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/hotdata.h> |
62 | #include <net/busy_poll.h> |
63 | #include <net/rstreason.h> |
64 | |
65 | #include <linux/proc_fs.h> |
66 | #include <linux/seq_file.h> |
67 | |
68 | #include <crypto/hash.h> |
69 | #include <linux/scatterlist.h> |
70 | |
71 | #include <trace/events/tcp.h> |
72 | |
73 | static void tcp_v6_send_reset(const struct sock *sk, struct sk_buff *skb, |
74 | enum sk_rst_reason reason); |
75 | static void tcp_v6_reqsk_send_ack(const struct sock *sk, struct sk_buff *skb, |
76 | struct request_sock *req); |
77 | |
78 | INDIRECT_CALLABLE_SCOPE int tcp_v6_do_rcv(struct sock *sk, struct sk_buff *skb); |
79 | |
80 | static const struct inet_connection_sock_af_ops ipv6_mapped; |
81 | const struct inet_connection_sock_af_ops ipv6_specific; |
82 | #if defined(CONFIG_TCP_MD5SIG) || defined(CONFIG_TCP_AO) |
83 | static const struct tcp_sock_af_ops tcp_sock_ipv6_specific; |
84 | static const struct tcp_sock_af_ops tcp_sock_ipv6_mapped_specific; |
85 | #endif |
86 | |
87 | /* Helper returning the inet6 address from a given tcp socket. |
88 | * It can be used in TCP stack instead of inet6_sk(sk). |
89 | * This avoids a dereference and allow compiler optimizations. |
90 | * It is a specialized version of inet6_sk_generic(). |
91 | */ |
92 | #define tcp_inet6_sk(sk) (&container_of_const(tcp_sk(sk), \ |
93 | struct tcp6_sock, tcp)->inet6) |
94 | |
95 | static void inet6_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb) |
96 | { |
97 | struct dst_entry *dst = skb_dst(skb); |
98 | |
99 | if (dst && dst_hold_safe(dst)) { |
100 | rcu_assign_pointer(sk->sk_rx_dst, dst); |
101 | sk->sk_rx_dst_ifindex = skb->skb_iif; |
102 | sk->sk_rx_dst_cookie = rt6_get_cookie(dst_rt6_info(dst)); |
103 | } |
104 | } |
105 | |
106 | static u32 tcp_v6_init_seq(const struct sk_buff *skb) |
107 | { |
108 | return secure_tcpv6_seq(saddr: ipv6_hdr(skb)->daddr.s6_addr32, |
109 | daddr: ipv6_hdr(skb)->saddr.s6_addr32, |
110 | sport: tcp_hdr(skb)->dest, |
111 | dport: tcp_hdr(skb)->source); |
112 | } |
113 | |
114 | static u32 tcp_v6_init_ts_off(const struct net *net, const struct sk_buff *skb) |
115 | { |
116 | return secure_tcpv6_ts_off(net, saddr: ipv6_hdr(skb)->daddr.s6_addr32, |
117 | daddr: ipv6_hdr(skb)->saddr.s6_addr32); |
118 | } |
119 | |
120 | static int tcp_v6_pre_connect(struct sock *sk, struct sockaddr *uaddr, |
121 | int addr_len) |
122 | { |
123 | /* This check is replicated from tcp_v6_connect() and intended to |
124 | * prevent BPF program called below from accessing bytes that are out |
125 | * of the bound specified by user in addr_len. |
126 | */ |
127 | if (addr_len < SIN6_LEN_RFC2133) |
128 | return -EINVAL; |
129 | |
130 | sock_owned_by_me(sk); |
131 | |
132 | return BPF_CGROUP_RUN_PROG_INET6_CONNECT(sk, uaddr, &addr_len); |
133 | } |
134 | |
135 | static int tcp_v6_connect(struct sock *sk, struct sockaddr *uaddr, |
136 | int addr_len) |
137 | { |
138 | struct sockaddr_in6 *usin = (struct sockaddr_in6 *) uaddr; |
139 | struct inet_connection_sock *icsk = inet_csk(sk); |
140 | struct in6_addr *saddr = NULL, *final_p, final; |
141 | struct inet_timewait_death_row *tcp_death_row; |
142 | struct ipv6_pinfo *np = tcp_inet6_sk(sk); |
143 | struct inet_sock *inet = inet_sk(sk); |
144 | struct tcp_sock *tp = tcp_sk(sk); |
145 | struct net *net = sock_net(sk); |
146 | struct ipv6_txoptions *opt; |
147 | struct dst_entry *dst; |
148 | struct flowi6 fl6; |
149 | int addr_type; |
150 | int err; |
151 | |
152 | if (addr_len < SIN6_LEN_RFC2133) |
153 | return -EINVAL; |
154 | |
155 | if (usin->sin6_family != AF_INET6) |
156 | return -EAFNOSUPPORT; |
157 | |
158 | memset(&fl6, 0, sizeof(fl6)); |
159 | |
160 | if (inet6_test_bit(SNDFLOW, sk)) { |
161 | fl6.flowlabel = usin->sin6_flowinfo&IPV6_FLOWINFO_MASK; |
162 | IP6_ECN_flow_init(fl6.flowlabel); |
163 | if (fl6.flowlabel&IPV6_FLOWLABEL_MASK) { |
164 | struct ip6_flowlabel *flowlabel; |
165 | flowlabel = fl6_sock_lookup(sk, label: fl6.flowlabel); |
166 | if (IS_ERR(ptr: flowlabel)) |
167 | return -EINVAL; |
168 | fl6_sock_release(fl: flowlabel); |
169 | } |
170 | } |
171 | |
172 | /* |
173 | * connect() to INADDR_ANY means loopback (BSD'ism). |
174 | */ |
175 | |
176 | if (ipv6_addr_any(a: &usin->sin6_addr)) { |
177 | if (ipv6_addr_v4mapped(a: &sk->sk_v6_rcv_saddr)) |
178 | ipv6_addr_set_v4mapped(htonl(INADDR_LOOPBACK), |
179 | v4mapped: &usin->sin6_addr); |
180 | else |
181 | usin->sin6_addr = in6addr_loopback; |
182 | } |
183 | |
184 | addr_type = ipv6_addr_type(addr: &usin->sin6_addr); |
185 | |
186 | if (addr_type & IPV6_ADDR_MULTICAST) |
187 | return -ENETUNREACH; |
188 | |
189 | if (addr_type&IPV6_ADDR_LINKLOCAL) { |
190 | if (addr_len >= sizeof(struct sockaddr_in6) && |
191 | usin->sin6_scope_id) { |
192 | /* If interface is set while binding, indices |
193 | * must coincide. |
194 | */ |
195 | if (!sk_dev_equal_l3scope(sk, dif: usin->sin6_scope_id)) |
196 | return -EINVAL; |
197 | |
198 | sk->sk_bound_dev_if = usin->sin6_scope_id; |
199 | } |
200 | |
201 | /* Connect to link-local address requires an interface */ |
202 | if (!sk->sk_bound_dev_if) |
203 | return -EINVAL; |
204 | } |
205 | |
206 | if (tp->rx_opt.ts_recent_stamp && |
207 | !ipv6_addr_equal(a1: &sk->sk_v6_daddr, a2: &usin->sin6_addr)) { |
208 | tp->rx_opt.ts_recent = 0; |
209 | tp->rx_opt.ts_recent_stamp = 0; |
210 | WRITE_ONCE(tp->write_seq, 0); |
211 | } |
212 | |
213 | sk->sk_v6_daddr = usin->sin6_addr; |
214 | np->flow_label = fl6.flowlabel; |
215 | |
216 | /* |
217 | * TCP over IPv4 |
218 | */ |
219 | |
220 | if (addr_type & IPV6_ADDR_MAPPED) { |
221 | u32 exthdrlen = icsk->icsk_ext_hdr_len; |
222 | struct sockaddr_in sin; |
223 | |
224 | if (ipv6_only_sock(sk)) |
225 | return -ENETUNREACH; |
226 | |
227 | sin.sin_family = AF_INET; |
228 | sin.sin_port = usin->sin6_port; |
229 | sin.sin_addr.s_addr = usin->sin6_addr.s6_addr32[3]; |
230 | |
231 | /* Paired with READ_ONCE() in tcp_(get|set)sockopt() */ |
232 | WRITE_ONCE(icsk->icsk_af_ops, &ipv6_mapped); |
233 | if (sk_is_mptcp(sk)) |
234 | mptcpv6_handle_mapped(sk, mapped: true); |
235 | sk->sk_backlog_rcv = tcp_v4_do_rcv; |
236 | #if defined(CONFIG_TCP_MD5SIG) || defined(CONFIG_TCP_AO) |
237 | tp->af_specific = &tcp_sock_ipv6_mapped_specific; |
238 | #endif |
239 | |
240 | err = tcp_v4_connect(sk, uaddr: (struct sockaddr *)&sin, addr_len: sizeof(sin)); |
241 | |
242 | if (err) { |
243 | icsk->icsk_ext_hdr_len = exthdrlen; |
244 | /* Paired with READ_ONCE() in tcp_(get|set)sockopt() */ |
245 | WRITE_ONCE(icsk->icsk_af_ops, &ipv6_specific); |
246 | if (sk_is_mptcp(sk)) |
247 | mptcpv6_handle_mapped(sk, mapped: false); |
248 | sk->sk_backlog_rcv = tcp_v6_do_rcv; |
249 | #if defined(CONFIG_TCP_MD5SIG) || defined(CONFIG_TCP_AO) |
250 | tp->af_specific = &tcp_sock_ipv6_specific; |
251 | #endif |
252 | goto failure; |
253 | } |
254 | np->saddr = sk->sk_v6_rcv_saddr; |
255 | |
256 | return err; |
257 | } |
258 | |
259 | if (!ipv6_addr_any(a: &sk->sk_v6_rcv_saddr)) |
260 | saddr = &sk->sk_v6_rcv_saddr; |
261 | |
262 | fl6.flowi6_proto = IPPROTO_TCP; |
263 | fl6.daddr = sk->sk_v6_daddr; |
264 | fl6.saddr = saddr ? *saddr : np->saddr; |
265 | fl6.flowlabel = ip6_make_flowinfo(tclass: np->tclass, flowlabel: np->flow_label); |
266 | fl6.flowi6_oif = sk->sk_bound_dev_if; |
267 | fl6.flowi6_mark = sk->sk_mark; |
268 | fl6.fl6_dport = usin->sin6_port; |
269 | fl6.fl6_sport = inet->inet_sport; |
270 | if (IS_ENABLED(CONFIG_IP_ROUTE_MULTIPATH) && !fl6.fl6_sport) |
271 | fl6.flowi6_flags = FLOWI_FLAG_ANY_SPORT; |
272 | fl6.flowi6_uid = sk->sk_uid; |
273 | |
274 | opt = rcu_dereference_protected(np->opt, lockdep_sock_is_held(sk)); |
275 | final_p = fl6_update_dst(fl6: &fl6, opt, orig: &final); |
276 | |
277 | security_sk_classify_flow(sk, flic: flowi6_to_flowi_common(fl6: &fl6)); |
278 | |
279 | dst = ip6_dst_lookup_flow(net, sk, fl6: &fl6, final_dst: final_p); |
280 | if (IS_ERR(ptr: dst)) { |
281 | err = PTR_ERR(ptr: dst); |
282 | goto failure; |
283 | } |
284 | |
285 | tp->tcp_usec_ts = dst_tcp_usec_ts(dst); |
286 | tcp_death_row = &sock_net(sk)->ipv4.tcp_death_row; |
287 | |
288 | if (!saddr) { |
289 | saddr = &fl6.saddr; |
290 | |
291 | err = inet_bhash2_update_saddr(sk, saddr, AF_INET6); |
292 | if (err) |
293 | goto failure; |
294 | } |
295 | |
296 | /* set the source address */ |
297 | np->saddr = *saddr; |
298 | inet->inet_rcv_saddr = LOOPBACK4_IPV6; |
299 | |
300 | sk->sk_gso_type = SKB_GSO_TCPV6; |
301 | ip6_dst_store(sk, dst, NULL, NULL); |
302 | |
303 | icsk->icsk_ext_hdr_len = 0; |
304 | if (opt) |
305 | icsk->icsk_ext_hdr_len = opt->opt_flen + |
306 | opt->opt_nflen; |
307 | |
308 | tp->rx_opt.mss_clamp = IPV6_MIN_MTU - sizeof(struct tcphdr) - sizeof(struct ipv6hdr); |
309 | |
310 | inet->inet_dport = usin->sin6_port; |
311 | |
312 | tcp_set_state(sk, state: TCP_SYN_SENT); |
313 | err = inet6_hash_connect(death_row: tcp_death_row, sk); |
314 | if (err) |
315 | goto late_failure; |
316 | |
317 | sk_set_txhash(sk); |
318 | |
319 | if (likely(!tp->repair)) { |
320 | if (!tp->write_seq) |
321 | WRITE_ONCE(tp->write_seq, |
322 | secure_tcpv6_seq(np->saddr.s6_addr32, |
323 | sk->sk_v6_daddr.s6_addr32, |
324 | inet->inet_sport, |
325 | inet->inet_dport)); |
326 | tp->tsoffset = secure_tcpv6_ts_off(net, saddr: np->saddr.s6_addr32, |
327 | daddr: sk->sk_v6_daddr.s6_addr32); |
328 | } |
329 | |
330 | if (tcp_fastopen_defer_connect(sk, err: &err)) |
331 | return err; |
332 | if (err) |
333 | goto late_failure; |
334 | |
335 | err = tcp_connect(sk); |
336 | if (err) |
337 | goto late_failure; |
338 | |
339 | return 0; |
340 | |
341 | late_failure: |
342 | tcp_set_state(sk, state: TCP_CLOSE); |
343 | inet_bhash2_reset_saddr(sk); |
344 | failure: |
345 | inet->inet_dport = 0; |
346 | sk->sk_route_caps = 0; |
347 | return err; |
348 | } |
349 | |
350 | static void tcp_v6_mtu_reduced(struct sock *sk) |
351 | { |
352 | struct dst_entry *dst; |
353 | u32 mtu; |
354 | |
355 | if ((1 << sk->sk_state) & (TCPF_LISTEN | TCPF_CLOSE)) |
356 | return; |
357 | |
358 | mtu = READ_ONCE(tcp_sk(sk)->mtu_info); |
359 | |
360 | /* Drop requests trying to increase our current mss. |
361 | * Check done in __ip6_rt_update_pmtu() is too late. |
