]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
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 | * Version: $Id: tcp_ipv4.c,v 1.240 2002/02/01 22:01:04 davem Exp $ | |
9 | * | |
10 | * IPv4 specific functions | |
11 | * | |
12 | * | |
13 | * code split from: | |
14 | * linux/ipv4/tcp.c | |
15 | * linux/ipv4/tcp_input.c | |
16 | * linux/ipv4/tcp_output.c | |
17 | * | |
18 | * See tcp.c for author information | |
19 | * | |
20 | * This program is free software; you can redistribute it and/or | |
21 | * modify it under the terms of the GNU General Public License | |
22 | * as published by the Free Software Foundation; either version | |
23 | * 2 of the License, or (at your option) any later version. | |
24 | */ | |
25 | ||
26 | /* | |
27 | * Changes: | |
28 | * David S. Miller : New socket lookup architecture. | |
29 | * This code is dedicated to John Dyson. | |
30 | * David S. Miller : Change semantics of established hash, | |
31 | * half is devoted to TIME_WAIT sockets | |
32 | * and the rest go in the other half. | |
33 | * Andi Kleen : Add support for syncookies and fixed | |
34 | * some bugs: ip options weren't passed to | |
35 | * the TCP layer, missed a check for an | |
36 | * ACK bit. | |
37 | * Andi Kleen : Implemented fast path mtu discovery. | |
38 | * Fixed many serious bugs in the | |
60236fdd | 39 | * request_sock handling and moved |
1da177e4 LT |
40 | * most of it into the af independent code. |
41 | * Added tail drop and some other bugfixes. | |
caa20d9a | 42 | * Added new listen semantics. |
1da177e4 LT |
43 | * Mike McLagan : Routing by source |
44 | * Juan Jose Ciarlante: ip_dynaddr bits | |
45 | * Andi Kleen: various fixes. | |
46 | * Vitaly E. Lavrov : Transparent proxy revived after year | |
47 | * coma. | |
48 | * Andi Kleen : Fix new listen. | |
49 | * Andi Kleen : Fix accept error reporting. | |
50 | * YOSHIFUJI Hideaki @USAGI and: Support IPV6_V6ONLY socket option, which | |
51 | * Alexey Kuznetsov allow both IPv4 and IPv6 sockets to bind | |
52 | * a single port at the same time. | |
53 | */ | |
54 | ||
1da177e4 LT |
55 | |
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 | ||
457c4cbc | 65 | #include <net/net_namespace.h> |
1da177e4 | 66 | #include <net/icmp.h> |
304a1618 | 67 | #include <net/inet_hashtables.h> |
1da177e4 | 68 | #include <net/tcp.h> |
20380731 | 69 | #include <net/transp_v6.h> |
1da177e4 LT |
70 | #include <net/ipv6.h> |
71 | #include <net/inet_common.h> | |
6d6ee43e | 72 | #include <net/timewait_sock.h> |
1da177e4 | 73 | #include <net/xfrm.h> |
1a2449a8 | 74 | #include <net/netdma.h> |
1da177e4 LT |
75 | |
76 | #include <linux/inet.h> | |
77 | #include <linux/ipv6.h> | |
78 | #include <linux/stddef.h> | |
79 | #include <linux/proc_fs.h> | |
80 | #include <linux/seq_file.h> | |
81 | ||
cfb6eeb4 YH |
82 | #include <linux/crypto.h> |
83 | #include <linux/scatterlist.h> | |
84 | ||
ab32ea5d BH |
85 | int sysctl_tcp_tw_reuse __read_mostly; |
86 | int sysctl_tcp_low_latency __read_mostly; | |
1da177e4 LT |
87 | |
88 | /* Check TCP sequence numbers in ICMP packets. */ | |
89 | #define ICMP_MIN_LENGTH 8 | |
90 | ||
8292a17a | 91 | void tcp_v4_send_check(struct sock *sk, int len, struct sk_buff *skb); |
1da177e4 | 92 | |
cfb6eeb4 | 93 | #ifdef CONFIG_TCP_MD5SIG |
7174259e ACM |
94 | static struct tcp_md5sig_key *tcp_v4_md5_do_lookup(struct sock *sk, |
95 | __be32 addr); | |
cfb6eeb4 | 96 | static int tcp_v4_do_calc_md5_hash(char *md5_hash, struct tcp_md5sig_key *key, |
7174259e ACM |
97 | __be32 saddr, __be32 daddr, |
98 | struct tcphdr *th, int protocol, | |
9cb5734e | 99 | unsigned int tcplen); |
cfb6eeb4 YH |
100 | #endif |
101 | ||
0f7ff927 | 102 | struct inet_hashinfo __cacheline_aligned tcp_hashinfo = { |
7174259e ACM |
103 | .lhash_lock = __RW_LOCK_UNLOCKED(tcp_hashinfo.lhash_lock), |
104 | .lhash_users = ATOMIC_INIT(0), | |
105 | .lhash_wait = __WAIT_QUEUE_HEAD_INITIALIZER(tcp_hashinfo.lhash_wait), | |
1da177e4 LT |
106 | }; |
107 | ||
a94f723d | 108 | static inline __u32 tcp_v4_init_sequence(struct sk_buff *skb) |
1da177e4 | 109 | { |
eddc9ec5 ACM |
110 | return secure_tcp_sequence_number(ip_hdr(skb)->daddr, |
111 | ip_hdr(skb)->saddr, | |
aa8223c7 ACM |
112 | tcp_hdr(skb)->dest, |
113 | tcp_hdr(skb)->source); | |
1da177e4 LT |
114 | } |
115 | ||
6d6ee43e ACM |
116 | int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp) |
117 | { | |
118 | const struct tcp_timewait_sock *tcptw = tcp_twsk(sktw); | |
119 | struct tcp_sock *tp = tcp_sk(sk); | |
120 | ||
121 | /* With PAWS, it is safe from the viewpoint | |
122 | of data integrity. Even without PAWS it is safe provided sequence | |
123 | spaces do not overlap i.e. at data rates <= 80Mbit/sec. | |
124 | ||
125 | Actually, the idea is close to VJ's one, only timestamp cache is | |
126 | held not per host, but per port pair and TW bucket is used as state | |
127 | holder. | |
128 | ||
129 | If TW bucket has been already destroyed we fall back to VJ's scheme | |
130 | and use initial timestamp retrieved from peer table. | |
131 | */ | |
132 | if (tcptw->tw_ts_recent_stamp && | |
133 | (twp == NULL || (sysctl_tcp_tw_reuse && | |
9d729f72 | 134 | get_seconds() - tcptw->tw_ts_recent_stamp > 1))) { |
6d6ee43e ACM |
135 | tp->write_seq = tcptw->tw_snd_nxt + 65535 + 2; |
136 | if (tp->write_seq == 0) | |
137 | tp->write_seq = 1; | |
138 | tp->rx_opt.ts_recent = tcptw->tw_ts_recent; | |
139 | tp->rx_opt.ts_recent_stamp = tcptw->tw_ts_recent_stamp; | |
140 | sock_hold(sktw); | |
141 | return 1; | |
142 | } | |
143 | ||
144 | return 0; | |
145 | } | |
146 | ||
147 | EXPORT_SYMBOL_GPL(tcp_twsk_unique); | |
148 | ||
1da177e4 LT |
149 | /* This will initiate an outgoing connection. */ |
150 | int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len) | |
151 | { | |
152 | struct inet_sock *inet = inet_sk(sk); | |
153 | struct tcp_sock *tp = tcp_sk(sk); | |
154 | struct sockaddr_in *usin = (struct sockaddr_in *)uaddr; | |
155 | struct rtable *rt; | |
bada8adc | 156 | __be32 daddr, nexthop; |
1da177e4 LT |
157 | int tmp; |
158 | int err; | |
159 | ||
160 | if (addr_len < sizeof(struct sockaddr_in)) | |
161 | return -EINVAL; | |
162 | ||
163 | if (usin->sin_family != AF_INET) | |
164 | return -EAFNOSUPPORT; | |
165 | ||
166 | nexthop = daddr = usin->sin_addr.s_addr; | |
167 | if (inet->opt && inet->opt->srr) { | |
168 | if (!daddr) | |
169 | return -EINVAL; | |
170 | nexthop = inet->opt->faddr; | |
171 | } | |
172 | ||
173 | tmp = ip_route_connect(&rt, nexthop, inet->saddr, | |
174 | RT_CONN_FLAGS(sk), sk->sk_bound_dev_if, | |
175 | IPPROTO_TCP, | |
8eb9086f | 176 | inet->sport, usin->sin_port, sk, 1); |
584bdf8c WD |
177 | if (tmp < 0) { |
178 | if (tmp == -ENETUNREACH) | |
179 | IP_INC_STATS_BH(IPSTATS_MIB_OUTNOROUTES); | |
1da177e4 | 180 | return tmp; |
584bdf8c | 181 | } |
1da177e4 LT |
182 | |
183 | if (rt->rt_flags & (RTCF_MULTICAST | RTCF_BROADCAST)) { | |
184 | ip_rt_put(rt); | |
185 | return -ENETUNREACH; | |
186 | } | |
187 | ||
188 | if (!inet->opt || !inet->opt->srr) | |
189 | daddr = rt->rt_dst; | |
190 | ||
191 | if (!inet->saddr) | |
192 | inet->saddr = rt->rt_src; | |
193 | inet->rcv_saddr = inet->saddr; | |
194 | ||
195 | if (tp->rx_opt.ts_recent_stamp && inet->daddr != daddr) { | |
196 | /* Reset inherited state */ | |
197 | tp->rx_opt.ts_recent = 0; | |
198 | tp->rx_opt.ts_recent_stamp = 0; | |
199 | tp->write_seq = 0; | |
200 | } | |
201 | ||
295ff7ed | 202 | if (tcp_death_row.sysctl_tw_recycle && |
1da177e4 LT |
203 | !tp->rx_opt.ts_recent_stamp && rt->rt_dst == daddr) { |
204 | struct inet_peer *peer = rt_get_peer(rt); | |
7174259e ACM |
205 | /* |
206 | * VJ's idea. We save last timestamp seen from | |
207 | * the destination in peer table, when entering state | |
208 | * TIME-WAIT * and initialize rx_opt.ts_recent from it, | |
209 | * when trying new connection. | |
1da177e4 | 210 | */ |
7174259e | 211 | if (peer != NULL && |
9d729f72 | 212 | peer->tcp_ts_stamp + TCP_PAWS_MSL >= get_seconds()) { |
1da177e4 LT |
213 | tp->rx_opt.ts_recent_stamp = peer->tcp_ts_stamp; |
214 | tp->rx_opt.ts_recent = peer->tcp_ts; | |
215 | } | |
216 | } | |
217 | ||
218 | inet->dport = usin->sin_port; | |
219 | inet->daddr = daddr; | |
220 | ||
d83d8461 | 221 | inet_csk(sk)->icsk_ext_hdr_len = 0; |
1da177e4 | 222 | if (inet->opt) |
d83d8461 | 223 | inet_csk(sk)->icsk_ext_hdr_len = inet->opt->optlen; |
1da177e4 LT |
224 | |
225 | tp->rx_opt.mss_clamp = 536; | |
226 | ||
227 | /* Socket identity is still unknown (sport may be zero). | |
228 | * However we set state to SYN-SENT and not releasing socket | |
229 | * lock select source port, enter ourselves into the hash tables and | |
230 | * complete initialization after this. | |
231 | */ | |
232 | tcp_set_state(sk, TCP_SYN_SENT); | |
a7f5e7f1 | 233 | err = inet_hash_connect(&tcp_death_row, sk); |
1da177e4 LT |
234 | if (err) |
235 | goto failure; | |
236 | ||
7174259e ACM |
237 | err = ip_route_newports(&rt, IPPROTO_TCP, |
238 | inet->sport, inet->dport, sk); | |
1da177e4 LT |
239 | if (err) |
240 | goto failure; | |
241 | ||
242 | /* OK, now commit destination to socket. */ | |
bcd76111 | 243 | sk->sk_gso_type = SKB_GSO_TCPV4; |
6cbb0df7 | 244 | sk_setup_caps(sk, &rt->u.dst); |
1da177e4 LT |
245 | |
246 | if (!tp->write_seq) | |
247 | tp->write_seq = secure_tcp_sequence_number(inet->saddr, | |
248 | inet->daddr, | |
249 | inet->sport, | |
250 | usin->sin_port); | |
251 | ||
252 | inet->id = tp->write_seq ^ jiffies; | |
253 | ||
254 | err = tcp_connect(sk); | |
255 | rt = NULL; | |
256 | if (err) | |
257 | goto failure; | |
258 | ||
259 | return 0; | |
260 | ||
261 | failure: | |
7174259e ACM |
262 | /* |
263 | * This unhashes the socket and releases the local port, | |
264 | * if necessary. | |
265 | */ | |
1da177e4 LT |
266 | tcp_set_state(sk, TCP_CLOSE); |
267 | ip_rt_put(rt); | |
268 | sk->sk_route_caps = 0; | |
269 | inet->dport = 0; | |
270 | return err; | |
271 | } | |
272 | ||
1da177e4 LT |
273 | /* |
274 | * This routine does path mtu discovery as defined in RFC1191. | |
275 | */ | |
40efc6fa | 276 | static void do_pmtu_discovery(struct sock *sk, struct iphdr *iph, u32 mtu) |
1da177e4 LT |
277 | { |
278 | struct dst_entry *dst; | |
279 | struct inet_sock *inet = inet_sk(sk); | |
1da177e4 LT |
280 | |
281 | /* We are not interested in TCP_LISTEN and open_requests (SYN-ACKs | |
282 | * send out by Linux are always <576bytes so they should go through | |
283 | * unfragmented). | |
284 | */ | |
285 | if (sk->sk_state == TCP_LISTEN) | |
286 | return; | |
287 | ||
288 | /* We don't check in the destentry if pmtu discovery is forbidden | |
289 | * on this route. We just assume that no packet_to_big packets | |
290 | * are send back when pmtu discovery is not active. | |
e905a9ed | 291 | * There is a small race when the user changes this flag in the |
1da177e4 LT |
292 | * route, but I think that's acceptable. |
293 | */ | |
294 | if ((dst = __sk_dst_check(sk, 0)) == NULL) | |
295 | return; | |
296 | ||
297 | dst->ops->update_pmtu(dst, mtu); | |
298 | ||
299 | /* Something is about to be wrong... Remember soft error | |
300 | * for the case, if this connection will not able to recover. | |
301 | */ | |
302 | if (mtu < dst_mtu(dst) && ip_dont_fragment(sk, dst)) | |
303 | sk->sk_err_soft = EMSGSIZE; | |
304 | ||
305 | mtu = dst_mtu(dst); | |
306 | ||
307 | if (inet->pmtudisc != IP_PMTUDISC_DONT && | |
d83d8461 | 308 | inet_csk(sk)->icsk_pmtu_cookie > mtu) { |
