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 */
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 rcu_assign_pointer(sk->sk_rx_dst, dst);
111 sk->sk_rx_dst_ifindex = skb->skb_iif;
112 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_connection_sock *icsk = inet_csk(sk);
150 struct in6_addr *saddr = NULL, *final_p, final;
151 struct inet_timewait_death_row *tcp_death_row;
152 struct ipv6_pinfo *np = tcp_inet6_sk(sk);
153 struct inet_sock *inet = inet_sk(sk);
154 struct tcp_sock *tp = tcp_sk(sk);
155 struct net *net = sock_net(sk);
156 struct ipv6_txoptions *opt;
157 struct dst_entry *dst;
158 struct flowi6 fl6;
159 int addr_type;
160 int err;
161
162 if (addr_len < SIN6_LEN_RFC2133)
163 return -EINVAL;
164
165 if (usin->sin6_family != AF_INET6)
166 return -EAFNOSUPPORT;
167
168 memset(&fl6, 0, sizeof(fl6));
169
170 if (np->sndflow) {
171 fl6.flowlabel = usin->sin6_flowinfo&IPV6_FLOWINFO_MASK;
172 IP6_ECN_flow_init(fl6.flowlabel);
173 if (fl6.flowlabel&IPV6_FLOWLABEL_MASK) {
174 struct ip6_flowlabel *flowlabel;
175 flowlabel = fl6_sock_lookup(sk, fl6.flowlabel);
176 if (IS_ERR(flowlabel))
177 return -EINVAL;
178 fl6_sock_release(flowlabel);
179 }
180 }
181
182 /*
183 * connect() to INADDR_ANY means loopback (BSD'ism).
184 */
185
186 if (ipv6_addr_any(&usin->sin6_addr)) {
187 if (ipv6_addr_v4mapped(&sk->sk_v6_rcv_saddr))
188 ipv6_addr_set_v4mapped(htonl(INADDR_LOOPBACK),
189 &usin->sin6_addr);
190 else
191 usin->sin6_addr = in6addr_loopback;
192 }
193
194 addr_type = ipv6_addr_type(&usin->sin6_addr);
195
196 if (addr_type & IPV6_ADDR_MULTICAST)
197 return -ENETUNREACH;
198
199 if (addr_type&IPV6_ADDR_LINKLOCAL) {
200 if (addr_len >= sizeof(struct sockaddr_in6) &&
201 usin->sin6_scope_id) {
202 /* If interface is set while binding, indices
203 * must coincide.
204 */
205 if (!sk_dev_equal_l3scope(sk, usin->sin6_scope_id))
206 return -EINVAL;
207
208 sk->sk_bound_dev_if = usin->sin6_scope_id;
209 }
210
211 /* Connect to link-local address requires an interface */
212 if (!sk->sk_bound_dev_if)
213 return -EINVAL;
214 }
215
216 if (tp->rx_opt.ts_recent_stamp &&
217 !ipv6_addr_equal(&sk->sk_v6_daddr, &usin->sin6_addr)) {
218 tp->rx_opt.ts_recent = 0;
219 tp->rx_opt.ts_recent_stamp = 0;
220 WRITE_ONCE(tp->write_seq, 0);
221 }
222
223 sk->sk_v6_daddr = usin->sin6_addr;
224 np->flow_label = fl6.flowlabel;
225
226 /*
227 * TCP over IPv4
228 */
229
230 if (addr_type & IPV6_ADDR_MAPPED) {
231 u32 exthdrlen = icsk->icsk_ext_hdr_len;
232 struct sockaddr_in sin;
233
234 if (ipv6_only_sock(sk))
235 return -ENETUNREACH;
236
237 sin.sin_family = AF_INET;
238 sin.sin_port = usin->sin6_port;
239 sin.sin_addr.s_addr = usin->sin6_addr.s6_addr32[3];
240
241 /* Paired with READ_ONCE() in tcp_(get|set)sockopt() */
242 WRITE_ONCE(icsk->icsk_af_ops, &ipv6_mapped);
243 if (sk_is_mptcp(sk))
244 mptcpv6_handle_mapped(sk, true);
245 sk->sk_backlog_rcv = tcp_v4_do_rcv;
246 #ifdef CONFIG_TCP_MD5SIG
247 tp->af_specific = &tcp_sock_ipv6_mapped_specific;
248 #endif
249
250 err = tcp_v4_connect(sk, (struct sockaddr *)&sin, sizeof(sin));
251
252 if (err) {
253 icsk->icsk_ext_hdr_len = exthdrlen;
254 /* Paired with READ_ONCE() in tcp_(get|set)sockopt() */
255 WRITE_ONCE(icsk->icsk_af_ops, &ipv6_specific);
256 if (sk_is_mptcp(sk))
257 mptcpv6_handle_mapped(sk, false);
258 sk->sk_backlog_rcv = tcp_v6_do_rcv;
259 #ifdef CONFIG_TCP_MD5SIG
260 tp->af_specific = &tcp_sock_ipv6_specific;
261 #endif
262 goto failure;
263 }
264 np->saddr = sk->sk_v6_rcv_saddr;
265
266 return err;
267 }
268
269 if (!ipv6_addr_any(&sk->sk_v6_rcv_saddr))
270 saddr = &sk->sk_v6_rcv_saddr;
271
272 fl6.flowi6_proto = IPPROTO_TCP;
273 fl6.daddr = sk->sk_v6_daddr;
274 fl6.saddr = saddr ? *saddr : np->saddr;
275 fl6.flowi6_oif = sk->sk_bound_dev_if;
276 fl6.flowi6_mark = sk->sk_mark;
277 fl6.fl6_dport = usin->sin6_port;
278 fl6.fl6_sport = inet->inet_sport;
279 fl6.flowi6_uid = sk->sk_uid;
280
281 opt = rcu_dereference_protected(np->opt, lockdep_sock_is_held(sk));
282 final_p = fl6_update_dst(&fl6, opt, &final);
283
284 security_sk_classify_flow(sk, flowi6_to_flowi_common(&fl6));
285
286 dst = ip6_dst_lookup_flow(net, sk, &fl6, final_p);
287 if (IS_ERR(dst)) {
288 err = PTR_ERR(dst);
289 goto failure;
290 }
291
292 tcp_death_row = &sock_net(sk)->ipv4.tcp_death_row;
293
294 if (!saddr) {
295 saddr = &fl6.saddr;
296
297 err = inet_bhash2_update_saddr(sk, saddr, AF_INET6);
298 if (err)
299 goto failure;
300 }
301
302 /* set the source address */
303 np->saddr = *saddr;
304 inet->inet_rcv_saddr = LOOPBACK4_IPV6;
305
306 sk->sk_gso_type = SKB_GSO_TCPV6;
307 ip6_dst_store(sk, dst, NULL, NULL);
308
309 icsk->icsk_ext_hdr_len = 0;
310 if (opt)
311 icsk->icsk_ext_hdr_len = opt->opt_flen +
312 opt->opt_nflen;
313
314 tp->rx_opt.mss_clamp = IPV6_MIN_MTU - sizeof(struct tcphdr) - sizeof(struct ipv6hdr);
315
316 inet->inet_dport = usin->sin6_port;
317
318 tcp_set_state(sk, TCP_SYN_SENT);
319 err = inet6_hash_connect(tcp_death_row, sk);
320 if (err)
321 goto late_failure;
322
323 sk_set_txhash(sk);
324
325 if (likely(!tp->repair)) {
326 if (!tp->write_seq)
327 WRITE_ONCE(tp->write_seq,
328 secure_tcpv6_seq(np->saddr.s6_addr32,
329 sk->sk_v6_daddr.s6_addr32,
330 inet->inet_sport,
331 inet->inet_dport));
332 tp->tsoffset = secure_tcpv6_ts_off(net, np->saddr.s6_addr32,
333 sk->sk_v6_daddr.s6_addr32);
334 }
335
336 if (tcp_fastopen_defer_connect(sk, &err))
337 return err;
338 if (err)
339 goto late_failure;
340
341 err = tcp_connect(sk);
342 if (err)
343 goto late_failure;
344
345 return 0;
346
347 late_failure:
348 tcp_set_state(sk, TCP_CLOSE);
349 inet_bhash2_reset_saddr(sk);
350 failure:
351 inet->inet_dport = 0;
352 sk->sk_route_caps = 0;
353 return err;
354 }
355
tcp_v6_mtu_reduced(struct sock * sk)356 static void tcp_v6_mtu_reduced(struct sock *sk)
357 {
358 struct dst_entry *dst;
359 u32 mtu;
360
361 if ((1 << sk->sk_state) & (TCPF_LISTEN | TCPF_CLOSE))
362 return;
363
364 mtu = READ_ONCE(tcp_sk(sk)->mtu_info);
365
366 /* Drop requests trying to increase our current mss.
367 * Check done in __ip6_rt_update_pmtu() is too late.
368 */
369 if (tcp_mtu_to_mss(sk, mtu) >= tcp_sk(sk)->mss_cache)
370 return;
371
372 dst = inet6_csk_update_pmtu(sk, mtu);
373 if (!dst)
374 return;
375
376 if (inet_csk(sk)->icsk_pmtu_cookie > dst_mtu(dst)) {
377 tcp_sync_mss(sk, dst_mtu(dst));
378 tcp_simple_retransmit(sk);
379 }
380 }
381
tcp_v6_err(struct sk_buff * skb,struct inet6_skb_parm * opt,u8 type,u8 code,int offset,__be32 info)382 static int tcp_v6_err(struct sk_buff *skb, struct inet6_skb_parm *opt,
383 u8 type, u8 code, int offset, __be32 info)
384 {
385 const struct ipv6hdr *hdr = (const struct ipv6hdr *)skb->data;
386 const struct tcphdr *th = (struct tcphdr *)(skb->data+offset);
387 struct net *net = dev_net(skb->dev);
388 struct request_sock *fastopen;
389 struct ipv6_pinfo *np;
390 struct tcp_sock *tp;
391 __u32 seq, snd_una;
392 struct sock *sk;
393 bool fatal;
394 int err;
395
396 sk = __inet6_lookup_established(net, net->ipv4.tcp_death_row.hashinfo,
397 &hdr->daddr, th->dest,
398 &hdr->saddr, ntohs(th->source),
399 skb->dev->ifindex, inet6_sdif(skb));
400
401 if (!sk) {
402 __ICMP6_INC_STATS(net, __in6_dev_get(skb->dev),
403 ICMP6_MIB_INERRORS);
404 return -ENOENT;
405 }
406
407 if (sk->sk_state == TCP_TIME_WAIT) {
408 inet_twsk_put(inet_twsk(sk));
409 return 0;
410 }
411 seq = ntohl(th->seq);
412 fatal = icmpv6_err_convert(type, code, &err);
413 if (sk->sk_state == TCP_NEW_SYN_RECV) {
414 tcp_req_err(sk, seq, fatal);
415 return 0;
416 }
417
418 bh_lock_sock(sk);
419 if (sock_owned_by_user(sk) && type != ICMPV6_PKT_TOOBIG)
420 __NET_INC_STATS(net, LINUX_MIB_LOCKDROPPEDICMPS);
421
422 if (sk->sk_state == TCP_CLOSE)
423 goto out;
424
425 if (static_branch_unlikely(&ip6_min_hopcount)) {
426 /* min_hopcount can be changed concurrently from do_ipv6_setsockopt() */
427 if (ipv6_hdr(skb)->hop_limit < READ_ONCE(tcp_inet6_sk(sk)->min_hopcount)) {
428 __NET_INC_STATS(net, LINUX_MIB_TCPMINTTLDROP);
429 goto out;
430 }
431 }
432
433 tp = tcp_sk(sk);
434 /* XXX (TFO) - tp->snd_una should be ISN (tcp_create_openreq_child() */
435 fastopen = rcu_dereference(tp->fastopen_rsk);
436 snd_una = fastopen ? tcp_rsk(fastopen)->snt_isn : tp->snd_una;
437 if (sk->sk_state != TCP_LISTEN &&
438 !between(seq, snd_una, tp->snd_nxt)) {
439 __NET_INC_STATS(net, LINUX_MIB_OUTOFWINDOWICMPS);
440 goto out;
441 }
442
443 np = tcp_inet6_sk(sk);
444
445 if (type == NDISC_REDIRECT) {
446 if (!sock_owned_by_user(sk)) {
447 struct dst_entry *dst = __sk_dst_check(sk, np->dst_cookie);
448
449 if (dst)
450 dst->ops->redirect(dst, sk, skb);
451 }
452 goto out;
453 }
454
455 if (type == ICMPV6_PKT_TOOBIG) {
456 u32 mtu = ntohl(info);
457
458 /* We are not interested in TCP_LISTEN and open_requests
459 * (SYN-ACKs send out by Linux are always <576bytes so
460 * they should go through unfragmented).