362 | */ |
363 | if (tcp_mtu_to_mss(sk, pmtu: mtu) >= tcp_sk(sk)->mss_cache) |
364 | return; |
365 | |
366 | dst = inet6_csk_update_pmtu(sk, mtu); |
367 | if (!dst) |
368 | return; |
369 | |
370 | if (inet_csk(sk)->icsk_pmtu_cookie > dst_mtu(dst)) { |
371 | tcp_sync_mss(sk, pmtu: dst_mtu(dst)); |
372 | tcp_simple_retransmit(sk); |
373 | } |
374 | } |
375 | |
376 | static int tcp_v6_err(struct sk_buff *skb, struct inet6_skb_parm *opt, |
377 | u8 type, u8 code, int offset, __be32 info) |
378 | { |
379 | const struct ipv6hdr *hdr = (const struct ipv6hdr *)skb->data; |
380 | const struct tcphdr *th = (struct tcphdr *)(skb->data+offset); |
381 | struct net *net = dev_net_rcu(dev: skb->dev); |
382 | struct request_sock *fastopen; |
383 | struct ipv6_pinfo *np; |
384 | struct tcp_sock *tp; |
385 | __u32 seq, snd_una; |
386 | struct sock *sk; |
387 | bool fatal; |
388 | int err; |
389 | |
390 | sk = __inet6_lookup_established(net, hashinfo: net->ipv4.tcp_death_row.hashinfo, |
391 | saddr: &hdr->daddr, sport: th->dest, |
392 | daddr: &hdr->saddr, ntohs(th->source), |
393 | dif: skb->dev->ifindex, sdif: inet6_sdif(skb)); |
394 | |
395 | if (!sk) { |
396 | __ICMP6_INC_STATS(net, __in6_dev_get(skb->dev), |
397 | ICMP6_MIB_INERRORS); |
398 | return -ENOENT; |
399 | } |
400 | |
401 | if (sk->sk_state == TCP_TIME_WAIT) { |
402 | /* To increase the counter of ignored icmps for TCP-AO */ |
403 | tcp_ao_ignore_icmp(sk, AF_INET6, type, code); |
404 | inet_twsk_put(tw: inet_twsk(sk)); |
405 | return 0; |
406 | } |
407 | seq = ntohl(th->seq); |
408 | fatal = icmpv6_err_convert(type, code, err: &err); |
409 | if (sk->sk_state == TCP_NEW_SYN_RECV) { |
410 | tcp_req_err(sk, seq, abort: fatal); |
411 | return 0; |
412 | } |
413 | |
414 | if (tcp_ao_ignore_icmp(sk, AF_INET6, type, code)) { |
415 | sock_put(sk); |
416 | return 0; |
417 | } |
418 | |
419 | bh_lock_sock(sk); |
420 | if (sock_owned_by_user(sk) && type != ICMPV6_PKT_TOOBIG) |
421 | __NET_INC_STATS(net, LINUX_MIB_LOCKDROPPEDICMPS); |
422 | |
423 | if (sk->sk_state == TCP_CLOSE) |
424 | goto out; |
425 | |
426 | if (static_branch_unlikely(&ip6_min_hopcount)) { |
427 | /* min_hopcount can be changed concurrently from do_ipv6_setsockopt() */ |
428 | if (ipv6_hdr(skb)->hop_limit < READ_ONCE(tcp_inet6_sk(sk)->min_hopcount)) { |
429 | __NET_INC_STATS(net, LINUX_MIB_TCPMINTTLDROP); |
430 | goto out; |
431 | } |
432 | } |
433 | |
434 | tp = tcp_sk(sk); |
435 | /* XXX (TFO) - tp->snd_una should be ISN (tcp_create_openreq_child() */ |
436 | fastopen = rcu_dereference(tp->fastopen_rsk); |
437 | snd_una = fastopen ? tcp_rsk(req: fastopen)->snt_isn : tp->snd_una; |
438 | if (sk->sk_state != TCP_LISTEN && |
439 | !between(seq1: seq, seq2: snd_una, seq3: tp->snd_nxt)) { |
440 | __NET_INC_STATS(net, LINUX_MIB_OUTOFWINDOWICMPS); |
441 | goto out; |
442 | } |
443 | |
444 | np = tcp_inet6_sk(sk); |
445 | |
446 | if (type == NDISC_REDIRECT) { |
447 | if (!sock_owned_by_user(sk)) { |
448 | struct dst_entry *dst = __sk_dst_check(sk, cookie: np->dst_cookie); |
449 | |
450 | if (dst) |
451 | dst->ops->redirect(dst, sk, skb); |
452 | } |
453 | goto out; |
454 | } |
455 | |
456 | if (type == ICMPV6_PKT_TOOBIG) { |
457 | u32 mtu = ntohl(info); |
458 | |
459 | /* We are not interested in TCP_LISTEN and open_requests |
460 | * (SYN-ACKs send out by Linux are always <576bytes so |
461 | * they should go through unfragmented). |
462 | */ |
463 | if (sk->sk_state == TCP_LISTEN) |
464 | goto out; |
465 | |
466 | if (!ip6_sk_accept_pmtu(sk)) |
467 | goto out; |
468 | |
469 | if (mtu < IPV6_MIN_MTU) |
470 | goto out; |
471 | |
472 | WRITE_ONCE(tp->mtu_info, mtu); |
473 | |
474 | if (!sock_owned_by_user(sk)) |
475 | tcp_v6_mtu_reduced(sk); |
476 | else if (!test_and_set_bit(nr: TCP_MTU_REDUCED_DEFERRED, |
477 | addr: &sk->sk_tsq_flags)) |
478 | sock_hold(sk); |
479 | goto out; |
480 | } |
481 | |
482 | |
483 | /* Might be for an request_sock */ |
484 | switch (sk->sk_state) { |
485 | case TCP_SYN_SENT: |
486 | case TCP_SYN_RECV: |
487 | /* Only in fast or simultaneous open. If a fast open socket is |
488 | * already accepted it is treated as a connected one below. |
489 | */ |
490 | if (fastopen && !fastopen->sk) |
491 | break; |
492 | |
493 | ipv6_icmp_error(sk, skb, err, port: th->dest, ntohl(info), payload: (u8 *)th); |
494 | |
495 | if (!sock_owned_by_user(sk)) |
496 | tcp_done_with_error(sk, err); |
497 | else |
498 | WRITE_ONCE(sk->sk_err_soft, err); |
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) && inet6_test_bit(RECVERR6, sk)) { |
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 | |
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(sk: 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, saddr: &ireq->ir_v6_loc_addr, |
548 | daddr: &ireq->ir_v6_rmt_addr); |
549 | |
550 | fl6->daddr = ireq->ir_v6_rmt_addr; |
551 | if (inet6_test_bit(REPFLOW, sk) && ireq->pktopts) |
552 | fl6->flowlabel = ip6_flowlabel(hdr: ipv6_hdr(skb: 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(dsfield: tclass) && |
560 | tcp_bpf_ca_needs_ecn(sk: (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, mark: skb->mark ? : READ_ONCE(sk->sk_mark), |
568 | opt, tclass, READ_ONCE(sk->sk_priority)); |
569 | rcu_read_unlock(); |
570 | err = net_xmit_eval(err); |
571 | } |
572 | |
573 | done: |
574 | return err; |
575 | } |
576 | |
577 | |
578 | static void tcp_v6_reqsk_destructor(struct request_sock *req) |
579 | { |
580 | kfree(objp: inet_rsk(sk: req)->ipv6_opt); |
581 | consume_skb(skb: inet_rsk(sk: req)->pktopts); |
582 | } |
583 | |
584 | #ifdef CONFIG_TCP_MD5SIG |
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 | addr: (union tcp_md5_addr *)addr, AF_INET6); |
591 | } |
592 | |
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(net: sock_net(sk), |
599 | ifindex: addr_sk->sk_bound_dev_if); |
600 | return tcp_v6_md5_do_lookup(sk, addr: &addr_sk->sk_v6_daddr, |
601 | l3index); |
602 | } |
603 | |
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 | union tcp_ao_addr *addr; |
610 | int l3index = 0; |
611 | u8 prefixlen; |
612 | bool l3flag; |
613 | u8 flags; |
614 | |
615 | if (optlen < sizeof(cmd)) |
616 | return -EINVAL; |
617 | |
618 | if (copy_from_sockptr(dst: &cmd, src: optval, size: 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 | l3flag = cmd.tcpm_flags & TCP_MD5SIG_FLAG_IFINDEX; |
626 | |
627 | if (optname == TCP_MD5SIG_EXT && |
628 | cmd.tcpm_flags & TCP_MD5SIG_FLAG_PREFIX) { |
629 | prefixlen = cmd.tcpm_prefixlen; |
630 | if (prefixlen > 128 || (ipv6_addr_v4mapped(a: &sin6->sin6_addr) && |
631 | prefixlen > 32)) |
632 | return -EINVAL; |
633 | } else { |
634 | prefixlen = ipv6_addr_v4mapped(a: &sin6->sin6_addr) ? 32 : 128; |
635 | } |
636 | |
637 | if (optname == TCP_MD5SIG_EXT && cmd.tcpm_ifindex && |
638 | cmd.tcpm_flags & TCP_MD5SIG_FLAG_IFINDEX) { |
639 | struct net_device *dev; |
640 | |
641 | rcu_read_lock(); |
642 | dev = dev_get_by_index_rcu(net: sock_net(sk), ifindex: cmd.tcpm_ifindex); |
643 | if (dev && netif_is_l3_master(dev)) |
644 | l3index = dev->ifindex; |
645 | rcu_read_unlock(); |
646 | |
647 | /* ok to reference set/not set outside of rcu; |
648 | * right now device MUST be an L3 master |
649 | */ |
650 | if (!dev || !l3index) |
651 | return -EINVAL; |
652 | } |
653 | |
654 | if (!cmd.tcpm_keylen) { |
655 | if (ipv6_addr_v4mapped(a: &sin6->sin6_addr)) |
656 | return tcp_md5_do_del(sk, addr: (union tcp_md5_addr *)&sin6->sin6_addr.s6_addr32[3], |
657 | AF_INET, prefixlen, |
658 | l3index, flags); |
659 | return tcp_md5_do_del(sk, addr: (union tcp_md5_addr *)&sin6->sin6_addr, |
660 | AF_INET6, prefixlen, l3index, flags); |
661 | } |
662 | |
663 | if (cmd.tcpm_keylen > TCP_MD5SIG_MAXKEYLEN) |
664 | return -EINVAL; |
665 | |
666 | if (ipv6_addr_v4mapped(a: &sin6->sin6_addr)) { |
667 | addr = (union tcp_md5_addr *)&sin6->sin6_addr.s6_addr32[3]; |
668 | |
669 | /* Don't allow keys for peers that have a matching TCP-AO key. |
670 | * See the comment in tcp_ao_add_cmd() |
671 | */ |
672 | if (tcp_ao_required(sk, saddr: addr, AF_INET, |
673 | l3index: l3flag ? l3index : -1, stat_inc: false)) |
674 | return -EKEYREJECTED; |
675 | return tcp_md5_do_add(sk, addr, |
676 | AF_INET, prefixlen, l3index, flags, |
677 | newkey: cmd.tcpm_key, newkeylen: cmd.tcpm_keylen); |
678 | } |
679 | |
680 | addr = (union tcp_md5_addr *)&sin6->sin6_addr; |
681 | |
682 | /* Don't allow keys for peers that have a matching TCP-AO key. |
683 | * See the comment in tcp_ao_add_cmd() |
684 | */ |
685 | if (tcp_ao_required(sk, saddr: addr, AF_INET6, l3index: l3flag ? l3index : -1, stat_inc: false)) |
686 | return -EKEYREJECTED; |
687 | |
688 | return tcp_md5_do_add(sk, addr, AF_INET6, prefixlen, l3index, flags, |
689 | newkey: cmd.tcpm_key, newkeylen: cmd.tcpm_keylen); |
690 | } |
691 | |
692 | static int tcp_v6_md5_hash_headers(struct tcp_sigpool *hp, |
693 | const struct in6_addr *daddr, |
694 | const struct in6_addr *saddr, |
695 | const struct tcphdr *th, int nbytes) |
696 | { |
697 | struct tcp6_pseudohdr *bp; |
698 | struct scatterlist sg; |
699 | struct tcphdr *_th; |
700 | |
701 | bp = hp->scratch; |
702 | /* 1. TCP pseudo-header (RFC2460) */ |
703 | bp->saddr = *saddr; |
704 | bp->daddr = *daddr; |
705 | bp->protocol = cpu_to_be32(IPPROTO_TCP); |
706 | bp->len = cpu_to_be32(nbytes); |
707 | |
708 | _th = (struct tcphdr *)(bp + 1); |
709 | memcpy(_th, th, sizeof(*th)); |
710 | _th->check = 0; |
711 | |
712 | sg_init_one(&sg, bp, sizeof(*bp) + sizeof(*th)); |
713 | ahash_request_set_crypt(req: hp->req, src: &sg, NULL, |
714 | nbytes: sizeof(*bp) + sizeof(*th)); |
715 | return crypto_ahash_update(req: hp->req); |
716 | } |
717 | |
718 | static int tcp_v6_md5_hash_hdr(char *md5_hash, const struct tcp_md5sig_key *key, |
719 | const struct in6_addr *daddr, struct in6_addr *saddr, |
720 | const struct tcphdr *th) |
721 | { |
722 | struct tcp_sigpool hp; |
723 | |
724 | if (tcp_sigpool_start(id: tcp_md5_sigpool_id, c: &hp)) |
725 | goto clear_hash_nostart; |
726 | |
727 | if (crypto_ahash_init(req: hp.req)) |
728 | goto clear_hash; |
729 | if (tcp_v6_md5_hash_headers(hp: &hp, daddr, saddr, th, nbytes: th->doff << 2)) |
730 | goto clear_hash; |
731 | if (tcp_md5_hash_key(hp: &hp, key)) |