1da177e4 LT |
309 | tcp_sync_mss(sk, mtu); |
310 | ||
311 | /* Resend the TCP packet because it's | |
312 | * clear that the old packet has been | |
313 | * dropped. This is the new "fast" path mtu | |
314 | * discovery. | |
315 | */ | |
316 | tcp_simple_retransmit(sk); | |
317 | } /* else let the usual retransmit timer handle it */ | |
318 | } | |
319 | ||
320 | /* | |
321 | * This routine is called by the ICMP module when it gets some | |
322 | * sort of error condition. If err < 0 then the socket should | |
323 | * be closed and the error returned to the user. If err > 0 | |
324 | * it's just the icmp type << 8 | icmp code. After adjustment | |
325 | * header points to the first 8 bytes of the tcp header. We need | |
326 | * to find the appropriate port. | |
327 | * | |
328 | * The locking strategy used here is very "optimistic". When | |
329 | * someone else accesses the socket the ICMP is just dropped | |
330 | * and for some paths there is no check at all. | |
331 | * A more general error queue to queue errors for later handling | |
332 | * is probably better. | |
333 | * | |
334 | */ | |
335 | ||
336 | void tcp_v4_err(struct sk_buff *skb, u32 info) | |
337 | { | |
338 | struct iphdr *iph = (struct iphdr *)skb->data; | |
339 | struct tcphdr *th = (struct tcphdr *)(skb->data + (iph->ihl << 2)); | |
340 | struct tcp_sock *tp; | |
341 | struct inet_sock *inet; | |
88c7664f ACM |
342 | const int type = icmp_hdr(skb)->type; |
343 | const int code = icmp_hdr(skb)->code; | |
1da177e4 LT |
344 | struct sock *sk; |
345 | __u32 seq; | |
346 | int err; | |
347 | ||
348 | if (skb->len < (iph->ihl << 2) + 8) { | |
349 | ICMP_INC_STATS_BH(ICMP_MIB_INERRORS); | |
350 | return; | |
351 | } | |
352 | ||
c346dca1 | 353 | sk = inet_lookup(dev_net(skb->dev), &tcp_hashinfo, iph->daddr, th->dest, |
c67499c0 | 354 | iph->saddr, th->source, inet_iif(skb)); |
1da177e4 LT |
355 | if (!sk) { |
356 | ICMP_INC_STATS_BH(ICMP_MIB_INERRORS); | |
357 | return; | |
358 | } | |
359 | if (sk->sk_state == TCP_TIME_WAIT) { | |
9469c7b4 | 360 | inet_twsk_put(inet_twsk(sk)); |
1da177e4 LT |
361 | return; |
362 | } | |
363 | ||
364 | bh_lock_sock(sk); | |
365 | /* If too many ICMPs get dropped on busy | |
366 | * servers this needs to be solved differently. | |
367 | */ | |
368 | if (sock_owned_by_user(sk)) | |
369 | NET_INC_STATS_BH(LINUX_MIB_LOCKDROPPEDICMPS); | |
370 | ||
371 | if (sk->sk_state == TCP_CLOSE) | |
372 | goto out; | |
373 | ||
374 | tp = tcp_sk(sk); | |
375 | seq = ntohl(th->seq); | |
376 | if (sk->sk_state != TCP_LISTEN && | |
377 | !between(seq, tp->snd_una, tp->snd_nxt)) { | |
06ca719f | 378 | NET_INC_STATS_BH(LINUX_MIB_OUTOFWINDOWICMPS); |
1da177e4 LT |
379 | goto out; |
380 | } | |
381 | ||
382 | switch (type) { | |
383 | case ICMP_SOURCE_QUENCH: | |
384 | /* Just silently ignore these. */ | |
385 | goto out; | |
386 | case ICMP_PARAMETERPROB: | |
387 | err = EPROTO; | |
388 | break; | |
389 | case ICMP_DEST_UNREACH: | |
390 | if (code > NR_ICMP_UNREACH) | |
391 | goto out; | |
392 | ||
393 | if (code == ICMP_FRAG_NEEDED) { /* PMTU discovery (RFC1191) */ | |
394 | if (!sock_owned_by_user(sk)) | |
395 | do_pmtu_discovery(sk, iph, info); | |
396 | goto out; | |
397 | } | |
398 | ||
399 | err = icmp_err_convert[code].errno; | |
400 | break; | |
401 | case ICMP_TIME_EXCEEDED: | |
402 | err = EHOSTUNREACH; | |
403 | break; | |
404 | default: | |
405 | goto out; | |
406 | } | |
407 | ||
408 | switch (sk->sk_state) { | |
60236fdd | 409 | struct request_sock *req, **prev; |
1da177e4 LT |
410 | case TCP_LISTEN: |
411 | if (sock_owned_by_user(sk)) | |
412 | goto out; | |
413 | ||
463c84b9 ACM |
414 | req = inet_csk_search_req(sk, &prev, th->dest, |
415 | iph->daddr, iph->saddr); | |
1da177e4 LT |
416 | if (!req) |
417 | goto out; | |
418 | ||
419 | /* ICMPs are not backlogged, hence we cannot get | |
420 | an established socket here. | |
421 | */ | |
422 | BUG_TRAP(!req->sk); | |
423 | ||
2e6599cb | 424 | if (seq != tcp_rsk(req)->snt_isn) { |
1da177e4 LT |
425 | NET_INC_STATS_BH(LINUX_MIB_OUTOFWINDOWICMPS); |
426 | goto out; | |
427 | } | |
428 | ||
429 | /* | |
430 | * Still in SYN_RECV, just remove it silently. | |
431 | * There is no good way to pass the error to the newly | |
432 | * created socket, and POSIX does not want network | |
433 | * errors returned from accept(). | |
434 | */ | |
463c84b9 | 435 | inet_csk_reqsk_queue_drop(sk, req, prev); |
1da177e4 LT |
436 | goto out; |
437 | ||
438 | case TCP_SYN_SENT: | |
439 | case TCP_SYN_RECV: /* Cannot happen. | |
440 | It can f.e. if SYNs crossed. | |
441 | */ | |
442 | if (!sock_owned_by_user(sk)) { | |
1da177e4 LT |
443 | sk->sk_err = err; |
444 | ||
445 | sk->sk_error_report(sk); | |
446 | ||
447 | tcp_done(sk); | |
448 | } else { | |
449 | sk->sk_err_soft = err; | |
450 | } | |
451 | goto out; | |
452 | } | |
453 | ||
454 | /* If we've already connected we will keep trying | |
455 | * until we time out, or the user gives up. | |
456 | * | |
457 | * rfc1122 4.2.3.9 allows to consider as hard errors | |
458 | * only PROTO_UNREACH and PORT_UNREACH (well, FRAG_FAILED too, | |
459 | * but it is obsoleted by pmtu discovery). | |
460 | * | |
461 | * Note, that in modern internet, where routing is unreliable | |
462 | * and in each dark corner broken firewalls sit, sending random | |
463 | * errors ordered by their masters even this two messages finally lose | |
464 | * their original sense (even Linux sends invalid PORT_UNREACHs) | |
465 | * | |
466 | * Now we are in compliance with RFCs. | |
467 | * --ANK (980905) | |
468 | */ | |
469 | ||
470 | inet = inet_sk(sk); | |
471 | if (!sock_owned_by_user(sk) && inet->recverr) { | |
472 | sk->sk_err = err; | |
473 | sk->sk_error_report(sk); | |
474 | } else { /* Only an error on timeout */ | |
475 | sk->sk_err_soft = err; | |
476 | } | |
477 | ||
478 | out: | |
479 | bh_unlock_sock(sk); | |
480 | sock_put(sk); | |
481 | } | |
482 | ||
483 | /* This routine computes an IPv4 TCP checksum. */ | |
8292a17a | 484 | void tcp_v4_send_check(struct sock *sk, int len, struct sk_buff *skb) |
1da177e4 LT |
485 | { |
486 | struct inet_sock *inet = inet_sk(sk); | |
aa8223c7 | 487 | struct tcphdr *th = tcp_hdr(skb); |
1da177e4 | 488 | |
84fa7933 | 489 | if (skb->ip_summed == CHECKSUM_PARTIAL) { |
ba7808ea FD |
490 | th->check = ~tcp_v4_check(len, inet->saddr, |
491 | inet->daddr, 0); | |
663ead3b | 492 | skb->csum_start = skb_transport_header(skb) - skb->head; |
ff1dcadb | 493 | skb->csum_offset = offsetof(struct tcphdr, check); |
1da177e4 | 494 | } else { |
ba7808ea | 495 | th->check = tcp_v4_check(len, inet->saddr, inet->daddr, |
1da177e4 LT |
496 | csum_partial((char *)th, |
497 | th->doff << 2, | |
498 | skb->csum)); | |
499 | } | |
500 | } | |
501 | ||
a430a43d HX |
502 | int tcp_v4_gso_send_check(struct sk_buff *skb) |
503 | { | |
eddc9ec5 | 504 | const struct iphdr *iph; |
a430a43d HX |
505 | struct tcphdr *th; |
506 | ||
507 | if (!pskb_may_pull(skb, sizeof(*th))) | |
508 | return -EINVAL; | |
509 | ||
eddc9ec5 | 510 | iph = ip_hdr(skb); |
aa8223c7 | 511 | th = tcp_hdr(skb); |
a430a43d HX |
512 | |
513 | th->check = 0; | |
ba7808ea | 514 | th->check = ~tcp_v4_check(skb->len, iph->saddr, iph->daddr, 0); |
663ead3b | 515 | skb->csum_start = skb_transport_header(skb) - skb->head; |
ff1dcadb | 516 | skb->csum_offset = offsetof(struct tcphdr, check); |
84fa7933 | 517 | skb->ip_summed = CHECKSUM_PARTIAL; |
a430a43d HX |
518 | return 0; |
519 | } | |
520 | ||
1da177e4 LT |
521 | /* |
522 | * This routine will send an RST to the other tcp. | |
523 | * | |
524 | * Someone asks: why I NEVER use socket parameters (TOS, TTL etc.) | |
525 | * for reset. | |
526 | * Answer: if a packet caused RST, it is not for a socket | |
527 | * existing in our system, if it is matched to a socket, | |
528 | * it is just duplicate segment or bug in other side's TCP. | |
529 | * So that we build reply only basing on parameters | |
530 | * arrived with segment. | |
531 | * Exception: precedence violation. We do not implement it in any case. | |
532 | */ | |
533 | ||
cfb6eeb4 | 534 | static void tcp_v4_send_reset(struct sock *sk, struct sk_buff *skb) |
1da177e4 | 535 | { |
aa8223c7 | 536 | struct tcphdr *th = tcp_hdr(skb); |
cfb6eeb4 YH |
537 | struct { |
538 | struct tcphdr th; | |
539 | #ifdef CONFIG_TCP_MD5SIG | |
714e85be | 540 | __be32 opt[(TCPOLEN_MD5SIG_ALIGNED >> 2)]; |
cfb6eeb4 YH |
541 | #endif |
542 | } rep; | |
1da177e4 | 543 | struct ip_reply_arg arg; |
cfb6eeb4 YH |
544 | #ifdef CONFIG_TCP_MD5SIG |
545 | struct tcp_md5sig_key *key; | |
546 | #endif | |
1da177e4 LT |
547 | |
548 | /* Never send a reset in response to a reset. */ | |
549 | if (th->rst) | |
550 | return; | |
551 | ||
ee6b9673 | 552 | if (skb->rtable->rt_type != RTN_LOCAL) |
1da177e4 LT |
553 | return; |
554 | ||
555 | /* Swap the send and the receive. */ | |
cfb6eeb4 YH |
556 | memset(&rep, 0, sizeof(rep)); |
557 | rep.th.dest = th->source; | |
558 | rep.th.source = th->dest; | |
559 | rep.th.doff = sizeof(struct tcphdr) / 4; | |
560 | rep.th.rst = 1; | |
1da177e4 LT |
561 | |
562 | if (th->ack) { | |
cfb6eeb4 | 563 | rep.th.seq = th->ack_seq; |
1da177e4 | 564 | } else { |
cfb6eeb4 YH |
565 | rep.th.ack = 1; |
566 | rep.th.ack_seq = htonl(ntohl(th->seq) + th->syn + th->fin + | |
567 | skb->len - (th->doff << 2)); | |
1da177e4 LT |
568 | } |
569 | ||
7174259e | 570 | memset(&arg, 0, sizeof(arg)); |
cfb6eeb4 YH |
571 | arg.iov[0].iov_base = (unsigned char *)&rep; |
572 | arg.iov[0].iov_len = sizeof(rep.th); | |
573 | ||
574 | #ifdef CONFIG_TCP_MD5SIG | |
eddc9ec5 | 575 | key = sk ? tcp_v4_md5_do_lookup(sk, ip_hdr(skb)->daddr) : NULL; |
cfb6eeb4 YH |
576 | if (key) { |
577 | rep.opt[0] = htonl((TCPOPT_NOP << 24) | | |
578 | (TCPOPT_NOP << 16) | | |
579 | (TCPOPT_MD5SIG << 8) | | |
580 | TCPOLEN_MD5SIG); | |
581 | /* Update length and the length the header thinks exists */ | |
582 | arg.iov[0].iov_len += TCPOLEN_MD5SIG_ALIGNED; | |
583 | rep.th.doff = arg.iov[0].iov_len / 4; | |
584 | ||
585 | tcp_v4_do_calc_md5_hash((__u8 *)&rep.opt[1], | |
586 | key, | |
eddc9ec5 ACM |
587 | ip_hdr(skb)->daddr, |
588 | ip_hdr(skb)->saddr, | |
cfb6eeb4 YH |
589 | &rep.th, IPPROTO_TCP, |
590 | arg.iov[0].iov_len); | |
591 | } | |
592 | #endif | |
eddc9ec5 ACM |
593 | arg.csum = csum_tcpudp_nofold(ip_hdr(skb)->daddr, |
594 | ip_hdr(skb)->saddr, /* XXX */ | |
1da177e4 LT |
595 | sizeof(struct tcphdr), IPPROTO_TCP, 0); |
596 | arg.csumoffset = offsetof(struct tcphdr, check) / 2; | |
597 | ||
7feb49c8 DL |
598 | ip_send_reply(dev_net(skb->dst->dev)->ipv4.tcp_sock, skb, |
599 | &arg, arg.iov[0].iov_len); | |
1da177e4 LT |
600 | |
601 | TCP_INC_STATS_BH(TCP_MIB_OUTSEGS); | |
602 | TCP_INC_STATS_BH(TCP_MIB_OUTRSTS); | |
603 | } | |
604 | ||
605 | /* The code following below sending ACKs in SYN-RECV and TIME-WAIT states | |
606 | outside socket context is ugly, certainly. What can I do? | |
607 | */ | |
608 | ||
cfb6eeb4 YH |
609 | static void tcp_v4_send_ack(struct tcp_timewait_sock *twsk, |
610 | struct sk_buff *skb, u32 seq, u32 ack, | |
1da177e4 LT |
611 | u32 win, u32 ts) |
612 | { | |
aa8223c7 | 613 | struct tcphdr *th = tcp_hdr(skb); |
1da177e4 LT |
614 | struct { |
615 | struct tcphdr th; | |
714e85be | 616 | __be32 opt[(TCPOLEN_TSTAMP_ALIGNED >> 2) |
cfb6eeb4 | 617 | #ifdef CONFIG_TCP_MD5SIG |
714e85be | 618 | + (TCPOLEN_MD5SIG_ALIGNED >> 2) |
cfb6eeb4 YH |
619 | #endif |
620 | ]; | |
1da177e4 LT |
