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 | ||
afd46503 | 53 | #define pr_fmt(fmt) "TCP: " fmt |
1da177e4 | 54 | |
eb4dea58 | 55 | #include <linux/bottom_half.h> |
1da177e4 LT |
56 | #include <linux/types.h> |
57 | #include <linux/fcntl.h> | |
58 | #include <linux/module.h> | |
59 | #include <linux/random.h> | |
60 | #include <linux/cache.h> | |
61 | #include <linux/jhash.h> | |
62 | #include <linux/init.h> | |
63 | #include <linux/times.h> | |
5a0e3ad6 | 64 | #include <linux/slab.h> |
1da177e4 | 65 | |
457c4cbc | 66 | #include <net/net_namespace.h> |
1da177e4 | 67 | #include <net/icmp.h> |
304a1618 | 68 | #include <net/inet_hashtables.h> |
1da177e4 | 69 | #include <net/tcp.h> |
20380731 | 70 | #include <net/transp_v6.h> |
1da177e4 LT |
71 | #include <net/ipv6.h> |
72 | #include <net/inet_common.h> | |
6d6ee43e | 73 | #include <net/timewait_sock.h> |
1da177e4 | 74 | #include <net/xfrm.h> |
1a2449a8 | 75 | #include <net/netdma.h> |
6e5714ea | 76 | #include <net/secure_seq.h> |
d1a4c0b3 | 77 | #include <net/tcp_memcontrol.h> |
076bb0c8 | 78 | #include <net/busy_poll.h> |
1da177e4 LT |
79 | |
80 | #include <linux/inet.h> | |
81 | #include <linux/ipv6.h> | |
82 | #include <linux/stddef.h> | |
83 | #include <linux/proc_fs.h> | |
84 | #include <linux/seq_file.h> | |
85 | ||
cfb6eeb4 YH |
86 | #include <linux/crypto.h> |
87 | #include <linux/scatterlist.h> | |
88 | ||
ab32ea5d BH |
89 | int sysctl_tcp_tw_reuse __read_mostly; |
90 | int sysctl_tcp_low_latency __read_mostly; | |
4bc2f18b | 91 | EXPORT_SYMBOL(sysctl_tcp_low_latency); |
1da177e4 | 92 | |
1da177e4 | 93 | |
cfb6eeb4 | 94 | #ifdef CONFIG_TCP_MD5SIG |
a915da9b | 95 | static int tcp_v4_md5_hash_hdr(char *md5_hash, const struct tcp_md5sig_key *key, |
318cf7aa | 96 | __be32 daddr, __be32 saddr, const struct tcphdr *th); |
cfb6eeb4 YH |
97 | #endif |
98 | ||
5caea4ea | 99 | struct inet_hashinfo tcp_hashinfo; |
4bc2f18b | 100 | EXPORT_SYMBOL(tcp_hashinfo); |
1da177e4 | 101 | |
936b8bdb | 102 | static __u32 tcp_v4_init_sequence(const struct sk_buff *skb) |
1da177e4 | 103 | { |
eddc9ec5 ACM |
104 | return secure_tcp_sequence_number(ip_hdr(skb)->daddr, |
105 | ip_hdr(skb)->saddr, | |
aa8223c7 ACM |
106 | tcp_hdr(skb)->dest, |
107 | tcp_hdr(skb)->source); | |
1da177e4 LT |
108 | } |
109 | ||
6d6ee43e ACM |
110 | int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp) |
111 | { | |
112 | const struct tcp_timewait_sock *tcptw = tcp_twsk(sktw); | |
113 | struct tcp_sock *tp = tcp_sk(sk); | |
114 | ||
115 | /* With PAWS, it is safe from the viewpoint | |
116 | of data integrity. Even without PAWS it is safe provided sequence | |
117 | spaces do not overlap i.e. at data rates <= 80Mbit/sec. | |
118 | ||
119 | Actually, the idea is close to VJ's one, only timestamp cache is | |
120 | held not per host, but per port pair and TW bucket is used as state | |
121 | holder. | |
122 | ||
123 | If TW bucket has been already destroyed we fall back to VJ's scheme | |
124 | and use initial timestamp retrieved from peer table. | |
125 | */ | |
126 | if (tcptw->tw_ts_recent_stamp && | |
127 | (twp == NULL || (sysctl_tcp_tw_reuse && | |
9d729f72 | 128 | get_seconds() - tcptw->tw_ts_recent_stamp > 1))) { |
6d6ee43e ACM |
129 | tp->write_seq = tcptw->tw_snd_nxt + 65535 + 2; |
130 | if (tp->write_seq == 0) | |
131 | tp->write_seq = 1; | |
132 | tp->rx_opt.ts_recent = tcptw->tw_ts_recent; | |
133 | tp->rx_opt.ts_recent_stamp = tcptw->tw_ts_recent_stamp; | |
134 | sock_hold(sktw); | |
135 | return 1; | |
136 | } | |
137 | ||
138 | return 0; | |
139 | } | |
6d6ee43e ACM |
140 | EXPORT_SYMBOL_GPL(tcp_twsk_unique); |
141 | ||
1da177e4 LT |
142 | /* This will initiate an outgoing connection. */ |
143 | int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len) | |
144 | { | |
2d7192d6 | 145 | struct sockaddr_in *usin = (struct sockaddr_in *)uaddr; |
1da177e4 LT |
146 | struct inet_sock *inet = inet_sk(sk); |
147 | struct tcp_sock *tp = tcp_sk(sk); | |
dca8b089 | 148 | __be16 orig_sport, orig_dport; |
bada8adc | 149 | __be32 daddr, nexthop; |
da905bd1 | 150 | struct flowi4 *fl4; |
2d7192d6 | 151 | struct rtable *rt; |
1da177e4 | 152 | int err; |
f6d8bd05 | 153 | struct ip_options_rcu *inet_opt; |
1da177e4 LT |
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; | |
f6d8bd05 ED |
162 | inet_opt = rcu_dereference_protected(inet->inet_opt, |
163 | sock_owned_by_user(sk)); | |
164 | if (inet_opt && inet_opt->opt.srr) { | |
1da177e4 LT |
165 | if (!daddr) |
166 | return -EINVAL; | |
f6d8bd05 | 167 | nexthop = inet_opt->opt.faddr; |
1da177e4 LT |
168 | } |
169 | ||
dca8b089 DM |
170 | orig_sport = inet->inet_sport; |
171 | orig_dport = usin->sin_port; | |
da905bd1 DM |
172 | fl4 = &inet->cork.fl.u.ip4; |
173 | rt = ip_route_connect(fl4, nexthop, inet->inet_saddr, | |
b23dd4fe DM |
174 | RT_CONN_FLAGS(sk), sk->sk_bound_dev_if, |
175 | IPPROTO_TCP, | |
0e0d44ab | 176 | orig_sport, orig_dport, sk); |
b23dd4fe DM |
177 | if (IS_ERR(rt)) { |
178 | err = PTR_ERR(rt); | |
179 | if (err == -ENETUNREACH) | |
f1d8cba6 | 180 | IP_INC_STATS(sock_net(sk), IPSTATS_MIB_OUTNOROUTES); |
b23dd4fe | 181 | return err; |
584bdf8c | 182 | } |
1da177e4 LT |
183 | |
184 | if (rt->rt_flags & (RTCF_MULTICAST | RTCF_BROADCAST)) { | |
185 | ip_rt_put(rt); | |
186 | return -ENETUNREACH; | |
187 | } | |
188 | ||
f6d8bd05 | 189 | if (!inet_opt || !inet_opt->opt.srr) |
da905bd1 | 190 | daddr = fl4->daddr; |
1da177e4 | 191 | |
c720c7e8 | 192 | if (!inet->inet_saddr) |
da905bd1 | 193 | inet->inet_saddr = fl4->saddr; |
c720c7e8 | 194 | inet->inet_rcv_saddr = inet->inet_saddr; |
1da177e4 | 195 | |
c720c7e8 | 196 | if (tp->rx_opt.ts_recent_stamp && inet->inet_daddr != daddr) { |
1da177e4 LT |
197 | /* Reset inherited state */ |
198 | tp->rx_opt.ts_recent = 0; | |
199 | tp->rx_opt.ts_recent_stamp = 0; | |
ee995283 PE |
200 | if (likely(!tp->repair)) |
201 | tp->write_seq = 0; | |
1da177e4 LT |
202 | } |
203 | ||
295ff7ed | 204 | if (tcp_death_row.sysctl_tw_recycle && |
81166dd6 DM |
205 | !tp->rx_opt.ts_recent_stamp && fl4->daddr == daddr) |
206 | tcp_fetch_timewait_stamp(sk, &rt->dst); | |
1da177e4 | 207 | |
c720c7e8 ED |
208 | inet->inet_dport = usin->sin_port; |
209 | inet->inet_daddr = daddr; | |
1da177e4 | 210 | |
d83d8461 | 211 | inet_csk(sk)->icsk_ext_hdr_len = 0; |
f6d8bd05 ED |
212 | if (inet_opt) |
213 | inet_csk(sk)->icsk_ext_hdr_len = inet_opt->opt.optlen; | |
1da177e4 | 214 | |
bee7ca9e | 215 | tp->rx_opt.mss_clamp = TCP_MSS_DEFAULT; |
1da177e4 LT |
216 | |
217 | /* Socket identity is still unknown (sport may be zero). | |
218 | * However we set state to SYN-SENT and not releasing socket | |
219 | * lock select source port, enter ourselves into the hash tables and | |
220 | * complete initialization after this. | |
221 | */ | |
222 | tcp_set_state(sk, TCP_SYN_SENT); | |
a7f5e7f1 | 223 | err = inet_hash_connect(&tcp_death_row, sk); |
1da177e4 LT |
224 | if (err) |
225 | goto failure; | |
226 | ||
da905bd1 | 227 | rt = ip_route_newports(fl4, rt, orig_sport, orig_dport, |
b23dd4fe DM |
228 | inet->inet_sport, inet->inet_dport, sk); |
229 | if (IS_ERR(rt)) { | |
230 | err = PTR_ERR(rt); | |
231 | rt = NULL; | |
1da177e4 | 232 | goto failure; |
b23dd4fe | 233 | } |
1da177e4 | 234 | /* OK, now commit destination to socket. */ |
bcd76111 | 235 | sk->sk_gso_type = SKB_GSO_TCPV4; |
d8d1f30b | 236 | sk_setup_caps(sk, &rt->dst); |
1da177e4 | 237 | |
ee995283 | 238 | if (!tp->write_seq && likely(!tp->repair)) |
c720c7e8 ED |
239 | tp->write_seq = secure_tcp_sequence_number(inet->inet_saddr, |
240 | inet->inet_daddr, | |
241 | inet->inet_sport, | |
1da177e4 LT |
242 | usin->sin_port); |
243 | ||
c720c7e8 | 244 | inet->inet_id = tp->write_seq ^ jiffies; |
1da177e4 | 245 | |
2b916477 | 246 | err = tcp_connect(sk); |
ee995283 | 247 | |
1da177e4 LT |
248 | rt = NULL; |
249 | if (err) | |
250 | goto failure; | |
251 | ||
252 | return 0; | |
253 | ||
254 | failure: | |
7174259e ACM |
255 | /* |
256 | * This unhashes the socket and releases the local port, | |
257 | * if necessary. | |
258 | */ | |
1da177e4 LT |
259 | tcp_set_state(sk, TCP_CLOSE); |
260 | ip_rt_put(rt); | |
261 | sk->sk_route_caps = 0; | |
c720c7e8 | 262 | inet->inet_dport = 0; |
1da177e4 LT |
263 | return err; |
264 | } | |
4bc2f18b | 265 | EXPORT_SYMBOL(tcp_v4_connect); |
1da177e4 | 266 | |
1da177e4 | 267 | /* |
563d34d0 ED |
268 | * This routine reacts to ICMP_FRAG_NEEDED mtu indications as defined in RFC1191. |
269 | * It can be called through tcp_release_cb() if socket was owned by user | |
270 | * at the time tcp_v4_err() was called to handle ICMP message. | |
1da177e4 | 271 | */ |
563d34d0 | 272 | static void tcp_v4_mtu_reduced(struct sock *sk) |
1da177e4 LT |
273 | { |
274 | struct dst_entry *dst; | |
275 | struct inet_sock *inet = inet_sk(sk); | |
563d34d0 | 276 | u32 mtu = tcp_sk(sk)->mtu_info; |
1da177e4 | 277 | |
80d0a69f DM |
278 | dst = inet_csk_update_pmtu(sk, mtu); |
279 | if (!dst) | |
1da177e4 LT |
280 | return; |
281 | ||
1da177e4 LT |
282 | /* Something is about to be wrong... Remember soft error |
283 | * for the case, if this connection will not able to recover. | |
284 | */ | |
285 | if (mtu < dst_mtu(dst) && ip_dont_fragment(sk, dst)) | |
286 | sk->sk_err_soft = EMSGSIZE; | |
287 | ||
288 | mtu = dst_mtu(dst); | |
289 | ||
290 | if (inet->pmtudisc != IP_PMTUDISC_DONT && | |
482fc609 | 291 | ip_sk_accept_pmtu(sk) && |
d83d8461 | 292 | inet_csk(sk)->icsk_pmtu_cookie > mtu) { |
1da177e4 LT |
293 | tcp_sync_mss(sk, mtu); |
294 | ||
295 | /* Resend the TCP packet because it's | |
296 | * clear that the old packet has been | |
297 | * dropped. This is the new "fast" path mtu | |
298 | * discovery. | |
299 | */ | |
300 | tcp_simple_retransmit(sk); | |
301 | } /* else let the usual retransmit timer handle it */ | |
302 | } | |
303 | ||
55be7a9c DM |
304 | static void do_redirect(struct sk_buff *skb, struct sock *sk) |
305 | { | |
306 | struct dst_entry *dst = __sk_dst_check(sk, 0); | |
307 | ||
1ed5c48f | 308 | if (dst) |
6700c270 | 309 | dst->ops->redirect(dst, sk, skb); |
55be7a9c DM |
310 | } |
311 | ||
1da177e4 LT |
312 | /* |
313 | * This routine is called by the ICMP module when it gets some | |
314 | * sort of error condition. If err < 0 then the socket should | |
315 | * be closed and the error returned to the user. If err > 0 | |
316 | * it's just the icmp type << 8 | icmp code. After adjustment | |
317 | * header points to the first 8 bytes of the tcp header. We need | |
318 | * to find the appropriate port. | |
319 | * | |
320 | * The locking strategy used here is very "optimistic". When | |
321 | * someone else accesses the socket the ICMP is just dropped | |
322 | * and for some paths there is no check at all. | |
323 | * A more general error queue to queue errors for later handling | |
324 | * is probably better. | |
325 | * | |
326 | */ | |
327 | ||
4d1a2d9e | 328 | void tcp_v4_err(struct sk_buff *icmp_skb, u32 info) |
1da177e4 | 329 | { |
b71d1d42 | 330 | const struct iphdr *iph = (const struct iphdr *)icmp_skb->data; |
4d1a2d9e | 331 | struct tcphdr *th = (struct tcphdr *)(icmp_skb->data + (iph->ihl << 2)); |
f1ecd5d9 | 332 | struct inet_connection_sock *icsk; |
1da177e4 LT |
333 | struct tcp_sock *tp; |
334 | struct inet_sock *inet; | |
4d1a2d9e DL |
335 | const int type = icmp_hdr(icmp_skb)->type; |
336 | const int code = icmp_hdr(icmp_skb)->code; | |
1da177e4 | 337 | struct sock *sk; |
f1ecd5d9 | 338 | struct sk_buff *skb; |
0a672f74 YC |
339 | struct request_sock *fastopen; |
340 | __u32 seq, snd_una; | |
f1ecd5d9 | 341 | __u32 remaining; |
1da177e4 | 342 | int err; |
4d1a2d9e | 343 | struct net *net = dev_net(icmp_skb->dev); |
1da177e4 | 344 | |
4d1a2d9e | 345 | if (icmp_skb->len < (iph->ihl << 2) + 8) { |
dcfc23ca | 346 | ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS); |
1da177e4 LT |
347 | return; |
348 | } | |
349 | ||
fd54d716 | 350 | sk = inet_lookup(net, &tcp_hashinfo, iph->daddr, th->dest, |
4d1a2d9e | 351 | iph->saddr, th->source, inet_iif(icmp_skb)); |
1da177e4 | 352 | if (!sk) { |
dcfc23ca | 353 | ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS); |
1da177e4 LT |
354 | return; |
355 | } | |
356 | if (sk->sk_state == TCP_TIME_WAIT) { | |
9469c7b4 | 357 | inet_twsk_put(inet_twsk(sk)); |
1da177e4 LT |
358 | return; |
359 | } | |
360 | ||
361 | bh_lock_sock(sk); | |
362 | /* If too many ICMPs get dropped on busy | |
363 | * servers this needs to be solved differently. | |
563d34d0 ED |
364 | * We do take care of PMTU discovery (RFC1191) special case : |
