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 INDIRECT_CALLABLE_SCOPE 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 */
96 #define tcp_inet6_sk(sk) (&container_of_const(tcp_sk(sk), \
97 struct tcp6_sock, tcp)->inet6)
98
inet6_sk_rx_dst_set(struct sock * sk,const struct sk_buff * skb)99 static void inet6_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb)
100 {
101 struct dst_entry *dst = skb_dst(skb);
102
103 if (dst && dst_hold_safe(dst)) {
104 const struct rt6_info *rt = (const struct rt6_info *)dst;
105
106 rcu_assign_pointer(sk->sk_rx_dst, dst);
107 sk->sk_rx_dst_ifindex = skb->skb_iif;
108 sk->sk_rx_dst_cookie = rt6_get_cookie(rt);
109 }
110 }
111
tcp_v6_init_seq(const struct sk_buff * skb)112 static u32 tcp_v6_init_seq(const struct sk_buff *skb)
113 {
114 return secure_tcpv6_seq(ipv6_hdr(skb)->daddr.s6_addr32,
115 ipv6_hdr(skb)->saddr.s6_addr32,
116 tcp_hdr(skb)->dest,
117 tcp_hdr(skb)->source);
118 }
119
tcp_v6_init_ts_off(const struct net * net,const struct sk_buff * skb)120 static u32 tcp_v6_init_ts_off(const struct net *net, const struct sk_buff *skb)
121 {
122 return secure_tcpv6_ts_off(net, ipv6_hdr(skb)->daddr.s6_addr32,
123 ipv6_hdr(skb)->saddr.s6_addr32);
124 }
125
tcp_v6_pre_connect(struct sock * sk,struct sockaddr * uaddr,int addr_len)126 static int tcp_v6_pre_connect(struct sock *sk, struct sockaddr *uaddr,
127 int addr_len)
128 {
129 /* This check is replicated from tcp_v6_connect() and intended to
130 * prevent BPF program called below from accessing bytes that are out
131 * of the bound specified by user in addr_len.
132 */
133 if (addr_len < SIN6_LEN_RFC2133)
134 return -EINVAL;
135
136 sock_owned_by_me(sk);
137
138 return BPF_CGROUP_RUN_PROG_INET6_CONNECT(sk, uaddr, &addr_len);
139 }
140
tcp_v6_connect(struct sock * sk,struct sockaddr * uaddr,int addr_len)141 static int tcp_v6_connect(struct sock *sk, struct sockaddr *uaddr,
142 int addr_len)
143 {
144 struct sockaddr_in6 *usin = (struct sockaddr_in6 *) uaddr;
145 struct inet_connection_sock *icsk = inet_csk(sk);
146 struct in6_addr *saddr = NULL, *final_p, final;
147 struct inet_timewait_death_row *tcp_death_row;
148 struct ipv6_pinfo *np = tcp_inet6_sk(sk);
149 struct inet_sock *inet = inet_sk(sk);
150 struct tcp_sock *tp = tcp_sk(sk);
151 struct net *net = sock_net(sk);
152 struct ipv6_txoptions *opt;
153 struct dst_entry *dst;
154 struct flowi6 fl6;
155 int addr_type;
156 int err;
157
158 if (addr_len < SIN6_LEN_RFC2133)
159 return -EINVAL;
160
161 if (usin->sin6_family != AF_INET6)
162 return -EAFNOSUPPORT;
163
164 memset(&fl6, 0, sizeof(fl6));
165
166 if (np->sndflow) {
167 fl6.flowlabel = usin->sin6_flowinfo&IPV6_FLOWINFO_MASK;
168 IP6_ECN_flow_init(fl6.flowlabel);
169 if (fl6.flowlabel&IPV6_FLOWLABEL_MASK) {
170 struct ip6_flowlabel *flowlabel;
171 flowlabel = fl6_sock_lookup(sk, fl6.flowlabel);
172 if (IS_ERR(flowlabel))
173 return -EINVAL;
174 fl6_sock_release(flowlabel);
175 }
176 }
177
178 /*
179 * connect() to INADDR_ANY means loopback (BSD'ism).
180 */
181
182 if (ipv6_addr_any(&usin->sin6_addr)) {
183 if (ipv6_addr_v4mapped(&sk->sk_v6_rcv_saddr))
184 ipv6_addr_set_v4mapped(htonl(INADDR_LOOPBACK),
185 &usin->sin6_addr);
186 else
187 usin->sin6_addr = in6addr_loopback;
188 }
189
190 addr_type = ipv6_addr_type(&usin->sin6_addr);
191
192 if (addr_type & IPV6_ADDR_MULTICAST)
193 return -ENETUNREACH;
194
195 if (addr_type&IPV6_ADDR_LINKLOCAL) {
196 if (addr_len >= sizeof(struct sockaddr_in6) &&
197 usin->sin6_scope_id) {
198 /* If interface is set while binding, indices
199 * must coincide.
200 */
201 if (!sk_dev_equal_l3scope(sk, usin->sin6_scope_id))
202 return -EINVAL;
203
204 sk->sk_bound_dev_if = usin->sin6_scope_id;
205 }
206
207 /* Connect to link-local address requires an interface */
208 if (!sk->sk_bound_dev_if)
209 return -EINVAL;
210 }
211
212 if (tp->rx_opt.ts_recent_stamp &&
213 !ipv6_addr_equal(&sk->sk_v6_daddr, &usin->sin6_addr)) {
214 tp->rx_opt.ts_recent = 0;
215 tp->rx_opt.ts_recent_stamp = 0;
216 WRITE_ONCE(tp->write_seq, 0);
217 }
218
219 sk->sk_v6_daddr = usin->sin6_addr;
220 np->flow_label = fl6.flowlabel;
221
222 /*
223 * TCP over IPv4
224 */
225
226 if (addr_type & IPV6_ADDR_MAPPED) {
227 u32 exthdrlen = icsk->icsk_ext_hdr_len;
228 struct sockaddr_in sin;
229
230 if (ipv6_only_sock(sk))
231 return -ENETUNREACH;
232
233 sin.sin_family = AF_INET;
234 sin.sin_port = usin->sin6_port;
235 sin.sin_addr.s_addr = usin->sin6_addr.s6_addr32[3];
236
237 /* Paired with READ_ONCE() in tcp_(get|set)sockopt() */
238 WRITE_ONCE(icsk->icsk_af_ops, &ipv6_mapped);
239 if (sk_is_mptcp(sk))
240 mptcpv6_handle_mapped(sk, true);
241 sk->sk_backlog_rcv = tcp_v4_do_rcv;
242 #ifdef CONFIG_TCP_MD5SIG
243 tp->af_specific = &tcp_sock_ipv6_mapped_specific;
244 #endif
245
246 err = tcp_v4_connect(sk, (struct sockaddr *)&sin, sizeof(sin));
247
248 if (err) {
249 icsk->icsk_ext_hdr_len = exthdrlen;
250 /* Paired with READ_ONCE() in tcp_(get|set)sockopt() */
251 WRITE_ONCE(icsk->icsk_af_ops, &ipv6_specific);
252 if (sk_is_mptcp(sk))
253 mptcpv6_handle_mapped(sk, false);
254 sk->sk_backlog_rcv = tcp_v6_do_rcv;
255 #ifdef CONFIG_TCP_MD5SIG
256 tp->af_specific = &tcp_sock_ipv6_specific;
257 #endif
258 goto failure;
259 }
260 np->saddr = sk->sk_v6_rcv_saddr;
261
262 return err;
263 }
264
265 if (!ipv6_addr_any(&sk->sk_v6_rcv_saddr))
266 saddr = &sk->sk_v6_rcv_saddr;
267
268 fl6.flowi6_proto = IPPROTO_TCP;
269 fl6.daddr = sk->sk_v6_daddr;
270 fl6.saddr = saddr ? *saddr : np->saddr;
271 fl6.flowlabel = ip6_make_flowinfo(np->tclass, np->flow_label);
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(net, sk, &fl6, final_p);
284 if (IS_ERR(dst)) {
285 err = PTR_ERR(dst);
286 goto failure;
287 }
288
289 tcp_death_row = &sock_net(sk)->ipv4.tcp_death_row;
290
291 if (!saddr) {
292 saddr = &fl6.saddr;
293
294 err = inet_bhash2_update_saddr(sk, saddr, AF_INET6);
295 if (err)
296 goto failure;
297 }
298
299 /* set the source address */
300 np->saddr = *saddr;
301 inet->inet_rcv_saddr = LOOPBACK4_IPV6;
302
303 sk->sk_gso_type = SKB_GSO_TCPV6;
304 ip6_dst_store(sk, dst, NULL, NULL);
305
306 icsk->icsk_ext_hdr_len = 0;
307 if (opt)
308 icsk->icsk_ext_hdr_len = opt->opt_flen +
309 opt->opt_nflen;
310
311 tp->rx_opt.mss_clamp = IPV6_MIN_MTU - sizeof(struct tcphdr) - sizeof(struct ipv6hdr);
312
313 inet->inet_dport = usin->sin6_port;
314
315 tcp_set_state(sk, TCP_SYN_SENT);
316 err = inet6_hash_connect(tcp_death_row, sk);
317 if (err)
318 goto late_failure;
319
320 sk_set_txhash(sk);
321
322 if (likely(!tp->repair)) {
323 if (!tp->write_seq)
324 WRITE_ONCE(tp->write_seq,
325 secure_tcpv6_seq(np->saddr.s6_addr32,
326 sk->sk_v6_daddr.s6_addr32,
327 inet->inet_sport,
328 inet->inet_dport));
329 tp->tsoffset = secure_tcpv6_ts_off(net, np->saddr.s6_addr32,
330 sk->sk_v6_daddr.s6_addr32);
331 }
332
333 if (tcp_fastopen_defer_connect(sk, &err))
334 return err;
335 if (err)
336 goto late_failure;
337
338 err = tcp_connect(sk);
339 if (err)
340 goto late_failure;
341
342 return 0;
343
344 late_failure:
345 tcp_set_state(sk, TCP_CLOSE);
346 inet_bhash2_reset_saddr(sk);
347 failure:
348 inet->inet_dport = 0;
349 sk->sk_route_caps = 0;
350 return err;
351 }
352
tcp_v6_mtu_reduced(struct sock * sk)353 static void tcp_v6_mtu_reduced(struct sock *sk)
354 {
355 struct dst_entry *dst;
356 u32 mtu;
357
358 if ((1 << sk->sk_state) & (TCPF_LISTEN | TCPF_CLOSE))
359 return;
360
361 mtu = READ_ONCE(tcp_sk(sk)->mtu_info);
362
363 /* Drop requests trying to increase our current mss.
364 * Check done in __ip6_rt_update_pmtu() is too late.
