Loading...
1// SPDX-License-Identifier: GPL-2.0-or-later
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 * IPv4 specific functions
10 *
11 * code split from:
12 * linux/ipv4/tcp.c
13 * linux/ipv4/tcp_input.c
14 * linux/ipv4/tcp_output.c
15 *
16 * See tcp.c for author information
17 */
18
19/*
20 * Changes:
21 * David S. Miller : New socket lookup architecture.
22 * This code is dedicated to John Dyson.
23 * David S. Miller : Change semantics of established hash,
24 * half is devoted to TIME_WAIT sockets
25 * and the rest go in the other half.
26 * Andi Kleen : Add support for syncookies and fixed
27 * some bugs: ip options weren't passed to
28 * the TCP layer, missed a check for an
29 * ACK bit.
30 * Andi Kleen : Implemented fast path mtu discovery.
31 * Fixed many serious bugs in the
32 * request_sock handling and moved
33 * most of it into the af independent code.
34 * Added tail drop and some other bugfixes.
35 * Added new listen semantics.
36 * Mike McLagan : Routing by source
37 * Juan Jose Ciarlante: ip_dynaddr bits
38 * Andi Kleen: various fixes.
39 * Vitaly E. Lavrov : Transparent proxy revived after year
40 * coma.
41 * Andi Kleen : Fix new listen.
42 * Andi Kleen : Fix accept error reporting.
43 * YOSHIFUJI Hideaki @USAGI and: Support IPV6_V6ONLY socket option, which
44 * Alexey Kuznetsov allow both IPv4 and IPv6 sockets to bind
45 * a single port at the same time.
46 */
47
48#define pr_fmt(fmt) "TCP: " fmt
49
50#include <linux/bottom_half.h>
51#include <linux/types.h>
52#include <linux/fcntl.h>
53#include <linux/module.h>
54#include <linux/random.h>
55#include <linux/cache.h>
56#include <linux/jhash.h>
57#include <linux/init.h>
58#include <linux/times.h>
59#include <linux/slab.h>
60
61#include <net/net_namespace.h>
62#include <net/icmp.h>
63#include <net/inet_hashtables.h>
64#include <net/tcp.h>
65#include <net/transp_v6.h>
66#include <net/ipv6.h>
67#include <net/inet_common.h>
68#include <net/timewait_sock.h>
69#include <net/xfrm.h>
70#include <net/secure_seq.h>
71#include <net/busy_poll.h>
72
73#include <linux/inet.h>
74#include <linux/ipv6.h>
75#include <linux/stddef.h>
76#include <linux/proc_fs.h>
77#include <linux/seq_file.h>
78#include <linux/inetdevice.h>
79
80#include <crypto/hash.h>
81#include <linux/scatterlist.h>
82
83#include <trace/events/tcp.h>
84
85#ifdef CONFIG_TCP_MD5SIG
86static int tcp_v4_md5_hash_hdr(char *md5_hash, const struct tcp_md5sig_key *key,
87 __be32 daddr, __be32 saddr, const struct tcphdr *th);
88#endif
89
90struct inet_hashinfo tcp_hashinfo;
91EXPORT_SYMBOL(tcp_hashinfo);
92
93static u32 tcp_v4_init_seq(const struct sk_buff *skb)
94{
95 return secure_tcp_seq(ip_hdr(skb)->daddr,
96 ip_hdr(skb)->saddr,
97 tcp_hdr(skb)->dest,
98 tcp_hdr(skb)->source);
99}
100
101static u32 tcp_v4_init_ts_off(const struct net *net, const struct sk_buff *skb)
102{
103 return secure_tcp_ts_off(net, ip_hdr(skb)->daddr, ip_hdr(skb)->saddr);
104}
105
106int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp)
107{
108 const struct inet_timewait_sock *tw = inet_twsk(sktw);
109 const struct tcp_timewait_sock *tcptw = tcp_twsk(sktw);
110 struct tcp_sock *tp = tcp_sk(sk);
111 int reuse = sock_net(sk)->ipv4.sysctl_tcp_tw_reuse;
112
113 if (reuse == 2) {
114 /* Still does not detect *everything* that goes through
115 * lo, since we require a loopback src or dst address
116 * or direct binding to 'lo' interface.
117 */
118 bool loopback = false;
119 if (tw->tw_bound_dev_if == LOOPBACK_IFINDEX)
120 loopback = true;
121#if IS_ENABLED(CONFIG_IPV6)
122 if (tw->tw_family == AF_INET6) {
123 if (ipv6_addr_loopback(&tw->tw_v6_daddr) ||
124 (ipv6_addr_v4mapped(&tw->tw_v6_daddr) &&
125 (tw->tw_v6_daddr.s6_addr[12] == 127)) ||
126 ipv6_addr_loopback(&tw->tw_v6_rcv_saddr) ||
127 (ipv6_addr_v4mapped(&tw->tw_v6_rcv_saddr) &&
128 (tw->tw_v6_rcv_saddr.s6_addr[12] == 127)))
129 loopback = true;
130 } else
131#endif
132 {
133 if (ipv4_is_loopback(tw->tw_daddr) ||
134 ipv4_is_loopback(tw->tw_rcv_saddr))
135 loopback = true;
136 }
137 if (!loopback)
138 reuse = 0;
139 }
140
141 /* With PAWS, it is safe from the viewpoint
142 of data integrity. Even without PAWS it is safe provided sequence
143 spaces do not overlap i.e. at data rates <= 80Mbit/sec.
144
145 Actually, the idea is close to VJ's one, only timestamp cache is
146 held not per host, but per port pair and TW bucket is used as state
147 holder.
148
149 If TW bucket has been already destroyed we fall back to VJ's scheme
150 and use initial timestamp retrieved from peer table.
151 */
152 if (tcptw->tw_ts_recent_stamp &&
153 (!twp || (reuse && time_after32(ktime_get_seconds(),
154 tcptw->tw_ts_recent_stamp)))) {
155 /* In case of repair and re-using TIME-WAIT sockets we still
156 * want to be sure that it is safe as above but honor the
157 * sequence numbers and time stamps set as part of the repair
158 * process.
159 *
160 * Without this check re-using a TIME-WAIT socket with TCP
161 * repair would accumulate a -1 on the repair assigned
162 * sequence number. The first time it is reused the sequence
163 * is -1, the second time -2, etc. This fixes that issue
164 * without appearing to create any others.
165 */
166 if (likely(!tp->repair)) {
167 u32 seq = tcptw->tw_snd_nxt + 65535 + 2;
168
169 if (!seq)
170 seq = 1;
171 WRITE_ONCE(tp->write_seq, seq);
172 tp->rx_opt.ts_recent = tcptw->tw_ts_recent;
173 tp->rx_opt.ts_recent_stamp = tcptw->tw_ts_recent_stamp;
174 }
175 sock_hold(sktw);
176 return 1;
177 }
178
179 return 0;
180}
181EXPORT_SYMBOL_GPL(tcp_twsk_unique);
182
183static int tcp_v4_pre_connect(struct sock *sk, struct sockaddr *uaddr,
184 int addr_len)
185{
186 /* This check is replicated from tcp_v4_connect() and intended to
187 * prevent BPF program called below from accessing bytes that are out
188 * of the bound specified by user in addr_len.
189 */
190 if (addr_len < sizeof(struct sockaddr_in))
191 return -EINVAL;
192
193 sock_owned_by_me(sk);
194
195 return BPF_CGROUP_RUN_PROG_INET4_CONNECT(sk, uaddr);
196}
197
198/* This will initiate an outgoing connection. */
199int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len)
200{
201 struct sockaddr_in *usin = (struct sockaddr_in *)uaddr;
202 struct inet_sock *inet = inet_sk(sk);
203 struct tcp_sock *tp = tcp_sk(sk);
204 __be16 orig_sport, orig_dport;
205 __be32 daddr, nexthop;
206 struct flowi4 *fl4;
207 struct rtable *rt;
208 int err;
209 struct ip_options_rcu *inet_opt;
210 struct inet_timewait_death_row *tcp_death_row = &sock_net(sk)->ipv4.tcp_death_row;
211
212 if (addr_len < sizeof(struct sockaddr_in))
213 return -EINVAL;
214
215 if (usin->sin_family != AF_INET)
216 return -EAFNOSUPPORT;
217
218 nexthop = daddr = usin->sin_addr.s_addr;
219 inet_opt = rcu_dereference_protected(inet->inet_opt,
220 lockdep_sock_is_held(sk));
221 if (inet_opt && inet_opt->opt.srr) {
222 if (!daddr)
223 return -EINVAL;
224 nexthop = inet_opt->opt.faddr;
225 }
226
227 orig_sport = inet->inet_sport;
228 orig_dport = usin->sin_port;
229 fl4 = &inet->cork.fl.u.ip4;
230 rt = ip_route_connect(fl4, nexthop, inet->inet_saddr,
231 RT_CONN_FLAGS(sk), sk->sk_bound_dev_if,
232 IPPROTO_TCP,
233 orig_sport, orig_dport, sk);
234 if (IS_ERR(rt)) {
235 err = PTR_ERR(rt);
236 if (err == -ENETUNREACH)
237 IP_INC_STATS(sock_net(sk), IPSTATS_MIB_OUTNOROUTES);
238 return err;
239 }
240
241 if (rt->rt_flags & (RTCF_MULTICAST | RTCF_BROADCAST)) {
242 ip_rt_put(rt);
243 return -ENETUNREACH;
244 }
245
246 if (!inet_opt || !inet_opt->opt.srr)
247 daddr = fl4->daddr;
248
249 if (!inet->inet_saddr)
250 inet->inet_saddr = fl4->saddr;
251 sk_rcv_saddr_set(sk, inet->inet_saddr);
252
253 if (tp->rx_opt.ts_recent_stamp && inet->inet_daddr != daddr) {
254 /* Reset inherited state */
255 tp->rx_opt.ts_recent = 0;
256 tp->rx_opt.ts_recent_stamp = 0;
257 if (likely(!tp->repair))
258 WRITE_ONCE(tp->write_seq, 0);
259 }
260
261 inet->inet_dport = usin->sin_port;
262 sk_daddr_set(sk, daddr);
263
264 inet_csk(sk)->icsk_ext_hdr_len = 0;
265 if (inet_opt)
266 inet_csk(sk)->icsk_ext_hdr_len = inet_opt->opt.optlen;
267
268 tp->rx_opt.mss_clamp = TCP_MSS_DEFAULT;
269
270 /* Socket identity is still unknown (sport may be zero).
271 * However we set state to SYN-SENT and not releasing socket
272 * lock select source port, enter ourselves into the hash tables and
273 * complete initialization after this.
274 */
275 tcp_set_state(sk, TCP_SYN_SENT);
276 err = inet_hash_connect(tcp_death_row, sk);
277 if (err)
278 goto failure;
279
280 sk_set_txhash(sk);
281
282 rt = ip_route_newports(fl4, rt, orig_sport, orig_dport,
283 inet->inet_sport, inet->inet_dport, sk);
284 if (IS_ERR(rt)) {
285 err = PTR_ERR(rt);
286 rt = NULL;
287 goto failure;
288 }
289 /* OK, now commit destination to socket. */
290 sk->sk_gso_type = SKB_GSO_TCPV4;
291 sk_setup_caps(sk, &rt->dst);
292 rt = NULL;
293
294 if (likely(!tp->repair)) {
295 if (!tp->write_seq)
296 WRITE_ONCE(tp->write_seq,
297 secure_tcp_seq(inet->inet_saddr,
298 inet->inet_daddr,
299 inet->inet_sport,
300 usin->sin_port));
301 tp->tsoffset = secure_tcp_ts_off(sock_net(sk),
302 inet->inet_saddr,
303 inet->inet_daddr);
304 }
305
306 inet->inet_id = prandom_u32();
307
308 if (tcp_fastopen_defer_connect(sk, &err))
309 return err;
310 if (err)
311 goto failure;
312
313 err = tcp_connect(sk);
314
315 if (err)
316 goto failure;
317
318 return 0;
319
320failure:
321 /*
322 * This unhashes the socket and releases the local port,
323 * if necessary.
324 */
325 tcp_set_state(sk, TCP_CLOSE);
326 ip_rt_put(rt);
327 sk->sk_route_caps = 0;
328 inet->inet_dport = 0;
329 return err;
330}
331EXPORT_SYMBOL(tcp_v4_connect);
332
333/*
334 * This routine reacts to ICMP_FRAG_NEEDED mtu indications as defined in RFC1191.
335 * It can be called through tcp_release_cb() if socket was owned by user
336 * at the time tcp_v4_err() was called to handle ICMP message.
337 */
338void tcp_v4_mtu_reduced(struct sock *sk)
339{
340 struct inet_sock *inet = inet_sk(sk);
341 struct dst_entry *dst;
342 u32 mtu;
343
344 if ((1 << sk->sk_state) & (TCPF_LISTEN | TCPF_CLOSE))
345 return;
346 mtu = tcp_sk(sk)->mtu_info;
347 dst = inet_csk_update_pmtu(sk, mtu);
348 if (!dst)
349 return;
350
351 /* Something is about to be wrong... Remember soft error
352 * for the case, if this connection will not able to recover.
353 */
354 if (mtu < dst_mtu(dst) && ip_dont_fragment(sk, dst))
355 sk->sk_err_soft = EMSGSIZE;
356
357 mtu = dst_mtu(dst);
358
359 if (inet->pmtudisc != IP_PMTUDISC_DONT &&
360 ip_sk_accept_pmtu(sk) &&
361 inet_csk(sk)->icsk_pmtu_cookie > mtu) {
362 tcp_sync_mss(sk, mtu);
363
364 /* Resend the TCP packet because it's
365 * clear that the old packet has been
366 * dropped. This is the new "fast" path mtu
367 * discovery.
368 */
369 tcp_simple_retransmit(sk);
370 } /* else let the usual retransmit timer handle it */
371}
372EXPORT_SYMBOL(tcp_v4_mtu_reduced);
373
374static void do_redirect(struct sk_buff *skb, struct sock *sk)
375{
376 struct dst_entry *dst = __sk_dst_check(sk, 0);
377
378 if (dst)
379 dst->ops->redirect(dst, sk, skb);
380}
381
382
383/* handle ICMP messages on TCP_NEW_SYN_RECV request sockets */
384void tcp_req_err(struct sock *sk, u32 seq, bool abort)
385{
386 struct request_sock *req = inet_reqsk(sk);
387 struct net *net = sock_net(sk);
388
389 /* ICMPs are not backlogged, hence we cannot get
390 * an established socket here.
391 */
392 if (seq != tcp_rsk(req)->snt_isn) {
393 __NET_INC_STATS(net, LINUX_MIB_OUTOFWINDOWICMPS);
394 } else if (abort) {
395 /*
396 * Still in SYN_RECV, just remove it silently.
397 * There is no good way to pass the error to the newly
398 * created socket, and POSIX does not want network
399 * errors returned from accept().
400 */
401 inet_csk_reqsk_queue_drop(req->rsk_listener, req);
402 tcp_listendrop(req->rsk_listener);
403 }
404 reqsk_put(req);
405}
406EXPORT_SYMBOL(tcp_req_err);
407
408/*
409 * This routine is called by the ICMP module when it gets some
410 * sort of error condition. If err < 0 then the socket should
411 * be closed and the error returned to the user. If err > 0
412 * it's just the icmp type << 8 | icmp code. After adjustment
413 * header points to the first 8 bytes of the tcp header. We need
414 * to find the appropriate port.
415 *
416 * The locking strategy used here is very "optimistic". When
417 * someone else accesses the socket the ICMP is just dropped
418 * and for some paths there is no check at all.
419 * A more general error queue to queue errors for later handling
420 * is probably better.
421 *
422 */
423
424int tcp_v4_err(struct sk_buff *icmp_skb, u32 info)
425{
426 const struct iphdr *iph = (const struct iphdr *)icmp_skb->data;
427 struct tcphdr *th = (struct tcphdr *)(icmp_skb->data + (iph->ihl << 2));
428 struct inet_connection_sock *icsk;
429 struct tcp_sock *tp;
430 struct inet_sock *inet;
431 const int type = icmp_hdr(icmp_skb)->type;
432 const int code = icmp_hdr(icmp_skb)->code;
433 struct sock *sk;
434 struct sk_buff *skb;
435 struct request_sock *fastopen;
436 u32 seq, snd_una;
437 s32 remaining;
438 u32 delta_us;
439 int err;
440 struct net *net = dev_net(icmp_skb->dev);
441
442 sk = __inet_lookup_established(net, &tcp_hashinfo, iph->daddr,
443 th->dest, iph->saddr, ntohs(th->source),
444 inet_iif(icmp_skb), 0);
445 if (!sk) {
446 __ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
447 return -ENOENT;
448 }
449 if (sk->sk_state == TCP_TIME_WAIT) {
450 inet_twsk_put(inet_twsk(sk));
451 return 0;
452 }
453 seq = ntohl(th->seq);
454 if (sk->sk_state == TCP_NEW_SYN_RECV) {
455 tcp_req_err(sk, seq, type == ICMP_PARAMETERPROB ||
456 type == ICMP_TIME_EXCEEDED ||
457 (type == ICMP_DEST_UNREACH &&
458 (code == ICMP_NET_UNREACH ||
459 code == ICMP_HOST_UNREACH)));
460 return 0;
461 }
462
463 bh_lock_sock(sk);
464 /* If too many ICMPs get dropped on busy
465 * servers this needs to be solved differently.
466 * We do take care of PMTU discovery (RFC1191) special case :
467 * we can receive locally generated ICMP messages while socket is held.
468 */
469 if (sock_owned_by_user(sk)) {
470 if (!(type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED))
471 __NET_INC_STATS(net, LINUX_MIB_LOCKDROPPEDICMPS);
472 }
473 if (sk->sk_state == TCP_CLOSE)
474 goto out;
475
476 if (unlikely(iph->ttl < inet_sk(sk)->min_ttl)) {
477 __NET_INC_STATS(net, LINUX_MIB_TCPMINTTLDROP);
478 goto out;
479 }
480
481 icsk = inet_csk(sk);
482 tp = tcp_sk(sk);
483 /* XXX (TFO) - tp->snd_una should be ISN (tcp_create_openreq_child() */
484 fastopen = rcu_dereference(tp->fastopen_rsk);
485 snd_una = fastopen ? tcp_rsk(fastopen)->snt_isn : tp->snd_una;
486 if (sk->sk_state != TCP_LISTEN &&
487 !between(seq, snd_una, tp->snd_nxt)) {
488 __NET_INC_STATS(net, LINUX_MIB_OUTOFWINDOWICMPS);
489 goto out;
490 }
491
492 switch (type) {
493 case ICMP_REDIRECT:
494 if (!sock_owned_by_user(sk))
495 do_redirect(icmp_skb, sk);
496 goto out;
497 case ICMP_SOURCE_QUENCH:
498 /* Just silently ignore these. */
499 goto out;
500 case ICMP_PARAMETERPROB:
501 err = EPROTO;
502 break;
503 case ICMP_DEST_UNREACH:
504 if (code > NR_ICMP_UNREACH)
505 goto out;
506
507 if (code == ICMP_FRAG_NEEDED) { /* PMTU discovery (RFC1191) */
508 /* We are not interested in TCP_LISTEN and open_requests
509 * (SYN-ACKs send out by Linux are always <576bytes so
510 * they should go through unfragmented).
511 */
512 if (sk->sk_state == TCP_LISTEN)
513 goto out;
514
515 tp->mtu_info = info;
516 if (!sock_owned_by_user(sk)) {
517 tcp_v4_mtu_reduced(sk);
518 } else {
519 if (!test_and_set_bit(TCP_MTU_REDUCED_DEFERRED, &sk->sk_tsq_flags))
520 sock_hold(sk);
521 }
522 goto out;
523 }
524
525 err = icmp_err_convert[code].errno;
526 /* check if icmp_skb allows revert of backoff
527 * (see draft-zimmermann-tcp-lcd) */
528 if (code != ICMP_NET_UNREACH && code != ICMP_HOST_UNREACH)
529 break;
530 if (seq != tp->snd_una || !icsk->icsk_retransmits ||
531 !icsk->icsk_backoff || fastopen)
532 break;
533
534 if (sock_owned_by_user(sk))
535 break;
536
537 skb = tcp_rtx_queue_head(sk);
538 if (WARN_ON_ONCE(!skb))
539 break;
540
541 icsk->icsk_backoff--;
542 icsk->icsk_rto = tp->srtt_us ? __tcp_set_rto(tp) :
543 TCP_TIMEOUT_INIT;
544 icsk->icsk_rto = inet_csk_rto_backoff(icsk, TCP_RTO_MAX);
545
546
547 tcp_mstamp_refresh(tp);
548 delta_us = (u32)(tp->tcp_mstamp - tcp_skb_timestamp_us(skb));
549 remaining = icsk->icsk_rto -
550 usecs_to_jiffies(delta_us);
551
552 if (remaining > 0) {
553 inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
554 remaining, TCP_RTO_MAX);
555 } else {
556 /* RTO revert clocked out retransmission.
557 * Will retransmit now */
558 tcp_retransmit_timer(sk);
559 }
560
561 break;
562 case ICMP_TIME_EXCEEDED:
563 err = EHOSTUNREACH;
564 break;
565 default:
566 goto out;
567 }
568
569 switch (sk->sk_state) {
570 case TCP_SYN_SENT:
571 case TCP_SYN_RECV:
572 /* Only in fast or simultaneous open. If a fast open socket is
573 * is already accepted it is treated as a connected one below.
574 */
575 if (fastopen && !fastopen->sk)
576 break;
577
578 if (!sock_owned_by_user(sk)) {
579 sk->sk_err = err;
580
581 sk->sk_error_report(sk);
582
583 tcp_done(sk);
584 } else {
585 sk->sk_err_soft = err;
586 }
587 goto out;
588 }
589
590 /* If we've already connected we will keep trying
591 * until we time out, or the user gives up.
592 *
593 * rfc1122 4.2.3.9 allows to consider as hard errors
594 * only PROTO_UNREACH and PORT_UNREACH (well, FRAG_FAILED too,
595 * but it is obsoleted by pmtu discovery).
596 *
597 * Note, that in modern internet, where routing is unreliable
598 * and in each dark corner broken firewalls sit, sending random
599 * errors ordered by their masters even this two messages finally lose
600 * their original sense (even Linux sends invalid PORT_UNREACHs)
601 *
602 * Now we are in compliance with RFCs.
603 * --ANK (980905)
604 */
605
606 inet = inet_sk(sk);
607 if (!sock_owned_by_user(sk) && inet->recverr) {
608 sk->sk_err = err;
609 sk->sk_error_report(sk);
610 } else { /* Only an error on timeout */
611 sk->sk_err_soft = err;
612 }
613
614out:
615 bh_unlock_sock(sk);
616 sock_put(sk);
617 return 0;
618}
619
620void __tcp_v4_send_check(struct sk_buff *skb, __be32 saddr, __be32 daddr)
621{
622 struct tcphdr *th = tcp_hdr(skb);
623
624 th->check = ~tcp_v4_check(skb->len, saddr, daddr, 0);
625 skb->csum_start = skb_transport_header(skb) - skb->head;
626 skb->csum_offset = offsetof(struct tcphdr, check);
627}
628
629/* This routine computes an IPv4 TCP checksum. */
630void tcp_v4_send_check(struct sock *sk, struct sk_buff *skb)
631{
632 const struct inet_sock *inet = inet_sk(sk);
633
634 __tcp_v4_send_check(skb, inet->inet_saddr, inet->inet_daddr);
635}
636EXPORT_SYMBOL(tcp_v4_send_check);
637
638/*
639 * This routine will send an RST to the other tcp.
640 *
641 * Someone asks: why I NEVER use socket parameters (TOS, TTL etc.)
642 * for reset.
643 * Answer: if a packet caused RST, it is not for a socket
644 * existing in our system, if it is matched to a socket,
645 * it is just duplicate segment or bug in other side's TCP.
646 * So that we build reply only basing on parameters
647 * arrived with segment.
648 * Exception: precedence violation. We do not implement it in any case.
649 */
650
651static void tcp_v4_send_reset(const struct sock *sk, struct sk_buff *skb)
652{
653 const struct tcphdr *th = tcp_hdr(skb);
654 struct {
655 struct tcphdr th;
656#ifdef CONFIG_TCP_MD5SIG
657 __be32 opt[(TCPOLEN_MD5SIG_ALIGNED >> 2)];
658#endif
659 } rep;
660 struct ip_reply_arg arg;
661#ifdef CONFIG_TCP_MD5SIG
662 struct tcp_md5sig_key *key = NULL;
663 const __u8 *hash_location = NULL;
664 unsigned char newhash[16];
665 int genhash;
666 struct sock *sk1 = NULL;
667#endif
668 u64 transmit_time = 0;
669 struct sock *ctl_sk;
670 struct net *net;
671
672 /* Never send a reset in response to a reset. */
673 if (th->rst)
674 return;
675
676 /* If sk not NULL, it means we did a successful lookup and incoming
677 * route had to be correct. prequeue might have dropped our dst.
678 */
679 if (!sk && skb_rtable(skb)->rt_type != RTN_LOCAL)
680 return;
681
682 /* Swap the send and the receive. */
683 memset(&rep, 0, sizeof(rep));
684 rep.th.dest = th->source;
685 rep.th.source = th->dest;
686 rep.th.doff = sizeof(struct tcphdr) / 4;
687 rep.th.rst = 1;
688
689 if (th->ack) {
690 rep.th.seq = th->ack_seq;
691 } else {
692 rep.th.ack = 1;
693 rep.th.ack_seq = htonl(ntohl(th->seq) + th->syn + th->fin +
694 skb->len - (th->doff << 2));
695 }
696
697 memset(&arg, 0, sizeof(arg));
698 arg.iov[0].iov_base = (unsigned char *)&rep;
699 arg.iov[0].iov_len = sizeof(rep.th);
700
701 net = sk ? sock_net(sk) : dev_net(skb_dst(skb)->dev);
702#ifdef CONFIG_TCP_MD5SIG
703 rcu_read_lock();
704 hash_location = tcp_parse_md5sig_option(th);
705 if (sk && sk_fullsock(sk)) {
706 key = tcp_md5_do_lookup(sk, (union tcp_md5_addr *)
707 &ip_hdr(skb)->saddr, AF_INET);
708 } else if (hash_location) {
709 /*
710 * active side is lost. Try to find listening socket through
711 * source port, and then find md5 key through listening socket.
712 * we are not loose security here:
713 * Incoming packet is checked with md5 hash with finding key,
714 * no RST generated if md5 hash doesn't match.
715 */
716 sk1 = __inet_lookup_listener(net, &tcp_hashinfo, NULL, 0,
717 ip_hdr(skb)->saddr,
718 th->source, ip_hdr(skb)->daddr,
719 ntohs(th->source), inet_iif(skb),
720 tcp_v4_sdif(skb));
721 /* don't send rst if it can't find key */
722 if (!sk1)
723 goto out;
724
725 key = tcp_md5_do_lookup(sk1, (union tcp_md5_addr *)
726 &ip_hdr(skb)->saddr, AF_INET);
727 if (!key)
728 goto out;
729
730
731 genhash = tcp_v4_md5_hash_skb(newhash, key, NULL, skb);
732 if (genhash || memcmp(hash_location, newhash, 16) != 0)
733 goto out;
734
735 }
736
737 if (key) {
738 rep.opt[0] = htonl((TCPOPT_NOP << 24) |
739 (TCPOPT_NOP << 16) |
740 (TCPOPT_MD5SIG << 8) |
741 TCPOLEN_MD5SIG);
742 /* Update length and the length the header thinks exists */
743 arg.iov[0].iov_len += TCPOLEN_MD5SIG_ALIGNED;
744 rep.th.doff = arg.iov[0].iov_len / 4;
745
746 tcp_v4_md5_hash_hdr((__u8 *) &rep.opt[1],
747 key, ip_hdr(skb)->saddr,
748 ip_hdr(skb)->daddr, &rep.th);
749 }
750#endif
751 arg.csum = csum_tcpudp_nofold(ip_hdr(skb)->daddr,
752 ip_hdr(skb)->saddr, /* XXX */
753 arg.iov[0].iov_len, IPPROTO_TCP, 0);
754 arg.csumoffset = offsetof(struct tcphdr, check) / 2;
755 arg.flags = (sk && inet_sk_transparent(sk)) ? IP_REPLY_ARG_NOSRCCHECK : 0;
756
757 /* When socket is gone, all binding information is lost.
758 * routing might fail in this case. No choice here, if we choose to force
759 * input interface, we will misroute in case of asymmetric route.
760 */
761 if (sk) {
762 arg.bound_dev_if = sk->sk_bound_dev_if;
763 if (sk_fullsock(sk))
764 trace_tcp_send_reset(sk, skb);
765 }
766
767 BUILD_BUG_ON(offsetof(struct sock, sk_bound_dev_if) !=
768 offsetof(struct inet_timewait_sock, tw_bound_dev_if));
769
770 arg.tos = ip_hdr(skb)->tos;
771 arg.uid = sock_net_uid(net, sk && sk_fullsock(sk) ? sk : NULL);
772 local_bh_disable();
773 ctl_sk = this_cpu_read(*net->ipv4.tcp_sk);
774 if (sk) {
775 ctl_sk->sk_mark = (sk->sk_state == TCP_TIME_WAIT) ?
776 inet_twsk(sk)->tw_mark : sk->sk_mark;
777 ctl_sk->sk_priority = (sk->sk_state == TCP_TIME_WAIT) ?
778 inet_twsk(sk)->tw_priority : sk->sk_priority;
779 transmit_time = tcp_transmit_time(sk);
780 }
781 ip_send_unicast_reply(ctl_sk,
782 skb, &TCP_SKB_CB(skb)->header.h4.opt,
783 ip_hdr(skb)->saddr, ip_hdr(skb)->daddr,
784 &arg, arg.iov[0].iov_len,
785 transmit_time);
786
787 ctl_sk->sk_mark = 0;
788 __TCP_INC_STATS(net, TCP_MIB_OUTSEGS);
789 __TCP_INC_STATS(net, TCP_MIB_OUTRSTS);
790 local_bh_enable();
791
792#ifdef CONFIG_TCP_MD5SIG
793out:
794 rcu_read_unlock();
795#endif
796}
797
798/* The code following below sending ACKs in SYN-RECV and TIME-WAIT states
799 outside socket context is ugly, certainly. What can I do?
800 */
801
802static void tcp_v4_send_ack(const struct sock *sk,
803 struct sk_buff *skb, u32 seq, u32 ack,
804 u32 win, u32 tsval, u32 tsecr, int oif,
805 struct tcp_md5sig_key *key,
806 int reply_flags, u8 tos)
807{
808 const struct tcphdr *th = tcp_hdr(skb);
809 struct {
810 struct tcphdr th;
811 __be32 opt[(TCPOLEN_TSTAMP_ALIGNED >> 2)
812#ifdef CONFIG_TCP_MD5SIG
813 + (TCPOLEN_MD5SIG_ALIGNED >> 2)
814#endif
815 ];
816 } rep;
817 struct net *net = sock_net(sk);
818 struct ip_reply_arg arg;
819 struct sock *ctl_sk;
820 u64 transmit_time;
821
822 memset(&rep.th, 0, sizeof(struct tcphdr));
823 memset(&arg, 0, sizeof(arg));
824
825 arg.iov[0].iov_base = (unsigned char *)&rep;
826 arg.iov[0].iov_len = sizeof(rep.th);
827 if (tsecr) {
828 rep.opt[0] = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
829 (TCPOPT_TIMESTAMP << 8) |
830 TCPOLEN_TIMESTAMP);
831 rep.opt[1] = htonl(tsval);
832 rep.opt[2] = htonl(tsecr);
833 arg.iov[0].iov_len += TCPOLEN_TSTAMP_ALIGNED;
834 }
835
836 /* Swap the send and the receive. */
837 rep.th.dest = th->source;
838 rep.th.source = th->dest;
839 rep.th.doff = arg.iov[0].iov_len / 4;
840 rep.th.seq = htonl(seq);
841 rep.th.ack_seq = htonl(ack);
842 rep.th.ack = 1;
843 rep.th.window = htons(win);
844
845#ifdef CONFIG_TCP_MD5SIG
846 if (key) {
847 int offset = (tsecr) ? 3 : 0;
848
849 rep.opt[offset++] = htonl((TCPOPT_NOP << 24) |
850 (TCPOPT_NOP << 16) |
851 (TCPOPT_MD5SIG << 8) |
852 TCPOLEN_MD5SIG);
853 arg.iov[0].iov_len += TCPOLEN_MD5SIG_ALIGNED;
854 rep.th.doff = arg.iov[0].iov_len/4;
855
856 tcp_v4_md5_hash_hdr((__u8 *) &rep.opt[offset],
857 key, ip_hdr(skb)->saddr,
858 ip_hdr(skb)->daddr, &rep.th);
859 }
860#endif
861 arg.flags = reply_flags;
862 arg.csum = csum_tcpudp_nofold(ip_hdr(skb)->daddr,
863 ip_hdr(skb)->saddr, /* XXX */
864 arg.iov[0].iov_len, IPPROTO_TCP, 0);
865 arg.csumoffset = offsetof(struct tcphdr, check) / 2;
866 if (oif)
867 arg.bound_dev_if = oif;
868 arg.tos = tos;
869 arg.uid = sock_net_uid(net, sk_fullsock(sk) ? sk : NULL);
870 local_bh_disable();
871 ctl_sk = this_cpu_read(*net->ipv4.tcp_sk);
872 ctl_sk->sk_mark = (sk->sk_state == TCP_TIME_WAIT) ?
