]>
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> |
6e5714ea | 75 | #include <net/secure_seq.h> |
076bb0c8 | 76 | #include <net/busy_poll.h> |
1da177e4 LT |
77 | |
78 | #include <linux/inet.h> | |
79 | #include <linux/ipv6.h> | |
80 | #include <linux/stddef.h> | |
81 | #include <linux/proc_fs.h> | |
82 | #include <linux/seq_file.h> | |
83 | ||
cf80e0e4 | 84 | #include <crypto/hash.h> |
cfb6eeb4 YH |
85 | #include <linux/scatterlist.h> |
86 | ||
ab32ea5d | 87 | int sysctl_tcp_low_latency __read_mostly; |
1da177e4 | 88 | |
cfb6eeb4 | 89 | #ifdef CONFIG_TCP_MD5SIG |
a915da9b | 90 | static int tcp_v4_md5_hash_hdr(char *md5_hash, const struct tcp_md5sig_key *key, |
318cf7aa | 91 | __be32 daddr, __be32 saddr, const struct tcphdr *th); |
cfb6eeb4 YH |
92 | #endif |
93 | ||
5caea4ea | 94 | struct inet_hashinfo tcp_hashinfo; |
4bc2f18b | 95 | EXPORT_SYMBOL(tcp_hashinfo); |
1da177e4 | 96 | |
95a22cae | 97 | static u32 tcp_v4_init_sequence(const struct sk_buff *skb, u32 *tsoff) |
1da177e4 | 98 | { |
eddc9ec5 ACM |
99 | return secure_tcp_sequence_number(ip_hdr(skb)->daddr, |
100 | ip_hdr(skb)->saddr, | |
aa8223c7 | 101 | tcp_hdr(skb)->dest, |
95a22cae | 102 | tcp_hdr(skb)->source, tsoff); |
1da177e4 LT |
103 | } |
104 | ||
6d6ee43e ACM |
105 | int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp) |
106 | { | |
107 | const struct tcp_timewait_sock *tcptw = tcp_twsk(sktw); | |
108 | struct tcp_sock *tp = tcp_sk(sk); | |
109 | ||
110 | /* With PAWS, it is safe from the viewpoint | |
111 | of data integrity. Even without PAWS it is safe provided sequence | |
112 | spaces do not overlap i.e. at data rates <= 80Mbit/sec. | |
113 | ||
114 | Actually, the idea is close to VJ's one, only timestamp cache is | |
115 | held not per host, but per port pair and TW bucket is used as state | |
116 | holder. | |
117 | ||
118 | If TW bucket has been already destroyed we fall back to VJ's scheme | |
119 | and use initial timestamp retrieved from peer table. | |
120 | */ | |
121 | if (tcptw->tw_ts_recent_stamp && | |
56ab6b93 | 122 | (!twp || (sock_net(sk)->ipv4.sysctl_tcp_tw_reuse && |
9d729f72 | 123 | get_seconds() - tcptw->tw_ts_recent_stamp > 1))) { |
6d6ee43e ACM |
124 | tp->write_seq = tcptw->tw_snd_nxt + 65535 + 2; |
125 | if (tp->write_seq == 0) | |
126 | tp->write_seq = 1; | |
127 | tp->rx_opt.ts_recent = tcptw->tw_ts_recent; | |
128 | tp->rx_opt.ts_recent_stamp = tcptw->tw_ts_recent_stamp; | |
129 | sock_hold(sktw); | |
130 | return 1; | |
131 | } | |
132 | ||
133 | return 0; | |
134 | } | |
6d6ee43e ACM |
135 | EXPORT_SYMBOL_GPL(tcp_twsk_unique); |
136 | ||
1da177e4 LT |
137 | /* This will initiate an outgoing connection. */ |
138 | int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len) | |
139 | { | |
2d7192d6 | 140 | struct sockaddr_in *usin = (struct sockaddr_in *)uaddr; |
1da177e4 LT |
141 | struct inet_sock *inet = inet_sk(sk); |
142 | struct tcp_sock *tp = tcp_sk(sk); | |
dca8b089 | 143 | __be16 orig_sport, orig_dport; |
bada8adc | 144 | __be32 daddr, nexthop; |
da905bd1 | 145 | struct flowi4 *fl4; |
2d7192d6 | 146 | struct rtable *rt; |
1da177e4 | 147 | int err; |
f6d8bd05 | 148 | struct ip_options_rcu *inet_opt; |
1da177e4 LT |
149 | |
150 | if (addr_len < sizeof(struct sockaddr_in)) | |
151 | return -EINVAL; | |
152 | ||
153 | if (usin->sin_family != AF_INET) | |
154 | return -EAFNOSUPPORT; | |
155 | ||
156 | nexthop = daddr = usin->sin_addr.s_addr; | |
f6d8bd05 | 157 | inet_opt = rcu_dereference_protected(inet->inet_opt, |
1e1d04e6 | 158 | lockdep_sock_is_held(sk)); |
f6d8bd05 | 159 | if (inet_opt && inet_opt->opt.srr) { |
1da177e4 LT |
160 | if (!daddr) |
161 | return -EINVAL; | |
f6d8bd05 | 162 | nexthop = inet_opt->opt.faddr; |
1da177e4 LT |
163 | } |
164 | ||
dca8b089 DM |
165 | orig_sport = inet->inet_sport; |
166 | orig_dport = usin->sin_port; | |
da905bd1 DM |
167 | fl4 = &inet->cork.fl.u.ip4; |
168 | rt = ip_route_connect(fl4, nexthop, inet->inet_saddr, | |
b23dd4fe DM |
169 | RT_CONN_FLAGS(sk), sk->sk_bound_dev_if, |
170 | IPPROTO_TCP, | |
0e0d44ab | 171 | orig_sport, orig_dport, sk); |
b23dd4fe DM |
172 | if (IS_ERR(rt)) { |
173 | err = PTR_ERR(rt); | |
174 | if (err == -ENETUNREACH) | |
f1d8cba6 | 175 | IP_INC_STATS(sock_net(sk), IPSTATS_MIB_OUTNOROUTES); |
b23dd4fe | 176 | return err; |
584bdf8c | 177 | } |
1da177e4 LT |
178 | |
179 | if (rt->rt_flags & (RTCF_MULTICAST | RTCF_BROADCAST)) { | |
180 | ip_rt_put(rt); | |
181 | return -ENETUNREACH; | |
182 | } | |
183 | ||
f6d8bd05 | 184 | if (!inet_opt || !inet_opt->opt.srr) |
da905bd1 | 185 | daddr = fl4->daddr; |
1da177e4 | 186 | |
c720c7e8 | 187 | if (!inet->inet_saddr) |
da905bd1 | 188 | inet->inet_saddr = fl4->saddr; |
d1e559d0 | 189 | sk_rcv_saddr_set(sk, inet->inet_saddr); |
1da177e4 | 190 | |
c720c7e8 | 191 | if (tp->rx_opt.ts_recent_stamp && inet->inet_daddr != daddr) { |
1da177e4 LT |
192 | /* Reset inherited state */ |
193 | tp->rx_opt.ts_recent = 0; | |
194 | tp->rx_opt.ts_recent_stamp = 0; | |
ee995283 PE |
195 | if (likely(!tp->repair)) |
196 | tp->write_seq = 0; | |
1da177e4 LT |
197 | } |
198 | ||
295ff7ed | 199 | if (tcp_death_row.sysctl_tw_recycle && |
