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#include <linux/btf_ids.h>
80
81#include <crypto/hash.h>
82#include <linux/scatterlist.h>
83
84#include <trace/events/tcp.h>
85
86#ifdef CONFIG_TCP_MD5SIG
87static int tcp_v4_md5_hash_hdr(char *md5_hash, const struct tcp_md5sig_key *key,
88 __be32 daddr, __be32 saddr, const struct tcphdr *th);
89#endif
90
91struct inet_hashinfo tcp_hashinfo;
92EXPORT_SYMBOL(tcp_hashinfo);
93
94static DEFINE_PER_CPU(struct sock *, ipv4_tcp_sk);
95
96static u32 tcp_v4_init_seq(const struct sk_buff *skb)
97{
98 return secure_tcp_seq(ip_hdr(skb)->daddr,
99 ip_hdr(skb)->saddr,
100 tcp_hdr(skb)->dest,
101 tcp_hdr(skb)->source);
102}
103
104static u32 tcp_v4_init_ts_off(const struct net *net, const struct sk_buff *skb)
105{
106 return secure_tcp_ts_off(net, ip_hdr(skb)->daddr, ip_hdr(skb)->saddr);
107}
108
109int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp)
110{
111 int reuse = READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_tw_reuse);
112 const struct inet_timewait_sock *tw = inet_twsk(sktw);
113 const struct tcp_timewait_sock *tcptw = tcp_twsk(sktw);
114 struct tcp_sock *tp = tcp_sk(sk);
115
116 if (reuse == 2) {
117 /* Still does not detect *everything* that goes through
118 * lo, since we require a loopback src or dst address
119 * or direct binding to 'lo' interface.
120 */
121 bool loopback = false;
122 if (tw->tw_bound_dev_if == LOOPBACK_IFINDEX)
123 loopback = true;
124#if IS_ENABLED(CONFIG_IPV6)
125 if (tw->tw_family == AF_INET6) {
126 if (ipv6_addr_loopback(&tw->tw_v6_daddr) ||
127 ipv6_addr_v4mapped_loopback(&tw->tw_v6_daddr) ||
128 ipv6_addr_loopback(&tw->tw_v6_rcv_saddr) ||
129 ipv6_addr_v4mapped_loopback(&tw->tw_v6_rcv_saddr))
130 loopback = true;
131 } else
132#endif
133 {
134 if (ipv4_is_loopback(tw->tw_daddr) ||
135 ipv4_is_loopback(tw->tw_rcv_saddr))
136 loopback = true;
137 }
138 if (!loopback)
139 reuse = 0;
140 }
141
142 /* With PAWS, it is safe from the viewpoint
143 of data integrity. Even without PAWS it is safe provided sequence
144 spaces do not overlap i.e. at data rates <= 80Mbit/sec.
145
146 Actually, the idea is close to VJ's one, only timestamp cache is
147 held not per host, but per port pair and TW bucket is used as state
148 holder.
149
150 If TW bucket has been already destroyed we fall back to VJ's scheme
151 and use initial timestamp retrieved from peer table.
152 */
153 if (tcptw->tw_ts_recent_stamp &&
154 (!twp || (reuse && time_after32(ktime_get_seconds(),
155 tcptw->tw_ts_recent_stamp)))) {
156 /* In case of repair and re-using TIME-WAIT sockets we still
157 * want to be sure that it is safe as above but honor the
158 * sequence numbers and time stamps set as part of the repair
159 * process.
160 *
161 * Without this check re-using a TIME-WAIT socket with TCP
162 * repair would accumulate a -1 on the repair assigned
163 * sequence number. The first time it is reused the sequence
164 * is -1, the second time -2, etc. This fixes that issue
165 * without appearing to create any others.
166 */
167 if (likely(!tp->repair)) {
168 u32 seq = tcptw->tw_snd_nxt + 65535 + 2;
169
170 if (!seq)
171 seq = 1;
172 WRITE_ONCE(tp->write_seq, seq);
173 tp->rx_opt.ts_recent = tcptw->tw_ts_recent;
174 tp->rx_opt.ts_recent_stamp = tcptw->tw_ts_recent_stamp;
175 }
176 sock_hold(sktw);
177 return 1;
178 }
179
180 return 0;
181}
182EXPORT_SYMBOL_GPL(tcp_twsk_unique);
183
184static int tcp_v4_pre_connect(struct sock *sk, struct sockaddr *uaddr,
185 int addr_len)
186{
187 /* This check is replicated from tcp_v4_connect() and intended to
188 * prevent BPF program called below from accessing bytes that are out
189 * of the bound specified by user in addr_len.
190 */
191 if (addr_len < sizeof(struct sockaddr_in))
192 return -EINVAL;
193
194 sock_owned_by_me(sk);
195
196 return BPF_CGROUP_RUN_PROG_INET4_CONNECT(sk, uaddr);
197}
198
199/* This will initiate an outgoing connection. */
200int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len)
201{
202 struct sockaddr_in *usin = (struct sockaddr_in *)uaddr;
203 struct inet_timewait_death_row *tcp_death_row;
204 struct inet_sock *inet = inet_sk(sk);
205 struct tcp_sock *tp = tcp_sk(sk);
206 struct ip_options_rcu *inet_opt;
207 struct net *net = sock_net(sk);
208 __be16 orig_sport, orig_dport;
209 __be32 daddr, nexthop;
210 struct flowi4 *fl4;
211 struct rtable *rt;
212 int err;
213
214 if (addr_len < sizeof(struct sockaddr_in))
215 return -EINVAL;
216
217 if (usin->sin_family != AF_INET)
218 return -EAFNOSUPPORT;
219
220 nexthop = daddr = usin->sin_addr.s_addr;
221 inet_opt = rcu_dereference_protected(inet->inet_opt,
222 lockdep_sock_is_held(sk));
223 if (inet_opt && inet_opt->opt.srr) {
224 if (!daddr)
225 return -EINVAL;
226 nexthop = inet_opt->opt.faddr;
227 }
228
229 orig_sport = inet->inet_sport;
230 orig_dport = usin->sin_port;
231 fl4 = &inet->cork.fl.u.ip4;
232 rt = ip_route_connect(fl4, nexthop, inet->inet_saddr,
233 sk->sk_bound_dev_if, IPPROTO_TCP, orig_sport,
234 orig_dport, sk);
235 if (IS_ERR(rt)) {
236 err = PTR_ERR(rt);
237 if (err == -ENETUNREACH)
238 IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES);
239 return err;
240 }
241
242 if (rt->rt_flags & (RTCF_MULTICAST | RTCF_BROADCAST)) {
243 ip_rt_put(rt);
244 return -ENETUNREACH;
245 }
246
247 if (!inet_opt || !inet_opt->opt.srr)
248 daddr = fl4->daddr;
249
250 tcp_death_row = &sock_net(sk)->ipv4.tcp_death_row;
251
252 if (!inet->inet_saddr) {
253 err = inet_bhash2_update_saddr(sk, &fl4->saddr, AF_INET);
254 if (err) {
255 ip_rt_put(rt);
256 return err;
257 }
258 } else {
259 sk_rcv_saddr_set(sk, inet->inet_saddr);
260 }
261
262 if (tp->rx_opt.ts_recent_stamp && inet->inet_daddr != daddr) {
263 /* Reset inherited state */
264 tp->rx_opt.ts_recent = 0;
265 tp->rx_opt.ts_recent_stamp = 0;
266 if (likely(!tp->repair))
267 WRITE_ONCE(tp->write_seq, 0);
268 }
269
270 inet->inet_dport = usin->sin_port;
271 sk_daddr_set(sk, daddr);
272
273 inet_csk(sk)->icsk_ext_hdr_len = 0;
274 if (inet_opt)
275 inet_csk(sk)->icsk_ext_hdr_len = inet_opt->opt.optlen;
276
277 tp->rx_opt.mss_clamp = TCP_MSS_DEFAULT;
278
279 /* Socket identity is still unknown (sport may be zero).
280 * However we set state to SYN-SENT and not releasing socket
281 * lock select source port, enter ourselves into the hash tables and
282 * complete initialization after this.
283 */
284 tcp_set_state(sk, TCP_SYN_SENT);
285 err = inet_hash_connect(tcp_death_row, sk);
286 if (err)
287 goto failure;
288
289 sk_set_txhash(sk);
290
291 rt = ip_route_newports(fl4, rt, orig_sport, orig_dport,
292 inet->inet_sport, inet->inet_dport, sk);
293 if (IS_ERR(rt)) {
294 err = PTR_ERR(rt);
295 rt = NULL;
296 goto failure;
297 }
298 /* OK, now commit destination to socket. */
299 sk->sk_gso_type = SKB_GSO_TCPV4;
300 sk_setup_caps(sk, &rt->dst);
301 rt = NULL;
302
303 if (likely(!tp->repair)) {
304 if (!tp->write_seq)
305 WRITE_ONCE(tp->write_seq,
306 secure_tcp_seq(inet->inet_saddr,
307 inet->inet_daddr,
308 inet->inet_sport,
309 usin->sin_port));
310 tp->tsoffset = secure_tcp_ts_off(net, inet->inet_saddr,
311 inet->inet_daddr);
312 }
313
314 inet->inet_id = get_random_u16();
315
316 if (tcp_fastopen_defer_connect(sk, &err))
317 return err;
318 if (err)
319 goto failure;
320
321 err = tcp_connect(sk);
322
323 if (err)
324 goto failure;
325
326 return 0;
327
328failure:
329 /*
330 * This unhashes the socket and releases the local port,
331 * if necessary.
332 */
333 tcp_set_state(sk, TCP_CLOSE);
334 inet_bhash2_reset_saddr(sk);
335 ip_rt_put(rt);
336 sk->sk_route_caps = 0;
337 inet->inet_dport = 0;
338 return err;
339}
340EXPORT_SYMBOL(tcp_v4_connect);
341
342/*
343 * This routine reacts to ICMP_FRAG_NEEDED mtu indications as defined in RFC1191.
344 * It can be called through tcp_release_cb() if socket was owned by user
345 * at the time tcp_v4_err() was called to handle ICMP message.
346 */
347void tcp_v4_mtu_reduced(struct sock *sk)
348{
349 struct inet_sock *inet = inet_sk(sk);
350 struct dst_entry *dst;
351 u32 mtu;
352
353 if ((1 << sk->sk_state) & (TCPF_LISTEN | TCPF_CLOSE))
354 return;
355 mtu = READ_ONCE(tcp_sk(sk)->mtu_info);
356 dst = inet_csk_update_pmtu(sk, mtu);
357 if (!dst)
358 return;
359
360 /* Something is about to be wrong... Remember soft error
361 * for the case, if this connection will not able to recover.
362 */
363 if (mtu < dst_mtu(dst) && ip_dont_fragment(sk, dst))
364 sk->sk_err_soft = EMSGSIZE;
365
366 mtu = dst_mtu(dst);
367
368 if (inet->pmtudisc != IP_PMTUDISC_DONT &&
369 ip_sk_accept_pmtu(sk) &&
370 inet_csk(sk)->icsk_pmtu_cookie > mtu) {
371 tcp_sync_mss(sk, mtu);
372
373 /* Resend the TCP packet because it's
374 * clear that the old packet has been
375 * dropped. This is the new "fast" path mtu
376 * discovery.
377 */
378 tcp_simple_retransmit(sk);
379 } /* else let the usual retransmit timer handle it */
380}
381EXPORT_SYMBOL(tcp_v4_mtu_reduced);
382
383static void do_redirect(struct sk_buff *skb, struct sock *sk)
384{
385 struct dst_entry *dst = __sk_dst_check(sk, 0);
386
387 if (dst)
388 dst->ops->redirect(dst, sk, skb);
389}
390
391
392/* handle ICMP messages on TCP_NEW_SYN_RECV request sockets */
393void tcp_req_err(struct sock *sk, u32 seq, bool abort)
394{
395 struct request_sock *req = inet_reqsk(sk);
396 struct net *net = sock_net(sk);
397
398 /* ICMPs are not backlogged, hence we cannot get
399 * an established socket here.
400 */
401 if (seq != tcp_rsk(req)->snt_isn) {
402 __NET_INC_STATS(net, LINUX_MIB_OUTOFWINDOWICMPS);
403 } else if (abort) {
404 /*
405 * Still in SYN_RECV, just remove it silently.
406 * There is no good way to pass the error to the newly
407 * created socket, and POSIX does not want network
408 * errors returned from accept().
409 */
410 inet_csk_reqsk_queue_drop(req->rsk_listener, req);
411 tcp_listendrop(req->rsk_listener);
412 }
413 reqsk_put(req);
414}
415EXPORT_SYMBOL(tcp_req_err);
416
417/* TCP-LD (RFC 6069) logic */
418void tcp_ld_RTO_revert(struct sock *sk, u32 seq)
419{
420 struct inet_connection_sock *icsk = inet_csk(sk);
421 struct tcp_sock *tp = tcp_sk(sk);
422 struct sk_buff *skb;
423 s32 remaining;
424 u32 delta_us;
425
426 if (sock_owned_by_user(sk))
427 return;
428
429 if (seq != tp->snd_una || !icsk->icsk_retransmits ||
430 !icsk->icsk_backoff)
431 return;
432
433 skb = tcp_rtx_queue_head(sk);
434 if (WARN_ON_ONCE(!skb))
435 return;
436
437 icsk->icsk_backoff--;
438 icsk->icsk_rto = tp->srtt_us ? __tcp_set_rto(tp) : TCP_TIMEOUT_INIT;
439 icsk->icsk_rto = inet_csk_rto_backoff(icsk, TCP_RTO_MAX);
440
441 tcp_mstamp_refresh(tp);
442 delta_us = (u32)(tp->tcp_mstamp - tcp_skb_timestamp_us(skb));
443 remaining = icsk->icsk_rto - usecs_to_jiffies(delta_us);
444
445 if (remaining > 0) {
446 inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
447 remaining, TCP_RTO_MAX);
448 } else {
449 /* RTO revert clocked out retransmission.
450 * Will retransmit now.
451 */
452 tcp_retransmit_timer(sk);
453 }
454}
455EXPORT_SYMBOL(tcp_ld_RTO_revert);
456
457/*
458 * This routine is called by the ICMP module when it gets some
459 * sort of error condition. If err < 0 then the socket should
460 * be closed and the error returned to the user. If err > 0
461 * it's just the icmp type << 8 | icmp code. After adjustment
462 * header points to the first 8 bytes of the tcp header. We need
463 * to find the appropriate port.
464 *
465 * The locking strategy used here is very "optimistic". When
466 * someone else accesses the socket the ICMP is just dropped
467 * and for some paths there is no check at all.
468 * A more general error queue to queue errors for later handling
469 * is probably better.
470 *
471 */
472
473int tcp_v4_err(struct sk_buff *skb, u32 info)
474{
475 const struct iphdr *iph = (const struct iphdr *)skb->data;
476 struct tcphdr *th = (struct tcphdr *)(skb->data + (iph->ihl << 2));
477 struct tcp_sock *tp;
478 struct inet_sock *inet;
479 const int type = icmp_hdr(skb)->type;
480 const int code = icmp_hdr(skb)->code;
481 struct sock *sk;
482 struct request_sock *fastopen;
483 u32 seq, snd_una;
484 int err;
485 struct net *net = dev_net(skb->dev);
486
487 sk = __inet_lookup_established(net, net->ipv4.tcp_death_row.hashinfo,
488 iph->daddr, th->dest, iph->saddr,
489 ntohs(th->source), inet_iif(skb), 0);
490 if (!sk) {
491 __ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
492 return -ENOENT;
493 }
494 if (sk->sk_state == TCP_TIME_WAIT) {
495 inet_twsk_put(inet_twsk(sk));
496 return 0;
497 }
498 seq = ntohl(th->seq);
499 if (sk->sk_state == TCP_NEW_SYN_RECV) {
500 tcp_req_err(sk, seq, type == ICMP_PARAMETERPROB ||
501 type == ICMP_TIME_EXCEEDED ||
502 (type == ICMP_DEST_UNREACH &&
503 (code == ICMP_NET_UNREACH ||
504 code == ICMP_HOST_UNREACH)));
505 return 0;
506 }
507
508 bh_lock_sock(sk);
509 /* If too many ICMPs get dropped on busy
510 * servers this needs to be solved differently.
511 * We do take care of PMTU discovery (RFC1191) special case :
512 * we can receive locally generated ICMP messages while socket is held.
513 */
514 if (sock_owned_by_user(sk)) {
515 if (!(type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED))
516 __NET_INC_STATS(net, LINUX_MIB_LOCKDROPPEDICMPS);
517 }
518 if (sk->sk_state == TCP_CLOSE)
519 goto out;
520
521 if (static_branch_unlikely(&ip4_min_ttl)) {
522 /* min_ttl can be changed concurrently from do_ip_setsockopt() */
523 if (unlikely(iph->ttl < READ_ONCE(inet_sk(sk)->min_ttl))) {
524 __NET_INC_STATS(net, LINUX_MIB_TCPMINTTLDROP);
525 goto out;
526 }
527 }
528
529 tp = tcp_sk(sk);
530 /* XXX (TFO) - tp->snd_una should be ISN (tcp_create_openreq_child() */
531 fastopen = rcu_dereference(tp->fastopen_rsk);
532 snd_una = fastopen ? tcp_rsk(fastopen)->snt_isn : tp->snd_una;
533 if (sk->sk_state != TCP_LISTEN &&
534 !between(seq, snd_una, tp->snd_nxt)) {
535 __NET_INC_STATS(net, LINUX_MIB_OUTOFWINDOWICMPS);
536 goto out;
537 }
538
539 switch (type) {
540 case ICMP_REDIRECT:
541 if (!sock_owned_by_user(sk))
542 do_redirect(skb, sk);
543 goto out;
544 case ICMP_SOURCE_QUENCH:
545 /* Just silently ignore these. */
546 goto out;
547 case ICMP_PARAMETERPROB:
548 err = EPROTO;
549 break;
550 case ICMP_DEST_UNREACH:
551 if (code > NR_ICMP_UNREACH)
552 goto out;
553
554 if (code == ICMP_FRAG_NEEDED) { /* PMTU discovery (RFC1191) */
555 /* We are not interested in TCP_LISTEN and open_requests
556 * (SYN-ACKs send out by Linux are always <576bytes so
557 * they should go through unfragmented).
558 */
559 if (sk->sk_state == TCP_LISTEN)
560 goto out;
561
562 WRITE_ONCE(tp->mtu_info, info);
563 if (!sock_owned_by_user(sk)) {
564 tcp_v4_mtu_reduced(sk);
565 } else {
566 if (!test_and_set_bit(TCP_MTU_REDUCED_DEFERRED, &sk->sk_tsq_flags))
567 sock_hold(sk);
568 }
569 goto out;
570 }
571
572 err = icmp_err_convert[code].errno;
573 /* check if this ICMP message allows revert of backoff.
574 * (see RFC 6069)
575 */
576 if (!fastopen &&
577 (code == ICMP_NET_UNREACH || code == ICMP_HOST_UNREACH))
578 tcp_ld_RTO_revert(sk, seq);
579 break;
580 case ICMP_TIME_EXCEEDED:
581 err = EHOSTUNREACH;
582 break;
583 default:
584 goto out;
585 }
586
587 switch (sk->sk_state) {
588 case TCP_SYN_SENT:
589 case TCP_SYN_RECV:
590 /* Only in fast or simultaneous open. If a fast open socket is
591 * already accepted it is treated as a connected one below.
592 */
593 if (fastopen && !fastopen->sk)
594 break;
595
596 ip_icmp_error(sk, skb, err, th->dest, info, (u8 *)th);
597
598 if (!sock_owned_by_user(sk)) {
599 sk->sk_err = err;
600
601 sk_error_report(sk);
602
603 tcp_done(sk);
604 } else {
605 sk->sk_err_soft = err;
606 }
607 goto out;
608 }
609
610 /* If we've already connected we will keep trying
611 * until we time out, or the user gives up.
612 *
613 * rfc1122 4.2.3.9 allows to consider as hard errors
614 * only PROTO_UNREACH and PORT_UNREACH (well, FRAG_FAILED too,
615 * but it is obsoleted by pmtu discovery).
616 *
617 * Note, that in modern internet, where routing is unreliable
618 * and in each dark corner broken firewalls sit, sending random
619 * errors ordered by their masters even this two messages finally lose
620 * their original sense (even Linux sends invalid PORT_UNREACHs)
621 *
622 * Now we are in compliance with RFCs.
623 * --ANK (980905)
624 */
625
626 inet = inet_sk(sk);
627 if (!sock_owned_by_user(sk) && inet->recverr) {
628 sk->sk_err = err;
629 sk_error_report(sk);
630 } else { /* Only an error on timeout */
631 sk->sk_err_soft = err;
632 }
633
634out:
635 bh_unlock_sock(sk);
636 sock_put(sk);
637 return 0;
638}
639
640void __tcp_v4_send_check(struct sk_buff *skb, __be32 saddr, __be32 daddr)
641{
642 struct tcphdr *th = tcp_hdr(skb);
643
644 th->check = ~tcp_v4_check(skb->len, saddr, daddr, 0);
645 skb->csum_start = skb_transport_header(skb) - skb->head;
646 skb->csum_offset = offsetof(struct tcphdr, check);
647}
648
649/* This routine computes an IPv4 TCP checksum. */
650void tcp_v4_send_check(struct sock *sk, struct sk_buff *skb)
651{
652 const struct inet_sock *inet = inet_sk(sk);
653
654 __tcp_v4_send_check(skb, inet->inet_saddr, inet->inet_daddr);
655}
656EXPORT_SYMBOL(tcp_v4_send_check);
657
658/*
659 * This routine will send an RST to the other tcp.
660 *
661 * Someone asks: why I NEVER use socket parameters (TOS, TTL etc.)
662 * for reset.
663 * Answer: if a packet caused RST, it is not for a socket
664 * existing in our system, if it is matched to a socket,
665 * it is just duplicate segment or bug in other side's TCP.
666 * So that we build reply only basing on parameters
667 * arrived with segment.
668 * Exception: precedence violation. We do not implement it in any case.
669 */
670
671#ifdef CONFIG_TCP_MD5SIG
672#define OPTION_BYTES TCPOLEN_MD5SIG_ALIGNED
673#else
674#define OPTION_BYTES sizeof(__be32)
675#endif
676
677static void tcp_v4_send_reset(const struct sock *sk, struct sk_buff *skb)
678{
679 const struct tcphdr *th = tcp_hdr(skb);
680 struct {
681 struct tcphdr th;
682 __be32 opt[OPTION_BYTES / sizeof(__be32)];
683 } rep;
684 struct ip_reply_arg arg;
685#ifdef CONFIG_TCP_MD5SIG
686 struct tcp_md5sig_key *key = NULL;
687 const __u8 *hash_location = NULL;
688 unsigned char newhash[16];
689 int genhash;
690 struct sock *sk1 = NULL;
691#endif
692 u64 transmit_time = 0;
693 struct sock *ctl_sk;
694 struct net *net;
695
696 /* Never send a reset in response to a reset. */
697 if (th->rst)
698 return;
699
700 /* If sk not NULL, it means we did a successful lookup and incoming
701 * route had to be correct. prequeue might have dropped our dst.
702 */
703 if (!sk && skb_rtable(skb)->rt_type != RTN_LOCAL)
704 return;
705
706 /* Swap the send and the receive. */
707 memset(&rep, 0, sizeof(rep));
708 rep.th.dest = th->source;
709 rep.th.source = th->dest;
710 rep.th.doff = sizeof(struct tcphdr) / 4;
711 rep.th.rst = 1;
712
713 if (th->ack) {
714 rep.th.seq = th->ack_seq;
715 } else {
716 rep.th.ack = 1;
717 rep.th.ack_seq = htonl(ntohl(th->seq) + th->syn + th->fin +
718 skb->len - (th->doff << 2));
719 }
720
721 memset(&arg, 0, sizeof(arg));
722 arg.iov[0].iov_base = (unsigned char *)&rep;
723 arg.iov[0].iov_len = sizeof(rep.th);
724
725 net = sk ? sock_net(sk) : dev_net(skb_dst(skb)->dev);
726#ifdef CONFIG_TCP_MD5SIG
727 rcu_read_lock();
728 hash_location = tcp_parse_md5sig_option(th);
729 if (sk && sk_fullsock(sk)) {
730 const union tcp_md5_addr *addr;
731 int l3index;
732
733 /* sdif set, means packet ingressed via a device
734 * in an L3 domain and inet_iif is set to it.
735 */
736 l3index = tcp_v4_sdif(skb) ? inet_iif(skb) : 0;
737 addr = (union tcp_md5_addr *)&ip_hdr(skb)->saddr;
738 key = tcp_md5_do_lookup(sk, l3index, addr, AF_INET);
739 } else if (hash_location) {
740 const union tcp_md5_addr *addr;
741 int sdif = tcp_v4_sdif(skb);
742 int dif = inet_iif(skb);
743 int l3index;
744
745 /*
746 * active side is lost. Try to find listening socket through
747 * source port, and then find md5 key through listening socket.
748 * we are not loose security here:
749 * Incoming packet is checked with md5 hash with finding key,
750 * no RST generated if md5 hash doesn't match.
751 */
752 sk1 = __inet_lookup_listener(net, net->ipv4.tcp_death_row.hashinfo,
753 NULL, 0, ip_hdr(skb)->saddr,
754 th->source, ip_hdr(skb)->daddr,
755 ntohs(th->source), dif, sdif);
756 /* don't send rst if it can't find key */
757 if (!sk1)
758 goto out;
759
760 /* sdif set, means packet ingressed via a device
761 * in an L3 domain and dif is set to it.
762 */
763 l3index = sdif ? dif : 0;
764 addr = (union tcp_md5_addr *)&ip_hdr(skb)->saddr;
765 key = tcp_md5_do_lookup(sk1, l3index, addr, AF_INET);
766 if (!key)
767 goto out;
768
769
770 genhash = tcp_v4_md5_hash_skb(newhash, key, NULL, skb);
771 if (genhash || memcmp(hash_location, newhash, 16) != 0)
772 goto out;
773
774 }
775
776 if (key) {
777 rep.opt[0] = htonl((TCPOPT_NOP << 24) |
778 (TCPOPT_NOP << 16) |
779 (TCPOPT_MD5SIG << 8) |
780 TCPOLEN_MD5SIG);
781 /* Update length and the length the header thinks exists */
782 arg.iov[0].iov_len += TCPOLEN_MD5SIG_ALIGNED;
783 rep.th.doff = arg.iov[0].iov_len / 4;
784
785 tcp_v4_md5_hash_hdr((__u8 *) &rep.opt[1],
786 key, ip_hdr(skb)->saddr,
787 ip_hdr(skb)->daddr, &rep.th);
788 }
789#endif
790 /* Can't co-exist with TCPMD5, hence check rep.opt[0] */
791 if (rep.opt[0] == 0) {
792 __be32 mrst = mptcp_reset_option(skb);
793
794 if (mrst) {
795 rep.opt[0] = mrst;
796 arg.iov[0].iov_len += sizeof(mrst);
797 rep.th.doff = arg.iov[0].iov_len / 4;
798 }
799 }
800
801 arg.csum = csum_tcpudp_nofold(ip_hdr(skb)->daddr,
802 ip_hdr(skb)->saddr, /* XXX */
803 arg.iov[0].iov_len, IPPROTO_TCP, 0);
804 arg.csumoffset = offsetof(struct tcphdr, check) / 2;
805 arg.flags = (sk && inet_sk_transparent(sk)) ? IP_REPLY_ARG_NOSRCCHECK : 0;
806
807 /* When socket is gone, all binding information is lost.
808 * routing might fail in this case. No choice here, if we choose to force
809 * input interface, we will misroute in case of asymmetric route.
810 */
811 if (sk) {
812 arg.bound_dev_if = sk->sk_bound_dev_if;
813 if (sk_fullsock(sk))
814 trace_tcp_send_reset(sk, skb);
815 }
816
817 BUILD_BUG_ON(offsetof(struct sock, sk_bound_dev_if) !=
818 offsetof(struct inet_timewait_sock, tw_bound_dev_if));
819
820 arg.tos = ip_hdr(skb)->tos;
821 arg.uid = sock_net_uid(net, sk && sk_fullsock(sk) ? sk : NULL);
822 local_bh_disable();
823 ctl_sk = this_cpu_read(ipv4_tcp_sk);
824 sock_net_set(ctl_sk, net);
825 if (sk) {
826 ctl_sk->sk_mark = (sk->sk_state == TCP_TIME_WAIT) ?
827 inet_twsk(sk)->tw_mark : sk->sk_mark;
828 ctl_sk->sk_priority = (sk->sk_state == TCP_TIME_WAIT) ?