461 */
462 if (sk->sk_state == TCP_LISTEN)
463 goto out;
464
465 if (!ip6_sk_accept_pmtu(sk))
466 goto out;
467
468 if (mtu < IPV6_MIN_MTU)
469 goto out;
470
471 WRITE_ONCE(tp->mtu_info, mtu);
472
473 if (!sock_owned_by_user(sk))
474 tcp_v6_mtu_reduced(sk);
475 else if (!test_and_set_bit(TCP_MTU_REDUCED_DEFERRED,
476 &sk->sk_tsq_flags))
477 sock_hold(sk);
478 goto out;
479 }
480
481
482 /* Might be for an request_sock */
483 switch (sk->sk_state) {
484 case TCP_SYN_SENT:
485 case TCP_SYN_RECV:
486 /* Only in fast or simultaneous open. If a fast open socket is
487 * already accepted it is treated as a connected one below.
488 */
489 if (fastopen && !fastopen->sk)
490 break;
491
492 ipv6_icmp_error(sk, skb, err, th->dest, ntohl(info), (u8 *)th);
493
494 if (!sock_owned_by_user(sk)) {
495 sk->sk_err = err;
496 sk_error_report(sk); /* Wake people up to see the error (see connect in sock.c) */
497
498 tcp_done(sk);
499 } else
500 sk->sk_err_soft = err;
501 goto out;
502 case TCP_LISTEN:
503 break;
504 default:
505 /* check if this ICMP message allows revert of backoff.
506 * (see RFC 6069)
507 */
508 if (!fastopen && type == ICMPV6_DEST_UNREACH &&
509 code == ICMPV6_NOROUTE)
510 tcp_ld_RTO_revert(sk, seq);
511 }
512
513 if (!sock_owned_by_user(sk) && np->recverr) {
514 sk->sk_err = err;
515 sk_error_report(sk);
516 } else
517 sk->sk_err_soft = err;
518
519 out:
520 bh_unlock_sock(sk);
521 sock_put(sk);
522 return 0;
523 }
524
525
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)526 static int tcp_v6_send_synack(const struct sock *sk, struct dst_entry *dst,
527 struct flowi *fl,
528 struct request_sock *req,
529 struct tcp_fastopen_cookie *foc,
530 enum tcp_synack_type synack_type,
531 struct sk_buff *syn_skb)
532 {
533 struct inet_request_sock *ireq = inet_rsk(req);
534 struct ipv6_pinfo *np = tcp_inet6_sk(sk);
535 struct ipv6_txoptions *opt;
536 struct flowi6 *fl6 = &fl->u.ip6;
537 struct sk_buff *skb;
538 int err = -ENOMEM;
539 u8 tclass;
540
541 /* First, grab a route. */
542 if (!dst && (dst = inet6_csk_route_req(sk, fl6, req,
543 IPPROTO_TCP)) == NULL)
544 goto done;
545
546 skb = tcp_make_synack(sk, dst, req, foc, synack_type, syn_skb);
547
548 if (skb) {
549 __tcp_v6_send_check(skb, &ireq->ir_v6_loc_addr,
550 &ireq->ir_v6_rmt_addr);
551
552 fl6->daddr = ireq->ir_v6_rmt_addr;
553 if (np->repflow && ireq->pktopts)
554 fl6->flowlabel = ip6_flowlabel(ipv6_hdr(ireq->pktopts));
555
556 tclass = READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_reflect_tos) ?
557 (tcp_rsk(req)->syn_tos & ~INET_ECN_MASK) |
558 (np->tclass & INET_ECN_MASK) :
559 np->tclass;
560
561 if (!INET_ECN_is_capable(tclass) &&
562 tcp_bpf_ca_needs_ecn((struct sock *)req))
563 tclass |= INET_ECN_ECT_0;
564
565 rcu_read_lock();
566 opt = ireq->ipv6_opt;
567 if (!opt)
568 opt = rcu_dereference(np->opt);
569 err = ip6_xmit(sk, skb, fl6, skb->mark ? : sk->sk_mark, opt,
570 tclass, sk->sk_priority);
571 rcu_read_unlock();
572 err = net_xmit_eval(err);
573 }
574
575 done:
576 return err;
577 }
578
579
tcp_v6_reqsk_destructor(struct request_sock * req)580 static void tcp_v6_reqsk_destructor(struct request_sock *req)
581 {
582 kfree(inet_rsk(req)->ipv6_opt);
583 consume_skb(inet_rsk(req)->pktopts);
584 }
585
586 #ifdef CONFIG_TCP_MD5SIG
tcp_v6_md5_do_lookup(const struct sock * sk,const struct in6_addr * addr,int l3index)587 static struct tcp_md5sig_key *tcp_v6_md5_do_lookup(const struct sock *sk,
588 const struct in6_addr *addr,
589 int l3index)
590 {
591 return tcp_md5_do_lookup(sk, l3index,
592 (union tcp_md5_addr *)addr, AF_INET6);
593 }
594
tcp_v6_md5_lookup(const struct sock * sk,const struct sock * addr_sk)595 static struct tcp_md5sig_key *tcp_v6_md5_lookup(const struct sock *sk,
596 const struct sock *addr_sk)
597 {
598 int l3index;
599
600 l3index = l3mdev_master_ifindex_by_index(sock_net(sk),
601 addr_sk->sk_bound_dev_if);
602 return tcp_v6_md5_do_lookup(sk, &addr_sk->sk_v6_daddr,
603 l3index);
604 }
605
tcp_v6_parse_md5_keys(struct sock * sk,int optname,sockptr_t optval,int optlen)606 static int tcp_v6_parse_md5_keys(struct sock *sk, int optname,
607 sockptr_t optval, int optlen)
608 {
609 struct tcp_md5sig cmd;
610 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)&cmd.tcpm_addr;
611 int l3index = 0;
612 u8 prefixlen;
613 u8 flags;
614
615 if (optlen < sizeof(cmd))
616 return -EINVAL;
617
618 if (copy_from_sockptr(&cmd, optval, sizeof(cmd)))
619 return -EFAULT;
620
621 if (sin6->sin6_family != AF_INET6)
622 return -EINVAL;
623
624 flags = cmd.tcpm_flags & TCP_MD5SIG_FLAG_IFINDEX;
625
626 if (optname == TCP_MD5SIG_EXT &&
627 cmd.tcpm_flags & TCP_MD5SIG_FLAG_PREFIX) {
628 prefixlen = cmd.tcpm_prefixlen;
629 if (prefixlen > 128 || (ipv6_addr_v4mapped(&sin6->sin6_addr) &&
630 prefixlen > 32))
631 return -EINVAL;
632 } else {
633 prefixlen = ipv6_addr_v4mapped(&sin6->sin6_addr) ? 32 : 128;
634 }
635
636 if (optname == TCP_MD5SIG_EXT && cmd.tcpm_ifindex &&
637 cmd.tcpm_flags & TCP_MD5SIG_FLAG_IFINDEX) {
638 struct net_device *dev;
639
640 rcu_read_lock();
641 dev = dev_get_by_index_rcu(sock_net(sk), cmd.tcpm_ifindex);
642 if (dev && netif_is_l3_master(dev))
643 l3index = dev->ifindex;
644 rcu_read_unlock();
645
646 /* ok to reference set/not set outside of rcu;
647 * right now device MUST be an L3 master
648 */
649 if (!dev || !l3index)
650 return -EINVAL;
651 }
652
653 if (!cmd.tcpm_keylen) {
654 if (ipv6_addr_v4mapped(&sin6->sin6_addr))
655 return tcp_md5_do_del(sk, (union tcp_md5_addr *)&sin6->sin6_addr.s6_addr32[3],
656 AF_INET, prefixlen,
657 l3index, flags);
658 return tcp_md5_do_del(sk, (union tcp_md5_addr *)&sin6->sin6_addr,
659 AF_INET6, prefixlen, l3index, flags);
660 }
661
662 if (cmd.tcpm_keylen > TCP_MD5SIG_MAXKEYLEN)
663 return -EINVAL;
664
665 if (ipv6_addr_v4mapped(&sin6->sin6_addr))
666 return tcp_md5_do_add(sk, (union tcp_md5_addr *)&sin6->sin6_addr.s6_addr32[3],
667 AF_INET, prefixlen, l3index, flags,
668 cmd.tcpm_key, cmd.tcpm_keylen,
669 GFP_KERNEL);
670
671 return tcp_md5_do_add(sk, (union tcp_md5_addr *)&sin6->sin6_addr,
672 AF_INET6, prefixlen, l3index, flags,
673 cmd.tcpm_key, cmd.tcpm_keylen, GFP_KERNEL);
674 }
675
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)676 static int tcp_v6_md5_hash_headers(struct tcp_md5sig_pool *hp,
677 const struct in6_addr *daddr,
678 const struct in6_addr *saddr,
679 const struct tcphdr *th, int nbytes)
680 {
681 struct tcp6_pseudohdr *bp;
682 struct scatterlist sg;
683 struct tcphdr *_th;
684
685 bp = hp->scratch;
686 /* 1. TCP pseudo-header (RFC2460) */
687 bp->saddr = *saddr;
688 bp->daddr = *daddr;
689 bp->protocol = cpu_to_be32(IPPROTO_TCP);
690 bp->len = cpu_to_be32(nbytes);
691
692 _th = (struct tcphdr *)(bp + 1);
693 memcpy(_th, th, sizeof(*th));
694 _th->check = 0;
695
696 sg_init_one(&sg, bp, sizeof(*bp) + sizeof(*th));
697 ahash_request_set_crypt(hp->md5_req, &sg, NULL,
698 sizeof(*bp) + sizeof(*th));
699 return crypto_ahash_update(hp->md5_req);
700 }
701
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)702 static int tcp_v6_md5_hash_hdr(char *md5_hash, const struct tcp_md5sig_key *key,
703 const struct in6_addr *daddr, struct in6_addr *saddr,
704 const struct tcphdr *th)
705 {
706 struct tcp_md5sig_pool *hp;
707 struct ahash_request *req;
708
709 hp = tcp_get_md5sig_pool();
710 if (!hp)
711 goto clear_hash_noput;
712 req = hp->md5_req;
713
714 if (crypto_ahash_init(req))
715 goto clear_hash;
716 if (tcp_v6_md5_hash_headers(hp, daddr, saddr, th, th->doff << 2))
717 goto clear_hash;
718 if (tcp_md5_hash_key(hp, key))
719 goto clear_hash;
720 ahash_request_set_crypt(req, NULL, md5_hash, 0);
721 if (crypto_ahash_final(req))
722 goto clear_hash;
723
724 tcp_put_md5sig_pool();
725 return 0;
726
727 clear_hash:
728 tcp_put_md5sig_pool();
729 clear_hash_noput:
730 memset(md5_hash, 0, 16);
731 return 1;
732 }
733
tcp_v6_md5_hash_skb(char * md5_hash,const struct tcp_md5sig_key * key,const struct sock * sk,const struct sk_buff * skb)734 static int tcp_v6_md5_hash_skb(char *md5_hash,