81166dd6 DM |
200 | !tp->rx_opt.ts_recent_stamp && fl4->daddr == daddr) |
201 | tcp_fetch_timewait_stamp(sk, &rt->dst); | |
1da177e4 | 202 | |
c720c7e8 | 203 | inet->inet_dport = usin->sin_port; |
d1e559d0 | 204 | sk_daddr_set(sk, daddr); |
1da177e4 | 205 | |
d83d8461 | 206 | inet_csk(sk)->icsk_ext_hdr_len = 0; |
f6d8bd05 ED |
207 | if (inet_opt) |
208 | inet_csk(sk)->icsk_ext_hdr_len = inet_opt->opt.optlen; | |
1da177e4 | 209 | |
bee7ca9e | 210 | tp->rx_opt.mss_clamp = TCP_MSS_DEFAULT; |
1da177e4 LT |
211 | |
212 | /* Socket identity is still unknown (sport may be zero). | |
213 | * However we set state to SYN-SENT and not releasing socket | |
214 | * lock select source port, enter ourselves into the hash tables and | |
215 | * complete initialization after this. | |
216 | */ | |
217 | tcp_set_state(sk, TCP_SYN_SENT); | |
a7f5e7f1 | 218 | err = inet_hash_connect(&tcp_death_row, sk); |
1da177e4 LT |
219 | if (err) |
220 | goto failure; | |
221 | ||
877d1f62 | 222 | sk_set_txhash(sk); |
9e7ceb06 | 223 | |
da905bd1 | 224 | rt = ip_route_newports(fl4, rt, orig_sport, orig_dport, |
b23dd4fe DM |
225 | inet->inet_sport, inet->inet_dport, sk); |
226 | if (IS_ERR(rt)) { | |
227 | err = PTR_ERR(rt); | |
228 | rt = NULL; | |
1da177e4 | 229 | goto failure; |
b23dd4fe | 230 | } |
1da177e4 | 231 | /* OK, now commit destination to socket. */ |
bcd76111 | 232 | sk->sk_gso_type = SKB_GSO_TCPV4; |
d8d1f30b | 233 | sk_setup_caps(sk, &rt->dst); |
1da177e4 | 234 | |
ee995283 | 235 | if (!tp->write_seq && likely(!tp->repair)) |
c720c7e8 ED |
236 | tp->write_seq = secure_tcp_sequence_number(inet->inet_saddr, |
237 | inet->inet_daddr, | |
238 | inet->inet_sport, | |
95a22cae FW |
239 | usin->sin_port, |
240 | &tp->tsoffset); | |
1da177e4 | 241 | |
c720c7e8 | 242 | inet->inet_id = tp->write_seq ^ jiffies; |
1da177e4 | 243 | |
2b916477 | 244 | err = tcp_connect(sk); |
ee995283 | 245 | |
1da177e4 LT |
246 | rt = NULL; |
247 | if (err) | |
248 | goto failure; | |
249 | ||
250 | return 0; | |
251 | ||
252 | failure: | |
7174259e ACM |
253 | /* |
254 | * This unhashes the socket and releases the local port, | |
255 | * if necessary. | |
256 | */ | |
1da177e4 LT |
257 | tcp_set_state(sk, TCP_CLOSE); |
258 | ip_rt_put(rt); | |
259 | sk->sk_route_caps = 0; | |
c720c7e8 | 260 | inet->inet_dport = 0; |
1da177e4 LT |
261 | return err; |
262 | } | |
4bc2f18b | 263 | EXPORT_SYMBOL(tcp_v4_connect); |
1da177e4 | 264 | |
1da177e4 | 265 | /* |
563d34d0 ED |
266 | * This routine reacts to ICMP_FRAG_NEEDED mtu indications as defined in RFC1191. |
267 | * It can be called through tcp_release_cb() if socket was owned by user | |
268 | * at the time tcp_v4_err() was called to handle ICMP message. | |
1da177e4 | 269 | */ |
4fab9071 | 270 | void tcp_v4_mtu_reduced(struct sock *sk) |
1da177e4 | 271 | { |
1da177e4 | 272 | struct inet_sock *inet = inet_sk(sk); |
76d06ff4 ED |
273 | struct dst_entry *dst; |
274 | u32 mtu; | |
1da177e4 | 275 | |
76d06ff4 ED |
276 | if ((1 << sk->sk_state) & (TCPF_LISTEN | TCPF_CLOSE)) |
277 | return; | |
278 | mtu = tcp_sk(sk)->mtu_info; | |
80d0a69f DM |
279 | dst = inet_csk_update_pmtu(sk, mtu); |
280 | if (!dst) | |
1da177e4 LT |
281 | return; |
282 | ||
1da177e4 LT |
283 | /* Something is about to be wrong... Remember soft error |
284 | * for the case, if this connection will not able to recover. | |
285 | */ | |
286 | if (mtu < dst_mtu(dst) && ip_dont_fragment(sk, dst)) | |
287 | sk->sk_err_soft = EMSGSIZE; | |
288 | ||
289 | mtu = dst_mtu(dst); | |
290 | ||
291 | if (inet->pmtudisc != IP_PMTUDISC_DONT && | |
482fc609 | 292 | ip_sk_accept_pmtu(sk) && |
d83d8461 | 293 | inet_csk(sk)->icsk_pmtu_cookie > mtu) { |
1da177e4 LT |
294 | tcp_sync_mss(sk, mtu); |
295 | ||
296 | /* Resend the TCP packet because it's | |
297 | * clear that the old packet has been | |
298 | * dropped. This is the new "fast" path mtu | |
299 | * discovery. | |
300 | */ | |
301 | tcp_simple_retransmit(sk); | |
302 | } /* else let the usual retransmit timer handle it */ | |
303 | } | |
4fab9071 | 304 | EXPORT_SYMBOL(tcp_v4_mtu_reduced); |
1da177e4 | 305 | |
55be7a9c DM |
306 | static void do_redirect(struct sk_buff *skb, struct sock *sk) |
307 | { | |
308 | struct dst_entry *dst = __sk_dst_check(sk, 0); | |
309 | ||
1ed5c48f | 310 | if (dst) |
6700c270 | 311 | dst->ops->redirect(dst, sk, skb); |
55be7a9c DM |
312 | } |
313 | ||
26e37360 ED |
314 | |
315 | /* handle ICMP messages on TCP_NEW_SYN_RECV request sockets */ | |
9cf74903 | 316 | void tcp_req_err(struct sock *sk, u32 seq, bool abort) |
26e37360 ED |
317 | { |
318 | struct request_sock *req = inet_reqsk(sk); | |
319 | struct net *net = sock_net(sk); | |
320 | ||
321 | /* ICMPs are not backlogged, hence we cannot get | |
322 | * an established socket here. | |
323 | */ | |
26e37360 | 324 | if (seq != tcp_rsk(req)->snt_isn) { |
02a1d6e7 | 325 | __NET_INC_STATS(net, LINUX_MIB_OUTOFWINDOWICMPS); |
9cf74903 | 326 | } else if (abort) { |
26e37360 ED |
327 | /* |
328 | * Still in SYN_RECV, just remove it silently. | |
329 | * There is no good way to pass the error to the newly | |
