1 /*
2 * INET An implementation of the TCP/IP protocol suite for the LINUX
3 * operating system. INET is implemented using the BSD Socket
4 * interface as the means of communication with the user level.
5 *
6 * Implementation of the Transmission Control Protocol(TCP).
7 *
8 * Authors: Ross Biro
9 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
10 * Mark Evans, <evansmp@uhura.aston.ac.uk>
11 * Corey Minyard <wf-rch!minyard@relay.EU.net>
12 * Florian La Roche, <flla@stud.uni-sb.de>
13 * Charles Hedrick, <hedrick@klinzhai.rutgers.edu>
14 * Linus Torvalds, <torvalds@cs.helsinki.fi>
15 * Alan Cox, <gw4pts@gw4pts.ampr.org>
16 * Matthew Dillon, <dillon@apollo.west.oic.com>
17 * Arnt Gulbrandsen, <agulbra@nvg.unit.no>
18 * Jorge Cwik, <jorge@laser.satlink.net>
19 */
20
21 #include <linux/mm.h>
22 #include <linux/module.h>
23 #include <linux/slab.h>
24 #include <linux/sysctl.h>
25 #include <linux/workqueue.h>
26 #include <net/tcp.h>
27 #include <net/inet_common.h>
28 #include <net/xfrm.h>
29 #include <net/busy_poll.h>
30
31 int sysctl_tcp_abort_on_overflow __read_mostly;
32
tcp_in_window(u32 seq,u32 end_seq,u32 s_win,u32 e_win)33 static bool tcp_in_window(u32 seq, u32 end_seq, u32 s_win, u32 e_win)
34 {
35 if (seq == s_win)
36 return true;
37 if (after(end_seq, s_win) && before(seq, e_win))
38 return true;
39 return seq == e_win && seq == end_seq;
40 }
41
42 static enum tcp_tw_status
tcp_timewait_check_oow_rate_limit(struct inet_timewait_sock * tw,const struct sk_buff * skb,int mib_idx)43 tcp_timewait_check_oow_rate_limit(struct inet_timewait_sock *tw,
44 const struct sk_buff *skb, int mib_idx)
45 {
46 struct tcp_timewait_sock *tcptw = tcp_twsk((struct sock *)tw);
47
48 if (!tcp_oow_rate_limited(twsk_net(tw), skb, mib_idx,
49 &tcptw->tw_last_oow_ack_time)) {
50 /* Send ACK. Note, we do not put the bucket,
51 * it will be released by caller.
52 */
53 return TCP_TW_ACK;
54 }
55
56 /* We are rate-limiting, so just release the tw sock and drop skb. */
57 inet_twsk_put(tw);
58 return TCP_TW_SUCCESS;
59 }
60
61 /*
62 * * Main purpose of TIME-WAIT state is to close connection gracefully,
63 * when one of ends sits in LAST-ACK or CLOSING retransmitting FIN
64 * (and, probably, tail of data) and one or more our ACKs are lost.
65 * * What is TIME-WAIT timeout? It is associated with maximal packet
66 * lifetime in the internet, which results in wrong conclusion, that
67 * it is set to catch "old duplicate segments" wandering out of their path.
68 * It is not quite correct. This timeout is calculated so that it exceeds
69 * maximal retransmission timeout enough to allow to lose one (or more)
70 * segments sent by peer and our ACKs. This time may be calculated from RTO.
71 * * When TIME-WAIT socket receives RST, it means that another end
72 * finally closed and we are allowed to kill TIME-WAIT too.
73 * * Second purpose of TIME-WAIT is catching old duplicate segments.
74 * Well, certainly it is pure paranoia, but if we load TIME-WAIT
75 * with this semantics, we MUST NOT kill TIME-WAIT state with RSTs.
76 * * If we invented some more clever way to catch duplicates
77 * (f.e. based on PAWS), we could truncate TIME-WAIT to several RTOs.
78 *
79 * The algorithm below is based on FORMAL INTERPRETATION of RFCs.
80 * When you compare it to RFCs, please, read section SEGMENT ARRIVES
81 * from the very beginning.