365 | * we can receive locally generated ICMP messages while socket is held. | |
1da177e4 | 366 | */ |
b74aa930 ED |
367 | if (sock_owned_by_user(sk)) { |
368 | if (!(type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED)) | |
369 | NET_INC_STATS_BH(net, LINUX_MIB_LOCKDROPPEDICMPS); | |
370 | } | |
1da177e4 LT |
371 | if (sk->sk_state == TCP_CLOSE) |
372 | goto out; | |
373 | ||
97e3ecd1 | 374 | if (unlikely(iph->ttl < inet_sk(sk)->min_ttl)) { |
375 | NET_INC_STATS_BH(net, LINUX_MIB_TCPMINTTLDROP); | |
376 | goto out; | |
377 | } | |
378 | ||
f1ecd5d9 | 379 | icsk = inet_csk(sk); |
1da177e4 LT |
380 | tp = tcp_sk(sk); |
381 | seq = ntohl(th->seq); | |
0a672f74 YC |
382 | /* XXX (TFO) - tp->snd_una should be ISN (tcp_create_openreq_child() */ |
383 | fastopen = tp->fastopen_rsk; | |
384 | snd_una = fastopen ? tcp_rsk(fastopen)->snt_isn : tp->snd_una; | |
1da177e4 | 385 | if (sk->sk_state != TCP_LISTEN && |
0a672f74 | 386 | !between(seq, snd_una, tp->snd_nxt)) { |
de0744af | 387 | NET_INC_STATS_BH(net, LINUX_MIB_OUTOFWINDOWICMPS); |
1da177e4 LT |
388 | goto out; |
389 | } | |
390 | ||
391 | switch (type) { | |
55be7a9c DM |
392 | case ICMP_REDIRECT: |
393 | do_redirect(icmp_skb, sk); | |
394 | goto out; | |
1da177e4 LT |
395 | case ICMP_SOURCE_QUENCH: |
396 | /* Just silently ignore these. */ | |
397 | goto out; | |
398 | case ICMP_PARAMETERPROB: | |
399 | err = EPROTO; | |
400 | break; | |
401 | case ICMP_DEST_UNREACH: | |
402 | if (code > NR_ICMP_UNREACH) | |
403 | goto out; | |
404 | ||
405 | if (code == ICMP_FRAG_NEEDED) { /* PMTU discovery (RFC1191) */ | |
0d4f0608 ED |
406 | /* We are not interested in TCP_LISTEN and open_requests |
407 | * (SYN-ACKs send out by Linux are always <576bytes so | |
408 | * they should go through unfragmented). | |
409 | */ | |
410 | if (sk->sk_state == TCP_LISTEN) | |
411 | goto out; | |
412 | ||
563d34d0 | 413 | tp->mtu_info = info; |
144d56e9 | 414 | if (!sock_owned_by_user(sk)) { |
563d34d0 | 415 | tcp_v4_mtu_reduced(sk); |
144d56e9 ED |
416 | } else { |
417 | if (!test_and_set_bit(TCP_MTU_REDUCED_DEFERRED, &tp->tsq_flags)) | |
418 | sock_hold(sk); | |
419 | } | |
1da177e4 LT |
420 | goto out; |
421 | } | |
422 | ||
423 | err = icmp_err_convert[code].errno; | |
f1ecd5d9 DL |
424 | /* check if icmp_skb allows revert of backoff |
425 | * (see draft-zimmermann-tcp-lcd) */ | |
426 | if (code != ICMP_NET_UNREACH && code != ICMP_HOST_UNREACH) | |
427 | break; | |
428 | if (seq != tp->snd_una || !icsk->icsk_retransmits || | |
0a672f74 | 429 | !icsk->icsk_backoff || fastopen) |
f1ecd5d9 DL |
430 | break; |
431 | ||
8f49c270 DM |
432 | if (sock_owned_by_user(sk)) |
433 | break; | |
434 | ||
f1ecd5d9 | 435 | icsk->icsk_backoff--; |
740b0f18 | 436 | inet_csk(sk)->icsk_rto = (tp->srtt_us ? __tcp_set_rto(tp) : |
9ad7c049 | 437 | TCP_TIMEOUT_INIT) << icsk->icsk_backoff; |
f1ecd5d9 DL |
438 | tcp_bound_rto(sk); |
439 | ||
440 | skb = tcp_write_queue_head(sk); | |
441 | BUG_ON(!skb); | |
442 | ||
443 | remaining = icsk->icsk_rto - min(icsk->icsk_rto, | |
444 | tcp_time_stamp - TCP_SKB_CB(skb)->when); | |
445 | ||
446 | if (remaining) { | |
447 | inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS, | |
448 | remaining, TCP_RTO_MAX); | |
f1ecd5d9 DL |
449 | } else { |
450 | /* RTO revert clocked out retransmission. | |
451 | * Will retransmit now */ | |
452 | tcp_retransmit_timer(sk); | |
453 | } | |
454 | ||
1da177e4 LT |
455 | break; |
456 | case ICMP_TIME_EXCEEDED: | |
457 | err = EHOSTUNREACH; | |
458 | break; | |
459 | default: | |
460 | goto out; | |
461 | } | |
462 | ||
463 | switch (sk->sk_state) { | |
60236fdd | 464 | struct request_sock *req, **prev; |
1da177e4 LT |
465 | case TCP_LISTEN: |
466 | if (sock_owned_by_user(sk)) | |
467 | goto out; | |
468 | ||
463c84b9 ACM |
469 | req = inet_csk_search_req(sk, &prev, th->dest, |
470 | iph->daddr, iph->saddr); | |
1da177e4 LT |
471 | if (!req) |
472 | goto out; | |
473 | ||
474 | /* ICMPs are not backlogged, hence we cannot get | |
475 | an established socket here. | |
476 | */ | |
547b792c | 477 | WARN_ON(req->sk); |
1da177e4 | 478 | |
2e6599cb | 479 | if (seq != tcp_rsk(req)->snt_isn) { |
de0744af | 480 | NET_INC_STATS_BH(net, LINUX_MIB_OUTOFWINDOWICMPS); |
1da177e4 LT |
481 | goto out; |
482 | } | |
483 | ||
484 | /* | |
485 | * Still in SYN_RECV, just remove it silently. | |
486 | * There is no good way to pass the error to the newly | |
487 | * created socket, and POSIX does not want network | |
488 | * errors returned from accept(). | |
489 | */ | |
463c84b9 | 490 | inet_csk_reqsk_queue_drop(sk, req, prev); |
848bf15f | 491 | NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENDROPS); |
1da177e4 LT |
492 | goto out; |
493 | ||
494 | case TCP_SYN_SENT: | |
0a672f74 YC |
495 | case TCP_SYN_RECV: |
496 | /* Only in fast or simultaneous open. If a fast open socket is | |
497 | * is already accepted it is treated as a connected one below. | |
498 | */ | |
499 | if (fastopen && fastopen->sk == NULL) | |
500 | break; | |
501 | ||
1da177e4 | 502 | if (!sock_owned_by_user(sk)) { |
1da177e4 LT |
503 | sk->sk_err = err; |
504 | ||
505 | sk->sk_error_report(sk); | |
506 | ||
507 | tcp_done(sk); | |
508 | } else { | |
509 | sk->sk_err_soft = err; | |
510 | } | |
511 | goto out; | |
512 | } | |
513 | ||
514 | /* If we've already connected we will keep trying | |
515 | * until we time out, or the user gives up. | |
516 | * | |
517 | * rfc1122 4.2.3.9 allows to consider as hard errors | |
518 | * only PROTO_UNREACH and PORT_UNREACH (well, FRAG_FAILED too, | |
519 | * but it is obsoleted by pmtu discovery). | |
520 | * | |
521 | * Note, that in modern internet, where routing is unreliable | |
522 | * and in each dark corner broken firewalls sit, sending random | |
523 | * errors ordered by their masters even this two messages finally lose | |
524 | * their original sense (even Linux sends invalid PORT_UNREACHs) | |
525 | * | |
526 | * Now we are in compliance with RFCs. | |
527 | * --ANK (980905) | |
528 | */ | |
529 | ||
530 | inet = inet_sk(sk); | |
531 | if (!sock_owned_by_user(sk) && inet->recverr) { | |
532 | sk->sk_err = err; | |
533 | sk->sk_error_report(sk); | |
534 | } else { /* Only an error on timeout */ | |
535 | sk->sk_err_soft = err; | |
536 | } | |
537 | ||
538 | out: | |
539 | bh_unlock_sock(sk); | |
540 | sock_put(sk); | |
541 | } | |
542 | ||
28850dc7 | 543 | void __tcp_v4_send_check(struct sk_buff *skb, __be32 saddr, __be32 daddr) |
1da177e4 | 544 | { |
aa8223c7 | 545 | struct tcphdr *th = tcp_hdr(skb); |
1da177e4 | 546 | |
84fa7933 | 547 | if (skb->ip_summed == CHECKSUM_PARTIAL) { |
419f9f89 | 548 | th->check = ~tcp_v4_check(skb->len, saddr, daddr, 0); |
663ead3b | 549 | skb->csum_start = skb_transport_header(skb) - skb->head; |
ff1dcadb | 550 | skb->csum_offset = offsetof(struct tcphdr, check); |
1da177e4 | 551 | } else { |
419f9f89 | 552 | th->check = tcp_v4_check(skb->len, saddr, daddr, |
07f0757a | 553 | csum_partial(th, |
1da177e4 LT |
554 | th->doff << 2, |
555 | skb->csum)); | |
556 | } | |
557 | } | |
558 | ||
419f9f89 | 559 | /* This routine computes an IPv4 TCP checksum. */ |
bb296246 | 560 | void tcp_v4_send_check(struct sock *sk, struct sk_buff *skb) |
419f9f89 | 561 | { |
cf533ea5 | 562 | const struct inet_sock *inet = inet_sk(sk); |
419f9f89 HX |
563 | |
564 | __tcp_v4_send_check(skb, inet->inet_saddr, inet->inet_daddr); | |
565 | } | |
4bc2f18b | 566 | EXPORT_SYMBOL(tcp_v4_send_check); |
419f9f89 | 567 | |
1da177e4 LT |
568 | /* |
569 | * This routine will send an RST to the other tcp. | |
570 | * | |
571 | * Someone asks: why I NEVER use socket parameters (TOS, TTL etc.) | |
572 | * for reset. | |
573 | * Answer: if a packet caused RST, it is not for a socket | |
574 | * existing in our system, if it is matched to a socket, | |
575 | * it is just duplicate segment or bug in other side's TCP. | |
576 | * So that we build reply only basing on parameters | |
577 | * arrived with segment. | |
578 | * Exception: precedence violation. We do not implement it in any case. | |
579 | */ | |
580 | ||
cfb6eeb4 | 581 | static void tcp_v4_send_reset(struct sock *sk, struct sk_buff *skb) |
1da177e4 | 582 | { |
cf533ea5 | 583 | const struct tcphdr *th = tcp_hdr(skb); |
cfb6eeb4 YH |
584 | struct { |
585 | struct tcphdr th; | |
586 | #ifdef CONFIG_TCP_MD5SIG | |
714e85be | 587 | __be32 opt[(TCPOLEN_MD5SIG_ALIGNED >> 2)]; |
cfb6eeb4 YH |
588 | #endif |
589 | } rep; | |
1da177e4 | 590 | struct ip_reply_arg arg; |
cfb6eeb4 YH |
591 | #ifdef CONFIG_TCP_MD5SIG |
592 | struct tcp_md5sig_key *key; | |
658ddaaf SL |
593 | const __u8 *hash_location = NULL; |
594 | unsigned char newhash[16]; | |
595 | int genhash; | |
596 | struct sock *sk1 = NULL; | |
cfb6eeb4 | 597 | #endif |
a86b1e30 | 598 | struct net *net; |
1da177e4 LT |
599 | |
600 | /* Never send a reset in response to a reset. */ | |
601 | if (th->rst) | |
602 | return; | |
603 | ||
511c3f92 | 604 | if (skb_rtable(skb)->rt_type != RTN_LOCAL) |
1da177e4 LT |
605 | return; |
606 | ||
607 | /* Swap the send and the receive. */ | |
cfb6eeb4 YH |
608 | memset(&rep, 0, sizeof(rep)); |
609 | rep.th.dest = th->source; | |
610 | rep.th.source = th->dest; | |
611 | rep.th.doff = sizeof(struct tcphdr) / 4; | |
612 | rep.th.rst = 1; | |
1da177e4 LT |
613 | |
614 | if (th->ack) { | |
cfb6eeb4 | 615 | rep.th.seq = th->ack_seq; |
1da177e4 | 616 | } else { |
cfb6eeb4 YH |
617 | rep.th.ack = 1; |
618 | rep.th.ack_seq = htonl(ntohl(th->seq) + th->syn + th->fin + | |
619 | skb->len - (th->doff << 2)); | |
1da177e4 LT |
620 | } |
621 | ||
7174259e | 622 | memset(&arg, 0, sizeof(arg)); |
cfb6eeb4 YH |
623 | arg.iov[0].iov_base = (unsigned char *)&rep; |
624 | arg.iov[0].iov_len = sizeof(rep.th); | |
625 | ||
626 | #ifdef CONFIG_TCP_MD5SIG | |
658ddaaf SL |
627 | hash_location = tcp_parse_md5sig_option(th); |
628 | if (!sk && hash_location) { | |
629 | /* | |
630 | * active side is lost. Try to find listening socket through | |
631 | * source port, and then find md5 key through listening socket. | |
632 | * we are not loose security here: | |
633 | * Incoming packet is checked with md5 hash with finding key, | |
634 | * no RST generated if md5 hash doesn't match. | |
635 | */ | |
636 | sk1 = __inet_lookup_listener(dev_net(skb_dst(skb)->dev), | |
da5e3630 TH |
637 | &tcp_hashinfo, ip_hdr(skb)->saddr, |
638 | th->source, ip_hdr(skb)->daddr, | |
658ddaaf SL |
639 | ntohs(th->source), inet_iif(skb)); |
640 | /* don't send rst if it can't find key */ | |
641 | if (!sk1) | |
642 | return; | |
643 | rcu_read_lock(); | |
644 | key = tcp_md5_do_lookup(sk1, (union tcp_md5_addr *) | |
645 | &ip_hdr(skb)->saddr, AF_INET); | |
646 | if (!key) | |
647 | goto release_sk1; | |
648 | ||
649 | genhash = tcp_v4_md5_hash_skb(newhash, key, NULL, NULL, skb); | |
650 | if (genhash || memcmp(hash_location, newhash, 16) != 0) | |
651 | goto release_sk1; | |
652 | } else { | |
653 | key = sk ? tcp_md5_do_lookup(sk, (union tcp_md5_addr *) | |
654 | &ip_hdr(skb)->saddr, | |
655 | AF_INET) : NULL; | |
656 | } | |
657 | ||
cfb6eeb4 YH |
658 | if (key) { |
659 | rep.opt[0] = htonl((TCPOPT_NOP << 24) | | |
660 | (TCPOPT_NOP << 16) | | |
661 | (TCPOPT_MD5SIG << 8) | | |
662 | TCPOLEN_MD5SIG); | |
663 | /* Update length and the length the header thinks exists */ | |
664 | arg.iov[0].iov_len += TCPOLEN_MD5SIG_ALIGNED; | |
665 | rep.th.doff = arg.iov[0].iov_len / 4; | |
666 | ||
49a72dfb | 667 | tcp_v4_md5_hash_hdr((__u8 *) &rep.opt[1], |
78e645cb IJ |
668 | key, ip_hdr(skb)->saddr, |
669 | ip_hdr(skb)->daddr, &rep.th); | |
cfb6eeb4 YH |
670 | } |
671 | #endif | |
eddc9ec5 ACM |
672 | arg.csum = csum_tcpudp_nofold(ip_hdr(skb)->daddr, |
673 | ip_hdr(skb)->saddr, /* XXX */ | |
52cd5750 | 674 | arg.iov[0].iov_len, IPPROTO_TCP, 0); |
1da177e4 | 675 | arg.csumoffset = offsetof(struct tcphdr, check) / 2; |
88ef4a5a | 676 | arg.flags = (sk && inet_sk(sk)->transparent) ? IP_REPLY_ARG_NOSRCCHECK : 0; |
e2446eaa | 677 | /* When socket is gone, all binding information is lost. |
4c675258 AK |
678 | * routing might fail in this case. No choice here, if we choose to force |
679 | * input interface, we will misroute in case of asymmetric route. | |
e2446eaa | 680 | */ |
4c675258 AK |
681 | if (sk) |
682 | arg.bound_dev_if = sk->sk_bound_dev_if; | |
1da177e4 | 683 | |
adf30907 | 684 | net = dev_net(skb_dst(skb)->dev); |
66b13d99 | 685 | arg.tos = ip_hdr(skb)->tos; |
be9f4a44 | 686 | ip_send_unicast_reply(net, skb, ip_hdr(skb)->saddr, |
70e73416 | 687 | ip_hdr(skb)->daddr, &arg, arg.iov[0].iov_len); |
1da177e4 | 688 | |
63231bdd PE |