735 const struct tcp_md5sig_key *key,
736 const struct sock *sk,
737 const struct sk_buff *skb)
738 {
739 const struct in6_addr *saddr, *daddr;
740 struct tcp_md5sig_pool *hp;
741 struct ahash_request *req;
742 const struct tcphdr *th = tcp_hdr(skb);
743
744 if (sk) { /* valid for establish/request sockets */
745 saddr = &sk->sk_v6_rcv_saddr;
746 daddr = &sk->sk_v6_daddr;
747 } else {
748 const struct ipv6hdr *ip6h = ipv6_hdr(skb);
749 saddr = &ip6h->saddr;
750 daddr = &ip6h->daddr;
751 }
752
753 hp = tcp_get_md5sig_pool();
754 if (!hp)
755 goto clear_hash_noput;
756 req = hp->md5_req;
757
758 if (crypto_ahash_init(req))
759 goto clear_hash;
760
761 if (tcp_v6_md5_hash_headers(hp, daddr, saddr, th, skb->len))
762 goto clear_hash;
763 if (tcp_md5_hash_skb_data(hp, skb, th->doff << 2))
764 goto clear_hash;
765 if (tcp_md5_hash_key(hp, key))
766 goto clear_hash;
767 ahash_request_set_crypt(req, NULL, md5_hash, 0);
768 if (crypto_ahash_final(req))
769 goto clear_hash;
770
771 tcp_put_md5sig_pool();
772 return 0;
773
774 clear_hash:
775 tcp_put_md5sig_pool();
776 clear_hash_noput:
777 memset(md5_hash, 0, 16);
778 return 1;
779 }
780
781 #endif
782
tcp_v6_init_req(struct request_sock * req,const struct sock * sk_listener,struct sk_buff * skb)783 static void tcp_v6_init_req(struct request_sock *req,
784 const struct sock *sk_listener,
785 struct sk_buff *skb)
786 {
787 bool l3_slave = ipv6_l3mdev_skb(TCP_SKB_CB(skb)->header.h6.flags);
788 struct inet_request_sock *ireq = inet_rsk(req);
789 const struct ipv6_pinfo *np = tcp_inet6_sk(sk_listener);
790
791 ireq->ir_v6_rmt_addr = ipv6_hdr(skb)->saddr;
792 ireq->ir_v6_loc_addr = ipv6_hdr(skb)->daddr;
793
794 /* So that link locals have meaning */
795 if ((!sk_listener->sk_bound_dev_if || l3_slave) &&
796 ipv6_addr_type(&ireq->ir_v6_rmt_addr) & IPV6_ADDR_LINKLOCAL)
797 ireq->ir_iif = tcp_v6_iif(skb);
798
799 if (!TCP_SKB_CB(skb)->tcp_tw_isn &&
800 (ipv6_opt_accepted(sk_listener, skb, &TCP_SKB_CB(skb)->header.h6) ||
801 np->rxopt.bits.rxinfo ||
802 np->rxopt.bits.rxoinfo || np->rxopt.bits.rxhlim ||
803 np->rxopt.bits.rxohlim || np->repflow)) {
804 refcount_inc(&skb->users);
805 ireq->pktopts = skb;
806 }
807 }
808
tcp_v6_route_req(const struct sock * sk,struct sk_buff * skb,struct flowi * fl,struct request_sock * req)809 static struct dst_entry *tcp_v6_route_req(const struct sock *sk,
810 struct sk_buff *skb,
811 struct flowi *fl,
812 struct request_sock *req)
813 {
814 tcp_v6_init_req(req, sk, skb);
815
816 if (security_inet_conn_request(sk, skb, req))
817 return NULL;
818
819 return inet6_csk_route_req(sk, &fl->u.ip6, req, IPPROTO_TCP);
820 }
821
822 struct request_sock_ops tcp6_request_sock_ops __read_mostly = {
823 .family = AF_INET6,
824 .obj_size = sizeof(struct tcp6_request_sock),
825 .rtx_syn_ack = tcp_rtx_synack,
826 .send_ack = tcp_v6_reqsk_send_ack,
827 .destructor = tcp_v6_reqsk_destructor,
828 .send_reset = tcp_v6_send_reset,
829 .syn_ack_timeout = tcp_syn_ack_timeout,
830 };
831
832 const struct tcp_request_sock_ops tcp_request_sock_ipv6_ops = {
833 .mss_clamp = IPV6_MIN_MTU - sizeof(struct tcphdr) -
834 sizeof(struct ipv6hdr),
835 #ifdef CONFIG_TCP_MD5SIG
836 .req_md5_lookup = tcp_v6_md5_lookup,
837 .calc_md5_hash = tcp_v6_md5_hash_skb,
838 #endif
839 #ifdef CONFIG_SYN_COOKIES
840 .cookie_init_seq = cookie_v6_init_sequence,
841 #endif
842 .route_req = tcp_v6_route_req,
843 .init_seq = tcp_v6_init_seq,
844 .init_ts_off = tcp_v6_init_ts_off,
845 .send_synack = tcp_v6_send_synack,
846 };
847
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)848 static void tcp_v6_send_response(const struct sock *sk, struct sk_buff *skb, u32 seq,
849 u32 ack, u32 win, u32 tsval, u32 tsecr,
850 int oif, struct tcp_md5sig_key *key, int rst,
851 u8 tclass, __be32 label, u32 priority, u32 txhash)
852 {
853 const struct tcphdr *th = tcp_hdr(skb);
854 struct tcphdr *t1;
855 struct sk_buff *buff;
856 struct flowi6 fl6;
857 struct net *net = sk ? sock_net(sk) : dev_net(skb_dst(skb)->dev);
858 struct sock *ctl_sk = net->ipv6.tcp_sk;
859 unsigned int tot_len = sizeof(struct tcphdr);
860 __be32 mrst = 0, *topt;
861 struct dst_entry *dst;
862 __u32 mark = 0;
863
864 if (tsecr)
865 tot_len += TCPOLEN_TSTAMP_ALIGNED;
866 #ifdef CONFIG_TCP_MD5SIG
867 if (key)
868 tot_len += TCPOLEN_MD5SIG_ALIGNED;
869 #endif
870
871 #ifdef CONFIG_MPTCP
872 if (rst && !key) {
873 mrst = mptcp_reset_option(skb);
874
875 if (mrst)
876 tot_len += sizeof(__be32);
877 }
878 #endif
879
880 buff = alloc_skb(MAX_TCP_HEADER, GFP_ATOMIC);
881 if (!buff)
882 return;
883
884 skb_reserve(buff, MAX_TCP_HEADER);
885
886 t1 = skb_push(buff, tot_len);
887 skb_reset_transport_header(buff);
888
889 /* Swap the send and the receive. */
890 memset(t1, 0, sizeof(*t1));
891 t1->dest = th->source;
892 t1->source = th->dest;
893 t1->doff = tot_len / 4;
894 t1->seq = htonl(seq);
895 t1->ack_seq = htonl(ack);
896 t1->ack = !rst || !th->ack;
897 t1->rst = rst;
898 t1->window = htons(win);
899
900 topt = (__be32 *)(t1 + 1);
901
902 if (tsecr) {
903 *topt++ = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
904 (TCPOPT_TIMESTAMP << 8) | TCPOLEN_TIMESTAMP);
905 *topt++ = htonl(tsval);
906 *topt++ = htonl(tsecr);
907 }
908
909 if (mrst)
910 *topt++ = mrst;
911
912 #ifdef CONFIG_TCP_MD5SIG
913 if (key) {
914 *topt++ = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
915 (TCPOPT_MD5SIG << 8) | TCPOLEN_MD5SIG);
916 tcp_v6_md5_hash_hdr((__u8 *)topt, key,
917 &ipv6_hdr(skb)->saddr,
918 &ipv6_hdr(skb)->daddr, t1);
919 }
920 #endif
921
922 memset(&fl6, 0, sizeof(fl6));
923 fl6.daddr = ipv6_hdr(skb)->saddr;
924 fl6.saddr = ipv6_hdr(skb)->daddr;
925 fl6.flowlabel = label;
926
927 buff->ip_summed = CHECKSUM_PARTIAL;
928
929 __tcp_v6_send_check(buff, &fl6.saddr, &fl6.daddr);
930
931 fl6.flowi6_proto = IPPROTO_TCP;
932 if (rt6_need_strict(&fl6.daddr) && !oif)
933 fl6.flowi6_oif = tcp_v6_iif(skb);
934 else {
935 if (!oif && netif_index_is_l3_master(net, skb->skb_iif))
936 oif = skb->skb_iif;
937
938 fl6.flowi6_oif = oif;
939 }
940
941 if (sk) {
942 if (sk->sk_state == TCP_TIME_WAIT)
943 mark = inet_twsk(sk)->tw_mark;
944 else
945 mark = sk->sk_mark;
946 skb_set_delivery_time(buff, tcp_transmit_time(sk), true);
947 }
948 if (txhash) {
949 /* autoflowlabel/skb_get_hash_flowi6 rely on buff->hash */
950 skb_set_hash(buff, txhash, PKT_HASH_TYPE_L4);
951 }
952 fl6.flowi6_mark = IP6_REPLY_MARK(net, skb->mark) ?: mark;
953 fl6.fl6_dport = t1->dest;
954 fl6.fl6_sport = t1->source;
955 fl6.flowi6_uid = sock_net_uid(net, sk && sk_fullsock(sk) ? sk : NULL);
956 security_skb_classify_flow(skb, flowi6_to_flowi_common(&fl6));
957
958 /* Pass a socket to ip6_dst_lookup either it is for RST
959 * Underlying function will use this to retrieve the network
960 * namespace
961 */
962 if (sk && sk->sk_state != TCP_TIME_WAIT)
963 dst = ip6_dst_lookup_flow(net, sk, &fl6, NULL); /*sk's xfrm_policy can be referred*/
964 else
965 dst = ip6_dst_lookup_flow(net, ctl_sk, &fl6, NULL);
966 if (!IS_ERR(dst)) {
967 skb_dst_set(buff, dst);
968 ip6_xmit(ctl_sk, buff, &fl6, fl6.flowi6_mark, NULL,
969 tclass & ~INET_ECN_MASK, priority);
970 TCP_INC_STATS(net, TCP_MIB_OUTSEGS);
971 if (rst)
972 TCP_INC_STATS(net, TCP_MIB_OUTRSTS);
973 return;
974 }
975
976 kfree_skb(buff);
977 }
978
tcp_v6_send_reset(const struct sock * sk,struct sk_buff * skb)979 static void tcp_v6_send_reset(const struct sock *sk, struct sk_buff *skb)
980 {
981 const struct tcphdr *th = tcp_hdr(skb);
982 struct ipv6hdr *ipv6h = ipv6_hdr(skb);
983 u32 seq = 0, ack_seq = 0;
984 struct tcp_md5sig_key *key = NULL;
985 #ifdef CONFIG_TCP_MD5SIG
986 const __u8 *hash_location = NULL;
987 unsigned char newhash[16];
988 int genhash;
989 struct sock *sk1 = NULL;
990 #endif
991 __be32 label = 0;
992 u32 priority = 0;
993 struct net *net;
994 u32 txhash = 0;
995 int oif = 0;
996
997 if (th->rst)
998 return;
999
1000 /* If sk not NULL, it means we did a successful lookup and incoming
1001 * route had to be correct. prequeue might have dropped our dst.