82 *
83 * NOTE. With recycling (and later with fin-wait-2) TW bucket
84 * is _not_ stateless. It means, that strictly speaking we must
85 * spinlock it. I do not want! Well, probability of misbehaviour
86 * is ridiculously low and, seems, we could use some mb() tricks
87 * to avoid misread sequence numbers, states etc. --ANK
88 *
89 * We don't need to initialize tmp_out.sack_ok as we don't use the results
90 */
91 enum tcp_tw_status
tcp_timewait_state_process(struct inet_timewait_sock * tw,struct sk_buff * skb,const struct tcphdr * th)92 tcp_timewait_state_process(struct inet_timewait_sock *tw, struct sk_buff *skb,
93 const struct tcphdr *th)
94 {
95 struct tcp_options_received tmp_opt;
96 struct tcp_timewait_sock *tcptw = tcp_twsk((struct sock *)tw);
97 bool paws_reject = false;
98
99 tmp_opt.saw_tstamp = 0;
100 if (th->doff > (sizeof(*th) >> 2) && tcptw->tw_ts_recent_stamp) {
101 tcp_parse_options(twsk_net(tw), skb, &tmp_opt, 0, NULL);
102
103 if (tmp_opt.saw_tstamp) {
104 if (tmp_opt.rcv_tsecr)
105 tmp_opt.rcv_tsecr -= tcptw->tw_ts_offset;
106 tmp_opt.ts_recent = tcptw->tw_ts_recent;
107 tmp_opt.ts_recent_stamp = tcptw->tw_ts_recent_stamp;
108 paws_reject = tcp_paws_reject(&tmp_opt, th->rst);
109 }
110 }
111
112 if (tw->tw_substate == TCP_FIN_WAIT2) {
113 /* Just repeat all the checks of tcp_rcv_state_process() */
114
115 /* Out of window, send ACK */
116 if (paws_reject ||
117 !tcp_in_window(TCP_SKB_CB(skb)->seq, TCP_SKB_CB(skb)->end_seq,
118 tcptw->tw_rcv_nxt,
119 tcptw->tw_rcv_nxt + tcptw->tw_rcv_wnd))
120 return tcp_timewait_check_oow_rate_limit(
121 tw, skb, LINUX_MIB_TCPACKSKIPPEDFINWAIT2);
122
123 if (th->rst)
124 goto kill;
125
126 if (th->syn && !before(TCP_SKB_CB(skb)->seq, tcptw->tw_rcv_nxt))
127 return TCP_TW_RST;
128
129 /* Dup ACK? */
130 if (!th->ack ||
131 !after(TCP_SKB_CB(skb)->end_seq, tcptw->tw_rcv_nxt) ||
132 TCP_SKB_CB(skb)->end_seq == TCP_SKB_CB(skb)->seq) {
133 inet_twsk_put(tw);
134 return TCP_TW_SUCCESS;
135 }
136
137 /* New data or FIN. If new data arrive after half-duplex close,
138 * reset.
139 */
140 if (!th->fin ||
141 TCP_SKB_CB(skb)->end_seq != tcptw->tw_rcv_nxt + 1)
142 return TCP_TW_RST;
143
144 /* FIN arrived, enter true time-wait state. */
145 tw->tw_substate = TCP_TIME_WAIT;
146 tcptw->tw_rcv_nxt = TCP_SKB_CB(skb)->end_seq;
147 if (tmp_opt.saw_tstamp) {
148 tcptw->tw_ts_recent_stamp = get_seconds();
149 tcptw->tw_ts_recent = tmp_opt.rcv_tsval;
150 }
151
152 inet_twsk_reschedule(tw, TCP_TIMEWAIT_LEN);
153 return TCP_TW_ACK;
154 }
155
156 /*
157 * Now real TIME-WAIT state.
158 *
159 * RFC 1122:
160 * "When a connection is [...] on TIME-WAIT state [...]
161 * [a TCP] MAY accept a new SYN from the remote TCP to
162 * reopen the connection directly, if it:
163 *
164 * (1) assigns its initial sequence number for the new
165 * connection to be larger than the largest sequence
166 * number it used on the previous connection incarnation,
167 * and
168 *
169 * (2) returns to TIME-WAIT state if the SYN turns out
170 * to be an old duplicate".
171 */
172
173 if (!paws_reject &&
174 (TCP_SKB_CB(skb)->seq == tcptw->tw_rcv_nxt &&
175 (TCP_SKB_CB(skb)->seq == TCP_SKB_CB(skb)->end_seq || th->rst))) {
176 /* In window segment, it may be only reset or bare ack. */
177
178 if (th->rst) {
179 /* This is TIME_WAIT assassination, in two flavors.
180 * Oh well... nobody has a sufficient solution to this
181 * protocol bug yet.
182 */
183 if (sysctl_tcp_rfc1337 == 0) {
184 kill:
185 inet_twsk_deschedule_put(tw);
186 return TCP_TW_SUCCESS;
187 }
188 } else {
189 inet_twsk_reschedule(tw, TCP_TIMEWAIT_LEN);
190 }
191
192 if (tmp_opt.saw_tstamp) {
193 tcptw->tw_ts_recent = tmp_opt.rcv_tsval;
194 tcptw->tw_ts_recent_stamp = get_seconds();
195 }
196
197 inet_twsk_put(tw);
198 return TCP_TW_SUCCESS;
199 }
200
201 /* Out of window segment.