621 | } rep; |
622 | struct ip_reply_arg arg; | |
cfb6eeb4 YH |
623 | #ifdef CONFIG_TCP_MD5SIG |
624 | struct tcp_md5sig_key *key; | |
625 | struct tcp_md5sig_key tw_key; | |
626 | #endif | |
1da177e4 LT |
627 | |
628 | memset(&rep.th, 0, sizeof(struct tcphdr)); | |
7174259e | 629 | memset(&arg, 0, sizeof(arg)); |
1da177e4 LT |
630 | |
631 | arg.iov[0].iov_base = (unsigned char *)&rep; | |
632 | arg.iov[0].iov_len = sizeof(rep.th); | |
633 | if (ts) { | |
cfb6eeb4 YH |
634 | rep.opt[0] = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) | |
635 | (TCPOPT_TIMESTAMP << 8) | | |
636 | TCPOLEN_TIMESTAMP); | |
637 | rep.opt[1] = htonl(tcp_time_stamp); | |
638 | rep.opt[2] = htonl(ts); | |
cb48cfe8 | 639 | arg.iov[0].iov_len += TCPOLEN_TSTAMP_ALIGNED; |
1da177e4 LT |
640 | } |
641 | ||
642 | /* Swap the send and the receive. */ | |
643 | rep.th.dest = th->source; | |
644 | rep.th.source = th->dest; | |
645 | rep.th.doff = arg.iov[0].iov_len / 4; | |
646 | rep.th.seq = htonl(seq); | |
647 | rep.th.ack_seq = htonl(ack); | |
648 | rep.th.ack = 1; | |
649 | rep.th.window = htons(win); | |
650 | ||
cfb6eeb4 YH |
651 | #ifdef CONFIG_TCP_MD5SIG |
652 | /* | |
653 | * The SKB holds an imcoming packet, but may not have a valid ->sk | |
654 | * pointer. This is especially the case when we're dealing with a | |
655 | * TIME_WAIT ack, because the sk structure is long gone, and only | |
656 | * the tcp_timewait_sock remains. So the md5 key is stashed in that | |
657 | * structure, and we use it in preference. I believe that (twsk || | |
658 | * skb->sk) holds true, but we program defensively. | |
659 | */ | |
660 | if (!twsk && skb->sk) { | |
eddc9ec5 | 661 | key = tcp_v4_md5_do_lookup(skb->sk, ip_hdr(skb)->daddr); |
cfb6eeb4 YH |
662 | } else if (twsk && twsk->tw_md5_keylen) { |
663 | tw_key.key = twsk->tw_md5_key; | |
664 | tw_key.keylen = twsk->tw_md5_keylen; | |
665 | key = &tw_key; | |
7174259e | 666 | } else |
cfb6eeb4 | 667 | key = NULL; |
cfb6eeb4 YH |
668 | |
669 | if (key) { | |
670 | int offset = (ts) ? 3 : 0; | |
671 | ||
672 | rep.opt[offset++] = htonl((TCPOPT_NOP << 24) | | |
673 | (TCPOPT_NOP << 16) | | |
674 | (TCPOPT_MD5SIG << 8) | | |
675 | TCPOLEN_MD5SIG); | |
676 | arg.iov[0].iov_len += TCPOLEN_MD5SIG_ALIGNED; | |
677 | rep.th.doff = arg.iov[0].iov_len/4; | |
678 | ||
679 | tcp_v4_do_calc_md5_hash((__u8 *)&rep.opt[offset], | |
680 | key, | |
eddc9ec5 ACM |
681 | ip_hdr(skb)->daddr, |
682 | ip_hdr(skb)->saddr, | |
cfb6eeb4 YH |
683 | &rep.th, IPPROTO_TCP, |
684 | arg.iov[0].iov_len); | |
685 | } | |
686 | #endif | |
eddc9ec5 ACM |
687 | arg.csum = csum_tcpudp_nofold(ip_hdr(skb)->daddr, |
688 | ip_hdr(skb)->saddr, /* XXX */ | |
1da177e4 LT |
689 | arg.iov[0].iov_len, IPPROTO_TCP, 0); |
690 | arg.csumoffset = offsetof(struct tcphdr, check) / 2; | |
f0e48dbf PM |
691 | if (twsk) |
692 | arg.bound_dev_if = twsk->tw_sk.tw_bound_dev_if; | |
1da177e4 | 693 | |
7feb49c8 DL |
694 | ip_send_reply(dev_net(skb->dev)->ipv4.tcp_sock, skb, |
695 | &arg, arg.iov[0].iov_len); | |
1da177e4 LT |
696 | |
697 | TCP_INC_STATS_BH(TCP_MIB_OUTSEGS); | |
698 | } | |
699 | ||
700 | static void tcp_v4_timewait_ack(struct sock *sk, struct sk_buff *skb) | |
701 | { | |
8feaf0c0 | 702 | struct inet_timewait_sock *tw = inet_twsk(sk); |
cfb6eeb4 | 703 | struct tcp_timewait_sock *tcptw = tcp_twsk(sk); |
1da177e4 | 704 | |
cfb6eeb4 | 705 | tcp_v4_send_ack(tcptw, skb, tcptw->tw_snd_nxt, tcptw->tw_rcv_nxt, |
7174259e ACM |
706 | tcptw->tw_rcv_wnd >> tw->tw_rcv_wscale, |
707 | tcptw->tw_ts_recent); | |
1da177e4 | 708 | |
8feaf0c0 | 709 | inet_twsk_put(tw); |
1da177e4 LT |
710 | } |
711 | ||
7174259e ACM |
712 | static void tcp_v4_reqsk_send_ack(struct sk_buff *skb, |
713 | struct request_sock *req) | |
1da177e4 | 714 | { |
cfb6eeb4 YH |
715 | tcp_v4_send_ack(NULL, skb, tcp_rsk(req)->snt_isn + 1, |
716 | tcp_rsk(req)->rcv_isn + 1, req->rcv_wnd, | |
1da177e4 LT |
717 | req->ts_recent); |
718 | } | |
719 | ||
1da177e4 | 720 | /* |
9bf1d83e | 721 | * Send a SYN-ACK after having received a SYN. |
60236fdd | 722 | * This still operates on a request_sock only, not on a big |
1da177e4 LT |
723 | * socket. |
724 | */ | |
fd80eb94 DL |
725 | static int __tcp_v4_send_synack(struct sock *sk, struct request_sock *req, |
726 | struct dst_entry *dst) | |
1da177e4 | 727 | { |
2e6599cb | 728 | const struct inet_request_sock *ireq = inet_rsk(req); |
1da177e4 LT |
729 | int err = -1; |
730 | struct sk_buff * skb; | |
731 | ||
732 | /* First, grab a route. */ | |
463c84b9 | 733 | if (!dst && (dst = inet_csk_route_req(sk, req)) == NULL) |
fd80eb94 | 734 | return -1; |
1da177e4 LT |
735 | |
736 | skb = tcp_make_synack(sk, dst, req); | |
737 | ||
738 | if (skb) { | |
aa8223c7 | 739 | struct tcphdr *th = tcp_hdr(skb); |
1da177e4 | 740 | |
ba7808ea | 741 | th->check = tcp_v4_check(skb->len, |
2e6599cb ACM |
742 | ireq->loc_addr, |
743 | ireq->rmt_addr, | |
1da177e4 LT |
744 | csum_partial((char *)th, skb->len, |
745 | skb->csum)); | |
746 | ||
2e6599cb ACM |
747 | err = ip_build_and_send_pkt(skb, sk, ireq->loc_addr, |
748 | ireq->rmt_addr, | |
749 | ireq->opt); | |
b9df3cb8 | 750 | err = net_xmit_eval(err); |
1da177e4 LT |
751 | } |
752 | ||
1da177e4 LT |
753 | dst_release(dst); |
754 | return err; | |
755 | } | |
756 | ||
fd80eb94 DL |
757 | static int tcp_v4_send_synack(struct sock *sk, struct request_sock *req) |
758 | { | |
759 | return __tcp_v4_send_synack(sk, req, NULL); | |
760 | } | |
761 | ||
1da177e4 | 762 | /* |
60236fdd | 763 | * IPv4 request_sock destructor. |
1da177e4 | 764 | */ |
60236fdd | 765 | static void tcp_v4_reqsk_destructor(struct request_sock *req) |
1da177e4 | 766 | { |
a51482bd | 767 | kfree(inet_rsk(req)->opt); |
1da177e4 LT |
768 | } |
769 | ||
80e40daa | 770 | #ifdef CONFIG_SYN_COOKIES |
40efc6fa | 771 | static void syn_flood_warning(struct sk_buff *skb) |
1da177e4 LT |
772 | { |
773 | static unsigned long warntime; | |
774 | ||
775 | if (time_after(jiffies, (warntime + HZ * 60))) { | |
776 | warntime = jiffies; | |
777 | printk(KERN_INFO | |
778 | "possible SYN flooding on port %d. Sending cookies.\n", | |
aa8223c7 | 779 | ntohs(tcp_hdr(skb)->dest)); |
1da177e4 LT |
780 | } |
781 | } | |
80e40daa | 782 | #endif |
1da177e4 LT |
783 | |
784 | /* | |
60236fdd | 785 | * Save and compile IPv4 options into the request_sock if needed. |
1da177e4 | 786 | */ |
40efc6fa SH |
787 | static struct ip_options *tcp_v4_save_options(struct sock *sk, |
788 | struct sk_buff *skb) | |
1da177e4 LT |
789 | { |
790 | struct ip_options *opt = &(IPCB(skb)->opt); | |
791 | struct ip_options *dopt = NULL; | |
792 | ||
793 | if (opt && opt->optlen) { | |
794 | int opt_size = optlength(opt); | |
795 | dopt = kmalloc(opt_size, GFP_ATOMIC); | |
796 | if (dopt) { | |
797 | if (ip_options_echo(dopt, skb)) { | |
798 | kfree(dopt); | |
799 | dopt = NULL; | |
800 | } | |
801 | } | |
802 | } | |
803 | return dopt; | |
804 | } | |
805 | ||
cfb6eeb4 YH |
806 | #ifdef CONFIG_TCP_MD5SIG |
807 | /* | |
808 | * RFC2385 MD5 checksumming requires a mapping of | |
809 | * IP address->MD5 Key. | |
810 | * We need to maintain these in the sk structure. | |
811 | */ | |
812 | ||
813 | /* Find the Key structure for an address. */ | |
7174259e ACM |
814 | static struct tcp_md5sig_key * |
815 | tcp_v4_md5_do_lookup(struct sock *sk, __be32 addr) | |
cfb6eeb4 YH |
816 | { |
817 | struct tcp_sock *tp = tcp_sk(sk); | |
818 | int i; | |
819 | ||
820 | if (!tp->md5sig_info || !tp->md5sig_info->entries4) | |
821 | return NULL; | |
822 | for (i = 0; i < tp->md5sig_info->entries4; i++) { | |
823 | if (tp->md5sig_info->keys4[i].addr == addr) | |
f8ab18d2 | 824 | return &tp->md5sig_info->keys4[i].base; |
cfb6eeb4 YH |
825 | } |
826 | return NULL; | |
827 | } | |
828 | ||
829 | struct tcp_md5sig_key *tcp_v4_md5_lookup(struct sock *sk, | |
830 | struct sock *addr_sk) | |
831 | { | |
832 | return tcp_v4_md5_do_lookup(sk, inet_sk(addr_sk)->daddr); | |
833 | } | |
834 | ||
835 | EXPORT_SYMBOL(tcp_v4_md5_lookup); | |
836 | ||
f5b99bcd AB |
837 | static struct tcp_md5sig_key *tcp_v4_reqsk_md5_lookup(struct sock *sk, |
838 | struct request_sock *req) | |
cfb6eeb4 YH |
839 | { |
840 | return tcp_v4_md5_do_lookup(sk, inet_rsk(req)->rmt_addr); | |
841 | } | |
842 | ||
843 | /* This can be called on a newly created socket, from other files */ | |
844 | int tcp_v4_md5_do_add(struct sock *sk, __be32 addr, | |
845 | u8 *newkey, u8 newkeylen) | |
846 | { | |
847 | /* Add Key to the list */ | |
b0a713e9 | 848 | struct tcp_md5sig_key *key; |
cfb6eeb4 YH |
849 | struct tcp_sock *tp = tcp_sk(sk); |
850 | struct tcp4_md5sig_key *keys; | |
851 | ||
b0a713e9 | 852 | key = tcp_v4_md5_do_lookup(sk, addr); |
cfb6eeb4 YH |
853 | if (key) { |
854 | /* Pre-existing entry - just update that one. */ | |
b0a713e9 MD |
855 | kfree(key->key); |
856 | key->key = newkey; | |
857 | key->keylen = newkeylen; | |
cfb6eeb4 | 858 | } else { |
f6685938 ACM |
859 | struct tcp_md5sig_info *md5sig; |
860 | ||
cfb6eeb4 | 861 | if (!tp->md5sig_info) { |
f6685938 ACM |
862 | tp->md5sig_info = kzalloc(sizeof(*tp->md5sig_info), |
863 | GFP_ATOMIC); | |
cfb6eeb4 YH |
864 | if (!tp->md5sig_info) { |
865 | kfree(newkey); | |
866 | return -ENOMEM; | |
867 | } | |
3d7dbeac | 868 | sk->sk_route_caps &= ~NETIF_F_GSO_MASK; |
cfb6eeb4 YH |
869 | } |
870 | if (tcp_alloc_md5sig_pool() == NULL) { | |
871 | kfree(newkey); | |
872 | return -ENOMEM; | |
873 | } | |
f6685938 ACM |
874 | md5sig = tp->md5sig_info; |
875 | ||
876 | if (md5sig->alloced4 == md5sig->entries4) { | |
877 | keys = kmalloc((sizeof(*keys) * | |
e905a9ed | 878 | (md5sig->entries4 + 1)), GFP_ATOMIC); |
cfb6eeb4 YH |
879 | if (!keys) { |
880 | kfree(newkey); | |
881 | tcp_free_md5sig_pool(); | |
882 | return -ENOMEM; | |
883 | } | |
884 | ||
f6685938 ACM |
885 | if (md5sig->entries4) |
886 | memcpy(keys, md5sig->keys4, | |
887 | sizeof(*keys) * md5sig->entries4); | |
cfb6eeb4 YH |
888 | |
889 | /* Free old key list, and reference new one */ | |
a80cc20d | 890 | kfree(md5sig->keys4); |
f6685938 ACM |
891 | md5sig->keys4 = keys; |
892 | md5sig->alloced4++; | |
cfb6eeb4 | 893 | } |
f6685938 | 894 | md5sig->entries4++; |
f8ab18d2 DM |
895 | md5sig->keys4[md5sig->entries4 - 1].addr = addr; |
896 | md5sig->keys4[md5sig->entries4 - 1].base.key = newkey; | |
897 | md5sig->keys4[md5sig->entries4 - 1].base.keylen = newkeylen; | |
cfb6eeb4 YH |
898 | } |
899 | return 0; | |
900 | } | |
901 | ||
902 | EXPORT_SYMBOL(tcp_v4_md5_do_add); | |
903 | ||
904 | static int tcp_v4_md5_add_func(struct sock *sk, struct sock *addr_sk, | |
905 | u8 *newkey, u8 newkeylen) | |
906 | { | |
907 | return tcp_v4_md5_do_add(sk, inet_sk(addr_sk)->daddr, | |
908 | newkey, newkeylen); | |
909 | } | |
910 | ||
911 | int tcp_v4_md5_do_del(struct sock *sk, __be32 addr) | |
912 | { | |
913 | struct tcp_sock *tp = tcp_sk(sk); | |
914 | int i; | |
915 | ||
916 | for (i = 0; i < tp->md5sig_info->entries4; i++) { | |
917 | if (tp->md5sig_info->keys4[i].addr == addr) { | |
918 | /* Free the key */ | |
f8ab18d2 | 919 | kfree(tp->md5sig_info->keys4[i].base.key); |
cfb6eeb4 YH |
920 | tp->md5sig_info->entries4--; |
921 | ||
922 | if (tp->md5sig_info->entries4 == 0) { | |
923 | kfree(tp->md5sig_info->keys4); | |
924 | tp->md5sig_info->keys4 = NULL; | |
8228a18d | 925 | tp->md5sig_info->alloced4 = 0; |
7174259e | 926 | } else if (tp->md5sig_info->entries4 != i) { |
cfb6eeb4 | 927 | /* Need to do some manipulation */ |
354faf09 YH |