365 */
366 if (tcp_mtu_to_mss(sk, mtu) >= tcp_sk(sk)->mss_cache)
367 return;
368
369 dst = inet6_csk_update_pmtu(sk, mtu);
370 if (!dst)
371 return;
372
373 if (inet_csk(sk)->icsk_pmtu_cookie > dst_mtu(dst)) {
374 tcp_sync_mss(sk, dst_mtu(dst));
375 tcp_simple_retransmit(sk);
376 }
377 }
378
tcp_v6_err(struct sk_buff * skb,struct inet6_skb_parm * opt,u8 type,u8 code,int offset,__be32 info)379 static int tcp_v6_err(struct sk_buff *skb, struct inet6_skb_parm *opt,
380 u8 type, u8 code, int offset, __be32 info)
381 {
382 const struct ipv6hdr *hdr = (const struct ipv6hdr *)skb->data;
383 const struct tcphdr *th = (struct tcphdr *)(skb->data+offset);
384 struct net *net = dev_net(skb->dev);
385 struct request_sock *fastopen;
386 struct ipv6_pinfo *np;
387 struct tcp_sock *tp;
388 __u32 seq, snd_una;
389 struct sock *sk;
390 bool fatal;
391 int err;
392
393 sk = __inet6_lookup_established(net, net->ipv4.tcp_death_row.hashinfo,
394 &hdr->daddr, th->dest,
395 &hdr->saddr, ntohs(th->source),
396 skb->dev->ifindex, inet6_sdif(skb));
397
398 if (!sk) {
399 __ICMP6_INC_STATS(net, __in6_dev_get(skb->dev),
400 ICMP6_MIB_INERRORS);
401 return -ENOENT;
402 }
403
404 if (sk->sk_state == TCP_TIME_WAIT) {
405 inet_twsk_put(inet_twsk(sk));
406 return 0;
407 }
408 seq = ntohl(th->seq);
409 fatal = icmpv6_err_convert(type, code, &err);
410 if (sk->sk_state == TCP_NEW_SYN_RECV) {
411 tcp_req_err(sk, seq, fatal);
412 return 0;
413 }
414
415 bh_lock_sock(sk);
416 if (sock_owned_by_user(sk) && type != ICMPV6_PKT_TOOBIG)
417 __NET_INC_STATS(net, LINUX_MIB_LOCKDROPPEDICMPS);
418
419 if (sk->sk_state == TCP_CLOSE)
420 goto out;
421
422 if (static_branch_unlikely(&ip6_min_hopcount)) {
423 /* min_hopcount can be changed concurrently from do_ipv6_setsockopt() */
424 if (ipv6_hdr(skb)->hop_limit < READ_ONCE(tcp_inet6_sk(sk)->min_hopcount)) {
425 __NET_INC_STATS(net, LINUX_MIB_TCPMINTTLDROP);
426 goto out;
427 }
428 }
429
430 tp = tcp_sk(sk);
431 /* XXX (TFO) - tp->snd_una should be ISN (tcp_create_openreq_child() */
432 fastopen = rcu_dereference(tp->fastopen_rsk);
433 snd_una = fastopen ? tcp_rsk(fastopen)->snt_isn : tp->snd_una;
434 if (sk->sk_state != TCP_LISTEN &&
435 !between(seq, snd_una, tp->snd_nxt)) {
436 __NET_INC_STATS(net, LINUX_MIB_OUTOFWINDOWICMPS);
437 goto out;
438 }
439
440 np = tcp_inet6_sk(sk);
441
442 if (type == NDISC_REDIRECT) {
443 if (!sock_owned_by_user(sk)) {
444 struct dst_entry *dst = __sk_dst_check(sk, np->dst_cookie);
445
446 if (dst)
447 dst->ops->redirect(dst, sk, skb);
448 }
449 goto out;
450 }
451
452 if (type == ICMPV6_PKT_TOOBIG) {
453 u32 mtu = ntohl(info);
454
455 /* We are not interested in TCP_LISTEN and open_requests
456 * (SYN-ACKs send out by Linux are always <576bytes so
457 * they should go through unfragmented).
458 */
459 if (sk->sk_state == TCP_LISTEN)
460 goto out;
461
462 if (!ip6_sk_accept_pmtu(sk))
463 goto out;
464
465 if (mtu < IPV6_MIN_MTU)
466 goto out;
467
468 WRITE_ONCE(tp->mtu_info, mtu);
469
470 if (!sock_owned_by_user(sk))
471 tcp_v6_mtu_reduced(sk);
472 else if (!test_and_set_bit(TCP_MTU_REDUCED_DEFERRED,
473 &sk->sk_tsq_flags))
474 sock_hold(sk);
475 goto out;
476 }
477
478
479 /* Might be for an request_sock */
480 switch (sk->sk_state) {
481 case TCP_SYN_SENT:
482 case TCP_SYN_RECV:
483 /* Only in fast or simultaneous open. If a fast open socket is
484 * already accepted it is treated as a connected one below.
485 */
486 if (fastopen && !fastopen->sk)
487 break;
488
489 ipv6_icmp_error(sk, skb, err, th->dest, ntohl(info), (u8 *)th);
490
491 if (!sock_owned_by_user(sk)) {
492 WRITE_ONCE(sk->sk_err, err);
493 sk_error_report(sk); /* Wake people up to see the error (see connect in sock.c) */
494
495 tcp_done(sk);
496 } else {
497 WRITE_ONCE(sk->sk_err_soft, err);
498 }
499 goto out;
500 case TCP_LISTEN:
501 break;
502 default:
503 /* check if this ICMP message allows revert of backoff.
504 * (see RFC 6069)
505 */
506 if (!fastopen && type == ICMPV6_DEST_UNREACH &&
507 code == ICMPV6_NOROUTE)
508 tcp_ld_RTO_revert(sk, seq);
509 }
510
511 if (!sock_owned_by_user(sk) && np->recverr) {
512 WRITE_ONCE(sk->sk_err, err);
513 sk_error_report(sk);
514 } else {
515 WRITE_ONCE(sk->sk_err_soft, err);
516 }
517 out:
518 bh_unlock_sock(sk);
519 sock_put(sk);
520 return 0;
521 }
522
523
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)524 static int tcp_v6_send_synack(const struct sock *sk, struct dst_entry *dst,
525 struct flowi *fl,
526 struct request_sock *req,
527 struct tcp_fastopen_cookie *foc,
528 enum tcp_synack_type synack_type,
529 struct sk_buff *syn_skb)
530 {
531 struct inet_request_sock *ireq = inet_rsk(req);
532 const struct ipv6_pinfo *np = tcp_inet6_sk(sk);
533 struct ipv6_txoptions *opt;
534 struct flowi6 *fl6 = &fl->u.ip6;
535 struct sk_buff *skb;
536 int err = -ENOMEM;
537 u8 tclass;
538
539 /* First, grab a route. */
540 if (!dst && (dst = inet6_csk_route_req(sk, fl6, req,
541 IPPROTO_TCP)) == NULL)
542 goto done;
543
544 skb = tcp_make_synack(sk, dst, req, foc, synack_type, syn_skb);
545
546 if (skb) {
547 __tcp_v6_send_check(skb, &ireq->ir_v6_loc_addr,
548 &ireq->ir_v6_rmt_addr);
549
550 fl6->daddr = ireq->ir_v6_rmt_addr;
551 if (np->repflow && ireq->pktopts)
552 fl6->flowlabel = ip6_flowlabel(ipv6_hdr(ireq->pktopts));
553
554 tclass = READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_reflect_tos) ?
555 (tcp_rsk(req)->syn_tos & ~INET_ECN_MASK) |
556 (np->tclass & INET_ECN_MASK) :
557 np->tclass;
558
559 if (!INET_ECN_is_capable(tclass) &&
560 tcp_bpf_ca_needs_ecn((struct sock *)req))
561 tclass |= INET_ECN_ECT_0;
562
563 rcu_read_lock();
564 opt = ireq->ipv6_opt;
565 if (!opt)
566 opt = rcu_dereference(np->opt);
567 err = ip6_xmit(sk, skb, fl6, skb->mark ? : READ_ONCE(sk->sk_mark),
568 opt, tclass, sk->sk_priority);
569 rcu_read_unlock();
570 err = net_xmit_eval(err);
571 }
572
573 done:
574 return err;
575 }
576
577
tcp_v6_reqsk_destructor(struct request_sock * req)578 static void tcp_v6_reqsk_destructor(struct request_sock *req)
579 {
580 kfree(inet_rsk(req)->ipv6_opt);
581 consume_skb(inet_rsk(req)->pktopts);
582 }
583
584 #ifdef CONFIG_TCP_MD5SIG
tcp_v6_md5_do_lookup(const struct sock * sk,const struct in6_addr * addr,int l3index)585 static struct tcp_md5sig_key *tcp_v6_md5_do_lookup(const struct sock *sk,
586 const struct in6_addr *addr,
587 int l3index)
588 {
589 return tcp_md5_do_lookup(sk, l3index,
590 (union tcp_md5_addr *)addr, AF_INET6);
591 }
592
tcp_v6_md5_lookup(const struct sock * sk,const struct sock * addr_sk)593 static struct tcp_md5sig_key *tcp_v6_md5_lookup(const struct sock *sk,
594 const struct sock *addr_sk)
595 {
596 int l3index;
597
598 l3index = l3mdev_master_ifindex_by_index(sock_net(sk),
599 addr_sk->sk_bound_dev_if);
600 return tcp_v6_md5_do_lookup(sk, &addr_sk->sk_v6_daddr,
601 l3index);
602 }
603
tcp_v6_parse_md5_keys(struct sock * sk,int optname,sockptr_t optval,int optlen)604 static int tcp_v6_parse_md5_keys(struct sock *sk, int optname,
605 sockptr_t optval, int optlen)
606 {
607 struct tcp_md5sig cmd;
608 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)&cmd.tcpm_addr;
609 int l3index = 0;
610 u8 prefixlen;
611 u8 flags;
612
613 if (optlen < sizeof(cmd))
614 return -EINVAL;
615
616 if (copy_from_sockptr(&cmd, optval, sizeof(cmd)))
617 return -EFAULT;
618
619 if (sin6->sin6_family != AF_INET6)
620 return -EINVAL;
621
622 flags = cmd.tcpm_flags & TCP_MD5SIG_FLAG_IFINDEX;
623
624 if (optname == TCP_MD5SIG_EXT &&
625 cmd.tcpm_flags & TCP_MD5SIG_FLAG_PREFIX) {
626 prefixlen = cmd.tcpm_prefixlen;
627 if (prefixlen > 128 || (ipv6_addr_v4mapped(&sin6->sin6_addr) &&
628 prefixlen > 32))
629 return -EINVAL;
630 } else {
631 prefixlen = ipv6_addr_v4mapped(&sin6->sin6_addr) ? 32 : 128;
632 }
633
634 if (optname == TCP_MD5SIG_EXT && cmd.tcpm_ifindex &&
635 cmd.tcpm_flags & TCP_MD5SIG_FLAG_IFINDEX) {
636 struct net_device *dev;
637
638 rcu_read_lock();
639 dev = dev_get_by_index_rcu(sock_net(sk), cmd.tcpm_ifindex);
640 if (dev && netif_is_l3_master(dev))
641 l3index = dev->ifindex;
642 rcu_read_unlock();
643
644 /* ok to reference set/not set outside of rcu;
645 * right now device MUST be an L3 master
646 */
647 if (!dev || !l3index)
648 return -EINVAL;
649 }
650
651 if (!cmd.tcpm_keylen) {
652 if (ipv6_addr_v4mapped(&sin6->sin6_addr))
653 return tcp_md5_do_del(sk, (union tcp_md5_addr *)&sin6->sin6_addr.s6_addr32[3],
654 AF_INET, prefixlen,
655 l3index, flags);
656 return tcp_md5_do_del(sk, (union tcp_md5_addr *)&sin6->sin6_addr,
657 AF_INET6, prefixlen, l3index, flags);
658 }
659
660 if (cmd.tcpm_keylen > TCP_MD5SIG_MAXKEYLEN)
661 return -EINVAL;
662
663 if (ipv6_addr_v4mapped(&sin6->sin6_addr))
664 return tcp_md5_do_add(sk, (union tcp_md5_addr *)&sin6->sin6_addr.s6_addr32[3],
665 AF_INET, prefixlen, l3index, flags,
666 cmd.tcpm_key, cmd.tcpm_keylen);
667
668 return tcp_md5_do_add(sk, (union tcp_md5_addr *)&sin6->sin6_addr,
669 AF_INET6, prefixlen, l3index, flags,
670 cmd.tcpm_key, cmd.tcpm_keylen);
671 }
672
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)673 static int tcp_v6_md5_hash_headers(struct tcp_md5sig_pool *hp,
674 const struct in6_addr *daddr,
675 const struct in6_addr *saddr,
676 const struct tcphdr *th, int nbytes)
677 {
678 struct tcp6_pseudohdr *bp;
679 struct scatterlist sg;
680 struct tcphdr *_th;
681
682 bp = hp->scratch;
683 /* 1. TCP pseudo-header (RFC2460) */
684 bp->saddr = *saddr;
685 bp->daddr = *daddr;
