]>
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 | * | |
1da177e4 LT |
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 | |
60236fdd | 37 | * request_sock handling and moved |
1da177e4 LT |
38 | * most of it into the af independent code. |
39 | * Added tail drop and some other bugfixes. | |
caa20d9a | 40 | * Added new listen semantics. |
1da177e4 LT |
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 | ||
1da177e4 | 53 | |
eb4dea58 | 54 | #include <linux/bottom_half.h> |
1da177e4 LT |
55 | #include <linux/types.h> |
56 | #include <linux/fcntl.h> | |
57 | #include <linux/module.h> | |
58 | #include <linux/random.h> | |
59 | #include <linux/cache.h> | |
60 | #include <linux/jhash.h> | |
61 | #include <linux/init.h> | |
62 | #include <linux/times.h> | |
63 | ||
457c4cbc | 64 | #include <net/net_namespace.h> |
1da177e4 | 65 | #include <net/icmp.h> |
304a1618 | 66 | #include <net/inet_hashtables.h> |
1da177e4 | 67 | #include <net/tcp.h> |
20380731 | 68 | #include <net/transp_v6.h> |
1da177e4 LT |
69 | #include <net/ipv6.h> |
70 | #include <net/inet_common.h> | |
6d6ee43e | 71 | #include <net/timewait_sock.h> |
1da177e4 | 72 | #include <net/xfrm.h> |
1a2449a8 | 73 | #include <net/netdma.h> |
1da177e4 LT |
74 | |
75 | #include <linux/inet.h> | |
76 | #include <linux/ipv6.h> | |
77 | #include <linux/stddef.h> | |
78 | #include <linux/proc_fs.h> | |
79 | #include <linux/seq_file.h> | |
80 | ||
cfb6eeb4 YH |
81 | #include <linux/crypto.h> |
82 | #include <linux/scatterlist.h> | |
83 | ||
ab32ea5d BH |
84 | int sysctl_tcp_tw_reuse __read_mostly; |
85 | int sysctl_tcp_low_latency __read_mostly; | |
1da177e4 | 86 | |
1da177e4 | 87 | |
cfb6eeb4 | 88 | #ifdef CONFIG_TCP_MD5SIG |
7174259e ACM |
89 | static struct tcp_md5sig_key *tcp_v4_md5_do_lookup(struct sock *sk, |
90 | __be32 addr); | |
49a72dfb AL |
91 | static int tcp_v4_md5_hash_hdr(char *md5_hash, struct tcp_md5sig_key *key, |
92 | __be32 daddr, __be32 saddr, struct tcphdr *th); | |
9501f972 YH |
93 | #else |
94 | static inline | |
95 | struct tcp_md5sig_key *tcp_v4_md5_do_lookup(struct sock *sk, __be32 addr) | |
96 | { | |
97 | return NULL; | |
98 | } | |
cfb6eeb4 YH |
99 | #endif |
100 | ||
5caea4ea | 101 | struct inet_hashinfo tcp_hashinfo; |
1da177e4 | 102 | |
a94f723d | 103 | static inline __u32 tcp_v4_init_sequence(struct sk_buff *skb) |
1da177e4 | 104 | { |
eddc9ec5 ACM |
105 | return secure_tcp_sequence_number(ip_hdr(skb)->daddr, |
106 | ip_hdr(skb)->saddr, | |
aa8223c7 ACM |
107 | tcp_hdr(skb)->dest, |
108 | tcp_hdr(skb)->source); | |
1da177e4 LT |
109 | } |
110 | ||
6d6ee43e ACM |
111 | int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp) |
112 | { | |
113 | const struct tcp_timewait_sock *tcptw = tcp_twsk(sktw); | |
114 | struct tcp_sock *tp = tcp_sk(sk); | |
115 | ||
116 | /* With PAWS, it is safe from the viewpoint | |
117 | of data integrity. Even without PAWS it is safe provided sequence | |
118 | spaces do not overlap i.e. at data rates <= 80Mbit/sec. | |
119 | ||
120 | Actually, the idea is close to VJ's one, only timestamp cache is | |
121 | held not per host, but per port pair and TW bucket is used as state | |
122 | holder. | |
123 | ||
124 | If TW bucket has been already destroyed we fall back to VJ's scheme | |
125 | and use initial timestamp retrieved from peer table. | |
126 | */ | |
127 | if (tcptw->tw_ts_recent_stamp && | |
128 | (twp == NULL || (sysctl_tcp_tw_reuse && | |
9d729f72 | 129 | get_seconds() - tcptw->tw_ts_recent_stamp > 1))) { |
6d6ee43e ACM |
130 | tp->write_seq = tcptw->tw_snd_nxt + 65535 + 2; |
131 | if (tp->write_seq == 0) | |
132 | tp->write_seq = 1; | |
133 | tp->rx_opt.ts_recent = tcptw->tw_ts_recent; | |
134 | tp->rx_opt.ts_recent_stamp = tcptw->tw_ts_recent_stamp; | |
135 | sock_hold(sktw); | |
136 | return 1; | |
137 | } | |
138 | ||
139 | return 0; | |
140 | } | |
141 | ||
142 | EXPORT_SYMBOL_GPL(tcp_twsk_unique); | |
143 | ||
1da177e4 LT |
144 | /* This will initiate an outgoing connection. */ |
145 | int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len) | |
146 | { | |
147 | struct inet_sock *inet = inet_sk(sk); | |
148 | struct tcp_sock *tp = tcp_sk(sk); | |
149 | struct sockaddr_in *usin = (struct sockaddr_in *)uaddr; | |
150 | struct rtable *rt; | |
bada8adc | 151 | __be32 daddr, nexthop; |
1da177e4 LT |
152 | int tmp; |
153 | int err; | |
154 | ||
155 | if (addr_len < sizeof(struct sockaddr_in)) | |
156 | return -EINVAL; | |
157 | ||
158 | if (usin->sin_family != AF_INET) | |
159 | return -EAFNOSUPPORT; | |
160 | ||
161 | nexthop = daddr = usin->sin_addr.s_addr; | |
162 | if (inet->opt && inet->opt->srr) { | |
163 | if (!daddr) | |
164 | return -EINVAL; | |
165 | nexthop = inet->opt->faddr; | |
166 | } | |
167 | ||
c720c7e8 | 168 | tmp = ip_route_connect(&rt, nexthop, inet->inet_saddr, |
1da177e4 LT |
169 | RT_CONN_FLAGS(sk), sk->sk_bound_dev_if, |
170 | IPPROTO_TCP, | |
c720c7e8 | 171 | inet->inet_sport, usin->sin_port, sk, 1); |
584bdf8c WD |
172 | if (tmp < 0) { |
173 | if (tmp == -ENETUNREACH) | |
7c73a6fa | 174 | IP_INC_STATS_BH(sock_net(sk), IPSTATS_MIB_OUTNOROUTES); |
1da177e4 | 175 | return tmp; |
584bdf8c | 176 | } |
1da177e4 LT |
177 | |
178 | if (rt->rt_flags & (RTCF_MULTICAST | RTCF_BROADCAST)) { | |
179 | ip_rt_put(rt); | |
180 | return -ENETUNREACH; | |
181 | } | |
182 | ||
183 | if (!inet->opt || !inet->opt->srr) | |
184 | daddr = rt->rt_dst; | |
185 | ||
c720c7e8 ED |
186 | if (!inet->inet_saddr) |
187 | inet->inet_saddr = rt->rt_src; | |
188 | inet->inet_rcv_saddr = inet->inet_saddr; | |
1da177e4 | 189 | |
c720c7e8 | 190 | if (tp->rx_opt.ts_recent_stamp && inet->inet_daddr != daddr) { |
1da177e4 LT |
191 | /* Reset inherited state */ |
192 | tp->rx_opt.ts_recent = 0; | |
193 | tp->rx_opt.ts_recent_stamp = 0; | |
194 | tp->write_seq = 0; | |
195 | } | |
196 | ||
295ff7ed | 197 | if (tcp_death_row.sysctl_tw_recycle && |
1da177e4 LT |
198 | !tp->rx_opt.ts_recent_stamp && rt->rt_dst == daddr) { |
199 | struct inet_peer *peer = rt_get_peer(rt); | |
7174259e ACM |
200 | /* |
201 | * VJ's idea. We save last timestamp seen from | |
202 | * the destination in peer table, when entering state | |
203 | * TIME-WAIT * and initialize rx_opt.ts_recent from it, | |
204 | * when trying new connection. | |
1da177e4 | 205 | */ |
7174259e | 206 | if (peer != NULL && |
2c1409a0 | 207 | (u32)get_seconds() - peer->tcp_ts_stamp <= TCP_PAWS_MSL) { |
1da177e4 LT |
208 | tp->rx_opt.ts_recent_stamp = peer->tcp_ts_stamp; |
209 | tp->rx_opt.ts_recent = peer->tcp_ts; | |
210 | } | |
211 | } | |
212 | ||
c720c7e8 ED |
213 | inet->inet_dport = usin->sin_port; |
214 | inet->inet_daddr = daddr; | |
1da177e4 | 215 | |
d83d8461 | 216 | inet_csk(sk)->icsk_ext_hdr_len = 0; |
1da177e4 | 217 | if (inet->opt) |
d83d8461 | 218 | inet_csk(sk)->icsk_ext_hdr_len = inet->opt->optlen; |
1da177e4 | 219 | |
bee7ca9e | 220 | tp->rx_opt.mss_clamp = TCP_MSS_DEFAULT; |
1da177e4 LT |
221 | |
222 | /* Socket identity is still unknown (sport may be zero). | |
223 | * However we set state to SYN-SENT and not releasing socket | |
224 | * lock select source port, enter ourselves into the hash tables and | |
225 | * complete initialization after this. | |
226 | */ | |
227 | tcp_set_state(sk, TCP_SYN_SENT); | |
a7f5e7f1 | 228 | err = inet_hash_connect(&tcp_death_row, sk); |
1da177e4 LT |
229 | if (err) |
230 | goto failure; | |
231 | ||
7174259e | 232 | err = ip_route_newports(&rt, IPPROTO_TCP, |
c720c7e8 | 233 | inet->inet_sport, inet->inet_dport, sk); |
1da177e4 LT |
234 | if (err) |
235 | goto failure; | |
236 | ||
237 | /* OK, now commit destination to socket. */ | |
bcd76111 | 238 | sk->sk_gso_type = SKB_GSO_TCPV4; |
6cbb0df7 | 239 | sk_setup_caps(sk, &rt->u.dst); |
1da177e4 LT |
240 | |
241 | if (!tp->write_seq) | |
c720c7e8 ED |
242 | tp->write_seq = secure_tcp_sequence_number(inet->inet_saddr, |
243 | inet->inet_daddr, | |
244 | inet->inet_sport, | |
1da177e4 LT |
245 | usin->sin_port); |
246 | ||
c720c7e8 | 247 | inet->inet_id = tp->write_seq ^ jiffies; |
1da177e4 LT |
248 | |
249 | err = tcp_connect(sk); | |
250 | rt = NULL; | |
251 | if (err) | |
252 | goto failure; | |
253 | ||
254 | return 0; | |
255 | ||
256 | failure: | |
7174259e ACM |
257 | /* |
258 | * This unhashes the socket and releases the local port, | |
259 | * if necessary. | |
260 | */ | |
1da177e4 LT |
261 | tcp_set_state(sk, TCP_CLOSE); |
262 | ip_rt_put(rt); | |
263 | sk->sk_route_caps = 0; | |
c720c7e8 | 264 | inet->inet_dport = 0; |
1da177e4 LT |
265 | return err; |
266 | } | |
267 | ||
1da177e4 LT |
268 | /* |
269 | * This routine does path mtu discovery as defined in RFC1191. | |
270 | */ | |
40efc6fa | 271 | static void do_pmtu_discovery(struct sock *sk, struct iphdr *iph, u32 mtu) |
1da177e4 LT |
272 | { |
273 | struct dst_entry *dst; | |
274 | struct inet_sock *inet = inet_sk(sk); | |
1da177e4 LT |
275 | |
276 | /* We are not interested in TCP_LISTEN and open_requests (SYN-ACKs | |
277 | * send out by Linux are always <576bytes so they should go through | |
278 | * unfragmented). | |
279 | */ | |
280 | if (sk->sk_state == TCP_LISTEN) | |
281 | return; | |
282 | ||
283 | /* We don't check in the destentry if pmtu discovery is forbidden | |
284 | * on this route. We just assume that no packet_to_big packets | |
285 | * are send back when pmtu discovery is not active. | |
e905a9ed | 286 | * There is a small race when the user changes this flag in the |
1da177e4 LT |
287 | * route, but I think that's acceptable. |
288 | */ | |
289 | if ((dst = __sk_dst_check(sk, 0)) == NULL) | |
290 | return; | |
291 | ||
292 | dst->ops->update_pmtu(dst, mtu); | |
293 | ||
294 | /* Something is about to be wrong... Remember soft error | |
295 | * for the case, if this connection will not able to recover. | |
296 | */ | |
297 | if (mtu < dst_mtu(dst) && ip_dont_fragment(sk, dst)) | |
298 | sk->sk_err_soft = EMSGSIZE; | |
299 | ||
300 | mtu = dst_mtu(dst); | |
301 | ||
302 | if (inet->pmtudisc != IP_PMTUDISC_DONT && | |
d83d8461 | 303 | inet_csk(sk)->icsk_pmtu_cookie > mtu) { |
1da177e4 LT |
304 | tcp_sync_mss(sk, mtu); |
305 | ||
306 | /* Resend the TCP packet because it's | |
307 | * clear that the old packet has been | |
308 | * dropped. This is the new "fast" path mtu | |
309 | * discovery. | |
310 | */ | |
311 | tcp_simple_retransmit(sk); | |
312 | } /* else let the usual retransmit timer handle it */ | |
313 | } | |
314 | ||
315 | /* | |
316 | * This routine is called by the ICMP module when it gets some | |
317 | * sort of error condition. If err < 0 then the socket should | |
318 | * be closed and the error returned to the user. If err > 0 | |
319 | * it's just the icmp type << 8 | icmp code. After adjustment | |
320 | * header points to the first 8 bytes of the tcp header. We need | |
321 | * to find the appropriate port. | |
322 | * | |
323 | * The locking strategy used here is very "optimistic". When | |
324 | * someone else accesses the socket the ICMP is just dropped | |
325 | * and for some paths there is no check at all. | |
326 | * A more general error queue to queue errors for later handling | |
327 | * is probably better. | |
328 | * | |
329 | */ | |
330 | ||
4d1a2d9e | 331 | void tcp_v4_err(struct sk_buff *icmp_skb, u32 info) |
1da177e4 | 332 | { |
4d1a2d9e DL |
333 | struct iphdr *iph = (struct iphdr *)icmp_skb->data; |
334 | struct tcphdr *th = (struct tcphdr *)(icmp_skb->data + (iph->ihl << 2)); | |
f1ecd5d9 | 335 | struct inet_connection_sock *icsk; |
1da177e4 LT |
336 | struct tcp_sock *tp; |
337 | struct inet_sock *inet; | |
4d1a2d9e DL |
338 | const int type = icmp_hdr(icmp_skb)->type; |
339 | const int code = icmp_hdr(icmp_skb)->code; | |
1da177e4 | 340 | struct sock *sk; |
f1ecd5d9 | 341 | struct sk_buff *skb; |
1da177e4 | 342 | __u32 seq; |
f1ecd5d9 | 343 | __u32 remaining; |
1da177e4 | 344 | int err; |
4d1a2d9e | 345 | struct net *net = dev_net(icmp_skb->dev); |
1da177e4 | 346 | |
4d1a2d9e | 347 | if (icmp_skb->len < (iph->ihl << 2) + 8) { |
dcfc23ca | 348 | ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS); |