330 | * created socket, and POSIX does not want network | |
331 | * errors returned from accept(). | |
332 | */ | |
c6973669 | 333 | inet_csk_reqsk_queue_drop(req->rsk_listener, req); |
9caad864 | 334 | tcp_listendrop(req->rsk_listener); |
26e37360 | 335 | } |
ef84d8ce | 336 | reqsk_put(req); |
26e37360 ED |
337 | } |
338 | EXPORT_SYMBOL(tcp_req_err); | |
339 | ||
1da177e4 LT |
340 | /* |
341 | * This routine is called by the ICMP module when it gets some | |
342 | * sort of error condition. If err < 0 then the socket should | |
343 | * be closed and the error returned to the user. If err > 0 | |
344 | * it's just the icmp type << 8 | icmp code. After adjustment | |
345 | * header points to the first 8 bytes of the tcp header. We need | |
346 | * to find the appropriate port. | |
347 | * | |
348 | * The locking strategy used here is very "optimistic". When | |
349 | * someone else accesses the socket the ICMP is just dropped | |
350 | * and for some paths there is no check at all. | |
351 | * A more general error queue to queue errors for later handling | |
352 | * is probably better. | |
353 | * | |
354 | */ | |
355 | ||
4d1a2d9e | 356 | void tcp_v4_err(struct sk_buff *icmp_skb, u32 info) |
1da177e4 | 357 | { |
b71d1d42 | 358 | const struct iphdr *iph = (const struct iphdr *)icmp_skb->data; |
4d1a2d9e | 359 | struct tcphdr *th = (struct tcphdr *)(icmp_skb->data + (iph->ihl << 2)); |
f1ecd5d9 | 360 | struct inet_connection_sock *icsk; |
1da177e4 LT |
361 | struct tcp_sock *tp; |
362 | struct inet_sock *inet; | |
4d1a2d9e DL |
363 | const int type = icmp_hdr(icmp_skb)->type; |
364 | const int code = icmp_hdr(icmp_skb)->code; | |
1da177e4 | 365 | struct sock *sk; |
f1ecd5d9 | 366 | struct sk_buff *skb; |
0a672f74 YC |
367 | struct request_sock *fastopen; |
368 | __u32 seq, snd_una; | |
f1ecd5d9 | 369 | __u32 remaining; |
1da177e4 | 370 | int err; |
4d1a2d9e | 371 | struct net *net = dev_net(icmp_skb->dev); |
1da177e4 | 372 | |
26e37360 ED |
373 | sk = __inet_lookup_established(net, &tcp_hashinfo, iph->daddr, |
374 | th->dest, iph->saddr, ntohs(th->source), | |
375 | inet_iif(icmp_skb)); | |
1da177e4 | 376 | if (!sk) { |
5d3848bc | 377 | __ICMP_INC_STATS(net, ICMP_MIB_INERRORS); |
1da177e4 LT |
378 | return; |
379 | } | |
380 | if (sk->sk_state == TCP_TIME_WAIT) { | |
9469c7b4 | 381 | inet_twsk_put(inet_twsk(sk)); |
1da177e4 LT |
382 | return; |
383 | } | |
26e37360 ED |
384 | seq = ntohl(th->seq); |
385 | if (sk->sk_state == TCP_NEW_SYN_RECV) | |
9cf74903 ED |
386 | return tcp_req_err(sk, seq, |
387 | type == ICMP_PARAMETERPROB || | |
388 | type == ICMP_TIME_EXCEEDED || | |
389 | (type == ICMP_DEST_UNREACH && | |
390 | (code == ICMP_NET_UNREACH || | |
391 | code == ICMP_HOST_UNREACH))); | |
1da177e4 LT |
392 | |
393 | bh_lock_sock(sk); | |
394 | /* If too many ICMPs get dropped on busy | |
395 | * servers this needs to be solved differently. | |
563d34d0 ED |
396 | * We do take care of PMTU discovery (RFC1191) special case : |
397 | * we can receive locally generated ICMP messages while socket is held. | |
1da177e4 | 398 | */ |
b74aa930 ED |
399 | if (sock_owned_by_user(sk)) { |
400 | if (!(type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED)) | |
02a1d6e7 | 401 | __NET_INC_STATS(net, LINUX_MIB_LOCKDROPPEDICMPS); |
b74aa930 | 402 | } |
1da177e4 LT |
403 | if (sk->sk_state == TCP_CLOSE) |
404 | goto out; | |
405 | ||
97e3ecd1 | 406 | if (unlikely(iph->ttl < inet_sk(sk)->min_ttl)) { |
02a1d6e7 | 407 | __NET_INC_STATS(net, LINUX_MIB_TCPMINTTLDROP); |
97e3ecd1 | 408 | goto out; |
409 | } | |
410 | ||
f1ecd5d9 | 411 | icsk = inet_csk(sk); |
1da177e4 | 412 | tp = tcp_sk(sk); |
0a672f74 YC |
413 | /* XXX (TFO) - tp->snd_una should be ISN (tcp_create_openreq_child() */ |
414 | fastopen = tp->fastopen_rsk; | |
415 | snd_una = fastopen ? tcp_rsk(fastopen)->snt_isn : tp->snd_una; | |
1da177e4 | 416 | if (sk->sk_state != TCP_LISTEN && |
0a672f74 | 417 | !between(seq, snd_una, tp->snd_nxt)) { |
02a1d6e7 | 418 | __NET_INC_STATS(net, LINUX_MIB_OUTOFWINDOWICMPS); |
1da177e4 LT |
419 | goto out; |
420 | } | |
421 | ||
422 | switch (type) { | |
55be7a9c | 423 | case ICMP_REDIRECT: |
0c244cdf JM |
424 | if (!sock_owned_by_user(sk)) |
425 | do_redirect(icmp_skb, sk); | |
55be7a9c | 426 | goto out; |
1da177e4 LT |
427 | case ICMP_SOURCE_QUENCH: |
428 | /* Just silently ignore these. */ | |
429 | goto out; | |
430 | case ICMP_PARAMETERPROB: | |
431 | err = EPROTO; | |
432 | break; | |
433 | case ICMP_DEST_UNREACH: | |
434 | if (code > NR_ICMP_UNREACH) | |
435 | goto out; | |
436 | ||
437 | if (code == ICMP_FRAG_NEEDED) { /* PMTU discovery (RFC1191) */ | |
0d4f0608 ED |
438 | /* We are not interested in TCP_LISTEN and open_requests |
439 | * (SYN-ACKs send out by Linux are always <576bytes so | |
440 | * they should go through unfragmented). | |
441 | */ | |
442 | if (sk->sk_state == TCP_LISTEN) | |
443 | goto out; | |
444 | ||
563d34d0 | 445 | tp->mtu_info = info; |
144d56e9 | 446 | if (!sock_owned_by_user(sk)) { |
563d34d0 | 447 | tcp_v4_mtu_reduced(sk); |
144d56e9 | 448 | } else { |
7aa5470c | 449 | if (!test_and_set_bit(TCP_MTU_REDUCED_DEFERRED, &sk->sk_tsq_flags)) |