873 inet_twsk(sk)->tw_mark : sk->sk_mark;
874 ctl_sk->sk_priority = (sk->sk_state == TCP_TIME_WAIT) ?
875 inet_twsk(sk)->tw_priority : sk->sk_priority;
876 transmit_time = tcp_transmit_time(sk);
877 ip_send_unicast_reply(ctl_sk,
878 skb, &TCP_SKB_CB(skb)->header.h4.opt,
879 ip_hdr(skb)->saddr, ip_hdr(skb)->daddr,
880 &arg, arg.iov[0].iov_len,
881 transmit_time);
882
883 ctl_sk->sk_mark = 0;
884 __TCP_INC_STATS(net, TCP_MIB_OUTSEGS);
885 local_bh_enable();
886}
887
888static void tcp_v4_timewait_ack(struct sock *sk, struct sk_buff *skb)
889{
890 struct inet_timewait_sock *tw = inet_twsk(sk);
891 struct tcp_timewait_sock *tcptw = tcp_twsk(sk);
892
893 tcp_v4_send_ack(sk, skb,
894 tcptw->tw_snd_nxt, tcptw->tw_rcv_nxt,
895 tcptw->tw_rcv_wnd >> tw->tw_rcv_wscale,
896 tcp_time_stamp_raw() + tcptw->tw_ts_offset,
897 tcptw->tw_ts_recent,
898 tw->tw_bound_dev_if,
899 tcp_twsk_md5_key(tcptw),
900 tw->tw_transparent ? IP_REPLY_ARG_NOSRCCHECK : 0,
901 tw->tw_tos
902 );
903
904 inet_twsk_put(tw);
905}
906
907static void tcp_v4_reqsk_send_ack(const struct sock *sk, struct sk_buff *skb,
908 struct request_sock *req)
909{
910 /* sk->sk_state == TCP_LISTEN -> for regular TCP_SYN_RECV
911 * sk->sk_state == TCP_SYN_RECV -> for Fast Open.
912 */
913 u32 seq = (sk->sk_state == TCP_LISTEN) ? tcp_rsk(req)->snt_isn + 1 :
914 tcp_sk(sk)->snd_nxt;
915
916 /* RFC 7323 2.3
917 * The window field (SEG.WND) of every outgoing segment, with the
918 * exception of <SYN> segments, MUST be right-shifted by
919 * Rcv.Wind.Shift bits:
920 */
921 tcp_v4_send_ack(sk, skb, seq,
922 tcp_rsk(req)->rcv_nxt,
923 req->rsk_rcv_wnd >> inet_rsk(req)->rcv_wscale,
924 tcp_time_stamp_raw() + tcp_rsk(req)->ts_off,
925 req->ts_recent,
926 0,
927 tcp_md5_do_lookup(sk, (union tcp_md5_addr *)&ip_hdr(skb)->saddr,
928 AF_INET),
929 inet_rsk(req)->no_srccheck ? IP_REPLY_ARG_NOSRCCHECK : 0,
930 ip_hdr(skb)->tos);
931}
932
933/*
934 * Send a SYN-ACK after having received a SYN.
935 * This still operates on a request_sock only, not on a big
936 * socket.
937 */
938static int tcp_v4_send_synack(const struct sock *sk, struct dst_entry *dst,
939 struct flowi *fl,
940 struct request_sock *req,
941 struct tcp_fastopen_cookie *foc,
942 enum tcp_synack_type synack_type)
943{
944 const struct inet_request_sock *ireq = inet_rsk(req);
945 struct flowi4 fl4;
946 int err = -1;
947 struct sk_buff *skb;
948
949 /* First, grab a route. */
950 if (!dst && (dst = inet_csk_route_req(sk, &fl4, req)) == NULL)
951 return -1;
952
953 skb = tcp_make_synack(sk, dst, req, foc, synack_type);
954
955 if (skb) {
956 __tcp_v4_send_check(skb, ireq->ir_loc_addr, ireq->ir_rmt_addr);
957
958 rcu_read_lock();
959 err = ip_build_and_send_pkt(skb, sk, ireq->ir_loc_addr,
960 ireq->ir_rmt_addr,
961 rcu_dereference(ireq->ireq_opt));
962 rcu_read_unlock();
963 err = net_xmit_eval(err);
964 }
965
966 return err;
967}
968
969/*
970 * IPv4 request_sock destructor.
971 */
972static void tcp_v4_reqsk_destructor(struct request_sock *req)
973{
974 kfree(rcu_dereference_protected(inet_rsk(req)->ireq_opt, 1));
975}
976
977#ifdef CONFIG_TCP_MD5SIG
978/*
979 * RFC2385 MD5 checksumming requires a mapping of
980 * IP address->MD5 Key.
981 * We need to maintain these in the sk structure.
982 */
983
984DEFINE_STATIC_KEY_FALSE(tcp_md5_needed);
985EXPORT_SYMBOL(tcp_md5_needed);
986
987/* Find the Key structure for an address. */
988struct tcp_md5sig_key *__tcp_md5_do_lookup(const struct sock *sk,
989 const union tcp_md5_addr *addr,
990 int family)
991{
992 const struct tcp_sock *tp = tcp_sk(sk);
993 struct tcp_md5sig_key *key;
994 const struct tcp_md5sig_info *md5sig;
995 __be32 mask;
996 struct tcp_md5sig_key *best_match = NULL;
997 bool match;
998
999 /* caller either holds rcu_read_lock() or socket lock */
1000 md5sig = rcu_dereference_check(tp->md5sig_info,
1001 lockdep_sock_is_held(sk));
1002 if (!md5sig)
1003 return NULL;
1004
1005 hlist_for_each_entry_rcu(key, &md5sig->head, node) {
1006 if (key->family != family)
1007 continue;
1008
1009 if (family == AF_INET) {
1010 mask = inet_make_mask(key->prefixlen);
1011 match = (key->addr.a4.s_addr & mask) ==
1012 (addr->a4.s_addr & mask);
1013#if IS_ENABLED(CONFIG_IPV6)
1014 } else if (family == AF_INET6) {
1015 match = ipv6_prefix_equal(&key->addr.a6, &addr->a6,
1016 key->prefixlen);
1017#endif
1018 } else {
1019 match = false;
1020 }
1021
1022 if (match && (!best_match ||
1023 key->prefixlen > best_match->prefixlen))
1024 best_match = key;
1025 }
1026 return best_match;
1027}
1028EXPORT_SYMBOL(__tcp_md5_do_lookup);
1029
1030static struct tcp_md5sig_key *tcp_md5_do_lookup_exact(const struct sock *sk,
1031 const union tcp_md5_addr *addr,
1032 int family, u8 prefixlen)
1033{
1034 const struct tcp_sock *tp = tcp_sk(sk);
1035 struct tcp_md5sig_key *key;
1036 unsigned int size = sizeof(struct in_addr);
1037 const struct tcp_md5sig_info *md5sig;
1038
1039 /* caller either holds rcu_read_lock() or socket lock */
1040 md5sig = rcu_dereference_check(tp->md5sig_info,
1041 lockdep_sock_is_held(sk));
1042 if (!md5sig)
1043 return NULL;
1044#if IS_ENABLED(CONFIG_IPV6)
1045 if (family == AF_INET6)
1046 size = sizeof(struct in6_addr);
1047#endif
1048 hlist_for_each_entry_rcu(key, &md5sig->head, node) {
1049 if (key->family != family)
1050 continue;
1051 if (!memcmp(&key->addr, addr, size) &&
1052 key->prefixlen == prefixlen)
1053 return key;
1054 }
1055 return NULL;
1056}
1057
1058struct tcp_md5sig_key *tcp_v4_md5_lookup(const struct sock *sk,
1059 const struct sock *addr_sk)
1060{
1061 const union tcp_md5_addr *addr;
1062
1063 addr = (const union tcp_md5_addr *)&addr_sk->sk_daddr;
1064 return tcp_md5_do_lookup(sk, addr, AF_INET);
1065}
1066EXPORT_SYMBOL(tcp_v4_md5_lookup);
1067
1068/* This can be called on a newly created socket, from other files */
1069int tcp_md5_do_add(struct sock *sk, const union tcp_md5_addr *addr,
1070 int family, u8 prefixlen, const u8 *newkey, u8 newkeylen,
1071 gfp_t gfp)
1072{
1073 /* Add Key to the list */
1074 struct tcp_md5sig_key *key;
1075 struct tcp_sock *tp = tcp_sk(sk);
1076 struct tcp_md5sig_info *md5sig;
1077
1078 key = tcp_md5_do_lookup_exact(sk, addr, family, prefixlen);
1079 if (key) {
1080 /* Pre-existing entry - just update that one. */
1081 memcpy(key->key, newkey, newkeylen);
1082 key->keylen = newkeylen;
1083 return 0;
1084 }
1085
1086 md5sig = rcu_dereference_protected(tp->md5sig_info,
1087 lockdep_sock_is_held(sk));
1088 if (!md5sig) {
1089 md5sig = kmalloc(sizeof(*md5sig), gfp);
1090 if (!md5sig)
1091 return -ENOMEM;
1092
1093 sk_nocaps_add(sk, NETIF_F_GSO_MASK);
1094 INIT_HLIST_HEAD(&md5sig->head);
1095 rcu_assign_pointer(tp->md5sig_info, md5sig);
1096 }
1097
1098 key = sock_kmalloc(sk, sizeof(*key), gfp);
1099 if (!key)
1100 return -ENOMEM;
1101 if (!tcp_alloc_md5sig_pool()) {
1102 sock_kfree_s(sk, key, sizeof(*key));
1103 return -ENOMEM;
1104 }
1105
1106 memcpy(key->key, newkey, newkeylen);
1107 key->keylen = newkeylen;
1108 key->family = family;
1109 key->prefixlen = prefixlen;
1110 memcpy(&key->addr, addr,
1111 (family == AF_INET6) ? sizeof(struct in6_addr) :
1112 sizeof(struct in_addr));
1113 hlist_add_head_rcu(&key->node, &md5sig->head);
1114 return 0;
1115}
1116EXPORT_SYMBOL(tcp_md5_do_add);
1117
1118int tcp_md5_do_del(struct sock *sk, const union tcp_md5_addr *addr, int family,
1119 u8 prefixlen)
1120{
1121 struct tcp_md5sig_key *key;
1122
1123 key = tcp_md5_do_lookup_exact(sk, addr, family, prefixlen);
1124 if (!key)
1125 return -ENOENT;
1126 hlist_del_rcu(&key->node);
1127 atomic_sub(sizeof(*key), &sk->sk_omem_alloc);
1128 kfree_rcu(key, rcu);
1129 return 0;
1130}
1131EXPORT_SYMBOL(tcp_md5_do_del);
1132
1133static void tcp_clear_md5_list(struct sock *sk)
1134{
1135 struct tcp_sock *tp = tcp_sk(sk);
1136 struct tcp_md5sig_key *key;
1137 struct hlist_node *n;
1138 struct tcp_md5sig_info *md5sig;
1139
1140 md5sig = rcu_dereference_protected(tp->md5sig_info, 1);
1141
1142 hlist_for_each_entry_safe(key, n, &md5sig->head, node) {
1143 hlist_del_rcu(&key->node);
1144 atomic_sub(sizeof(*key), &sk->sk_omem_alloc);
1145 kfree_rcu(key, rcu);
1146 }
1147}
1148
1149static int tcp_v4_parse_md5_keys(struct sock *sk, int optname,
1150 char __user *optval, int optlen)
1151{
1152 struct tcp_md5sig cmd;
1153 struct sockaddr_in *sin = (struct sockaddr_in *)&cmd.tcpm_addr;
1154 u8 prefixlen = 32;
1155
1156 if (optlen < sizeof(cmd))
1157 return -EINVAL;
1158
1159 if (copy_from_user(&cmd, optval, sizeof(cmd)))
1160 return -EFAULT;
1161
1162 if (sin->sin_family != AF_INET)
1163 return -EINVAL;
1164
1165 if (optname == TCP_MD5SIG_EXT &&
1166 cmd.tcpm_flags & TCP_MD5SIG_FLAG_PREFIX) {
1167 prefixlen = cmd.tcpm_prefixlen;
1168 if (prefixlen > 32)
1169 return -EINVAL;
1170 }
1171
1172 if (!cmd.tcpm_keylen)
1173 return tcp_md5_do_del(sk, (union tcp_md5_addr *)&sin->sin_addr.s_addr,
1174 AF_INET, prefixlen);
1175
1176 if (cmd.tcpm_keylen > TCP_MD5SIG_MAXKEYLEN)
1177 return -EINVAL;
1178
1179 return tcp_md5_do_add(sk, (union tcp_md5_addr *)&sin->sin_addr.s_addr,
1180 AF_INET, prefixlen, cmd.tcpm_key, cmd.tcpm_keylen,
1181 GFP_KERNEL);
1182}
1183
1184static int tcp_v4_md5_hash_headers(struct tcp_md5sig_pool *hp,
1185 __be32 daddr, __be32 saddr,
1186 const struct tcphdr *th, int nbytes)
1187{
1188 struct tcp4_pseudohdr *bp;
1189 struct scatterlist sg;
1190 struct tcphdr *_th;
1191
1192 bp = hp->scratch;
1193 bp->saddr = saddr;
1194 bp->daddr = daddr;
1195 bp->pad = 0;
1196 bp->protocol = IPPROTO_TCP;
1197 bp->len = cpu_to_be16(nbytes);
1198
1199 _th = (struct tcphdr *)(bp + 1);
1200 memcpy(_th, th, sizeof(*th));
1201 _th->check = 0;
1202
1203 sg_init_one(&sg, bp, sizeof(*bp) + sizeof(*th));
1204 ahash_request_set_crypt(hp->md5_req, &sg, NULL,
1205 sizeof(*bp) + sizeof(*th));
1206 return crypto_ahash_update(hp->md5_req);
1207}
1208
1209static int tcp_v4_md5_hash_hdr(char *md5_hash, const struct tcp_md5sig_key *key,
1210 __be32 daddr, __be32 saddr, const struct tcphdr *th)
1211{
1212 struct tcp_md5sig_pool *hp;
1213 struct ahash_request *req;
1214
1215 hp = tcp_get_md5sig_pool();
1216 if (!hp)
1217 goto clear_hash_noput;
1218 req = hp->md5_req;
1219
1220 if (crypto_ahash_init(req))
1221 goto clear_hash;
1222 if (tcp_v4_md5_hash_headers(hp, daddr, saddr, th, th->doff << 2))
1223 goto clear_hash;
1224 if (tcp_md5_hash_key(hp, key))
1225 goto clear_hash;
1226 ahash_request_set_crypt(req, NULL, md5_hash, 0);
1227 if (crypto_ahash_final(req))
1228 goto clear_hash;
1229
1230 tcp_put_md5sig_pool();
1231 return 0;
1232
1233clear_hash:
1234 tcp_put_md5sig_pool();
1235clear_hash_noput:
1236 memset(md5_hash, 0, 16);
1237 return 1;
1238}
1239
1240int tcp_v4_md5_hash_skb(char *md5_hash, const struct tcp_md5sig_key *key,
1241 const struct sock *sk,
1242 const struct sk_buff *skb)
1243{
1244 struct tcp_md5sig_pool *hp;
1245 struct ahash_request *req;
1246 const struct tcphdr *th = tcp_hdr(skb);
1247 __be32 saddr, daddr;
1248
1249 if (sk) { /* valid for establish/request sockets */
1250 saddr = sk->sk_rcv_saddr;
1251 daddr = sk->sk_daddr;
1252 } else {
1253 const struct iphdr *iph = ip_hdr(skb);
1254 saddr = iph->saddr;
1255 daddr = iph->daddr;
1256 }
1257
1258 hp = tcp_get_md5sig_pool();
1259 if (!hp)
1260 goto clear_hash_noput;
1261 req = hp->md5_req;
1262
1263 if (crypto_ahash_init(req))
1264 goto clear_hash;
1265
1266 if (tcp_v4_md5_hash_headers(hp, daddr, saddr, th, skb->len))
1267 goto clear_hash;
1268 if (tcp_md5_hash_skb_data(hp, skb, th->doff << 2))
1269 goto clear_hash;
1270 if (tcp_md5_hash_key(hp, key))
1271 goto clear_hash;
1272 ahash_request_set_crypt(req, NULL, md5_hash, 0);
1273 if (crypto_ahash_final(req))
1274 goto clear_hash;
1275
1276 tcp_put_md5sig_pool();
1277 return 0;
1278
1279clear_hash:
1280 tcp_put_md5sig_pool();
1281clear_hash_noput:
1282 memset(md5_hash, 0, 16);
1283 return 1;
1284}
1285EXPORT_SYMBOL(tcp_v4_md5_hash_skb);
1286
1287#endif
1288
1289/* Called with rcu_read_lock() */
1290static bool tcp_v4_inbound_md5_hash(const struct sock *sk,
1291 const struct sk_buff *skb)
1292{
1293#ifdef CONFIG_TCP_MD5SIG
1294 /*
1295 * This gets called for each TCP segment that arrives
1296 * so we want to be efficient.
1297 * We have 3 drop cases:
1298 * o No MD5 hash and one expected.
1299 * o MD5 hash and we're not expecting one.
1300 * o MD5 hash and its wrong.
1301 */
1302 const __u8 *hash_location = NULL;
1303 struct tcp_md5sig_key *hash_expected;
1304 const struct iphdr *iph = ip_hdr(skb);
1305 const struct tcphdr *th = tcp_hdr(skb);
1306 int genhash;
1307 unsigned char newhash[16];
1308
1309 hash_expected = tcp_md5_do_lookup(sk, (union tcp_md5_addr *)&iph->saddr,
1310 AF_INET);
1311 hash_location = tcp_parse_md5sig_option(th);
1312
1313 /* We've parsed the options - do we have a hash? */
1314 if (!hash_expected && !hash_location)
1315 return false;
1316
1317 if (hash_expected && !hash_location) {
1318 NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPMD5NOTFOUND);
1319 return true;
1320 }
1321
1322 if (!hash_expected && hash_location) {
1323 NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPMD5UNEXPECTED);
1324 return true;
1325 }
1326
1327 /* Okay, so this is hash_expected and hash_location -
1328 * so we need to calculate the checksum.
1329 */
1330 genhash = tcp_v4_md5_hash_skb(newhash,
1331 hash_expected,
1332 NULL, skb);
1333
1334 if (genhash || memcmp(hash_location, newhash, 16) != 0) {
1335 NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPMD5FAILURE);
1336 net_info_ratelimited("MD5 Hash failed for (%pI4, %d)->(%pI4, %d)%s\n",
1337 &iph->saddr, ntohs(th->source),
1338 &iph->daddr, ntohs(th->dest),
1339 genhash ? " tcp_v4_calc_md5_hash failed"
1340 : "");
1341 return true;
1342 }
1343 return false;
1344#endif
1345 return false;
1346}
1347
1348static void tcp_v4_init_req(struct request_sock *req,
1349 const struct sock *sk_listener,
1350 struct sk_buff *skb)
1351{
1352 struct inet_request_sock *ireq = inet_rsk(req);
1353 struct net *net = sock_net(sk_listener);
1354
1355 sk_rcv_saddr_set(req_to_sk(req), ip_hdr(skb)->daddr);
1356 sk_daddr_set(req_to_sk(req), ip_hdr(skb)->saddr);
1357 RCU_INIT_POINTER(ireq->ireq_opt, tcp_v4_save_options(net, skb));
1358}
1359
1360static struct dst_entry *tcp_v4_route_req(const struct sock *sk,
1361 struct flowi *fl,
1362 const struct request_sock *req)
1363{
1364 return inet_csk_route_req(sk, &fl->u.ip4, req);
1365}
1366
1367struct request_sock_ops tcp_request_sock_ops __read_mostly = {
1368 .family = PF_INET,
1369 .obj_size = sizeof(struct tcp_request_sock),
1370 .rtx_syn_ack = tcp_rtx_synack,
1371 .send_ack = tcp_v4_reqsk_send_ack,
1372 .destructor = tcp_v4_reqsk_destructor,
1373 .send_reset = tcp_v4_send_reset,
1374 .syn_ack_timeout = tcp_syn_ack_timeout,
1375};
1376
1377static const struct tcp_request_sock_ops tcp_request_sock_ipv4_ops = {
1378 .mss_clamp = TCP_MSS_DEFAULT,
1379#ifdef CONFIG_TCP_MD5SIG
1380 .req_md5_lookup = tcp_v4_md5_lookup,
1381 .calc_md5_hash = tcp_v4_md5_hash_skb,
1382#endif
1383 .init_req = tcp_v4_init_req,
1384#ifdef CONFIG_SYN_COOKIES
1385 .cookie_init_seq = cookie_v4_init_sequence,
1386#endif
1387 .route_req = tcp_v4_route_req,
1388 .init_seq = tcp_v4_init_seq,
1389 .init_ts_off = tcp_v4_init_ts_off,
1390 .send_synack = tcp_v4_send_synack,
1391};
1392
1393int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb)
1394{
1395 /* Never answer to SYNs send to broadcast or multicast */
1396 if (skb_rtable(skb)->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))
1397 goto drop;
1398
1399 return tcp_conn_request(&tcp_request_sock_ops,
1400 &tcp_request_sock_ipv4_ops, sk, skb);
1401
1402drop:
1403 tcp_listendrop(sk);
1404 return 0;
1405}
1406EXPORT_SYMBOL(tcp_v4_conn_request);
1407
1408
1409/*
1410 * The three way handshake has completed - we got a valid synack -
1411 * now create the new socket.
1412 */
1413struct sock *tcp_v4_syn_recv_sock(const struct sock *sk, struct sk_buff *skb,
1414 struct request_sock *req,
1415 struct dst_entry *dst,
1416 struct request_sock *req_unhash,
1417 bool *own_req)
1418{
1419 struct inet_request_sock *ireq;
1420 struct inet_sock *newinet;
1421 struct tcp_sock *newtp;
1422 struct sock *newsk;
1423#ifdef CONFIG_TCP_MD5SIG
1424 struct tcp_md5sig_key *key;
1425#endif
1426 struct ip_options_rcu *inet_opt;
1427
1428 if (sk_acceptq_is_full(sk))
1429 goto exit_overflow;
1430
1431 newsk = tcp_create_openreq_child(sk, req, skb);
1432 if (!newsk)
1433 goto exit_nonewsk;
1434
1435 newsk->sk_gso_type = SKB_GSO_TCPV4;
1436 inet_sk_rx_dst_set(newsk, skb);
1437
1438 newtp = tcp_sk(newsk);
1439 newinet = inet_sk(newsk);
1440 ireq = inet_rsk(req);
1441 sk_daddr_set(newsk, ireq->ir_rmt_addr);
1442 sk_rcv_saddr_set(newsk, ireq->ir_loc_addr);
1443 newsk->sk_bound_dev_if = ireq->ir_iif;
1444 newinet->inet_saddr = ireq->ir_loc_addr;
1445 inet_opt = rcu_dereference(ireq->ireq_opt);
1446 RCU_INIT_POINTER(newinet->inet_opt, inet_opt);
1447 newinet->mc_index = inet_iif(skb);
1448 newinet->mc_ttl = ip_hdr(skb)->ttl;
1449 newinet->rcv_tos = ip_hdr(skb)->tos;
1450 inet_csk(newsk)->icsk_ext_hdr_len = 0;
1451 if (inet_opt)
1452 inet_csk(newsk)->icsk_ext_hdr_len = inet_opt->opt.optlen;
1453 newinet->inet_id = prandom_u32();
1454
1455 if (!dst) {
1456 dst = inet_csk_route_child_sock(sk, newsk, req);
1457 if (!dst)
1458 goto put_and_exit;
1459 } else {
1460 /* syncookie case : see end of cookie_v4_check() */
1461 }
1462 sk_setup_caps(newsk, dst);
1463
1464 tcp_ca_openreq_child(newsk, dst);
1465
1466 tcp_sync_mss(newsk, dst_mtu(dst));
1467 newtp->advmss = tcp_mss_clamp(tcp_sk(sk), dst_metric_advmss(dst));
1468
1469 tcp_initialize_rcv_mss(newsk);
1470
1471#ifdef CONFIG_TCP_MD5SIG
1472 /* Copy over the MD5 key from the original socket */
1473 key = tcp_md5_do_lookup(sk, (union tcp_md5_addr *)&newinet->inet_daddr,
1474 AF_INET);
1475 if (key) {
1476 /*
1477 * We're using one, so create a matching key
1478 * on the newsk structure. If we fail to get
1479 * memory, then we end up not copying the key
1480 * across. Shucks.
1481 */
1482 tcp_md5_do_add(newsk, (union tcp_md5_addr *)&newinet->inet_daddr,
1483 AF_INET, 32, key->key, key->keylen, GFP_ATOMIC);
1484 sk_nocaps_add(newsk, NETIF_F_GSO_MASK);
1485 }
1486#endif
1487
1488 if (__inet_inherit_port(sk, newsk) < 0)
1489 goto put_and_exit;
1490 *own_req = inet_ehash_nolisten(newsk, req_to_sk(req_unhash));
1491 if (likely(*own_req)) {
1492 tcp_move_syn(newtp, req);
1493 ireq->ireq_opt = NULL;
1494 } else {
1495 newinet->inet_opt = NULL;
1496 }
1497 return newsk;
1498
1499exit_overflow:
1500 NET_INC_STATS(sock_net(sk), LINUX_MIB_LISTENOVERFLOWS);
1501exit_nonewsk:
1502 dst_release(dst);
1503exit:
1504 tcp_listendrop(sk);
1505 return NULL;
1506put_and_exit:
1507 newinet->inet_opt = NULL;
1508 inet_csk_prepare_forced_close(newsk);
1509 tcp_done(newsk);
1510 goto exit;
1511}
1512EXPORT_SYMBOL(tcp_v4_syn_recv_sock);
1513
1514static struct sock *tcp_v4_cookie_check(struct sock *sk, struct sk_buff *skb)
1515{
1516#ifdef CONFIG_SYN_COOKIES
1517 const struct tcphdr *th = tcp_hdr(skb);
1518
1519 if (!th->syn)
1520 sk = cookie_v4_check(sk, skb);
1521#endif
1522 return sk;
1523}
1524
1525u16 tcp_v4_get_syncookie(struct sock *sk, struct iphdr *iph,
1526 struct tcphdr *th, u32 *cookie)
1527{
1528 u16 mss = 0;
1529#ifdef CONFIG_SYN_COOKIES
1530 mss = tcp_get_syncookie_mss(&tcp_request_sock_ops,
1531 &tcp_request_sock_ipv4_ops, sk, th);
1532 if (mss) {
1533 *cookie = __cookie_v4_init_sequence(iph, th, &mss);
1534 tcp_synq_overflow(sk);
1535 }
1536#endif
1537 return mss;
1538}
1539
1540/* The socket must have it's spinlock held when we get
1541 * here, unless it is a TCP_LISTEN socket.
1542 *
1543 * We have a potential double-lock case here, so even when
1544 * doing backlog processing we use the BH locking scheme.
1545 * This is because we cannot sleep with the original spinlock
1546 * held.
1547 */
1548int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb)
1549{
1550 struct sock *rsk;
1551
1552 if (sk->sk_state == TCP_ESTABLISHED) { /* Fast path */
1553 struct dst_entry *dst = sk->sk_rx_dst;
1554
1555 sock_rps_save_rxhash(sk, skb);
1556 sk_mark_napi_id(sk, skb);
1557 if (dst) {
1558 if (inet_sk(sk)->rx_dst_ifindex != skb->skb_iif ||
1559 !dst->ops->check(dst, 0)) {
1560 dst_release(dst);
1561 sk->sk_rx_dst = NULL;
1562 }
1563 }
1564 tcp_rcv_established(sk, skb);
1565 return 0;
1566 }
1567
1568 if (tcp_checksum_complete(skb))
1569 goto csum_err;
1570
1571 if (sk->sk_state == TCP_LISTEN) {
1572 struct sock *nsk = tcp_v4_cookie_check(sk, skb);
1573
1574 if (!nsk)
1575 goto discard;
1576 if (nsk != sk) {
1577 if (tcp_child_process(sk, nsk, skb)) {
1578 rsk = nsk;
1579 goto reset;
1580 }
1581 return 0;
1582 }
1583 } else
1584 sock_rps_save_rxhash(sk, skb);
1585
1586 if (tcp_rcv_state_process(sk, skb)) {
1587 rsk = sk;
1588 goto reset;
1589 }
1590 return 0;
1591
1592reset:
1593 tcp_v4_send_reset(rsk, skb);
1594discard:
1595 kfree_skb(skb);
1596 /* Be careful here. If this function gets more complicated and
1597 * gcc suffers from register pressure on the x86, sk (in %ebx)
1598 * might be destroyed here. This current version compiles correctly,
1599 * but you have been warned.
1600 */
1601 return 0;
1602
1603csum_err:
1604 TCP_INC_STATS(sock_net(sk), TCP_MIB_CSUMERRORS);
1605 TCP_INC_STATS(sock_net(sk), TCP_MIB_INERRS);
1606 goto discard;
1607}
1608EXPORT_SYMBOL(tcp_v4_do_rcv);
1609
1610int tcp_v4_early_demux(struct sk_buff *skb)
1611{
1612 const struct iphdr *iph;
1613 const struct tcphdr *th;
1614 struct sock *sk;
1615
1616 if (skb->pkt_type != PACKET_HOST)
1617 return 0;
1618
1619 if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct tcphdr)))
1620 return 0;
1621
1622 iph = ip_hdr(skb);
1623 th = tcp_hdr(skb);
1624
1625 if (th->doff < sizeof(struct tcphdr) / 4)
1626 return 0;
1627
1628 sk = __inet_lookup_established(dev_net(skb->dev), &tcp_hashinfo,
1629 iph->saddr, th->source,
1630 iph->daddr, ntohs(th->dest),
1631 skb->skb_iif, inet_sdif(skb));
1632 if (sk) {
1633 skb->sk = sk;
1634 skb->destructor = sock_edemux;
1635 if (sk_fullsock(sk)) {
1636 struct dst_entry *dst = READ_ONCE(sk->sk_rx_dst);
1637
1638 if (dst)
1639 dst = dst_check(dst, 0);
1640 if (dst &&
1641 inet_sk(sk)->rx_dst_ifindex == skb->skb_iif)
1642 skb_dst_set_noref(skb, dst);
1643 }
1644 }
1645 return 0;
1646}
1647
1648bool tcp_add_backlog(struct sock *sk, struct sk_buff *skb)
1649{
1650 u32 limit = READ_ONCE(sk->sk_rcvbuf) + READ_ONCE(sk->sk_sndbuf);
1651 struct skb_shared_info *shinfo;
1652 const struct tcphdr *th;
1653 struct tcphdr *thtail;
1654 struct sk_buff *tail;
1655 unsigned int hdrlen;
1656 bool fragstolen;
1657 u32 gso_segs;
1658 int delta;
1659
1660 /* In case all data was pulled from skb frags (in __pskb_pull_tail()),
1661 * we can fix skb->truesize to its real value to avoid future drops.