689 | TCP_INC_STATS_BH(net, TCP_MIB_OUTSEGS); |
690 | TCP_INC_STATS_BH(net, TCP_MIB_OUTRSTS); | |
658ddaaf SL |
691 | |
692 | #ifdef CONFIG_TCP_MD5SIG | |
693 | release_sk1: | |
694 | if (sk1) { | |
695 | rcu_read_unlock(); | |
696 | sock_put(sk1); | |
697 | } | |
698 | #endif | |
1da177e4 LT |
699 | } |
700 | ||
701 | /* The code following below sending ACKs in SYN-RECV and TIME-WAIT states | |
702 | outside socket context is ugly, certainly. What can I do? | |
703 | */ | |
704 | ||
9501f972 | 705 | static void tcp_v4_send_ack(struct sk_buff *skb, u32 seq, u32 ack, |
ee684b6f | 706 | u32 win, u32 tsval, u32 tsecr, int oif, |
88ef4a5a | 707 | struct tcp_md5sig_key *key, |
66b13d99 | 708 | int reply_flags, u8 tos) |
1da177e4 | 709 | { |
cf533ea5 | 710 | const struct tcphdr *th = tcp_hdr(skb); |
1da177e4 LT |
711 | struct { |
712 | struct tcphdr th; | |
714e85be | 713 | __be32 opt[(TCPOLEN_TSTAMP_ALIGNED >> 2) |
cfb6eeb4 | 714 | #ifdef CONFIG_TCP_MD5SIG |
714e85be | 715 | + (TCPOLEN_MD5SIG_ALIGNED >> 2) |
cfb6eeb4 YH |
716 | #endif |
717 | ]; | |
1da177e4 LT |
718 | } rep; |
719 | struct ip_reply_arg arg; | |
adf30907 | 720 | struct net *net = dev_net(skb_dst(skb)->dev); |
1da177e4 LT |
721 | |
722 | memset(&rep.th, 0, sizeof(struct tcphdr)); | |
7174259e | 723 | memset(&arg, 0, sizeof(arg)); |
1da177e4 LT |
724 | |
725 | arg.iov[0].iov_base = (unsigned char *)&rep; | |
726 | arg.iov[0].iov_len = sizeof(rep.th); | |
ee684b6f | 727 | if (tsecr) { |
cfb6eeb4 YH |
728 | rep.opt[0] = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) | |
729 | (TCPOPT_TIMESTAMP << 8) | | |
730 | TCPOLEN_TIMESTAMP); | |
ee684b6f AV |
731 | rep.opt[1] = htonl(tsval); |
732 | rep.opt[2] = htonl(tsecr); | |
cb48cfe8 | 733 | arg.iov[0].iov_len += TCPOLEN_TSTAMP_ALIGNED; |
1da177e4 LT |
734 | } |
735 | ||
736 | /* Swap the send and the receive. */ | |
737 | rep.th.dest = th->source; | |
738 | rep.th.source = th->dest; | |
739 | rep.th.doff = arg.iov[0].iov_len / 4; | |
740 | rep.th.seq = htonl(seq); | |
741 | rep.th.ack_seq = htonl(ack); | |
742 | rep.th.ack = 1; | |
743 | rep.th.window = htons(win); | |
744 | ||
cfb6eeb4 | 745 | #ifdef CONFIG_TCP_MD5SIG |
cfb6eeb4 | 746 | if (key) { |
ee684b6f | 747 | int offset = (tsecr) ? 3 : 0; |
cfb6eeb4 YH |
748 | |
749 | rep.opt[offset++] = htonl((TCPOPT_NOP << 24) | | |
750 | (TCPOPT_NOP << 16) | | |
751 | (TCPOPT_MD5SIG << 8) | | |
752 | TCPOLEN_MD5SIG); | |
753 | arg.iov[0].iov_len += TCPOLEN_MD5SIG_ALIGNED; | |
754 | rep.th.doff = arg.iov[0].iov_len/4; | |
755 | ||
49a72dfb | 756 | tcp_v4_md5_hash_hdr((__u8 *) &rep.opt[offset], |
90b7e112 AL |
757 | key, ip_hdr(skb)->saddr, |
758 | ip_hdr(skb)->daddr, &rep.th); | |
cfb6eeb4 YH |
759 | } |
760 | #endif | |
88ef4a5a | 761 | arg.flags = reply_flags; |
eddc9ec5 ACM |
762 | arg.csum = csum_tcpudp_nofold(ip_hdr(skb)->daddr, |
763 | ip_hdr(skb)->saddr, /* XXX */ | |
1da177e4 LT |
764 | arg.iov[0].iov_len, IPPROTO_TCP, 0); |
765 | arg.csumoffset = offsetof(struct tcphdr, check) / 2; | |
9501f972 YH |
766 | if (oif) |
767 | arg.bound_dev_if = oif; | |
66b13d99 | 768 | arg.tos = tos; |
be9f4a44 | 769 | ip_send_unicast_reply(net, skb, ip_hdr(skb)->saddr, |
70e73416 | 770 | ip_hdr(skb)->daddr, &arg, arg.iov[0].iov_len); |
1da177e4 | 771 | |
63231bdd | 772 | TCP_INC_STATS_BH(net, TCP_MIB_OUTSEGS); |
1da177e4 LT |
773 | } |
774 | ||
775 | static void tcp_v4_timewait_ack(struct sock *sk, struct sk_buff *skb) | |
776 | { | |
8feaf0c0 | 777 | struct inet_timewait_sock *tw = inet_twsk(sk); |
cfb6eeb4 | 778 | struct tcp_timewait_sock *tcptw = tcp_twsk(sk); |
1da177e4 | 779 | |
9501f972 | 780 | tcp_v4_send_ack(skb, tcptw->tw_snd_nxt, tcptw->tw_rcv_nxt, |
7174259e | 781 | tcptw->tw_rcv_wnd >> tw->tw_rcv_wscale, |
ee684b6f | 782 | tcp_time_stamp + tcptw->tw_ts_offset, |
9501f972 YH |
783 | tcptw->tw_ts_recent, |
784 | tw->tw_bound_dev_if, | |
88ef4a5a | 785 | tcp_twsk_md5_key(tcptw), |
66b13d99 ED |
786 | tw->tw_transparent ? IP_REPLY_ARG_NOSRCCHECK : 0, |
787 | tw->tw_tos | |
9501f972 | 788 | ); |
1da177e4 | 789 | |
8feaf0c0 | 790 | inet_twsk_put(tw); |
1da177e4 LT |
791 | } |
792 | ||
6edafaaf | 793 | static void tcp_v4_reqsk_send_ack(struct sock *sk, struct sk_buff *skb, |
7174259e | 794 | struct request_sock *req) |
1da177e4 | 795 | { |
168a8f58 JC |
796 | /* sk->sk_state == TCP_LISTEN -> for regular TCP_SYN_RECV |
797 | * sk->sk_state == TCP_SYN_RECV -> for Fast Open. | |
798 | */ | |
799 | tcp_v4_send_ack(skb, (sk->sk_state == TCP_LISTEN) ? | |
800 | tcp_rsk(req)->snt_isn + 1 : tcp_sk(sk)->snd_nxt, | |
801 | tcp_rsk(req)->rcv_nxt, req->rcv_wnd, | |
ee684b6f | 802 | tcp_time_stamp, |
9501f972 YH |
803 | req->ts_recent, |
804 | 0, | |
a915da9b ED |
805 | tcp_md5_do_lookup(sk, (union tcp_md5_addr *)&ip_hdr(skb)->daddr, |
806 | AF_INET), | |
66b13d99 ED |
807 | inet_rsk(req)->no_srccheck ? IP_REPLY_ARG_NOSRCCHECK : 0, |
808 | ip_hdr(skb)->tos); | |
1da177e4 LT |
809 | } |
810 | ||
1da177e4 | 811 | /* |
9bf1d83e | 812 | * Send a SYN-ACK after having received a SYN. |
60236fdd | 813 | * This still operates on a request_sock only, not on a big |
1da177e4 LT |
814 | * socket. |
815 | */ | |
72659ecc | 816 | static int tcp_v4_send_synack(struct sock *sk, struct dst_entry *dst, |
d6274bd8 | 817 | struct flowi *fl, |
72659ecc | 818 | struct request_sock *req, |
843f4a55 YC |
819 | u16 queue_mapping, |
820 | struct tcp_fastopen_cookie *foc) | |
1da177e4 | 821 | { |
2e6599cb | 822 | const struct inet_request_sock *ireq = inet_rsk(req); |
6bd023f3 | 823 | struct flowi4 fl4; |
1da177e4 | 824 | int err = -1; |
d41db5af | 825 | struct sk_buff *skb; |
1da177e4 LT |
826 | |
827 | /* First, grab a route. */ | |
ba3f7f04 | 828 | if (!dst && (dst = inet_csk_route_req(sk, &fl4, req)) == NULL) |
fd80eb94 | 829 | return -1; |
1da177e4 | 830 | |
843f4a55 | 831 | skb = tcp_make_synack(sk, dst, req, foc); |
1da177e4 LT |
832 | |
833 | if (skb) { | |
634fb979 | 834 | __tcp_v4_send_check(skb, ireq->ir_loc_addr, ireq->ir_rmt_addr); |
1da177e4 | 835 | |
fff32699 | 836 | skb_set_queue_mapping(skb, queue_mapping); |
634fb979 ED |
837 | err = ip_build_and_send_pkt(skb, sk, ireq->ir_loc_addr, |
838 | ireq->ir_rmt_addr, | |
2e6599cb | 839 | ireq->opt); |
b9df3cb8 | 840 | err = net_xmit_eval(err); |
016818d0 NC |
841 | if (!tcp_rsk(req)->snt_synack && !err) |
842 | tcp_rsk(req)->snt_synack = tcp_time_stamp; | |
1da177e4 LT |
843 | } |
844 | ||
1da177e4 LT |
845 | return err; |
846 | } | |
847 | ||
848 | /* | |
60236fdd | 849 | * IPv4 request_sock destructor. |
1da177e4 | 850 | */ |
60236fdd | 851 | static void tcp_v4_reqsk_destructor(struct request_sock *req) |
1da177e4 | 852 | { |
a51482bd | 853 | kfree(inet_rsk(req)->opt); |
1da177e4 LT |
854 | } |
855 | ||
946cedcc | 856 | /* |
a2a385d6 | 857 | * Return true if a syncookie should be sent |
946cedcc | 858 | */ |
a2a385d6 | 859 | bool tcp_syn_flood_action(struct sock *sk, |
946cedcc ED |
860 | const struct sk_buff *skb, |
861 | const char *proto) | |
1da177e4 | 862 | { |
946cedcc | 863 | const char *msg = "Dropping request"; |
a2a385d6 | 864 | bool want_cookie = false; |
946cedcc ED |
865 | struct listen_sock *lopt; |
866 | ||
2a1d4bd4 | 867 | #ifdef CONFIG_SYN_COOKIES |
946cedcc | 868 | if (sysctl_tcp_syncookies) { |
2a1d4bd4 | 869 | msg = "Sending cookies"; |
a2a385d6 | 870 | want_cookie = true; |
946cedcc ED |
871 | NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPREQQFULLDOCOOKIES); |
872 | } else | |
80e40daa | 873 | #endif |
946cedcc ED |
874 | NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPREQQFULLDROP); |
875 | ||
876 | lopt = inet_csk(sk)->icsk_accept_queue.listen_opt; | |
5ad37d5d | 877 | if (!lopt->synflood_warned && sysctl_tcp_syncookies != 2) { |
946cedcc | 878 | lopt->synflood_warned = 1; |
afd46503 | 879 | pr_info("%s: Possible SYN flooding on port %d. %s. Check SNMP counters.\n", |
946cedcc ED |
880 | proto, ntohs(tcp_hdr(skb)->dest), msg); |
881 | } | |
882 | return want_cookie; | |
2a1d4bd4 | 883 | } |
946cedcc | 884 | EXPORT_SYMBOL(tcp_syn_flood_action); |
1da177e4 LT |
885 | |
886 | /* | |
60236fdd | 887 | * Save and compile IPv4 options into the request_sock if needed. |
1da177e4 | 888 | */ |
5dff747b | 889 | static struct ip_options_rcu *tcp_v4_save_options(struct sk_buff *skb) |
1da177e4 | 890 | { |
f6d8bd05 ED |
891 | const struct ip_options *opt = &(IPCB(skb)->opt); |
892 | struct ip_options_rcu *dopt = NULL; | |
1da177e4 LT |
893 | |
894 | if (opt && opt->optlen) { | |
f6d8bd05 ED |
895 | int opt_size = sizeof(*dopt) + opt->optlen; |
896 | ||
1da177e4 LT |
897 | dopt = kmalloc(opt_size, GFP_ATOMIC); |
898 | if (dopt) { | |
f6d8bd05 | 899 | if (ip_options_echo(&dopt->opt, skb)) { |
1da177e4 LT |
900 | kfree(dopt); |
901 | dopt = NULL; | |
902 | } | |
903 | } | |
904 | } | |
905 | return dopt; | |
906 | } | |
907 | ||
cfb6eeb4 YH |
908 | #ifdef CONFIG_TCP_MD5SIG |
909 | /* | |
910 | * RFC2385 MD5 checksumming requires a mapping of | |
911 | * IP address->MD5 Key. | |
912 | * We need to maintain these in the sk structure. | |
913 | */ | |
914 | ||
915 | /* Find the Key structure for an address. */ | |
a915da9b ED |
916 | struct tcp_md5sig_key *tcp_md5_do_lookup(struct sock *sk, |
917 | const union tcp_md5_addr *addr, | |
918 | int family) | |
cfb6eeb4 YH |
919 | { |
920 | struct tcp_sock *tp = tcp_sk(sk); | |
a915da9b | 921 | struct tcp_md5sig_key *key; |
a915da9b | 922 | unsigned int size = sizeof(struct in_addr); |
a8afca03 | 923 | struct tcp_md5sig_info *md5sig; |
cfb6eeb4 | 924 | |
a8afca03 ED |
925 | /* caller either holds rcu_read_lock() or socket lock */ |
926 | md5sig = rcu_dereference_check(tp->md5sig_info, | |
b4fb05ea ED |
927 | sock_owned_by_user(sk) || |
928 | lockdep_is_held(&sk->sk_lock.slock)); | |
a8afca03 | 929 | if (!md5sig) |
cfb6eeb4 | 930 | return NULL; |
a915da9b ED |
931 | #if IS_ENABLED(CONFIG_IPV6) |
932 | if (family == AF_INET6) | |
933 | size = sizeof(struct in6_addr); | |
934 | #endif | |
b67bfe0d | 935 | hlist_for_each_entry_rcu(key, &md5sig->head, node) { |
a915da9b ED |
936 | if (key->family != family) |
937 | continue; | |
938 | if (!memcmp(&key->addr, addr, size)) | |
939 | return key; | |
cfb6eeb4 YH |
940 | } |
941 | return NULL; | |
942 | } | |
a915da9b | 943 | EXPORT_SYMBOL(tcp_md5_do_lookup); |
cfb6eeb4 YH |
944 | |
945 | struct tcp_md5sig_key *tcp_v4_md5_lookup(struct sock *sk, | |
946 | struct sock *addr_sk) | |
947 | { | |
a915da9b ED |
948 | union tcp_md5_addr *addr; |
949 | ||
950 | addr = (union tcp_md5_addr *)&inet_sk(addr_sk)->inet_daddr; | |
951 | return tcp_md5_do_lookup(sk, addr, AF_INET); | |
cfb6eeb4 | 952 | } |
cfb6eeb4 YH |
953 | EXPORT_SYMBOL(tcp_v4_md5_lookup); |
954 | ||
f5b99bcd AB |
955 | static struct tcp_md5sig_key *tcp_v4_reqsk_md5_lookup(struct sock *sk, |
956 | struct request_sock *req) | |
cfb6eeb4 | 957 | { |
a915da9b ED |
958 | union tcp_md5_addr *addr; |
959 | ||
634fb979 | 960 | addr = (union tcp_md5_addr *)&inet_rsk(req)->ir_rmt_addr; |
a915da9b | 961 | return tcp_md5_do_lookup(sk, addr, AF_INET); |
cfb6eeb4 YH |
962 | } |
963 | ||
964 | /* This can be called on a newly created socket, from other files */ | |
a915da9b ED |
965 | int tcp_md5_do_add(struct sock *sk, const union tcp_md5_addr *addr, |
966 | int family, const u8 *newkey, u8 newkeylen, gfp_t gfp) | |
cfb6eeb4 YH |
967 | { |
968 | /* Add Key to the list */ | |
b0a713e9 | 969 | struct tcp_md5sig_key *key; |
cfb6eeb4 | 970 | struct tcp_sock *tp = tcp_sk(sk); |
a915da9b | 971 | struct tcp_md5sig_info *md5sig; |
cfb6eeb4 | 972 | |
c0353c7b | 973 | key = tcp_md5_do_lookup(sk, addr, family); |
cfb6eeb4 YH |
974 | if (key) { |
975 | /* Pre-existing entry - just update that one. */ | |
a915da9b | 976 | memcpy(key->key, newkey, newkeylen); |
b0a713e9 | 977 | key->keylen = newkeylen; |
a915da9b ED |
978 | return 0; |
979 | } | |
260fcbeb | 980 | |
a8afca03 ED |
981 | md5sig = rcu_dereference_protected(tp->md5sig_info, |
982 | sock_owned_by_user(sk)); | |
a915da9b ED |
983 | if (!md5sig) { |
984 | md5sig = kmalloc(sizeof(*md5sig), gfp); | |
985 | if (!md5sig) | |
cfb6eeb4 | 986 | return -ENOMEM; |
cfb6eeb4 | 987 | |
a915da9b ED |
988 | sk_nocaps_add(sk, NETIF_F_GSO_MASK); |
989 | INIT_HLIST_HEAD(&md5sig->head); | |
a8afca03 | 990 | rcu_assign_pointer(tp->md5sig_info, md5sig); |
a915da9b | 991 | } |
cfb6eeb4 | 992 | |
5f3d9cb2 | 993 | key = sock_kmalloc(sk, sizeof(*key), gfp); |
a915da9b ED |
994 | if (!key) |
995 | return -ENOMEM; | |
71cea17e | 996 | if (!tcp_alloc_md5sig_pool()) { |
5f3d9cb2 | 997 | sock_kfree_s(sk, key, sizeof(*key)); |
a915da9b | 998 | return -ENOMEM; |
cfb6eeb4 | 999 | } |