686 bp->protocol = cpu_to_be32(IPPROTO_TCP);
687 bp->len = cpu_to_be32(nbytes);
688
689 _th = (struct tcphdr *)(bp + 1);
690 memcpy(_th, th, sizeof(*th));
691 _th->check = 0;
692
693 sg_init_one(&sg, bp, sizeof(*bp) + sizeof(*th));
694 ahash_request_set_crypt(hp->md5_req, &sg, NULL,
695 sizeof(*bp) + sizeof(*th));
696 return crypto_ahash_update(hp->md5_req);
697 }
698
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)699 static int tcp_v6_md5_hash_hdr(char *md5_hash, const struct tcp_md5sig_key *key,
700 const struct in6_addr *daddr, struct in6_addr *saddr,
701 const struct tcphdr *th)
702 {
703 struct tcp_md5sig_pool *hp;
704 struct ahash_request *req;
705
706 hp = tcp_get_md5sig_pool();
707 if (!hp)
708 goto clear_hash_noput;
709 req = hp->md5_req;
710
711 if (crypto_ahash_init(req))
712 goto clear_hash;
713 if (tcp_v6_md5_hash_headers(hp, daddr, saddr, th, th->doff << 2))
714 goto clear_hash;
715 if (tcp_md5_hash_key(hp, key))
716 goto clear_hash;
717 ahash_request_set_crypt(req, NULL, md5_hash, 0);
718 if (crypto_ahash_final(req))
719 goto clear_hash;
720
721 tcp_put_md5sig_pool();
722 return 0;
723
724 clear_hash:
725 tcp_put_md5sig_pool();
726 clear_hash_noput:
727 memset(md5_hash, 0, 16);
728 return 1;
729 }
730
tcp_v6_md5_hash_skb(char * md5_hash,const struct tcp_md5sig_key * key,const struct sock * sk,const struct sk_buff * skb)731 static int tcp_v6_md5_hash_skb(char *md5_hash,
732 const struct tcp_md5sig_key *key,
733 const struct sock *sk,
734 const struct sk_buff *skb)
735 {
736 const struct in6_addr *saddr, *daddr;
737 struct tcp_md5sig_pool *hp;
738 struct ahash_request *req;
739 const struct tcphdr *th = tcp_hdr(skb);
740
741 if (sk) { /* valid for establish/request sockets */
742 saddr = &sk->sk_v6_rcv_saddr;
743 daddr = &sk->sk_v6_daddr;
744 } else {
745 const struct ipv6hdr *ip6h = ipv6_hdr(skb);
746 saddr = &ip6h->saddr;
747 daddr = &ip6h->daddr;
748 }
749
750 hp = tcp_get_md5sig_pool();
751 if (!hp)
752 goto clear_hash_noput;
753 req = hp->md5_req;
754
755 if (crypto_ahash_init(req))
756 goto clear_hash;
757
758 if (tcp_v6_md5_hash_headers(hp, daddr, saddr, th, skb->len))
759 goto clear_hash;
760 if (tcp_md5_hash_skb_data(hp, skb, th->doff << 2))
761 goto clear_hash;
762 if (tcp_md5_hash_key(hp, key))
763 goto clear_hash;
764 ahash_request_set_crypt(req, NULL, md5_hash, 0);
765 if (crypto_ahash_final(req))
766 goto clear_hash;
767
768 tcp_put_md5sig_pool();
769 return 0;
770
771 clear_hash:
772 tcp_put_md5sig_pool();
773 clear_hash_noput:
774 memset(md5_hash, 0, 16);
775 return 1;
776 }
777
778 #endif
779
tcp_v6_init_req(struct request_sock * req,const struct sock * sk_listener,struct sk_buff * skb)780 static void tcp_v6_init_req(struct request_sock *req,
781 const struct sock *sk_listener,
782 struct sk_buff *skb)
783 {
784 bool l3_slave = ipv6_l3mdev_skb(TCP_SKB_CB(skb)->header.h6.flags);
785 struct inet_request_sock *ireq = inet_rsk(req);
786 const struct ipv6_pinfo *np = tcp_inet6_sk(sk_listener);
787
788 ireq->ir_v6_rmt_addr = ipv6_hdr(skb)->saddr;
789 ireq->ir_v6_loc_addr = ipv6_hdr(skb)->daddr;
790
791 /* So that link locals have meaning */
792 if ((!sk_listener->sk_bound_dev_if || l3_slave) &&
793 ipv6_addr_type(&ireq->ir_v6_rmt_addr) & IPV6_ADDR_LINKLOCAL)
794 ireq->ir_iif = tcp_v6_iif(skb);
795
796 if (!TCP_SKB_CB(skb)->tcp_tw_isn &&
797 (ipv6_opt_accepted(sk_listener, skb, &TCP_SKB_CB(skb)->header.h6) ||
798 np->rxopt.bits.rxinfo ||
799 np->rxopt.bits.rxoinfo || np->rxopt.bits.rxhlim ||
800 np->rxopt.bits.rxohlim || np->repflow)) {
801 refcount_inc(&skb->users);
802 ireq->pktopts = skb;
803 }
804 }
805
tcp_v6_route_req(const struct sock * sk,struct sk_buff * skb,struct flowi * fl,struct request_sock * req)806 static struct dst_entry *tcp_v6_route_req(const struct sock *sk,
807 struct sk_buff *skb,
808 struct flowi *fl,
809 struct request_sock *req)
810 {
811 tcp_v6_init_req(req, sk, skb);
812
813 if (security_inet_conn_request(sk, skb, req))
814 return NULL;
815
816 return inet6_csk_route_req(sk, &fl->u.ip6, req, IPPROTO_TCP);
817 }
818
819 struct request_sock_ops tcp6_request_sock_ops __read_mostly = {
820 .family = AF_INET6,
821 .obj_size = sizeof(struct tcp6_request_sock),
822 .rtx_syn_ack = tcp_rtx_synack,
823 .send_ack = tcp_v6_reqsk_send_ack,
824 .destructor = tcp_v6_reqsk_destructor,
825 .send_reset = tcp_v6_send_reset,
826 .syn_ack_timeout = tcp_syn_ack_timeout,
827 };
828
829 const struct tcp_request_sock_ops tcp_request_sock_ipv6_ops = {
830 .mss_clamp = IPV6_MIN_MTU - sizeof(struct tcphdr) -
831 sizeof(struct ipv6hdr),
832 #ifdef CONFIG_TCP_MD5SIG
833 .req_md5_lookup = tcp_v6_md5_lookup,
834 .calc_md5_hash = tcp_v6_md5_hash_skb,
835 #endif
836 #ifdef CONFIG_SYN_COOKIES
837 .cookie_init_seq = cookie_v6_init_sequence,
838 #endif
839 .route_req = tcp_v6_route_req,
840 .init_seq = tcp_v6_init_seq,
841 .init_ts_off = tcp_v6_init_ts_off,
842 .send_synack = tcp_v6_send_synack,
843 };
844
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,u32 txhash)845 static void tcp_v6_send_response(const struct sock *sk, struct sk_buff *skb, u32 seq,
846 u32 ack, u32 win, u32 tsval, u32 tsecr,
847 int oif, struct tcp_md5sig_key *key, int rst,
848 u8 tclass, __be32 label, u32 priority, u32 txhash)
849 {
850 const struct tcphdr *th = tcp_hdr(skb);
851 struct tcphdr *t1;
852 struct sk_buff *buff;
853 struct flowi6 fl6;
854 struct net *net = sk ? sock_net(sk) : dev_net(skb_dst(skb)->dev);
855 struct sock *ctl_sk = net->ipv6.tcp_sk;
856 unsigned int tot_len = sizeof(struct tcphdr);
857 __be32 mrst = 0, *topt;
858 struct dst_entry *dst;
859 __u32 mark = 0;
860
861 if (tsecr)
862 tot_len += TCPOLEN_TSTAMP_ALIGNED;
863 #ifdef CONFIG_TCP_MD5SIG
864 if (key)
865 tot_len += TCPOLEN_MD5SIG_ALIGNED;
866 #endif
867
868 #ifdef CONFIG_MPTCP
869 if (rst && !key) {
870 mrst = mptcp_reset_option(skb);
871
872 if (mrst)
873 tot_len += sizeof(__be32);
874 }
875 #endif
876
877 buff = alloc_skb(MAX_TCP_HEADER, GFP_ATOMIC);
878 if (!buff)
879 return;
880
881 skb_reserve(buff, MAX_TCP_HEADER);
882
883 t1 = skb_push(buff, tot_len);
884 skb_reset_transport_header(buff);
885
886 /* Swap the send and the receive. */
887 memset(t1, 0, sizeof(*t1));
888 t1->dest = th->source;
889 t1->source = th->dest;
890 t1->doff = tot_len / 4;
891 t1->seq = htonl(seq);
892 t1->ack_seq = htonl(ack);
893 t1->ack = !rst || !th->ack;
894 t1->rst = rst;
895 t1->window = htons(win);
896
897 topt = (__be32 *)(t1 + 1);
898
899 if (tsecr) {
900 *topt++ = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
901 (TCPOPT_TIMESTAMP << 8) | TCPOLEN_TIMESTAMP);
902 *topt++ = htonl(tsval);
903 *topt++ = htonl(tsecr);
904 }
905
906 if (mrst)
907 *topt++ = mrst;
908
909 #ifdef CONFIG_TCP_MD5SIG
910 if (key) {
911 *topt++ = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
912 (TCPOPT_MD5SIG << 8) | TCPOLEN_MD5SIG);
913 tcp_v6_md5_hash_hdr((__u8 *)topt, key,
914 &ipv6_hdr(skb)->saddr,
915 &ipv6_hdr(skb)->daddr, t1);
916 }
917 #endif
918
919 memset(&fl6, 0, sizeof(fl6));
920 fl6.daddr = ipv6_hdr(skb)->saddr;
921 fl6.saddr = ipv6_hdr(skb)->daddr;
922 fl6.flowlabel = label;
923
924 buff->ip_summed = CHECKSUM_PARTIAL;
925
926 __tcp_v6_send_check(buff, &fl6.saddr, &fl6.daddr);
927
928 fl6.flowi6_proto = IPPROTO_TCP;
929 if (rt6_need_strict(&fl6.daddr) && !oif)
930 fl6.flowi6_oif = tcp_v6_iif(skb);
931 else {
932 if (!oif && netif_index_is_l3_master(net, skb->skb_iif))
933 oif = skb->skb_iif;
934
935 fl6.flowi6_oif = oif;
936 }
937
938 if (sk) {
939 if (sk->sk_state == TCP_TIME_WAIT)
940 mark = inet_twsk(sk)->tw_mark;
941 else
942 mark = READ_ONCE(sk->sk_mark);
943 skb_set_delivery_time(buff, tcp_transmit_time(sk), true);
944 }
945 if (txhash) {
946 /* autoflowlabel/skb_get_hash_flowi6 rely on buff->hash */
947 skb_set_hash(buff, txhash, PKT_HASH_TYPE_L4);
948 }
949 fl6.flowi6_mark = IP6_REPLY_MARK(net, skb->mark) ?: mark;
950 fl6.fl6_dport = t1->dest;
951 fl6.fl6_sport = t1->source;
952 fl6.flowi6_uid = sock_net_uid(net, sk && sk_fullsock(sk) ? sk : NULL);
953 security_skb_classify_flow(skb, flowi6_to_flowi_common(&fl6));
954
955 /* Pass a socket to ip6_dst_lookup either it is for RST
956 * Underlying function will use this to retrieve the network
957 * namespace
958 */
959 if (sk && sk->sk_state != TCP_TIME_WAIT)
960 dst = ip6_dst_lookup_flow(net, sk, &fl6, NULL); /*sk's xfrm_policy can be referred*/
961 else
962 dst = ip6_dst_lookup_flow(net, ctl_sk, &fl6, NULL);
963 if (!IS_ERR(dst)) {
964 skb_dst_set(buff, dst);
965 ip6_xmit(ctl_sk, buff, &fl6, fl6.flowi6_mark, NULL,
966 tclass & ~INET_ECN_MASK, priority);
967 TCP_INC_STATS(net, TCP_MIB_OUTSEGS);
968 if (rst)
969 TCP_INC_STATS(net, TCP_MIB_OUTRSTS);
970 return;
971 }
972
973 kfree_skb(buff);
974 }
975
tcp_v6_send_reset(const struct sock * sk,struct sk_buff * skb)976 static void tcp_v6_send_reset(const struct sock *sk, struct sk_buff *skb)
977 {
978 const struct tcphdr *th = tcp_hdr(skb);
979 struct ipv6hdr *ipv6h = ipv6_hdr(skb);
980 u32 seq = 0, ack_seq = 0;
981 struct tcp_md5sig_key *key = NULL;
982 #ifdef CONFIG_TCP_MD5SIG
983 const __u8 *hash_location = NULL;
984 unsigned char newhash[16];
985 int genhash;
986 struct sock *sk1 = NULL;
987 #endif
988 __be32 label = 0;
989 u32 priority = 0;
990 struct net *net;
991 u32 txhash = 0;
992 int oif = 0;
993
994 if (th->rst)
995 return;
996
997 /* If sk not NULL, it means we did a successful lookup and incoming
998 * route had to be correct. prequeue might have dropped our dst.