829 inet_twsk(sk)->tw_priority : sk->sk_priority;
830 transmit_time = tcp_transmit_time(sk);
831 xfrm_sk_clone_policy(ctl_sk, sk);
832 }
833 ip_send_unicast_reply(ctl_sk,
834 skb, &TCP_SKB_CB(skb)->header.h4.opt,
835 ip_hdr(skb)->saddr, ip_hdr(skb)->daddr,
836 &arg, arg.iov[0].iov_len,
837 transmit_time);
838
839 ctl_sk->sk_mark = 0;
840 xfrm_sk_free_policy(ctl_sk);
841 sock_net_set(ctl_sk, &init_net);
842 __TCP_INC_STATS(net, TCP_MIB_OUTSEGS);
843 __TCP_INC_STATS(net, TCP_MIB_OUTRSTS);
844 local_bh_enable();
845
846#ifdef CONFIG_TCP_MD5SIG
847out:
848 rcu_read_unlock();
849#endif
850}
851
852/* The code following below sending ACKs in SYN-RECV and TIME-WAIT states
853 outside socket context is ugly, certainly. What can I do?
854 */
855
856static void tcp_v4_send_ack(const struct sock *sk,
857 struct sk_buff *skb, u32 seq, u32 ack,
858 u32 win, u32 tsval, u32 tsecr, int oif,
859 struct tcp_md5sig_key *key,
860 int reply_flags, u8 tos)
861{
862 const struct tcphdr *th = tcp_hdr(skb);
863 struct {
864 struct tcphdr th;
865 __be32 opt[(TCPOLEN_TSTAMP_ALIGNED >> 2)
866#ifdef CONFIG_TCP_MD5SIG
867 + (TCPOLEN_MD5SIG_ALIGNED >> 2)
868#endif
869 ];
870 } rep;
871 struct net *net = sock_net(sk);
872 struct ip_reply_arg arg;
873 struct sock *ctl_sk;
874 u64 transmit_time;
875
876 memset(&rep.th, 0, sizeof(struct tcphdr));
877 memset(&arg, 0, sizeof(arg));
878
879 arg.iov[0].iov_base = (unsigned char *)&rep;
880 arg.iov[0].iov_len = sizeof(rep.th);
881 if (tsecr) {
882 rep.opt[0] = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
883 (TCPOPT_TIMESTAMP << 8) |
884 TCPOLEN_TIMESTAMP);
885 rep.opt[1] = htonl(tsval);
886 rep.opt[2] = htonl(tsecr);
887 arg.iov[0].iov_len += TCPOLEN_TSTAMP_ALIGNED;
888 }
889
890 /* Swap the send and the receive. */
891 rep.th.dest = th->source;
892 rep.th.source = th->dest;
893 rep.th.doff = arg.iov[0].iov_len / 4;
894 rep.th.seq = htonl(seq);
895 rep.th.ack_seq = htonl(ack);
896 rep.th.ack = 1;
897 rep.th.window = htons(win);
898
899#ifdef CONFIG_TCP_MD5SIG
900 if (key) {
901 int offset = (tsecr) ? 3 : 0;
902
903 rep.opt[offset++] = htonl((TCPOPT_NOP << 24) |
904 (TCPOPT_NOP << 16) |
905 (TCPOPT_MD5SIG << 8) |
906 TCPOLEN_MD5SIG);
907 arg.iov[0].iov_len += TCPOLEN_MD5SIG_ALIGNED;
908 rep.th.doff = arg.iov[0].iov_len/4;
909
910 tcp_v4_md5_hash_hdr((__u8 *) &rep.opt[offset],
911 key, ip_hdr(skb)->saddr,
912 ip_hdr(skb)->daddr, &rep.th);
913 }
914#endif
915 arg.flags = reply_flags;
916 arg.csum = csum_tcpudp_nofold(ip_hdr(skb)->daddr,
917 ip_hdr(skb)->saddr, /* XXX */
918 arg.iov[0].iov_len, IPPROTO_TCP, 0);
919 arg.csumoffset = offsetof(struct tcphdr, check) / 2;
920 if (oif)
921 arg.bound_dev_if = oif;
922 arg.tos = tos;
923 arg.uid = sock_net_uid(net, sk_fullsock(sk) ? sk : NULL);
924 local_bh_disable();
925 ctl_sk = this_cpu_read(ipv4_tcp_sk);
926 sock_net_set(ctl_sk, net);
927 ctl_sk->sk_mark = (sk->sk_state == TCP_TIME_WAIT) ?
928 inet_twsk(sk)->tw_mark : sk->sk_mark;
929 ctl_sk->sk_priority = (sk->sk_state == TCP_TIME_WAIT) ?
930 inet_twsk(sk)->tw_priority : sk->sk_priority;
931 transmit_time = tcp_transmit_time(sk);
932 ip_send_unicast_reply(ctl_sk,
933 skb, &TCP_SKB_CB(skb)->header.h4.opt,
934 ip_hdr(skb)->saddr, ip_hdr(skb)->daddr,
935 &arg, arg.iov[0].iov_len,
936 transmit_time);
937
938 ctl_sk->sk_mark = 0;
939 sock_net_set(ctl_sk, &init_net);
940 __TCP_INC_STATS(net, TCP_MIB_OUTSEGS);
941 local_bh_enable();
942}
943
944static void tcp_v4_timewait_ack(struct sock *sk, struct sk_buff *skb)
945{
946 struct inet_timewait_sock *tw = inet_twsk(sk);
947 struct tcp_timewait_sock *tcptw = tcp_twsk(sk);
948
949 tcp_v4_send_ack(sk, skb,
950 tcptw->tw_snd_nxt, tcptw->tw_rcv_nxt,
951 tcptw->tw_rcv_wnd >> tw->tw_rcv_wscale,
952 tcp_time_stamp_raw() + tcptw->tw_ts_offset,
953 tcptw->tw_ts_recent,
954 tw->tw_bound_dev_if,
955 tcp_twsk_md5_key(tcptw),
956 tw->tw_transparent ? IP_REPLY_ARG_NOSRCCHECK : 0,
957 tw->tw_tos
958 );
959
960 inet_twsk_put(tw);
961}
962
963static void tcp_v4_reqsk_send_ack(const struct sock *sk, struct sk_buff *skb,
964 struct request_sock *req)
965{
966 const union tcp_md5_addr *addr;
967 int l3index;
968
969 /* sk->sk_state == TCP_LISTEN -> for regular TCP_SYN_RECV
970 * sk->sk_state == TCP_SYN_RECV -> for Fast Open.
971 */
972 u32 seq = (sk->sk_state == TCP_LISTEN) ? tcp_rsk(req)->snt_isn + 1 :
973 tcp_sk(sk)->snd_nxt;
974
975 /* RFC 7323 2.3
976 * The window field (SEG.WND) of every outgoing segment, with the
977 * exception of <SYN> segments, MUST be right-shifted by
978 * Rcv.Wind.Shift bits:
979 */
980 addr = (union tcp_md5_addr *)&ip_hdr(skb)->saddr;
981 l3index = tcp_v4_sdif(skb) ? inet_iif(skb) : 0;
982 tcp_v4_send_ack(sk, skb, seq,
983 tcp_rsk(req)->rcv_nxt,
984 req->rsk_rcv_wnd >> inet_rsk(req)->rcv_wscale,
985 tcp_time_stamp_raw() + tcp_rsk(req)->ts_off,
986 req->ts_recent,
987 0,
988 tcp_md5_do_lookup(sk, l3index, addr, AF_INET),
989 inet_rsk(req)->no_srccheck ? IP_REPLY_ARG_NOSRCCHECK : 0,
990 ip_hdr(skb)->tos);
991}
992
993/*
994 * Send a SYN-ACK after having received a SYN.
995 * This still operates on a request_sock only, not on a big
996 * socket.
997 */
998static int tcp_v4_send_synack(const struct sock *sk, struct dst_entry *dst,
999 struct flowi *fl,
1000 struct request_sock *req,
1001 struct tcp_fastopen_cookie *foc,
1002 enum tcp_synack_type synack_type,
1003 struct sk_buff *syn_skb)
1004{
1005 const struct inet_request_sock *ireq = inet_rsk(req);
1006 struct flowi4 fl4;
1007 int err = -1;
1008 struct sk_buff *skb;
1009 u8 tos;
1010
1011 /* First, grab a route. */
1012 if (!dst && (dst = inet_csk_route_req(sk, &fl4, req)) == NULL)
1013 return -1;
1014
1015 skb = tcp_make_synack(sk, dst, req, foc, synack_type, syn_skb);
1016
1017 if (skb) {
1018 __tcp_v4_send_check(skb, ireq->ir_loc_addr, ireq->ir_rmt_addr);
1019
1020 tos = READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_reflect_tos) ?
1021 (tcp_rsk(req)->syn_tos & ~INET_ECN_MASK) |
1022 (inet_sk(sk)->tos & INET_ECN_MASK) :
1023 inet_sk(sk)->tos;
1024
1025 if (!INET_ECN_is_capable(tos) &&
1026 tcp_bpf_ca_needs_ecn((struct sock *)req))
1027 tos |= INET_ECN_ECT_0;
1028
1029 rcu_read_lock();
1030 err = ip_build_and_send_pkt(skb, sk, ireq->ir_loc_addr,
1031 ireq->ir_rmt_addr,
1032 rcu_dereference(ireq->ireq_opt),
1033 tos);
1034 rcu_read_unlock();
1035 err = net_xmit_eval(err);
1036 }
1037
1038 return err;
1039}
1040
1041/*
1042 * IPv4 request_sock destructor.
1043 */
1044static void tcp_v4_reqsk_destructor(struct request_sock *req)
1045{
1046 kfree(rcu_dereference_protected(inet_rsk(req)->ireq_opt, 1));
1047}
1048
1049#ifdef CONFIG_TCP_MD5SIG
1050/*
1051 * RFC2385 MD5 checksumming requires a mapping of
1052 * IP address->MD5 Key.
1053 * We need to maintain these in the sk structure.
1054 */
1055
1056DEFINE_STATIC_KEY_DEFERRED_FALSE(tcp_md5_needed, HZ);
1057EXPORT_SYMBOL(tcp_md5_needed);
1058
1059static bool better_md5_match(struct tcp_md5sig_key *old, struct tcp_md5sig_key *new)
1060{
1061 if (!old)
1062 return true;
1063
1064 /* l3index always overrides non-l3index */
1065 if (old->l3index && new->l3index == 0)
1066 return false;
1067 if (old->l3index == 0 && new->l3index)
1068 return true;
1069
1070 return old->prefixlen < new->prefixlen;
1071}
1072
1073/* Find the Key structure for an address. */
1074struct tcp_md5sig_key *__tcp_md5_do_lookup(const struct sock *sk, int l3index,
1075 const union tcp_md5_addr *addr,
1076 int family)
1077{
1078 const struct tcp_sock *tp = tcp_sk(sk);
1079 struct tcp_md5sig_key *key;
1080 const struct tcp_md5sig_info *md5sig;
1081 __be32 mask;
1082 struct tcp_md5sig_key *best_match = NULL;
1083 bool match;
1084
1085 /* caller either holds rcu_read_lock() or socket lock */
1086 md5sig = rcu_dereference_check(tp->md5sig_info,
1087 lockdep_sock_is_held(sk));
1088 if (!md5sig)
1089 return NULL;
1090
1091 hlist_for_each_entry_rcu(key, &md5sig->head, node,
1092 lockdep_sock_is_held(sk)) {
1093 if (key->family != family)
1094 continue;
1095 if (key->flags & TCP_MD5SIG_FLAG_IFINDEX && key->l3index != l3index)
1096 continue;
1097 if (family == AF_INET) {
1098 mask = inet_make_mask(key->prefixlen);
1099 match = (key->addr.a4.s_addr & mask) ==
1100 (addr->a4.s_addr & mask);
1101#if IS_ENABLED(CONFIG_IPV6)
1102 } else if (family == AF_INET6) {
1103 match = ipv6_prefix_equal(&key->addr.a6, &addr->a6,
1104 key->prefixlen);
1105#endif
1106 } else {
1107 match = false;
1108 }
1109
1110 if (match && better_md5_match(best_match, key))
1111 best_match = key;
1112 }
1113 return best_match;
1114}
1115EXPORT_SYMBOL(__tcp_md5_do_lookup);
1116
1117static struct tcp_md5sig_key *tcp_md5_do_lookup_exact(const struct sock *sk,
1118 const union tcp_md5_addr *addr,
1119 int family, u8 prefixlen,
1120 int l3index, u8 flags)
1121{
1122 const struct tcp_sock *tp = tcp_sk(sk);
1123 struct tcp_md5sig_key *key;
1124 unsigned int size = sizeof(struct in_addr);
1125 const struct tcp_md5sig_info *md5sig;
1126
1127 /* caller either holds rcu_read_lock() or socket lock */
1128 md5sig = rcu_dereference_check(tp->md5sig_info,
1129 lockdep_sock_is_held(sk));
1130 if (!md5sig)
1131 return NULL;
1132#if IS_ENABLED(CONFIG_IPV6)
1133 if (family == AF_INET6)
1134 size = sizeof(struct in6_addr);
1135#endif
1136 hlist_for_each_entry_rcu(key, &md5sig->head, node,
1137 lockdep_sock_is_held(sk)) {
1138 if (key->family != family)
1139 continue;
1140 if ((key->flags & TCP_MD5SIG_FLAG_IFINDEX) != (flags & TCP_MD5SIG_FLAG_IFINDEX))
1141 continue;
1142 if (key->l3index != l3index)
1143 continue;
1144 if (!memcmp(&key->addr, addr, size) &&
1145 key->prefixlen == prefixlen)
1146 return key;
1147 }
1148 return NULL;
1149}
1150
1151struct tcp_md5sig_key *tcp_v4_md5_lookup(const struct sock *sk,
1152 const struct sock *addr_sk)
1153{
1154 const union tcp_md5_addr *addr;
1155 int l3index;
1156
1157 l3index = l3mdev_master_ifindex_by_index(sock_net(sk),
1158 addr_sk->sk_bound_dev_if);
1159 addr = (const union tcp_md5_addr *)&addr_sk->sk_daddr;
1160 return tcp_md5_do_lookup(sk, l3index, addr, AF_INET);
1161}
1162EXPORT_SYMBOL(tcp_v4_md5_lookup);
1163
1164static int tcp_md5sig_info_add(struct sock *sk, gfp_t gfp)
1165{
1166 struct tcp_sock *tp = tcp_sk(sk);
1167 struct tcp_md5sig_info *md5sig;
1168
1169 md5sig = kmalloc(sizeof(*md5sig), gfp);
1170 if (!md5sig)
1171 return -ENOMEM;
1172
1173 sk_gso_disable(sk);
1174 INIT_HLIST_HEAD(&md5sig->head);
1175 rcu_assign_pointer(tp->md5sig_info, md5sig);
1176 return 0;
1177}
1178
1179/* This can be called on a newly created socket, from other files */
1180static int __tcp_md5_do_add(struct sock *sk, const union tcp_md5_addr *addr,
1181 int family, u8 prefixlen, int l3index, u8 flags,
1182 const u8 *newkey, u8 newkeylen, gfp_t gfp)
1183{
1184 /* Add Key to the list */
1185 struct tcp_md5sig_key *key;
1186 struct tcp_sock *tp = tcp_sk(sk);
1187 struct tcp_md5sig_info *md5sig;
1188
1189 key = tcp_md5_do_lookup_exact(sk, addr, family, prefixlen, l3index, flags);
1190 if (key) {
1191 /* Pre-existing entry - just update that one.
1192 * Note that the key might be used concurrently.
1193 * data_race() is telling kcsan that we do not care of
1194 * key mismatches, since changing MD5 key on live flows
1195 * can lead to packet drops.
1196 */
1197 data_race(memcpy(key->key, newkey, newkeylen));
1198
1199 /* Pairs with READ_ONCE() in tcp_md5_hash_key().
1200 * Also note that a reader could catch new key->keylen value
1201 * but old key->key[], this is the reason we use __GFP_ZERO
1202 * at sock_kmalloc() time below these lines.
1203 */
1204 WRITE_ONCE(key->keylen, newkeylen);
1205
1206 return 0;
1207 }
1208
1209 md5sig = rcu_dereference_protected(tp->md5sig_info,
1210 lockdep_sock_is_held(sk));
1211
1212 key = sock_kmalloc(sk, sizeof(*key), gfp | __GFP_ZERO);
1213 if (!key)
1214 return -ENOMEM;
1215 if (!tcp_alloc_md5sig_pool()) {
1216 sock_kfree_s(sk, key, sizeof(*key));
1217 return -ENOMEM;
1218 }
1219
1220 memcpy(key->key, newkey, newkeylen);
1221 key->keylen = newkeylen;
1222 key->family = family;
1223 key->prefixlen = prefixlen;
1224 key->l3index = l3index;
1225 key->flags = flags;
1226 memcpy(&key->addr, addr,
1227 (IS_ENABLED(CONFIG_IPV6) && family == AF_INET6) ? sizeof(struct in6_addr) :
1228 sizeof(struct in_addr));
1229 hlist_add_head_rcu(&key->node, &md5sig->head);
1230 return 0;
1231}
1232
1233int tcp_md5_do_add(struct sock *sk, const union tcp_md5_addr *addr,
1234 int family, u8 prefixlen, int l3index, u8 flags,
1235 const u8 *newkey, u8 newkeylen)
1236{
1237 struct tcp_sock *tp = tcp_sk(sk);
1238
1239 if (!rcu_dereference_protected(tp->md5sig_info, lockdep_sock_is_held(sk))) {
1240 if (tcp_md5sig_info_add(sk, GFP_KERNEL))
1241 return -ENOMEM;
1242
1243 if (!static_branch_inc(&tcp_md5_needed.key)) {
1244 struct tcp_md5sig_info *md5sig;
1245
1246 md5sig = rcu_dereference_protected(tp->md5sig_info, lockdep_sock_is_held(sk));
1247 rcu_assign_pointer(tp->md5sig_info, NULL);
1248 kfree_rcu(md5sig, rcu);
1249 return -EUSERS;
1250 }
1251 }
1252
1253 return __tcp_md5_do_add(sk, addr, family, prefixlen, l3index, flags,
1254 newkey, newkeylen, GFP_KERNEL);
1255}
1256EXPORT_SYMBOL(tcp_md5_do_add);
1257
1258int tcp_md5_key_copy(struct sock *sk, const union tcp_md5_addr *addr,
1259 int family, u8 prefixlen, int l3index,
1260 struct tcp_md5sig_key *key)
1261{
1262 struct tcp_sock *tp = tcp_sk(sk);
1263
1264 if (!rcu_dereference_protected(tp->md5sig_info, lockdep_sock_is_held(sk))) {
1265 if (tcp_md5sig_info_add(sk, sk_gfp_mask(sk, GFP_ATOMIC)))
1266 return -ENOMEM;
1267
1268 if (!static_key_fast_inc_not_disabled(&tcp_md5_needed.key.key)) {
1269 struct tcp_md5sig_info *md5sig;
1270
1271 md5sig = rcu_dereference_protected(tp->md5sig_info, lockdep_sock_is_held(sk));
1272 net_warn_ratelimited("Too many TCP-MD5 keys in the system\n");
1273 rcu_assign_pointer(tp->md5sig_info, NULL);
1274 kfree_rcu(md5sig, rcu);
1275 return -EUSERS;
1276 }
1277 }
1278
1279 return __tcp_md5_do_add(sk, addr, family, prefixlen, l3index,
1280 key->flags, key->key, key->keylen,
1281 sk_gfp_mask(sk, GFP_ATOMIC));
1282}
1283EXPORT_SYMBOL(tcp_md5_key_copy);
1284
1285int tcp_md5_do_del(struct sock *sk, const union tcp_md5_addr *addr, int family,
1286 u8 prefixlen, int l3index, u8 flags)
1287{
1288 struct tcp_md5sig_key *key;
1289
1290 key = tcp_md5_do_lookup_exact(sk, addr, family, prefixlen, l3index, flags);
1291 if (!key)
1292 return -ENOENT;
1293 hlist_del_rcu(&key->node);
1294 atomic_sub(sizeof(*key), &sk->sk_omem_alloc);
1295 kfree_rcu(key, rcu);
1296 return 0;
1297}
1298EXPORT_SYMBOL(tcp_md5_do_del);
1299
1300static void tcp_clear_md5_list(struct sock *sk)
1301{
1302 struct tcp_sock *tp = tcp_sk(sk);
1303 struct tcp_md5sig_key *key;
1304 struct hlist_node *n;
1305 struct tcp_md5sig_info *md5sig;
1306
1307 md5sig = rcu_dereference_protected(tp->md5sig_info, 1);
1308
1309 hlist_for_each_entry_safe(key, n, &md5sig->head, node) {
1310 hlist_del_rcu(&key->node);
1311 atomic_sub(sizeof(*key), &sk->sk_omem_alloc);
1312 kfree_rcu(key, rcu);
1313 }
1314}
1315
1316static int tcp_v4_parse_md5_keys(struct sock *sk, int optname,
1317 sockptr_t optval, int optlen)
1318{
1319 struct tcp_md5sig cmd;
1320 struct sockaddr_in *sin = (struct sockaddr_in *)&cmd.tcpm_addr;
1321 const union tcp_md5_addr *addr;
1322 u8 prefixlen = 32;
1323 int l3index = 0;
1324 u8 flags;
1325
1326 if (optlen < sizeof(cmd))
1327 return -EINVAL;
1328
1329 if (copy_from_sockptr(&cmd, optval, sizeof(cmd)))
1330 return -EFAULT;
1331
1332 if (sin->sin_family != AF_INET)
1333 return -EINVAL;
1334
1335 flags = cmd.tcpm_flags & TCP_MD5SIG_FLAG_IFINDEX;
1336
1337 if (optname == TCP_MD5SIG_EXT &&
1338 cmd.tcpm_flags & TCP_MD5SIG_FLAG_PREFIX) {
1339 prefixlen = cmd.tcpm_prefixlen;
1340 if (prefixlen > 32)
1341 return -EINVAL;
1342 }
1343
1344 if (optname == TCP_MD5SIG_EXT && cmd.tcpm_ifindex &&
1345 cmd.tcpm_flags & TCP_MD5SIG_FLAG_IFINDEX) {
1346 struct net_device *dev;
1347
1348 rcu_read_lock();
1349 dev = dev_get_by_index_rcu(sock_net(sk), cmd.tcpm_ifindex);
1350 if (dev && netif_is_l3_master(dev))
1351 l3index = dev->ifindex;
1352
1353 rcu_read_unlock();
1354
1355 /* ok to reference set/not set outside of rcu;
1356 * right now device MUST be an L3 master
1357 */
1358 if (!dev || !l3index)
1359 return -EINVAL;
1360 }
1361
1362 addr = (union tcp_md5_addr *)&sin->sin_addr.s_addr;
1363
1364 if (!cmd.tcpm_keylen)
1365 return tcp_md5_do_del(sk, addr, AF_INET, prefixlen, l3index, flags);
1366
1367 if (cmd.tcpm_keylen > TCP_MD5SIG_MAXKEYLEN)
1368 return -EINVAL;
1369
1370 return tcp_md5_do_add(sk, addr, AF_INET, prefixlen, l3index, flags,
1371 cmd.tcpm_key, cmd.tcpm_keylen);
1372}
1373
1374static int tcp_v4_md5_hash_headers(struct tcp_md5sig_pool *hp,
1375 __be32 daddr, __be32 saddr,
1376 const struct tcphdr *th, int nbytes)
1377{
1378 struct tcp4_pseudohdr *bp;
1379 struct scatterlist sg;
1380 struct tcphdr *_th;
1381
1382 bp = hp->scratch;
1383 bp->saddr = saddr;
1384 bp->daddr = daddr;
1385 bp->pad = 0;
1386 bp->protocol = IPPROTO_TCP;
1387 bp->len = cpu_to_be16(nbytes);
1388
1389 _th = (struct tcphdr *)(bp + 1);
1390 memcpy(_th, th, sizeof(*th));
1391 _th->check = 0;
1392
1393 sg_init_one(&sg, bp, sizeof(*bp) + sizeof(*th));
1394 ahash_request_set_crypt(hp->md5_req, &sg, NULL,
1395 sizeof(*bp) + sizeof(*th));
1396 return crypto_ahash_update(hp->md5_req);
1397}
1398
1399static int tcp_v4_md5_hash_hdr(char *md5_hash, const struct tcp_md5sig_key *key,
1400 __be32 daddr, __be32 saddr, const struct tcphdr *th)
1401{
1402 struct tcp_md5sig_pool *hp;
1403 struct ahash_request *req;
1404
1405 hp = tcp_get_md5sig_pool();
1406 if (!hp)
1407 goto clear_hash_noput;
1408 req = hp->md5_req;
1409
1410 if (crypto_ahash_init(req))
1411 goto clear_hash;
1412 if (tcp_v4_md5_hash_headers(hp, daddr, saddr, th, th->doff << 2))
1413 goto clear_hash;
1414 if (tcp_md5_hash_key(hp, key))
1415 goto clear_hash;
1416 ahash_request_set_crypt(req, NULL, md5_hash, 0);
1417 if (crypto_ahash_final(req))
1418 goto clear_hash;
1419
1420 tcp_put_md5sig_pool();
1421 return 0;
1422
1423clear_hash:
1424 tcp_put_md5sig_pool();
1425clear_hash_noput:
1426 memset(md5_hash, 0, 16);
1427 return 1;
1428}
1429
1430int tcp_v4_md5_hash_skb(char *md5_hash, const struct tcp_md5sig_key *key,
1431 const struct sock *sk,
1432 const struct sk_buff *skb)
1433{
1434 struct tcp_md5sig_pool *hp;
1435 struct ahash_request *req;
1436 const struct tcphdr *th = tcp_hdr(skb);
1437 __be32 saddr, daddr;
1438
1439 if (sk) { /* valid for establish/request sockets */
1440 saddr = sk->sk_rcv_saddr;
1441 daddr = sk->sk_daddr;
1442 } else {
1443 const struct iphdr *iph = ip_hdr(skb);
1444 saddr = iph->saddr;
1445 daddr = iph->daddr;
1446 }
1447
1448 hp = tcp_get_md5sig_pool();
1449 if (!hp)
1450 goto clear_hash_noput;
1451 req = hp->md5_req;
1452
1453 if (crypto_ahash_init(req))
1454 goto clear_hash;
1455
1456 if (tcp_v4_md5_hash_headers(hp, daddr, saddr, th, skb->len))
1457 goto clear_hash;
1458 if (tcp_md5_hash_skb_data(hp, skb, th->doff << 2))
1459 goto clear_hash;
1460 if (tcp_md5_hash_key(hp, key))
1461 goto clear_hash;
1462 ahash_request_set_crypt(req, NULL, md5_hash, 0);
1463 if (crypto_ahash_final(req))
1464 goto clear_hash;
1465
1466 tcp_put_md5sig_pool();
1467 return 0;
1468
1469clear_hash:
1470 tcp_put_md5sig_pool();
1471clear_hash_noput:
1472 memset(md5_hash, 0, 16);
1473 return 1;
1474}
1475EXPORT_SYMBOL(tcp_v4_md5_hash_skb);
1476
1477#endif
1478
1479static void tcp_v4_init_req(struct request_sock *req,
1480 const struct sock *sk_listener,
1481 struct sk_buff *skb)
1482{
1483 struct inet_request_sock *ireq = inet_rsk(req);
1484 struct net *net = sock_net(sk_listener);
1485
1486 sk_rcv_saddr_set(req_to_sk(req), ip_hdr(skb)->daddr);
1487 sk_daddr_set(req_to_sk(req), ip_hdr(skb)->saddr);
1488 RCU_INIT_POINTER(ireq->ireq_opt, tcp_v4_save_options(net, skb));
1489}
1490
1491static struct dst_entry *tcp_v4_route_req(const struct sock *sk,
1492 struct sk_buff *skb,
1493 struct flowi *fl,
1494 struct request_sock *req)
1495{
1496 tcp_v4_init_req(req, sk, skb);
1497
1498 if (security_inet_conn_request(sk, skb, req))
1499 return NULL;
1500
1501 return inet_csk_route_req(sk, &fl->u.ip4, req);
1502}
1503
1504struct request_sock_ops tcp_request_sock_ops __read_mostly = {
1505 .family = PF_INET,
1506 .obj_size = sizeof(struct tcp_request_sock),
1507 .rtx_syn_ack = tcp_rtx_synack,
1508 .send_ack = tcp_v4_reqsk_send_ack,
1509 .destructor = tcp_v4_reqsk_destructor,
1510 .send_reset = tcp_v4_send_reset,
1511 .syn_ack_timeout = tcp_syn_ack_timeout,
1512};
1513
1514const struct tcp_request_sock_ops tcp_request_sock_ipv4_ops = {
1515 .mss_clamp = TCP_MSS_DEFAULT,
1516#ifdef CONFIG_TCP_MD5SIG
1517 .req_md5_lookup = tcp_v4_md5_lookup,
1518 .calc_md5_hash = tcp_v4_md5_hash_skb,
1519#endif
1520#ifdef CONFIG_SYN_COOKIES
1521 .cookie_init_seq = cookie_v4_init_sequence,
1522#endif
1523 .route_req = tcp_v4_route_req,
1524 .init_seq = tcp_v4_init_seq,
1525 .init_ts_off = tcp_v4_init_ts_off,
1526 .send_synack = tcp_v4_send_synack,
1527};
1528
1529int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb)
1530{
1531 /* Never answer to SYNs send to broadcast or multicast */
1532 if (skb_rtable(skb)->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))
1533 goto drop;
1534
1535 return tcp_conn_request(&tcp_request_sock_ops,
1536 &tcp_request_sock_ipv4_ops, sk, skb);
1537
1538drop:
1539 tcp_listendrop(sk);
1540 return 0;
1541}
1542EXPORT_SYMBOL(tcp_v4_conn_request);
1543
1544
1545/*
1546 * The three way handshake has completed - we got a valid synack -
1547 * now create the new socket.