928 | memmove(&tp->md5sig_info->keys4[i], |
929 | &tp->md5sig_info->keys4[i+1], | |
930 | (tp->md5sig_info->entries4 - i) * | |
931 | sizeof(struct tcp4_md5sig_key)); | |
cfb6eeb4 YH |
932 | } |
933 | tcp_free_md5sig_pool(); | |
934 | return 0; | |
935 | } | |
936 | } | |
937 | return -ENOENT; | |
938 | } | |
939 | ||
940 | EXPORT_SYMBOL(tcp_v4_md5_do_del); | |
941 | ||
7174259e | 942 | static void tcp_v4_clear_md5_list(struct sock *sk) |
cfb6eeb4 YH |
943 | { |
944 | struct tcp_sock *tp = tcp_sk(sk); | |
945 | ||
946 | /* Free each key, then the set of key keys, | |
947 | * the crypto element, and then decrement our | |
948 | * hold on the last resort crypto. | |
949 | */ | |
950 | if (tp->md5sig_info->entries4) { | |
951 | int i; | |
952 | for (i = 0; i < tp->md5sig_info->entries4; i++) | |
f8ab18d2 | 953 | kfree(tp->md5sig_info->keys4[i].base.key); |
cfb6eeb4 YH |
954 | tp->md5sig_info->entries4 = 0; |
955 | tcp_free_md5sig_pool(); | |
956 | } | |
957 | if (tp->md5sig_info->keys4) { | |
958 | kfree(tp->md5sig_info->keys4); | |
959 | tp->md5sig_info->keys4 = NULL; | |
960 | tp->md5sig_info->alloced4 = 0; | |
961 | } | |
962 | } | |
963 | ||
7174259e ACM |
964 | static int tcp_v4_parse_md5_keys(struct sock *sk, char __user *optval, |
965 | int optlen) | |
cfb6eeb4 YH |
966 | { |
967 | struct tcp_md5sig cmd; | |
968 | struct sockaddr_in *sin = (struct sockaddr_in *)&cmd.tcpm_addr; | |
969 | u8 *newkey; | |
970 | ||
971 | if (optlen < sizeof(cmd)) | |
972 | return -EINVAL; | |
973 | ||
7174259e | 974 | if (copy_from_user(&cmd, optval, sizeof(cmd))) |
cfb6eeb4 YH |
975 | return -EFAULT; |
976 | ||
977 | if (sin->sin_family != AF_INET) | |
978 | return -EINVAL; | |
979 | ||
980 | if (!cmd.tcpm_key || !cmd.tcpm_keylen) { | |
981 | if (!tcp_sk(sk)->md5sig_info) | |
982 | return -ENOENT; | |
983 | return tcp_v4_md5_do_del(sk, sin->sin_addr.s_addr); | |
984 | } | |
985 | ||
986 | if (cmd.tcpm_keylen > TCP_MD5SIG_MAXKEYLEN) | |
987 | return -EINVAL; | |
988 | ||
989 | if (!tcp_sk(sk)->md5sig_info) { | |
990 | struct tcp_sock *tp = tcp_sk(sk); | |
7174259e | 991 | struct tcp_md5sig_info *p = kzalloc(sizeof(*p), GFP_KERNEL); |
cfb6eeb4 | 992 | |
cfb6eeb4 YH |
993 | if (!p) |
994 | return -EINVAL; | |
995 | ||
996 | tp->md5sig_info = p; | |
3d7dbeac | 997 | sk->sk_route_caps &= ~NETIF_F_GSO_MASK; |
cfb6eeb4 YH |
998 | } |
999 | ||
f6685938 | 1000 | newkey = kmemdup(cmd.tcpm_key, cmd.tcpm_keylen, GFP_KERNEL); |
cfb6eeb4 YH |
1001 | if (!newkey) |
1002 | return -ENOMEM; | |
cfb6eeb4 YH |
1003 | return tcp_v4_md5_do_add(sk, sin->sin_addr.s_addr, |
1004 | newkey, cmd.tcpm_keylen); | |
1005 | } | |
1006 | ||
1007 | static int tcp_v4_do_calc_md5_hash(char *md5_hash, struct tcp_md5sig_key *key, | |
1008 | __be32 saddr, __be32 daddr, | |
1009 | struct tcphdr *th, int protocol, | |
9cb5734e | 1010 | unsigned int tcplen) |
cfb6eeb4 YH |
1011 | { |
1012 | struct scatterlist sg[4]; | |
1013 | __u16 data_len; | |
1014 | int block = 0; | |
8e5200f5 | 1015 | __sum16 old_checksum; |
cfb6eeb4 YH |
1016 | struct tcp_md5sig_pool *hp; |
1017 | struct tcp4_pseudohdr *bp; | |
1018 | struct hash_desc *desc; | |
1019 | int err; | |
1020 | unsigned int nbytes = 0; | |
1021 | ||
1022 | /* | |
1023 | * Okay, so RFC2385 is turned on for this connection, | |
1024 | * so we need to generate the MD5 hash for the packet now. | |
1025 | */ | |
1026 | ||
1027 | hp = tcp_get_md5sig_pool(); | |
1028 | if (!hp) | |
1029 | goto clear_hash_noput; | |
1030 | ||
1031 | bp = &hp->md5_blk.ip4; | |
1032 | desc = &hp->md5_desc; | |
1033 | ||
1034 | /* | |
1035 | * 1. the TCP pseudo-header (in the order: source IP address, | |
1036 | * destination IP address, zero-padded protocol number, and | |
1037 | * segment length) | |
1038 | */ | |
1039 | bp->saddr = saddr; | |
1040 | bp->daddr = daddr; | |
1041 | bp->pad = 0; | |
1042 | bp->protocol = protocol; | |
1043 | bp->len = htons(tcplen); | |
c7da57a1 DM |
1044 | |
1045 | sg_init_table(sg, 4); | |
1046 | ||
cfb6eeb4 YH |
1047 | sg_set_buf(&sg[block++], bp, sizeof(*bp)); |
1048 | nbytes += sizeof(*bp); | |
1049 | ||
cfb6eeb4 YH |
1050 | /* 2. the TCP header, excluding options, and assuming a |
1051 | * checksum of zero/ | |
1052 | */ | |
1053 | old_checksum = th->check; | |
1054 | th->check = 0; | |
1055 | sg_set_buf(&sg[block++], th, sizeof(struct tcphdr)); | |
1056 | nbytes += sizeof(struct tcphdr); | |
08dd1a50 | 1057 | |
cfb6eeb4 YH |
1058 | /* 3. the TCP segment data (if any) */ |
1059 | data_len = tcplen - (th->doff << 2); | |
1060 | if (data_len > 0) { | |
1061 | unsigned char *data = (unsigned char *)th + (th->doff << 2); | |
1062 | sg_set_buf(&sg[block++], data, data_len); | |
1063 | nbytes += data_len; | |
1064 | } | |
1065 | ||
1066 | /* 4. an independently-specified key or password, known to both | |
1067 | * TCPs and presumably connection-specific | |
1068 | */ | |
1069 | sg_set_buf(&sg[block++], key->key, key->keylen); | |
1070 | nbytes += key->keylen; | |
1071 | ||
c46f2334 | 1072 | sg_mark_end(&sg[block - 1]); |
c7da57a1 | 1073 | |
cfb6eeb4 YH |
1074 | /* Now store the Hash into the packet */ |
1075 | err = crypto_hash_init(desc); | |
1076 | if (err) | |
1077 | goto clear_hash; | |
1078 | err = crypto_hash_update(desc, sg, nbytes); | |
1079 | if (err) | |
1080 | goto clear_hash; | |
1081 | err = crypto_hash_final(desc, md5_hash); | |
1082 | if (err) | |
1083 | goto clear_hash; | |
1084 | ||
1085 | /* Reset header, and free up the crypto */ | |
1086 | tcp_put_md5sig_pool(); | |
1087 | th->check = old_checksum; | |
1088 | ||
1089 | out: | |
cfb6eeb4 YH |
1090 | return 0; |
1091 | clear_hash: | |
1092 | tcp_put_md5sig_pool(); | |
1093 | clear_hash_noput: | |
1094 | memset(md5_hash, 0, 16); | |
1095 | goto out; | |
1096 | } | |
1097 | ||
1098 | int tcp_v4_calc_md5_hash(char *md5_hash, struct tcp_md5sig_key *key, | |
1099 | struct sock *sk, | |
1100 | struct dst_entry *dst, | |
1101 | struct request_sock *req, | |
1102 | struct tcphdr *th, int protocol, | |
9cb5734e | 1103 | unsigned int tcplen) |
cfb6eeb4 YH |
1104 | { |
1105 | __be32 saddr, daddr; | |
1106 | ||
1107 | if (sk) { | |
1108 | saddr = inet_sk(sk)->saddr; | |
1109 | daddr = inet_sk(sk)->daddr; | |
1110 | } else { | |
1111 | struct rtable *rt = (struct rtable *)dst; | |
1112 | BUG_ON(!rt); | |
1113 | saddr = rt->rt_src; | |
1114 | daddr = rt->rt_dst; | |
1115 | } | |
1116 | return tcp_v4_do_calc_md5_hash(md5_hash, key, | |
1117 | saddr, daddr, | |
1118 | th, protocol, tcplen); | |
1119 | } | |
1120 | ||
1121 | EXPORT_SYMBOL(tcp_v4_calc_md5_hash); | |
1122 | ||
7174259e | 1123 | static int tcp_v4_inbound_md5_hash(struct sock *sk, struct sk_buff *skb) |
cfb6eeb4 YH |
1124 | { |
1125 | /* | |
1126 | * This gets called for each TCP segment that arrives | |
1127 | * so we want to be efficient. | |
1128 | * We have 3 drop cases: | |
1129 | * o No MD5 hash and one expected. | |
1130 | * o MD5 hash and we're not expecting one. | |
1131 | * o MD5 hash and its wrong. | |
1132 | */ | |
1133 | __u8 *hash_location = NULL; | |
1134 | struct tcp_md5sig_key *hash_expected; | |
eddc9ec5 | 1135 | const struct iphdr *iph = ip_hdr(skb); |
aa8223c7 | 1136 | struct tcphdr *th = tcp_hdr(skb); |
cfb6eeb4 | 1137 | int genhash; |
cfb6eeb4 YH |
1138 | unsigned char newhash[16]; |
1139 | ||
1140 | hash_expected = tcp_v4_md5_do_lookup(sk, iph->saddr); | |
7d5d5525 | 1141 | hash_location = tcp_parse_md5sig_option(th); |
cfb6eeb4 | 1142 | |
cfb6eeb4 YH |
1143 | /* We've parsed the options - do we have a hash? */ |
1144 | if (!hash_expected && !hash_location) | |
1145 | return 0; | |
1146 | ||
1147 | if (hash_expected && !hash_location) { | |
a9fc00cc | 1148 | LIMIT_NETDEBUG(KERN_INFO "MD5 Hash expected but NOT found " |
cfb6eeb4 | 1149 | "(" NIPQUAD_FMT ", %d)->(" NIPQUAD_FMT ", %d)\n", |
7174259e ACM |
1150 | NIPQUAD(iph->saddr), ntohs(th->source), |
1151 | NIPQUAD(iph->daddr), ntohs(th->dest)); | |
cfb6eeb4 YH |
1152 | return 1; |
1153 | } | |
1154 | ||
1155 | if (!hash_expected && hash_location) { | |
7174259e | 1156 | LIMIT_NETDEBUG(KERN_INFO "MD5 Hash NOT expected but found " |
cfb6eeb4 | 1157 | "(" NIPQUAD_FMT ", %d)->(" NIPQUAD_FMT ", %d)\n", |
7174259e ACM |
1158 | NIPQUAD(iph->saddr), ntohs(th->source), |
1159 | NIPQUAD(iph->daddr), ntohs(th->dest)); | |
cfb6eeb4 YH |
1160 | return 1; |
1161 | } | |
1162 | ||
1163 | /* Okay, so this is hash_expected and hash_location - | |
1164 | * so we need to calculate the checksum. | |
1165 | */ | |
1166 | genhash = tcp_v4_do_calc_md5_hash(newhash, | |
1167 | hash_expected, | |
1168 | iph->saddr, iph->daddr, | |
1169 | th, sk->sk_protocol, | |
1170 | skb->len); | |
1171 | ||
1172 | if (genhash || memcmp(hash_location, newhash, 16) != 0) { | |
1173 | if (net_ratelimit()) { | |
1174 | printk(KERN_INFO "MD5 Hash failed for " | |
1175 | "(" NIPQUAD_FMT ", %d)->(" NIPQUAD_FMT ", %d)%s\n", | |
7174259e ACM |
1176 | NIPQUAD(iph->saddr), ntohs(th->source), |
1177 | NIPQUAD(iph->daddr), ntohs(th->dest), | |
cfb6eeb4 | 1178 | genhash ? " tcp_v4_calc_md5_hash failed" : ""); |
cfb6eeb4 YH |
1179 | } |
1180 | return 1; | |
1181 | } | |
1182 | return 0; | |
1183 | } | |
1184 | ||
1185 | #endif | |
1186 | ||
72a3effa | 1187 | struct request_sock_ops tcp_request_sock_ops __read_mostly = { |
1da177e4 | 1188 | .family = PF_INET, |
2e6599cb | 1189 | .obj_size = sizeof(struct tcp_request_sock), |
1da177e4 | 1190 | .rtx_syn_ack = tcp_v4_send_synack, |
60236fdd ACM |
1191 | .send_ack = tcp_v4_reqsk_send_ack, |
1192 | .destructor = tcp_v4_reqsk_destructor, | |
1da177e4 LT |
1193 | .send_reset = tcp_v4_send_reset, |
1194 | }; | |
1195 | ||
cfb6eeb4 | 1196 | #ifdef CONFIG_TCP_MD5SIG |
b6332e6c | 1197 | static struct tcp_request_sock_ops tcp_request_sock_ipv4_ops = { |
cfb6eeb4 | 1198 | .md5_lookup = tcp_v4_reqsk_md5_lookup, |
cfb6eeb4 | 1199 | }; |
b6332e6c | 1200 | #endif |
cfb6eeb4 | 1201 | |
6d6ee43e ACM |
1202 | static struct timewait_sock_ops tcp_timewait_sock_ops = { |
1203 | .twsk_obj_size = sizeof(struct tcp_timewait_sock), | |
1204 | .twsk_unique = tcp_twsk_unique, | |
cfb6eeb4 | 1205 | .twsk_destructor= tcp_twsk_destructor, |
6d6ee43e ACM |
1206 | }; |
1207 | ||
1da177e4 LT |
1208 | int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb) |
1209 | { | |
2e6599cb | 1210 | struct inet_request_sock *ireq; |
1da177e4 | 1211 | struct tcp_options_received tmp_opt; |
60236fdd | 1212 | struct request_sock *req; |
eddc9ec5 ACM |
1213 | __be32 saddr = ip_hdr(skb)->saddr; |
1214 | __be32 daddr = ip_hdr(skb)->daddr; | |
1da177e4 LT |
1215 | __u32 isn = TCP_SKB_CB(skb)->when; |
1216 | struct dst_entry *dst = NULL; | |
1217 | #ifdef CONFIG_SYN_COOKIES | |
1218 | int want_cookie = 0; | |
1219 | #else | |
1220 | #define want_cookie 0 /* Argh, why doesn't gcc optimize this :( */ | |
1221 | #endif | |
1222 | ||
1223 | /* Never answer to SYNs send to broadcast or multicast */ | |
ee6b9673 | 1224 | if (skb->rtable->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST)) |
1da177e4 LT |
1225 | goto drop; |
1226 | ||
1227 | /* TW buckets are converted to open requests without | |
1228 | * limitations, they conserve resources and peer is | |
1229 | * evidently real one. | |
1230 | */ | |
463c84b9 | 1231 | if (inet_csk_reqsk_queue_is_full(sk) && !isn) { |
1da177e4 LT |
1232 | #ifdef CONFIG_SYN_COOKIES |
1233 | if (sysctl_tcp_syncookies) { | |
1234 | want_cookie = 1; | |
1235 | } else | |
1236 | #endif | |
1237 | goto drop; | |
1238 | } | |
1239 | ||
