned_by_me(sk);
141
142 return BPF_CGROUP_RUN_PROG_INET6_CONNECT(sk, uaddr);
143 }
144
tcp_v6_connect(struct sock * sk,struct sockaddr * uaddr,int addr_len)145 static int tcp_v6_connect(struct sock *sk, struct sockaddr *uaddr,
146 int addr_len)
147 {
148 struct sockaddr_in6 *usin = (struct sockaddr_in6 *) uaddr;
149 struct inet_sock *inet = inet_sk(sk);
150 struct inet_connection_sock *icsk = inet_csk(sk);
151 struct ipv6_pinfo *np = tcp_inet6_sk(sk);
152 struct tcp_sock *tp = tcp_sk(sk);
153 struct in6_addr *saddr = NULL, *final_p, final;
154 struct ipv6_txoptions *opt;
155 struct flowi6 fl6;
156 struct dst_entry *dst;
157 int addr_type;
158 int err;
159 struct inet_timewait_death_row *tcp_death_row = &sock_net(sk)->ipv4.tcp_death_row;
160
161 if (addr_len < SIN6_LEN_RFC2133)
162 return -EINVAL;
163
164 if (usin->sin6_family != AF_INET6)
165 return -EAFNOSUPPORT;
166
167 memset(&fl6, 0, sizeof(fl6));
168
169 if (np->sndflow) {
170 fl6.flowlabel = usin->sin6_flowinfo&IPV6_FLOWINFO_MASK;
171 IP6_ECN_flow_init(fl6.flowlabel);
172 if (fl6.flowlabel&IPV6_FLOWLABEL_MASK) {
173 struct ip6_flowlabel *flowlabel;
174 flowlabel = fl6_sock_lookup(sk, fl6.flowlabel);
175 if (IS_ERR(flowlabel))
176 return -EINVAL;
177 fl6_sock_release(flowlabel);
178 }
179 }
180
181 /*
182 * connect() to INADDR_ANY means loopback (BSD'ism).
183 */
184
185 if (ipv6_addr_any(&usin->sin6_addr)) {
186 if (ipv6_addr_v4mapped(&sk->sk_v6_rcv_saddr))
187 ipv6_addr_set_v4mapped(htonl(INADDR_LOOPBACK),
188 &usin->sin6_addr);
189 else
190 usin->sin6_addr = in6addr_loopback;
191 }
192
193 addr_type = ipv6_addr_type(&usin->sin6_addr);
194
195 if (addr_type & IPV6_ADDR_MULTICAST)
196 return -ENETUNREACH;
197
198 if (addr_type&IPV6_ADDR_LINKLOCAL) {
199 if (addr_len >= sizeof(struct sockaddr_in6) &&
200 usin->sin6_scope_id) {
201 /* If interface is set while binding, indices
202 * must coincide.
203 */
204 if (!sk_dev_equal_l3scope(sk, usin->sin6_scope_id))
205 return -EINVAL;
206
207 sk->sk_bound_dev_if = usin->sin6_scope_id;
208 }
209
210 /* Connect to link-local address requires an interface */
211 if (!sk->sk_bound_dev_if)
212 return -EINVAL;
213 }
214
215 if (tp->rx_opt.ts_recent_stamp &&
216 !ipv6_addr_equal(&sk->sk_v6_daddr, &usin->sin6_addr)) {
217 tp->rx_opt.ts_recent = 0;
218 tp->rx_opt.ts_recent_stamp = 0;
219 WRITE_ONCE(tp->write_seq, 0);
220 }
221
222 sk->sk_v6_daddr = usin->sin6_addr;
223 np->flow_label = fl6.flowlabel;
224
225 /*
226 * TCP over IPv4
227 */
228
229 if (addr_type & IPV6_ADDR_MAPPED) {
230 u32 exthdrlen = icsk->icsk_ext_hdr_len;
231 struct sockaddr_in sin;
232
233 if (__ipv6_only_sock(sk))
234 return -ENETUNREACH;
235
236 sin.sin_family = AF_INET;
237 sin.sin_port = usin->sin6_port;
238 sin.sin_addr.s_addr = usin->sin6_addr.s6_addr32[3];
239
240 icsk->icsk_af_ops = &ipv6_mapped;
241 if (sk_is_mptcp(sk))
242 mptcpv6_handle_mapped(sk, true);
243 sk->sk_backlog_rcv = tcp_v4_do_rcv;
244 #ifdef CONFIG_TCP_MD5SIG
245 tp->af_specific = &tcp_sock_ipv6_mapped_specific;
246 #endif
247
248 err = tcp_v4_connect(sk, (struct sockaddr *)&sin, sizeof(sin));
249
250 if (err) {
251 icsk->icsk_ext_hdr_len = exthdrlen;
252 icsk->icsk_af_ops = &ipv6_specific;
253 if (sk_is_mptcp(sk))
254 mptcpv6_handle_mapped(sk, false);
255 sk->sk_backlog_rcv = tcp_v6_do_rcv;
256 #ifdef CONFIG_TCP_MD5SIG
257 tp->af_specific = &tcp_sock_ipv6_specific;
258 #endif
259 goto failure;
260 }
261 np->saddr = sk->sk_v6_rcv_saddr;
262
263 return err;
264 }
265
266 if (!ipv6_addr_any(&sk->sk_v6_rcv_saddr))
267 saddr = &sk->sk_v6_rcv_saddr;
268
269 fl6.flowi6_proto = IPPROTO_TCP;
270 fl6.daddr = sk->sk_v6_daddr;
271 fl6.saddr = saddr ? *saddr : np->saddr;
272 fl6.flowi6_oif = sk->sk_bound_dev_if;
273 fl6.flowi6_mark = sk->sk_mark;
274 fl6.fl6_dport = usin->sin6_port;
275 fl6.fl6_sport = inet->inet_sport;
276 fl6.flowi6_uid = sk->sk_uid;
277
278 opt = rcu_dereference_protected(np->opt, lockdep_sock_is_held(sk));
279 final_p = fl6_update_dst(&fl6, opt, &final);
280
281 security_sk_classify_flow(sk, flowi6_to_flowi_common(&fl6));
282
283 dst = ip6_dst_lookup_flow(sock_net(sk), sk, &fl6, final_p);
284 if (IS_ERR(dst)) {
285 err = PTR_ERR(dst);
286 goto failure;
287 }
288
289 if (!saddr) {
290 saddr = &fl6.saddr;
291 sk->sk_v6_rcv_saddr = *saddr;
292 }
293
294 /* set the source address */
295 np->saddr = *saddr;
296 inet->inet_rcv_saddr = LOOPBACK4_IPV6;
297
298 sk->sk_gso_type = SKB_GSO_TCPV6;
299 ip6_dst_store(sk, dst, NULL, NULL);
300
301 icsk->icsk_ext_hdr_len = 0;
302 if (opt)
303 icsk->icsk_ext_hdr_len = opt->opt_flen +
304 opt->opt_nflen;
305
306 tp->rx_opt.mss_clamp = IPV6_MIN_MTU - sizeof(struct tcphdr) - sizeof(struct ipv6hdr);
307
308 inet->inet_dport = usin->sin6_port;
309
310 tcp_set_state(sk, TCP_SYN_SENT);
311 err = inet6_hash_connect(tcp_death_row, sk);
312 if (err)
313 goto late_failure;
314
315 sk_set_txhash(sk);
316
317 if (likely(!tp->repair)) {
318 if (!tp->write_seq)
319 WRITE_ONCE(tp->write_seq,
320 secure_tcpv6_seq(np->saddr.s6_addr32,
321 sk->sk_v6_daddr.s6_addr32,
322 inet->inet_sport,
323 inet->inet_dport));
324 tp->tsoffset = secure_tcpv6_ts_off(sock_net(sk),
325 np->saddr.s6_addr32,
326 sk->sk_v6_daddr.s6_addr32);
327 }
328
329 if (tcp_fastopen_defer_connect(sk, &err))
330 return err;
331 if (err)
332 goto late_failure;
333
334 err = tcp_connect(sk);
335 if (err)
336 goto late_failure;
337
338 return 0;
339
340 late_failure:
341 tcp_set_state(sk, TCP_CLOSE);
342 failure:
343 inet->inet_dport = 0;
344 sk->sk_route_caps = 0;
345 return err;
346 }
347
tcp_v6_mtu_reduced(struct sock * sk)348 static void tcp_v6_mtu_reduced(struct sock *sk)
349 {
350 struct dst_entry *dst;
351
352 if ((1 << sk->sk_state) & (TCPF_LISTEN | TCPF_CLOSE))
353 return;
354
355 dst = inet6_csk_update_pmtu(sk, tcp_sk(sk)->mtu_info);
356 if (!dst)
357 return;
358
359 if (inet_csk(sk)->icsk_pmtu_cookie > dst_mtu(dst)) {
360 tcp_sync_mss(sk, dst_mtu(dst));
361 tcp_simple_retransmit(sk);
362 }
363 }
364
tcp_v6_err(struct sk_buff * skb,struct inet6_skb_parm * opt,u8 type,u8 code,int offset,__be32 info)365 static int tcp_v6_err(struct sk_buff *skb, struct inet6_skb_parm *opt,
366 u8 type, u8 code, int offset, __be32 info)
367 {
368 const struct ipv6hdr *hdr = (const struct ipv6hdr *)skb->data;
369 const struct tcphdr *th = (struct tcphdr *)(skb->data+offset);
370 struct net *net = dev_net(skb->dev);
371 struct request_sock *fastopen;
372 struct ipv6_pinfo *np;
373 struct tcp_sock *tp;
374 __u32 seq, snd_una;
375 struct sock *sk;
376 bool fatal;
377 int err;
378
379 sk = __inet6_lookup_established(net, &tcp_hashinfo,
380 &hdr->daddr, th->dest,
381 &hdr->saddr, ntohs(th->source),
382 skb->dev->ifindex, inet6_sdif(skb));
383
384 if (!sk) {
385 __ICMP6_INC_STATS(net, __in6_dev_get(skb->dev),
386 ICMP6_MIB_INERRORS);
387 return -ENOENT;
388 }
389
390 if (sk->sk_state == TCP_TIME_WAIT) {
391 inet_twsk_put(inet_twsk(sk));
392 return 0;
393 }
394 seq = ntohl(th->seq);
395 fatal = icmpv6_err_convert(type, code, &err);
396 if (sk->sk_state == TCP_NEW_SYN_RECV) {
397 tcp_req_err(sk, seq, fatal);
398 return 0;
399 }
400
401 bh_lock_sock(sk);
402 if (sock_owned_by_user(sk) && type != ICMPV6_PKT_TOOBIG)
403 __NET_INC_STATS(net, LINUX_MIB_LOCKDROPPEDICMPS);
404
405 if (sk->sk_state == TCP_CLOSE)
406 goto out;
407
408 if (ipv6_hdr(skb)->hop_limit < tcp_inet6_sk(sk)->min_hopcount) {
409 __NET_INC_STATS(net, LINUX_MIB_TCPMINTTLDROP);
410 goto out;
411 }
412
413 tp = tcp_sk(sk);
414 /* XXX (TFO) - tp->snd_una should be ISN (tcp_create_openreq_child() */
415 fastopen = rcu_dereference(tp->fastopen_rsk);
416 snd_una = fastopen ? tcp_rsk(fastopen)->snt_isn : tp->snd_una;
417 if (sk->sk_state != TCP_LISTEN &&
418 !between(seq, snd_una, tp->snd_nxt)) {
419 __NET_INC_STATS(net, LINUX_MIB_OUTOFWINDOWICMPS);
420 goto out;
421 }
422
423 np = tcp_inet6_sk(sk);
424
425 if (type == NDISC_REDIRECT) {
426 if (!sock_owned_by_user(sk)) {
427 struct dst_entry *dst = __sk_dst_check(sk, np->dst_cookie);
428
429 if (dst)
430 dst->ops->redirect(dst, sk, skb);
431 }
432 goto out;
433 }
434
435 if (type == ICMPV6_PKT_TOOBIG) {
436 /* We are not interested in TCP_LISTEN and open_requests
437 * (SYN-ACKs send out by Linux are always <576bytes so
438 * they should go through unfragmented).