1548 */
1549struct sock *tcp_v4_syn_recv_sock(const struct sock *sk, struct sk_buff *skb,
1550 struct request_sock *req,
1551 struct dst_entry *dst,
1552 struct request_sock *req_unhash,
1553 bool *own_req)
1554{
1555 struct inet_request_sock *ireq;
1556 bool found_dup_sk = false;
1557 struct inet_sock *newinet;
1558 struct tcp_sock *newtp;
1559 struct sock *newsk;
1560#ifdef CONFIG_TCP_MD5SIG
1561 const union tcp_md5_addr *addr;
1562 struct tcp_md5sig_key *key;
1563 int l3index;
1564#endif
1565 struct ip_options_rcu *inet_opt;
1566
1567 if (sk_acceptq_is_full(sk))
1568 goto exit_overflow;
1569
1570 newsk = tcp_create_openreq_child(sk, req, skb);
1571 if (!newsk)
1572 goto exit_nonewsk;
1573
1574 newsk->sk_gso_type = SKB_GSO_TCPV4;
1575 inet_sk_rx_dst_set(newsk, skb);
1576
1577 newtp = tcp_sk(newsk);
1578 newinet = inet_sk(newsk);
1579 ireq = inet_rsk(req);
1580 sk_daddr_set(newsk, ireq->ir_rmt_addr);
1581 sk_rcv_saddr_set(newsk, ireq->ir_loc_addr);
1582 newsk->sk_bound_dev_if = ireq->ir_iif;
1583 newinet->inet_saddr = ireq->ir_loc_addr;
1584 inet_opt = rcu_dereference(ireq->ireq_opt);
1585 RCU_INIT_POINTER(newinet->inet_opt, inet_opt);
1586 newinet->mc_index = inet_iif(skb);
1587 newinet->mc_ttl = ip_hdr(skb)->ttl;
1588 newinet->rcv_tos = ip_hdr(skb)->tos;
1589 inet_csk(newsk)->icsk_ext_hdr_len = 0;
1590 if (inet_opt)
1591 inet_csk(newsk)->icsk_ext_hdr_len = inet_opt->opt.optlen;
1592 newinet->inet_id = get_random_u16();
1593
1594 /* Set ToS of the new socket based upon the value of incoming SYN.
1595 * ECT bits are set later in tcp_init_transfer().
1596 */
1597 if (READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_reflect_tos))
1598 newinet->tos = tcp_rsk(req)->syn_tos & ~INET_ECN_MASK;
1599
1600 if (!dst) {
1601 dst = inet_csk_route_child_sock(sk, newsk, req);
1602 if (!dst)
1603 goto put_and_exit;
1604 } else {
1605 /* syncookie case : see end of cookie_v4_check() */
1606 }
1607 sk_setup_caps(newsk, dst);
1608
1609 tcp_ca_openreq_child(newsk, dst);
1610
1611 tcp_sync_mss(newsk, dst_mtu(dst));
1612 newtp->advmss = tcp_mss_clamp(tcp_sk(sk), dst_metric_advmss(dst));
1613
1614 tcp_initialize_rcv_mss(newsk);
1615
1616#ifdef CONFIG_TCP_MD5SIG
1617 l3index = l3mdev_master_ifindex_by_index(sock_net(sk), ireq->ir_iif);
1618 /* Copy over the MD5 key from the original socket */
1619 addr = (union tcp_md5_addr *)&newinet->inet_daddr;
1620 key = tcp_md5_do_lookup(sk, l3index, addr, AF_INET);
1621 if (key) {
1622 if (tcp_md5_key_copy(newsk, addr, AF_INET, 32, l3index, key))
1623 goto put_and_exit;
1624 sk_gso_disable(newsk);
1625 }
1626#endif
1627
1628 if (__inet_inherit_port(sk, newsk) < 0)
1629 goto put_and_exit;
1630 *own_req = inet_ehash_nolisten(newsk, req_to_sk(req_unhash),
1631 &found_dup_sk);
1632 if (likely(*own_req)) {
1633 tcp_move_syn(newtp, req);
1634 ireq->ireq_opt = NULL;
1635 } else {
1636 newinet->inet_opt = NULL;
1637
1638 if (!req_unhash && found_dup_sk) {
1639 /* This code path should only be executed in the
1640 * syncookie case only
1641 */
1642 bh_unlock_sock(newsk);
1643 sock_put(newsk);
1644 newsk = NULL;
1645 }
1646 }
1647 return newsk;
1648
1649exit_overflow:
1650 NET_INC_STATS(sock_net(sk), LINUX_MIB_LISTENOVERFLOWS);
1651exit_nonewsk:
1652 dst_release(dst);
1653exit:
1654 tcp_listendrop(sk);
1655 return NULL;
1656put_and_exit:
1657 newinet->inet_opt = NULL;
1658 inet_csk_prepare_forced_close(newsk);
1659 tcp_done(newsk);
1660 goto exit;
1661}
1662EXPORT_SYMBOL(tcp_v4_syn_recv_sock);
1663
1664static struct sock *tcp_v4_cookie_check(struct sock *sk, struct sk_buff *skb)
1665{
1666#ifdef CONFIG_SYN_COOKIES
1667 const struct tcphdr *th = tcp_hdr(skb);
1668
1669 if (!th->syn)
1670 sk = cookie_v4_check(sk, skb);
1671#endif
1672 return sk;
1673}
1674
1675u16 tcp_v4_get_syncookie(struct sock *sk, struct iphdr *iph,
1676 struct tcphdr *th, u32 *cookie)
1677{
1678 u16 mss = 0;
1679#ifdef CONFIG_SYN_COOKIES
1680 mss = tcp_get_syncookie_mss(&tcp_request_sock_ops,
1681 &tcp_request_sock_ipv4_ops, sk, th);
1682 if (mss) {
1683 *cookie = __cookie_v4_init_sequence(iph, th, &mss);
1684 tcp_synq_overflow(sk);
1685 }
1686#endif
1687 return mss;
1688}
1689
1690INDIRECT_CALLABLE_DECLARE(struct dst_entry *ipv4_dst_check(struct dst_entry *,
1691 u32));
1692/* The socket must have it's spinlock held when we get
1693 * here, unless it is a TCP_LISTEN socket.
1694 *
1695 * We have a potential double-lock case here, so even when
1696 * doing backlog processing we use the BH locking scheme.
1697 * This is because we cannot sleep with the original spinlock
1698 * held.
1699 */
1700int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb)
1701{
1702 enum skb_drop_reason reason;
1703 struct sock *rsk;
1704
1705 if (sk->sk_state == TCP_ESTABLISHED) { /* Fast path */
1706 struct dst_entry *dst;
1707
1708 dst = rcu_dereference_protected(sk->sk_rx_dst,
1709 lockdep_sock_is_held(sk));
1710
1711 sock_rps_save_rxhash(sk, skb);
1712 sk_mark_napi_id(sk, skb);
1713 if (dst) {
1714 if (sk->sk_rx_dst_ifindex != skb->skb_iif ||
1715 !INDIRECT_CALL_1(dst->ops->check, ipv4_dst_check,
1716 dst, 0)) {
1717 RCU_INIT_POINTER(sk->sk_rx_dst, NULL);
1718 dst_release(dst);
1719 }
1720 }
1721 tcp_rcv_established(sk, skb);
1722 return 0;
1723 }
1724
1725 reason = SKB_DROP_REASON_NOT_SPECIFIED;
1726 if (tcp_checksum_complete(skb))
1727 goto csum_err;
1728
1729 if (sk->sk_state == TCP_LISTEN) {
1730 struct sock *nsk = tcp_v4_cookie_check(sk, skb);
1731
1732 if (!nsk)
1733 goto discard;
1734 if (nsk != sk) {
1735 if (tcp_child_process(sk, nsk, skb)) {
1736 rsk = nsk;
1737 goto reset;
1738 }
1739 return 0;
1740 }
1741 } else
1742 sock_rps_save_rxhash(sk, skb);
1743
1744 if (tcp_rcv_state_process(sk, skb)) {
1745 rsk = sk;
1746 goto reset;
1747 }
1748 return 0;
1749
1750reset:
1751 tcp_v4_send_reset(rsk, skb);
1752discard:
1753 kfree_skb_reason(skb, reason);
1754 /* Be careful here. If this function gets more complicated and
1755 * gcc suffers from register pressure on the x86, sk (in %ebx)
1756 * might be destroyed here. This current version compiles correctly,
1757 * but you have been warned.
1758 */
1759 return 0;
1760
1761csum_err:
1762 reason = SKB_DROP_REASON_TCP_CSUM;
1763 trace_tcp_bad_csum(skb);
1764 TCP_INC_STATS(sock_net(sk), TCP_MIB_CSUMERRORS);
1765 TCP_INC_STATS(sock_net(sk), TCP_MIB_INERRS);
1766 goto discard;
1767}
1768EXPORT_SYMBOL(tcp_v4_do_rcv);
1769
1770int tcp_v4_early_demux(struct sk_buff *skb)
1771{
1772 struct net *net = dev_net(skb->dev);
1773 const struct iphdr *iph;
1774 const struct tcphdr *th;
1775 struct sock *sk;
1776
1777 if (skb->pkt_type != PACKET_HOST)
1778 return 0;
1779
1780 if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct tcphdr)))
1781 return 0;
1782
1783 iph = ip_hdr(skb);
1784 th = tcp_hdr(skb);
1785
1786 if (th->doff < sizeof(struct tcphdr) / 4)
1787 return 0;
1788
1789 sk = __inet_lookup_established(net, net->ipv4.tcp_death_row.hashinfo,
1790 iph->saddr, th->source,
1791 iph->daddr, ntohs(th->dest),
1792 skb->skb_iif, inet_sdif(skb));
1793 if (sk) {
1794 skb->sk = sk;
1795 skb->destructor = sock_edemux;
1796 if (sk_fullsock(sk)) {
1797 struct dst_entry *dst = rcu_dereference(sk->sk_rx_dst);
1798
1799 if (dst)
1800 dst = dst_check(dst, 0);
1801 if (dst &&
1802 sk->sk_rx_dst_ifindex == skb->skb_iif)
1803 skb_dst_set_noref(skb, dst);
1804 }
1805 }
1806 return 0;
1807}
1808
1809bool tcp_add_backlog(struct sock *sk, struct sk_buff *skb,
1810 enum skb_drop_reason *reason)
1811{
1812 u32 limit, tail_gso_size, tail_gso_segs;
1813 struct skb_shared_info *shinfo;
1814 const struct tcphdr *th;
1815 struct tcphdr *thtail;
1816 struct sk_buff *tail;
1817 unsigned int hdrlen;
1818 bool fragstolen;
1819 u32 gso_segs;
1820 u32 gso_size;
1821 int delta;
1822
1823 /* In case all data was pulled from skb frags (in __pskb_pull_tail()),
1824 * we can fix skb->truesize to its real value to avoid future drops.
1825 * This is valid because skb is not yet charged to the socket.
1826 * It has been noticed pure SACK packets were sometimes dropped
1827 * (if cooked by drivers without copybreak feature).
1828 */
1829 skb_condense(skb);
1830
1831 skb_dst_drop(skb);
1832
1833 if (unlikely(tcp_checksum_complete(skb))) {
1834 bh_unlock_sock(sk);
1835 trace_tcp_bad_csum(skb);
1836 *reason = SKB_DROP_REASON_TCP_CSUM;
1837 __TCP_INC_STATS(sock_net(sk), TCP_MIB_CSUMERRORS);
1838 __TCP_INC_STATS(sock_net(sk), TCP_MIB_INERRS);
1839 return true;
1840 }
1841
1842 /* Attempt coalescing to last skb in backlog, even if we are
1843 * above the limits.
1844 * This is okay because skb capacity is limited to MAX_SKB_FRAGS.
1845 */
1846 th = (const struct tcphdr *)skb->data;
1847 hdrlen = th->doff * 4;
1848
1849 tail = sk->sk_backlog.tail;
1850 if (!tail)
1851 goto no_coalesce;
1852 thtail = (struct tcphdr *)tail->data;
1853
1854 if (TCP_SKB_CB(tail)->end_seq != TCP_SKB_CB(skb)->seq ||
1855 TCP_SKB_CB(tail)->ip_dsfield != TCP_SKB_CB(skb)->ip_dsfield ||
1856 ((TCP_SKB_CB(tail)->tcp_flags |
1857 TCP_SKB_CB(skb)->tcp_flags) & (TCPHDR_SYN | TCPHDR_RST | TCPHDR_URG)) ||
1858 !((TCP_SKB_CB(tail)->tcp_flags &
1859 TCP_SKB_CB(skb)->tcp_flags) & TCPHDR_ACK) ||
1860 ((TCP_SKB_CB(tail)->tcp_flags ^
1861 TCP_SKB_CB(skb)->tcp_flags) & (TCPHDR_ECE | TCPHDR_CWR)) ||
1862#ifdef CONFIG_TLS_DEVICE
1863 tail->decrypted != skb->decrypted ||
1864#endif
1865 thtail->doff != th->doff ||
1866 memcmp(thtail + 1, th + 1, hdrlen - sizeof(*th)))
1867 goto no_coalesce;
1868
1869 __skb_pull(skb, hdrlen);
1870
1871 shinfo = skb_shinfo(skb);
1872 gso_size = shinfo->gso_size ?: skb->len;
1873 gso_segs = shinfo->gso_segs ?: 1;
1874
1875 shinfo = skb_shinfo(tail);
1876 tail_gso_size = shinfo->gso_size ?: (tail->len - hdrlen);
1877 tail_gso_segs = shinfo->gso_segs ?: 1;
1878
1879 if (skb_try_coalesce(tail, skb, &fragstolen, &delta)) {
1880 TCP_SKB_CB(tail)->end_seq = TCP_SKB_CB(skb)->end_seq;
1881
1882 if (likely(!before(TCP_SKB_CB(skb)->ack_seq, TCP_SKB_CB(tail)->ack_seq))) {
1883 TCP_SKB_CB(tail)->ack_seq = TCP_SKB_CB(skb)->ack_seq;
1884 thtail->window = th->window;
1885 }
1886
1887 /* We have to update both TCP_SKB_CB(tail)->tcp_flags and
1888 * thtail->fin, so that the fast path in tcp_rcv_established()
1889 * is not entered if we append a packet with a FIN.
1890 * SYN, RST, URG are not present.
1891 * ACK is set on both packets.
1892 * PSH : we do not really care in TCP stack,
1893 * at least for 'GRO' packets.
1894 */
1895 thtail->fin |= th->fin;
1896 TCP_SKB_CB(tail)->tcp_flags |= TCP_SKB_CB(skb)->tcp_flags;
1897
1898 if (TCP_SKB_CB(skb)->has_rxtstamp) {
1899 TCP_SKB_CB(tail)->has_rxtstamp = true;
1900 tail->tstamp = skb->tstamp;
1901 skb_hwtstamps(tail)->hwtstamp = skb_hwtstamps(skb)->hwtstamp;
1902 }
1903
1904 /* Not as strict as GRO. We only need to carry mss max value */
1905 shinfo->gso_size = max(gso_size, tail_gso_size);
1906 shinfo->gso_segs = min_t(u32, gso_segs + tail_gso_segs, 0xFFFF);
1907
1908 sk->sk_backlog.len += delta;
1909 __NET_INC_STATS(sock_net(sk),
1910 LINUX_MIB_TCPBACKLOGCOALESCE);
1911 kfree_skb_partial(skb, fragstolen);
1912 return false;
1913 }
1914 __skb_push(skb, hdrlen);
1915
1916no_coalesce:
1917 limit = (u32)READ_ONCE(sk->sk_rcvbuf) + (u32)(READ_ONCE(sk->sk_sndbuf) >> 1);
1918
1919 /* Only socket owner can try to collapse/prune rx queues
1920 * to reduce memory overhead, so add a little headroom here.
1921 * Few sockets backlog are possibly concurrently non empty.
1922 */
1923 limit += 64 * 1024;
1924
1925 if (unlikely(sk_add_backlog(sk, skb, limit))) {
1926 bh_unlock_sock(sk);
1927 *reason = SKB_DROP_REASON_SOCKET_BACKLOG;
1928 __NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPBACKLOGDROP);
1929 return true;
1930 }
1931 return false;
1932}
1933EXPORT_SYMBOL(tcp_add_backlog);
1934
1935int tcp_filter(struct sock *sk, struct sk_buff *skb)
1936{
1937 struct tcphdr *th = (struct tcphdr *)skb->data;
1938
1939 return sk_filter_trim_cap(sk, skb, th->doff * 4);
1940}
1941EXPORT_SYMBOL(tcp_filter);
1942
1943static void tcp_v4_restore_cb(struct sk_buff *skb)
1944{
1945 memmove(IPCB(skb), &TCP_SKB_CB(skb)->header.h4,
1946 sizeof(struct inet_skb_parm));
1947}
1948
1949static void tcp_v4_fill_cb(struct sk_buff *skb, const struct iphdr *iph,
1950 const struct tcphdr *th)
1951{
1952 /* This is tricky : We move IPCB at its correct location into TCP_SKB_CB()
1953 * barrier() makes sure compiler wont play fool^Waliasing games.
1954 */
1955 memmove(&TCP_SKB_CB(skb)->header.h4, IPCB(skb),
1956 sizeof(struct inet_skb_parm));
1957 barrier();
1958
1959 TCP_SKB_CB(skb)->seq = ntohl(th->seq);
1960 TCP_SKB_CB(skb)->end_seq = (TCP_SKB_CB(skb)->seq + th->syn + th->fin +
1961 skb->len - th->doff * 4);
1962 TCP_SKB_CB(skb)->ack_seq = ntohl(th->ack_seq);
1963 TCP_SKB_CB(skb)->tcp_flags = tcp_flag_byte(th);
1964 TCP_SKB_CB(skb)->tcp_tw_isn = 0;
1965 TCP_SKB_CB(skb)->ip_dsfield = ipv4_get_dsfield(iph);
1966 TCP_SKB_CB(skb)->sacked = 0;
1967 TCP_SKB_CB(skb)->has_rxtstamp =
1968 skb->tstamp || skb_hwtstamps(skb)->hwtstamp;
1969}
1970
1971/*
1972 * From tcp_input.c
1973 */
1974
1975int tcp_v4_rcv(struct sk_buff *skb)
1976{
1977 struct net *net = dev_net(skb->dev);
1978 enum skb_drop_reason drop_reason;
1979 int sdif = inet_sdif(skb);
1980 int dif = inet_iif(skb);
1981 const struct iphdr *iph;
1982 const struct tcphdr *th;
1983 bool refcounted;
1984 struct sock *sk;
1985 int ret;
1986
1987 drop_reason = SKB_DROP_REASON_NOT_SPECIFIED;
1988 if (skb->pkt_type != PACKET_HOST)
1989 goto discard_it;
1990
1991 /* Count it even if it's bad */
1992 __TCP_INC_STATS(net, TCP_MIB_INSEGS);
1993
1994 if (!pskb_may_pull(skb, sizeof(struct tcphdr)))
1995 goto discard_it;
1996
1997 th = (const struct tcphdr *)skb->data;
1998
1999 if (unlikely(th->doff < sizeof(struct tcphdr) / 4)) {
2000 drop_reason = SKB_DROP_REASON_PKT_TOO_SMALL;
2001 goto bad_packet;
2002 }
2003 if (!pskb_may_pull(skb, th->doff * 4))
2004 goto discard_it;
2005
2006 /* An explanation is required here, I think.
2007 * Packet length and doff are validated by header prediction,
2008 * provided case of th->doff==0 is eliminated.
2009 * So, we defer the checks. */
2010
2011 if (skb_checksum_init(skb, IPPROTO_TCP, inet_compute_pseudo))
2012 goto csum_error;
2013
2014 th = (const struct tcphdr *)skb->data;
2015 iph = ip_hdr(skb);
2016lookup:
2017 sk = __inet_lookup_skb(net->ipv4.tcp_death_row.hashinfo,
2018 skb, __tcp_hdrlen(th), th->source,
2019 th->dest, sdif, &refcounted);
2020 if (!sk)
2021 goto no_tcp_socket;
2022
2023process:
2024 if (sk->sk_state == TCP_TIME_WAIT)
2025 goto do_time_wait;
2026
2027 if (sk->sk_state == TCP_NEW_SYN_RECV) {
2028 struct request_sock *req = inet_reqsk(sk);
2029 bool req_stolen = false;
2030 struct sock *nsk;
2031
2032 sk = req->rsk_listener;
2033 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
2034 drop_reason = SKB_DROP_REASON_XFRM_POLICY;
2035 else
2036 drop_reason = tcp_inbound_md5_hash(sk, skb,
2037 &iph->saddr, &iph->daddr,
2038 AF_INET, dif, sdif);
2039 if (unlikely(drop_reason)) {
2040 sk_drops_add(sk, skb);
2041 reqsk_put(req);
2042 goto discard_it;
2043 }
2044 if (tcp_checksum_complete(skb)) {
2045 reqsk_put(req);
2046 goto csum_error;
2047 }
2048 if (unlikely(sk->sk_state != TCP_LISTEN)) {
2049 nsk = reuseport_migrate_sock(sk, req_to_sk(req), skb);
2050 if (!nsk) {
2051 inet_csk_reqsk_queue_drop_and_put(sk, req);
2052 goto lookup;
2053 }
2054 sk = nsk;
2055 /* reuseport_migrate_sock() has already held one sk_refcnt
2056 * before returning.
2057 */
2058 } else {
2059 /* We own a reference on the listener, increase it again
2060 * as we might lose it too soon.
2061 */
2062 sock_hold(sk);
2063 }
2064 refcounted = true;
2065 nsk = NULL;
2066 if (!tcp_filter(sk, skb)) {
2067 th = (const struct tcphdr *)skb->data;
2068 iph = ip_hdr(skb);
2069 tcp_v4_fill_cb(skb, iph, th);
2070 nsk = tcp_check_req(sk, skb, req, false, &req_stolen);
2071 } else {
2072 drop_reason = SKB_DROP_REASON_SOCKET_FILTER;
2073 }
2074 if (!nsk) {
2075 reqsk_put(req);
2076 if (req_stolen) {
2077 /* Another cpu got exclusive access to req
2078 * and created a full blown socket.
2079 * Try to feed this packet to this socket
2080 * instead of discarding it.
2081 */
2082 tcp_v4_restore_cb(skb);
2083 sock_put(sk);
2084 goto lookup;
2085 }
2086 goto discard_and_relse;
2087 }
2088 nf_reset_ct(skb);
2089 if (nsk == sk) {
2090 reqsk_put(req);
2091 tcp_v4_restore_cb(skb);
2092 } else if (tcp_child_process(sk, nsk, skb)) {
2093 tcp_v4_send_reset(nsk, skb);
2094 goto discard_and_relse;
2095 } else {
2096 sock_put(sk);
2097 return 0;
2098 }
2099 }
2100
2101 if (static_branch_unlikely(&ip4_min_ttl)) {
2102 /* min_ttl can be changed concurrently from do_ip_setsockopt() */
2103 if (unlikely(iph->ttl < READ_ONCE(inet_sk(sk)->min_ttl))) {
2104 __NET_INC_STATS(net, LINUX_MIB_TCPMINTTLDROP);
2105 goto discard_and_relse;
2106 }
2107 }
2108
2109 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb)) {
2110 drop_reason = SKB_DROP_REASON_XFRM_POLICY;
2111 goto discard_and_relse;
2112 }
2113
2114 drop_reason = tcp_inbound_md5_hash(sk, skb, &iph->saddr,
2115 &iph->daddr, AF_INET, dif, sdif);
2116 if (drop_reason)
2117 goto discard_and_relse;
2118
2119 nf_reset_ct(skb);
2120
2121 if (tcp_filter(sk, skb)) {
2122 drop_reason = SKB_DROP_REASON_SOCKET_FILTER;
2123 goto discard_and_relse;
2124 }
2125 th = (const struct tcphdr *)skb->data;
2126 iph = ip_hdr(skb);
2127 tcp_v4_fill_cb(skb, iph, th);
2128
2129 skb->dev = NULL;
2130
2131 if (sk->sk_state == TCP_LISTEN) {
2132 ret = tcp_v4_do_rcv(sk, skb);
2133 goto put_and_return;
2134 }
2135
2136 sk_incoming_cpu_update(sk);
2137
2138 bh_lock_sock_nested(sk);
2139 tcp_segs_in(tcp_sk(sk), skb);
2140 ret = 0;
2141 if (!sock_owned_by_user(sk)) {
2142 ret = tcp_v4_do_rcv(sk, skb);
2143 } else {
2144 if (tcp_add_backlog(sk, skb, &drop_reason))
2145 goto discard_and_relse;
2146 }
2147 bh_unlock_sock(sk);
2148
2149put_and_return:
2150 if (refcounted)
2151 sock_put(sk);
2152
2153 return ret;
2154
2155no_tcp_socket:
2156 drop_reason = SKB_DROP_REASON_NO_SOCKET;
2157 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
2158 goto discard_it;
2159
2160 tcp_v4_fill_cb(skb, iph, th);
2161
2162 if (tcp_checksum_complete(skb)) {
2163csum_error:
2164 drop_reason = SKB_DROP_REASON_TCP_CSUM;
2165 trace_tcp_bad_csum(skb);
2166 __TCP_INC_STATS(net, TCP_MIB_CSUMERRORS);
2167bad_packet:
2168 __TCP_INC_STATS(net, TCP_MIB_INERRS);
2169 } else {
2170 tcp_v4_send_reset(NULL, skb);
2171 }
2172
2173discard_it:
2174 SKB_DR_OR(drop_reason, NOT_SPECIFIED);
2175 /* Discard frame. */
2176 kfree_skb_reason(skb, drop_reason);
2177 return 0;
2178
2179discard_and_relse:
2180 sk_drops_add(sk, skb);
2181 if (refcounted)
2182 sock_put(sk);
2183 goto discard_it;
2184
2185do_time_wait:
2186 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) {
2187 drop_reason = SKB_DROP_REASON_XFRM_POLICY;
2188 inet_twsk_put(inet_twsk(sk));
2189 goto discard_it;
2190 }
2191
2192 tcp_v4_fill_cb(skb, iph, th);
2193
2194 if (tcp_checksum_complete(skb)) {
2195 inet_twsk_put(inet_twsk(sk));
2196 goto csum_error;
2197 }
2198 switch (tcp_timewait_state_process(inet_twsk(sk), skb, th)) {
2199 case TCP_TW_SYN: {
2200 struct sock *sk2 = inet_lookup_listener(net,
2201 net->ipv4.tcp_death_row.hashinfo,
2202 skb, __tcp_hdrlen(th),
2203 iph->saddr, th->source,
2204 iph->daddr, th->dest,
2205 inet_iif(skb),
2206 sdif);
2207 if (sk2) {
2208 inet_twsk_deschedule_put(inet_twsk(sk));
2209 sk = sk2;
2210 tcp_v4_restore_cb(skb);
2211 refcounted = false;
2212 goto process;
2213 }
2214 }
2215 /* to ACK */
2216 fallthrough;
2217 case TCP_TW_ACK:
2218 tcp_v4_timewait_ack(sk, skb);
2219 break;
2220 case TCP_TW_RST:
2221 tcp_v4_send_reset(sk, skb);
2222 inet_twsk_deschedule_put(inet_twsk(sk));
2223 goto discard_it;
2224 case TCP_TW_SUCCESS:;
2225 }
2226 goto discard_it;
2227}
2228
2229static struct timewait_sock_ops tcp_timewait_sock_ops = {
2230 .twsk_obj_size = sizeof(struct tcp_timewait_sock),
2231 .twsk_unique = tcp_twsk_unique,
2232 .twsk_destructor= tcp_twsk_destructor,
2233};
2234
2235void inet_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb)
2236{
2237 struct dst_entry *dst = skb_dst(skb);
2238
2239 if (dst && dst_hold_safe(dst)) {
2240 rcu_assign_pointer(sk->sk_rx_dst, dst);
2241 sk->sk_rx_dst_ifindex = skb->skb_iif;
2242 }
2243}
2244EXPORT_SYMBOL(inet_sk_rx_dst_set);
2245
2246const struct inet_connection_sock_af_ops ipv4_specific = {
2247 .queue_xmit = ip_queue_xmit,
2248 .send_check = tcp_v4_send_check,
2249 .rebuild_header = inet_sk_rebuild_header,
2250 .sk_rx_dst_set = inet_sk_rx_dst_set,
2251 .conn_request = tcp_v4_conn_request,
2252 .syn_recv_sock = tcp_v4_syn_recv_sock,
2253 .net_header_len = sizeof(struct iphdr),
2254 .setsockopt = ip_setsockopt,
2255 .getsockopt = ip_getsockopt,
2256 .addr2sockaddr = inet_csk_addr2sockaddr,
2257 .sockaddr_len = sizeof(struct sockaddr_in),
2258 .mtu_reduced = tcp_v4_mtu_reduced,
2259};
2260EXPORT_SYMBOL(ipv4_specific);
2261
2262#ifdef CONFIG_TCP_MD5SIG
2263static const struct tcp_sock_af_ops tcp_sock_ipv4_specific = {
2264 .md5_lookup = tcp_v4_md5_lookup,
2265 .calc_md5_hash = tcp_v4_md5_hash_skb,
2266 .md5_parse = tcp_v4_parse_md5_keys,
2267};
2268#endif
2269
2270/* NOTE: A lot of things set to zero explicitly by call to
2271 * sk_alloc() so need not be done here.