1002 */
1003 if (!sk && !ipv6_unicast_destination(skb))
1004 return;
1005
1006 net = sk ? sock_net(sk) : dev_net(skb_dst(skb)->dev);
1007 #ifdef CONFIG_TCP_MD5SIG
1008 rcu_read_lock();
1009 hash_location = tcp_parse_md5sig_option(th);
1010 if (sk && sk_fullsock(sk)) {
1011 int l3index;
1012
1013 /* sdif set, means packet ingressed via a device
1014 * in an L3 domain and inet_iif is set to it.
1015 */
1016 l3index = tcp_v6_sdif(skb) ? tcp_v6_iif_l3_slave(skb) : 0;
1017 key = tcp_v6_md5_do_lookup(sk, &ipv6h->saddr, l3index);
1018 } else if (hash_location) {
1019 int dif = tcp_v6_iif_l3_slave(skb);
1020 int sdif = tcp_v6_sdif(skb);
1021 int l3index;
1022
1023 /*
1024 * active side is lost. Try to find listening socket through
1025 * source port, and then find md5 key through listening socket.
1026 * we are not loose security here:
1027 * Incoming packet is checked with md5 hash with finding key,
1028 * no RST generated if md5 hash doesn't match.
1029 */
1030 sk1 = inet6_lookup_listener(net, net->ipv4.tcp_death_row.hashinfo,
1031 NULL, 0, &ipv6h->saddr, th->source,
1032 &ipv6h->daddr, ntohs(th->source),
1033 dif, sdif);
1034 if (!sk1)
1035 goto out;
1036
1037 /* sdif set, means packet ingressed via a device
1038 * in an L3 domain and dif is set to it.
1039 */
1040 l3index = tcp_v6_sdif(skb) ? dif : 0;
1041
1042 key = tcp_v6_md5_do_lookup(sk1, &ipv6h->saddr, l3index);
1043 if (!key)
1044 goto out;
1045
1046 genhash = tcp_v6_md5_hash_skb(newhash, key, NULL, skb);
1047 if (genhash || memcmp(hash_location, newhash, 16) != 0)
1048 goto out;
1049 }
1050 #endif
1051
1052 if (th->ack)
1053 seq = ntohl(th->ack_seq);
1054 else
1055 ack_seq = ntohl(th->seq) + th->syn + th->fin + skb->len -
1056 (th->doff << 2);
1057
1058 if (sk) {
1059 oif = sk->sk_bound_dev_if;
1060 if (sk_fullsock(sk)) {
1061 const struct ipv6_pinfo *np = tcp_inet6_sk(sk);
1062
1063 trace_tcp_send_reset(sk, skb);
1064 if (np->repflow)
1065 label = ip6_flowlabel(ipv6h);
1066 priority = sk->sk_priority;
1067 txhash = sk->sk_hash;
1068 }
1069 if (sk->sk_state == TCP_TIME_WAIT) {
1070 label = cpu_to_be32(inet_twsk(sk)->tw_flowlabel);
1071 priority = inet_twsk(sk)->tw_priority;
1072 txhash = inet_twsk(sk)->tw_txhash;
1073 }
1074 } else {
1075 if (net->ipv6.sysctl.flowlabel_reflect & FLOWLABEL_REFLECT_TCP_RESET)
1076 label = ip6_flowlabel(ipv6h);
1077 }
1078
1079 tcp_v6_send_response(sk, skb, seq, ack_seq, 0, 0, 0, oif, key, 1,
1080 ipv6_get_dsfield(ipv6h), label, priority, txhash);
1081
1082 #ifdef CONFIG_TCP_MD5SIG
1083 out:
1084 rcu_read_unlock();
1085 #endif
1086 }
1087
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)1088 static void tcp_v6_send_ack(const struct sock *sk, struct sk_buff *skb, u32 seq,
1089 u32 ack, u32 win, u32 tsval, u32 tsecr, int oif,
1090 struct tcp_md5sig_key *key, u8 tclass,
1091 __be32 label, u32 priority, u32 txhash)
1092 {
1093 tcp_v6_send_response(sk, skb, seq, ack, win, tsval, tsecr, oif, key, 0,
1094 tclass, label, priority, txhash);
1095 }
1096
tcp_v6_timewait_ack(struct sock * sk,struct sk_buff * skb)1097 static void tcp_v6_timewait_ack(struct sock *sk, struct sk_buff *skb)
1098 {
1099 struct inet_timewait_sock *tw = inet_twsk(sk);
1100 struct tcp_timewait_sock *tcptw = tcp_twsk(sk);
1101
1102 tcp_v6_send_ack(sk, skb, tcptw->tw_snd_nxt, tcptw->tw_rcv_nxt,
1103 tcptw->tw_rcv_wnd >> tw->tw_rcv_wscale,
1104 tcp_time_stamp_raw() + tcptw->tw_ts_offset,
1105 tcptw->tw_ts_recent, tw->tw_bound_dev_if, tcp_twsk_md5_key(tcptw),
1106 tw->tw_tclass, cpu_to_be32(tw->tw_flowlabel), tw->tw_priority,
1107 tw->tw_txhash);
1108
1109 inet_twsk_put(tw);
1110 }
1111
tcp_v6_reqsk_send_ack(const struct sock * sk,struct sk_buff * skb,struct request_sock * req)1112 static void tcp_v6_reqsk_send_ack(const struct sock *sk, struct sk_buff *skb,
1113 struct request_sock *req)
1114 {
1115 int l3index;
1116
1117 l3index = tcp_v6_sdif(skb) ? tcp_v6_iif_l3_slave(skb) : 0;
1118
1119 /* sk->sk_state == TCP_LISTEN -> for regular TCP_SYN_RECV
1120 * sk->sk_state == TCP_SYN_RECV -> for Fast Open.
1121 */
1122 /* RFC 7323 2.3
1123 * The window field (SEG.WND) of every outgoing segment, with the
1124 * exception of <SYN> segments, MUST be right-shifted by
1125 * Rcv.Wind.Shift bits:
1126 */
1127 tcp_v6_send_ack(sk, skb, (sk->sk_state == TCP_LISTEN) ?
1128 tcp_rsk(req)->snt_isn + 1 : tcp_sk(sk)->snd_nxt,
1129 tcp_rsk(req)->rcv_nxt,
1130 req->rsk_rcv_wnd >> inet_rsk(req)->rcv_wscale,
1131 tcp_time_stamp_raw() + tcp_rsk(req)->ts_off,
1132 req->ts_recent, sk->sk_bound_dev_if,
1133 tcp_v6_md5_do_lookup(sk, &ipv6_hdr(skb)->saddr, l3index),
1134 ipv6_get_dsfield(ipv6_hdr(skb)), 0, sk->sk_priority,
1135 tcp_rsk(req)->txhash);
1136 }
1137
1138
tcp_v6_cookie_check(struct sock * sk,struct sk_buff * skb)1139 static struct sock *tcp_v6_cookie_check(struct sock *sk, struct sk_buff *skb)
1140 {
1141 #ifdef CONFIG_SYN_COOKIES
1142 const struct tcphdr *th = tcp_hdr(skb);
1143
1144 if (!th->syn)
1145 sk = cookie_v6_check(sk, skb);
1146 #endif
1147 return sk;
1148 }
1149
tcp_v6_get_syncookie(struct sock * sk,struct ipv6hdr * iph,struct tcphdr * th,u32 * cookie)1150 u16 tcp_v6_get_syncookie(struct sock *sk, struct ipv6hdr *iph,
1151 struct tcphdr *th, u32 *cookie)
1152 {
1153 u16 mss = 0;
1154 #ifdef CONFIG_SYN_COOKIES
1155 mss = tcp_get_syncookie_mss(&tcp6_request_sock_ops,
1156 &tcp_request_sock_ipv6_ops, sk, th);
1157 if (mss) {
1158 *cookie = __cookie_v6_init_sequence(iph, th, &mss);
1159 tcp_synq_overflow(sk);
1160 }
1161 #endif
1162 return mss;
1163 }
1164
tcp_v6_conn_request(struct sock * sk,struct sk_buff * skb)1165 static int tcp_v6_conn_request(struct sock *sk, struct sk_buff *skb)
1166 {
1167 if (skb->protocol == htons(ETH_P_IP))
1168 return tcp_v4_conn_request(sk, skb);
1169
1170 if (!ipv6_unicast_destination(skb))
1171 goto drop;
1172
1173 if (ipv6_addr_v4mapped(&ipv6_hdr(skb)->saddr)) {
1174 __IP6_INC_STATS(sock_net(sk), NULL, IPSTATS_MIB_INHDRERRORS);
1175 return 0;
1176 }
1177
1178 return tcp_conn_request(&tcp6_request_sock_ops,
1179 &tcp_request_sock_ipv6_ops, sk, skb);
1180
1181 drop:
1182 tcp_listendrop(sk);
1183 return 0; /* don't send reset */
1184 }
1185
tcp_v6_restore_cb(struct sk_buff * skb)1186 static void tcp_v6_restore_cb(struct sk_buff *skb)
1187 {
1188 /* We need to move header back to the beginning if xfrm6_policy_check()
1189 * and tcp_v6_fill_cb() are going to be called again.
1190 * ip6_datagram_recv_specific_ctl() also expects IP6CB to be there.