1240 | /* Accept backlog is full. If we have already queued enough | |
1241 | * of warm entries in syn queue, drop request. It is better than | |
1242 | * clogging syn queue with openreqs with exponentially increasing | |
1243 | * timeout. | |
1244 | */ | |
463c84b9 | 1245 | if (sk_acceptq_is_full(sk) && inet_csk_reqsk_queue_young(sk) > 1) |
1da177e4 LT |
1246 | goto drop; |
1247 | ||
60236fdd | 1248 | req = reqsk_alloc(&tcp_request_sock_ops); |
1da177e4 LT |
1249 | if (!req) |
1250 | goto drop; | |
1251 | ||
cfb6eeb4 YH |
1252 | #ifdef CONFIG_TCP_MD5SIG |
1253 | tcp_rsk(req)->af_specific = &tcp_request_sock_ipv4_ops; | |
1254 | #endif | |
1255 | ||
1da177e4 LT |
1256 | tcp_clear_options(&tmp_opt); |
1257 | tmp_opt.mss_clamp = 536; | |
1258 | tmp_opt.user_mss = tcp_sk(sk)->rx_opt.user_mss; | |
1259 | ||
1260 | tcp_parse_options(skb, &tmp_opt, 0); | |
1261 | ||
4dfc2817 | 1262 | if (want_cookie && !tmp_opt.saw_tstamp) |
1da177e4 | 1263 | tcp_clear_options(&tmp_opt); |
1da177e4 LT |
1264 | |
1265 | if (tmp_opt.saw_tstamp && !tmp_opt.rcv_tsval) { | |
1266 | /* Some OSes (unknown ones, but I see them on web server, which | |
1267 | * contains information interesting only for windows' | |
1268 | * users) do not send their stamp in SYN. It is easy case. | |
1269 | * We simply do not advertise TS support. | |
1270 | */ | |
1271 | tmp_opt.saw_tstamp = 0; | |
1272 | tmp_opt.tstamp_ok = 0; | |
1273 | } | |
1274 | tmp_opt.tstamp_ok = tmp_opt.saw_tstamp; | |
1275 | ||
1276 | tcp_openreq_init(req, &tmp_opt, skb); | |
1277 | ||
4237c75c VY |
1278 | if (security_inet_conn_request(sk, skb, req)) |
1279 | goto drop_and_free; | |
1280 | ||
2e6599cb ACM |
1281 | ireq = inet_rsk(req); |
1282 | ireq->loc_addr = daddr; | |
1283 | ireq->rmt_addr = saddr; | |
1284 | ireq->opt = tcp_v4_save_options(sk, skb); | |
1da177e4 | 1285 | if (!want_cookie) |
aa8223c7 | 1286 | TCP_ECN_create_request(req, tcp_hdr(skb)); |
1da177e4 LT |
1287 | |
1288 | if (want_cookie) { | |
1289 | #ifdef CONFIG_SYN_COOKIES | |
1290 | syn_flood_warning(skb); | |
4dfc2817 | 1291 | req->cookie_ts = tmp_opt.tstamp_ok; |
1da177e4 LT |
1292 | #endif |
1293 | isn = cookie_v4_init_sequence(sk, skb, &req->mss); | |
1294 | } else if (!isn) { | |
1295 | struct inet_peer *peer = NULL; | |
1296 | ||
1297 | /* VJ's idea. We save last timestamp seen | |
1298 | * from the destination in peer table, when entering | |
1299 | * state TIME-WAIT, and check against it before | |
1300 | * accepting new connection request. | |
1301 | * | |
1302 | * If "isn" is not zero, this request hit alive | |
1303 | * timewait bucket, so that all the necessary checks | |
1304 | * are made in the function processing timewait state. | |
1305 | */ | |
1306 | if (tmp_opt.saw_tstamp && | |
295ff7ed | 1307 | tcp_death_row.sysctl_tw_recycle && |
463c84b9 | 1308 | (dst = inet_csk_route_req(sk, req)) != NULL && |
1da177e4 LT |
1309 | (peer = rt_get_peer((struct rtable *)dst)) != NULL && |
1310 | peer->v4daddr == saddr) { | |
9d729f72 | 1311 | if (get_seconds() < peer->tcp_ts_stamp + TCP_PAWS_MSL && |
1da177e4 LT |
1312 | (s32)(peer->tcp_ts - req->ts_recent) > |
1313 | TCP_PAWS_WINDOW) { | |
1314 | NET_INC_STATS_BH(LINUX_MIB_PAWSPASSIVEREJECTED); | |
7cd04fa7 | 1315 | goto drop_and_release; |
1da177e4 LT |
1316 | } |
1317 | } | |
1318 | /* Kill the following clause, if you dislike this way. */ | |
1319 | else if (!sysctl_tcp_syncookies && | |
463c84b9 | 1320 | (sysctl_max_syn_backlog - inet_csk_reqsk_queue_len(sk) < |
1da177e4 LT |
1321 | (sysctl_max_syn_backlog >> 2)) && |
1322 | (!peer || !peer->tcp_ts_stamp) && | |
1323 | (!dst || !dst_metric(dst, RTAX_RTT))) { | |
1324 | /* Without syncookies last quarter of | |
1325 | * backlog is filled with destinations, | |
1326 | * proven to be alive. | |
1327 | * It means that we continue to communicate | |
1328 | * to destinations, already remembered | |
1329 | * to the moment of synflood. | |
1330 | */ | |
64ce2073 | 1331 | LIMIT_NETDEBUG(KERN_DEBUG "TCP: drop open " |
a7d632b6 | 1332 | "request from " NIPQUAD_FMT "/%u\n", |
64ce2073 | 1333 | NIPQUAD(saddr), |
aa8223c7 | 1334 | ntohs(tcp_hdr(skb)->source)); |
7cd04fa7 | 1335 | goto drop_and_release; |
1da177e4 LT |
1336 | } |
1337 | ||
a94f723d | 1338 | isn = tcp_v4_init_sequence(skb); |
1da177e4 | 1339 | } |
2e6599cb | 1340 | tcp_rsk(req)->snt_isn = isn; |
1da177e4 | 1341 | |
7cd04fa7 | 1342 | if (__tcp_v4_send_synack(sk, req, dst) || want_cookie) |
1da177e4 LT |
1343 | goto drop_and_free; |
1344 | ||
7cd04fa7 | 1345 | inet_csk_reqsk_queue_hash_add(sk, req, TCP_TIMEOUT_INIT); |
1da177e4 LT |
1346 | return 0; |
1347 | ||
7cd04fa7 DL |
1348 | drop_and_release: |
1349 | dst_release(dst); | |
1da177e4 | 1350 | drop_and_free: |
60236fdd | 1351 | reqsk_free(req); |
1da177e4 | 1352 | drop: |
1da177e4 LT |
1353 | return 0; |
1354 | } | |
1355 | ||
1356 | ||
1357 | /* | |
1358 | * The three way handshake has completed - we got a valid synack - | |
1359 | * now create the new socket. | |
1360 | */ | |
1361 | struct sock *tcp_v4_syn_recv_sock(struct sock *sk, struct sk_buff *skb, | |
60236fdd | 1362 | struct request_sock *req, |
1da177e4 LT |
1363 | struct dst_entry *dst) |
1364 | { | |
2e6599cb | 1365 | struct inet_request_sock *ireq; |
1da177e4 LT |
1366 | struct inet_sock *newinet; |
1367 | struct tcp_sock *newtp; | |
1368 | struct sock *newsk; | |
cfb6eeb4 YH |
1369 | #ifdef CONFIG_TCP_MD5SIG |
1370 | struct tcp_md5sig_key *key; | |
1371 | #endif | |
1da177e4 LT |
1372 | |
1373 | if (sk_acceptq_is_full(sk)) | |
1374 | goto exit_overflow; | |
1375 | ||
463c84b9 | 1376 | if (!dst && (dst = inet_csk_route_req(sk, req)) == NULL) |
1da177e4 LT |
1377 | goto exit; |
1378 | ||
1379 | newsk = tcp_create_openreq_child(sk, req, skb); | |
1380 | if (!newsk) | |
1381 | goto exit; | |
1382 | ||
bcd76111 | 1383 | newsk->sk_gso_type = SKB_GSO_TCPV4; |
6cbb0df7 | 1384 | sk_setup_caps(newsk, dst); |
1da177e4 LT |
1385 | |
1386 | newtp = tcp_sk(newsk); | |
1387 | newinet = inet_sk(newsk); | |
2e6599cb ACM |
1388 | ireq = inet_rsk(req); |
1389 | newinet->daddr = ireq->rmt_addr; | |
1390 | newinet->rcv_saddr = ireq->loc_addr; | |
1391 | newinet->saddr = ireq->loc_addr; | |
1392 | newinet->opt = ireq->opt; | |
1393 | ireq->opt = NULL; | |
463c84b9 | 1394 | newinet->mc_index = inet_iif(skb); |
eddc9ec5 | 1395 | newinet->mc_ttl = ip_hdr(skb)->ttl; |
d83d8461 | 1396 | inet_csk(newsk)->icsk_ext_hdr_len = 0; |
1da177e4 | 1397 | if (newinet->opt) |
d83d8461 | 1398 | inet_csk(newsk)->icsk_ext_hdr_len = newinet->opt->optlen; |
1da177e4 LT |
1399 | newinet->id = newtp->write_seq ^ jiffies; |
1400 | ||
5d424d5a | 1401 | tcp_mtup_init(newsk); |
1da177e4 LT |
1402 | tcp_sync_mss(newsk, dst_mtu(dst)); |
1403 | newtp->advmss = dst_metric(dst, RTAX_ADVMSS); | |
1404 | tcp_initialize_rcv_mss(newsk); | |
1405 | ||
cfb6eeb4 YH |
1406 | #ifdef CONFIG_TCP_MD5SIG |
1407 | /* Copy over the MD5 key from the original socket */ | |
1408 | if ((key = tcp_v4_md5_do_lookup(sk, newinet->daddr)) != NULL) { | |
1409 | /* | |
1410 | * We're using one, so create a matching key | |
1411 | * on the newsk structure. If we fail to get | |
1412 | * memory, then we end up not copying the key | |
1413 | * across. Shucks. | |
1414 | */ | |
f6685938 ACM |
1415 | char *newkey = kmemdup(key->key, key->keylen, GFP_ATOMIC); |
1416 | if (newkey != NULL) | |
cfb6eeb4 YH |
1417 | tcp_v4_md5_do_add(newsk, inet_sk(sk)->daddr, |
1418 | newkey, key->keylen); | |
cfb6eeb4 YH |
1419 | } |
1420 | #endif | |
1421 | ||
ab1e0a13 ACM |
1422 | __inet_hash_nolisten(newsk); |
1423 | __inet_inherit_port(sk, newsk); | |
1da177e4 LT |
1424 | |
1425 | return newsk; | |
1426 | ||
1427 | exit_overflow: | |
1428 | NET_INC_STATS_BH(LINUX_MIB_LISTENOVERFLOWS); | |
1429 | exit: | |
1430 | NET_INC_STATS_BH(LINUX_MIB_LISTENDROPS); | |
1431 | dst_release(dst); | |
1432 | return NULL; | |
1433 | } | |
1434 | ||
1435 | static struct sock *tcp_v4_hnd_req(struct sock *sk, struct sk_buff *skb) | |
1436 | { | |
aa8223c7 | 1437 | struct tcphdr *th = tcp_hdr(skb); |
eddc9ec5 | 1438 | const struct iphdr *iph = ip_hdr(skb); |
1da177e4 | 1439 | struct sock *nsk; |
60236fdd | 1440 | struct request_sock **prev; |
1da177e4 | 1441 | /* Find possible connection requests. */ |
463c84b9 ACM |
1442 | struct request_sock *req = inet_csk_search_req(sk, &prev, th->source, |
1443 | iph->saddr, iph->daddr); | |
1da177e4 LT |
1444 | if (req) |
1445 | return tcp_check_req(sk, skb, req, prev); | |
1446 | ||
3b1e0a65 | 1447 | nsk = inet_lookup_established(sock_net(sk), &tcp_hashinfo, iph->saddr, |
c67499c0 | 1448 | th->source, iph->daddr, th->dest, inet_iif(skb)); |
1da177e4 LT |
1449 | |
1450 | if (nsk) { | |
1451 | if (nsk->sk_state != TCP_TIME_WAIT) { | |
1452 | bh_lock_sock(nsk); | |
1453 | return nsk; | |
1454 | } | |
9469c7b4 | 1455 | inet_twsk_put(inet_twsk(nsk)); |
1da177e4 LT |
1456 | return NULL; |
1457 | } | |
1458 | ||
1459 | #ifdef CONFIG_SYN_COOKIES | |
1460 | if (!th->rst && !th->syn && th->ack) | |
1461 | sk = cookie_v4_check(sk, skb, &(IPCB(skb)->opt)); | |
1462 | #endif | |
1463 | return sk; | |
1464 | } | |
1465 | ||
b51655b9 | 1466 | static __sum16 tcp_v4_checksum_init(struct sk_buff *skb) |
1da177e4 | 1467 | { |
eddc9ec5 ACM |
1468 | const struct iphdr *iph = ip_hdr(skb); |
1469 | ||
84fa7933 | 1470 | if (skb->ip_summed == CHECKSUM_COMPLETE) { |
eddc9ec5 ACM |
1471 | if (!tcp_v4_check(skb->len, iph->saddr, |
1472 | iph->daddr, skb->csum)) { | |
fb286bb2 | 1473 | skb->ip_summed = CHECKSUM_UNNECESSARY; |
1da177e4 | 1474 | return 0; |
fb286bb2 | 1475 | } |
1da177e4 | 1476 | } |
fb286bb2 | 1477 | |
eddc9ec5 | 1478 | skb->csum = csum_tcpudp_nofold(iph->saddr, iph->daddr, |
fb286bb2 HX |
1479 | skb->len, IPPROTO_TCP, 0); |
1480 | ||
1da177e4 | 1481 | if (skb->len <= 76) { |
fb286bb2 | 1482 | return __skb_checksum_complete(skb); |
1da177e4 LT |
1483 | } |
1484 | return 0; | |
1485 | } | |
1486 | ||
1487 | ||
1488 | /* The socket must have it's spinlock held when we get | |
1489 | * here. | |
1490 | * | |
1491 | * We have a potential double-lock case here, so even when | |
1492 | * doing backlog processing we use the BH locking scheme. | |
1493 | * This is because we cannot sleep with the original spinlock | |
1494 | * held. | |
1495 | */ | |
1496 | int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb) | |
1497 | { | |
cfb6eeb4 YH |
1498 | struct sock *rsk; |
1499 | #ifdef CONFIG_TCP_MD5SIG | |
1500 | /* | |
1501 | * We really want to reject the packet as early as possible | |
1502 | * if: | |
1503 | * o We're expecting an MD5'd packet and this is no MD5 tcp option | |
1504 | * o There is an MD5 option and we're not expecting one | |
1505 | */ | |
7174259e | 1506 | if (tcp_v4_inbound_md5_hash(sk, skb)) |
cfb6eeb4 YH |
1507 | goto discard; |
1508 | #endif | |
1509 | ||
1da177e4 LT |
1510 | if (sk->sk_state == TCP_ESTABLISHED) { /* Fast path */ |
1511 | TCP_CHECK_TIMER(sk); | |
aa8223c7 | 1512 | if (tcp_rcv_established(sk, skb, tcp_hdr(skb), skb->len)) { |
cfb6eeb4 | 1513 | rsk = sk; |
1da177e4 | 1514 | goto reset; |
cfb6eeb4 | 1515 | } |
1da177e4 LT |
1516 | TCP_CHECK_TIMER(sk); |
1517 | return 0; | |
1518 | } | |
1519 | ||
ab6a5bb6 | 1520 | if (skb->len < tcp_hdrlen(skb) || tcp_checksum_complete(skb)) |
1da177e4 LT |
1521 | goto csum_err; |
1522 | ||
1523 | if (sk->sk_state == TCP_LISTEN) { | |
1524 | struct sock *nsk = tcp_v4_hnd_req(sk, skb); | |
1525 | if (!nsk) | |
1526 | goto discard; | |
1527 | ||
1528 | if (nsk != sk) { | |