999 */
1000 if (!sk && !ipv6_unicast_destination(skb))
1001 return;
1002
1003 net = sk ? sock_net(sk) : dev_net(skb_dst(skb)->dev);
1004 #ifdef CONFIG_TCP_MD5SIG
1005 rcu_read_lock();
1006 hash_location = tcp_parse_md5sig_option(th);
1007 if (sk && sk_fullsock(sk)) {
1008 int l3index;
1009
1010 /* sdif set, means packet ingressed via a device
1011 * in an L3 domain and inet_iif is set to it.
1012 */
1013 l3index = tcp_v6_sdif(skb) ? tcp_v6_iif_l3_slave(skb) : 0;
1014 key = tcp_v6_md5_do_lookup(sk, &ipv6h->saddr, l3index);
1015 } else if (hash_location) {
1016 int dif = tcp_v6_iif_l3_slave(skb);
1017 int sdif = tcp_v6_sdif(skb);
1018 int l3index;
1019
1020 /*
1021 * active side is lost. Try to find listening socket through
1022 * source port, and then find md5 key through listening socket.
1023 * we are not loose security here:
1024 * Incoming packet is checked with md5 hash with finding key,
1025 * no RST generated if md5 hash doesn't match.
1026 */
1027 sk1 = inet6_lookup_listener(net, net->ipv4.tcp_death_row.hashinfo,
1028 NULL, 0, &ipv6h->saddr, th->source,
1029 &ipv6h->daddr, ntohs(th->source),
1030 dif, sdif);
1031 if (!sk1)
1032 goto out;
1033
1034 /* sdif set, means packet ingressed via a device
1035 * in an L3 domain and dif is set to it.
1036 */
1037 l3index = tcp_v6_sdif(skb) ? dif : 0;
1038
1039 key = tcp_v6_md5_do_lookup(sk1, &ipv6h->saddr, l3index);
1040 if (!key)
1041 goto out;
1042
1043 genhash = tcp_v6_md5_hash_skb(newhash, key, NULL, skb);
1044 if (genhash || memcmp(hash_location, newhash, 16) != 0)
1045 goto out;
1046 }
1047 #endif
1048
1049 if (th->ack)
1050 seq = ntohl(th->ack_seq);
1051 else
1052 ack_seq = ntohl(th->seq) + th->syn + th->fin + skb->len -
1053 (th->doff << 2);
1054
1055 if (sk) {
1056 oif = sk->sk_bound_dev_if;
1057 if (sk_fullsock(sk)) {
1058 const struct ipv6_pinfo *np = tcp_inet6_sk(sk);
1059
1060 trace_tcp_send_reset(sk, skb);
1061 if (np->repflow)
1062 label = ip6_flowlabel(ipv6h);
1063 priority = sk->sk_priority;
1064 txhash = sk->sk_txhash;
1065 }
1066 if (sk->sk_state == TCP_TIME_WAIT) {
1067 label = cpu_to_be32(inet_twsk(sk)->tw_flowlabel);
1068 priority = inet_twsk(sk)->tw_priority;
1069 txhash = inet_twsk(sk)->tw_txhash;
1070 }
1071 } else {
1072 if (net->ipv6.sysctl.flowlabel_reflect & FLOWLABEL_REFLECT_TCP_RESET)
1073 label = ip6_flowlabel(ipv6h);
1074 }
1075
1076 tcp_v6_send_response(sk, skb, seq, ack_seq, 0, 0, 0, oif, key, 1,
1077 ipv6_get_dsfield(ipv6h), label, priority, txhash);
1078
1079 #ifdef CONFIG_TCP_MD5SIG
1080 out:
1081 rcu_read_unlock();
1082 #endif
1083 }
1084
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,u32 txhash)1085 static void tcp_v6_send_ack(const struct sock *sk, struct sk_buff *skb, u32 seq,
1086 u32 ack, u32 win, u32 tsval, u32 tsecr, int oif,
1087 struct tcp_md5sig_key *key, u8 tclass,
1088 __be32 label, u32 priority, u32 txhash)
1089 {
1090 tcp_v6_send_response(sk, skb, seq, ack, win, tsval, tsecr, oif, key, 0,
1091 tclass, label, priority, txhash);
1092 }
1093
tcp_v6_timewait_ack(struct sock * sk,struct sk_buff * skb)1094 static void tcp_v6_timewait_ack(struct sock *sk, struct sk_buff *skb)
1095 {
1096 struct inet_timewait_sock *tw = inet_twsk(sk);
1097 struct tcp_timewait_sock *tcptw = tcp_twsk(sk);
1098
1099 tcp_v6_send_ack(sk, skb, tcptw->tw_snd_nxt, tcptw->tw_rcv_nxt,
1100 tcptw->tw_rcv_wnd >> tw->tw_rcv_wscale,
1101 tcp_time_stamp_raw() + tcptw->tw_ts_offset,
1102 tcptw->tw_ts_recent, tw->tw_bound_dev_if, tcp_twsk_md5_key(tcptw),
1103 tw->tw_tclass, cpu_to_be32(tw->tw_flowlabel), tw->tw_priority,
1104 tw->tw_txhash);
1105
1106 inet_twsk_put(tw);
1107 }
1108
tcp_v6_reqsk_send_ack(const struct sock * sk,struct sk_buff * skb,struct request_sock * req)1109 static void tcp_v6_reqsk_send_ack(const struct sock *sk, struct sk_buff *skb,
1110 struct request_sock *req)
1111 {
1112 int l3index;
1113
1114 l3index = tcp_v6_sdif(skb) ? tcp_v6_iif_l3_slave(skb) : 0;
1115
1116 /* sk->sk_state == TCP_LISTEN -> for regular TCP_SYN_RECV
1117 * sk->sk_state == TCP_SYN_RECV -> for Fast Open.
1118 */
1119 /* RFC 7323 2.3
1120 * The window field (SEG.WND) of every outgoing segment, with the
1121 * exception of <SYN> segments, MUST be right-shifted by
1122 * Rcv.Wind.Shift bits:
1123 */
1124 tcp_v6_send_ack(sk, skb, (sk->sk_state == TCP_LISTEN) ?
1125 tcp_rsk(req)->snt_isn + 1 : tcp_sk(sk)->snd_nxt,
1126 tcp_rsk(req)->rcv_nxt,
1127 req->rsk_rcv_wnd >> inet_rsk(req)->rcv_wscale,
1128 tcp_time_stamp_raw() + tcp_rsk(req)->ts_off,
1129 READ_ONCE(req->ts_recent), sk->sk_bound_dev_if,
1130 tcp_v6_md5_do_lookup(sk, &ipv6_hdr(skb)->saddr, l3index),
1131 ipv6_get_dsfield(ipv6_hdr(skb)), 0,
1132 READ_ONCE(sk->sk_priority),
1133 READ_ONCE(tcp_rsk(req)->txhash));
1134 }
1135
1136
tcp_v6_cookie_check(struct sock * sk,struct sk_buff * skb)1137 static struct sock *tcp_v6_cookie_check(struct sock *sk, struct sk_buff *skb)
1138 {
1139 #ifdef CONFIG_SYN_COOKIES
1140 const struct tcphdr *th = tcp_hdr(skb);
1141
1142 if (!th->syn)
1143 sk = cookie_v6_check(sk, skb);
1144 #endif
1145 return sk;
1146 }
1147
tcp_v6_get_syncookie(struct sock * sk,struct ipv6hdr * iph,struct tcphdr * th,u32 * cookie)1148 u16 tcp_v6_get_syncookie(struct sock *sk, struct ipv6hdr *iph,
1149 struct tcphdr *th, u32 *cookie)
1150 {
1151 u16 mss = 0;
1152 #ifdef CONFIG_SYN_COOKIES
1153 mss = tcp_get_syncookie_mss(&tcp6_request_sock_ops,
1154 &tcp_request_sock_ipv6_ops, sk, th);
1155 if (mss) {
1156 *cookie = __cookie_v6_init_sequence(iph, th, &mss);
1157 tcp_synq_overflow(sk);
1158 }
1159 #endif
1160 return mss;
1161 }
1162
tcp_v6_conn_request(struct sock * sk,struct sk_buff * skb)1163 static int tcp_v6_conn_request(struct sock *sk, struct sk_buff *skb)
1164 {
1165 if (skb->protocol == htons(ETH_P_IP))
1166 return tcp_v4_conn_request(sk, skb);
1167
1168 if (!ipv6_unicast_destination(skb))
1169 goto drop;
1170
1171 if (ipv6_addr_v4mapped(&ipv6_hdr(skb)->saddr)) {
1172 __IP6_INC_STATS(sock_net(sk), NULL, IPSTATS_MIB_INHDRERRORS);
1173 return 0;
1174 }
1175
1176 return tcp_conn_request(&tcp6_request_sock_ops,
1177 &tcp_request_sock_ipv6_ops, sk, skb);
1178
1179 drop:
1180 tcp_listendrop(sk);
1181 return 0; /* don't send reset */
1182 }
1183
tcp_v6_restore_cb(struct sk_buff * skb)1184 static void tcp_v6_restore_cb(struct sk_buff *skb)
1185 {
1186 /* We need to move header back to the beginning if xfrm6_policy_check()
1187 * and tcp_v6_fill_cb() are going to be called again.
1188 * ip6_datagram_recv_specific_ctl() also expects IP6CB to be there.