202
203 All the segments are ACKed immediately.
204
205 The only exception is new SYN. We accept it, if it is
206 not old duplicate and we are not in danger to be killed
207 by delayed old duplicates. RFC check is that it has
208 newer sequence number works at rates <40Mbit/sec.
209 However, if paws works, it is reliable AND even more,
210 we even may relax silly seq space cutoff.
211
212 RED-PEN: we violate main RFC requirement, if this SYN will appear
213 old duplicate (i.e. we receive RST in reply to SYN-ACK),
214 we must return socket to time-wait state. It is not good,
215 but not fatal yet.
216 */
217
218 if (th->syn && !th->rst && !th->ack && !paws_reject &&
219 (after(TCP_SKB_CB(skb)->seq, tcptw->tw_rcv_nxt) ||
220 (tmp_opt.saw_tstamp &&
221 (s32)(tcptw->tw_ts_recent - tmp_opt.rcv_tsval) < 0))) {
222 u32 isn = tcptw->tw_snd_nxt + 65535 + 2;
223 if (isn == 0)
224 isn++;
225 TCP_SKB_CB(skb)->tcp_tw_isn = isn;
226 return TCP_TW_SYN;
227 }
228
229 if (paws_reject)
230 __NET_INC_STATS(twsk_net(tw), LINUX_MIB_PAWSESTABREJECTED);
231
232 if (!th->rst) {
233 /* In this case we must reset the TIMEWAIT timer.
234 *
235 * If it is ACKless SYN it may be both old duplicate
236 * and new good SYN with random sequence number <rcv_nxt.
237 * Do not reschedule in the last case.
238 */
239 if (paws_reject || th->ack)
240 inet_twsk_reschedule(tw, TCP_TIMEWAIT_LEN);
241
242 return tcp_timewait_check_oow_rate_limit(
243 tw, skb, LINUX_MIB_TCPACKSKIPPEDTIMEWAIT);
244 }
245 inet_twsk_put(tw);
246 return TCP_TW_SUCCESS;
247 }
248 EXPORT_SYMBOL(tcp_timewait_state_process);
249
250 /*
251 * Move a socket to time-wait or dead fin-wait-2 state.
252 */
tcp_time_wait(struct sock * sk,int state,int timeo)253 void tcp_time_wait(struct sock *sk, int state, int timeo)
254 {
255 const struct inet_connection_sock *icsk = inet_csk(sk);
256 const struct tcp_sock *tp = tcp_sk(sk);
257 struct inet_timewait_sock *tw;
258 struct inet_timewait_death_row *tcp_death_row = &sock_net(sk)->ipv4.tcp_death_row;
259
260 tw = inet_twsk_alloc(sk, tcp_death_row, state);
261
262 if (tw) {
263 struct tcp_timewait_sock *tcptw = tcp_twsk((struct sock *)tw);
264 const int rto = (icsk->icsk_rto << 2) - (icsk->icsk_rto >> 1);
265 struct inet_sock *inet = inet_sk(sk);
266
267 tw->tw_transparent = inet->transparent;
268 tw->tw_rcv_wscale = tp->rx_opt.rcv_wscale;
269 tcptw->tw_rcv_nxt = tp->rcv_nxt;
270 tcptw->tw_snd_nxt = tp->snd_nxt;
271 tcptw->tw_rcv_wnd = tcp_receive_window(tp);
272 tcptw->tw_ts_recent = tp->rx_opt.ts_recent;
273 tcptw->tw_ts_recent_stamp = tp->rx_opt.ts_recent_stamp;
274 tcptw->tw_ts_offset = tp->tsoffset;
275 tcptw->tw_last_oow_ack_time = 0;
276
277 #if IS_ENABLED(CONFIG_IPV6)
278 if (tw->tw_family == PF_INET6) {
279 struct ipv6_pinfo *np = inet6_sk(sk);
280
281 tw->tw_v6_daddr = sk->sk_v6_daddr;
282 tw->tw_v6_rcv_saddr = sk->sk_v6_rcv_saddr;
283 tw->tw_tclass = np->tclass;
284 tw->tw_flowlabel = be32_to_cpu(np->flow_label & IPV6_FLOWLABEL_MASK);
285 tw->tw_ipv6only = sk->sk_ipv6only;
286 }
287 #endif
288
289 #ifdef CONFIG_TCP_MD5SIG
290 /*
291 * The timewait bucket does not have the key DB from the
292 * sock structure. We just make a quick copy of the
293 * md5 key being used (if indeed we are using one)
294 * so the timewait ack generating code has the key.