cfb6eeb4 YH |
1529 | if (tcp_child_process(sk, nsk, skb)) { |
1530 | rsk = nsk; | |
1da177e4 | 1531 | goto reset; |
cfb6eeb4 | 1532 | } |
1da177e4 LT |
1533 | return 0; |
1534 | } | |
1535 | } | |
1536 | ||
1537 | TCP_CHECK_TIMER(sk); | |
aa8223c7 | 1538 | if (tcp_rcv_state_process(sk, skb, tcp_hdr(skb), skb->len)) { |
cfb6eeb4 | 1539 | rsk = sk; |
1da177e4 | 1540 | goto reset; |
cfb6eeb4 | 1541 | } |
1da177e4 LT |
1542 | TCP_CHECK_TIMER(sk); |
1543 | return 0; | |
1544 | ||
1545 | reset: | |
cfb6eeb4 | 1546 | tcp_v4_send_reset(rsk, skb); |
1da177e4 LT |
1547 | discard: |
1548 | kfree_skb(skb); | |
1549 | /* Be careful here. If this function gets more complicated and | |
1550 | * gcc suffers from register pressure on the x86, sk (in %ebx) | |
1551 | * might be destroyed here. This current version compiles correctly, | |
1552 | * but you have been warned. | |
1553 | */ | |
1554 | return 0; | |
1555 | ||
1556 | csum_err: | |
1557 | TCP_INC_STATS_BH(TCP_MIB_INERRS); | |
1558 | goto discard; | |
1559 | } | |
1560 | ||
1561 | /* | |
1562 | * From tcp_input.c | |
1563 | */ | |
1564 | ||
1565 | int tcp_v4_rcv(struct sk_buff *skb) | |
1566 | { | |
eddc9ec5 | 1567 | const struct iphdr *iph; |
1da177e4 LT |
1568 | struct tcphdr *th; |
1569 | struct sock *sk; | |
1570 | int ret; | |
1571 | ||
1572 | if (skb->pkt_type != PACKET_HOST) | |
1573 | goto discard_it; | |
1574 | ||
1575 | /* Count it even if it's bad */ | |
1576 | TCP_INC_STATS_BH(TCP_MIB_INSEGS); | |
1577 | ||
1578 | if (!pskb_may_pull(skb, sizeof(struct tcphdr))) | |
1579 | goto discard_it; | |
1580 | ||
aa8223c7 | 1581 | th = tcp_hdr(skb); |
1da177e4 LT |
1582 | |
1583 | if (th->doff < sizeof(struct tcphdr) / 4) | |
1584 | goto bad_packet; | |
1585 | if (!pskb_may_pull(skb, th->doff * 4)) | |
1586 | goto discard_it; | |
1587 | ||
1588 | /* An explanation is required here, I think. | |
1589 | * Packet length and doff are validated by header prediction, | |
caa20d9a | 1590 | * provided case of th->doff==0 is eliminated. |
1da177e4 | 1591 | * So, we defer the checks. */ |
60476372 | 1592 | if (!skb_csum_unnecessary(skb) && tcp_v4_checksum_init(skb)) |
1da177e4 LT |
1593 | goto bad_packet; |
1594 | ||
aa8223c7 | 1595 | th = tcp_hdr(skb); |
eddc9ec5 | 1596 | iph = ip_hdr(skb); |
1da177e4 LT |
1597 | TCP_SKB_CB(skb)->seq = ntohl(th->seq); |
1598 | TCP_SKB_CB(skb)->end_seq = (TCP_SKB_CB(skb)->seq + th->syn + th->fin + | |
1599 | skb->len - th->doff * 4); | |
1600 | TCP_SKB_CB(skb)->ack_seq = ntohl(th->ack_seq); | |
1601 | TCP_SKB_CB(skb)->when = 0; | |
eddc9ec5 | 1602 | TCP_SKB_CB(skb)->flags = iph->tos; |
1da177e4 LT |
1603 | TCP_SKB_CB(skb)->sacked = 0; |
1604 | ||
c346dca1 | 1605 | sk = __inet_lookup(dev_net(skb->dev), &tcp_hashinfo, iph->saddr, |
c67499c0 | 1606 | th->source, iph->daddr, th->dest, inet_iif(skb)); |
1da177e4 LT |
1607 | if (!sk) |
1608 | goto no_tcp_socket; | |
1609 | ||
1610 | process: | |
1611 | if (sk->sk_state == TCP_TIME_WAIT) | |
1612 | goto do_time_wait; | |
1613 | ||
1614 | if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb)) | |
1615 | goto discard_and_relse; | |
b59c2701 | 1616 | nf_reset(skb); |
1da177e4 | 1617 | |
fda9ef5d | 1618 | if (sk_filter(sk, skb)) |
1da177e4 LT |
1619 | goto discard_and_relse; |
1620 | ||
1621 | skb->dev = NULL; | |
1622 | ||
c6366184 | 1623 | bh_lock_sock_nested(sk); |
1da177e4 LT |
1624 | ret = 0; |
1625 | if (!sock_owned_by_user(sk)) { | |
1a2449a8 CL |
1626 | #ifdef CONFIG_NET_DMA |
1627 | struct tcp_sock *tp = tcp_sk(sk); | |
1628 | if (!tp->ucopy.dma_chan && tp->ucopy.pinned_list) | |
1629 | tp->ucopy.dma_chan = get_softnet_dma(); | |
1630 | if (tp->ucopy.dma_chan) | |
1da177e4 | 1631 | ret = tcp_v4_do_rcv(sk, skb); |
1a2449a8 CL |
1632 | else |
1633 | #endif | |
1634 | { | |
1635 | if (!tcp_prequeue(sk, skb)) | |
1636 | ret = tcp_v4_do_rcv(sk, skb); | |
1637 | } | |
1da177e4 LT |
1638 | } else |
1639 | sk_add_backlog(sk, skb); | |
1640 | bh_unlock_sock(sk); | |
1641 | ||
1642 | sock_put(sk); | |
1643 | ||
1644 | return ret; | |
1645 | ||
1646 | no_tcp_socket: | |
1647 | if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) | |
1648 | goto discard_it; | |
1649 | ||
1650 | if (skb->len < (th->doff << 2) || tcp_checksum_complete(skb)) { | |
1651 | bad_packet: | |
1652 | TCP_INC_STATS_BH(TCP_MIB_INERRS); | |
1653 | } else { | |
cfb6eeb4 | 1654 | tcp_v4_send_reset(NULL, skb); |
1da177e4 LT |
1655 | } |
1656 | ||
1657 | discard_it: | |
1658 | /* Discard frame. */ | |
1659 | kfree_skb(skb); | |
e905a9ed | 1660 | return 0; |
1da177e4 LT |
1661 | |
1662 | discard_and_relse: | |
1663 | sock_put(sk); | |
1664 | goto discard_it; | |
1665 | ||
1666 | do_time_wait: | |
1667 | if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) { | |
9469c7b4 | 1668 | inet_twsk_put(inet_twsk(sk)); |
1da177e4 LT |
1669 | goto discard_it; |
1670 | } | |
1671 | ||
1672 | if (skb->len < (th->doff << 2) || tcp_checksum_complete(skb)) { | |
1673 | TCP_INC_STATS_BH(TCP_MIB_INERRS); | |
9469c7b4 | 1674 | inet_twsk_put(inet_twsk(sk)); |
1da177e4 LT |
1675 | goto discard_it; |
1676 | } | |
9469c7b4 | 1677 | switch (tcp_timewait_state_process(inet_twsk(sk), skb, th)) { |
1da177e4 | 1678 | case TCP_TW_SYN: { |
c346dca1 | 1679 | struct sock *sk2 = inet_lookup_listener(dev_net(skb->dev), |
c67499c0 | 1680 | &tcp_hashinfo, |
eddc9ec5 | 1681 | iph->daddr, th->dest, |
463c84b9 | 1682 | inet_iif(skb)); |
1da177e4 | 1683 | if (sk2) { |
9469c7b4 YH |
1684 | inet_twsk_deschedule(inet_twsk(sk), &tcp_death_row); |
1685 | inet_twsk_put(inet_twsk(sk)); | |
1da177e4 LT |
1686 | sk = sk2; |
1687 | goto process; | |
1688 | } | |
1689 | /* Fall through to ACK */ | |
1690 | } | |
1691 | case TCP_TW_ACK: | |
1692 | tcp_v4_timewait_ack(sk, skb); | |
1693 | break; | |
1694 | case TCP_TW_RST: | |
1695 | goto no_tcp_socket; | |
1696 | case TCP_TW_SUCCESS:; | |
1697 | } | |
1698 | goto discard_it; | |
1699 | } | |
1700 | ||
1da177e4 LT |
1701 | /* VJ's idea. Save last timestamp seen from this destination |
1702 | * and hold it at least for normal timewait interval to use for duplicate | |
1703 | * segment detection in subsequent connections, before they enter synchronized | |
1704 | * state. | |
1705 | */ | |
1706 | ||
1707 | int tcp_v4_remember_stamp(struct sock *sk) | |
1708 | { | |
1709 | struct inet_sock *inet = inet_sk(sk); | |
1710 | struct tcp_sock *tp = tcp_sk(sk); | |
1711 | struct rtable *rt = (struct rtable *)__sk_dst_get(sk); | |
1712 | struct inet_peer *peer = NULL; | |
1713 | int release_it = 0; | |
1714 | ||
1715 | if (!rt || rt->rt_dst != inet->daddr) { | |
1716 | peer = inet_getpeer(inet->daddr, 1); | |
1717 | release_it = 1; | |
1718 | } else { | |
1719 | if (!rt->peer) | |
1720 | rt_bind_peer(rt, 1); | |
1721 | peer = rt->peer; | |
1722 | } | |
1723 | ||
1724 | if (peer) { | |
1725 | if ((s32)(peer->tcp_ts - tp->rx_opt.ts_recent) <= 0 || | |
9d729f72 | 1726 | (peer->tcp_ts_stamp + TCP_PAWS_MSL < get_seconds() && |
1da177e4 LT |
1727 | peer->tcp_ts_stamp <= tp->rx_opt.ts_recent_stamp)) { |
1728 | peer->tcp_ts_stamp = tp->rx_opt.ts_recent_stamp; | |
1729 | peer->tcp_ts = tp->rx_opt.ts_recent; | |
1730 | } | |
1731 | if (release_it) | |
1732 | inet_putpeer(peer); | |
1733 | return 1; | |
1734 | } | |
1735 | ||
1736 | return 0; | |
1737 | } | |
1738 | ||
8feaf0c0 | 1739 | int tcp_v4_tw_remember_stamp(struct inet_timewait_sock *tw) |
1da177e4 | 1740 | { |
8feaf0c0 | 1741 | struct inet_peer *peer = inet_getpeer(tw->tw_daddr, 1); |
1da177e4 LT |
1742 | |
1743 | if (peer) { | |
8feaf0c0 ACM |
1744 | const struct tcp_timewait_sock *tcptw = tcp_twsk((struct sock *)tw); |
1745 | ||
1746 | if ((s32)(peer->tcp_ts - tcptw->tw_ts_recent) <= 0 || | |
9d729f72 | 1747 | (peer->tcp_ts_stamp + TCP_PAWS_MSL < get_seconds() && |
8feaf0c0 ACM |
1748 | peer->tcp_ts_stamp <= tcptw->tw_ts_recent_stamp)) { |
1749 | peer->tcp_ts_stamp = tcptw->tw_ts_recent_stamp; | |
1750 | peer->tcp_ts = tcptw->tw_ts_recent; | |
1da177e4 LT |
1751 | } |
1752 | inet_putpeer(peer); | |
1753 | return 1; | |
1754 | } | |
1755 | ||
1756 | return 0; | |
1757 | } | |
1758 | ||
8292a17a | 1759 | struct inet_connection_sock_af_ops ipv4_specific = { |
543d9cfe ACM |
1760 | .queue_xmit = ip_queue_xmit, |
1761 | .send_check = tcp_v4_send_check, | |
1762 | .rebuild_header = inet_sk_rebuild_header, | |
1763 | .conn_request = tcp_v4_conn_request, | |
1764 | .syn_recv_sock = tcp_v4_syn_recv_sock, | |
1765 | .remember_stamp = tcp_v4_remember_stamp, | |
1766 | .net_header_len = sizeof(struct iphdr), | |
1767 | .setsockopt = ip_setsockopt, | |
1768 | .getsockopt = ip_getsockopt, | |
1769 | .addr2sockaddr = inet_csk_addr2sockaddr, | |
1770 | .sockaddr_len = sizeof(struct sockaddr_in), | |
ab1e0a13 | 1771 | .bind_conflict = inet_csk_bind_conflict, |
3fdadf7d | 1772 | #ifdef CONFIG_COMPAT |
543d9cfe ACM |
1773 | .compat_setsockopt = compat_ip_setsockopt, |
1774 | .compat_getsockopt = compat_ip_getsockopt, | |
3fdadf7d | 1775 | #endif |
1da177e4 LT |
1776 | }; |
1777 | ||
cfb6eeb4 | 1778 | #ifdef CONFIG_TCP_MD5SIG |
b6332e6c | 1779 | static struct tcp_sock_af_ops tcp_sock_ipv4_specific = { |
cfb6eeb4 YH |
1780 | .md5_lookup = tcp_v4_md5_lookup, |
1781 | .calc_md5_hash = tcp_v4_calc_md5_hash, | |
1782 | .md5_add = tcp_v4_md5_add_func, | |
1783 | .md5_parse = tcp_v4_parse_md5_keys, | |
cfb6eeb4 | 1784 | }; |
b6332e6c | 1785 | #endif |
cfb6eeb4 | 1786 | |
1da177e4 LT |
1787 | /* NOTE: A lot of things set to zero explicitly by call to |
1788 | * sk_alloc() so need not be done here. | |
1789 | */ | |
1790 | static int tcp_v4_init_sock(struct sock *sk) | |
1791 | { | |
6687e988 | 1792 | struct inet_connection_sock *icsk = inet_csk(sk); |
1da177e4 LT |
1793 | struct tcp_sock *tp = tcp_sk(sk); |
1794 | ||
1795 | skb_queue_head_init(&tp->out_of_order_queue); | |
1796 | tcp_init_xmit_timers(sk); | |
1797 | tcp_prequeue_init(tp); | |
1798 | ||
6687e988 | 1799 | icsk->icsk_rto = TCP_TIMEOUT_INIT; |
1da177e4 LT |
1800 | tp->mdev = TCP_TIMEOUT_INIT; |
1801 | ||
1802 | /* So many TCP implementations out there (incorrectly) count the | |
1803 | * initial SYN frame in their delayed-ACK and congestion control | |
1804 | * algorithms that we must have the following bandaid to talk | |
1805 | * efficiently to them. -DaveM | |
1806 | */ | |
1807 | tp->snd_cwnd = 2; | |
1808 | ||
1809 | /* See draft-stevens-tcpca-spec-01 for discussion of the | |
1810 | * initialization of these values. | |
1811 | */ | |
1812 | tp->snd_ssthresh = 0x7fffffff; /* Infinity */ | |
1813 | tp->snd_cwnd_clamp = ~0; | |
c1b4a7e6 | 1814 | tp->mss_cache = 536; |
1da177e4 LT |
1815 | |
1816 | tp->reordering = sysctl_tcp_reordering; | |
6687e988 | 1817 | icsk->icsk_ca_ops = &tcp_init_congestion_ops; |
1da177e4 LT |
1818 | |
1819 | sk->sk_state = TCP_CLOSE; | |
1820 | ||
1821 | sk->sk_write_space = sk_stream_write_space; | |
1822 | sock_set_flag(sk, SOCK_USE_WRITE_QUEUE); | |
1823 | ||
8292a17a | 1824 | icsk->icsk_af_ops = &ipv4_specific; |
d83d8461 | 1825 | icsk->icsk_sync_mss = tcp_sync_mss; |
cfb6eeb4 YH |
1826 | #ifdef CONFIG_TCP_MD5SIG |
1827 | tp->af_specific = &tcp_sock_ipv4_specific; | |
1828 | #endif | |
1da177e4 LT |
1829 | |
1830 | sk->sk_sndbuf = sysctl_tcp_wmem[1]; | |
1831 | sk->sk_rcvbuf = sysctl_tcp_rmem[1]; | |
1832 | ||
1833 | atomic_inc(&tcp_sockets_allocated); | |
1834 | ||
1835 | return 0; | |
1836 | } | |
1837 | ||
1838 | int tcp_v4_destroy_sock(struct sock *sk) | |
1839 | { | |
1840 | struct tcp_sock *tp = tcp_sk(sk); | |