2272 */
2273static int tcp_v4_init_sock(struct sock *sk)
2274{
2275 struct inet_connection_sock *icsk = inet_csk(sk);
2276
2277 tcp_init_sock(sk);
2278
2279 icsk->icsk_af_ops = &ipv4_specific;
2280
2281#ifdef CONFIG_TCP_MD5SIG
2282 tcp_sk(sk)->af_specific = &tcp_sock_ipv4_specific;
2283#endif
2284
2285 return 0;
2286}
2287
2288void tcp_v4_destroy_sock(struct sock *sk)
2289{
2290 struct tcp_sock *tp = tcp_sk(sk);
2291
2292 trace_tcp_destroy_sock(sk);
2293
2294 tcp_clear_xmit_timers(sk);
2295
2296 tcp_cleanup_congestion_control(sk);
2297
2298 tcp_cleanup_ulp(sk);
2299
2300 /* Cleanup up the write buffer. */
2301 tcp_write_queue_purge(sk);
2302
2303 /* Check if we want to disable active TFO */
2304 tcp_fastopen_active_disable_ofo_check(sk);
2305
2306 /* Cleans up our, hopefully empty, out_of_order_queue. */
2307 skb_rbtree_purge(&tp->out_of_order_queue);
2308
2309#ifdef CONFIG_TCP_MD5SIG
2310 /* Clean up the MD5 key list, if any */
2311 if (tp->md5sig_info) {
2312 tcp_clear_md5_list(sk);
2313 kfree_rcu(rcu_dereference_protected(tp->md5sig_info, 1), rcu);
2314 tp->md5sig_info = NULL;
2315 static_branch_slow_dec_deferred(&tcp_md5_needed);
2316 }
2317#endif
2318
2319 /* Clean up a referenced TCP bind bucket. */
2320 if (inet_csk(sk)->icsk_bind_hash)
2321 inet_put_port(sk);
2322
2323 BUG_ON(rcu_access_pointer(tp->fastopen_rsk));
2324
2325 /* If socket is aborted during connect operation */
2326 tcp_free_fastopen_req(tp);
2327 tcp_fastopen_destroy_cipher(sk);
2328 tcp_saved_syn_free(tp);
2329
2330 sk_sockets_allocated_dec(sk);
2331}
2332EXPORT_SYMBOL(tcp_v4_destroy_sock);
2333
2334#ifdef CONFIG_PROC_FS
2335/* Proc filesystem TCP sock list dumping. */
2336
2337static unsigned short seq_file_family(const struct seq_file *seq);
2338
2339static bool seq_sk_match(struct seq_file *seq, const struct sock *sk)
2340{
2341 unsigned short family = seq_file_family(seq);
2342
2343 /* AF_UNSPEC is used as a match all */
2344 return ((family == AF_UNSPEC || family == sk->sk_family) &&
2345 net_eq(sock_net(sk), seq_file_net(seq)));
2346}
2347
2348/* Find a non empty bucket (starting from st->bucket)
2349 * and return the first sk from it.
2350 */
2351static void *listening_get_first(struct seq_file *seq)
2352{
2353 struct inet_hashinfo *hinfo = seq_file_net(seq)->ipv4.tcp_death_row.hashinfo;
2354 struct tcp_iter_state *st = seq->private;
2355
2356 st->offset = 0;
2357 for (; st->bucket <= hinfo->lhash2_mask; st->bucket++) {
2358 struct inet_listen_hashbucket *ilb2;
2359 struct hlist_nulls_node *node;
2360 struct sock *sk;
2361
2362 ilb2 = &hinfo->lhash2[st->bucket];
2363 if (hlist_nulls_empty(&ilb2->nulls_head))
2364 continue;
2365
2366 spin_lock(&ilb2->lock);
2367 sk_nulls_for_each(sk, node, &ilb2->nulls_head) {
2368 if (seq_sk_match(seq, sk))
2369 return sk;
2370 }
2371 spin_unlock(&ilb2->lock);
2372 }
2373
2374 return NULL;
2375}
2376
2377/* Find the next sk of "cur" within the same bucket (i.e. st->bucket).
2378 * If "cur" is the last one in the st->bucket,
2379 * call listening_get_first() to return the first sk of the next
2380 * non empty bucket.
2381 */
2382static void *listening_get_next(struct seq_file *seq, void *cur)
2383{
2384 struct tcp_iter_state *st = seq->private;
2385 struct inet_listen_hashbucket *ilb2;
2386 struct hlist_nulls_node *node;
2387 struct inet_hashinfo *hinfo;
2388 struct sock *sk = cur;
2389
2390 ++st->num;
2391 ++st->offset;
2392
2393 sk = sk_nulls_next(sk);
2394 sk_nulls_for_each_from(sk, node) {
2395 if (seq_sk_match(seq, sk))
2396 return sk;
2397 }
2398
2399 hinfo = seq_file_net(seq)->ipv4.tcp_death_row.hashinfo;
2400 ilb2 = &hinfo->lhash2[st->bucket];
2401 spin_unlock(&ilb2->lock);
2402 ++st->bucket;
2403 return listening_get_first(seq);
2404}
2405
2406static void *listening_get_idx(struct seq_file *seq, loff_t *pos)
2407{
2408 struct tcp_iter_state *st = seq->private;
2409 void *rc;
2410
2411 st->bucket = 0;
2412 st->offset = 0;
2413 rc = listening_get_first(seq);
2414
2415 while (rc && *pos) {
2416 rc = listening_get_next(seq, rc);
2417 --*pos;
2418 }
2419 return rc;
2420}
2421
2422static inline bool empty_bucket(struct inet_hashinfo *hinfo,
2423 const struct tcp_iter_state *st)
2424{
2425 return hlist_nulls_empty(&hinfo->ehash[st->bucket].chain);
2426}
2427
2428/*
2429 * Get first established socket starting from bucket given in st->bucket.
2430 * If st->bucket is zero, the very first socket in the hash is returned.
2431 */
2432static void *established_get_first(struct seq_file *seq)
2433{
2434 struct inet_hashinfo *hinfo = seq_file_net(seq)->ipv4.tcp_death_row.hashinfo;
2435 struct tcp_iter_state *st = seq->private;
2436
2437 st->offset = 0;
2438 for (; st->bucket <= hinfo->ehash_mask; ++st->bucket) {
2439 struct sock *sk;
2440 struct hlist_nulls_node *node;
2441 spinlock_t *lock = inet_ehash_lockp(hinfo, st->bucket);
2442
2443 /* Lockless fast path for the common case of empty buckets */
2444 if (empty_bucket(hinfo, st))
2445 continue;
2446
2447 spin_lock_bh(lock);
2448 sk_nulls_for_each(sk, node, &hinfo->ehash[st->bucket].chain) {
2449 if (seq_sk_match(seq, sk))
2450 return sk;
2451 }
2452 spin_unlock_bh(lock);
2453 }
2454
2455 return NULL;
2456}
2457
2458static void *established_get_next(struct seq_file *seq, void *cur)
2459{
2460 struct inet_hashinfo *hinfo = seq_file_net(seq)->ipv4.tcp_death_row.hashinfo;
2461 struct tcp_iter_state *st = seq->private;
2462 struct hlist_nulls_node *node;
2463 struct sock *sk = cur;
2464
2465 ++st->num;
2466 ++st->offset;
2467
2468 sk = sk_nulls_next(sk);
2469
2470 sk_nulls_for_each_from(sk, node) {
2471 if (seq_sk_match(seq, sk))
2472 return sk;
2473 }
2474
2475 spin_unlock_bh(inet_ehash_lockp(hinfo, st->bucket));
2476 ++st->bucket;
2477 return established_get_first(seq);
2478}
2479
2480static void *established_get_idx(struct seq_file *seq, loff_t pos)
2481{
2482 struct tcp_iter_state *st = seq->private;
2483 void *rc;
2484
2485 st->bucket = 0;
2486 rc = established_get_first(seq);
2487
2488 while (rc && pos) {
2489 rc = established_get_next(seq, rc);
2490 --pos;
2491 }
2492 return rc;
2493}
2494
2495static void *tcp_get_idx(struct seq_file *seq, loff_t pos)
2496{
2497 void *rc;
2498 struct tcp_iter_state *st = seq->private;
2499
2500 st->state = TCP_SEQ_STATE_LISTENING;
2501 rc = listening_get_idx(seq, &pos);
2502
2503 if (!rc) {
2504 st->state = TCP_SEQ_STATE_ESTABLISHED;
2505 rc = established_get_idx(seq, pos);
2506 }
2507
2508 return rc;
2509}
2510
2511static void *tcp_seek_last_pos(struct seq_file *seq)
2512{
2513 struct inet_hashinfo *hinfo = seq_file_net(seq)->ipv4.tcp_death_row.hashinfo;
2514 struct tcp_iter_state *st = seq->private;
2515 int bucket = st->bucket;
2516 int offset = st->offset;
2517 int orig_num = st->num;
2518 void *rc = NULL;
2519
2520 switch (st->state) {
2521 case TCP_SEQ_STATE_LISTENING:
2522 if (st->bucket > hinfo->lhash2_mask)
2523 break;
2524 rc = listening_get_first(seq);
2525 while (offset-- && rc && bucket == st->bucket)
2526 rc = listening_get_next(seq, rc);
2527 if (rc)
2528 break;
2529 st->bucket = 0;
2530 st->state = TCP_SEQ_STATE_ESTABLISHED;
2531 fallthrough;
2532 case TCP_SEQ_STATE_ESTABLISHED:
2533 if (st->bucket > hinfo->ehash_mask)
2534 break;
2535 rc = established_get_first(seq);
2536 while (offset-- && rc && bucket == st->bucket)
2537 rc = established_get_next(seq, rc);
2538 }
2539
2540 st->num = orig_num;
2541
2542 return rc;
2543}
2544
2545void *tcp_seq_start(struct seq_file *seq, loff_t *pos)
2546{
2547 struct tcp_iter_state *st = seq->private;
2548 void *rc;
2549
2550 if (*pos && *pos == st->last_pos) {
2551 rc = tcp_seek_last_pos(seq);
2552 if (rc)
2553 goto out;
2554 }
2555
2556 st->state = TCP_SEQ_STATE_LISTENING;
2557 st->num = 0;
2558 st->bucket = 0;
2559 st->offset = 0;
2560 rc = *pos ? tcp_get_idx(seq, *pos - 1) : SEQ_START_TOKEN;
2561
2562out:
2563 st->last_pos = *pos;
2564 return rc;
2565}
2566EXPORT_SYMBOL(tcp_seq_start);
2567
2568void *tcp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2569{
2570 struct tcp_iter_state *st = seq->private;
2571 void *rc = NULL;
2572
2573 if (v == SEQ_START_TOKEN) {
2574 rc = tcp_get_idx(seq, 0);
2575 goto out;
2576 }
2577
2578 switch (st->state) {
2579 case TCP_SEQ_STATE_LISTENING:
2580 rc = listening_get_next(seq, v);
2581 if (!rc) {
2582 st->state = TCP_SEQ_STATE_ESTABLISHED;
2583 st->bucket = 0;
2584 st->offset = 0;
2585 rc = established_get_first(seq);
2586 }
2587 break;
2588 case TCP_SEQ_STATE_ESTABLISHED:
2589 rc = established_get_next(seq, v);
2590 break;
2591 }
2592out:
2593 ++*pos;
2594 st->last_pos = *pos;
2595 return rc;
2596}
2597EXPORT_SYMBOL(tcp_seq_next);
2598
2599void tcp_seq_stop(struct seq_file *seq, void *v)
2600{
2601 struct inet_hashinfo *hinfo = seq_file_net(seq)->ipv4.tcp_death_row.hashinfo;
2602 struct tcp_iter_state *st = seq->private;
2603
2604 switch (st->state) {
2605 case TCP_SEQ_STATE_LISTENING:
2606 if (v != SEQ_START_TOKEN)
2607 spin_unlock(&hinfo->lhash2[st->bucket].lock);
2608 break;
2609 case TCP_SEQ_STATE_ESTABLISHED:
2610 if (v)
2611 spin_unlock_bh(inet_ehash_lockp(hinfo, st->bucket));
2612 break;
2613 }
2614}
2615EXPORT_SYMBOL(tcp_seq_stop);
2616
2617static void get_openreq4(const struct request_sock *req,
2618 struct seq_file *f, int i)
2619{
2620 const struct inet_request_sock *ireq = inet_rsk(req);
2621 long delta = req->rsk_timer.expires - jiffies;
2622
2623 seq_printf(f, "%4d: %08X:%04X %08X:%04X"
2624 " %02X %08X:%08X %02X:%08lX %08X %5u %8d %u %d %pK",
2625 i,
2626 ireq->ir_loc_addr,
2627 ireq->ir_num,
2628 ireq->ir_rmt_addr,
2629 ntohs(ireq->ir_rmt_port),
2630 TCP_SYN_RECV,
2631 0, 0, /* could print option size, but that is af dependent. */
2632 1, /* timers active (only the expire timer) */
2633 jiffies_delta_to_clock_t(delta),
2634 req->num_timeout,
2635 from_kuid_munged(seq_user_ns(f),
2636 sock_i_uid(req->rsk_listener)),
2637 0, /* non standard timer */
2638 0, /* open_requests have no inode */
2639 0,
2640 req);
2641}
2642
2643static void get_tcp4_sock(struct sock *sk, struct seq_file *f, int i)
2644{
2645 int timer_active;
2646 unsigned long timer_expires;
2647 const struct tcp_sock *tp = tcp_sk(sk);
2648 const struct inet_connection_sock *icsk = inet_csk(sk);
2649 const struct inet_sock *inet = inet_sk(sk);
2650 const struct fastopen_queue *fastopenq = &icsk->icsk_accept_queue.fastopenq;
2651 __be32 dest = inet->inet_daddr;
2652 __be32 src = inet->inet_rcv_saddr;
2653 __u16 destp = ntohs(inet->inet_dport);
2654 __u16 srcp = ntohs(inet->inet_sport);
2655 int rx_queue;
2656 int state;
2657
2658 if (icsk->icsk_pending == ICSK_TIME_RETRANS ||
2659 icsk->icsk_pending == ICSK_TIME_REO_TIMEOUT ||
2660 icsk->icsk_pending == ICSK_TIME_LOSS_PROBE) {
2661 timer_active = 1;
2662 timer_expires = icsk->icsk_timeout;
2663 } else if (icsk->icsk_pending == ICSK_TIME_PROBE0) {
2664 timer_active = 4;
2665 timer_expires = icsk->icsk_timeout;
2666 } else if (timer_pending(&sk->sk_timer)) {
2667 timer_active = 2;
2668 timer_expires = sk->sk_timer.expires;
2669 } else {
2670 timer_active = 0;
2671 timer_expires = jiffies;
2672 }
2673
2674 state = inet_sk_state_load(sk);
2675 if (state == TCP_LISTEN)
2676 rx_queue = READ_ONCE(sk->sk_ack_backlog);
2677 else
2678 /* Because we don't lock the socket,
2679 * we might find a transient negative value.
2680 */
2681 rx_queue = max_t(int, READ_ONCE(tp->rcv_nxt) -
2682 READ_ONCE(tp->copied_seq), 0);
2683
2684 seq_printf(f, "%4d: %08X:%04X %08X:%04X %02X %08X:%08X %02X:%08lX "
2685 "%08X %5u %8d %lu %d %pK %lu %lu %u %u %d",
2686 i, src, srcp, dest, destp, state,
2687 READ_ONCE(tp->write_seq) - tp->snd_una,
2688 rx_queue,
2689 timer_active,
2690 jiffies_delta_to_clock_t(timer_expires - jiffies),
2691 icsk->icsk_retransmits,
2692 from_kuid_munged(seq_user_ns(f), sock_i_uid(sk)),
2693 icsk->icsk_probes_out,
2694 sock_i_ino(sk),
2695 refcount_read(&sk->sk_refcnt), sk,
2696 jiffies_to_clock_t(icsk->icsk_rto),
2697 jiffies_to_clock_t(icsk->icsk_ack.ato),
2698 (icsk->icsk_ack.quick << 1) | inet_csk_in_pingpong_mode(sk),
2699 tcp_snd_cwnd(tp),
2700 state == TCP_LISTEN ?
2701 fastopenq->max_qlen :
2702 (tcp_in_initial_slowstart(tp) ? -1 : tp->snd_ssthresh));
2703}
2704
2705static void get_timewait4_sock(const struct inet_timewait_sock *tw,
2706 struct seq_file *f, int i)
2707{
2708 long delta = tw->tw_timer.expires - jiffies;
2709 __be32 dest, src;
2710 __u16 destp, srcp;
2711
2712 dest = tw->tw_daddr;
2713 src = tw->tw_rcv_saddr;
2714 destp = ntohs(tw->tw_dport);
2715 srcp = ntohs(tw->tw_sport);
2716
2717 seq_printf(f, "%4d: %08X:%04X %08X:%04X"
2718 " %02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %pK",
2719 i, src, srcp, dest, destp, tw->tw_substate, 0, 0,
2720 3, jiffies_delta_to_clock_t(delta), 0, 0, 0, 0,
2721 refcount_read(&tw->tw_refcnt), tw);
2722}
2723
2724#define TMPSZ 150
2725
2726static int tcp4_seq_show(struct seq_file *seq, void *v)
2727{
2728 struct tcp_iter_state *st;
2729 struct sock *sk = v;
2730
2731 seq_setwidth(seq, TMPSZ - 1);
2732 if (v == SEQ_START_TOKEN) {
2733 seq_puts(seq, " sl local_address rem_address st tx_queue "
2734 "rx_queue tr tm->when retrnsmt uid timeout "
2735 "inode");
2736 goto out;
2737 }
2738 st = seq->private;
2739
2740 if (sk->sk_state == TCP_TIME_WAIT)
2741 get_timewait4_sock(v, seq, st->num);
2742 else if (sk->sk_state == TCP_NEW_SYN_RECV)
2743 get_openreq4(v, seq, st->num);
2744 else
2745 get_tcp4_sock(v, seq, st->num);
2746out:
2747 seq_pad(seq, '\n');
2748 return 0;
2749}
2750
2751#ifdef CONFIG_BPF_SYSCALL
2752struct bpf_tcp_iter_state {
2753 struct tcp_iter_state state;
2754 unsigned int cur_sk;
2755 unsigned int end_sk;
2756 unsigned int max_sk;
2757 struct sock **batch;
2758 bool st_bucket_done;
2759};
2760
2761struct bpf_iter__tcp {
2762 __bpf_md_ptr(struct bpf_iter_meta *, meta);
2763 __bpf_md_ptr(struct sock_common *, sk_common);
2764 uid_t uid __aligned(8);
2765};
2766
2767static int tcp_prog_seq_show(struct bpf_prog *prog, struct bpf_iter_meta *meta,
2768 struct sock_common *sk_common, uid_t uid)
2769{
2770 struct bpf_iter__tcp ctx;
2771
2772 meta->seq_num--; /* skip SEQ_START_TOKEN */
2773 ctx.meta = meta;
2774 ctx.sk_common = sk_common;
2775 ctx.uid = uid;
2776 return bpf_iter_run_prog(prog, &ctx);
2777}
2778
2779static void bpf_iter_tcp_put_batch(struct bpf_tcp_iter_state *iter)
2780{
2781 while (iter->cur_sk < iter->end_sk)
2782 sock_put(iter->batch[iter->cur_sk++]);
2783}
2784
2785static int bpf_iter_tcp_realloc_batch(struct bpf_tcp_iter_state *iter,
2786 unsigned int new_batch_sz)
2787{
2788 struct sock **new_batch;
2789
2790 new_batch = kvmalloc(sizeof(*new_batch) * new_batch_sz,
2791 GFP_USER | __GFP_NOWARN);
2792 if (!new_batch)
2793 return -ENOMEM;
2794
2795 bpf_iter_tcp_put_batch(iter);
2796 kvfree(iter->batch);
2797 iter->batch = new_batch;
2798 iter->max_sk = new_batch_sz;
2799
2800 return 0;
2801}
2802
2803static unsigned int bpf_iter_tcp_listening_batch(struct seq_file *seq,
2804 struct sock *start_sk)
2805{
2806 struct inet_hashinfo *hinfo = seq_file_net(seq)->ipv4.tcp_death_row.hashinfo;
2807 struct bpf_tcp_iter_state *iter = seq->private;
2808 struct tcp_iter_state *st = &iter->state;
2809 struct hlist_nulls_node *node;
2810 unsigned int expected = 1;
2811 struct sock *sk;
2812
2813 sock_hold(start_sk);
2814 iter->batch[iter->end_sk++] = start_sk;
2815
2816 sk = sk_nulls_next(start_sk);
2817 sk_nulls_for_each_from(sk, node) {
2818 if (seq_sk_match(seq, sk)) {
2819 if (iter->end_sk < iter->max_sk) {
2820 sock_hold(sk);
2821 iter->batch[iter->end_sk++] = sk;
2822 }
2823 expected++;
2824 }
2825 }
2826 spin_unlock(&hinfo->lhash2[st->bucket].lock);
2827
2828 return expected;
2829}
2830
2831static unsigned int bpf_iter_tcp_established_batch(struct seq_file *seq,
2832 struct sock *start_sk)
2833{
2834 struct inet_hashinfo *hinfo = seq_file_net(seq)->ipv4.tcp_death_row.hashinfo;
2835 struct bpf_tcp_iter_state *iter = seq->private;
2836 struct tcp_iter_state *st = &iter->state;
2837 struct hlist_nulls_node *node;
2838 unsigned int expected = 1;
2839 struct sock *sk;
2840
2841 sock_hold(start_sk);
2842 iter->batch[iter->end_sk++] = start_sk;
2843
2844 sk = sk_nulls_next(start_sk);
2845 sk_nulls_for_each_from(sk, node) {
2846 if (seq_sk_match(seq, sk)) {
2847 if (iter->end_sk < iter->max_sk) {
2848 sock_hold(sk);
2849 iter->batch[iter->end_sk++] = sk;
2850 }
2851 expected++;
2852 }
2853 }
2854 spin_unlock_bh(inet_ehash_lockp(hinfo, st->bucket));
2855
2856 return expected;
2857}
2858
2859static struct sock *bpf_iter_tcp_batch(struct seq_file *seq)
2860{
2861 struct inet_hashinfo *hinfo = seq_file_net(seq)->ipv4.tcp_death_row.hashinfo;
2862 struct bpf_tcp_iter_state *iter = seq->private;
2863 struct tcp_iter_state *st = &iter->state;
2864 unsigned int expected;
2865 bool resized = false;
2866 struct sock *sk;
2867
2868 /* The st->bucket is done. Directly advance to the next
2869 * bucket instead of having the tcp_seek_last_pos() to skip
2870 * one by one in the current bucket and eventually find out
2871 * it has to advance to the next bucket.
2872 */
2873 if (iter->st_bucket_done) {
2874 st->offset = 0;
2875 st->bucket++;
2876 if (st->state == TCP_SEQ_STATE_LISTENING &&
2877 st->bucket > hinfo->lhash2_mask) {
2878 st->state = TCP_SEQ_STATE_ESTABLISHED;
2879 st->bucket = 0;
2880 }
2881 }
2882
2883again:
2884 /* Get a new batch */
2885 iter->cur_sk = 0;
2886 iter->end_sk = 0;
2887 iter->st_bucket_done = false;
2888
2889 sk = tcp_seek_last_pos(seq);
2890 if (!sk)
2891 return NULL; /* Done */
2892
2893 if (st->state == TCP_SEQ_STATE_LISTENING)
2894 expected = bpf_iter_tcp_listening_batch(seq, sk);
2895 else
2896 expected = bpf_iter_tcp_established_batch(seq, sk);
2897
2898 if (iter->end_sk == expected) {
2899 iter->st_bucket_done = true;
2900 return sk;
2901 }
2902
2903 if (!resized && !bpf_iter_tcp_realloc_batch(iter, expected * 3 / 2)) {
2904 resized = true;
2905 goto again;
2906 }
2907
2908 return sk;
2909}
2910
2911static void *bpf_iter_tcp_seq_start(struct seq_file *seq, loff_t *pos)
2912{
2913 /* bpf iter does not support lseek, so it always
2914 * continue from where it was stop()-ped.
2915 */
2916 if (*pos)
2917 return bpf_iter_tcp_batch(seq);
2918
2919 return SEQ_START_TOKEN;
2920}
2921
2922static void *bpf_iter_tcp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2923{
2924 struct bpf_tcp_iter_state *iter = seq->private;
2925 struct tcp_iter_state *st = &iter->state;
2926 struct sock *sk;
2927
2928 /* Whenever seq_next() is called, the iter->cur_sk is
2929 * done with seq_show(), so advance to the next sk in
2930 * the batch.
2931 */
2932 if (iter->cur_sk < iter->end_sk) {
2933 /* Keeping st->num consistent in tcp_iter_state.
2934 * bpf_iter_tcp does not use st->num.
2935 * meta.seq_num is used instead.
2936 */
2937 st->num++;
2938 /* Move st->offset to the next sk in the bucket such that
2939 * the future start() will resume at st->offset in
2940 * st->bucket. See tcp_seek_last_pos().
2941 */
2942 st->offset++;
2943 sock_put(iter->batch[iter->cur_sk++]);
2944 }
2945
2946 if (iter->cur_sk < iter->end_sk)
2947 sk = iter->batch[iter->cur_sk];
2948 else
2949 sk = bpf_iter_tcp_batch(seq);
2950
2951 ++*pos;
2952 /* Keeping st->last_pos consistent in tcp_iter_state.
2953 * bpf iter does not do lseek, so st->last_pos always equals to *pos.