144d56e9 ED |
450 | sock_hold(sk); |
451 | } | |
1da177e4 LT |
452 | goto out; |
453 | } | |
454 | ||
455 | err = icmp_err_convert[code].errno; | |
f1ecd5d9 DL |
456 | /* check if icmp_skb allows revert of backoff |
457 | * (see draft-zimmermann-tcp-lcd) */ | |
458 | if (code != ICMP_NET_UNREACH && code != ICMP_HOST_UNREACH) | |
459 | break; | |
460 | if (seq != tp->snd_una || !icsk->icsk_retransmits || | |
0a672f74 | 461 | !icsk->icsk_backoff || fastopen) |
f1ecd5d9 DL |
462 | break; |
463 | ||
8f49c270 DM |
464 | if (sock_owned_by_user(sk)) |
465 | break; | |
466 | ||
f1ecd5d9 | 467 | icsk->icsk_backoff--; |
fcdd1cf4 ED |
468 | icsk->icsk_rto = tp->srtt_us ? __tcp_set_rto(tp) : |
469 | TCP_TIMEOUT_INIT; | |
470 | icsk->icsk_rto = inet_csk_rto_backoff(icsk, TCP_RTO_MAX); | |
f1ecd5d9 DL |
471 | |
472 | skb = tcp_write_queue_head(sk); | |
473 | BUG_ON(!skb); | |
474 | ||
7faee5c0 ED |
475 | remaining = icsk->icsk_rto - |
476 | min(icsk->icsk_rto, | |
477 | tcp_time_stamp - tcp_skb_timestamp(skb)); | |
f1ecd5d9 DL |
478 | |
479 | if (remaining) { | |
480 | inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS, | |
481 | remaining, TCP_RTO_MAX); | |
f1ecd5d9 DL |
482 | } else { |
483 | /* RTO revert clocked out retransmission. | |
484 | * Will retransmit now */ | |
485 | tcp_retransmit_timer(sk); | |
486 | } | |
487 | ||
1da177e4 LT |
488 | break; |
489 | case ICMP_TIME_EXCEEDED: | |
490 | err = EHOSTUNREACH; | |
491 | break; | |
492 | default: | |
493 | goto out; | |
494 | } | |
495 | ||
496 | switch (sk->sk_state) { | |
1da177e4 | 497 | case TCP_SYN_SENT: |
0a672f74 YC |
498 | case TCP_SYN_RECV: |
499 | /* Only in fast or simultaneous open. If a fast open socket is | |
500 | * is already accepted it is treated as a connected one below. | |
501 | */ | |
51456b29 | 502 | if (fastopen && !fastopen->sk) |
0a672f74 YC |
503 | break; |
504 | ||
1da177e4 | 505 | if (!sock_owned_by_user(sk)) { |
1da177e4 LT |
506 | sk->sk_err = err; |
507 | ||
508 | sk->sk_error_report(sk); | |
509 | ||
510 | tcp_done(sk); | |
511 | } else { | |
512 | sk->sk_err_soft = err; | |
513 | } | |
514 | goto out; | |
515 | } | |
516 | ||
517 | /* If we've already connected we will keep trying | |
518 | * until we time out, or the user gives up. | |
519 | * | |
520 | * rfc1122 4.2.3.9 allows to consider as hard errors | |
521 | * only PROTO_UNREACH and PORT_UNREACH (well, FRAG_FAILED too, | |
522 | * but it is obsoleted by pmtu discovery). | |
523 | * | |
524 | * Note, that in modern internet, where routing is unreliable | |
525 | * and in each dark corner broken firewalls sit, sending random | |
526 | * errors ordered by their masters even this two messages finally lose | |
527 | * their original sense (even Linux sends invalid PORT_UNREACHs) | |
528 | * | |
529 | * Now we are in compliance with RFCs. | |
530 | * --ANK (980905) | |
531 | */ | |
532 | ||
533 | inet = inet_sk(sk); | |
534 | if (!sock_owned_by_user(sk) && inet->recverr) { | |
535 | sk->sk_err = err; | |
536 | sk->sk_error_report(sk); | |
537 | } else { /* Only an error on timeout */ | |
538 | sk->sk_err_soft = err; | |
539 | } | |
540 | ||
541 | out: | |
542 | bh_unlock_sock(sk); | |
543 | sock_put(sk); | |
544 | } | |
545 | ||
28850dc7 | 546 | void __tcp_v4_send_check(struct sk_buff *skb, __be32 saddr, __be32 daddr) |
1da177e4 | 547 | { |
aa8223c7 | 548 | struct tcphdr *th = tcp_hdr(skb); |
1da177e4 | 549 | |
84fa7933 | 550 | if (skb->ip_summed == CHECKSUM_PARTIAL) { |
419f9f89 | 551 | th->check = ~tcp_v4_check(skb->len, saddr, daddr, 0); |
663ead3b | 552 | skb->csum_start = skb_transport_header(skb) - skb->head; |
ff1dcadb | 553 | skb->csum_offset = offsetof(struct tcphdr, check); |
1da177e4 | 554 | } else { |
419f9f89 | 555 | th->check = tcp_v4_check(skb->len, saddr, daddr, |
07f0757a | 556 | csum_partial(th, |
1da177e4 LT |
557 | th->doff << 2, |
558 | skb->csum)); | |
559 | } | |
560 | } | |
561 | ||
419f9f89 | 562 | /* This routine computes an IPv4 TCP checksum. */ |
bb296246 | 563 | void tcp_v4_send_check(struct sock *sk, struct sk_buff *skb) |
419f9f89 | 564 | { |
cf533ea5 | 565 | const struct inet_sock *inet = inet_sk(sk); |
419f9f89 HX |
566 | |
567 | __tcp_v4_send_check(skb, inet->inet_saddr, inet->inet_daddr); | |
568 | } | |
4bc2f18b | 569 | EXPORT_SYMBOL(tcp_v4_send_check); |
419f9f89 | 570 | |
1da177e4 LT |
571 | /* |
572 | * This routine will send an RST to the other tcp. | |
573 | * | |
574 | * Someone asks: why I NEVER use socket parameters (TOS, TTL etc.) | |
575 | * for reset. | |
576 | * Answer: if a packet caused RST, it is not for a socket | |
577 | * existing in our system, if it is matched to a socket, | |
578 | * it is just duplicate segment or bug in other side's TCP. | |
579 | * So that we build reply only basing on parameters | |
580 | * arrived with segment. | |
581 | * Exception: precedence violation. We do not implement it in any case. | |
582 | */ | |
583 | ||
a00e7444 | 584 | static void tcp_v4_send_reset(const struct sock *sk, struct sk_buff *skb) |
1da177e4 | 585 | { |
cf533ea5 | 586 | const struct tcphdr *th = tcp_hdr(skb); |