1191 */
1192 memmove(IP6CB(skb), &TCP_SKB_CB(skb)->header.h6,
1193 sizeof(struct inet6_skb_parm));
1194 }
1195
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)1196 static struct sock *tcp_v6_syn_recv_sock(const struct sock *sk, struct sk_buff *skb,
1197 struct request_sock *req,
1198 struct dst_entry *dst,
1199 struct request_sock *req_unhash,
1200 bool *own_req)
1201 {
1202 struct inet_request_sock *ireq;
1203 struct ipv6_pinfo *newnp;
1204 const struct ipv6_pinfo *np = tcp_inet6_sk(sk);
1205 struct ipv6_txoptions *opt;
1206 struct inet_sock *newinet;
1207 bool found_dup_sk = false;
1208 struct tcp_sock *newtp;
1209 struct sock *newsk;
1210 #ifdef CONFIG_TCP_MD5SIG
1211 struct tcp_md5sig_key *key;
1212 int l3index;
1213 #endif
1214 struct flowi6 fl6;
1215
1216 if (skb->protocol == htons(ETH_P_IP)) {
1217 /*
1218 * v6 mapped
1219 */
1220
1221 newsk = tcp_v4_syn_recv_sock(sk, skb, req, dst,
1222 req_unhash, own_req);
1223
1224 if (!newsk)
1225 return NULL;
1226
1227 inet_sk(newsk)->pinet6 = tcp_inet6_sk(newsk);
1228
1229 newnp = tcp_inet6_sk(newsk);
1230 newtp = tcp_sk(newsk);
1231
1232 memcpy(newnp, np, sizeof(struct ipv6_pinfo));
1233
1234 newnp->saddr = newsk->sk_v6_rcv_saddr;
1235
1236 inet_csk(newsk)->icsk_af_ops = &ipv6_mapped;
1237 if (sk_is_mptcp(newsk))
1238 mptcpv6_handle_mapped(newsk, true);
1239 newsk->sk_backlog_rcv = tcp_v4_do_rcv;
1240 #ifdef CONFIG_TCP_MD5SIG
1241 newtp->af_specific = &tcp_sock_ipv6_mapped_specific;
1242 #endif
1243
1244 newnp->ipv6_mc_list = NULL;
1245 newnp->ipv6_ac_list = NULL;
1246 newnp->ipv6_fl_list = NULL;
1247 newnp->pktoptions = NULL;
1248 newnp->opt = NULL;
1249 newnp->mcast_oif = inet_iif(skb);
1250 newnp->mcast_hops = ip_hdr(skb)->ttl;
1251 newnp->rcv_flowinfo = 0;
1252 if (np->repflow)
1253 newnp->flow_label = 0;
1254
1255 /*
1256 * No need to charge this sock to the relevant IPv6 refcnt debug socks count
1257 * here, tcp_create_openreq_child now does this for us, see the comment in
1258 * that function for the gory details. -acme
1259 */
1260
1261 /* It is tricky place. Until this moment IPv4 tcp
1262 worked with IPv6 icsk.icsk_af_ops.
1263 Sync it now.
1264 */
1265 tcp_sync_mss(newsk, inet_csk(newsk)->icsk_pmtu_cookie);
1266
1267 return newsk;
1268 }
1269
1270 ireq = inet_rsk(req);
1271
1272 if (sk_acceptq_is_full(sk))
1273 goto out_overflow;
1274
1275 if (!dst) {
1276 dst = inet6_csk_route_req(sk, &fl6, req, IPPROTO_TCP);
1277 if (!dst)
1278 goto out;
1279 }
1280
1281 newsk = tcp_create_openreq_child(sk, req, skb);
1282 if (!newsk)
1283 goto out_nonewsk;
1284
1285 /*
1286 * No need to charge this sock to the relevant IPv6 refcnt debug socks
1287 * count here, tcp_create_openreq_child now does this for us, see the
1288 * comment in that function for the gory details. -acme
1289 */
1290
1291 newsk->sk_gso_type = SKB_GSO_TCPV6;
1292 ip6_dst_store(newsk, dst, NULL, NULL);
1293 inet6_sk_rx_dst_set(newsk, skb);
1294
1295 inet_sk(newsk)->pinet6 = tcp_inet6_sk(newsk);
1296
1297 newtp = tcp_sk(newsk);
1298 newinet = inet_sk(newsk);
1299 newnp = tcp_inet6_sk(newsk);
1300
1301 memcpy(newnp, np, sizeof(struct ipv6_pinfo));
1302
1303 newsk->sk_v6_daddr = ireq->ir_v6_rmt_addr;
1304 newnp->saddr = ireq->ir_v6_loc_addr;
1305 newsk->sk_v6_rcv_saddr = ireq->ir_v6_loc_addr;
1306 newsk->sk_bound_dev_if = ireq->ir_iif;
1307
1308 /* Now IPv6 options...
1309
1310 First: no IPv4 options.
1311 */
1312 newinet->inet_opt = NULL;
1313 newnp->ipv6_mc_list = NULL;
1314 newnp->ipv6_ac_list = NULL;
1315 newnp->ipv6_fl_list = NULL;
1316
1317 /* Clone RX bits */
1318 newnp->rxopt.all = np->rxopt.all;
1319
1320 newnp->pktoptions = NULL;
1321 newnp->opt = NULL;
1322 newnp->mcast_oif = tcp_v6_iif(skb);
1323 newnp->mcast_hops = ipv6_hdr(skb)->hop_limit;
1324 newnp->rcv_flowinfo = ip6_flowinfo(ipv6_hdr(skb));
1325 if (np->repflow)
1326 newnp->flow_label = ip6_flowlabel(ipv6_hdr(skb));
1327
1328 /* Set ToS of the new socket based upon the value of incoming SYN.
1329 * ECT bits are set later in tcp_init_transfer().
1330 */
1331 if (READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_reflect_tos))
1332 newnp->tclass = tcp_rsk(req)->syn_tos & ~INET_ECN_MASK;
1333
1334 /* Clone native IPv6 options from listening socket (if any)
1335
1336 Yes, keeping reference count would be much more clever,
1337 but we make one more one thing there: reattach optmem
1338 to newsk.
1339 */
1340 opt = ireq->ipv6_opt;
1341 if (!opt)
1342 opt = rcu_dereference(np->opt);
1343 if (opt) {
1344 opt = ipv6_dup_options(newsk, opt);
1345 RCU_INIT_POINTER(newnp->opt, opt);
1346 }
1347 inet_csk(newsk)->icsk_ext_hdr_len = 0;
1348 if (opt)
1349 inet_csk(newsk)->icsk_ext_hdr_len = opt->opt_nflen +
1350 opt->opt_flen;
1351
1352 tcp_ca_openreq_child(newsk, dst);
1353
1354 tcp_sync_mss(newsk, dst_mtu(dst));
1355 newtp->advmss = tcp_mss_clamp(tcp_sk(sk), dst_metric_advmss(dst));
1356
1357 tcp_initialize_rcv_mss(newsk);
1358
1359 newinet->inet_daddr = newinet->inet_saddr = LOOPBACK4_IPV6;
1360 newinet->inet_rcv_saddr = LOOPBACK4_IPV6;
1361
1362 #ifdef CONFIG_TCP_MD5SIG
1363 l3index = l3mdev_master_ifindex_by_index(sock_net(sk), ireq->ir_iif);
1364
1365 /* Copy over the MD5 key from the original socket */
1366 key = tcp_v6_md5_do_lookup(sk, &newsk->sk_v6_daddr, l3index);
1367 if (key) {
1368 /* We're using one, so create a matching key
1369 * on the newsk structure. If we fail to get
1370 * memory, then we end up not copying the key
1371 * across. Shucks.
1372 */
1373 tcp_md5_do_add(newsk, (union tcp_md5_addr *)&newsk->sk_v6_daddr,
1374 AF_INET6, 128, l3index, key->flags, key->key, key->keylen,
1375 sk_gfp_mask(sk, GFP_ATOMIC));
1376 }
1377 #endif
1378
1379 if (__inet_inherit_port(sk, newsk) < 0) {
1380 inet_csk_prepare_forced_close(newsk);
1381 tcp_done(newsk);
1382 goto out;
1383 }
1384 *own_req = inet_ehash_nolisten(newsk, req_to_sk(req_unhash),
1385 &found_dup_sk);
1386 if (*own_req) {
1387 tcp_move_syn(newtp, req);
1388
1389 /* Clone pktoptions received with SYN, if we own the req */
1390 if (ireq->pktopts) {
1391 newnp->pktoptions = skb_clone(ireq->pktopts,
1392 sk_gfp_mask(sk, GFP_ATOMIC));
1393 consume_skb(ireq->pktopts);
1394 ireq->pktopts = NULL;
1395 if (newnp->pktoptions) {
1396 tcp_v6_restore_cb(newnp->pktoptions);
1397 skb_set_owner_r(newnp->pktoptions, newsk);
1398 }
1399 }
1400 } else {
1401 if (!req_unhash && found_dup_sk) {
1402 /* This code path should only be executed in the
1403 * syncookie case only
1404 */
1405 bh_unlock_sock(newsk);
1406 sock_put(newsk);
1407 newsk = NULL;
1408 }
1409 }
1410
1411 return newsk;
1412
1413 out_overflow:
1414 __NET_INC_STATS(sock_net(sk), LINUX_MIB_LISTENOVERFLOWS);
1415 out_nonewsk:
1416 dst_release(dst);
1417 out:
1418 tcp_listendrop(sk);
1419 return NULL;
1420 }
1421
1422 INDIRECT_CALLABLE_DECLARE(struct dst_entry *ipv4_dst_check(struct dst_entry *,
1423 u32));
1424 /* The socket must have it's spinlock held when we get
1425 * here, unless it is a TCP_LISTEN socket.
1426 *
1427 * We have a potential double-lock case here, so even when
1428 * doing backlog processing we use the BH locking scheme.
1429 * This is because we cannot sleep with the original spinlock
1430 * held.
1431 */
1432 INDIRECT_CALLABLE_SCOPE
tcp_v6_do_rcv(struct sock * sk,struct sk_buff * skb)1433 int tcp_v6_do_rcv(struct sock *sk, struct sk_buff *skb)
1434 {
1435 struct ipv6_pinfo *np = tcp_inet6_sk(sk);
1436 struct sk_buff *opt_skb = NULL;
1437 enum skb_drop_reason reason;
1438 struct tcp_sock *tp;
1439
1440 /* Imagine: socket is IPv6. IPv4 packet arrives,
1441 goes to IPv4 receive handler and backlogged.
1442 From backlog it always goes here. Kerboom...
1443 Fortunately, tcp_rcv_established and rcv_established
1444 handle them correctly, but it is not case with
1445 tcp_v6_hnd_req and tcp_v6_send_reset(). --ANK
1446 */
1447
1448 if (skb->protocol == htons(ETH_P_IP))
1449 return tcp_v4_do_rcv(sk, skb);
1450
1451 /*
1452 * socket locking is here for SMP purposes as backlog rcv
1453 * is currently called with bh processing disabled.
1454 */
1455
1456 /* Do Stevens' IPV6_PKTOPTIONS.
1457
1458 Yes, guys, it is the only place in our code, where we
1459 may make it not affecting IPv4.
1460 The rest of code is protocol independent,
1461 and I do not like idea to uglify IPv4.
1462
1463 Actually, all the idea behind IPV6_PKTOPTIONS
1464 looks not very well thought. For now we latch
1465 options, received in the last packet, enqueued
1466 by tcp. Feel free to propose better solution.
