1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3 * INET An implementation of the TCP/IP protocol suite for the LINUX
4 * operating system. INET is implemented using the BSD Socket
5 * interface as the means of communication with the user level.
6 *
7 * Implementation of the Transmission Control Protocol(TCP).
8 *
9 * Authors: Ross Biro
10 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
11 * Mark Evans, <evansmp@uhura.aston.ac.uk>
12 * Corey Minyard <wf-rch!minyard@relay.EU.net>
13 * Florian La Roche, <flla@stud.uni-sb.de>
14 * Charles Hedrick, <hedrick@klinzhai.rutgers.edu>
15 * Linus Torvalds, <torvalds@cs.helsinki.fi>
16 * Alan Cox, <gw4pts@gw4pts.ampr.org>
17 * Matthew Dillon, <dillon@apollo.west.oic.com>
18 * Arnt Gulbrandsen, <agulbra@nvg.unit.no>
19 * Jorge Cwik, <jorge@laser.satlink.net>
20 */
21
22 #include <net/tcp.h>
23 #include <net/xfrm.h>
24 #include <net/busy_poll.h>
25 #include <net/rstreason.h>
26
27 static bool tcp_in_window(u32 seq, u32 end_seq, u32 s_win, u32 e_win)
28 {
29 if (seq == s_win)
30 return true;
31 if (after(end_seq, s_win) && before(seq, e_win))
32 return true;
33 return seq == e_win && seq == end_seq;
34 }
35
36 static enum tcp_tw_status
37 tcp_timewait_check_oow_rate_limit(struct inet_timewait_sock *tw,
38 const struct sk_buff *skb, int mib_idx)
39 {
40 struct tcp_timewait_sock *tcptw = tcp_twsk((struct sock *)tw);
41
42 if (!tcp_oow_rate_limited(twsk_net(tw), skb, mib_idx,
43 &tcptw->tw_last_oow_ack_time)) {
44 /* Send ACK. Note, we do not put the bucket,
45 * it will be released by caller.
46 */
47 return TCP_TW_ACK;
48 }
49
50 /* We are rate-limiting, so just release the tw sock and drop skb. */
51 inet_twsk_put(tw);
52 return TCP_TW_SUCCESS;
53 }
54
55 static void twsk_rcv_nxt_update(struct tcp_timewait_sock *tcptw, u32 seq)
56 {
57 #ifdef CONFIG_TCP_AO
58 struct tcp_ao_info *ao;
59
60 ao = rcu_dereference(tcptw->ao_info);
61 if (unlikely(ao && seq < tcptw->tw_rcv_nxt))
62 WRITE_ONCE(ao->rcv_sne, ao->rcv_sne + 1);
63 #endif
64 tcptw->tw_rcv_nxt = seq;
65 }
66
67 /*
68 * * Main purpose of TIME-WAIT state is to close connection gracefully,
69 * when one of ends sits in LAST-ACK or CLOSING retransmitting FIN
70 * (and, probably, tail of data) and one or more our ACKs are lost.
71 * * What is TIME-WAIT timeout? It is associated with maximal packet
72 * lifetime in the internet, which results in wrong conclusion, that
73 * it is set to catch "old duplicate segments" wandering out of their path.
74 * It is not quite correct. This timeout is calculated so that it exceeds
75 * maximal retransmission timeout enough to allow to lose one (or more)
76 * segments sent by peer and our ACKs. This time may be calculated from RTO.
77 * * When TIME-WAIT socket receives RST, it means that another end
78 * finally closed and we are allowed to kill TIME-WAIT too.
79 * * Second purpose of TIME-WAIT is catching old duplicate segments.
80 * Well, certainly it is pure paranoia, but if we load TIME-WAIT
81 * with this semantics, we MUST NOT kill TIME-WAIT state with RSTs.
82 * * If we invented some more clever way to catch duplicates
83 * (f.e. based on PAWS), we could truncate TIME-WAIT to several RTOs.
84 *
85 * The algorithm below is based on FORMAL INTERPRETATION of RFCs.
86 * When you compare it to RFCs, please, read section SEGMENT ARRIVES
87 * from the very beginning.