1da177e4 LT |
349 | return; |
350 | } | |
351 | ||
fd54d716 | 352 | sk = inet_lookup(net, &tcp_hashinfo, iph->daddr, th->dest, |
4d1a2d9e | 353 | iph->saddr, th->source, inet_iif(icmp_skb)); |
1da177e4 | 354 | if (!sk) { |
dcfc23ca | 355 | ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS); |
1da177e4 LT |
356 | return; |
357 | } | |
358 | if (sk->sk_state == TCP_TIME_WAIT) { | |
9469c7b4 | 359 | inet_twsk_put(inet_twsk(sk)); |
1da177e4 LT |
360 | return; |
361 | } | |
362 | ||
363 | bh_lock_sock(sk); | |
364 | /* If too many ICMPs get dropped on busy | |
365 | * servers this needs to be solved differently. | |
366 | */ | |
367 | if (sock_owned_by_user(sk)) | |
de0744af | 368 | NET_INC_STATS_BH(net, LINUX_MIB_LOCKDROPPEDICMPS); |
1da177e4 LT |
369 | |
370 | if (sk->sk_state == TCP_CLOSE) | |
371 | goto out; | |
372 | ||
f1ecd5d9 | 373 | icsk = inet_csk(sk); |
1da177e4 LT |
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)) { | |
de0744af | 378 | NET_INC_STATS_BH(net, 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; | |
f1ecd5d9 DL |
400 | /* check if icmp_skb allows revert of backoff |
401 | * (see draft-zimmermann-tcp-lcd) */ | |
402 | if (code != ICMP_NET_UNREACH && code != ICMP_HOST_UNREACH) | |
403 | break; | |
404 | if (seq != tp->snd_una || !icsk->icsk_retransmits || | |
405 | !icsk->icsk_backoff) | |
406 | break; | |
407 | ||
408 | icsk->icsk_backoff--; | |
409 | inet_csk(sk)->icsk_rto = __tcp_set_rto(tp) << | |
410 | icsk->icsk_backoff; | |
411 | tcp_bound_rto(sk); | |
412 | ||
413 | skb = tcp_write_queue_head(sk); | |
414 | BUG_ON(!skb); | |
415 | ||
416 | remaining = icsk->icsk_rto - min(icsk->icsk_rto, | |
417 | tcp_time_stamp - TCP_SKB_CB(skb)->when); | |
418 | ||
419 | if (remaining) { | |
420 | inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS, | |
421 | remaining, TCP_RTO_MAX); | |
422 | } else if (sock_owned_by_user(sk)) { | |
423 | /* RTO revert clocked out retransmission, | |
424 | * but socket is locked. Will defer. */ | |
425 | inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS, | |
426 | HZ/20, TCP_RTO_MAX); | |
427 | } else { | |
428 | /* RTO revert clocked out retransmission. | |
429 | * Will retransmit now */ | |
430 | tcp_retransmit_timer(sk); | |
431 | } | |
432 | ||
1da177e4 LT |
433 | break; |
434 | case ICMP_TIME_EXCEEDED: | |
435 | err = EHOSTUNREACH; | |
436 | break; | |
437 | default: | |
438 | goto out; | |
439 | } | |
440 | ||
441 | switch (sk->sk_state) { | |
60236fdd | 442 | struct request_sock *req, **prev; |
1da177e4 LT |
443 | case TCP_LISTEN: |
444 | if (sock_owned_by_user(sk)) | |
445 | goto out; | |
446 | ||
463c84b9 ACM |
447 | req = inet_csk_search_req(sk, &prev, th->dest, |
448 | iph->daddr, iph->saddr); | |
1da177e4 LT |
449 | if (!req) |
450 | goto out; | |
451 | ||
452 | /* ICMPs are not backlogged, hence we cannot get | |
453 | an established socket here. | |
454 | */ | |
547b792c | 455 | WARN_ON(req->sk); |
1da177e4 | 456 | |
2e6599cb | 457 | if (seq != tcp_rsk(req)->snt_isn) { |
de0744af | 458 | NET_INC_STATS_BH(net, LINUX_MIB_OUTOFWINDOWICMPS); |
1da177e4 LT |
459 | goto out; |
460 | } | |
461 | ||
462 | /* | |
463 | * Still in SYN_RECV, just remove it silently. | |
464 | * There is no good way to pass the error to the newly | |
465 | * created socket, and POSIX does not want network | |
466 | * errors returned from accept(). | |
467 | */ | |
463c84b9 | 468 | inet_csk_reqsk_queue_drop(sk, req, prev); |
1da177e4 LT |
469 | goto out; |
470 | ||
471 | case TCP_SYN_SENT: | |
472 | case TCP_SYN_RECV: /* Cannot happen. | |
473 | It can f.e. if SYNs crossed. | |
474 | */ | |
475 | if (!sock_owned_by_user(sk)) { | |
1da177e4 LT |
476 | sk->sk_err = err; |
477 | ||
478 | sk->sk_error_report(sk); | |
479 | ||
480 | tcp_done(sk); | |
481 | } else { | |
482 | sk->sk_err_soft = err; | |
483 | } | |
484 | goto out; | |
485 | } | |
486 | ||
487 | /* If we've already connected we will keep trying | |
488 | * until we time out, or the user gives up. | |
489 | * | |
490 | * rfc1122 4.2.3.9 allows to consider as hard errors | |
491 | * only PROTO_UNREACH and PORT_UNREACH (well, FRAG_FAILED too, | |
492 | * but it is obsoleted by pmtu discovery). | |
493 | * | |
494 | * Note, that in modern internet, where routing is unreliable | |
495 | * and in each dark corner broken firewalls sit, sending random | |
496 | * errors ordered by their masters even this two messages finally lose | |
497 | * their original sense (even Linux sends invalid PORT_UNREACHs) | |
498 | * | |
499 | * Now we are in compliance with RFCs. | |
500 | * --ANK (980905) | |
501 | */ | |
502 | ||
503 | inet = inet_sk(sk); | |
504 | if (!sock_owned_by_user(sk) && inet->recverr) { | |
505 | sk->sk_err = err; | |
506 | sk->sk_error_report(sk); | |
507 | } else { /* Only an error on timeout */ | |
508 | sk->sk_err_soft = err; | |
509 | } | |
510 | ||
511 | out: | |
512 | bh_unlock_sock(sk); | |
513 | sock_put(sk); | |
514 | } | |
515 | ||
516 | /* This routine computes an IPv4 TCP checksum. */ | |
8292a17a | 517 | void tcp_v4_send_check(struct sock *sk, int len, struct sk_buff *skb) |
1da177e4 LT |
518 | { |
519 | struct inet_sock *inet = inet_sk(sk); | |
aa8223c7 | 520 | struct tcphdr *th = tcp_hdr(skb); |
1da177e4 | 521 | |
84fa7933 | 522 | if (skb->ip_summed == CHECKSUM_PARTIAL) { |
c720c7e8 ED |
523 | th->check = ~tcp_v4_check(len, inet->inet_saddr, |
524 | inet->inet_daddr, 0); | |
663ead3b | 525 | skb->csum_start = skb_transport_header(skb) - skb->head; |
ff1dcadb | 526 | skb->csum_offset = offsetof(struct tcphdr, check); |
1da177e4 | 527 | } else { |
c720c7e8 ED |
528 | th->check = tcp_v4_check(len, inet->inet_saddr, |
529 | inet->inet_daddr, | |
07f0757a | 530 | csum_partial(th, |
1da177e4 LT |
531 | th->doff << 2, |
532 | skb->csum)); | |
533 | } | |
534 | } | |
535 | ||
a430a43d HX |
536 | int tcp_v4_gso_send_check(struct sk_buff *skb) |
537 | { | |
eddc9ec5 | 538 | const struct iphdr *iph; |
a430a43d HX |
539 | struct tcphdr *th; |
540 | ||
541 | if (!pskb_may_pull(skb, sizeof(*th))) | |
542 | return -EINVAL; | |
543 | ||
eddc9ec5 | 544 | iph = ip_hdr(skb); |
aa8223c7 | 545 | th = tcp_hdr(skb); |
a430a43d HX |
546 | |
547 | th->check = 0; | |
ba7808ea | 548 | th->check = ~tcp_v4_check(skb->len, iph->saddr, iph->daddr, 0); |
663ead3b | 549 | skb->csum_start = skb_transport_header(skb) - skb->head; |
ff1dcadb | 550 | skb->csum_offset = offsetof(struct tcphdr, check); |
84fa7933 | 551 | skb->ip_summed = CHECKSUM_PARTIAL; |
a430a43d HX |
552 | return 0; |
553 | } | |
554 | ||
1da177e4 LT |
555 | /* |
556 | * This routine will send an RST to the other tcp. | |
557 | * | |
558 | * Someone asks: why I NEVER use socket parameters (TOS, TTL etc.) | |
559 | * for reset. | |
560 | * Answer: if a packet caused RST, it is not for a socket | |
561 | * existing in our system, if it is matched to a socket, | |
562 | * it is just duplicate segment or bug in other side's TCP. | |
563 | * So that we build reply only basing on parameters | |
564 | * arrived with segment. | |
565 | * Exception: precedence violation. We do not implement it in any case. | |
566 | */ | |
567 | ||
cfb6eeb4 | 568 | static void tcp_v4_send_reset(struct sock *sk, struct sk_buff *skb) |
1da177e4 | 569 | { |
aa8223c7 | 570 | struct tcphdr *th = tcp_hdr(skb); |
cfb6eeb4 YH |
571 | struct { |
572 | struct tcphdr th; | |
573 | #ifdef CONFIG_TCP_MD5SIG | |
714e85be | 574 | __be32 opt[(TCPOLEN_MD5SIG_ALIGNED >> 2)]; |
cfb6eeb4 YH |
575 | #endif |
576 | } rep; | |
1da177e4 | 577 | struct ip_reply_arg arg; |
cfb6eeb4 YH |
578 | #ifdef CONFIG_TCP_MD5SIG |
579 | struct tcp_md5sig_key *key; | |
580 | #endif | |
a86b1e30 | 581 | struct net *net; |
1da177e4 LT |
582 | |
583 | /* Never send a reset in response to a reset. */ | |
584 | if (th->rst) | |
585 | return; | |
586 | ||
511c3f92 | 587 | if (skb_rtable(skb)->rt_type != RTN_LOCAL) |
1da177e4 LT |
588 | return; |
589 | ||
590 | /* Swap the send and the receive. */ | |
cfb6eeb4 YH |
591 | memset(&rep, 0, sizeof(rep)); |
592 | rep.th.dest = th->source; | |
593 | rep.th.source = th->dest; | |
594 | rep.th.doff = sizeof(struct tcphdr) / 4; | |
595 | rep.th.rst = 1; | |
1da177e4 LT |
596 | |
597 | if (th->ack) { | |
cfb6eeb4 | 598 | rep.th.seq = th->ack_seq; |
1da177e4 | 599 | } else { |
cfb6eeb4 YH |
600 | rep.th.ack = 1; |
601 | rep.th.ack_seq = htonl(ntohl(th->seq) + th->syn + th->fin + | |
602 | skb->len - (th->doff << 2)); | |
1da177e4 LT |
603 | } |
604 | ||
7174259e | 605 | memset(&arg, 0, sizeof(arg)); |
cfb6eeb4 YH |
606 | arg.iov[0].iov_base = (unsigned char *)&rep; |
607 | arg.iov[0].iov_len = sizeof(rep.th); | |
608 | ||
609 | #ifdef CONFIG_TCP_MD5SIG | |
eddc9ec5 | 610 | key = sk ? tcp_v4_md5_do_lookup(sk, ip_hdr(skb)->daddr) : NULL; |
cfb6eeb4 YH |
611 | if (key) { |
612 | rep.opt[0] = htonl((TCPOPT_NOP << 24) | | |
613 | (TCPOPT_NOP << 16) | | |
614 | (TCPOPT_MD5SIG << 8) | | |
615 | TCPOLEN_MD5SIG); | |
616 | /* Update length and the length the header thinks exists */ | |
617 | arg.iov[0].iov_len += TCPOLEN_MD5SIG_ALIGNED; | |
618 | rep.th.doff = arg.iov[0].iov_len / 4; | |
619 | ||
49a72dfb | 620 | tcp_v4_md5_hash_hdr((__u8 *) &rep.opt[1], |
78e645cb IJ |
621 | key, ip_hdr(skb)->saddr, |
622 | ip_hdr(skb)->daddr, &rep.th); | |
cfb6eeb4 YH |
623 | } |
624 | #endif | |
eddc9ec5 ACM |
625 | arg.csum = csum_tcpudp_nofold(ip_hdr(skb)->daddr, |
626 | ip_hdr(skb)->saddr, /* XXX */ | |
52cd5750 | 627 | arg.iov[0].iov_len, IPPROTO_TCP, 0); |
1da177e4 | 628 | arg.csumoffset = offsetof(struct tcphdr, check) / 2; |
88ef4a5a | 629 | arg.flags = (sk && inet_sk(sk)->transparent) ? IP_REPLY_ARG_NOSRCCHECK : 0; |
1da177e4 | 630 | |
adf30907 | 631 | net = dev_net(skb_dst(skb)->dev); |
a86b1e30 | 632 | ip_send_reply(net->ipv4.tcp_sock, skb, |
7feb49c8 | 633 | &arg, arg.iov[0].iov_len); |
1da177e4 | 634 | |
63231bdd PE |
635 | TCP_INC_STATS_BH(net, TCP_MIB_OUTSEGS); |
636 | TCP_INC_STATS_BH(net, TCP_MIB_OUTRSTS); | |
1da177e4 LT |
637 | } |
638 | ||
639 | /* The code following below sending ACKs in SYN-RECV and TIME-WAIT states | |
640 | outside socket context is ugly, certainly. What can I do? | |
641 | */ | |
642 | ||
9501f972 YH |
643 | static void tcp_v4_send_ack(struct sk_buff *skb, u32 seq, u32 ack, |
644 | u32 win, u32 ts, int oif, | |
88ef4a5a KK |
645 | struct tcp_md5sig_key *key, |
646 | int reply_flags) | |
1da177e4 | 647 | { |
aa8223c7 | 648 | struct tcphdr *th = tcp_hdr(skb); |
1da177e4 LT |
649 | struct { |
650 | struct tcphdr th; | |
714e85be | 651 | __be32 opt[(TCPOLEN_TSTAMP_ALIGNED >> 2) |
cfb6eeb4 | 652 | #ifdef CONFIG_TCP_MD5SIG |
714e85be | 653 | + (TCPOLEN_MD5SIG_ALIGNED >> 2) |
cfb6eeb4 YH |
654 | #endif |
655 | ]; | |
1da177e4 LT |
656 | } rep; |
657 | struct ip_reply_arg arg; | |
adf30907 | 658 | struct net *net = dev_net(skb_dst(skb)->dev); |
1da177e4 LT |
659 | |
660 | memset(&rep.th, 0, sizeof(struct tcphdr)); | |
7174259e | 661 | memset(&arg, 0, sizeof(arg)); |
1da177e4 LT |
662 | |
663 | arg.iov[0].iov_base = (unsigned char *)&rep; | |
664 | arg.iov[0].iov_len = sizeof(rep.th); | |
665 | if (ts) { | |
cfb6eeb4 YH |
666 | rep.opt[0] = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) | |
667 | (TCPOPT_TIMESTAMP << 8) | | |
668 | TCPOLEN_TIMESTAMP); | |
669 | rep.opt[1] = htonl(tcp_time_stamp); | |
670 | rep.opt[2] = htonl(ts); | |
cb48cfe8 | 671 | arg.iov[0].iov_len += TCPOLEN_TSTAMP_ALIGNED; |
1da177e4 LT |
672 | } |
673 | ||
674 | /* Swap the send and the receive. */ | |
675 | rep.th.dest = th->source; | |
676 | rep.th.source = th->dest; | |
677 | rep.th.doff = arg.iov[0].iov_len / 4; | |
678 | rep.th.seq = htonl(seq); | |
679 | rep.th.ack_seq = htonl(ack); | |
680 | rep.th.ack = 1; | |
681 | rep.th.window = htons(win); | |
682 | ||
cfb6eeb4 | 683 | #ifdef CONFIG_TCP_MD5SIG |
cfb6eeb4 YH |
684 | if (key) { |
685 | int offset = (ts) ? 3 : 0; | |
686 | ||
687 | rep.opt[offset++] = htonl((TCPOPT_NOP << 24) | | |
688 | (TCPOPT_NOP << 16) | | |
689 | (TCPOPT_MD5SIG << 8) | | |
690 | TCPOLEN_MD5SIG); | |
691 | arg.iov[0].iov_len += TCPOLEN_MD5SIG_ALIGNED; | |