1662 * This is valid because skb is not yet charged to the socket.
1663 * It has been noticed pure SACK packets were sometimes dropped
1664 * (if cooked by drivers without copybreak feature).
1665 */
1666 skb_condense(skb);
1667
1668 skb_dst_drop(skb);
1669
1670 if (unlikely(tcp_checksum_complete(skb))) {
1671 bh_unlock_sock(sk);
1672 __TCP_INC_STATS(sock_net(sk), TCP_MIB_CSUMERRORS);
1673 __TCP_INC_STATS(sock_net(sk), TCP_MIB_INERRS);
1674 return true;
1675 }
1676
1677 /* Attempt coalescing to last skb in backlog, even if we are
1678 * above the limits.
1679 * This is okay because skb capacity is limited to MAX_SKB_FRAGS.
1680 */
1681 th = (const struct tcphdr *)skb->data;
1682 hdrlen = th->doff * 4;
1683 shinfo = skb_shinfo(skb);
1684
1685 if (!shinfo->gso_size)
1686 shinfo->gso_size = skb->len - hdrlen;
1687
1688 if (!shinfo->gso_segs)
1689 shinfo->gso_segs = 1;
1690
1691 tail = sk->sk_backlog.tail;
1692 if (!tail)
1693 goto no_coalesce;
1694 thtail = (struct tcphdr *)tail->data;
1695
1696 if (TCP_SKB_CB(tail)->end_seq != TCP_SKB_CB(skb)->seq ||
1697 TCP_SKB_CB(tail)->ip_dsfield != TCP_SKB_CB(skb)->ip_dsfield ||
1698 ((TCP_SKB_CB(tail)->tcp_flags |
1699 TCP_SKB_CB(skb)->tcp_flags) & (TCPHDR_SYN | TCPHDR_RST | TCPHDR_URG)) ||
1700 !((TCP_SKB_CB(tail)->tcp_flags &
1701 TCP_SKB_CB(skb)->tcp_flags) & TCPHDR_ACK) ||
1702 ((TCP_SKB_CB(tail)->tcp_flags ^
1703 TCP_SKB_CB(skb)->tcp_flags) & (TCPHDR_ECE | TCPHDR_CWR)) ||
1704#ifdef CONFIG_TLS_DEVICE
1705 tail->decrypted != skb->decrypted ||
1706#endif
1707 thtail->doff != th->doff ||
1708 memcmp(thtail + 1, th + 1, hdrlen - sizeof(*th)))
1709 goto no_coalesce;
1710
1711 __skb_pull(skb, hdrlen);
1712 if (skb_try_coalesce(tail, skb, &fragstolen, &delta)) {
1713 thtail->window = th->window;
1714
1715 TCP_SKB_CB(tail)->end_seq = TCP_SKB_CB(skb)->end_seq;
1716
1717 if (after(TCP_SKB_CB(skb)->ack_seq, TCP_SKB_CB(tail)->ack_seq))
1718 TCP_SKB_CB(tail)->ack_seq = TCP_SKB_CB(skb)->ack_seq;
1719
1720 /* We have to update both TCP_SKB_CB(tail)->tcp_flags and
1721 * thtail->fin, so that the fast path in tcp_rcv_established()
1722 * is not entered if we append a packet with a FIN.
1723 * SYN, RST, URG are not present.
1724 * ACK is set on both packets.
1725 * PSH : we do not really care in TCP stack,
1726 * at least for 'GRO' packets.
1727 */
1728 thtail->fin |= th->fin;
1729 TCP_SKB_CB(tail)->tcp_flags |= TCP_SKB_CB(skb)->tcp_flags;
1730
1731 if (TCP_SKB_CB(skb)->has_rxtstamp) {
1732 TCP_SKB_CB(tail)->has_rxtstamp = true;
1733 tail->tstamp = skb->tstamp;
1734 skb_hwtstamps(tail)->hwtstamp = skb_hwtstamps(skb)->hwtstamp;
1735 }
1736
1737 /* Not as strict as GRO. We only need to carry mss max value */
1738 skb_shinfo(tail)->gso_size = max(shinfo->gso_size,
1739 skb_shinfo(tail)->gso_size);
1740
1741 gso_segs = skb_shinfo(tail)->gso_segs + shinfo->gso_segs;
1742 skb_shinfo(tail)->gso_segs = min_t(u32, gso_segs, 0xFFFF);
1743
1744 sk->sk_backlog.len += delta;
1745 __NET_INC_STATS(sock_net(sk),
1746 LINUX_MIB_TCPBACKLOGCOALESCE);
1747 kfree_skb_partial(skb, fragstolen);
1748 return false;
1749 }
1750 __skb_push(skb, hdrlen);
1751
1752no_coalesce:
1753 /* Only socket owner can try to collapse/prune rx queues
1754 * to reduce memory overhead, so add a little headroom here.
1755 * Few sockets backlog are possibly concurrently non empty.
1756 */
1757 limit += 64*1024;
1758
1759 if (unlikely(sk_add_backlog(sk, skb, limit))) {
1760 bh_unlock_sock(sk);
1761 __NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPBACKLOGDROP);
1762 return true;
1763 }
1764 return false;
1765}
1766EXPORT_SYMBOL(tcp_add_backlog);
1767
1768int tcp_filter(struct sock *sk, struct sk_buff *skb)
1769{
1770 struct tcphdr *th = (struct tcphdr *)skb->data;
1771
1772 return sk_filter_trim_cap(sk, skb, th->doff * 4);
1773}
1774EXPORT_SYMBOL(tcp_filter);
1775
1776static void tcp_v4_restore_cb(struct sk_buff *skb)
1777{
1778 memmove(IPCB(skb), &TCP_SKB_CB(skb)->header.h4,
1779 sizeof(struct inet_skb_parm));
1780}
1781
1782static void tcp_v4_fill_cb(struct sk_buff *skb, const struct iphdr *iph,
1783 const struct tcphdr *th)
1784{
1785 /* This is tricky : We move IPCB at its correct location into TCP_SKB_CB()
1786 * barrier() makes sure compiler wont play fool^Waliasing games.
1787 */
1788 memmove(&TCP_SKB_CB(skb)->header.h4, IPCB(skb),
1789 sizeof(struct inet_skb_parm));
1790 barrier();
1791
1792 TCP_SKB_CB(skb)->seq = ntohl(th->seq);
1793 TCP_SKB_CB(skb)->end_seq = (TCP_SKB_CB(skb)->seq + th->syn + th->fin +
1794 skb->len - th->doff * 4);
1795 TCP_SKB_CB(skb)->ack_seq = ntohl(th->ack_seq);
1796 TCP_SKB_CB(skb)->tcp_flags = tcp_flag_byte(th);
1797 TCP_SKB_CB(skb)->tcp_tw_isn = 0;
1798 TCP_SKB_CB(skb)->ip_dsfield = ipv4_get_dsfield(iph);
1799 TCP_SKB_CB(skb)->sacked = 0;
1800 TCP_SKB_CB(skb)->has_rxtstamp =
1801 skb->tstamp || skb_hwtstamps(skb)->hwtstamp;
1802}
1803
1804/*
1805 * From tcp_input.c
1806 */
1807
1808int tcp_v4_rcv(struct sk_buff *skb)
1809{
1810 struct net *net = dev_net(skb->dev);
1811 struct sk_buff *skb_to_free;
1812 int sdif = inet_sdif(skb);
1813 const struct iphdr *iph;
1814 const struct tcphdr *th;
1815 bool refcounted;
1816 struct sock *sk;
1817 int ret;
1818
1819 if (skb->pkt_type != PACKET_HOST)
1820 goto discard_it;
1821
1822 /* Count it even if it's bad */
1823 __TCP_INC_STATS(net, TCP_MIB_INSEGS);
1824
1825 if (!pskb_may_pull(skb, sizeof(struct tcphdr)))
1826 goto discard_it;
1827
1828 th = (const struct tcphdr *)skb->data;
1829
1830 if (unlikely(th->doff < sizeof(struct tcphdr) / 4))
1831 goto bad_packet;
1832 if (!pskb_may_pull(skb, th->doff * 4))
1833 goto discard_it;
1834
1835 /* An explanation is required here, I think.
1836 * Packet length and doff are validated by header prediction,
1837 * provided case of th->doff==0 is eliminated.
1838 * So, we defer the checks. */
1839
1840 if (skb_checksum_init(skb, IPPROTO_TCP, inet_compute_pseudo))
1841 goto csum_error;
1842
1843 th = (const struct tcphdr *)skb->data;
1844 iph = ip_hdr(skb);
1845lookup:
1846 sk = __inet_lookup_skb(&tcp_hashinfo, skb, __tcp_hdrlen(th), th->source,
1847 th->dest, sdif, &refcounted);
1848 if (!sk)
1849 goto no_tcp_socket;
1850
1851process:
1852 if (sk->sk_state == TCP_TIME_WAIT)
1853 goto do_time_wait;
1854
1855 if (sk->sk_state == TCP_NEW_SYN_RECV) {
1856 struct request_sock *req = inet_reqsk(sk);
1857 bool req_stolen = false;
1858 struct sock *nsk;
1859
1860 sk = req->rsk_listener;
1861 if (unlikely(tcp_v4_inbound_md5_hash(sk, skb))) {
1862 sk_drops_add(sk, skb);
1863 reqsk_put(req);
1864 goto discard_it;
1865 }
1866 if (tcp_checksum_complete(skb)) {
1867 reqsk_put(req);
1868 goto csum_error;
1869 }
1870 if (unlikely(sk->sk_state != TCP_LISTEN)) {
1871 inet_csk_reqsk_queue_drop_and_put(sk, req);
1872 goto lookup;
1873 }
1874 /* We own a reference on the listener, increase it again
1875 * as we might lose it too soon.
1876 */
1877 sock_hold(sk);
1878 refcounted = true;
1879 nsk = NULL;
1880 if (!tcp_filter(sk, skb)) {
1881 th = (const struct tcphdr *)skb->data;
1882 iph = ip_hdr(skb);
1883 tcp_v4_fill_cb(skb, iph, th);
1884 nsk = tcp_check_req(sk, skb, req, false, &req_stolen);
1885 }
1886 if (!nsk) {
1887 reqsk_put(req);
1888 if (req_stolen) {
1889 /* Another cpu got exclusive access to req
1890 * and created a full blown socket.
1891 * Try to feed this packet to this socket
1892 * instead of discarding it.
1893 */
1894 tcp_v4_restore_cb(skb);
1895 sock_put(sk);
1896 goto lookup;
1897 }
1898 goto discard_and_relse;
1899 }
1900 if (nsk == sk) {
1901 reqsk_put(req);
1902 tcp_v4_restore_cb(skb);
1903 } else if (tcp_child_process(sk, nsk, skb)) {
1904 tcp_v4_send_reset(nsk, skb);
1905 goto discard_and_relse;
1906 } else {
1907 sock_put(sk);
1908 return 0;
1909 }
1910 }
1911 if (unlikely(iph->ttl < inet_sk(sk)->min_ttl)) {
1912 __NET_INC_STATS(net, LINUX_MIB_TCPMINTTLDROP);
1913 goto discard_and_relse;
1914 }
1915
1916 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
1917 goto discard_and_relse;
1918
1919 if (tcp_v4_inbound_md5_hash(sk, skb))
1920 goto discard_and_relse;
1921
1922 nf_reset_ct(skb);
1923
1924 if (tcp_filter(sk, skb))
1925 goto discard_and_relse;
1926 th = (const struct tcphdr *)skb->data;
1927 iph = ip_hdr(skb);
1928 tcp_v4_fill_cb(skb, iph, th);
1929
1930 skb->dev = NULL;
1931
1932 if (sk->sk_state == TCP_LISTEN) {
1933 ret = tcp_v4_do_rcv(sk, skb);
1934 goto put_and_return;
1935 }
1936
1937 sk_incoming_cpu_update(sk);
1938
1939 bh_lock_sock_nested(sk);
1940 tcp_segs_in(tcp_sk(sk), skb);
1941 ret = 0;
1942 if (!sock_owned_by_user(sk)) {
1943 skb_to_free = sk->sk_rx_skb_cache;
1944 sk->sk_rx_skb_cache = NULL;
1945 ret = tcp_v4_do_rcv(sk, skb);
1946 } else {
1947 if (tcp_add_backlog(sk, skb))
1948 goto discard_and_relse;
1949 skb_to_free = NULL;
1950 }
1951 bh_unlock_sock(sk);
1952 if (skb_to_free)
1953 __kfree_skb(skb_to_free);
1954
1955put_and_return:
1956 if (refcounted)
1957 sock_put(sk);
1958
1959 return ret;
1960
1961no_tcp_socket:
1962 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
1963 goto discard_it;
1964
1965 tcp_v4_fill_cb(skb, iph, th);
1966
1967 if (tcp_checksum_complete(skb)) {
1968csum_error:
1969 __TCP_INC_STATS(net, TCP_MIB_CSUMERRORS);
1970bad_packet:
1971 __TCP_INC_STATS(net, TCP_MIB_INERRS);
1972 } else {
1973 tcp_v4_send_reset(NULL, skb);
1974 }
1975
1976discard_it:
1977 /* Discard frame. */
1978 kfree_skb(skb);
1979 return 0;
1980
1981discard_and_relse:
1982 sk_drops_add(sk, skb);
1983 if (refcounted)
1984 sock_put(sk);
1985 goto discard_it;
1986
1987do_time_wait:
1988 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) {
1989 inet_twsk_put(inet_twsk(sk));
1990 goto discard_it;
1991 }
1992
1993 tcp_v4_fill_cb(skb, iph, th);
1994
1995 if (tcp_checksum_complete(skb)) {
1996 inet_twsk_put(inet_twsk(sk));
1997 goto csum_error;
1998 }
1999 switch (tcp_timewait_state_process(inet_twsk(sk), skb, th)) {
2000 case TCP_TW_SYN: {
2001 struct sock *sk2 = inet_lookup_listener(dev_net(skb->dev),
2002 &tcp_hashinfo, skb,
2003 __tcp_hdrlen(th),
2004 iph->saddr, th->source,
2005 iph->daddr, th->dest,
2006 inet_iif(skb),
2007 sdif);
2008 if (sk2) {
2009 inet_twsk_deschedule_put(inet_twsk(sk));
2010 sk = sk2;
2011 tcp_v4_restore_cb(skb);
2012 refcounted = false;
2013 goto process;
2014 }
2015 }
2016 /* to ACK */
2017 /* fall through */
2018 case TCP_TW_ACK:
2019 tcp_v4_timewait_ack(sk, skb);
2020 break;
2021 case TCP_TW_RST:
2022 tcp_v4_send_reset(sk, skb);
2023 inet_twsk_deschedule_put(inet_twsk(sk));
2024 goto discard_it;
2025 case TCP_TW_SUCCESS:;
2026 }
2027 goto discard_it;
2028}
2029
2030static struct timewait_sock_ops tcp_timewait_sock_ops = {
2031 .twsk_obj_size = sizeof(struct tcp_timewait_sock),
2032 .twsk_unique = tcp_twsk_unique,
2033 .twsk_destructor= tcp_twsk_destructor,
2034};
2035
2036void inet_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb)
2037{
2038 struct dst_entry *dst = skb_dst(skb);
2039
2040 if (dst && dst_hold_safe(dst)) {
2041 sk->sk_rx_dst = dst;
2042 inet_sk(sk)->rx_dst_ifindex = skb->skb_iif;
2043 }
2044}
2045EXPORT_SYMBOL(inet_sk_rx_dst_set);
2046
2047const struct inet_connection_sock_af_ops ipv4_specific = {
2048 .queue_xmit = ip_queue_xmit,
2049 .send_check = tcp_v4_send_check,
2050 .rebuild_header = inet_sk_rebuild_header,
2051 .sk_rx_dst_set = inet_sk_rx_dst_set,
2052 .conn_request = tcp_v4_conn_request,
2053 .syn_recv_sock = tcp_v4_syn_recv_sock,
2054 .net_header_len = sizeof(struct iphdr),
2055 .setsockopt = ip_setsockopt,
2056 .getsockopt = ip_getsockopt,
2057 .addr2sockaddr = inet_csk_addr2sockaddr,
2058 .sockaddr_len = sizeof(struct sockaddr_in),
2059#ifdef CONFIG_COMPAT
2060 .compat_setsockopt = compat_ip_setsockopt,
2061 .compat_getsockopt = compat_ip_getsockopt,
2062#endif
2063 .mtu_reduced = tcp_v4_mtu_reduced,
2064};
2065EXPORT_SYMBOL(ipv4_specific);
2066
2067#ifdef CONFIG_TCP_MD5SIG
2068static const struct tcp_sock_af_ops tcp_sock_ipv4_specific = {
2069 .md5_lookup = tcp_v4_md5_lookup,
2070 .calc_md5_hash = tcp_v4_md5_hash_skb,
2071 .md5_parse = tcp_v4_parse_md5_keys,
2072};
2073#endif
2074
2075/* NOTE: A lot of things set to zero explicitly by call to
2076 * sk_alloc() so need not be done here.
2077 */
2078static int tcp_v4_init_sock(struct sock *sk)
2079{
2080 struct inet_connection_sock *icsk = inet_csk(sk);
2081
2082 tcp_init_sock(sk);
2083
2084 icsk->icsk_af_ops = &ipv4_specific;
2085
2086#ifdef CONFIG_TCP_MD5SIG
2087 tcp_sk(sk)->af_specific = &tcp_sock_ipv4_specific;
2088#endif
2089
2090 return 0;
2091}
2092
2093void tcp_v4_destroy_sock(struct sock *sk)
2094{
2095 struct tcp_sock *tp = tcp_sk(sk);
2096
2097 trace_tcp_destroy_sock(sk);
2098
2099 tcp_clear_xmit_timers(sk);
2100
2101 tcp_cleanup_congestion_control(sk);
2102
2103 tcp_cleanup_ulp(sk);
2104
2105 /* Cleanup up the write buffer. */
2106 tcp_write_queue_purge(sk);
2107
2108 /* Check if we want to disable active TFO */
2109 tcp_fastopen_active_disable_ofo_check(sk);
2110
2111 /* Cleans up our, hopefully empty, out_of_order_queue. */
2112 skb_rbtree_purge(&tp->out_of_order_queue);
2113
2114#ifdef CONFIG_TCP_MD5SIG
2115 /* Clean up the MD5 key list, if any */
2116 if (tp->md5sig_info) {
2117 tcp_clear_md5_list(sk);
2118 kfree_rcu(rcu_dereference_protected(tp->md5sig_info, 1), rcu);
2119 tp->md5sig_info = NULL;
2120 }
2121#endif
2122
2123 /* Clean up a referenced TCP bind bucket. */
2124 if (inet_csk(sk)->icsk_bind_hash)
2125 inet_put_port(sk);
2126
2127 BUG_ON(rcu_access_pointer(tp->fastopen_rsk));
2128
2129 /* If socket is aborted during connect operation */
2130 tcp_free_fastopen_req(tp);
2131 tcp_fastopen_destroy_cipher(sk);
2132 tcp_saved_syn_free(tp);
2133
2134 sk_sockets_allocated_dec(sk);
2135}
2136EXPORT_SYMBOL(tcp_v4_destroy_sock);
2137
2138#ifdef CONFIG_PROC_FS
2139/* Proc filesystem TCP sock list dumping. */
2140
2141/*
2142 * Get next listener socket follow cur. If cur is NULL, get first socket
2143 * starting from bucket given in st->bucket; when st->bucket is zero the
2144 * very first socket in the hash table is returned.
2145 */
2146static void *listening_get_next(struct seq_file *seq, void *cur)
2147{
2148 struct tcp_seq_afinfo *afinfo = PDE_DATA(file_inode(seq->file));
2149 struct tcp_iter_state *st = seq->private;
2150 struct net *net = seq_file_net(seq);
2151 struct inet_listen_hashbucket *ilb;
2152 struct sock *sk = cur;
2153
2154 if (!sk) {
2155get_head:
2156 ilb = &tcp_hashinfo.listening_hash[st->bucket];
2157 spin_lock(&ilb->lock);
2158 sk = sk_head(&ilb->head);
2159 st->offset = 0;
2160 goto get_sk;
2161 }
2162 ilb = &tcp_hashinfo.listening_hash[st->bucket];
2163 ++st->num;
2164 ++st->offset;
2165
2166 sk = sk_next(sk);
2167get_sk:
2168 sk_for_each_from(sk) {
2169 if (!net_eq(sock_net(sk), net))
2170 continue;
2171 if (sk->sk_family == afinfo->family)
2172 return sk;
2173 }
2174 spin_unlock(&ilb->lock);
2175 st->offset = 0;
2176 if (++st->bucket < INET_LHTABLE_SIZE)
2177 goto get_head;
2178 return NULL;
2179}
2180
2181static void *listening_get_idx(struct seq_file *seq, loff_t *pos)
2182{
2183 struct tcp_iter_state *st = seq->private;
2184 void *rc;
2185
2186 st->bucket = 0;
2187 st->offset = 0;
2188 rc = listening_get_next(seq, NULL);
2189
2190 while (rc && *pos) {
2191 rc = listening_get_next(seq, rc);
2192 --*pos;
2193 }
2194 return rc;
2195}
2196
2197static inline bool empty_bucket(const struct tcp_iter_state *st)
2198{
2199 return hlist_nulls_empty(&tcp_hashinfo.ehash[st->bucket].chain);
2200}
2201
2202/*
2203 * Get first established socket starting from bucket given in st->bucket.
2204 * If st->bucket is zero, the very first socket in the hash is returned.
2205 */
2206static void *established_get_first(struct seq_file *seq)
2207{
2208 struct tcp_seq_afinfo *afinfo = PDE_DATA(file_inode(seq->file));
2209 struct tcp_iter_state *st = seq->private;
2210 struct net *net = seq_file_net(seq);
2211 void *rc = NULL;
2212
2213 st->offset = 0;
2214 for (; st->bucket <= tcp_hashinfo.ehash_mask; ++st->bucket) {
2215 struct sock *sk;
2216 struct hlist_nulls_node *node;
2217 spinlock_t *lock = inet_ehash_lockp(&tcp_hashinfo, st->bucket);
2218
2219 /* Lockless fast path for the common case of empty buckets */
2220 if (empty_bucket(st))
2221 continue;
2222
2223 spin_lock_bh(lock);
2224 sk_nulls_for_each(sk, node, &tcp_hashinfo.ehash[st->bucket].chain) {
2225 if (sk->sk_family != afinfo->family ||
2226 !net_eq(sock_net(sk), net)) {
2227 continue;
2228 }
2229 rc = sk;
2230 goto out;
2231 }
2232 spin_unlock_bh(lock);
2233 }
2234out:
2235 return rc;
2236}
2237
2238static void *established_get_next(struct seq_file *seq, void *cur)
2239{
2240 struct tcp_seq_afinfo *afinfo = PDE_DATA(file_inode(seq->file));
2241 struct sock *sk = cur;
2242 struct hlist_nulls_node *node;
2243 struct tcp_iter_state *st = seq->private;
2244 struct net *net = seq_file_net(seq);
2245
2246 ++st->num;
2247 ++st->offset;
2248
2249 sk = sk_nulls_next(sk);
2250
2251 sk_nulls_for_each_from(sk, node) {
2252 if (sk->sk_family == afinfo->family &&
2253 net_eq(sock_net(sk), net))
2254 return sk;
2255 }
2256
2257 spin_unlock_bh(inet_ehash_lockp(&tcp_hashinfo, st->bucket));
2258 ++st->bucket;
2259 return established_get_first(seq);
2260}
2261
2262static void *established_get_idx(struct seq_file *seq, loff_t pos)
2263{
2264 struct tcp_iter_state *st = seq->private;
2265 void *rc;
2266
2267 st->bucket = 0;
2268 rc = established_get_first(seq);
2269
2270 while (rc && pos) {
2271 rc = established_get_next(seq, rc);
2272 --pos;
2273 }
2274 return rc;
2275}
2276
2277static void *tcp_get_idx(struct seq_file *seq, loff_t pos)
2278{
2279 void *rc;
2280 struct tcp_iter_state *st = seq->private;
2281
2282 st->state = TCP_SEQ_STATE_LISTENING;
2283 rc = listening_get_idx(seq, &pos);
2284
2285 if (!rc) {
2286 st->state = TCP_SEQ_STATE_ESTABLISHED;
2287 rc = established_get_idx(seq, pos);
2288 }
2289
2290 return rc;
2291}
2292
2293static void *tcp_seek_last_pos(struct seq_file *seq)
2294{
2295 struct tcp_iter_state *st = seq->private;
2296 int offset = st->offset;
2297 int orig_num = st->num;
2298 void *rc = NULL;
2299
2300 switch (st->state) {
2301 case TCP_SEQ_STATE_LISTENING:
2302 if (st->bucket >= INET_LHTABLE_SIZE)
2303 break;
2304 st->state = TCP_SEQ_STATE_LISTENING;
2305 rc = listening_get_next(seq, NULL);
2306 while (offset-- && rc)
2307 rc = listening_get_next(seq, rc);
2308 if (rc)
2309 break;
2310 st->bucket = 0;
2311 st->state = TCP_SEQ_STATE_ESTABLISHED;
2312 /* Fallthrough */
2313 case TCP_SEQ_STATE_ESTABLISHED:
2314 if (st->bucket > tcp_hashinfo.ehash_mask)
2315 break;
2316 rc = established_get_first(seq);
2317 while (offset-- && rc)
2318 rc = established_get_next(seq, rc);
2319 }
2320
2321 st->num = orig_num;
2322
2323 return rc;
2324}
2325
2326void *tcp_seq_start(struct seq_file *seq, loff_t *pos)
2327{
2328 struct tcp_iter_state *st = seq->private;
2329 void *rc;
2330
2331 if (*pos && *pos == st->last_pos) {
2332 rc = tcp_seek_last_pos(seq);
2333 if (rc)
2334 goto out;
2335 }
2336
2337 st->state = TCP_SEQ_STATE_LISTENING;
2338 st->num = 0;
2339 st->bucket = 0;
2340 st->offset = 0;
2341 rc = *pos ? tcp_get_idx(seq, *pos - 1) : SEQ_START_TOKEN;
2342
2343out:
2344 st->last_pos = *pos;
2345 return rc;
2346}
2347EXPORT_SYMBOL(tcp_seq_start);
2348
2349void *tcp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2350{
2351 struct tcp_iter_state *st = seq->private;
2352 void *rc = NULL;
2353
2354 if (v == SEQ_START_TOKEN) {
2355 rc = tcp_get_idx(seq, 0);
2356 goto out;
2357 }
2358
2359 switch (st->state) {
2360 case TCP_SEQ_STATE_LISTENING:
2361 rc = listening_get_next(seq, v);
2362 if (!rc) {
2363 st->state = TCP_SEQ_STATE_ESTABLISHED;
2364 st->bucket = 0;
2365 st->offset = 0;
2366 rc = established_get_first(seq);
2367 }
2368 break;
2369 case TCP_SEQ_STATE_ESTABLISHED:
2370 rc = established_get_next(seq, v);
2371 break;
2372 }
2373out:
2374 ++*pos;
2375 st->last_pos = *pos;
2376 return rc;
2377}
2378EXPORT_SYMBOL(tcp_seq_next);
2379
2380void tcp_seq_stop(struct seq_file *seq, void *v)
2381{
2382 struct tcp_iter_state *st = seq->private;
2383
2384 switch (st->state) {
2385 case TCP_SEQ_STATE_LISTENING:
2386 if (v != SEQ_START_TOKEN)
2387 spin_unlock(&tcp_hashinfo.listening_hash[st->bucket].lock);
2388 break;
2389 case TCP_SEQ_STATE_ESTABLISHED:
2390 if (v)
2391 spin_unlock_bh(inet_ehash_lockp(&tcp_hashinfo, st->bucket));
2392 break;
2393 }
2394}
2395EXPORT_SYMBOL(tcp_seq_stop);
2396
2397static void get_openreq4(const struct request_sock *req,
2398 struct seq_file *f, int i)
2399{
2400 const struct inet_request_sock *ireq = inet_rsk(req);
2401 long delta = req->rsk_timer.expires - jiffies;
2402
2403 seq_printf(f, "%4d: %08X:%04X %08X:%04X"
2404 " %02X %08X:%08X %02X:%08lX %08X %5u %8d %u %d %pK",
2405 i,
2406 ireq->ir_loc_addr,
2407 ireq->ir_num,
2408 ireq->ir_rmt_addr,
2409 ntohs(ireq->ir_rmt_port),
2410 TCP_SYN_RECV,
2411 0, 0, /* could print option size, but that is af dependent. */
2412 1, /* timers active (only the expire timer) */
2413 jiffies_delta_to_clock_t(delta),
2414 req->num_timeout,
2415 from_kuid_munged(seq_user_ns(f),
2416 sock_i_uid(req->rsk_listener)),
2417 0, /* non standard timer */
2418 0, /* open_requests have no inode */
2419 0,
2420 req);
2421}
2422
2423static void get_tcp4_sock(struct sock *sk, struct seq_file *f, int i)
2424{
2425 int timer_active;
2426 unsigned long timer_expires;
2427 const struct tcp_sock *tp = tcp_sk(sk);
2428 const struct inet_connection_sock *icsk = inet_csk(sk);
2429 const struct inet_sock *inet = inet_sk(sk);
2430 const struct fastopen_queue *fastopenq = &icsk->icsk_accept_queue.fastopenq;
2431 __be32 dest = inet->inet_daddr;
2432 __be32 src = inet->inet_rcv_saddr;
2433 __u16 destp = ntohs(inet->inet_dport);
2434 __u16 srcp = ntohs(inet->inet_sport);
2435 int rx_queue;
2436 int state;
2437
2438 if (icsk->icsk_pending == ICSK_TIME_RETRANS ||
2439 icsk->icsk_pending == ICSK_TIME_REO_TIMEOUT ||
2440 icsk->icsk_pending == ICSK_TIME_LOSS_PROBE) {
2441 timer_active = 1;
2442 timer_expires = icsk->icsk_timeout;
2443 } else if (icsk->icsk_pending == ICSK_TIME_PROBE0) {
2444 timer_active = 4;
2445 timer_expires = icsk->icsk_timeout;
2446 } else if (timer_pending(&sk->sk_timer)) {
2447 timer_active = 2;
2448 timer_expires = sk->sk_timer.expires;
2449 } else {
2450 timer_active = 0;
2451 timer_expires = jiffies;
2452 }
2453
2454 state = inet_sk_state_load(sk);
2455 if (state == TCP_LISTEN)
2456 rx_queue = sk->sk_ack_backlog;
2457 else
2458 /* Because we don't lock the socket,
2459 * we might find a transient negative value.
2460 */
2461 rx_queue = max_t(int, READ_ONCE(tp->rcv_nxt) -
2462 READ_ONCE(tp->copied_seq), 0);
2463
2464 seq_printf(f, "%4d: %08X:%04X %08X:%04X %02X %08X:%08X %02X:%08lX "
2465 "%08X %5u %8d %lu %d %pK %lu %lu %u %u %d",
2466 i, src, srcp, dest, destp, state,
2467 READ_ONCE(tp->write_seq) - tp->snd_una,
2468 rx_queue,
2469 timer_active,
2470 jiffies_delta_to_clock_t(timer_expires - jiffies),
2471 icsk->icsk_retransmits,
2472 from_kuid_munged(seq_user_ns(f), sock_i_uid(sk)),
2473 icsk->icsk_probes_out,
2474 sock_i_ino(sk),
2475 refcount_read(&sk->sk_refcnt), sk,
2476 jiffies_to_clock_t(icsk->icsk_rto),
2477 jiffies_to_clock_t(icsk->icsk_ack.ato),
2478 (icsk->icsk_ack.quick << 1) | inet_csk_in_pingpong_mode(sk),
2479 tp->snd_cwnd,
2480 state == TCP_LISTEN ?