88 *
89 * NOTE. With recycling (and later with fin-wait-2) TW bucket
90 * is _not_ stateless. It means, that strictly speaking we must
91 * spinlock it. I do not want! Well, probability of misbehaviour
92 * is ridiculously low and, seems, we could use some mb() tricks
93 * to avoid misread sequence numbers, states etc. --ANK
94 *
95 * We don't need to initialize tmp_out.sack_ok as we don't use the results
96 */
97 enum tcp_tw_status
98 tcp_timewait_state_process(struct inet_timewait_sock *tw, struct sk_buff *skb,
99 const struct tcphdr *th, u32 *tw_isn)
100 {
101 struct tcp_options_received tmp_opt;
102 struct tcp_timewait_sock *tcptw = tcp_twsk((struct sock *)tw);
103 bool paws_reject = false;
104 int ts_recent_stamp;
105
106 tmp_opt.saw_tstamp = 0;
107 ts_recent_stamp = READ_ONCE(tcptw->tw_ts_recent_stamp);
108 if (th->doff > (sizeof(*th) >> 2) && ts_recent_stamp) {
109 tcp_parse_options(twsk_net(tw), skb, &tmp_opt, 0, NULL);
110
111 if (tmp_opt.saw_tstamp) {
112 if (tmp_opt.rcv_tsecr)
113 tmp_opt.rcv_tsecr -= tcptw->tw_ts_offset;
114 tmp_opt.ts_recent = READ_ONCE(tcptw->tw_ts_recent);
115 tmp_opt.ts_recent_stamp = ts_recent_stamp;
116 paws_reject = tcp_paws_reject(&tmp_opt, th->rst);
117 }
118 }
119
120 if (tw->tw_substate == TCP_FIN_WAIT2) {
121 /* Just repeat all the checks of tcp_rcv_state_process() */
122
123 /* Out of window, send ACK */
124 if (paws_reject ||
125 !tcp_in_window(TCP_SKB_CB(skb)->seq, TCP_SKB_CB(skb)->end_seq,
126 tcptw->tw_rcv_nxt,
127 tcptw->tw_rcv_nxt + tcptw->tw_rcv_wnd))
128 return tcp_timewait_check_oow_rate_limit(
129 tw, skb, LINUX_MIB_TCPACKSKIPPEDFINWAIT2);
130
131 if (th->rst)
132 goto kill;
133
134 if (th->syn && !before(TCP_SKB_CB(skb)->seq, tcptw->tw_rcv_nxt))
135 return TCP_TW_RST;
136
137 /* Dup ACK? */
138 if (!th->ack ||
139 !after(TCP_SKB_CB(skb)->end_seq, tcptw->tw_rcv_nxt) ||
140 TCP_SKB_CB(skb)->end_seq == TCP_SKB_CB(skb)->seq) {
141 inet_twsk_put(tw);
142 return TCP_TW_SUCCESS;
143 }
144
145 /* New data or FIN. If new data arrive after half-duplex close,
146 * reset.
147 */
148 if (!th->fin ||
149 TCP_SKB_CB(skb)->end_seq != tcptw->tw_rcv_nxt + 1)
150 return TCP_TW_RST;
151
152 /* FIN arrived, enter true time-wait state. */
153 tw->tw_substate = TCP_TIME_WAIT;
154 twsk_rcv_nxt_update(tcptw, TCP_SKB_CB(skb)->end_seq);
155
156 if (tmp_opt.saw_tstamp) {
157 WRITE_ONCE(tcptw->tw_ts_recent_stamp,
158 ktime_get_seconds());
159 WRITE_ONCE(tcptw->tw_ts_recent,
160 tmp_opt.rcv_tsval);
161 }
162
163 inet_twsk_reschedule(tw, TCP_TIMEWAIT_LEN);
164 return TCP_TW_ACK;
165 }
166
167 /*
168 * Now real TIME-WAIT state.
169 *
170 * RFC 1122:
171 * "When a connection is [...] on TIME-WAIT state [...]
172 * [a TCP] MAY accept a new SYN from the remote TCP to
173 * reopen the connection directly, if it:
174 *
175 * (1) assigns its initial sequence number for the new
176 * connection to be larger than the largest sequence
177 * number it used on the previous connection incarnation,
178 * and
179 *
180 * (2) returns to TIME-WAIT state if the SYN turns out
181 * to be an old duplicate".
182 */
183
184 if (!paws_reject &&
185 (TCP_SKB_CB(skb)->seq == tcptw->tw_rcv_nxt &&
186 (TCP_SKB_CB(skb)->seq == TCP_SKB_CB(skb)->end_seq || th->rst))) {
187 /* In window segment, it may be only reset or bare ack. */
188
189 if (th->rst) {
190 /* This is TIME_WAIT assassination, in two flavors.
191 * Oh well... nobody has a sufficient solution to this
192 * protocol bug yet.
193 */
194 if (!READ_ONCE(twsk_net(tw)->ipv4.sysctl_tcp_rfc1337)) {
195 kill:
196 inet_twsk_deschedule_put(tw);
197 return TCP_TW_SUCCESS;
198 }
199 } else {
200 inet_twsk_reschedule(tw, TCP_TIMEWAIT_LEN);
201 }
202
203 if (tmp_opt.saw_tstamp) {
204 WRITE_ONCE(tcptw->tw_ts_recent,
205 tmp_opt.rcv_tsval);
206 WRITE_ONCE(tcptw->tw_ts_recent_stamp,
207 ktime_get_seconds());
208 }
209
210 inet_twsk_put(tw);
211 return TCP_TW_SUCCESS;
212 }
213
214 /* Out of window segment.
215
216 All the segments are ACKed immediately.
217
218 The only exception is new SYN. We accept it, if it is
219 not old duplicate and we are not in danger to be killed
220 by delayed old duplicates. RFC check is that it has
221 newer sequence number works at rates <40Mbit/sec.
222 However, if paws works, it is reliable AND even more,
223 we even may relax silly seq space cutoff.
224
225 RED-PEN: we violate main RFC requirement, if this SYN will appear
226 old duplicate (i.e. we receive RST in reply to SYN-ACK),
227 we must return socket to time-wait state. It is not good,
228 but not fatal yet.