732 | goto clear_hash; |
733 | ahash_request_set_crypt(req: hp.req, NULL, result: md5_hash, nbytes: 0); |
734 | if (crypto_ahash_final(req: hp.req)) |
735 | goto clear_hash; |
736 | |
737 | tcp_sigpool_end(c: &hp); |
738 | return 0; |
739 | |
740 | clear_hash: |
741 | tcp_sigpool_end(c: &hp); |
742 | clear_hash_nostart: |
743 | memset(md5_hash, 0, 16); |
744 | return 1; |
745 | } |
746 | |
747 | static int tcp_v6_md5_hash_skb(char *md5_hash, |
748 | const struct tcp_md5sig_key *key, |
749 | const struct sock *sk, |
750 | const struct sk_buff *skb) |
751 | { |
752 | const struct tcphdr *th = tcp_hdr(skb); |
753 | const struct in6_addr *saddr, *daddr; |
754 | struct tcp_sigpool hp; |
755 | |
756 | if (sk) { /* valid for establish/request sockets */ |
757 | saddr = &sk->sk_v6_rcv_saddr; |
758 | daddr = &sk->sk_v6_daddr; |
759 | } else { |
760 | const struct ipv6hdr *ip6h = ipv6_hdr(skb); |
761 | saddr = &ip6h->saddr; |
762 | daddr = &ip6h->daddr; |
763 | } |
764 | |
765 | if (tcp_sigpool_start(id: tcp_md5_sigpool_id, c: &hp)) |
766 | goto clear_hash_nostart; |
767 | |
768 | if (crypto_ahash_init(req: hp.req)) |
769 | goto clear_hash; |
770 | |
771 | if (tcp_v6_md5_hash_headers(hp: &hp, daddr, saddr, th, nbytes: skb->len)) |
772 | goto clear_hash; |
773 | if (tcp_sigpool_hash_skb_data(hp: &hp, skb, header_len: th->doff << 2)) |
774 | goto clear_hash; |
775 | if (tcp_md5_hash_key(hp: &hp, key)) |
776 | goto clear_hash; |
777 | ahash_request_set_crypt(req: hp.req, NULL, result: md5_hash, nbytes: 0); |
778 | if (crypto_ahash_final(req: hp.req)) |
779 | goto clear_hash; |
780 | |
781 | tcp_sigpool_end(c: &hp); |
782 | return 0; |
783 | |
784 | clear_hash: |
785 | tcp_sigpool_end(c: &hp); |
786 | clear_hash_nostart: |
787 | memset(md5_hash, 0, 16); |
788 | return 1; |
789 | } |
790 | #endif |
791 | |
792 | static void tcp_v6_init_req(struct request_sock *req, |
793 | const struct sock *sk_listener, |
794 | struct sk_buff *skb, |
795 | u32 tw_isn) |
796 | { |
797 | bool l3_slave = ipv6_l3mdev_skb(TCP_SKB_CB(skb)->header.h6.flags); |
798 | struct inet_request_sock *ireq = inet_rsk(sk: req); |
799 | const struct ipv6_pinfo *np = tcp_inet6_sk(sk_listener); |
800 | |
801 | ireq->ir_v6_rmt_addr = ipv6_hdr(skb)->saddr; |
802 | ireq->ir_v6_loc_addr = ipv6_hdr(skb)->daddr; |
803 | ireq->ir_rmt_addr = LOOPBACK4_IPV6; |
804 | ireq->ir_loc_addr = LOOPBACK4_IPV6; |
805 | |
806 | /* So that link locals have meaning */ |
807 | if ((!sk_listener->sk_bound_dev_if || l3_slave) && |
808 | ipv6_addr_type(addr: &ireq->ir_v6_rmt_addr) & IPV6_ADDR_LINKLOCAL) |
809 | ireq->ir_iif = tcp_v6_iif(skb); |
810 | |
811 | if (!tw_isn && |
812 | (ipv6_opt_accepted(sk: sk_listener, skb, opt: &TCP_SKB_CB(skb)->header.h6) || |
813 | np->rxopt.bits.rxinfo || |
814 | np->rxopt.bits.rxoinfo || np->rxopt.bits.rxhlim || |
815 | np->rxopt.bits.rxohlim || inet6_test_bit(REPFLOW, sk_listener))) { |
816 | refcount_inc(r: &skb->users); |
817 | ireq->pktopts = skb; |
818 | } |
819 | } |
820 | |
821 | static struct dst_entry *tcp_v6_route_req(const struct sock *sk, |
822 | struct sk_buff *skb, |
823 | struct flowi *fl, |
824 | struct request_sock *req, |
825 | u32 tw_isn) |
826 | { |
827 | tcp_v6_init_req(req, sk_listener: sk, skb, tw_isn); |
828 | |
829 | if (security_inet_conn_request(sk, skb, req)) |
830 | return NULL; |
831 | |
832 | return inet6_csk_route_req(sk, fl6: &fl->u.ip6, req, IPPROTO_TCP); |
833 | } |
834 | |
835 | struct request_sock_ops tcp6_request_sock_ops __read_mostly = { |
836 | .family = AF_INET6, |
837 | .obj_size = sizeof(struct tcp6_request_sock), |
838 | .rtx_syn_ack = tcp_rtx_synack, |
839 | .send_ack = tcp_v6_reqsk_send_ack, |
840 | .destructor = tcp_v6_reqsk_destructor, |
841 | .send_reset = tcp_v6_send_reset, |
842 | .syn_ack_timeout = tcp_syn_ack_timeout, |
843 | }; |
844 | |
845 | const struct tcp_request_sock_ops tcp_request_sock_ipv6_ops = { |
846 | .mss_clamp = IPV6_MIN_MTU - sizeof(struct tcphdr) - |
847 | sizeof(struct ipv6hdr), |
848 | #ifdef CONFIG_TCP_MD5SIG |
849 | .req_md5_lookup = tcp_v6_md5_lookup, |
850 | .calc_md5_hash = tcp_v6_md5_hash_skb, |
851 | #endif |
852 | #ifdef CONFIG_TCP_AO |
853 | .ao_lookup = tcp_v6_ao_lookup_rsk, |
854 | .ao_calc_key = tcp_v6_ao_calc_key_rsk, |
855 | .ao_synack_hash = tcp_v6_ao_synack_hash, |
856 | #endif |
857 | #ifdef CONFIG_SYN_COOKIES |
858 | .cookie_init_seq = cookie_v6_init_sequence, |
859 | #endif |
860 | .route_req = tcp_v6_route_req, |
861 | .init_seq = tcp_v6_init_seq, |
862 | .init_ts_off = tcp_v6_init_ts_off, |
863 | .send_synack = tcp_v6_send_synack, |
864 | }; |
865 | |
866 | static void tcp_v6_send_response(const struct sock *sk, struct sk_buff *skb, u32 seq, |
867 | u32 ack, u32 win, u32 tsval, u32 tsecr, |
868 | int oif, int rst, u8 tclass, __be32 label, |
869 | u32 priority, u32 txhash, struct tcp_key *key) |
870 | { |
871 | struct net *net = sk ? sock_net(sk) : dev_net_rcu(dev: skb_dst(skb)->dev); |
872 | unsigned int tot_len = sizeof(struct tcphdr); |
873 | struct sock *ctl_sk = net->ipv6.tcp_sk; |
874 | const struct tcphdr *th = tcp_hdr(skb); |
875 | __be32 mrst = 0, *topt; |
876 | struct dst_entry *dst; |
877 | struct sk_buff *buff; |
878 | struct tcphdr *t1; |
879 | struct flowi6 fl6; |
880 | u32 mark = 0; |
881 | |
882 | if (tsecr) |
883 | tot_len += TCPOLEN_TSTAMP_ALIGNED; |
884 | if (tcp_key_is_md5(key)) |
885 | tot_len += TCPOLEN_MD5SIG_ALIGNED; |
886 | if (tcp_key_is_ao(key)) |
887 | tot_len += tcp_ao_len_aligned(key: key->ao_key); |
888 | |
889 | #ifdef CONFIG_MPTCP |
890 | if (rst && !tcp_key_is_md5(key)) { |
891 | mrst = mptcp_reset_option(skb); |
892 | |
893 | if (mrst) |
894 | tot_len += sizeof(__be32); |
895 | } |
896 | #endif |
897 | |
898 | buff = alloc_skb(MAX_TCP_HEADER, GFP_ATOMIC); |
899 | if (!buff) |
900 | return; |
901 | |
902 | skb_reserve(skb: buff, MAX_TCP_HEADER); |
903 | |
904 | t1 = skb_push(skb: buff, len: tot_len); |
905 | skb_reset_transport_header(skb: buff); |
906 | |
907 | /* Swap the send and the receive. */ |
908 | memset(t1, 0, sizeof(*t1)); |
909 | t1->dest = th->source; |
910 | t1->source = th->dest; |
911 | t1->doff = tot_len / 4; |
912 | t1->seq = htonl(seq); |
913 | t1->ack_seq = htonl(ack); |
914 | t1->ack = !rst || !th->ack; |
915 | t1->rst = rst; |
916 | t1->window = htons(win); |
917 | |
918 | topt = (__be32 *)(t1 + 1); |
919 | |
920 | if (tsecr) { |
921 | *topt++ = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) | |
922 | (TCPOPT_TIMESTAMP << 8) | TCPOLEN_TIMESTAMP); |
923 | *topt++ = htonl(tsval); |
924 | *topt++ = htonl(tsecr); |
925 | } |
926 | |
927 | if (mrst) |
928 | *topt++ = mrst; |
929 | |
930 | #ifdef CONFIG_TCP_MD5SIG |
931 | if (tcp_key_is_md5(key)) { |
932 | *topt++ = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) | |
933 | (TCPOPT_MD5SIG << 8) | TCPOLEN_MD5SIG); |
934 | tcp_v6_md5_hash_hdr(md5_hash: (__u8 *)topt, key: key->md5_key, |
935 | daddr: &ipv6_hdr(skb)->saddr, |
936 | saddr: &ipv6_hdr(skb)->daddr, th: t1); |
937 | } |
938 | #endif |
939 | #ifdef CONFIG_TCP_AO |
940 | if (tcp_key_is_ao(key)) { |
941 | *topt++ = htonl((TCPOPT_AO << 24) | |
942 | (tcp_ao_len(key->ao_key) << 16) | |
943 | (key->ao_key->sndid << 8) | |
944 | (key->rcv_next)); |
945 | |
946 | tcp_ao_hash_hdr(AF_INET6, ao_hash: (char *)topt, key: key->ao_key, |
947 | tkey: key->traffic_key, |
948 | daddr: (union tcp_ao_addr *)&ipv6_hdr(skb)->saddr, |
949 | saddr: (union tcp_ao_addr *)&ipv6_hdr(skb)->daddr, |
950 | th: t1, sne: key->sne); |
951 | } |
952 | #endif |
953 | |
954 | memset(&fl6, 0, sizeof(fl6)); |
955 | fl6.daddr = ipv6_hdr(skb)->saddr; |
956 | fl6.saddr = ipv6_hdr(skb)->daddr; |
957 | fl6.flowlabel = label; |
958 | |
959 | buff->ip_summed = CHECKSUM_PARTIAL; |
960 | |
961 | __tcp_v6_send_check(skb: buff, saddr: &fl6.saddr, daddr: &fl6.daddr); |
962 | |
963 | fl6.flowi6_proto = IPPROTO_TCP; |
964 | if (rt6_need_strict(daddr: &fl6.daddr) && !oif) |
965 | fl6.flowi6_oif = tcp_v6_iif(skb); |
966 | else { |
967 | if (!oif && netif_index_is_l3_master(net, ifindex: skb->skb_iif)) |
968 | oif = skb->skb_iif; |
969 | |
970 | fl6.flowi6_oif = oif; |
971 | } |
972 | |
973 | if (sk) { |
974 | /* unconstify the socket only to attach it to buff with care. */ |
975 | skb_set_owner_edemux(skb: buff, sk: (struct sock *)sk); |
976 | |
977 | if (sk->sk_state == TCP_TIME_WAIT) |
978 | mark = inet_twsk(sk)->tw_mark; |
979 | else |
980 | mark = READ_ONCE(sk->sk_mark); |
981 | skb_set_delivery_time(skb: buff, kt: tcp_transmit_time(sk), tstamp_type: SKB_CLOCK_MONOTONIC); |
982 | } |
983 | if (txhash) { |
984 | /* autoflowlabel/skb_get_hash_flowi6 rely on buff->hash */ |
985 | skb_set_hash(skb: buff, hash: txhash, type: PKT_HASH_TYPE_L4); |
986 | } |
987 | fl6.flowi6_mark = IP6_REPLY_MARK(net, skb->mark) ?: mark; |
988 | fl6.fl6_dport = t1->dest; |
989 | fl6.fl6_sport = t1->source; |
990 | fl6.flowi6_uid = sock_net_uid(net, sk: sk && sk_fullsock(sk) ? sk : NULL); |
991 | security_skb_classify_flow(skb, flic: flowi6_to_flowi_common(fl6: &fl6)); |
992 | |
993 | /* Pass a socket to ip6_dst_lookup either it is for RST |
994 | * Underlying function will use this to retrieve the network |
995 | * namespace |
996 | */ |
997 | if (sk && sk->sk_state != TCP_TIME_WAIT) |
998 | dst = ip6_dst_lookup_flow(net, sk, fl6: &fl6, NULL); /*sk's xfrm_policy can be referred*/ |
999 | else |
1000 | dst = ip6_dst_lookup_flow(net, sk: ctl_sk, fl6: &fl6, NULL); |
1001 | if (!IS_ERR(ptr: dst)) { |
1002 | skb_dst_set(skb: buff, dst); |
1003 | ip6_xmit(sk: ctl_sk, skb: buff, fl6: &fl6, mark: fl6.flowi6_mark, NULL, |
1004 | tclass, priority); |
1005 | TCP_INC_STATS(net, TCP_MIB_OUTSEGS); |
1006 | if (rst) |
1007 | TCP_INC_STATS(net, TCP_MIB_OUTRSTS); |
1008 | return; |
1009 | } |
1010 | |
1011 | kfree_skb(skb: buff); |
1012 | } |
1013 | |
1014 | static void tcp_v6_send_reset(const struct sock *sk, struct sk_buff *skb, |
1015 | enum sk_rst_reason reason) |
1016 | { |
1017 | const struct tcphdr *th = tcp_hdr(skb); |
1018 | struct ipv6hdr *ipv6h = ipv6_hdr(skb); |
1019 | const __u8 *md5_hash_location = NULL; |
1020 | #if defined(CONFIG_TCP_MD5SIG) || defined(CONFIG_TCP_AO) |
1021 | bool allocated_traffic_key = false; |
1022 | #endif |