692 | rep.th.doff = arg.iov[0].iov_len/4; | |
693 | ||
49a72dfb | 694 | tcp_v4_md5_hash_hdr((__u8 *) &rep.opt[offset], |
90b7e112 AL |
695 | key, ip_hdr(skb)->saddr, |
696 | ip_hdr(skb)->daddr, &rep.th); | |
cfb6eeb4 YH |
697 | } |
698 | #endif | |
88ef4a5a | 699 | arg.flags = reply_flags; |
eddc9ec5 ACM |
700 | arg.csum = csum_tcpudp_nofold(ip_hdr(skb)->daddr, |
701 | ip_hdr(skb)->saddr, /* XXX */ | |
1da177e4 LT |
702 | arg.iov[0].iov_len, IPPROTO_TCP, 0); |
703 | arg.csumoffset = offsetof(struct tcphdr, check) / 2; | |
9501f972 YH |
704 | if (oif) |
705 | arg.bound_dev_if = oif; | |
1da177e4 | 706 | |
a86b1e30 | 707 | ip_send_reply(net->ipv4.tcp_sock, skb, |
7feb49c8 | 708 | &arg, arg.iov[0].iov_len); |
1da177e4 | 709 | |
63231bdd | 710 | TCP_INC_STATS_BH(net, TCP_MIB_OUTSEGS); |
1da177e4 LT |
711 | } |
712 | ||
713 | static void tcp_v4_timewait_ack(struct sock *sk, struct sk_buff *skb) | |
714 | { | |
8feaf0c0 | 715 | struct inet_timewait_sock *tw = inet_twsk(sk); |
cfb6eeb4 | 716 | struct tcp_timewait_sock *tcptw = tcp_twsk(sk); |
1da177e4 | 717 | |
9501f972 | 718 | tcp_v4_send_ack(skb, tcptw->tw_snd_nxt, tcptw->tw_rcv_nxt, |
7174259e | 719 | tcptw->tw_rcv_wnd >> tw->tw_rcv_wscale, |
9501f972 YH |
720 | tcptw->tw_ts_recent, |
721 | tw->tw_bound_dev_if, | |
88ef4a5a KK |
722 | tcp_twsk_md5_key(tcptw), |
723 | tw->tw_transparent ? IP_REPLY_ARG_NOSRCCHECK : 0 | |
9501f972 | 724 | ); |
1da177e4 | 725 | |
8feaf0c0 | 726 | inet_twsk_put(tw); |
1da177e4 LT |
727 | } |
728 | ||
6edafaaf | 729 | static void tcp_v4_reqsk_send_ack(struct sock *sk, struct sk_buff *skb, |
7174259e | 730 | struct request_sock *req) |
1da177e4 | 731 | { |
9501f972 | 732 | tcp_v4_send_ack(skb, tcp_rsk(req)->snt_isn + 1, |
cfb6eeb4 | 733 | tcp_rsk(req)->rcv_isn + 1, req->rcv_wnd, |
9501f972 YH |
734 | req->ts_recent, |
735 | 0, | |
88ef4a5a KK |
736 | tcp_v4_md5_do_lookup(sk, ip_hdr(skb)->daddr), |
737 | inet_rsk(req)->no_srccheck ? IP_REPLY_ARG_NOSRCCHECK : 0); | |
1da177e4 LT |
738 | } |
739 | ||
1da177e4 | 740 | /* |
9bf1d83e | 741 | * Send a SYN-ACK after having received a SYN. |
60236fdd | 742 | * This still operates on a request_sock only, not on a big |
1da177e4 LT |
743 | * socket. |
744 | */ | |
e6b4d113 WAS |
745 | static int __tcp_v4_send_synack(struct sock *sk, struct dst_entry *dst, |
746 | struct request_sock *req, | |
747 | struct request_values *rvp) | |
1da177e4 | 748 | { |
2e6599cb | 749 | const struct inet_request_sock *ireq = inet_rsk(req); |
1da177e4 LT |
750 | int err = -1; |
751 | struct sk_buff * skb; | |
752 | ||
753 | /* First, grab a route. */ | |
463c84b9 | 754 | if (!dst && (dst = inet_csk_route_req(sk, req)) == NULL) |
fd80eb94 | 755 | return -1; |
1da177e4 | 756 | |
e6b4d113 | 757 | skb = tcp_make_synack(sk, dst, req, rvp); |
1da177e4 LT |
758 | |
759 | if (skb) { | |
aa8223c7 | 760 | struct tcphdr *th = tcp_hdr(skb); |
1da177e4 | 761 | |
ba7808ea | 762 | th->check = tcp_v4_check(skb->len, |
2e6599cb ACM |
763 | ireq->loc_addr, |
764 | ireq->rmt_addr, | |
07f0757a | 765 | csum_partial(th, skb->len, |
1da177e4 LT |
766 | skb->csum)); |
767 | ||
2e6599cb ACM |
768 | err = ip_build_and_send_pkt(skb, sk, ireq->loc_addr, |
769 | ireq->rmt_addr, | |
770 | ireq->opt); | |
b9df3cb8 | 771 | err = net_xmit_eval(err); |
1da177e4 LT |
772 | } |
773 | ||
1da177e4 LT |
774 | dst_release(dst); |
775 | return err; | |
776 | } | |
777 | ||
e6b4d113 WAS |
778 | static int tcp_v4_send_synack(struct sock *sk, struct request_sock *req, |
779 | struct request_values *rvp) | |
fd80eb94 | 780 | { |
e6b4d113 | 781 | return __tcp_v4_send_synack(sk, NULL, req, rvp); |
fd80eb94 DL |
782 | } |
783 | ||
1da177e4 | 784 | /* |
60236fdd | 785 | * IPv4 request_sock destructor. |
1da177e4 | 786 | */ |
60236fdd | 787 | static void tcp_v4_reqsk_destructor(struct request_sock *req) |
1da177e4 | 788 | { |
a51482bd | 789 | kfree(inet_rsk(req)->opt); |
1da177e4 LT |
790 | } |
791 | ||
80e40daa | 792 | #ifdef CONFIG_SYN_COOKIES |
40efc6fa | 793 | static void syn_flood_warning(struct sk_buff *skb) |
1da177e4 LT |
794 | { |
795 | static unsigned long warntime; | |
796 | ||
797 | if (time_after(jiffies, (warntime + HZ * 60))) { | |
798 | warntime = jiffies; | |
799 | printk(KERN_INFO | |
800 | "possible SYN flooding on port %d. Sending cookies.\n", | |
aa8223c7 | 801 | ntohs(tcp_hdr(skb)->dest)); |
1da177e4 LT |
802 | } |
803 | } | |
80e40daa | 804 | #endif |
1da177e4 LT |
805 | |
806 | /* | |
60236fdd | 807 | * Save and compile IPv4 options into the request_sock if needed. |
1da177e4 | 808 | */ |
40efc6fa SH |
809 | static struct ip_options *tcp_v4_save_options(struct sock *sk, |
810 | struct sk_buff *skb) | |
1da177e4 LT |
811 | { |
812 | struct ip_options *opt = &(IPCB(skb)->opt); | |
813 | struct ip_options *dopt = NULL; | |
814 | ||
815 | if (opt && opt->optlen) { | |
816 | int opt_size = optlength(opt); | |
817 | dopt = kmalloc(opt_size, GFP_ATOMIC); | |
818 | if (dopt) { | |
819 | if (ip_options_echo(dopt, skb)) { | |
820 | kfree(dopt); | |
821 | dopt = NULL; | |
822 | } | |
823 | } | |
824 | } | |
825 | return dopt; | |
826 | } | |
827 | ||
cfb6eeb4 YH |
828 | #ifdef CONFIG_TCP_MD5SIG |
829 | /* | |
830 | * RFC2385 MD5 checksumming requires a mapping of | |
831 | * IP address->MD5 Key. | |
832 | * We need to maintain these in the sk structure. | |
833 | */ | |
834 | ||
835 | /* Find the Key structure for an address. */ | |
7174259e ACM |
836 | static struct tcp_md5sig_key * |
837 | tcp_v4_md5_do_lookup(struct sock *sk, __be32 addr) | |
cfb6eeb4 YH |
838 | { |
839 | struct tcp_sock *tp = tcp_sk(sk); | |
840 | int i; | |
841 | ||
842 | if (!tp->md5sig_info || !tp->md5sig_info->entries4) | |
843 | return NULL; | |
844 | for (i = 0; i < tp->md5sig_info->entries4; i++) { | |
845 | if (tp->md5sig_info->keys4[i].addr == addr) | |
f8ab18d2 | 846 | return &tp->md5sig_info->keys4[i].base; |
cfb6eeb4 YH |
847 | } |
848 | return NULL; | |
849 | } | |
850 | ||
851 | struct tcp_md5sig_key *tcp_v4_md5_lookup(struct sock *sk, | |
852 | struct sock *addr_sk) | |
853 | { | |
c720c7e8 | 854 | return tcp_v4_md5_do_lookup(sk, inet_sk(addr_sk)->inet_daddr); |
cfb6eeb4 YH |
855 | } |
856 | ||
857 | EXPORT_SYMBOL(tcp_v4_md5_lookup); | |
858 | ||
f5b99bcd AB |
859 | static struct tcp_md5sig_key *tcp_v4_reqsk_md5_lookup(struct sock *sk, |
860 | struct request_sock *req) | |
cfb6eeb4 YH |
861 | { |
862 | return tcp_v4_md5_do_lookup(sk, inet_rsk(req)->rmt_addr); | |
863 | } | |
864 | ||
865 | /* This can be called on a newly created socket, from other files */ | |
866 | int tcp_v4_md5_do_add(struct sock *sk, __be32 addr, | |
867 | u8 *newkey, u8 newkeylen) | |
868 | { | |
869 | /* Add Key to the list */ | |
b0a713e9 | 870 | struct tcp_md5sig_key *key; |
cfb6eeb4 YH |
871 | struct tcp_sock *tp = tcp_sk(sk); |
872 | struct tcp4_md5sig_key *keys; | |
873 | ||
b0a713e9 | 874 | key = tcp_v4_md5_do_lookup(sk, addr); |
cfb6eeb4 YH |
875 | if (key) { |
876 | /* Pre-existing entry - just update that one. */ | |
b0a713e9 MD |
877 | kfree(key->key); |
878 | key->key = newkey; | |
879 | key->keylen = newkeylen; | |
cfb6eeb4 | 880 | } else { |
f6685938 ACM |
881 | struct tcp_md5sig_info *md5sig; |
882 | ||
cfb6eeb4 | 883 | if (!tp->md5sig_info) { |
f6685938 ACM |
884 | tp->md5sig_info = kzalloc(sizeof(*tp->md5sig_info), |
885 | GFP_ATOMIC); | |
cfb6eeb4 YH |
886 | if (!tp->md5sig_info) { |
887 | kfree(newkey); | |
888 | return -ENOMEM; | |
889 | } | |
3d7dbeac | 890 | sk->sk_route_caps &= ~NETIF_F_GSO_MASK; |
cfb6eeb4 | 891 | } |
aa133076 | 892 | if (tcp_alloc_md5sig_pool(sk) == NULL) { |
cfb6eeb4 YH |
893 | kfree(newkey); |
894 | return -ENOMEM; | |
895 | } | |
f6685938 ACM |
896 | md5sig = tp->md5sig_info; |
897 | ||
898 | if (md5sig->alloced4 == md5sig->entries4) { | |
899 | keys = kmalloc((sizeof(*keys) * | |
e905a9ed | 900 | (md5sig->entries4 + 1)), GFP_ATOMIC); |
cfb6eeb4 YH |
901 | if (!keys) { |
902 | kfree(newkey); | |
903 | tcp_free_md5sig_pool(); | |
904 | return -ENOMEM; | |
905 | } | |
906 | ||
f6685938 ACM |
907 | if (md5sig->entries4) |
908 | memcpy(keys, md5sig->keys4, | |
909 | sizeof(*keys) * md5sig->entries4); | |
cfb6eeb4 YH |
910 | |
911 | /* Free old key list, and reference new one */ | |
a80cc20d | 912 | kfree(md5sig->keys4); |
f6685938 ACM |
913 | md5sig->keys4 = keys; |
914 | md5sig->alloced4++; | |
cfb6eeb4 | 915 | } |
f6685938 | 916 | md5sig->entries4++; |
f8ab18d2 DM |
917 | md5sig->keys4[md5sig->entries4 - 1].addr = addr; |
918 | md5sig->keys4[md5sig->entries4 - 1].base.key = newkey; | |
919 | md5sig->keys4[md5sig->entries4 - 1].base.keylen = newkeylen; | |
cfb6eeb4 YH |
920 | } |
921 | return 0; | |
922 | } | |
923 | ||
924 | EXPORT_SYMBOL(tcp_v4_md5_do_add); | |
925 | ||
926 | static int tcp_v4_md5_add_func(struct sock *sk, struct sock *addr_sk, | |
927 | u8 *newkey, u8 newkeylen) | |
928 | { | |
c720c7e8 | 929 | return tcp_v4_md5_do_add(sk, inet_sk(addr_sk)->inet_daddr, |
cfb6eeb4 YH |
930 | newkey, newkeylen); |
931 | } | |
932 | ||
933 | int tcp_v4_md5_do_del(struct sock *sk, __be32 addr) | |
934 | { | |
935 | struct tcp_sock *tp = tcp_sk(sk); | |
936 | int i; | |
937 | ||
938 | for (i = 0; i < tp->md5sig_info->entries4; i++) { | |
939 | if (tp->md5sig_info->keys4[i].addr == addr) { | |
940 | /* Free the key */ | |
f8ab18d2 | 941 | kfree(tp->md5sig_info->keys4[i].base.key); |
cfb6eeb4 YH |
942 | tp->md5sig_info->entries4--; |
943 | ||
944 | if (tp->md5sig_info->entries4 == 0) { | |
945 | kfree(tp->md5sig_info->keys4); | |
946 | tp->md5sig_info->keys4 = NULL; | |
8228a18d | 947 | tp->md5sig_info->alloced4 = 0; |
7174259e | 948 | } else if (tp->md5sig_info->entries4 != i) { |
cfb6eeb4 | 949 | /* Need to do some manipulation */ |
354faf09 YH |
950 | memmove(&tp->md5sig_info->keys4[i], |
951 | &tp->md5sig_info->keys4[i+1], | |
952 | (tp->md5sig_info->entries4 - i) * | |
953 | sizeof(struct tcp4_md5sig_key)); | |
cfb6eeb4 YH |
954 | } |
955 | tcp_free_md5sig_pool(); | |
956 | return 0; | |
957 | } | |
958 | } | |
959 | return -ENOENT; | |
960 | } | |
961 | ||
962 | EXPORT_SYMBOL(tcp_v4_md5_do_del); | |
963 | ||
7174259e | 964 | static void tcp_v4_clear_md5_list(struct sock *sk) |
cfb6eeb4 YH |
965 | { |
966 | struct tcp_sock *tp = tcp_sk(sk); | |
967 | ||
968 | /* Free each key, then the set of key keys, | |
969 | * the crypto element, and then decrement our | |
970 | * hold on the last resort crypto. | |
971 | */ | |
972 | if (tp->md5sig_info->entries4) { | |
973 | int i; | |
974 | for (i = 0; i < tp->md5sig_info->entries4; i++) | |
f8ab18d2 | 975 | kfree(tp->md5sig_info->keys4[i].base.key); |
cfb6eeb4 YH |
976 | tp->md5sig_info->entries4 = 0; |
977 | tcp_free_md5sig_pool(); | |
978 | } | |
979 | if (tp->md5sig_info->keys4) { | |
980 | kfree(tp->md5sig_info->keys4); | |
981 | tp->md5sig_info->keys4 = NULL; | |
982 | tp->md5sig_info->alloced4 = 0; | |
983 | } | |
984 | } | |
985 | ||
7174259e ACM |
986 | static int tcp_v4_parse_md5_keys(struct sock *sk, char __user *optval, |
987 | int optlen) | |
cfb6eeb4 YH |
988 | { |
989 | struct tcp_md5sig cmd; | |
990 | struct sockaddr_in *sin = (struct sockaddr_in *)&cmd.tcpm_addr; | |
991 | u8 *newkey; | |
992 | ||
993 | if (optlen < sizeof(cmd)) | |
994 | return -EINVAL; | |
995 | ||
7174259e | 996 | if (copy_from_user(&cmd, optval, sizeof(cmd))) |
cfb6eeb4 YH |
997 | return -EFAULT; |
998 | ||
999 | if (sin->sin_family != AF_INET) | |
1000 | return -EINVAL; | |
1001 | ||
1002 | if (!cmd.tcpm_key || !cmd.tcpm_keylen) { | |
1003 | if (!tcp_sk(sk)->md5sig_info) | |
1004 | return -ENOENT; | |
1005 | return tcp_v4_md5_do_del(sk, sin->sin_addr.s_addr); | |
1006 | } | |
1007 | ||
1008 | if (cmd.tcpm_keylen > TCP_MD5SIG_MAXKEYLEN) | |
1009 | return -EINVAL; | |
1010 | ||
1011 | if (!tcp_sk(sk)->md5sig_info) { | |
1012 | struct tcp_sock *tp = tcp_sk(sk); | |
aa133076 | 1013 | struct tcp_md5sig_info *p; |
cfb6eeb4 | 1014 | |
aa133076 | 1015 | p = kzalloc(sizeof(*p), sk->sk_allocation); |
cfb6eeb4 YH |
1016 | if (!p) |
1017 | return -EINVAL; | |
1018 | ||
1019 | tp->md5sig_info = p; | |