2954 */
2955 st->last_pos = *pos;
2956 return sk;
2957}
2958
2959static int bpf_iter_tcp_seq_show(struct seq_file *seq, void *v)
2960{
2961 struct bpf_iter_meta meta;
2962 struct bpf_prog *prog;
2963 struct sock *sk = v;
2964 bool slow;
2965 uid_t uid;
2966 int ret;
2967
2968 if (v == SEQ_START_TOKEN)
2969 return 0;
2970
2971 if (sk_fullsock(sk))
2972 slow = lock_sock_fast(sk);
2973
2974 if (unlikely(sk_unhashed(sk))) {
2975 ret = SEQ_SKIP;
2976 goto unlock;
2977 }
2978
2979 if (sk->sk_state == TCP_TIME_WAIT) {
2980 uid = 0;
2981 } else if (sk->sk_state == TCP_NEW_SYN_RECV) {
2982 const struct request_sock *req = v;
2983
2984 uid = from_kuid_munged(seq_user_ns(seq),
2985 sock_i_uid(req->rsk_listener));
2986 } else {
2987 uid = from_kuid_munged(seq_user_ns(seq), sock_i_uid(sk));
2988 }
2989
2990 meta.seq = seq;
2991 prog = bpf_iter_get_info(&meta, false);
2992 ret = tcp_prog_seq_show(prog, &meta, v, uid);
2993
2994unlock:
2995 if (sk_fullsock(sk))
2996 unlock_sock_fast(sk, slow);
2997 return ret;
2998
2999}
3000
3001static void bpf_iter_tcp_seq_stop(struct seq_file *seq, void *v)
3002{
3003 struct bpf_tcp_iter_state *iter = seq->private;
3004 struct bpf_iter_meta meta;
3005 struct bpf_prog *prog;
3006
3007 if (!v) {
3008 meta.seq = seq;
3009 prog = bpf_iter_get_info(&meta, true);
3010 if (prog)
3011 (void)tcp_prog_seq_show(prog, &meta, v, 0);
3012 }
3013
3014 if (iter->cur_sk < iter->end_sk) {
3015 bpf_iter_tcp_put_batch(iter);
3016 iter->st_bucket_done = false;
3017 }
3018}
3019
3020static const struct seq_operations bpf_iter_tcp_seq_ops = {
3021 .show = bpf_iter_tcp_seq_show,
3022 .start = bpf_iter_tcp_seq_start,
3023 .next = bpf_iter_tcp_seq_next,
3024 .stop = bpf_iter_tcp_seq_stop,
3025};
3026#endif
3027static unsigned short seq_file_family(const struct seq_file *seq)
3028{
3029 const struct tcp_seq_afinfo *afinfo;
3030
3031#ifdef CONFIG_BPF_SYSCALL
3032 /* Iterated from bpf_iter. Let the bpf prog to filter instead. */
3033 if (seq->op == &bpf_iter_tcp_seq_ops)
3034 return AF_UNSPEC;
3035#endif
3036
3037 /* Iterated from proc fs */
3038 afinfo = pde_data(file_inode(seq->file));
3039 return afinfo->family;
3040}
3041
3042static const struct seq_operations tcp4_seq_ops = {
3043 .show = tcp4_seq_show,
3044 .start = tcp_seq_start,
3045 .next = tcp_seq_next,
3046 .stop = tcp_seq_stop,
3047};
3048
3049static struct tcp_seq_afinfo tcp4_seq_afinfo = {
3050 .family = AF_INET,
3051};
3052
3053static int __net_init tcp4_proc_init_net(struct net *net)
3054{
3055 if (!proc_create_net_data("tcp", 0444, net->proc_net, &tcp4_seq_ops,
3056 sizeof(struct tcp_iter_state), &tcp4_seq_afinfo))
3057 return -ENOMEM;
3058 return 0;
3059}
3060
3061static void __net_exit tcp4_proc_exit_net(struct net *net)
3062{
3063 remove_proc_entry("tcp", net->proc_net);
3064}
3065
3066static struct pernet_operations tcp4_net_ops = {
3067 .init = tcp4_proc_init_net,
3068 .exit = tcp4_proc_exit_net,
3069};
3070
3071int __init tcp4_proc_init(void)
3072{
3073 return register_pernet_subsys(&tcp4_net_ops);
3074}
3075
3076void tcp4_proc_exit(void)
3077{
3078 unregister_pernet_subsys(&tcp4_net_ops);
3079}
3080#endif /* CONFIG_PROC_FS */
3081
3082/* @wake is one when sk_stream_write_space() calls us.
3083 * This sends EPOLLOUT only if notsent_bytes is half the limit.
3084 * This mimics the strategy used in sock_def_write_space().
3085 */
3086bool tcp_stream_memory_free(const struct sock *sk, int wake)
3087{
3088 const struct tcp_sock *tp = tcp_sk(sk);
3089 u32 notsent_bytes = READ_ONCE(tp->write_seq) -
3090 READ_ONCE(tp->snd_nxt);
3091
3092 return (notsent_bytes << wake) < tcp_notsent_lowat(tp);
3093}
3094EXPORT_SYMBOL(tcp_stream_memory_free);
3095
3096struct proto tcp_prot = {
3097 .name = "TCP",
3098 .owner = THIS_MODULE,
3099 .close = tcp_close,
3100 .pre_connect = tcp_v4_pre_connect,
3101 .connect = tcp_v4_connect,
3102 .disconnect = tcp_disconnect,
3103 .accept = inet_csk_accept,
3104 .ioctl = tcp_ioctl,
3105 .init = tcp_v4_init_sock,
3106 .destroy = tcp_v4_destroy_sock,
3107 .shutdown = tcp_shutdown,
3108 .setsockopt = tcp_setsockopt,
3109 .getsockopt = tcp_getsockopt,
3110 .bpf_bypass_getsockopt = tcp_bpf_bypass_getsockopt,
3111 .keepalive = tcp_set_keepalive,
3112 .recvmsg = tcp_recvmsg,
3113 .sendmsg = tcp_sendmsg,
3114 .sendpage = tcp_sendpage,
3115 .backlog_rcv = tcp_v4_do_rcv,
3116 .release_cb = tcp_release_cb,
3117 .hash = inet_hash,
3118 .unhash = inet_unhash,
3119 .get_port = inet_csk_get_port,
3120 .put_port = inet_put_port,
3121#ifdef CONFIG_BPF_SYSCALL
3122 .psock_update_sk_prot = tcp_bpf_update_proto,
3123#endif
3124 .enter_memory_pressure = tcp_enter_memory_pressure,
3125 .leave_memory_pressure = tcp_leave_memory_pressure,
3126 .stream_memory_free = tcp_stream_memory_free,
3127 .sockets_allocated = &tcp_sockets_allocated,
3128 .orphan_count = &tcp_orphan_count,
3129
3130 .memory_allocated = &tcp_memory_allocated,
3131 .per_cpu_fw_alloc = &tcp_memory_per_cpu_fw_alloc,
3132
3133 .memory_pressure = &tcp_memory_pressure,
3134 .sysctl_mem = sysctl_tcp_mem,
3135 .sysctl_wmem_offset = offsetof(struct net, ipv4.sysctl_tcp_wmem),
3136 .sysctl_rmem_offset = offsetof(struct net, ipv4.sysctl_tcp_rmem),
3137 .max_header = MAX_TCP_HEADER,
3138 .obj_size = sizeof(struct tcp_sock),
3139 .slab_flags = SLAB_TYPESAFE_BY_RCU,
3140 .twsk_prot = &tcp_timewait_sock_ops,
3141 .rsk_prot = &tcp_request_sock_ops,
3142 .h.hashinfo = NULL,
3143 .no_autobind = true,
3144 .diag_destroy = tcp_abort,
3145};
3146EXPORT_SYMBOL(tcp_prot);
3147
3148static void __net_exit tcp_sk_exit(struct net *net)
3149{
3150 if (net->ipv4.tcp_congestion_control)
3151 bpf_module_put(net->ipv4.tcp_congestion_control,
3152 net->ipv4.tcp_congestion_control->owner);
3153}
3154
3155static void __net_init tcp_set_hashinfo(struct net *net)
3156{
3157 struct inet_hashinfo *hinfo;
3158 unsigned int ehash_entries;
3159 struct net *old_net;
3160
3161 if (net_eq(net, &init_net))
3162 goto fallback;
3163
3164 old_net = current->nsproxy->net_ns;
3165 ehash_entries = READ_ONCE(old_net->ipv4.sysctl_tcp_child_ehash_entries);
3166 if (!ehash_entries)
3167 goto fallback;
3168
3169 ehash_entries = roundup_pow_of_two(ehash_entries);
3170 hinfo = inet_pernet_hashinfo_alloc(&tcp_hashinfo, ehash_entries);
3171 if (!hinfo) {
3172 pr_warn("Failed to allocate TCP ehash (entries: %u) "
3173 "for a netns, fallback to the global one\n",
3174 ehash_entries);
3175fallback:
3176 hinfo = &tcp_hashinfo;
3177 ehash_entries = tcp_hashinfo.ehash_mask + 1;
3178 }
3179
3180 net->ipv4.tcp_death_row.hashinfo = hinfo;
3181 net->ipv4.tcp_death_row.sysctl_max_tw_buckets = ehash_entries / 2;
3182 net->ipv4.sysctl_max_syn_backlog = max(128U, ehash_entries / 128);
3183}
3184
3185static int __net_init tcp_sk_init(struct net *net)
3186{
3187 net->ipv4.sysctl_tcp_ecn = 2;
3188 net->ipv4.sysctl_tcp_ecn_fallback = 1;
3189
3190 net->ipv4.sysctl_tcp_base_mss = TCP_BASE_MSS;
3191 net->ipv4.sysctl_tcp_min_snd_mss = TCP_MIN_SND_MSS;
3192 net->ipv4.sysctl_tcp_probe_threshold = TCP_PROBE_THRESHOLD;
3193 net->ipv4.sysctl_tcp_probe_interval = TCP_PROBE_INTERVAL;
3194 net->ipv4.sysctl_tcp_mtu_probe_floor = TCP_MIN_SND_MSS;
3195
3196 net->ipv4.sysctl_tcp_keepalive_time = TCP_KEEPALIVE_TIME;
3197 net->ipv4.sysctl_tcp_keepalive_probes = TCP_KEEPALIVE_PROBES;
3198 net->ipv4.sysctl_tcp_keepalive_intvl = TCP_KEEPALIVE_INTVL;
3199
3200 net->ipv4.sysctl_tcp_syn_retries = TCP_SYN_RETRIES;
3201 net->ipv4.sysctl_tcp_synack_retries = TCP_SYNACK_RETRIES;
3202 net->ipv4.sysctl_tcp_syncookies = 1;
3203 net->ipv4.sysctl_tcp_reordering = TCP_FASTRETRANS_THRESH;
3204 net->ipv4.sysctl_tcp_retries1 = TCP_RETR1;
3205 net->ipv4.sysctl_tcp_retries2 = TCP_RETR2;
3206 net->ipv4.sysctl_tcp_orphan_retries = 0;
3207 net->ipv4.sysctl_tcp_fin_timeout = TCP_FIN_TIMEOUT;
3208 net->ipv4.sysctl_tcp_notsent_lowat = UINT_MAX;
3209 net->ipv4.sysctl_tcp_tw_reuse = 2;
3210 net->ipv4.sysctl_tcp_no_ssthresh_metrics_save = 1;
3211
3212 refcount_set(&net->ipv4.tcp_death_row.tw_refcount, 1);
3213 tcp_set_hashinfo(net);
3214
3215 net->ipv4.sysctl_tcp_sack = 1;
3216 net->ipv4.sysctl_tcp_window_scaling = 1;
3217 net->ipv4.sysctl_tcp_timestamps = 1;
3218 net->ipv4.sysctl_tcp_early_retrans = 3;
3219 net->ipv4.sysctl_tcp_recovery = TCP_RACK_LOSS_DETECTION;
3220 net->ipv4.sysctl_tcp_slow_start_after_idle = 1; /* By default, RFC2861 behavior. */
3221 net->ipv4.sysctl_tcp_retrans_collapse = 1;
3222 net->ipv4.sysctl_tcp_max_reordering = 300;
3223 net->ipv4.sysctl_tcp_dsack = 1;
3224 net->ipv4.sysctl_tcp_app_win = 31;
3225 net->ipv4.sysctl_tcp_adv_win_scale = 1;
3226 net->ipv4.sysctl_tcp_frto = 2;
3227 net->ipv4.sysctl_tcp_moderate_rcvbuf = 1;
3228 /* This limits the percentage of the congestion window which we
3229 * will allow a single TSO frame to consume. Building TSO frames
3230 * which are too large can cause TCP streams to be bursty.
3231 */
3232 net->ipv4.sysctl_tcp_tso_win_divisor = 3;
3233 /* Default TSQ limit of 16 TSO segments */
3234 net->ipv4.sysctl_tcp_limit_output_bytes = 16 * 65536;
3235
3236 /* rfc5961 challenge ack rate limiting, per net-ns, disabled by default. */
3237 net->ipv4.sysctl_tcp_challenge_ack_limit = INT_MAX;
3238
3239 net->ipv4.sysctl_tcp_min_tso_segs = 2;
3240 net->ipv4.sysctl_tcp_tso_rtt_log = 9; /* 2^9 = 512 usec */
3241 net->ipv4.sysctl_tcp_min_rtt_wlen = 300;
3242 net->ipv4.sysctl_tcp_autocorking = 1;
3243 net->ipv4.sysctl_tcp_invalid_ratelimit = HZ/2;
3244 net->ipv4.sysctl_tcp_pacing_ss_ratio = 200;
3245 net->ipv4.sysctl_tcp_pacing_ca_ratio = 120;
3246 if (net != &init_net) {
3247 memcpy(net->ipv4.sysctl_tcp_rmem,
3248 init_net.ipv4.sysctl_tcp_rmem,
3249 sizeof(init_net.ipv4.sysctl_tcp_rmem));
3250 memcpy(net->ipv4.sysctl_tcp_wmem,
3251 init_net.ipv4.sysctl_tcp_wmem,
3252 sizeof(init_net.ipv4.sysctl_tcp_wmem));
3253 }
3254 net->ipv4.sysctl_tcp_comp_sack_delay_ns = NSEC_PER_MSEC;
3255 net->ipv4.sysctl_tcp_comp_sack_slack_ns = 100 * NSEC_PER_USEC;
3256 net->ipv4.sysctl_tcp_comp_sack_nr = 44;
3257 net->ipv4.sysctl_tcp_fastopen = TFO_CLIENT_ENABLE;
3258 net->ipv4.sysctl_tcp_fastopen_blackhole_timeout = 0;
3259 atomic_set(&net->ipv4.tfo_active_disable_times, 0);
3260
3261 /* Set default values for PLB */
3262 net->ipv4.sysctl_tcp_plb_enabled = 0; /* Disabled by default */
3263 net->ipv4.sysctl_tcp_plb_idle_rehash_rounds = 3;
3264 net->ipv4.sysctl_tcp_plb_rehash_rounds = 12;
3265 net->ipv4.sysctl_tcp_plb_suspend_rto_sec = 60;
3266 /* Default congestion threshold for PLB to mark a round is 50% */
3267 net->ipv4.sysctl_tcp_plb_cong_thresh = (1 << TCP_PLB_SCALE) / 2;
3268
3269 /* Reno is always built in */
3270 if (!net_eq(net, &init_net) &&
3271 bpf_try_module_get(init_net.ipv4.tcp_congestion_control,
3272 init_net.ipv4.tcp_congestion_control->owner))
3273 net->ipv4.tcp_congestion_control = init_net.ipv4.tcp_congestion_control;
3274 else
3275 net->ipv4.tcp_congestion_control = &tcp_reno;
3276
3277 return 0;
3278}
3279
3280static void __net_exit tcp_sk_exit_batch(struct list_head *net_exit_list)
3281{
3282 struct net *net;
3283
3284 tcp_twsk_purge(net_exit_list, AF_INET);
3285
3286 list_for_each_entry(net, net_exit_list, exit_list) {
3287 inet_pernet_hashinfo_free(net->ipv4.tcp_death_row.hashinfo);
3288 WARN_ON_ONCE(!refcount_dec_and_test(&net->ipv4.tcp_death_row.tw_refcount));
3289 tcp_fastopen_ctx_destroy(net);
3290 }
3291}
3292
3293static struct pernet_operations __net_initdata tcp_sk_ops = {
3294 .init = tcp_sk_init,
3295 .exit = tcp_sk_exit,
3296 .exit_batch = tcp_sk_exit_batch,
3297};
3298
3299#if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
3300DEFINE_BPF_ITER_FUNC(tcp, struct bpf_iter_meta *meta,
3301 struct sock_common *sk_common, uid_t uid)
3302
3303#define INIT_BATCH_SZ 16
3304
3305static int bpf_iter_init_tcp(void *priv_data, struct bpf_iter_aux_info *aux)
3306{
3307 struct bpf_tcp_iter_state *iter = priv_data;
3308 int err;
3309
3310 err = bpf_iter_init_seq_net(priv_data, aux);
3311 if (err)
3312 return err;
3313
3314 err = bpf_iter_tcp_realloc_batch(iter, INIT_BATCH_SZ);
3315 if (err) {
3316 bpf_iter_fini_seq_net(priv_data);
3317 return err;
3318 }
3319
3320 return 0;
3321}
3322
3323static void bpf_iter_fini_tcp(void *priv_data)
3324{
3325 struct bpf_tcp_iter_state *iter = priv_data;
3326
3327 bpf_iter_fini_seq_net(priv_data);
3328 kvfree(iter->batch);
3329}
3330
3331static const struct bpf_iter_seq_info tcp_seq_info = {
3332 .seq_ops = &bpf_iter_tcp_seq_ops,
3333 .init_seq_private = bpf_iter_init_tcp,
3334 .fini_seq_private = bpf_iter_fini_tcp,
3335 .seq_priv_size = sizeof(struct bpf_tcp_iter_state),
3336};
3337
3338static const struct bpf_func_proto *
3339bpf_iter_tcp_get_func_proto(enum bpf_func_id func_id,
3340 const struct bpf_prog *prog)
3341{
3342 switch (func_id) {
3343 case BPF_FUNC_setsockopt:
3344 return &bpf_sk_setsockopt_proto;
3345 case BPF_FUNC_getsockopt:
3346 return &bpf_sk_getsockopt_proto;
3347 default:
3348 return NULL;
3349 }
3350}
3351
3352static struct bpf_iter_reg tcp_reg_info = {
3353 .target = "tcp",
3354 .ctx_arg_info_size = 1,
3355 .ctx_arg_info = {
3356 { offsetof(struct bpf_iter__tcp, sk_common),
3357 PTR_TO_BTF_ID_OR_NULL },
3358 },
3359 .get_func_proto = bpf_iter_tcp_get_func_proto,
3360 .seq_info = &tcp_seq_info,
3361};
3362
3363static void __init bpf_iter_register(void)
3364{
3365 tcp_reg_info.ctx_arg_info[0].btf_id = btf_sock_ids[BTF_SOCK_TYPE_SOCK_COMMON];
3366 if (bpf_iter_reg_target(&tcp_reg_info))
3367 pr_warn("Warning: could not register bpf iterator tcp\n");
3368}
3369
3370#endif
3371
3372void __init tcp_v4_init(void)
3373{
3374 int cpu, res;
3375
3376 for_each_possible_cpu(cpu) {
3377 struct sock *sk;
3378
3379 res = inet_ctl_sock_create(&sk, PF_INET, SOCK_RAW,
3380 IPPROTO_TCP, &init_net);
3381 if (res)
3382 panic("Failed to create the TCP control socket.\n");
3383 sock_set_flag(sk, SOCK_USE_WRITE_QUEUE);
3384
3385 /* Please enforce IP_DF and IPID==0 for RST and
3386 * ACK sent in SYN-RECV and TIME-WAIT state.
3387 */
3388 inet_sk(sk)->pmtudisc = IP_PMTUDISC_DO;
3389
3390 per_cpu(ipv4_tcp_sk, cpu) = sk;
3391 }
3392 if (register_pernet_subsys(&tcp_sk_ops))
3393 panic("Failed to create the TCP control socket.\n");
3394
3395#if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
3396 bpf_iter_register();
3397#endif
3398}
1/*
2 * INET An implementation of the TCP/IP protocol suite for the LINUX
3 * operating system. INET is implemented using the BSD Socket
4 * interface as the means of communication with the user level.
5 *
6 * Implementation of the Transmission Control Protocol(TCP).
7 *
8 * IPv4 specific functions
9 *
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 * This program is free software; you can redistribute it and/or
19 * modify it under the terms of the GNU General Public License
20 * as published by the Free Software Foundation; either version
21 * 2 of the License, or (at your option) any later version.
22 */
23
24/*
25 * Changes:
26 * David S. Miller : New socket lookup architecture.
27 * This code is dedicated to John Dyson.
28 * David S. Miller : Change semantics of established hash,
29 * half is devoted to TIME_WAIT sockets
30 * and the rest go in the other half.
31 * Andi Kleen : Add support for syncookies and fixed
32 * some bugs: ip options weren't passed to
33 * the TCP layer, missed a check for an
34 * ACK bit.
35 * Andi Kleen : Implemented fast path mtu discovery.
36 * Fixed many serious bugs in the
37 * request_sock handling and moved
38 * most of it into the af independent code.
39 * Added tail drop and some other bugfixes.
40 * Added new listen semantics.
41 * Mike McLagan : Routing by source
42 * Juan Jose Ciarlante: ip_dynaddr bits
43 * Andi Kleen: various fixes.
44 * Vitaly E. Lavrov : Transparent proxy revived after year
45 * coma.
46 * Andi Kleen : Fix new listen.
47 * Andi Kleen : Fix accept error reporting.
48 * YOSHIFUJI Hideaki @USAGI and: Support IPV6_V6ONLY socket option, which
49 * Alexey Kuznetsov allow both IPv4 and IPv6 sockets to bind
50 * a single port at the same time.
51 */
52
53#define pr_fmt(fmt) "TCP: " fmt
54
55#include <linux/bottom_half.h>
56#include <linux/types.h>
57#include <linux/fcntl.h>
58#include <linux/module.h>
59#include <linux/random.h>
60#include <linux/cache.h>
61#include <linux/jhash.h>
62#include <linux/init.h>
63#include <linux/times.h>
64#include <linux/slab.h>
65
66#include <net/net_namespace.h>
67#include <net/icmp.h>
68#include <net/inet_hashtables.h>
69#include <net/tcp.h>
70#include <net/transp_v6.h>
71#include <net/ipv6.h>
72#include <net/inet_common.h>
73#include <net/timewait_sock.h>
74#include <net/xfrm.h>
75#include <net/secure_seq.h>
76#include <net/busy_poll.h>
77
78#include <linux/inet.h>
79#include <linux/ipv6.h>
80#include <linux/stddef.h>
81#include <linux/proc_fs.h>
82#include <linux/seq_file.h>
83
84#include <crypto/hash.h>
85#include <linux/scatterlist.h>
86
87int sysctl_tcp_low_latency __read_mostly;
88
89#ifdef CONFIG_TCP_MD5SIG
90static int tcp_v4_md5_hash_hdr(char *md5_hash, const struct tcp_md5sig_key *key,
91 __be32 daddr, __be32 saddr, const struct tcphdr *th);
92#endif
93
94struct inet_hashinfo tcp_hashinfo;
95EXPORT_SYMBOL(tcp_hashinfo);
96
97static u32 tcp_v4_init_sequence(const struct sk_buff *skb, u32 *tsoff)
98{
99 return secure_tcp_sequence_number(ip_hdr(skb)->daddr,
100 ip_hdr(skb)->saddr,
101 tcp_hdr(skb)->dest,
102 tcp_hdr(skb)->source, tsoff);
103}
104
105int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp)
106{
107 const struct tcp_timewait_sock *tcptw = tcp_twsk(sktw);
108 struct tcp_sock *tp = tcp_sk(sk);
109
110 /* With PAWS, it is safe from the viewpoint
111 of data integrity. Even without PAWS it is safe provided sequence
112 spaces do not overlap i.e. at data rates <= 80Mbit/sec.
113
114 Actually, the idea is close to VJ's one, only timestamp cache is
115 held not per host, but per port pair and TW bucket is used as state
116 holder.
117
118 If TW bucket has been already destroyed we fall back to VJ's scheme
119 and use initial timestamp retrieved from peer table.
120 */
121 if (tcptw->tw_ts_recent_stamp &&
122 (!twp || (sock_net(sk)->ipv4.sysctl_tcp_tw_reuse &&
123 get_seconds() - tcptw->tw_ts_recent_stamp > 1))) {
124 tp->write_seq = tcptw->tw_snd_nxt + 65535 + 2;
125 if (tp->write_seq == 0)
126 tp->write_seq = 1;
127 tp->rx_opt.ts_recent = tcptw->tw_ts_recent;
128 tp->rx_opt.ts_recent_stamp = tcptw->tw_ts_recent_stamp;
129 sock_hold(sktw);
130 return 1;
131 }
132
133 return 0;
134}
135EXPORT_SYMBOL_GPL(tcp_twsk_unique);
136
137/* This will initiate an outgoing connection. */
138int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len)
139{
140 struct sockaddr_in *usin = (struct sockaddr_in *)uaddr;
141 struct inet_sock *inet = inet_sk(sk);
142 struct tcp_sock *tp = tcp_sk(sk);
143 __be16 orig_sport, orig_dport;
144 __be32 daddr, nexthop;
145 struct flowi4 *fl4;
146 struct rtable *rt;
147 int err;
148 struct ip_options_rcu *inet_opt;
149
150 if (addr_len < sizeof(struct sockaddr_in))
151 return -EINVAL;
152
153 if (usin->sin_family != AF_INET)
154 return -EAFNOSUPPORT;
155
156 nexthop = daddr = usin->sin_addr.s_addr;
157 inet_opt = rcu_dereference_protected(inet->inet_opt,
158 lockdep_sock_is_held(sk));
159 if (inet_opt && inet_opt->opt.srr) {
160 if (!daddr)
161 return -EINVAL;
162 nexthop = inet_opt->opt.faddr;
163 }
164
165 orig_sport = inet->inet_sport;
166 orig_dport = usin->sin_port;
167 fl4 = &inet->cork.fl.u.ip4;
168 rt = ip_route_connect(fl4, nexthop, inet->inet_saddr,
169 RT_CONN_FLAGS(sk), sk->sk_bound_dev_if,
170 IPPROTO_TCP,
171 orig_sport, orig_dport, sk);
172 if (IS_ERR(rt)) {
173 err = PTR_ERR(rt);
174 if (err == -ENETUNREACH)
175 IP_INC_STATS(sock_net(sk), IPSTATS_MIB_OUTNOROUTES);
176 return err;
177 }
178
179 if (rt->rt_flags & (RTCF_MULTICAST | RTCF_BROADCAST)) {
180 ip_rt_put(rt);
181 return -ENETUNREACH;
182 }
183
184 if (!inet_opt || !inet_opt->opt.srr)
185 daddr = fl4->daddr;
186
187 if (!inet->inet_saddr)
188 inet->inet_saddr = fl4->saddr;
189 sk_rcv_saddr_set(sk, inet->inet_saddr);
190
191 if (tp->rx_opt.ts_recent_stamp && inet->inet_daddr != daddr) {
192 /* Reset inherited state */
193 tp->rx_opt.ts_recent = 0;
194 tp->rx_opt.ts_recent_stamp = 0;
195 if (likely(!tp->repair))
196 tp->write_seq = 0;
197 }
198
199 if (tcp_death_row.sysctl_tw_recycle &&
200 !tp->rx_opt.ts_recent_stamp && fl4->daddr == daddr)
201 tcp_fetch_timewait_stamp(sk, &rt->dst);
202
203 inet->inet_dport = usin->sin_port;
204 sk_daddr_set(sk, daddr);
205
206 inet_csk(sk)->icsk_ext_hdr_len = 0;
207 if (inet_opt)
208 inet_csk(sk)->icsk_ext_hdr_len = inet_opt->opt.optlen;
209
210 tp->rx_opt.mss_clamp = TCP_MSS_DEFAULT;
211
212 /* Socket identity is still unknown (sport may be zero).
213 * However we set state to SYN-SENT and not releasing socket
214 * lock select source port, enter ourselves into the hash tables and
215 * complete initialization after this.
216 */
217 tcp_set_state(sk, TCP_SYN_SENT);
218 err = inet_hash_connect(&tcp_death_row, sk);
219 if (err)
220 goto failure;
221
222 sk_set_txhash(sk);
223
224 rt = ip_route_newports(fl4, rt, orig_sport, orig_dport,
225 inet->inet_sport, inet->inet_dport, sk);
226 if (IS_ERR(rt)) {
227 err = PTR_ERR(rt);
228 rt = NULL;
229 goto failure;
230 }
231 /* OK, now commit destination to socket. */
232 sk->sk_gso_type = SKB_GSO_TCPV4;
233 sk_setup_caps(sk, &rt->dst);
234
235 if (!tp->write_seq && likely(!tp->repair))
236 tp->write_seq = secure_tcp_sequence_number(inet->inet_saddr,
237 inet->inet_daddr,
238 inet->inet_sport,
239 usin->sin_port,
240 &tp->tsoffset);
241
242 inet->inet_id = tp->write_seq ^ jiffies;
243
244 err = tcp_connect(sk);
245
246 rt = NULL;
247 if (err)
248 goto failure;
249
250 return 0;
251
252failure:
253 /*
254 * This unhashes the socket and releases the local port,
255 * if necessary.
256 */
257 tcp_set_state(sk, TCP_CLOSE);
258 ip_rt_put(rt);
259 sk->sk_route_caps = 0;
260 inet->inet_dport = 0;
261 return err;
262}
263EXPORT_SYMBOL(tcp_v4_connect);
264
265/*
266 * This routine reacts to ICMP_FRAG_NEEDED mtu indications as defined in RFC1191.