2481 fastopenq->max_qlen :
2482 (tcp_in_initial_slowstart(tp) ? -1 : tp->snd_ssthresh));
2483}
2484
2485static void get_timewait4_sock(const struct inet_timewait_sock *tw,
2486 struct seq_file *f, int i)
2487{
2488 long delta = tw->tw_timer.expires - jiffies;
2489 __be32 dest, src;
2490 __u16 destp, srcp;
2491
2492 dest = tw->tw_daddr;
2493 src = tw->tw_rcv_saddr;
2494 destp = ntohs(tw->tw_dport);
2495 srcp = ntohs(tw->tw_sport);
2496
2497 seq_printf(f, "%4d: %08X:%04X %08X:%04X"
2498 " %02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %pK",
2499 i, src, srcp, dest, destp, tw->tw_substate, 0, 0,
2500 3, jiffies_delta_to_clock_t(delta), 0, 0, 0, 0,
2501 refcount_read(&tw->tw_refcnt), tw);
2502}
2503
2504#define TMPSZ 150
2505
2506static int tcp4_seq_show(struct seq_file *seq, void *v)
2507{
2508 struct tcp_iter_state *st;
2509 struct sock *sk = v;
2510
2511 seq_setwidth(seq, TMPSZ - 1);
2512 if (v == SEQ_START_TOKEN) {
2513 seq_puts(seq, " sl local_address rem_address st tx_queue "
2514 "rx_queue tr tm->when retrnsmt uid timeout "
2515 "inode");
2516 goto out;
2517 }
2518 st = seq->private;
2519
2520 if (sk->sk_state == TCP_TIME_WAIT)
2521 get_timewait4_sock(v, seq, st->num);
2522 else if (sk->sk_state == TCP_NEW_SYN_RECV)
2523 get_openreq4(v, seq, st->num);
2524 else
2525 get_tcp4_sock(v, seq, st->num);
2526out:
2527 seq_pad(seq, '\n');
2528 return 0;
2529}
2530
2531static const struct seq_operations tcp4_seq_ops = {
2532 .show = tcp4_seq_show,
2533 .start = tcp_seq_start,
2534 .next = tcp_seq_next,
2535 .stop = tcp_seq_stop,
2536};
2537
2538static struct tcp_seq_afinfo tcp4_seq_afinfo = {
2539 .family = AF_INET,
2540};
2541
2542static int __net_init tcp4_proc_init_net(struct net *net)
2543{
2544 if (!proc_create_net_data("tcp", 0444, net->proc_net, &tcp4_seq_ops,
2545 sizeof(struct tcp_iter_state), &tcp4_seq_afinfo))
2546 return -ENOMEM;
2547 return 0;
2548}
2549
2550static void __net_exit tcp4_proc_exit_net(struct net *net)
2551{
2552 remove_proc_entry("tcp", net->proc_net);
2553}
2554
2555static struct pernet_operations tcp4_net_ops = {
2556 .init = tcp4_proc_init_net,
2557 .exit = tcp4_proc_exit_net,
2558};
2559
2560int __init tcp4_proc_init(void)
2561{
2562 return register_pernet_subsys(&tcp4_net_ops);
2563}
2564
2565void tcp4_proc_exit(void)
2566{
2567 unregister_pernet_subsys(&tcp4_net_ops);
2568}
2569#endif /* CONFIG_PROC_FS */
2570
2571struct proto tcp_prot = {
2572 .name = "TCP",
2573 .owner = THIS_MODULE,
2574 .close = tcp_close,
2575 .pre_connect = tcp_v4_pre_connect,
2576 .connect = tcp_v4_connect,
2577 .disconnect = tcp_disconnect,
2578 .accept = inet_csk_accept,
2579 .ioctl = tcp_ioctl,
2580 .init = tcp_v4_init_sock,
2581 .destroy = tcp_v4_destroy_sock,
2582 .shutdown = tcp_shutdown,
2583 .setsockopt = tcp_setsockopt,
2584 .getsockopt = tcp_getsockopt,
2585 .keepalive = tcp_set_keepalive,
2586 .recvmsg = tcp_recvmsg,
2587 .sendmsg = tcp_sendmsg,
2588 .sendpage = tcp_sendpage,
2589 .backlog_rcv = tcp_v4_do_rcv,
2590 .release_cb = tcp_release_cb,
2591 .hash = inet_hash,
2592 .unhash = inet_unhash,
2593 .get_port = inet_csk_get_port,
2594 .enter_memory_pressure = tcp_enter_memory_pressure,
2595 .leave_memory_pressure = tcp_leave_memory_pressure,
2596 .stream_memory_free = tcp_stream_memory_free,
2597 .sockets_allocated = &tcp_sockets_allocated,
2598 .orphan_count = &tcp_orphan_count,
2599 .memory_allocated = &tcp_memory_allocated,
2600 .memory_pressure = &tcp_memory_pressure,
2601 .sysctl_mem = sysctl_tcp_mem,
2602 .sysctl_wmem_offset = offsetof(struct net, ipv4.sysctl_tcp_wmem),
2603 .sysctl_rmem_offset = offsetof(struct net, ipv4.sysctl_tcp_rmem),
2604 .max_header = MAX_TCP_HEADER,
2605 .obj_size = sizeof(struct tcp_sock),
2606 .slab_flags = SLAB_TYPESAFE_BY_RCU,
2607 .twsk_prot = &tcp_timewait_sock_ops,
2608 .rsk_prot = &tcp_request_sock_ops,
2609 .h.hashinfo = &tcp_hashinfo,
2610 .no_autobind = true,
2611#ifdef CONFIG_COMPAT
2612 .compat_setsockopt = compat_tcp_setsockopt,
2613 .compat_getsockopt = compat_tcp_getsockopt,
2614#endif
2615 .diag_destroy = tcp_abort,
2616};
2617EXPORT_SYMBOL(tcp_prot);
2618
2619static void __net_exit tcp_sk_exit(struct net *net)
2620{
2621 int cpu;
2622
2623 if (net->ipv4.tcp_congestion_control)
2624 module_put(net->ipv4.tcp_congestion_control->owner);
2625
2626 for_each_possible_cpu(cpu)
2627 inet_ctl_sock_destroy(*per_cpu_ptr(net->ipv4.tcp_sk, cpu));
2628 free_percpu(net->ipv4.tcp_sk);
2629}
2630
2631static int __net_init tcp_sk_init(struct net *net)
2632{
2633 int res, cpu, cnt;
2634
2635 net->ipv4.tcp_sk = alloc_percpu(struct sock *);
2636 if (!net->ipv4.tcp_sk)
2637 return -ENOMEM;
2638
2639 for_each_possible_cpu(cpu) {
2640 struct sock *sk;
2641
2642 res = inet_ctl_sock_create(&sk, PF_INET, SOCK_RAW,
2643 IPPROTO_TCP, net);
2644 if (res)
2645 goto fail;
2646 sock_set_flag(sk, SOCK_USE_WRITE_QUEUE);
2647
2648 /* Please enforce IP_DF and IPID==0 for RST and
2649 * ACK sent in SYN-RECV and TIME-WAIT state.
2650 */
2651 inet_sk(sk)->pmtudisc = IP_PMTUDISC_DO;
2652
2653 *per_cpu_ptr(net->ipv4.tcp_sk, cpu) = sk;
2654 }
2655
2656 net->ipv4.sysctl_tcp_ecn = 2;
2657 net->ipv4.sysctl_tcp_ecn_fallback = 1;
2658
2659 net->ipv4.sysctl_tcp_base_mss = TCP_BASE_MSS;
2660 net->ipv4.sysctl_tcp_min_snd_mss = TCP_MIN_SND_MSS;
2661 net->ipv4.sysctl_tcp_probe_threshold = TCP_PROBE_THRESHOLD;
2662 net->ipv4.sysctl_tcp_probe_interval = TCP_PROBE_INTERVAL;
2663 net->ipv4.sysctl_tcp_mtu_probe_floor = TCP_MIN_SND_MSS;
2664
2665 net->ipv4.sysctl_tcp_keepalive_time = TCP_KEEPALIVE_TIME;
2666 net->ipv4.sysctl_tcp_keepalive_probes = TCP_KEEPALIVE_PROBES;
2667 net->ipv4.sysctl_tcp_keepalive_intvl = TCP_KEEPALIVE_INTVL;
2668
2669 net->ipv4.sysctl_tcp_syn_retries = TCP_SYN_RETRIES;
2670 net->ipv4.sysctl_tcp_synack_retries = TCP_SYNACK_RETRIES;
2671 net->ipv4.sysctl_tcp_syncookies = 1;
2672 net->ipv4.sysctl_tcp_reordering = TCP_FASTRETRANS_THRESH;
2673 net->ipv4.sysctl_tcp_retries1 = TCP_RETR1;
2674 net->ipv4.sysctl_tcp_retries2 = TCP_RETR2;
2675 net->ipv4.sysctl_tcp_orphan_retries = 0;
2676 net->ipv4.sysctl_tcp_fin_timeout = TCP_FIN_TIMEOUT;
2677 net->ipv4.sysctl_tcp_notsent_lowat = UINT_MAX;
2678 net->ipv4.sysctl_tcp_tw_reuse = 2;
2679
2680 cnt = tcp_hashinfo.ehash_mask + 1;
2681 net->ipv4.tcp_death_row.sysctl_max_tw_buckets = cnt / 2;
2682 net->ipv4.tcp_death_row.hashinfo = &tcp_hashinfo;
2683
2684 net->ipv4.sysctl_max_syn_backlog = max(128, cnt / 128);
2685 net->ipv4.sysctl_tcp_sack = 1;
2686 net->ipv4.sysctl_tcp_window_scaling = 1;
2687 net->ipv4.sysctl_tcp_timestamps = 1;
2688 net->ipv4.sysctl_tcp_early_retrans = 3;
2689 net->ipv4.sysctl_tcp_recovery = TCP_RACK_LOSS_DETECTION;
2690 net->ipv4.sysctl_tcp_slow_start_after_idle = 1; /* By default, RFC2861 behavior. */
2691 net->ipv4.sysctl_tcp_retrans_collapse = 1;
2692 net->ipv4.sysctl_tcp_max_reordering = 300;
2693 net->ipv4.sysctl_tcp_dsack = 1;
2694 net->ipv4.sysctl_tcp_app_win = 31;
2695 net->ipv4.sysctl_tcp_adv_win_scale = 1;
2696 net->ipv4.sysctl_tcp_frto = 2;
2697 net->ipv4.sysctl_tcp_moderate_rcvbuf = 1;
2698 /* This limits the percentage of the congestion window which we
2699 * will allow a single TSO frame to consume. Building TSO frames
2700 * which are too large can cause TCP streams to be bursty.
2701 */
2702 net->ipv4.sysctl_tcp_tso_win_divisor = 3;
2703 /* Default TSQ limit of 16 TSO segments */
2704 net->ipv4.sysctl_tcp_limit_output_bytes = 16 * 65536;
2705 /* rfc5961 challenge ack rate limiting */
2706 net->ipv4.sysctl_tcp_challenge_ack_limit = 1000;
2707 net->ipv4.sysctl_tcp_min_tso_segs = 2;
2708 net->ipv4.sysctl_tcp_min_rtt_wlen = 300;
2709 net->ipv4.sysctl_tcp_autocorking = 1;
2710 net->ipv4.sysctl_tcp_invalid_ratelimit = HZ/2;
2711 net->ipv4.sysctl_tcp_pacing_ss_ratio = 200;
2712 net->ipv4.sysctl_tcp_pacing_ca_ratio = 120;
2713 if (net != &init_net) {
2714 memcpy(net->ipv4.sysctl_tcp_rmem,
2715 init_net.ipv4.sysctl_tcp_rmem,
2716 sizeof(init_net.ipv4.sysctl_tcp_rmem));
2717 memcpy(net->ipv4.sysctl_tcp_wmem,
2718 init_net.ipv4.sysctl_tcp_wmem,
2719 sizeof(init_net.ipv4.sysctl_tcp_wmem));
2720 }
2721 net->ipv4.sysctl_tcp_comp_sack_delay_ns = NSEC_PER_MSEC;
2722 net->ipv4.sysctl_tcp_comp_sack_nr = 44;
2723 net->ipv4.sysctl_tcp_fastopen = TFO_CLIENT_ENABLE;
2724 spin_lock_init(&net->ipv4.tcp_fastopen_ctx_lock);
2725 net->ipv4.sysctl_tcp_fastopen_blackhole_timeout = 60 * 60;
2726 atomic_set(&net->ipv4.tfo_active_disable_times, 0);
2727
2728 /* Reno is always built in */
2729 if (!net_eq(net, &init_net) &&
2730 try_module_get(init_net.ipv4.tcp_congestion_control->owner))
2731 net->ipv4.tcp_congestion_control = init_net.ipv4.tcp_congestion_control;
2732 else
2733 net->ipv4.tcp_congestion_control = &tcp_reno;
2734
2735 return 0;
2736fail:
2737 tcp_sk_exit(net);
2738
2739 return res;
2740}
2741
2742static void __net_exit tcp_sk_exit_batch(struct list_head *net_exit_list)
2743{
2744 struct net *net;
2745
2746 inet_twsk_purge(&tcp_hashinfo, AF_INET);
2747
2748 list_for_each_entry(net, net_exit_list, exit_list)
2749 tcp_fastopen_ctx_destroy(net);
2750}
2751
2752static struct pernet_operations __net_initdata tcp_sk_ops = {
2753 .init = tcp_sk_init,
2754 .exit = tcp_sk_exit,
2755 .exit_batch = tcp_sk_exit_batch,
2756};
2757
2758void __init tcp_v4_init(void)
2759{
2760 if (register_pernet_subsys(&tcp_sk_ops))
2761 panic("Failed to create the TCP control socket.\n");
2762}
1// SPDX-License-Identifier: GPL-2.0-or-later
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 * IPv4 specific functions
10 *
11 * code split from:
12 * linux/ipv4/tcp.c
13 * linux/ipv4/tcp_input.c
14 * linux/ipv4/tcp_output.c
15 *
16 * See tcp.c for author information
17 */
18
19/*
20 * Changes:
21 * David S. Miller : New socket lookup architecture.
22 * This code is dedicated to John Dyson.
23 * David S. Miller : Change semantics of established hash,
24 * half is devoted to TIME_WAIT sockets
25 * and the rest go in the other half.
26 * Andi Kleen : Add support for syncookies and fixed
27 * some bugs: ip options weren't passed to
28 * the TCP layer, missed a check for an
29 * ACK bit.
30 * Andi Kleen : Implemented fast path mtu discovery.
31 * Fixed many serious bugs in the
32 * request_sock handling and moved
33 * most of it into the af independent code.
34 * Added tail drop and some other bugfixes.
35 * Added new listen semantics.
36 * Mike McLagan : Routing by source
37 * Juan Jose Ciarlante: ip_dynaddr bits
38 * Andi Kleen: various fixes.
39 * Vitaly E. Lavrov : Transparent proxy revived after year
40 * coma.
41 * Andi Kleen : Fix new listen.
42 * Andi Kleen : Fix accept error reporting.
43 * YOSHIFUJI Hideaki @USAGI and: Support IPV6_V6ONLY socket option, which
44 * Alexey Kuznetsov allow both IPv4 and IPv6 sockets to bind
45 * a single port at the same time.
46 */
47
48#define pr_fmt(fmt) "TCP: " fmt
49
50#include <linux/bottom_half.h>
51#include <linux/types.h>
52#include <linux/fcntl.h>
53#include <linux/module.h>
54#include <linux/random.h>
55#include <linux/cache.h>
56#include <linux/jhash.h>
57#include <linux/init.h>
58#include <linux/times.h>
59#include <linux/slab.h>
60#include <linux/sched.h>
61
62#include <net/net_namespace.h>
63#include <net/icmp.h>
64#include <net/inet_hashtables.h>
65#include <net/tcp.h>
66#include <net/transp_v6.h>
67#include <net/ipv6.h>
68#include <net/inet_common.h>
69#include <net/timewait_sock.h>
70#include <net/xfrm.h>
71#include <net/secure_seq.h>
72#include <net/busy_poll.h>
73
74#include <linux/inet.h>
75#include <linux/ipv6.h>
76#include <linux/stddef.h>
77#include <linux/proc_fs.h>
78#include <linux/seq_file.h>
79#include <linux/inetdevice.h>
80#include <linux/btf_ids.h>
81
82#include <crypto/hash.h>
83#include <linux/scatterlist.h>
84
85#include <trace/events/tcp.h>
86
87#ifdef CONFIG_TCP_MD5SIG
88static int tcp_v4_md5_hash_hdr(char *md5_hash, const struct tcp_md5sig_key *key,
89 __be32 daddr, __be32 saddr, const struct tcphdr *th);
90#endif
91
92struct inet_hashinfo tcp_hashinfo;
93EXPORT_SYMBOL(tcp_hashinfo);
94
95static DEFINE_PER_CPU(struct sock *, ipv4_tcp_sk);
96
97static u32 tcp_v4_init_seq(const struct sk_buff *skb)
98{
99 return secure_tcp_seq(ip_hdr(skb)->daddr,
100 ip_hdr(skb)->saddr,
101 tcp_hdr(skb)->dest,
102 tcp_hdr(skb)->source);
103}
104
105static u32 tcp_v4_init_ts_off(const struct net *net, const struct sk_buff *skb)
106{
107 return secure_tcp_ts_off(net, ip_hdr(skb)->daddr, ip_hdr(skb)->saddr);
108}
109
110int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp)
111{
112 int reuse = READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_tw_reuse);
113 const struct inet_timewait_sock *tw = inet_twsk(sktw);
114 const struct tcp_timewait_sock *tcptw = tcp_twsk(sktw);
115 struct tcp_sock *tp = tcp_sk(sk);
116
117 if (reuse == 2) {
118 /* Still does not detect *everything* that goes through
119 * lo, since we require a loopback src or dst address
120 * or direct binding to 'lo' interface.
121 */
122 bool loopback = false;
123 if (tw->tw_bound_dev_if == LOOPBACK_IFINDEX)
124 loopback = true;
125#if IS_ENABLED(CONFIG_IPV6)
126 if (tw->tw_family == AF_INET6) {
127 if (ipv6_addr_loopback(&tw->tw_v6_daddr) ||
128 ipv6_addr_v4mapped_loopback(&tw->tw_v6_daddr) ||
129 ipv6_addr_loopback(&tw->tw_v6_rcv_saddr) ||
130 ipv6_addr_v4mapped_loopback(&tw->tw_v6_rcv_saddr))
131 loopback = true;
132 } else
133#endif
134 {
135 if (ipv4_is_loopback(tw->tw_daddr) ||
136 ipv4_is_loopback(tw->tw_rcv_saddr))
137 loopback = true;
138 }
139 if (!loopback)
140 reuse = 0;
141 }
142
143 /* With PAWS, it is safe from the viewpoint
144 of data integrity. Even without PAWS it is safe provided sequence
145 spaces do not overlap i.e. at data rates <= 80Mbit/sec.
146
147 Actually, the idea is close to VJ's one, only timestamp cache is
148 held not per host, but per port pair and TW bucket is used as state
149 holder.
150
151 If TW bucket has been already destroyed we fall back to VJ's scheme
152 and use initial timestamp retrieved from peer table.
153 */
154 if (tcptw->tw_ts_recent_stamp &&
155 (!twp || (reuse && time_after32(ktime_get_seconds(),
156 tcptw->tw_ts_recent_stamp)))) {
157 /* inet_twsk_hashdance() sets sk_refcnt after putting twsk
158 * and releasing the bucket lock.
159 */
160 if (unlikely(!refcount_inc_not_zero(&sktw->sk_refcnt)))
161 return 0;
162
163 /* In case of repair and re-using TIME-WAIT sockets we still
164 * want to be sure that it is safe as above but honor the
165 * sequence numbers and time stamps set as part of the repair
166 * process.
167 *
168 * Without this check re-using a TIME-WAIT socket with TCP
169 * repair would accumulate a -1 on the repair assigned
170 * sequence number. The first time it is reused the sequence
171 * is -1, the second time -2, etc. This fixes that issue
172 * without appearing to create any others.
173 */
174 if (likely(!tp->repair)) {
175 u32 seq = tcptw->tw_snd_nxt + 65535 + 2;
176
177 if (!seq)
178 seq = 1;
179 WRITE_ONCE(tp->write_seq, seq);
180 tp->rx_opt.ts_recent = tcptw->tw_ts_recent;
181 tp->rx_opt.ts_recent_stamp = tcptw->tw_ts_recent_stamp;
182 }
183
184 return 1;
185 }
186
187 return 0;
188}
189EXPORT_SYMBOL_GPL(tcp_twsk_unique);
190
191static int tcp_v4_pre_connect(struct sock *sk, struct sockaddr *uaddr,
192 int addr_len)
193{
194 /* This check is replicated from tcp_v4_connect() and intended to
195 * prevent BPF program called below from accessing bytes that are out
196 * of the bound specified by user in addr_len.
197 */
198 if (addr_len < sizeof(struct sockaddr_in))
199 return -EINVAL;
200
201 sock_owned_by_me(sk);
202
203 return BPF_CGROUP_RUN_PROG_INET4_CONNECT(sk, uaddr, &addr_len);
204}
205
206/* This will initiate an outgoing connection. */
207int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len)
208{
209 struct sockaddr_in *usin = (struct sockaddr_in *)uaddr;
210 struct inet_timewait_death_row *tcp_death_row;
211 struct inet_sock *inet = inet_sk(sk);
212 struct tcp_sock *tp = tcp_sk(sk);
213 struct ip_options_rcu *inet_opt;
214 struct net *net = sock_net(sk);
215 __be16 orig_sport, orig_dport;
216 __be32 daddr, nexthop;
217 struct flowi4 *fl4;
218 struct rtable *rt;
219 int err;
220
221 if (addr_len < sizeof(struct sockaddr_in))
222 return -EINVAL;
223
224 if (usin->sin_family != AF_INET)
225 return -EAFNOSUPPORT;
226
227 nexthop = daddr = usin->sin_addr.s_addr;
228 inet_opt = rcu_dereference_protected(inet->inet_opt,
229 lockdep_sock_is_held(sk));
230 if (inet_opt && inet_opt->opt.srr) {
231 if (!daddr)
232 return -EINVAL;
233 nexthop = inet_opt->opt.faddr;
234 }
235
236 orig_sport = inet->inet_sport;
237 orig_dport = usin->sin_port;
238 fl4 = &inet->cork.fl.u.ip4;
239 rt = ip_route_connect(fl4, nexthop, inet->inet_saddr,
240 sk->sk_bound_dev_if, IPPROTO_TCP, orig_sport,
241 orig_dport, sk);
242 if (IS_ERR(rt)) {
243 err = PTR_ERR(rt);
244 if (err == -ENETUNREACH)
245 IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES);
246 return err;
247 }
248
249 if (rt->rt_flags & (RTCF_MULTICAST | RTCF_BROADCAST)) {
250 ip_rt_put(rt);
251 return -ENETUNREACH;
252 }
253
254 if (!inet_opt || !inet_opt->opt.srr)
255 daddr = fl4->daddr;
256
257 tcp_death_row = &sock_net(sk)->ipv4.tcp_death_row;
258
259 if (!inet->inet_saddr) {
260 err = inet_bhash2_update_saddr(sk, &fl4->saddr, AF_INET);
261 if (err) {
262 ip_rt_put(rt);
263 return err;
264 }
265 } else {
266 sk_rcv_saddr_set(sk, inet->inet_saddr);
267 }
268
269 if (tp->rx_opt.ts_recent_stamp && inet->inet_daddr != daddr) {
270 /* Reset inherited state */
271 tp->rx_opt.ts_recent = 0;
272 tp->rx_opt.ts_recent_stamp = 0;
273 if (likely(!tp->repair))
274 WRITE_ONCE(tp->write_seq, 0);
275 }
276
277 inet->inet_dport = usin->sin_port;
278 sk_daddr_set(sk, daddr);
279
280 inet_csk(sk)->icsk_ext_hdr_len = 0;
281 if (inet_opt)
282 inet_csk(sk)->icsk_ext_hdr_len = inet_opt->opt.optlen;
283
284 tp->rx_opt.mss_clamp = TCP_MSS_DEFAULT;
285
286 /* Socket identity is still unknown (sport may be zero).
287 * However we set state to SYN-SENT and not releasing socket
288 * lock select source port, enter ourselves into the hash tables and
289 * complete initialization after this.
290 */
291 tcp_set_state(sk, TCP_SYN_SENT);
292 err = inet_hash_connect(tcp_death_row, sk);
293 if (err)
294 goto failure;
295
296 sk_set_txhash(sk);
297
298 rt = ip_route_newports(fl4, rt, orig_sport, orig_dport,
299 inet->inet_sport, inet->inet_dport, sk);
300 if (IS_ERR(rt)) {
301 err = PTR_ERR(rt);
302 rt = NULL;
303 goto failure;
304 }
305 tp->tcp_usec_ts = dst_tcp_usec_ts(&rt->dst);
306 /* OK, now commit destination to socket. */
307 sk->sk_gso_type = SKB_GSO_TCPV4;
308 sk_setup_caps(sk, &rt->dst);
309 rt = NULL;
310
311 if (likely(!tp->repair)) {
312 if (!tp->write_seq)
313 WRITE_ONCE(tp->write_seq,
314 secure_tcp_seq(inet->inet_saddr,
315 inet->inet_daddr,
316 inet->inet_sport,
317 usin->sin_port));
318 WRITE_ONCE(tp->tsoffset,
319 secure_tcp_ts_off(net, inet->inet_saddr,
320 inet->inet_daddr));
321 }
322
323 atomic_set(&inet->inet_id, get_random_u16());
324
325 if (tcp_fastopen_defer_connect(sk, &err))
326 return err;
327 if (err)
328 goto failure;
329
330 err = tcp_connect(sk);
331
332 if (err)
333 goto failure;
334
335 return 0;
336
337failure:
338 /*
339 * This unhashes the socket and releases the local port,
340 * if necessary.
341 */
342 tcp_set_state(sk, TCP_CLOSE);
343 inet_bhash2_reset_saddr(sk);
344 ip_rt_put(rt);
345 sk->sk_route_caps = 0;
346 inet->inet_dport = 0;
347 return err;
348}
349EXPORT_SYMBOL(tcp_v4_connect);
350
351/*
352 * This routine reacts to ICMP_FRAG_NEEDED mtu indications as defined in RFC1191.
353 * It can be called through tcp_release_cb() if socket was owned by user
354 * at the time tcp_v4_err() was called to handle ICMP message.
355 */
356void tcp_v4_mtu_reduced(struct sock *sk)
357{
358 struct inet_sock *inet = inet_sk(sk);
359 struct dst_entry *dst;
360 u32 mtu;
361
362 if ((1 << sk->sk_state) & (TCPF_LISTEN | TCPF_CLOSE))
363 return;
364 mtu = READ_ONCE(tcp_sk(sk)->mtu_info);
365 dst = inet_csk_update_pmtu(sk, mtu);
366 if (!dst)
367 return;
368
369 /* Something is about to be wrong... Remember soft error
370 * for the case, if this connection will not able to recover.
371 */
372 if (mtu < dst_mtu(dst) && ip_dont_fragment(sk, dst))
373 WRITE_ONCE(sk->sk_err_soft, EMSGSIZE);
374
375 mtu = dst_mtu(dst);
376
377 if (inet->pmtudisc != IP_PMTUDISC_DONT &&
378 ip_sk_accept_pmtu(sk) &&
379 inet_csk(sk)->icsk_pmtu_cookie > mtu) {
380 tcp_sync_mss(sk, mtu);
381
382 /* Resend the TCP packet because it's
383 * clear that the old packet has been
384 * dropped. This is the new "fast" path mtu
385 * discovery.
386 */
387 tcp_simple_retransmit(sk);
388 } /* else let the usual retransmit timer handle it */
389}
390EXPORT_SYMBOL(tcp_v4_mtu_reduced);
391
392static void do_redirect(struct sk_buff *skb, struct sock *sk)
393{
394 struct dst_entry *dst = __sk_dst_check(sk, 0);
395
396 if (dst)
397 dst->ops->redirect(dst, sk, skb);
398}
399
400
401/* handle ICMP messages on TCP_NEW_SYN_RECV request sockets */
402void tcp_req_err(struct sock *sk, u32 seq, bool abort)
403{
404 struct request_sock *req = inet_reqsk(sk);
405 struct net *net = sock_net(sk);
406
407 /* ICMPs are not backlogged, hence we cannot get
408 * an established socket here.
409 */
410 if (seq != tcp_rsk(req)->snt_isn) {
411 __NET_INC_STATS(net, LINUX_MIB_OUTOFWINDOWICMPS);
412 } else if (abort) {
413 /*
414 * Still in SYN_RECV, just remove it silently.
415 * There is no good way to pass the error to the newly
416 * created socket, and POSIX does not want network
417 * errors returned from accept().
418 */
419 inet_csk_reqsk_queue_drop(req->rsk_listener, req);
420 tcp_listendrop(req->rsk_listener);
421 }
422 reqsk_put(req);
423}
424EXPORT_SYMBOL(tcp_req_err);
425
426/* TCP-LD (RFC 6069) logic */
427void tcp_ld_RTO_revert(struct sock *sk, u32 seq)
428{
429 struct inet_connection_sock *icsk = inet_csk(sk);
430 struct tcp_sock *tp = tcp_sk(sk);
431 struct sk_buff *skb;
432 s32 remaining;
433 u32 delta_us;
434
435 if (sock_owned_by_user(sk))
436 return;
437
438 if (seq != tp->snd_una || !icsk->icsk_retransmits ||
439 !icsk->icsk_backoff)
440 return;
441
442 skb = tcp_rtx_queue_head(sk);
443 if (WARN_ON_ONCE(!skb))
444 return;
445
446 icsk->icsk_backoff--;
447 icsk->icsk_rto = tp->srtt_us ? __tcp_set_rto(tp) : TCP_TIMEOUT_INIT;
448 icsk->icsk_rto = inet_csk_rto_backoff(icsk, TCP_RTO_MAX);
449
450 tcp_mstamp_refresh(tp);
451 delta_us = (u32)(tp->tcp_mstamp - tcp_skb_timestamp_us(skb));
452 remaining = icsk->icsk_rto - usecs_to_jiffies(delta_us);
453
454 if (remaining > 0) {
455 inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
456 remaining, TCP_RTO_MAX);
457 } else {
458 /* RTO revert clocked out retransmission.
459 * Will retransmit now.
460 */
461 tcp_retransmit_timer(sk);
462 }
463}
464EXPORT_SYMBOL(tcp_ld_RTO_revert);
465
466/*
467 * This routine is called by the ICMP module when it gets some
468 * sort of error condition. If err < 0 then the socket should
469 * be closed and the error returned to the user. If err > 0
470 * it's just the icmp type << 8 | icmp code. After adjustment
471 * header points to the first 8 bytes of the tcp header. We need
472 * to find the appropriate port.
473 *
474 * The locking strategy used here is very "optimistic". When
475 * someone else accesses the socket the ICMP is just dropped
476 * and for some paths there is no check at all.
477 * A more general error queue to queue errors for later handling
478 * is probably better.
479 *
480 */
481
482int tcp_v4_err(struct sk_buff *skb, u32 info)
483{
484 const struct iphdr *iph = (const struct iphdr *)skb->data;
485 struct tcphdr *th = (struct tcphdr *)(skb->data + (iph->ihl << 2));
486 struct tcp_sock *tp;
487 const int type = icmp_hdr(skb)->type;
488 const int code = icmp_hdr(skb)->code;
489 struct sock *sk;
490 struct request_sock *fastopen;
491 u32 seq, snd_una;
492 int err;
493 struct net *net = dev_net(skb->dev);
494
495 sk = __inet_lookup_established(net, net->ipv4.tcp_death_row.hashinfo,
496 iph->daddr, th->dest, iph->saddr,
497 ntohs(th->source), inet_iif(skb), 0);
498 if (!sk) {
499 __ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
500 return -ENOENT;
501 }
502 if (sk->sk_state == TCP_TIME_WAIT) {
503 /* To increase the counter of ignored icmps for TCP-AO */
504 tcp_ao_ignore_icmp(sk, AF_INET, type, code);
505 inet_twsk_put(inet_twsk(sk));
506 return 0;
507 }
508 seq = ntohl(th->seq);
509 if (sk->sk_state == TCP_NEW_SYN_RECV) {
510 tcp_req_err(sk, seq, type == ICMP_PARAMETERPROB ||
511 type == ICMP_TIME_EXCEEDED ||
512 (type == ICMP_DEST_UNREACH &&
513 (code == ICMP_NET_UNREACH ||
514 code == ICMP_HOST_UNREACH)));
515 return 0;
516 }
517
518 if (tcp_ao_ignore_icmp(sk, AF_INET, type, code)) {
519 sock_put(sk);
520 return 0;
521 }
522
523 bh_lock_sock(sk);
524 /* If too many ICMPs get dropped on busy
525 * servers this needs to be solved differently.