3d7dbeac | 1020 | sk->sk_route_caps &= ~NETIF_F_GSO_MASK; |
cfb6eeb4 YH |
1021 | } |
1022 | ||
aa133076 | 1023 | newkey = kmemdup(cmd.tcpm_key, cmd.tcpm_keylen, sk->sk_allocation); |
cfb6eeb4 YH |
1024 | if (!newkey) |
1025 | return -ENOMEM; | |
cfb6eeb4 YH |
1026 | return tcp_v4_md5_do_add(sk, sin->sin_addr.s_addr, |
1027 | newkey, cmd.tcpm_keylen); | |
1028 | } | |
1029 | ||
49a72dfb AL |
1030 | static int tcp_v4_md5_hash_pseudoheader(struct tcp_md5sig_pool *hp, |
1031 | __be32 daddr, __be32 saddr, int nbytes) | |
cfb6eeb4 | 1032 | { |
cfb6eeb4 | 1033 | struct tcp4_pseudohdr *bp; |
49a72dfb | 1034 | struct scatterlist sg; |
cfb6eeb4 YH |
1035 | |
1036 | bp = &hp->md5_blk.ip4; | |
cfb6eeb4 YH |
1037 | |
1038 | /* | |
49a72dfb | 1039 | * 1. the TCP pseudo-header (in the order: source IP address, |
cfb6eeb4 YH |
1040 | * destination IP address, zero-padded protocol number, and |
1041 | * segment length) | |
1042 | */ | |
1043 | bp->saddr = saddr; | |
1044 | bp->daddr = daddr; | |
1045 | bp->pad = 0; | |
076fb722 | 1046 | bp->protocol = IPPROTO_TCP; |
49a72dfb | 1047 | bp->len = cpu_to_be16(nbytes); |
c7da57a1 | 1048 | |
49a72dfb AL |
1049 | sg_init_one(&sg, bp, sizeof(*bp)); |
1050 | return crypto_hash_update(&hp->md5_desc, &sg, sizeof(*bp)); | |
1051 | } | |
1052 | ||
1053 | static int tcp_v4_md5_hash_hdr(char *md5_hash, struct tcp_md5sig_key *key, | |
1054 | __be32 daddr, __be32 saddr, struct tcphdr *th) | |
1055 | { | |
1056 | struct tcp_md5sig_pool *hp; | |
1057 | struct hash_desc *desc; | |
1058 | ||
1059 | hp = tcp_get_md5sig_pool(); | |
1060 | if (!hp) | |
1061 | goto clear_hash_noput; | |
1062 | desc = &hp->md5_desc; | |
1063 | ||
1064 | if (crypto_hash_init(desc)) | |
1065 | goto clear_hash; | |
1066 | if (tcp_v4_md5_hash_pseudoheader(hp, daddr, saddr, th->doff << 2)) | |
1067 | goto clear_hash; | |
1068 | if (tcp_md5_hash_header(hp, th)) | |
1069 | goto clear_hash; | |
1070 | if (tcp_md5_hash_key(hp, key)) | |
1071 | goto clear_hash; | |
1072 | if (crypto_hash_final(desc, md5_hash)) | |
cfb6eeb4 YH |
1073 | goto clear_hash; |
1074 | ||
cfb6eeb4 | 1075 | tcp_put_md5sig_pool(); |
cfb6eeb4 | 1076 | return 0; |
49a72dfb | 1077 | |
cfb6eeb4 YH |
1078 | clear_hash: |
1079 | tcp_put_md5sig_pool(); | |
1080 | clear_hash_noput: | |
1081 | memset(md5_hash, 0, 16); | |
49a72dfb | 1082 | return 1; |
cfb6eeb4 YH |
1083 | } |
1084 | ||
49a72dfb AL |
1085 | int tcp_v4_md5_hash_skb(char *md5_hash, struct tcp_md5sig_key *key, |
1086 | struct sock *sk, struct request_sock *req, | |
1087 | struct sk_buff *skb) | |
cfb6eeb4 | 1088 | { |
49a72dfb AL |
1089 | struct tcp_md5sig_pool *hp; |
1090 | struct hash_desc *desc; | |
1091 | struct tcphdr *th = tcp_hdr(skb); | |
cfb6eeb4 YH |
1092 | __be32 saddr, daddr; |
1093 | ||
1094 | if (sk) { | |
c720c7e8 ED |
1095 | saddr = inet_sk(sk)->inet_saddr; |
1096 | daddr = inet_sk(sk)->inet_daddr; | |
49a72dfb AL |
1097 | } else if (req) { |
1098 | saddr = inet_rsk(req)->loc_addr; | |
1099 | daddr = inet_rsk(req)->rmt_addr; | |
cfb6eeb4 | 1100 | } else { |
49a72dfb AL |
1101 | const struct iphdr *iph = ip_hdr(skb); |
1102 | saddr = iph->saddr; | |
1103 | daddr = iph->daddr; | |
cfb6eeb4 | 1104 | } |
49a72dfb AL |
1105 | |
1106 | hp = tcp_get_md5sig_pool(); | |
1107 | if (!hp) | |
1108 | goto clear_hash_noput; | |
1109 | desc = &hp->md5_desc; | |
1110 | ||
1111 | if (crypto_hash_init(desc)) | |
1112 | goto clear_hash; | |
1113 | ||
1114 | if (tcp_v4_md5_hash_pseudoheader(hp, daddr, saddr, skb->len)) | |
1115 | goto clear_hash; | |
1116 | if (tcp_md5_hash_header(hp, th)) | |
1117 | goto clear_hash; | |
1118 | if (tcp_md5_hash_skb_data(hp, skb, th->doff << 2)) | |
1119 | goto clear_hash; | |
1120 | if (tcp_md5_hash_key(hp, key)) | |
1121 | goto clear_hash; | |
1122 | if (crypto_hash_final(desc, md5_hash)) | |
1123 | goto clear_hash; | |
1124 | ||
1125 | tcp_put_md5sig_pool(); | |
1126 | return 0; | |
1127 | ||
1128 | clear_hash: | |
1129 | tcp_put_md5sig_pool(); | |
1130 | clear_hash_noput: | |
1131 | memset(md5_hash, 0, 16); | |
1132 | return 1; | |
cfb6eeb4 YH |
1133 | } |
1134 | ||
49a72dfb | 1135 | EXPORT_SYMBOL(tcp_v4_md5_hash_skb); |
cfb6eeb4 | 1136 | |
7174259e | 1137 | static int tcp_v4_inbound_md5_hash(struct sock *sk, struct sk_buff *skb) |
cfb6eeb4 YH |
1138 | { |
1139 | /* | |
1140 | * This gets called for each TCP segment that arrives | |
1141 | * so we want to be efficient. | |
1142 | * We have 3 drop cases: | |
1143 | * o No MD5 hash and one expected. | |
1144 | * o MD5 hash and we're not expecting one. | |
1145 | * o MD5 hash and its wrong. | |
1146 | */ | |
1147 | __u8 *hash_location = NULL; | |
1148 | struct tcp_md5sig_key *hash_expected; | |
eddc9ec5 | 1149 | const struct iphdr *iph = ip_hdr(skb); |
aa8223c7 | 1150 | struct tcphdr *th = tcp_hdr(skb); |
cfb6eeb4 | 1151 | int genhash; |
cfb6eeb4 YH |
1152 | unsigned char newhash[16]; |
1153 | ||
1154 | hash_expected = tcp_v4_md5_do_lookup(sk, iph->saddr); | |
7d5d5525 | 1155 | hash_location = tcp_parse_md5sig_option(th); |
cfb6eeb4 | 1156 | |
cfb6eeb4 YH |
1157 | /* We've parsed the options - do we have a hash? */ |
1158 | if (!hash_expected && !hash_location) | |
1159 | return 0; | |
1160 | ||
1161 | if (hash_expected && !hash_location) { | |
785957d3 | 1162 | NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPMD5NOTFOUND); |
cfb6eeb4 YH |
1163 | return 1; |
1164 | } | |
1165 | ||
1166 | if (!hash_expected && hash_location) { | |
785957d3 | 1167 | NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPMD5UNEXPECTED); |
cfb6eeb4 YH |
1168 | return 1; |
1169 | } | |
1170 | ||
1171 | /* Okay, so this is hash_expected and hash_location - | |
1172 | * so we need to calculate the checksum. | |
1173 | */ | |
49a72dfb AL |
1174 | genhash = tcp_v4_md5_hash_skb(newhash, |
1175 | hash_expected, | |
1176 | NULL, NULL, skb); | |
cfb6eeb4 YH |
1177 | |
1178 | if (genhash || memcmp(hash_location, newhash, 16) != 0) { | |
1179 | if (net_ratelimit()) { | |
673d57e7 HH |
1180 | printk(KERN_INFO "MD5 Hash failed for (%pI4, %d)->(%pI4, %d)%s\n", |
1181 | &iph->saddr, ntohs(th->source), | |
1182 | &iph->daddr, ntohs(th->dest), | |
cfb6eeb4 | 1183 | genhash ? " tcp_v4_calc_md5_hash failed" : ""); |
cfb6eeb4 YH |
1184 | } |
1185 | return 1; | |
1186 | } | |
1187 | return 0; | |
1188 | } | |
1189 | ||
1190 | #endif | |
1191 | ||
72a3effa | 1192 | struct request_sock_ops tcp_request_sock_ops __read_mostly = { |
1da177e4 | 1193 | .family = PF_INET, |
2e6599cb | 1194 | .obj_size = sizeof(struct tcp_request_sock), |
1da177e4 | 1195 | .rtx_syn_ack = tcp_v4_send_synack, |
60236fdd ACM |
1196 | .send_ack = tcp_v4_reqsk_send_ack, |
1197 | .destructor = tcp_v4_reqsk_destructor, | |
1da177e4 LT |
1198 | .send_reset = tcp_v4_send_reset, |
1199 | }; | |
1200 | ||
cfb6eeb4 | 1201 | #ifdef CONFIG_TCP_MD5SIG |
b2e4b3de | 1202 | static const struct tcp_request_sock_ops tcp_request_sock_ipv4_ops = { |
cfb6eeb4 | 1203 | .md5_lookup = tcp_v4_reqsk_md5_lookup, |
e3afe7b7 | 1204 | .calc_md5_hash = tcp_v4_md5_hash_skb, |
cfb6eeb4 | 1205 | }; |
b6332e6c | 1206 | #endif |
cfb6eeb4 | 1207 | |
6d6ee43e ACM |
1208 | static struct timewait_sock_ops tcp_timewait_sock_ops = { |
1209 | .twsk_obj_size = sizeof(struct tcp_timewait_sock), | |
1210 | .twsk_unique = tcp_twsk_unique, | |
cfb6eeb4 | 1211 | .twsk_destructor= tcp_twsk_destructor, |
6d6ee43e ACM |
1212 | }; |
1213 | ||
1da177e4 LT |
1214 | int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb) |
1215 | { | |
4957faad | 1216 | struct tcp_extend_values tmp_ext; |
1da177e4 | 1217 | struct tcp_options_received tmp_opt; |
4957faad | 1218 | u8 *hash_location; |
60236fdd | 1219 | struct request_sock *req; |
e6b4d113 | 1220 | struct inet_request_sock *ireq; |
4957faad | 1221 | struct tcp_sock *tp = tcp_sk(sk); |
e6b4d113 | 1222 | struct dst_entry *dst = NULL; |
eddc9ec5 ACM |
1223 | __be32 saddr = ip_hdr(skb)->saddr; |
1224 | __be32 daddr = ip_hdr(skb)->daddr; | |
1da177e4 | 1225 | __u32 isn = TCP_SKB_CB(skb)->when; |
1da177e4 LT |
1226 | #ifdef CONFIG_SYN_COOKIES |
1227 | int want_cookie = 0; | |
1228 | #else | |
1229 | #define want_cookie 0 /* Argh, why doesn't gcc optimize this :( */ | |
1230 | #endif | |
1231 | ||
1232 | /* Never answer to SYNs send to broadcast or multicast */ | |
511c3f92 | 1233 | if (skb_rtable(skb)->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST)) |
1da177e4 LT |
1234 | goto drop; |
1235 | ||
1236 | /* TW buckets are converted to open requests without | |
1237 | * limitations, they conserve resources and peer is | |
1238 | * evidently real one. | |
1239 | */ | |
463c84b9 | 1240 | if (inet_csk_reqsk_queue_is_full(sk) && !isn) { |
1da177e4 LT |
1241 | #ifdef CONFIG_SYN_COOKIES |
1242 | if (sysctl_tcp_syncookies) { | |
1243 | want_cookie = 1; | |
1244 | } else | |
1245 | #endif | |
1246 | goto drop; | |
1247 | } | |
1248 | ||
1249 | /* Accept backlog is full. If we have already queued enough | |
1250 | * of warm entries in syn queue, drop request. It is better than | |
1251 | * clogging syn queue with openreqs with exponentially increasing | |
1252 | * timeout. | |
1253 | */ | |
463c84b9 | 1254 | if (sk_acceptq_is_full(sk) && inet_csk_reqsk_queue_young(sk) > 1) |
1da177e4 LT |
1255 | goto drop; |
1256 | ||
ce4a7d0d | 1257 | req = inet_reqsk_alloc(&tcp_request_sock_ops); |
1da177e4 LT |
1258 | if (!req) |
1259 | goto drop; | |
1260 | ||
cfb6eeb4 YH |
1261 | #ifdef CONFIG_TCP_MD5SIG |
1262 | tcp_rsk(req)->af_specific = &tcp_request_sock_ipv4_ops; | |
1263 | #endif | |
1264 | ||
1da177e4 | 1265 | tcp_clear_options(&tmp_opt); |
bee7ca9e | 1266 | tmp_opt.mss_clamp = TCP_MSS_DEFAULT; |
4957faad | 1267 | tmp_opt.user_mss = tp->rx_opt.user_mss; |
bb5b7c11 | 1268 | tcp_parse_options(skb, &tmp_opt, &hash_location, 0); |
4957faad WAS |
1269 | |
1270 | if (tmp_opt.cookie_plus > 0 && | |
1271 | tmp_opt.saw_tstamp && | |
1272 | !tp->rx_opt.cookie_out_never && | |
1273 | (sysctl_tcp_cookie_size > 0 || | |
1274 | (tp->cookie_values != NULL && | |
1275 | tp->cookie_values->cookie_desired > 0))) { | |
1276 | u8 *c; | |
1277 | u32 *mess = &tmp_ext.cookie_bakery[COOKIE_DIGEST_WORDS]; | |
1278 | int l = tmp_opt.cookie_plus - TCPOLEN_COOKIE_BASE; | |
1279 | ||
1280 | if (tcp_cookie_generator(&tmp_ext.cookie_bakery[0]) != 0) | |
1281 | goto drop_and_release; | |
1282 | ||
1283 | /* Secret recipe starts with IP addresses */ | |
1284 | *mess++ ^= daddr; | |
1285 | *mess++ ^= saddr; | |
1da177e4 | 1286 | |
4957faad WAS |
1287 | /* plus variable length Initiator Cookie */ |
1288 | c = (u8 *)mess; | |
1289 | while (l-- > 0) | |
1290 | *c++ ^= *hash_location++; | |
1291 | ||
1292 | #ifdef CONFIG_SYN_COOKIES | |
1293 | want_cookie = 0; /* not our kind of cookie */ | |
1294 | #endif | |
1295 | tmp_ext.cookie_out_never = 0; /* false */ | |
1296 | tmp_ext.cookie_plus = tmp_opt.cookie_plus; | |
1297 | } else if (!tp->rx_opt.cookie_in_always) { | |
1298 | /* redundant indications, but ensure initialization. */ | |
1299 | tmp_ext.cookie_out_never = 1; /* true */ | |
1300 | tmp_ext.cookie_plus = 0; | |
1301 | } else { | |
1302 | goto drop_and_release; | |
1303 | } | |
1304 | tmp_ext.cookie_in_always = tp->rx_opt.cookie_in_always; | |
1da177e4 | 1305 | |
4dfc2817 | 1306 | if (want_cookie && !tmp_opt.saw_tstamp) |
1da177e4 | 1307 | tcp_clear_options(&tmp_opt); |
1da177e4 | 1308 | |
1da177e4 | 1309 | tmp_opt.tstamp_ok = tmp_opt.saw_tstamp; |
1da177e4 LT |
1310 | tcp_openreq_init(req, &tmp_opt, skb); |
1311 | ||
bb5b7c11 DM |
1312 | ireq = inet_rsk(req); |
1313 | ireq->loc_addr = daddr; | |
1314 | ireq->rmt_addr = saddr; | |
1315 | ireq->no_srccheck = inet_sk(sk)->transparent; | |
1316 | ireq->opt = tcp_v4_save_options(sk, skb); | |
1317 | ||
284904aa | 1318 | if (security_inet_conn_request(sk, skb, req)) |
bb5b7c11 | 1319 | goto drop_and_free; |
284904aa | 1320 | |
1da177e4 | 1321 | if (!want_cookie) |
aa8223c7 | 1322 | TCP_ECN_create_request(req, tcp_hdr(skb)); |
1da177e4 LT |
1323 | |
1324 | if (want_cookie) { | |
1325 | #ifdef CONFIG_SYN_COOKIES | |
1326 | syn_flood_warning(skb); | |