229 */
230
231 if (th->syn && !th->rst && !th->ack && !paws_reject &&
232 (after(TCP_SKB_CB(skb)->seq, tcptw->tw_rcv_nxt) ||
233 (tmp_opt.saw_tstamp &&
234 (s32)(READ_ONCE(tcptw->tw_ts_recent) - tmp_opt.rcv_tsval) < 0))) {
235 u32 isn = tcptw->tw_snd_nxt + 65535 + 2;
236 if (isn == 0)
237 isn++;
238 *tw_isn = isn;
239 return TCP_TW_SYN;
240 }
241
242 if (paws_reject)
243 __NET_INC_STATS(twsk_net(tw), LINUX_MIB_PAWSESTABREJECTED);
244
245 if (!th->rst) {
246 /* In this case we must reset the TIMEWAIT timer.
247 *
248 * If it is ACKless SYN it may be both old duplicate
249 * and new good SYN with random sequence number <rcv_nxt.
250 * Do not reschedule in the last case.
251 */
252 if (paws_reject || th->ack)
253 inet_twsk_reschedule(tw, TCP_TIMEWAIT_LEN);
254
255 return tcp_timewait_check_oow_rate_limit(
256 tw, skb, LINUX_MIB_TCPACKSKIPPEDTIMEWAIT);
257 }
258 inet_twsk_put(tw);
259 return TCP_TW_SUCCESS;
260 }
261 EXPORT_SYMBOL(tcp_timewait_state_process);
262
263 static void tcp_time_wait_init(struct sock *sk, struct tcp_timewait_sock *tcptw)
264 {
265 #ifdef CONFIG_TCP_MD5SIG
266 const struct tcp_sock *tp = tcp_sk(sk);
267 struct tcp_md5sig_key *key;
268
269 /*
270 * The timewait bucket does not have the key DB from the
271 * sock structure. We just make a quick copy of the
272 * md5 key being used (if indeed we are using one)
273 * so the timewait ack generating code has the key.
274 */
275 tcptw->tw_md5_key = NULL;
276 if (!static_branch_unlikely(&tcp_md5_needed.key))
277 return;
278
279 key = tp->af_specific->md5_lookup(sk, sk);
280 if (key) {
281 tcptw->tw_md5_key = kmemdup(key, sizeof(*key), GFP_ATOMIC);
282 if (!tcptw->tw_md5_key)
283 return;
284 if (!static_key_fast_inc_not_disabled(&tcp_md5_needed.key.key))
285 goto out_free;
286 tcp_md5_add_sigpool();
287 }
288 return;
289 out_free:
290 WARN_ON_ONCE(1);
291 kfree(tcptw->tw_md5_key);
292 tcptw->tw_md5_key = NULL;
293 #endif
294 }
295
296 /*
297 * Move a socket to time-wait or dead fin-wait-2 state.
298 */
299 void tcp_time_wait(struct sock *sk, int state, int timeo)
300 {
301 const struct inet_connection_sock *icsk = inet_csk(sk);
302 struct tcp_sock *tp = tcp_sk(sk);
303 struct net *net = sock_net(sk);
304 struct inet_timewait_sock *tw;
305
306 tw = inet_twsk_alloc(sk, &net->ipv4.tcp_death_row, state);
307
308 if (tw) {
309 struct tcp_timewait_sock *tcptw = tcp_twsk((struct sock *)tw);
310 const int rto = (icsk->icsk_rto << 2) - (icsk->icsk_rto >> 1);
311
312 tw->tw_transparent = inet_test_bit(TRANSPARENT, sk);
313 tw->tw_mark = sk->sk_mark;
314 tw->tw_priority = READ_ONCE(sk->sk_priority);
315 tw->tw_rcv_wscale = tp->rx_opt.rcv_wscale;
316 tcptw->tw_rcv_nxt = tp->rcv_nxt;
317 tcptw->tw_snd_nxt = tp->snd_nxt;
318 tcptw->tw_rcv_wnd = tcp_receive_window(tp);
319 tcptw->tw_ts_recent = tp->rx_opt.ts_recent;
320 tcptw->tw_ts_recent_stamp = tp->rx_opt.ts_recent_stamp;
321 tcptw->tw_ts_offset = tp->tsoffset;
322 tw->tw_usec_ts = tp->tcp_usec_ts;
323 tcptw->tw_last_oow_ack_time = 0;
324 tcptw->tw_tx_delay = tp->tcp_tx_delay;
325 tw->tw_txhash = sk->sk_txhash;
326 #if IS_ENABLED(CONFIG_IPV6)
327 if (tw->tw_family == PF_INET6) {
328 struct ipv6_pinfo *np = inet6_sk(sk);
329
330 tw->tw_v6_daddr = sk->sk_v6_daddr;
331 tw->tw_v6_rcv_saddr = sk->sk_v6_rcv_saddr;
332 tw->tw_tclass = np->tclass;
333 tw->tw_flowlabel = be32_to_cpu(np->flow_label & IPV6_FLOWLABEL_MASK);
334 tw->tw_ipv6only = sk->sk_ipv6only;
335 }
336 #endif
337
338 tcp_time_wait_init(sk, tcptw);
339 tcp_ao_time_wait(tcptw, tp);
340
341 /* Get the TIME_WAIT timeout firing. */
342 if (timeo < rto)
343 timeo = rto;
344
345 if (state == TCP_TIME_WAIT)
346 timeo = TCP_TIMEWAIT_LEN;
347
348 /* Linkage updates.