1841 | ||
1842 | tcp_clear_xmit_timers(sk); | |
1843 | ||
6687e988 | 1844 | tcp_cleanup_congestion_control(sk); |
317a76f9 | 1845 | |
1da177e4 | 1846 | /* Cleanup up the write buffer. */ |
fe067e8a | 1847 | tcp_write_queue_purge(sk); |
1da177e4 LT |
1848 | |
1849 | /* Cleans up our, hopefully empty, out_of_order_queue. */ | |
e905a9ed | 1850 | __skb_queue_purge(&tp->out_of_order_queue); |
1da177e4 | 1851 | |
cfb6eeb4 YH |
1852 | #ifdef CONFIG_TCP_MD5SIG |
1853 | /* Clean up the MD5 key list, if any */ | |
1854 | if (tp->md5sig_info) { | |
1855 | tcp_v4_clear_md5_list(sk); | |
1856 | kfree(tp->md5sig_info); | |
1857 | tp->md5sig_info = NULL; | |
1858 | } | |
1859 | #endif | |
1860 | ||
1a2449a8 CL |
1861 | #ifdef CONFIG_NET_DMA |
1862 | /* Cleans up our sk_async_wait_queue */ | |
e905a9ed | 1863 | __skb_queue_purge(&sk->sk_async_wait_queue); |
1a2449a8 CL |
1864 | #endif |
1865 | ||
1da177e4 LT |
1866 | /* Clean prequeue, it must be empty really */ |
1867 | __skb_queue_purge(&tp->ucopy.prequeue); | |
1868 | ||
1869 | /* Clean up a referenced TCP bind bucket. */ | |
463c84b9 | 1870 | if (inet_csk(sk)->icsk_bind_hash) |
ab1e0a13 | 1871 | inet_put_port(sk); |
1da177e4 LT |
1872 | |
1873 | /* | |
1874 | * If sendmsg cached page exists, toss it. | |
1875 | */ | |
1876 | if (sk->sk_sndmsg_page) { | |
1877 | __free_page(sk->sk_sndmsg_page); | |
1878 | sk->sk_sndmsg_page = NULL; | |
1879 | } | |
1880 | ||
ec3c0982 PM |
1881 | if (tp->defer_tcp_accept.request) { |
1882 | reqsk_free(tp->defer_tcp_accept.request); | |
1883 | sock_put(tp->defer_tcp_accept.listen_sk); | |
1884 | sock_put(sk); | |
1885 | tp->defer_tcp_accept.listen_sk = NULL; | |
1886 | tp->defer_tcp_accept.request = NULL; | |
1887 | } | |
1888 | ||
1da177e4 LT |
1889 | atomic_dec(&tcp_sockets_allocated); |
1890 | ||
1891 | return 0; | |
1892 | } | |
1893 | ||
1894 | EXPORT_SYMBOL(tcp_v4_destroy_sock); | |
1895 | ||
1896 | #ifdef CONFIG_PROC_FS | |
1897 | /* Proc filesystem TCP sock list dumping. */ | |
1898 | ||
8feaf0c0 | 1899 | static inline struct inet_timewait_sock *tw_head(struct hlist_head *head) |
1da177e4 LT |
1900 | { |
1901 | return hlist_empty(head) ? NULL : | |
8feaf0c0 | 1902 | list_entry(head->first, struct inet_timewait_sock, tw_node); |
1da177e4 LT |
1903 | } |
1904 | ||
8feaf0c0 | 1905 | static inline struct inet_timewait_sock *tw_next(struct inet_timewait_sock *tw) |
1da177e4 LT |
1906 | { |
1907 | return tw->tw_node.next ? | |
1908 | hlist_entry(tw->tw_node.next, typeof(*tw), tw_node) : NULL; | |
1909 | } | |
1910 | ||
1911 | static void *listening_get_next(struct seq_file *seq, void *cur) | |
1912 | { | |
463c84b9 | 1913 | struct inet_connection_sock *icsk; |
1da177e4 LT |
1914 | struct hlist_node *node; |
1915 | struct sock *sk = cur; | |
1916 | struct tcp_iter_state* st = seq->private; | |
a4146b1b | 1917 | struct net *net = seq_file_net(seq); |
1da177e4 LT |
1918 | |
1919 | if (!sk) { | |
1920 | st->bucket = 0; | |
6e04e021 | 1921 | sk = sk_head(&tcp_hashinfo.listening_hash[0]); |
1da177e4 LT |
1922 | goto get_sk; |
1923 | } | |
1924 | ||
1925 | ++st->num; | |
1926 | ||
1927 | if (st->state == TCP_SEQ_STATE_OPENREQ) { | |
60236fdd | 1928 | struct request_sock *req = cur; |
1da177e4 | 1929 | |
72a3effa | 1930 | icsk = inet_csk(st->syn_wait_sk); |
1da177e4 LT |
1931 | req = req->dl_next; |
1932 | while (1) { | |
1933 | while (req) { | |
f40c8174 | 1934 | if (req->rsk_ops->family == st->family && |
878628fb | 1935 | net_eq(sock_net(req->sk), net)) { |
1da177e4 LT |
1936 | cur = req; |
1937 | goto out; | |
1938 | } | |
1939 | req = req->dl_next; | |
1940 | } | |
72a3effa | 1941 | if (++st->sbucket >= icsk->icsk_accept_queue.listen_opt->nr_table_entries) |
1da177e4 LT |
1942 | break; |
1943 | get_req: | |
463c84b9 | 1944 | req = icsk->icsk_accept_queue.listen_opt->syn_table[st->sbucket]; |
1da177e4 LT |
1945 | } |
1946 | sk = sk_next(st->syn_wait_sk); | |
1947 | st->state = TCP_SEQ_STATE_LISTENING; | |
463c84b9 | 1948 | read_unlock_bh(&icsk->icsk_accept_queue.syn_wait_lock); |
1da177e4 | 1949 | } else { |
e905a9ed | 1950 | icsk = inet_csk(sk); |
463c84b9 ACM |
1951 | read_lock_bh(&icsk->icsk_accept_queue.syn_wait_lock); |
1952 | if (reqsk_queue_len(&icsk->icsk_accept_queue)) | |
1da177e4 | 1953 | goto start_req; |
463c84b9 | 1954 | read_unlock_bh(&icsk->icsk_accept_queue.syn_wait_lock); |
1da177e4 LT |
1955 | sk = sk_next(sk); |
1956 | } | |
1957 | get_sk: | |
1958 | sk_for_each_from(sk, node) { | |
878628fb | 1959 | if (sk->sk_family == st->family && net_eq(sock_net(sk), net)) { |
1da177e4 LT |
1960 | cur = sk; |
1961 | goto out; | |
1962 | } | |
e905a9ed | 1963 | icsk = inet_csk(sk); |
463c84b9 ACM |
1964 | read_lock_bh(&icsk->icsk_accept_queue.syn_wait_lock); |
1965 | if (reqsk_queue_len(&icsk->icsk_accept_queue)) { | |
1da177e4 LT |
1966 | start_req: |
1967 | st->uid = sock_i_uid(sk); | |
1968 | st->syn_wait_sk = sk; | |
1969 | st->state = TCP_SEQ_STATE_OPENREQ; | |
1970 | st->sbucket = 0; | |
1971 | goto get_req; | |
1972 | } | |
463c84b9 | 1973 | read_unlock_bh(&icsk->icsk_accept_queue.syn_wait_lock); |
1da177e4 | 1974 | } |
0f7ff927 | 1975 | if (++st->bucket < INET_LHTABLE_SIZE) { |
6e04e021 | 1976 | sk = sk_head(&tcp_hashinfo.listening_hash[st->bucket]); |
1da177e4 LT |
1977 | goto get_sk; |
1978 | } | |
1979 | cur = NULL; | |
1980 | out: | |
1981 | return cur; | |
1982 | } | |
1983 | ||
1984 | static void *listening_get_idx(struct seq_file *seq, loff_t *pos) | |
1985 | { | |
1986 | void *rc = listening_get_next(seq, NULL); | |
1987 | ||
1988 | while (rc && *pos) { | |
1989 | rc = listening_get_next(seq, rc); | |
1990 | --*pos; | |
1991 | } | |
1992 | return rc; | |
1993 | } | |
1994 | ||
1995 | static void *established_get_first(struct seq_file *seq) | |
1996 | { | |
1997 | struct tcp_iter_state* st = seq->private; | |
a4146b1b | 1998 | struct net *net = seq_file_net(seq); |
1da177e4 LT |
1999 | void *rc = NULL; |
2000 | ||
6e04e021 | 2001 | for (st->bucket = 0; st->bucket < tcp_hashinfo.ehash_size; ++st->bucket) { |
1da177e4 LT |
2002 | struct sock *sk; |
2003 | struct hlist_node *node; | |
8feaf0c0 | 2004 | struct inet_timewait_sock *tw; |
230140cf | 2005 | rwlock_t *lock = inet_ehash_lockp(&tcp_hashinfo, st->bucket); |
1da177e4 | 2006 | |
230140cf | 2007 | read_lock_bh(lock); |
6e04e021 | 2008 | sk_for_each(sk, node, &tcp_hashinfo.ehash[st->bucket].chain) { |
f40c8174 | 2009 | if (sk->sk_family != st->family || |
878628fb | 2010 | !net_eq(sock_net(sk), net)) { |
1da177e4 LT |
2011 | continue; |
2012 | } | |
2013 | rc = sk; | |
2014 | goto out; | |
2015 | } | |
2016 | st->state = TCP_SEQ_STATE_TIME_WAIT; | |
8feaf0c0 | 2017 | inet_twsk_for_each(tw, node, |
dbca9b27 | 2018 | &tcp_hashinfo.ehash[st->bucket].twchain) { |
28518fc1 | 2019 | if (tw->tw_family != st->family || |
878628fb | 2020 | !net_eq(twsk_net(tw), net)) { |
1da177e4 LT |
2021 | continue; |
2022 | } | |
2023 | rc = tw; | |
2024 | goto out; | |
2025 | } | |
230140cf | 2026 | read_unlock_bh(lock); |
1da177e4 LT |
2027 | st->state = TCP_SEQ_STATE_ESTABLISHED; |
2028 | } | |
2029 | out: | |
2030 | return rc; | |
2031 | } | |
2032 | ||
2033 | static void *established_get_next(struct seq_file *seq, void *cur) | |
2034 | { | |
2035 | struct sock *sk = cur; | |
8feaf0c0 | 2036 | struct inet_timewait_sock *tw; |
1da177e4 LT |
2037 | struct hlist_node *node; |
2038 | struct tcp_iter_state* st = seq->private; | |
a4146b1b | 2039 | struct net *net = seq_file_net(seq); |
1da177e4 LT |
2040 | |
2041 | ++st->num; | |
2042 | ||
2043 | if (st->state == TCP_SEQ_STATE_TIME_WAIT) { | |
2044 | tw = cur; | |
2045 | tw = tw_next(tw); | |
2046 | get_tw: | |
878628fb | 2047 | while (tw && (tw->tw_family != st->family || !net_eq(twsk_net(tw), net))) { |
1da177e4 LT |
2048 | tw = tw_next(tw); |
2049 | } | |
2050 | if (tw) { | |
2051 | cur = tw; | |
2052 | goto out; | |
2053 | } | |
230140cf | 2054 | read_unlock_bh(inet_ehash_lockp(&tcp_hashinfo, st->bucket)); |
1da177e4 LT |
2055 | st->state = TCP_SEQ_STATE_ESTABLISHED; |
2056 | ||
6e04e021 | 2057 | if (++st->bucket < tcp_hashinfo.ehash_size) { |
230140cf | 2058 | read_lock_bh(inet_ehash_lockp(&tcp_hashinfo, st->bucket)); |
6e04e021 | 2059 | sk = sk_head(&tcp_hashinfo.ehash[st->bucket].chain); |
1da177e4 LT |
2060 | } else { |
2061 | cur = NULL; | |
2062 | goto out; | |
2063 | } | |
2064 | } else | |
2065 | sk = sk_next(sk); | |
2066 | ||
2067 | sk_for_each_from(sk, node) { | |
878628fb | 2068 | if (sk->sk_family == st->family && net_eq(sock_net(sk), net)) |
1da177e4 LT |
2069 | goto found; |
2070 | } | |
2071 | ||
2072 | st->state = TCP_SEQ_STATE_TIME_WAIT; | |
dbca9b27 | 2073 | tw = tw_head(&tcp_hashinfo.ehash[st->bucket].twchain); |
1da177e4 LT |
2074 | goto get_tw; |
2075 | found: | |
2076 | cur = sk; | |
2077 | out: | |
2078 | return cur; | |
2079 | } | |
2080 | ||
2081 | static void *established_get_idx(struct seq_file *seq, loff_t pos) | |
2082 | { | |
2083 | void *rc = established_get_first(seq); | |
2084 | ||
2085 | while (rc && pos) { | |
2086 | rc = established_get_next(seq, rc); | |
2087 | --pos; | |
7174259e | 2088 | } |
1da177e4 LT |
2089 | return rc; |
2090 | } | |
2091 | ||
2092 | static void *tcp_get_idx(struct seq_file *seq, loff_t pos) | |
2093 | { | |
2094 | void *rc; | |
2095 | struct tcp_iter_state* st = seq->private; | |
2096 | ||
f3f05f70 | 2097 | inet_listen_lock(&tcp_hashinfo); |
1da177e4 LT |
2098 | st->state = TCP_SEQ_STATE_LISTENING; |
2099 | rc = listening_get_idx(seq, &pos); | |
2100 | ||
2101 | if (!rc) { | |
f3f05f70 | 2102 | inet_listen_unlock(&tcp_hashinfo); |
1da177e4 LT |
2103 | st->state = TCP_SEQ_STATE_ESTABLISHED; |
2104 | rc = established_get_idx(seq, pos); | |
2105 | } | |
2106 | ||
2107 | return rc; | |
2108 | } | |
2109 | ||
2110 | static void *tcp_seq_start(struct seq_file *seq, loff_t *pos) | |
2111 | { | |
2112 | struct tcp_iter_state* st = seq->private; | |
2113 | st->state = TCP_SEQ_STATE_LISTENING; | |
2114 | st->num = 0; | |
2115 | return *pos ? tcp_get_idx(seq, *pos - 1) : SEQ_START_TOKEN; | |
2116 | } | |
2117 | ||
2118 | static void *tcp_seq_next(struct seq_file *seq, void *v, loff_t *pos) | |
2119 | { | |
2120 | void *rc = NULL; | |
2121 | struct tcp_iter_state* st; | |
2122 | ||
2123 | if (v == SEQ_START_TOKEN) { | |
2124 | rc = tcp_get_idx(seq, 0); | |
2125 | goto out; | |
2126 | } | |
2127 | st = seq->private; | |
2128 | ||
2129 | switch (st->state) { | |
2130 | case TCP_SEQ_STATE_OPENREQ: | |
2131 | case TCP_SEQ_STATE_LISTENING: | |
2132 | rc = listening_get_next(seq, v); | |
2133 | if (!rc) { | |
f3f05f70 | 2134 | inet_listen_unlock(&tcp_hashinfo); |
1da177e4 LT |
2135 | st->state = TCP_SEQ_STATE_ESTABLISHED; |
2136 | rc = established_get_first(seq); | |
2137 | } | |
2138 | break; | |
2139 | case TCP_SEQ_STATE_ESTABLISHED: | |
2140 | case TCP_SEQ_STATE_TIME_WAIT: | |
2141 | rc = established_get_next(seq, v); | |
2142 | break; | |
2143 | } | |
2144 | out: | |
2145 | ++*pos; | |
2146 | return rc; | |
2147 | } | |
2148 | ||
2149 | static void tcp_seq_stop(struct seq_file *seq, void *v) | |
2150 | { | |
2151 | struct tcp_iter_state* st = seq->private; | |
2152 | ||
2153 | switch (st->state) { | |
2154 | case TCP_SEQ_STATE_OPENREQ: | |
2155 | if (v) { | |
463c84b9 ACM |
2156 | struct inet_connection_sock *icsk = inet_csk(st->syn_wait_sk); |
2157 | read_unlock_bh(&icsk->icsk_accept_queue.syn_wait_lock); | |
1da177e4 LT |
2158 | } |
2159 | case TCP_SEQ_STATE_LISTENING: | |