439 */
440 if (sk->sk_state == TCP_LISTEN)
441 goto out;
442
443 if (!ip6_sk_accept_pmtu(sk))
444 goto out;
445
446 tp->mtu_info = ntohl(info);
447 if (!sock_owned_by_user(sk))
448 tcp_v6_mtu_reduced(sk);
449 else if (!test_and_set_bit(TCP_MTU_REDUCED_DEFERRED,
450 &sk->sk_tsq_flags))
451 sock_hold(sk);
452 goto out;
453 }
454
455
456 /* Might be for an request_sock */
457 switch (sk->sk_state) {
458 case TCP_SYN_SENT:
459 case TCP_SYN_RECV:
460 /* Only in fast or simultaneous open. If a fast open socket is
461 * already accepted it is treated as a connected one below.
462 */
463 if (fastopen && !fastopen->sk)
464 break;
465
466 ipv6_icmp_error(sk, skb, err, th->dest, ntohl(info), (u8 *)th);
467
468 if (!sock_owned_by_user(sk)) {
469 sk->sk_err = err;
470 sk->sk_error_report(sk); /* Wake people up to see the error (see connect in sock.c) */
471
472 tcp_done(sk);
473 } else
474 sk->sk_err_soft = err;
475 goto out;
476 case TCP_LISTEN:
477 break;
478 default:
479 /* check if this ICMP message allows revert of backoff.
480 * (see RFC 6069)
481 */
482 if (!fastopen && type == ICMPV6_DEST_UNREACH &&
483 code == ICMPV6_NOROUTE)
484 tcp_ld_RTO_revert(sk, seq);
485 }
486
487 if (!sock_owned_by_user(sk) && np->recverr) {
488 sk->sk_err = err;
489 sk->sk_error_report(sk);
490 } else
491 sk->sk_err_soft = err;
492
493 out:
494 bh_unlock_sock(sk);
495 sock_put(sk);
496 return 0;
497 }
498
499
tcp_v6_send_synack(const struct sock * sk,struct dst_entry * dst,struct flowi * fl,struct request_sock * req,struct tcp_fastopen_cookie * foc,enum tcp_synack_type synack_type,struct sk_buff * syn_skb)500 static int tcp_v6_send_synack(const struct sock *sk, struct dst_entry *dst,
501 struct flowi *fl,
502 struct request_sock *req,
503 struct tcp_fastopen_cookie *foc,
504 enum tcp_synack_type synack_type,
505 struct sk_buff *syn_skb)
506 {
507 struct inet_request_sock *ireq = inet_rsk(req);
508 struct ipv6_pinfo *np = tcp_inet6_sk(sk);
509 struct ipv6_txoptions *opt;
510 struct flowi6 *fl6 = &fl->u.ip6;
511 struct sk_buff *skb;
512 int err = -ENOMEM;
513 u8 tclass;
514
515 /* First, grab a route. */
516 if (!dst && (dst = inet6_csk_route_req(sk, fl6, req,
517 IPPROTO_TCP)) == NULL)
518 goto done;
519
520 skb = tcp_make_synack(sk, dst, req, foc, synack_type, syn_skb);
521
522 if (skb) {
523 __tcp_v6_send_check(skb, &ireq->ir_v6_loc_addr,
524 &ireq->ir_v6_rmt_addr);
525
526 fl6->daddr = ireq->ir_v6_rmt_addr;
527 if (np->repflow && ireq->pktopts)
528 fl6->flowlabel = ip6_flowlabel(ipv6_hdr(ireq->pktopts));
529
530 tclass = sock_net(sk)->ipv4.sysctl_tcp_reflect_tos ?
531 (tcp_rsk(req)->syn_tos & ~INET_ECN_MASK) |
532 (np->tclass & INET_ECN_MASK) :
533 np->tclass;
534
535 if (!INET_ECN_is_capable(tclass) &&
536 tcp_bpf_ca_needs_ecn((struct sock *)req))
537 tclass |= INET_ECN_ECT_0;
538
539 rcu_read_lock();
540 opt = ireq->ipv6_opt;
541 if (!opt)
542 opt = rcu_dereference(np->opt);
543 err = ip6_xmit(sk, skb, fl6, sk->sk_mark, opt,
544 tclass, sk->sk_priority);
545 rcu_read_unlock();
546 err = net_xmit_eval(err);
547 }
548
549 done:
550 return err;
551 }
552
553
tcp_v6_reqsk_destructor(struct request_sock * req)554 static void tcp_v6_reqsk_destructor(struct request_sock *req)
555 {
556 kfree(inet_rsk(req)->ipv6_opt);
557 kfree_skb(inet_rsk(req)->pktopts);
558 }
559
560 #ifdef CONFIG_TCP_MD5SIG
tcp_v6_md5_do_lookup(const struct sock * sk,const struct in6_addr * addr,int l3index)561 static struct tcp_md5sig_key *tcp_v6_md5_do_lookup(const struct sock *sk,
562 const struct in6_addr *addr,
563 int l3index)
564 {
565 return tcp_md5_do_lookup(sk, l3index,
566 (union tcp_md5_addr *)addr, AF_INET6);
567 }
568
tcp_v6_md5_lookup(const struct sock * sk,const struct sock * addr_sk)569 static struct tcp_md5sig_key *tcp_v6_md5_lookup(const struct sock *sk,
570 const struct sock *addr_sk)
571 {
572 int l3index;
573
574 l3index = l3mdev_master_ifindex_by_index(sock_net(sk),
575 addr_sk->sk_bound_dev_if);
576 return tcp_v6_md5_do_lookup(sk, &addr_sk->sk_v6_daddr,
577 l3index);
578 }
579
tcp_v6_parse_md5_keys(struct sock * sk,int optname,sockptr_t optval,int optlen)580 static int tcp_v6_parse_md5_keys(struct sock *sk, int optname,
581 sockptr_t optval, int optlen)
582 {
583 struct tcp_md5sig cmd;
584 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)&cmd.tcpm_addr;
585 int l3index = 0;
586 u8 prefixlen;
587
588 if (optlen < sizeof(cmd))
589 return -EINVAL;
590
591 if (copy_from_sockptr(&cmd, optval, sizeof(cmd)))
592 return -EFAULT;
593
594 if (sin6->sin6_family != AF_INET6)
595 return -EINVAL;
596
597 if (optname == TCP_MD5SIG_EXT &&
598 cmd.tcpm_flags & TCP_MD5SIG_FLAG_PREFIX) {
599 prefixlen = cmd.tcpm_prefixlen;
600 if (prefixlen > 128 || (ipv6_addr_v4mapped(&sin6->sin6_addr) &&
601 prefixlen > 32))
602 return -EINVAL;
603 } else {
604 prefixlen = ipv6_addr_v4mapped(&sin6->sin6_addr) ? 32 : 128;
605 }
606
607 if (optname == TCP_MD5SIG_EXT &&
608 cmd.tcpm_flags & TCP_MD5SIG_FLAG_IFINDEX) {
609 struct net_device *dev;
610
611 rcu_read_lock();
612 dev = dev_get_by_index_rcu(sock_net(sk), cmd.tcpm_ifindex);
613 if (dev && netif_is_l3_master(dev))
614 l3index = dev->ifindex;
615 rcu_read_unlock();
616
617 /* ok to reference set/not set outside of rcu;
618 * right now device MUST be an L3 master
619 */
620 if (!dev || !l3index)
621 return -EINVAL;
622 }
623
624 if (!cmd.tcpm_keylen) {
625 if (ipv6_addr_v4mapped(&sin6->sin6_addr))
626 return tcp_md5_do_del(sk, (union tcp_md5_addr *)&sin6->sin6_addr.s6_addr32[3],
627 AF_INET, prefixlen,
628 l3index);
629 return tcp_md5_do_del(sk, (union tcp_md5_addr *)&sin6->sin6_addr,
630 AF_INET6, prefixlen, l3index);
631 }
632
633 if (cmd.tcpm_keylen > TCP_MD5SIG_MAXKEYLEN)
634 return -EINVAL;
635
636 if (ipv6_addr_v4mapped(&sin6->sin6_addr))
637 return tcp_md5_do_add(sk, (union tcp_md5_addr *)&sin6->sin6_addr.s6_addr32[3],
638 AF_INET, prefixlen, l3index,
639 cmd.tcpm_key, cmd.tcpm_keylen,
640 GFP_KERNEL);
641
642 return tcp_md5_do_add(sk, (union tcp_md5_addr *)&sin6->sin6_addr,
643 AF_INET6, prefixlen, l3index,
644 cmd.tcpm_key, cmd.tcpm_keylen, GFP_KERNEL);
645 }
646
tcp_v6_md5_hash_headers(struct tcp_md5sig_pool * hp,const struct in6_addr * daddr,const struct in6_addr * saddr,const struct tcphdr * th,int nbytes)647 static int tcp_v6_md5_hash_headers(struct tcp_md5sig_pool *hp,
648 const struct in6_addr *daddr,
649 const struct in6_addr *saddr,
650 const struct tcphdr *th, int nbytes)
651 {
652 struct tcp6_pseudohdr *bp;
653 struct scatterlist sg;
654 struct tcphdr *_th;
655
656 bp = hp->scratch;
657 /* 1. TCP pseudo-header (RFC2460) */
658 bp->saddr = *saddr;
659 bp->daddr = *daddr;
660 bp->protocol = cpu_to_be32(IPPROTO_TCP);
661 bp->len = cpu_to_be32(nbytes);
662
663 _th = (struct tcphdr *)(bp + 1);
664 memcpy(_th, th, sizeof(*th));
665 _th->check = 0;
666
667 sg_init_one(&sg, bp, sizeof(*bp) + sizeof(*th));
668 ahash_request_set_crypt(hp->md5_req, &sg, NULL,
669 sizeof(*bp) + sizeof(*th));
670 return crypto_ahash_update(hp->md5_req);
671 }
672
tcp_v6_md5_hash_hdr(char * md5_hash,const struct tcp_md5sig_key * key,const struct in6_addr * daddr,struct in6_addr * saddr,const struct tcphdr * th)673 static int tcp_v6_md5_hash_hdr(char *md5_hash, const struct tcp_md5sig_key *key,
674 const struct in6_addr *daddr, struct in6_addr *saddr,
675 const struct tcphdr *th)
676 {
677 struct tcp_md5sig_pool *hp;
678 struct ahash_request *req;
679
680 hp = tcp_get_md5sig_pool();
681 if (!hp)
682 goto clear_hash_noput;
683 req = hp->md5_req;
684
685 if (crypto_ahash_init(req))
686 goto clear_hash;
687 if (tcp_v6_md5_hash_headers(hp, daddr, saddr, th, th->doff << 2))
688 goto clear_hash;
689 if (tcp_md5_hash_key(hp, key))
690 goto clear_hash;
691 ahash_request_set_crypt(req, NULL, md5_hash, 0);
692 if (crypto_ahash_final(req))
693 goto clear_hash;
694
695 tcp_put_md5sig_pool();
696 return 0;
697
698 clear_hash:
699 tcp_put_md5sig_pool();
700 clear_hash_noput:
701 memset(md5_hash, 0, 16);
702 return 1;
703 }
704
tcp_v6_md5_hash_skb(char * md5_hash,const struct tcp_md5sig_key * key,const struct sock * sk,const struct sk_buff * skb)705 static int tcp_v6_md5_hash_skb(char *md5_hash,
706 const struct tcp_md5sig_key *key,
707 const struct sock *sk,
708 const struct sk_buff *skb)
709 {
710 const struct in6_addr *saddr, *daddr;
711 struct tcp_md5sig_pool *hp;
712 struct ahash_request *req;
713 const struct tcphdr *th = tcp_hdr(skb);
714
715 if (sk) { /* valid for establish/request sockets */
716 saddr = &sk->sk_v6_rcv_saddr;
717 daddr = &sk->sk_v6_daddr;
718 } else {
719 const struct ipv6hdr *ip6h = ipv6_hdr(skb);
720 saddr = &ip6h->saddr;
721 daddr = &ip6h->daddr;
722 }
723
724 hp = tcp_get_md5sig_pool();
725 if (!hp)
726 goto clear_hash_noput;
727 req = hp->md5_req;
728
729 if (crypto_ahash_init(req))
730 goto clear_hash;
731
732 if (tcp_v6_md5_hash_headers(hp, daddr, saddr, th, skb->len))
733 goto clear_hash;
734 if (tcp_md5_hash_skb_data(hp, skb, th->doff << 2))