349 * Note that access to tw after this point is illegal.
350 */
351 inet_twsk_hashdance_schedule(tw, sk, net->ipv4.tcp_death_row.hashinfo, timeo);
352 } else {
353 /* Sorry, if we're out of memory, just CLOSE this
354 * socket up. We've got bigger problems than
355 * non-graceful socket closings.
356 */
357 NET_INC_STATS(net, LINUX_MIB_TCPTIMEWAITOVERFLOW);
358 }
359
360 tcp_update_metrics(sk);
361 tcp_done(sk);
362 }
363 EXPORT_SYMBOL(tcp_time_wait);
364
365 #ifdef CONFIG_TCP_MD5SIG
366 static void tcp_md5_twsk_free_rcu(struct rcu_head *head)
367 {
368 struct tcp_md5sig_key *key;
369
370 key = container_of(head, struct tcp_md5sig_key, rcu);
371 kfree(key);
372 static_branch_slow_dec_deferred(&tcp_md5_needed);
373 tcp_md5_release_sigpool();
374 }
375 #endif
376
377 void tcp_twsk_destructor(struct sock *sk)
378 {
379 #ifdef CONFIG_TCP_MD5SIG
380 if (static_branch_unlikely(&tcp_md5_needed.key)) {
381 struct tcp_timewait_sock *twsk = tcp_twsk(sk);
382
383 if (twsk->tw_md5_key)
384 call_rcu(&twsk->tw_md5_key->rcu, tcp_md5_twsk_free_rcu);
385 }
386 #endif
387 tcp_ao_destroy_sock(sk, true);
388 }
389 EXPORT_SYMBOL_GPL(tcp_twsk_destructor);
390
391 void tcp_twsk_purge(struct list_head *net_exit_list)
392 {
393 bool purged_once = false;
394 struct net *net;
395
396 list_for_each_entry(net, net_exit_list, exit_list) {
397 if (net->ipv4.tcp_death_row.hashinfo->pernet) {
398 /* Even if tw_refcount == 1, we must clean up kernel reqsk */
399 inet_twsk_purge(net->ipv4.tcp_death_row.hashinfo);
400 } else if (!purged_once) {
401 inet_twsk_purge(&tcp_hashinfo);
402 purged_once = true;
403 }
404 }
405 }
406
407 /* Warning : This function is called without sk_listener being locked.
408 * Be sure to read socket fields once, as their value could change under us.
409 */
410 void tcp_openreq_init_rwin(struct request_sock *req,
411 const struct sock *sk_listener,
412 const struct dst_entry *dst)
413 {
414 struct inet_request_sock *ireq = inet_rsk(req);
415 const struct tcp_sock *tp = tcp_sk(sk_listener);
416 int full_space = tcp_full_space(sk_listener);
417 u32 window_clamp;
418 __u8 rcv_wscale;
419 u32 rcv_wnd;
420 int mss;
421
422 mss = tcp_mss_clamp(tp, dst_metric_advmss(dst));
423 window_clamp = READ_ONCE(tp->window_clamp);
424 /* Set this up on the first call only */
425 req->rsk_window_clamp = window_clamp ? : dst_metric(dst, RTAX_WINDOW);
426
427 /* limit the window selection if the user enforce a smaller rx buffer */
428 if (sk_listener->sk_userlocks & SOCK_RCVBUF_LOCK &&
429 (req->rsk_window_clamp > full_space || req->rsk_window_clamp == 0))
430 req->rsk_window_clamp = full_space;
431
432 rcv_wnd = tcp_rwnd_init_bpf((struct sock *)req);
433 if (rcv_wnd == 0)
434 rcv_wnd = dst_metric(dst, RTAX_INITRWND);
435 else if (full_space < rcv_wnd * mss)
436 full_space = rcv_wnd * mss;
437
438 /* tcp_full_space because it is guaranteed to be the first packet */
439 tcp_select_initial_window(sk_listener, full_space,
440 mss - (ireq->tstamp_ok ? TCPOLEN_TSTAMP_ALIGNED : 0),
441 &req->rsk_rcv_wnd,
442 &req->rsk_window_clamp,
443 ireq->wscale_ok,
444 &rcv_wscale,
445 rcv_wnd);
446 ireq->rcv_wscale = rcv_wscale;
447 }
448 EXPORT_SYMBOL(tcp_openreq_init_rwin);
449
450 static void tcp_ecn_openreq_child(struct tcp_sock *tp,
451 const struct request_sock *req)
452 {
453 tp->ecn_flags = inet_rsk(req)->ecn_ok ? TCP_ECN_OK : 0;
454 }
455
456 void tcp_ca_openreq_child(struct sock *sk, const struct dst_entry *dst)
457 {
458 struct inet_connection_sock *icsk = inet_csk(sk);
459 u32 ca_key = dst_metric(dst, RTAX_CC_ALGO);
460 bool ca_got_dst = false;
461
462 if (ca_key != TCP_CA_UNSPEC) {
463 const struct tcp_congestion_ops *ca;
464
465 rcu_read_lock();
466 ca = tcp_ca_find_key(ca_key);