526 * We do take care of PMTU discovery (RFC1191) special case :
527 * we can receive locally generated ICMP messages while socket is held.
528 */
529 if (sock_owned_by_user(sk)) {
530 if (!(type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED))
531 __NET_INC_STATS(net, LINUX_MIB_LOCKDROPPEDICMPS);
532 }
533 if (sk->sk_state == TCP_CLOSE)
534 goto out;
535
536 if (static_branch_unlikely(&ip4_min_ttl)) {
537 /* min_ttl can be changed concurrently from do_ip_setsockopt() */
538 if (unlikely(iph->ttl < READ_ONCE(inet_sk(sk)->min_ttl))) {
539 __NET_INC_STATS(net, LINUX_MIB_TCPMINTTLDROP);
540 goto out;
541 }
542 }
543
544 tp = tcp_sk(sk);
545 /* XXX (TFO) - tp->snd_una should be ISN (tcp_create_openreq_child() */
546 fastopen = rcu_dereference(tp->fastopen_rsk);
547 snd_una = fastopen ? tcp_rsk(fastopen)->snt_isn : tp->snd_una;
548 if (sk->sk_state != TCP_LISTEN &&
549 !between(seq, snd_una, tp->snd_nxt)) {
550 __NET_INC_STATS(net, LINUX_MIB_OUTOFWINDOWICMPS);
551 goto out;
552 }
553
554 switch (type) {
555 case ICMP_REDIRECT:
556 if (!sock_owned_by_user(sk))
557 do_redirect(skb, sk);
558 goto out;
559 case ICMP_SOURCE_QUENCH:
560 /* Just silently ignore these. */
561 goto out;
562 case ICMP_PARAMETERPROB:
563 err = EPROTO;
564 break;
565 case ICMP_DEST_UNREACH:
566 if (code > NR_ICMP_UNREACH)
567 goto out;
568
569 if (code == ICMP_FRAG_NEEDED) { /* PMTU discovery (RFC1191) */
570 /* We are not interested in TCP_LISTEN and open_requests
571 * (SYN-ACKs send out by Linux are always <576bytes so
572 * they should go through unfragmented).
573 */
574 if (sk->sk_state == TCP_LISTEN)
575 goto out;
576
577 WRITE_ONCE(tp->mtu_info, info);
578 if (!sock_owned_by_user(sk)) {
579 tcp_v4_mtu_reduced(sk);
580 } else {
581 if (!test_and_set_bit(TCP_MTU_REDUCED_DEFERRED, &sk->sk_tsq_flags))
582 sock_hold(sk);
583 }
584 goto out;
585 }
586
587 err = icmp_err_convert[code].errno;
588 /* check if this ICMP message allows revert of backoff.
589 * (see RFC 6069)
590 */
591 if (!fastopen &&
592 (code == ICMP_NET_UNREACH || code == ICMP_HOST_UNREACH))
593 tcp_ld_RTO_revert(sk, seq);
594 break;
595 case ICMP_TIME_EXCEEDED:
596 err = EHOSTUNREACH;
597 break;
598 default:
599 goto out;
600 }
601
602 switch (sk->sk_state) {
603 case TCP_SYN_SENT:
604 case TCP_SYN_RECV:
605 /* Only in fast or simultaneous open. If a fast open socket is
606 * already accepted it is treated as a connected one below.
607 */
608 if (fastopen && !fastopen->sk)
609 break;
610
611 ip_icmp_error(sk, skb, err, th->dest, info, (u8 *)th);
612
613 if (!sock_owned_by_user(sk)) {
614 WRITE_ONCE(sk->sk_err, err);
615
616 sk_error_report(sk);
617
618 tcp_done(sk);
619 } else {
620 WRITE_ONCE(sk->sk_err_soft, err);
621 }
622 goto out;
623 }
624
625 /* If we've already connected we will keep trying
626 * until we time out, or the user gives up.
627 *
628 * rfc1122 4.2.3.9 allows to consider as hard errors
629 * only PROTO_UNREACH and PORT_UNREACH (well, FRAG_FAILED too,
630 * but it is obsoleted by pmtu discovery).
631 *
632 * Note, that in modern internet, where routing is unreliable
633 * and in each dark corner broken firewalls sit, sending random
634 * errors ordered by their masters even this two messages finally lose
635 * their original sense (even Linux sends invalid PORT_UNREACHs)
636 *
637 * Now we are in compliance with RFCs.
638 * --ANK (980905)
639 */
640
641 if (!sock_owned_by_user(sk) &&
642 inet_test_bit(RECVERR, sk)) {
643 WRITE_ONCE(sk->sk_err, err);
644 sk_error_report(sk);
645 } else { /* Only an error on timeout */
646 WRITE_ONCE(sk->sk_err_soft, err);
647 }
648
649out:
650 bh_unlock_sock(sk);
651 sock_put(sk);
652 return 0;
653}
654
655void __tcp_v4_send_check(struct sk_buff *skb, __be32 saddr, __be32 daddr)
656{
657 struct tcphdr *th = tcp_hdr(skb);
658
659 th->check = ~tcp_v4_check(skb->len, saddr, daddr, 0);
660 skb->csum_start = skb_transport_header(skb) - skb->head;
661 skb->csum_offset = offsetof(struct tcphdr, check);
662}
663
664/* This routine computes an IPv4 TCP checksum. */
665void tcp_v4_send_check(struct sock *sk, struct sk_buff *skb)
666{
667 const struct inet_sock *inet = inet_sk(sk);
668
669 __tcp_v4_send_check(skb, inet->inet_saddr, inet->inet_daddr);
670}
671EXPORT_SYMBOL(tcp_v4_send_check);
672
673#define REPLY_OPTIONS_LEN (MAX_TCP_OPTION_SPACE / sizeof(__be32))
674
675static bool tcp_v4_ao_sign_reset(const struct sock *sk, struct sk_buff *skb,
676 const struct tcp_ao_hdr *aoh,
677 struct ip_reply_arg *arg, struct tcphdr *reply,
678 __be32 reply_options[REPLY_OPTIONS_LEN])
679{
680#ifdef CONFIG_TCP_AO
681 int sdif = tcp_v4_sdif(skb);
682 int dif = inet_iif(skb);
683 int l3index = sdif ? dif : 0;
684 bool allocated_traffic_key;
685 struct tcp_ao_key *key;
686 char *traffic_key;
687 bool drop = true;
688 u32 ao_sne = 0;
689 u8 keyid;
690
691 rcu_read_lock();
692 if (tcp_ao_prepare_reset(sk, skb, aoh, l3index, ntohl(reply->seq),
693 &key, &traffic_key, &allocated_traffic_key,
694 &keyid, &ao_sne))
695 goto out;
696
697 reply_options[0] = htonl((TCPOPT_AO << 24) | (tcp_ao_len(key) << 16) |
698 (aoh->rnext_keyid << 8) | keyid);
699 arg->iov[0].iov_len += tcp_ao_len_aligned(key);
700 reply->doff = arg->iov[0].iov_len / 4;
701
702 if (tcp_ao_hash_hdr(AF_INET, (char *)&reply_options[1],
703 key, traffic_key,
704 (union tcp_ao_addr *)&ip_hdr(skb)->saddr,
705 (union tcp_ao_addr *)&ip_hdr(skb)->daddr,
706 reply, ao_sne))
707 goto out;
708 drop = false;
709out:
710 rcu_read_unlock();
711 if (allocated_traffic_key)
712 kfree(traffic_key);
713 return drop;
714#else
715 return true;
716#endif
717}
718
719/*
720 * This routine will send an RST to the other tcp.
721 *
722 * Someone asks: why I NEVER use socket parameters (TOS, TTL etc.)
723 * for reset.
724 * Answer: if a packet caused RST, it is not for a socket
725 * existing in our system, if it is matched to a socket,
726 * it is just duplicate segment or bug in other side's TCP.
727 * So that we build reply only basing on parameters
728 * arrived with segment.
729 * Exception: precedence violation. We do not implement it in any case.
730 */
731
732static void tcp_v4_send_reset(const struct sock *sk, struct sk_buff *skb)
733{
734 const struct tcphdr *th = tcp_hdr(skb);
735 struct {
736 struct tcphdr th;
737 __be32 opt[REPLY_OPTIONS_LEN];
738 } rep;
739 const __u8 *md5_hash_location = NULL;
740 const struct tcp_ao_hdr *aoh;
741 struct ip_reply_arg arg;
742#ifdef CONFIG_TCP_MD5SIG
743 struct tcp_md5sig_key *key = NULL;
744 unsigned char newhash[16];
745 struct sock *sk1 = NULL;
746 int genhash;
747#endif
748 u64 transmit_time = 0;
749 struct sock *ctl_sk;
750 struct net *net;
751 u32 txhash = 0;
752
753 /* Never send a reset in response to a reset. */
754 if (th->rst)
755 return;
756
757 /* If sk not NULL, it means we did a successful lookup and incoming
758 * route had to be correct. prequeue might have dropped our dst.
759 */
760 if (!sk && skb_rtable(skb)->rt_type != RTN_LOCAL)
761 return;
762
763 /* Swap the send and the receive. */
764 memset(&rep, 0, sizeof(rep));
765 rep.th.dest = th->source;
766 rep.th.source = th->dest;
767 rep.th.doff = sizeof(struct tcphdr) / 4;
768 rep.th.rst = 1;
769
770 if (th->ack) {
771 rep.th.seq = th->ack_seq;
772 } else {
773 rep.th.ack = 1;
774 rep.th.ack_seq = htonl(ntohl(th->seq) + th->syn + th->fin +
775 skb->len - (th->doff << 2));
776 }
777
778 memset(&arg, 0, sizeof(arg));
779 arg.iov[0].iov_base = (unsigned char *)&rep;
780 arg.iov[0].iov_len = sizeof(rep.th);
781
782 net = sk ? sock_net(sk) : dev_net(skb_dst(skb)->dev);
783
784 /* Invalid TCP option size or twice included auth */
785 if (tcp_parse_auth_options(tcp_hdr(skb), &md5_hash_location, &aoh))
786 return;
787
788 if (aoh && tcp_v4_ao_sign_reset(sk, skb, aoh, &arg, &rep.th, rep.opt))
789 return;
790
791#ifdef CONFIG_TCP_MD5SIG
792 rcu_read_lock();
793 if (sk && sk_fullsock(sk)) {
794 const union tcp_md5_addr *addr;
795 int l3index;
796
797 /* sdif set, means packet ingressed via a device
798 * in an L3 domain and inet_iif is set to it.
799 */
800 l3index = tcp_v4_sdif(skb) ? inet_iif(skb) : 0;
801 addr = (union tcp_md5_addr *)&ip_hdr(skb)->saddr;
802 key = tcp_md5_do_lookup(sk, l3index, addr, AF_INET);
803 } else if (md5_hash_location) {
804 const union tcp_md5_addr *addr;
805 int sdif = tcp_v4_sdif(skb);
806 int dif = inet_iif(skb);
807 int l3index;
808
809 /*
810 * active side is lost. Try to find listening socket through
811 * source port, and then find md5 key through listening socket.
812 * we are not loose security here:
813 * Incoming packet is checked with md5 hash with finding key,
814 * no RST generated if md5 hash doesn't match.
815 */
816 sk1 = __inet_lookup_listener(net, net->ipv4.tcp_death_row.hashinfo,
817 NULL, 0, ip_hdr(skb)->saddr,
818 th->source, ip_hdr(skb)->daddr,
819 ntohs(th->source), dif, sdif);
820 /* don't send rst if it can't find key */
821 if (!sk1)
822 goto out;
823
824 /* sdif set, means packet ingressed via a device
825 * in an L3 domain and dif is set to it.
826 */
827 l3index = sdif ? dif : 0;
828 addr = (union tcp_md5_addr *)&ip_hdr(skb)->saddr;
829 key = tcp_md5_do_lookup(sk1, l3index, addr, AF_INET);
830 if (!key)
831 goto out;
832
833
834 genhash = tcp_v4_md5_hash_skb(newhash, key, NULL, skb);
835 if (genhash || memcmp(md5_hash_location, newhash, 16) != 0)
836 goto out;
837
838 }
839
840 if (key) {
841 rep.opt[0] = htonl((TCPOPT_NOP << 24) |
842 (TCPOPT_NOP << 16) |
843 (TCPOPT_MD5SIG << 8) |
844 TCPOLEN_MD5SIG);
845 /* Update length and the length the header thinks exists */
846 arg.iov[0].iov_len += TCPOLEN_MD5SIG_ALIGNED;
847 rep.th.doff = arg.iov[0].iov_len / 4;
848
849 tcp_v4_md5_hash_hdr((__u8 *) &rep.opt[1],
850 key, ip_hdr(skb)->saddr,
851 ip_hdr(skb)->daddr, &rep.th);
852 }
853#endif
854 /* Can't co-exist with TCPMD5, hence check rep.opt[0] */
855 if (rep.opt[0] == 0) {
856 __be32 mrst = mptcp_reset_option(skb);
857
858 if (mrst) {
859 rep.opt[0] = mrst;
860 arg.iov[0].iov_len += sizeof(mrst);
861 rep.th.doff = arg.iov[0].iov_len / 4;
862 }
863 }
864
865 arg.csum = csum_tcpudp_nofold(ip_hdr(skb)->daddr,
866 ip_hdr(skb)->saddr, /* XXX */
867 arg.iov[0].iov_len, IPPROTO_TCP, 0);
868 arg.csumoffset = offsetof(struct tcphdr, check) / 2;
869 arg.flags = (sk && inet_sk_transparent(sk)) ? IP_REPLY_ARG_NOSRCCHECK : 0;
870
871 /* When socket is gone, all binding information is lost.
872 * routing might fail in this case. No choice here, if we choose to force
873 * input interface, we will misroute in case of asymmetric route.
874 */
875 if (sk) {
876 arg.bound_dev_if = sk->sk_bound_dev_if;
877 if (sk_fullsock(sk))
878 trace_tcp_send_reset(sk, skb);
879 }
880
881 BUILD_BUG_ON(offsetof(struct sock, sk_bound_dev_if) !=
882 offsetof(struct inet_timewait_sock, tw_bound_dev_if));
883
884 arg.tos = ip_hdr(skb)->tos;
885 arg.uid = sock_net_uid(net, sk && sk_fullsock(sk) ? sk : NULL);
886 local_bh_disable();
887 ctl_sk = this_cpu_read(ipv4_tcp_sk);
888 sock_net_set(ctl_sk, net);
889 if (sk) {
890 ctl_sk->sk_mark = (sk->sk_state == TCP_TIME_WAIT) ?
891 inet_twsk(sk)->tw_mark : sk->sk_mark;
892 ctl_sk->sk_priority = (sk->sk_state == TCP_TIME_WAIT) ?
893 inet_twsk(sk)->tw_priority : READ_ONCE(sk->sk_priority);
894 transmit_time = tcp_transmit_time(sk);
895 xfrm_sk_clone_policy(ctl_sk, sk);
896 txhash = (sk->sk_state == TCP_TIME_WAIT) ?
897 inet_twsk(sk)->tw_txhash : sk->sk_txhash;
898 } else {
899 ctl_sk->sk_mark = 0;
900 ctl_sk->sk_priority = 0;
901 }
902 ip_send_unicast_reply(ctl_sk,
903 skb, &TCP_SKB_CB(skb)->header.h4.opt,
904 ip_hdr(skb)->saddr, ip_hdr(skb)->daddr,
905 &arg, arg.iov[0].iov_len,
906 transmit_time, txhash);
907
908 xfrm_sk_free_policy(ctl_sk);
909 sock_net_set(ctl_sk, &init_net);
910 __TCP_INC_STATS(net, TCP_MIB_OUTSEGS);
911 __TCP_INC_STATS(net, TCP_MIB_OUTRSTS);
912 local_bh_enable();
913
914#ifdef CONFIG_TCP_MD5SIG
915out:
916 rcu_read_unlock();
917#endif
918}
919
920/* The code following below sending ACKs in SYN-RECV and TIME-WAIT states
921 outside socket context is ugly, certainly. What can I do?
922 */
923
924static void tcp_v4_send_ack(const struct sock *sk,
925 struct sk_buff *skb, u32 seq, u32 ack,
926 u32 win, u32 tsval, u32 tsecr, int oif,
927 struct tcp_key *key,
928 int reply_flags, u8 tos, u32 txhash)
929{
930 const struct tcphdr *th = tcp_hdr(skb);
931 struct {
932 struct tcphdr th;
933 __be32 opt[(MAX_TCP_OPTION_SPACE >> 2)];
934 } rep;
935 struct net *net = sock_net(sk);
936 struct ip_reply_arg arg;
937 struct sock *ctl_sk;
938 u64 transmit_time;
939
940 memset(&rep.th, 0, sizeof(struct tcphdr));
941 memset(&arg, 0, sizeof(arg));
942
943 arg.iov[0].iov_base = (unsigned char *)&rep;
944 arg.iov[0].iov_len = sizeof(rep.th);
945 if (tsecr) {
946 rep.opt[0] = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
947 (TCPOPT_TIMESTAMP << 8) |
948 TCPOLEN_TIMESTAMP);
949 rep.opt[1] = htonl(tsval);
950 rep.opt[2] = htonl(tsecr);
951 arg.iov[0].iov_len += TCPOLEN_TSTAMP_ALIGNED;
952 }
953
954 /* Swap the send and the receive. */
955 rep.th.dest = th->source;
956 rep.th.source = th->dest;
957 rep.th.doff = arg.iov[0].iov_len / 4;
958 rep.th.seq = htonl(seq);
959 rep.th.ack_seq = htonl(ack);
960 rep.th.ack = 1;
961 rep.th.window = htons(win);
962
963#ifdef CONFIG_TCP_MD5SIG
964 if (tcp_key_is_md5(key)) {
965 int offset = (tsecr) ? 3 : 0;
966
967 rep.opt[offset++] = htonl((TCPOPT_NOP << 24) |
968 (TCPOPT_NOP << 16) |
969 (TCPOPT_MD5SIG << 8) |
970 TCPOLEN_MD5SIG);
971 arg.iov[0].iov_len += TCPOLEN_MD5SIG_ALIGNED;
972 rep.th.doff = arg.iov[0].iov_len/4;
973
974 tcp_v4_md5_hash_hdr((__u8 *) &rep.opt[offset],
975 key->md5_key, ip_hdr(skb)->saddr,
976 ip_hdr(skb)->daddr, &rep.th);
977 }
978#endif
979#ifdef CONFIG_TCP_AO
980 if (tcp_key_is_ao(key)) {
981 int offset = (tsecr) ? 3 : 0;
982
983 rep.opt[offset++] = htonl((TCPOPT_AO << 24) |
984 (tcp_ao_len(key->ao_key) << 16) |
985 (key->ao_key->sndid << 8) |
986 key->rcv_next);
987 arg.iov[0].iov_len += tcp_ao_len_aligned(key->ao_key);
988 rep.th.doff = arg.iov[0].iov_len / 4;
989
990 tcp_ao_hash_hdr(AF_INET, (char *)&rep.opt[offset],
991 key->ao_key, key->traffic_key,
992 (union tcp_ao_addr *)&ip_hdr(skb)->saddr,
993 (union tcp_ao_addr *)&ip_hdr(skb)->daddr,
994 &rep.th, key->sne);
995 }
996#endif
997 arg.flags = reply_flags;
998 arg.csum = csum_tcpudp_nofold(ip_hdr(skb)->daddr,
999 ip_hdr(skb)->saddr, /* XXX */
1000 arg.iov[0].iov_len, IPPROTO_TCP, 0);
1001 arg.csumoffset = offsetof(struct tcphdr, check) / 2;
1002 if (oif)
1003 arg.bound_dev_if = oif;
1004 arg.tos = tos;
1005 arg.uid = sock_net_uid(net, sk_fullsock(sk) ? sk : NULL);
1006 local_bh_disable();
1007 ctl_sk = this_cpu_read(ipv4_tcp_sk);
1008 sock_net_set(ctl_sk, net);
1009 ctl_sk->sk_mark = (sk->sk_state == TCP_TIME_WAIT) ?
1010 inet_twsk(sk)->tw_mark : READ_ONCE(sk->sk_mark);
1011 ctl_sk->sk_priority = (sk->sk_state == TCP_TIME_WAIT) ?
1012 inet_twsk(sk)->tw_priority : READ_ONCE(sk->sk_priority);
1013 transmit_time = tcp_transmit_time(sk);
1014 ip_send_unicast_reply(ctl_sk,
1015 skb, &TCP_SKB_CB(skb)->header.h4.opt,
1016 ip_hdr(skb)->saddr, ip_hdr(skb)->daddr,
1017 &arg, arg.iov[0].iov_len,
1018 transmit_time, txhash);
1019
1020 sock_net_set(ctl_sk, &init_net);
1021 __TCP_INC_STATS(net, TCP_MIB_OUTSEGS);
1022 local_bh_enable();
1023}
1024
1025static void tcp_v4_timewait_ack(struct sock *sk, struct sk_buff *skb)
1026{
1027 struct inet_timewait_sock *tw = inet_twsk(sk);
1028 struct tcp_timewait_sock *tcptw = tcp_twsk(sk);
1029 struct tcp_key key = {};
1030#ifdef CONFIG_TCP_AO
1031 struct tcp_ao_info *ao_info;
1032
1033 if (static_branch_unlikely(&tcp_ao_needed.key)) {
1034 /* FIXME: the segment to-be-acked is not verified yet */
1035 ao_info = rcu_dereference(tcptw->ao_info);
1036 if (ao_info) {
1037 const struct tcp_ao_hdr *aoh;
1038
1039 if (tcp_parse_auth_options(tcp_hdr(skb), NULL, &aoh)) {
1040 inet_twsk_put(tw);
1041 return;
1042 }
1043
1044 if (aoh)
1045 key.ao_key = tcp_ao_established_key(ao_info, aoh->rnext_keyid, -1);
1046 }
1047 }
1048 if (key.ao_key) {
1049 struct tcp_ao_key *rnext_key;
1050
1051 key.traffic_key = snd_other_key(key.ao_key);
1052 key.sne = READ_ONCE(ao_info->snd_sne);
1053 rnext_key = READ_ONCE(ao_info->rnext_key);
1054 key.rcv_next = rnext_key->rcvid;
1055 key.type = TCP_KEY_AO;
1056#else
1057 if (0) {
1058#endif
1059#ifdef CONFIG_TCP_MD5SIG
1060 } else if (static_branch_unlikely(&tcp_md5_needed.key)) {
1061 key.md5_key = tcp_twsk_md5_key(tcptw);
1062 if (key.md5_key)
1063 key.type = TCP_KEY_MD5;
1064#endif
1065 }
1066
1067 tcp_v4_send_ack(sk, skb,
1068 tcptw->tw_snd_nxt, tcptw->tw_rcv_nxt,
1069 tcptw->tw_rcv_wnd >> tw->tw_rcv_wscale,
1070 tcp_tw_tsval(tcptw),
1071 tcptw->tw_ts_recent,
1072 tw->tw_bound_dev_if, &key,
1073 tw->tw_transparent ? IP_REPLY_ARG_NOSRCCHECK : 0,
1074 tw->tw_tos,
1075 tw->tw_txhash);
1076
1077 inet_twsk_put(tw);
1078}
1079
1080static void tcp_v4_reqsk_send_ack(const struct sock *sk, struct sk_buff *skb,
1081 struct request_sock *req)
1082{
1083 struct tcp_key key = {};
1084
1085 /* sk->sk_state == TCP_LISTEN -> for regular TCP_SYN_RECV
1086 * sk->sk_state == TCP_SYN_RECV -> for Fast Open.
1087 */
1088 u32 seq = (sk->sk_state == TCP_LISTEN) ? tcp_rsk(req)->snt_isn + 1 :
1089 tcp_sk(sk)->snd_nxt;
1090
1091#ifdef CONFIG_TCP_AO
1092 if (static_branch_unlikely(&tcp_ao_needed.key) &&
1093 tcp_rsk_used_ao(req)) {
1094 const union tcp_md5_addr *addr;
1095 const struct tcp_ao_hdr *aoh;
1096 int l3index;
1097
1098 /* Invalid TCP option size or twice included auth */
1099 if (tcp_parse_auth_options(tcp_hdr(skb), NULL, &aoh))
1100 return;
1101 if (!aoh)
1102 return;
1103
1104 addr = (union tcp_md5_addr *)&ip_hdr(skb)->saddr;
1105 l3index = tcp_v4_sdif(skb) ? inet_iif(skb) : 0;
1106 key.ao_key = tcp_ao_do_lookup(sk, l3index, addr, AF_INET,
1107 aoh->rnext_keyid, -1);
1108 if (unlikely(!key.ao_key)) {
1109 /* Send ACK with any matching MKT for the peer */
1110 key.ao_key = tcp_ao_do_lookup(sk, l3index, addr, AF_INET, -1, -1);
1111 /* Matching key disappeared (user removed the key?)
1112 * let the handshake timeout.
1113 */
1114 if (!key.ao_key) {
1115 net_info_ratelimited("TCP-AO key for (%pI4, %d)->(%pI4, %d) suddenly disappeared, won't ACK new connection\n",
1116 addr,
1117 ntohs(tcp_hdr(skb)->source),
1118 &ip_hdr(skb)->daddr,
1119 ntohs(tcp_hdr(skb)->dest));
1120 return;
1121 }
1122 }
1123 key.traffic_key = kmalloc(tcp_ao_digest_size(key.ao_key), GFP_ATOMIC);
1124 if (!key.traffic_key)
1125 return;
1126
1127 key.type = TCP_KEY_AO;
1128 key.rcv_next = aoh->keyid;
1129 tcp_v4_ao_calc_key_rsk(key.ao_key, key.traffic_key, req);
1130#else
1131 if (0) {
1132#endif
1133#ifdef CONFIG_TCP_MD5SIG
1134 } else if (static_branch_unlikely(&tcp_md5_needed.key)) {
1135 const union tcp_md5_addr *addr;
1136 int l3index;
1137
1138 addr = (union tcp_md5_addr *)&ip_hdr(skb)->saddr;
1139 l3index = tcp_v4_sdif(skb) ? inet_iif(skb) : 0;
1140 key.md5_key = tcp_md5_do_lookup(sk, l3index, addr, AF_INET);
1141 if (key.md5_key)
1142 key.type = TCP_KEY_MD5;
1143#endif
1144 }
1145
1146 /* RFC 7323 2.3
1147 * The window field (SEG.WND) of every outgoing segment, with the
1148 * exception of <SYN> segments, MUST be right-shifted by
1149 * Rcv.Wind.Shift bits:
1150 */
1151 tcp_v4_send_ack(sk, skb, seq,
1152 tcp_rsk(req)->rcv_nxt,
1153 req->rsk_rcv_wnd >> inet_rsk(req)->rcv_wscale,
1154 tcp_rsk_tsval(tcp_rsk(req)),
1155 READ_ONCE(req->ts_recent),
1156 0, &key,
1157 inet_rsk(req)->no_srccheck ? IP_REPLY_ARG_NOSRCCHECK : 0,
1158 ip_hdr(skb)->tos,
1159 READ_ONCE(tcp_rsk(req)->txhash));
1160 if (tcp_key_is_ao(&key))
1161 kfree(key.traffic_key);
1162}
1163
1164/*
1165 * Send a SYN-ACK after having received a SYN.
1166 * This still operates on a request_sock only, not on a big
1167 * socket.
1168 */
1169static int tcp_v4_send_synack(const struct sock *sk, struct dst_entry *dst,
1170 struct flowi *fl,
1171 struct request_sock *req,
1172 struct tcp_fastopen_cookie *foc,
1173 enum tcp_synack_type synack_type,
1174 struct sk_buff *syn_skb)
1175{
1176 const struct inet_request_sock *ireq = inet_rsk(req);
1177 struct flowi4 fl4;
1178 int err = -1;
1179 struct sk_buff *skb;
1180 u8 tos;
1181
1182 /* First, grab a route. */
1183 if (!dst && (dst = inet_csk_route_req(sk, &fl4, req)) == NULL)
1184 return -1;
1185
1186 skb = tcp_make_synack(sk, dst, req, foc, synack_type, syn_skb);
1187
1188 if (skb) {
1189 __tcp_v4_send_check(skb, ireq->ir_loc_addr, ireq->ir_rmt_addr);
1190
1191 tos = READ_ONCE(inet_sk(sk)->tos);
1192
1193 if (READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_reflect_tos))
1194 tos = (tcp_rsk(req)->syn_tos & ~INET_ECN_MASK) |
1195 (tos & INET_ECN_MASK);
1196
1197 if (!INET_ECN_is_capable(tos) &&
1198 tcp_bpf_ca_needs_ecn((struct sock *)req))
1199 tos |= INET_ECN_ECT_0;
1200
1201 rcu_read_lock();
1202 err = ip_build_and_send_pkt(skb, sk, ireq->ir_loc_addr,
1203 ireq->ir_rmt_addr,
1204 rcu_dereference(ireq->ireq_opt),
1205 tos);
1206 rcu_read_unlock();
1207 err = net_xmit_eval(err);
1208 }
1209
1210 return err;
1211}
1212
1213/*
1214 * IPv4 request_sock destructor.
1215 */
1216static void tcp_v4_reqsk_destructor(struct request_sock *req)
1217{
1218 kfree(rcu_dereference_protected(inet_rsk(req)->ireq_opt, 1));
1219}
1220
1221#ifdef CONFIG_TCP_MD5SIG
1222/*
1223 * RFC2385 MD5 checksumming requires a mapping of
1224 * IP address->MD5 Key.
1225 * We need to maintain these in the sk structure.