4dfc2817 | 1327 | req->cookie_ts = tmp_opt.tstamp_ok; |
1da177e4 LT |
1328 | #endif |
1329 | isn = cookie_v4_init_sequence(sk, skb, &req->mss); | |
1330 | } else if (!isn) { | |
1331 | struct inet_peer *peer = NULL; | |
1332 | ||
1333 | /* VJ's idea. We save last timestamp seen | |
1334 | * from the destination in peer table, when entering | |
1335 | * state TIME-WAIT, and check against it before | |
1336 | * accepting new connection request. | |
1337 | * | |
1338 | * If "isn" is not zero, this request hit alive | |
1339 | * timewait bucket, so that all the necessary checks | |
1340 | * are made in the function processing timewait state. | |
1341 | */ | |
1342 | if (tmp_opt.saw_tstamp && | |
295ff7ed | 1343 | tcp_death_row.sysctl_tw_recycle && |
bb5b7c11 | 1344 | (dst = inet_csk_route_req(sk, req)) != NULL && |
1da177e4 LT |
1345 | (peer = rt_get_peer((struct rtable *)dst)) != NULL && |
1346 | peer->v4daddr == saddr) { | |
2c1409a0 | 1347 | if ((u32)get_seconds() - peer->tcp_ts_stamp < TCP_PAWS_MSL && |
1da177e4 LT |
1348 | (s32)(peer->tcp_ts - req->ts_recent) > |
1349 | TCP_PAWS_WINDOW) { | |
de0744af | 1350 | NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_PAWSPASSIVEREJECTED); |
7cd04fa7 | 1351 | goto drop_and_release; |
1da177e4 LT |
1352 | } |
1353 | } | |
1354 | /* Kill the following clause, if you dislike this way. */ | |
1355 | else if (!sysctl_tcp_syncookies && | |
463c84b9 | 1356 | (sysctl_max_syn_backlog - inet_csk_reqsk_queue_len(sk) < |
1da177e4 LT |
1357 | (sysctl_max_syn_backlog >> 2)) && |
1358 | (!peer || !peer->tcp_ts_stamp) && | |
1359 | (!dst || !dst_metric(dst, RTAX_RTT))) { | |
1360 | /* Without syncookies last quarter of | |
1361 | * backlog is filled with destinations, | |
1362 | * proven to be alive. | |
1363 | * It means that we continue to communicate | |
1364 | * to destinations, already remembered | |
1365 | * to the moment of synflood. | |
1366 | */ | |
673d57e7 HH |
1367 | LIMIT_NETDEBUG(KERN_DEBUG "TCP: drop open request from %pI4/%u\n", |
1368 | &saddr, ntohs(tcp_hdr(skb)->source)); | |
7cd04fa7 | 1369 | goto drop_and_release; |
1da177e4 LT |
1370 | } |
1371 | ||
a94f723d | 1372 | isn = tcp_v4_init_sequence(skb); |
1da177e4 | 1373 | } |
2e6599cb | 1374 | tcp_rsk(req)->snt_isn = isn; |
1da177e4 | 1375 | |
4957faad WAS |
1376 | if (__tcp_v4_send_synack(sk, dst, req, |
1377 | (struct request_values *)&tmp_ext) || | |
1378 | want_cookie) | |
1da177e4 LT |
1379 | goto drop_and_free; |
1380 | ||
7cd04fa7 | 1381 | inet_csk_reqsk_queue_hash_add(sk, req, TCP_TIMEOUT_INIT); |
1da177e4 LT |
1382 | return 0; |
1383 | ||
7cd04fa7 DL |
1384 | drop_and_release: |
1385 | dst_release(dst); | |
1da177e4 | 1386 | drop_and_free: |
60236fdd | 1387 | reqsk_free(req); |
1da177e4 | 1388 | drop: |
1da177e4 LT |
1389 | return 0; |
1390 | } | |
1391 | ||
1392 | ||
1393 | /* | |
1394 | * The three way handshake has completed - we got a valid synack - | |
1395 | * now create the new socket. | |
1396 | */ | |
1397 | struct sock *tcp_v4_syn_recv_sock(struct sock *sk, struct sk_buff *skb, | |
60236fdd | 1398 | struct request_sock *req, |
1da177e4 LT |
1399 | struct dst_entry *dst) |
1400 | { | |
2e6599cb | 1401 | struct inet_request_sock *ireq; |
1da177e4 LT |
1402 | struct inet_sock *newinet; |
1403 | struct tcp_sock *newtp; | |
1404 | struct sock *newsk; | |
cfb6eeb4 YH |
1405 | #ifdef CONFIG_TCP_MD5SIG |
1406 | struct tcp_md5sig_key *key; | |
1407 | #endif | |
1da177e4 LT |
1408 | |
1409 | if (sk_acceptq_is_full(sk)) | |
1410 | goto exit_overflow; | |
1411 | ||
463c84b9 | 1412 | if (!dst && (dst = inet_csk_route_req(sk, req)) == NULL) |
1da177e4 LT |
1413 | goto exit; |
1414 | ||
1415 | newsk = tcp_create_openreq_child(sk, req, skb); | |
1416 | if (!newsk) | |
1417 | goto exit; | |
1418 | ||
bcd76111 | 1419 | newsk->sk_gso_type = SKB_GSO_TCPV4; |
6cbb0df7 | 1420 | sk_setup_caps(newsk, dst); |
1da177e4 LT |
1421 | |
1422 | newtp = tcp_sk(newsk); | |
1423 | newinet = inet_sk(newsk); | |
2e6599cb | 1424 | ireq = inet_rsk(req); |
c720c7e8 ED |
1425 | newinet->inet_daddr = ireq->rmt_addr; |
1426 | newinet->inet_rcv_saddr = ireq->loc_addr; | |
1427 | newinet->inet_saddr = ireq->loc_addr; | |
2e6599cb ACM |
1428 | newinet->opt = ireq->opt; |
1429 | ireq->opt = NULL; | |
463c84b9 | 1430 | newinet->mc_index = inet_iif(skb); |
eddc9ec5 | 1431 | newinet->mc_ttl = ip_hdr(skb)->ttl; |
d83d8461 | 1432 | inet_csk(newsk)->icsk_ext_hdr_len = 0; |
1da177e4 | 1433 | if (newinet->opt) |
d83d8461 | 1434 | inet_csk(newsk)->icsk_ext_hdr_len = newinet->opt->optlen; |
c720c7e8 | 1435 | newinet->inet_id = newtp->write_seq ^ jiffies; |
1da177e4 | 1436 | |
5d424d5a | 1437 | tcp_mtup_init(newsk); |
1da177e4 LT |
1438 | tcp_sync_mss(newsk, dst_mtu(dst)); |
1439 | newtp->advmss = dst_metric(dst, RTAX_ADVMSS); | |
f5fff5dc TQ |
1440 | if (tcp_sk(sk)->rx_opt.user_mss && |
1441 | tcp_sk(sk)->rx_opt.user_mss < newtp->advmss) | |
1442 | newtp->advmss = tcp_sk(sk)->rx_opt.user_mss; | |
1443 | ||
1da177e4 LT |
1444 | tcp_initialize_rcv_mss(newsk); |
1445 | ||
cfb6eeb4 YH |
1446 | #ifdef CONFIG_TCP_MD5SIG |
1447 | /* Copy over the MD5 key from the original socket */ | |
c720c7e8 ED |
1448 | key = tcp_v4_md5_do_lookup(sk, newinet->inet_daddr); |
1449 | if (key != NULL) { | |
cfb6eeb4 YH |
1450 | /* |
1451 | * We're using one, so create a matching key | |
1452 | * on the newsk structure. If we fail to get | |
1453 | * memory, then we end up not copying the key | |
1454 | * across. Shucks. | |
1455 | */ | |
f6685938 ACM |
1456 | char *newkey = kmemdup(key->key, key->keylen, GFP_ATOMIC); |
1457 | if (newkey != NULL) | |
c720c7e8 | 1458 | tcp_v4_md5_do_add(newsk, newinet->inet_daddr, |
cfb6eeb4 | 1459 | newkey, key->keylen); |
49a72dfb | 1460 | newsk->sk_route_caps &= ~NETIF_F_GSO_MASK; |
cfb6eeb4 YH |
1461 | } |
1462 | #endif | |
1463 | ||
9327f705 | 1464 | __inet_hash_nolisten(newsk, NULL); |
ab1e0a13 | 1465 | __inet_inherit_port(sk, newsk); |
1da177e4 LT |
1466 | |
1467 | return newsk; | |
1468 | ||
1469 | exit_overflow: | |
de0744af | 1470 | NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENOVERFLOWS); |
1da177e4 | 1471 | exit: |
de0744af | 1472 | NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENDROPS); |
1da177e4 LT |
1473 | dst_release(dst); |
1474 | return NULL; | |
1475 | } | |
1476 | ||
1477 | static struct sock *tcp_v4_hnd_req(struct sock *sk, struct sk_buff *skb) | |
1478 | { | |
aa8223c7 | 1479 | struct tcphdr *th = tcp_hdr(skb); |
eddc9ec5 | 1480 | const struct iphdr *iph = ip_hdr(skb); |
1da177e4 | 1481 | struct sock *nsk; |
60236fdd | 1482 | struct request_sock **prev; |
1da177e4 | 1483 | /* Find possible connection requests. */ |
463c84b9 ACM |
1484 | struct request_sock *req = inet_csk_search_req(sk, &prev, th->source, |
1485 | iph->saddr, iph->daddr); | |
1da177e4 LT |
1486 | if (req) |
1487 | return tcp_check_req(sk, skb, req, prev); | |
1488 | ||
3b1e0a65 | 1489 | nsk = inet_lookup_established(sock_net(sk), &tcp_hashinfo, iph->saddr, |
c67499c0 | 1490 | th->source, iph->daddr, th->dest, inet_iif(skb)); |
1da177e4 LT |
1491 | |
1492 | if (nsk) { | |
1493 | if (nsk->sk_state != TCP_TIME_WAIT) { | |
1494 | bh_lock_sock(nsk); | |
1495 | return nsk; | |
1496 | } | |
9469c7b4 | 1497 | inet_twsk_put(inet_twsk(nsk)); |
1da177e4 LT |
1498 | return NULL; |
1499 | } | |
1500 | ||
1501 | #ifdef CONFIG_SYN_COOKIES | |
1502 | if (!th->rst && !th->syn && th->ack) | |
1503 | sk = cookie_v4_check(sk, skb, &(IPCB(skb)->opt)); | |
1504 | #endif | |
1505 | return sk; | |
1506 | } | |
1507 | ||
b51655b9 | 1508 | static __sum16 tcp_v4_checksum_init(struct sk_buff *skb) |
1da177e4 | 1509 | { |
eddc9ec5 ACM |
1510 | const struct iphdr *iph = ip_hdr(skb); |
1511 | ||
84fa7933 | 1512 | if (skb->ip_summed == CHECKSUM_COMPLETE) { |
eddc9ec5 ACM |
1513 | if (!tcp_v4_check(skb->len, iph->saddr, |
1514 | iph->daddr, skb->csum)) { | |
fb286bb2 | 1515 | skb->ip_summed = CHECKSUM_UNNECESSARY; |
1da177e4 | 1516 | return 0; |
fb286bb2 | 1517 | } |
1da177e4 | 1518 | } |
fb286bb2 | 1519 | |
eddc9ec5 | 1520 | skb->csum = csum_tcpudp_nofold(iph->saddr, iph->daddr, |
fb286bb2 HX |
1521 | skb->len, IPPROTO_TCP, 0); |
1522 | ||
1da177e4 | 1523 | if (skb->len <= 76) { |
fb286bb2 | 1524 | return __skb_checksum_complete(skb); |
1da177e4 LT |
1525 | } |
1526 | return 0; | |
1527 | } | |
1528 | ||
1529 | ||
1530 | /* The socket must have it's spinlock held when we get | |
1531 | * here. | |
1532 | * | |
1533 | * We have a potential double-lock case here, so even when | |
1534 | * doing backlog processing we use the BH locking scheme. | |
1535 | * This is because we cannot sleep with the original spinlock | |
1536 | * held. | |
1537 | */ | |
1538 | int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb) | |
1539 | { | |
cfb6eeb4 YH |
1540 | struct sock *rsk; |
1541 | #ifdef CONFIG_TCP_MD5SIG | |
1542 | /* | |
1543 | * We really want to reject the packet as early as possible | |
1544 | * if: | |
1545 | * o We're expecting an MD5'd packet and this is no MD5 tcp option | |
1546 | * o There is an MD5 option and we're not expecting one | |
1547 | */ | |
7174259e | 1548 | if (tcp_v4_inbound_md5_hash(sk, skb)) |
cfb6eeb4 YH |
1549 | goto discard; |
1550 | #endif | |
1551 | ||
1da177e4 LT |
1552 | if (sk->sk_state == TCP_ESTABLISHED) { /* Fast path */ |
1553 | TCP_CHECK_TIMER(sk); | |
aa8223c7 | 1554 | if (tcp_rcv_established(sk, skb, tcp_hdr(skb), skb->len)) { |
cfb6eeb4 | 1555 | rsk = sk; |
1da177e4 | 1556 | goto reset; |
cfb6eeb4 | 1557 | } |
1da177e4 LT |
1558 | TCP_CHECK_TIMER(sk); |
1559 | return 0; | |
1560 | } | |
1561 | ||
ab6a5bb6 | 1562 | if (skb->len < tcp_hdrlen(skb) || tcp_checksum_complete(skb)) |
1da177e4 LT |
1563 | goto csum_err; |
1564 | ||
1565 | if (sk->sk_state == TCP_LISTEN) { | |
1566 | struct sock *nsk = tcp_v4_hnd_req(sk, skb); | |
1567 | if (!nsk) | |
1568 | goto discard; | |
1569 | ||
1570 | if (nsk != sk) { | |
cfb6eeb4 YH |
1571 | if (tcp_child_process(sk, nsk, skb)) { |
1572 | rsk = nsk; | |
1da177e4 | 1573 | goto reset; |
cfb6eeb4 | 1574 | } |
1da177e4 LT |
1575 | return 0; |
1576 | } | |
1577 | } | |
1578 | ||
1579 | TCP_CHECK_TIMER(sk); | |
aa8223c7 | 1580 | if (tcp_rcv_state_process(sk, skb, tcp_hdr(skb), skb->len)) { |
cfb6eeb4 | 1581 | rsk = sk; |
1da177e4 | 1582 | goto reset; |
cfb6eeb4 | 1583 | } |
1da177e4 LT |
1584 | TCP_CHECK_TIMER(sk); |
1585 | return 0; | |
1586 | ||
1587 | reset: | |
cfb6eeb4 | 1588 | tcp_v4_send_reset(rsk, skb); |
1da177e4 LT |
1589 | discard: |
1590 | kfree_skb(skb); | |
1591 | /* Be careful here. If this function gets more complicated and | |
1592 | * gcc suffers from register pressure on the x86, sk (in %ebx) | |
1593 | * might be destroyed here. This current version compiles correctly, | |
1594 | * but you have been warned. | |
1595 | */ | |
1596 | return 0; | |
1597 | ||
1598 | csum_err: | |
63231bdd | 1599 | TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_INERRS); |
1da177e4 LT |
1600 | goto discard; |
1601 | } | |
1602 | ||
1603 | /* | |
1604 | * From tcp_input.c | |
1605 | */ | |
1606 | ||
1607 | int tcp_v4_rcv(struct sk_buff *skb) | |
1608 | { | |
eddc9ec5 | 1609 | const struct iphdr *iph; |
1da177e4 LT |
1610 | struct tcphdr *th; |
1611 | struct sock *sk; | |
1612 | int ret; | |
a86b1e30 | 1613 | struct net *net = dev_net(skb->dev); |
1da177e4 LT |
1614 | |
1615 | if (skb->pkt_type != PACKET_HOST) | |
1616 | goto discard_it; | |
1617 | ||
1618 | /* Count it even if it's bad */ | |
63231bdd | 1619 | TCP_INC_STATS_BH(net, TCP_MIB_INSEGS); |
1da177e4 LT |
1620 | |
1621 | if (!pskb_may_pull(skb, sizeof(struct tcphdr))) | |
1622 | goto discard_it; | |
1623 | ||
aa8223c7 | 1624 | th = tcp_hdr(skb); |
1da177e4 LT |
1625 | |
1626 | if (th->doff < sizeof(struct tcphdr) / 4) | |
1627 | goto bad_packet; | |
1628 | if (!pskb_may_pull(skb, th->doff * 4)) | |
1629 | goto discard_it; | |
1630 | ||
1631 | /* An explanation is required here, I think. | |
1632 | * Packet length and doff are validated by header prediction, | |
caa20d9a | 1633 | * provided case of th->doff==0 is eliminated. |