295 */
296 do {
297 struct tcp_md5sig_key *key;
298 tcptw->tw_md5_key = NULL;
299 key = tp->af_specific->md5_lookup(sk, sk);
300 if (key) {
301 tcptw->tw_md5_key = kmemdup(key, sizeof(*key), GFP_ATOMIC);
302 if (tcptw->tw_md5_key && !tcp_alloc_md5sig_pool())
303 BUG();
304 }
305 } while (0);
306 #endif
307
308 /* Get the TIME_WAIT timeout firing. */
309 if (timeo < rto)
310 timeo = rto;
311
312 tw->tw_timeout = TCP_TIMEWAIT_LEN;
313 if (state == TCP_TIME_WAIT)
314 timeo = TCP_TIMEWAIT_LEN;
315
316 /* tw_timer is pinned, so we need to make sure BH are disabled
317 * in following section, otherwise timer handler could run before
318 * we complete the initialization.
319 */
320 local_bh_disable();
321 inet_twsk_schedule(tw, timeo);
322 /* Linkage updates. */
323 __inet_twsk_hashdance(tw, sk, &tcp_hashinfo);
324 inet_twsk_put(tw);
325 local_bh_enable();
326 } else {
327 /* Sorry, if we're out of memory, just CLOSE this
328 * socket up. We've got bigger problems than
329 * non-graceful socket closings.
330 */
331 NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPTIMEWAITOVERFLOW);
332 }
333
334 tcp_update_metrics(sk);
335 tcp_done(sk);
336 }
337
tcp_twsk_destructor(struct sock * sk)338 void tcp_twsk_destructor(struct sock *sk)
339 {
340 #ifdef CONFIG_TCP_MD5SIG
341 struct tcp_timewait_sock *twsk = tcp_twsk(sk);
342
343 if (twsk->tw_md5_key)
344 kfree_rcu(twsk->tw_md5_key, rcu);
345 #endif
346 }
347 EXPORT_SYMBOL_GPL(tcp_twsk_destructor);
348
349 /* Warning : This function is called without sk_listener being locked.
350 * Be sure to read socket fields once, as their value could change under us.
351 */
tcp_openreq_init_rwin(struct request_sock * req,const struct sock * sk_listener,const struct dst_entry * dst)352 void tcp_openreq_init_rwin(struct request_sock *req,
353 const struct sock *sk_listener,
354 const struct dst_entry *dst)
355 {
356 struct inet_request_sock *ireq = inet_rsk(req);
357 const struct tcp_sock *tp = tcp_sk(sk_listener);
358 int full_space = tcp_full_space(sk_listener);
359 u32 window_clamp;
360 __u8 rcv_wscale;
361 u32 rcv_wnd;
362 int mss;
363
364 mss = tcp_mss_clamp(tp, dst_metric_advmss(dst));
365 window_clamp = READ_ONCE(tp->window_clamp);
366 /* Set this up on the first call only */
367 req->rsk_window_clamp = window_clamp ? : dst_metric(dst, RTAX_WINDOW);
368
369 /* limit the window selection if the user enforce a smaller rx buffer */
370 if (sk_listener->sk_userlocks & SOCK_RCVBUF_LOCK &&
371 (req->rsk_window_clamp > full_space || req->rsk_window_clamp == 0))
372 req->rsk_window_clamp = full_space;
373
374 rcv_wnd = tcp_rwnd_init_bpf((struct sock *)req);
375 if (rcv_wnd == 0)
376 rcv_wnd = dst_metric(dst, RTAX_INITRWND);
377 else if (full_space < rcv_wnd * mss)
378 full_space = rcv_wnd * mss;
379
380 /* tcp_full_space because it is guaranteed to be the first packet */
381 tcp_select_initial_window(full_space,
382 mss - (ireq->tstamp_ok ? TCPOLEN_TSTAMP_ALIGNED : 0),
383 &req->rsk_rcv_wnd,
384 &req->rsk_window_clamp,
385 ireq->wscale_ok,
386 &rcv_wscale,
387 rcv_wnd);
388 ireq->rcv_wscale = rcv_wscale;
389 }
390 EXPORT_SYMBOL(tcp_openreq_init_rwin);
391
tcp_ecn_openreq_child(struct tcp_sock * tp,const struct request_sock * req)392 static void tcp_ecn_openreq_child(struct tcp_sock *tp,
393 const struct request_sock *req)
394 {
395 tp->ecn_flags = inet_rsk(req)->ecn_ok ? TCP_ECN_OK : 0;
396 }
397
tcp_ca_openreq_child(struct sock * sk,const struct dst_entry * dst)398 void tcp_ca_openreq_child(struct sock *sk, const struct dst_entry *dst)
399 {
400 struct inet_connection_sock *icsk = inet_csk(sk);
401 u32 ca_key = dst_metric(dst, RTAX_CC_ALGO);
402 bool ca_got_dst = false;
403
404 if (ca_key != TCP_CA_UNSPEC) {
405 const struct tcp_congestion_ops *ca;
406
407 rcu_read_lock();
408 ca = tcp_ca_find_key(ca_key);
409 if (likely(ca && try_module_get(ca->owner))) {
410 icsk->icsk_ca_dst_locked = tcp_ca_dst_locked(dst);
411 icsk->icsk_ca_ops = ca;
412 ca_got_dst = true;
413 }
414 rcu_read_unlock();
415 }
416
417 /* If no valid choice made yet, assign current system default ca. */
418 if (!ca_got_dst &&
419 (!icsk->icsk_ca_setsockopt ||
420 !try_module_get(icsk->icsk_ca_ops->owner)))
421 tcp_assign_congestion_control(sk);
422
423 tcp_set_ca_state(sk, TCP_CA_Open);
424 }
425 EXPORT_SYMBOL_GPL(tcp_ca_openreq_child);
426
427 /* This is not only more efficient than what we used to do, it eliminates
428 * a lot of code duplication between IPv4/IPv6 SYN recv processing. -DaveM
429 *
430 * Actually, we could lots of memory writes here. tp of listening
431 * socket contains all necessary default parameters.