735 goto clear_hash;
736 if (tcp_md5_hash_key(hp, key))
737 goto clear_hash;
738 ahash_request_set_crypt(req, NULL, md5_hash, 0);
739 if (crypto_ahash_final(req))
740 goto clear_hash;
741
742 tcp_put_md5sig_pool();
743 return 0;
744
745 clear_hash:
746 tcp_put_md5sig_pool();
747 clear_hash_noput:
748 memset(md5_hash, 0, 16);
749 return 1;
750 }
751
752 #endif
753
tcp_v6_inbound_md5_hash(const struct sock * sk,const struct sk_buff * skb,int dif,int sdif)754 static bool tcp_v6_inbound_md5_hash(const struct sock *sk,
755 const struct sk_buff *skb,
756 int dif, int sdif)
757 {
758 #ifdef CONFIG_TCP_MD5SIG
759 const __u8 *hash_location = NULL;
760 struct tcp_md5sig_key *hash_expected;
761 const struct ipv6hdr *ip6h = ipv6_hdr(skb);
762 const struct tcphdr *th = tcp_hdr(skb);
763 int genhash, l3index;
764 u8 newhash[16];
765
766 /* sdif set, means packet ingressed via a device
767 * in an L3 domain and dif is set to the l3mdev
768 */
769 l3index = sdif ? dif : 0;
770
771 hash_expected = tcp_v6_md5_do_lookup(sk, &ip6h->saddr, l3index);
772 hash_location = tcp_parse_md5sig_option(th);
773
774 /* We've parsed the options - do we have a hash? */
775 if (!hash_expected && !hash_location)
776 return false;
777
778 if (hash_expected && !hash_location) {
779 NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPMD5NOTFOUND);
780 return true;
781 }
782
783 if (!hash_expected && hash_location) {
784 NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPMD5UNEXPECTED);
785 return true;
786 }
787
788 /* check the signature */
789 genhash = tcp_v6_md5_hash_skb(newhash,
790 hash_expected,
791 NULL, skb);
792
793 if (genhash || memcmp(hash_location, newhash, 16) != 0) {
794 NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPMD5FAILURE);
795 net_info_ratelimited("MD5 Hash %s for [%pI6c]:%u->[%pI6c]:%u L3 index %d\n",
796 genhash ? "failed" : "mismatch",
797 &ip6h->saddr, ntohs(th->source),
798 &ip6h->daddr, ntohs(th->dest), l3index);
799 return true;
800 }
801 #endif
802 return false;
803 }
804
tcp_v6_init_req(struct request_sock * req,const struct sock * sk_listener,struct sk_buff * skb)805 static void tcp_v6_init_req(struct request_sock *req,
806 const struct sock *sk_listener,
807 struct sk_buff *skb)
808 {
809 bool l3_slave = ipv6_l3mdev_skb(TCP_SKB_CB(skb)->header.h6.flags);
810 struct inet_request_sock *ireq = inet_rsk(req);
811 const struct ipv6_pinfo *np = tcp_inet6_sk(sk_listener);
812
813 ireq->ir_v6_rmt_addr = ipv6_hdr(skb)->saddr;
814 ireq->ir_v6_loc_addr = ipv6_hdr(skb)->daddr;
815
816 /* So that link locals have meaning */
817 if ((!sk_listener->sk_bound_dev_if || l3_slave) &&
818 ipv6_addr_type(&ireq->ir_v6_rmt_addr) & IPV6_ADDR_LINKLOCAL)
819 ireq->ir_iif = tcp_v6_iif(skb);
820
821 if (!TCP_SKB_CB(skb)->tcp_tw_isn &&
822 (ipv6_opt_accepted(sk_listener, skb, &TCP_SKB_CB(skb)->header.h6) ||
823 np->rxopt.bits.rxinfo ||
824 np->rxopt.bits.rxoinfo || np->rxopt.bits.rxhlim ||
825 np->rxopt.bits.rxohlim || np->repflow)) {
826 refcount_inc(&skb->users);
827 ireq->pktopts = skb;
828 }
829 }
830
tcp_v6_route_req(const struct sock * sk,struct sk_buff * skb,struct flowi * fl,struct request_sock * req)831 static struct dst_entry *tcp_v6_route_req(const struct sock *sk,
832 struct sk_buff *skb,
833 struct flowi *fl,
834 struct request_sock *req)
835 {
836 tcp_v6_init_req(req, sk, skb);
837
838 if (security_inet_conn_request(sk, skb, req))
839 return NULL;
840
841 return inet6_csk_route_req(sk, &fl->u.ip6, req, IPPROTO_TCP);
842 }
843
844 struct request_sock_ops tcp6_request_sock_ops __read_mostly = {
845 .family = AF_INET6,
846 .obj_size = sizeof(struct tcp6_request_sock),
847 .rtx_syn_ack = tcp_rtx_synack,
848 .send_ack = tcp_v6_reqsk_send_ack,
849 .destructor = tcp_v6_reqsk_destructor,
850 .send_reset = tcp_v6_send_reset,
851 .syn_ack_timeout = tcp_syn_ack_timeout,
852 };
853
854 const struct tcp_request_sock_ops tcp_request_sock_ipv6_ops = {
855 .mss_clamp = IPV6_MIN_MTU - sizeof(struct tcphdr) -
856 sizeof(struct ipv6hdr),
857 #ifdef CONFIG_TCP_MD5SIG
858 .req_md5_lookup = tcp_v6_md5_lookup,
859 .calc_md5_hash = tcp_v6_md5_hash_skb,
860 #endif
861 #ifdef CONFIG_SYN_COOKIES
862 .cookie_init_seq = cookie_v6_init_sequence,
863 #endif
864 .route_req = tcp_v6_route_req,
865 .init_seq = tcp_v6_init_seq,
866 .init_ts_off = tcp_v6_init_ts_off,
867 .send_synack = tcp_v6_send_synack,
868 };
869
tcp_v6_send_response(const struct sock * sk,struct sk_buff * skb,u32 seq,u32 ack,u32 win,u32 tsval,u32 tsecr,int oif,struct tcp_md5sig_key * key,int rst,u8 tclass,__be32 label,u32 priority)870 static void tcp_v6_send_response(const struct sock *sk, struct sk_buff *skb, u32 seq,
871 u32 ack, u32 win, u32 tsval, u32 tsecr,
872 int oif, struct tcp_md5sig_key *key, int rst,
873 u8 tclass, __be32 label, u32 priority)
874 {
875 const struct tcphdr *th = tcp_hdr(skb);
876 struct tcphdr *t1;
877 struct sk_buff *buff;
878 struct flowi6 fl6;
879 struct net *net = sk ? sock_net(sk) : dev_net(skb_dst(skb)->dev);
880 struct sock *ctl_sk = net->ipv6.tcp_sk;
881 unsigned int tot_len = sizeof(struct tcphdr);
882 __be32 mrst = 0, *topt;
883 struct dst_entry *dst;
884 __u32 mark = 0;
885
886 if (tsecr)
887 tot_len += TCPOLEN_TSTAMP_ALIGNED;
888 #ifdef CONFIG_TCP_MD5SIG
889 if (key)
890 tot_len += TCPOLEN_MD5SIG_ALIGNED;
891 #endif
892
893 #ifdef CONFIG_MPTCP
894 if (rst && !key) {
895 mrst = mptcp_reset_option(skb);
896
897 if (mrst)
898 tot_len += sizeof(__be32);
899 }
900 #endif
901
902 buff = alloc_skb(MAX_HEADER + sizeof(struct ipv6hdr) + tot_len,
903 GFP_ATOMIC);
904 if (!buff)
905 return;
906
907 skb_reserve(buff, MAX_HEADER + sizeof(struct ipv6hdr) + tot_len);
908
909 t1 = skb_push(buff, tot_len);
910 skb_reset_transport_header(buff);
911
912 /* Swap the send and the receive. */
913 memset(t1, 0, sizeof(*t1));
914 t1->dest = th->source;
915 t1->source = th->dest;
916 t1->doff = tot_len / 4;
917 t1->seq = htonl(seq);
918 t1->ack_seq = htonl(ack);
919 t1->ack = !rst || !th->ack;
920 t1->rst = rst;
921 t1->window = htons(win);
922
923 topt = (__be32 *)(t1 + 1);
924
925 if (tsecr) {
926 *topt++ = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
927 (TCPOPT_TIMESTAMP << 8) | TCPOLEN_TIMESTAMP);
928 *topt++ = htonl(tsval);
929 *topt++ = htonl(tsecr);
930 }
931
932 if (mrst)
933 *topt++ = mrst;
934
935 #ifdef CONFIG_TCP_MD5SIG
936 if (key) {
937 *topt++ = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
938 (TCPOPT_MD5SIG << 8) | TCPOLEN_MD5SIG);
939 tcp_v6_md5_hash_hdr((__u8 *)topt, key,
940 &ipv6_hdr(skb)->saddr,
941 &ipv6_hdr(skb)->daddr, t1);
942 }
943 #endif
944
945 memset(&fl6, 0, sizeof(fl6));
946 fl6.daddr = ipv6_hdr(skb)->saddr;
947 fl6.saddr = ipv6_hdr(skb)->daddr;
948 fl6.flowlabel = label;
949
950 buff->ip_summed = CHECKSUM_PARTIAL;
951 buff->csum = 0;
952
953 __tcp_v6_send_check(buff, &fl6.saddr, &fl6.daddr);
954
955 fl6.flowi6_proto = IPPROTO_TCP;
956 if (rt6_need_strict(&fl6.daddr) && !oif)
957 fl6.flowi6_oif = tcp_v6_iif(skb);
958 else {
959 if (!oif && netif_index_is_l3_master(net, skb->skb_iif))
960 oif = skb->skb_iif;
961
962 fl6.flowi6_oif = oif;
963 }
964
965 if (sk) {
966 if (sk->sk_state == TCP_TIME_WAIT) {
967 mark = inet_twsk(sk)->tw_mark;
968 /* autoflowlabel relies on buff->hash */
969 skb_set_hash(buff, inet_twsk(sk)->tw_txhash,
970 PKT_HASH_TYPE_L4);
971 } else {
972 mark = sk->sk_mark;
973 }
974 buff->tstamp = tcp_transmit_time(sk);
975 }
976 fl6.flowi6_mark = IP6_REPLY_MARK(net, skb->mark) ?: mark;
977 fl6.fl6_dport = t1->dest;
978 fl6.fl6_sport = t1->source;
979 fl6.flowi6_uid = sock_net_uid(net, sk && sk_fullsock(sk) ? sk : NULL);
980 security_skb_classify_flow(skb, flowi6_to_flowi_common(&fl6));
981
982 /* Pass a socket to ip6_dst_lookup either it is for RST
983 * Underlying function will use this to retrieve the network
984 * namespace
985 */
986 dst = ip6_dst_lookup_flow(sock_net(ctl_sk), ctl_sk, &fl6, NULL);
987 if (!IS_ERR(dst)) {
988 skb_dst_set(buff, dst);
989 ip6_xmit(ctl_sk, buff, &fl6, fl6.flowi6_mark, NULL,
990 tclass & ~INET_ECN_MASK, priority);
991 TCP_INC_STATS(net, TCP_MIB_OUTSEGS);
992 if (rst)
993 TCP_INC_STATS(net, TCP_MIB_OUTRSTS);
994 return;
995 }
996
997 kfree_skb(buff);
998 }
999
tcp_v6_send_reset(const struct sock * sk,struct sk_buff * skb)1000 static void tcp_v6_send_reset(const struct sock *sk, struct sk_buff *skb)
1001 {
1002 const struct tcphdr *th = tcp_hdr(skb);
1003 struct ipv6hdr *ipv6h = ipv6_hdr(skb);
1004 u32 seq = 0, ack_seq = 0;
1005 struct tcp_md5sig_key *key = NULL;
1006 #ifdef CONFIG_TCP_MD5SIG
1007 const __u8 *hash_location = NULL;
1008 unsigned char newhash[16];
1009 int genhash;
1010 struct sock *sk1 = NULL;
1011 #endif
1012 __be32 label = 0;
1013 u32 priority = 0;
1014 struct net *net;
1015 int oif = 0;
1016
1017 if (th->rst)
1018 return;
1019
1020 /* If sk not NULL, it means we did a successful lookup and incoming
1021 * route had to be correct. prequeue might have dropped our dst.