467 if (likely(ca && bpf_try_module_get(ca, ca->owner))) {
468 icsk->icsk_ca_dst_locked = tcp_ca_dst_locked(dst);
469 icsk->icsk_ca_ops = ca;
470 ca_got_dst = true;
471 }
472 rcu_read_unlock();
473 }
474
475 /* If no valid choice made yet, assign current system default ca. */
476 if (!ca_got_dst &&
477 (!icsk->icsk_ca_setsockopt ||
478 !bpf_try_module_get(icsk->icsk_ca_ops, icsk->icsk_ca_ops->owner)))
479 tcp_assign_congestion_control(sk);
480
481 tcp_set_ca_state(sk, TCP_CA_Open);
482 }
483 EXPORT_SYMBOL_GPL(tcp_ca_openreq_child);
484
485 static void smc_check_reset_syn_req(const struct tcp_sock *oldtp,
486 struct request_sock *req,
487 struct tcp_sock *newtp)
488 {
489 #if IS_ENABLED(CONFIG_SMC)
490 struct inet_request_sock *ireq;
491
492 if (static_branch_unlikely(&tcp_have_smc)) {
493 ireq = inet_rsk(req);
494 if (oldtp->syn_smc && !ireq->smc_ok)
495 newtp->syn_smc = 0;
496 }
497 #endif
498 }
499
500 /* This is not only more efficient than what we used to do, it eliminates
501 * a lot of code duplication between IPv4/IPv6 SYN recv processing. -DaveM
502 *
503 * Actually, we could lots of memory writes here. tp of listening
504 * socket contains all necessary default parameters.
505 */
506 struct sock *tcp_create_openreq_child(const struct sock *sk,
507 struct request_sock *req,
508 struct sk_buff *skb)
509 {
510 struct sock *newsk = inet_csk_clone_lock(sk, req, GFP_ATOMIC);
511 const struct inet_request_sock *ireq = inet_rsk(req);
512 struct tcp_request_sock *treq = tcp_rsk(req);
513 struct inet_connection_sock *newicsk;
514 const struct tcp_sock *oldtp;
515 struct tcp_sock *newtp;
516 u32 seq;
517
518 if (!newsk)
519 return NULL;
520
521 newicsk = inet_csk(newsk);
522 newtp = tcp_sk(newsk);
523 oldtp = tcp_sk(sk);
524
525 smc_check_reset_syn_req(oldtp, req, newtp);
526
527 /* Now setup tcp_sock */
528 newtp->pred_flags = 0;
529
530 seq = treq->rcv_isn + 1;
531 newtp->rcv_wup = seq;
532 WRITE_ONCE(newtp->copied_seq, seq);
533 WRITE_ONCE(newtp->rcv_nxt, seq);
534 newtp->segs_in = 1;
535
536 seq = treq->snt_isn + 1;
537 newtp->snd_sml = newtp->snd_una = seq;
538 WRITE_ONCE(newtp->snd_nxt, seq);
539 newtp->snd_up = seq;
540
541 INIT_LIST_HEAD(&newtp->tsq_node);
542 INIT_LIST_HEAD(&newtp->tsorted_sent_queue);
543
544 tcp_init_wl(newtp, treq->rcv_isn);
545
546 minmax_reset(&newtp->rtt_min, tcp_jiffies32, ~0U);
547 newicsk->icsk_ack.lrcvtime = tcp_jiffies32;
548
549 newtp->lsndtime = tcp_jiffies32;
550 newsk->sk_txhash = READ_ONCE(treq->txhash);
551 newtp->total_retrans = req->num_retrans;
552
553 tcp_init_xmit_timers(newsk);
554 WRITE_ONCE(newtp->write_seq, newtp->pushed_seq = treq->snt_isn + 1);
555
556 if (sock_flag(newsk, SOCK_KEEPOPEN))
557 inet_csk_reset_keepalive_timer(newsk,
558 keepalive_time_when(newtp));
559
560 newtp->rx_opt.tstamp_ok = ireq->tstamp_ok;
561 newtp->rx_opt.sack_ok = ireq->sack_ok;
562 newtp->window_clamp = req->rsk_window_clamp;
563 newtp->rcv_ssthresh = req->rsk_rcv_wnd;
564 newtp->rcv_wnd = req->rsk_rcv_wnd;
565 newtp->rx_opt.wscale_ok = ireq->wscale_ok;
566 if (newtp->rx_opt.wscale_ok) {
567 newtp->rx_opt.snd_wscale = ireq->snd_wscale;
568 newtp->rx_opt.rcv_wscale = ireq->rcv_wscale;
569 } else {
570 newtp->rx_opt.snd_wscale = newtp->rx_opt.rcv_wscale = 0;
571 newtp->window_clamp = min(newtp->window_clamp, 65535U);
572 }
573 newtp->snd_wnd = ntohs(tcp_hdr(skb)->window) << newtp->rx_opt.snd_wscale;
574 newtp->max_window = newtp->snd_wnd;
575
576 if (newtp->rx_opt.tstamp_ok) {
577 newtp->tcp_usec_ts = treq->req_usec_ts;
578 newtp->rx_opt.ts_recent = READ_ONCE(req->ts_recent);
579 newtp->rx_opt.ts_recent_stamp = ktime_get_seconds();
580 newtp->tcp_header_len = sizeof(struct tcphdr) + TCPOLEN_TSTAMP_ALIGNED;
581 } else {