432 */
tcp_create_openreq_child(const struct sock * sk,struct request_sock * req,struct sk_buff * skb)433 struct sock *tcp_create_openreq_child(const struct sock *sk,
434 struct request_sock *req,
435 struct sk_buff *skb)
436 {
437 struct sock *newsk = inet_csk_clone_lock(sk, req, GFP_ATOMIC);
438
439 if (newsk) {
440 const struct inet_request_sock *ireq = inet_rsk(req);
441 struct tcp_request_sock *treq = tcp_rsk(req);
442 struct inet_connection_sock *newicsk = inet_csk(newsk);
443 struct tcp_sock *newtp = tcp_sk(newsk);
444
445 /* Now setup tcp_sock */
446 newtp->pred_flags = 0;
447
448 newtp->rcv_wup = newtp->copied_seq =
449 newtp->rcv_nxt = treq->rcv_isn + 1;
450 newtp->segs_in = 1;
451
452 newtp->snd_sml = newtp->snd_una =
453 newtp->snd_nxt = newtp->snd_up = treq->snt_isn + 1;
454
455 INIT_LIST_HEAD(&newtp->tsq_node);
456
457 tcp_init_wl(newtp, treq->rcv_isn);
458
459 newtp->srtt_us = 0;
460 newtp->mdev_us = jiffies_to_usecs(TCP_TIMEOUT_INIT);
461 minmax_reset(&newtp->rtt_min, tcp_jiffies32, ~0U);
462 newicsk->icsk_rto = TCP_TIMEOUT_INIT;
463 newicsk->icsk_ack.lrcvtime = tcp_jiffies32;
464
465 newtp->packets_out = 0;
466 newtp->retrans_out = 0;
467 newtp->sacked_out = 0;
468 newtp->fackets_out = 0;
469 newtp->snd_ssthresh = TCP_INFINITE_SSTHRESH;
470 newtp->tlp_high_seq = 0;
471 newtp->lsndtime = tcp_jiffies32;
472 newsk->sk_txhash = treq->txhash;
473 newtp->last_oow_ack_time = 0;
474 newtp->total_retrans = req->num_retrans;
475
476 /* So many TCP implementations out there (incorrectly) count the
477 * initial SYN frame in their delayed-ACK and congestion control
478 * algorithms that we must have the following bandaid to talk
479 * efficiently to them. -DaveM
480 */
481 newtp->snd_cwnd = TCP_INIT_CWND;
482 newtp->snd_cwnd_cnt = 0;
483
484 /* There's a bubble in the pipe until at least the first ACK. */
485 newtp->app_limited = ~0U;
486
487 tcp_init_xmit_timers(newsk);
488 newtp->write_seq = newtp->pushed_seq = treq->snt_isn + 1;
489
490 newtp->rx_opt.saw_tstamp = 0;
491
492 newtp->rx_opt.dsack = 0;
493 newtp->rx_opt.num_sacks = 0;
494
495 newtp->urg_data = 0;
496
497 if (sock_flag(newsk, SOCK_KEEPOPEN))
498 inet_csk_reset_keepalive_timer(newsk,
499 keepalive_time_when(newtp));
500
501 newtp->rx_opt.tstamp_ok = ireq->tstamp_ok;
502 if ((newtp->rx_opt.sack_ok = ireq->sack_ok) != 0) {
503 if (sysctl_tcp_fack)
504 tcp_enable_fack(newtp);
505 }
506 newtp->window_clamp = req->rsk_window_clamp;
507 newtp->rcv_ssthresh = req->rsk_rcv_wnd;
508 newtp->rcv_wnd = req->rsk_rcv_wnd;
509 newtp->rx_opt.wscale_ok = ireq->wscale_ok;
510 if (newtp->rx_opt.wscale_ok) {
511 newtp->rx_opt.snd_wscale = ireq->snd_wscale;
512 newtp->rx_opt.rcv_wscale = ireq->rcv_wscale;
513 } else {
514 newtp->rx_opt.snd_wscale = newtp->rx_opt.rcv_wscale = 0;
515 newtp->window_clamp = min(newtp->window_clamp, 65535U);
516 }
517 newtp->snd_wnd = (ntohs(tcp_hdr(skb)->window) <<
518 newtp->rx_opt.snd_wscale);
519 newtp->max_window = newtp->snd_wnd;
520
521 if (newtp->rx_opt.tstamp_ok) {
522 newtp->rx_opt.ts_recent = req->ts_recent;