1022 */
1023 if (!sk && !ipv6_unicast_destination(skb))
1024 return;
1025
1026 net = sk ? sock_net(sk) : dev_net(skb_dst(skb)->dev);
1027 #ifdef CONFIG_TCP_MD5SIG
1028 rcu_read_lock();
1029 hash_location = tcp_parse_md5sig_option(th);
1030 if (sk && sk_fullsock(sk)) {
1031 int l3index;
1032
1033 /* sdif set, means packet ingressed via a device
1034 * in an L3 domain and inet_iif is set to it.
1035 */
1036 l3index = tcp_v6_sdif(skb) ? tcp_v6_iif_l3_slave(skb) : 0;
1037 key = tcp_v6_md5_do_lookup(sk, &ipv6h->saddr, l3index);
1038 } else if (hash_location) {
1039 int dif = tcp_v6_iif_l3_slave(skb);
1040 int sdif = tcp_v6_sdif(skb);
1041 int l3index;
1042
1043 /*
1044 * active side is lost. Try to find listening socket through
1045 * source port, and then find md5 key through listening socket.
1046 * we are not loose security here:
1047 * Incoming packet is checked with md5 hash with finding key,
1048 * no RST generated if md5 hash doesn't match.
1049 */
1050 sk1 = inet6_lookup_listener(net,
1051 &tcp_hashinfo, NULL, 0,
1052 &ipv6h->saddr,
1053 th->source, &ipv6h->daddr,
1054 ntohs(th->source), dif, sdif);
1055 if (!sk1)
1056 goto out;
1057
1058 /* sdif set, means packet ingressed via a device
1059 * in an L3 domain and dif is set to it.
1060 */
1061 l3index = tcp_v6_sdif(skb) ? dif : 0;
1062
1063 key = tcp_v6_md5_do_lookup(sk1, &ipv6h->saddr, l3index);
1064 if (!key)
1065 goto out;
1066
1067 genhash = tcp_v6_md5_hash_skb(newhash, key, NULL, skb);
1068 if (genhash || memcmp(hash_location, newhash, 16) != 0)
1069 goto out;
1070 }
1071 #endif
1072
1073 if (th->ack)
1074 seq = ntohl(th->ack_seq);
1075 else
1076 ack_seq = ntohl(th->seq) + th->syn + th->fin + skb->len -
1077 (th->doff << 2);
1078
1079 if (sk) {
1080 oif = sk->sk_bound_dev_if;
1081 if (sk_fullsock(sk)) {
1082 const struct ipv6_pinfo *np = tcp_inet6_sk(sk);
1083
1084 trace_tcp_send_reset(sk, skb);
1085 if (np->repflow)
1086 label = ip6_flowlabel(ipv6h);
1087 priority = sk->sk_priority;
1088 }
1089 if (sk->sk_state == TCP_TIME_WAIT) {
1090 label = cpu_to_be32(inet_twsk(sk)->tw_flowlabel);
1091 priority = inet_twsk(sk)->tw_priority;
1092 }
1093 } else {
1094 if (net->ipv6.sysctl.flowlabel_reflect & FLOWLABEL_REFLECT_TCP_RESET)
1095 label = ip6_flowlabel(ipv6h);
1096 }
1097
1098 tcp_v6_send_response(sk, skb, seq, ack_seq, 0, 0, 0, oif, key, 1,
1099 ipv6_get_dsfield(ipv6h), label, priority);
1100
1101 #ifdef CONFIG_TCP_MD5SIG
1102 out:
1103 rcu_read_unlock();
1104 #endif
1105 }
1106
tcp_v6_send_ack(const struct sock * sk,struct sk_buff * skb,u32 seq,u32 ack,u32 win,u32 tsval,u32 tsecr,int oif,struct tcp_md5sig_key * key,u8 tclass,__be32 label,u32 priority)1107 static void tcp_v6_send_ack(const struct sock *sk, struct sk_buff *skb, u32 seq,
1108 u32 ack, u32 win, u32 tsval, u32 tsecr, int oif,
1109 struct tcp_md5sig_key *key, u8 tclass,
1110 __be32 label, u32 priority)
1111 {
1112 tcp_v6_send_response(sk, skb, seq, ack, win, tsval, tsecr, oif, key, 0,
1113 tclass, label, priority);
1114 }
1115
tcp_v6_timewait_ack(struct sock * sk,struct sk_buff * skb)1116 static void tcp_v6_timewait_ack(struct sock *sk, struct sk_buff *skb)
1117 {
1118 struct inet_timewait_sock *tw = inet_twsk(sk);
1119 struct tcp_timewait_sock *tcptw = tcp_twsk(sk);
1120
1121 tcp_v6_send_ack(sk, skb, tcptw->tw_snd_nxt, tcptw->tw_rcv_nxt,
1122 tcptw->tw_rcv_wnd >> tw->tw_rcv_wscale,
1123 tcp_time_stamp_raw() + tcptw->tw_ts_offset,
1124 tcptw->tw_ts_recent, tw->tw_bound_dev_if, tcp_twsk_md5_key(tcptw),
1125 tw->tw_tclass, cpu_to_be32(tw->tw_flowlabel), tw->tw_priority);
1126
1127 inet_twsk_put(tw);
1128 }
1129
tcp_v6_reqsk_send_ack(const struct sock * sk,struct sk_buff * skb,struct request_sock * req)1130 static void tcp_v6_reqsk_send_ack(const struct sock *sk, struct sk_buff *skb,
1131 struct request_sock *req)
1132 {
1133 int l3index;
1134
1135 l3index = tcp_v6_sdif(skb) ? tcp_v6_iif_l3_slave(skb) : 0;
1136
1137 /* sk->sk_state == TCP_LISTEN -> for regular TCP_SYN_RECV
1138 * sk->sk_state == TCP_SYN_RECV -> for Fast Open.
1139 */
1140 /* RFC 7323 2.3
1141 * The window field (SEG.WND) of every outgoing segment, with the
1142 * exception of <SYN> segments, MUST be right-shifted by
1143 * Rcv.Wind.Shift bits:
1144 */
1145 tcp_v6_send_ack(sk, skb, (sk->sk_state == TCP_LISTEN) ?
1146 tcp_rsk(req)->snt_isn + 1 : tcp_sk(sk)->snd_nxt,
1147 tcp_rsk(req)->rcv_nxt,
1148 req->rsk_rcv_wnd >> inet_rsk(req)->rcv_wscale,
1149 tcp_time_stamp_raw() + tcp_rsk(req)->ts_off,
1150 req->ts_recent, sk->sk_bound_dev_if,
1151 tcp_v6_md5_do_lookup(sk, &ipv6_hdr(skb)->saddr, l3index),
1152 ipv6_get_dsfield(ipv6_hdr(skb)), 0, sk->sk_priority);
1153 }
1154
1155
tcp_v6_cookie_check(struct sock * sk,struct sk_buff * skb)1156 static struct sock *tcp_v6_cookie_check(struct sock *sk, struct sk_buff *skb)
1157 {
1158 #ifdef CONFIG_SYN_COOKIES
1159 const struct tcphdr *th = tcp_hdr(skb);
1160
1161 if (!th->syn)
1162 sk = cookie_v6_check(sk, skb);
1163 #endif
1164 return sk;
1165 }
1166
tcp_v6_get_syncookie(struct sock * sk,struct ipv6hdr * iph,struct tcphdr * th,u32 * cookie)1167 u16 tcp_v6_get_syncookie(struct sock *sk, struct ipv6hdr *iph,
1168 struct tcphdr *th, u32 *cookie)
1169 {
1170 u16 mss = 0;
1171 #ifdef CONFIG_SYN_COOKIES
1172 mss = tcp_get_syncookie_mss(&tcp6_request_sock_ops,
1173 &tcp_request_sock_ipv6_ops, sk, th);
1174 if (mss) {
1175 *cookie = __cookie_v6_init_sequence(iph, th, &mss);
1176 tcp_synq_overflow(sk);
1177 }
1178 #endif
1179 return mss;
1180 }
1181
tcp_v6_conn_request(struct sock * sk,struct sk_buff * skb)1182 static int tcp_v6_conn_request(struct sock *sk, struct sk_buff *skb)
1183 {
1184 if (skb->protocol == htons(ETH_P_IP))
1185 return tcp_v4_conn_request(sk, skb);
1186
1187 if (!ipv6_unicast_destination(skb))
1188 goto drop;
1189
1190 if (ipv6_addr_v4mapped(&ipv6_hdr(skb)->saddr)) {
1191 __IP6_INC_STATS(sock_net(sk), NULL, IPSTATS_MIB_INHDRERRORS);
1192 return 0;
1193 }
1194
1195 return tcp_conn_request(&tcp6_request_sock_ops,
1196 &tcp_request_sock_ipv6_ops, sk, skb);
1197
1198 drop:
1199 tcp_listendrop(sk);
1200 return 0; /* don't send reset */
1201 }
1202
tcp_v6_restore_cb(struct sk_buff * skb)1203 static void tcp_v6_restore_cb(struct sk_buff *skb)
1204 {
1205 /* We need to move header back to the beginning if xfrm6_policy_check()
1206 * and tcp_v6_fill_cb() are going to be called again.