cfb6eeb4 YH |
587 | struct { |
588 | struct tcphdr th; | |
589 | #ifdef CONFIG_TCP_MD5SIG | |
714e85be | 590 | __be32 opt[(TCPOLEN_MD5SIG_ALIGNED >> 2)]; |
cfb6eeb4 YH |
591 | #endif |
592 | } rep; | |
1da177e4 | 593 | struct ip_reply_arg arg; |
cfb6eeb4 | 594 | #ifdef CONFIG_TCP_MD5SIG |
e46787f0 | 595 | struct tcp_md5sig_key *key = NULL; |
658ddaaf SL |
596 | const __u8 *hash_location = NULL; |
597 | unsigned char newhash[16]; | |
598 | int genhash; | |
599 | struct sock *sk1 = NULL; | |
cfb6eeb4 | 600 | #endif |
a86b1e30 | 601 | struct net *net; |
1da177e4 LT |
602 | |
603 | /* Never send a reset in response to a reset. */ | |
604 | if (th->rst) | |
605 | return; | |
606 | ||
c3658e8d ED |
607 | /* If sk not NULL, it means we did a successful lookup and incoming |
608 | * route had to be correct. prequeue might have dropped our dst. | |
609 | */ | |
610 | if (!sk && skb_rtable(skb)->rt_type != RTN_LOCAL) | |
1da177e4 LT |
611 | return; |
612 | ||
613 | /* Swap the send and the receive. */ | |
cfb6eeb4 YH |
614 | memset(&rep, 0, sizeof(rep)); |
615 | rep.th.dest = th->source; | |
616 | rep.th.source = th->dest; | |
617 | rep.th.doff = sizeof(struct tcphdr) / 4; | |
618 | rep.th.rst = 1; | |
1da177e4 LT |
619 | |
620 | if (th->ack) { | |
cfb6eeb4 | 621 | rep.th.seq = th->ack_seq; |
1da177e4 | 622 | } else { |
cfb6eeb4 YH |
623 | rep.th.ack = 1; |
624 | rep.th.ack_seq = htonl(ntohl(th->seq) + th->syn + th->fin + | |
625 | skb->len - (th->doff << 2)); | |
1da177e4 LT |
626 | } |
627 | ||
7174259e | 628 | memset(&arg, 0, sizeof(arg)); |
cfb6eeb4 YH |
629 | arg.iov[0].iov_base = (unsigned char *)&rep; |
630 | arg.iov[0].iov_len = sizeof(rep.th); | |
631 | ||
0f85feae | 632 | net = sk ? sock_net(sk) : dev_net(skb_dst(skb)->dev); |
cfb6eeb4 | 633 | #ifdef CONFIG_TCP_MD5SIG |
3b24d854 | 634 | rcu_read_lock(); |
658ddaaf | 635 | hash_location = tcp_parse_md5sig_option(th); |
271c3b9b | 636 | if (sk && sk_fullsock(sk)) { |
e46787f0 FW |
637 | key = tcp_md5_do_lookup(sk, (union tcp_md5_addr *) |
638 | &ip_hdr(skb)->saddr, AF_INET); | |
639 | } else if (hash_location) { | |
658ddaaf SL |
640 | /* |
641 | * active side is lost. Try to find listening socket through | |
642 | * source port, and then find md5 key through listening socket. | |
643 | * we are not loose security here: | |
644 | * Incoming packet is checked with md5 hash with finding key, | |
645 | * no RST generated if md5 hash doesn't match. | |
646 | */ | |
a583636a CG |
647 | sk1 = __inet_lookup_listener(net, &tcp_hashinfo, NULL, 0, |
648 | ip_hdr(skb)->saddr, | |
da5e3630 | 649 | th->source, ip_hdr(skb)->daddr, |
658ddaaf SL |
650 | ntohs(th->source), inet_iif(skb)); |
651 | /* don't send rst if it can't find key */ | |
652 | if (!sk1) | |
3b24d854 ED |
653 | goto out; |
654 | ||
658ddaaf SL |
655 | key = tcp_md5_do_lookup(sk1, (union tcp_md5_addr *) |
656 | &ip_hdr(skb)->saddr, AF_INET); | |
657 | if (!key) | |
3b24d854 ED |
658 | goto out; |
659 | ||
658ddaaf | 660 | |
39f8e58e | 661 | genhash = tcp_v4_md5_hash_skb(newhash, key, NULL, skb); |
658ddaaf | 662 | if (genhash || memcmp(hash_location, newhash, 16) != 0) |
3b24d854 ED |
663 | goto out; |
664 | ||
658ddaaf SL |
665 | } |
666 | ||
cfb6eeb4 YH |
667 | if (key) { |
668 | rep.opt[0] = htonl((TCPOPT_NOP << 24) | | |
669 | (TCPOPT_NOP << 16) | | |
670 | (TCPOPT_MD5SIG << 8) | | |
671 | TCPOLEN_MD5SIG); | |
672 | /* Update length and the length the header thinks exists */ | |
673 | arg.iov[0].iov_len += TCPOLEN_MD5SIG_ALIGNED; | |
674 | rep.th.doff = arg.iov[0].iov_len / 4; | |
675 | ||
49a72dfb | 676 | tcp_v4_md5_hash_hdr((__u8 *) &rep.opt[1], |
78e645cb IJ |
677 | key, ip_hdr(skb)->saddr, |
678 | ip_hdr(skb)->daddr, &rep.th); | |
cfb6eeb4 YH |
679 | } |
680 | #endif | |
eddc9ec5 ACM |
681 | arg.csum = csum_tcpudp_nofold(ip_hdr(skb)->daddr, |
682 | ip_hdr(skb)->saddr, /* XXX */ | |
52cd5750 | 683 | arg.iov[0].iov_len, IPPROTO_TCP, 0); |
1da177e4 | 684 | arg.csumoffset = offsetof(struct tcphdr, check) / 2; |
271c3b9b FW |
685 | arg.flags = (sk && inet_sk_transparent(sk)) ? IP_REPLY_ARG_NOSRCCHECK : 0; |
686 | ||
e2446eaa | 687 | /* When socket is gone, all binding information is lost. |
4c675258 AK |
688 | * routing might fail in this case. No choice here, if we choose to force |
689 | * input interface, we will misroute in case of asymmetric route. | |
e2446eaa | 690 | */ |
4c675258 AK |
691 | if (sk) |
692 | arg.bound_dev_if = sk->sk_bound_dev_if; | |
1da177e4 | 693 | |
271c3b9b FW |
694 | BUILD_BUG_ON(offsetof(struct sock, sk_bound_dev_if) != |
695 | offsetof(struct inet_timewait_sock, tw_bound_dev_if)); | |
696 | ||
66b13d99 | 697 | arg.tos = ip_hdr(skb)->tos; |
e2d118a1 | 698 | arg.uid = sock_net_uid(net, sk && sk_fullsock(sk) ? sk : NULL); |
47dcc20a | 699 | local_bh_disable(); |
bdbbb852 ED |
700 | ip_send_unicast_reply(*this_cpu_ptr(net->ipv4.tcp_sk), |
701 | skb, &TCP_SKB_CB(skb)->header.h4.opt, | |
24a2d43d ED |
702 | ip_hdr(skb)->saddr, ip_hdr(skb)->daddr, |
703 | &arg, arg.iov[0].iov_len); | |
1da177e4 | 704 | |
90bbcc60 ED |
705 | __TCP_INC_STATS(net, TCP_MIB_OUTSEGS); |