1023 | const struct tcp_ao_hdr *aoh; |
1024 | struct tcp_key key = {}; |
1025 | u32 seq = 0, ack_seq = 0; |
1026 | __be32 label = 0; |
1027 | u32 priority = 0; |
1028 | struct net *net; |
1029 | u32 txhash = 0; |
1030 | int oif = 0; |
1031 | #ifdef CONFIG_TCP_MD5SIG |
1032 | unsigned char newhash[16]; |
1033 | int genhash; |
1034 | struct sock *sk1 = NULL; |
1035 | #endif |
1036 | |
1037 | if (th->rst) |
1038 | return; |
1039 | |
1040 | /* If sk not NULL, it means we did a successful lookup and incoming |
1041 | * route had to be correct. prequeue might have dropped our dst. |
1042 | */ |
1043 | if (!sk && !ipv6_unicast_destination(skb)) |
1044 | return; |
1045 | |
1046 | net = sk ? sock_net(sk) : dev_net_rcu(dev: skb_dst(skb)->dev); |
1047 | /* Invalid TCP option size or twice included auth */ |
1048 | if (tcp_parse_auth_options(th, md5_hash: &md5_hash_location, aoh: &aoh)) |
1049 | return; |
1050 | #if defined(CONFIG_TCP_MD5SIG) || defined(CONFIG_TCP_AO) |
1051 | rcu_read_lock(); |
1052 | #endif |
1053 | #ifdef CONFIG_TCP_MD5SIG |
1054 | if (sk && sk_fullsock(sk)) { |
1055 | int l3index; |
1056 | |
1057 | /* sdif set, means packet ingressed via a device |
1058 | * in an L3 domain and inet_iif is set to it. |
1059 | */ |
1060 | l3index = tcp_v6_sdif(skb) ? tcp_v6_iif_l3_slave(skb) : 0; |
1061 | key.md5_key = tcp_v6_md5_do_lookup(sk, addr: &ipv6h->saddr, l3index); |
1062 | if (key.md5_key) |
1063 | key.type = TCP_KEY_MD5; |
1064 | } else if (md5_hash_location) { |
1065 | int dif = tcp_v6_iif_l3_slave(skb); |
1066 | int sdif = tcp_v6_sdif(skb); |
1067 | int l3index; |
1068 | |
1069 | /* |
1070 | * active side is lost. Try to find listening socket through |
1071 | * source port, and then find md5 key through listening socket. |
1072 | * we are not loose security here: |
1073 | * Incoming packet is checked with md5 hash with finding key, |
1074 | * no RST generated if md5 hash doesn't match. |
1075 | */ |
1076 | sk1 = inet6_lookup_listener(net, hashinfo: net->ipv4.tcp_death_row.hashinfo, |
1077 | NULL, doff: 0, saddr: &ipv6h->saddr, sport: th->source, |
1078 | daddr: &ipv6h->daddr, ntohs(th->source), |
1079 | dif, sdif); |
1080 | if (!sk1) |
1081 | goto out; |
1082 | |
1083 | /* sdif set, means packet ingressed via a device |
1084 | * in an L3 domain and dif is set to it. |
1085 | */ |
1086 | l3index = tcp_v6_sdif(skb) ? dif : 0; |
1087 | |
1088 | key.md5_key = tcp_v6_md5_do_lookup(sk: sk1, addr: &ipv6h->saddr, l3index); |
1089 | if (!key.md5_key) |
1090 | goto out; |
1091 | key.type = TCP_KEY_MD5; |
1092 | |
1093 | genhash = tcp_v6_md5_hash_skb(md5_hash: newhash, key: key.md5_key, NULL, skb); |
1094 | if (genhash || memcmp(p: md5_hash_location, q: newhash, size: 16) != 0) |
1095 | goto out; |
1096 | } |
1097 | #endif |
1098 | |
1099 | if (th->ack) |
1100 | seq = ntohl(th->ack_seq); |
1101 | else |
1102 | ack_seq = ntohl(th->seq) + th->syn + th->fin + skb->len - |
1103 | (th->doff << 2); |
1104 | |
1105 | #ifdef CONFIG_TCP_AO |
1106 | if (aoh) { |
1107 | int l3index; |
1108 | |
1109 | l3index = tcp_v6_sdif(skb) ? tcp_v6_iif_l3_slave(skb) : 0; |
1110 | if (tcp_ao_prepare_reset(sk, skb, aoh, l3index, seq, |
1111 | key: &key.ao_key, traffic_key: &key.traffic_key, |
1112 | allocated_traffic_key: &allocated_traffic_key, |
1113 | keyid: &key.rcv_next, sne: &key.sne)) |
1114 | goto out; |
1115 | key.type = TCP_KEY_AO; |
1116 | } |
1117 | #endif |
1118 | |
1119 | if (sk) { |
1120 | oif = sk->sk_bound_dev_if; |
1121 | if (sk_fullsock(sk)) { |
1122 | if (inet6_test_bit(REPFLOW, sk)) |
1123 | label = ip6_flowlabel(hdr: ipv6h); |
1124 | priority = READ_ONCE(sk->sk_priority); |
1125 | txhash = sk->sk_txhash; |
1126 | } |
1127 | if (sk->sk_state == TCP_TIME_WAIT) { |
1128 | label = cpu_to_be32(inet_twsk(sk)->tw_flowlabel); |
1129 | priority = inet_twsk(sk)->tw_priority; |
1130 | txhash = inet_twsk(sk)->tw_txhash; |
1131 | } |
1132 | } else { |
1133 | if (net->ipv6.sysctl.flowlabel_reflect & FLOWLABEL_REFLECT_TCP_RESET) |
1134 | label = ip6_flowlabel(hdr: ipv6h); |
1135 | } |
1136 | |
1137 | trace_tcp_send_reset(sk, skb__nullable: skb, reason); |
1138 | |
1139 | tcp_v6_send_response(sk, skb, seq, ack: ack_seq, win: 0, tsval: 0, tsecr: 0, oif, rst: 1, |
1140 | tclass: ipv6_get_dsfield(ipv6h) & ~INET_ECN_MASK, |
1141 | label, priority, txhash, |
1142 | key: &key); |
1143 | |
1144 | #if defined(CONFIG_TCP_MD5SIG) || defined(CONFIG_TCP_AO) |
1145 | out: |
1146 | if (allocated_traffic_key) |
1147 | kfree(objp: key.traffic_key); |
1148 | rcu_read_unlock(); |
1149 | #endif |
1150 | } |
1151 | |
1152 | static void tcp_v6_send_ack(const struct sock *sk, struct sk_buff *skb, u32 seq, |
1153 | u32 ack, u32 win, u32 tsval, u32 tsecr, int oif, |
1154 | struct tcp_key *key, u8 tclass, |
1155 | __be32 label, u32 priority, u32 txhash) |
1156 | { |
1157 | tcp_v6_send_response(sk, skb, seq, ack, win, tsval, tsecr, oif, rst: 0, |
1158 | tclass, label, priority, txhash, key); |
1159 | } |
1160 | |
1161 | static void tcp_v6_timewait_ack(struct sock *sk, struct sk_buff *skb, |
1162 | enum tcp_tw_status tw_status) |
1163 | { |
1164 | struct inet_timewait_sock *tw = inet_twsk(sk); |
1165 | struct tcp_timewait_sock *tcptw = tcp_twsk(sk); |
1166 | u8 tclass = tw->tw_tclass; |
1167 | struct tcp_key key = {}; |
1168 | |
1169 | if (tw_status == TCP_TW_ACK_OOW) |
1170 | tclass &= ~INET_ECN_MASK; |
1171 | #ifdef CONFIG_TCP_AO |
1172 | struct tcp_ao_info *ao_info; |
1173 | |
1174 | if (static_branch_unlikely(&tcp_ao_needed.key)) { |
1175 | |
1176 | /* FIXME: the segment to-be-acked is not verified yet */ |
1177 | ao_info = rcu_dereference(tcptw->ao_info); |
1178 | if (ao_info) { |
1179 | const struct tcp_ao_hdr *aoh; |
1180 | |
1181 | /* Invalid TCP option size or twice included auth */ |
1182 | if (tcp_parse_auth_options(th: tcp_hdr(skb), NULL, aoh: &aoh)) |
1183 | goto out; |
1184 | if (aoh) |
1185 | key.ao_key = tcp_ao_established_key(sk, ao: ao_info, |
1186 | sndid: aoh->rnext_keyid, rcvid: -1); |
1187 | } |
1188 | } |
1189 | if (key.ao_key) { |
1190 | struct tcp_ao_key *rnext_key; |
1191 | |
1192 | key.traffic_key = snd_other_key(key: key.ao_key); |
1193 | /* rcv_next switches to our rcv_next */ |
1194 | rnext_key = READ_ONCE(ao_info->rnext_key); |
1195 | key.rcv_next = rnext_key->rcvid; |
1196 | key.sne = READ_ONCE(ao_info->snd_sne); |
1197 | key.type = TCP_KEY_AO; |
1198 | #else |
1199 | if (0) { |
1200 | #endif |
1201 | #ifdef CONFIG_TCP_MD5SIG |
1202 | } else if (static_branch_unlikely(&tcp_md5_needed.key)) { |
1203 | key.md5_key = tcp_twsk_md5_key(tcptw); |
1204 | if (key.md5_key) |
1205 | key.type = TCP_KEY_MD5; |
1206 | #endif |
1207 | } |
1208 | |
1209 | tcp_v6_send_ack(sk, skb, seq: tcptw->tw_snd_nxt, |
1210 | READ_ONCE(tcptw->tw_rcv_nxt), |
1211 | win: tcptw->tw_rcv_wnd >> tw->tw_rcv_wscale, |
1212 | tsval: tcp_tw_tsval(tcptw), |
1213 | READ_ONCE(tcptw->tw_ts_recent), oif: tw->tw_bound_dev_if, |
1214 | key: &key, tclass, cpu_to_be32(tw->tw_flowlabel), |
1215 | priority: tw->tw_priority, txhash: tw->tw_txhash); |
1216 | |
1217 | #ifdef CONFIG_TCP_AO |
1218 | out: |
1219 | #endif |
1220 | inet_twsk_put(tw); |
1221 | } |
1222 | |
1223 | static void tcp_v6_reqsk_send_ack(const struct sock *sk, struct sk_buff *skb, |
1224 | struct request_sock *req) |
1225 | { |
1226 | struct tcp_key key = {}; |
1227 | |
1228 | #ifdef CONFIG_TCP_AO |
1229 | if (static_branch_unlikely(&tcp_ao_needed.key) && |
1230 | tcp_rsk_used_ao(req)) { |
1231 | const struct in6_addr *addr = &ipv6_hdr(skb)->saddr; |
1232 | const struct tcp_ao_hdr *aoh; |
1233 | int l3index; |
1234 | |
1235 | l3index = tcp_v6_sdif(skb) ? tcp_v6_iif_l3_slave(skb) : 0; |
1236 | /* Invalid TCP option size or twice included auth */ |
1237 | if (tcp_parse_auth_options(th: tcp_hdr(skb), NULL, aoh: &aoh)) |
1238 | return; |
1239 | if (!aoh) |
1240 | return; |
1241 | key.ao_key = tcp_ao_do_lookup(sk, l3index, |
1242 | addr: (union tcp_ao_addr *)addr, |
1243 | AF_INET6, sndid: aoh->rnext_keyid, rcvid: -1); |
1244 | if (unlikely(!key.ao_key)) { |
1245 | /* Send ACK with any matching MKT for the peer */ |
1246 | key.ao_key = tcp_ao_do_lookup(sk, l3index, |
1247 | addr: (union tcp_ao_addr *)addr, |
1248 | AF_INET6, sndid: -1, rcvid: -1); |
1249 | /* Matching key disappeared (user removed the key?) |
1250 | * let the handshake timeout. |
1251 | */ |
1252 | if (!key.ao_key) { |
1253 | net_info_ratelimited("TCP-AO key for (%pI6, %d)->(%pI6, %d) suddenly disappeared, won't ACK new connection\n", |
1254 | addr, |
1255 | ntohs(tcp_hdr(skb)->source), |
1256 | &ipv6_hdr(skb)->daddr, |
1257 | ntohs(tcp_hdr(skb)->dest)); |
1258 | return; |
1259 | } |
1260 | } |
1261 | key.traffic_key = kmalloc(tcp_ao_digest_size(key.ao_key), GFP_ATOMIC); |
1262 | if (!key.traffic_key) |
1263 | return; |
1264 | |
1265 | key.type = TCP_KEY_AO; |
1266 | key.rcv_next = aoh->keyid; |
1267 | tcp_v6_ao_calc_key_rsk(mkt: key.ao_key, key: key.traffic_key, req); |
1268 | #else |
1269 | if (0) { |
1270 | #endif |
1271 | #ifdef CONFIG_TCP_MD5SIG |
1272 | } else if (static_branch_unlikely(&tcp_md5_needed.key)) { |
1273 | int l3index = tcp_v6_sdif(skb) ? tcp_v6_iif_l3_slave(skb) : 0; |
1274 | |
1275 | key.md5_key = tcp_v6_md5_do_lookup(sk, addr: &ipv6_hdr(skb)->saddr, |
1276 | l3index); |
1277 | if (key.md5_key) |
1278 | key.type = TCP_KEY_MD5; |
1279 | #endif |
1280 | } |
1281 | |
1282 | /* sk->sk_state == TCP_LISTEN -> for regular TCP_SYN_RECV |
1283 | * sk->sk_state == TCP_SYN_RECV -> for Fast Open. |
1284 | */ |
1285 | tcp_v6_send_ack(sk, skb, seq: (sk->sk_state == TCP_LISTEN) ? |
1286 | tcp_rsk(req)->snt_isn + 1 : tcp_sk(sk)->snd_nxt, |
1287 | ack: tcp_rsk(req)->rcv_nxt, |
1288 | win: tcp_synack_window(req) >> inet_rsk(sk: req)->rcv_wscale, |
1289 | tsval: tcp_rsk_tsval(treq: tcp_rsk(req)), |
1290 | tsecr: req->ts_recent, oif: sk->sk_bound_dev_if, |
1291 | key: &key, tclass: ipv6_get_dsfield(ipv6h: ipv6_hdr(skb)) & ~INET_ECN_MASK, |
1292 | label: 0, |
1293 | READ_ONCE(sk->sk_priority), |
1294 | READ_ONCE(tcp_rsk(req)->txhash)); |
1295 | if (tcp_key_is_ao(key: &key)) |
1296 | kfree(objp: key.traffic_key); |
1297 | } |
1298 | |
1299 | |
1300 | static struct sock *tcp_v6_cookie_check(struct sock *sk, struct sk_buff *skb) |
1301 | { |
1302 | #ifdef CONFIG_SYN_COOKIES |