2160 | if (v != SEQ_START_TOKEN) | |
f3f05f70 | 2161 | inet_listen_unlock(&tcp_hashinfo); |
1da177e4 LT |
2162 | break; |
2163 | case TCP_SEQ_STATE_TIME_WAIT: | |
2164 | case TCP_SEQ_STATE_ESTABLISHED: | |
2165 | if (v) | |
230140cf | 2166 | read_unlock_bh(inet_ehash_lockp(&tcp_hashinfo, st->bucket)); |
1da177e4 LT |
2167 | break; |
2168 | } | |
2169 | } | |
2170 | ||
2171 | static int tcp_seq_open(struct inode *inode, struct file *file) | |
2172 | { | |
2173 | struct tcp_seq_afinfo *afinfo = PDE(inode)->data; | |
1da177e4 | 2174 | struct tcp_iter_state *s; |
52d6f3f1 | 2175 | int err; |
1da177e4 | 2176 | |
52d6f3f1 DL |
2177 | err = seq_open_net(inode, file, &afinfo->seq_ops, |
2178 | sizeof(struct tcp_iter_state)); | |
2179 | if (err < 0) | |
2180 | return err; | |
f40c8174 | 2181 | |
52d6f3f1 | 2182 | s = ((struct seq_file *)file->private_data)->private; |
1da177e4 | 2183 | s->family = afinfo->family; |
f40c8174 DL |
2184 | return 0; |
2185 | } | |
2186 | ||
6f8b13bc | 2187 | int tcp_proc_register(struct net *net, struct tcp_seq_afinfo *afinfo) |
1da177e4 LT |
2188 | { |
2189 | int rc = 0; | |
2190 | struct proc_dir_entry *p; | |
2191 | ||
68fcadd1 DL |
2192 | afinfo->seq_fops.open = tcp_seq_open; |
2193 | afinfo->seq_fops.read = seq_read; | |
2194 | afinfo->seq_fops.llseek = seq_lseek; | |
2195 | afinfo->seq_fops.release = seq_release_net; | |
7174259e | 2196 | |
9427c4b3 DL |
2197 | afinfo->seq_ops.start = tcp_seq_start; |
2198 | afinfo->seq_ops.next = tcp_seq_next; | |
2199 | afinfo->seq_ops.stop = tcp_seq_stop; | |
2200 | ||
84841c3c DL |
2201 | p = proc_create_data(afinfo->name, S_IRUGO, net->proc_net, |
2202 | &afinfo->seq_fops, afinfo); | |
2203 | if (!p) | |
1da177e4 LT |
2204 | rc = -ENOMEM; |
2205 | return rc; | |
2206 | } | |
2207 | ||
6f8b13bc | 2208 | void tcp_proc_unregister(struct net *net, struct tcp_seq_afinfo *afinfo) |
1da177e4 | 2209 | { |
6f8b13bc | 2210 | proc_net_remove(net, afinfo->name); |
1da177e4 LT |
2211 | } |
2212 | ||
60236fdd | 2213 | static void get_openreq4(struct sock *sk, struct request_sock *req, |
5e659e4c | 2214 | struct seq_file *f, int i, int uid, int *len) |
1da177e4 | 2215 | { |
2e6599cb | 2216 | const struct inet_request_sock *ireq = inet_rsk(req); |
1da177e4 LT |
2217 | int ttd = req->expires - jiffies; |
2218 | ||
5e659e4c PE |
2219 | seq_printf(f, "%4d: %08X:%04X %08X:%04X" |
2220 | " %02X %08X:%08X %02X:%08lX %08X %5d %8d %u %d %p%n", | |
1da177e4 | 2221 | i, |
2e6599cb | 2222 | ireq->loc_addr, |
1da177e4 | 2223 | ntohs(inet_sk(sk)->sport), |
2e6599cb ACM |
2224 | ireq->rmt_addr, |
2225 | ntohs(ireq->rmt_port), | |
1da177e4 LT |
2226 | TCP_SYN_RECV, |
2227 | 0, 0, /* could print option size, but that is af dependent. */ | |
2228 | 1, /* timers active (only the expire timer) */ | |
2229 | jiffies_to_clock_t(ttd), | |
2230 | req->retrans, | |
2231 | uid, | |
2232 | 0, /* non standard timer */ | |
2233 | 0, /* open_requests have no inode */ | |
2234 | atomic_read(&sk->sk_refcnt), | |
5e659e4c PE |
2235 | req, |
2236 | len); | |
1da177e4 LT |
2237 | } |
2238 | ||
5e659e4c | 2239 | static void get_tcp4_sock(struct sock *sk, struct seq_file *f, int i, int *len) |
1da177e4 LT |
2240 | { |
2241 | int timer_active; | |
2242 | unsigned long timer_expires; | |
cf4c6bf8 IJ |
2243 | struct tcp_sock *tp = tcp_sk(sk); |
2244 | const struct inet_connection_sock *icsk = inet_csk(sk); | |
2245 | struct inet_sock *inet = inet_sk(sk); | |
714e85be AV |
2246 | __be32 dest = inet->daddr; |
2247 | __be32 src = inet->rcv_saddr; | |
1da177e4 LT |
2248 | __u16 destp = ntohs(inet->dport); |
2249 | __u16 srcp = ntohs(inet->sport); | |
2250 | ||
463c84b9 | 2251 | if (icsk->icsk_pending == ICSK_TIME_RETRANS) { |
1da177e4 | 2252 | timer_active = 1; |
463c84b9 ACM |
2253 | timer_expires = icsk->icsk_timeout; |
2254 | } else if (icsk->icsk_pending == ICSK_TIME_PROBE0) { | |
1da177e4 | 2255 | timer_active = 4; |
463c84b9 | 2256 | timer_expires = icsk->icsk_timeout; |
cf4c6bf8 | 2257 | } else if (timer_pending(&sk->sk_timer)) { |
1da177e4 | 2258 | timer_active = 2; |
cf4c6bf8 | 2259 | timer_expires = sk->sk_timer.expires; |
1da177e4 LT |
2260 | } else { |
2261 | timer_active = 0; | |
2262 | timer_expires = jiffies; | |
2263 | } | |
2264 | ||
5e659e4c PE |
2265 | seq_printf(f, "%4d: %08X:%04X %08X:%04X %02X %08X:%08X %02X:%08lX " |
2266 | "%08X %5d %8d %lu %d %p %u %u %u %u %d%n", | |
cf4c6bf8 | 2267 | i, src, srcp, dest, destp, sk->sk_state, |
47da8ee6 | 2268 | tp->write_seq - tp->snd_una, |
cf4c6bf8 | 2269 | sk->sk_state == TCP_LISTEN ? sk->sk_ack_backlog : |
7174259e | 2270 | (tp->rcv_nxt - tp->copied_seq), |
1da177e4 LT |
2271 | timer_active, |
2272 | jiffies_to_clock_t(timer_expires - jiffies), | |
463c84b9 | 2273 | icsk->icsk_retransmits, |
cf4c6bf8 | 2274 | sock_i_uid(sk), |
6687e988 | 2275 | icsk->icsk_probes_out, |
cf4c6bf8 IJ |
2276 | sock_i_ino(sk), |
2277 | atomic_read(&sk->sk_refcnt), sk, | |
463c84b9 ACM |
2278 | icsk->icsk_rto, |
2279 | icsk->icsk_ack.ato, | |
2280 | (icsk->icsk_ack.quick << 1) | icsk->icsk_ack.pingpong, | |
1da177e4 | 2281 | tp->snd_cwnd, |
5e659e4c PE |
2282 | tp->snd_ssthresh >= 0xFFFF ? -1 : tp->snd_ssthresh, |
2283 | len); | |
1da177e4 LT |
2284 | } |
2285 | ||
7174259e | 2286 | static void get_timewait4_sock(struct inet_timewait_sock *tw, |
5e659e4c | 2287 | struct seq_file *f, int i, int *len) |
1da177e4 | 2288 | { |
23f33c2d | 2289 | __be32 dest, src; |
1da177e4 LT |
2290 | __u16 destp, srcp; |
2291 | int ttd = tw->tw_ttd - jiffies; | |
2292 | ||
2293 | if (ttd < 0) | |
2294 | ttd = 0; | |
2295 | ||
2296 | dest = tw->tw_daddr; | |
2297 | src = tw->tw_rcv_saddr; | |
2298 | destp = ntohs(tw->tw_dport); | |
2299 | srcp = ntohs(tw->tw_sport); | |
2300 | ||
5e659e4c PE |
2301 | seq_printf(f, "%4d: %08X:%04X %08X:%04X" |
2302 | " %02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %p%n", | |
1da177e4 LT |
2303 | i, src, srcp, dest, destp, tw->tw_substate, 0, 0, |
2304 | 3, jiffies_to_clock_t(ttd), 0, 0, 0, 0, | |
5e659e4c | 2305 | atomic_read(&tw->tw_refcnt), tw, len); |
1da177e4 LT |
2306 | } |
2307 | ||
2308 | #define TMPSZ 150 | |
2309 | ||
2310 | static int tcp4_seq_show(struct seq_file *seq, void *v) | |
2311 | { | |
2312 | struct tcp_iter_state* st; | |
5e659e4c | 2313 | int len; |
1da177e4 LT |
2314 | |
2315 | if (v == SEQ_START_TOKEN) { | |
2316 | seq_printf(seq, "%-*s\n", TMPSZ - 1, | |
2317 | " sl local_address rem_address st tx_queue " | |
2318 | "rx_queue tr tm->when retrnsmt uid timeout " | |
2319 | "inode"); | |
2320 | goto out; | |
2321 | } | |
2322 | st = seq->private; | |
2323 | ||
2324 | switch (st->state) { | |
2325 | case TCP_SEQ_STATE_LISTENING: | |
2326 | case TCP_SEQ_STATE_ESTABLISHED: | |
5e659e4c | 2327 | get_tcp4_sock(v, seq, st->num, &len); |
1da177e4 LT |
2328 | break; |
2329 | case TCP_SEQ_STATE_OPENREQ: | |
5e659e4c | 2330 | get_openreq4(st->syn_wait_sk, v, seq, st->num, st->uid, &len); |
1da177e4 LT |
2331 | break; |
2332 | case TCP_SEQ_STATE_TIME_WAIT: | |
5e659e4c | 2333 | get_timewait4_sock(v, seq, st->num, &len); |
1da177e4 LT |
2334 | break; |
2335 | } | |
5e659e4c | 2336 | seq_printf(seq, "%*s\n", TMPSZ - 1 - len, ""); |
1da177e4 LT |
2337 | out: |
2338 | return 0; | |
2339 | } | |
2340 | ||
1da177e4 | 2341 | static struct tcp_seq_afinfo tcp4_seq_afinfo = { |
1da177e4 LT |
2342 | .name = "tcp", |
2343 | .family = AF_INET, | |
5f4472c5 DL |
2344 | .seq_fops = { |
2345 | .owner = THIS_MODULE, | |
2346 | }, | |
9427c4b3 DL |
2347 | .seq_ops = { |
2348 | .show = tcp4_seq_show, | |
2349 | }, | |
1da177e4 LT |
2350 | }; |
2351 | ||
757764f6 PE |
2352 | static int tcp4_proc_init_net(struct net *net) |
2353 | { | |
2354 | return tcp_proc_register(net, &tcp4_seq_afinfo); | |
2355 | } | |
2356 | ||
2357 | static void tcp4_proc_exit_net(struct net *net) | |
2358 | { | |
2359 | tcp_proc_unregister(net, &tcp4_seq_afinfo); | |
2360 | } | |
2361 | ||
2362 | static struct pernet_operations tcp4_net_ops = { | |
2363 | .init = tcp4_proc_init_net, | |
2364 | .exit = tcp4_proc_exit_net, | |
2365 | }; | |
2366 | ||
1da177e4 LT |
2367 | int __init tcp4_proc_init(void) |
2368 | { | |
757764f6 | 2369 | return register_pernet_subsys(&tcp4_net_ops); |
1da177e4 LT |
2370 | } |
2371 | ||
2372 | void tcp4_proc_exit(void) | |
2373 | { | |
757764f6 | 2374 | unregister_pernet_subsys(&tcp4_net_ops); |
1da177e4 LT |
2375 | } |
2376 | #endif /* CONFIG_PROC_FS */ | |
2377 | ||
2378 | struct proto tcp_prot = { | |
2379 | .name = "TCP", | |
2380 | .owner = THIS_MODULE, | |
2381 | .close = tcp_close, | |
2382 | .connect = tcp_v4_connect, | |
2383 | .disconnect = tcp_disconnect, | |
463c84b9 | 2384 | .accept = inet_csk_accept, |
1da177e4 LT |
2385 | .ioctl = tcp_ioctl, |
2386 | .init = tcp_v4_init_sock, | |
2387 | .destroy = tcp_v4_destroy_sock, | |
2388 | .shutdown = tcp_shutdown, | |
2389 | .setsockopt = tcp_setsockopt, | |
2390 | .getsockopt = tcp_getsockopt, | |
1da177e4 LT |
2391 | .recvmsg = tcp_recvmsg, |
2392 | .backlog_rcv = tcp_v4_do_rcv, | |
ab1e0a13 ACM |
2393 | .hash = inet_hash, |
2394 | .unhash = inet_unhash, | |
2395 | .get_port = inet_csk_get_port, | |
1da177e4 LT |
2396 | .enter_memory_pressure = tcp_enter_memory_pressure, |
2397 | .sockets_allocated = &tcp_sockets_allocated, | |
0a5578cf | 2398 | .orphan_count = &tcp_orphan_count, |
1da177e4 LT |
2399 | .memory_allocated = &tcp_memory_allocated, |
2400 | .memory_pressure = &tcp_memory_pressure, | |
2401 | .sysctl_mem = sysctl_tcp_mem, | |
2402 | .sysctl_wmem = sysctl_tcp_wmem, | |
2403 | .sysctl_rmem = sysctl_tcp_rmem, | |
2404 | .max_header = MAX_TCP_HEADER, | |
2405 | .obj_size = sizeof(struct tcp_sock), | |
6d6ee43e | 2406 | .twsk_prot = &tcp_timewait_sock_ops, |
60236fdd | 2407 | .rsk_prot = &tcp_request_sock_ops, |
39d8cda7 | 2408 | .h.hashinfo = &tcp_hashinfo, |
543d9cfe ACM |
2409 | #ifdef CONFIG_COMPAT |
2410 | .compat_setsockopt = compat_tcp_setsockopt, | |
2411 | .compat_getsockopt = compat_tcp_getsockopt, | |
2412 | #endif | |
1da177e4 LT |
2413 | }; |
2414 | ||
046ee902 DL |
2415 | |
2416 | static int __net_init tcp_sk_init(struct net *net) | |
2417 | { | |
2418 | return inet_ctl_sock_create(&net->ipv4.tcp_sock, | |
2419 | PF_INET, SOCK_RAW, IPPROTO_TCP, net); | |
2420 | } | |
2421 | ||
2422 | static void __net_exit tcp_sk_exit(struct net *net) | |
2423 | { | |
2424 | inet_ctl_sock_destroy(net->ipv4.tcp_sock); | |
2425 | } | |
2426 | ||
2427 | static struct pernet_operations __net_initdata tcp_sk_ops = { | |
2428 | .init = tcp_sk_init, | |
2429 | .exit = tcp_sk_exit, | |
2430 | }; | |
2431 | ||
9b0f976f | 2432 | void __init tcp_v4_init(void) |
1da177e4 | 2433 | { |
046ee902 | 2434 | if (register_pernet_device(&tcp_sk_ops)) |
1da177e4 | 2435 | panic("Failed to create the TCP control socket.\n"); |
1da177e4 LT |
2436 | } |
2437 | ||
2438 | EXPORT_SYMBOL(ipv4_specific); | |
1da177e4 | 2439 | EXPORT_SYMBOL(tcp_hashinfo); |
1da177e4 | 2440 | EXPORT_SYMBOL(tcp_prot); |
1da177e4 LT |
2441 | EXPORT_SYMBOL(tcp_v4_conn_request); |
2442 | EXPORT_SYMBOL(tcp_v4_connect); | |
2443 | EXPORT_SYMBOL(tcp_v4_do_rcv); | |
1da177e4 LT |
2444 | EXPORT_SYMBOL(tcp_v4_remember_stamp); |
2445 | EXPORT_SYMBOL(tcp_v4_send_check); | |
2446 | EXPORT_SYMBOL(tcp_v4_syn_recv_sock); | |
2447 | ||
2448 | #ifdef CONFIG_PROC_FS | |
2449 | EXPORT_SYMBOL(tcp_proc_register); | |
2450 | EXPORT_SYMBOL(tcp_proc_unregister); | |
2451 | #endif | |
1da177e4 | 2452 | EXPORT_SYMBOL(sysctl_tcp_low_latency); |
1da177e4 | 2453 |