1226 */
1227
1228DEFINE_STATIC_KEY_DEFERRED_FALSE(tcp_md5_needed, HZ);
1229EXPORT_SYMBOL(tcp_md5_needed);
1230
1231static bool better_md5_match(struct tcp_md5sig_key *old, struct tcp_md5sig_key *new)
1232{
1233 if (!old)
1234 return true;
1235
1236 /* l3index always overrides non-l3index */
1237 if (old->l3index && new->l3index == 0)
1238 return false;
1239 if (old->l3index == 0 && new->l3index)
1240 return true;
1241
1242 return old->prefixlen < new->prefixlen;
1243}
1244
1245/* Find the Key structure for an address. */
1246struct tcp_md5sig_key *__tcp_md5_do_lookup(const struct sock *sk, int l3index,
1247 const union tcp_md5_addr *addr,
1248 int family, bool any_l3index)
1249{
1250 const struct tcp_sock *tp = tcp_sk(sk);
1251 struct tcp_md5sig_key *key;
1252 const struct tcp_md5sig_info *md5sig;
1253 __be32 mask;
1254 struct tcp_md5sig_key *best_match = NULL;
1255 bool match;
1256
1257 /* caller either holds rcu_read_lock() or socket lock */
1258 md5sig = rcu_dereference_check(tp->md5sig_info,
1259 lockdep_sock_is_held(sk));
1260 if (!md5sig)
1261 return NULL;
1262
1263 hlist_for_each_entry_rcu(key, &md5sig->head, node,
1264 lockdep_sock_is_held(sk)) {
1265 if (key->family != family)
1266 continue;
1267 if (!any_l3index && key->flags & TCP_MD5SIG_FLAG_IFINDEX &&
1268 key->l3index != l3index)
1269 continue;
1270 if (family == AF_INET) {
1271 mask = inet_make_mask(key->prefixlen);
1272 match = (key->addr.a4.s_addr & mask) ==
1273 (addr->a4.s_addr & mask);
1274#if IS_ENABLED(CONFIG_IPV6)
1275 } else if (family == AF_INET6) {
1276 match = ipv6_prefix_equal(&key->addr.a6, &addr->a6,
1277 key->prefixlen);
1278#endif
1279 } else {
1280 match = false;
1281 }
1282
1283 if (match && better_md5_match(best_match, key))
1284 best_match = key;
1285 }
1286 return best_match;
1287}
1288EXPORT_SYMBOL(__tcp_md5_do_lookup);
1289
1290static struct tcp_md5sig_key *tcp_md5_do_lookup_exact(const struct sock *sk,
1291 const union tcp_md5_addr *addr,
1292 int family, u8 prefixlen,
1293 int l3index, u8 flags)
1294{
1295 const struct tcp_sock *tp = tcp_sk(sk);
1296 struct tcp_md5sig_key *key;
1297 unsigned int size = sizeof(struct in_addr);
1298 const struct tcp_md5sig_info *md5sig;
1299
1300 /* caller either holds rcu_read_lock() or socket lock */
1301 md5sig = rcu_dereference_check(tp->md5sig_info,
1302 lockdep_sock_is_held(sk));
1303 if (!md5sig)
1304 return NULL;
1305#if IS_ENABLED(CONFIG_IPV6)
1306 if (family == AF_INET6)
1307 size = sizeof(struct in6_addr);
1308#endif
1309 hlist_for_each_entry_rcu(key, &md5sig->head, node,
1310 lockdep_sock_is_held(sk)) {
1311 if (key->family != family)
1312 continue;
1313 if ((key->flags & TCP_MD5SIG_FLAG_IFINDEX) != (flags & TCP_MD5SIG_FLAG_IFINDEX))
1314 continue;
1315 if (key->l3index != l3index)
1316 continue;
1317 if (!memcmp(&key->addr, addr, size) &&
1318 key->prefixlen == prefixlen)
1319 return key;
1320 }
1321 return NULL;
1322}
1323
1324struct tcp_md5sig_key *tcp_v4_md5_lookup(const struct sock *sk,
1325 const struct sock *addr_sk)
1326{
1327 const union tcp_md5_addr *addr;
1328 int l3index;
1329
1330 l3index = l3mdev_master_ifindex_by_index(sock_net(sk),
1331 addr_sk->sk_bound_dev_if);
1332 addr = (const union tcp_md5_addr *)&addr_sk->sk_daddr;
1333 return tcp_md5_do_lookup(sk, l3index, addr, AF_INET);
1334}
1335EXPORT_SYMBOL(tcp_v4_md5_lookup);
1336
1337static int tcp_md5sig_info_add(struct sock *sk, gfp_t gfp)
1338{
1339 struct tcp_sock *tp = tcp_sk(sk);
1340 struct tcp_md5sig_info *md5sig;
1341
1342 md5sig = kmalloc(sizeof(*md5sig), gfp);
1343 if (!md5sig)
1344 return -ENOMEM;
1345
1346 sk_gso_disable(sk);
1347 INIT_HLIST_HEAD(&md5sig->head);
1348 rcu_assign_pointer(tp->md5sig_info, md5sig);
1349 return 0;
1350}
1351
1352/* This can be called on a newly created socket, from other files */
1353static int __tcp_md5_do_add(struct sock *sk, const union tcp_md5_addr *addr,
1354 int family, u8 prefixlen, int l3index, u8 flags,
1355 const u8 *newkey, u8 newkeylen, gfp_t gfp)
1356{
1357 /* Add Key to the list */
1358 struct tcp_md5sig_key *key;
1359 struct tcp_sock *tp = tcp_sk(sk);
1360 struct tcp_md5sig_info *md5sig;
1361
1362 key = tcp_md5_do_lookup_exact(sk, addr, family, prefixlen, l3index, flags);
1363 if (key) {
1364 /* Pre-existing entry - just update that one.
1365 * Note that the key might be used concurrently.
1366 * data_race() is telling kcsan that we do not care of
1367 * key mismatches, since changing MD5 key on live flows
1368 * can lead to packet drops.
1369 */
1370 data_race(memcpy(key->key, newkey, newkeylen));
1371
1372 /* Pairs with READ_ONCE() in tcp_md5_hash_key().
1373 * Also note that a reader could catch new key->keylen value
1374 * but old key->key[], this is the reason we use __GFP_ZERO
1375 * at sock_kmalloc() time below these lines.
1376 */
1377 WRITE_ONCE(key->keylen, newkeylen);
1378
1379 return 0;
1380 }
1381
1382 md5sig = rcu_dereference_protected(tp->md5sig_info,
1383 lockdep_sock_is_held(sk));
1384
1385 key = sock_kmalloc(sk, sizeof(*key), gfp | __GFP_ZERO);
1386 if (!key)
1387 return -ENOMEM;
1388
1389 memcpy(key->key, newkey, newkeylen);
1390 key->keylen = newkeylen;
1391 key->family = family;
1392 key->prefixlen = prefixlen;
1393 key->l3index = l3index;
1394 key->flags = flags;
1395 memcpy(&key->addr, addr,
1396 (IS_ENABLED(CONFIG_IPV6) && family == AF_INET6) ? sizeof(struct in6_addr) :
1397 sizeof(struct in_addr));
1398 hlist_add_head_rcu(&key->node, &md5sig->head);
1399 return 0;
1400}
1401
1402int tcp_md5_do_add(struct sock *sk, const union tcp_md5_addr *addr,
1403 int family, u8 prefixlen, int l3index, u8 flags,
1404 const u8 *newkey, u8 newkeylen)
1405{
1406 struct tcp_sock *tp = tcp_sk(sk);
1407
1408 if (!rcu_dereference_protected(tp->md5sig_info, lockdep_sock_is_held(sk))) {
1409 if (tcp_md5_alloc_sigpool())
1410 return -ENOMEM;
1411
1412 if (tcp_md5sig_info_add(sk, GFP_KERNEL)) {
1413 tcp_md5_release_sigpool();
1414 return -ENOMEM;
1415 }
1416
1417 if (!static_branch_inc(&tcp_md5_needed.key)) {
1418 struct tcp_md5sig_info *md5sig;
1419
1420 md5sig = rcu_dereference_protected(tp->md5sig_info, lockdep_sock_is_held(sk));
1421 rcu_assign_pointer(tp->md5sig_info, NULL);
1422 kfree_rcu(md5sig, rcu);
1423 tcp_md5_release_sigpool();
1424 return -EUSERS;
1425 }
1426 }
1427
1428 return __tcp_md5_do_add(sk, addr, family, prefixlen, l3index, flags,
1429 newkey, newkeylen, GFP_KERNEL);
1430}
1431EXPORT_SYMBOL(tcp_md5_do_add);
1432
1433int tcp_md5_key_copy(struct sock *sk, const union tcp_md5_addr *addr,
1434 int family, u8 prefixlen, int l3index,
1435 struct tcp_md5sig_key *key)
1436{
1437 struct tcp_sock *tp = tcp_sk(sk);
1438
1439 if (!rcu_dereference_protected(tp->md5sig_info, lockdep_sock_is_held(sk))) {
1440 tcp_md5_add_sigpool();
1441
1442 if (tcp_md5sig_info_add(sk, sk_gfp_mask(sk, GFP_ATOMIC))) {
1443 tcp_md5_release_sigpool();
1444 return -ENOMEM;
1445 }
1446
1447 if (!static_key_fast_inc_not_disabled(&tcp_md5_needed.key.key)) {
1448 struct tcp_md5sig_info *md5sig;
1449
1450 md5sig = rcu_dereference_protected(tp->md5sig_info, lockdep_sock_is_held(sk));
1451 net_warn_ratelimited("Too many TCP-MD5 keys in the system\n");
1452 rcu_assign_pointer(tp->md5sig_info, NULL);
1453 kfree_rcu(md5sig, rcu);
1454 tcp_md5_release_sigpool();
1455 return -EUSERS;
1456 }
1457 }
1458
1459 return __tcp_md5_do_add(sk, addr, family, prefixlen, l3index,
1460 key->flags, key->key, key->keylen,
1461 sk_gfp_mask(sk, GFP_ATOMIC));
1462}
1463EXPORT_SYMBOL(tcp_md5_key_copy);
1464
1465int tcp_md5_do_del(struct sock *sk, const union tcp_md5_addr *addr, int family,
1466 u8 prefixlen, int l3index, u8 flags)
1467{
1468 struct tcp_md5sig_key *key;
1469
1470 key = tcp_md5_do_lookup_exact(sk, addr, family, prefixlen, l3index, flags);
1471 if (!key)
1472 return -ENOENT;
1473 hlist_del_rcu(&key->node);
1474 atomic_sub(sizeof(*key), &sk->sk_omem_alloc);
1475 kfree_rcu(key, rcu);
1476 return 0;
1477}
1478EXPORT_SYMBOL(tcp_md5_do_del);
1479
1480void tcp_clear_md5_list(struct sock *sk)
1481{
1482 struct tcp_sock *tp = tcp_sk(sk);
1483 struct tcp_md5sig_key *key;
1484 struct hlist_node *n;
1485 struct tcp_md5sig_info *md5sig;
1486
1487 md5sig = rcu_dereference_protected(tp->md5sig_info, 1);
1488
1489 hlist_for_each_entry_safe(key, n, &md5sig->head, node) {
1490 hlist_del_rcu(&key->node);
1491 atomic_sub(sizeof(*key), &sk->sk_omem_alloc);
1492 kfree_rcu(key, rcu);
1493 }
1494}
1495
1496static int tcp_v4_parse_md5_keys(struct sock *sk, int optname,
1497 sockptr_t optval, int optlen)
1498{
1499 struct tcp_md5sig cmd;
1500 struct sockaddr_in *sin = (struct sockaddr_in *)&cmd.tcpm_addr;
1501 const union tcp_md5_addr *addr;
1502 u8 prefixlen = 32;
1503 int l3index = 0;
1504 bool l3flag;
1505 u8 flags;
1506
1507 if (optlen < sizeof(cmd))
1508 return -EINVAL;
1509
1510 if (copy_from_sockptr(&cmd, optval, sizeof(cmd)))
1511 return -EFAULT;
1512
1513 if (sin->sin_family != AF_INET)
1514 return -EINVAL;
1515
1516 flags = cmd.tcpm_flags & TCP_MD5SIG_FLAG_IFINDEX;
1517 l3flag = cmd.tcpm_flags & TCP_MD5SIG_FLAG_IFINDEX;
1518
1519 if (optname == TCP_MD5SIG_EXT &&
1520 cmd.tcpm_flags & TCP_MD5SIG_FLAG_PREFIX) {
1521 prefixlen = cmd.tcpm_prefixlen;
1522 if (prefixlen > 32)
1523 return -EINVAL;
1524 }
1525
1526 if (optname == TCP_MD5SIG_EXT && cmd.tcpm_ifindex &&
1527 cmd.tcpm_flags & TCP_MD5SIG_FLAG_IFINDEX) {
1528 struct net_device *dev;
1529
1530 rcu_read_lock();
1531 dev = dev_get_by_index_rcu(sock_net(sk), cmd.tcpm_ifindex);
1532 if (dev && netif_is_l3_master(dev))
1533 l3index = dev->ifindex;
1534
1535 rcu_read_unlock();
1536
1537 /* ok to reference set/not set outside of rcu;
1538 * right now device MUST be an L3 master
1539 */
1540 if (!dev || !l3index)
1541 return -EINVAL;
1542 }
1543
1544 addr = (union tcp_md5_addr *)&sin->sin_addr.s_addr;
1545
1546 if (!cmd.tcpm_keylen)
1547 return tcp_md5_do_del(sk, addr, AF_INET, prefixlen, l3index, flags);
1548
1549 if (cmd.tcpm_keylen > TCP_MD5SIG_MAXKEYLEN)
1550 return -EINVAL;
1551
1552 /* Don't allow keys for peers that have a matching TCP-AO key.
1553 * See the comment in tcp_ao_add_cmd()
1554 */
1555 if (tcp_ao_required(sk, addr, AF_INET, l3flag ? l3index : -1, false))
1556 return -EKEYREJECTED;
1557
1558 return tcp_md5_do_add(sk, addr, AF_INET, prefixlen, l3index, flags,
1559 cmd.tcpm_key, cmd.tcpm_keylen);
1560}
1561
1562static int tcp_v4_md5_hash_headers(struct tcp_sigpool *hp,
1563 __be32 daddr, __be32 saddr,
1564 const struct tcphdr *th, int nbytes)
1565{
1566 struct tcp4_pseudohdr *bp;
1567 struct scatterlist sg;
1568 struct tcphdr *_th;
1569
1570 bp = hp->scratch;
1571 bp->saddr = saddr;
1572 bp->daddr = daddr;
1573 bp->pad = 0;
1574 bp->protocol = IPPROTO_TCP;
1575 bp->len = cpu_to_be16(nbytes);
1576
1577 _th = (struct tcphdr *)(bp + 1);
1578 memcpy(_th, th, sizeof(*th));
1579 _th->check = 0;
1580
1581 sg_init_one(&sg, bp, sizeof(*bp) + sizeof(*th));
1582 ahash_request_set_crypt(hp->req, &sg, NULL,
1583 sizeof(*bp) + sizeof(*th));
1584 return crypto_ahash_update(hp->req);
1585}
1586
1587static int tcp_v4_md5_hash_hdr(char *md5_hash, const struct tcp_md5sig_key *key,
1588 __be32 daddr, __be32 saddr, const struct tcphdr *th)
1589{
1590 struct tcp_sigpool hp;
1591
1592 if (tcp_sigpool_start(tcp_md5_sigpool_id, &hp))
1593 goto clear_hash_nostart;
1594
1595 if (crypto_ahash_init(hp.req))
1596 goto clear_hash;
1597 if (tcp_v4_md5_hash_headers(&hp, daddr, saddr, th, th->doff << 2))
1598 goto clear_hash;
1599 if (tcp_md5_hash_key(&hp, key))
1600 goto clear_hash;
1601 ahash_request_set_crypt(hp.req, NULL, md5_hash, 0);
1602 if (crypto_ahash_final(hp.req))
1603 goto clear_hash;
1604
1605 tcp_sigpool_end(&hp);
1606 return 0;
1607
1608clear_hash:
1609 tcp_sigpool_end(&hp);
1610clear_hash_nostart:
1611 memset(md5_hash, 0, 16);
1612 return 1;
1613}
1614
1615int tcp_v4_md5_hash_skb(char *md5_hash, const struct tcp_md5sig_key *key,
1616 const struct sock *sk,
1617 const struct sk_buff *skb)
1618{
1619 const struct tcphdr *th = tcp_hdr(skb);
1620 struct tcp_sigpool hp;
1621 __be32 saddr, daddr;
1622
1623 if (sk) { /* valid for establish/request sockets */
1624 saddr = sk->sk_rcv_saddr;
1625 daddr = sk->sk_daddr;
1626 } else {
1627 const struct iphdr *iph = ip_hdr(skb);
1628 saddr = iph->saddr;
1629 daddr = iph->daddr;
1630 }
1631
1632 if (tcp_sigpool_start(tcp_md5_sigpool_id, &hp))
1633 goto clear_hash_nostart;
1634
1635 if (crypto_ahash_init(hp.req))
1636 goto clear_hash;
1637
1638 if (tcp_v4_md5_hash_headers(&hp, daddr, saddr, th, skb->len))
1639 goto clear_hash;
1640 if (tcp_sigpool_hash_skb_data(&hp, skb, th->doff << 2))
1641 goto clear_hash;
1642 if (tcp_md5_hash_key(&hp, key))
1643 goto clear_hash;
1644 ahash_request_set_crypt(hp.req, NULL, md5_hash, 0);
1645 if (crypto_ahash_final(hp.req))
1646 goto clear_hash;
1647
1648 tcp_sigpool_end(&hp);
1649 return 0;
1650
1651clear_hash:
1652 tcp_sigpool_end(&hp);
1653clear_hash_nostart:
1654 memset(md5_hash, 0, 16);
1655 return 1;
1656}
1657EXPORT_SYMBOL(tcp_v4_md5_hash_skb);
1658
1659#endif
1660
1661static void tcp_v4_init_req(struct request_sock *req,
1662 const struct sock *sk_listener,
1663 struct sk_buff *skb)
1664{
1665 struct inet_request_sock *ireq = inet_rsk(req);
1666 struct net *net = sock_net(sk_listener);
1667
1668 sk_rcv_saddr_set(req_to_sk(req), ip_hdr(skb)->daddr);
1669 sk_daddr_set(req_to_sk(req), ip_hdr(skb)->saddr);
1670 RCU_INIT_POINTER(ireq->ireq_opt, tcp_v4_save_options(net, skb));
1671}
1672
1673static struct dst_entry *tcp_v4_route_req(const struct sock *sk,
1674 struct sk_buff *skb,
1675 struct flowi *fl,
1676 struct request_sock *req)
1677{
1678 tcp_v4_init_req(req, sk, skb);
1679
1680 if (security_inet_conn_request(sk, skb, req))
1681 return NULL;
1682
1683 return inet_csk_route_req(sk, &fl->u.ip4, req);
1684}
1685
1686struct request_sock_ops tcp_request_sock_ops __read_mostly = {
1687 .family = PF_INET,
1688 .obj_size = sizeof(struct tcp_request_sock),
1689 .rtx_syn_ack = tcp_rtx_synack,
1690 .send_ack = tcp_v4_reqsk_send_ack,
1691 .destructor = tcp_v4_reqsk_destructor,
1692 .send_reset = tcp_v4_send_reset,
1693 .syn_ack_timeout = tcp_syn_ack_timeout,
1694};
1695
1696const struct tcp_request_sock_ops tcp_request_sock_ipv4_ops = {
1697 .mss_clamp = TCP_MSS_DEFAULT,
1698#ifdef CONFIG_TCP_MD5SIG
1699 .req_md5_lookup = tcp_v4_md5_lookup,
1700 .calc_md5_hash = tcp_v4_md5_hash_skb,
1701#endif
1702#ifdef CONFIG_TCP_AO
1703 .ao_lookup = tcp_v4_ao_lookup_rsk,
1704 .ao_calc_key = tcp_v4_ao_calc_key_rsk,
1705 .ao_synack_hash = tcp_v4_ao_synack_hash,
1706#endif
1707#ifdef CONFIG_SYN_COOKIES
1708 .cookie_init_seq = cookie_v4_init_sequence,
1709#endif
1710 .route_req = tcp_v4_route_req,
1711 .init_seq = tcp_v4_init_seq,
1712 .init_ts_off = tcp_v4_init_ts_off,
1713 .send_synack = tcp_v4_send_synack,
1714};
1715
1716int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb)
1717{
1718 /* Never answer to SYNs send to broadcast or multicast */
1719 if (skb_rtable(skb)->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))
1720 goto drop;
1721
1722 return tcp_conn_request(&tcp_request_sock_ops,
1723 &tcp_request_sock_ipv4_ops, sk, skb);
1724
1725drop:
1726 tcp_listendrop(sk);
1727 return 0;
1728}
1729EXPORT_SYMBOL(tcp_v4_conn_request);
1730
1731
1732/*
1733 * The three way handshake has completed - we got a valid synack -
1734 * now create the new socket.
1735 */
1736struct sock *tcp_v4_syn_recv_sock(const struct sock *sk, struct sk_buff *skb,
1737 struct request_sock *req,
1738 struct dst_entry *dst,
1739 struct request_sock *req_unhash,
1740 bool *own_req)
1741{
1742 struct inet_request_sock *ireq;
1743 bool found_dup_sk = false;
1744 struct inet_sock *newinet;
1745 struct tcp_sock *newtp;
1746 struct sock *newsk;
1747#ifdef CONFIG_TCP_MD5SIG
1748 const union tcp_md5_addr *addr;
1749 struct tcp_md5sig_key *key;
1750 int l3index;
1751#endif
1752 struct ip_options_rcu *inet_opt;
1753
1754 if (sk_acceptq_is_full(sk))
1755 goto exit_overflow;
1756
1757 newsk = tcp_create_openreq_child(sk, req, skb);
1758 if (!newsk)
1759 goto exit_nonewsk;
1760
1761 newsk->sk_gso_type = SKB_GSO_TCPV4;
1762 inet_sk_rx_dst_set(newsk, skb);
1763
1764 newtp = tcp_sk(newsk);
1765 newinet = inet_sk(newsk);
1766 ireq = inet_rsk(req);
1767 sk_daddr_set(newsk, ireq->ir_rmt_addr);
1768 sk_rcv_saddr_set(newsk, ireq->ir_loc_addr);
1769 newsk->sk_bound_dev_if = ireq->ir_iif;
1770 newinet->inet_saddr = ireq->ir_loc_addr;
1771 inet_opt = rcu_dereference(ireq->ireq_opt);
1772 RCU_INIT_POINTER(newinet->inet_opt, inet_opt);
1773 newinet->mc_index = inet_iif(skb);
1774 newinet->mc_ttl = ip_hdr(skb)->ttl;
1775 newinet->rcv_tos = ip_hdr(skb)->tos;
1776 inet_csk(newsk)->icsk_ext_hdr_len = 0;
1777 if (inet_opt)
1778 inet_csk(newsk)->icsk_ext_hdr_len = inet_opt->opt.optlen;
1779 atomic_set(&newinet->inet_id, get_random_u16());
1780
1781 /* Set ToS of the new socket based upon the value of incoming SYN.
1782 * ECT bits are set later in tcp_init_transfer().
1783 */
1784 if (READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_reflect_tos))
1785 newinet->tos = tcp_rsk(req)->syn_tos & ~INET_ECN_MASK;
1786
1787 if (!dst) {
1788 dst = inet_csk_route_child_sock(sk, newsk, req);
1789 if (!dst)
1790 goto put_and_exit;
1791 } else {
1792 /* syncookie case : see end of cookie_v4_check() */
1793 }
1794 sk_setup_caps(newsk, dst);
1795
1796 tcp_ca_openreq_child(newsk, dst);
1797
1798 tcp_sync_mss(newsk, dst_mtu(dst));
1799 newtp->advmss = tcp_mss_clamp(tcp_sk(sk), dst_metric_advmss(dst));
1800
1801 tcp_initialize_rcv_mss(newsk);
1802
1803#ifdef CONFIG_TCP_MD5SIG
1804 l3index = l3mdev_master_ifindex_by_index(sock_net(sk), ireq->ir_iif);
1805 /* Copy over the MD5 key from the original socket */
1806 addr = (union tcp_md5_addr *)&newinet->inet_daddr;
1807 key = tcp_md5_do_lookup(sk, l3index, addr, AF_INET);
1808 if (key && !tcp_rsk_used_ao(req)) {
1809 if (tcp_md5_key_copy(newsk, addr, AF_INET, 32, l3index, key))
1810 goto put_and_exit;
1811 sk_gso_disable(newsk);
1812 }
1813#endif
1814#ifdef CONFIG_TCP_AO
1815 if (tcp_ao_copy_all_matching(sk, newsk, req, skb, AF_INET))
1816 goto put_and_exit; /* OOM, release back memory */
1817#endif
1818
1819 if (__inet_inherit_port(sk, newsk) < 0)
1820 goto put_and_exit;
1821 *own_req = inet_ehash_nolisten(newsk, req_to_sk(req_unhash),
1822 &found_dup_sk);
1823 if (likely(*own_req)) {
1824 tcp_move_syn(newtp, req);
1825 ireq->ireq_opt = NULL;
1826 } else {
1827 newinet->inet_opt = NULL;
1828
1829 if (!req_unhash && found_dup_sk) {
1830 /* This code path should only be executed in the
1831 * syncookie case only
1832 */
1833 bh_unlock_sock(newsk);
1834 sock_put(newsk);
1835 newsk = NULL;
1836 }
1837 }
1838 return newsk;
1839
1840exit_overflow:
1841 NET_INC_STATS(sock_net(sk), LINUX_MIB_LISTENOVERFLOWS);
1842exit_nonewsk:
1843 dst_release(dst);
1844exit:
1845 tcp_listendrop(sk);
1846 return NULL;
1847put_and_exit:
1848 newinet->inet_opt = NULL;
1849 inet_csk_prepare_forced_close(newsk);
1850 tcp_done(newsk);
1851 goto exit;
1852}
1853EXPORT_SYMBOL(tcp_v4_syn_recv_sock);
1854
1855static struct sock *tcp_v4_cookie_check(struct sock *sk, struct sk_buff *skb)
1856{
1857#ifdef CONFIG_SYN_COOKIES
1858 const struct tcphdr *th = tcp_hdr(skb);
1859
1860 if (!th->syn)
1861 sk = cookie_v4_check(sk, skb);
1862#endif
1863 return sk;
1864}
1865
1866u16 tcp_v4_get_syncookie(struct sock *sk, struct iphdr *iph,
1867 struct tcphdr *th, u32 *cookie)
1868{
1869 u16 mss = 0;
1870#ifdef CONFIG_SYN_COOKIES
1871 mss = tcp_get_syncookie_mss(&tcp_request_sock_ops,
1872 &tcp_request_sock_ipv4_ops, sk, th);
1873 if (mss) {
1874 *cookie = __cookie_v4_init_sequence(iph, th, &mss);
1875 tcp_synq_overflow(sk);
1876 }
1877#endif
1878 return mss;
1879}
1880
1881INDIRECT_CALLABLE_DECLARE(struct dst_entry *ipv4_dst_check(struct dst_entry *,
1882 u32));
1883/* The socket must have it's spinlock held when we get
1884 * here, unless it is a TCP_LISTEN socket.
1885 *
1886 * We have a potential double-lock case here, so even when
1887 * doing backlog processing we use the BH locking scheme.
1888 * This is because we cannot sleep with the original spinlock
1889 * held.
1890 */
1891int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb)
1892{
1893 enum skb_drop_reason reason;
1894 struct sock *rsk;
1895
1896 if (sk->sk_state == TCP_ESTABLISHED) { /* Fast path */
1897 struct dst_entry *dst;
1898
1899 dst = rcu_dereference_protected(sk->sk_rx_dst,
1900 lockdep_sock_is_held(sk));
1901
1902 sock_rps_save_rxhash(sk, skb);
1903 sk_mark_napi_id(sk, skb);
1904 if (dst) {
1905 if (sk->sk_rx_dst_ifindex != skb->skb_iif ||
1906 !INDIRECT_CALL_1(dst->ops->check, ipv4_dst_check,
1907 dst, 0)) {
1908 RCU_INIT_POINTER(sk->sk_rx_dst, NULL);
1909 dst_release(dst);
1910 }
1911 }
1912 tcp_rcv_established(sk, skb);
1913 return 0;
1914 }
1915
1916 if (tcp_checksum_complete(skb))
1917 goto csum_err;
1918
1919 if (sk->sk_state == TCP_LISTEN) {
1920 struct sock *nsk = tcp_v4_cookie_check(sk, skb);
1921
1922 if (!nsk)
1923 return 0;
1924 if (nsk != sk) {
1925 reason = tcp_child_process(sk, nsk, skb);
1926 if (reason) {
1927 rsk = nsk;
1928 goto reset;
1929 }
1930 return 0;
1931 }
1932 } else
1933 sock_rps_save_rxhash(sk, skb);
1934
1935 reason = tcp_rcv_state_process(sk, skb);
1936 if (reason) {
1937 rsk = sk;
1938 goto reset;
1939 }
1940 return 0;
1941
1942reset:
1943 tcp_v4_send_reset(rsk, skb);
1944discard:
1945 kfree_skb_reason(skb, reason);
1946 /* Be careful here. If this function gets more complicated and
1947 * gcc suffers from register pressure on the x86, sk (in %ebx)
1948 * might be destroyed here. This current version compiles correctly,
1949 * but you have been warned.
1950 */
1951 return 0;
1952
1953csum_err:
1954 reason = SKB_DROP_REASON_TCP_CSUM;
1955 trace_tcp_bad_csum(skb);
1956 TCP_INC_STATS(sock_net(sk), TCP_MIB_CSUMERRORS);
1957 TCP_INC_STATS(sock_net(sk), TCP_MIB_INERRS);
1958 goto discard;
1959}
1960EXPORT_SYMBOL(tcp_v4_do_rcv);
1961
1962int tcp_v4_early_demux(struct sk_buff *skb)
1963{
1964 struct net *net = dev_net(skb->dev);
1965 const struct iphdr *iph;
1966 const struct tcphdr *th;
1967 struct sock *sk;
1968
1969 if (skb->pkt_type != PACKET_HOST)
1970 return 0;
1971
1972 if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct tcphdr)))
1973 return 0;
1974
1975 iph = ip_hdr(skb);
1976 th = tcp_hdr(skb);
1977
1978 if (th->doff < sizeof(struct tcphdr) / 4)
1979 return 0;
1980
1981 sk = __inet_lookup_established(net, net->ipv4.tcp_death_row.hashinfo,
1982 iph->saddr, th->source,
1983 iph->daddr, ntohs(th->dest),
1984 skb->skb_iif, inet_sdif(skb));
1985 if (sk) {
1986 skb->sk = sk;
1987 skb->destructor = sock_edemux;
1988 if (sk_fullsock(sk)) {
1989 struct dst_entry *dst = rcu_dereference(sk->sk_rx_dst);
1990
1991 if (dst)
1992 dst = dst_check(dst, 0);
1993 if (dst &&
1994 sk->sk_rx_dst_ifindex == skb->skb_iif)
1995 skb_dst_set_noref(skb, dst);
1996 }
1997 }
1998 return 0;
1999}
2000
2001bool tcp_add_backlog(struct sock *sk, struct sk_buff *skb,
2002 enum skb_drop_reason *reason)
2003{
2004 u32 tail_gso_size, tail_gso_segs;
2005 struct skb_shared_info *shinfo;
2006 const struct tcphdr *th;
2007 struct tcphdr *thtail;
2008 struct sk_buff *tail;
2009 unsigned int hdrlen;
2010 bool fragstolen;
2011 u32 gso_segs;
2012 u32 gso_size;
2013 u64 limit;
2014 int delta;
2015
2016 /* In case all data was pulled from skb frags (in __pskb_pull_tail()),
2017 * we can fix skb->truesize to its real value to avoid future drops.
2018 * This is valid because skb is not yet charged to the socket.
2019 * It has been noticed pure SACK packets were sometimes dropped
2020 * (if cooked by drivers without copybreak feature).
2021 */
2022 skb_condense(skb);
2023
2024 skb_dst_drop(skb);
2025
2026 if (unlikely(tcp_checksum_complete(skb))) {
2027 bh_unlock_sock(sk);
2028 trace_tcp_bad_csum(skb);
2029 *reason = SKB_DROP_REASON_TCP_CSUM;
2030 __TCP_INC_STATS(sock_net(sk), TCP_MIB_CSUMERRORS);
2031 __TCP_INC_STATS(sock_net(sk), TCP_MIB_INERRS);
2032 return true;
2033 }
2034
2035 /* Attempt coalescing to last skb in backlog, even if we are
2036 * above the limits.
2037 * This is okay because skb capacity is limited to MAX_SKB_FRAGS.