1467 --ANK (980728)
1468 */
1469 if (np->rxopt.all)
1470 opt_skb = skb_clone(skb, sk_gfp_mask(sk, GFP_ATOMIC));
1471
1472 reason = SKB_DROP_REASON_NOT_SPECIFIED;
1473 if (sk->sk_state == TCP_ESTABLISHED) { /* Fast path */
1474 struct dst_entry *dst;
1475
1476 dst = rcu_dereference_protected(sk->sk_rx_dst,
1477 lockdep_sock_is_held(sk));
1478
1479 sock_rps_save_rxhash(sk, skb);
1480 sk_mark_napi_id(sk, skb);
1481 if (dst) {
1482 if (sk->sk_rx_dst_ifindex != skb->skb_iif ||
1483 INDIRECT_CALL_1(dst->ops->check, ip6_dst_check,
1484 dst, sk->sk_rx_dst_cookie) == NULL) {
1485 RCU_INIT_POINTER(sk->sk_rx_dst, NULL);
1486 dst_release(dst);
1487 }
1488 }
1489
1490 tcp_rcv_established(sk, skb);
1491 if (opt_skb)
1492 goto ipv6_pktoptions;
1493 return 0;
1494 }
1495
1496 if (tcp_checksum_complete(skb))
1497 goto csum_err;
1498
1499 if (sk->sk_state == TCP_LISTEN) {
1500 struct sock *nsk = tcp_v6_cookie_check(sk, skb);
1501
1502 if (!nsk)
1503 goto discard;
1504
1505 if (nsk != sk) {
1506 if (tcp_child_process(sk, nsk, skb))
1507 goto reset;
1508 if (opt_skb)
1509 __kfree_skb(opt_skb);
1510 return 0;
1511 }
1512 } else
1513 sock_rps_save_rxhash(sk, skb);
1514
1515 if (tcp_rcv_state_process(sk, skb))
1516 goto reset;
1517 if (opt_skb)
1518 goto ipv6_pktoptions;
1519 return 0;
1520
1521 reset:
1522 tcp_v6_send_reset(sk, skb);
1523 discard:
1524 if (opt_skb)
1525 __kfree_skb(opt_skb);
1526 kfree_skb_reason(skb, reason);
1527 return 0;
1528 csum_err:
1529 reason = SKB_DROP_REASON_TCP_CSUM;
1530 trace_tcp_bad_csum(skb);
1531 TCP_INC_STATS(sock_net(sk), TCP_MIB_CSUMERRORS);
1532 TCP_INC_STATS(sock_net(sk), TCP_MIB_INERRS);
1533 goto discard;
1534
1535
1536 ipv6_pktoptions:
1537 /* Do you ask, what is it?
1538
1539 1. skb was enqueued by tcp.
1540 2. skb is added to tail of read queue, rather than out of order.
1541 3. socket is not in passive state.
1542 4. Finally, it really contains options, which user wants to receive.
1543 */
1544 tp = tcp_sk(sk);
1545 if (TCP_SKB_CB(opt_skb)->end_seq == tp->rcv_nxt &&
1546 !((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_LISTEN))) {
1547 if (np->rxopt.bits.rxinfo || np->rxopt.bits.rxoinfo)
1548 np->mcast_oif = tcp_v6_iif(opt_skb);
1549 if (np->rxopt.bits.rxhlim || np->rxopt.bits.rxohlim)
1550 np->mcast_hops = ipv6_hdr(opt_skb)->hop_limit;
1551 if (np->rxopt.bits.rxflow || np->rxopt.bits.rxtclass)
1552 np->rcv_flowinfo = ip6_flowinfo(ipv6_hdr(opt_skb));
1553 if (np->repflow)
1554 np->flow_label = ip6_flowlabel(ipv6_hdr(opt_skb));
1555 if (ipv6_opt_accepted(sk, opt_skb, &TCP_SKB_CB(opt_skb)->header.h6)) {
1556 skb_set_owner_r(opt_skb, sk);
1557 tcp_v6_restore_cb(opt_skb);
1558 opt_skb = xchg(&np->pktoptions, opt_skb);
1559 } else {
1560 __kfree_skb(opt_skb);
1561 opt_skb = xchg(&np->pktoptions, NULL);
1562 }
1563 }
1564
1565 consume_skb(opt_skb);
1566 return 0;
1567 }
1568
tcp_v6_fill_cb(struct sk_buff * skb,const struct ipv6hdr * hdr,const struct tcphdr * th)1569 static void tcp_v6_fill_cb(struct sk_buff *skb, const struct ipv6hdr *hdr,
1570 const struct tcphdr *th)
1571 {
1572 /* This is tricky: we move IP6CB at its correct location into
1573 * TCP_SKB_CB(). It must be done after xfrm6_policy_check(), because
1574 * _decode_session6() uses IP6CB().
1575 * barrier() makes sure compiler won't play aliasing games.
1576 */
1577 memmove(&TCP_SKB_CB(skb)->header.h6, IP6CB(skb),
1578 sizeof(struct inet6_skb_parm));
1579 barrier();
1580
1581 TCP_SKB_CB(skb)->seq = ntohl(th->seq);
1582 TCP_SKB_CB(skb)->end_seq = (TCP_SKB_CB(skb)->seq + th->syn + th->fin +
1583 skb->len - th->doff*4);
1584 TCP_SKB_CB(skb)->ack_seq = ntohl(th->ack_seq);
1585 TCP_SKB_CB(skb)->tcp_flags = tcp_flag_byte(th);
1586 TCP_SKB_CB(skb)->tcp_tw_isn = 0;
1587 TCP_SKB_CB(skb)->ip_dsfield = ipv6_get_dsfield(hdr);
1588 TCP_SKB_CB(skb)->sacked = 0;
1589 TCP_SKB_CB(skb)->has_rxtstamp =
1590 skb->tstamp || skb_hwtstamps(skb)->hwtstamp;
1591 }
1592
tcp_v6_rcv(struct sk_buff * skb)1593 INDIRECT_CALLABLE_SCOPE int tcp_v6_rcv(struct sk_buff *skb)
1594 {
1595 enum skb_drop_reason drop_reason;
1596 int sdif = inet6_sdif(skb);
1597 int dif = inet6_iif(skb);
1598 const struct tcphdr *th;
1599 const struct ipv6hdr *hdr;
1600 bool refcounted;
1601 struct sock *sk;
1602 int ret;
1603 struct net *net = dev_net(skb->dev);
1604
1605 drop_reason = SKB_DROP_REASON_NOT_SPECIFIED;
1606 if (skb->pkt_type != PACKET_HOST)
1607 goto discard_it;
1608
1609 /*
1610 * Count it even if it's bad.
1611 */
1612 __TCP_INC_STATS(net, TCP_MIB_INSEGS);
1613
1614 if (!pskb_may_pull(skb, sizeof(struct tcphdr)))
1615 goto discard_it;
1616
1617 th = (const struct tcphdr *)skb->data;
1618
1619 if (unlikely(th->doff < sizeof(struct tcphdr) / 4)) {
1620 drop_reason = SKB_DROP_REASON_PKT_TOO_SMALL;
1621 goto bad_packet;
1622 }
1623 if (!pskb_may_pull(skb, th->doff*4))
1624 goto discard_it;
1625
1626 if (skb_checksum_init(skb, IPPROTO_TCP, ip6_compute_pseudo))
1627 goto csum_error;
1628
1629 th = (const struct tcphdr *)skb->data;
1630 hdr = ipv6_hdr(skb);
1631
1632 lookup:
1633 sk = __inet6_lookup_skb(net->ipv4.tcp_death_row.hashinfo, skb, __tcp_hdrlen(th),
1634 th->source, th->dest, inet6_iif(skb), sdif,
1635 &refcounted);
1636 if (!sk)
1637 goto no_tcp_socket;
1638
1639 process:
1640 if (sk->sk_state == TCP_TIME_WAIT)
1641 goto do_time_wait;
1642
1643 if (sk->sk_state == TCP_NEW_SYN_RECV) {
1644 struct request_sock *req = inet_reqsk(sk);
1645 bool req_stolen = false;
1646 struct sock *nsk;
1647
1648 sk = req->rsk_listener;
1649 drop_reason = tcp_inbound_md5_hash(sk, skb,
1650 &hdr->saddr, &hdr->daddr,
1651 AF_INET6, dif, sdif);
1652 if (drop_reason) {
1653 sk_drops_add(sk, skb);
1654 reqsk_put(req);
1655 goto discard_it;
1656 }
1657 if (tcp_checksum_complete(skb)) {
1658 reqsk_put(req);
1659 goto csum_error;
1660 }
1661 if (unlikely(sk->sk_state != TCP_LISTEN)) {
1662 nsk = reuseport_migrate_sock(sk, req_to_sk(req), skb);
1663 if (!nsk) {
1664 inet_csk_reqsk_queue_drop_and_put(sk, req);
1665 goto lookup;
1666 }
1667 sk = nsk;
1668 /* reuseport_migrate_sock() has already held one sk_refcnt
1669 * before returning.
1670 */
1671 } else {
1672 sock_hold(sk);
1673 }
1674 refcounted = true;
1675 nsk = NULL;
1676 if (!tcp_filter(sk, skb)) {
1677 th = (const struct tcphdr *)skb->data;
1678 hdr = ipv6_hdr(skb);
1679 tcp_v6_fill_cb(skb, hdr, th);
1680 nsk = tcp_check_req(sk, skb, req, false, &req_stolen);
1681 } else {
1682 drop_reason = SKB_DROP_REASON_SOCKET_FILTER;
1683 }
1684 if (!nsk) {
1685 reqsk_put(req);
1686 if (req_stolen) {
1687 /* Another cpu got exclusive access to req
1688 * and created a full blown socket.
1689 * Try to feed this packet to this socket
1690 * instead of discarding it.