706 | __TCP_INC_STATS(net, TCP_MIB_OUTRSTS); | |
47dcc20a | 707 | local_bh_enable(); |
658ddaaf SL |
708 | |
709 | #ifdef CONFIG_TCP_MD5SIG | |
3b24d854 ED |
710 | out: |
711 | rcu_read_unlock(); | |
658ddaaf | 712 | #endif |
1da177e4 LT |
713 | } |
714 | ||
715 | /* The code following below sending ACKs in SYN-RECV and TIME-WAIT states | |
716 | outside socket context is ugly, certainly. What can I do? | |
717 | */ | |
718 | ||
e2d118a1 | 719 | static void tcp_v4_send_ack(const struct sock *sk, |
e62a123b | 720 | struct sk_buff *skb, u32 seq, u32 ack, |
ee684b6f | 721 | u32 win, u32 tsval, u32 tsecr, int oif, |
88ef4a5a | 722 | struct tcp_md5sig_key *key, |
66b13d99 | 723 | int reply_flags, u8 tos) |
1da177e4 | 724 | { |
cf533ea5 | 725 | const struct tcphdr *th = tcp_hdr(skb); |
1da177e4 LT |
726 | struct { |
727 | struct tcphdr th; | |
714e85be | 728 | __be32 opt[(TCPOLEN_TSTAMP_ALIGNED >> 2) |
cfb6eeb4 | 729 | #ifdef CONFIG_TCP_MD5SIG |
714e85be | 730 | + (TCPOLEN_MD5SIG_ALIGNED >> 2) |
cfb6eeb4 YH |
731 | #endif |
732 | ]; | |
1da177e4 | 733 | } rep; |
e2d118a1 | 734 | struct net *net = sock_net(sk); |
1da177e4 LT |
735 | struct ip_reply_arg arg; |
736 | ||
737 | memset(&rep.th, 0, sizeof(struct tcphdr)); | |
7174259e | 738 | memset(&arg, 0, sizeof(arg)); |
1da177e4 LT |
739 | |
740 | arg.iov[0].iov_base = (unsigned char *)&rep; | |
741 | arg.iov[0].iov_len = sizeof(rep.th); | |
ee684b6f | 742 | if (tsecr) { |
cfb6eeb4 YH |
743 | rep.opt[0] = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) | |
744 | (TCPOPT_TIMESTAMP << 8) | | |
745 | TCPOLEN_TIMESTAMP); | |
ee684b6f AV |
746 | rep.opt[1] = htonl(tsval); |
747 | rep.opt[2] = htonl(tsecr); | |
cb48cfe8 | 748 | arg.iov[0].iov_len += TCPOLEN_TSTAMP_ALIGNED; |
1da177e4 LT |
749 | } |
750 | ||
751 | /* Swap the send and the receive. */ | |
752 | rep.th.dest = th->source; | |
753 | rep.th.source = th->dest; | |
754 | rep.th.doff = arg.iov[0].iov_len / 4; | |
755 | rep.th.seq = htonl(seq); | |
756 | rep.th.ack_seq = htonl(ack); | |
757 | rep.th.ack = 1; | |
758 | rep.th.window = htons(win); | |
759 | ||
cfb6eeb4 | 760 | #ifdef CONFIG_TCP_MD5SIG |
cfb6eeb4 | 761 | if (key) { |
ee684b6f | 762 | int offset = (tsecr) ? 3 : 0; |
cfb6eeb4 YH |
763 | |
764 | rep.opt[offset++] = htonl((TCPOPT_NOP << 24) | | |
765 | (TCPOPT_NOP << 16) | | |
766 | (TCPOPT_MD5SIG << 8) | | |
767 | TCPOLEN_MD5SIG); | |
768 | arg.iov[0].iov_len += TCPOLEN_MD5SIG_ALIGNED; | |
769 | rep.th.doff = arg.iov[0].iov_len/4; | |
770 | ||
49a72dfb | 771 | tcp_v4_md5_hash_hdr((__u8 *) &rep.opt[offset], |
90b7e112 AL |
772 | key, ip_hdr(skb)->saddr, |
773 | ip_hdr(skb)->daddr, &rep.th); | |
cfb6eeb4 YH |
774 | } |
775 | #endif | |
88ef4a5a | 776 | arg.flags = reply_flags; |
eddc9ec5 ACM |
777 | arg.csum = csum_tcpudp_nofold(ip_hdr(skb)->daddr, |
778 | ip_hdr(skb)->saddr, /* XXX */ | |
1da177e4 LT |
779 | arg.iov[0].iov_len, IPPROTO_TCP, 0); |
780 | arg.csumoffset = offsetof(struct tcphdr, check) / 2; | |
9501f972 YH |
781 | if (oif) |
782 | arg.bound_dev_if = oif; | |
66b13d99 | 783 | arg.tos = tos; |
e2d118a1 | 784 | arg.uid = sock_net_uid(net, sk_fullsock(sk) ? sk : NULL); |
47dcc20a | 785 | local_bh_disable(); |
bdbbb852 ED |
786 | ip_send_unicast_reply(*this_cpu_ptr(net->ipv4.tcp_sk), |
787 | skb, &TCP_SKB_CB(skb)->header.h4.opt, | |
24a2d43d ED |
788 | ip_hdr(skb)->saddr, ip_hdr(skb)->daddr, |
789 | &arg, arg.iov[0].iov_len); | |
1da177e4 | 790 | |
90bbcc60 | 791 | __TCP_INC_STATS(net, TCP_MIB_OUTSEGS); |
47dcc20a | 792 | local_bh_enable(); |
1da177e4 LT |
793 | } |
794 | ||
795 | static void tcp_v4_timewait_ack(struct sock *sk, struct sk_buff *skb) | |
796 | { | |
8feaf0c0 | 797 | struct inet_timewait_sock *tw = inet_twsk(sk); |
cfb6eeb4 | 798 | struct tcp_timewait_sock *tcptw = tcp_twsk(sk); |
1da177e4 | 799 | |
e2d118a1 | 800 | tcp_v4_send_ack(sk, skb, |
e62a123b | 801 | tcptw->tw_snd_nxt, tcptw->tw_rcv_nxt, |
7174259e | 802 | tcptw->tw_rcv_wnd >> tw->tw_rcv_wscale, |
ee684b6f | 803 | tcp_time_stamp + tcptw->tw_ts_offset, |
9501f972 YH |
804 | tcptw->tw_ts_recent, |
805 | tw->tw_bound_dev_if, | |
88ef4a5a | 806 | tcp_twsk_md5_key(tcptw), |
66b13d99 ED |
807 | tw->tw_transparent ? IP_REPLY_ARG_NOSRCCHECK : 0, |
808 | tw->tw_tos | |
9501f972 | 809 | ); |
1da177e4 | 810 | |
8feaf0c0 | 811 | inet_twsk_put(tw); |
1da177e4 LT |
812 | } |
813 | ||
a00e7444 | 814 | static void tcp_v4_reqsk_send_ack(const struct sock *sk, struct sk_buff *skb, |
7174259e | 815 | struct request_sock *req) |
1da177e4 | 816 | { |
168a8f58 JC |
817 | /* sk->sk_state == TCP_LISTEN -> for regular TCP_SYN_RECV |
818 | * sk->sk_state == TCP_SYN_RECV -> for Fast Open. | |
819 | */ | |
e62a123b ED |
820 | u32 seq = (sk->sk_state == TCP_LISTEN) ? tcp_rsk(req)->snt_isn + 1 : |
821 | tcp_sk(sk)->snd_nxt; | |
822 | ||
20a2b49f ED |
823 | /* RFC 7323 2.3 |
824 | * The window field (SEG.WND) of every outgoing segment, with the | |
825 | * exception of <SYN> segments, MUST be right-shifted by | |
826 | * Rcv.Wind.Shift bits: | |
827 | */ | |
e2d118a1 | 828 | tcp_v4_send_ack(sk, skb, seq, |
20a2b49f ED |
829 | tcp_rsk(req)->rcv_nxt, |