1da177e4 | 1634 | * So, we defer the checks. */ |
60476372 | 1635 | if (!skb_csum_unnecessary(skb) && tcp_v4_checksum_init(skb)) |
1da177e4 LT |
1636 | goto bad_packet; |
1637 | ||
aa8223c7 | 1638 | th = tcp_hdr(skb); |
eddc9ec5 | 1639 | iph = ip_hdr(skb); |
1da177e4 LT |
1640 | TCP_SKB_CB(skb)->seq = ntohl(th->seq); |
1641 | TCP_SKB_CB(skb)->end_seq = (TCP_SKB_CB(skb)->seq + th->syn + th->fin + | |
1642 | skb->len - th->doff * 4); | |
1643 | TCP_SKB_CB(skb)->ack_seq = ntohl(th->ack_seq); | |
1644 | TCP_SKB_CB(skb)->when = 0; | |
eddc9ec5 | 1645 | TCP_SKB_CB(skb)->flags = iph->tos; |
1da177e4 LT |
1646 | TCP_SKB_CB(skb)->sacked = 0; |
1647 | ||
9a1f27c4 | 1648 | sk = __inet_lookup_skb(&tcp_hashinfo, skb, th->source, th->dest); |
1da177e4 LT |
1649 | if (!sk) |
1650 | goto no_tcp_socket; | |
1651 | ||
d218d111 SH |
1652 | if (iph->ttl < inet_sk(sk)->min_ttl) |
1653 | goto discard_and_relse; | |
1654 | ||
1da177e4 LT |
1655 | process: |
1656 | if (sk->sk_state == TCP_TIME_WAIT) | |
1657 | goto do_time_wait; | |
1658 | ||
1659 | if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb)) | |
1660 | goto discard_and_relse; | |
b59c2701 | 1661 | nf_reset(skb); |
1da177e4 | 1662 | |
fda9ef5d | 1663 | if (sk_filter(sk, skb)) |
1da177e4 LT |
1664 | goto discard_and_relse; |
1665 | ||
1666 | skb->dev = NULL; | |
1667 | ||
c6366184 | 1668 | bh_lock_sock_nested(sk); |
1da177e4 LT |
1669 | ret = 0; |
1670 | if (!sock_owned_by_user(sk)) { | |
1a2449a8 CL |
1671 | #ifdef CONFIG_NET_DMA |
1672 | struct tcp_sock *tp = tcp_sk(sk); | |
1673 | if (!tp->ucopy.dma_chan && tp->ucopy.pinned_list) | |
f67b4599 | 1674 | tp->ucopy.dma_chan = dma_find_channel(DMA_MEMCPY); |
1a2449a8 | 1675 | if (tp->ucopy.dma_chan) |
1da177e4 | 1676 | ret = tcp_v4_do_rcv(sk, skb); |
1a2449a8 CL |
1677 | else |
1678 | #endif | |
1679 | { | |
1680 | if (!tcp_prequeue(sk, skb)) | |
ae8d7f88 | 1681 | ret = tcp_v4_do_rcv(sk, skb); |
1a2449a8 | 1682 | } |
1da177e4 LT |
1683 | } else |
1684 | sk_add_backlog(sk, skb); | |
1685 | bh_unlock_sock(sk); | |
1686 | ||
1687 | sock_put(sk); | |
1688 | ||
1689 | return ret; | |
1690 | ||
1691 | no_tcp_socket: | |
1692 | if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) | |
1693 | goto discard_it; | |
1694 | ||
1695 | if (skb->len < (th->doff << 2) || tcp_checksum_complete(skb)) { | |
1696 | bad_packet: | |
63231bdd | 1697 | TCP_INC_STATS_BH(net, TCP_MIB_INERRS); |
1da177e4 | 1698 | } else { |
cfb6eeb4 | 1699 | tcp_v4_send_reset(NULL, skb); |
1da177e4 LT |
1700 | } |
1701 | ||
1702 | discard_it: | |
1703 | /* Discard frame. */ | |
1704 | kfree_skb(skb); | |
e905a9ed | 1705 | return 0; |
1da177e4 LT |
1706 | |
1707 | discard_and_relse: | |
1708 | sock_put(sk); | |
1709 | goto discard_it; | |
1710 | ||
1711 | do_time_wait: | |
1712 | if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) { | |
9469c7b4 | 1713 | inet_twsk_put(inet_twsk(sk)); |
1da177e4 LT |
1714 | goto discard_it; |
1715 | } | |
1716 | ||
1717 | if (skb->len < (th->doff << 2) || tcp_checksum_complete(skb)) { | |
63231bdd | 1718 | TCP_INC_STATS_BH(net, TCP_MIB_INERRS); |
9469c7b4 | 1719 | inet_twsk_put(inet_twsk(sk)); |
1da177e4 LT |
1720 | goto discard_it; |
1721 | } | |
9469c7b4 | 1722 | switch (tcp_timewait_state_process(inet_twsk(sk), skb, th)) { |
1da177e4 | 1723 | case TCP_TW_SYN: { |
c346dca1 | 1724 | struct sock *sk2 = inet_lookup_listener(dev_net(skb->dev), |
c67499c0 | 1725 | &tcp_hashinfo, |
eddc9ec5 | 1726 | iph->daddr, th->dest, |
463c84b9 | 1727 | inet_iif(skb)); |
1da177e4 | 1728 | if (sk2) { |
9469c7b4 YH |
1729 | inet_twsk_deschedule(inet_twsk(sk), &tcp_death_row); |
1730 | inet_twsk_put(inet_twsk(sk)); | |
1da177e4 LT |
1731 | sk = sk2; |
1732 | goto process; | |
1733 | } | |
1734 | /* Fall through to ACK */ | |
1735 | } | |
1736 | case TCP_TW_ACK: | |
1737 | tcp_v4_timewait_ack(sk, skb); | |
1738 | break; | |
1739 | case TCP_TW_RST: | |
1740 | goto no_tcp_socket; | |
1741 | case TCP_TW_SUCCESS:; | |
1742 | } | |
1743 | goto discard_it; | |
1744 | } | |
1745 | ||
1da177e4 LT |
1746 | /* VJ's idea. Save last timestamp seen from this destination |
1747 | * and hold it at least for normal timewait interval to use for duplicate | |
1748 | * segment detection in subsequent connections, before they enter synchronized | |
1749 | * state. | |
1750 | */ | |
1751 | ||
1752 | int tcp_v4_remember_stamp(struct sock *sk) | |
1753 | { | |
1754 | struct inet_sock *inet = inet_sk(sk); | |
1755 | struct tcp_sock *tp = tcp_sk(sk); | |
1756 | struct rtable *rt = (struct rtable *)__sk_dst_get(sk); | |
1757 | struct inet_peer *peer = NULL; | |
1758 | int release_it = 0; | |
1759 | ||
c720c7e8 ED |
1760 | if (!rt || rt->rt_dst != inet->inet_daddr) { |
1761 | peer = inet_getpeer(inet->inet_daddr, 1); | |
1da177e4 LT |
1762 | release_it = 1; |
1763 | } else { | |
1764 | if (!rt->peer) | |
1765 | rt_bind_peer(rt, 1); | |
1766 | peer = rt->peer; | |
1767 | } | |
1768 | ||
1769 | if (peer) { | |
1770 | if ((s32)(peer->tcp_ts - tp->rx_opt.ts_recent) <= 0 || | |
2c1409a0 ED |
1771 | ((u32)get_seconds() - peer->tcp_ts_stamp > TCP_PAWS_MSL && |
1772 | peer->tcp_ts_stamp <= (u32)tp->rx_opt.ts_recent_stamp)) { | |
1773 | peer->tcp_ts_stamp = (u32)tp->rx_opt.ts_recent_stamp; | |
1da177e4 LT |
1774 | peer->tcp_ts = tp->rx_opt.ts_recent; |
1775 | } | |
1776 | if (release_it) | |
1777 | inet_putpeer(peer); | |
1778 | return 1; | |
1779 | } | |
1780 | ||
1781 | return 0; | |
1782 | } | |
1783 | ||
8feaf0c0 | 1784 | int tcp_v4_tw_remember_stamp(struct inet_timewait_sock *tw) |
1da177e4 | 1785 | { |
8feaf0c0 | 1786 | struct inet_peer *peer = inet_getpeer(tw->tw_daddr, 1); |
1da177e4 LT |
1787 | |
1788 | if (peer) { | |
8feaf0c0 ACM |
1789 | const struct tcp_timewait_sock *tcptw = tcp_twsk((struct sock *)tw); |
1790 | ||
1791 | if ((s32)(peer->tcp_ts - tcptw->tw_ts_recent) <= 0 || | |
2c1409a0 ED |
1792 | ((u32)get_seconds() - peer->tcp_ts_stamp > TCP_PAWS_MSL && |
1793 | peer->tcp_ts_stamp <= (u32)tcptw->tw_ts_recent_stamp)) { | |
1794 | peer->tcp_ts_stamp = (u32)tcptw->tw_ts_recent_stamp; | |
8feaf0c0 | 1795 | peer->tcp_ts = tcptw->tw_ts_recent; |
1da177e4 LT |
1796 | } |
1797 | inet_putpeer(peer); | |
1798 | return 1; | |
1799 | } | |
1800 | ||
1801 | return 0; | |
1802 | } | |
1803 | ||
3b401a81 | 1804 | const struct inet_connection_sock_af_ops ipv4_specific = { |
543d9cfe ACM |
1805 | .queue_xmit = ip_queue_xmit, |
1806 | .send_check = tcp_v4_send_check, | |
1807 | .rebuild_header = inet_sk_rebuild_header, | |
1808 | .conn_request = tcp_v4_conn_request, | |
1809 | .syn_recv_sock = tcp_v4_syn_recv_sock, | |
1810 | .remember_stamp = tcp_v4_remember_stamp, | |
1811 | .net_header_len = sizeof(struct iphdr), | |
1812 | .setsockopt = ip_setsockopt, | |
1813 | .getsockopt = ip_getsockopt, | |
1814 | .addr2sockaddr = inet_csk_addr2sockaddr, | |
1815 | .sockaddr_len = sizeof(struct sockaddr_in), | |
ab1e0a13 | 1816 | .bind_conflict = inet_csk_bind_conflict, |
3fdadf7d | 1817 | #ifdef CONFIG_COMPAT |
543d9cfe ACM |
1818 | .compat_setsockopt = compat_ip_setsockopt, |
1819 | .compat_getsockopt = compat_ip_getsockopt, | |
3fdadf7d | 1820 | #endif |
1da177e4 LT |
1821 | }; |
1822 | ||
cfb6eeb4 | 1823 | #ifdef CONFIG_TCP_MD5SIG |
b2e4b3de | 1824 | static const struct tcp_sock_af_ops tcp_sock_ipv4_specific = { |
cfb6eeb4 | 1825 | .md5_lookup = tcp_v4_md5_lookup, |
49a72dfb | 1826 | .calc_md5_hash = tcp_v4_md5_hash_skb, |
cfb6eeb4 YH |
1827 | .md5_add = tcp_v4_md5_add_func, |
1828 | .md5_parse = tcp_v4_parse_md5_keys, | |
cfb6eeb4 | 1829 | }; |
b6332e6c | 1830 | #endif |
cfb6eeb4 | 1831 | |
1da177e4 LT |
1832 | /* NOTE: A lot of things set to zero explicitly by call to |
1833 | * sk_alloc() so need not be done here. | |
1834 | */ | |
1835 | static int tcp_v4_init_sock(struct sock *sk) | |
1836 | { | |
6687e988 | 1837 | struct inet_connection_sock *icsk = inet_csk(sk); |
1da177e4 LT |
1838 | struct tcp_sock *tp = tcp_sk(sk); |
1839 | ||
1840 | skb_queue_head_init(&tp->out_of_order_queue); | |
1841 | tcp_init_xmit_timers(sk); | |
1842 | tcp_prequeue_init(tp); | |
1843 | ||
6687e988 | 1844 | icsk->icsk_rto = TCP_TIMEOUT_INIT; |
1da177e4 LT |
1845 | tp->mdev = TCP_TIMEOUT_INIT; |
1846 | ||
1847 | /* So many TCP implementations out there (incorrectly) count the | |
1848 | * initial SYN frame in their delayed-ACK and congestion control | |
1849 | * algorithms that we must have the following bandaid to talk | |
1850 | * efficiently to them. -DaveM | |
1851 | */ | |
1852 | tp->snd_cwnd = 2; | |
1853 | ||
1854 | /* See draft-stevens-tcpca-spec-01 for discussion of the | |
1855 | * initialization of these values. | |
1856 | */ | |
0b6a05c1 | 1857 | tp->snd_ssthresh = TCP_INFINITE_SSTHRESH; |
1da177e4 | 1858 | tp->snd_cwnd_clamp = ~0; |
bee7ca9e | 1859 | tp->mss_cache = TCP_MSS_DEFAULT; |
1da177e4 LT |
1860 | |
1861 | tp->reordering = sysctl_tcp_reordering; | |
6687e988 | 1862 | icsk->icsk_ca_ops = &tcp_init_congestion_ops; |
1da177e4 LT |
1863 | |
1864 | sk->sk_state = TCP_CLOSE; | |
1865 | ||
1866 | sk->sk_write_space = sk_stream_write_space; | |
1867 | sock_set_flag(sk, SOCK_USE_WRITE_QUEUE); | |
1868 | ||
8292a17a | 1869 | icsk->icsk_af_ops = &ipv4_specific; |
d83d8461 | 1870 | icsk->icsk_sync_mss = tcp_sync_mss; |
cfb6eeb4 YH |
1871 | #ifdef CONFIG_TCP_MD5SIG |
1872 | tp->af_specific = &tcp_sock_ipv4_specific; | |
1873 | #endif | |
1da177e4 | 1874 | |
435cf559 WAS |
1875 | /* TCP Cookie Transactions */ |
1876 | if (sysctl_tcp_cookie_size > 0) { | |
1877 | /* Default, cookies without s_data_payload. */ | |
1878 | tp->cookie_values = | |
1879 | kzalloc(sizeof(*tp->cookie_values), | |
1880 | sk->sk_allocation); | |
1881 | if (tp->cookie_values != NULL) | |
1882 | kref_init(&tp->cookie_values->kref); | |
1883 | } | |
1884 | /* Presumed zeroed, in order of appearance: | |
1885 | * cookie_in_always, cookie_out_never, | |
1886 | * s_data_constant, s_data_in, s_data_out | |
1887 | */ | |
1da177e4 LT |
1888 | sk->sk_sndbuf = sysctl_tcp_wmem[1]; |
1889 | sk->sk_rcvbuf = sysctl_tcp_rmem[1]; | |
1890 | ||
eb4dea58 | 1891 | local_bh_disable(); |
1748376b | 1892 | percpu_counter_inc(&tcp_sockets_allocated); |
eb4dea58 | 1893 | local_bh_enable(); |
1da177e4 LT |
1894 | |
1895 | return 0; | |
1896 | } | |
1897 | ||
7d06b2e0 | 1898 | void tcp_v4_destroy_sock(struct sock *sk) |
1da177e4 LT |
1899 | { |
1900 | struct tcp_sock *tp = tcp_sk(sk); | |
1901 | ||
1902 | tcp_clear_xmit_timers(sk); | |
1903 | ||
6687e988 | 1904 | tcp_cleanup_congestion_control(sk); |
317a76f9 | 1905 | |
1da177e4 | 1906 | /* Cleanup up the write buffer. */ |
fe067e8a | 1907 | tcp_write_queue_purge(sk); |
1da177e4 LT |
1908 | |
1909 | /* Cleans up our, hopefully empty, out_of_order_queue. */ | |
e905a9ed | 1910 | __skb_queue_purge(&tp->out_of_order_queue); |
1da177e4 | 1911 | |
cfb6eeb4 YH |
1912 | #ifdef CONFIG_TCP_MD5SIG |
1913 | /* Clean up the MD5 key list, if any */ | |
1914 | if (tp->md5sig_info) { | |
1915 | tcp_v4_clear_md5_list(sk); | |
1916 | kfree(tp->md5sig_info); | |
1917 | tp->md5sig_info = NULL; | |
1918 | } | |
1919 | #endif | |
1920 | ||
1a2449a8 CL |
1921 | #ifdef CONFIG_NET_DMA |
1922 | /* Cleans up our sk_async_wait_queue */ | |
e905a9ed | 1923 | __skb_queue_purge(&sk->sk_async_wait_queue); |
1a2449a8 CL |
1924 | #endif |
1925 | ||
1da177e4 LT |
1926 | /* Clean prequeue, it must be empty really */ |
1927 | __skb_queue_purge(&tp->ucopy.prequeue); | |
1928 | ||
1929 | /* Clean up a referenced TCP bind bucket. */ | |
463c84b9 | 1930 | if (inet_csk(sk)->icsk_bind_hash) |
ab1e0a13 | 1931 | inet_put_port(sk); |
1da177e4 LT |
1932 | |
1933 | /* | |
1934 | * If sendmsg cached page exists, toss it. | |
1935 | */ | |
1936 | if (sk->sk_sndmsg_page) { | |
1937 | __free_page(sk->sk_sndmsg_page); | |
1938 | sk->sk_sndmsg_page = NULL; | |
1939 | } | |
1940 | ||
435cf559 WAS |
1941 | /* TCP Cookie Transactions */ |
1942 | if (tp->cookie_values != NULL) { | |
1943 | kref_put(&tp->cookie_values->kref, | |
1944 | tcp_cookie_values_release); | |
1945 | tp->cookie_values = NULL; | |
1946 | } | |
1947 | ||