1189 */
1190 memmove(IP6CB(skb), &TCP_SKB_CB(skb)->header.h6,
1191 sizeof(struct inet6_skb_parm));
1192 }
1193
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)1194 static struct sock *tcp_v6_syn_recv_sock(const struct sock *sk, struct sk_buff *skb,
1195 struct request_sock *req,
1196 struct dst_entry *dst,
1197 struct request_sock *req_unhash,
1198 bool *own_req)
1199 {
1200 struct inet_request_sock *ireq;
1201 struct ipv6_pinfo *newnp;
1202 const struct ipv6_pinfo *np = tcp_inet6_sk(sk);
1203 struct ipv6_txoptions *opt;
1204 struct inet_sock *newinet;
1205 bool found_dup_sk = false;
1206 struct tcp_sock *newtp;
1207 struct sock *newsk;
1208 #ifdef CONFIG_TCP_MD5SIG
1209 struct tcp_md5sig_key *key;
1210 int l3index;
1211 #endif
1212 struct flowi6 fl6;
1213
1214 if (skb->protocol == htons(ETH_P_IP)) {
1215 /*
1216 * v6 mapped
1217 */
1218
1219 newsk = tcp_v4_syn_recv_sock(sk, skb, req, dst,
1220 req_unhash, own_req);
1221
1222 if (!newsk)
1223 return NULL;
1224
1225 inet_sk(newsk)->pinet6 = tcp_inet6_sk(newsk);
1226
1227 newnp = tcp_inet6_sk(newsk);
1228 newtp = tcp_sk(newsk);
1229
1230 memcpy(newnp, np, sizeof(struct ipv6_pinfo));
1231
1232 newnp->saddr = newsk->sk_v6_rcv_saddr;
1233
1234 inet_csk(newsk)->icsk_af_ops = &ipv6_mapped;
1235 if (sk_is_mptcp(newsk))
1236 mptcpv6_handle_mapped(newsk, true);
1237 newsk->sk_backlog_rcv = tcp_v4_do_rcv;
1238 #ifdef CONFIG_TCP_MD5SIG
1239 newtp->af_specific = &tcp_sock_ipv6_mapped_specific;
1240 #endif
1241
1242 newnp->ipv6_mc_list = NULL;
1243 newnp->ipv6_ac_list = NULL;
1244 newnp->ipv6_fl_list = NULL;
1245 newnp->pktoptions = NULL;
1246 newnp->opt = NULL;
1247 newnp->mcast_oif = inet_iif(skb);
1248 newnp->mcast_hops = ip_hdr(skb)->ttl;
1249 newnp->rcv_flowinfo = 0;
1250 if (np->repflow)
1251 newnp->flow_label = 0;
1252
1253 /*
1254 * No need to charge this sock to the relevant IPv6 refcnt debug socks count
1255 * here, tcp_create_openreq_child now does this for us, see the comment in
1256 * that function for the gory details. -acme
1257 */
1258
1259 /* It is tricky place. Until this moment IPv4 tcp
1260 worked with IPv6 icsk.icsk_af_ops.
1261 Sync it now.
1262 */
1263 tcp_sync_mss(newsk, inet_csk(newsk)->icsk_pmtu_cookie);
1264
1265 return newsk;
1266 }
1267
1268 ireq = inet_rsk(req);
1269
1270 if (sk_acceptq_is_full(sk))
1271 goto out_overflow;
1272
1273 if (!dst) {
1274 dst = inet6_csk_route_req(sk, &fl6, req, IPPROTO_TCP);
1275 if (!dst)
1276 goto out;
1277 }
1278
1279 newsk = tcp_create_openreq_child(sk, req, skb);
1280 if (!newsk)
1281 goto out_nonewsk;
1282
1283 /*
1284 * No need to charge this sock to the relevant IPv6 refcnt debug socks
1285 * count here, tcp_create_openreq_child now does this for us, see the
1286 * comment in that function for the gory details. -acme
1287 */
1288
1289 newsk->sk_gso_type = SKB_GSO_TCPV6;
1290 ip6_dst_store(newsk, dst, NULL, NULL);
1291 inet6_sk_rx_dst_set(newsk, skb);
1292
1293 inet_sk(newsk)->pinet6 = tcp_inet6_sk(newsk);
1294
1295 newtp = tcp_sk(newsk);
1296 newinet = inet_sk(newsk);
1297 newnp = tcp_inet6_sk(newsk);
1298
1299 memcpy(newnp, np, sizeof(struct ipv6_pinfo));
1300
1301 newsk->sk_v6_daddr = ireq->ir_v6_rmt_addr;
1302 newnp->saddr = ireq->ir_v6_loc_addr;
1303 newsk->sk_v6_rcv_saddr = ireq->ir_v6_loc_addr;
1304 newsk->sk_bound_dev_if = ireq->ir_iif;
1305
1306 /* Now IPv6 options...
1307
1308 First: no IPv4 options.
1309 */
1310 newinet->inet_opt = NULL;
1311 newnp->ipv6_mc_list = NULL;
1312 newnp->ipv6_ac_list = NULL;
1313 newnp->ipv6_fl_list = NULL;
1314
1315 /* Clone RX bits */
1316 newnp->rxopt.all = np->rxopt.all;
1317
1318 newnp->pktoptions = NULL;
1319 newnp->opt = NULL;
1320 newnp->mcast_oif = tcp_v6_iif(skb);
1321 newnp->mcast_hops = ipv6_hdr(skb)->hop_limit;
1322 newnp->rcv_flowinfo = ip6_flowinfo(ipv6_hdr(skb));
1323 if (np->repflow)
1324 newnp->flow_label = ip6_flowlabel(ipv6_hdr(skb));
1325
1326 /* Set ToS of the new socket based upon the value of incoming SYN.
1327 * ECT bits are set later in tcp_init_transfer().
1328 */
1329 if (READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_reflect_tos))
1330 newnp->tclass = tcp_rsk(req)->syn_tos & ~INET_ECN_MASK;
1331
1332 /* Clone native IPv6 options from listening socket (if any)
1333
1334 Yes, keeping reference count would be much more clever,
1335 but we make one more one thing there: reattach optmem
1336 to newsk.
1337 */
1338 opt = ireq->ipv6_opt;
1339 if (!opt)
1340 opt = rcu_dereference(np->opt);
1341 if (opt) {
1342 opt = ipv6_dup_options(newsk, opt);
1343 RCU_INIT_POINTER(newnp->opt, opt);
1344 }
1345 inet_csk(newsk)->icsk_ext_hdr_len = 0;
1346 if (opt)
1347 inet_csk(newsk)->icsk_ext_hdr_len = opt->opt_nflen +
1348 opt->opt_flen;
1349
1350 tcp_ca_openreq_child(newsk, dst);
1351
1352 tcp_sync_mss(newsk, dst_mtu(dst));
1353 newtp->advmss = tcp_mss_clamp(tcp_sk(sk), dst_metric_advmss(dst));
1354
1355 tcp_initialize_rcv_mss(newsk);
1356
1357 newinet->inet_daddr = newinet->inet_saddr = LOOPBACK4_IPV6;
1358 newinet->inet_rcv_saddr = LOOPBACK4_IPV6;
1359
1360 #ifdef CONFIG_TCP_MD5SIG
1361 l3index = l3mdev_master_ifindex_by_index(sock_net(sk), ireq->ir_iif);
1362
1363 /* Copy over the MD5 key from the original socket */
1364 key = tcp_v6_md5_do_lookup(sk, &newsk->sk_v6_daddr, l3index);
1365 if (key) {
1366 const union tcp_md5_addr *addr;
1367
1368 addr = (union tcp_md5_addr *)&newsk->sk_v6_daddr;
1369 if (tcp_md5_key_copy(newsk, addr, AF_INET6, 128, l3index, key)) {
1370 inet_csk_prepare_forced_close(newsk);
1371 tcp_done(newsk);
1372 goto out;
1373 }
1374 }
1375 #endif
1376
1377 if (__inet_inherit_port(sk, newsk) < 0) {
1378 inet_csk_prepare_forced_close(newsk);
1379 tcp_done(newsk);
1380 goto out;
1381 }
1382 *own_req = inet_ehash_nolisten(newsk, req_to_sk(req_unhash),
1383 &found_dup_sk);
1384 if (*own_req) {
1385 tcp_move_syn(newtp, req);
1386
1387 /* Clone pktoptions received with SYN, if we own the req */
1388 if (ireq->pktopts) {
1389 newnp->pktoptions = skb_clone_and_charge_r(ireq->pktopts, newsk);
1390 consume_skb(ireq->pktopts);
1391 ireq->pktopts = NULL;
1392 if (newnp->pktoptions)
1393 tcp_v6_restore_cb(newnp->pktoptions);
1394 }
1395 } else {
1396 if (!req_unhash && found_dup_sk) {
1397 /* This code path should only be executed in the
1398 * syncookie case only
1399 */
1400 bh_unlock_sock(newsk);
1401 sock_put(newsk);
1402 newsk = NULL;
1403 }
1404 }
1405
1406 return newsk;
1407
1408 out_overflow:
1409 __NET_INC_STATS(sock_net(sk), LINUX_MIB_LISTENOVERFLOWS);
1410 out_nonewsk:
1411 dst_release(dst);
1412 out:
1413 tcp_listendrop(sk);
1414 return NULL;
1415 }
1416
1417 INDIRECT_CALLABLE_DECLARE(struct dst_entry *ipv4_dst_check(struct dst_entry *,
1418 u32));
1419 /* The socket must have it's spinlock held when we get
1420 * here, unless it is a TCP_LISTEN socket.
1421 *
1422 * We have a potential double-lock case here, so even when
1423 * doing backlog processing we use the BH locking scheme.
1424 * This is because we cannot sleep with the original spinlock
1425 * held.
1426 */
1427 INDIRECT_CALLABLE_SCOPE
tcp_v6_do_rcv(struct sock * sk,struct sk_buff * skb)1428 int tcp_v6_do_rcv(struct sock *sk, struct sk_buff *skb)
1429 {
1430 struct ipv6_pinfo *np = tcp_inet6_sk(sk);
1431 struct sk_buff *opt_skb = NULL;
1432 enum skb_drop_reason reason;
1433 struct tcp_sock *tp;
1434
1435 /* Imagine: socket is IPv6. IPv4 packet arrives,
1436 goes to IPv4 receive handler and backlogged.
1437 From backlog it always goes here. Kerboom...
1438 Fortunately, tcp_rcv_established and rcv_established
1439 handle them correctly, but it is not case with
1440 tcp_v6_hnd_req and tcp_v6_send_reset(). --ANK
1441 */
1442
1443 if (skb->protocol == htons(ETH_P_IP))
1444 return tcp_v4_do_rcv(sk, skb);
1445
1446 /*
1447 * socket locking is here for SMP purposes as backlog rcv
1448 * is currently called with bh processing disabled.
1449 */
1450
1451 /* Do Stevens' IPV6_PKTOPTIONS.
1452
1453 Yes, guys, it is the only place in our code, where we
1454 may make it not affecting IPv4.
1455 The rest of code is protocol independent,
1456 and I do not like idea to uglify IPv4.
1457
1458 Actually, all the idea behind IPV6_PKTOPTIONS
1459 looks not very well thought. For now we latch
1460 options, received in the last packet, enqueued
1461 by tcp. Feel free to propose better solution.