830 | req->rsk_rcv_wnd >> inet_rsk(req)->rcv_wscale, | |
95a22cae | 831 | tcp_time_stamp + tcp_rsk(req)->ts_off, |
9501f972 YH |
832 | req->ts_recent, |
833 | 0, | |
a915da9b ED |
834 | tcp_md5_do_lookup(sk, (union tcp_md5_addr *)&ip_hdr(skb)->daddr, |
835 | AF_INET), | |
66b13d99 ED |
836 | inet_rsk(req)->no_srccheck ? IP_REPLY_ARG_NOSRCCHECK : 0, |
837 | ip_hdr(skb)->tos); | |
1da177e4 LT |
838 | } |
839 | ||
1da177e4 | 840 | /* |
9bf1d83e | 841 | * Send a SYN-ACK after having received a SYN. |
60236fdd | 842 | * This still operates on a request_sock only, not on a big |
1da177e4 LT |
843 | * socket. |
844 | */ | |
0f935dbe | 845 | static int tcp_v4_send_synack(const struct sock *sk, struct dst_entry *dst, |
d6274bd8 | 846 | struct flowi *fl, |
72659ecc | 847 | struct request_sock *req, |
ca6fb065 | 848 | struct tcp_fastopen_cookie *foc, |
b3d05147 | 849 | enum tcp_synack_type synack_type) |
1da177e4 | 850 | { |
2e6599cb | 851 | const struct inet_request_sock *ireq = inet_rsk(req); |
6bd023f3 | 852 | struct flowi4 fl4; |
1da177e4 | 853 | int err = -1; |
d41db5af | 854 | struct sk_buff *skb; |
1da177e4 LT |
855 | |
856 | /* First, grab a route. */ | |
ba3f7f04 | 857 | if (!dst && (dst = inet_csk_route_req(sk, &fl4, req)) == NULL) |
fd80eb94 | 858 | return -1; |
1da177e4 | 859 | |
b3d05147 | 860 | skb = tcp_make_synack(sk, dst, req, foc, synack_type); |
1da177e4 LT |
861 | |
862 | if (skb) { | |
634fb979 | 863 | __tcp_v4_send_check(skb, ireq->ir_loc_addr, ireq->ir_rmt_addr); |
1da177e4 | 864 | |
634fb979 ED |
865 | err = ip_build_and_send_pkt(skb, sk, ireq->ir_loc_addr, |
866 | ireq->ir_rmt_addr, | |
2e6599cb | 867 | ireq->opt); |
b9df3cb8 | 868 | err = net_xmit_eval(err); |
1da177e4 LT |
869 | } |
870 | ||
1da177e4 LT |
871 | return err; |
872 | } | |
873 | ||
874 | /* | |
60236fdd | 875 | * IPv4 request_sock destructor. |
1da177e4 | 876 | */ |
60236fdd | 877 | static void tcp_v4_reqsk_destructor(struct request_sock *req) |
1da177e4 | 878 | { |
a51482bd | 879 | kfree(inet_rsk(req)->opt); |
1da177e4 LT |
880 | } |
881 | ||
cfb6eeb4 YH |
882 | #ifdef CONFIG_TCP_MD5SIG |
883 | /* | |
884 | * RFC2385 MD5 checksumming requires a mapping of | |
885 | * IP address->MD5 Key. | |
886 | * We need to maintain these in the sk structure. | |
887 | */ | |
888 | ||
889 | /* Find the Key structure for an address. */ | |
b83e3deb | 890 | struct tcp_md5sig_key *tcp_md5_do_lookup(const struct sock *sk, |
a915da9b ED |
891 | const union tcp_md5_addr *addr, |
892 | int family) | |
cfb6eeb4 | 893 | { |
fd3a154a | 894 | const struct tcp_sock *tp = tcp_sk(sk); |
a915da9b | 895 | struct tcp_md5sig_key *key; |
a915da9b | 896 | unsigned int size = sizeof(struct in_addr); |
fd3a154a | 897 | const struct tcp_md5sig_info *md5sig; |
cfb6eeb4 | 898 | |
a8afca03 ED |
899 | /* caller either holds rcu_read_lock() or socket lock */ |
900 | md5sig = rcu_dereference_check(tp->md5sig_info, | |
1e1d04e6 | 901 | lockdep_sock_is_held(sk)); |
a8afca03 | 902 | if (!md5sig) |
cfb6eeb4 | 903 | return NULL; |
a915da9b ED |
904 | #if IS_ENABLED(CONFIG_IPV6) |
905 | if (family == AF_INET6) | |
906 | size = sizeof(struct in6_addr); | |
907 | #endif | |
b67bfe0d | 908 | hlist_for_each_entry_rcu(key, &md5sig->head, node) { |
a915da9b ED |
909 | if (key->family != family) |
910 | continue; | |
911 | if (!memcmp(&key->addr, addr, size)) | |
912 | return key; | |
cfb6eeb4 YH |
913 | } |
914 | return NULL; | |
915 | } | |
a915da9b | 916 | EXPORT_SYMBOL(tcp_md5_do_lookup); |
cfb6eeb4 | 917 | |
b83e3deb | 918 | struct tcp_md5sig_key *tcp_v4_md5_lookup(const struct sock *sk, |
fd3a154a | 919 | const struct sock *addr_sk) |
cfb6eeb4 | 920 | { |
b52e6921 | 921 | const union tcp_md5_addr *addr; |
a915da9b | 922 | |
b52e6921 | 923 | addr = (const union tcp_md5_addr *)&addr_sk->sk_daddr; |
a915da9b | 924 | return tcp_md5_do_lookup(sk, addr, AF_INET); |
cfb6eeb4 | 925 | } |
cfb6eeb4 YH |
926 | EXPORT_SYMBOL(tcp_v4_md5_lookup); |
927 | ||
cfb6eeb4 | 928 | /* This can be called on a newly created socket, from other files */ |
a915da9b ED |
929 | int tcp_md5_do_add(struct sock *sk, const union tcp_md5_addr *addr, |
930 | int family, const u8 *newkey, u8 newkeylen, gfp_t gfp) | |
cfb6eeb4 YH |
931 | { |
932 | /* Add Key to the list */ | |
b0a713e9 | 933 | struct tcp_md5sig_key *key; |
cfb6eeb4 | 934 | struct tcp_sock *tp = tcp_sk(sk); |
a915da9b | 935 | struct tcp_md5sig_info *md5sig; |
cfb6eeb4 | 936 | |
c0353c7b | 937 | key = tcp_md5_do_lookup(sk, addr, family); |
cfb6eeb4 YH |
938 | if (key) { |
939 | /* Pre-existing entry - just update that one. */ | |
a915da9b | 940 | memcpy(key->key, newkey, newkeylen); |
b0a713e9 | 941 | key->keylen = newkeylen; |
a915da9b ED |
942 | return 0; |
943 | } | |
260fcbeb | 944 | |
a8afca03 | 945 | md5sig = rcu_dereference_protected(tp->md5sig_info, |
1e1d04e6 | 946 | lockdep_sock_is_held(sk)); |
a915da9b ED |
947 | if (!md5sig) { |
948 | md5sig = kmalloc(sizeof(*md5sig), gfp); | |
949 | if (!md5sig) | |
cfb6eeb4 | 950 | return -ENOMEM; |
cfb6eeb4 | 951 | |
a915da9b ED |
952 | sk_nocaps_add(sk, NETIF_F_GSO_MASK); |
953 | INIT_HLIST_HEAD(&md5sig->head); | |