1691 */
1692 tcp_v6_restore_cb(skb);
1693 sock_put(sk);
1694 goto lookup;
1695 }
1696 goto discard_and_relse;
1697 }
1698 if (nsk == sk) {
1699 reqsk_put(req);
1700 tcp_v6_restore_cb(skb);
1701 } else if (tcp_child_process(sk, nsk, skb)) {
1702 tcp_v6_send_reset(nsk, skb);
1703 goto discard_and_relse;
1704 } else {
1705 sock_put(sk);
1706 return 0;
1707 }
1708 }
1709
1710 if (static_branch_unlikely(&ip6_min_hopcount)) {
1711 /* min_hopcount can be changed concurrently from do_ipv6_setsockopt() */
1712 if (hdr->hop_limit < READ_ONCE(tcp_inet6_sk(sk)->min_hopcount)) {
1713 __NET_INC_STATS(net, LINUX_MIB_TCPMINTTLDROP);
1714 goto discard_and_relse;
1715 }
1716 }
1717
1718 if (!xfrm6_policy_check(sk, XFRM_POLICY_IN, skb)) {
1719 drop_reason = SKB_DROP_REASON_XFRM_POLICY;
1720 goto discard_and_relse;
1721 }
1722
1723 drop_reason = tcp_inbound_md5_hash(sk, skb, &hdr->saddr, &hdr->daddr,
1724 AF_INET6, dif, sdif);
1725 if (drop_reason)
1726 goto discard_and_relse;
1727
1728 if (tcp_filter(sk, skb)) {
1729 drop_reason = SKB_DROP_REASON_SOCKET_FILTER;
1730 goto discard_and_relse;
1731 }
1732 th = (const struct tcphdr *)skb->data;
1733 hdr = ipv6_hdr(skb);
1734 tcp_v6_fill_cb(skb, hdr, th);
1735
1736 skb->dev = NULL;
1737
1738 if (sk->sk_state == TCP_LISTEN) {
1739 ret = tcp_v6_do_rcv(sk, skb);
1740 goto put_and_return;
1741 }
1742
1743 sk_incoming_cpu_update(sk);
1744
1745 bh_lock_sock_nested(sk);
1746 tcp_segs_in(tcp_sk(sk), skb);
1747 ret = 0;
1748 if (!sock_owned_by_user(sk)) {
1749 ret = tcp_v6_do_rcv(sk, skb);
1750 } else {
1751 if (tcp_add_backlog(sk, skb, &drop_reason))
1752 goto discard_and_relse;
1753 }
1754 bh_unlock_sock(sk);
1755 put_and_return:
1756 if (refcounted)
1757 sock_put(sk);
1758 return ret ? -1 : 0;
1759
1760 no_tcp_socket:
1761 drop_reason = SKB_DROP_REASON_NO_SOCKET;
1762 if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb))
1763 goto discard_it;
1764
1765 tcp_v6_fill_cb(skb, hdr, th);
1766
1767 if (tcp_checksum_complete(skb)) {
1768 csum_error:
1769 drop_reason = SKB_DROP_REASON_TCP_CSUM;
1770 trace_tcp_bad_csum(skb);
1771 __TCP_INC_STATS(net, TCP_MIB_CSUMERRORS);
1772 bad_packet:
1773 __TCP_INC_STATS(net, TCP_MIB_INERRS);
1774 } else {
1775 tcp_v6_send_reset(NULL, skb);
1776 }
1777
1778 discard_it:
1779 SKB_DR_OR(drop_reason, NOT_SPECIFIED);
1780 kfree_skb_reason(skb, drop_reason);
1781 return 0;
1782
1783 discard_and_relse:
1784 sk_drops_add(sk, skb);
1785 if (refcounted)
1786 sock_put(sk);
1787 goto discard_it;
1788
1789 do_time_wait:
1790 if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb)) {
1791 drop_reason = SKB_DROP_REASON_XFRM_POLICY;
1792 inet_twsk_put(inet_twsk(sk));
1793 goto discard_it;
1794 }
1795
1796 tcp_v6_fill_cb(skb, hdr, th);
1797
1798 if (tcp_checksum_complete(skb)) {
1799 inet_twsk_put(inet_twsk(sk));
1800 goto csum_error;
1801 }
1802
1803 switch (tcp_timewait_state_process(inet_twsk(sk), skb, th)) {
1804 case TCP_TW_SYN:
1805 {
1806 struct sock *sk2;
1807
1808 sk2 = inet6_lookup_listener(net, net->ipv4.tcp_death_row.hashinfo,
1809 skb, __tcp_hdrlen(th),
1810 &ipv6_hdr(skb)->saddr, th->source,
1811 &ipv6_hdr(skb)->daddr,
1812 ntohs(th->dest),
1813 tcp_v6_iif_l3_slave(skb),
1814 sdif);
1815 if (sk2) {
1816 struct inet_timewait_sock *tw = inet_twsk(sk);
1817 inet_twsk_deschedule_put(tw);
1818 sk = sk2;
1819 tcp_v6_restore_cb(skb);
1820 refcounted = false;
1821 goto process;
1822 }
1823 }
1824 /* to ACK */
1825 fallthrough;
1826 case TCP_TW_ACK:
1827 tcp_v6_timewait_ack(sk, skb);
1828 break;
1829 case TCP_TW_RST:
1830 tcp_v6_send_reset(sk, skb);
1831 inet_twsk_deschedule_put(inet_twsk(sk));
1832 goto discard_it;
1833 case TCP_TW_SUCCESS:
1834 ;
1835 }
1836 goto discard_it;
1837 }
1838
tcp_v6_early_demux(struct sk_buff * skb)1839 void tcp_v6_early_demux(struct sk_buff *skb)
1840 {
1841 struct net *net = dev_net(skb->dev);
1842 const struct ipv6hdr *hdr;
1843 const struct tcphdr *th;
1844 struct sock *sk;
1845
1846 if (skb->pkt_type != PACKET_HOST)
1847 return;
1848
1849 if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct tcphdr)))
1850 return;
1851
1852 hdr = ipv6_hdr(skb);
1853 th = tcp_hdr(skb);
1854
1855 if (th->doff < sizeof(struct tcphdr) / 4)
1856 return;
1857
1858 /* Note : We use inet6_iif() here, not tcp_v6_iif() */
1859 sk = __inet6_lookup_established(net, net->ipv4.tcp_death_row.hashinfo,
1860 &hdr->saddr, th->source,
1861 &hdr->daddr, ntohs(th->dest),
1862 inet6_iif(skb), inet6_sdif(skb));
1863 if (sk) {
1864 skb->sk = sk;
1865 skb->destructor = sock_edemux;
1866 if (sk_fullsock(sk)) {
1867 struct dst_entry *dst = rcu_dereference(sk->sk_rx_dst);
1868
1869 if (dst)
1870 dst = dst_check(dst, sk->sk_rx_dst_cookie);
1871 if (dst &&
1872 sk->sk_rx_dst_ifindex == skb->skb_iif)
1873 skb_dst_set_noref(skb, dst);
1874 }
1875 }
1876 }
1877
1878 static struct timewait_sock_ops tcp6_timewait_sock_ops = {
1879 .twsk_obj_size = sizeof(struct tcp6_timewait_sock),
1880 .twsk_unique = tcp_twsk_unique,
1881 .twsk_destructor = tcp_twsk_destructor,
1882 };
1883
tcp_v6_send_check(struct sock * sk,struct sk_buff * skb)1884 INDIRECT_CALLABLE_SCOPE void tcp_v6_send_check(struct sock *sk, struct sk_buff *skb)
1885 {
1886 __tcp_v6_send_check(skb, &sk->sk_v6_rcv_saddr, &sk->sk_v6_daddr);
1887 }
1888
1889 const struct inet_connection_sock_af_ops ipv6_specific = {
1890 .queue_xmit = inet6_csk_xmit,
1891 .send_check = tcp_v6_send_check,
1892 .rebuild_header = inet6_sk_rebuild_header,
1893 .sk_rx_dst_set = inet6_sk_rx_dst_set,
1894 .conn_request = tcp_v6_conn_request,
1895 .syn_recv_sock = tcp_v6_syn_recv_sock,
1896 .net_header_len = sizeof(struct ipv6hdr),
1897 .net_frag_header_len = sizeof(struct frag_hdr),
1898 .setsockopt = ipv6_setsockopt,
1899 .getsockopt = ipv6_getsockopt,
1900 .addr2sockaddr = inet6_csk_addr2sockaddr,
1901 .sockaddr_len = sizeof(struct sockaddr_in6),
1902 .mtu_reduced = tcp_v6_mtu_reduced,
1903 };
1904
1905 #ifdef CONFIG_TCP_MD5SIG
1906 static const struct tcp_sock_af_ops tcp_sock_ipv6_specific = {
1907 .md5_lookup = tcp_v6_md5_lookup,
1908 .calc_md5_hash = tcp_v6_md5_hash_skb,
1909 .md5_parse = tcp_v6_parse_md5_keys,
1910 };
1911 #endif
1912
1913 /*
1914 * TCP over IPv4 via INET6 API
1915 */
1916 static const struct inet_connection_sock_af_ops ipv6_mapped = {
1917 .queue_xmit = ip_queue_xmit,
1918 .send_check = tcp_v4_send_check,
1919 .rebuild_header = inet_sk_rebuild_header,
1920 .sk_rx_dst_set = inet_sk_rx_dst_set,
1921 .conn_request = tcp_v6_conn_request,
1922 .syn_recv_sock = tcp_v6_syn_recv_sock,
1923 .net_header_len = sizeof(struct iphdr),
1924 .setsockopt = ipv6_setsockopt,
1925 .getsockopt = ipv6_getsockopt,
1926 .addr2sockaddr = inet6_csk_addr2sockaddr,
1927 .sockaddr_len = sizeof(struct sockaddr_in6),
1928 .mtu_reduced = tcp_v4_mtu_reduced,
1929 };
1930
1931 #ifdef CONFIG_TCP_MD5SIG
1932 static const struct tcp_sock_af_ops tcp_sock_ipv6_mapped_specific = {
1933 .md5_lookup = tcp_v4_md5_lookup,
1934 .calc_md5_hash = tcp_v4_md5_hash_skb,
1935 .md5_parse = tcp_v6_parse_md5_keys,
1936 };
1937 #endif
1938
1939 /* NOTE: A lot of things set to zero explicitly by call to
1940 * sk_alloc() so need not be done here.
1941 */
tcp_v6_init_sock(struct sock * sk)1942 static int tcp_v6_init_sock(struct sock *sk)
1943 {
1944 struct inet_connection_sock *icsk = inet_csk(sk);
1945
1946 tcp_init_sock(sk);
1947
1948 icsk->icsk_af_ops = &ipv6_specific;
1949
1950 #ifdef CONFIG_TCP_MD5SIG
1951 tcp_sk(sk)->af_specific = &tcp_sock_ipv6_specific;
1952 #endif
1953
1954 return 0;
1955 }
1956
tcp_v6_destroy_sock(struct sock * sk)1957 static void tcp_v6_destroy_sock(struct sock *sk)
1958 {
1959 tcp_v4_destroy_sock(sk);
1960 inet6_destroy_sock(sk);
1961 }
1962
1963 #ifdef CONFIG_PROC_FS
1964 /* Proc filesystem TCPv6 sock list dumping. */
get_openreq6(struct seq_file * seq,const struct request_sock * req,int i)1965 static void get_openreq6(struct seq_file *seq,
1966 const struct request_sock *req, int i)
1967 {
1968 long ttd = req->rsk_timer.expires - jiffies;
1969 const struct in6_addr *src = &inet_rsk(req)->ir_v6_loc_addr;
1970 const struct in6_addr *dest = &inet_rsk(req)->ir_v6_rmt_addr;
1971
1972 if (ttd < 0)
1973 ttd = 0;
1974
1975 seq_printf(seq,
1976 "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
1977 "%02X %08X:%08X %02X:%08lX %08X %5u %8d %d %d %pK\n",
1978 i,
1979 src->s6_addr32[0], src->s6_addr32[1],
1980 src->s6_addr32[2], src->s6_addr32[3],
1981 inet_rsk(req)->ir_num,
1982 dest->s6_addr32[0], dest->s6_addr32[1],
1983 dest->s6_addr32[2], dest->s6_addr32[3],
1984 ntohs(inet_rsk(req)->ir_rmt_port),
1985 TCP_SYN_RECV,
1986 0, 0, /* could print option size, but that is af dependent. */
1987 1, /* timers active (only the expire timer) */
1988 jiffies_to_clock_t(ttd),
1989 req->num_timeout,
1990 from_kuid_munged(seq_user_ns(seq),
1991 sock_i_uid(req->rsk_listener)),
1992 0, /* non standard timer */
1993 0, /* open_requests have no inode */
1994 0, req);
1995 }
1996
get_tcp6_sock(struct seq_file * seq,struct sock * sp,int i)1997 static void get_tcp6_sock(struct seq_file *seq, struct sock *sp, int i)
1998 {
1999 const struct in6_addr *dest, *src;
2000 __u16 destp, srcp;
2001 int timer_active;
2002 unsigned long timer_expires;
2003 const struct inet_sock *inet = inet_sk(sp);
2004 const struct tcp_sock *tp = tcp_sk(sp);
2005 const struct inet_connection_sock *icsk = inet_csk(sp);
2006 const struct fastopen_queue *fastopenq = &icsk->icsk_accept_queue.fastopenq;
2007 int rx_queue;
2008 int state;
2009
2010 dest = &sp->sk_v6_daddr;
2011 src = &sp->sk_v6_rcv_saddr;
2012 destp = ntohs(inet->inet_dport);
2013 srcp = ntohs(inet->inet_sport);
2014
2015 if (icsk->icsk_pending == ICSK_TIME_RETRANS ||
2016 icsk->icsk_pending == ICSK_TIME_REO_TIMEOUT ||
2017 icsk->icsk_pending == ICSK_TIME_LOSS_PROBE) {
2018 timer_active = 1;
2019 timer_expires = icsk->icsk_timeout;
2020 } else if (icsk->icsk_pending == ICSK_TIME_PROBE0) {
2021 timer_active = 4;
2022 timer_expires = icsk->icsk_timeout;
2023 } else if (timer_pending(&sp->sk_timer)) {
2024 timer_active = 2;
2025 timer_expires = sp->sk_timer.expires;
2026 } else {
2027 timer_active = 0;
2028 timer_expires = jiffies;
2029 }
2030
2031 state = inet_sk_state_load(sp);
2032 if (state == TCP_LISTEN)
2033 rx_queue = READ_ONCE(sp->sk_ack_backlog);
2034 else
2035 /* Because we don't lock the socket,
2036 * we might find a transient negative value.