1462 --ANK (980728)
1463 */
1464 if (np->rxopt.all)
1465 opt_skb = skb_clone_and_charge_r(skb, sk);
1466
1467 reason = SKB_DROP_REASON_NOT_SPECIFIED;
1468 if (sk->sk_state == TCP_ESTABLISHED) { /* Fast path */
1469 struct dst_entry *dst;
1470
1471 dst = rcu_dereference_protected(sk->sk_rx_dst,
1472 lockdep_sock_is_held(sk));
1473
1474 sock_rps_save_rxhash(sk, skb);
1475 sk_mark_napi_id(sk, skb);
1476 if (dst) {
1477 if (sk->sk_rx_dst_ifindex != skb->skb_iif ||
1478 INDIRECT_CALL_1(dst->ops->check, ip6_dst_check,
1479 dst, sk->sk_rx_dst_cookie) == NULL) {
1480 RCU_INIT_POINTER(sk->sk_rx_dst, NULL);
1481 dst_release(dst);
1482 }
1483 }
1484
1485 tcp_rcv_established(sk, skb);
1486 if (opt_skb)
1487 goto ipv6_pktoptions;
1488 return 0;
1489 }
1490
1491 if (tcp_checksum_complete(skb))
1492 goto csum_err;
1493
1494 if (sk->sk_state == TCP_LISTEN) {
1495 struct sock *nsk = tcp_v6_cookie_check(sk, skb);
1496
1497 if (!nsk)
1498 goto discard;
1499
1500 if (nsk != sk) {
1501 if (tcp_child_process(sk, nsk, skb))
1502 goto reset;
1503 if (opt_skb)
1504 __kfree_skb(opt_skb);
1505 return 0;
1506 }
1507 } else
1508 sock_rps_save_rxhash(sk, skb);
1509
1510 if (tcp_rcv_state_process(sk, skb))
1511 goto reset;
1512 if (opt_skb)
1513 goto ipv6_pktoptions;
1514 return 0;
1515
1516 reset:
1517 tcp_v6_send_reset(sk, skb);
1518 discard:
1519 if (opt_skb)
1520 __kfree_skb(opt_skb);
1521 kfree_skb_reason(skb, reason);
1522 return 0;
1523 csum_err:
1524 reason = SKB_DROP_REASON_TCP_CSUM;
1525 trace_tcp_bad_csum(skb);
1526 TCP_INC_STATS(sock_net(sk), TCP_MIB_CSUMERRORS);
1527 TCP_INC_STATS(sock_net(sk), TCP_MIB_INERRS);
1528 goto discard;
1529
1530
1531 ipv6_pktoptions:
1532 /* Do you ask, what is it?
1533
1534 1. skb was enqueued by tcp.
1535 2. skb is added to tail of read queue, rather than out of order.
1536 3. socket is not in passive state.
1537 4. Finally, it really contains options, which user wants to receive.
1538 */
1539 tp = tcp_sk(sk);
1540 if (TCP_SKB_CB(opt_skb)->end_seq == tp->rcv_nxt &&
1541 !((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_LISTEN))) {
1542 if (np->rxopt.bits.rxinfo || np->rxopt.bits.rxoinfo)
1543 np->mcast_oif = tcp_v6_iif(opt_skb);
1544 if (np->rxopt.bits.rxhlim || np->rxopt.bits.rxohlim)
1545 np->mcast_hops = ipv6_hdr(opt_skb)->hop_limit;
1546 if (np->rxopt.bits.rxflow || np->rxopt.bits.rxtclass)
1547 np->rcv_flowinfo = ip6_flowinfo(ipv6_hdr(opt_skb));
1548 if (np->repflow)
1549 np->flow_label = ip6_flowlabel(ipv6_hdr(opt_skb));
1550 if (ipv6_opt_accepted(sk, opt_skb, &TCP_SKB_CB(opt_skb)->header.h6)) {
1551 tcp_v6_restore_cb(opt_skb);
1552 opt_skb = xchg(&np->pktoptions, opt_skb);
1553 } else {
1554 __kfree_skb(opt_skb);
1555 opt_skb = xchg(&np->pktoptions, NULL);
1556 }
1557 }
1558
1559 consume_skb(opt_skb);
1560 return 0;
1561 }
1562
tcp_v6_fill_cb(struct sk_buff * skb,const struct ipv6hdr * hdr,const struct tcphdr * th)1563 static void tcp_v6_fill_cb(struct sk_buff *skb, const struct ipv6hdr *hdr,
1564 const struct tcphdr *th)
1565 {
1566 /* This is tricky: we move IP6CB at its correct location into
1567 * TCP_SKB_CB(). It must be done after xfrm6_policy_check(), because
1568 * _decode_session6() uses IP6CB().
1569 * barrier() makes sure compiler won't play aliasing games.
1570 */
1571 memmove(&TCP_SKB_CB(skb)->header.h6, IP6CB(skb),
1572 sizeof(struct inet6_skb_parm));
1573 barrier();
1574
1575 TCP_SKB_CB(skb)->seq = ntohl(th->seq);
1576 TCP_SKB_CB(skb)->end_seq = (TCP_SKB_CB(skb)->seq + th->syn + th->fin +
1577 skb->len - th->doff*4);
1578 TCP_SKB_CB(skb)->ack_seq = ntohl(th->ack_seq);
1579 TCP_SKB_CB(skb)->tcp_flags = tcp_flag_byte(th);
1580 TCP_SKB_CB(skb)->tcp_tw_isn = 0;
1581 TCP_SKB_CB(skb)->ip_dsfield = ipv6_get_dsfield(hdr);
1582 TCP_SKB_CB(skb)->sacked = 0;
1583 TCP_SKB_CB(skb)->has_rxtstamp =
1584 skb->tstamp || skb_hwtstamps(skb)->hwtstamp;
1585 }
1586
tcp_v6_rcv(struct sk_buff * skb)1587 INDIRECT_CALLABLE_SCOPE int tcp_v6_rcv(struct sk_buff *skb)
1588 {
1589 enum skb_drop_reason drop_reason;
1590 int sdif = inet6_sdif(skb);
1591 int dif = inet6_iif(skb);
1592 const struct tcphdr *th;
1593 const struct ipv6hdr *hdr;
1594 bool refcounted;
1595 struct sock *sk;
1596 int ret;
1597 struct net *net = dev_net(skb->dev);
1598
1599 drop_reason = SKB_DROP_REASON_NOT_SPECIFIED;
1600 if (skb->pkt_type != PACKET_HOST)
1601 goto discard_it;
1602
1603 /*
1604 * Count it even if it's bad.
1605 */
1606 __TCP_INC_STATS(net, TCP_MIB_INSEGS);
1607
1608 if (!pskb_may_pull(skb, sizeof(struct tcphdr)))
1609 goto discard_it;
1610
1611 th = (const struct tcphdr *)skb->data;
1612
1613 if (unlikely(th->doff < sizeof(struct tcphdr) / 4)) {
1614 drop_reason = SKB_DROP_REASON_PKT_TOO_SMALL;
1615 goto bad_packet;
1616 }
1617 if (!pskb_may_pull(skb, th->doff*4))
1618 goto discard_it;
1619
1620 if (skb_checksum_init(skb, IPPROTO_TCP, ip6_compute_pseudo))
1621 goto csum_error;
1622
1623 th = (const struct tcphdr *)skb->data;
1624 hdr = ipv6_hdr(skb);
1625
1626 lookup:
1627 sk = __inet6_lookup_skb(net->ipv4.tcp_death_row.hashinfo, skb, __tcp_hdrlen(th),
1628 th->source, th->dest, inet6_iif(skb), sdif,
1629 &refcounted);
1630 if (!sk)
1631 goto no_tcp_socket;
1632
1633 process:
1634 if (sk->sk_state == TCP_TIME_WAIT)
1635 goto do_time_wait;
1636
1637 if (sk->sk_state == TCP_NEW_SYN_RECV) {
1638 struct request_sock *req = inet_reqsk(sk);
1639 bool req_stolen = false;
1640 struct sock *nsk;
1641
1642 sk = req->rsk_listener;
1643 if (!xfrm6_policy_check(sk, XFRM_POLICY_IN, skb))
1644 drop_reason = SKB_DROP_REASON_XFRM_POLICY;
1645 else
1646 drop_reason = tcp_inbound_md5_hash(sk, skb,
1647 &hdr->saddr, &hdr->daddr,
1648 AF_INET6, dif, sdif);
1649 if (drop_reason) {
1650 sk_drops_add(sk, skb);
1651 reqsk_put(req);
1652 goto discard_it;
1653 }
1654 if (tcp_checksum_complete(skb)) {
1655 reqsk_put(req);
1656 goto csum_error;
1657 }
1658 if (unlikely(sk->sk_state != TCP_LISTEN)) {
1659 nsk = reuseport_migrate_sock(sk, req_to_sk(req), skb);
1660 if (!nsk) {
1661 inet_csk_reqsk_queue_drop_and_put(sk, req);
1662 goto lookup;
1663 }
1664 sk = nsk;
1665 /* reuseport_migrate_sock() has already held one sk_refcnt
1666 * before returning.
1667 */
1668 } else {
1669 sock_hold(sk);
1670 }
1671 refcounted = true;
1672 nsk = NULL;
1673 if (!tcp_filter(sk, skb)) {
1674 th = (const struct tcphdr *)skb->data;
1675 hdr = ipv6_hdr(skb);
1676 tcp_v6_fill_cb(skb, hdr, th);
1677 nsk = tcp_check_req(sk, skb, req, false, &req_stolen);
1678 } else {
1679 drop_reason = SKB_DROP_REASON_SOCKET_FILTER;
1680 }
1681 if (!nsk) {
1682 reqsk_put(req);
1683 if (req_stolen) {
1684 /* Another cpu got exclusive access to req
1685 * and created a full blown socket.
1686 * Try to feed this packet to this socket
1687 * instead of discarding it.