267 * It can be called through tcp_release_cb() if socket was owned by user
268 * at the time tcp_v4_err() was called to handle ICMP message.
269 */
270void tcp_v4_mtu_reduced(struct sock *sk)
271{
272 struct inet_sock *inet = inet_sk(sk);
273 struct dst_entry *dst;
274 u32 mtu;
275
276 if ((1 << sk->sk_state) & (TCPF_LISTEN | TCPF_CLOSE))
277 return;
278 mtu = tcp_sk(sk)->mtu_info;
279 dst = inet_csk_update_pmtu(sk, mtu);
280 if (!dst)
281 return;
282
283 /* Something is about to be wrong... Remember soft error
284 * for the case, if this connection will not able to recover.
285 */
286 if (mtu < dst_mtu(dst) && ip_dont_fragment(sk, dst))
287 sk->sk_err_soft = EMSGSIZE;
288
289 mtu = dst_mtu(dst);
290
291 if (inet->pmtudisc != IP_PMTUDISC_DONT &&
292 ip_sk_accept_pmtu(sk) &&
293 inet_csk(sk)->icsk_pmtu_cookie > mtu) {
294 tcp_sync_mss(sk, mtu);
295
296 /* Resend the TCP packet because it's
297 * clear that the old packet has been
298 * dropped. This is the new "fast" path mtu
299 * discovery.
300 */
301 tcp_simple_retransmit(sk);
302 } /* else let the usual retransmit timer handle it */
303}
304EXPORT_SYMBOL(tcp_v4_mtu_reduced);
305
306static void do_redirect(struct sk_buff *skb, struct sock *sk)
307{
308 struct dst_entry *dst = __sk_dst_check(sk, 0);
309
310 if (dst)
311 dst->ops->redirect(dst, sk, skb);
312}
313
314
315/* handle ICMP messages on TCP_NEW_SYN_RECV request sockets */
316void tcp_req_err(struct sock *sk, u32 seq, bool abort)
317{
318 struct request_sock *req = inet_reqsk(sk);
319 struct net *net = sock_net(sk);
320
321 /* ICMPs are not backlogged, hence we cannot get
322 * an established socket here.
323 */
324 if (seq != tcp_rsk(req)->snt_isn) {
325 __NET_INC_STATS(net, LINUX_MIB_OUTOFWINDOWICMPS);
326 } else if (abort) {
327 /*
328 * Still in SYN_RECV, just remove it silently.
329 * There is no good way to pass the error to the newly
330 * created socket, and POSIX does not want network
331 * errors returned from accept().
332 */
333 inet_csk_reqsk_queue_drop(req->rsk_listener, req);
334 tcp_listendrop(req->rsk_listener);
335 }
336 reqsk_put(req);
337}
338EXPORT_SYMBOL(tcp_req_err);
339
340/*
341 * This routine is called by the ICMP module when it gets some
342 * sort of error condition. If err < 0 then the socket should
343 * be closed and the error returned to the user. If err > 0
344 * it's just the icmp type << 8 | icmp code. After adjustment
345 * header points to the first 8 bytes of the tcp header. We need
346 * to find the appropriate port.
347 *
348 * The locking strategy used here is very "optimistic". When
349 * someone else accesses the socket the ICMP is just dropped
350 * and for some paths there is no check at all.
351 * A more general error queue to queue errors for later handling
352 * is probably better.
353 *
354 */
355
356void tcp_v4_err(struct sk_buff *icmp_skb, u32 info)
357{
358 const struct iphdr *iph = (const struct iphdr *)icmp_skb->data;
359 struct tcphdr *th = (struct tcphdr *)(icmp_skb->data + (iph->ihl << 2));
360 struct inet_connection_sock *icsk;
361 struct tcp_sock *tp;
362 struct inet_sock *inet;
363 const int type = icmp_hdr(icmp_skb)->type;
364 const int code = icmp_hdr(icmp_skb)->code;
365 struct sock *sk;
366 struct sk_buff *skb;
367 struct request_sock *fastopen;
368 __u32 seq, snd_una;
369 __u32 remaining;
370 int err;
371 struct net *net = dev_net(icmp_skb->dev);
372
373 sk = __inet_lookup_established(net, &tcp_hashinfo, iph->daddr,
374 th->dest, iph->saddr, ntohs(th->source),
375 inet_iif(icmp_skb));
376 if (!sk) {
377 __ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
378 return;
379 }
380 if (sk->sk_state == TCP_TIME_WAIT) {
381 inet_twsk_put(inet_twsk(sk));
382 return;
383 }
384 seq = ntohl(th->seq);
385 if (sk->sk_state == TCP_NEW_SYN_RECV)
386 return tcp_req_err(sk, seq,
387 type == ICMP_PARAMETERPROB ||
388 type == ICMP_TIME_EXCEEDED ||
389 (type == ICMP_DEST_UNREACH &&
390 (code == ICMP_NET_UNREACH ||
391 code == ICMP_HOST_UNREACH)));
392
393 bh_lock_sock(sk);
394 /* If too many ICMPs get dropped on busy
395 * servers this needs to be solved differently.
396 * We do take care of PMTU discovery (RFC1191) special case :
397 * we can receive locally generated ICMP messages while socket is held.
398 */
399 if (sock_owned_by_user(sk)) {
400 if (!(type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED))
401 __NET_INC_STATS(net, LINUX_MIB_LOCKDROPPEDICMPS);
402 }
403 if (sk->sk_state == TCP_CLOSE)
404 goto out;
405
406 if (unlikely(iph->ttl < inet_sk(sk)->min_ttl)) {
407 __NET_INC_STATS(net, LINUX_MIB_TCPMINTTLDROP);
408 goto out;
409 }
410
411 icsk = inet_csk(sk);
412 tp = tcp_sk(sk);
413 /* XXX (TFO) - tp->snd_una should be ISN (tcp_create_openreq_child() */
414 fastopen = tp->fastopen_rsk;
415 snd_una = fastopen ? tcp_rsk(fastopen)->snt_isn : tp->snd_una;
416 if (sk->sk_state != TCP_LISTEN &&
417 !between(seq, snd_una, tp->snd_nxt)) {
418 __NET_INC_STATS(net, LINUX_MIB_OUTOFWINDOWICMPS);
419 goto out;
420 }
421
422 switch (type) {
423 case ICMP_REDIRECT:
424 if (!sock_owned_by_user(sk))
425 do_redirect(icmp_skb, sk);
426 goto out;
427 case ICMP_SOURCE_QUENCH:
428 /* Just silently ignore these. */
429 goto out;
430 case ICMP_PARAMETERPROB:
431 err = EPROTO;
432 break;
433 case ICMP_DEST_UNREACH:
434 if (code > NR_ICMP_UNREACH)
435 goto out;
436
437 if (code == ICMP_FRAG_NEEDED) { /* PMTU discovery (RFC1191) */
438 /* We are not interested in TCP_LISTEN and open_requests
439 * (SYN-ACKs send out by Linux are always <576bytes so
440 * they should go through unfragmented).
441 */
442 if (sk->sk_state == TCP_LISTEN)
443 goto out;
444
445 tp->mtu_info = info;
446 if (!sock_owned_by_user(sk)) {
447 tcp_v4_mtu_reduced(sk);
448 } else {
449 if (!test_and_set_bit(TCP_MTU_REDUCED_DEFERRED, &sk->sk_tsq_flags))
450 sock_hold(sk);
451 }
452 goto out;
453 }
454
455 err = icmp_err_convert[code].errno;
456 /* check if icmp_skb allows revert of backoff
457 * (see draft-zimmermann-tcp-lcd) */
458 if (code != ICMP_NET_UNREACH && code != ICMP_HOST_UNREACH)
459 break;
460 if (seq != tp->snd_una || !icsk->icsk_retransmits ||
461 !icsk->icsk_backoff || fastopen)
462 break;
463
464 if (sock_owned_by_user(sk))
465 break;
466
467 icsk->icsk_backoff--;
468 icsk->icsk_rto = tp->srtt_us ? __tcp_set_rto(tp) :
469 TCP_TIMEOUT_INIT;
470 icsk->icsk_rto = inet_csk_rto_backoff(icsk, TCP_RTO_MAX);
471
472 skb = tcp_write_queue_head(sk);
473 BUG_ON(!skb);
474
475 remaining = icsk->icsk_rto -
476 min(icsk->icsk_rto,
477 tcp_time_stamp - tcp_skb_timestamp(skb));
478
479 if (remaining) {
480 inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
481 remaining, TCP_RTO_MAX);
482 } else {
483 /* RTO revert clocked out retransmission.
484 * Will retransmit now */
485 tcp_retransmit_timer(sk);
486 }
487
488 break;
489 case ICMP_TIME_EXCEEDED:
490 err = EHOSTUNREACH;
491 break;
492 default:
493 goto out;
494 }
495
496 switch (sk->sk_state) {
497 case TCP_SYN_SENT:
498 case TCP_SYN_RECV:
499 /* Only in fast or simultaneous open. If a fast open socket is
500 * is already accepted it is treated as a connected one below.
501 */
502 if (fastopen && !fastopen->sk)
503 break;
504
505 if (!sock_owned_by_user(sk)) {
506 sk->sk_err = err;
507
508 sk->sk_error_report(sk);
509
510 tcp_done(sk);
511 } else {
512 sk->sk_err_soft = err;
513 }
514 goto out;
515 }
516
517 /* If we've already connected we will keep trying
518 * until we time out, or the user gives up.
519 *
520 * rfc1122 4.2.3.9 allows to consider as hard errors
521 * only PROTO_UNREACH and PORT_UNREACH (well, FRAG_FAILED too,
522 * but it is obsoleted by pmtu discovery).
523 *
524 * Note, that in modern internet, where routing is unreliable
525 * and in each dark corner broken firewalls sit, sending random
526 * errors ordered by their masters even this two messages finally lose
527 * their original sense (even Linux sends invalid PORT_UNREACHs)
528 *
529 * Now we are in compliance with RFCs.
530 * --ANK (980905)
531 */
532
533 inet = inet_sk(sk);
534 if (!sock_owned_by_user(sk) && inet->recverr) {
535 sk->sk_err = err;
536 sk->sk_error_report(sk);
537 } else { /* Only an error on timeout */
538 sk->sk_err_soft = err;
539 }
540
541out:
542 bh_unlock_sock(sk);
543 sock_put(sk);
544}
545
546void __tcp_v4_send_check(struct sk_buff *skb, __be32 saddr, __be32 daddr)
547{
548 struct tcphdr *th = tcp_hdr(skb);
549
550 if (skb->ip_summed == CHECKSUM_PARTIAL) {
551 th->check = ~tcp_v4_check(skb->len, saddr, daddr, 0);
552 skb->csum_start = skb_transport_header(skb) - skb->head;
553 skb->csum_offset = offsetof(struct tcphdr, check);
554 } else {
555 th->check = tcp_v4_check(skb->len, saddr, daddr,
556 csum_partial(th,
557 th->doff << 2,
558 skb->csum));
559 }
560}
561
562/* This routine computes an IPv4 TCP checksum. */
563void tcp_v4_send_check(struct sock *sk, struct sk_buff *skb)
564{
565 const struct inet_sock *inet = inet_sk(sk);
566
567 __tcp_v4_send_check(skb, inet->inet_saddr, inet->inet_daddr);
568}
569EXPORT_SYMBOL(tcp_v4_send_check);
570
571/*
572 * This routine will send an RST to the other tcp.
573 *
574 * Someone asks: why I NEVER use socket parameters (TOS, TTL etc.)
575 * for reset.
576 * Answer: if a packet caused RST, it is not for a socket
577 * existing in our system, if it is matched to a socket,
578 * it is just duplicate segment or bug in other side's TCP.
579 * So that we build reply only basing on parameters
580 * arrived with segment.
581 * Exception: precedence violation. We do not implement it in any case.
582 */
583
584static void tcp_v4_send_reset(const struct sock *sk, struct sk_buff *skb)
585{
586 const struct tcphdr *th = tcp_hdr(skb);
587 struct {
588 struct tcphdr th;
589#ifdef CONFIG_TCP_MD5SIG
590 __be32 opt[(TCPOLEN_MD5SIG_ALIGNED >> 2)];
591#endif
592 } rep;
593 struct ip_reply_arg arg;
594#ifdef CONFIG_TCP_MD5SIG
595 struct tcp_md5sig_key *key = NULL;
596 const __u8 *hash_location = NULL;
597 unsigned char newhash[16];
598 int genhash;
599 struct sock *sk1 = NULL;
600#endif
601 struct net *net;
602
603 /* Never send a reset in response to a reset. */
604 if (th->rst)
605 return;
606
607 /* If sk not NULL, it means we did a successful lookup and incoming
608 * route had to be correct. prequeue might have dropped our dst.
609 */
610 if (!sk && skb_rtable(skb)->rt_type != RTN_LOCAL)
611 return;
612
613 /* Swap the send and the receive. */
614 memset(&rep, 0, sizeof(rep));
615 rep.th.dest = th->source;
616 rep.th.source = th->dest;
617 rep.th.doff = sizeof(struct tcphdr) / 4;
618 rep.th.rst = 1;
619
620 if (th->ack) {
621 rep.th.seq = th->ack_seq;
622 } else {
623 rep.th.ack = 1;
624 rep.th.ack_seq = htonl(ntohl(th->seq) + th->syn + th->fin +
625 skb->len - (th->doff << 2));
626 }
627
628 memset(&arg, 0, sizeof(arg));
629 arg.iov[0].iov_base = (unsigned char *)&rep;
630 arg.iov[0].iov_len = sizeof(rep.th);
631
632 net = sk ? sock_net(sk) : dev_net(skb_dst(skb)->dev);
633#ifdef CONFIG_TCP_MD5SIG
634 rcu_read_lock();
635 hash_location = tcp_parse_md5sig_option(th);
636 if (sk && sk_fullsock(sk)) {
637 key = tcp_md5_do_lookup(sk, (union tcp_md5_addr *)
638 &ip_hdr(skb)->saddr, AF_INET);
639 } else if (hash_location) {
640 /*
641 * active side is lost. Try to find listening socket through
642 * source port, and then find md5 key through listening socket.
643 * we are not loose security here:
644 * Incoming packet is checked with md5 hash with finding key,
645 * no RST generated if md5 hash doesn't match.
646 */
647 sk1 = __inet_lookup_listener(net, &tcp_hashinfo, NULL, 0,
648 ip_hdr(skb)->saddr,
649 th->source, ip_hdr(skb)->daddr,
650 ntohs(th->source), inet_iif(skb));
651 /* don't send rst if it can't find key */
652 if (!sk1)
653 goto out;
654
655 key = tcp_md5_do_lookup(sk1, (union tcp_md5_addr *)
656 &ip_hdr(skb)->saddr, AF_INET);
657 if (!key)
658 goto out;
659
660
661 genhash = tcp_v4_md5_hash_skb(newhash, key, NULL, skb);
662 if (genhash || memcmp(hash_location, newhash, 16) != 0)
663 goto out;
664
665 }
666
667 if (key) {
668 rep.opt[0] = htonl((TCPOPT_NOP << 24) |
669 (TCPOPT_NOP << 16) |
670 (TCPOPT_MD5SIG << 8) |
671 TCPOLEN_MD5SIG);
672 /* Update length and the length the header thinks exists */
673 arg.iov[0].iov_len += TCPOLEN_MD5SIG_ALIGNED;
674 rep.th.doff = arg.iov[0].iov_len / 4;
675
676 tcp_v4_md5_hash_hdr((__u8 *) &rep.opt[1],
677 key, ip_hdr(skb)->saddr,
678 ip_hdr(skb)->daddr, &rep.th);
679 }
680#endif
681 arg.csum = csum_tcpudp_nofold(ip_hdr(skb)->daddr,
682 ip_hdr(skb)->saddr, /* XXX */
683 arg.iov[0].iov_len, IPPROTO_TCP, 0);
684 arg.csumoffset = offsetof(struct tcphdr, check) / 2;
685 arg.flags = (sk && inet_sk_transparent(sk)) ? IP_REPLY_ARG_NOSRCCHECK : 0;
686
687 /* When socket is gone, all binding information is lost.
688 * routing might fail in this case. No choice here, if we choose to force
689 * input interface, we will misroute in case of asymmetric route.
690 */
691 if (sk)
692 arg.bound_dev_if = sk->sk_bound_dev_if;
693
694 BUILD_BUG_ON(offsetof(struct sock, sk_bound_dev_if) !=
695 offsetof(struct inet_timewait_sock, tw_bound_dev_if));
696
697 arg.tos = ip_hdr(skb)->tos;
698 arg.uid = sock_net_uid(net, sk && sk_fullsock(sk) ? sk : NULL);
699 local_bh_disable();
700 ip_send_unicast_reply(*this_cpu_ptr(net->ipv4.tcp_sk),
701 skb, &TCP_SKB_CB(skb)->header.h4.opt,
702 ip_hdr(skb)->saddr, ip_hdr(skb)->daddr,
703 &arg, arg.iov[0].iov_len);
704
705 __TCP_INC_STATS(net, TCP_MIB_OUTSEGS);
706 __TCP_INC_STATS(net, TCP_MIB_OUTRSTS);
707 local_bh_enable();
708
709#ifdef CONFIG_TCP_MD5SIG
710out:
711 rcu_read_unlock();
712#endif
713}
714
715/* The code following below sending ACKs in SYN-RECV and TIME-WAIT states
716 outside socket context is ugly, certainly. What can I do?
717 */
718
719static void tcp_v4_send_ack(const struct sock *sk,
720 struct sk_buff *skb, u32 seq, u32 ack,
721 u32 win, u32 tsval, u32 tsecr, int oif,
722 struct tcp_md5sig_key *key,
723 int reply_flags, u8 tos)
724{
725 const struct tcphdr *th = tcp_hdr(skb);
726 struct {
727 struct tcphdr th;
728 __be32 opt[(TCPOLEN_TSTAMP_ALIGNED >> 2)
729#ifdef CONFIG_TCP_MD5SIG
730 + (TCPOLEN_MD5SIG_ALIGNED >> 2)
731#endif
732 ];
733 } rep;
734 struct net *net = sock_net(sk);
735 struct ip_reply_arg arg;
736
737 memset(&rep.th, 0, sizeof(struct tcphdr));
738 memset(&arg, 0, sizeof(arg));
739
740 arg.iov[0].iov_base = (unsigned char *)&rep;
741 arg.iov[0].iov_len = sizeof(rep.th);
742 if (tsecr) {
743 rep.opt[0] = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
744 (TCPOPT_TIMESTAMP << 8) |
745 TCPOLEN_TIMESTAMP);
746 rep.opt[1] = htonl(tsval);
747 rep.opt[2] = htonl(tsecr);
748 arg.iov[0].iov_len += TCPOLEN_TSTAMP_ALIGNED;
749 }
750
751 /* Swap the send and the receive. */
752 rep.th.dest = th->source;
753 rep.th.source = th->dest;
754 rep.th.doff = arg.iov[0].iov_len / 4;
755 rep.th.seq = htonl(seq);
756 rep.th.ack_seq = htonl(ack);
757 rep.th.ack = 1;
758 rep.th.window = htons(win);
759
760#ifdef CONFIG_TCP_MD5SIG
761 if (key) {
762 int offset = (tsecr) ? 3 : 0;
763
764 rep.opt[offset++] = htonl((TCPOPT_NOP << 24) |
765 (TCPOPT_NOP << 16) |
766 (TCPOPT_MD5SIG << 8) |
767 TCPOLEN_MD5SIG);
768 arg.iov[0].iov_len += TCPOLEN_MD5SIG_ALIGNED;
769 rep.th.doff = arg.iov[0].iov_len/4;
770
771 tcp_v4_md5_hash_hdr((__u8 *) &rep.opt[offset],
772 key, ip_hdr(skb)->saddr,
773 ip_hdr(skb)->daddr, &rep.th);
774 }
775#endif
776 arg.flags = reply_flags;
777 arg.csum = csum_tcpudp_nofold(ip_hdr(skb)->daddr,
778 ip_hdr(skb)->saddr, /* XXX */
779 arg.iov[0].iov_len, IPPROTO_TCP, 0);
780 arg.csumoffset = offsetof(struct tcphdr, check) / 2;
781 if (oif)
782 arg.bound_dev_if = oif;
783 arg.tos = tos;
784 arg.uid = sock_net_uid(net, sk_fullsock(sk) ? sk : NULL);
785 local_bh_disable();
786 ip_send_unicast_reply(*this_cpu_ptr(net->ipv4.tcp_sk),
787 skb, &TCP_SKB_CB(skb)->header.h4.opt,
788 ip_hdr(skb)->saddr, ip_hdr(skb)->daddr,
789 &arg, arg.iov[0].iov_len);
790
791 __TCP_INC_STATS(net, TCP_MIB_OUTSEGS);
792 local_bh_enable();
793}
794
795static void tcp_v4_timewait_ack(struct sock *sk, struct sk_buff *skb)
796{
797 struct inet_timewait_sock *tw = inet_twsk(sk);
798 struct tcp_timewait_sock *tcptw = tcp_twsk(sk);
799
800 tcp_v4_send_ack(sk, skb,
801 tcptw->tw_snd_nxt, tcptw->tw_rcv_nxt,
802 tcptw->tw_rcv_wnd >> tw->tw_rcv_wscale,
803 tcp_time_stamp + tcptw->tw_ts_offset,
804 tcptw->tw_ts_recent,
805 tw->tw_bound_dev_if,
806 tcp_twsk_md5_key(tcptw),
807 tw->tw_transparent ? IP_REPLY_ARG_NOSRCCHECK : 0,
808 tw->tw_tos
809 );
810
811 inet_twsk_put(tw);
812}
813
814static void tcp_v4_reqsk_send_ack(const struct sock *sk, struct sk_buff *skb,
815 struct request_sock *req)
816{
817 /* sk->sk_state == TCP_LISTEN -> for regular TCP_SYN_RECV
818 * sk->sk_state == TCP_SYN_RECV -> for Fast Open.
819 */
820 u32 seq = (sk->sk_state == TCP_LISTEN) ? tcp_rsk(req)->snt_isn + 1 :
821 tcp_sk(sk)->snd_nxt;
822
823 /* RFC 7323 2.3
824 * The window field (SEG.WND) of every outgoing segment, with the
825 * exception of <SYN> segments, MUST be right-shifted by
826 * Rcv.Wind.Shift bits:
827 */
828 tcp_v4_send_ack(sk, skb, seq,
829 tcp_rsk(req)->rcv_nxt,
830 req->rsk_rcv_wnd >> inet_rsk(req)->rcv_wscale,
831 tcp_time_stamp + tcp_rsk(req)->ts_off,
832 req->ts_recent,
833 0,
834 tcp_md5_do_lookup(sk, (union tcp_md5_addr *)&ip_hdr(skb)->daddr,
835 AF_INET),
836 inet_rsk(req)->no_srccheck ? IP_REPLY_ARG_NOSRCCHECK : 0,
837 ip_hdr(skb)->tos);
838}
839
840/*
841 * Send a SYN-ACK after having received a SYN.
842 * This still operates on a request_sock only, not on a big
843 * socket.
844 */
845static int tcp_v4_send_synack(const struct sock *sk, struct dst_entry *dst,
846 struct flowi *fl,
847 struct request_sock *req,
848 struct tcp_fastopen_cookie *foc,
849 enum tcp_synack_type synack_type)
850{
851 const struct inet_request_sock *ireq = inet_rsk(req);
852 struct flowi4 fl4;
853 int err = -1;
854 struct sk_buff *skb;
855
856 /* First, grab a route. */
857 if (!dst && (dst = inet_csk_route_req(sk, &fl4, req)) == NULL)
858 return -1;
859
860 skb = tcp_make_synack(sk, dst, req, foc, synack_type);
861
862 if (skb) {
863 __tcp_v4_send_check(skb, ireq->ir_loc_addr, ireq->ir_rmt_addr);
864
865 err = ip_build_and_send_pkt(skb, sk, ireq->ir_loc_addr,
866 ireq->ir_rmt_addr,
867 ireq->opt);
868 err = net_xmit_eval(err);
869 }
870
871 return err;
872}
873
874/*
875 * IPv4 request_sock destructor.
876 */
877static void tcp_v4_reqsk_destructor(struct request_sock *req)
878{
879 kfree(inet_rsk(req)->opt);
880}
881
882#ifdef CONFIG_TCP_MD5SIG
883/*
884 * RFC2385 MD5 checksumming requires a mapping of
885 * IP address->MD5 Key.
886 * We need to maintain these in the sk structure.