2038 */
2039 th = (const struct tcphdr *)skb->data;
2040 hdrlen = th->doff * 4;
2041
2042 tail = sk->sk_backlog.tail;
2043 if (!tail)
2044 goto no_coalesce;
2045 thtail = (struct tcphdr *)tail->data;
2046
2047 if (TCP_SKB_CB(tail)->end_seq != TCP_SKB_CB(skb)->seq ||
2048 TCP_SKB_CB(tail)->ip_dsfield != TCP_SKB_CB(skb)->ip_dsfield ||
2049 ((TCP_SKB_CB(tail)->tcp_flags |
2050 TCP_SKB_CB(skb)->tcp_flags) & (TCPHDR_SYN | TCPHDR_RST | TCPHDR_URG)) ||
2051 !((TCP_SKB_CB(tail)->tcp_flags &
2052 TCP_SKB_CB(skb)->tcp_flags) & TCPHDR_ACK) ||
2053 ((TCP_SKB_CB(tail)->tcp_flags ^
2054 TCP_SKB_CB(skb)->tcp_flags) & (TCPHDR_ECE | TCPHDR_CWR)) ||
2055#ifdef CONFIG_TLS_DEVICE
2056 tail->decrypted != skb->decrypted ||
2057#endif
2058 !mptcp_skb_can_collapse(tail, skb) ||
2059 thtail->doff != th->doff ||
2060 memcmp(thtail + 1, th + 1, hdrlen - sizeof(*th)))
2061 goto no_coalesce;
2062
2063 __skb_pull(skb, hdrlen);
2064
2065 shinfo = skb_shinfo(skb);
2066 gso_size = shinfo->gso_size ?: skb->len;
2067 gso_segs = shinfo->gso_segs ?: 1;
2068
2069 shinfo = skb_shinfo(tail);
2070 tail_gso_size = shinfo->gso_size ?: (tail->len - hdrlen);
2071 tail_gso_segs = shinfo->gso_segs ?: 1;
2072
2073 if (skb_try_coalesce(tail, skb, &fragstolen, &delta)) {
2074 TCP_SKB_CB(tail)->end_seq = TCP_SKB_CB(skb)->end_seq;
2075
2076 if (likely(!before(TCP_SKB_CB(skb)->ack_seq, TCP_SKB_CB(tail)->ack_seq))) {
2077 TCP_SKB_CB(tail)->ack_seq = TCP_SKB_CB(skb)->ack_seq;
2078 thtail->window = th->window;
2079 }
2080
2081 /* We have to update both TCP_SKB_CB(tail)->tcp_flags and
2082 * thtail->fin, so that the fast path in tcp_rcv_established()
2083 * is not entered if we append a packet with a FIN.
2084 * SYN, RST, URG are not present.
2085 * ACK is set on both packets.
2086 * PSH : we do not really care in TCP stack,
2087 * at least for 'GRO' packets.
2088 */
2089 thtail->fin |= th->fin;
2090 TCP_SKB_CB(tail)->tcp_flags |= TCP_SKB_CB(skb)->tcp_flags;
2091
2092 if (TCP_SKB_CB(skb)->has_rxtstamp) {
2093 TCP_SKB_CB(tail)->has_rxtstamp = true;
2094 tail->tstamp = skb->tstamp;
2095 skb_hwtstamps(tail)->hwtstamp = skb_hwtstamps(skb)->hwtstamp;
2096 }
2097
2098 /* Not as strict as GRO. We only need to carry mss max value */
2099 shinfo->gso_size = max(gso_size, tail_gso_size);
2100 shinfo->gso_segs = min_t(u32, gso_segs + tail_gso_segs, 0xFFFF);
2101
2102 sk->sk_backlog.len += delta;
2103 __NET_INC_STATS(sock_net(sk),
2104 LINUX_MIB_TCPBACKLOGCOALESCE);
2105 kfree_skb_partial(skb, fragstolen);
2106 return false;
2107 }
2108 __skb_push(skb, hdrlen);
2109
2110no_coalesce:
2111 /* sk->sk_backlog.len is reset only at the end of __release_sock().
2112 * Both sk->sk_backlog.len and sk->sk_rmem_alloc could reach
2113 * sk_rcvbuf in normal conditions.
2114 */
2115 limit = ((u64)READ_ONCE(sk->sk_rcvbuf)) << 1;
2116
2117 limit += ((u32)READ_ONCE(sk->sk_sndbuf)) >> 1;
2118
2119 /* Only socket owner can try to collapse/prune rx queues
2120 * to reduce memory overhead, so add a little headroom here.
2121 * Few sockets backlog are possibly concurrently non empty.
2122 */
2123 limit += 64 * 1024;
2124
2125 limit = min_t(u64, limit, UINT_MAX);
2126
2127 if (unlikely(sk_add_backlog(sk, skb, limit))) {
2128 bh_unlock_sock(sk);
2129 *reason = SKB_DROP_REASON_SOCKET_BACKLOG;
2130 __NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPBACKLOGDROP);
2131 return true;
2132 }
2133 return false;
2134}
2135EXPORT_SYMBOL(tcp_add_backlog);
2136
2137int tcp_filter(struct sock *sk, struct sk_buff *skb)
2138{
2139 struct tcphdr *th = (struct tcphdr *)skb->data;
2140
2141 return sk_filter_trim_cap(sk, skb, th->doff * 4);
2142}
2143EXPORT_SYMBOL(tcp_filter);
2144
2145static void tcp_v4_restore_cb(struct sk_buff *skb)
2146{
2147 memmove(IPCB(skb), &TCP_SKB_CB(skb)->header.h4,
2148 sizeof(struct inet_skb_parm));
2149}
2150
2151static void tcp_v4_fill_cb(struct sk_buff *skb, const struct iphdr *iph,
2152 const struct tcphdr *th)
2153{
2154 /* This is tricky : We move IPCB at its correct location into TCP_SKB_CB()
2155 * barrier() makes sure compiler wont play fool^Waliasing games.
2156 */
2157 memmove(&TCP_SKB_CB(skb)->header.h4, IPCB(skb),
2158 sizeof(struct inet_skb_parm));
2159 barrier();
2160
2161 TCP_SKB_CB(skb)->seq = ntohl(th->seq);
2162 TCP_SKB_CB(skb)->end_seq = (TCP_SKB_CB(skb)->seq + th->syn + th->fin +
2163 skb->len - th->doff * 4);
2164 TCP_SKB_CB(skb)->ack_seq = ntohl(th->ack_seq);
2165 TCP_SKB_CB(skb)->tcp_flags = tcp_flag_byte(th);
2166 TCP_SKB_CB(skb)->tcp_tw_isn = 0;
2167 TCP_SKB_CB(skb)->ip_dsfield = ipv4_get_dsfield(iph);
2168 TCP_SKB_CB(skb)->sacked = 0;
2169 TCP_SKB_CB(skb)->has_rxtstamp =
2170 skb->tstamp || skb_hwtstamps(skb)->hwtstamp;
2171}
2172
2173/*
2174 * From tcp_input.c
2175 */
2176
2177int tcp_v4_rcv(struct sk_buff *skb)
2178{
2179 struct net *net = dev_net(skb->dev);
2180 enum skb_drop_reason drop_reason;
2181 int sdif = inet_sdif(skb);
2182 int dif = inet_iif(skb);
2183 const struct iphdr *iph;
2184 const struct tcphdr *th;
2185 bool refcounted;
2186 struct sock *sk;
2187 int ret;
2188
2189 drop_reason = SKB_DROP_REASON_NOT_SPECIFIED;
2190 if (skb->pkt_type != PACKET_HOST)
2191 goto discard_it;
2192
2193 /* Count it even if it's bad */
2194 __TCP_INC_STATS(net, TCP_MIB_INSEGS);
2195
2196 if (!pskb_may_pull(skb, sizeof(struct tcphdr)))
2197 goto discard_it;
2198
2199 th = (const struct tcphdr *)skb->data;
2200
2201 if (unlikely(th->doff < sizeof(struct tcphdr) / 4)) {
2202 drop_reason = SKB_DROP_REASON_PKT_TOO_SMALL;
2203 goto bad_packet;
2204 }
2205 if (!pskb_may_pull(skb, th->doff * 4))
2206 goto discard_it;
2207
2208 /* An explanation is required here, I think.
2209 * Packet length and doff are validated by header prediction,
2210 * provided case of th->doff==0 is eliminated.
2211 * So, we defer the checks. */
2212
2213 if (skb_checksum_init(skb, IPPROTO_TCP, inet_compute_pseudo))
2214 goto csum_error;
2215
2216 th = (const struct tcphdr *)skb->data;
2217 iph = ip_hdr(skb);
2218lookup:
2219 sk = __inet_lookup_skb(net->ipv4.tcp_death_row.hashinfo,
2220 skb, __tcp_hdrlen(th), th->source,
2221 th->dest, sdif, &refcounted);
2222 if (!sk)
2223 goto no_tcp_socket;
2224
2225process:
2226 if (sk->sk_state == TCP_TIME_WAIT)
2227 goto do_time_wait;
2228
2229 if (sk->sk_state == TCP_NEW_SYN_RECV) {
2230 struct request_sock *req = inet_reqsk(sk);
2231 bool req_stolen = false;
2232 struct sock *nsk;
2233
2234 sk = req->rsk_listener;
2235 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
2236 drop_reason = SKB_DROP_REASON_XFRM_POLICY;
2237 else
2238 drop_reason = tcp_inbound_hash(sk, req, skb,
2239 &iph->saddr, &iph->daddr,
2240 AF_INET, dif, sdif);
2241 if (unlikely(drop_reason)) {
2242 sk_drops_add(sk, skb);
2243 reqsk_put(req);
2244 goto discard_it;
2245 }
2246 if (tcp_checksum_complete(skb)) {
2247 reqsk_put(req);
2248 goto csum_error;
2249 }
2250 if (unlikely(sk->sk_state != TCP_LISTEN)) {
2251 nsk = reuseport_migrate_sock(sk, req_to_sk(req), skb);
2252 if (!nsk) {
2253 inet_csk_reqsk_queue_drop_and_put(sk, req);
2254 goto lookup;
2255 }
2256 sk = nsk;
2257 /* reuseport_migrate_sock() has already held one sk_refcnt
2258 * before returning.
2259 */
2260 } else {
2261 /* We own a reference on the listener, increase it again
2262 * as we might lose it too soon.
2263 */
2264 sock_hold(sk);
2265 }
2266 refcounted = true;
2267 nsk = NULL;
2268 if (!tcp_filter(sk, skb)) {
2269 th = (const struct tcphdr *)skb->data;
2270 iph = ip_hdr(skb);
2271 tcp_v4_fill_cb(skb, iph, th);
2272 nsk = tcp_check_req(sk, skb, req, false, &req_stolen);
2273 } else {
2274 drop_reason = SKB_DROP_REASON_SOCKET_FILTER;
2275 }
2276 if (!nsk) {
2277 reqsk_put(req);
2278 if (req_stolen) {
2279 /* Another cpu got exclusive access to req
2280 * and created a full blown socket.
2281 * Try to feed this packet to this socket
2282 * instead of discarding it.
2283 */
2284 tcp_v4_restore_cb(skb);
2285 sock_put(sk);
2286 goto lookup;
2287 }
2288 goto discard_and_relse;
2289 }
2290 nf_reset_ct(skb);
2291 if (nsk == sk) {
2292 reqsk_put(req);
2293 tcp_v4_restore_cb(skb);
2294 } else {
2295 drop_reason = tcp_child_process(sk, nsk, skb);
2296 if (drop_reason) {
2297 tcp_v4_send_reset(nsk, skb);
2298 goto discard_and_relse;
2299 }
2300 sock_put(sk);
2301 return 0;
2302 }
2303 }
2304
2305 if (static_branch_unlikely(&ip4_min_ttl)) {
2306 /* min_ttl can be changed concurrently from do_ip_setsockopt() */
2307 if (unlikely(iph->ttl < READ_ONCE(inet_sk(sk)->min_ttl))) {
2308 __NET_INC_STATS(net, LINUX_MIB_TCPMINTTLDROP);
2309 drop_reason = SKB_DROP_REASON_TCP_MINTTL;
2310 goto discard_and_relse;
2311 }
2312 }
2313
2314 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb)) {
2315 drop_reason = SKB_DROP_REASON_XFRM_POLICY;
2316 goto discard_and_relse;
2317 }
2318
2319 drop_reason = tcp_inbound_hash(sk, NULL, skb, &iph->saddr, &iph->daddr,
2320 AF_INET, dif, sdif);
2321 if (drop_reason)
2322 goto discard_and_relse;
2323
2324 nf_reset_ct(skb);
2325
2326 if (tcp_filter(sk, skb)) {
2327 drop_reason = SKB_DROP_REASON_SOCKET_FILTER;
2328 goto discard_and_relse;
2329 }
2330 th = (const struct tcphdr *)skb->data;
2331 iph = ip_hdr(skb);
2332 tcp_v4_fill_cb(skb, iph, th);
2333
2334 skb->dev = NULL;
2335
2336 if (sk->sk_state == TCP_LISTEN) {
2337 ret = tcp_v4_do_rcv(sk, skb);
2338 goto put_and_return;
2339 }
2340
2341 sk_incoming_cpu_update(sk);
2342
2343 bh_lock_sock_nested(sk);
2344 tcp_segs_in(tcp_sk(sk), skb);
2345 ret = 0;
2346 if (!sock_owned_by_user(sk)) {
2347 ret = tcp_v4_do_rcv(sk, skb);
2348 } else {
2349 if (tcp_add_backlog(sk, skb, &drop_reason))
2350 goto discard_and_relse;
2351 }
2352 bh_unlock_sock(sk);
2353
2354put_and_return:
2355 if (refcounted)
2356 sock_put(sk);
2357
2358 return ret;
2359
2360no_tcp_socket:
2361 drop_reason = SKB_DROP_REASON_NO_SOCKET;
2362 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
2363 goto discard_it;
2364
2365 tcp_v4_fill_cb(skb, iph, th);
2366
2367 if (tcp_checksum_complete(skb)) {
2368csum_error:
2369 drop_reason = SKB_DROP_REASON_TCP_CSUM;
2370 trace_tcp_bad_csum(skb);
2371 __TCP_INC_STATS(net, TCP_MIB_CSUMERRORS);
2372bad_packet:
2373 __TCP_INC_STATS(net, TCP_MIB_INERRS);
2374 } else {
2375 tcp_v4_send_reset(NULL, skb);
2376 }
2377
2378discard_it:
2379 SKB_DR_OR(drop_reason, NOT_SPECIFIED);
2380 /* Discard frame. */
2381 kfree_skb_reason(skb, drop_reason);
2382 return 0;
2383
2384discard_and_relse:
2385 sk_drops_add(sk, skb);
2386 if (refcounted)
2387 sock_put(sk);
2388 goto discard_it;
2389
2390do_time_wait:
2391 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) {
2392 drop_reason = SKB_DROP_REASON_XFRM_POLICY;
2393 inet_twsk_put(inet_twsk(sk));
2394 goto discard_it;
2395 }
2396
2397 tcp_v4_fill_cb(skb, iph, th);
2398
2399 if (tcp_checksum_complete(skb)) {
2400 inet_twsk_put(inet_twsk(sk));
2401 goto csum_error;
2402 }
2403 switch (tcp_timewait_state_process(inet_twsk(sk), skb, th)) {
2404 case TCP_TW_SYN: {
2405 struct sock *sk2 = inet_lookup_listener(net,
2406 net->ipv4.tcp_death_row.hashinfo,
2407 skb, __tcp_hdrlen(th),
2408 iph->saddr, th->source,
2409 iph->daddr, th->dest,
2410 inet_iif(skb),
2411 sdif);
2412 if (sk2) {
2413 inet_twsk_deschedule_put(inet_twsk(sk));
2414 sk = sk2;
2415 tcp_v4_restore_cb(skb);
2416 refcounted = false;
2417 goto process;
2418 }
2419 }
2420 /* to ACK */
2421 fallthrough;
2422 case TCP_TW_ACK:
2423 tcp_v4_timewait_ack(sk, skb);
2424 break;
2425 case TCP_TW_RST:
2426 tcp_v4_send_reset(sk, skb);
2427 inet_twsk_deschedule_put(inet_twsk(sk));
2428 goto discard_it;
2429 case TCP_TW_SUCCESS:;
2430 }
2431 goto discard_it;
2432}
2433
2434static struct timewait_sock_ops tcp_timewait_sock_ops = {
2435 .twsk_obj_size = sizeof(struct tcp_timewait_sock),
2436 .twsk_unique = tcp_twsk_unique,
2437 .twsk_destructor= tcp_twsk_destructor,
2438};
2439
2440void inet_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb)
2441{
2442 struct dst_entry *dst = skb_dst(skb);
2443
2444 if (dst && dst_hold_safe(dst)) {
2445 rcu_assign_pointer(sk->sk_rx_dst, dst);
2446 sk->sk_rx_dst_ifindex = skb->skb_iif;
2447 }
2448}
2449EXPORT_SYMBOL(inet_sk_rx_dst_set);
2450
2451const struct inet_connection_sock_af_ops ipv4_specific = {
2452 .queue_xmit = ip_queue_xmit,
2453 .send_check = tcp_v4_send_check,
2454 .rebuild_header = inet_sk_rebuild_header,
2455 .sk_rx_dst_set = inet_sk_rx_dst_set,
2456 .conn_request = tcp_v4_conn_request,
2457 .syn_recv_sock = tcp_v4_syn_recv_sock,
2458 .net_header_len = sizeof(struct iphdr),
2459 .setsockopt = ip_setsockopt,
2460 .getsockopt = ip_getsockopt,
2461 .addr2sockaddr = inet_csk_addr2sockaddr,
2462 .sockaddr_len = sizeof(struct sockaddr_in),
2463 .mtu_reduced = tcp_v4_mtu_reduced,
2464};
2465EXPORT_SYMBOL(ipv4_specific);
2466
2467#if defined(CONFIG_TCP_MD5SIG) || defined(CONFIG_TCP_AO)
2468static const struct tcp_sock_af_ops tcp_sock_ipv4_specific = {
2469#ifdef CONFIG_TCP_MD5SIG
2470 .md5_lookup = tcp_v4_md5_lookup,
2471 .calc_md5_hash = tcp_v4_md5_hash_skb,
2472 .md5_parse = tcp_v4_parse_md5_keys,
2473#endif
2474#ifdef CONFIG_TCP_AO
2475 .ao_lookup = tcp_v4_ao_lookup,
2476 .calc_ao_hash = tcp_v4_ao_hash_skb,
2477 .ao_parse = tcp_v4_parse_ao,
2478 .ao_calc_key_sk = tcp_v4_ao_calc_key_sk,
2479#endif
2480};
2481#endif
2482
2483/* NOTE: A lot of things set to zero explicitly by call to
2484 * sk_alloc() so need not be done here.
2485 */
2486static int tcp_v4_init_sock(struct sock *sk)
2487{
2488 struct inet_connection_sock *icsk = inet_csk(sk);
2489
2490 tcp_init_sock(sk);
2491
2492 icsk->icsk_af_ops = &ipv4_specific;
2493
2494#if defined(CONFIG_TCP_MD5SIG) || defined(CONFIG_TCP_AO)
2495 tcp_sk(sk)->af_specific = &tcp_sock_ipv4_specific;
2496#endif
2497
2498 return 0;
2499}
2500
2501#ifdef CONFIG_TCP_MD5SIG
2502static void tcp_md5sig_info_free_rcu(struct rcu_head *head)
2503{
2504 struct tcp_md5sig_info *md5sig;
2505
2506 md5sig = container_of(head, struct tcp_md5sig_info, rcu);
2507 kfree(md5sig);
2508 static_branch_slow_dec_deferred(&tcp_md5_needed);
2509 tcp_md5_release_sigpool();
2510}
2511#endif
2512
2513void tcp_v4_destroy_sock(struct sock *sk)
2514{
2515 struct tcp_sock *tp = tcp_sk(sk);
2516
2517 trace_tcp_destroy_sock(sk);
2518
2519 tcp_clear_xmit_timers(sk);
2520
2521 tcp_cleanup_congestion_control(sk);
2522
2523 tcp_cleanup_ulp(sk);
2524
2525 /* Cleanup up the write buffer. */
2526 tcp_write_queue_purge(sk);
2527
2528 /* Check if we want to disable active TFO */
2529 tcp_fastopen_active_disable_ofo_check(sk);
2530
2531 /* Cleans up our, hopefully empty, out_of_order_queue. */
2532 skb_rbtree_purge(&tp->out_of_order_queue);
2533
2534#ifdef CONFIG_TCP_MD5SIG
2535 /* Clean up the MD5 key list, if any */
2536 if (tp->md5sig_info) {
2537 struct tcp_md5sig_info *md5sig;
2538
2539 md5sig = rcu_dereference_protected(tp->md5sig_info, 1);
2540 tcp_clear_md5_list(sk);
2541 call_rcu(&md5sig->rcu, tcp_md5sig_info_free_rcu);
2542 rcu_assign_pointer(tp->md5sig_info, NULL);
2543 }
2544#endif
2545 tcp_ao_destroy_sock(sk, false);
2546
2547 /* Clean up a referenced TCP bind bucket. */
2548 if (inet_csk(sk)->icsk_bind_hash)
2549 inet_put_port(sk);
2550
2551 BUG_ON(rcu_access_pointer(tp->fastopen_rsk));
2552
2553 /* If socket is aborted during connect operation */
2554 tcp_free_fastopen_req(tp);
2555 tcp_fastopen_destroy_cipher(sk);
2556 tcp_saved_syn_free(tp);
2557
2558 sk_sockets_allocated_dec(sk);
2559}
2560EXPORT_SYMBOL(tcp_v4_destroy_sock);
2561
2562#ifdef CONFIG_PROC_FS
2563/* Proc filesystem TCP sock list dumping. */
2564
2565static unsigned short seq_file_family(const struct seq_file *seq);
2566
2567static bool seq_sk_match(struct seq_file *seq, const struct sock *sk)
2568{
2569 unsigned short family = seq_file_family(seq);
2570
2571 /* AF_UNSPEC is used as a match all */
2572 return ((family == AF_UNSPEC || family == sk->sk_family) &&
2573 net_eq(sock_net(sk), seq_file_net(seq)));
2574}
2575
2576/* Find a non empty bucket (starting from st->bucket)
2577 * and return the first sk from it.
2578 */
2579static void *listening_get_first(struct seq_file *seq)
2580{
2581 struct inet_hashinfo *hinfo = seq_file_net(seq)->ipv4.tcp_death_row.hashinfo;
2582 struct tcp_iter_state *st = seq->private;
2583
2584 st->offset = 0;
2585 for (; st->bucket <= hinfo->lhash2_mask; st->bucket++) {
2586 struct inet_listen_hashbucket *ilb2;
2587 struct hlist_nulls_node *node;
2588 struct sock *sk;
2589
2590 ilb2 = &hinfo->lhash2[st->bucket];
2591 if (hlist_nulls_empty(&ilb2->nulls_head))
2592 continue;
2593
2594 spin_lock(&ilb2->lock);
2595 sk_nulls_for_each(sk, node, &ilb2->nulls_head) {
2596 if (seq_sk_match(seq, sk))
2597 return sk;
2598 }
2599 spin_unlock(&ilb2->lock);
2600 }
2601
2602 return NULL;
2603}
2604
2605/* Find the next sk of "cur" within the same bucket (i.e. st->bucket).
2606 * If "cur" is the last one in the st->bucket,
2607 * call listening_get_first() to return the first sk of the next
2608 * non empty bucket.
2609 */
2610static void *listening_get_next(struct seq_file *seq, void *cur)
2611{
2612 struct tcp_iter_state *st = seq->private;
2613 struct inet_listen_hashbucket *ilb2;
2614 struct hlist_nulls_node *node;
2615 struct inet_hashinfo *hinfo;
2616 struct sock *sk = cur;
2617
2618 ++st->num;
2619 ++st->offset;
2620
2621 sk = sk_nulls_next(sk);
2622 sk_nulls_for_each_from(sk, node) {
2623 if (seq_sk_match(seq, sk))
2624 return sk;
2625 }
2626
2627 hinfo = seq_file_net(seq)->ipv4.tcp_death_row.hashinfo;
2628 ilb2 = &hinfo->lhash2[st->bucket];
2629 spin_unlock(&ilb2->lock);
2630 ++st->bucket;
2631 return listening_get_first(seq);
2632}
2633
2634static void *listening_get_idx(struct seq_file *seq, loff_t *pos)
2635{
2636 struct tcp_iter_state *st = seq->private;
2637 void *rc;
2638
2639 st->bucket = 0;
2640 st->offset = 0;
2641 rc = listening_get_first(seq);
2642
2643 while (rc && *pos) {
2644 rc = listening_get_next(seq, rc);
2645 --*pos;
2646 }
2647 return rc;
2648}
2649
2650static inline bool empty_bucket(struct inet_hashinfo *hinfo,
2651 const struct tcp_iter_state *st)
2652{
2653 return hlist_nulls_empty(&hinfo->ehash[st->bucket].chain);
2654}
2655
2656/*
2657 * Get first established socket starting from bucket given in st->bucket.
2658 * If st->bucket is zero, the very first socket in the hash is returned.
2659 */
2660static void *established_get_first(struct seq_file *seq)
2661{
2662 struct inet_hashinfo *hinfo = seq_file_net(seq)->ipv4.tcp_death_row.hashinfo;
2663 struct tcp_iter_state *st = seq->private;
2664
2665 st->offset = 0;
2666 for (; st->bucket <= hinfo->ehash_mask; ++st->bucket) {
2667 struct sock *sk;
2668 struct hlist_nulls_node *node;
2669 spinlock_t *lock = inet_ehash_lockp(hinfo, st->bucket);
2670
2671 cond_resched();
2672
2673 /* Lockless fast path for the common case of empty buckets */
2674 if (empty_bucket(hinfo, st))
2675 continue;
2676
2677 spin_lock_bh(lock);
2678 sk_nulls_for_each(sk, node, &hinfo->ehash[st->bucket].chain) {
2679 if (seq_sk_match(seq, sk))
2680 return sk;
2681 }
2682 spin_unlock_bh(lock);
2683 }
2684
2685 return NULL;
2686}
2687
2688static void *established_get_next(struct seq_file *seq, void *cur)
2689{
2690 struct inet_hashinfo *hinfo = seq_file_net(seq)->ipv4.tcp_death_row.hashinfo;
2691 struct tcp_iter_state *st = seq->private;
2692 struct hlist_nulls_node *node;
2693 struct sock *sk = cur;
2694
2695 ++st->num;
2696 ++st->offset;
2697
2698 sk = sk_nulls_next(sk);
2699
2700 sk_nulls_for_each_from(sk, node) {
2701 if (seq_sk_match(seq, sk))
2702 return sk;
2703 }
2704
2705 spin_unlock_bh(inet_ehash_lockp(hinfo, st->bucket));
2706 ++st->bucket;
2707 return established_get_first(seq);
2708}
2709
2710static void *established_get_idx(struct seq_file *seq, loff_t pos)
2711{
2712 struct tcp_iter_state *st = seq->private;
2713 void *rc;
2714
2715 st->bucket = 0;
2716 rc = established_get_first(seq);
2717
2718 while (rc && pos) {
2719 rc = established_get_next(seq, rc);
2720 --pos;
2721 }
2722 return rc;
2723}
2724
2725static void *tcp_get_idx(struct seq_file *seq, loff_t pos)
2726{
2727 void *rc;
2728 struct tcp_iter_state *st = seq->private;
2729
2730 st->state = TCP_SEQ_STATE_LISTENING;
2731 rc = listening_get_idx(seq, &pos);
2732
2733 if (!rc) {
2734 st->state = TCP_SEQ_STATE_ESTABLISHED;
2735 rc = established_get_idx(seq, pos);
2736 }
2737
2738 return rc;
2739}
2740
2741static void *tcp_seek_last_pos(struct seq_file *seq)
2742{
2743 struct inet_hashinfo *hinfo = seq_file_net(seq)->ipv4.tcp_death_row.hashinfo;
2744 struct tcp_iter_state *st = seq->private;
2745 int bucket = st->bucket;
2746 int offset = st->offset;
2747 int orig_num = st->num;
2748 void *rc = NULL;
2749
2750 switch (st->state) {
2751 case TCP_SEQ_STATE_LISTENING:
2752 if (st->bucket > hinfo->lhash2_mask)
2753 break;
2754 rc = listening_get_first(seq);
2755 while (offset-- && rc && bucket == st->bucket)
2756 rc = listening_get_next(seq, rc);
2757 if (rc)
2758 break;
2759 st->bucket = 0;
2760 st->state = TCP_SEQ_STATE_ESTABLISHED;
2761 fallthrough;
2762 case TCP_SEQ_STATE_ESTABLISHED:
2763 if (st->bucket > hinfo->ehash_mask)
2764 break;
2765 rc = established_get_first(seq);
2766 while (offset-- && rc && bucket == st->bucket)
2767 rc = established_get_next(seq, rc);
2768 }
2769
2770 st->num = orig_num;
2771
2772 return rc;
2773}
2774
2775void *tcp_seq_start(struct seq_file *seq, loff_t *pos)
2776{
2777 struct tcp_iter_state *st = seq->private;
2778 void *rc;
2779
2780 if (*pos && *pos == st->last_pos) {
2781 rc = tcp_seek_last_pos(seq);
2782 if (rc)
2783 goto out;
2784 }
2785
2786 st->state = TCP_SEQ_STATE_LISTENING;
2787 st->num = 0;
2788 st->bucket = 0;
2789 st->offset = 0;
2790 rc = *pos ? tcp_get_idx(seq, *pos - 1) : SEQ_START_TOKEN;
2791
2792out:
2793 st->last_pos = *pos;
2794 return rc;
2795}
2796EXPORT_SYMBOL(tcp_seq_start);
2797
2798void *tcp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2799{
2800 struct tcp_iter_state *st = seq->private;
2801 void *rc = NULL;
2802
2803 if (v == SEQ_START_TOKEN) {
2804 rc = tcp_get_idx(seq, 0);
2805 goto out;
2806 }
2807
2808 switch (st->state) {
2809 case TCP_SEQ_STATE_LISTENING:
2810 rc = listening_get_next(seq, v);
2811 if (!rc) {
2812 st->state = TCP_SEQ_STATE_ESTABLISHED;
2813 st->bucket = 0;
2814 st->offset = 0;
2815 rc = established_get_first(seq);
2816 }
2817 break;
2818 case TCP_SEQ_STATE_ESTABLISHED:
2819 rc = established_get_next(seq, v);
2820 break;
2821 }
2822out:
2823 ++*pos;
2824 st->last_pos = *pos;
2825 return rc;
2826}
2827EXPORT_SYMBOL(tcp_seq_next);
2828
2829void tcp_seq_stop(struct seq_file *seq, void *v)
2830{
2831 struct inet_hashinfo *hinfo = seq_file_net(seq)->ipv4.tcp_death_row.hashinfo;
2832 struct tcp_iter_state *st = seq->private;
2833
2834 switch (st->state) {
2835 case TCP_SEQ_STATE_LISTENING:
2836 if (v != SEQ_START_TOKEN)
2837 spin_unlock(&hinfo->lhash2[st->bucket].lock);
2838 break;
2839 case TCP_SEQ_STATE_ESTABLISHED:
2840 if (v)
2841 spin_unlock_bh(inet_ehash_lockp(hinfo, st->bucket));
2842 break;
2843 }
2844}
2845EXPORT_SYMBOL(tcp_seq_stop);
2846
2847static void get_openreq4(const struct request_sock *req,
2848 struct seq_file *f, int i)
2849{
2850 const struct inet_request_sock *ireq = inet_rsk(req);
2851 long delta = req->rsk_timer.expires - jiffies;
2852
2853 seq_printf(f, "%4d: %08X:%04X %08X:%04X"
2854 " %02X %08X:%08X %02X:%08lX %08X %5u %8d %u %d %pK",
2855 i,
2856 ireq->ir_loc_addr,
2857 ireq->ir_num,
2858 ireq->ir_rmt_addr,
2859 ntohs(ireq->ir_rmt_port),
2860 TCP_SYN_RECV,
2861 0, 0, /* could print option size, but that is af dependent. */
2862 1, /* timers active (only the expire timer) */
2863 jiffies_delta_to_clock_t(delta),
2864 req->num_timeout,
2865 from_kuid_munged(seq_user_ns(f),
2866 sock_i_uid(req->rsk_listener)),
2867 0, /* non standard timer */
2868 0, /* open_requests have no inode */
2869 0,
2870 req);
2871}
2872
2873static void get_tcp4_sock(struct sock *sk, struct seq_file *f, int i)
2874{
2875 int timer_active;
2876 unsigned long timer_expires;
2877 const struct tcp_sock *tp = tcp_sk(sk);
2878 const struct inet_connection_sock *icsk = inet_csk(sk);
2879 const struct inet_sock *inet = inet_sk(sk);
2880 const struct fastopen_queue *fastopenq = &icsk->icsk_accept_queue.fastopenq;
2881 __be32 dest = inet->inet_daddr;
2882 __be32 src = inet->inet_rcv_saddr;
2883 __u16 destp = ntohs(inet->inet_dport);
2884 __u16 srcp = ntohs(inet->inet_sport);
2885 int rx_queue;
2886 int state;
2887
2888 if (icsk->icsk_pending == ICSK_TIME_RETRANS ||
2889 icsk->icsk_pending == ICSK_TIME_REO_TIMEOUT ||
2890 icsk->icsk_pending == ICSK_TIME_LOSS_PROBE) {
2891 timer_active = 1;
2892 timer_expires = icsk->icsk_timeout;
2893 } else if (icsk->icsk_pending == ICSK_TIME_PROBE0) {
2894 timer_active = 4;
2895 timer_expires = icsk->icsk_timeout;
2896 } else if (timer_pending(&sk->sk_timer)) {
2897 timer_active = 2;
2898 timer_expires = sk->sk_timer.expires;
2899 } else {
2900 timer_active = 0;
2901 timer_expires = jiffies;
2902 }
2903
2904 state = inet_sk_state_load(sk);
2905 if (state == TCP_LISTEN)
2906 rx_queue = READ_ONCE(sk->sk_ack_backlog);
2907 else
2908 /* Because we don't lock the socket,
2909 * we might find a transient negative value.