523 newtp->rx_opt.ts_recent_stamp = get_seconds();
524 newtp->tcp_header_len = sizeof(struct tcphdr) + TCPOLEN_TSTAMP_ALIGNED;
525 } else {
526 newtp->rx_opt.ts_recent_stamp = 0;
527 newtp->tcp_header_len = sizeof(struct tcphdr);
528 }
529 newtp->tsoffset = treq->ts_off;
530 #ifdef CONFIG_TCP_MD5SIG
531 newtp->md5sig_info = NULL; /*XXX*/
532 if (newtp->af_specific->md5_lookup(sk, newsk))
533 newtp->tcp_header_len += TCPOLEN_MD5SIG_ALIGNED;
534 #endif
535 if (skb->len >= TCP_MSS_DEFAULT + newtp->tcp_header_len)
536 newicsk->icsk_ack.last_seg_size = skb->len - newtp->tcp_header_len;
537 newtp->rx_opt.mss_clamp = req->mss;
538 tcp_ecn_openreq_child(newtp, req);
539 newtp->fastopen_req = NULL;
540 newtp->fastopen_rsk = NULL;
541 newtp->syn_data_acked = 0;
542 newtp->rack.mstamp = 0;
543 newtp->rack.advanced = 0;
544
545 __TCP_INC_STATS(sock_net(sk), TCP_MIB_PASSIVEOPENS);
546 }
547 return newsk;
548 }
549 EXPORT_SYMBOL(tcp_create_openreq_child);
550
551 /*
552 * Process an incoming packet for SYN_RECV sockets represented as a
553 * request_sock. Normally sk is the listener socket but for TFO it
554 * points to the child socket.
555 *
556 * XXX (TFO) - The current impl contains a special check for ack
557 * validation and inside tcp_v4_reqsk_send_ack(). Can we do better?
558 *
559 * We don't need to initialize tmp_opt.sack_ok as we don't use the results
560 */
561
tcp_check_req(struct sock * sk,struct sk_buff * skb,struct request_sock * req,bool fastopen)562 struct sock *tcp_check_req(struct sock *sk, struct sk_buff *skb,
563 struct request_sock *req,
564 bool fastopen)
565 {
566 struct tcp_options_received tmp_opt;
567 struct sock *child;
568 const struct tcphdr *th = tcp_hdr(skb);
569 __be32 flg = tcp_flag_word(th) & (TCP_FLAG_RST|TCP_FLAG_SYN|TCP_FLAG_ACK);
570 bool paws_reject = false;
571 bool own_req;
572
573 tmp_opt.saw_tstamp = 0;
574 if (th->doff > (sizeof(struct tcphdr)>>2)) {
575 tcp_parse_options(sock_net(sk), skb, &tmp_opt, 0, NULL);
576
577 if (tmp_opt.saw_tstamp) {
578 tmp_opt.ts_recent = req->ts_recent;
579 if (tmp_opt.rcv_tsecr)
580 tmp_opt.rcv_tsecr -= tcp_rsk(req)->ts_off;
581 /* We do not store true stamp, but it is not required,
582 * it can be estimated (approximately)
583 * from another data.
584 */
585 tmp_opt.ts_recent_stamp = get_seconds() - ((TCP_TIMEOUT_INIT/HZ)<<req->num_timeout);
586 paws_reject = tcp_paws_reject(&tmp_opt, th->rst);
587 }
588 }
589
590 /* Check for pure retransmitted SYN. */
591 if (TCP_SKB_CB(skb)->seq == tcp_rsk(req)->rcv_isn &&
592 flg == TCP_FLAG_SYN &&
593 !paws_reject) {
594 /*
595 * RFC793 draws (Incorrectly! It was fixed in RFC1122)
596 * this case on figure 6 and figure 8, but formal
597 * protocol description says NOTHING.
598 * To be more exact, it says that we should send ACK,
599 * because this segment (at least, if it has no data)
600 * is out of window.
601 *
602 * CONCLUSION: RFC793 (even with RFC1122) DOES NOT
603 * describe SYN-RECV state. All the description
604 * is wrong, we cannot believe to it and should
605 * rely only on common sense and implementation
606 * experience.