582 newtp->tcp_usec_ts = 0;
583 newtp->rx_opt.ts_recent_stamp = 0;
584 newtp->tcp_header_len = sizeof(struct tcphdr);
585 }
586 if (req->num_timeout) {
587 newtp->total_rto = req->num_timeout;
588 newtp->undo_marker = treq->snt_isn;
589 if (newtp->tcp_usec_ts) {
590 newtp->retrans_stamp = treq->snt_synack;
591 newtp->total_rto_time = (u32)(tcp_clock_us() -
592 newtp->retrans_stamp) / USEC_PER_MSEC;
593 } else {
594 newtp->retrans_stamp = div_u64(treq->snt_synack,
595 USEC_PER_SEC / TCP_TS_HZ);
596 newtp->total_rto_time = tcp_clock_ms() -
597 newtp->retrans_stamp;
598 }
599 newtp->total_rto_recoveries = 1;
600 }
601 newtp->tsoffset = treq->ts_off;
602 #ifdef CONFIG_TCP_MD5SIG
603 newtp->md5sig_info = NULL; /*XXX*/
604 #endif
605 #ifdef CONFIG_TCP_AO
606 newtp->ao_info = NULL;
607
608 if (tcp_rsk_used_ao(req)) {
609 struct tcp_ao_key *ao_key;
610
611 ao_key = treq->af_specific->ao_lookup(sk, req, tcp_rsk(req)->ao_keyid, -1);
612 if (ao_key)
613 newtp->tcp_header_len += tcp_ao_len_aligned(ao_key);
614 }
615 #endif
616 if (skb->len >= TCP_MSS_DEFAULT + newtp->tcp_header_len)
617 newicsk->icsk_ack.last_seg_size = skb->len - newtp->tcp_header_len;
618 newtp->rx_opt.mss_clamp = req->mss;
619 tcp_ecn_openreq_child(newtp, req);
620 newtp->fastopen_req = NULL;
621 RCU_INIT_POINTER(newtp->fastopen_rsk, NULL);
622
623 newtp->bpf_chg_cc_inprogress = 0;
624 tcp_bpf_clone(sk, newsk);
625
626 __TCP_INC_STATS(sock_net(sk), TCP_MIB_PASSIVEOPENS);
627
628 return newsk;
629 }
630 EXPORT_SYMBOL(tcp_create_openreq_child);
631
632 /*
633 * Process an incoming packet for SYN_RECV sockets represented as a
634 * request_sock. Normally sk is the listener socket but for TFO it
635 * points to the child socket.
636 *
637 * XXX (TFO) - The current impl contains a special check for ack
638 * validation and inside tcp_v4_reqsk_send_ack(). Can we do better?
639 *
640 * We don't need to initialize tmp_opt.sack_ok as we don't use the results
641 *
642 * Note: If @fastopen is true, this can be called from process context.
643 * Otherwise, this is from BH context.
644 */
645
646 struct sock *tcp_check_req(struct sock *sk, struct sk_buff *skb,
647 struct request_sock *req,
648 bool fastopen, bool *req_stolen)
649 {
650 struct tcp_options_received tmp_opt;
651 struct sock *child;
652 const struct tcphdr *th = tcp_hdr(skb);
653 __be32 flg = tcp_flag_word(th) & (TCP_FLAG_RST|TCP_FLAG_SYN|TCP_FLAG_ACK);
654 bool paws_reject = false;
655 bool own_req;
656
657 tmp_opt.saw_tstamp = 0;
658 if (th->doff > (sizeof(struct tcphdr)>>2)) {
659 tcp_parse_options(sock_net(sk), skb, &tmp_opt, 0, NULL);
660
661 if (tmp_opt.saw_tstamp) {
662 tmp_opt.ts_recent = READ_ONCE(req->ts_recent);
663 if (tmp_opt.rcv_tsecr)
664 tmp_opt.rcv_tsecr -= tcp_rsk(req)->ts_off;
665 /* We do not store true stamp, but it is not required,
666 * it can be estimated (approximately)
667 * from another data.
668 */
669 tmp_opt.ts_recent_stamp = ktime_get_seconds() - reqsk_timeout(req, TCP_RTO_MAX) / HZ;
670 paws_reject = tcp_paws_reject(&tmp_opt, th->rst);
671 }
672 }
673
674 /* Check for pure retransmitted SYN. */
675 if (TCP_SKB_CB(skb)->seq == tcp_rsk(req)->rcv_isn &&
676 flg == TCP_FLAG_SYN &&
677 !paws_reject) {
678 /*
679 * RFC793 draws (Incorrectly! It was fixed in RFC1122)
680 * this case on figure 6 and figure 8, but formal
681 * protocol description says NOTHING.
682 * To be more exact, it says that we should send ACK,
683 * because this segment (at least, if it has no data)
684 * is out of window.
685 *
686 * CONCLUSION: RFC793 (even with RFC1122) DOES NOT
687 * describe SYN-RECV state. All the description
688 * is wrong, we cannot believe to it and should
689 * rely only on common sense and implementation
690 * experience.