1207 * ip6_datagram_recv_specific_ctl() also expects IP6CB to be there.
1208 */
1209 memmove(IP6CB(skb), &TCP_SKB_CB(skb)->header.h6,
1210 sizeof(struct inet6_skb_parm));
1211 }
1212
tcp_v6_syn_recv_sock(const struct sock * sk,struct sk_buff * skb,struct request_sock * req,struct dst_entry * dst,struct request_sock * req_unhash,bool * own_req)1213 static struct sock *tcp_v6_syn_recv_sock(const struct sock *sk, struct sk_buff *skb,
1214 struct request_sock *req,
1215 struct dst_entry *dst,
1216 struct request_sock *req_unhash,
1217 bool *own_req)
1218 {
1219 struct inet_request_sock *ireq;
1220 struct ipv6_pinfo *newnp;
1221 const struct ipv6_pinfo *np = tcp_inet6_sk(sk);
1222 struct ipv6_txoptions *opt;
1223 struct inet_sock *newinet;
1224 bool found_dup_sk = false;
1225 struct tcp_sock *newtp;
1226 struct sock *newsk;
1227 #ifdef CONFIG_TCP_MD5SIG
1228 struct tcp_md5sig_key *key;
1229 int l3index;
1230 #endif
1231 struct flowi6 fl6;
1232
1233 if (skb->protocol == htons(ETH_P_IP)) {
1234 /*
1235 * v6 mapped
1236 */
1237
1238 newsk = tcp_v4_syn_recv_sock(sk, skb, req, dst,
1239 req_unhash, own_req);
1240
1241 if (!newsk)
1242 return NULL;
1243
1244 inet_sk(newsk)->pinet6 = tcp_inet6_sk(newsk);
1245
1246 newinet = inet_sk(newsk);
1247 newnp = tcp_inet6_sk(newsk);
1248 newtp = tcp_sk(newsk);
1249
1250 memcpy(newnp, np, sizeof(struct ipv6_pinfo));
1251
1252 newnp->saddr = newsk->sk_v6_rcv_saddr;
1253
1254 inet_csk(newsk)->icsk_af_ops = &ipv6_mapped;
1255 if (sk_is_mptcp(newsk))
1256 mptcpv6_handle_mapped(newsk, true);
1257 newsk->sk_backlog_rcv = tcp_v4_do_rcv;
1258 #ifdef CONFIG_TCP_MD5SIG
1259 newtp->af_specific = &tcp_sock_ipv6_mapped_specific;
1260 #endif
1261
1262 newnp->ipv6_mc_list = NULL;
1263 newnp->ipv6_ac_list = NULL;
1264 newnp->ipv6_fl_list = NULL;
1265 newnp->pktoptions = NULL;
1266 newnp->opt = NULL;
1267 newnp->mcast_oif = inet_iif(skb);
1268 newnp->mcast_hops = ip_hdr(skb)->ttl;
1269 newnp->rcv_flowinfo = 0;
1270 if (np->repflow)
1271 newnp->flow_label = 0;
1272
1273 /*
1274 * No need to charge this sock to the relevant IPv6 refcnt debug socks count
1275 * here, tcp_create_openreq_child now does this for us, see the comment in
1276 * that function for the gory details. -acme
1277 */
1278
1279 /* It is tricky place. Until this moment IPv4 tcp
1280 worked with IPv6 icsk.icsk_af_ops.
1281 Sync it now.
1282 */
1283 tcp_sync_mss(newsk, inet_csk(newsk)->icsk_pmtu_cookie);
1284
1285 return newsk;
1286 }
1287
1288 ireq = inet_rsk(req);
1289
1290 if (sk_acceptq_is_full(sk))
1291 goto out_overflow;
1292
1293 if (!dst) {
1294 dst = inet6_csk_route_req(sk, &fl6, req, IPPROTO_TCP);
1295 if (!dst)
1296 goto out;
1297 }
1298
1299 newsk = tcp_create_openreq_child(sk, req, skb);
1300 if (!newsk)
1301 goto out_nonewsk;
1302
1303 /*
1304 * No need to charge this sock to the relevant IPv6 refcnt debug socks
1305 * count here, tcp_create_openreq_child now does this for us, see the
1306 * comment in that function for the gory details. -acme
1307 */
1308
1309 newsk->sk_gso_type = SKB_GSO_TCPV6;
1310 ip6_dst_store(newsk, dst, NULL, NULL);
1311 inet6_sk_rx_dst_set(newsk, skb);
1312
1313 inet_sk(newsk)->pinet6 = tcp_inet6_sk(newsk);
1314
1315 newtp = tcp_sk(newsk);
1316 newinet = inet_sk(newsk);
1317 newnp = tcp_inet6_sk(newsk);
1318
1319 memcpy(newnp, np, sizeof(struct ipv6_pinfo));
1320
1321 newsk->sk_v6_daddr = ireq->ir_v6_rmt_addr;
1322 newnp->saddr = ireq->ir_v6_loc_addr;
1323 newsk->sk_v6_rcv_saddr = ireq->ir_v6_loc_addr;
1324 newsk->sk_bound_dev_if = ireq->ir_iif;
1325
1326 /* Now IPv6 options...
1327
1328 First: no IPv4 options.
1329 */
1330 newinet->inet_opt = NULL;
1331 newnp->ipv6_mc_list = NULL;
1332 newnp->ipv6_ac_list = NULL;
1333 newnp->ipv6_fl_list = NULL;
1334
1335 /* Clone RX bits */
1336 newnp->rxopt.all = np->rxopt.all;
1337
1338 newnp->pktoptions = NULL;
1339 newnp->opt = NULL;
1340 newnp->mcast_oif = tcp_v6_iif(skb);
1341 newnp->mcast_hops = ipv6_hdr(skb)->hop_limit;
1342 newnp->rcv_flowinfo = ip6_flowinfo(ipv6_hdr(skb));
1343 if (np->repflow)
1344 newnp->flow_label = ip6_flowlabel(ipv6_hdr(skb));
1345
1346 /* Set ToS of the new socket based upon the value of incoming SYN.
1347 * ECT bits are set later in tcp_init_transfer().
1348 */
1349 if (sock_net(sk)->ipv4.sysctl_tcp_reflect_tos)
1350 newnp->tclass = tcp_rsk(req)->syn_tos & ~INET_ECN_MASK;
1351
1352 /* Clone native IPv6 options from listening socket (if any)
1353
1354 Yes, keeping reference count would be much more clever,
1355 but we make one more one thing there: reattach optmem
1356 to newsk.
1357 */
1358 opt = ireq->ipv6_opt;
1359 if (!opt)
1360 opt = rcu_dereference(np->opt);
1361 if (opt) {
1362 opt = ipv6_dup_options(newsk, opt);
1363 RCU_INIT_POINTER(newnp->opt, opt);
1364 }
1365 inet_csk(newsk)->icsk_ext_hdr_len = 0;
1366 if (opt)
1367 inet_csk(newsk)->icsk_ext_hdr_len = opt->opt_nflen +
1368 opt->opt_flen;
1369
1370 tcp_ca_openreq_child(newsk, dst);
1371
1372 tcp_sync_mss(newsk, dst_mtu(dst));
1373 newtp->advmss = tcp_mss_clamp(tcp_sk(sk), dst_metric_advmss(dst));
1374
1375 tcp_initialize_rcv_mss(newsk);
1376
1377 newinet->inet_daddr = newinet->inet_saddr = LOOPBACK4_IPV6;
1378 newinet->inet_rcv_saddr = LOOPBACK4_IPV6;
1379
1380 #ifdef CONFIG_TCP_MD5SIG
1381 l3index = l3mdev_master_ifindex_by_index(sock_net(sk), ireq->ir_iif);
1382
1383 /* Copy over the MD5 key from the original socket */
1384 key = tcp_v6_md5_do_lookup(sk, &newsk->sk_v6_daddr, l3index);
1385 if (key) {
1386 /* We're using one, so create a matching key
1387 * on the newsk structure. If we fail to get
1388 * memory, then we end up not copying the key
1389 * across. Shucks.
1390 */
1391 tcp_md5_do_add(newsk, (union tcp_md5_addr *)&newsk->sk_v6_daddr,
1392 AF_INET6, 128, l3index, key->key, key->keylen,
1393 sk_gfp_mask(sk, GFP_ATOMIC));
1394 }
1395 #endif
1396
1397 if (__inet_inherit_port(sk, newsk) < 0) {
1398 inet_csk_prepare_forced_close(newsk);
1399 tcp_done(newsk);
1400 goto out;
1401 }
1402 *own_req = inet_ehash_nolisten(newsk, req_to_sk(req_unhash),
1403 &found_dup_sk);
1404 if (*own_req) {
1405 tcp_move_syn(newtp, req);
1406
1407 /* Clone pktoptions received with SYN, if we own the req */
1408 if (ireq->pktopts) {
1409 newnp->pktoptions = skb_clone(ireq->pktopts,
1410 sk_gfp_mask(sk, GFP_ATOMIC));
1411 consume_skb(ireq->pktopts);
1412 ireq->pktopts = NULL;
1413 if (newnp->pktoptions) {
1414 tcp_v6_restore_cb(newnp->pktoptions);
1415 skb_set_owner_r(newnp->pktoptions, newsk);
1416 }
1417 }
1418 } else {
1419 if (!req_unhash && found_dup_sk) {
1420 /* This code path should only be executed in the
1421 * syncookie case only
1422 */
1423 bh_unlock_sock(newsk);
1424 sock_put(newsk);
1425 newsk = NULL;
1426 }
1427 }
1428
1429 return newsk;
1430
1431 out_overflow:
1432 __NET_INC_STATS(sock_net(sk), LINUX_MIB_LISTENOVERFLOWS);
1433 out_nonewsk:
1434 dst_release(dst);
1435 out:
1436 tcp_listendrop(sk);
1437 return NULL;
1438 }
1439
1440 INDIRECT_CALLABLE_DECLARE(struct dst_entry *ipv4_dst_check(struct dst_entry *,
1441 u32));
1442 /* The socket must have it's spinlock held when we get
1443 * here, unless it is a TCP_LISTEN socket.
1444 *
1445 * We have a potential double-lock case here, so even when
1446 * doing backlog processing we use the BH locking scheme.
1447 * This is because we cannot sleep with the original spinlock
1448 * held.
1449 */
tcp_v6_do_rcv(struct sock * sk,struct sk_buff * skb)1450 static int tcp_v6_do_rcv(struct sock *sk, struct sk_buff *skb)
1451 {
1452 struct ipv6_pinfo *np = tcp_inet6_sk(sk);
1453 struct sk_buff *opt_skb = NULL;
1454 struct tcp_sock *tp;
1455
1456 /* Imagine: socket is IPv6. IPv4 packet arrives,
1457 goes to IPv4 receive handler and backlogged.