607 *
608 * Enforce "SYN-ACK" according to figure 8, figure 6
609 * of RFC793, fixed by RFC1122.
610 *
611 * Note that even if there is new data in the SYN packet
612 * they will be thrown away too.
613 *
614 * Reset timer after retransmitting SYNACK, similar to
615 * the idea of fast retransmit in recovery.
616 */
617 if (!tcp_oow_rate_limited(sock_net(sk), skb,
618 LINUX_MIB_TCPACKSKIPPEDSYNRECV,
619 &tcp_rsk(req)->last_oow_ack_time) &&
620
621 !inet_rtx_syn_ack(sk, req)) {
622 unsigned long expires = jiffies;
623
624 expires += min(TCP_TIMEOUT_INIT << req->num_timeout,
625 TCP_RTO_MAX);
626 if (!fastopen)
627 mod_timer_pending(&req->rsk_timer, expires);
628 else
629 req->rsk_timer.expires = expires;
630 }
631 return NULL;
632 }
633
634 /* Further reproduces section "SEGMENT ARRIVES"
635 for state SYN-RECEIVED of RFC793.
636 It is broken, however, it does not work only
637 when SYNs are crossed.
638
639 You would think that SYN crossing is impossible here, since
640 we should have a SYN_SENT socket (from connect()) on our end,
641 but this is not true if the crossed SYNs were sent to both
642 ends by a malicious third party. We must defend against this,
643 and to do that we first verify the ACK (as per RFC793, page
644 36) and reset if it is invalid. Is this a true full defense?
645 To convince ourselves, let us consider a way in which the ACK
646 test can still pass in this 'malicious crossed SYNs' case.
647 Malicious sender sends identical SYNs (and thus identical sequence
648 numbers) to both A and B:
649
650 A: gets SYN, seq=7
651 B: gets SYN, seq=7
652
653 By our good fortune, both A and B select the same initial
654 send sequence number of seven :-)
655
656 A: sends SYN|ACK, seq=7, ack_seq=8
657 B: sends SYN|ACK, seq=7, ack_seq=8
658
659 So we are now A eating this SYN|ACK, ACK test passes. So
660 does sequence test, SYN is truncated, and thus we consider
661 it a bare ACK.
662
663 If icsk->icsk_accept_queue.rskq_defer_accept, we silently drop this
664 bare ACK. Otherwise, we create an established connection. Both
665 ends (listening sockets) accept the new incoming connection and try
666 to talk to each other. 8-)
667
668 Note: This case is both harmless, and rare. Possibility is about the
669 same as us discovering intelligent life on another plant tomorrow.
670
671 But generally, we should (RFC lies!) to accept ACK
672 from SYNACK both here and in tcp_rcv_state_process().
673 tcp_rcv_state_process() does not, hence, we do not too.
674
675 Note that the case is absolutely generic:
676 we cannot optimize anything here without
677 violating protocol. All the checks must be made
678 before attempt to create socket.
679 */
680
681 /* RFC793 page 36: "If the connection is in any non-synchronized state ...
682 * and the incoming segment acknowledges something not yet
683 * sent (the segment carries an unacceptable ACK) ...
684 * a reset is sent."
685 *
686 * Invalid ACK: reset will be sent by listening socket.
687 * Note that the ACK validity check for a Fast Open socket is done
688 * elsewhere and is checked directly against the child socket rather
689 * than req because user data may have been sent out.
690 */
691 if ((flg & TCP_FLAG_ACK) && !fastopen &&
692 (TCP_SKB_CB(skb)->ack_seq !=
693 tcp_rsk(req)->snt_isn + 1))
694 return sk;
695
696 /* Also, it would be not so bad idea to check rcv_tsecr, which
697 * is essentially ACK extension and too early or too late values
698 * should cause reset in unsynchronized states.
699 */
700
701 /* RFC793: "first check sequence number". */
702
703 if (paws_reject || !tcp_in_window(TCP_SKB_CB(skb)->seq, TCP_SKB_CB(skb)->end_seq,
704 tcp_rsk(req)->rcv_nxt, tcp_rsk(req)->rcv_nxt + req->rsk_rcv_wnd)) {
705 /* Out of window: send ACK and drop. */
706 if (!(flg & TCP_FLAG_RST) &&
707 !tcp_oow_rate_limited(sock_net(sk), skb,
708 LINUX_MIB_TCPACKSKIPPEDSYNRECV,
709 &tcp_rsk(req)->last_oow_ack_time))
710 req->rsk_ops->send_ack(sk, skb, req);
711 if (paws_reject)
712 __NET_INC_STATS(sock_net(sk), LINUX_MIB_PAWSESTABREJECTED);
713 return NULL;
714 }
715
716 /* In sequence, PAWS is OK. */
717
718 if (tmp_opt.saw_tstamp && !after(TCP_SKB_CB(skb)->seq, tcp_rsk(req)->rcv_nxt))
719 req->ts_recent = tmp_opt.rcv_tsval;
720
721 if (TCP_SKB_CB(skb)->seq == tcp_rsk(req)->rcv_isn) {
722 /* Truncate SYN, it is out of window starting
723 at tcp_rsk(req)->rcv_isn + 1. */
724 flg &= ~TCP_FLAG_SYN;
725 }
726
727 /* RFC793: "second check the RST bit" and
728 * "fourth, check the SYN bit"
729 */
730 if (flg & (TCP_FLAG_RST|TCP_FLAG_SYN)) {
731 __TCP_INC_STATS(sock_net(sk), TCP_MIB_ATTEMPTFAILS);
732 goto embryonic_reset;
733 }
734
735 /* ACK sequence verified above, just make sure ACK is
736 * set. If ACK not set, just silently drop the packet.