1688 */
1689 tcp_v6_restore_cb(skb);
1690 sock_put(sk);
1691 goto lookup;
1692 }
1693 goto discard_and_relse;
1694 }
1695 nf_reset_ct(skb);
1696 if (nsk == sk) {
1697 reqsk_put(req);
1698 tcp_v6_restore_cb(skb);
1699 } else if (tcp_child_process(sk, nsk, skb)) {
1700 tcp_v6_send_reset(nsk, skb);
1701 goto discard_and_relse;
1702 } else {
1703 sock_put(sk);
1704 return 0;
1705 }
1706 }
1707
1708 if (static_branch_unlikely(&ip6_min_hopcount)) {
1709 /* min_hopcount can be changed concurrently from do_ipv6_setsockopt() */
1710 if (unlikely(hdr->hop_limit < READ_ONCE(tcp_inet6_sk(sk)->min_hopcount))) {
1711 __NET_INC_STATS(net, LINUX_MIB_TCPMINTTLDROP);
1712 drop_reason = SKB_DROP_REASON_TCP_MINTTL;
1713 goto discard_and_relse;
1714 }
1715 }
1716
1717 if (!xfrm6_policy_check(sk, XFRM_POLICY_IN, skb)) {
1718 drop_reason = SKB_DROP_REASON_XFRM_POLICY;
1719 goto discard_and_relse;
1720 }
1721
1722 drop_reason = tcp_inbound_md5_hash(sk, skb, &hdr->saddr, &hdr->daddr,
1723 AF_INET6, dif, sdif);
1724 if (drop_reason)
1725 goto discard_and_relse;
1726
1727 nf_reset_ct(skb);
1728
1729 if (tcp_filter(sk, skb)) {
1730 drop_reason = SKB_DROP_REASON_SOCKET_FILTER;
1731 goto discard_and_relse;
1732 }
1733 th = (const struct tcphdr *)skb->data;
1734 hdr = ipv6_hdr(skb);
1735 tcp_v6_fill_cb(skb, hdr, th);
1736
1737 skb->dev = NULL;
1738
1739 if (sk->sk_state == TCP_LISTEN) {
1740 ret = tcp_v6_do_rcv(sk, skb);
1741 goto put_and_return;
1742 }
1743
1744 sk_incoming_cpu_update(sk);
1745
1746 bh_lock_sock_nested(sk);
1747 tcp_segs_in(tcp_sk(sk), skb);
1748 ret = 0;
1749 if (!sock_owned_by_user(sk)) {
1750 ret = tcp_v6_do_rcv(sk, skb);
1751 } else {
1752 if (tcp_add_backlog(sk, skb, &drop_reason))
1753 goto discard_and_relse;
1754 }
1755 bh_unlock_sock(sk);
1756 put_and_return:
1757 if (refcounted)
1758 sock_put(sk);
1759 return ret ? -1 : 0;
1760
1761 no_tcp_socket:
1762 drop_reason = SKB_DROP_REASON_NO_SOCKET;
1763 if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb))
1764 goto discard_it;
1765
1766 tcp_v6_fill_cb(skb, hdr, th);
1767
1768 if (tcp_checksum_complete(skb)) {
1769 csum_error:
1770 drop_reason = SKB_DROP_REASON_TCP_CSUM;
1771 trace_tcp_bad_csum(skb);
1772 __TCP_INC_STATS(net, TCP_MIB_CSUMERRORS);
1773 bad_packet:
1774 __TCP_INC_STATS(net, TCP_MIB_INERRS);
1775 } else {
1776 tcp_v6_send_reset(NULL, skb);
1777 }
1778
1779 discard_it:
1780 SKB_DR_OR(drop_reason, NOT_SPECIFIED);
1781 kfree_skb_reason(skb, drop_reason);
1782 return 0;
1783
1784 discard_and_relse:
1785 sk_drops_add(sk, skb);
1786 if (refcounted)
1787 sock_put(sk);
1788 goto discard_it;
1789
1790 do_time_wait:
1791 if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb)) {
1792 drop_reason = SKB_DROP_REASON_XFRM_POLICY;
1793 inet_twsk_put(inet_twsk(sk));
1794 goto discard_it;
1795 }
1796
1797 tcp_v6_fill_cb(skb, hdr, th);
1798
1799 if (tcp_checksum_complete(skb)) {
1800 inet_twsk_put(inet_twsk(sk));
1801 goto csum_error;
1802 }
1803
1804 switch (tcp_timewait_state_process(inet_twsk(sk), skb, th)) {
1805 case TCP_TW_SYN:
1806 {
1807 struct sock *sk2;
1808
1809 sk2 = inet6_lookup_listener(net, net->ipv4.tcp_death_row.hashinfo,
1810 skb, __tcp_hdrlen(th),
1811 &ipv6_hdr(skb)->saddr, th->source,
1812 &ipv6_hdr(skb)->daddr,
1813 ntohs(th->dest),
1814 tcp_v6_iif_l3_slave(skb),
1815 sdif);
1816 if (sk2) {
1817 struct inet_timewait_sock *tw = inet_twsk(sk);
1818 inet_twsk_deschedule_put(tw);
1819 sk = sk2;
1820 tcp_v6_restore_cb(skb);
1821 refcounted = false;
1822 goto process;
1823 }
1824 }
1825 /* to ACK */
1826 fallthrough;
1827 case TCP_TW_ACK:
1828 tcp_v6_timewait_ack(sk, skb);
1829 break;
1830 case TCP_TW_RST:
1831 tcp_v6_send_reset(sk, skb);
1832 inet_twsk_deschedule_put(inet_twsk(sk));
1833 goto discard_it;
1834 case TCP_TW_SUCCESS:
1835 ;
1836 }
1837 goto discard_it;
1838 }
1839
tcp_v6_early_demux(struct sk_buff * skb)1840 void tcp_v6_early_demux(struct sk_buff *skb)
1841 {
1842 struct net *net = dev_net(skb->dev);
1843 const struct ipv6hdr *hdr;
1844 const struct tcphdr *th;
1845 struct sock *sk;
1846
1847 if (skb->pkt_type != PACKET_HOST)
1848 return;
1849
1850 if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct tcphdr)))
1851 return;
1852
1853 hdr = ipv6_hdr(skb);
1854 th = tcp_hdr(skb);
1855
1856 if (th->doff < sizeof(struct tcphdr) / 4)
1857 return;
1858
1859 /* Note : We use inet6_iif() here, not tcp_v6_iif() */
1860 sk = __inet6_lookup_established(net, net->ipv4.tcp_death_row.hashinfo,
1861 &hdr->saddr, th->source,
1862 &hdr->daddr, ntohs(th->dest),
1863 inet6_iif(skb), inet6_sdif(skb));
1864 if (sk) {
1865 skb->sk = sk;
1866 skb->destructor = sock_edemux;
1867 if (sk_fullsock(sk)) {
1868 struct dst_entry *dst = rcu_dereference(sk->sk_rx_dst);
1869
1870 if (dst)
1871 dst = dst_check(dst, sk->sk_rx_dst_cookie);
1872 if (dst &&
1873 sk->sk_rx_dst_ifindex == skb->skb_iif)
1874 skb_dst_set_noref(skb, dst);
1875 }
1876 }
1877 }
1878
1879 static struct timewait_sock_ops tcp6_timewait_sock_ops = {
1880 .twsk_obj_size = sizeof(struct tcp6_timewait_sock),
1881 .twsk_unique = tcp_twsk_unique,
1882 .twsk_destructor = tcp_twsk_destructor,
1883 };
1884
tcp_v6_send_check(struct sock * sk,struct sk_buff * skb)1885 INDIRECT_CALLABLE_SCOPE void tcp_v6_send_check(struct sock *sk, struct sk_buff *skb)
1886 {
1887 __tcp_v6_send_check(skb, &sk->sk_v6_rcv_saddr, &sk->sk_v6_daddr);
1888 }
1889
1890 const struct inet_connection_sock_af_ops ipv6_specific = {
1891 .queue_xmit = inet6_csk_xmit,
1892 .send_check = tcp_v6_send_check,
1893 .rebuild_header = inet6_sk_rebuild_header,
1894 .sk_rx_dst_set = inet6_sk_rx_dst_set,
1895 .conn_request = tcp_v6_conn_request,
1896 .syn_recv_sock = tcp_v6_syn_recv_sock,
1897 .net_header_len = sizeof(struct ipv6hdr),
1898 .net_frag_header_len = sizeof(struct frag_hdr),
1899 .setsockopt = ipv6_setsockopt,
1900 .getsockopt = ipv6_getsockopt,
1901 .addr2sockaddr = inet6_csk_addr2sockaddr,
1902 .sockaddr_len = sizeof(struct sockaddr_in6),
1903 .mtu_reduced = tcp_v6_mtu_reduced,
1904 };
1905
1906 #ifdef CONFIG_TCP_MD5SIG
1907 static const struct tcp_sock_af_ops tcp_sock_ipv6_specific = {
1908 .md5_lookup = tcp_v6_md5_lookup,
1909 .calc_md5_hash = tcp_v6_md5_hash_skb,
1910 .md5_parse = tcp_v6_parse_md5_keys,
1911 };
1912 #endif
1913
1914 /*
1915 * TCP over IPv4 via INET6 API
1916 */
1917 static const struct inet_connection_sock_af_ops ipv6_mapped = {
1918 .queue_xmit = ip_queue_xmit,
1919 .send_check = tcp_v4_send_check,
1920 .rebuild_header = inet_sk_rebuild_header,
1921 .sk_rx_dst_set = inet_sk_rx_dst_set,
1922 .conn_request = tcp_v6_conn_request,
1923 .syn_recv_sock = tcp_v6_syn_recv_sock,
1924 .net_header_len = sizeof(struct iphdr),
1925 .setsockopt = ipv6_setsockopt,
1926 .getsockopt = ipv6_getsockopt,
1927 .addr2sockaddr = inet6_csk_addr2sockaddr,
1928 .sockaddr_len = sizeof(struct sockaddr_in6),
1929 .mtu_reduced = tcp_v4_mtu_reduced,
1930 };
1931
1932 #ifdef CONFIG_TCP_MD5SIG
1933 static const struct tcp_sock_af_ops tcp_sock_ipv6_mapped_specific = {
1934 .md5_lookup = tcp_v4_md5_lookup,
1935 .calc_md5_hash = tcp_v4_md5_hash_skb,
1936 .md5_parse = tcp_v6_parse_md5_keys,
1937 };
1938 #endif
1939
1940 /* NOTE: A lot of things set to zero explicitly by call to
1941 * sk_alloc() so need not be done here.
1942 */
tcp_v6_init_sock(struct sock * sk)1943 static int tcp_v6_init_sock(struct sock *sk)
1944 {
1945 struct inet_connection_sock *icsk = inet_csk(sk);
1946
1947 tcp_init_sock(sk);
1948
1949 icsk->icsk_af_ops = &ipv6_specific;
1950
1951 #ifdef CONFIG_TCP_MD5SIG
1952 tcp_sk(sk)->af_specific = &tcp_sock_ipv6_specific;
1953 #endif
1954
1955 return 0;
1956 }
1957
1958 #ifdef CONFIG_PROC_FS
1959 /* Proc filesystem TCPv6 sock list dumping. */
get_openreq6(struct seq_file * seq,const struct request_sock * req,int i)1960 static void get_openreq6(struct seq_file *seq,
1961 const struct request_sock *req, int i)
1962 {
1963 long ttd = req->rsk_timer.expires - jiffies;
1964 const struct in6_addr *src = &inet_rsk(req)->ir_v6_loc_addr;
1965 const struct in6_addr *dest = &inet_rsk(req)->ir_v6_rmt_addr;
1966
1967 if (ttd < 0)
1968 ttd = 0;
1969
1970 seq_printf(seq,
1971 "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
1972 "%02X %08X:%08X %02X:%08lX %08X %5u %8d %d %d %pK\n",
1973 i,
1974 src->s6_addr32[0], src->s6_addr32[1],
1975 src->s6_addr32[2], src->s6_addr32[3],
1976 inet_rsk(req)->ir_num,
1977 dest->s6_addr32[0], dest->s6_addr32[1],
1978 dest->s6_addr32[2], dest->s6_addr32[3],
1979 ntohs(inet_rsk(req)->ir_rmt_port),
1980 TCP_SYN_RECV,
1981 0, 0, /* could print option size, but that is af dependent. */
1982 1, /* timers active (only the expire timer) */
1983 jiffies_to_clock_t(ttd),
1984 req->num_timeout,
1985 from_kuid_munged(seq_user_ns(seq),
1986 sock_i_uid(req->rsk_listener)),
1987 0, /* non standard timer */
1988 0, /* open_requests have no inode */
1989 0, req);
1990 }
1991
get_tcp6_sock(struct seq_file * seq,struct sock * sp,int i)1992 static void get_tcp6_sock(struct seq_file *seq, struct sock *sp, int i)
1993 {
1994 const struct in6_addr *dest, *src;
1995 __u16 destp, srcp;
1996 int timer_active;
1997 unsigned long timer_expires;
1998 const struct inet_sock *inet = inet_sk(sp);
1999 const struct tcp_sock *tp = tcp_sk(sp);
2000 const struct inet_connection_sock *icsk = inet_csk(sp);
2001 const struct fastopen_queue *fastopenq = &icsk->icsk_accept_queue.fastopenq;
2002 int rx_queue;
2003 int state;
2004
2005 dest = &sp->sk_v6_daddr;
2006 src = &sp->sk_v6_rcv_saddr;
2007 destp = ntohs(inet->inet_dport);
2008 srcp = ntohs(inet->inet_sport);
2009
2010 if (icsk->icsk_pending == ICSK_TIME_RETRANS ||
2011 icsk->icsk_pending == ICSK_TIME_REO_TIMEOUT ||
2012 icsk->icsk_pending == ICSK_TIME_LOSS_PROBE) {
2013 timer_active = 1;
2014 timer_expires = icsk->icsk_timeout;
2015 } else if (icsk->icsk_pending == ICSK_TIME_PROBE0) {
2016 timer_active = 4;
2017 timer_expires = icsk->icsk_timeout;
2018 } else if (timer_pending(&sp->sk_timer)) {
2019 timer_active = 2;
2020 timer_expires = sp->sk_timer.expires;
2021 } else {
2022 timer_active = 0;
2023 timer_expires = jiffies;
2024 }
2025
2026 state = inet_sk_state_load(sp);
2027 if (state == TCP_LISTEN)
2028 rx_queue = READ_ONCE(sp->sk_ack_backlog);
2029 else
2030 /* Because we don't lock the socket,
2031 * we might find a transient negative value.