1748376b | 1948 | percpu_counter_dec(&tcp_sockets_allocated); |
1da177e4 LT |
1949 | } |
1950 | ||
1951 | EXPORT_SYMBOL(tcp_v4_destroy_sock); | |
1952 | ||
1953 | #ifdef CONFIG_PROC_FS | |
1954 | /* Proc filesystem TCP sock list dumping. */ | |
1955 | ||
3ab5aee7 | 1956 | static inline struct inet_timewait_sock *tw_head(struct hlist_nulls_head *head) |
1da177e4 | 1957 | { |
3ab5aee7 | 1958 | return hlist_nulls_empty(head) ? NULL : |
8feaf0c0 | 1959 | list_entry(head->first, struct inet_timewait_sock, tw_node); |
1da177e4 LT |
1960 | } |
1961 | ||
8feaf0c0 | 1962 | static inline struct inet_timewait_sock *tw_next(struct inet_timewait_sock *tw) |
1da177e4 | 1963 | { |
3ab5aee7 ED |
1964 | return !is_a_nulls(tw->tw_node.next) ? |
1965 | hlist_nulls_entry(tw->tw_node.next, typeof(*tw), tw_node) : NULL; | |
1da177e4 LT |
1966 | } |
1967 | ||
1968 | static void *listening_get_next(struct seq_file *seq, void *cur) | |
1969 | { | |
463c84b9 | 1970 | struct inet_connection_sock *icsk; |
c25eb3bf | 1971 | struct hlist_nulls_node *node; |
1da177e4 | 1972 | struct sock *sk = cur; |
5caea4ea | 1973 | struct inet_listen_hashbucket *ilb; |
5799de0b | 1974 | struct tcp_iter_state *st = seq->private; |
a4146b1b | 1975 | struct net *net = seq_file_net(seq); |
1da177e4 LT |
1976 | |
1977 | if (!sk) { | |
1978 | st->bucket = 0; | |
5caea4ea ED |
1979 | ilb = &tcp_hashinfo.listening_hash[0]; |
1980 | spin_lock_bh(&ilb->lock); | |
c25eb3bf | 1981 | sk = sk_nulls_head(&ilb->head); |
1da177e4 LT |
1982 | goto get_sk; |
1983 | } | |
5caea4ea | 1984 | ilb = &tcp_hashinfo.listening_hash[st->bucket]; |
1da177e4 LT |
1985 | ++st->num; |
1986 | ||
1987 | if (st->state == TCP_SEQ_STATE_OPENREQ) { | |
60236fdd | 1988 | struct request_sock *req = cur; |
1da177e4 | 1989 | |
72a3effa | 1990 | icsk = inet_csk(st->syn_wait_sk); |
1da177e4 LT |
1991 | req = req->dl_next; |
1992 | while (1) { | |
1993 | while (req) { | |
bdccc4ca | 1994 | if (req->rsk_ops->family == st->family) { |
1da177e4 LT |
1995 | cur = req; |
1996 | goto out; | |
1997 | } | |
1998 | req = req->dl_next; | |
1999 | } | |
72a3effa | 2000 | if (++st->sbucket >= icsk->icsk_accept_queue.listen_opt->nr_table_entries) |
1da177e4 LT |
2001 | break; |
2002 | get_req: | |
463c84b9 | 2003 | req = icsk->icsk_accept_queue.listen_opt->syn_table[st->sbucket]; |
1da177e4 LT |
2004 | } |
2005 | sk = sk_next(st->syn_wait_sk); | |
2006 | st->state = TCP_SEQ_STATE_LISTENING; | |
463c84b9 | 2007 | read_unlock_bh(&icsk->icsk_accept_queue.syn_wait_lock); |
1da177e4 | 2008 | } else { |
e905a9ed | 2009 | icsk = inet_csk(sk); |
463c84b9 ACM |
2010 | read_lock_bh(&icsk->icsk_accept_queue.syn_wait_lock); |
2011 | if (reqsk_queue_len(&icsk->icsk_accept_queue)) | |
1da177e4 | 2012 | goto start_req; |
463c84b9 | 2013 | read_unlock_bh(&icsk->icsk_accept_queue.syn_wait_lock); |
1da177e4 LT |
2014 | sk = sk_next(sk); |
2015 | } | |
2016 | get_sk: | |
c25eb3bf | 2017 | sk_nulls_for_each_from(sk, node) { |
878628fb | 2018 | if (sk->sk_family == st->family && net_eq(sock_net(sk), net)) { |
1da177e4 LT |
2019 | cur = sk; |
2020 | goto out; | |
2021 | } | |
e905a9ed | 2022 | icsk = inet_csk(sk); |
463c84b9 ACM |
2023 | read_lock_bh(&icsk->icsk_accept_queue.syn_wait_lock); |
2024 | if (reqsk_queue_len(&icsk->icsk_accept_queue)) { | |
1da177e4 LT |
2025 | start_req: |
2026 | st->uid = sock_i_uid(sk); | |
2027 | st->syn_wait_sk = sk; | |
2028 | st->state = TCP_SEQ_STATE_OPENREQ; | |
2029 | st->sbucket = 0; | |
2030 | goto get_req; | |
2031 | } | |
463c84b9 | 2032 | read_unlock_bh(&icsk->icsk_accept_queue.syn_wait_lock); |
1da177e4 | 2033 | } |
5caea4ea | 2034 | spin_unlock_bh(&ilb->lock); |
0f7ff927 | 2035 | if (++st->bucket < INET_LHTABLE_SIZE) { |
5caea4ea ED |
2036 | ilb = &tcp_hashinfo.listening_hash[st->bucket]; |
2037 | spin_lock_bh(&ilb->lock); | |
c25eb3bf | 2038 | sk = sk_nulls_head(&ilb->head); |
1da177e4 LT |
2039 | goto get_sk; |
2040 | } | |
2041 | cur = NULL; | |
2042 | out: | |
2043 | return cur; | |
2044 | } | |
2045 | ||
2046 | static void *listening_get_idx(struct seq_file *seq, loff_t *pos) | |
2047 | { | |
2048 | void *rc = listening_get_next(seq, NULL); | |
2049 | ||
2050 | while (rc && *pos) { | |
2051 | rc = listening_get_next(seq, rc); | |
2052 | --*pos; | |
2053 | } | |
2054 | return rc; | |
2055 | } | |
2056 | ||
6eac5604 AK |
2057 | static inline int empty_bucket(struct tcp_iter_state *st) |
2058 | { | |
3ab5aee7 ED |
2059 | return hlist_nulls_empty(&tcp_hashinfo.ehash[st->bucket].chain) && |
2060 | hlist_nulls_empty(&tcp_hashinfo.ehash[st->bucket].twchain); | |
6eac5604 AK |
2061 | } |
2062 | ||
1da177e4 LT |
2063 | static void *established_get_first(struct seq_file *seq) |
2064 | { | |
5799de0b | 2065 | struct tcp_iter_state *st = seq->private; |
a4146b1b | 2066 | struct net *net = seq_file_net(seq); |
1da177e4 LT |
2067 | void *rc = NULL; |
2068 | ||
f373b53b | 2069 | for (st->bucket = 0; st->bucket <= tcp_hashinfo.ehash_mask; ++st->bucket) { |
1da177e4 | 2070 | struct sock *sk; |
3ab5aee7 | 2071 | struct hlist_nulls_node *node; |
8feaf0c0 | 2072 | struct inet_timewait_sock *tw; |
9db66bdc | 2073 | spinlock_t *lock = inet_ehash_lockp(&tcp_hashinfo, st->bucket); |
1da177e4 | 2074 | |
6eac5604 AK |
2075 | /* Lockless fast path for the common case of empty buckets */ |
2076 | if (empty_bucket(st)) | |
2077 | continue; | |
2078 | ||
9db66bdc | 2079 | spin_lock_bh(lock); |
3ab5aee7 | 2080 | sk_nulls_for_each(sk, node, &tcp_hashinfo.ehash[st->bucket].chain) { |
f40c8174 | 2081 | if (sk->sk_family != st->family || |
878628fb | 2082 | !net_eq(sock_net(sk), net)) { |
1da177e4 LT |
2083 | continue; |
2084 | } | |
2085 | rc = sk; | |
2086 | goto out; | |
2087 | } | |
2088 | st->state = TCP_SEQ_STATE_TIME_WAIT; | |
8feaf0c0 | 2089 | inet_twsk_for_each(tw, node, |
dbca9b27 | 2090 | &tcp_hashinfo.ehash[st->bucket].twchain) { |
28518fc1 | 2091 | if (tw->tw_family != st->family || |
878628fb | 2092 | !net_eq(twsk_net(tw), net)) { |
1da177e4 LT |
2093 | continue; |
2094 | } | |
2095 | rc = tw; | |
2096 | goto out; | |
2097 | } | |
9db66bdc | 2098 | spin_unlock_bh(lock); |
1da177e4 LT |
2099 | st->state = TCP_SEQ_STATE_ESTABLISHED; |
2100 | } | |
2101 | out: | |
2102 | return rc; | |
2103 | } | |
2104 | ||
2105 | static void *established_get_next(struct seq_file *seq, void *cur) | |
2106 | { | |
2107 | struct sock *sk = cur; | |
8feaf0c0 | 2108 | struct inet_timewait_sock *tw; |
3ab5aee7 | 2109 | struct hlist_nulls_node *node; |
5799de0b | 2110 | struct tcp_iter_state *st = seq->private; |
a4146b1b | 2111 | struct net *net = seq_file_net(seq); |
1da177e4 LT |
2112 | |
2113 | ++st->num; | |
2114 | ||
2115 | if (st->state == TCP_SEQ_STATE_TIME_WAIT) { | |
2116 | tw = cur; | |
2117 | tw = tw_next(tw); | |
2118 | get_tw: | |
878628fb | 2119 | while (tw && (tw->tw_family != st->family || !net_eq(twsk_net(tw), net))) { |
1da177e4 LT |
2120 | tw = tw_next(tw); |
2121 | } | |
2122 | if (tw) { | |
2123 | cur = tw; | |
2124 | goto out; | |
2125 | } | |
9db66bdc | 2126 | spin_unlock_bh(inet_ehash_lockp(&tcp_hashinfo, st->bucket)); |
1da177e4 LT |
2127 | st->state = TCP_SEQ_STATE_ESTABLISHED; |
2128 | ||
6eac5604 | 2129 | /* Look for next non empty bucket */ |
f373b53b | 2130 | while (++st->bucket <= tcp_hashinfo.ehash_mask && |
6eac5604 AK |
2131 | empty_bucket(st)) |
2132 | ; | |
f373b53b | 2133 | if (st->bucket > tcp_hashinfo.ehash_mask) |
6eac5604 AK |
2134 | return NULL; |
2135 | ||
9db66bdc | 2136 | spin_lock_bh(inet_ehash_lockp(&tcp_hashinfo, st->bucket)); |
3ab5aee7 | 2137 | sk = sk_nulls_head(&tcp_hashinfo.ehash[st->bucket].chain); |
1da177e4 | 2138 | } else |
3ab5aee7 | 2139 | sk = sk_nulls_next(sk); |
1da177e4 | 2140 | |
3ab5aee7 | 2141 | sk_nulls_for_each_from(sk, node) { |
878628fb | 2142 | if (sk->sk_family == st->family && net_eq(sock_net(sk), net)) |
1da177e4 LT |
2143 | goto found; |
2144 | } | |
2145 | ||
2146 | st->state = TCP_SEQ_STATE_TIME_WAIT; | |
dbca9b27 | 2147 | tw = tw_head(&tcp_hashinfo.ehash[st->bucket].twchain); |
1da177e4 LT |
2148 | goto get_tw; |
2149 | found: | |
2150 | cur = sk; | |
2151 | out: | |
2152 | return cur; | |
2153 | } | |
2154 | ||
2155 | static void *established_get_idx(struct seq_file *seq, loff_t pos) | |
2156 | { | |
2157 | void *rc = established_get_first(seq); | |
2158 | ||
2159 | while (rc && pos) { | |
2160 | rc = established_get_next(seq, rc); | |
2161 | --pos; | |
7174259e | 2162 | } |
1da177e4 LT |
2163 | return rc; |
2164 | } | |
2165 | ||
2166 | static void *tcp_get_idx(struct seq_file *seq, loff_t pos) | |
2167 | { | |
2168 | void *rc; | |
5799de0b | 2169 | struct tcp_iter_state *st = seq->private; |
1da177e4 | 2170 | |
1da177e4 LT |
2171 | st->state = TCP_SEQ_STATE_LISTENING; |
2172 | rc = listening_get_idx(seq, &pos); | |
2173 | ||
2174 | if (!rc) { | |
1da177e4 LT |
2175 | st->state = TCP_SEQ_STATE_ESTABLISHED; |
2176 | rc = established_get_idx(seq, pos); | |
2177 | } | |
2178 | ||
2179 | return rc; | |
2180 | } | |
2181 | ||
2182 | static void *tcp_seq_start(struct seq_file *seq, loff_t *pos) | |
2183 | { | |
5799de0b | 2184 | struct tcp_iter_state *st = seq->private; |
1da177e4 LT |
2185 | st->state = TCP_SEQ_STATE_LISTENING; |
2186 | st->num = 0; | |
2187 | return *pos ? tcp_get_idx(seq, *pos - 1) : SEQ_START_TOKEN; | |
2188 | } | |
2189 | ||
2190 | static void *tcp_seq_next(struct seq_file *seq, void *v, loff_t *pos) | |
2191 | { | |
2192 | void *rc = NULL; | |
5799de0b | 2193 | struct tcp_iter_state *st; |
1da177e4 LT |
2194 | |
2195 | if (v == SEQ_START_TOKEN) { | |
2196 | rc = tcp_get_idx(seq, 0); | |
2197 | goto out; | |
2198 | } | |
2199 | st = seq->private; | |
2200 | ||
2201 | switch (st->state) { | |
2202 | case TCP_SEQ_STATE_OPENREQ: | |
2203 | case TCP_SEQ_STATE_LISTENING: | |
2204 | rc = listening_get_next(seq, v); | |
2205 | if (!rc) { | |
1da177e4 LT |
2206 | st->state = TCP_SEQ_STATE_ESTABLISHED; |
2207 | rc = established_get_first(seq); | |
2208 | } | |
2209 | break; | |
2210 | case TCP_SEQ_STATE_ESTABLISHED: | |
2211 | case TCP_SEQ_STATE_TIME_WAIT: | |
2212 | rc = established_get_next(seq, v); | |
2213 | break; | |
2214 | } | |
2215 | out: | |
2216 | ++*pos; | |
2217 | return rc; | |
2218 | } | |
2219 | ||
2220 | static void tcp_seq_stop(struct seq_file *seq, void *v) | |
2221 | { | |
5799de0b | 2222 | struct tcp_iter_state *st = seq->private; |
1da177e4 LT |
2223 | |
2224 | switch (st->state) { | |
2225 | case TCP_SEQ_STATE_OPENREQ: | |
2226 | if (v) { | |
463c84b9 ACM |
2227 | struct inet_connection_sock *icsk = inet_csk(st->syn_wait_sk); |
2228 | read_unlock_bh(&icsk->icsk_accept_queue.syn_wait_lock); | |
1da177e4 LT |
2229 | } |
2230 | case TCP_SEQ_STATE_LISTENING: | |
2231 | if (v != SEQ_START_TOKEN) | |
5caea4ea | 2232 | spin_unlock_bh(&tcp_hashinfo.listening_hash[st->bucket].lock); |
1da177e4 LT |
2233 | break; |
2234 | case TCP_SEQ_STATE_TIME_WAIT: | |
2235 | case TCP_SEQ_STATE_ESTABLISHED: | |
2236 | if (v) | |
9db66bdc | 2237 | spin_unlock_bh(inet_ehash_lockp(&tcp_hashinfo, st->bucket)); |
1da177e4 LT |
2238 | break; |
2239 | } | |
2240 | } | |
2241 | ||
2242 | static int tcp_seq_open(struct inode *inode, struct file *file) | |
2243 | { | |
2244 | struct tcp_seq_afinfo *afinfo = PDE(inode)->data; | |
1da177e4 | 2245 | struct tcp_iter_state *s; |
52d6f3f1 | 2246 | int err; |
1da177e4 | 2247 | |
52d6f3f1 DL |
2248 | err = seq_open_net(inode, file, &afinfo->seq_ops, |
2249 | sizeof(struct tcp_iter_state)); | |
2250 | if (err < 0) | |
2251 | return err; | |
f40c8174 | 2252 | |
52d6f3f1 | 2253 | s = ((struct seq_file *)file->private_data)->private; |
1da177e4 | 2254 | s->family = afinfo->family; |
f40c8174 DL |
2255 | return 0; |
2256 | } | |
2257 | ||
6f8b13bc | 2258 | int tcp_proc_register(struct net *net, struct tcp_seq_afinfo *afinfo) |
1da177e4 LT |
2259 | { |
2260 | int rc = 0; | |
2261 | struct proc_dir_entry *p; | |
2262 | ||
68fcadd1 DL |
2263 | afinfo->seq_fops.open = tcp_seq_open; |
2264 | afinfo->seq_fops.read = seq_read; | |
2265 | afinfo->seq_fops.llseek = seq_lseek; | |
2266 | afinfo->seq_fops.release = seq_release_net; | |