2910 */
2911 rx_queue = max_t(int, READ_ONCE(tp->rcv_nxt) -
2912 READ_ONCE(tp->copied_seq), 0);
2913
2914 seq_printf(f, "%4d: %08X:%04X %08X:%04X %02X %08X:%08X %02X:%08lX "
2915 "%08X %5u %8d %lu %d %pK %lu %lu %u %u %d",
2916 i, src, srcp, dest, destp, state,
2917 READ_ONCE(tp->write_seq) - tp->snd_una,
2918 rx_queue,
2919 timer_active,
2920 jiffies_delta_to_clock_t(timer_expires - jiffies),
2921 icsk->icsk_retransmits,
2922 from_kuid_munged(seq_user_ns(f), sock_i_uid(sk)),
2923 icsk->icsk_probes_out,
2924 sock_i_ino(sk),
2925 refcount_read(&sk->sk_refcnt), sk,
2926 jiffies_to_clock_t(icsk->icsk_rto),
2927 jiffies_to_clock_t(icsk->icsk_ack.ato),
2928 (icsk->icsk_ack.quick << 1) | inet_csk_in_pingpong_mode(sk),
2929 tcp_snd_cwnd(tp),
2930 state == TCP_LISTEN ?
2931 fastopenq->max_qlen :
2932 (tcp_in_initial_slowstart(tp) ? -1 : tp->snd_ssthresh));
2933}
2934
2935static void get_timewait4_sock(const struct inet_timewait_sock *tw,
2936 struct seq_file *f, int i)
2937{
2938 long delta = tw->tw_timer.expires - jiffies;
2939 __be32 dest, src;
2940 __u16 destp, srcp;
2941
2942 dest = tw->tw_daddr;
2943 src = tw->tw_rcv_saddr;
2944 destp = ntohs(tw->tw_dport);
2945 srcp = ntohs(tw->tw_sport);
2946
2947 seq_printf(f, "%4d: %08X:%04X %08X:%04X"
2948 " %02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %pK",
2949 i, src, srcp, dest, destp, tw->tw_substate, 0, 0,
2950 3, jiffies_delta_to_clock_t(delta), 0, 0, 0, 0,
2951 refcount_read(&tw->tw_refcnt), tw);
2952}
2953
2954#define TMPSZ 150
2955
2956static int tcp4_seq_show(struct seq_file *seq, void *v)
2957{
2958 struct tcp_iter_state *st;
2959 struct sock *sk = v;
2960
2961 seq_setwidth(seq, TMPSZ - 1);
2962 if (v == SEQ_START_TOKEN) {
2963 seq_puts(seq, " sl local_address rem_address st tx_queue "
2964 "rx_queue tr tm->when retrnsmt uid timeout "
2965 "inode");
2966 goto out;
2967 }
2968 st = seq->private;
2969
2970 if (sk->sk_state == TCP_TIME_WAIT)
2971 get_timewait4_sock(v, seq, st->num);
2972 else if (sk->sk_state == TCP_NEW_SYN_RECV)
2973 get_openreq4(v, seq, st->num);
2974 else
2975 get_tcp4_sock(v, seq, st->num);
2976out:
2977 seq_pad(seq, '\n');
2978 return 0;
2979}
2980
2981#ifdef CONFIG_BPF_SYSCALL
2982struct bpf_tcp_iter_state {
2983 struct tcp_iter_state state;
2984 unsigned int cur_sk;
2985 unsigned int end_sk;
2986 unsigned int max_sk;
2987 struct sock **batch;
2988 bool st_bucket_done;
2989};
2990
2991struct bpf_iter__tcp {
2992 __bpf_md_ptr(struct bpf_iter_meta *, meta);
2993 __bpf_md_ptr(struct sock_common *, sk_common);
2994 uid_t uid __aligned(8);
2995};
2996
2997static int tcp_prog_seq_show(struct bpf_prog *prog, struct bpf_iter_meta *meta,
2998 struct sock_common *sk_common, uid_t uid)
2999{
3000 struct bpf_iter__tcp ctx;
3001
3002 meta->seq_num--; /* skip SEQ_START_TOKEN */
3003 ctx.meta = meta;
3004 ctx.sk_common = sk_common;
3005 ctx.uid = uid;
3006 return bpf_iter_run_prog(prog, &ctx);
3007}
3008
3009static void bpf_iter_tcp_put_batch(struct bpf_tcp_iter_state *iter)
3010{
3011 while (iter->cur_sk < iter->end_sk)
3012 sock_gen_put(iter->batch[iter->cur_sk++]);
3013}
3014
3015static int bpf_iter_tcp_realloc_batch(struct bpf_tcp_iter_state *iter,
3016 unsigned int new_batch_sz)
3017{
3018 struct sock **new_batch;
3019
3020 new_batch = kvmalloc(sizeof(*new_batch) * new_batch_sz,
3021 GFP_USER | __GFP_NOWARN);
3022 if (!new_batch)
3023 return -ENOMEM;
3024
3025 bpf_iter_tcp_put_batch(iter);
3026 kvfree(iter->batch);
3027 iter->batch = new_batch;
3028 iter->max_sk = new_batch_sz;
3029
3030 return 0;
3031}
3032
3033static unsigned int bpf_iter_tcp_listening_batch(struct seq_file *seq,
3034 struct sock *start_sk)
3035{
3036 struct inet_hashinfo *hinfo = seq_file_net(seq)->ipv4.tcp_death_row.hashinfo;
3037 struct bpf_tcp_iter_state *iter = seq->private;
3038 struct tcp_iter_state *st = &iter->state;
3039 struct hlist_nulls_node *node;
3040 unsigned int expected = 1;
3041 struct sock *sk;
3042
3043 sock_hold(start_sk);
3044 iter->batch[iter->end_sk++] = start_sk;
3045
3046 sk = sk_nulls_next(start_sk);
3047 sk_nulls_for_each_from(sk, node) {
3048 if (seq_sk_match(seq, sk)) {
3049 if (iter->end_sk < iter->max_sk) {
3050 sock_hold(sk);
3051 iter->batch[iter->end_sk++] = sk;
3052 }
3053 expected++;
3054 }
3055 }
3056 spin_unlock(&hinfo->lhash2[st->bucket].lock);
3057
3058 return expected;
3059}
3060
3061static unsigned int bpf_iter_tcp_established_batch(struct seq_file *seq,
3062 struct sock *start_sk)
3063{
3064 struct inet_hashinfo *hinfo = seq_file_net(seq)->ipv4.tcp_death_row.hashinfo;
3065 struct bpf_tcp_iter_state *iter = seq->private;
3066 struct tcp_iter_state *st = &iter->state;
3067 struct hlist_nulls_node *node;
3068 unsigned int expected = 1;
3069 struct sock *sk;
3070
3071 sock_hold(start_sk);
3072 iter->batch[iter->end_sk++] = start_sk;
3073
3074 sk = sk_nulls_next(start_sk);
3075 sk_nulls_for_each_from(sk, node) {
3076 if (seq_sk_match(seq, sk)) {
3077 if (iter->end_sk < iter->max_sk) {
3078 sock_hold(sk);
3079 iter->batch[iter->end_sk++] = sk;
3080 }
3081 expected++;
3082 }
3083 }
3084 spin_unlock_bh(inet_ehash_lockp(hinfo, st->bucket));
3085
3086 return expected;
3087}
3088
3089static struct sock *bpf_iter_tcp_batch(struct seq_file *seq)
3090{
3091 struct inet_hashinfo *hinfo = seq_file_net(seq)->ipv4.tcp_death_row.hashinfo;
3092 struct bpf_tcp_iter_state *iter = seq->private;
3093 struct tcp_iter_state *st = &iter->state;
3094 unsigned int expected;
3095 bool resized = false;
3096 struct sock *sk;
3097
3098 /* The st->bucket is done. Directly advance to the next
3099 * bucket instead of having the tcp_seek_last_pos() to skip
3100 * one by one in the current bucket and eventually find out
3101 * it has to advance to the next bucket.
3102 */
3103 if (iter->st_bucket_done) {
3104 st->offset = 0;
3105 st->bucket++;
3106 if (st->state == TCP_SEQ_STATE_LISTENING &&
3107 st->bucket > hinfo->lhash2_mask) {
3108 st->state = TCP_SEQ_STATE_ESTABLISHED;
3109 st->bucket = 0;
3110 }
3111 }
3112
3113again:
3114 /* Get a new batch */
3115 iter->cur_sk = 0;
3116 iter->end_sk = 0;
3117 iter->st_bucket_done = false;
3118
3119 sk = tcp_seek_last_pos(seq);
3120 if (!sk)
3121 return NULL; /* Done */
3122
3123 if (st->state == TCP_SEQ_STATE_LISTENING)
3124 expected = bpf_iter_tcp_listening_batch(seq, sk);
3125 else
3126 expected = bpf_iter_tcp_established_batch(seq, sk);
3127
3128 if (iter->end_sk == expected) {
3129 iter->st_bucket_done = true;
3130 return sk;
3131 }
3132
3133 if (!resized && !bpf_iter_tcp_realloc_batch(iter, expected * 3 / 2)) {
3134 resized = true;
3135 goto again;
3136 }
3137
3138 return sk;
3139}
3140
3141static void *bpf_iter_tcp_seq_start(struct seq_file *seq, loff_t *pos)
3142{
3143 /* bpf iter does not support lseek, so it always
3144 * continue from where it was stop()-ped.
3145 */
3146 if (*pos)
3147 return bpf_iter_tcp_batch(seq);
3148
3149 return SEQ_START_TOKEN;
3150}
3151
3152static void *bpf_iter_tcp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3153{
3154 struct bpf_tcp_iter_state *iter = seq->private;
3155 struct tcp_iter_state *st = &iter->state;
3156 struct sock *sk;
3157
3158 /* Whenever seq_next() is called, the iter->cur_sk is
3159 * done with seq_show(), so advance to the next sk in
3160 * the batch.
3161 */
3162 if (iter->cur_sk < iter->end_sk) {
3163 /* Keeping st->num consistent in tcp_iter_state.
3164 * bpf_iter_tcp does not use st->num.
3165 * meta.seq_num is used instead.
3166 */
3167 st->num++;
3168 /* Move st->offset to the next sk in the bucket such that
3169 * the future start() will resume at st->offset in
3170 * st->bucket. See tcp_seek_last_pos().
3171 */
3172 st->offset++;
3173 sock_gen_put(iter->batch[iter->cur_sk++]);
3174 }
3175
3176 if (iter->cur_sk < iter->end_sk)
3177 sk = iter->batch[iter->cur_sk];
3178 else
3179 sk = bpf_iter_tcp_batch(seq);
3180
3181 ++*pos;
3182 /* Keeping st->last_pos consistent in tcp_iter_state.
3183 * bpf iter does not do lseek, so st->last_pos always equals to *pos.
3184 */
3185 st->last_pos = *pos;
3186 return sk;
3187}
3188
3189static int bpf_iter_tcp_seq_show(struct seq_file *seq, void *v)
3190{
3191 struct bpf_iter_meta meta;
3192 struct bpf_prog *prog;
3193 struct sock *sk = v;
3194 uid_t uid;
3195 int ret;
3196
3197 if (v == SEQ_START_TOKEN)
3198 return 0;
3199
3200 if (sk_fullsock(sk))
3201 lock_sock(sk);
3202
3203 if (unlikely(sk_unhashed(sk))) {
3204 ret = SEQ_SKIP;
3205 goto unlock;
3206 }
3207
3208 if (sk->sk_state == TCP_TIME_WAIT) {
3209 uid = 0;
3210 } else if (sk->sk_state == TCP_NEW_SYN_RECV) {
3211 const struct request_sock *req = v;
3212
3213 uid = from_kuid_munged(seq_user_ns(seq),
3214 sock_i_uid(req->rsk_listener));
3215 } else {
3216 uid = from_kuid_munged(seq_user_ns(seq), sock_i_uid(sk));
3217 }
3218
3219 meta.seq = seq;
3220 prog = bpf_iter_get_info(&meta, false);
3221 ret = tcp_prog_seq_show(prog, &meta, v, uid);
3222
3223unlock:
3224 if (sk_fullsock(sk))
3225 release_sock(sk);
3226 return ret;
3227
3228}
3229
3230static void bpf_iter_tcp_seq_stop(struct seq_file *seq, void *v)
3231{
3232 struct bpf_tcp_iter_state *iter = seq->private;
3233 struct bpf_iter_meta meta;
3234 struct bpf_prog *prog;
3235
3236 if (!v) {
3237 meta.seq = seq;
3238 prog = bpf_iter_get_info(&meta, true);
3239 if (prog)
3240 (void)tcp_prog_seq_show(prog, &meta, v, 0);
3241 }
3242
3243 if (iter->cur_sk < iter->end_sk) {
3244 bpf_iter_tcp_put_batch(iter);
3245 iter->st_bucket_done = false;
3246 }
3247}
3248
3249static const struct seq_operations bpf_iter_tcp_seq_ops = {
3250 .show = bpf_iter_tcp_seq_show,
3251 .start = bpf_iter_tcp_seq_start,
3252 .next = bpf_iter_tcp_seq_next,
3253 .stop = bpf_iter_tcp_seq_stop,
3254};
3255#endif
3256static unsigned short seq_file_family(const struct seq_file *seq)
3257{
3258 const struct tcp_seq_afinfo *afinfo;
3259
3260#ifdef CONFIG_BPF_SYSCALL
3261 /* Iterated from bpf_iter. Let the bpf prog to filter instead. */
3262 if (seq->op == &bpf_iter_tcp_seq_ops)
3263 return AF_UNSPEC;
3264#endif
3265
3266 /* Iterated from proc fs */
3267 afinfo = pde_data(file_inode(seq->file));
3268 return afinfo->family;
3269}
3270
3271static const struct seq_operations tcp4_seq_ops = {
3272 .show = tcp4_seq_show,
3273 .start = tcp_seq_start,
3274 .next = tcp_seq_next,
3275 .stop = tcp_seq_stop,
3276};
3277
3278static struct tcp_seq_afinfo tcp4_seq_afinfo = {
3279 .family = AF_INET,
3280};
3281
3282static int __net_init tcp4_proc_init_net(struct net *net)
3283{
3284 if (!proc_create_net_data("tcp", 0444, net->proc_net, &tcp4_seq_ops,
3285 sizeof(struct tcp_iter_state), &tcp4_seq_afinfo))
3286 return -ENOMEM;
3287 return 0;
3288}
3289
3290static void __net_exit tcp4_proc_exit_net(struct net *net)
3291{
3292 remove_proc_entry("tcp", net->proc_net);
3293}
3294
3295static struct pernet_operations tcp4_net_ops = {
3296 .init = tcp4_proc_init_net,
3297 .exit = tcp4_proc_exit_net,
3298};
3299
3300int __init tcp4_proc_init(void)
3301{
3302 return register_pernet_subsys(&tcp4_net_ops);
3303}
3304
3305void tcp4_proc_exit(void)
3306{
3307 unregister_pernet_subsys(&tcp4_net_ops);
3308}
3309#endif /* CONFIG_PROC_FS */
3310
3311/* @wake is one when sk_stream_write_space() calls us.
3312 * This sends EPOLLOUT only if notsent_bytes is half the limit.
3313 * This mimics the strategy used in sock_def_write_space().
3314 */
3315bool tcp_stream_memory_free(const struct sock *sk, int wake)
3316{
3317 const struct tcp_sock *tp = tcp_sk(sk);
3318 u32 notsent_bytes = READ_ONCE(tp->write_seq) -
3319 READ_ONCE(tp->snd_nxt);
3320
3321 return (notsent_bytes << wake) < tcp_notsent_lowat(tp);
3322}
3323EXPORT_SYMBOL(tcp_stream_memory_free);
3324
3325struct proto tcp_prot = {
3326 .name = "TCP",
3327 .owner = THIS_MODULE,
3328 .close = tcp_close,
3329 .pre_connect = tcp_v4_pre_connect,
3330 .connect = tcp_v4_connect,
3331 .disconnect = tcp_disconnect,
3332 .accept = inet_csk_accept,
3333 .ioctl = tcp_ioctl,
3334 .init = tcp_v4_init_sock,
3335 .destroy = tcp_v4_destroy_sock,
3336 .shutdown = tcp_shutdown,
3337 .setsockopt = tcp_setsockopt,
3338 .getsockopt = tcp_getsockopt,
3339 .bpf_bypass_getsockopt = tcp_bpf_bypass_getsockopt,
3340 .keepalive = tcp_set_keepalive,
3341 .recvmsg = tcp_recvmsg,
3342 .sendmsg = tcp_sendmsg,
3343 .splice_eof = tcp_splice_eof,
3344 .backlog_rcv = tcp_v4_do_rcv,
3345 .release_cb = tcp_release_cb,
3346 .hash = inet_hash,
3347 .unhash = inet_unhash,
3348 .get_port = inet_csk_get_port,
3349 .put_port = inet_put_port,
3350#ifdef CONFIG_BPF_SYSCALL
3351 .psock_update_sk_prot = tcp_bpf_update_proto,
3352#endif
3353 .enter_memory_pressure = tcp_enter_memory_pressure,
3354 .leave_memory_pressure = tcp_leave_memory_pressure,
3355 .stream_memory_free = tcp_stream_memory_free,
3356 .sockets_allocated = &tcp_sockets_allocated,
3357 .orphan_count = &tcp_orphan_count,
3358
3359 .memory_allocated = &tcp_memory_allocated,
3360 .per_cpu_fw_alloc = &tcp_memory_per_cpu_fw_alloc,
3361
3362 .memory_pressure = &tcp_memory_pressure,
3363 .sysctl_mem = sysctl_tcp_mem,
3364 .sysctl_wmem_offset = offsetof(struct net, ipv4.sysctl_tcp_wmem),
3365 .sysctl_rmem_offset = offsetof(struct net, ipv4.sysctl_tcp_rmem),
3366 .max_header = MAX_TCP_HEADER,
3367 .obj_size = sizeof(struct tcp_sock),
3368 .slab_flags = SLAB_TYPESAFE_BY_RCU,
3369 .twsk_prot = &tcp_timewait_sock_ops,
3370 .rsk_prot = &tcp_request_sock_ops,
3371 .h.hashinfo = NULL,
3372 .no_autobind = true,
3373 .diag_destroy = tcp_abort,
3374};
3375EXPORT_SYMBOL(tcp_prot);
3376
3377static void __net_exit tcp_sk_exit(struct net *net)
3378{
3379 if (net->ipv4.tcp_congestion_control)
3380 bpf_module_put(net->ipv4.tcp_congestion_control,
3381 net->ipv4.tcp_congestion_control->owner);
3382}
3383
3384static void __net_init tcp_set_hashinfo(struct net *net)
3385{
3386 struct inet_hashinfo *hinfo;
3387 unsigned int ehash_entries;
3388 struct net *old_net;
3389
3390 if (net_eq(net, &init_net))
3391 goto fallback;
3392
3393 old_net = current->nsproxy->net_ns;
3394 ehash_entries = READ_ONCE(old_net->ipv4.sysctl_tcp_child_ehash_entries);
3395 if (!ehash_entries)
3396 goto fallback;
3397
3398 ehash_entries = roundup_pow_of_two(ehash_entries);
3399 hinfo = inet_pernet_hashinfo_alloc(&tcp_hashinfo, ehash_entries);
3400 if (!hinfo) {
3401 pr_warn("Failed to allocate TCP ehash (entries: %u) "
3402 "for a netns, fallback to the global one\n",
3403 ehash_entries);
3404fallback:
3405 hinfo = &tcp_hashinfo;
3406 ehash_entries = tcp_hashinfo.ehash_mask + 1;
3407 }
3408
3409 net->ipv4.tcp_death_row.hashinfo = hinfo;
3410 net->ipv4.tcp_death_row.sysctl_max_tw_buckets = ehash_entries / 2;
3411 net->ipv4.sysctl_max_syn_backlog = max(128U, ehash_entries / 128);
3412}
3413
3414static int __net_init tcp_sk_init(struct net *net)
3415{
3416 net->ipv4.sysctl_tcp_ecn = 2;
3417 net->ipv4.sysctl_tcp_ecn_fallback = 1;
3418
3419 net->ipv4.sysctl_tcp_base_mss = TCP_BASE_MSS;
3420 net->ipv4.sysctl_tcp_min_snd_mss = TCP_MIN_SND_MSS;
3421 net->ipv4.sysctl_tcp_probe_threshold = TCP_PROBE_THRESHOLD;
3422 net->ipv4.sysctl_tcp_probe_interval = TCP_PROBE_INTERVAL;
3423 net->ipv4.sysctl_tcp_mtu_probe_floor = TCP_MIN_SND_MSS;
3424
3425 net->ipv4.sysctl_tcp_keepalive_time = TCP_KEEPALIVE_TIME;
3426 net->ipv4.sysctl_tcp_keepalive_probes = TCP_KEEPALIVE_PROBES;
3427 net->ipv4.sysctl_tcp_keepalive_intvl = TCP_KEEPALIVE_INTVL;
3428
3429 net->ipv4.sysctl_tcp_syn_retries = TCP_SYN_RETRIES;
3430 net->ipv4.sysctl_tcp_synack_retries = TCP_SYNACK_RETRIES;
3431 net->ipv4.sysctl_tcp_syncookies = 1;
3432 net->ipv4.sysctl_tcp_reordering = TCP_FASTRETRANS_THRESH;
3433 net->ipv4.sysctl_tcp_retries1 = TCP_RETR1;
3434 net->ipv4.sysctl_tcp_retries2 = TCP_RETR2;
3435 net->ipv4.sysctl_tcp_orphan_retries = 0;
3436 net->ipv4.sysctl_tcp_fin_timeout = TCP_FIN_TIMEOUT;
3437 net->ipv4.sysctl_tcp_notsent_lowat = UINT_MAX;
3438 net->ipv4.sysctl_tcp_tw_reuse = 2;
3439 net->ipv4.sysctl_tcp_no_ssthresh_metrics_save = 1;
3440
3441 refcount_set(&net->ipv4.tcp_death_row.tw_refcount, 1);
3442 tcp_set_hashinfo(net);
3443
3444 net->ipv4.sysctl_tcp_sack = 1;
3445 net->ipv4.sysctl_tcp_window_scaling = 1;
3446 net->ipv4.sysctl_tcp_timestamps = 1;
3447 net->ipv4.sysctl_tcp_early_retrans = 3;
3448 net->ipv4.sysctl_tcp_recovery = TCP_RACK_LOSS_DETECTION;
3449 net->ipv4.sysctl_tcp_slow_start_after_idle = 1; /* By default, RFC2861 behavior. */
3450 net->ipv4.sysctl_tcp_retrans_collapse = 1;
3451 net->ipv4.sysctl_tcp_max_reordering = 300;
3452 net->ipv4.sysctl_tcp_dsack = 1;
3453 net->ipv4.sysctl_tcp_app_win = 31;
3454 net->ipv4.sysctl_tcp_adv_win_scale = 1;
3455 net->ipv4.sysctl_tcp_frto = 2;
3456 net->ipv4.sysctl_tcp_moderate_rcvbuf = 1;
3457 /* This limits the percentage of the congestion window which we
3458 * will allow a single TSO frame to consume. Building TSO frames
3459 * which are too large can cause TCP streams to be bursty.
3460 */
3461 net->ipv4.sysctl_tcp_tso_win_divisor = 3;
3462 /* Default TSQ limit of 16 TSO segments */
3463 net->ipv4.sysctl_tcp_limit_output_bytes = 16 * 65536;
3464
3465 /* rfc5961 challenge ack rate limiting, per net-ns, disabled by default. */
3466 net->ipv4.sysctl_tcp_challenge_ack_limit = INT_MAX;
3467
3468 net->ipv4.sysctl_tcp_min_tso_segs = 2;
3469 net->ipv4.sysctl_tcp_tso_rtt_log = 9; /* 2^9 = 512 usec */
3470 net->ipv4.sysctl_tcp_min_rtt_wlen = 300;
3471 net->ipv4.sysctl_tcp_autocorking = 1;
3472 net->ipv4.sysctl_tcp_invalid_ratelimit = HZ/2;
3473 net->ipv4.sysctl_tcp_pacing_ss_ratio = 200;
3474 net->ipv4.sysctl_tcp_pacing_ca_ratio = 120;
3475 if (net != &init_net) {
3476 memcpy(net->ipv4.sysctl_tcp_rmem,
3477 init_net.ipv4.sysctl_tcp_rmem,
3478 sizeof(init_net.ipv4.sysctl_tcp_rmem));
3479 memcpy(net->ipv4.sysctl_tcp_wmem,
3480 init_net.ipv4.sysctl_tcp_wmem,
3481 sizeof(init_net.ipv4.sysctl_tcp_wmem));
3482 }
3483 net->ipv4.sysctl_tcp_comp_sack_delay_ns = NSEC_PER_MSEC;
3484 net->ipv4.sysctl_tcp_comp_sack_slack_ns = 100 * NSEC_PER_USEC;
3485 net->ipv4.sysctl_tcp_comp_sack_nr = 44;
3486 net->ipv4.sysctl_tcp_backlog_ack_defer = 1;
3487 net->ipv4.sysctl_tcp_fastopen = TFO_CLIENT_ENABLE;
3488 net->ipv4.sysctl_tcp_fastopen_blackhole_timeout = 0;
3489 atomic_set(&net->ipv4.tfo_active_disable_times, 0);
3490
3491 /* Set default values for PLB */
3492 net->ipv4.sysctl_tcp_plb_enabled = 0; /* Disabled by default */
3493 net->ipv4.sysctl_tcp_plb_idle_rehash_rounds = 3;
3494 net->ipv4.sysctl_tcp_plb_rehash_rounds = 12;
3495 net->ipv4.sysctl_tcp_plb_suspend_rto_sec = 60;
3496 /* Default congestion threshold for PLB to mark a round is 50% */
3497 net->ipv4.sysctl_tcp_plb_cong_thresh = (1 << TCP_PLB_SCALE) / 2;
3498
3499 /* Reno is always built in */
3500 if (!net_eq(net, &init_net) &&
3501 bpf_try_module_get(init_net.ipv4.tcp_congestion_control,
3502 init_net.ipv4.tcp_congestion_control->owner))
3503 net->ipv4.tcp_congestion_control = init_net.ipv4.tcp_congestion_control;
3504 else
3505 net->ipv4.tcp_congestion_control = &tcp_reno;
3506
3507 net->ipv4.sysctl_tcp_syn_linear_timeouts = 4;
3508 net->ipv4.sysctl_tcp_shrink_window = 0;
3509
3510 net->ipv4.sysctl_tcp_pingpong_thresh = 1;
3511
3512 return 0;
3513}
3514
3515static void __net_exit tcp_sk_exit_batch(struct list_head *net_exit_list)
3516{
3517 struct net *net;
3518
3519 tcp_twsk_purge(net_exit_list, AF_INET);
3520
3521 list_for_each_entry(net, net_exit_list, exit_list) {
3522 inet_pernet_hashinfo_free(net->ipv4.tcp_death_row.hashinfo);
3523 WARN_ON_ONCE(!refcount_dec_and_test(&net->ipv4.tcp_death_row.tw_refcount));
3524 tcp_fastopen_ctx_destroy(net);
3525 }
3526}
3527
3528static struct pernet_operations __net_initdata tcp_sk_ops = {
3529 .init = tcp_sk_init,
3530 .exit = tcp_sk_exit,
3531 .exit_batch = tcp_sk_exit_batch,
3532};
3533
3534#if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
3535DEFINE_BPF_ITER_FUNC(tcp, struct bpf_iter_meta *meta,
3536 struct sock_common *sk_common, uid_t uid)
3537
3538#define INIT_BATCH_SZ 16
3539
3540static int bpf_iter_init_tcp(void *priv_data, struct bpf_iter_aux_info *aux)
3541{
3542 struct bpf_tcp_iter_state *iter = priv_data;
3543 int err;
3544
3545 err = bpf_iter_init_seq_net(priv_data, aux);
3546 if (err)
3547 return err;
3548
3549 err = bpf_iter_tcp_realloc_batch(iter, INIT_BATCH_SZ);
3550 if (err) {
3551 bpf_iter_fini_seq_net(priv_data);
3552 return err;
3553 }
3554
3555 return 0;
3556}
3557
3558static void bpf_iter_fini_tcp(void *priv_data)
3559{
3560 struct bpf_tcp_iter_state *iter = priv_data;
3561
3562 bpf_iter_fini_seq_net(priv_data);
3563 kvfree(iter->batch);
3564}
3565
3566static const struct bpf_iter_seq_info tcp_seq_info = {
3567 .seq_ops = &bpf_iter_tcp_seq_ops,
3568 .init_seq_private = bpf_iter_init_tcp,
3569 .fini_seq_private = bpf_iter_fini_tcp,
3570 .seq_priv_size = sizeof(struct bpf_tcp_iter_state),
3571};
3572
3573static const struct bpf_func_proto *
3574bpf_iter_tcp_get_func_proto(enum bpf_func_id func_id,
3575 const struct bpf_prog *prog)
3576{
3577 switch (func_id) {
3578 case BPF_FUNC_setsockopt:
3579 return &bpf_sk_setsockopt_proto;
3580 case BPF_FUNC_getsockopt:
3581 return &bpf_sk_getsockopt_proto;
3582 default:
3583 return NULL;
3584 }
3585}
3586
3587static struct bpf_iter_reg tcp_reg_info = {
3588 .target = "tcp",
3589 .ctx_arg_info_size = 1,
3590 .ctx_arg_info = {
3591 { offsetof(struct bpf_iter__tcp, sk_common),
3592 PTR_TO_BTF_ID_OR_NULL | PTR_TRUSTED },
3593 },
3594 .get_func_proto = bpf_iter_tcp_get_func_proto,
3595 .seq_info = &tcp_seq_info,
3596};
3597
3598static void __init bpf_iter_register(void)
3599{
3600 tcp_reg_info.ctx_arg_info[0].btf_id = btf_sock_ids[BTF_SOCK_TYPE_SOCK_COMMON];
3601 if (bpf_iter_reg_target(&tcp_reg_info))
3602 pr_warn("Warning: could not register bpf iterator tcp\n");
3603}
3604
3605#endif
3606
3607void __init tcp_v4_init(void)
3608{
3609 int cpu, res;
3610
3611 for_each_possible_cpu(cpu) {
3612 struct sock *sk;
3613
3614 res = inet_ctl_sock_create(&sk, PF_INET, SOCK_RAW,
3615 IPPROTO_TCP, &init_net);
3616 if (res)
3617 panic("Failed to create the TCP control socket.\n");
3618 sock_set_flag(sk, SOCK_USE_WRITE_QUEUE);
3619
3620 /* Please enforce IP_DF and IPID==0 for RST and
3621 * ACK sent in SYN-RECV and TIME-WAIT state.
3622 */
3623 inet_sk(sk)->pmtudisc = IP_PMTUDISC_DO;
3624
3625 per_cpu(ipv4_tcp_sk, cpu) = sk;
3626 }
3627 if (register_pernet_subsys(&tcp_sk_ops))
3628 panic("Failed to create the TCP control socket.\n");
3629
3630#if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
3631 bpf_iter_register();
3632#endif
3633}