1303 | const struct tcphdr *th = tcp_hdr(skb); |
1304 | |
1305 | if (!th->syn) |
1306 | sk = cookie_v6_check(sk, skb); |
1307 | #endif |
1308 | return sk; |
1309 | } |
1310 | |
1311 | u16 tcp_v6_get_syncookie(struct sock *sk, struct ipv6hdr *iph, |
1312 | struct tcphdr *th, u32 *cookie) |
1313 | { |
1314 | u16 mss = 0; |
1315 | #ifdef CONFIG_SYN_COOKIES |
1316 | mss = tcp_get_syncookie_mss(rsk_ops: &tcp6_request_sock_ops, |
1317 | af_ops: &tcp_request_sock_ipv6_ops, sk, th); |
1318 | if (mss) { |
1319 | *cookie = __cookie_v6_init_sequence(iph, th, mssp: &mss); |
1320 | tcp_synq_overflow(sk); |
1321 | } |
1322 | #endif |
1323 | return mss; |
1324 | } |
1325 | |
1326 | static int tcp_v6_conn_request(struct sock *sk, struct sk_buff *skb) |
1327 | { |
1328 | if (skb->protocol == htons(ETH_P_IP)) |
1329 | return tcp_v4_conn_request(sk, skb); |
1330 | |
1331 | if (!ipv6_unicast_destination(skb)) |
1332 | goto drop; |
1333 | |
1334 | if (ipv6_addr_v4mapped(a: &ipv6_hdr(skb)->saddr)) { |
1335 | __IP6_INC_STATS(sock_net(sk), NULL, IPSTATS_MIB_INHDRERRORS); |
1336 | return 0; |
1337 | } |
1338 | |
1339 | return tcp_conn_request(rsk_ops: &tcp6_request_sock_ops, |
1340 | af_ops: &tcp_request_sock_ipv6_ops, sk, skb); |
1341 | |
1342 | drop: |
1343 | tcp_listendrop(sk); |
1344 | return 0; /* don't send reset */ |
1345 | } |
1346 | |
1347 | static void tcp_v6_restore_cb(struct sk_buff *skb) |
1348 | { |
1349 | /* We need to move header back to the beginning if xfrm6_policy_check() |
1350 | * and tcp_v6_fill_cb() are going to be called again. |
1351 | * ip6_datagram_recv_specific_ctl() also expects IP6CB to be there. |
1352 | */ |
1353 | memmove(IP6CB(skb), &TCP_SKB_CB(skb)->header.h6, |
1354 | sizeof(struct inet6_skb_parm)); |
1355 | } |
1356 | |
1357 | static struct sock *tcp_v6_syn_recv_sock(const struct sock *sk, struct sk_buff *skb, |
1358 | struct request_sock *req, |
1359 | struct dst_entry *dst, |
1360 | struct request_sock *req_unhash, |
1361 | bool *own_req) |
1362 | { |
1363 | struct inet_request_sock *ireq; |
1364 | struct ipv6_pinfo *newnp; |
1365 | const struct ipv6_pinfo *np = tcp_inet6_sk(sk); |
1366 | struct ipv6_txoptions *opt; |
1367 | struct inet_sock *newinet; |
1368 | bool found_dup_sk = false; |
1369 | struct tcp_sock *newtp; |
1370 | struct sock *newsk; |
1371 | #ifdef CONFIG_TCP_MD5SIG |
1372 | struct tcp_md5sig_key *key; |
1373 | int l3index; |
1374 | #endif |
1375 | struct flowi6 fl6; |
1376 | |
1377 | if (skb->protocol == htons(ETH_P_IP)) { |
1378 | /* |
1379 | * v6 mapped |
1380 | */ |
1381 | |
1382 | newsk = tcp_v4_syn_recv_sock(sk, skb, req, dst, |
1383 | req_unhash, own_req); |
1384 | |
1385 | if (!newsk) |
1386 | return NULL; |
1387 | |
1388 | inet_sk(newsk)->pinet6 = tcp_inet6_sk(newsk); |
1389 | |
1390 | newnp = tcp_inet6_sk(newsk); |
1391 | newtp = tcp_sk(newsk); |
1392 | |
1393 | memcpy(newnp, np, sizeof(struct ipv6_pinfo)); |
1394 | |
1395 | newnp->saddr = newsk->sk_v6_rcv_saddr; |
1396 | |
1397 | inet_csk(newsk)->icsk_af_ops = &ipv6_mapped; |
1398 | if (sk_is_mptcp(sk: newsk)) |
1399 | mptcpv6_handle_mapped(sk: newsk, mapped: true); |
1400 | newsk->sk_backlog_rcv = tcp_v4_do_rcv; |
1401 | #if defined(CONFIG_TCP_MD5SIG) || defined(CONFIG_TCP_AO) |
1402 | newtp->af_specific = &tcp_sock_ipv6_mapped_specific; |
1403 | #endif |
1404 | |
1405 | newnp->ipv6_mc_list = NULL; |
1406 | newnp->ipv6_ac_list = NULL; |
1407 | newnp->ipv6_fl_list = NULL; |
1408 | newnp->pktoptions = NULL; |
1409 | newnp->opt = NULL; |
1410 | newnp->mcast_oif = inet_iif(skb); |
1411 | newnp->mcast_hops = ip_hdr(skb)->ttl; |
1412 | newnp->rcv_flowinfo = 0; |
1413 | if (inet6_test_bit(REPFLOW, sk)) |
1414 | newnp->flow_label = 0; |
1415 | |
1416 | /* |
1417 | * No need to charge this sock to the relevant IPv6 refcnt debug socks count |
1418 | * here, tcp_create_openreq_child now does this for us, see the comment in |
1419 | * that function for the gory details. -acme |
1420 | */ |
1421 | |
1422 | /* It is tricky place. Until this moment IPv4 tcp |
1423 | worked with IPv6 icsk.icsk_af_ops. |
1424 | Sync it now. |
1425 | */ |
1426 | tcp_sync_mss(sk: newsk, inet_csk(newsk)->icsk_pmtu_cookie); |
1427 | |
1428 | return newsk; |
1429 | } |
1430 | |
1431 | ireq = inet_rsk(sk: req); |
1432 | |
1433 | if (sk_acceptq_is_full(sk)) |
1434 | goto out_overflow; |
1435 | |
1436 | if (!dst) { |
1437 | dst = inet6_csk_route_req(sk, fl6: &fl6, req, IPPROTO_TCP); |
1438 | if (!dst) |
1439 | goto out; |
1440 | } |
1441 | |
1442 | newsk = tcp_create_openreq_child(sk, req, skb); |
1443 | if (!newsk) |
1444 | goto out_nonewsk; |
1445 | |
1446 | /* |
1447 | * No need to charge this sock to the relevant IPv6 refcnt debug socks |
1448 | * count here, tcp_create_openreq_child now does this for us, see the |
1449 | * comment in that function for the gory details. -acme |
1450 | */ |
1451 | |
1452 | newsk->sk_gso_type = SKB_GSO_TCPV6; |
1453 | inet6_sk_rx_dst_set(sk: newsk, skb); |
1454 | |
1455 | inet_sk(newsk)->pinet6 = tcp_inet6_sk(newsk); |
1456 | |
1457 | newtp = tcp_sk(newsk); |
1458 | newinet = inet_sk(newsk); |
1459 | newnp = tcp_inet6_sk(newsk); |
1460 | |
1461 | memcpy(newnp, np, sizeof(struct ipv6_pinfo)); |
1462 | |
1463 | ip6_dst_store(sk: newsk, dst, NULL, NULL); |
1464 | |
1465 | newnp->saddr = ireq->ir_v6_loc_addr; |
1466 | |
1467 | /* Now IPv6 options... |
1468 | |
1469 | First: no IPv4 options. |
1470 | */ |
1471 | newinet->inet_opt = NULL; |
1472 | newnp->ipv6_mc_list = NULL; |
1473 | newnp->ipv6_ac_list = NULL; |
1474 | newnp->ipv6_fl_list = NULL; |
1475 | |
1476 | /* Clone RX bits */ |
1477 | newnp->rxopt.all = np->rxopt.all; |
1478 | |
1479 | newnp->pktoptions = NULL; |
1480 | newnp->opt = NULL; |
1481 | newnp->mcast_oif = tcp_v6_iif(skb); |
1482 | newnp->mcast_hops = ipv6_hdr(skb)->hop_limit; |
1483 | newnp->rcv_flowinfo = ip6_flowinfo(hdr: ipv6_hdr(skb)); |
1484 | if (inet6_test_bit(REPFLOW, sk)) |
1485 | newnp->flow_label = ip6_flowlabel(hdr: ipv6_hdr(skb)); |
1486 | |
1487 | /* Set ToS of the new socket based upon the value of incoming SYN. |
1488 | * ECT bits are set later in tcp_init_transfer(). |
1489 | */ |
1490 | if (READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_reflect_tos)) |
1491 | newnp->tclass = tcp_rsk(req)->syn_tos & ~INET_ECN_MASK; |
1492 | |
1493 | /* Clone native IPv6 options from listening socket (if any) |
1494 | |
1495 | Yes, keeping reference count would be much more clever, |
1496 | but we make one more one thing there: reattach optmem |
1497 | to newsk. |
1498 | */ |
1499 | opt = ireq->ipv6_opt; |
1500 | if (!opt) |
1501 | opt = rcu_dereference(np->opt); |
1502 | if (opt) { |
1503 | opt = ipv6_dup_options(sk: newsk, opt); |
1504 | RCU_INIT_POINTER(newnp->opt, opt); |
1505 | } |
1506 | inet_csk(newsk)->icsk_ext_hdr_len = 0; |
1507 | if (opt) |
1508 | inet_csk(newsk)->icsk_ext_hdr_len = opt->opt_nflen + |
1509 | opt->opt_flen; |
1510 | |
1511 | tcp_ca_openreq_child(sk: newsk, dst); |
1512 | |
1513 | tcp_sync_mss(sk: newsk, pmtu: dst_mtu(dst)); |
1514 | newtp->advmss = tcp_mss_clamp(tcp_sk(sk), mss: dst_metric_advmss(dst)); |
1515 | |
1516 | tcp_initialize_rcv_mss(sk: newsk); |
1517 | |
1518 | #ifdef CONFIG_TCP_MD5SIG |
1519 | l3index = l3mdev_master_ifindex_by_index(net: sock_net(sk), ifindex: ireq->ir_iif); |
1520 | |
1521 | if (!tcp_rsk_used_ao(req)) { |
1522 | /* Copy over the MD5 key from the original socket */ |
1523 | key = tcp_v6_md5_do_lookup(sk, addr: &newsk->sk_v6_daddr, l3index); |
1524 | if (key) { |
1525 | const union tcp_md5_addr *addr; |
1526 | |
1527 | addr = (union tcp_md5_addr *)&newsk->sk_v6_daddr; |
1528 | if (tcp_md5_key_copy(sk: newsk, addr, AF_INET6, prefixlen: 128, l3index, key)) { |
1529 | inet_csk_prepare_forced_close(sk: newsk); |
1530 | tcp_done(sk: newsk); |
1531 | goto out; |
1532 | } |
1533 | } |
1534 | } |
1535 | #endif |
1536 | #ifdef CONFIG_TCP_AO |
1537 | /* Copy over tcp_ao_info if any */ |
1538 | if (tcp_ao_copy_all_matching(sk, newsk, req, skb, AF_INET6)) |
1539 | goto out; /* OOM */ |
1540 | #endif |
1541 | |
1542 | if (__inet_inherit_port(sk, child: newsk) < 0) { |
1543 | inet_csk_prepare_forced_close(sk: newsk); |
1544 | tcp_done(sk: newsk); |
1545 | goto out; |
1546 | } |
1547 | *own_req = inet_ehash_nolisten(sk: newsk, osk: req_to_sk(req: req_unhash), |
1548 | found_dup_sk: &found_dup_sk); |
1549 | if (*own_req) { |
1550 | tcp_move_syn(tp: newtp, req); |
1551 | |
1552 | /* Clone pktoptions received with SYN, if we own the req */ |
1553 | if (ireq->pktopts) { |
1554 | newnp->pktoptions = skb_clone_and_charge_r(skb: ireq->pktopts, sk: newsk); |
1555 | consume_skb(skb: ireq->pktopts); |
1556 | ireq->pktopts = NULL; |
1557 | if (newnp->pktoptions) |
1558 | tcp_v6_restore_cb(skb: newnp->pktoptions); |
1559 | } |
1560 | } else { |
1561 | if (!req_unhash && found_dup_sk) { |
1562 | /* This code path should only be executed in the |
1563 | * syncookie case only |
1564 | */ |
1565 | bh_unlock_sock(newsk); |
1566 | sock_put(sk: newsk); |
1567 | newsk = NULL; |
1568 | } |
1569 | } |
1570 | |
1571 | return newsk; |
1572 | |
1573 | out_overflow: |
1574 | __NET_INC_STATS(sock_net(sk), LINUX_MIB_LISTENOVERFLOWS); |
1575 | out_nonewsk: |
1576 | dst_release(dst); |
1577 | out: |
1578 | tcp_listendrop(sk); |
1579 | return NULL; |
1580 | } |
1581 | |
1582 | INDIRECT_CALLABLE_DECLARE(struct dst_entry *ipv4_dst_check(struct dst_entry *, |
1583 | u32)); |
1584 | /* The socket must have it's spinlock held when we get |
1585 | * here, unless it is a TCP_LISTEN socket. |
1586 | * |
1587 | * We have a potential double-lock case here, so even when |
1588 | * doing backlog processing we use the BH locking scheme. |
1589 | * This is because we cannot sleep with the original spinlock |
1590 | * held. |
1591 | */ |
1592 | INDIRECT_CALLABLE_SCOPE |
1593 | int tcp_v6_do_rcv(struct sock *sk, struct sk_buff *skb) |
1594 | { |
1595 | struct ipv6_pinfo *np = tcp_inet6_sk(sk); |
1596 | struct sk_buff *opt_skb = NULL; |
1597 | enum skb_drop_reason reason; |
1598 | struct tcp_sock *tp; |
1599 | |
1600 | /* Imagine: socket is IPv6. IPv4 packet arrives, |
1601 | goes to IPv4 receive handler and backlogged. |
1602 | From backlog it always goes here. Kerboom... |