1458 From backlog it always goes here. Kerboom...
1459 Fortunately, tcp_rcv_established and rcv_established
1460 handle them correctly, but it is not case with
1461 tcp_v6_hnd_req and tcp_v6_send_reset(). --ANK
1462 */
1463
1464 if (skb->protocol == htons(ETH_P_IP))
1465 return tcp_v4_do_rcv(sk, skb);
1466
1467 /*
1468 * socket locking is here for SMP purposes as backlog rcv
1469 * is currently called with bh processing disabled.
1470 */
1471
1472 /* Do Stevens' IPV6_PKTOPTIONS.
1473
1474 Yes, guys, it is the only place in our code, where we
1475 may make it not affecting IPv4.
1476 The rest of code is protocol independent,
1477 and I do not like idea to uglify IPv4.
1478
1479 Actually, all the idea behind IPV6_PKTOPTIONS
1480 looks not very well thought. For now we latch
1481 options, received in the last packet, enqueued
1482 by tcp. Feel free to propose better solution.
1483 --ANK (980728)
1484 */
1485 if (np->rxopt.all)
1486 opt_skb = skb_clone(skb, sk_gfp_mask(sk, GFP_ATOMIC));
1487
1488 if (sk->sk_state == TCP_ESTABLISHED) { /* Fast path */
1489 struct dst_entry *dst = sk->sk_rx_dst;
1490
1491 sock_rps_save_rxhash(sk, skb);
1492 sk_mark_napi_id(sk, skb);
1493 if (dst) {
1494 if (inet_sk(sk)->rx_dst_ifindex != skb->skb_iif ||
1495 INDIRECT_CALL_1(dst->ops->check, ip6_dst_check,
1496 dst, np->rx_dst_cookie) == NULL) {
1497 dst_release(dst);
1498 sk->sk_rx_dst = NULL;
1499 }
1500 }
1501
1502 tcp_rcv_established(sk, skb);
1503 if (opt_skb)
1504 goto ipv6_pktoptions;
1505 return 0;
1506 }
1507
1508 if (tcp_checksum_complete(skb))
1509 goto csum_err;
1510
1511 if (sk->sk_state == TCP_LISTEN) {
1512 struct sock *nsk = tcp_v6_cookie_check(sk, skb);
1513
1514 if (!nsk)
1515 goto discard;
1516
1517 if (nsk != sk) {
1518 if (tcp_child_process(sk, nsk, skb))
1519 goto reset;
1520 if (opt_skb)
1521 __kfree_skb(opt_skb);
1522 return 0;
1523 }
1524 } else
1525 sock_rps_save_rxhash(sk, skb);
1526
1527 if (tcp_rcv_state_process(sk, skb))
1528 goto reset;
1529 if (opt_skb)
1530 goto ipv6_pktoptions;
1531 return 0;
1532
1533 reset:
1534 tcp_v6_send_reset(sk, skb);
1535 discard:
1536 if (opt_skb)
1537 __kfree_skb(opt_skb);
1538 kfree_skb(skb);
1539 return 0;
1540 csum_err:
1541 TCP_INC_STATS(sock_net(sk), TCP_MIB_CSUMERRORS);
1542 TCP_INC_STATS(sock_net(sk), TCP_MIB_INERRS);
1543 goto discard;
1544
1545
1546 ipv6_pktoptions:
1547 /* Do you ask, what is it?
1548
1549 1. skb was enqueued by tcp.
1550 2. skb is added to tail of read queue, rather than out of order.
1551 3. socket is not in passive state.
1552 4. Finally, it really contains options, which user wants to receive.
1553 */
1554 tp = tcp_sk(sk);
1555 if (TCP_SKB_CB(opt_skb)->end_seq == tp->rcv_nxt &&
1556 !((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_LISTEN))) {
1557 if (np->rxopt.bits.rxinfo || np->rxopt.bits.rxoinfo)
1558 np->mcast_oif = tcp_v6_iif(opt_skb);
1559 if (np->rxopt.bits.rxhlim || np->rxopt.bits.rxohlim)
1560 np->mcast_hops = ipv6_hdr(opt_skb)->hop_limit;
1561 if (np->rxopt.bits.rxflow || np->rxopt.bits.rxtclass)
1562 np->rcv_flowinfo = ip6_flowinfo(ipv6_hdr(opt_skb));
1563 if (np->repflow)
1564 np->flow_label = ip6_flowlabel(ipv6_hdr(opt_skb));
1565 if (ipv6_opt_accepted(sk, opt_skb, &TCP_SKB_CB(opt_skb)->header.h6)) {
1566 skb_set_owner_r(opt_skb, sk);
1567 tcp_v6_restore_cb(opt_skb);
1568 opt_skb = xchg(&np->pktoptions, opt_skb);
1569 } else {
1570 __kfree_skb(opt_skb);
1571 opt_skb = xchg(&np->pktoptions, NULL);
1572 }
1573 }
1574
1575 kfree_skb(opt_skb);
1576 return 0;
1577 }
1578
tcp_v6_fill_cb(struct sk_buff * skb,const struct ipv6hdr * hdr,const struct tcphdr * th)1579 static void tcp_v6_fill_cb(struct sk_buff *skb, const struct ipv6hdr *hdr,
1580 const struct tcphdr *th)
1581 {
1582 /* This is tricky: we move IP6CB at its correct location into
1583 * TCP_SKB_CB(). It must be done after xfrm6_policy_check(), because
1584 * _decode_session6() uses IP6CB().
1585 * barrier() makes sure compiler won't play aliasing games.
1586 */
1587 memmove(&TCP_SKB_CB(skb)->header.h6, IP6CB(skb),
1588 sizeof(struct inet6_skb_parm));
1589 barrier();
1590
1591 TCP_SKB_CB(skb)->seq = ntohl(th->seq);
1592 TCP_SKB_CB(skb)->end_seq = (TCP_SKB_CB(skb)->seq + th->syn + th->fin +
1593 skb->len - th->doff*4);
1594 TCP_SKB_CB(skb)->ack_seq = ntohl(th->ack_seq);
1595 TCP_SKB_CB(skb)->tcp_flags = tcp_flag_byte(th);
1596 TCP_SKB_CB(skb)->tcp_tw_isn = 0;
1597 TCP_SKB_CB(skb)->ip_dsfield = ipv6_get_dsfield(hdr);
1598 TCP_SKB_CB(skb)->sacked = 0;
1599 TCP_SKB_CB(skb)->has_rxtstamp =
1600 skb->tstamp || skb_hwtstamps(skb)->hwtstamp;
1601 }
1602
tcp_v6_rcv(struct sk_buff * skb)1603 INDIRECT_CALLABLE_SCOPE int tcp_v6_rcv(struct sk_buff *skb)
1604 {
1605 struct sk_buff *skb_to_free;
1606 int sdif = inet6_sdif(skb);
1607 int dif = inet6_iif(skb);
1608 const struct tcphdr *th;
1609 const struct ipv6hdr *hdr;
1610 bool refcounted;
1611 struct sock *sk;
1612 int ret;
1613 struct net *net = dev_net(skb->dev);
1614
1615 if (skb->pkt_type != PACKET_HOST)
1616 goto discard_it;
1617
1618 /*
1619 * Count it even if it's bad.
1620 */
1621 __TCP_INC_STATS(net, TCP_MIB_INSEGS);
1622
1623 if (!pskb_may_pull(skb, sizeof(struct tcphdr)))
1624 goto discard_it;
1625
1626 th = (const struct tcphdr *)skb->data;
1627
1628 if (unlikely(th->doff < sizeof(struct tcphdr)/4))
1629 goto bad_packet;
1630 if (!pskb_may_pull(skb, th->doff*4))
1631 goto discard_it;
1632
1633 if (skb_checksum_init(skb, IPPROTO_TCP, ip6_compute_pseudo))
1634 goto csum_error;
1635
1636 th = (const struct tcphdr *)skb->data;
1637 hdr = ipv6_hdr(skb);
1638
1639 lookup:
1640 sk = __inet6_lookup_skb(&tcp_hashinfo, skb, __tcp_hdrlen(th),
1641 th->source, th->dest, inet6_iif(skb), sdif,
1642 &refcounted);
1643 if (!sk)
1644 goto no_tcp_socket;
1645
1646 process:
1647 if (sk->sk_state == TCP_TIME_WAIT)
1648 goto do_time_wait;
1649
1650 if (sk->sk_state == TCP_NEW_SYN_RECV) {
1651 struct request_sock *req = inet_reqsk(sk);
1652 bool req_stolen = false;
1653 struct sock *nsk;
1654
1655 sk = req->rsk_listener;
1656 if (tcp_v6_inbound_md5_hash(sk, skb, dif, sdif)) {
1657 sk_drops_add(sk, skb);
1658 reqsk_put(req);
1659 goto discard_it;
1660 }
1661 if (tcp_checksum_complete(skb)) {
1662 reqsk_put(req);
1663 goto csum_error;
1664 }
1665 if (unlikely(sk->sk_state != TCP_LISTEN)) {
1666 inet_csk_reqsk_queue_drop_and_put(sk, req);
1667 goto lookup;
1668 }
1669 sock_hold(sk);
1670 refcounted = true;
1671 nsk = NULL;
1672 if (!tcp_filter(sk, skb)) {
1673 th = (const struct tcphdr *)skb->data;
1674 hdr = ipv6_hdr(skb);
1675 tcp_v6_fill_cb(skb, hdr, th);
1676 nsk = tcp_check_req(sk, skb, req, false, &req_stolen);
1677 }
1678 if (!nsk) {
1679 reqsk_put(req);
1680 if (req_stolen) {
1681 /* Another cpu got exclusive access to req
1682 * and created a full blown socket.
1683 * Try to feed this packet to this socket
1684 * instead of discarding it.