2037 */
2038 rx_queue = max_t(int, READ_ONCE(tp->rcv_nxt) -
2039 READ_ONCE(tp->copied_seq), 0);
2040
2041 seq_printf(seq,
2042 "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
2043 "%02X %08X:%08X %02X:%08lX %08X %5u %8d %lu %d %pK %lu %lu %u %u %d\n",
2044 i,
2045 src->s6_addr32[0], src->s6_addr32[1],
2046 src->s6_addr32[2], src->s6_addr32[3], srcp,
2047 dest->s6_addr32[0], dest->s6_addr32[1],
2048 dest->s6_addr32[2], dest->s6_addr32[3], destp,
2049 state,
2050 READ_ONCE(tp->write_seq) - tp->snd_una,
2051 rx_queue,
2052 timer_active,
2053 jiffies_delta_to_clock_t(timer_expires - jiffies),
2054 icsk->icsk_retransmits,
2055 from_kuid_munged(seq_user_ns(seq), sock_i_uid(sp)),
2056 icsk->icsk_probes_out,
2057 sock_i_ino(sp),
2058 refcount_read(&sp->sk_refcnt), sp,
2059 jiffies_to_clock_t(icsk->icsk_rto),
2060 jiffies_to_clock_t(icsk->icsk_ack.ato),
2061 (icsk->icsk_ack.quick << 1) | inet_csk_in_pingpong_mode(sp),
2062 tcp_snd_cwnd(tp),
2063 state == TCP_LISTEN ?
2064 fastopenq->max_qlen :
2065 (tcp_in_initial_slowstart(tp) ? -1 : tp->snd_ssthresh)
2066 );
2067 }
2068
get_timewait6_sock(struct seq_file * seq,struct inet_timewait_sock * tw,int i)2069 static void get_timewait6_sock(struct seq_file *seq,
2070 struct inet_timewait_sock *tw, int i)
2071 {
2072 long delta = tw->tw_timer.expires - jiffies;
2073 const struct in6_addr *dest, *src;
2074 __u16 destp, srcp;
2075
2076 dest = &tw->tw_v6_daddr;
2077 src = &tw->tw_v6_rcv_saddr;
2078 destp = ntohs(tw->tw_dport);
2079 srcp = ntohs(tw->tw_sport);
2080
2081 seq_printf(seq,
2082 "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X "
2083 "%02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %pK\n",
2084 i,
2085 src->s6_addr32[0], src->s6_addr32[1],
2086 src->s6_addr32[2], src->s6_addr32[3], srcp,
2087 dest->s6_addr32[0], dest->s6_addr32[1],
2088 dest->s6_addr32[2], dest->s6_addr32[3], destp,
2089 tw->tw_substate, 0, 0,
2090 3, jiffies_delta_to_clock_t(delta), 0, 0, 0, 0,
2091 refcount_read(&tw->tw_refcnt), tw);
2092 }
2093
tcp6_seq_show(struct seq_file * seq,void * v)2094 static int tcp6_seq_show(struct seq_file *seq, void *v)
2095 {
2096 struct tcp_iter_state *st;
2097 struct sock *sk = v;
2098
2099 if (v == SEQ_START_TOKEN) {
2100 seq_puts(seq,
2101 " sl "
2102 "local_address "
2103 "remote_address "
2104 "st tx_queue rx_queue tr tm->when retrnsmt"
2105 " uid timeout inode\n");
2106 goto out;
2107 }
2108 st = seq->private;
2109
2110 if (sk->sk_state == TCP_TIME_WAIT)
2111 get_timewait6_sock(seq, v, st->num);
2112 else if (sk->sk_state == TCP_NEW_SYN_RECV)
2113 get_openreq6(seq, v, st->num);
2114 else
2115 get_tcp6_sock(seq, v, st->num);
2116 out:
2117 return 0;
2118 }
2119
2120 static const struct seq_operations tcp6_seq_ops = {
2121 .show = tcp6_seq_show,
2122 .start = tcp_seq_start,
2123 .next = tcp_seq_next,
2124 .stop = tcp_seq_stop,
2125 };
2126
2127 static struct tcp_seq_afinfo tcp6_seq_afinfo = {
2128 .family = AF_INET6,
2129 };
2130
tcp6_proc_init(struct net * net)2131 int __net_init tcp6_proc_init(struct net *net)
2132 {
2133 if (!proc_create_net_data("tcp6", 0444, net->proc_net, &tcp6_seq_ops,
2134 sizeof(struct tcp_iter_state), &tcp6_seq_afinfo))
2135 return -ENOMEM;
2136 return 0;
2137 }
2138
tcp6_proc_exit(struct net * net)2139 void tcp6_proc_exit(struct net *net)
2140 {
2141 remove_proc_entry("tcp6", net->proc_net);
2142 }
2143 #endif
2144
2145 struct proto tcpv6_prot = {
2146 .name = "TCPv6",
2147 .owner = THIS_MODULE,
2148 .close = tcp_close,
2149 .pre_connect = tcp_v6_pre_connect,
2150 .connect = tcp_v6_connect,
2151 .disconnect = tcp_disconnect,
2152 .accept = inet_csk_accept,
2153 .ioctl = tcp_ioctl,
2154 .init = tcp_v6_init_sock,
2155 .destroy = tcp_v6_destroy_sock,
2156 .shutdown = tcp_shutdown,
2157 .setsockopt = tcp_setsockopt,
2158 .getsockopt = tcp_getsockopt,
2159 .bpf_bypass_getsockopt = tcp_bpf_bypass_getsockopt,
2160 .keepalive = tcp_set_keepalive,
2161 .recvmsg = tcp_recvmsg,
2162 .sendmsg = tcp_sendmsg,
2163 .sendpage = tcp_sendpage,
2164 .backlog_rcv = tcp_v6_do_rcv,
2165 .release_cb = tcp_release_cb,
2166 .hash = inet6_hash,
2167 .unhash = inet_unhash,
2168 .get_port = inet_csk_get_port,
2169 .put_port = inet_put_port,
2170 #ifdef CONFIG_BPF_SYSCALL
2171 .psock_update_sk_prot = tcp_bpf_update_proto,
2172 #endif
2173 .enter_memory_pressure = tcp_enter_memory_pressure,
2174 .leave_memory_pressure = tcp_leave_memory_pressure,
2175 .stream_memory_free = tcp_stream_memory_free,
2176 .sockets_allocated = &tcp_sockets_allocated,
2177
2178 .memory_allocated = &tcp_memory_allocated,
2179 .per_cpu_fw_alloc = &tcp_memory_per_cpu_fw_alloc,
2180
2181 .memory_pressure = &tcp_memory_pressure,
2182 .orphan_count = &tcp_orphan_count,
2183 .sysctl_mem = sysctl_tcp_mem,
2184 .sysctl_wmem_offset = offsetof(struct net, ipv4.sysctl_tcp_wmem),
2185 .sysctl_rmem_offset = offsetof(struct net, ipv4.sysctl_tcp_rmem),
2186 .max_header = MAX_TCP_HEADER,
2187 .obj_size = sizeof(struct tcp6_sock),
2188 .slab_flags = SLAB_TYPESAFE_BY_RCU,
2189 .twsk_prot = &tcp6_timewait_sock_ops,
2190 .rsk_prot = &tcp6_request_sock_ops,
2191 .h.hashinfo = NULL,
2192 .no_autobind = true,
2193 .diag_destroy = tcp_abort,
2194 };
2195 EXPORT_SYMBOL_GPL(tcpv6_prot);
2196
2197 static const struct inet6_protocol tcpv6_protocol = {
2198 .handler = tcp_v6_rcv,
2199 .err_handler = tcp_v6_err,
2200 .flags = INET6_PROTO_NOPOLICY|INET6_PROTO_FINAL,
2201 };
2202
2203 static struct inet_protosw tcpv6_protosw = {
2204 .type = SOCK_STREAM,
2205 .protocol = IPPROTO_TCP,
2206 .prot = &tcpv6_prot,
2207 .ops = &inet6_stream_ops,
2208 .flags = INET_PROTOSW_PERMANENT |
2209 INET_PROTOSW_ICSK,
2210 };
2211
tcpv6_net_init(struct net * net)2212 static int __net_init tcpv6_net_init(struct net *net)
2213 {
2214 return inet_ctl_sock_create(&net->ipv6.tcp_sk, PF_INET6,
2215 SOCK_RAW, IPPROTO_TCP, net);
2216 }
2217
tcpv6_net_exit(struct net * net)2218 static void __net_exit tcpv6_net_exit(struct net *net)
2219 {
2220 inet_ctl_sock_destroy(net->ipv6.tcp_sk);
2221 }
2222
tcpv6_net_exit_batch(struct list_head * net_exit_list)2223 static void __net_exit tcpv6_net_exit_batch(struct list_head *net_exit_list)
2224 {
2225 tcp_twsk_purge(net_exit_list, AF_INET6);
2226 }
2227
2228 static struct pernet_operations tcpv6_net_ops = {
2229 .init = tcpv6_net_init,
2230 .exit = tcpv6_net_exit,
2231 .exit_batch = tcpv6_net_exit_batch,
2232 };
2233
tcpv6_init(void)2234 int __init tcpv6_init(void)
2235 {
2236 int ret;
2237
2238 ret = inet6_add_protocol(&tcpv6_protocol, IPPROTO_TCP);
2239 if (ret)
2240 goto out;
2241
2242 /* register inet6 protocol */
2243 ret = inet6_register_protosw(&tcpv6_protosw);
2244 if (ret)
2245 goto out_tcpv6_protocol;
2246
2247 ret = register_pernet_subsys(&tcpv6_net_ops);
2248 if (ret)
2249 goto out_tcpv6_protosw;
2250
2251 ret = mptcpv6_init();
2252 if (ret)
2253 goto out_tcpv6_pernet_subsys;
2254
2255 out:
2256 return ret;
2257
2258 out_tcpv6_pernet_subsys:
2259 unregister_pernet_subsys(&tcpv6_net_ops);
2260 out_tcpv6_protosw:
2261 inet6_unregister_protosw(&tcpv6_protosw);
2262 out_tcpv6_protocol:
2263 inet6_del_protocol(&tcpv6_protocol, IPPROTO_TCP);
2264 goto out;
2265 }
2266
tcpv6_exit(void)2267 void tcpv6_exit(void)
2268 {
2269 unregister_pernet_subsys(&tcpv6_net_ops);
2270 inet6_unregister_protosw(&tcpv6_protosw);
2271 inet6_del_protocol(&tcpv6_protocol, IPPROTO_TCP);
2272 }
2273