1603 | Fortunately, tcp_rcv_established and rcv_established |
1604 | handle them correctly, but it is not case with |
1605 | tcp_v6_hnd_req and tcp_v6_send_reset(). --ANK |
1606 | */ |
1607 | |
1608 | if (skb->protocol == htons(ETH_P_IP)) |
1609 | return tcp_v4_do_rcv(sk, skb); |
1610 | |
1611 | /* |
1612 | * socket locking is here for SMP purposes as backlog rcv |
1613 | * is currently called with bh processing disabled. |
1614 | */ |
1615 | |
1616 | /* Do Stevens' IPV6_PKTOPTIONS. |
1617 | |
1618 | Yes, guys, it is the only place in our code, where we |
1619 | may make it not affecting IPv4. |
1620 | The rest of code is protocol independent, |
1621 | and I do not like idea to uglify IPv4. |
1622 | |
1623 | Actually, all the idea behind IPV6_PKTOPTIONS |
1624 | looks not very well thought. For now we latch |
1625 | options, received in the last packet, enqueued |
1626 | by tcp. Feel free to propose better solution. |
1627 | --ANK (980728) |
1628 | */ |
1629 | if (np->rxopt.all && sk->sk_state != TCP_LISTEN) |
1630 | opt_skb = skb_clone_and_charge_r(skb, sk); |
1631 | |
1632 | if (sk->sk_state == TCP_ESTABLISHED) { /* Fast path */ |
1633 | struct dst_entry *dst; |
1634 | |
1635 | dst = rcu_dereference_protected(sk->sk_rx_dst, |
1636 | lockdep_sock_is_held(sk)); |
1637 | |
1638 | sock_rps_save_rxhash(sk, skb); |
1639 | sk_mark_napi_id(sk, skb); |
1640 | if (dst) { |
1641 | if (sk->sk_rx_dst_ifindex != skb->skb_iif || |
1642 | INDIRECT_CALL_1(dst->ops->check, ip6_dst_check, |
1643 | dst, sk->sk_rx_dst_cookie) == NULL) { |
1644 | RCU_INIT_POINTER(sk->sk_rx_dst, NULL); |
1645 | dst_release(dst); |
1646 | } |
1647 | } |
1648 | |
1649 | tcp_rcv_established(sk, skb); |
1650 | if (opt_skb) |
1651 | goto ipv6_pktoptions; |
1652 | return 0; |
1653 | } |
1654 | |
1655 | if (tcp_checksum_complete(skb)) |
1656 | goto csum_err; |
1657 | |
1658 | if (sk->sk_state == TCP_LISTEN) { |
1659 | struct sock *nsk = tcp_v6_cookie_check(sk, skb); |
1660 | |
1661 | if (nsk != sk) { |
1662 | if (nsk) { |
1663 | reason = tcp_child_process(parent: sk, child: nsk, skb); |
1664 | if (reason) |
1665 | goto reset; |
1666 | } |
1667 | return 0; |
1668 | } |
1669 | } else |
1670 | sock_rps_save_rxhash(sk, skb); |
1671 | |
1672 | reason = tcp_rcv_state_process(sk, skb); |
1673 | if (reason) |
1674 | goto reset; |
1675 | if (opt_skb) |
1676 | goto ipv6_pktoptions; |
1677 | return 0; |
1678 | |
1679 | reset: |
1680 | tcp_v6_send_reset(sk, skb, reason: sk_rst_convert_drop_reason(reason)); |
1681 | discard: |
1682 | if (opt_skb) |
1683 | __kfree_skb(skb: opt_skb); |
1684 | sk_skb_reason_drop(sk, skb, reason); |
1685 | return 0; |
1686 | csum_err: |
1687 | reason = SKB_DROP_REASON_TCP_CSUM; |
1688 | trace_tcp_bad_csum(skb); |
1689 | TCP_INC_STATS(sock_net(sk), TCP_MIB_CSUMERRORS); |
1690 | TCP_INC_STATS(sock_net(sk), TCP_MIB_INERRS); |
1691 | goto discard; |
1692 | |
1693 | |
1694 | ipv6_pktoptions: |
1695 | /* Do you ask, what is it? |
1696 | |
1697 | 1. skb was enqueued by tcp. |
1698 | 2. skb is added to tail of read queue, rather than out of order. |
1699 | 3. socket is not in passive state. |
1700 | 4. Finally, it really contains options, which user wants to receive. |
1701 | */ |
1702 | tp = tcp_sk(sk); |
1703 | if (TCP_SKB_CB(opt_skb)->end_seq == tp->rcv_nxt && |
1704 | !((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_LISTEN))) { |
1705 | if (np->rxopt.bits.rxinfo || np->rxopt.bits.rxoinfo) |
1706 | WRITE_ONCE(np->mcast_oif, tcp_v6_iif(opt_skb)); |
1707 | if (np->rxopt.bits.rxhlim || np->rxopt.bits.rxohlim) |
1708 | WRITE_ONCE(np->mcast_hops, |
1709 | ipv6_hdr(opt_skb)->hop_limit); |
1710 | if (np->rxopt.bits.rxflow || np->rxopt.bits.rxtclass) |
1711 | np->rcv_flowinfo = ip6_flowinfo(hdr: ipv6_hdr(skb: opt_skb)); |
1712 | if (inet6_test_bit(REPFLOW, sk)) |
1713 | np->flow_label = ip6_flowlabel(hdr: ipv6_hdr(skb: opt_skb)); |
1714 | if (ipv6_opt_accepted(sk, skb: opt_skb, opt: &TCP_SKB_CB(opt_skb)->header.h6)) { |
1715 | tcp_v6_restore_cb(skb: opt_skb); |
1716 | opt_skb = xchg(&np->pktoptions, opt_skb); |
1717 | } else { |
1718 | __kfree_skb(skb: opt_skb); |
1719 | opt_skb = xchg(&np->pktoptions, NULL); |
1720 | } |
1721 | } |
1722 | |
1723 | consume_skb(skb: opt_skb); |
1724 | return 0; |
1725 | } |
1726 | |
1727 | static void tcp_v6_fill_cb(struct sk_buff *skb, const struct ipv6hdr *hdr, |
1728 | const struct tcphdr *th) |
1729 | { |
1730 | /* This is tricky: we move IP6CB at its correct location into |
1731 | * TCP_SKB_CB(). It must be done after xfrm6_policy_check(), because |
1732 | * _decode_session6() uses IP6CB(). |
1733 | * barrier() makes sure compiler won't play aliasing games. |
1734 | */ |
1735 | memmove(&TCP_SKB_CB(skb)->header.h6, IP6CB(skb), |
1736 | sizeof(struct inet6_skb_parm)); |
1737 | barrier(); |
1738 | |
1739 | TCP_SKB_CB(skb)->seq = ntohl(th->seq); |
1740 | TCP_SKB_CB(skb)->end_seq = (TCP_SKB_CB(skb)->seq + th->syn + th->fin + |
1741 | skb->len - th->doff*4); |
1742 | TCP_SKB_CB(skb)->ack_seq = ntohl(th->ack_seq); |
1743 | TCP_SKB_CB(skb)->tcp_flags = tcp_flags_ntohs(th); |
1744 | TCP_SKB_CB(skb)->ip_dsfield = ipv6_get_dsfield(ipv6h: hdr); |
1745 | TCP_SKB_CB(skb)->sacked = 0; |
1746 | TCP_SKB_CB(skb)->has_rxtstamp = |
1747 | skb->tstamp || skb_hwtstamps(skb)->hwtstamp; |
1748 | } |
1749 | |
1750 | INDIRECT_CALLABLE_SCOPE int tcp_v6_rcv(struct sk_buff *skb) |
1751 | { |
1752 | struct net *net = dev_net_rcu(dev: skb->dev); |
1753 | enum skb_drop_reason drop_reason; |
1754 | enum tcp_tw_status tw_status; |
1755 | int sdif = inet6_sdif(skb); |
1756 | int dif = inet6_iif(skb); |
1757 | const struct tcphdr *th; |
1758 | const struct ipv6hdr *hdr; |
1759 | struct sock *sk = NULL; |
1760 | bool refcounted; |
1761 | int ret; |
1762 | u32 isn; |
1763 | |
1764 | drop_reason = SKB_DROP_REASON_NOT_SPECIFIED; |
1765 | if (skb->pkt_type != PACKET_HOST) |
1766 | goto discard_it; |
1767 | |
1768 | /* |
1769 | * Count it even if it's bad. |
1770 | */ |
1771 | __TCP_INC_STATS(net, TCP_MIB_INSEGS); |
1772 | |
1773 | if (!pskb_may_pull(skb, len: sizeof(struct tcphdr))) |
1774 | goto discard_it; |
1775 | |
1776 | th = (const struct tcphdr *)skb->data; |
1777 | |
1778 | if (unlikely(th->doff < sizeof(struct tcphdr) / 4)) { |
1779 | drop_reason = SKB_DROP_REASON_PKT_TOO_SMALL; |
1780 | goto bad_packet; |
1781 | } |
1782 | if (!pskb_may_pull(skb, len: th->doff*4)) |
1783 | goto discard_it; |
1784 | |
1785 | if (skb_checksum_init(skb, IPPROTO_TCP, ip6_compute_pseudo)) |
1786 | goto csum_error; |
1787 | |
1788 | th = (const struct tcphdr *)skb->data; |
1789 | hdr = ipv6_hdr(skb); |
1790 | |
1791 | lookup: |
1792 | sk = __inet6_lookup_skb(hashinfo: net->ipv4.tcp_death_row.hashinfo, skb, doff: __tcp_hdrlen(th), |
1793 | sport: th->source, dport: th->dest, iif: inet6_iif(skb), sdif, |
1794 | refcounted: &refcounted); |
1795 | if (!sk) |
1796 | goto no_tcp_socket; |
1797 | |
1798 | if (sk->sk_state == TCP_TIME_WAIT) |
1799 | goto do_time_wait; |
1800 | |
1801 | if (sk->sk_state == TCP_NEW_SYN_RECV) { |
1802 | struct request_sock *req = inet_reqsk(sk); |
1803 | bool req_stolen = false; |
1804 | struct sock *nsk; |
1805 | |
1806 | sk = req->rsk_listener; |
1807 | if (!xfrm6_policy_check(sk, dir: XFRM_POLICY_IN, skb)) |
1808 | drop_reason = SKB_DROP_REASON_XFRM_POLICY; |
1809 | else |
1810 | drop_reason = tcp_inbound_hash(sk, req, skb, |
1811 | saddr: &hdr->saddr, daddr: &hdr->daddr, |
1812 | AF_INET6, dif, sdif); |
1813 | if (drop_reason) { |
1814 | sk_drops_add(sk, skb); |
1815 | reqsk_put(req); |
1816 | goto discard_it; |
1817 | } |
1818 | if (tcp_checksum_complete(skb)) { |
1819 | reqsk_put(req); |
1820 | goto csum_error; |
1821 | } |
1822 | if (unlikely(sk->sk_state != TCP_LISTEN)) { |
1823 | nsk = reuseport_migrate_sock(sk, migrating_sk: req_to_sk(req), skb); |
1824 | if (!nsk) { |
1825 | inet_csk_reqsk_queue_drop_and_put(sk, req); |
1826 | goto lookup; |
1827 | } |
1828 | sk = nsk; |
1829 | /* reuseport_migrate_sock() has already held one sk_refcnt |
1830 | * before returning. |
1831 | */ |
1832 | } else { |
1833 | sock_hold(sk); |
1834 | } |
1835 | refcounted = true; |
1836 | nsk = NULL; |
1837 | if (!tcp_filter(sk, skb)) { |
1838 | th = (const struct tcphdr *)skb->data; |
1839 | hdr = ipv6_hdr(skb); |
1840 | tcp_v6_fill_cb(skb, hdr, th); |
1841 | nsk = tcp_check_req(sk, skb, req, fastopen: false, lost_race: &req_stolen, |
1842 | drop_reason: &drop_reason); |
1843 | } else { |
1844 | drop_reason = SKB_DROP_REASON_SOCKET_FILTER; |
1845 | } |
1846 | if (!nsk) { |
1847 | reqsk_put(req); |
1848 | if (req_stolen) { |
1849 | /* Another cpu got exclusive access to req |
1850 | * and created a full blown socket. |
1851 | * Try to feed this packet to this socket |
1852 | * instead of discarding it. |
1853 | */ |
1854 | tcp_v6_restore_cb(skb); |
1855 | sock_put(sk); |
1856 | goto lookup; |
1857 | } |
1858 | goto discard_and_relse; |
1859 | } |
1860 | nf_reset_ct(skb); |
1861 | if (nsk == sk) { |
1862 | reqsk_put(req); |
1863 | tcp_v6_restore_cb(skb); |
1864 | } else { |
1865 | drop_reason = tcp_child_process(parent: sk, child: nsk, skb); |
1866 | if (drop_reason) { |
1867 | enum sk_rst_reason rst_reason; |
1868 | |
1869 | rst_reason = sk_rst_convert_drop_reason(reason: drop_reason); |
1870 | tcp_v6_send_reset(sk: nsk, skb, reason: rst_reason); |
1871 | goto discard_and_relse; |
1872 | } |
1873 | sock_put(sk); |
1874 | return 0; |
1875 | } |
1876 | } |
1877 | |
1878 | process: |
1879 | if (static_branch_unlikely(&ip6_min_hopcount)) { |
1880 | /* min_hopcount can be changed concurrently from do_ipv6_setsockopt() */ |
1881 | if (unlikely(hdr->hop_limit < READ_ONCE(tcp_inet6_sk(sk)->min_hopcount))) { |
1882 | __NET_INC_STATS(net, LINUX_MIB_TCPMINTTLDROP); |
1883 | drop_reason = SKB_DROP_REASON_TCP_MINTTL; |
1884 | goto discard_and_relse; |
1885 | } |
1886 | } |
1887 | |
1888 | if (!xfrm6_policy_check(sk, dir: XFRM_POLICY_IN, skb)) { |
1889 | drop_reason = SKB_DROP_REASON_XFRM_POLICY; |
1890 | goto discard_and_relse; |
1891 | } |
1892 | |
1893 | drop_reason = tcp_inbound_hash(sk, NULL, skb, saddr: &hdr->saddr, daddr: &hdr->daddr, |
1894 | AF_INET6, dif, sdif); |
1895 | if (drop_reason) |
1896 | goto discard_and_relse; |
1897 | |
1898 | nf_reset_ct(skb); |