737 *
738 * XXX (TFO) - if we ever allow "data after SYN", the
739 * following check needs to be removed.
740 */
741 if (!(flg & TCP_FLAG_ACK))
742 return NULL;
743
744 /* For Fast Open no more processing is needed (sk is the
745 * child socket).
746 */
747 if (fastopen)
748 return sk;
749
750 /* While TCP_DEFER_ACCEPT is active, drop bare ACK. */
751 if (req->num_timeout < inet_csk(sk)->icsk_accept_queue.rskq_defer_accept &&
752 TCP_SKB_CB(skb)->end_seq == tcp_rsk(req)->rcv_isn + 1) {
753 inet_rsk(req)->acked = 1;
754 __NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPDEFERACCEPTDROP);
755 return NULL;
756 }
757
758 /* OK, ACK is valid, create big socket and
759 * feed this segment to it. It will repeat all
760 * the tests. THIS SEGMENT MUST MOVE SOCKET TO
761 * ESTABLISHED STATE. If it will be dropped after
762 * socket is created, wait for troubles.
763 */
764 child = inet_csk(sk)->icsk_af_ops->syn_recv_sock(sk, skb, req, NULL,
765 req, &own_req);
766 if (!child)
767 goto listen_overflow;
768
769 sock_rps_save_rxhash(child, skb);
770 tcp_synack_rtt_meas(child, req);
771 return inet_csk_complete_hashdance(sk, child, req, own_req);
772
773 listen_overflow:
774 if (!sysctl_tcp_abort_on_overflow) {
775 inet_rsk(req)->acked = 1;
776 return NULL;
777 }
778
779 embryonic_reset:
780 if (!(flg & TCP_FLAG_RST)) {
781 /* Received a bad SYN pkt - for TFO We try not to reset
782 * the local connection unless it's really necessary to
783 * avoid becoming vulnerable to outside attack aiming at
784 * resetting legit local connections.
785 */
786 req->rsk_ops->send_reset(sk, skb);
787 } else if (fastopen) { /* received a valid RST pkt */
788 reqsk_fastopen_remove(sk, req, true);
789 tcp_reset(sk);
790 }
791 if (!fastopen) {
792 inet_csk_reqsk_queue_drop(sk, req);
793 __NET_INC_STATS(sock_net(sk), LINUX_MIB_EMBRYONICRSTS);
794 }
795 return NULL;
796 }
797 EXPORT_SYMBOL(tcp_check_req);
798
799 /*
800 * Queue segment on the new socket if the new socket is active,
801 * otherwise we just shortcircuit this and continue with
802 * the new socket.
803 *
804 * For the vast majority of cases child->sk_state will be TCP_SYN_RECV
805 * when entering. But other states are possible due to a race condition
806 * where after __inet_lookup_established() fails but before the listener
807 * locked is obtained, other packets cause the same connection to
808 * be created.
809 */
810
tcp_child_process(struct sock * parent,struct sock * child,struct sk_buff * skb)811 int tcp_child_process(struct sock *parent, struct sock *child,
812 struct sk_buff *skb)
813 {
814 int ret = 0;
815 int state = child->sk_state;
816
817 /* record NAPI ID of child */
818 sk_mark_napi_id(child, skb);
819
820 tcp_segs_in(tcp_sk(child), skb);
821 if (!sock_owned_by_user(child)) {
822 ret = tcp_rcv_state_process(child, skb);
823 /* Wakeup parent, send SIGIO */
824 if (state == TCP_SYN_RECV && child->sk_state != state)
825 parent->sk_data_ready(parent);
826 } else {
827 /* Alas, it is possible again, because we do lookup
828 * in main socket hash table and lock on listening
829 * socket does not protect us more.
830 */
831 __sk_add_backlog(child, skb);
832 }
833
834 bh_unlock_sock(child);
835 sock_put(child);
836 return ret;
837 }
838 EXPORT_SYMBOL(tcp_child_process);
839