a915da9b ED |
1000 | |
1001 | memcpy(key->key, newkey, newkeylen); | |
1002 | key->keylen = newkeylen; | |
1003 | key->family = family; | |
1004 | memcpy(&key->addr, addr, | |
1005 | (family == AF_INET6) ? sizeof(struct in6_addr) : | |
1006 | sizeof(struct in_addr)); | |
1007 | hlist_add_head_rcu(&key->node, &md5sig->head); | |
cfb6eeb4 YH |
1008 | return 0; |
1009 | } | |
a915da9b | 1010 | EXPORT_SYMBOL(tcp_md5_do_add); |
cfb6eeb4 | 1011 | |
a915da9b | 1012 | int tcp_md5_do_del(struct sock *sk, const union tcp_md5_addr *addr, int family) |
cfb6eeb4 | 1013 | { |
a915da9b ED |
1014 | struct tcp_md5sig_key *key; |
1015 | ||
c0353c7b | 1016 | key = tcp_md5_do_lookup(sk, addr, family); |
a915da9b ED |
1017 | if (!key) |
1018 | return -ENOENT; | |
1019 | hlist_del_rcu(&key->node); | |
5f3d9cb2 | 1020 | atomic_sub(sizeof(*key), &sk->sk_omem_alloc); |
a915da9b | 1021 | kfree_rcu(key, rcu); |
a915da9b | 1022 | return 0; |
cfb6eeb4 | 1023 | } |
a915da9b | 1024 | EXPORT_SYMBOL(tcp_md5_do_del); |
cfb6eeb4 | 1025 | |
e0683e70 | 1026 | static void tcp_clear_md5_list(struct sock *sk) |
cfb6eeb4 YH |
1027 | { |
1028 | struct tcp_sock *tp = tcp_sk(sk); | |
a915da9b | 1029 | struct tcp_md5sig_key *key; |
b67bfe0d | 1030 | struct hlist_node *n; |
a8afca03 | 1031 | struct tcp_md5sig_info *md5sig; |
cfb6eeb4 | 1032 | |
a8afca03 ED |
1033 | md5sig = rcu_dereference_protected(tp->md5sig_info, 1); |
1034 | ||
b67bfe0d | 1035 | hlist_for_each_entry_safe(key, n, &md5sig->head, node) { |
a915da9b | 1036 | hlist_del_rcu(&key->node); |
5f3d9cb2 | 1037 | atomic_sub(sizeof(*key), &sk->sk_omem_alloc); |
a915da9b | 1038 | kfree_rcu(key, rcu); |
cfb6eeb4 YH |
1039 | } |
1040 | } | |
1041 | ||
7174259e ACM |
1042 | static int tcp_v4_parse_md5_keys(struct sock *sk, char __user *optval, |
1043 | int optlen) | |
cfb6eeb4 YH |
1044 | { |
1045 | struct tcp_md5sig cmd; | |
1046 | struct sockaddr_in *sin = (struct sockaddr_in *)&cmd.tcpm_addr; | |
cfb6eeb4 YH |
1047 | |
1048 | if (optlen < sizeof(cmd)) | |
1049 | return -EINVAL; | |
1050 | ||
7174259e | 1051 | if (copy_from_user(&cmd, optval, sizeof(cmd))) |
cfb6eeb4 YH |
1052 | return -EFAULT; |
1053 | ||
1054 | if (sin->sin_family != AF_INET) | |
1055 | return -EINVAL; | |
1056 | ||
a8afca03 | 1057 | if (!cmd.tcpm_key || !cmd.tcpm_keylen) |
a915da9b ED |
1058 | return tcp_md5_do_del(sk, (union tcp_md5_addr *)&sin->sin_addr.s_addr, |
1059 | AF_INET); | |
cfb6eeb4 YH |
1060 | |
1061 | if (cmd.tcpm_keylen > TCP_MD5SIG_MAXKEYLEN) | |
1062 | return -EINVAL; | |
1063 | ||
a915da9b ED |
1064 | return tcp_md5_do_add(sk, (union tcp_md5_addr *)&sin->sin_addr.s_addr, |
1065 | AF_INET, cmd.tcpm_key, cmd.tcpm_keylen, | |
1066 | GFP_KERNEL); | |
cfb6eeb4 YH |
1067 | } |
1068 | ||
49a72dfb AL |
1069 | static int tcp_v4_md5_hash_pseudoheader(struct tcp_md5sig_pool *hp, |
1070 | __be32 daddr, __be32 saddr, int nbytes) | |
cfb6eeb4 | 1071 | { |
cfb6eeb4 | 1072 | struct tcp4_pseudohdr *bp; |
49a72dfb | 1073 | struct scatterlist sg; |
cfb6eeb4 YH |
1074 | |
1075 | bp = &hp->md5_blk.ip4; | |
cfb6eeb4 YH |
1076 | |
1077 | /* | |
49a72dfb | 1078 | * 1. the TCP pseudo-header (in the order: source IP address, |
cfb6eeb4 YH |
1079 | * destination IP address, zero-padded protocol number, and |
1080 | * segment length) | |
1081 | */ | |
1082 | bp->saddr = saddr; | |
1083 | bp->daddr = daddr; | |
1084 | bp->pad = 0; | |
076fb722 | 1085 | bp->protocol = IPPROTO_TCP; |
49a72dfb | 1086 | bp->len = cpu_to_be16(nbytes); |
c7da57a1 | 1087 | |
49a72dfb AL |
1088 | sg_init_one(&sg, bp, sizeof(*bp)); |
1089 | return crypto_hash_update(&hp->md5_desc, &sg, sizeof(*bp)); | |
1090 | } | |
1091 | ||
a915da9b | 1092 | static int tcp_v4_md5_hash_hdr(char *md5_hash, const struct tcp_md5sig_key *key, |
318cf7aa | 1093 | __be32 daddr, __be32 saddr, const struct tcphdr *th) |
49a72dfb AL |
1094 | { |
1095 | struct tcp_md5sig_pool *hp; | |
1096 | struct hash_desc *desc; | |
1097 | ||
1098 | hp = tcp_get_md5sig_pool(); | |
1099 | if (!hp) | |
1100 | goto clear_hash_noput; | |
1101 | desc = &hp->md5_desc; | |
1102 | ||
1103 | if (crypto_hash_init(desc)) | |
1104 | goto clear_hash; | |
1105 | if (tcp_v4_md5_hash_pseudoheader(hp, daddr, saddr, th->doff << 2)) | |
1106 | goto clear_hash; | |
1107 | if (tcp_md5_hash_header(hp, th)) | |
1108 | goto clear_hash; | |
1109 | if (tcp_md5_hash_key(hp, key)) | |
1110 | goto clear_hash; | |
1111 | if (crypto_hash_final(desc, md5_hash)) | |
cfb6eeb4 YH |
1112 | goto clear_hash; |
1113 | ||
cfb6eeb4 | 1114 | tcp_put_md5sig_pool(); |
cfb6eeb4 | 1115 | return 0; |
49a72dfb | 1116 | |
cfb6eeb4 YH |
1117 | clear_hash: |
1118 | tcp_put_md5sig_pool(); | |
1119 | clear_hash_noput: | |
1120 | memset(md5_hash, 0, 16); | |
49a72dfb | 1121 | return 1; |
cfb6eeb4 YH |
1122 | } |
1123 | ||
49a72dfb | 1124 | int tcp_v4_md5_hash_skb(char *md5_hash, struct tcp_md5sig_key *key, |
318cf7aa ED |
1125 | const struct sock *sk, const struct request_sock *req, |
1126 | const struct sk_buff *skb) | |
cfb6eeb4 | 1127 | { |
49a72dfb AL |
1128 | struct tcp_md5sig_pool *hp; |
1129 | struct hash_desc *desc; | |
318cf7aa | 1130 | const struct tcphdr *th = tcp_hdr(skb); |
cfb6eeb4 YH |
1131 | __be32 saddr, daddr; |
1132 | ||
1133 | if (sk) { | |
c720c7e8 ED |
1134 | saddr = inet_sk(sk)->inet_saddr; |
1135 | daddr = inet_sk(sk)->inet_daddr; | |
49a72dfb | 1136 | } else if (req) { |
634fb979 ED |
1137 | saddr = inet_rsk(req)->ir_loc_addr; |
1138 | daddr = inet_rsk(req)->ir_rmt_addr; | |
cfb6eeb4 | 1139 | } else { |
49a72dfb AL |
1140 | const struct iphdr *iph = ip_hdr(skb); |
1141 | saddr = iph->saddr; | |
1142 | daddr = iph->daddr; | |
cfb6eeb4 | 1143 | } |
49a72dfb AL |
1144 | |
1145 | hp = tcp_get_md5sig_pool(); | |
1146 | if (!hp) | |
1147 | goto clear_hash_noput; | |
1148 | desc = &hp->md5_desc; | |
1149 | ||
1150 | if (crypto_hash_init(desc)) | |
1151 | goto clear_hash; | |
1152 | ||
1153 | if (tcp_v4_md5_hash_pseudoheader(hp, daddr, saddr, skb->len)) | |
1154 | goto clear_hash; | |
1155 | if (tcp_md5_hash_header(hp, th)) | |
1156 | goto clear_hash; | |
1157 | if (tcp_md5_hash_skb_data(hp, skb, th->doff << 2)) | |
1158 | goto clear_hash; | |
1159 | if (tcp_md5_hash_key(hp, key)) | |
1160 | goto clear_hash; | |
1161 | if (crypto_hash_final(desc, md5_hash)) | |
1162 | goto clear_hash; | |
1163 | ||
1164 | tcp_put_md5sig_pool(); | |
1165 | return 0; | |
1166 | ||
1167 | clear_hash: | |
1168 | tcp_put_md5sig_pool(); | |
1169 | clear_hash_noput: | |
1170 | memset(md5_hash, 0, 16); | |
1171 | return 1; | |
cfb6eeb4 | 1172 | } |
49a72dfb | 1173 | EXPORT_SYMBOL(tcp_v4_md5_hash_skb); |
cfb6eeb4 | 1174 | |
a2a385d6 | 1175 | static bool tcp_v4_inbound_md5_hash(struct sock *sk, const struct sk_buff *skb) |
cfb6eeb4 YH |
1176 | { |
1177 | /* | |
1178 | * This gets called for each TCP segment that arrives | |
1179 | * so we want to be efficient. | |
1180 | * We have 3 drop cases: | |
1181 | * o No MD5 hash and one expected. | |
1182 | * o MD5 hash and we're not expecting one. | |
1183 | * o MD5 hash and its wrong. | |
1184 | */ | |
cf533ea5 | 1185 | const __u8 *hash_location = NULL; |
cfb6eeb4 | 1186 | struct tcp_md5sig_key *hash_expected; |
eddc9ec5 | 1187 | const struct iphdr *iph = ip_hdr(skb); |
cf533ea5 | 1188 | const struct tcphdr *th = tcp_hdr(skb); |
cfb6eeb4 | 1189 | int genhash; |
cfb6eeb4 YH |
1190 | unsigned char newhash[16]; |
1191 | ||
a915da9b ED |
1192 | hash_expected = tcp_md5_do_lookup(sk, (union tcp_md5_addr *)&iph->saddr, |
1193 | AF_INET); | |
7d5d5525 | 1194 | hash_location = tcp_parse_md5sig_option(th); |
cfb6eeb4 | 1195 | |
cfb6eeb4 YH |
1196 | /* We've parsed the options - do we have a hash? */ |
1197 | if (!hash_expected && !hash_location) | |
a2a385d6 | 1198 | return false; |
cfb6eeb4 YH |
1199 | |
1200 | if (hash_expected && !hash_location) { | |
785957d3 | 1201 | NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPMD5NOTFOUND); |
a2a385d6 | 1202 | return true; |
cfb6eeb4 YH |
1203 | } |
1204 | ||
1205 | if (!hash_expected && hash_location) { | |
785957d3 | 1206 | NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPMD5UNEXPECTED); |
a2a385d6 | 1207 | return true; |
cfb6eeb4 YH |
1208 | } |
1209 | ||
1210 | /* Okay, so this is hash_expected and hash_location - | |
1211 | * so we need to calculate the checksum. | |
1212 | */ | |
49a72dfb AL |
1213 | genhash = tcp_v4_md5_hash_skb(newhash, |
1214 | hash_expected, | |
1215 | NULL, NULL, skb); | |
cfb6eeb4 YH |
1216 | |
1217 | if (genhash || memcmp(hash_location, newhash, 16) != 0) { | |
e87cc472 JP |
1218 | net_info_ratelimited("MD5 Hash failed for (%pI4, %d)->(%pI4, %d)%s\n", |
1219 | &iph->saddr, ntohs(th->source), | |
1220 | &iph->daddr, ntohs(th->dest), | |
1221 | genhash ? " tcp_v4_calc_md5_hash failed" | |
1222 | : ""); | |
a2a385d6 | 1223 | return true; |
cfb6eeb4 | 1224 | } |
a2a385d6 | 1225 | return false; |
cfb6eeb4 YH |
1226 | } |
1227 | ||
1228 | #endif | |
1229 | ||
16bea70a OP |
1230 | static void tcp_v4_init_req(struct request_sock *req, struct sock *sk, |
1231 | struct sk_buff *skb) | |
1232 | { | |
1233 | struct inet_request_sock *ireq = inet_rsk(req); | |
1234 | ||
1235 | ireq->ir_loc_addr = ip_hdr(skb)->daddr; | |
1236 | ireq->ir_rmt_addr = ip_hdr(skb)->saddr; | |
1237 | ireq->no_srccheck = inet_sk(sk)->transparent; | |
1238 | ireq->opt = tcp_v4_save_options(skb); | |
1239 | } | |
1240 | ||
d94e0417 OP |
1241 | static struct dst_entry *tcp_v4_route_req(struct sock *sk, struct flowi *fl, |
1242 | const struct request_sock *req, | |
1243 | bool *strict) | |
1244 | { | |
1245 | struct dst_entry *dst = inet_csk_route_req(sk, &fl->u.ip4, req); | |
1246 | ||
1247 | if (strict) { | |
1248 | if (fl->u.ip4.daddr == inet_rsk(req)->ir_rmt_addr) | |
1249 | *strict = true; | |
1250 | else | |
1251 | *strict = false; | |
1252 | } | |
1253 | ||
1254 | return dst; | |
1255 | } | |
1256 | ||
72a3effa | 1257 | struct request_sock_ops tcp_request_sock_ops __read_mostly = { |
1da177e4 | 1258 | .family = PF_INET, |
2e6599cb | 1259 | .obj_size = sizeof(struct tcp_request_sock), |
5db92c99 | 1260 | .rtx_syn_ack = tcp_rtx_synack, |
60236fdd ACM |
1261 | .send_ack = tcp_v4_reqsk_send_ack, |
1262 | .destructor = tcp_v4_reqsk_destructor, | |
1da177e4 | 1263 | .send_reset = tcp_v4_send_reset, |
72659ecc | 1264 | .syn_ack_timeout = tcp_syn_ack_timeout, |
1da177e4 LT |
1265 | }; |
1266 | ||
b2e4b3de | 1267 | static const struct tcp_request_sock_ops tcp_request_sock_ipv4_ops = { |
2aec4a29 | 1268 | .mss_clamp = TCP_MSS_DEFAULT, |
16bea70a | 1269 | #ifdef CONFIG_TCP_MD5SIG |
cfb6eeb4 | 1270 | .md5_lookup = tcp_v4_reqsk_md5_lookup, |
e3afe7b7 | 1271 | .calc_md5_hash = tcp_v4_md5_hash_skb, |
b6332e6c | 1272 | #endif |
16bea70a | 1273 | .init_req = tcp_v4_init_req, |
fb7b37a7 OP |
1274 | #ifdef CONFIG_SYN_COOKIES |
1275 | .cookie_init_seq = cookie_v4_init_sequence, | |
1276 | #endif | |
d94e0417 | 1277 | .route_req = tcp_v4_route_req, |
936b8bdb | 1278 | .init_seq = tcp_v4_init_sequence, |
d6274bd8 | 1279 | .send_synack = tcp_v4_send_synack, |
695da14e | 1280 | .queue_hash_add = inet_csk_reqsk_queue_hash_add, |
16bea70a | 1281 | }; |
cfb6eeb4 | 1282 | |
1da177e4 LT |
1283 | int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb) |
1284 | { | |
1285 | struct tcp_options_received tmp_opt; | |
60236fdd | 1286 | struct request_sock *req; |
4957faad | 1287 | struct tcp_sock *tp = tcp_sk(sk); |
e6b4d113 | 1288 | struct dst_entry *dst = NULL; |
eddc9ec5 | 1289 | __be32 saddr = ip_hdr(skb)->saddr; |
1da177e4 | 1290 | __u32 isn = TCP_SKB_CB(skb)->when; |
843f4a55 | 1291 | bool want_cookie = false, fastopen; |
168a8f58 JC |
1292 | struct flowi4 fl4; |
1293 | struct tcp_fastopen_cookie foc = { .len = -1 }; | |
16bea70a | 1294 | const struct tcp_request_sock_ops *af_ops; |
843f4a55 | 1295 | int err; |
1da177e4 LT |
1296 | |
1297 | /* Never answer to SYNs send to broadcast or multicast */ | |
511c3f92 | 1298 | if (skb_rtable(skb)->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST)) |
1da177e4 LT |
1299 | goto drop; |
1300 | ||
1301 | /* TW buckets are converted to open requests without | |
1302 | * limitations, they conserve resources and peer is | |
1303 | * evidently real one. | |
1304 | */ | |
5ad37d5d HFS |
1305 | if ((sysctl_tcp_syncookies == 2 || |
1306 | inet_csk_reqsk_queue_is_full(sk)) && !isn) { | |
946cedcc ED |
1307 | want_cookie = tcp_syn_flood_action(sk, skb, "TCP"); |
1308 | if (!want_cookie) | |
1309 | goto drop; | |
1da177e4 LT |
1310 | } |
1311 | ||
1312 | /* Accept backlog is full. If we have already queued enough | |
1313 | * of warm entries in syn queue, drop request. It is better than | |