691 *
692 * Enforce "SYN-ACK" according to figure 8, figure 6
693 * of RFC793, fixed by RFC1122.
694 *
695 * Note that even if there is new data in the SYN packet
696 * they will be thrown away too.
697 *
698 * Reset timer after retransmitting SYNACK, similar to
699 * the idea of fast retransmit in recovery.
700 */
701 if (!tcp_oow_rate_limited(sock_net(sk), skb,
702 LINUX_MIB_TCPACKSKIPPEDSYNRECV,
703 &tcp_rsk(req)->last_oow_ack_time) &&
704
705 !inet_rtx_syn_ack(sk, req)) {
706 unsigned long expires = jiffies;
707
708 expires += reqsk_timeout(req, TCP_RTO_MAX);
709 if (!fastopen)
710 mod_timer_pending(&req->rsk_timer, expires);
711 else
712 req->rsk_timer.expires = expires;
713 }
714 return NULL;
715 }
716
717 /* Further reproduces section "SEGMENT ARRIVES"
718 for state SYN-RECEIVED of RFC793.
719 It is broken, however, it does not work only
720 when SYNs are crossed.
721
722 You would think that SYN crossing is impossible here, since
723 we should have a SYN_SENT socket (from connect()) on our end,
724 but this is not true if the crossed SYNs were sent to both
725 ends by a malicious third party. We must defend against this,
726 and to do that we first verify the ACK (as per RFC793, page
727 36) and reset if it is invalid. Is this a true full defense?
728 To convince ourselves, let us consider a way in which the ACK
729 test can still pass in this 'malicious crossed SYNs' case.
730 Malicious sender sends identical SYNs (and thus identical sequence
731 numbers) to both A and B:
732
733 A: gets SYN, seq=7
734 B: gets SYN, seq=7
735
736 By our good fortune, both A and B select the same initial
737 send sequence number of seven :-)
738
739 A: sends SYN|ACK, seq=7, ack_seq=8
740 B: sends SYN|ACK, seq=7, ack_seq=8
741
742 So we are now A eating this SYN|ACK, ACK test passes. So
743 does sequence test, SYN is truncated, and thus we consider
744 it a bare ACK.
745
746 If icsk->icsk_accept_queue.rskq_defer_accept, we silently drop this
747 bare ACK. Otherwise, we create an established connection. Both
748 ends (listening sockets) accept the new incoming connection and try
749 to talk to each other. 8-)
750
751 Note: This case is both harmless, and rare. Possibility is about the
752 same as us discovering intelligent life on another plant tomorrow.
753
754 But generally, we should (RFC lies!) to accept ACK
755 from SYNACK both here and in tcp_rcv_state_process().
756 tcp_rcv_state_process() does not, hence, we do not too.
757
758 Note that the case is absolutely generic:
759 we cannot optimize anything here without
760 violating protocol. All the checks must be made
761 before attempt to create socket.
762 */
763
764 /* RFC793 page 36: "If the connection is in any non-synchronized state ...
765 * and the incoming segment acknowledges something not yet
766 * sent (the segment carries an unacceptable ACK) ...
767 * a reset is sent."
768 *
769 * Invalid ACK: reset will be sent by listening socket.
770 * Note that the ACK validity check for a Fast Open socket is done
771 * elsewhere and is checked directly against the child socket rather
772 * than req because user data may have been sent out.
773 */
774 if ((flg & TCP_FLAG_ACK) && !fastopen &&
775 (TCP_SKB_CB(skb)->ack_seq !=
776 tcp_rsk(req)->snt_isn + 1))
777 return sk;
778
779 /* Also, it would be not so bad idea to check rcv_tsecr, which
780 * is essentially ACK extension and too early or too late values
781 * should cause reset in unsynchronized states.
782 */
783
784 /* RFC793: "first check sequence number". */
785
786 if (paws_reject || !tcp_in_window(TCP_SKB_CB(skb)->seq,
787 TCP_SKB_CB(skb)->end_seq,
788 tcp_rsk(req)->rcv_nxt,
789 tcp_rsk(req)->rcv_nxt +
790 tcp_synack_window(req))) {
791 /* Out of window: send ACK and drop. */
792 if (!(flg & TCP_FLAG_RST) &&
793 !tcp_oow_rate_limited(sock_net(sk), skb,
794 LINUX_MIB_TCPACKSKIPPEDSYNRECV,
795 &tcp_rsk(req)->last_oow_ack_time))
796 req->rsk_ops->send_ack(sk, skb, req);
797 if (paws_reject)
798 NET_INC_STATS(sock_net(sk), LINUX_MIB_PAWSESTABREJECTED);
799 return NULL;
800 }
801
802 /* In sequence, PAWS is OK. */
803
804 /* TODO: We probably should defer ts_recent change once
805 * we take ownership of @req.