1685 */
1686 tcp_v6_restore_cb(skb);
1687 sock_put(sk);
1688 goto lookup;
1689 }
1690 goto discard_and_relse;
1691 }
1692 if (nsk == sk) {
1693 reqsk_put(req);
1694 tcp_v6_restore_cb(skb);
1695 } else if (tcp_child_process(sk, nsk, skb)) {
1696 tcp_v6_send_reset(nsk, skb);
1697 goto discard_and_relse;
1698 } else {
1699 sock_put(sk);
1700 return 0;
1701 }
1702 }
1703 if (hdr->hop_limit < tcp_inet6_sk(sk)->min_hopcount) {
1704 __NET_INC_STATS(net, LINUX_MIB_TCPMINTTLDROP);
1705 goto discard_and_relse;
1706 }
1707
1708 if (!xfrm6_policy_check(sk, XFRM_POLICY_IN, skb))
1709 goto discard_and_relse;
1710
1711 if (tcp_v6_inbound_md5_hash(sk, skb, dif, sdif))
1712 goto discard_and_relse;
1713
1714 if (tcp_filter(sk, skb))
1715 goto discard_and_relse;
1716 th = (const struct tcphdr *)skb->data;
1717 hdr = ipv6_hdr(skb);
1718 tcp_v6_fill_cb(skb, hdr, th);
1719
1720 skb->dev = NULL;
1721
1722 if (sk->sk_state == TCP_LISTEN) {
1723 ret = tcp_v6_do_rcv(sk, skb);
1724 goto put_and_return;
1725 }
1726
1727 sk_incoming_cpu_update(sk);
1728
1729 bh_lock_sock_nested(sk);
1730 tcp_segs_in(tcp_sk(sk), skb);
1731 ret = 0;
1732 if (!sock_owned_by_user(sk)) {
1733 skb_to_free = sk->sk_rx_skb_cache;
1734 sk->sk_rx_skb_cache = NULL;
1735 ret = tcp_v6_do_rcv(sk, skb);
1736 } else {
1737 if (tcp_add_backlog(sk, skb))
1738 goto discard_and_relse;
1739 skb_to_free = NULL;
1740 }
1741 bh_unlock_sock(sk);
1742 if (skb_to_free)
1743 __kfree_skb(skb_to_free);
1744 put_and_return:
1745 if (refcounted)
1746 sock_put(sk);
1747 return ret ? -1 : 0;
1748
1749 no_tcp_socket:
1750 if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb))
1751 goto discard_it;
1752
1753 tcp_v6_fill_cb(skb, hdr, th);
1754
1755 if (tcp_checksum_complete(skb)) {
1756 csum_error:
1757 __TCP_INC_STATS(net, TCP_MIB_CSUMERRORS);
1758 bad_packet:
1759 __TCP_INC_STATS(net, TCP_MIB_INERRS);
1760 } else {
1761 tcp_v6_send_reset(NULL, skb);
1762 }
1763
1764 discard_it:
1765 kfree_skb(skb);
1766 return 0;
1767
1768 discard_and_relse:
1769 sk_drops_add(sk, skb);
1770 if (refcounted)
1771 sock_put(sk);
1772 goto discard_it;
1773
1774 do_time_wait:
1775 if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb)) {
1776 inet_twsk_put(inet_twsk(sk));
1777 goto discard_it;
1778 }
1779
1780 tcp_v6_fill_cb(skb, hdr, th);
1781
1782 if (tcp_checksum_complete(skb)) {
1783 inet_twsk_put(inet_twsk(sk));
1784 goto csum_error;
1785 }
1786
1787 switch (tcp_timewait_state_process(inet_twsk(sk), skb, th)) {
1788 case TCP_TW_SYN:
1789 {
1790 struct sock *sk2;
1791
1792 sk2 = inet6_lookup_listener(dev_net(skb->dev), &tcp_hashinfo,
1793 skb, __tcp_hdrlen(th),
1794 &ipv6_hdr(skb)->saddr, th->source,
1795 &ipv6_hdr(skb)->daddr,
1796 ntohs(th->dest),
1797 tcp_v6_iif_l3_slave(skb),
1798 sdif);
1799 if (sk2) {
1800 struct inet_timewait_sock *tw = inet_twsk(sk);
1801 inet_twsk_deschedule_put(tw);
1802 sk = sk2;
1803 tcp_v6_restore_cb(skb);
1804 refcounted = false;
1805 goto process;
1806 }
1807 }
1808 /* to ACK */
1809 fallthrough;
1810 case TCP_TW_ACK:
1811 tcp_v6_timewait_ack(sk, skb);
1812 break;
1813 case TCP_TW_RST:
1814 tcp_v6_send_reset(sk, skb);
1815 inet_twsk_deschedule_put(inet_twsk(sk));
1816 goto discard_it;
1817 case TCP_TW_SUCCESS:
1818 ;
1819 }
1820 goto discard_it;
1821 }
1822
tcp_v6_early_demux(struct sk_buff * skb)1823 INDIRECT_CALLABLE_SCOPE void tcp_v6_early_demux(struct sk_buff *skb)
1824 {
1825 const struct ipv6hdr *hdr;
1826 const struct tcphdr *th;
1827 struct sock *sk;
1828
1829 if (skb->pkt_type != PACKET_HOST)
1830 return;
1831
1832 if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct tcphdr)))
1833 return;
1834
1835 hdr = ipv6_hdr(skb);
1836 th = tcp_hdr(skb);
1837
1838 if (th->doff < sizeof(struct tcphdr) / 4)
1839 return;
1840
1841 /* Note : We use inet6_iif() here, not tcp_v6_iif() */
1842 sk = __inet6_lookup_established(dev_net(skb->dev), &tcp_hashinfo,
1843 &hdr->saddr, th->source,
1844 &hdr->daddr, ntohs(th->dest),
1845 inet6_iif(skb), inet6_sdif(skb));
1846 if (sk) {
1847 skb->sk = sk;
1848 skb->destructor = sock_edemux;
1849 if (sk_fullsock(sk)) {
1850 struct dst_entry *dst = READ_ONCE(sk->sk_rx_dst);
1851
1852 if (dst)
1853 dst = dst_check(dst, tcp_inet6_sk(sk)->rx_dst_cookie);
1854 if (dst &&
1855 inet_sk(sk)->rx_dst_ifindex == skb->skb_iif)
1856 skb_dst_set_noref(skb, dst);
1857 }
1858 }
1859 }
1860
1861 static struct timewait_sock_ops tcp6_timewait_sock_ops = {
1862 .twsk_obj_size = sizeof(struct tcp6_timewait_sock),
1863 .twsk_unique = tcp_twsk_unique,
1864 .twsk_destructor = tcp_twsk_destructor,
1865 };
1866
tcp_v6_send_check(struct sock * sk,struct sk_buff * skb)1867 INDIRECT_CALLABLE_SCOPE void tcp_v6_send_check(struct sock *sk, struct sk_buff *skb)
1868 {
1869 struct ipv6_pinfo *np = inet6_sk(sk);
1870
1871 __tcp_v6_send_check(skb, &np->saddr, &sk->sk_v6_daddr);
1872 }
1873
1874 const struct inet_connection_sock_af_ops ipv6_specific = {
1875 .queue_xmit = inet6_csk_xmit,
1876 .send_check = tcp_v6_send_check,
1877 .rebuild_header = inet6_sk_rebuild_header,
1878 .sk_rx_dst_set = inet6_sk_rx_dst_set,
1879 .conn_request = tcp_v6_conn_request,
1880 .syn_recv_sock = tcp_v6_syn_recv_sock,
1881 .net_header_len = sizeof(struct ipv6hdr),
1882 .net_frag_header_len = sizeof(struct frag_hdr),
1883 .setsockopt = ipv6_setsockopt,
1884 .getsockopt = ipv6_getsockopt,
1885 .addr2sockaddr = inet6_csk_addr2sockaddr,
1886 .sockaddr_len = sizeof(struct sockaddr_in6),
1887 .mtu_reduced = tcp_v6_mtu_reduced,
1888 };
1889
1890 #ifdef CONFIG_TCP_MD5SIG
1891 static const struct tcp_sock_af_ops tcp_sock_ipv6_specific = {
1892 .md5_lookup = tcp_v6_md5_lookup,
1893 .calc_md5_hash = tcp_v6_md5_hash_skb,
1894 .md5_parse = tcp_v6_parse_md5_keys,
1895 };
1896 #endif
1897
1898 /*
1899 * TCP over IPv4 via INET6 API
1900 */
1901 static const struct inet_connection_sock_af_ops ipv6_mapped = {
1902 .queue_xmit = ip_queue_xmit,
1903 .send_check = tcp_v4_send_check,
1904 .rebuild_header = inet_sk_rebuild_header,
1905 .sk_rx_dst_set = inet_sk_rx_dst_set,
1906 .conn_request = tcp_v6_conn_request,
1907 .syn_recv_sock = tcp_v6_syn_recv_sock,
1908 .net_header_len = sizeof(struct iphdr),
1909 .setsockopt = ipv6_setsockopt,
1910 .getsockopt = ipv6_getsockopt,
1911 .addr2sockaddr = inet6_csk_addr2sockaddr,
1912 .sockaddr_len = sizeof(struct sockaddr_in6),
1913 .mtu_reduced = tcp_v4_mtu_reduced,
1914 };
1915
1916 #ifdef CONFIG_TCP_MD5SIG
1917 static const struct tcp_sock_af_ops tcp_sock_ipv6_mapped_specific = {
1918 .md5_lookup = tcp_v4_md5_lookup,
1919 .calc_md5_hash = tcp_v4_md5_hash_skb,
1920 .md5_parse = tcp_v6_parse_md5_keys,
1921 };
1922 #endif
1923
1924 /* NOTE: A lot of things set to zero explicitly by call to
1925 * sk_alloc() so need not be done here.
1926 */
tcp_v6_init_sock(struct sock * sk)1927 static int tcp_v6_init_sock(struct sock *sk)
1928 {
1929 struct inet_connection_sock *icsk = inet_csk(sk);
1930
1931 tcp_init_sock(sk);
1932
1933 icsk->icsk_af_ops = &ipv6_specific;
1934
1935 #ifdef CONFIG_TCP_MD5SIG
1936 tcp_sk(sk)->af_specific = &tcp_sock_ipv6_specific;
1937 #endif
1938
1939 return 0;
1940 }
1941
tcp_v6_destroy_sock(struct sock * sk)1942 static void tcp_v6_destroy_sock(struct sock *sk)
1943 {
1944 tcp_v4_destroy_sock(sk);
1945 inet6_destroy_sock(sk);
1946 }
1947
1948 #ifdef CONFIG_PROC_FS
1949 /* Proc filesystem TCPv6 sock list dumping. */
get_openreq6(struct seq_file * seq,const struct request_sock * req,int i)1950 static void get_openreq6(struct seq_file *seq,
1951 const struct request_sock *req, int i)
1952 {
1953 long ttd = req->rsk_timer.expires - jiffies;
1954 const struct in6_addr *src = &inet_rsk(req)->ir_v6_loc_addr;
1955 const struct in6_addr *dest = &inet_rsk(req)->ir_v6_rmt_addr;
1956
1957 if (ttd < 0)
1958 ttd = 0;
1959
1960 seq_printf(seq,
1961 "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
1962 "%02X %08X:%08X %02X:%08lX %08X %5u %8d %d %d %pK\n",
1963 i,
1964 src->s6_addr32[0], src->s6_addr32[1],
1965 src->s6_addr32[2], src->s6_addr32[3],
1966 inet_rsk(req)->ir_num,
1967 dest->s6_addr32[0], dest->s6_addr32[1],
1968 dest->s6_addr32[2], dest->s6_addr32[3],
1969 ntohs(inet_rsk(req)->ir_rmt_port),
1970 TCP_SYN_RECV,
1971 0, 0, /* could print option size, but that is af dependent. */
1972 1, /* timers active (only the expire timer) */
1973 jiffies_to_clock_t(ttd),
1974 req->num_timeout,
1975 from_kuid_munged(seq_user_ns(seq),
1976 sock_i_uid(req->rsk_listener)),
1977 0, /* non standard timer */
1978 0, /* open_requests have no inode */
1979 0, req);
1980 }
1981
get_tcp6_sock(struct seq_file * seq,struct sock * sp,int i)1982 static void get_tcp6_sock(struct seq_file *seq, struct sock *sp, int i)
1983 {
1984 const struct in6_addr *dest, *src;
1985 __u16 destp, srcp;
1986 int timer_active;
1987 unsigned long timer_expires;
1988 const struct inet_sock *inet = inet_sk(sp);
1989 const struct tcp_sock *tp = tcp_sk(sp);
1990 const struct inet_connection_sock *icsk = inet_csk(sp);
1991 const struct fastopen_queue *fastopenq = &icsk->icsk_accept_queue.fastopenq;
1992 int rx_queue;
1993 int state;
1994
1995 dest = &sp->sk_v6_daddr;
1996 src = &sp->sk_v6_rcv_saddr;
1997 destp = ntohs(inet->inet_dport);
1998 srcp = ntohs(inet->inet_sport);
1999
2000 if (icsk->icsk_pending == ICSK_TIME_RETRANS ||
2001 icsk->icsk_pending == ICSK_TIME_REO_TIMEOUT ||
2002 icsk->icsk_pending == ICSK_TIME_LOSS_PROBE) {
2003 timer_active = 1;
2004 timer_expires = icsk->icsk_timeout;
2005 } else if (icsk->icsk_pending == ICSK_TIME_PROBE0) {
2006 timer_active = 4;
2007 timer_expires = icsk->icsk_timeout;
2008 } else if (timer_pending(&sp->sk_timer)) {
2009 timer_active = 2;
2010 timer_expires = sp->sk_timer.expires;
2011 } else {
2012 timer_active = 0;
2013 timer_expires = jiffies;
2014 }
2015
2016 state = inet_sk_state_load(sp);
2017 if (state == TCP_LISTEN)
2018 rx_queue = READ_ONCE(sp->sk_ack_backlog);
2019 else
2020 /* Because we don't lock the socket,
2021 * we might find a transient negative value.