1314 | * clogging syn queue with openreqs with exponentially increasing | |
1315 | * timeout. | |
1316 | */ | |
2aeef18d NS |
1317 | if (sk_acceptq_is_full(sk) && inet_csk_reqsk_queue_young(sk) > 1) { |
1318 | NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENOVERFLOWS); | |
1da177e4 | 1319 | goto drop; |
2aeef18d | 1320 | } |
1da177e4 | 1321 | |
ce4a7d0d | 1322 | req = inet_reqsk_alloc(&tcp_request_sock_ops); |
1da177e4 LT |
1323 | if (!req) |
1324 | goto drop; | |
1325 | ||
16bea70a | 1326 | af_ops = tcp_rsk(req)->af_specific = &tcp_request_sock_ipv4_ops; |
cfb6eeb4 | 1327 | |
1da177e4 | 1328 | tcp_clear_options(&tmp_opt); |
2aec4a29 | 1329 | tmp_opt.mss_clamp = af_ops->mss_clamp; |
4957faad | 1330 | tmp_opt.user_mss = tp->rx_opt.user_mss; |
1a2c6181 | 1331 | tcp_parse_options(skb, &tmp_opt, 0, want_cookie ? NULL : &foc); |
1da177e4 | 1332 | |
4dfc2817 | 1333 | if (want_cookie && !tmp_opt.saw_tstamp) |
1da177e4 | 1334 | tcp_clear_options(&tmp_opt); |
1da177e4 | 1335 | |
1da177e4 | 1336 | tmp_opt.tstamp_ok = tmp_opt.saw_tstamp; |
e0f802fb | 1337 | tcp_openreq_init(req, &tmp_opt, skb, sk); |
1da177e4 | 1338 | |
16bea70a | 1339 | af_ops->init_req(req, sk, skb); |
bb5b7c11 | 1340 | |
284904aa | 1341 | if (security_inet_conn_request(sk, skb, req)) |
bb5b7c11 | 1342 | goto drop_and_free; |
284904aa | 1343 | |
172d69e6 | 1344 | if (!want_cookie || tmp_opt.tstamp_ok) |
5d134f1c | 1345 | TCP_ECN_create_request(req, skb, sock_net(sk)); |
1da177e4 LT |
1346 | |
1347 | if (want_cookie) { | |
fb7b37a7 | 1348 | isn = cookie_init_sequence(af_ops, sk, skb, &req->mss); |
172d69e6 | 1349 | req->cookie_ts = tmp_opt.tstamp_ok; |
1da177e4 | 1350 | } else if (!isn) { |
1da177e4 LT |
1351 | /* VJ's idea. We save last timestamp seen |
1352 | * from the destination in peer table, when entering | |
1353 | * state TIME-WAIT, and check against it before | |
1354 | * accepting new connection request. | |
1355 | * | |
1356 | * If "isn" is not zero, this request hit alive | |
1357 | * timewait bucket, so that all the necessary checks | |
1358 | * are made in the function processing timewait state. | |
1359 | */ | |
d94e0417 OP |
1360 | if (tmp_opt.saw_tstamp && tcp_death_row.sysctl_tw_recycle) { |
1361 | bool strict; | |
1362 | ||
1363 | dst = af_ops->route_req(sk, (struct flowi *)&fl4, req, | |
1364 | &strict); | |
1365 | if (dst && strict && | |
1366 | !tcp_peer_is_proven(req, dst, true)) { | |
de0744af | 1367 | NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_PAWSPASSIVEREJECTED); |
7cd04fa7 | 1368 | goto drop_and_release; |
1da177e4 LT |
1369 | } |
1370 | } | |
1371 | /* Kill the following clause, if you dislike this way. */ | |
1372 | else if (!sysctl_tcp_syncookies && | |
463c84b9 | 1373 | (sysctl_max_syn_backlog - inet_csk_reqsk_queue_len(sk) < |
1da177e4 | 1374 | (sysctl_max_syn_backlog >> 2)) && |
81166dd6 | 1375 | !tcp_peer_is_proven(req, dst, false)) { |
1da177e4 LT |
1376 | /* Without syncookies last quarter of |
1377 | * backlog is filled with destinations, | |
1378 | * proven to be alive. | |
1379 | * It means that we continue to communicate | |
1380 | * to destinations, already remembered | |
1381 | * to the moment of synflood. | |
1382 | */ | |
afd46503 | 1383 | LIMIT_NETDEBUG(KERN_DEBUG pr_fmt("drop open request from %pI4/%u\n"), |
673d57e7 | 1384 | &saddr, ntohs(tcp_hdr(skb)->source)); |
7cd04fa7 | 1385 | goto drop_and_release; |
1da177e4 LT |
1386 | } |
1387 | ||
936b8bdb | 1388 | isn = af_ops->init_seq(skb); |
1da177e4 | 1389 | } |
d94e0417 OP |
1390 | if (!dst) { |
1391 | dst = af_ops->route_req(sk, (struct flowi *)&fl4, req, NULL); | |
1392 | if (!dst) | |
1393 | goto drop_and_free; | |
1394 | } | |
168a8f58 | 1395 | |
843f4a55 | 1396 | tcp_rsk(req)->snt_isn = isn; |
843f4a55 YC |
1397 | tcp_openreq_init_rwin(req, sk, dst); |
1398 | fastopen = !want_cookie && | |
1399 | tcp_try_fastopen(sk, skb, req, &foc, dst); | |
d6274bd8 OP |
1400 | err = af_ops->send_synack(sk, dst, NULL, req, |
1401 | skb_get_queue_mapping(skb), &foc); | |
843f4a55 | 1402 | if (!fastopen) { |
168a8f58 JC |
1403 | if (err || want_cookie) |
1404 | goto drop_and_free; | |
1405 | ||
1406 | tcp_rsk(req)->listener = NULL; | |
695da14e | 1407 | af_ops->queue_hash_add(sk, req, TCP_TIMEOUT_INIT); |
843f4a55 | 1408 | } |
1da177e4 | 1409 | |
1da177e4 LT |
1410 | return 0; |
1411 | ||
7cd04fa7 DL |
1412 | drop_and_release: |
1413 | dst_release(dst); | |
1da177e4 | 1414 | drop_and_free: |
60236fdd | 1415 | reqsk_free(req); |
1da177e4 | 1416 | drop: |
848bf15f | 1417 | NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENDROPS); |
1da177e4 LT |
1418 | return 0; |
1419 | } | |
4bc2f18b | 1420 | EXPORT_SYMBOL(tcp_v4_conn_request); |
1da177e4 LT |
1421 | |
1422 | ||
1423 | /* | |
1424 | * The three way handshake has completed - we got a valid synack - | |
1425 | * now create the new socket. | |
1426 | */ | |
1427 | struct sock *tcp_v4_syn_recv_sock(struct sock *sk, struct sk_buff *skb, | |
60236fdd | 1428 | struct request_sock *req, |
1da177e4 LT |
1429 | struct dst_entry *dst) |
1430 | { | |
2e6599cb | 1431 | struct inet_request_sock *ireq; |
1da177e4 LT |
1432 | struct inet_sock *newinet; |
1433 | struct tcp_sock *newtp; | |
1434 | struct sock *newsk; | |
cfb6eeb4 YH |
1435 | #ifdef CONFIG_TCP_MD5SIG |
1436 | struct tcp_md5sig_key *key; | |
1437 | #endif | |
f6d8bd05 | 1438 | struct ip_options_rcu *inet_opt; |
1da177e4 LT |
1439 | |
1440 | if (sk_acceptq_is_full(sk)) | |
1441 | goto exit_overflow; | |
1442 | ||
1da177e4 LT |
1443 | newsk = tcp_create_openreq_child(sk, req, skb); |
1444 | if (!newsk) | |
093d2823 | 1445 | goto exit_nonewsk; |
1da177e4 | 1446 | |
bcd76111 | 1447 | newsk->sk_gso_type = SKB_GSO_TCPV4; |
fae6ef87 | 1448 | inet_sk_rx_dst_set(newsk, skb); |
1da177e4 LT |
1449 | |
1450 | newtp = tcp_sk(newsk); | |
1451 | newinet = inet_sk(newsk); | |
2e6599cb | 1452 | ireq = inet_rsk(req); |
634fb979 ED |
1453 | newinet->inet_daddr = ireq->ir_rmt_addr; |
1454 | newinet->inet_rcv_saddr = ireq->ir_loc_addr; | |
1455 | newinet->inet_saddr = ireq->ir_loc_addr; | |
f6d8bd05 ED |
1456 | inet_opt = ireq->opt; |
1457 | rcu_assign_pointer(newinet->inet_opt, inet_opt); | |
2e6599cb | 1458 | ireq->opt = NULL; |
463c84b9 | 1459 | newinet->mc_index = inet_iif(skb); |
eddc9ec5 | 1460 | newinet->mc_ttl = ip_hdr(skb)->ttl; |
4c507d28 | 1461 | newinet->rcv_tos = ip_hdr(skb)->tos; |
d83d8461 | 1462 | inet_csk(newsk)->icsk_ext_hdr_len = 0; |
f6d8bd05 ED |
1463 | if (inet_opt) |
1464 | inet_csk(newsk)->icsk_ext_hdr_len = inet_opt->opt.optlen; | |
c720c7e8 | 1465 | newinet->inet_id = newtp->write_seq ^ jiffies; |
1da177e4 | 1466 | |
dfd25fff ED |
1467 | if (!dst) { |
1468 | dst = inet_csk_route_child_sock(sk, newsk, req); | |
1469 | if (!dst) | |
1470 | goto put_and_exit; | |
1471 | } else { | |
1472 | /* syncookie case : see end of cookie_v4_check() */ | |
1473 | } | |
0e734419 DM |
1474 | sk_setup_caps(newsk, dst); |
1475 | ||
1da177e4 | 1476 | tcp_sync_mss(newsk, dst_mtu(dst)); |
0dbaee3b | 1477 | newtp->advmss = dst_metric_advmss(dst); |
f5fff5dc TQ |
1478 | if (tcp_sk(sk)->rx_opt.user_mss && |
1479 | tcp_sk(sk)->rx_opt.user_mss < newtp->advmss) | |
1480 | newtp->advmss = tcp_sk(sk)->rx_opt.user_mss; | |
1481 | ||
1da177e4 LT |
1482 | tcp_initialize_rcv_mss(newsk); |
1483 | ||
cfb6eeb4 YH |
1484 | #ifdef CONFIG_TCP_MD5SIG |
1485 | /* Copy over the MD5 key from the original socket */ | |
a915da9b ED |
1486 | key = tcp_md5_do_lookup(sk, (union tcp_md5_addr *)&newinet->inet_daddr, |
1487 | AF_INET); | |
c720c7e8 | 1488 | if (key != NULL) { |
cfb6eeb4 YH |
1489 | /* |
1490 | * We're using one, so create a matching key | |
1491 | * on the newsk structure. If we fail to get | |
1492 | * memory, then we end up not copying the key | |
1493 | * across. Shucks. | |
1494 | */ | |
a915da9b ED |
1495 | tcp_md5_do_add(newsk, (union tcp_md5_addr *)&newinet->inet_daddr, |
1496 | AF_INET, key->key, key->keylen, GFP_ATOMIC); | |
a465419b | 1497 | sk_nocaps_add(newsk, NETIF_F_GSO_MASK); |
cfb6eeb4 YH |
1498 | } |
1499 | #endif | |
1500 | ||
0e734419 DM |
1501 | if (__inet_inherit_port(sk, newsk) < 0) |
1502 | goto put_and_exit; | |
9327f705 | 1503 | __inet_hash_nolisten(newsk, NULL); |
1da177e4 LT |
1504 | |
1505 | return newsk; | |
1506 | ||
1507 | exit_overflow: | |
de0744af | 1508 | NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENOVERFLOWS); |
093d2823 BS |
1509 | exit_nonewsk: |
1510 | dst_release(dst); | |
1da177e4 | 1511 | exit: |
de0744af | 1512 | NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENDROPS); |
1da177e4 | 1513 | return NULL; |
0e734419 | 1514 | put_and_exit: |
e337e24d CP |
1515 | inet_csk_prepare_forced_close(newsk); |
1516 | tcp_done(newsk); | |
0e734419 | 1517 | goto exit; |
1da177e4 | 1518 | } |
4bc2f18b | 1519 | EXPORT_SYMBOL(tcp_v4_syn_recv_sock); |
1da177e4 LT |
1520 | |
1521 | static struct sock *tcp_v4_hnd_req(struct sock *sk, struct sk_buff *skb) | |
1522 | { | |
aa8223c7 | 1523 | struct tcphdr *th = tcp_hdr(skb); |
eddc9ec5 | 1524 | const struct iphdr *iph = ip_hdr(skb); |
1da177e4 | 1525 | struct sock *nsk; |
60236fdd | 1526 | struct request_sock **prev; |
1da177e4 | 1527 | /* Find possible connection requests. */ |
463c84b9 ACM |
1528 | struct request_sock *req = inet_csk_search_req(sk, &prev, th->source, |
1529 | iph->saddr, iph->daddr); | |
1da177e4 | 1530 | if (req) |
8336886f | 1531 | return tcp_check_req(sk, skb, req, prev, false); |
1da177e4 | 1532 | |
3b1e0a65 | 1533 | nsk = inet_lookup_established(sock_net(sk), &tcp_hashinfo, iph->saddr, |
c67499c0 | 1534 | th->source, iph->daddr, th->dest, inet_iif(skb)); |
1da177e4 LT |
1535 | |
1536 | if (nsk) { | |
1537 | if (nsk->sk_state != TCP_TIME_WAIT) { | |
1538 | bh_lock_sock(nsk); | |
1539 | return nsk; | |
1540 | } | |
9469c7b4 | 1541 | inet_twsk_put(inet_twsk(nsk)); |
1da177e4 LT |
1542 | return NULL; |
1543 | } | |
1544 | ||
1545 | #ifdef CONFIG_SYN_COOKIES | |
af9b4738 | 1546 | if (!th->syn) |
1da177e4 LT |
1547 | sk = cookie_v4_check(sk, skb, &(IPCB(skb)->opt)); |
1548 | #endif | |
1549 | return sk; | |
1550 | } | |
1551 | ||
1da177e4 LT |
1552 | /* The socket must have it's spinlock held when we get |
1553 | * here. | |
1554 | * | |
1555 | * We have a potential double-lock case here, so even when | |
1556 | * doing backlog processing we use the BH locking scheme. | |
1557 | * This is because we cannot sleep with the original spinlock | |
1558 | * held. | |
1559 | */ | |
1560 | int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb) | |
1561 | { | |
cfb6eeb4 YH |
1562 | struct sock *rsk; |
1563 | #ifdef CONFIG_TCP_MD5SIG | |
1564 | /* | |
1565 | * We really want to reject the packet as early as possible | |
1566 | * if: | |
1567 | * o We're expecting an MD5'd packet and this is no MD5 tcp option | |
1568 | * o There is an MD5 option and we're not expecting one | |
1569 | */ | |
7174259e | 1570 | if (tcp_v4_inbound_md5_hash(sk, skb)) |
cfb6eeb4 YH |
1571 | goto discard; |
1572 | #endif | |
1573 | ||
1da177e4 | 1574 | if (sk->sk_state == TCP_ESTABLISHED) { /* Fast path */ |
404e0a8b ED |
1575 | struct dst_entry *dst = sk->sk_rx_dst; |
1576 | ||
bdeab991 | 1577 | sock_rps_save_rxhash(sk, skb); |
404e0a8b | 1578 | if (dst) { |
505fbcf0 ED |
1579 | if (inet_sk(sk)->rx_dst_ifindex != skb->skb_iif || |
1580 | dst->ops->check(dst, 0) == NULL) { | |
92101b3b DM |
1581 | dst_release(dst); |
1582 | sk->sk_rx_dst = NULL; | |
1583 | } | |
1584 | } | |
c995ae22 | 1585 | tcp_rcv_established(sk, skb, tcp_hdr(skb), skb->len); |
1da177e4 LT |
1586 | return 0; |
1587 | } | |
1588 | ||
ab6a5bb6 | 1589 | if (skb->len < tcp_hdrlen(skb) || tcp_checksum_complete(skb)) |
1da177e4 LT |
1590 | goto csum_err; |
1591 | ||
1592 | if (sk->sk_state == TCP_LISTEN) { | |
1593 | struct sock *nsk = tcp_v4_hnd_req(sk, skb); | |
1594 | if (!nsk) | |
1595 | goto discard; | |
1596 | ||
1597 | if (nsk != sk) { | |
bdeab991 | 1598 | sock_rps_save_rxhash(nsk, skb); |
cfb6eeb4 YH |
1599 | if (tcp_child_process(sk, nsk, skb)) { |
1600 | rsk = nsk; | |
1da177e4 | 1601 | goto reset; |
cfb6eeb4 | 1602 | } |
1da177e4 LT |
1603 | return 0; |
1604 | } | |
ca55158c | 1605 | } else |
bdeab991 | 1606 | sock_rps_save_rxhash(sk, skb); |
ca55158c | 1607 | |
aa8223c7 | 1608 | if (tcp_rcv_state_process(sk, skb, tcp_hdr(skb), skb->len)) { |
cfb6eeb4 | 1609 | rsk = sk; |
1da177e4 | 1610 | goto reset; |
cfb6eeb4 | 1611 | } |
1da177e4 LT |
1612 | return 0; |
1613 | ||
1614 | reset: | |
cfb6eeb4 | 1615 | tcp_v4_send_reset(rsk, skb); |
1da177e4 LT |
1616 | discard: |