887 */
888
889/* Find the Key structure for an address. */
890struct tcp_md5sig_key *tcp_md5_do_lookup(const struct sock *sk,
891 const union tcp_md5_addr *addr,
892 int family)
893{
894 const struct tcp_sock *tp = tcp_sk(sk);
895 struct tcp_md5sig_key *key;
896 unsigned int size = sizeof(struct in_addr);
897 const struct tcp_md5sig_info *md5sig;
898
899 /* caller either holds rcu_read_lock() or socket lock */
900 md5sig = rcu_dereference_check(tp->md5sig_info,
901 lockdep_sock_is_held(sk));
902 if (!md5sig)
903 return NULL;
904#if IS_ENABLED(CONFIG_IPV6)
905 if (family == AF_INET6)
906 size = sizeof(struct in6_addr);
907#endif
908 hlist_for_each_entry_rcu(key, &md5sig->head, node) {
909 if (key->family != family)
910 continue;
911 if (!memcmp(&key->addr, addr, size))
912 return key;
913 }
914 return NULL;
915}
916EXPORT_SYMBOL(tcp_md5_do_lookup);
917
918struct tcp_md5sig_key *tcp_v4_md5_lookup(const struct sock *sk,
919 const struct sock *addr_sk)
920{
921 const union tcp_md5_addr *addr;
922
923 addr = (const union tcp_md5_addr *)&addr_sk->sk_daddr;
924 return tcp_md5_do_lookup(sk, addr, AF_INET);
925}
926EXPORT_SYMBOL(tcp_v4_md5_lookup);
927
928/* This can be called on a newly created socket, from other files */
929int tcp_md5_do_add(struct sock *sk, const union tcp_md5_addr *addr,
930 int family, const u8 *newkey, u8 newkeylen, gfp_t gfp)
931{
932 /* Add Key to the list */
933 struct tcp_md5sig_key *key;
934 struct tcp_sock *tp = tcp_sk(sk);
935 struct tcp_md5sig_info *md5sig;
936
937 key = tcp_md5_do_lookup(sk, addr, family);
938 if (key) {
939 /* Pre-existing entry - just update that one. */
940 memcpy(key->key, newkey, newkeylen);
941 key->keylen = newkeylen;
942 return 0;
943 }
944
945 md5sig = rcu_dereference_protected(tp->md5sig_info,
946 lockdep_sock_is_held(sk));
947 if (!md5sig) {
948 md5sig = kmalloc(sizeof(*md5sig), gfp);
949 if (!md5sig)
950 return -ENOMEM;
951
952 sk_nocaps_add(sk, NETIF_F_GSO_MASK);
953 INIT_HLIST_HEAD(&md5sig->head);
954 rcu_assign_pointer(tp->md5sig_info, md5sig);
955 }
956
957 key = sock_kmalloc(sk, sizeof(*key), gfp);
958 if (!key)
959 return -ENOMEM;
960 if (!tcp_alloc_md5sig_pool()) {
961 sock_kfree_s(sk, key, sizeof(*key));
962 return -ENOMEM;
963 }
964
965 memcpy(key->key, newkey, newkeylen);
966 key->keylen = newkeylen;
967 key->family = family;
968 memcpy(&key->addr, addr,
969 (family == AF_INET6) ? sizeof(struct in6_addr) :
970 sizeof(struct in_addr));
971 hlist_add_head_rcu(&key->node, &md5sig->head);
972 return 0;
973}
974EXPORT_SYMBOL(tcp_md5_do_add);
975
976int tcp_md5_do_del(struct sock *sk, const union tcp_md5_addr *addr, int family)
977{
978 struct tcp_md5sig_key *key;
979
980 key = tcp_md5_do_lookup(sk, addr, family);
981 if (!key)
982 return -ENOENT;
983 hlist_del_rcu(&key->node);
984 atomic_sub(sizeof(*key), &sk->sk_omem_alloc);
985 kfree_rcu(key, rcu);
986 return 0;
987}
988EXPORT_SYMBOL(tcp_md5_do_del);
989
990static void tcp_clear_md5_list(struct sock *sk)
991{
992 struct tcp_sock *tp = tcp_sk(sk);
993 struct tcp_md5sig_key *key;
994 struct hlist_node *n;
995 struct tcp_md5sig_info *md5sig;
996
997 md5sig = rcu_dereference_protected(tp->md5sig_info, 1);
998
999 hlist_for_each_entry_safe(key, n, &md5sig->head, node) {
1000 hlist_del_rcu(&key->node);
1001 atomic_sub(sizeof(*key), &sk->sk_omem_alloc);
1002 kfree_rcu(key, rcu);
1003 }
1004}
1005
1006static int tcp_v4_parse_md5_keys(struct sock *sk, char __user *optval,
1007 int optlen)
1008{
1009 struct tcp_md5sig cmd;
1010 struct sockaddr_in *sin = (struct sockaddr_in *)&cmd.tcpm_addr;
1011
1012 if (optlen < sizeof(cmd))
1013 return -EINVAL;
1014
1015 if (copy_from_user(&cmd, optval, sizeof(cmd)))
1016 return -EFAULT;
1017
1018 if (sin->sin_family != AF_INET)
1019 return -EINVAL;
1020
1021 if (!cmd.tcpm_keylen)
1022 return tcp_md5_do_del(sk, (union tcp_md5_addr *)&sin->sin_addr.s_addr,
1023 AF_INET);
1024
1025 if (cmd.tcpm_keylen > TCP_MD5SIG_MAXKEYLEN)
1026 return -EINVAL;
1027
1028 return tcp_md5_do_add(sk, (union tcp_md5_addr *)&sin->sin_addr.s_addr,
1029 AF_INET, cmd.tcpm_key, cmd.tcpm_keylen,
1030 GFP_KERNEL);
1031}
1032
1033static int tcp_v4_md5_hash_headers(struct tcp_md5sig_pool *hp,
1034 __be32 daddr, __be32 saddr,
1035 const struct tcphdr *th, int nbytes)
1036{
1037 struct tcp4_pseudohdr *bp;
1038 struct scatterlist sg;
1039 struct tcphdr *_th;
1040
1041 bp = hp->scratch;
1042 bp->saddr = saddr;
1043 bp->daddr = daddr;
1044 bp->pad = 0;
1045 bp->protocol = IPPROTO_TCP;
1046 bp->len = cpu_to_be16(nbytes);
1047
1048 _th = (struct tcphdr *)(bp + 1);
1049 memcpy(_th, th, sizeof(*th));
1050 _th->check = 0;
1051
1052 sg_init_one(&sg, bp, sizeof(*bp) + sizeof(*th));
1053 ahash_request_set_crypt(hp->md5_req, &sg, NULL,
1054 sizeof(*bp) + sizeof(*th));
1055 return crypto_ahash_update(hp->md5_req);
1056}
1057
1058static int tcp_v4_md5_hash_hdr(char *md5_hash, const struct tcp_md5sig_key *key,
1059 __be32 daddr, __be32 saddr, const struct tcphdr *th)
1060{
1061 struct tcp_md5sig_pool *hp;
1062 struct ahash_request *req;
1063
1064 hp = tcp_get_md5sig_pool();
1065 if (!hp)
1066 goto clear_hash_noput;
1067 req = hp->md5_req;
1068
1069 if (crypto_ahash_init(req))
1070 goto clear_hash;
1071 if (tcp_v4_md5_hash_headers(hp, daddr, saddr, th, th->doff << 2))
1072 goto clear_hash;
1073 if (tcp_md5_hash_key(hp, key))
1074 goto clear_hash;
1075 ahash_request_set_crypt(req, NULL, md5_hash, 0);
1076 if (crypto_ahash_final(req))
1077 goto clear_hash;
1078
1079 tcp_put_md5sig_pool();
1080 return 0;
1081
1082clear_hash:
1083 tcp_put_md5sig_pool();
1084clear_hash_noput:
1085 memset(md5_hash, 0, 16);
1086 return 1;
1087}
1088
1089int tcp_v4_md5_hash_skb(char *md5_hash, const struct tcp_md5sig_key *key,
1090 const struct sock *sk,
1091 const struct sk_buff *skb)
1092{
1093 struct tcp_md5sig_pool *hp;
1094 struct ahash_request *req;
1095 const struct tcphdr *th = tcp_hdr(skb);
1096 __be32 saddr, daddr;
1097
1098 if (sk) { /* valid for establish/request sockets */
1099 saddr = sk->sk_rcv_saddr;
1100 daddr = sk->sk_daddr;
1101 } else {
1102 const struct iphdr *iph = ip_hdr(skb);
1103 saddr = iph->saddr;
1104 daddr = iph->daddr;
1105 }
1106
1107 hp = tcp_get_md5sig_pool();
1108 if (!hp)
1109 goto clear_hash_noput;
1110 req = hp->md5_req;
1111
1112 if (crypto_ahash_init(req))
1113 goto clear_hash;
1114
1115 if (tcp_v4_md5_hash_headers(hp, daddr, saddr, th, skb->len))
1116 goto clear_hash;
1117 if (tcp_md5_hash_skb_data(hp, skb, th->doff << 2))
1118 goto clear_hash;
1119 if (tcp_md5_hash_key(hp, key))
1120 goto clear_hash;
1121 ahash_request_set_crypt(req, NULL, md5_hash, 0);
1122 if (crypto_ahash_final(req))
1123 goto clear_hash;
1124
1125 tcp_put_md5sig_pool();
1126 return 0;
1127
1128clear_hash:
1129 tcp_put_md5sig_pool();
1130clear_hash_noput:
1131 memset(md5_hash, 0, 16);
1132 return 1;
1133}
1134EXPORT_SYMBOL(tcp_v4_md5_hash_skb);
1135
1136#endif
1137
1138/* Called with rcu_read_lock() */
1139static bool tcp_v4_inbound_md5_hash(const struct sock *sk,
1140 const struct sk_buff *skb)
1141{
1142#ifdef CONFIG_TCP_MD5SIG
1143 /*
1144 * This gets called for each TCP segment that arrives
1145 * so we want to be efficient.
1146 * We have 3 drop cases:
1147 * o No MD5 hash and one expected.
1148 * o MD5 hash and we're not expecting one.
1149 * o MD5 hash and its wrong.
1150 */
1151 const __u8 *hash_location = NULL;
1152 struct tcp_md5sig_key *hash_expected;
1153 const struct iphdr *iph = ip_hdr(skb);
1154 const struct tcphdr *th = tcp_hdr(skb);
1155 int genhash;
1156 unsigned char newhash[16];
1157
1158 hash_expected = tcp_md5_do_lookup(sk, (union tcp_md5_addr *)&iph->saddr,
1159 AF_INET);
1160 hash_location = tcp_parse_md5sig_option(th);
1161
1162 /* We've parsed the options - do we have a hash? */
1163 if (!hash_expected && !hash_location)
1164 return false;
1165
1166 if (hash_expected && !hash_location) {
1167 NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPMD5NOTFOUND);
1168 return true;
1169 }
1170
1171 if (!hash_expected && hash_location) {
1172 NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPMD5UNEXPECTED);
1173 return true;
1174 }
1175
1176 /* Okay, so this is hash_expected and hash_location -
1177 * so we need to calculate the checksum.
1178 */
1179 genhash = tcp_v4_md5_hash_skb(newhash,
1180 hash_expected,
1181 NULL, skb);
1182
1183 if (genhash || memcmp(hash_location, newhash, 16) != 0) {
1184 NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPMD5FAILURE);
1185 net_info_ratelimited("MD5 Hash failed for (%pI4, %d)->(%pI4, %d)%s\n",
1186 &iph->saddr, ntohs(th->source),
1187 &iph->daddr, ntohs(th->dest),
1188 genhash ? " tcp_v4_calc_md5_hash failed"
1189 : "");
1190 return true;
1191 }
1192 return false;
1193#endif
1194 return false;
1195}
1196
1197static void tcp_v4_init_req(struct request_sock *req,
1198 const struct sock *sk_listener,
1199 struct sk_buff *skb)
1200{
1201 struct inet_request_sock *ireq = inet_rsk(req);
1202
1203 sk_rcv_saddr_set(req_to_sk(req), ip_hdr(skb)->daddr);
1204 sk_daddr_set(req_to_sk(req), ip_hdr(skb)->saddr);
1205 ireq->opt = tcp_v4_save_options(skb);
1206}
1207
1208static struct dst_entry *tcp_v4_route_req(const struct sock *sk,
1209 struct flowi *fl,
1210 const struct request_sock *req,
1211 bool *strict)
1212{
1213 struct dst_entry *dst = inet_csk_route_req(sk, &fl->u.ip4, req);
1214
1215 if (strict) {
1216 if (fl->u.ip4.daddr == inet_rsk(req)->ir_rmt_addr)
1217 *strict = true;
1218 else
1219 *strict = false;
1220 }
1221
1222 return dst;
1223}
1224
1225struct request_sock_ops tcp_request_sock_ops __read_mostly = {
1226 .family = PF_INET,
1227 .obj_size = sizeof(struct tcp_request_sock),
1228 .rtx_syn_ack = tcp_rtx_synack,
1229 .send_ack = tcp_v4_reqsk_send_ack,
1230 .destructor = tcp_v4_reqsk_destructor,
1231 .send_reset = tcp_v4_send_reset,
1232 .syn_ack_timeout = tcp_syn_ack_timeout,
1233};
1234
1235static const struct tcp_request_sock_ops tcp_request_sock_ipv4_ops = {
1236 .mss_clamp = TCP_MSS_DEFAULT,
1237#ifdef CONFIG_TCP_MD5SIG
1238 .req_md5_lookup = tcp_v4_md5_lookup,
1239 .calc_md5_hash = tcp_v4_md5_hash_skb,
1240#endif
1241 .init_req = tcp_v4_init_req,
1242#ifdef CONFIG_SYN_COOKIES
1243 .cookie_init_seq = cookie_v4_init_sequence,
1244#endif
1245 .route_req = tcp_v4_route_req,
1246 .init_seq = tcp_v4_init_sequence,
1247 .send_synack = tcp_v4_send_synack,
1248};
1249
1250int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb)
1251{
1252 /* Never answer to SYNs send to broadcast or multicast */
1253 if (skb_rtable(skb)->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST))
1254 goto drop;
1255
1256 return tcp_conn_request(&tcp_request_sock_ops,
1257 &tcp_request_sock_ipv4_ops, sk, skb);
1258
1259drop:
1260 tcp_listendrop(sk);
1261 return 0;
1262}
1263EXPORT_SYMBOL(tcp_v4_conn_request);
1264
1265
1266/*
1267 * The three way handshake has completed - we got a valid synack -
1268 * now create the new socket.
1269 */
1270struct sock *tcp_v4_syn_recv_sock(const struct sock *sk, struct sk_buff *skb,
1271 struct request_sock *req,
1272 struct dst_entry *dst,
1273 struct request_sock *req_unhash,
1274 bool *own_req)
1275{
1276 struct inet_request_sock *ireq;
1277 struct inet_sock *newinet;
1278 struct tcp_sock *newtp;
1279 struct sock *newsk;
1280#ifdef CONFIG_TCP_MD5SIG
1281 struct tcp_md5sig_key *key;
1282#endif
1283 struct ip_options_rcu *inet_opt;
1284
1285 if (sk_acceptq_is_full(sk))
1286 goto exit_overflow;
1287
1288 newsk = tcp_create_openreq_child(sk, req, skb);
1289 if (!newsk)
1290 goto exit_nonewsk;
1291
1292 newsk->sk_gso_type = SKB_GSO_TCPV4;
1293 inet_sk_rx_dst_set(newsk, skb);
1294
1295 newtp = tcp_sk(newsk);
1296 newinet = inet_sk(newsk);
1297 ireq = inet_rsk(req);
1298 sk_daddr_set(newsk, ireq->ir_rmt_addr);
1299 sk_rcv_saddr_set(newsk, ireq->ir_loc_addr);
1300 newsk->sk_bound_dev_if = ireq->ir_iif;
1301 newinet->inet_saddr = ireq->ir_loc_addr;
1302 inet_opt = ireq->opt;
1303 rcu_assign_pointer(newinet->inet_opt, inet_opt);
1304 ireq->opt = NULL;
1305 newinet->mc_index = inet_iif(skb);
1306 newinet->mc_ttl = ip_hdr(skb)->ttl;
1307 newinet->rcv_tos = ip_hdr(skb)->tos;
1308 inet_csk(newsk)->icsk_ext_hdr_len = 0;
1309 if (inet_opt)
1310 inet_csk(newsk)->icsk_ext_hdr_len = inet_opt->opt.optlen;
1311 newinet->inet_id = newtp->write_seq ^ jiffies;
1312
1313 if (!dst) {
1314 dst = inet_csk_route_child_sock(sk, newsk, req);
1315 if (!dst)
1316 goto put_and_exit;
1317 } else {
1318 /* syncookie case : see end of cookie_v4_check() */
1319 }
1320 sk_setup_caps(newsk, dst);
1321
1322 tcp_ca_openreq_child(newsk, dst);
1323
1324 tcp_sync_mss(newsk, dst_mtu(dst));
1325 newtp->advmss = dst_metric_advmss(dst);
1326 if (tcp_sk(sk)->rx_opt.user_mss &&
1327 tcp_sk(sk)->rx_opt.user_mss < newtp->advmss)
1328 newtp->advmss = tcp_sk(sk)->rx_opt.user_mss;
1329
1330 tcp_initialize_rcv_mss(newsk);
1331
1332#ifdef CONFIG_TCP_MD5SIG
1333 /* Copy over the MD5 key from the original socket */
1334 key = tcp_md5_do_lookup(sk, (union tcp_md5_addr *)&newinet->inet_daddr,
1335 AF_INET);
1336 if (key) {
1337 /*
1338 * We're using one, so create a matching key
1339 * on the newsk structure. If we fail to get
1340 * memory, then we end up not copying the key
1341 * across. Shucks.
1342 */
1343 tcp_md5_do_add(newsk, (union tcp_md5_addr *)&newinet->inet_daddr,
1344 AF_INET, key->key, key->keylen, GFP_ATOMIC);
1345 sk_nocaps_add(newsk, NETIF_F_GSO_MASK);
1346 }
1347#endif
1348
1349 if (__inet_inherit_port(sk, newsk) < 0)
1350 goto put_and_exit;
1351 *own_req = inet_ehash_nolisten(newsk, req_to_sk(req_unhash));
1352 if (*own_req)
1353 tcp_move_syn(newtp, req);
1354
1355 return newsk;
1356
1357exit_overflow:
1358 NET_INC_STATS(sock_net(sk), LINUX_MIB_LISTENOVERFLOWS);
1359exit_nonewsk:
1360 dst_release(dst);
1361exit:
1362 tcp_listendrop(sk);
1363 return NULL;
1364put_and_exit:
1365 inet_csk_prepare_forced_close(newsk);
1366 tcp_done(newsk);
1367 goto exit;
1368}
1369EXPORT_SYMBOL(tcp_v4_syn_recv_sock);
1370
1371static struct sock *tcp_v4_cookie_check(struct sock *sk, struct sk_buff *skb)
1372{
1373#ifdef CONFIG_SYN_COOKIES
1374 const struct tcphdr *th = tcp_hdr(skb);
1375
1376 if (!th->syn)
1377 sk = cookie_v4_check(sk, skb);
1378#endif
1379 return sk;
1380}
1381
1382/* The socket must have it's spinlock held when we get
1383 * here, unless it is a TCP_LISTEN socket.
1384 *
1385 * We have a potential double-lock case here, so even when
1386 * doing backlog processing we use the BH locking scheme.
1387 * This is because we cannot sleep with the original spinlock
1388 * held.
1389 */
1390int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb)
1391{
1392 struct sock *rsk;
1393
1394 if (sk->sk_state == TCP_ESTABLISHED) { /* Fast path */
1395 struct dst_entry *dst = sk->sk_rx_dst;
1396
1397 sock_rps_save_rxhash(sk, skb);
1398 sk_mark_napi_id(sk, skb);
1399 if (dst) {
1400 if (inet_sk(sk)->rx_dst_ifindex != skb->skb_iif ||
1401 !dst->ops->check(dst, 0)) {
1402 dst_release(dst);
1403 sk->sk_rx_dst = NULL;
1404 }
1405 }
1406 tcp_rcv_established(sk, skb, tcp_hdr(skb), skb->len);
1407 return 0;
1408 }
1409
1410 if (tcp_checksum_complete(skb))
1411 goto csum_err;
1412
1413 if (sk->sk_state == TCP_LISTEN) {
1414 struct sock *nsk = tcp_v4_cookie_check(sk, skb);
1415
1416 if (!nsk)
1417 goto discard;
1418 if (nsk != sk) {
1419 sock_rps_save_rxhash(nsk, skb);
1420 sk_mark_napi_id(nsk, skb);
1421 if (tcp_child_process(sk, nsk, skb)) {
1422 rsk = nsk;
1423 goto reset;
1424 }
1425 return 0;
1426 }
1427 } else
1428 sock_rps_save_rxhash(sk, skb);
1429
1430 if (tcp_rcv_state_process(sk, skb)) {
1431 rsk = sk;
1432 goto reset;
1433 }
1434 return 0;
1435
1436reset:
1437 tcp_v4_send_reset(rsk, skb);
1438discard:
1439 kfree_skb(skb);
1440 /* Be careful here. If this function gets more complicated and
1441 * gcc suffers from register pressure on the x86, sk (in %ebx)
1442 * might be destroyed here. This current version compiles correctly,
1443 * but you have been warned.
1444 */
1445 return 0;
1446
1447csum_err:
1448 TCP_INC_STATS(sock_net(sk), TCP_MIB_CSUMERRORS);
1449 TCP_INC_STATS(sock_net(sk), TCP_MIB_INERRS);
1450 goto discard;
1451}
1452EXPORT_SYMBOL(tcp_v4_do_rcv);
1453
1454void tcp_v4_early_demux(struct sk_buff *skb)
1455{
1456 const struct iphdr *iph;
1457 const struct tcphdr *th;
1458 struct sock *sk;
1459
1460 if (skb->pkt_type != PACKET_HOST)
1461 return;
1462
1463 if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct tcphdr)))
1464 return;
1465
1466 iph = ip_hdr(skb);
1467 th = tcp_hdr(skb);
1468
1469 if (th->doff < sizeof(struct tcphdr) / 4)
1470 return;
1471
1472 sk = __inet_lookup_established(dev_net(skb->dev), &tcp_hashinfo,
1473 iph->saddr, th->source,
1474 iph->daddr, ntohs(th->dest),
1475 skb->skb_iif);
1476 if (sk) {
1477 skb->sk = sk;
1478 skb->destructor = sock_edemux;
1479 if (sk_fullsock(sk)) {
1480 struct dst_entry *dst = READ_ONCE(sk->sk_rx_dst);
1481
1482 if (dst)
1483 dst = dst_check(dst, 0);
1484 if (dst &&
1485 inet_sk(sk)->rx_dst_ifindex == skb->skb_iif)
1486 skb_dst_set_noref(skb, dst);
1487 }
1488 }
1489}
1490
1491/* Packet is added to VJ-style prequeue for processing in process
1492 * context, if a reader task is waiting. Apparently, this exciting
1493 * idea (VJ's mail "Re: query about TCP header on tcp-ip" of 07 Sep 93)
1494 * failed somewhere. Latency? Burstiness? Well, at least now we will
1495 * see, why it failed. 8)8) --ANK
1496 *
1497 */
1498bool tcp_prequeue(struct sock *sk, struct sk_buff *skb)
1499{
1500 struct tcp_sock *tp = tcp_sk(sk);
1501
1502 if (sysctl_tcp_low_latency || !tp->ucopy.task)
1503 return false;
1504
1505 if (skb->len <= tcp_hdrlen(skb) &&
1506 skb_queue_len(&tp->ucopy.prequeue) == 0)
1507 return false;
1508
1509 /* Before escaping RCU protected region, we need to take care of skb
1510 * dst. Prequeue is only enabled for established sockets.
1511 * For such sockets, we might need the skb dst only to set sk->sk_rx_dst
1512 * Instead of doing full sk_rx_dst validity here, let's perform
1513 * an optimistic check.
1514 */
1515 if (likely(sk->sk_rx_dst))
1516 skb_dst_drop(skb);
1517 else
1518 skb_dst_force_safe(skb);
1519
1520 __skb_queue_tail(&tp->ucopy.prequeue, skb);
1521 tp->ucopy.memory += skb->truesize;
1522 if (skb_queue_len(&tp->ucopy.prequeue) >= 32 ||
1523 tp->ucopy.memory + atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf) {
1524 struct sk_buff *skb1;
1525
1526 BUG_ON(sock_owned_by_user(sk));
1527 __NET_ADD_STATS(sock_net(sk), LINUX_MIB_TCPPREQUEUEDROPPED,
1528 skb_queue_len(&tp->ucopy.prequeue));
1529
1530 while ((skb1 = __skb_dequeue(&tp->ucopy.prequeue)) != NULL)
1531 sk_backlog_rcv(sk, skb1);
1532
1533 tp->ucopy.memory = 0;
1534 } else if (skb_queue_len(&tp->ucopy.prequeue) == 1) {
1535 wake_up_interruptible_sync_poll(sk_sleep(sk),
1536 POLLIN | POLLRDNORM | POLLRDBAND);
1537 if (!inet_csk_ack_scheduled(sk))
1538 inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK,
1539 (3 * tcp_rto_min(sk)) / 4,
1540 TCP_RTO_MAX);
1541 }
1542 return true;
1543}
1544EXPORT_SYMBOL(tcp_prequeue);
1545
1546bool tcp_add_backlog(struct sock *sk, struct sk_buff *skb)
1547{
1548 u32 limit = sk->sk_rcvbuf + sk->sk_sndbuf;
1549
1550 /* Only socket owner can try to collapse/prune rx queues
1551 * to reduce memory overhead, so add a little headroom here.
1552 * Few sockets backlog are possibly concurrently non empty.
1553 */
1554 limit += 64*1024;
1555
1556 /* In case all data was pulled from skb frags (in __pskb_pull_tail()),
1557 * we can fix skb->truesize to its real value to avoid future drops.
1558 * This is valid because skb is not yet charged to the socket.
1559 * It has been noticed pure SACK packets were sometimes dropped
1560 * (if cooked by drivers without copybreak feature).
1561 */
1562 if (!skb->data_len)
1563 skb->truesize = SKB_TRUESIZE(skb_end_offset(skb));
1564
1565 if (unlikely(sk_add_backlog(sk, skb, limit))) {
1566 bh_unlock_sock(sk);
1567 __NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPBACKLOGDROP);
1568 return true;
1569 }
1570 return false;
1571}
1572EXPORT_SYMBOL(tcp_add_backlog);
1573
1574int tcp_filter(struct sock *sk, struct sk_buff *skb)
1575{
1576 struct tcphdr *th = (struct tcphdr *)skb->data;
1577 unsigned int eaten = skb->len;
1578 int err;
1579
1580 err = sk_filter_trim_cap(sk, skb, th->doff * 4);
1581 if (!err) {
1582 eaten -= skb->len;
1583 TCP_SKB_CB(skb)->end_seq -= eaten;
1584 }
1585 return err;
1586}
1587EXPORT_SYMBOL(tcp_filter);
1588
1589/*
1590 * From tcp_input.c
1591 */
1592
1593int tcp_v4_rcv(struct sk_buff *skb)
1594{
1595 struct net *net = dev_net(skb->dev);
1596 const struct iphdr *iph;
1597 const struct tcphdr *th;
1598 bool refcounted;
1599 struct sock *sk;
1600 int ret;
1601
1602 if (skb->pkt_type != PACKET_HOST)
1603 goto discard_it;
1604
1605 /* Count it even if it's bad */
1606 __TCP_INC_STATS(net, TCP_MIB_INSEGS);
1607
1608 if (!pskb_may_pull(skb, sizeof(struct tcphdr)))
1609 goto discard_it;
1610
1611 th = (const struct tcphdr *)skb->data;
1612
1613 if (unlikely(th->doff < sizeof(struct tcphdr) / 4))
1614 goto bad_packet;
1615 if (!pskb_may_pull(skb, th->doff * 4))
1616 goto discard_it;
1617
1618 /* An explanation is required here, I think.
1619 * Packet length and doff are validated by header prediction,
1620 * provided case of th->doff==0 is eliminated.
1621 * So, we defer the checks. */
1622
1623 if (skb_checksum_init(skb, IPPROTO_TCP, inet_compute_pseudo))
1624 goto csum_error;
1625
1626 th = (const struct tcphdr *)skb->data;
1627 iph = ip_hdr(skb);
1628 /* This is tricky : We move IPCB at its correct location into TCP_SKB_CB()
1629 * barrier() makes sure compiler wont play fool^Waliasing games.
1630 */
1631 memmove(&TCP_SKB_CB(skb)->header.h4, IPCB(skb),
1632 sizeof(struct inet_skb_parm));
1633 barrier();
1634
1635 TCP_SKB_CB(skb)->seq = ntohl(th->seq);
1636 TCP_SKB_CB(skb)->end_seq = (TCP_SKB_CB(skb)->seq + th->syn + th->fin +
1637 skb->len - th->doff * 4);
1638 TCP_SKB_CB(skb)->ack_seq = ntohl(th->ack_seq);
1639 TCP_SKB_CB(skb)->tcp_flags = tcp_flag_byte(th);
1640 TCP_SKB_CB(skb)->tcp_tw_isn = 0;
1641 TCP_SKB_CB(skb)->ip_dsfield = ipv4_get_dsfield(iph);
1642 TCP_SKB_CB(skb)->sacked = 0;
1643
1644lookup:
1645 sk = __inet_lookup_skb(&tcp_hashinfo, skb, __tcp_hdrlen(th), th->source,
1646 th->dest, &refcounted);
1647 if (!sk)
1648 goto no_tcp_socket;
1649
1650process:
1651 if (sk->sk_state == TCP_TIME_WAIT)
1652 goto do_time_wait;
1653
1654 if (sk->sk_state == TCP_NEW_SYN_RECV) {
1655 struct request_sock *req = inet_reqsk(sk);
1656 struct sock *nsk;
1657
1658 sk = req->rsk_listener;
1659 if (unlikely(tcp_v4_inbound_md5_hash(sk, skb))) {
1660 sk_drops_add(sk, skb);
1661 reqsk_put(req);
1662 goto discard_it;
1663 }
1664 if (unlikely(sk->sk_state != TCP_LISTEN)) {
1665 inet_csk_reqsk_queue_drop_and_put(sk, req);
1666 goto lookup;
1667 }
1668 /* We own a reference on the listener, increase it again
1669 * as we might lose it too soon.