2022 */
2023 rx_queue = max_t(int, READ_ONCE(tp->rcv_nxt) -
2024 READ_ONCE(tp->copied_seq), 0);
2025
2026 seq_printf(seq,
2027 "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
2028 "%02X %08X:%08X %02X:%08lX %08X %5u %8d %lu %d %pK %lu %lu %u %u %d\n",
2029 i,
2030 src->s6_addr32[0], src->s6_addr32[1],
2031 src->s6_addr32[2], src->s6_addr32[3], srcp,
2032 dest->s6_addr32[0], dest->s6_addr32[1],
2033 dest->s6_addr32[2], dest->s6_addr32[3], destp,
2034 state,
2035 READ_ONCE(tp->write_seq) - tp->snd_una,
2036 rx_queue,
2037 timer_active,
2038 jiffies_delta_to_clock_t(timer_expires - jiffies),
2039 icsk->icsk_retransmits,
2040 from_kuid_munged(seq_user_ns(seq), sock_i_uid(sp)),
2041 icsk->icsk_probes_out,
2042 sock_i_ino(sp),
2043 refcount_read(&sp->sk_refcnt), sp,
2044 jiffies_to_clock_t(icsk->icsk_rto),
2045 jiffies_to_clock_t(icsk->icsk_ack.ato),
2046 (icsk->icsk_ack.quick << 1) | inet_csk_in_pingpong_mode(sp),
2047 tp->snd_cwnd,
2048 state == TCP_LISTEN ?
2049 fastopenq->max_qlen :
2050 (tcp_in_initial_slowstart(tp) ? -1 : tp->snd_ssthresh)
2051 );
2052 }
2053
get_timewait6_sock(struct seq_file * seq,struct inet_timewait_sock * tw,int i)2054 static void get_timewait6_sock(struct seq_file *seq,
2055 struct inet_timewait_sock *tw, int i)
2056 {
2057 long delta = tw->tw_timer.expires - jiffies;
2058 const struct in6_addr *dest, *src;
2059 __u16 destp, srcp;
2060
2061 dest = &tw->tw_v6_daddr;
2062 src = &tw->tw_v6_rcv_saddr;
2063 destp = ntohs(tw->tw_dport);
2064 srcp = ntohs(tw->tw_sport);
2065
2066 seq_printf(seq,
2067 "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
2068 "%02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %pK\n",
2069 i,
2070 src->s6_addr32[0], src->s6_addr32[1],
2071 src->s6_addr32[2], src->s6_addr32[3], srcp,
2072 dest->s6_addr32[0], dest->s6_addr32[1],
2073 dest->s6_addr32[2], dest->s6_addr32[3], destp,
2074 tw->tw_substate, 0, 0,
2075 3, jiffies_delta_to_clock_t(delta), 0, 0, 0, 0,
2076 refcount_read(&tw->tw_refcnt), tw);
2077 }
2078
tcp6_seq_show(struct seq_file * seq,void * v)2079 static int tcp6_seq_show(struct seq_file *seq, void *v)
2080 {
2081 struct tcp_iter_state *st;
2082 struct sock *sk = v;
2083
2084 if (v == SEQ_START_TOKEN) {
2085 seq_puts(seq,
2086 " sl "
2087 "local_address "
2088 "remote_address "
2089 "st tx_queue rx_queue tr tm->when retrnsmt"
2090 " uid timeout inode\n");
2091 goto out;
2092 }
2093 st = seq->private;
2094
2095 if (sk->sk_state == TCP_TIME_WAIT)
2096 get_timewait6_sock(seq, v, st->num);
2097 else if (sk->sk_state == TCP_NEW_SYN_RECV)
2098 get_openreq6(seq, v, st->num);
2099 else
2100 get_tcp6_sock(seq, v, st->num);
2101 out:
2102 return 0;
2103 }
2104
2105 static const struct seq_operations tcp6_seq_ops = {
2106 .show = tcp6_seq_show,
2107 .start = tcp_seq_start,
2108 .next = tcp_seq_next,
2109 .stop = tcp_seq_stop,
2110 };
2111
2112 static struct tcp_seq_afinfo tcp6_seq_afinfo = {
2113 .family = AF_INET6,
2114 };
2115
tcp6_proc_init(struct net * net)2116 int __net_init tcp6_proc_init(struct net *net)
2117 {
2118 if (!proc_create_net_data("tcp6", 0444, net->proc_net, &tcp6_seq_ops,
2119 sizeof(struct tcp_iter_state), &tcp6_seq_afinfo))
2120 return -ENOMEM;
2121 return 0;
2122 }
2123
tcp6_proc_exit(struct net * net)2124 void tcp6_proc_exit(struct net *net)
2125 {
2126 remove_proc_entry("tcp6", net->proc_net);
2127 }
2128 #endif
2129
2130 struct proto tcpv6_prot = {
2131 .name = "TCPv6",
2132 .owner = THIS_MODULE,
2133 .close = tcp_close,
2134 .pre_connect = tcp_v6_pre_connect,
2135 .connect = tcp_v6_connect,
2136 .disconnect = tcp_disconnect,
2137 .accept = inet_csk_accept,
2138 .ioctl = tcp_ioctl,
2139 .init = tcp_v6_init_sock,
2140 .destroy = tcp_v6_destroy_sock,
2141 .shutdown = tcp_shutdown,
2142 .setsockopt = tcp_setsockopt,
2143 .getsockopt = tcp_getsockopt,
2144 .bpf_bypass_getsockopt = tcp_bpf_bypass_getsockopt,
2145 .keepalive = tcp_set_keepalive,
2146 .recvmsg = tcp_recvmsg,
2147 .sendmsg = tcp_sendmsg,
2148 .sendpage = tcp_sendpage,
2149 .backlog_rcv = tcp_v6_do_rcv,
2150 .release_cb = tcp_release_cb,
2151 .hash = inet6_hash,
2152 .unhash = inet_unhash,
2153 .get_port = inet_csk_get_port,
2154 #ifdef CONFIG_BPF_SYSCALL
2155 .psock_update_sk_prot = tcp_bpf_update_proto,
2156 #endif
2157 .enter_memory_pressure = tcp_enter_memory_pressure,
2158 .leave_memory_pressure = tcp_leave_memory_pressure,
2159 .stream_memory_free = tcp_stream_memory_free,
2160 .sockets_allocated = &tcp_sockets_allocated,
2161 .memory_allocated = &tcp_memory_allocated,
2162 .memory_pressure = &tcp_memory_pressure,
2163 .orphan_count = &tcp_orphan_count,
2164 .sysctl_mem = sysctl_tcp_mem,
2165 .sysctl_wmem_offset = offsetof(struct net, ipv4.sysctl_tcp_wmem),
2166 .sysctl_rmem_offset = offsetof(struct net, ipv4.sysctl_tcp_rmem),
2167 .max_header = MAX_TCP_HEADER,
2168 .obj_size = sizeof(struct tcp6_sock),
2169 .slab_flags = SLAB_TYPESAFE_BY_RCU,
2170 .twsk_prot = &tcp6_timewait_sock_ops,
2171 .rsk_prot = &tcp6_request_sock_ops,
2172 .h.hashinfo = &tcp_hashinfo,
2173 .no_autobind = true,
2174 .diag_destroy = tcp_abort,
2175 };
2176 EXPORT_SYMBOL_GPL(tcpv6_prot);
2177
2178 /* thinking of making this const? Don't.
2179 * early_demux can change based on sysctl.
2180 */
2181 static struct inet6_protocol tcpv6_protocol = {
2182 .early_demux = tcp_v6_early_demux,
2183 .early_demux_handler = tcp_v6_early_demux,
2184 .handler = tcp_v6_rcv,
2185 .err_handler = tcp_v6_err,
2186 .flags = INET6_PROTO_NOPOLICY|INET6_PROTO_FINAL,
2187 };
2188
2189 static struct inet_protosw tcpv6_protosw = {
2190 .type = SOCK_STREAM,
2191 .protocol = IPPROTO_TCP,
2192 .prot = &tcpv6_prot,
2193 .ops = &inet6_stream_ops,
2194 .flags = INET_PROTOSW_PERMANENT |
2195 INET_PROTOSW_ICSK,
2196 };
2197
tcpv6_net_init(struct net * net)2198 static int __net_init tcpv6_net_init(struct net *net)
2199 {
2200 return inet_ctl_sock_create(&net->ipv6.tcp_sk, PF_INET6,
2201 SOCK_RAW, IPPROTO_TCP, net);
2202 }
2203
tcpv6_net_exit(struct net * net)2204 static void __net_exit tcpv6_net_exit(struct net *net)
2205 {
2206 inet_ctl_sock_destroy(net->ipv6.tcp_sk);
2207 }
2208
tcpv6_net_exit_batch(struct list_head * net_exit_list)2209 static void __net_exit tcpv6_net_exit_batch(struct list_head *net_exit_list)
2210 {
2211 inet_twsk_purge(&tcp_hashinfo, AF_INET6);
2212 }
2213
2214 static struct pernet_operations tcpv6_net_ops = {
2215 .init = tcpv6_net_init,
2216 .exit = tcpv6_net_exit,
2217 .exit_batch = tcpv6_net_exit_batch,
2218 };
2219
tcpv6_init(void)2220 int __init tcpv6_init(void)
2221 {
2222 int ret;
2223
2224 ret = inet6_add_protocol(&tcpv6_protocol, IPPROTO_TCP);
2225 if (ret)
2226 goto out;
2227
2228 /* register inet6 protocol */
2229 ret = inet6_register_protosw(&tcpv6_protosw);
2230 if (ret)
2231 goto out_tcpv6_protocol;
2232
2233 ret = register_pernet_subsys(&tcpv6_net_ops);
2234 if (ret)
2235 goto out_tcpv6_protosw;
2236
2237 ret = mptcpv6_init();
2238 if (ret)
2239 goto out_tcpv6_pernet_subsys;
2240
2241 out:
2242 return ret;
2243
2244 out_tcpv6_pernet_subsys:
2245 unregister_pernet_subsys(&tcpv6_net_ops);
2246 out_tcpv6_protosw:
2247 inet6_unregister_protosw(&tcpv6_protosw);
2248 out_tcpv6_protocol:
2249 inet6_del_protocol(&tcpv6_protocol, IPPROTO_TCP);
2250 goto out;
2251 }
2252
tcpv6_exit(void)2253 void tcpv6_exit(void)
2254 {
2255 unregister_pernet_subsys(&tcpv6_net_ops);
2256 inet6_unregister_protosw(&tcpv6_protosw);
2257 inet6_del_protocol(&tcpv6_protocol, IPPROTO_TCP);
2258 }
2259