1617 | kfree_skb(skb); | |
1618 | /* Be careful here. If this function gets more complicated and | |
1619 | * gcc suffers from register pressure on the x86, sk (in %ebx) | |
1620 | * might be destroyed here. This current version compiles correctly, | |
1621 | * but you have been warned. | |
1622 | */ | |
1623 | return 0; | |
1624 | ||
1625 | csum_err: | |
6a5dc9e5 | 1626 | TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_CSUMERRORS); |
63231bdd | 1627 | TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_INERRS); |
1da177e4 LT |
1628 | goto discard; |
1629 | } | |
4bc2f18b | 1630 | EXPORT_SYMBOL(tcp_v4_do_rcv); |
1da177e4 | 1631 | |
160eb5a6 | 1632 | void tcp_v4_early_demux(struct sk_buff *skb) |
41063e9d | 1633 | { |
41063e9d DM |
1634 | const struct iphdr *iph; |
1635 | const struct tcphdr *th; | |
1636 | struct sock *sk; | |
41063e9d | 1637 | |
41063e9d | 1638 | if (skb->pkt_type != PACKET_HOST) |
160eb5a6 | 1639 | return; |
41063e9d | 1640 | |
45f00f99 | 1641 | if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct tcphdr))) |
160eb5a6 | 1642 | return; |
41063e9d DM |
1643 | |
1644 | iph = ip_hdr(skb); | |
45f00f99 | 1645 | th = tcp_hdr(skb); |
41063e9d DM |
1646 | |
1647 | if (th->doff < sizeof(struct tcphdr) / 4) | |
160eb5a6 | 1648 | return; |
41063e9d | 1649 | |
45f00f99 | 1650 | sk = __inet_lookup_established(dev_net(skb->dev), &tcp_hashinfo, |
41063e9d | 1651 | iph->saddr, th->source, |
7011d085 | 1652 | iph->daddr, ntohs(th->dest), |
9cb429d6 | 1653 | skb->skb_iif); |
41063e9d DM |
1654 | if (sk) { |
1655 | skb->sk = sk; | |
1656 | skb->destructor = sock_edemux; | |
1657 | if (sk->sk_state != TCP_TIME_WAIT) { | |
1658 | struct dst_entry *dst = sk->sk_rx_dst; | |
505fbcf0 | 1659 | |
41063e9d DM |
1660 | if (dst) |
1661 | dst = dst_check(dst, 0); | |
92101b3b | 1662 | if (dst && |
505fbcf0 | 1663 | inet_sk(sk)->rx_dst_ifindex == skb->skb_iif) |
92101b3b | 1664 | skb_dst_set_noref(skb, dst); |
41063e9d DM |
1665 | } |
1666 | } | |
41063e9d DM |
1667 | } |
1668 | ||
b2fb4f54 ED |
1669 | /* Packet is added to VJ-style prequeue for processing in process |
1670 | * context, if a reader task is waiting. Apparently, this exciting | |
1671 | * idea (VJ's mail "Re: query about TCP header on tcp-ip" of 07 Sep 93) | |
1672 | * failed somewhere. Latency? Burstiness? Well, at least now we will | |
1673 | * see, why it failed. 8)8) --ANK | |
1674 | * | |
1675 | */ | |
1676 | bool tcp_prequeue(struct sock *sk, struct sk_buff *skb) | |
1677 | { | |
1678 | struct tcp_sock *tp = tcp_sk(sk); | |
1679 | ||
1680 | if (sysctl_tcp_low_latency || !tp->ucopy.task) | |
1681 | return false; | |
1682 | ||
1683 | if (skb->len <= tcp_hdrlen(skb) && | |
1684 | skb_queue_len(&tp->ucopy.prequeue) == 0) | |
1685 | return false; | |
1686 | ||
58717686 | 1687 | skb_dst_force(skb); |
b2fb4f54 ED |
1688 | __skb_queue_tail(&tp->ucopy.prequeue, skb); |
1689 | tp->ucopy.memory += skb->truesize; | |
1690 | if (tp->ucopy.memory > sk->sk_rcvbuf) { | |
1691 | struct sk_buff *skb1; | |
1692 | ||
1693 | BUG_ON(sock_owned_by_user(sk)); | |
1694 | ||
1695 | while ((skb1 = __skb_dequeue(&tp->ucopy.prequeue)) != NULL) { | |
1696 | sk_backlog_rcv(sk, skb1); | |
1697 | NET_INC_STATS_BH(sock_net(sk), | |
1698 | LINUX_MIB_TCPPREQUEUEDROPPED); | |
1699 | } | |
1700 | ||
1701 | tp->ucopy.memory = 0; | |
1702 | } else if (skb_queue_len(&tp->ucopy.prequeue) == 1) { | |
1703 | wake_up_interruptible_sync_poll(sk_sleep(sk), | |
1704 | POLLIN | POLLRDNORM | POLLRDBAND); | |
1705 | if (!inet_csk_ack_scheduled(sk)) | |
1706 | inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK, | |
1707 | (3 * tcp_rto_min(sk)) / 4, | |
1708 | TCP_RTO_MAX); | |
1709 | } | |
1710 | return true; | |
1711 | } | |
1712 | EXPORT_SYMBOL(tcp_prequeue); | |
1713 | ||
1da177e4 LT |
1714 | /* |
1715 | * From tcp_input.c | |
1716 | */ | |
1717 | ||
1718 | int tcp_v4_rcv(struct sk_buff *skb) | |
1719 | { | |
eddc9ec5 | 1720 | const struct iphdr *iph; |
cf533ea5 | 1721 | const struct tcphdr *th; |
1da177e4 LT |
1722 | struct sock *sk; |
1723 | int ret; | |
a86b1e30 | 1724 | struct net *net = dev_net(skb->dev); |
1da177e4 LT |
1725 | |
1726 | if (skb->pkt_type != PACKET_HOST) | |
1727 | goto discard_it; | |
1728 | ||
1729 | /* Count it even if it's bad */ | |
63231bdd | 1730 | TCP_INC_STATS_BH(net, TCP_MIB_INSEGS); |
1da177e4 LT |
1731 | |
1732 | if (!pskb_may_pull(skb, sizeof(struct tcphdr))) | |
1733 | goto discard_it; | |
1734 | ||
aa8223c7 | 1735 | th = tcp_hdr(skb); |
1da177e4 LT |
1736 | |
1737 | if (th->doff < sizeof(struct tcphdr) / 4) | |
1738 | goto bad_packet; | |
1739 | if (!pskb_may_pull(skb, th->doff * 4)) | |
1740 | goto discard_it; | |
1741 | ||
1742 | /* An explanation is required here, I think. | |
1743 | * Packet length and doff are validated by header prediction, | |
caa20d9a | 1744 | * provided case of th->doff==0 is eliminated. |
1da177e4 | 1745 | * So, we defer the checks. */ |
ed70fcfc TH |
1746 | |
1747 | if (skb_checksum_init(skb, IPPROTO_TCP, inet_compute_pseudo)) | |
6a5dc9e5 | 1748 | goto csum_error; |
1da177e4 | 1749 | |
aa8223c7 | 1750 | th = tcp_hdr(skb); |
eddc9ec5 | 1751 | iph = ip_hdr(skb); |
1da177e4 LT |
1752 | TCP_SKB_CB(skb)->seq = ntohl(th->seq); |
1753 | TCP_SKB_CB(skb)->end_seq = (TCP_SKB_CB(skb)->seq + th->syn + th->fin + | |
1754 | skb->len - th->doff * 4); | |
1755 | TCP_SKB_CB(skb)->ack_seq = ntohl(th->ack_seq); | |
1756 | TCP_SKB_CB(skb)->when = 0; | |
b82d1bb4 | 1757 | TCP_SKB_CB(skb)->ip_dsfield = ipv4_get_dsfield(iph); |
1da177e4 LT |
1758 | TCP_SKB_CB(skb)->sacked = 0; |
1759 | ||
9a1f27c4 | 1760 | sk = __inet_lookup_skb(&tcp_hashinfo, skb, th->source, th->dest); |
1da177e4 LT |
1761 | if (!sk) |
1762 | goto no_tcp_socket; | |
1763 | ||
bb134d5d ED |
1764 | process: |
1765 | if (sk->sk_state == TCP_TIME_WAIT) | |
1766 | goto do_time_wait; | |
1767 | ||
6cce09f8 ED |
1768 | if (unlikely(iph->ttl < inet_sk(sk)->min_ttl)) { |
1769 | NET_INC_STATS_BH(net, LINUX_MIB_TCPMINTTLDROP); | |
d218d111 | 1770 | goto discard_and_relse; |
6cce09f8 | 1771 | } |
d218d111 | 1772 | |
1da177e4 LT |
1773 | if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb)) |
1774 | goto discard_and_relse; | |
b59c2701 | 1775 | nf_reset(skb); |
1da177e4 | 1776 | |
fda9ef5d | 1777 | if (sk_filter(sk, skb)) |
1da177e4 LT |
1778 | goto discard_and_relse; |
1779 | ||
8b80cda5 | 1780 | sk_mark_napi_id(sk, skb); |
1da177e4 LT |
1781 | skb->dev = NULL; |
1782 | ||
c6366184 | 1783 | bh_lock_sock_nested(sk); |
1da177e4 LT |
1784 | ret = 0; |
1785 | if (!sock_owned_by_user(sk)) { | |
1a2449a8 CL |
1786 | #ifdef CONFIG_NET_DMA |
1787 | struct tcp_sock *tp = tcp_sk(sk); | |
1788 | if (!tp->ucopy.dma_chan && tp->ucopy.pinned_list) | |
a2bd1140 | 1789 | tp->ucopy.dma_chan = net_dma_find_channel(); |
1a2449a8 | 1790 | if (tp->ucopy.dma_chan) |
1da177e4 | 1791 | ret = tcp_v4_do_rcv(sk, skb); |
1a2449a8 CL |
1792 | else |
1793 | #endif | |
1794 | { | |
1795 | if (!tcp_prequeue(sk, skb)) | |
ae8d7f88 | 1796 | ret = tcp_v4_do_rcv(sk, skb); |
1a2449a8 | 1797 | } |
da882c1f ED |
1798 | } else if (unlikely(sk_add_backlog(sk, skb, |
1799 | sk->sk_rcvbuf + sk->sk_sndbuf))) { | |
6b03a53a | 1800 | bh_unlock_sock(sk); |
6cce09f8 | 1801 | NET_INC_STATS_BH(net, LINUX_MIB_TCPBACKLOGDROP); |
6b03a53a ZY |
1802 | goto discard_and_relse; |
1803 | } | |
1da177e4 LT |
1804 | bh_unlock_sock(sk); |
1805 | ||
1806 | sock_put(sk); | |
1807 | ||
1808 | return ret; | |
1809 | ||
1810 | no_tcp_socket: | |
1811 | if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) | |
1812 | goto discard_it; | |
1813 | ||
1814 | if (skb->len < (th->doff << 2) || tcp_checksum_complete(skb)) { | |
6a5dc9e5 ED |
1815 | csum_error: |
1816 | TCP_INC_STATS_BH(net, TCP_MIB_CSUMERRORS); | |
1da177e4 | 1817 | bad_packet: |
63231bdd | 1818 | TCP_INC_STATS_BH(net, TCP_MIB_INERRS); |
1da177e4 | 1819 | } else { |
cfb6eeb4 | 1820 | tcp_v4_send_reset(NULL, skb); |
1da177e4 LT |
1821 | } |
1822 | ||
1823 | discard_it: | |
1824 | /* Discard frame. */ | |
1825 | kfree_skb(skb); | |
e905a9ed | 1826 | return 0; |
1da177e4 LT |
1827 | |
1828 | discard_and_relse: | |
1829 | sock_put(sk); | |
1830 | goto discard_it; | |
1831 | ||
1832 | do_time_wait: | |
1833 | if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) { | |
9469c7b4 | 1834 | inet_twsk_put(inet_twsk(sk)); |
1da177e4 LT |
1835 | goto discard_it; |
1836 | } | |
1837 | ||
6a5dc9e5 | 1838 | if (skb->len < (th->doff << 2)) { |
9469c7b4 | 1839 | inet_twsk_put(inet_twsk(sk)); |
6a5dc9e5 ED |
1840 | goto bad_packet; |
1841 | } | |
1842 | if (tcp_checksum_complete(skb)) { | |
1843 | inet_twsk_put(inet_twsk(sk)); | |
1844 | goto csum_error; | |
1da177e4 | 1845 | } |
9469c7b4 | 1846 | switch (tcp_timewait_state_process(inet_twsk(sk), skb, th)) { |
1da177e4 | 1847 | case TCP_TW_SYN: { |
c346dca1 | 1848 | struct sock *sk2 = inet_lookup_listener(dev_net(skb->dev), |
c67499c0 | 1849 | &tcp_hashinfo, |
da5e3630 | 1850 | iph->saddr, th->source, |
eddc9ec5 | 1851 | iph->daddr, th->dest, |
463c84b9 | 1852 | inet_iif(skb)); |
1da177e4 | 1853 | if (sk2) { |
9469c7b4 YH |
1854 | inet_twsk_deschedule(inet_twsk(sk), &tcp_death_row); |
1855 | inet_twsk_put(inet_twsk(sk)); | |
1da177e4 LT |
1856 | sk = sk2; |
1857 | goto process; | |
1858 | } | |
1859 | /* Fall through to ACK */ | |
1860 | } | |
1861 | case TCP_TW_ACK: | |
1862 | tcp_v4_timewait_ack(sk, skb); | |
1863 | break; | |
1864 | case TCP_TW_RST: | |
1865 | goto no_tcp_socket; | |
1866 | case TCP_TW_SUCCESS:; | |
1867 | } | |
1868 | goto discard_it; | |
1869 | } | |
1870 | ||
ccb7c410 DM |
1871 | static struct timewait_sock_ops tcp_timewait_sock_ops = { |
1872 | .twsk_obj_size = sizeof(struct tcp_timewait_sock), | |
1873 | .twsk_unique = tcp_twsk_unique, | |
1874 | .twsk_destructor= tcp_twsk_destructor, | |
ccb7c410 | 1875 | }; |
1da177e4 | 1876 | |
63d02d15 | 1877 | void inet_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb) |
5d299f3d ED |
1878 | { |
1879 | struct dst_entry *dst = skb_dst(skb); | |
1880 | ||
1881 | dst_hold(dst); | |
1882 | sk->sk_rx_dst = dst; | |
1883 | inet_sk(sk)->rx_dst_ifindex = skb->skb_iif; | |
1884 | } | |
63d02d15 | 1885 | EXPORT_SYMBOL(inet_sk_rx_dst_set); |
5d299f3d | 1886 | |
3b401a81 | 1887 | const struct inet_connection_sock_af_ops ipv4_specific = { |
543d9cfe ACM |
1888 | .queue_xmit = ip_queue_xmit, |
1889 | .send_check = tcp_v4_send_check, | |
1890 | .rebuild_header = inet_sk_rebuild_header, | |
5d299f3d | 1891 | .sk_rx_dst_set = inet_sk_rx_dst_set, |
543d9cfe ACM |
1892 | .conn_request = tcp_v4_conn_request, |
1893 | .syn_recv_sock = tcp_v4_syn_recv_sock, | |
543d9cfe ACM |
1894 | .net_header_len = sizeof(struct iphdr), |
1895 | .setsockopt = ip_setsockopt, | |
1896 | .getsockopt = ip_getsockopt, | |
1897 | .addr2sockaddr = inet_csk_addr2sockaddr, | |
1898 | .sockaddr_len = sizeof(struct sockaddr_in), | |
ab1e0a13 | 1899 | .bind_conflict = inet_csk_bind_conflict, |
3fdadf7d | 1900 | #ifdef CONFIG_COMPAT |
543d9cfe ACM |
1901 | .compat_setsockopt = compat_ip_setsockopt, |
1902 | .compat_getsockopt = compat_ip_getsockopt, | |
3fdadf7d | 1903 | #endif |
1da177e4 | 1904 | }; |
4bc2f18b | 1905 | EXPORT_SYMBOL(ipv4_specific); |
1da177e4 | 1906 | |
cfb6eeb4 | 1907 | #ifdef CONFIG_TCP_MD5SIG |
b2e4b3de | 1908 | static const struct tcp_sock_af_ops tcp_sock_ipv4_specific = { |
cfb6eeb4 | 1909 | .md5_lookup = tcp_v4_md5_lookup, |
49a72dfb | 1910 | .calc_md5_hash = tcp_v4_md5_hash_skb, |
cfb6eeb4 | 1911 | .md5_parse = tcp_v4_parse_md5_keys, |
cfb6eeb4 | 1912 | }; |
b6332e6c | 1913 | #endif |
cfb6eeb4 | 1914 | |
1da177e4 LT |
1915 | /* NOTE: A lot of things set to zero explicitly by call to |
1916 | * sk_alloc() so need not be done here. | |
1917 | */ | |
1918 | static int tcp_v4_init_sock(struct sock *sk) | |
1919 | { | |
6687e988 | 1920 | struct inet_connection_sock *icsk = inet_csk(sk); |
1da177e4 | 1921 | |
900f65d3 | 1922 | tcp_init_sock(sk); |
1da177e4 | 1923 | |
8292a17a | 1924 | icsk->icsk_af_ops = &ipv4_specific; |
900f65d3 | 1925 | |
cfb6eeb4 | 1926 | #ifdef CONFIG_TCP_MD5SIG |
ac807fa8 | 1927 | tcp_sk(sk)->af_specific = &tcp_sock_ipv4_specific; |
cfb6eeb4 | 1928 | #endif |
1da177e4 | 1929 | |
1da177e4 LT |
1930 | return 0; |
1931 | } | |
1932 | ||
7d06b2e0 | 1933 | void tcp_v4_destroy_sock(struct sock *sk) |
1da177e4 LT |
1934 | { |
1935 | struct tcp_sock *tp = tcp_sk(sk); | |
1936 | ||
1937 | tcp_clear_xmit_timers(sk); | |
1938 | ||
6687e988 | 1939 | tcp_cleanup_congestion_control(sk); |
317a76f9 | 1940 | |
1da177e4 | 1941 | /* Cleanup up the write buffer. */ |
fe067e8a | 1942 | tcp_write_queue_purge(sk); |