1670 */
1671 sock_hold(sk);
1672 refcounted = true;
1673 nsk = tcp_check_req(sk, skb, req, false);
1674 if (!nsk) {
1675 reqsk_put(req);
1676 goto discard_and_relse;
1677 }
1678 if (nsk == sk) {
1679 reqsk_put(req);
1680 } else if (tcp_child_process(sk, nsk, skb)) {
1681 tcp_v4_send_reset(nsk, skb);
1682 goto discard_and_relse;
1683 } else {
1684 sock_put(sk);
1685 return 0;
1686 }
1687 }
1688 if (unlikely(iph->ttl < inet_sk(sk)->min_ttl)) {
1689 __NET_INC_STATS(net, LINUX_MIB_TCPMINTTLDROP);
1690 goto discard_and_relse;
1691 }
1692
1693 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
1694 goto discard_and_relse;
1695
1696 if (tcp_v4_inbound_md5_hash(sk, skb))
1697 goto discard_and_relse;
1698
1699 nf_reset(skb);
1700
1701 if (tcp_filter(sk, skb))
1702 goto discard_and_relse;
1703 th = (const struct tcphdr *)skb->data;
1704 iph = ip_hdr(skb);
1705
1706 skb->dev = NULL;
1707
1708 if (sk->sk_state == TCP_LISTEN) {
1709 ret = tcp_v4_do_rcv(sk, skb);
1710 goto put_and_return;
1711 }
1712
1713 sk_incoming_cpu_update(sk);
1714
1715 bh_lock_sock_nested(sk);
1716 tcp_segs_in(tcp_sk(sk), skb);
1717 ret = 0;
1718 if (!sock_owned_by_user(sk)) {
1719 if (!tcp_prequeue(sk, skb))
1720 ret = tcp_v4_do_rcv(sk, skb);
1721 } else if (tcp_add_backlog(sk, skb)) {
1722 goto discard_and_relse;
1723 }
1724 bh_unlock_sock(sk);
1725
1726put_and_return:
1727 if (refcounted)
1728 sock_put(sk);
1729
1730 return ret;
1731
1732no_tcp_socket:
1733 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
1734 goto discard_it;
1735
1736 if (tcp_checksum_complete(skb)) {
1737csum_error:
1738 __TCP_INC_STATS(net, TCP_MIB_CSUMERRORS);
1739bad_packet:
1740 __TCP_INC_STATS(net, TCP_MIB_INERRS);
1741 } else {
1742 tcp_v4_send_reset(NULL, skb);
1743 }
1744
1745discard_it:
1746 /* Discard frame. */
1747 kfree_skb(skb);
1748 return 0;
1749
1750discard_and_relse:
1751 sk_drops_add(sk, skb);
1752 if (refcounted)
1753 sock_put(sk);
1754 goto discard_it;
1755
1756do_time_wait:
1757 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) {
1758 inet_twsk_put(inet_twsk(sk));
1759 goto discard_it;
1760 }
1761
1762 if (tcp_checksum_complete(skb)) {
1763 inet_twsk_put(inet_twsk(sk));
1764 goto csum_error;
1765 }
1766 switch (tcp_timewait_state_process(inet_twsk(sk), skb, th)) {
1767 case TCP_TW_SYN: {
1768 struct sock *sk2 = inet_lookup_listener(dev_net(skb->dev),
1769 &tcp_hashinfo, skb,
1770 __tcp_hdrlen(th),
1771 iph->saddr, th->source,
1772 iph->daddr, th->dest,
1773 inet_iif(skb));
1774 if (sk2) {
1775 inet_twsk_deschedule_put(inet_twsk(sk));
1776 sk = sk2;
1777 refcounted = false;
1778 goto process;
1779 }
1780 /* Fall through to ACK */
1781 }
1782 case TCP_TW_ACK:
1783 tcp_v4_timewait_ack(sk, skb);
1784 break;
1785 case TCP_TW_RST:
1786 tcp_v4_send_reset(sk, skb);
1787 inet_twsk_deschedule_put(inet_twsk(sk));
1788 goto discard_it;
1789 case TCP_TW_SUCCESS:;
1790 }
1791 goto discard_it;
1792}
1793
1794static struct timewait_sock_ops tcp_timewait_sock_ops = {
1795 .twsk_obj_size = sizeof(struct tcp_timewait_sock),
1796 .twsk_unique = tcp_twsk_unique,
1797 .twsk_destructor= tcp_twsk_destructor,
1798};
1799
1800void inet_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb)
1801{
1802 struct dst_entry *dst = skb_dst(skb);
1803
1804 if (dst && dst_hold_safe(dst)) {
1805 sk->sk_rx_dst = dst;
1806 inet_sk(sk)->rx_dst_ifindex = skb->skb_iif;
1807 }
1808}
1809EXPORT_SYMBOL(inet_sk_rx_dst_set);
1810
1811const struct inet_connection_sock_af_ops ipv4_specific = {
1812 .queue_xmit = ip_queue_xmit,
1813 .send_check = tcp_v4_send_check,
1814 .rebuild_header = inet_sk_rebuild_header,
1815 .sk_rx_dst_set = inet_sk_rx_dst_set,
1816 .conn_request = tcp_v4_conn_request,
1817 .syn_recv_sock = tcp_v4_syn_recv_sock,
1818 .net_header_len = sizeof(struct iphdr),
1819 .setsockopt = ip_setsockopt,
1820 .getsockopt = ip_getsockopt,
1821 .addr2sockaddr = inet_csk_addr2sockaddr,
1822 .sockaddr_len = sizeof(struct sockaddr_in),
1823 .bind_conflict = inet_csk_bind_conflict,
1824#ifdef CONFIG_COMPAT
1825 .compat_setsockopt = compat_ip_setsockopt,
1826 .compat_getsockopt = compat_ip_getsockopt,
1827#endif
1828 .mtu_reduced = tcp_v4_mtu_reduced,
1829};
1830EXPORT_SYMBOL(ipv4_specific);
1831
1832#ifdef CONFIG_TCP_MD5SIG
1833static const struct tcp_sock_af_ops tcp_sock_ipv4_specific = {
1834 .md5_lookup = tcp_v4_md5_lookup,
1835 .calc_md5_hash = tcp_v4_md5_hash_skb,
1836 .md5_parse = tcp_v4_parse_md5_keys,
1837};
1838#endif
1839
1840/* NOTE: A lot of things set to zero explicitly by call to
1841 * sk_alloc() so need not be done here.
1842 */
1843static int tcp_v4_init_sock(struct sock *sk)
1844{
1845 struct inet_connection_sock *icsk = inet_csk(sk);
1846
1847 tcp_init_sock(sk);
1848
1849 icsk->icsk_af_ops = &ipv4_specific;
1850
1851#ifdef CONFIG_TCP_MD5SIG
1852 tcp_sk(sk)->af_specific = &tcp_sock_ipv4_specific;
1853#endif
1854
1855 return 0;
1856}
1857
1858void tcp_v4_destroy_sock(struct sock *sk)
1859{
1860 struct tcp_sock *tp = tcp_sk(sk);
1861
1862 tcp_clear_xmit_timers(sk);
1863
1864 tcp_cleanup_congestion_control(sk);
1865
1866 /* Cleanup up the write buffer. */
1867 tcp_write_queue_purge(sk);
1868
1869 /* Cleans up our, hopefully empty, out_of_order_queue. */
1870 skb_rbtree_purge(&tp->out_of_order_queue);
1871
1872#ifdef CONFIG_TCP_MD5SIG
1873 /* Clean up the MD5 key list, if any */
1874 if (tp->md5sig_info) {
1875 tcp_clear_md5_list(sk);
1876 kfree_rcu(tp->md5sig_info, rcu);
1877 tp->md5sig_info = NULL;
1878 }
1879#endif
1880
1881 /* Clean prequeue, it must be empty really */
1882 __skb_queue_purge(&tp->ucopy.prequeue);
1883
1884 /* Clean up a referenced TCP bind bucket. */
1885 if (inet_csk(sk)->icsk_bind_hash)
1886 inet_put_port(sk);
1887
1888 BUG_ON(tp->fastopen_rsk);
1889
1890 /* If socket is aborted during connect operation */
1891 tcp_free_fastopen_req(tp);
1892 tcp_saved_syn_free(tp);
1893
1894 local_bh_disable();
1895 sk_sockets_allocated_dec(sk);
1896 local_bh_enable();
1897}
1898EXPORT_SYMBOL(tcp_v4_destroy_sock);
1899
1900#ifdef CONFIG_PROC_FS
1901/* Proc filesystem TCP sock list dumping. */
1902
1903/*
1904 * Get next listener socket follow cur. If cur is NULL, get first socket
1905 * starting from bucket given in st->bucket; when st->bucket is zero the
1906 * very first socket in the hash table is returned.
1907 */
1908static void *listening_get_next(struct seq_file *seq, void *cur)
1909{
1910 struct tcp_iter_state *st = seq->private;
1911 struct net *net = seq_file_net(seq);
1912 struct inet_listen_hashbucket *ilb;
1913 struct sock *sk = cur;
1914
1915 if (!sk) {
1916get_head:
1917 ilb = &tcp_hashinfo.listening_hash[st->bucket];
1918 spin_lock(&ilb->lock);
1919 sk = sk_head(&ilb->head);
1920 st->offset = 0;
1921 goto get_sk;
1922 }
1923 ilb = &tcp_hashinfo.listening_hash[st->bucket];
1924 ++st->num;
1925 ++st->offset;
1926
1927 sk = sk_next(sk);
1928get_sk:
1929 sk_for_each_from(sk) {
1930 if (!net_eq(sock_net(sk), net))
1931 continue;
1932 if (sk->sk_family == st->family)
1933 return sk;
1934 }
1935 spin_unlock(&ilb->lock);
1936 st->offset = 0;
1937 if (++st->bucket < INET_LHTABLE_SIZE)
1938 goto get_head;
1939 return NULL;
1940}
1941
1942static void *listening_get_idx(struct seq_file *seq, loff_t *pos)
1943{
1944 struct tcp_iter_state *st = seq->private;
1945 void *rc;
1946
1947 st->bucket = 0;
1948 st->offset = 0;
1949 rc = listening_get_next(seq, NULL);
1950
1951 while (rc && *pos) {
1952 rc = listening_get_next(seq, rc);
1953 --*pos;
1954 }
1955 return rc;
1956}
1957
1958static inline bool empty_bucket(const struct tcp_iter_state *st)
1959{
1960 return hlist_nulls_empty(&tcp_hashinfo.ehash[st->bucket].chain);
1961}
1962
1963/*
1964 * Get first established socket starting from bucket given in st->bucket.
1965 * If st->bucket is zero, the very first socket in the hash is returned.
1966 */
1967static void *established_get_first(struct seq_file *seq)
1968{
1969 struct tcp_iter_state *st = seq->private;
1970 struct net *net = seq_file_net(seq);
1971 void *rc = NULL;
1972
1973 st->offset = 0;
1974 for (; st->bucket <= tcp_hashinfo.ehash_mask; ++st->bucket) {
1975 struct sock *sk;
1976 struct hlist_nulls_node *node;
1977 spinlock_t *lock = inet_ehash_lockp(&tcp_hashinfo, st->bucket);
1978
1979 /* Lockless fast path for the common case of empty buckets */
1980 if (empty_bucket(st))
1981 continue;
1982
1983 spin_lock_bh(lock);
1984 sk_nulls_for_each(sk, node, &tcp_hashinfo.ehash[st->bucket].chain) {
1985 if (sk->sk_family != st->family ||
1986 !net_eq(sock_net(sk), net)) {
1987 continue;
1988 }
1989 rc = sk;
1990 goto out;
1991 }
1992 spin_unlock_bh(lock);
1993 }
1994out:
1995 return rc;
1996}
1997
1998static void *established_get_next(struct seq_file *seq, void *cur)
1999{
2000 struct sock *sk = cur;
2001 struct hlist_nulls_node *node;
2002 struct tcp_iter_state *st = seq->private;
2003 struct net *net = seq_file_net(seq);
2004
2005 ++st->num;
2006 ++st->offset;
2007
2008 sk = sk_nulls_next(sk);
2009
2010 sk_nulls_for_each_from(sk, node) {
2011 if (sk->sk_family == st->family && net_eq(sock_net(sk), net))
2012 return sk;
2013 }
2014
2015 spin_unlock_bh(inet_ehash_lockp(&tcp_hashinfo, st->bucket));
2016 ++st->bucket;
2017 return established_get_first(seq);
2018}
2019
2020static void *established_get_idx(struct seq_file *seq, loff_t pos)
2021{
2022 struct tcp_iter_state *st = seq->private;
2023 void *rc;
2024
2025 st->bucket = 0;
2026 rc = established_get_first(seq);
2027
2028 while (rc && pos) {
2029 rc = established_get_next(seq, rc);
2030 --pos;
2031 }
2032 return rc;
2033}
2034
2035static void *tcp_get_idx(struct seq_file *seq, loff_t pos)
2036{
2037 void *rc;
2038 struct tcp_iter_state *st = seq->private;
2039
2040 st->state = TCP_SEQ_STATE_LISTENING;
2041 rc = listening_get_idx(seq, &pos);
2042
2043 if (!rc) {
2044 st->state = TCP_SEQ_STATE_ESTABLISHED;
2045 rc = established_get_idx(seq, pos);
2046 }
2047
2048 return rc;
2049}
2050
2051static void *tcp_seek_last_pos(struct seq_file *seq)
2052{
2053 struct tcp_iter_state *st = seq->private;
2054 int offset = st->offset;
2055 int orig_num = st->num;
2056 void *rc = NULL;
2057
2058 switch (st->state) {
2059 case TCP_SEQ_STATE_LISTENING:
2060 if (st->bucket >= INET_LHTABLE_SIZE)
2061 break;
2062 st->state = TCP_SEQ_STATE_LISTENING;
2063 rc = listening_get_next(seq, NULL);
2064 while (offset-- && rc)
2065 rc = listening_get_next(seq, rc);
2066 if (rc)
2067 break;
2068 st->bucket = 0;
2069 st->state = TCP_SEQ_STATE_ESTABLISHED;
2070 /* Fallthrough */
2071 case TCP_SEQ_STATE_ESTABLISHED:
2072 if (st->bucket > tcp_hashinfo.ehash_mask)
2073 break;
2074 rc = established_get_first(seq);
2075 while (offset-- && rc)
2076 rc = established_get_next(seq, rc);
2077 }
2078
2079 st->num = orig_num;
2080
2081 return rc;
2082}
2083
2084static void *tcp_seq_start(struct seq_file *seq, loff_t *pos)
2085{
2086 struct tcp_iter_state *st = seq->private;
2087 void *rc;
2088
2089 if (*pos && *pos == st->last_pos) {
2090 rc = tcp_seek_last_pos(seq);
2091 if (rc)
2092 goto out;
2093 }
2094
2095 st->state = TCP_SEQ_STATE_LISTENING;
2096 st->num = 0;
2097 st->bucket = 0;
2098 st->offset = 0;
2099 rc = *pos ? tcp_get_idx(seq, *pos - 1) : SEQ_START_TOKEN;
2100
2101out:
2102 st->last_pos = *pos;
2103 return rc;
2104}
2105
2106static void *tcp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2107{
2108 struct tcp_iter_state *st = seq->private;
2109 void *rc = NULL;
2110
2111 if (v == SEQ_START_TOKEN) {
2112 rc = tcp_get_idx(seq, 0);
2113 goto out;
2114 }
2115
2116 switch (st->state) {
2117 case TCP_SEQ_STATE_LISTENING:
2118 rc = listening_get_next(seq, v);
2119 if (!rc) {
2120 st->state = TCP_SEQ_STATE_ESTABLISHED;
2121 st->bucket = 0;
2122 st->offset = 0;
2123 rc = established_get_first(seq);
2124 }
2125 break;
2126 case TCP_SEQ_STATE_ESTABLISHED:
2127 rc = established_get_next(seq, v);
2128 break;
2129 }
2130out:
2131 ++*pos;
2132 st->last_pos = *pos;
2133 return rc;
2134}
2135
2136static void tcp_seq_stop(struct seq_file *seq, void *v)
2137{
2138 struct tcp_iter_state *st = seq->private;
2139
2140 switch (st->state) {
2141 case TCP_SEQ_STATE_LISTENING:
2142 if (v != SEQ_START_TOKEN)
2143 spin_unlock(&tcp_hashinfo.listening_hash[st->bucket].lock);
2144 break;
2145 case TCP_SEQ_STATE_ESTABLISHED:
2146 if (v)
2147 spin_unlock_bh(inet_ehash_lockp(&tcp_hashinfo, st->bucket));
2148 break;
2149 }
2150}
2151
2152int tcp_seq_open(struct inode *inode, struct file *file)
2153{
2154 struct tcp_seq_afinfo *afinfo = PDE_DATA(inode);
2155 struct tcp_iter_state *s;
2156 int err;
2157
2158 err = seq_open_net(inode, file, &afinfo->seq_ops,
2159 sizeof(struct tcp_iter_state));
2160 if (err < 0)
2161 return err;
2162
2163 s = ((struct seq_file *)file->private_data)->private;
2164 s->family = afinfo->family;
2165 s->last_pos = 0;
2166 return 0;
2167}
2168EXPORT_SYMBOL(tcp_seq_open);
2169
2170int tcp_proc_register(struct net *net, struct tcp_seq_afinfo *afinfo)
2171{
2172 int rc = 0;
2173 struct proc_dir_entry *p;
2174
2175 afinfo->seq_ops.start = tcp_seq_start;
2176 afinfo->seq_ops.next = tcp_seq_next;
2177 afinfo->seq_ops.stop = tcp_seq_stop;
2178
2179 p = proc_create_data(afinfo->name, S_IRUGO, net->proc_net,
2180 afinfo->seq_fops, afinfo);
2181 if (!p)
2182 rc = -ENOMEM;
2183 return rc;
2184}
2185EXPORT_SYMBOL(tcp_proc_register);
2186
2187void tcp_proc_unregister(struct net *net, struct tcp_seq_afinfo *afinfo)
2188{
2189 remove_proc_entry(afinfo->name, net->proc_net);
2190}
2191EXPORT_SYMBOL(tcp_proc_unregister);
2192
2193static void get_openreq4(const struct request_sock *req,
2194 struct seq_file *f, int i)
2195{
2196 const struct inet_request_sock *ireq = inet_rsk(req);
2197 long delta = req->rsk_timer.expires - jiffies;
2198
2199 seq_printf(f, "%4d: %08X:%04X %08X:%04X"
2200 " %02X %08X:%08X %02X:%08lX %08X %5u %8d %u %d %pK",
2201 i,
2202 ireq->ir_loc_addr,
2203 ireq->ir_num,
2204 ireq->ir_rmt_addr,
2205 ntohs(ireq->ir_rmt_port),
2206 TCP_SYN_RECV,
2207 0, 0, /* could print option size, but that is af dependent. */
2208 1, /* timers active (only the expire timer) */
2209 jiffies_delta_to_clock_t(delta),
2210 req->num_timeout,
2211 from_kuid_munged(seq_user_ns(f),
2212 sock_i_uid(req->rsk_listener)),
2213 0, /* non standard timer */
2214 0, /* open_requests have no inode */
2215 0,
2216 req);
2217}
2218
2219static void get_tcp4_sock(struct sock *sk, struct seq_file *f, int i)
2220{
2221 int timer_active;
2222 unsigned long timer_expires;
2223 const struct tcp_sock *tp = tcp_sk(sk);
2224 const struct inet_connection_sock *icsk = inet_csk(sk);
2225 const struct inet_sock *inet = inet_sk(sk);
2226 const struct fastopen_queue *fastopenq = &icsk->icsk_accept_queue.fastopenq;
2227 __be32 dest = inet->inet_daddr;
2228 __be32 src = inet->inet_rcv_saddr;
2229 __u16 destp = ntohs(inet->inet_dport);
2230 __u16 srcp = ntohs(inet->inet_sport);
2231 int rx_queue;
2232 int state;
2233
2234 if (icsk->icsk_pending == ICSK_TIME_RETRANS ||
2235 icsk->icsk_pending == ICSK_TIME_EARLY_RETRANS ||
2236 icsk->icsk_pending == ICSK_TIME_LOSS_PROBE) {
2237 timer_active = 1;
2238 timer_expires = icsk->icsk_timeout;
2239 } else if (icsk->icsk_pending == ICSK_TIME_PROBE0) {
2240 timer_active = 4;
2241 timer_expires = icsk->icsk_timeout;
2242 } else if (timer_pending(&sk->sk_timer)) {
2243 timer_active = 2;
2244 timer_expires = sk->sk_timer.expires;
2245 } else {
2246 timer_active = 0;
2247 timer_expires = jiffies;
2248 }
2249
2250 state = sk_state_load(sk);
2251 if (state == TCP_LISTEN)
2252 rx_queue = sk->sk_ack_backlog;
2253 else
2254 /* Because we don't lock the socket,
2255 * we might find a transient negative value.
2256 */
2257 rx_queue = max_t(int, tp->rcv_nxt - tp->copied_seq, 0);
2258
2259 seq_printf(f, "%4d: %08X:%04X %08X:%04X %02X %08X:%08X %02X:%08lX "
2260 "%08X %5u %8d %lu %d %pK %lu %lu %u %u %d",
2261 i, src, srcp, dest, destp, state,
2262 tp->write_seq - tp->snd_una,
2263 rx_queue,
2264 timer_active,
2265 jiffies_delta_to_clock_t(timer_expires - jiffies),
2266 icsk->icsk_retransmits,
2267 from_kuid_munged(seq_user_ns(f), sock_i_uid(sk)),
2268 icsk->icsk_probes_out,
2269 sock_i_ino(sk),
2270 atomic_read(&sk->sk_refcnt), sk,
2271 jiffies_to_clock_t(icsk->icsk_rto),
2272 jiffies_to_clock_t(icsk->icsk_ack.ato),
2273 (icsk->icsk_ack.quick << 1) | icsk->icsk_ack.pingpong,
2274 tp->snd_cwnd,
2275 state == TCP_LISTEN ?
2276 fastopenq->max_qlen :
2277 (tcp_in_initial_slowstart(tp) ? -1 : tp->snd_ssthresh));
2278}
2279
2280static void get_timewait4_sock(const struct inet_timewait_sock *tw,
2281 struct seq_file *f, int i)
2282{
2283 long delta = tw->tw_timer.expires - jiffies;
2284 __be32 dest, src;
2285 __u16 destp, srcp;
2286
2287 dest = tw->tw_daddr;
2288 src = tw->tw_rcv_saddr;
2289 destp = ntohs(tw->tw_dport);
2290 srcp = ntohs(tw->tw_sport);
2291
2292 seq_printf(f, "%4d: %08X:%04X %08X:%04X"
2293 " %02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %pK",
2294 i, src, srcp, dest, destp, tw->tw_substate, 0, 0,
2295 3, jiffies_delta_to_clock_t(delta), 0, 0, 0, 0,
2296 atomic_read(&tw->tw_refcnt), tw);
2297}
2298
2299#define TMPSZ 150
2300
2301static int tcp4_seq_show(struct seq_file *seq, void *v)
2302{
2303 struct tcp_iter_state *st;
2304 struct sock *sk = v;
2305
2306 seq_setwidth(seq, TMPSZ - 1);
2307 if (v == SEQ_START_TOKEN) {
2308 seq_puts(seq, " sl local_address rem_address st tx_queue "
2309 "rx_queue tr tm->when retrnsmt uid timeout "
2310 "inode");
2311 goto out;
2312 }
2313 st = seq->private;
2314
2315 if (sk->sk_state == TCP_TIME_WAIT)
2316 get_timewait4_sock(v, seq, st->num);
2317 else if (sk->sk_state == TCP_NEW_SYN_RECV)
2318 get_openreq4(v, seq, st->num);
2319 else
2320 get_tcp4_sock(v, seq, st->num);
2321out:
2322 seq_pad(seq, '\n');
2323 return 0;
2324}
2325
2326static const struct file_operations tcp_afinfo_seq_fops = {
2327 .owner = THIS_MODULE,
2328 .open = tcp_seq_open,
2329 .read = seq_read,
2330 .llseek = seq_lseek,
2331 .release = seq_release_net
2332};
2333
2334static struct tcp_seq_afinfo tcp4_seq_afinfo = {
2335 .name = "tcp",
2336 .family = AF_INET,
2337 .seq_fops = &tcp_afinfo_seq_fops,
2338 .seq_ops = {
2339 .show = tcp4_seq_show,
2340 },
2341};
2342
2343static int __net_init tcp4_proc_init_net(struct net *net)
2344{
2345 return tcp_proc_register(net, &tcp4_seq_afinfo);
2346}
2347
2348static void __net_exit tcp4_proc_exit_net(struct net *net)
2349{
2350 tcp_proc_unregister(net, &tcp4_seq_afinfo);
2351}
2352
2353static struct pernet_operations tcp4_net_ops = {
2354 .init = tcp4_proc_init_net,
2355 .exit = tcp4_proc_exit_net,
2356};
2357
2358int __init tcp4_proc_init(void)
2359{
2360 return register_pernet_subsys(&tcp4_net_ops);
2361}
2362
2363void tcp4_proc_exit(void)
2364{
2365 unregister_pernet_subsys(&tcp4_net_ops);
2366}
2367#endif /* CONFIG_PROC_FS */
2368
2369struct proto tcp_prot = {
2370 .name = "TCP",
2371 .owner = THIS_MODULE,
2372 .close = tcp_close,
2373 .connect = tcp_v4_connect,
2374 .disconnect = tcp_disconnect,
2375 .accept = inet_csk_accept,
2376 .ioctl = tcp_ioctl,
2377 .init = tcp_v4_init_sock,
2378 .destroy = tcp_v4_destroy_sock,
2379 .shutdown = tcp_shutdown,
2380 .setsockopt = tcp_setsockopt,
2381 .getsockopt = tcp_getsockopt,
2382 .recvmsg = tcp_recvmsg,
2383 .sendmsg = tcp_sendmsg,
2384 .sendpage = tcp_sendpage,
2385 .backlog_rcv = tcp_v4_do_rcv,
2386 .release_cb = tcp_release_cb,
2387 .hash = inet_hash,
2388 .unhash = inet_unhash,
2389 .get_port = inet_csk_get_port,
2390 .enter_memory_pressure = tcp_enter_memory_pressure,
2391 .stream_memory_free = tcp_stream_memory_free,
2392 .sockets_allocated = &tcp_sockets_allocated,
2393 .orphan_count = &tcp_orphan_count,
2394 .memory_allocated = &tcp_memory_allocated,
2395 .memory_pressure = &tcp_memory_pressure,
2396 .sysctl_mem = sysctl_tcp_mem,
2397 .sysctl_wmem = sysctl_tcp_wmem,
2398 .sysctl_rmem = sysctl_tcp_rmem,
2399 .max_header = MAX_TCP_HEADER,
2400 .obj_size = sizeof(struct tcp_sock),
2401 .slab_flags = SLAB_DESTROY_BY_RCU,
2402 .twsk_prot = &tcp_timewait_sock_ops,
2403 .rsk_prot = &tcp_request_sock_ops,
2404 .h.hashinfo = &tcp_hashinfo,
2405 .no_autobind = true,
2406#ifdef CONFIG_COMPAT
2407 .compat_setsockopt = compat_tcp_setsockopt,
2408 .compat_getsockopt = compat_tcp_getsockopt,
2409#endif
2410 .diag_destroy = tcp_abort,
2411};
2412EXPORT_SYMBOL(tcp_prot);
2413
2414static void __net_exit tcp_sk_exit(struct net *net)
2415{
2416 int cpu;
2417
2418 for_each_possible_cpu(cpu)
2419 inet_ctl_sock_destroy(*per_cpu_ptr(net->ipv4.tcp_sk, cpu));
2420 free_percpu(net->ipv4.tcp_sk);
2421}
2422
2423static int __net_init tcp_sk_init(struct net *net)
2424{
2425 int res, cpu;
2426
2427 net->ipv4.tcp_sk = alloc_percpu(struct sock *);
2428 if (!net->ipv4.tcp_sk)
2429 return -ENOMEM;
2430
2431 for_each_possible_cpu(cpu) {
2432 struct sock *sk;
2433
2434 res = inet_ctl_sock_create(&sk, PF_INET, SOCK_RAW,
2435 IPPROTO_TCP, net);
2436 if (res)
2437 goto fail;
2438 sock_set_flag(sk, SOCK_USE_WRITE_QUEUE);
2439 *per_cpu_ptr(net->ipv4.tcp_sk, cpu) = sk;
2440 }
2441
2442 net->ipv4.sysctl_tcp_ecn = 2;
2443 net->ipv4.sysctl_tcp_ecn_fallback = 1;
2444
2445 net->ipv4.sysctl_tcp_base_mss = TCP_BASE_MSS;
2446 net->ipv4.sysctl_tcp_probe_threshold = TCP_PROBE_THRESHOLD;
2447 net->ipv4.sysctl_tcp_probe_interval = TCP_PROBE_INTERVAL;
2448
2449 net->ipv4.sysctl_tcp_keepalive_time = TCP_KEEPALIVE_TIME;
2450 net->ipv4.sysctl_tcp_keepalive_probes = TCP_KEEPALIVE_PROBES;
2451 net->ipv4.sysctl_tcp_keepalive_intvl = TCP_KEEPALIVE_INTVL;
2452
2453 net->ipv4.sysctl_tcp_syn_retries = TCP_SYN_RETRIES;
2454 net->ipv4.sysctl_tcp_synack_retries = TCP_SYNACK_RETRIES;
2455 net->ipv4.sysctl_tcp_syncookies = 1;
2456 net->ipv4.sysctl_tcp_reordering = TCP_FASTRETRANS_THRESH;
2457 net->ipv4.sysctl_tcp_retries1 = TCP_RETR1;
2458 net->ipv4.sysctl_tcp_retries2 = TCP_RETR2;
2459 net->ipv4.sysctl_tcp_orphan_retries = 0;
2460 net->ipv4.sysctl_tcp_fin_timeout = TCP_FIN_TIMEOUT;
2461 net->ipv4.sysctl_tcp_notsent_lowat = UINT_MAX;
2462 net->ipv4.sysctl_tcp_tw_reuse = 0;
2463
2464 return 0;
2465fail:
2466 tcp_sk_exit(net);
2467
2468 return res;
2469}
2470
2471static void __net_exit tcp_sk_exit_batch(struct list_head *net_exit_list)
2472{
2473 inet_twsk_purge(&tcp_hashinfo, &tcp_death_row, AF_INET);
2474}
2475
2476static struct pernet_operations __net_initdata tcp_sk_ops = {
2477 .init = tcp_sk_init,
2478 .exit = tcp_sk_exit,
2479 .exit_batch = tcp_sk_exit_batch,
2480};
2481
2482void __init tcp_v4_init(void)
2483{
2484 inet_hashinfo_init(&tcp_hashinfo);
2485 if (register_pernet_subsys(&tcp_sk_ops))
2486 panic("Failed to create the TCP control socket.\n");
2487}