7174259e | 2267 | |
9427c4b3 DL |
2268 | afinfo->seq_ops.start = tcp_seq_start; |
2269 | afinfo->seq_ops.next = tcp_seq_next; | |
2270 | afinfo->seq_ops.stop = tcp_seq_stop; | |
2271 | ||
84841c3c DL |
2272 | p = proc_create_data(afinfo->name, S_IRUGO, net->proc_net, |
2273 | &afinfo->seq_fops, afinfo); | |
2274 | if (!p) | |
1da177e4 LT |
2275 | rc = -ENOMEM; |
2276 | return rc; | |
2277 | } | |
2278 | ||
6f8b13bc | 2279 | void tcp_proc_unregister(struct net *net, struct tcp_seq_afinfo *afinfo) |
1da177e4 | 2280 | { |
6f8b13bc | 2281 | proc_net_remove(net, afinfo->name); |
1da177e4 LT |
2282 | } |
2283 | ||
60236fdd | 2284 | static void get_openreq4(struct sock *sk, struct request_sock *req, |
5e659e4c | 2285 | struct seq_file *f, int i, int uid, int *len) |
1da177e4 | 2286 | { |
2e6599cb | 2287 | const struct inet_request_sock *ireq = inet_rsk(req); |
1da177e4 LT |
2288 | int ttd = req->expires - jiffies; |
2289 | ||
5e659e4c PE |
2290 | seq_printf(f, "%4d: %08X:%04X %08X:%04X" |
2291 | " %02X %08X:%08X %02X:%08lX %08X %5d %8d %u %d %p%n", | |
1da177e4 | 2292 | i, |
2e6599cb | 2293 | ireq->loc_addr, |
c720c7e8 | 2294 | ntohs(inet_sk(sk)->inet_sport), |
2e6599cb ACM |
2295 | ireq->rmt_addr, |
2296 | ntohs(ireq->rmt_port), | |
1da177e4 LT |
2297 | TCP_SYN_RECV, |
2298 | 0, 0, /* could print option size, but that is af dependent. */ | |
2299 | 1, /* timers active (only the expire timer) */ | |
2300 | jiffies_to_clock_t(ttd), | |
2301 | req->retrans, | |
2302 | uid, | |
2303 | 0, /* non standard timer */ | |
2304 | 0, /* open_requests have no inode */ | |
2305 | atomic_read(&sk->sk_refcnt), | |
5e659e4c PE |
2306 | req, |
2307 | len); | |
1da177e4 LT |
2308 | } |
2309 | ||
5e659e4c | 2310 | static void get_tcp4_sock(struct sock *sk, struct seq_file *f, int i, int *len) |
1da177e4 LT |
2311 | { |
2312 | int timer_active; | |
2313 | unsigned long timer_expires; | |
cf4c6bf8 IJ |
2314 | struct tcp_sock *tp = tcp_sk(sk); |
2315 | const struct inet_connection_sock *icsk = inet_csk(sk); | |
2316 | struct inet_sock *inet = inet_sk(sk); | |
c720c7e8 ED |
2317 | __be32 dest = inet->inet_daddr; |
2318 | __be32 src = inet->inet_rcv_saddr; | |
2319 | __u16 destp = ntohs(inet->inet_dport); | |
2320 | __u16 srcp = ntohs(inet->inet_sport); | |
49d09007 | 2321 | int rx_queue; |
1da177e4 | 2322 | |
463c84b9 | 2323 | if (icsk->icsk_pending == ICSK_TIME_RETRANS) { |
1da177e4 | 2324 | timer_active = 1; |
463c84b9 ACM |
2325 | timer_expires = icsk->icsk_timeout; |
2326 | } else if (icsk->icsk_pending == ICSK_TIME_PROBE0) { | |
1da177e4 | 2327 | timer_active = 4; |
463c84b9 | 2328 | timer_expires = icsk->icsk_timeout; |
cf4c6bf8 | 2329 | } else if (timer_pending(&sk->sk_timer)) { |
1da177e4 | 2330 | timer_active = 2; |
cf4c6bf8 | 2331 | timer_expires = sk->sk_timer.expires; |
1da177e4 LT |
2332 | } else { |
2333 | timer_active = 0; | |
2334 | timer_expires = jiffies; | |
2335 | } | |
2336 | ||
49d09007 ED |
2337 | if (sk->sk_state == TCP_LISTEN) |
2338 | rx_queue = sk->sk_ack_backlog; | |
2339 | else | |
2340 | /* | |
2341 | * because we dont lock socket, we might find a transient negative value | |
2342 | */ | |
2343 | rx_queue = max_t(int, tp->rcv_nxt - tp->copied_seq, 0); | |
2344 | ||
5e659e4c | 2345 | seq_printf(f, "%4d: %08X:%04X %08X:%04X %02X %08X:%08X %02X:%08lX " |
7be87351 | 2346 | "%08X %5d %8d %lu %d %p %lu %lu %u %u %d%n", |
cf4c6bf8 | 2347 | i, src, srcp, dest, destp, sk->sk_state, |
47da8ee6 | 2348 | tp->write_seq - tp->snd_una, |
49d09007 | 2349 | rx_queue, |
1da177e4 LT |
2350 | timer_active, |
2351 | jiffies_to_clock_t(timer_expires - jiffies), | |
463c84b9 | 2352 | icsk->icsk_retransmits, |
cf4c6bf8 | 2353 | sock_i_uid(sk), |
6687e988 | 2354 | icsk->icsk_probes_out, |
cf4c6bf8 IJ |
2355 | sock_i_ino(sk), |
2356 | atomic_read(&sk->sk_refcnt), sk, | |
7be87351 SH |
2357 | jiffies_to_clock_t(icsk->icsk_rto), |
2358 | jiffies_to_clock_t(icsk->icsk_ack.ato), | |
463c84b9 | 2359 | (icsk->icsk_ack.quick << 1) | icsk->icsk_ack.pingpong, |
1da177e4 | 2360 | tp->snd_cwnd, |
0b6a05c1 | 2361 | tcp_in_initial_slowstart(tp) ? -1 : tp->snd_ssthresh, |
5e659e4c | 2362 | len); |
1da177e4 LT |
2363 | } |
2364 | ||
7174259e | 2365 | static void get_timewait4_sock(struct inet_timewait_sock *tw, |
5e659e4c | 2366 | struct seq_file *f, int i, int *len) |
1da177e4 | 2367 | { |
23f33c2d | 2368 | __be32 dest, src; |
1da177e4 LT |
2369 | __u16 destp, srcp; |
2370 | int ttd = tw->tw_ttd - jiffies; | |
2371 | ||
2372 | if (ttd < 0) | |
2373 | ttd = 0; | |
2374 | ||
2375 | dest = tw->tw_daddr; | |
2376 | src = tw->tw_rcv_saddr; | |
2377 | destp = ntohs(tw->tw_dport); | |
2378 | srcp = ntohs(tw->tw_sport); | |
2379 | ||
5e659e4c PE |
2380 | seq_printf(f, "%4d: %08X:%04X %08X:%04X" |
2381 | " %02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %p%n", | |
1da177e4 LT |
2382 | i, src, srcp, dest, destp, tw->tw_substate, 0, 0, |
2383 | 3, jiffies_to_clock_t(ttd), 0, 0, 0, 0, | |
5e659e4c | 2384 | atomic_read(&tw->tw_refcnt), tw, len); |
1da177e4 LT |
2385 | } |
2386 | ||
2387 | #define TMPSZ 150 | |
2388 | ||
2389 | static int tcp4_seq_show(struct seq_file *seq, void *v) | |
2390 | { | |
5799de0b | 2391 | struct tcp_iter_state *st; |
5e659e4c | 2392 | int len; |
1da177e4 LT |
2393 | |
2394 | if (v == SEQ_START_TOKEN) { | |
2395 | seq_printf(seq, "%-*s\n", TMPSZ - 1, | |
2396 | " sl local_address rem_address st tx_queue " | |
2397 | "rx_queue tr tm->when retrnsmt uid timeout " | |
2398 | "inode"); | |
2399 | goto out; | |
2400 | } | |
2401 | st = seq->private; | |
2402 | ||
2403 | switch (st->state) { | |
2404 | case TCP_SEQ_STATE_LISTENING: | |
2405 | case TCP_SEQ_STATE_ESTABLISHED: | |
5e659e4c | 2406 | get_tcp4_sock(v, seq, st->num, &len); |
1da177e4 LT |
2407 | break; |
2408 | case TCP_SEQ_STATE_OPENREQ: | |
5e659e4c | 2409 | get_openreq4(st->syn_wait_sk, v, seq, st->num, st->uid, &len); |
1da177e4 LT |
2410 | break; |
2411 | case TCP_SEQ_STATE_TIME_WAIT: | |
5e659e4c | 2412 | get_timewait4_sock(v, seq, st->num, &len); |
1da177e4 LT |
2413 | break; |
2414 | } | |
5e659e4c | 2415 | seq_printf(seq, "%*s\n", TMPSZ - 1 - len, ""); |
1da177e4 LT |
2416 | out: |
2417 | return 0; | |
2418 | } | |
2419 | ||
1da177e4 | 2420 | static struct tcp_seq_afinfo tcp4_seq_afinfo = { |
1da177e4 LT |
2421 | .name = "tcp", |
2422 | .family = AF_INET, | |
5f4472c5 DL |
2423 | .seq_fops = { |
2424 | .owner = THIS_MODULE, | |
2425 | }, | |
9427c4b3 DL |
2426 | .seq_ops = { |
2427 | .show = tcp4_seq_show, | |
2428 | }, | |
1da177e4 LT |
2429 | }; |
2430 | ||
757764f6 PE |
2431 | static int tcp4_proc_init_net(struct net *net) |
2432 | { | |
2433 | return tcp_proc_register(net, &tcp4_seq_afinfo); | |
2434 | } | |
2435 | ||
2436 | static void tcp4_proc_exit_net(struct net *net) | |
2437 | { | |
2438 | tcp_proc_unregister(net, &tcp4_seq_afinfo); | |
2439 | } | |
2440 | ||
2441 | static struct pernet_operations tcp4_net_ops = { | |
2442 | .init = tcp4_proc_init_net, | |
2443 | .exit = tcp4_proc_exit_net, | |
2444 | }; | |
2445 | ||
1da177e4 LT |
2446 | int __init tcp4_proc_init(void) |
2447 | { | |
757764f6 | 2448 | return register_pernet_subsys(&tcp4_net_ops); |
1da177e4 LT |
2449 | } |
2450 | ||
2451 | void tcp4_proc_exit(void) | |
2452 | { | |
757764f6 | 2453 | unregister_pernet_subsys(&tcp4_net_ops); |
1da177e4 LT |
2454 | } |
2455 | #endif /* CONFIG_PROC_FS */ | |
2456 | ||
bf296b12 HX |
2457 | struct sk_buff **tcp4_gro_receive(struct sk_buff **head, struct sk_buff *skb) |
2458 | { | |
36e7b1b8 | 2459 | struct iphdr *iph = skb_gro_network_header(skb); |
bf296b12 HX |
2460 | |
2461 | switch (skb->ip_summed) { | |
2462 | case CHECKSUM_COMPLETE: | |
86911732 | 2463 | if (!tcp_v4_check(skb_gro_len(skb), iph->saddr, iph->daddr, |
bf296b12 HX |
2464 | skb->csum)) { |
2465 | skb->ip_summed = CHECKSUM_UNNECESSARY; | |
2466 | break; | |
2467 | } | |
2468 | ||
2469 | /* fall through */ | |
2470 | case CHECKSUM_NONE: | |
2471 | NAPI_GRO_CB(skb)->flush = 1; | |
2472 | return NULL; | |
2473 | } | |
2474 | ||
2475 | return tcp_gro_receive(head, skb); | |
2476 | } | |
2477 | EXPORT_SYMBOL(tcp4_gro_receive); | |
2478 | ||
2479 | int tcp4_gro_complete(struct sk_buff *skb) | |
2480 | { | |
2481 | struct iphdr *iph = ip_hdr(skb); | |
2482 | struct tcphdr *th = tcp_hdr(skb); | |
2483 | ||
2484 | th->check = ~tcp_v4_check(skb->len - skb_transport_offset(skb), | |
2485 | iph->saddr, iph->daddr, 0); | |
2486 | skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4; | |
2487 | ||
2488 | return tcp_gro_complete(skb); | |
2489 | } | |
2490 | EXPORT_SYMBOL(tcp4_gro_complete); | |
2491 | ||
1da177e4 LT |
2492 | struct proto tcp_prot = { |
2493 | .name = "TCP", | |
2494 | .owner = THIS_MODULE, | |
2495 | .close = tcp_close, | |
2496 | .connect = tcp_v4_connect, | |
2497 | .disconnect = tcp_disconnect, | |
463c84b9 | 2498 | .accept = inet_csk_accept, |
1da177e4 LT |
2499 | .ioctl = tcp_ioctl, |
2500 | .init = tcp_v4_init_sock, | |
2501 | .destroy = tcp_v4_destroy_sock, | |
2502 | .shutdown = tcp_shutdown, | |
2503 | .setsockopt = tcp_setsockopt, | |
2504 | .getsockopt = tcp_getsockopt, | |
1da177e4 LT |
2505 | .recvmsg = tcp_recvmsg, |
2506 | .backlog_rcv = tcp_v4_do_rcv, | |
ab1e0a13 ACM |
2507 | .hash = inet_hash, |
2508 | .unhash = inet_unhash, | |
2509 | .get_port = inet_csk_get_port, | |
1da177e4 LT |
2510 | .enter_memory_pressure = tcp_enter_memory_pressure, |
2511 | .sockets_allocated = &tcp_sockets_allocated, | |
0a5578cf | 2512 | .orphan_count = &tcp_orphan_count, |
1da177e4 LT |
2513 | .memory_allocated = &tcp_memory_allocated, |
2514 | .memory_pressure = &tcp_memory_pressure, | |
2515 | .sysctl_mem = sysctl_tcp_mem, | |
2516 | .sysctl_wmem = sysctl_tcp_wmem, | |
2517 | .sysctl_rmem = sysctl_tcp_rmem, | |
2518 | .max_header = MAX_TCP_HEADER, | |
2519 | .obj_size = sizeof(struct tcp_sock), | |
3ab5aee7 | 2520 | .slab_flags = SLAB_DESTROY_BY_RCU, |
6d6ee43e | 2521 | .twsk_prot = &tcp_timewait_sock_ops, |
60236fdd | 2522 | .rsk_prot = &tcp_request_sock_ops, |
39d8cda7 | 2523 | .h.hashinfo = &tcp_hashinfo, |
543d9cfe ACM |
2524 | #ifdef CONFIG_COMPAT |
2525 | .compat_setsockopt = compat_tcp_setsockopt, | |
2526 | .compat_getsockopt = compat_tcp_getsockopt, | |
2527 | #endif | |
1da177e4 LT |
2528 | }; |
2529 | ||
046ee902 DL |
2530 | |
2531 | static int __net_init tcp_sk_init(struct net *net) | |
2532 | { | |
2533 | return inet_ctl_sock_create(&net->ipv4.tcp_sock, | |
2534 | PF_INET, SOCK_RAW, IPPROTO_TCP, net); | |
2535 | } | |
2536 | ||
2537 | static void __net_exit tcp_sk_exit(struct net *net) | |
2538 | { | |
2539 | inet_ctl_sock_destroy(net->ipv4.tcp_sock); | |
b099ce26 EB |
2540 | } |
2541 | ||
2542 | static void __net_exit tcp_sk_exit_batch(struct list_head *net_exit_list) | |
2543 | { | |
2544 | inet_twsk_purge(&tcp_hashinfo, &tcp_death_row, AF_INET); | |
046ee902 DL |
2545 | } |
2546 | ||
2547 | static struct pernet_operations __net_initdata tcp_sk_ops = { | |
b099ce26 EB |
2548 | .init = tcp_sk_init, |
2549 | .exit = tcp_sk_exit, | |
2550 | .exit_batch = tcp_sk_exit_batch, | |
046ee902 DL |
2551 | }; |
2552 | ||
9b0f976f | 2553 | void __init tcp_v4_init(void) |
1da177e4 | 2554 | { |
5caea4ea | 2555 | inet_hashinfo_init(&tcp_hashinfo); |
6a1b3054 | 2556 | if (register_pernet_subsys(&tcp_sk_ops)) |
1da177e4 | 2557 | panic("Failed to create the TCP control socket.\n"); |
1da177e4 LT |
2558 | } |
2559 | ||
2560 | EXPORT_SYMBOL(ipv4_specific); | |
1da177e4 | 2561 | EXPORT_SYMBOL(tcp_hashinfo); |
1da177e4 | 2562 | EXPORT_SYMBOL(tcp_prot); |
1da177e4 LT |
2563 | EXPORT_SYMBOL(tcp_v4_conn_request); |
2564 | EXPORT_SYMBOL(tcp_v4_connect); | |
2565 | EXPORT_SYMBOL(tcp_v4_do_rcv); | |
1da177e4 LT |
2566 | EXPORT_SYMBOL(tcp_v4_remember_stamp); |
2567 | EXPORT_SYMBOL(tcp_v4_send_check); | |
2568 | EXPORT_SYMBOL(tcp_v4_syn_recv_sock); | |
2569 | ||
2570 | #ifdef CONFIG_PROC_FS | |
2571 | EXPORT_SYMBOL(tcp_proc_register); | |
2572 | EXPORT_SYMBOL(tcp_proc_unregister); | |
2573 | #endif | |
1da177e4 | 2574 | EXPORT_SYMBOL(sysctl_tcp_low_latency); |
1da177e4 | 2575 |