1da177e4 LT |
1943 | |
1944 | /* Cleans up our, hopefully empty, out_of_order_queue. */ | |
e905a9ed | 1945 | __skb_queue_purge(&tp->out_of_order_queue); |
1da177e4 | 1946 | |
cfb6eeb4 YH |
1947 | #ifdef CONFIG_TCP_MD5SIG |
1948 | /* Clean up the MD5 key list, if any */ | |
1949 | if (tp->md5sig_info) { | |
a915da9b | 1950 | tcp_clear_md5_list(sk); |
a8afca03 | 1951 | kfree_rcu(tp->md5sig_info, rcu); |
cfb6eeb4 YH |
1952 | tp->md5sig_info = NULL; |
1953 | } | |
1954 | #endif | |
1955 | ||
1a2449a8 CL |
1956 | #ifdef CONFIG_NET_DMA |
1957 | /* Cleans up our sk_async_wait_queue */ | |
e905a9ed | 1958 | __skb_queue_purge(&sk->sk_async_wait_queue); |
1a2449a8 CL |
1959 | #endif |
1960 | ||
1da177e4 LT |
1961 | /* Clean prequeue, it must be empty really */ |
1962 | __skb_queue_purge(&tp->ucopy.prequeue); | |
1963 | ||
1964 | /* Clean up a referenced TCP bind bucket. */ | |
463c84b9 | 1965 | if (inet_csk(sk)->icsk_bind_hash) |
ab1e0a13 | 1966 | inet_put_port(sk); |
1da177e4 | 1967 | |
168a8f58 | 1968 | BUG_ON(tp->fastopen_rsk != NULL); |
435cf559 | 1969 | |
cf60af03 YC |
1970 | /* If socket is aborted during connect operation */ |
1971 | tcp_free_fastopen_req(tp); | |
1972 | ||
180d8cd9 | 1973 | sk_sockets_allocated_dec(sk); |
d1a4c0b3 | 1974 | sock_release_memcg(sk); |
1da177e4 | 1975 | } |
1da177e4 LT |
1976 | EXPORT_SYMBOL(tcp_v4_destroy_sock); |
1977 | ||
1978 | #ifdef CONFIG_PROC_FS | |
1979 | /* Proc filesystem TCP sock list dumping. */ | |
1980 | ||
a8b690f9 TH |
1981 | /* |
1982 | * Get next listener socket follow cur. If cur is NULL, get first socket | |
1983 | * starting from bucket given in st->bucket; when st->bucket is zero the | |
1984 | * very first socket in the hash table is returned. | |
1985 | */ | |
1da177e4 LT |
1986 | static void *listening_get_next(struct seq_file *seq, void *cur) |
1987 | { | |
463c84b9 | 1988 | struct inet_connection_sock *icsk; |
c25eb3bf | 1989 | struct hlist_nulls_node *node; |
1da177e4 | 1990 | struct sock *sk = cur; |
5caea4ea | 1991 | struct inet_listen_hashbucket *ilb; |
5799de0b | 1992 | struct tcp_iter_state *st = seq->private; |
a4146b1b | 1993 | struct net *net = seq_file_net(seq); |
1da177e4 LT |
1994 | |
1995 | if (!sk) { | |
a8b690f9 | 1996 | ilb = &tcp_hashinfo.listening_hash[st->bucket]; |
5caea4ea | 1997 | spin_lock_bh(&ilb->lock); |
c25eb3bf | 1998 | sk = sk_nulls_head(&ilb->head); |
a8b690f9 | 1999 | st->offset = 0; |
1da177e4 LT |
2000 | goto get_sk; |
2001 | } | |
5caea4ea | 2002 | ilb = &tcp_hashinfo.listening_hash[st->bucket]; |
1da177e4 | 2003 | ++st->num; |
a8b690f9 | 2004 | ++st->offset; |
1da177e4 LT |
2005 | |
2006 | if (st->state == TCP_SEQ_STATE_OPENREQ) { | |
60236fdd | 2007 | struct request_sock *req = cur; |
1da177e4 | 2008 | |
72a3effa | 2009 | icsk = inet_csk(st->syn_wait_sk); |
1da177e4 LT |
2010 | req = req->dl_next; |
2011 | while (1) { | |
2012 | while (req) { | |
bdccc4ca | 2013 | if (req->rsk_ops->family == st->family) { |
1da177e4 LT |
2014 | cur = req; |
2015 | goto out; | |
2016 | } | |
2017 | req = req->dl_next; | |
2018 | } | |
72a3effa | 2019 | if (++st->sbucket >= icsk->icsk_accept_queue.listen_opt->nr_table_entries) |
1da177e4 LT |
2020 | break; |
2021 | get_req: | |
463c84b9 | 2022 | req = icsk->icsk_accept_queue.listen_opt->syn_table[st->sbucket]; |
1da177e4 | 2023 | } |
1bde5ac4 | 2024 | sk = sk_nulls_next(st->syn_wait_sk); |
1da177e4 | 2025 | st->state = TCP_SEQ_STATE_LISTENING; |
463c84b9 | 2026 | read_unlock_bh(&icsk->icsk_accept_queue.syn_wait_lock); |
1da177e4 | 2027 | } else { |
e905a9ed | 2028 | icsk = inet_csk(sk); |
463c84b9 ACM |
2029 | read_lock_bh(&icsk->icsk_accept_queue.syn_wait_lock); |
2030 | if (reqsk_queue_len(&icsk->icsk_accept_queue)) | |
1da177e4 | 2031 | goto start_req; |
463c84b9 | 2032 | read_unlock_bh(&icsk->icsk_accept_queue.syn_wait_lock); |
1bde5ac4 | 2033 | sk = sk_nulls_next(sk); |
1da177e4 LT |
2034 | } |
2035 | get_sk: | |
c25eb3bf | 2036 | sk_nulls_for_each_from(sk, node) { |
8475ef9f PE |
2037 | if (!net_eq(sock_net(sk), net)) |
2038 | continue; | |
2039 | if (sk->sk_family == st->family) { | |
1da177e4 LT |
2040 | cur = sk; |
2041 | goto out; | |
2042 | } | |
e905a9ed | 2043 | icsk = inet_csk(sk); |
463c84b9 ACM |
2044 | read_lock_bh(&icsk->icsk_accept_queue.syn_wait_lock); |
2045 | if (reqsk_queue_len(&icsk->icsk_accept_queue)) { | |
1da177e4 LT |
2046 | start_req: |
2047 | st->uid = sock_i_uid(sk); | |
2048 | st->syn_wait_sk = sk; | |
2049 | st->state = TCP_SEQ_STATE_OPENREQ; | |
2050 | st->sbucket = 0; | |
2051 | goto get_req; | |
2052 | } | |
463c84b9 | 2053 | read_unlock_bh(&icsk->icsk_accept_queue.syn_wait_lock); |
1da177e4 | 2054 | } |
5caea4ea | 2055 | spin_unlock_bh(&ilb->lock); |
a8b690f9 | 2056 | st->offset = 0; |
0f7ff927 | 2057 | if (++st->bucket < INET_LHTABLE_SIZE) { |
5caea4ea ED |
2058 | ilb = &tcp_hashinfo.listening_hash[st->bucket]; |
2059 | spin_lock_bh(&ilb->lock); | |
c25eb3bf | 2060 | sk = sk_nulls_head(&ilb->head); |
1da177e4 LT |
2061 | goto get_sk; |
2062 | } | |
2063 | cur = NULL; | |
2064 | out: | |
2065 | return cur; | |
2066 | } | |
2067 | ||
2068 | static void *listening_get_idx(struct seq_file *seq, loff_t *pos) | |
2069 | { | |
a8b690f9 TH |
2070 | struct tcp_iter_state *st = seq->private; |
2071 | void *rc; | |
2072 | ||
2073 | st->bucket = 0; | |
2074 | st->offset = 0; | |
2075 | rc = listening_get_next(seq, NULL); | |
1da177e4 LT |
2076 | |
2077 | while (rc && *pos) { | |
2078 | rc = listening_get_next(seq, rc); | |
2079 | --*pos; | |
2080 | } | |
2081 | return rc; | |
2082 | } | |
2083 | ||
05dbc7b5 | 2084 | static inline bool empty_bucket(const struct tcp_iter_state *st) |
6eac5604 | 2085 | { |
05dbc7b5 | 2086 | return hlist_nulls_empty(&tcp_hashinfo.ehash[st->bucket].chain); |
6eac5604 AK |
2087 | } |
2088 | ||
a8b690f9 TH |
2089 | /* |
2090 | * Get first established socket starting from bucket given in st->bucket. | |
2091 | * If st->bucket is zero, the very first socket in the hash is returned. | |
2092 | */ | |
1da177e4 LT |
2093 | static void *established_get_first(struct seq_file *seq) |
2094 | { | |
5799de0b | 2095 | struct tcp_iter_state *st = seq->private; |
a4146b1b | 2096 | struct net *net = seq_file_net(seq); |
1da177e4 LT |
2097 | void *rc = NULL; |
2098 | ||
a8b690f9 TH |
2099 | st->offset = 0; |
2100 | for (; st->bucket <= tcp_hashinfo.ehash_mask; ++st->bucket) { | |
1da177e4 | 2101 | struct sock *sk; |
3ab5aee7 | 2102 | struct hlist_nulls_node *node; |
9db66bdc | 2103 | spinlock_t *lock = inet_ehash_lockp(&tcp_hashinfo, st->bucket); |
1da177e4 | 2104 | |
6eac5604 AK |
2105 | /* Lockless fast path for the common case of empty buckets */ |
2106 | if (empty_bucket(st)) | |
2107 | continue; | |
2108 | ||
9db66bdc | 2109 | spin_lock_bh(lock); |
3ab5aee7 | 2110 | sk_nulls_for_each(sk, node, &tcp_hashinfo.ehash[st->bucket].chain) { |
f40c8174 | 2111 | if (sk->sk_family != st->family || |
878628fb | 2112 | !net_eq(sock_net(sk), net)) { |
1da177e4 LT |
2113 | continue; |
2114 | } | |
2115 | rc = sk; | |
2116 | goto out; | |
2117 | } | |
9db66bdc | 2118 | spin_unlock_bh(lock); |
1da177e4 LT |
2119 | } |
2120 | out: | |
2121 | return rc; | |
2122 | } | |
2123 | ||
2124 | static void *established_get_next(struct seq_file *seq, void *cur) | |
2125 | { | |
2126 | struct sock *sk = cur; | |
3ab5aee7 | 2127 | struct hlist_nulls_node *node; |
5799de0b | 2128 | struct tcp_iter_state *st = seq->private; |
a4146b1b | 2129 | struct net *net = seq_file_net(seq); |
1da177e4 LT |
2130 | |
2131 | ++st->num; | |
a8b690f9 | 2132 | ++st->offset; |
1da177e4 | 2133 | |
05dbc7b5 | 2134 | sk = sk_nulls_next(sk); |
1da177e4 | 2135 | |
3ab5aee7 | 2136 | sk_nulls_for_each_from(sk, node) { |
878628fb | 2137 | if (sk->sk_family == st->family && net_eq(sock_net(sk), net)) |
05dbc7b5 | 2138 | return sk; |
1da177e4 LT |
2139 | } |
2140 | ||
05dbc7b5 ED |
2141 | spin_unlock_bh(inet_ehash_lockp(&tcp_hashinfo, st->bucket)); |
2142 | ++st->bucket; | |
2143 | return established_get_first(seq); | |
1da177e4 LT |
2144 | } |
2145 | ||
2146 | static void *established_get_idx(struct seq_file *seq, loff_t pos) | |
2147 | { | |
a8b690f9 TH |
2148 | struct tcp_iter_state *st = seq->private; |
2149 | void *rc; | |
2150 | ||
2151 | st->bucket = 0; | |
2152 | rc = established_get_first(seq); | |
1da177e4 LT |
2153 | |
2154 | while (rc && pos) { | |
2155 | rc = established_get_next(seq, rc); | |
2156 | --pos; | |
7174259e | 2157 | } |
1da177e4 LT |
2158 | return rc; |
2159 | } | |
2160 | ||
2161 | static void *tcp_get_idx(struct seq_file *seq, loff_t pos) | |
2162 | { | |
2163 | void *rc; | |
5799de0b | 2164 | struct tcp_iter_state *st = seq->private; |
1da177e4 | 2165 | |
1da177e4 LT |
2166 | st->state = TCP_SEQ_STATE_LISTENING; |
2167 | rc = listening_get_idx(seq, &pos); | |
2168 | ||
2169 | if (!rc) { | |
1da177e4 LT |
2170 | st->state = TCP_SEQ_STATE_ESTABLISHED; |
2171 | rc = established_get_idx(seq, pos); | |
2172 | } | |
2173 | ||
2174 | return rc; | |
2175 | } | |
2176 | ||
a8b690f9 TH |
2177 | static void *tcp_seek_last_pos(struct seq_file *seq) |
2178 | { | |
2179 | struct tcp_iter_state *st = seq->private; | |
2180 | int offset = st->offset; | |
2181 | int orig_num = st->num; | |
2182 | void *rc = NULL; | |
2183 | ||
2184 | switch (st->state) { | |
2185 | case TCP_SEQ_STATE_OPENREQ: | |
2186 | case TCP_SEQ_STATE_LISTENING: | |
2187 | if (st->bucket >= INET_LHTABLE_SIZE) | |
2188 | break; | |
2189 | st->state = TCP_SEQ_STATE_LISTENING; | |
2190 | rc = listening_get_next(seq, NULL); | |
2191 | while (offset-- && rc) | |
2192 | rc = listening_get_next(seq, rc); | |
2193 | if (rc) | |
2194 | break; | |
2195 | st->bucket = 0; | |
05dbc7b5 | 2196 | st->state = TCP_SEQ_STATE_ESTABLISHED; |
a8b690f9 TH |
2197 | /* Fallthrough */ |
2198 | case TCP_SEQ_STATE_ESTABLISHED: | |
a8b690f9 TH |
2199 | if (st->bucket > tcp_hashinfo.ehash_mask) |
2200 | break; | |
2201 | rc = established_get_first(seq); | |
2202 | while (offset-- && rc) | |
2203 | rc = established_get_next(seq, rc); | |
2204 | } | |
2205 | ||
2206 | st->num = orig_num; | |
2207 | ||
2208 | return rc; | |
2209 | } | |
2210 | ||
1da177e4 LT |
2211 | static void *tcp_seq_start(struct seq_file *seq, loff_t *pos) |
2212 | { | |
5799de0b | 2213 | struct tcp_iter_state *st = seq->private; |
a8b690f9 TH |
2214 | void *rc; |
2215 | ||
2216 | if (*pos && *pos == st->last_pos) { | |
2217 | rc = tcp_seek_last_pos(seq); | |
2218 | if (rc) | |
2219 | goto out; | |
2220 | } | |
2221 | ||
1da177e4 LT |
2222 | st->state = TCP_SEQ_STATE_LISTENING; |
2223 | st->num = 0; | |
a8b690f9 TH |
2224 | st->bucket = 0; |
2225 | st->offset = 0; | |
2226 | rc = *pos ? tcp_get_idx(seq, *pos - 1) : SEQ_START_TOKEN; | |
2227 | ||
2228 | out: | |
2229 | st->last_pos = *pos; | |
2230 | return rc; | |
1da177e4 LT |
2231 | } |
2232 | ||
2233 | static void *tcp_seq_next(struct seq_file *seq, void *v, loff_t *pos) | |
2234 | { | |
a8b690f9 | 2235 | struct tcp_iter_state *st = seq->private; |
1da177e4 | 2236 | void *rc = NULL; |
1da177e4 LT |
2237 | |
2238 | if (v == SEQ_START_TOKEN) { | |
2239 | rc = tcp_get_idx(seq, 0); | |
2240 | goto out; | |
2241 | } | |
1da177e4 LT |
2242 | |
2243 | switch (st->state) { | |
2244 | case TCP_SEQ_STATE_OPENREQ: | |
2245 | case TCP_SEQ_STATE_LISTENING: | |
2246 | rc = listening_get_next(seq, v); | |
2247 | if (!rc) { | |
1da177e4 | 2248 | st->state = TCP_SEQ_STATE_ESTABLISHED; |
a8b690f9 TH |
2249 | st->bucket = 0; |
2250 | st->offset = 0; | |
1da177e4 LT |
2251 | rc = established_get_first(seq); |
2252 | } | |
2253 | break; | |
2254 | case TCP_SEQ_STATE_ESTABLISHED: | |
1da177e4 LT |
2255 | rc = established_get_next(seq, v); |
2256 | break; | |
2257 | } | |
2258 | out: | |
2259 | ++*pos; | |
a8b690f9 | 2260 | st->last_pos = *pos; |
1da177e4 LT |
2261 | return rc; |
2262 | } | |
2263 | ||
2264 | static void tcp_seq_stop(struct seq_file *seq, void *v) | |
2265 | { | |
5799de0b | 2266 | struct tcp_iter_state *st = seq->private; |
1da177e4 LT |
2267 | |
2268 | switch (st->state) { | |
2269 | case TCP_SEQ_STATE_OPENREQ: | |
2270 | if (v) { | |
463c84b9 ACM |
2271 | struct inet_connection_sock *icsk = inet_csk(st->syn_wait_sk); |
2272 | read_unlock_bh(&icsk->icsk_accept_queue.syn_wait_lock); | |
1da177e4 LT |
2273 | } |
2274 | case TCP_SEQ_STATE_LISTENING: | |
2275 | if (v != SEQ_START_TOKEN) | |
5caea4ea | 2276 | spin_unlock_bh(&tcp_hashinfo.listening_hash[st->bucket].lock); |
1da177e4 | 2277 | break; |
1da177e4 LT |
2278 | case TCP_SEQ_STATE_ESTABLISHED: |
2279 | if (v) | |