2032 */
2033 rx_queue = max_t(int, READ_ONCE(tp->rcv_nxt) -
2034 READ_ONCE(tp->copied_seq), 0);
2035
2036 seq_printf(seq,
2037 "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
2038 "%02X %08X:%08X %02X:%08lX %08X %5u %8d %lu %d %pK %lu %lu %u %u %d\n",
2039 i,
2040 src->s6_addr32[0], src->s6_addr32[1],
2041 src->s6_addr32[2], src->s6_addr32[3], srcp,
2042 dest->s6_addr32[0], dest->s6_addr32[1],
2043 dest->s6_addr32[2], dest->s6_addr32[3], destp,
2044 state,
2045 READ_ONCE(tp->write_seq) - tp->snd_una,
2046 rx_queue,
2047 timer_active,
2048 jiffies_delta_to_clock_t(timer_expires - jiffies),
2049 icsk->icsk_retransmits,
2050 from_kuid_munged(seq_user_ns(seq), sock_i_uid(sp)),
2051 icsk->icsk_probes_out,
2052 sock_i_ino(sp),
2053 refcount_read(&sp->sk_refcnt), sp,
2054 jiffies_to_clock_t(icsk->icsk_rto),
2055 jiffies_to_clock_t(icsk->icsk_ack.ato),
2056 (icsk->icsk_ack.quick << 1) | inet_csk_in_pingpong_mode(sp),
2057 tcp_snd_cwnd(tp),
2058 state == TCP_LISTEN ?
2059 fastopenq->max_qlen :
2060 (tcp_in_initial_slowstart(tp) ? -1 : tp->snd_ssthresh)
2061 );
2062 }
2063
get_timewait6_sock(struct seq_file * seq,struct inet_timewait_sock * tw,int i)2064 static void get_timewait6_sock(struct seq_file *seq,
2065 struct inet_timewait_sock *tw, int i)
2066 {
2067 long delta = tw->tw_timer.expires - jiffies;
2068 const struct in6_addr *dest, *src;
2069 __u16 destp, srcp;
2070
2071 dest = &tw->tw_v6_daddr;
2072 src = &tw->tw_v6_rcv_saddr;
2073 destp = ntohs(tw->tw_dport);
2074 srcp = ntohs(tw->tw_sport);
2075
2076 seq_printf(seq,
2077 "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
2078 "%02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %pK\n",
2079 i,
2080 src->s6_addr32[0], src->s6_addr32[1],
2081 src->s6_addr32[2], src->s6_addr32[3], srcp,
2082 dest->s6_addr32[0], dest->s6_addr32[1],
2083 dest->s6_addr32[2], dest->s6_addr32[3], destp,
2084 tw->tw_substate, 0, 0,
2085 3, jiffies_delta_to_clock_t(delta), 0, 0, 0, 0,
2086 refcount_read(&tw->tw_refcnt), tw);
2087 }
2088
tcp6_seq_show(struct seq_file * seq,void * v)2089 static int tcp6_seq_show(struct seq_file *seq, void *v)
2090 {
2091 struct tcp_iter_state *st;
2092 struct sock *sk = v;
2093
2094 if (v == SEQ_START_TOKEN) {
2095 seq_puts(seq,
2096 " sl "
2097 "local_address "
2098 "remote_address "
2099 "st tx_queue rx_queue tr tm->when retrnsmt"
2100 " uid timeout inode\n");
2101 goto out;
2102 }
2103 st = seq->private;
2104
2105 if (sk->sk_state == TCP_TIME_WAIT)
2106 get_timewait6_sock(seq, v, st->num);
2107 else if (sk->sk_state == TCP_NEW_SYN_RECV)
2108 get_openreq6(seq, v, st->num);
2109 else
2110 get_tcp6_sock(seq, v, st->num);
2111 out:
2112 return 0;
2113 }
2114
2115 static const struct seq_operations tcp6_seq_ops = {
2116 .show = tcp6_seq_show,
2117 .start = tcp_seq_start,
2118 .next = tcp_seq_next,
2119 .stop = tcp_seq_stop,
2120 };
2121
2122 static struct tcp_seq_afinfo tcp6_seq_afinfo = {
2123 .family = AF_INET6,
2124 };
2125
tcp6_proc_init(struct net * net)2126 int __net_init tcp6_proc_init(struct net *net)
2127 {
2128 if (!proc_create_net_data("tcp6", 0444, net->proc_net, &tcp6_seq_ops,
2129 sizeof(struct tcp_iter_state), &tcp6_seq_afinfo))
2130 return -ENOMEM;
2131 return 0;
2132 }
2133
tcp6_proc_exit(struct net * net)2134 void tcp6_proc_exit(struct net *net)
2135 {
2136 remove_proc_entry("tcp6", net->proc_net);
2137 }
2138 #endif
2139
2140 struct proto tcpv6_prot = {
2141 .name = "TCPv6",
2142 .owner = THIS_MODULE,
2143 .close = tcp_close,
2144 .pre_connect = tcp_v6_pre_connect,
2145 .connect = tcp_v6_connect,
2146 .disconnect = tcp_disconnect,
2147 .accept = inet_csk_accept,
2148 .ioctl = tcp_ioctl,
2149 .init = tcp_v6_init_sock,
2150 .destroy = tcp_v4_destroy_sock,
2151 .shutdown = tcp_shutdown,
2152 .setsockopt = tcp_setsockopt,
2153 .getsockopt = tcp_getsockopt,
2154 .bpf_bypass_getsockopt = tcp_bpf_bypass_getsockopt,
2155 .keepalive = tcp_set_keepalive,
2156 .recvmsg = tcp_recvmsg,
2157 .sendmsg = tcp_sendmsg,
2158 .splice_eof = tcp_splice_eof,
2159 .backlog_rcv = tcp_v6_do_rcv,
2160 .release_cb = tcp_release_cb,
2161 .hash = inet6_hash,
2162 .unhash = inet_unhash,
2163 .get_port = inet_csk_get_port,
2164 .put_port = inet_put_port,
2165 #ifdef CONFIG_BPF_SYSCALL
2166 .psock_update_sk_prot = tcp_bpf_update_proto,
2167 #endif
2168 .enter_memory_pressure = tcp_enter_memory_pressure,
2169 .leave_memory_pressure = tcp_leave_memory_pressure,
2170 .stream_memory_free = tcp_stream_memory_free,
2171 .sockets_allocated = &tcp_sockets_allocated,
2172
2173 .memory_allocated = &tcp_memory_allocated,
2174 .per_cpu_fw_alloc = &tcp_memory_per_cpu_fw_alloc,
2175
2176 .memory_pressure = &tcp_memory_pressure,
2177 .orphan_count = &tcp_orphan_count,
2178 .sysctl_mem = sysctl_tcp_mem,
2179 .sysctl_wmem_offset = offsetof(struct net, ipv4.sysctl_tcp_wmem),
2180 .sysctl_rmem_offset = offsetof(struct net, ipv4.sysctl_tcp_rmem),
2181 .max_header = MAX_TCP_HEADER,
2182 .obj_size = sizeof(struct tcp6_sock),
2183 .ipv6_pinfo_offset = offsetof(struct tcp6_sock, inet6),
2184 .slab_flags = SLAB_TYPESAFE_BY_RCU,
2185 .twsk_prot = &tcp6_timewait_sock_ops,
2186 .rsk_prot = &tcp6_request_sock_ops,
2187 .h.hashinfo = NULL,
2188 .no_autobind = true,
2189 .diag_destroy = tcp_abort,
2190 };
2191 EXPORT_SYMBOL_GPL(tcpv6_prot);
2192
2193 static const struct inet6_protocol tcpv6_protocol = {
2194 .handler = tcp_v6_rcv,
2195 .err_handler = tcp_v6_err,
2196 .flags = INET6_PROTO_NOPOLICY|INET6_PROTO_FINAL,
2197 };
2198
2199 static struct inet_protosw tcpv6_protosw = {
2200 .type = SOCK_STREAM,
2201 .protocol = IPPROTO_TCP,
2202 .prot = &tcpv6_prot,
2203 .ops = &inet6_stream_ops,
2204 .flags = INET_PROTOSW_PERMANENT |
2205 INET_PROTOSW_ICSK,
2206 };
2207
tcpv6_net_init(struct net * net)2208 static int __net_init tcpv6_net_init(struct net *net)
2209 {
2210 return inet_ctl_sock_create(&net->ipv6.tcp_sk, PF_INET6,
2211 SOCK_RAW, IPPROTO_TCP, net);
2212 }
2213
tcpv6_net_exit(struct net * net)2214 static void __net_exit tcpv6_net_exit(struct net *net)
2215 {
2216 inet_ctl_sock_destroy(net->ipv6.tcp_sk);
2217 }
2218
tcpv6_net_exit_batch(struct list_head * net_exit_list)2219 static void __net_exit tcpv6_net_exit_batch(struct list_head *net_exit_list)
2220 {
2221 tcp_twsk_purge(net_exit_list, AF_INET6);
2222 }
2223
2224 static struct pernet_operations tcpv6_net_ops = {
2225 .init = tcpv6_net_init,
2226 .exit = tcpv6_net_exit,
2227 .exit_batch = tcpv6_net_exit_batch,
2228 };
2229
tcpv6_init(void)2230 int __init tcpv6_init(void)
2231 {
2232 int ret;
2233
2234 ret = inet6_add_protocol(&tcpv6_protocol, IPPROTO_TCP);
2235 if (ret)
2236 goto out;
2237
2238 /* register inet6 protocol */
2239 ret = inet6_register_protosw(&tcpv6_protosw);
2240 if (ret)
2241 goto out_tcpv6_protocol;
2242
2243 ret = register_pernet_subsys(&tcpv6_net_ops);
2244 if (ret)
2245 goto out_tcpv6_protosw;
2246
2247 ret = mptcpv6_init();
2248 if (ret)
2249 goto out_tcpv6_pernet_subsys;
2250
2251 out:
2252 return ret;
2253
2254 out_tcpv6_pernet_subsys:
2255 unregister_pernet_subsys(&tcpv6_net_ops);
2256 out_tcpv6_protosw:
2257 inet6_unregister_protosw(&tcpv6_protosw);
2258 out_tcpv6_protocol:
2259 inet6_del_protocol(&tcpv6_protocol, IPPROTO_TCP);
2260 goto out;
2261 }
2262
tcpv6_exit(void)2263 void tcpv6_exit(void)
2264 {
2265 unregister_pernet_subsys(&tcpv6_net_ops);
2266 inet6_unregister_protosw(&tcpv6_protosw);
2267 inet6_del_protocol(&tcpv6_protocol, IPPROTO_TCP);
2268 }
2269