1899 | |
1900 | if (tcp_filter(sk, skb)) { |
1901 | drop_reason = SKB_DROP_REASON_SOCKET_FILTER; |
1902 | goto discard_and_relse; |
1903 | } |
1904 | th = (const struct tcphdr *)skb->data; |
1905 | hdr = ipv6_hdr(skb); |
1906 | tcp_v6_fill_cb(skb, hdr, th); |
1907 | |
1908 | skb->dev = NULL; |
1909 | |
1910 | if (sk->sk_state == TCP_LISTEN) { |
1911 | ret = tcp_v6_do_rcv(sk, skb); |
1912 | goto put_and_return; |
1913 | } |
1914 | |
1915 | sk_incoming_cpu_update(sk); |
1916 | |
1917 | bh_lock_sock_nested(sk); |
1918 | tcp_segs_in(tcp_sk(sk), skb); |
1919 | ret = 0; |
1920 | if (!sock_owned_by_user(sk)) { |
1921 | ret = tcp_v6_do_rcv(sk, skb); |
1922 | } else { |
1923 | if (tcp_add_backlog(sk, skb, reason: &drop_reason)) |
1924 | goto discard_and_relse; |
1925 | } |
1926 | bh_unlock_sock(sk); |
1927 | put_and_return: |
1928 | if (refcounted) |
1929 | sock_put(sk); |
1930 | return ret ? -1 : 0; |
1931 | |
1932 | no_tcp_socket: |
1933 | drop_reason = SKB_DROP_REASON_NO_SOCKET; |
1934 | if (!xfrm6_policy_check(NULL, dir: XFRM_POLICY_IN, skb)) |
1935 | goto discard_it; |
1936 | |
1937 | tcp_v6_fill_cb(skb, hdr, th); |
1938 | |
1939 | if (tcp_checksum_complete(skb)) { |
1940 | csum_error: |
1941 | drop_reason = SKB_DROP_REASON_TCP_CSUM; |
1942 | trace_tcp_bad_csum(skb); |
1943 | __TCP_INC_STATS(net, TCP_MIB_CSUMERRORS); |
1944 | bad_packet: |
1945 | __TCP_INC_STATS(net, TCP_MIB_INERRS); |
1946 | } else { |
1947 | tcp_v6_send_reset(NULL, skb, reason: sk_rst_convert_drop_reason(reason: drop_reason)); |
1948 | } |
1949 | |
1950 | discard_it: |
1951 | SKB_DR_OR(drop_reason, NOT_SPECIFIED); |
1952 | sk_skb_reason_drop(sk, skb, reason: drop_reason); |
1953 | return 0; |
1954 | |
1955 | discard_and_relse: |
1956 | sk_drops_add(sk, skb); |
1957 | if (refcounted) |
1958 | sock_put(sk); |
1959 | goto discard_it; |
1960 | |
1961 | do_time_wait: |
1962 | if (!xfrm6_policy_check(NULL, dir: XFRM_POLICY_IN, skb)) { |
1963 | drop_reason = SKB_DROP_REASON_XFRM_POLICY; |
1964 | inet_twsk_put(tw: inet_twsk(sk)); |
1965 | goto discard_it; |
1966 | } |
1967 | |
1968 | tcp_v6_fill_cb(skb, hdr, th); |
1969 | |
1970 | if (tcp_checksum_complete(skb)) { |
1971 | inet_twsk_put(tw: inet_twsk(sk)); |
1972 | goto csum_error; |
1973 | } |
1974 | |
1975 | tw_status = tcp_timewait_state_process(tw: inet_twsk(sk), skb, th, tw_isn: &isn, |
1976 | drop_reason: &drop_reason); |
1977 | switch (tw_status) { |
1978 | case TCP_TW_SYN: |
1979 | { |
1980 | struct sock *sk2; |
1981 | |
1982 | sk2 = inet6_lookup_listener(net, hashinfo: net->ipv4.tcp_death_row.hashinfo, |
1983 | skb, doff: __tcp_hdrlen(th), |
1984 | saddr: &ipv6_hdr(skb)->saddr, sport: th->source, |
1985 | daddr: &ipv6_hdr(skb)->daddr, |
1986 | ntohs(th->dest), |
1987 | dif: tcp_v6_iif_l3_slave(skb), |
1988 | sdif); |
1989 | if (sk2) { |
1990 | struct inet_timewait_sock *tw = inet_twsk(sk); |
1991 | inet_twsk_deschedule_put(tw); |
1992 | sk = sk2; |
1993 | tcp_v6_restore_cb(skb); |
1994 | refcounted = false; |
1995 | __this_cpu_write(tcp_tw_isn, isn); |
1996 | goto process; |
1997 | } |
1998 | } |
1999 | /* to ACK */ |
2000 | fallthrough; |
2001 | case TCP_TW_ACK: |
2002 | case TCP_TW_ACK_OOW: |
2003 | tcp_v6_timewait_ack(sk, skb, tw_status); |
2004 | break; |
2005 | case TCP_TW_RST: |
2006 | tcp_v6_send_reset(sk, skb, reason: SK_RST_REASON_TCP_TIMEWAIT_SOCKET); |
2007 | inet_twsk_deschedule_put(tw: inet_twsk(sk)); |
2008 | goto discard_it; |
2009 | case TCP_TW_SUCCESS: |
2010 | ; |
2011 | } |
2012 | goto discard_it; |
2013 | } |
2014 | |
2015 | void tcp_v6_early_demux(struct sk_buff *skb) |
2016 | { |
2017 | struct net *net = dev_net_rcu(dev: skb->dev); |
2018 | const struct ipv6hdr *hdr; |
2019 | const struct tcphdr *th; |
2020 | struct sock *sk; |
2021 | |
2022 | if (skb->pkt_type != PACKET_HOST) |
2023 | return; |
2024 | |
2025 | if (!pskb_may_pull(skb, len: skb_transport_offset(skb) + sizeof(struct tcphdr))) |
2026 | return; |
2027 | |
2028 | hdr = ipv6_hdr(skb); |
2029 | th = tcp_hdr(skb); |
2030 | |
2031 | if (th->doff < sizeof(struct tcphdr) / 4) |
2032 | return; |
2033 | |
2034 | /* Note : We use inet6_iif() here, not tcp_v6_iif() */ |
2035 | sk = __inet6_lookup_established(net, hashinfo: net->ipv4.tcp_death_row.hashinfo, |
2036 | saddr: &hdr->saddr, sport: th->source, |
2037 | daddr: &hdr->daddr, ntohs(th->dest), |
2038 | dif: inet6_iif(skb), sdif: inet6_sdif(skb)); |
2039 | if (sk) { |
2040 | skb->sk = sk; |
2041 | skb->destructor = sock_edemux; |
2042 | if (sk_fullsock(sk)) { |
2043 | struct dst_entry *dst = rcu_dereference(sk->sk_rx_dst); |
2044 | |
2045 | if (dst) |
2046 | dst = dst_check(dst, cookie: sk->sk_rx_dst_cookie); |
2047 | if (dst && |
2048 | sk->sk_rx_dst_ifindex == skb->skb_iif) |
2049 | skb_dst_set_noref(skb, dst); |
2050 | } |
2051 | } |
2052 | } |
2053 | |
2054 | static struct timewait_sock_ops tcp6_timewait_sock_ops = { |
2055 | .twsk_obj_size = sizeof(struct tcp6_timewait_sock), |
2056 | .twsk_destructor = tcp_twsk_destructor, |
2057 | }; |
2058 | |
2059 | INDIRECT_CALLABLE_SCOPE void tcp_v6_send_check(struct sock *sk, struct sk_buff *skb) |
2060 | { |
2061 | __tcp_v6_send_check(skb, saddr: &sk->sk_v6_rcv_saddr, daddr: &sk->sk_v6_daddr); |
2062 | } |
2063 | |
2064 | const struct inet_connection_sock_af_ops ipv6_specific = { |
2065 | .queue_xmit = inet6_csk_xmit, |
2066 | .send_check = tcp_v6_send_check, |
2067 | .rebuild_header = inet6_sk_rebuild_header, |
2068 | .sk_rx_dst_set = inet6_sk_rx_dst_set, |
2069 | .conn_request = tcp_v6_conn_request, |
2070 | .syn_recv_sock = tcp_v6_syn_recv_sock, |
2071 | .net_header_len = sizeof(struct ipv6hdr), |
2072 | .setsockopt = ipv6_setsockopt, |
2073 | .getsockopt = ipv6_getsockopt, |
2074 | .mtu_reduced = tcp_v6_mtu_reduced, |
2075 | }; |
2076 | |
2077 | #if defined(CONFIG_TCP_MD5SIG) || defined(CONFIG_TCP_AO) |
2078 | static const struct tcp_sock_af_ops tcp_sock_ipv6_specific = { |
2079 | #ifdef CONFIG_TCP_MD5SIG |
2080 | .md5_lookup = tcp_v6_md5_lookup, |
2081 | .calc_md5_hash = tcp_v6_md5_hash_skb, |
2082 | .md5_parse = tcp_v6_parse_md5_keys, |
2083 | #endif |
2084 | #ifdef CONFIG_TCP_AO |
2085 | .ao_lookup = tcp_v6_ao_lookup, |
2086 | .calc_ao_hash = tcp_v6_ao_hash_skb, |
2087 | .ao_parse = tcp_v6_parse_ao, |
2088 | .ao_calc_key_sk = tcp_v6_ao_calc_key_sk, |
2089 | #endif |
2090 | }; |
2091 | #endif |
2092 | |
2093 | /* |
2094 | * TCP over IPv4 via INET6 API |
2095 | */ |
2096 | static const struct inet_connection_sock_af_ops ipv6_mapped = { |
2097 | .queue_xmit = ip_queue_xmit, |
2098 | .send_check = tcp_v4_send_check, |
2099 | .rebuild_header = inet_sk_rebuild_header, |
2100 | .sk_rx_dst_set = inet_sk_rx_dst_set, |
2101 | .conn_request = tcp_v6_conn_request, |
2102 | .syn_recv_sock = tcp_v6_syn_recv_sock, |
2103 | .net_header_len = sizeof(struct iphdr), |
2104 | .setsockopt = ipv6_setsockopt, |
2105 | .getsockopt = ipv6_getsockopt, |
2106 | .mtu_reduced = tcp_v4_mtu_reduced, |
2107 | }; |
2108 | |
2109 | #if defined(CONFIG_TCP_MD5SIG) || defined(CONFIG_TCP_AO) |
2110 | static const struct tcp_sock_af_ops tcp_sock_ipv6_mapped_specific = { |
2111 | #ifdef CONFIG_TCP_MD5SIG |
2112 | .md5_lookup = tcp_v4_md5_lookup, |
2113 | .calc_md5_hash = tcp_v4_md5_hash_skb, |
2114 | .md5_parse = tcp_v6_parse_md5_keys, |
2115 | #endif |
2116 | #ifdef CONFIG_TCP_AO |
2117 | .ao_lookup = tcp_v6_ao_lookup, |
2118 | .calc_ao_hash = tcp_v4_ao_hash_skb, |
2119 | .ao_parse = tcp_v6_parse_ao, |
2120 | .ao_calc_key_sk = tcp_v4_ao_calc_key_sk, |
2121 | #endif |
2122 | }; |
2123 | #endif |
2124 | |
2125 | /* NOTE: A lot of things set to zero explicitly by call to |
2126 | * sk_alloc() so need not be done here. |
2127 | */ |
2128 | static int tcp_v6_init_sock(struct sock *sk) |
2129 | { |
2130 | struct inet_connection_sock *icsk = inet_csk(sk); |
2131 | |
2132 | tcp_init_sock(sk); |
2133 | |
2134 | icsk->icsk_af_ops = &ipv6_specific; |
2135 | |
2136 | #if defined(CONFIG_TCP_MD5SIG) || defined(CONFIG_TCP_AO) |
2137 | tcp_sk(sk)->af_specific = &tcp_sock_ipv6_specific; |
2138 | #endif |
2139 | |
2140 | return 0; |
2141 | } |
2142 | |
2143 | #ifdef CONFIG_PROC_FS |
2144 | /* Proc filesystem TCPv6 sock list dumping. */ |
2145 | static void get_openreq6(struct seq_file *seq, |
2146 | const struct request_sock *req, int i) |
2147 | { |
2148 | long ttd = req->rsk_timer.expires - jiffies; |
2149 | const struct in6_addr *src = &inet_rsk(sk: req)->ir_v6_loc_addr; |
2150 | const struct in6_addr *dest = &inet_rsk(sk: req)->ir_v6_rmt_addr; |
2151 | |
2152 | if (ttd < 0) |
2153 | ttd = 0; |
2154 | |
2155 | seq_printf(m: seq, |
2156 | fmt: "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X " |
2157 | "%02X %08X:%08X %02X:%08lX %08X %5u %8d %d %d %pK\n", |
2158 | i, |
2159 | src->s6_addr32[0], src->s6_addr32[1], |
2160 | src->s6_addr32[2], src->s6_addr32[3], |
2161 | inet_rsk(sk: req)->ir_num, |
2162 | dest->s6_addr32[0], dest->s6_addr32[1], |
2163 | dest->s6_addr32[2], dest->s6_addr32[3], |
2164 | ntohs(inet_rsk(req)->ir_rmt_port), |
2165 | TCP_SYN_RECV, |
2166 | 0, 0, /* could print option size, but that is af dependent. */ |
2167 | 1, /* timers active (only the expire timer) */ |
2168 | jiffies_to_clock_t(x: ttd), |
2169 | req->num_timeout, |
2170 | from_kuid_munged(to: seq_user_ns(seq), |
2171 | uid: sock_i_uid(sk: req->rsk_listener)), |
2172 | 0, /* non standard timer */ |
2173 | 0, /* open_requests have no inode */ |
2174 | 0, req); |
2175 | } |
2176 | |
2177 | static void get_tcp6_sock(struct seq_file *seq, struct sock *sp, int i) |
2178 | { |
2179 | const struct in6_addr *dest, *src; |
2180 | __u16 destp, srcp; |
2181 | int timer_active; |
2182 | unsigned long timer_expires; |
2183 | const struct inet_sock *inet = inet_sk(sp); |
2184 | const struct tcp_sock *tp = tcp_sk(sp); |
2185 | const struct inet_connection_sock *icsk = inet_csk(sp); |
2186 | const struct fastopen_queue *fastopenq = &icsk->icsk_accept_queue.fastopenq; |
2187 | u8 icsk_pending; |
2188 | int rx_queue; |
2189 | int state; |
2190 | |
2191 | dest = &sp->sk_v6_daddr; |
2192 | src = &sp->sk_v6_rcv_saddr; |
2193 | destp = ntohs(inet->inet_dport); |
2194 | srcp = ntohs(inet->inet_sport); |
2195 | |
2196 | icsk_pending = smp_load_acquire(&icsk->icsk_pending); |