9db66bdc | 2280 | spin_unlock_bh(inet_ehash_lockp(&tcp_hashinfo, st->bucket)); |
1da177e4 LT |
2281 | break; |
2282 | } | |
2283 | } | |
2284 | ||
73cb88ec | 2285 | int tcp_seq_open(struct inode *inode, struct file *file) |
1da177e4 | 2286 | { |
d9dda78b | 2287 | struct tcp_seq_afinfo *afinfo = PDE_DATA(inode); |
1da177e4 | 2288 | struct tcp_iter_state *s; |
52d6f3f1 | 2289 | int err; |
1da177e4 | 2290 | |
52d6f3f1 DL |
2291 | err = seq_open_net(inode, file, &afinfo->seq_ops, |
2292 | sizeof(struct tcp_iter_state)); | |
2293 | if (err < 0) | |
2294 | return err; | |
f40c8174 | 2295 | |
52d6f3f1 | 2296 | s = ((struct seq_file *)file->private_data)->private; |
1da177e4 | 2297 | s->family = afinfo->family; |
a8b690f9 | 2298 | s->last_pos = 0; |
f40c8174 DL |
2299 | return 0; |
2300 | } | |
73cb88ec | 2301 | EXPORT_SYMBOL(tcp_seq_open); |
f40c8174 | 2302 | |
6f8b13bc | 2303 | int tcp_proc_register(struct net *net, struct tcp_seq_afinfo *afinfo) |
1da177e4 LT |
2304 | { |
2305 | int rc = 0; | |
2306 | struct proc_dir_entry *p; | |
2307 | ||
9427c4b3 DL |
2308 | afinfo->seq_ops.start = tcp_seq_start; |
2309 | afinfo->seq_ops.next = tcp_seq_next; | |
2310 | afinfo->seq_ops.stop = tcp_seq_stop; | |
2311 | ||
84841c3c | 2312 | p = proc_create_data(afinfo->name, S_IRUGO, net->proc_net, |
73cb88ec | 2313 | afinfo->seq_fops, afinfo); |
84841c3c | 2314 | if (!p) |
1da177e4 LT |
2315 | rc = -ENOMEM; |
2316 | return rc; | |
2317 | } | |
4bc2f18b | 2318 | EXPORT_SYMBOL(tcp_proc_register); |
1da177e4 | 2319 | |
6f8b13bc | 2320 | void tcp_proc_unregister(struct net *net, struct tcp_seq_afinfo *afinfo) |
1da177e4 | 2321 | { |
ece31ffd | 2322 | remove_proc_entry(afinfo->name, net->proc_net); |
1da177e4 | 2323 | } |
4bc2f18b | 2324 | EXPORT_SYMBOL(tcp_proc_unregister); |
1da177e4 | 2325 | |
cf533ea5 | 2326 | static void get_openreq4(const struct sock *sk, const struct request_sock *req, |
652586df | 2327 | struct seq_file *f, int i, kuid_t uid) |
1da177e4 | 2328 | { |
2e6599cb | 2329 | const struct inet_request_sock *ireq = inet_rsk(req); |
a399a805 | 2330 | long delta = req->expires - jiffies; |
1da177e4 | 2331 | |
5e659e4c | 2332 | seq_printf(f, "%4d: %08X:%04X %08X:%04X" |
652586df | 2333 | " %02X %08X:%08X %02X:%08lX %08X %5u %8d %u %d %pK", |
1da177e4 | 2334 | i, |
634fb979 | 2335 | ireq->ir_loc_addr, |
c720c7e8 | 2336 | ntohs(inet_sk(sk)->inet_sport), |
634fb979 ED |
2337 | ireq->ir_rmt_addr, |
2338 | ntohs(ireq->ir_rmt_port), | |
1da177e4 LT |
2339 | TCP_SYN_RECV, |
2340 | 0, 0, /* could print option size, but that is af dependent. */ | |
2341 | 1, /* timers active (only the expire timer) */ | |
a399a805 | 2342 | jiffies_delta_to_clock_t(delta), |
e6c022a4 | 2343 | req->num_timeout, |
a7cb5a49 | 2344 | from_kuid_munged(seq_user_ns(f), uid), |
1da177e4 LT |
2345 | 0, /* non standard timer */ |
2346 | 0, /* open_requests have no inode */ | |
2347 | atomic_read(&sk->sk_refcnt), | |
652586df | 2348 | req); |
1da177e4 LT |
2349 | } |
2350 | ||
652586df | 2351 | static void get_tcp4_sock(struct sock *sk, struct seq_file *f, int i) |
1da177e4 LT |
2352 | { |
2353 | int timer_active; | |
2354 | unsigned long timer_expires; | |
cf533ea5 | 2355 | const struct tcp_sock *tp = tcp_sk(sk); |
cf4c6bf8 | 2356 | const struct inet_connection_sock *icsk = inet_csk(sk); |
cf533ea5 | 2357 | const struct inet_sock *inet = inet_sk(sk); |
168a8f58 | 2358 | struct fastopen_queue *fastopenq = icsk->icsk_accept_queue.fastopenq; |
c720c7e8 ED |
2359 | __be32 dest = inet->inet_daddr; |
2360 | __be32 src = inet->inet_rcv_saddr; | |
2361 | __u16 destp = ntohs(inet->inet_dport); | |
2362 | __u16 srcp = ntohs(inet->inet_sport); | |
49d09007 | 2363 | int rx_queue; |
1da177e4 | 2364 | |
6ba8a3b1 ND |
2365 | if (icsk->icsk_pending == ICSK_TIME_RETRANS || |
2366 | icsk->icsk_pending == ICSK_TIME_EARLY_RETRANS || | |
2367 | icsk->icsk_pending == ICSK_TIME_LOSS_PROBE) { | |
1da177e4 | 2368 | timer_active = 1; |
463c84b9 ACM |
2369 | timer_expires = icsk->icsk_timeout; |
2370 | } else if (icsk->icsk_pending == ICSK_TIME_PROBE0) { | |
1da177e4 | 2371 | timer_active = 4; |
463c84b9 | 2372 | timer_expires = icsk->icsk_timeout; |
cf4c6bf8 | 2373 | } else if (timer_pending(&sk->sk_timer)) { |
1da177e4 | 2374 | timer_active = 2; |
cf4c6bf8 | 2375 | timer_expires = sk->sk_timer.expires; |
1da177e4 LT |
2376 | } else { |
2377 | timer_active = 0; | |
2378 | timer_expires = jiffies; | |
2379 | } | |
2380 | ||
49d09007 ED |
2381 | if (sk->sk_state == TCP_LISTEN) |
2382 | rx_queue = sk->sk_ack_backlog; | |
2383 | else | |
2384 | /* | |
2385 | * because we dont lock socket, we might find a transient negative value | |
2386 | */ | |
2387 | rx_queue = max_t(int, tp->rcv_nxt - tp->copied_seq, 0); | |
2388 | ||
5e659e4c | 2389 | seq_printf(f, "%4d: %08X:%04X %08X:%04X %02X %08X:%08X %02X:%08lX " |
652586df | 2390 | "%08X %5u %8d %lu %d %pK %lu %lu %u %u %d", |
cf4c6bf8 | 2391 | i, src, srcp, dest, destp, sk->sk_state, |
47da8ee6 | 2392 | tp->write_seq - tp->snd_una, |
49d09007 | 2393 | rx_queue, |
1da177e4 | 2394 | timer_active, |
a399a805 | 2395 | jiffies_delta_to_clock_t(timer_expires - jiffies), |
463c84b9 | 2396 | icsk->icsk_retransmits, |
a7cb5a49 | 2397 | from_kuid_munged(seq_user_ns(f), sock_i_uid(sk)), |
6687e988 | 2398 | icsk->icsk_probes_out, |
cf4c6bf8 IJ |
2399 | sock_i_ino(sk), |
2400 | atomic_read(&sk->sk_refcnt), sk, | |
7be87351 SH |
2401 | jiffies_to_clock_t(icsk->icsk_rto), |
2402 | jiffies_to_clock_t(icsk->icsk_ack.ato), | |
463c84b9 | 2403 | (icsk->icsk_ack.quick << 1) | icsk->icsk_ack.pingpong, |
1da177e4 | 2404 | tp->snd_cwnd, |
168a8f58 JC |
2405 | sk->sk_state == TCP_LISTEN ? |
2406 | (fastopenq ? fastopenq->max_qlen : 0) : | |
652586df | 2407 | (tcp_in_initial_slowstart(tp) ? -1 : tp->snd_ssthresh)); |
1da177e4 LT |
2408 | } |
2409 | ||
cf533ea5 | 2410 | static void get_timewait4_sock(const struct inet_timewait_sock *tw, |
652586df | 2411 | struct seq_file *f, int i) |
1da177e4 | 2412 | { |
23f33c2d | 2413 | __be32 dest, src; |
1da177e4 | 2414 | __u16 destp, srcp; |
e2a1d3e4 | 2415 | s32 delta = tw->tw_ttd - inet_tw_time_stamp(); |
1da177e4 LT |
2416 | |
2417 | dest = tw->tw_daddr; | |
2418 | src = tw->tw_rcv_saddr; | |
2419 | destp = ntohs(tw->tw_dport); | |
2420 | srcp = ntohs(tw->tw_sport); | |
2421 | ||
5e659e4c | 2422 | seq_printf(f, "%4d: %08X:%04X %08X:%04X" |
652586df | 2423 | " %02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %pK", |
1da177e4 | 2424 | i, src, srcp, dest, destp, tw->tw_substate, 0, 0, |
a399a805 | 2425 | 3, jiffies_delta_to_clock_t(delta), 0, 0, 0, 0, |
652586df | 2426 | atomic_read(&tw->tw_refcnt), tw); |
1da177e4 LT |
2427 | } |
2428 | ||
2429 | #define TMPSZ 150 | |
2430 | ||
2431 | static int tcp4_seq_show(struct seq_file *seq, void *v) | |
2432 | { | |
5799de0b | 2433 | struct tcp_iter_state *st; |
05dbc7b5 | 2434 | struct sock *sk = v; |
1da177e4 | 2435 | |
652586df | 2436 | seq_setwidth(seq, TMPSZ - 1); |
1da177e4 | 2437 | if (v == SEQ_START_TOKEN) { |
652586df | 2438 | seq_puts(seq, " sl local_address rem_address st tx_queue " |
1da177e4 LT |
2439 | "rx_queue tr tm->when retrnsmt uid timeout " |
2440 | "inode"); | |
2441 | goto out; | |
2442 | } | |
2443 | st = seq->private; | |
2444 | ||
2445 | switch (st->state) { | |
2446 | case TCP_SEQ_STATE_LISTENING: | |
2447 | case TCP_SEQ_STATE_ESTABLISHED: | |
05dbc7b5 | 2448 | if (sk->sk_state == TCP_TIME_WAIT) |
652586df | 2449 | get_timewait4_sock(v, seq, st->num); |
05dbc7b5 | 2450 | else |
652586df | 2451 | get_tcp4_sock(v, seq, st->num); |
1da177e4 LT |
2452 | break; |
2453 | case TCP_SEQ_STATE_OPENREQ: | |
652586df | 2454 | get_openreq4(st->syn_wait_sk, v, seq, st->num, st->uid); |
1da177e4 LT |
2455 | break; |
2456 | } | |
1da177e4 | 2457 | out: |
652586df | 2458 | seq_pad(seq, '\n'); |
1da177e4 LT |
2459 | return 0; |
2460 | } | |
2461 | ||
73cb88ec AV |
2462 | static const struct file_operations tcp_afinfo_seq_fops = { |
2463 | .owner = THIS_MODULE, | |
2464 | .open = tcp_seq_open, | |
2465 | .read = seq_read, | |
2466 | .llseek = seq_lseek, | |
2467 | .release = seq_release_net | |
2468 | }; | |
2469 | ||
1da177e4 | 2470 | static struct tcp_seq_afinfo tcp4_seq_afinfo = { |
1da177e4 LT |
2471 | .name = "tcp", |
2472 | .family = AF_INET, | |
73cb88ec | 2473 | .seq_fops = &tcp_afinfo_seq_fops, |
9427c4b3 DL |
2474 | .seq_ops = { |
2475 | .show = tcp4_seq_show, | |
2476 | }, | |
1da177e4 LT |
2477 | }; |
2478 | ||
2c8c1e72 | 2479 | static int __net_init tcp4_proc_init_net(struct net *net) |
757764f6 PE |
2480 | { |
2481 | return tcp_proc_register(net, &tcp4_seq_afinfo); | |
2482 | } | |
2483 | ||
2c8c1e72 | 2484 | static void __net_exit tcp4_proc_exit_net(struct net *net) |
757764f6 PE |
2485 | { |
2486 | tcp_proc_unregister(net, &tcp4_seq_afinfo); | |
2487 | } | |
2488 | ||
2489 | static struct pernet_operations tcp4_net_ops = { | |
2490 | .init = tcp4_proc_init_net, | |
2491 | .exit = tcp4_proc_exit_net, | |
2492 | }; | |
2493 | ||
1da177e4 LT |
2494 | int __init tcp4_proc_init(void) |
2495 | { | |
757764f6 | 2496 | return register_pernet_subsys(&tcp4_net_ops); |
1da177e4 LT |
2497 | } |
2498 | ||
2499 | void tcp4_proc_exit(void) | |
2500 | { | |
757764f6 | 2501 | unregister_pernet_subsys(&tcp4_net_ops); |
1da177e4 LT |
2502 | } |
2503 | #endif /* CONFIG_PROC_FS */ | |
2504 | ||
2505 | struct proto tcp_prot = { | |
2506 | .name = "TCP", | |
2507 | .owner = THIS_MODULE, | |
2508 | .close = tcp_close, | |
2509 | .connect = tcp_v4_connect, | |
2510 | .disconnect = tcp_disconnect, | |
463c84b9 | 2511 | .accept = inet_csk_accept, |
1da177e4 LT |
2512 | .ioctl = tcp_ioctl, |
2513 | .init = tcp_v4_init_sock, | |
2514 | .destroy = tcp_v4_destroy_sock, | |
2515 | .shutdown = tcp_shutdown, | |
2516 | .setsockopt = tcp_setsockopt, | |
2517 | .getsockopt = tcp_getsockopt, | |
1da177e4 | 2518 | .recvmsg = tcp_recvmsg, |
7ba42910 CG |
2519 | .sendmsg = tcp_sendmsg, |
2520 | .sendpage = tcp_sendpage, | |
1da177e4 | 2521 | .backlog_rcv = tcp_v4_do_rcv, |
46d3ceab | 2522 | .release_cb = tcp_release_cb, |
563d34d0 | 2523 | .mtu_reduced = tcp_v4_mtu_reduced, |
ab1e0a13 ACM |
2524 | .hash = inet_hash, |
2525 | .unhash = inet_unhash, | |
2526 | .get_port = inet_csk_get_port, | |
1da177e4 | 2527 | .enter_memory_pressure = tcp_enter_memory_pressure, |
c9bee3b7 | 2528 | .stream_memory_free = tcp_stream_memory_free, |
1da177e4 | 2529 | .sockets_allocated = &tcp_sockets_allocated, |
0a5578cf | 2530 | .orphan_count = &tcp_orphan_count, |
1da177e4 LT |
2531 | .memory_allocated = &tcp_memory_allocated, |
2532 | .memory_pressure = &tcp_memory_pressure, | |
a4fe34bf | 2533 | .sysctl_mem = sysctl_tcp_mem, |
1da177e4 LT |
2534 | .sysctl_wmem = sysctl_tcp_wmem, |
2535 | .sysctl_rmem = sysctl_tcp_rmem, | |
2536 | .max_header = MAX_TCP_HEADER, | |
2537 | .obj_size = sizeof(struct tcp_sock), | |
3ab5aee7 | 2538 | .slab_flags = SLAB_DESTROY_BY_RCU, |
6d6ee43e | 2539 | .twsk_prot = &tcp_timewait_sock_ops, |
60236fdd | 2540 | .rsk_prot = &tcp_request_sock_ops, |
39d8cda7 | 2541 | .h.hashinfo = &tcp_hashinfo, |
7ba42910 | 2542 | .no_autobind = true, |
543d9cfe ACM |
2543 | #ifdef CONFIG_COMPAT |
2544 | .compat_setsockopt = compat_tcp_setsockopt, | |
2545 | .compat_getsockopt = compat_tcp_getsockopt, | |
2546 | #endif | |
c255a458 | 2547 | #ifdef CONFIG_MEMCG_KMEM |
d1a4c0b3 GC |
2548 | .init_cgroup = tcp_init_cgroup, |
2549 | .destroy_cgroup = tcp_destroy_cgroup, | |
2550 | .proto_cgroup = tcp_proto_cgroup, | |
2551 | #endif | |
1da177e4 | 2552 | }; |
4bc2f18b | 2553 | EXPORT_SYMBOL(tcp_prot); |
1da177e4 | 2554 | |
046ee902 DL |
2555 | static int __net_init tcp_sk_init(struct net *net) |
2556 | { | |
5d134f1c | 2557 | net->ipv4.sysctl_tcp_ecn = 2; |
be9f4a44 | 2558 | return 0; |
046ee902 DL |
2559 | } |
2560 | ||
2561 | static void __net_exit tcp_sk_exit(struct net *net) | |
2562 | { | |
b099ce26 EB |
2563 | } |
2564 | ||
2565 | static void __net_exit tcp_sk_exit_batch(struct list_head *net_exit_list) | |
2566 | { | |
2567 | inet_twsk_purge(&tcp_hashinfo, &tcp_death_row, AF_INET); | |
046ee902 DL |
2568 | } |
2569 | ||
2570 | static struct pernet_operations __net_initdata tcp_sk_ops = { | |
b099ce26 EB |
2571 | .init = tcp_sk_init, |
2572 | .exit = tcp_sk_exit, | |
2573 | .exit_batch = tcp_sk_exit_batch, | |
046ee902 DL |
2574 | }; |
2575 | ||
9b0f976f | 2576 | void __init tcp_v4_init(void) |
1da177e4 | 2577 | { |
5caea4ea | 2578 | inet_hashinfo_init(&tcp_hashinfo); |
6a1b3054 | 2579 | if (register_pernet_subsys(&tcp_sk_ops)) |
1da177e4 | 2580 | panic("Failed to create the TCP control socket.\n"); |
1da177e4 | 2581 | } |