806 */
807 if (tmp_opt.saw_tstamp && !after(TCP_SKB_CB(skb)->seq, tcp_rsk(req)->rcv_nxt))
808 WRITE_ONCE(req->ts_recent, tmp_opt.rcv_tsval);
809
810 if (TCP_SKB_CB(skb)->seq == tcp_rsk(req)->rcv_isn) {
811 /* Truncate SYN, it is out of window starting
812 at tcp_rsk(req)->rcv_isn + 1. */
813 flg &= ~TCP_FLAG_SYN;
814 }
815
816 /* RFC793: "second check the RST bit" and
817 * "fourth, check the SYN bit"
818 */
819 if (flg & (TCP_FLAG_RST|TCP_FLAG_SYN)) {
820 TCP_INC_STATS(sock_net(sk), TCP_MIB_ATTEMPTFAILS);
821 goto embryonic_reset;
822 }
823
824 /* ACK sequence verified above, just make sure ACK is
825 * set. If ACK not set, just silently drop the packet.
826 *
827 * XXX (TFO) - if we ever allow "data after SYN", the
828 * following check needs to be removed.
829 */
830 if (!(flg & TCP_FLAG_ACK))
831 return NULL;
832
833 /* For Fast Open no more processing is needed (sk is the
834 * child socket).
835 */
836 if (fastopen)
837 return sk;
838
839 /* While TCP_DEFER_ACCEPT is active, drop bare ACK. */
840 if (req->num_timeout < READ_ONCE(inet_csk(sk)->icsk_accept_queue.rskq_defer_accept) &&
841 TCP_SKB_CB(skb)->end_seq == tcp_rsk(req)->rcv_isn + 1) {
842 inet_rsk(req)->acked = 1;
843 __NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPDEFERACCEPTDROP);
844 return NULL;
845 }
846
847 /* OK, ACK is valid, create big socket and
848 * feed this segment to it. It will repeat all
849 * the tests. THIS SEGMENT MUST MOVE SOCKET TO
850 * ESTABLISHED STATE. If it will be dropped after
851 * socket is created, wait for troubles.
852 */
853 child = inet_csk(sk)->icsk_af_ops->syn_recv_sock(sk, skb, req, NULL,
854 req, &own_req);
855 if (!child)
856 goto listen_overflow;
857
858 if (own_req && rsk_drop_req(req)) {
859 reqsk_queue_removed(&inet_csk(req->rsk_listener)->icsk_accept_queue, req);
860 inet_csk_reqsk_queue_drop_and_put(req->rsk_listener, req);
861 return child;
862 }
863
864 sock_rps_save_rxhash(child, skb);
865 tcp_synack_rtt_meas(child, req);
866 *req_stolen = !own_req;
867 return inet_csk_complete_hashdance(sk, child, req, own_req);
868
869 listen_overflow:
870 if (sk != req->rsk_listener)
871 __NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPMIGRATEREQFAILURE);
872
873 if (!READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_abort_on_overflow)) {
874 inet_rsk(req)->acked = 1;
875 return NULL;
876 }
877
878 embryonic_reset:
879 if (!(flg & TCP_FLAG_RST)) {
880 /* Received a bad SYN pkt - for TFO We try not to reset
881 * the local connection unless it's really necessary to
882 * avoid becoming vulnerable to outside attack aiming at
883 * resetting legit local connections.
884 */
885 req->rsk_ops->send_reset(sk, skb, SK_RST_REASON_INVALID_SYN);
886 } else if (fastopen) { /* received a valid RST pkt */
887 reqsk_fastopen_remove(sk, req, true);
888 tcp_reset(sk, skb);
889 }
890 if (!fastopen) {
891 bool unlinked = inet_csk_reqsk_queue_drop(sk, req);
892
893 if (unlinked)
894 __NET_INC_STATS(sock_net(sk), LINUX_MIB_EMBRYONICRSTS);
895 *req_stolen = !unlinked;
896 }
897 return NULL;
898 }
899 EXPORT_SYMBOL(tcp_check_req);
900
901 /*
902 * Queue segment on the new socket if the new socket is active,
903 * otherwise we just shortcircuit this and continue with
904 * the new socket.
905 *
906 * For the vast majority of cases child->sk_state will be TCP_SYN_RECV
907 * when entering. But other states are possible due to a race condition
908 * where after __inet_lookup_established() fails but before the listener
909 * locked is obtained, other packets cause the same connection to
910 * be created.
911 */
912
913 enum skb_drop_reason tcp_child_process(struct sock *parent, struct sock *child,
914 struct sk_buff *skb)
915 __releases(&((child)->sk_lock.slock))
916 {
917 enum skb_drop_reason reason = SKB_NOT_DROPPED_YET;
918 int state = child->sk_state;
919
920 /* record sk_napi_id and sk_rx_queue_mapping of child. */
921 sk_mark_napi_id_set(child, skb);
922
923 tcp_segs_in(tcp_sk(child), skb);
924 if (!sock_owned_by_user(child)) {
925 reason = tcp_rcv_state_process(child, skb);
926 /* Wakeup parent, send SIGIO */
927 if (state == TCP_SYN_RECV && child->sk_state != state)
928 parent->sk_data_ready(parent);
929 } else {
930 /* Alas, it is possible again, because we do lookup
931 * in main socket hash table and lock on listening
932 * socket does not protect us more.
933 */
934 __sk_add_backlog(child, skb);
935 }
936
937 bh_unlock_sock(child);
938 sock_put(child);
939 return reason;
940 }
941 EXPORT_SYMBOL(tcp_child_process);
942
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.