2197 | if (icsk_pending == ICSK_TIME_RETRANS || |
2198 | icsk_pending == ICSK_TIME_REO_TIMEOUT || |
2199 | icsk_pending == ICSK_TIME_LOSS_PROBE) { |
2200 | timer_active = 1; |
2201 | timer_expires = icsk_timeout(icsk); |
2202 | } else if (icsk_pending == ICSK_TIME_PROBE0) { |
2203 | timer_active = 4; |
2204 | timer_expires = icsk_timeout(icsk); |
2205 | } else if (timer_pending(timer: &sp->sk_timer)) { |
2206 | timer_active = 2; |
2207 | timer_expires = sp->sk_timer.expires; |
2208 | } else { |
2209 | timer_active = 0; |
2210 | timer_expires = jiffies; |
2211 | } |
2212 | |
2213 | state = inet_sk_state_load(sk: sp); |
2214 | if (state == TCP_LISTEN) |
2215 | rx_queue = READ_ONCE(sp->sk_ack_backlog); |
2216 | else |
2217 | /* Because we don't lock the socket, |
2218 | * we might find a transient negative value. |
2219 | */ |
2220 | rx_queue = max_t(int, READ_ONCE(tp->rcv_nxt) - |
2221 | READ_ONCE(tp->copied_seq), 0); |
2222 | |
2223 | seq_printf(m: seq, |
2224 | fmt: "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X " |
2225 | "%02X %08X:%08X %02X:%08lX %08X %5u %8d %lu %d %pK %lu %lu %u %u %d\n", |
2226 | i, |
2227 | src->s6_addr32[0], src->s6_addr32[1], |
2228 | src->s6_addr32[2], src->s6_addr32[3], srcp, |
2229 | dest->s6_addr32[0], dest->s6_addr32[1], |
2230 | dest->s6_addr32[2], dest->s6_addr32[3], destp, |
2231 | state, |
2232 | READ_ONCE(tp->write_seq) - tp->snd_una, |
2233 | rx_queue, |
2234 | timer_active, |
2235 | jiffies_delta_to_clock_t(delta: timer_expires - jiffies), |
2236 | icsk->icsk_retransmits, |
2237 | from_kuid_munged(to: seq_user_ns(seq), uid: sock_i_uid(sk: sp)), |
2238 | icsk->icsk_probes_out, |
2239 | sock_i_ino(sk: sp), |
2240 | refcount_read(r: &sp->sk_refcnt), sp, |
2241 | jiffies_to_clock_t(x: icsk->icsk_rto), |
2242 | jiffies_to_clock_t(x: icsk->icsk_ack.ato), |
2243 | (icsk->icsk_ack.quick << 1) | inet_csk_in_pingpong_mode(sk: sp), |
2244 | tcp_snd_cwnd(tp), |
2245 | state == TCP_LISTEN ? |
2246 | fastopenq->max_qlen : |
2247 | (tcp_in_initial_slowstart(tp) ? -1 : tp->snd_ssthresh) |
2248 | ); |
2249 | } |
2250 | |
2251 | static void get_timewait6_sock(struct seq_file *seq, |
2252 | struct inet_timewait_sock *tw, int i) |
2253 | { |
2254 | long delta = tw->tw_timer.expires - jiffies; |
2255 | const struct in6_addr *dest, *src; |
2256 | __u16 destp, srcp; |
2257 | |
2258 | dest = &tw->tw_v6_daddr; |
2259 | src = &tw->tw_v6_rcv_saddr; |
2260 | destp = ntohs(tw->tw_dport); |
2261 | srcp = ntohs(tw->tw_sport); |
2262 | |
2263 | seq_printf(m: seq, |
2264 | fmt: "%4d: %08X%08X%08X%08X:%04X %08X%08X%08X%08X:%04X " |
2265 | "%02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %pK\n", |
2266 | i, |
2267 | src->s6_addr32[0], src->s6_addr32[1], |
2268 | src->s6_addr32[2], src->s6_addr32[3], srcp, |
2269 | dest->s6_addr32[0], dest->s6_addr32[1], |
2270 | dest->s6_addr32[2], dest->s6_addr32[3], destp, |
2271 | READ_ONCE(tw->tw_substate), 0, 0, |
2272 | 3, jiffies_delta_to_clock_t(delta), 0, 0, 0, 0, |
2273 | refcount_read(r: &tw->tw_refcnt), tw); |
2274 | } |
2275 | |
2276 | static int tcp6_seq_show(struct seq_file *seq, void *v) |
2277 | { |
2278 | struct tcp_iter_state *st; |
2279 | struct sock *sk = v; |
2280 | |
2281 | if (v == SEQ_START_TOKEN) { |
2282 | seq_puts(m: seq, |
2283 | s: " sl " |
2284 | "local_address " |
2285 | "remote_address " |
2286 | "st tx_queue rx_queue tr tm->when retrnsmt" |
2287 | " uid timeout inode\n"); |
2288 | goto out; |
2289 | } |
2290 | st = seq->private; |
2291 | |
2292 | if (sk->sk_state == TCP_TIME_WAIT) |
2293 | get_timewait6_sock(seq, tw: v, i: st->num); |
2294 | else if (sk->sk_state == TCP_NEW_SYN_RECV) |
2295 | get_openreq6(seq, req: v, i: st->num); |
2296 | else |
2297 | get_tcp6_sock(seq, sp: v, i: st->num); |
2298 | out: |
2299 | return 0; |
2300 | } |
2301 | |
2302 | static const struct seq_operations tcp6_seq_ops = { |
2303 | .show = tcp6_seq_show, |
2304 | .start = tcp_seq_start, |
2305 | .next = tcp_seq_next, |
2306 | .stop = tcp_seq_stop, |
2307 | }; |
2308 | |
2309 | static struct tcp_seq_afinfo tcp6_seq_afinfo = { |
2310 | .family = AF_INET6, |
2311 | }; |
2312 | |
2313 | int __net_init tcp6_proc_init(struct net *net) |
2314 | { |
2315 | if (!proc_create_net_data(name: "tcp6", mode: 0444, parent: net->proc_net, ops: &tcp6_seq_ops, |
2316 | state_size: sizeof(struct tcp_iter_state), data: &tcp6_seq_afinfo)) |
2317 | return -ENOMEM; |
2318 | return 0; |
2319 | } |
2320 | |
2321 | void tcp6_proc_exit(struct net *net) |
2322 | { |
2323 | remove_proc_entry("tcp6", net->proc_net); |
2324 | } |
2325 | #endif |
2326 | |
2327 | struct proto tcpv6_prot = { |
2328 | .name = "TCPv6", |
2329 | .owner = THIS_MODULE, |
2330 | .close = tcp_close, |
2331 | .pre_connect = tcp_v6_pre_connect, |
2332 | .connect = tcp_v6_connect, |
2333 | .disconnect = tcp_disconnect, |
2334 | .accept = inet_csk_accept, |
2335 | .ioctl = tcp_ioctl, |
2336 | .init = tcp_v6_init_sock, |
2337 | .destroy = tcp_v4_destroy_sock, |
2338 | .shutdown = tcp_shutdown, |
2339 | .setsockopt = tcp_setsockopt, |
2340 | .getsockopt = tcp_getsockopt, |
2341 | .bpf_bypass_getsockopt = tcp_bpf_bypass_getsockopt, |
2342 | .keepalive = tcp_set_keepalive, |
2343 | .recvmsg = tcp_recvmsg, |
2344 | .sendmsg = tcp_sendmsg, |
2345 | .splice_eof = tcp_splice_eof, |
2346 | .backlog_rcv = tcp_v6_do_rcv, |
2347 | .release_cb = tcp_release_cb, |
2348 | .hash = inet6_hash, |
2349 | .unhash = inet_unhash, |
2350 | .get_port = inet_csk_get_port, |
2351 | .put_port = inet_put_port, |
2352 | #ifdef CONFIG_BPF_SYSCALL |
2353 | .psock_update_sk_prot = tcp_bpf_update_proto, |
2354 | #endif |
2355 | .enter_memory_pressure = tcp_enter_memory_pressure, |
2356 | .leave_memory_pressure = tcp_leave_memory_pressure, |
2357 | .stream_memory_free = tcp_stream_memory_free, |
2358 | .sockets_allocated = &tcp_sockets_allocated, |
2359 | |
2360 | .memory_allocated = &tcp_memory_allocated, |
2361 | .per_cpu_fw_alloc = &tcp_memory_per_cpu_fw_alloc, |
2362 | |
2363 | .memory_pressure = &tcp_memory_pressure, |
2364 | .orphan_count = &tcp_orphan_count, |
2365 | .sysctl_mem = sysctl_tcp_mem, |
2366 | .sysctl_wmem_offset = offsetof(struct net, ipv4.sysctl_tcp_wmem), |
2367 | .sysctl_rmem_offset = offsetof(struct net, ipv4.sysctl_tcp_rmem), |
2368 | .max_header = MAX_TCP_HEADER, |
2369 | .obj_size = sizeof(struct tcp6_sock), |
2370 | .ipv6_pinfo_offset = offsetof(struct tcp6_sock, inet6), |
2371 | .slab_flags = SLAB_TYPESAFE_BY_RCU, |
2372 | .twsk_prot = &tcp6_timewait_sock_ops, |
2373 | .rsk_prot = &tcp6_request_sock_ops, |
2374 | .h.hashinfo = NULL, |
2375 | .no_autobind = true, |
2376 | .diag_destroy = tcp_abort, |
2377 | }; |
2378 | EXPORT_SYMBOL_GPL(tcpv6_prot); |
2379 | |
2380 | |
2381 | static struct inet_protosw tcpv6_protosw = { |
2382 | .type = SOCK_STREAM, |
2383 | .protocol = IPPROTO_TCP, |
2384 | .prot = &tcpv6_prot, |
2385 | .ops = &inet6_stream_ops, |
2386 | .flags = INET_PROTOSW_PERMANENT | |
2387 | INET_PROTOSW_ICSK, |
2388 | }; |
2389 | |
2390 | static int __net_init tcpv6_net_init(struct net *net) |
2391 | { |
2392 | int res; |
2393 | |
2394 | res = inet_ctl_sock_create(sk: &net->ipv6.tcp_sk, PF_INET6, |
2395 | type: SOCK_RAW, IPPROTO_TCP, net); |
2396 | if (!res) |
2397 | net->ipv6.tcp_sk->sk_clockid = CLOCK_MONOTONIC; |
2398 | |
2399 | return res; |
2400 | } |
2401 | |
2402 | static void __net_exit tcpv6_net_exit(struct net *net) |
2403 | { |
2404 | inet_ctl_sock_destroy(sk: net->ipv6.tcp_sk); |
2405 | } |
2406 | |
2407 | static struct pernet_operations tcpv6_net_ops = { |
2408 | .init = tcpv6_net_init, |
2409 | .exit = tcpv6_net_exit, |
2410 | }; |
2411 | |
2412 | int __init tcpv6_init(void) |
2413 | { |
2414 | int ret; |
2415 | |
2416 | net_hotdata.tcpv6_protocol = (struct inet6_protocol) { |
2417 | .handler = tcp_v6_rcv, |
2418 | .err_handler = tcp_v6_err, |
2419 | .flags = INET6_PROTO_NOPOLICY | INET6_PROTO_FINAL, |
2420 | }; |
2421 | ret = inet6_add_protocol(prot: &net_hotdata.tcpv6_protocol, IPPROTO_TCP); |
2422 | if (ret) |
2423 | goto out; |
2424 | |
2425 | /* register inet6 protocol */ |
2426 | ret = inet6_register_protosw(p: &tcpv6_protosw); |
2427 | if (ret) |
2428 | goto out_tcpv6_protocol; |
2429 | |
2430 | ret = register_pernet_subsys(&tcpv6_net_ops); |
2431 | if (ret) |
2432 | goto out_tcpv6_protosw; |
2433 | |
2434 | ret = mptcpv6_init(); |
2435 | if (ret) |
2436 | goto out_tcpv6_pernet_subsys; |
2437 | |
2438 | out: |
2439 | return ret; |
2440 | |
2441 | out_tcpv6_pernet_subsys: |
2442 | unregister_pernet_subsys(&tcpv6_net_ops); |
2443 | out_tcpv6_protosw: |
2444 | inet6_unregister_protosw(p: &tcpv6_protosw); |
2445 | out_tcpv6_protocol: |
2446 | inet6_del_protocol(prot: &net_hotdata.tcpv6_protocol, IPPROTO_TCP); |
2447 | goto out; |
2448 | } |
2449 | |
2450 | void tcpv6_exit(void) |
2451 | { |
2452 | unregister_pernet_subsys(&tcpv6_net_ops); |
2453 | inet6_unregister_protosw(p: &tcpv6_protosw); |
2454 | inet6_del_protocol(prot: &net_hotdata.tcpv6_protocol, IPPROTO_TCP); |
2455 | } |
2456 |
Definitions
- ipv6_mapped
- ipv6_specific
- tcp_sock_ipv6_specific
- tcp_sock_ipv6_mapped_specific
- inet6_sk_rx_dst_set
- tcp_v6_init_seq
- tcp_v6_init_ts_off
- tcp_v6_pre_connect
- tcp_v6_connect
- tcp_v6_mtu_reduced
- tcp_v6_err
- tcp_v6_send_synack
- tcp_v6_reqsk_destructor
- tcp_v6_md5_do_lookup
- tcp_v6_md5_lookup
- tcp_v6_parse_md5_keys
- tcp_v6_md5_hash_headers
- tcp_v6_md5_hash_hdr
- tcp_v6_md5_hash_skb
- tcp_v6_init_req
- tcp_v6_route_req
- tcp6_request_sock_ops
- tcp_request_sock_ipv6_ops
- tcp_v6_send_response
- tcp_v6_send_reset
- tcp_v6_send_ack
- tcp_v6_timewait_ack
- tcp_v6_reqsk_send_ack
- tcp_v6_cookie_check
- tcp_v6_get_syncookie
- tcp_v6_conn_request
- tcp_v6_restore_cb
- tcp_v6_syn_recv_sock
- tcp_v6_do_rcv
- tcp_v6_fill_cb
- tcp_v6_rcv
- tcp_v6_early_demux
- tcp6_timewait_sock_ops
- tcp_v6_send_check
- ipv6_specific
- tcp_sock_ipv6_specific
- ipv6_mapped
- tcp_sock_ipv6_mapped_specific
- tcp_v6_init_sock
- get_openreq6
- get_tcp6_sock
- get_timewait6_sock
- tcp6_seq_show
- tcp6_seq_ops
- tcp6_seq_afinfo
- tcp6_proc_init
- tcp6_proc_exit
- tcpv6_prot
- tcpv6_protosw
- tcpv6_net_init
- tcpv6_net_exit
- tcpv6_net_ops
- tcpv6_init
Improve your Profiling and Debugging skills
Find out more