a8afca03 | 954 | rcu_assign_pointer(tp->md5sig_info, md5sig); |
a915da9b | 955 | } |
cfb6eeb4 | 956 | |
5f3d9cb2 | 957 | key = sock_kmalloc(sk, sizeof(*key), gfp); |
a915da9b ED |
958 | if (!key) |
959 | return -ENOMEM; | |
71cea17e | 960 | if (!tcp_alloc_md5sig_pool()) { |
5f3d9cb2 | 961 | sock_kfree_s(sk, key, sizeof(*key)); |
a915da9b | 962 | return -ENOMEM; |
cfb6eeb4 | 963 | } |
a915da9b ED |
964 | |
965 | memcpy(key->key, newkey, newkeylen); | |
966 | key->keylen = newkeylen; | |
967 | key->family = family; | |
968 | memcpy(&key->addr, addr, | |
969 | (family == AF_INET6) ? sizeof(struct in6_addr) : | |
970 | sizeof(struct in_addr)); | |
971 | hlist_add_head_rcu(&key->node, &md5sig->head); | |
cfb6eeb4 YH |
972 | return 0; |
973 | } | |
a915da9b | 974 | EXPORT_SYMBOL(tcp_md5_do_add); |
cfb6eeb4 | 975 | |
a915da9b | 976 | int tcp_md5_do_del(struct sock *sk, const union tcp_md5_addr *addr, int family) |
cfb6eeb4 | 977 | { |
a915da9b ED |
978 | struct tcp_md5sig_key *key; |
979 | ||
c0353c7b | 980 | key = tcp_md5_do_lookup(sk, addr, family); |
a915da9b ED |
981 | if (!key) |
982 | return -ENOENT; | |
983 | hlist_del_rcu(&key->node); | |
5f3d9cb2 | 984 | atomic_sub(sizeof(*key), &sk->sk_omem_alloc); |
a915da9b | 985 | kfree_rcu(key, rcu); |
a915da9b | 986 | return 0; |
cfb6eeb4 | 987 | } |
a915da9b | 988 | EXPORT_SYMBOL(tcp_md5_do_del); |
cfb6eeb4 | 989 | |
e0683e70 | 990 | static void tcp_clear_md5_list(struct sock *sk) |
cfb6eeb4 YH |
991 | { |
992 | struct tcp_sock *tp = tcp_sk(sk); | |
a915da9b | 993 | struct tcp_md5sig_key *key; |
b67bfe0d | 994 | struct hlist_node *n; |
a8afca03 | 995 | struct tcp_md5sig_info *md5sig; |
cfb6eeb4 | 996 | |
a8afca03 ED |
997 | md5sig = rcu_dereference_protected(tp->md5sig_info, 1); |
998 | ||
b67bfe0d | 999 | hlist_for_each_entry_safe(key, n, &md5sig->head, node) { |
a915da9b | 1000 | hlist_del_rcu(&key->node); |
5f3d9cb2 | 1001 | atomic_sub(sizeof(*key), &sk->sk_omem_alloc); |
a915da9b | 1002 | kfree_rcu(key, rcu); |
cfb6eeb4 YH |
1003 | } |
1004 | } | |
1005 | ||
7174259e ACM |
1006 | static int tcp_v4_parse_md5_keys(struct sock *sk, char __user *optval, |
1007 | int optlen) | |
cfb6eeb4 YH |
1008 | { |
1009 | struct tcp_md5sig cmd; | |
1010 | struct sockaddr_in *sin = (struct sockaddr_in *)&cmd.tcpm_addr; | |
cfb6eeb4 YH |
1011 | |
1012 | if (optlen < sizeof(cmd)) | |
1013 | return -EINVAL; | |
1014 | ||
7174259e | 1015 | if (copy_from_user(&cmd, optval, sizeof(cmd))) |
cfb6eeb4 YH |
1016 | return -EFAULT; |
1017 | ||
1018 | if (sin->sin_family != AF_INET) | |
1019 | return -EINVAL; | |
1020 | ||
64a124ed | 1021 | if (!cmd.tcpm_keylen) |
a915da9b ED |
1022 | return tcp_md5_do_del(sk, (union tcp_md5_addr *)&sin->sin_addr.s_addr, |
1023 | AF_INET); | |
cfb6eeb4 YH |
1024 | |
1025 | if (cmd.tcpm_keylen > TCP_MD5SIG_MAXKEYLEN) | |
1026 | return -EINVAL; | |
1027 | ||
a915da9b ED |
1028 | return tcp_md5_do_add(sk, (union tcp_md5_addr *)&sin->sin_addr.s_addr, |
1029 | AF_INET, cmd.tcpm_key, cmd.tcpm_keylen, | |
1030 | GFP_KERNEL); | |
cfb6eeb4 YH |
1031 | } |
1032 | ||
19689e38 ED |
1033 | static int tcp_v4_md5_hash_headers(struct tcp_md5sig_pool *hp, |
1034 | __be32 daddr, __be32 saddr, | |
1035 | const struct tcphdr *th, int nbytes) | |
cfb6eeb4 | 1036 | { |
cfb6eeb4 | 1037 | struct tcp4_pseudohdr *bp; |
49a72dfb | 1038 | struct scatterlist sg; |
19689e38 | 1039 | struct tcphdr *_th; |
cfb6eeb4 | 1040 | |
19689e38 | 1041 | bp = hp->scratch; |
cfb6eeb4 YH |
1042 | bp->saddr = saddr; |
1043 | bp->daddr = daddr; | |
1044 | bp->pad = 0; | |
076fb722 | 1045 | bp->protocol = IPPROTO_TCP; |
49a72dfb | 1046 | bp->len = cpu_to_be16(nbytes); |
c7da57a1 | 1047 | |
19689e38 ED |
1048 | _th = (struct tcphdr *)(bp + 1); |
1049 | memcpy(_th, th, sizeof(*th)); | |
1050 | _th->check = 0; | |
1051 | ||
1052 | sg_init_one(&sg, bp, sizeof(*bp) + sizeof(*th)); | |
1053 | ahash_request_set_crypt(hp->md5_req, &sg, NULL, | |
1054 | sizeof(*bp) + sizeof(*th)); | |
cf80e0e4 | 1055 | return crypto_ahash_update(hp->md5_req); |
49a72dfb AL |
1056 | } |
1057 | ||
a915da9b | 1058 | static int tcp_v4_md5_hash_hdr(char *md5_hash, const struct tcp_md5sig_key *key, |
318cf7aa | 1059 | __be32 daddr, __be32 saddr, const struct tcphdr *th) |
49a72dfb AL |
1060 | { |
1061 | struct tcp_md5sig_pool *hp; | |
cf80e0e4 | 1062 | struct ahash_request *req; |
49a72dfb AL |
1063 | |
1064 | hp = tcp_get_md5sig_pool(); | |
1065 | if (!hp) | |
1066 | goto clear_hash_noput; | |
cf80e0e4 | 1067 | req = hp->md5_req; |
49a72dfb | 1068 | |
cf80e0e4 | 1069 | if (crypto_ahash_init(req)) |
49a72dfb | 1070 | goto clear_hash; |
19689e38 | 1071 | if (tcp_v4_md5_hash_headers(hp, daddr, saddr, th, th->doff << 2)) |
49a72dfb AL |
1072 | goto clear_hash; |
1073 | if (tcp_md5_hash_key(hp, key)) | |
1074 | goto clear_hash; | |
cf80e0e4 HX |
1075 | ahash_request_set_crypt(req, NULL, md5_hash, 0); |
1076 | if (crypto_ahash_final(req)) | |
cfb6eeb4 YH |
1077 | goto clear_hash; |
1078 | ||
cfb6eeb4 | 1079 | tcp_put_md5sig_pool(); |
cfb6eeb4 | 1080 | return 0; |
49a72dfb | 1081 | |
cfb6eeb4 YH |
1082 | clear_hash: |
1083 | tcp_put_md5sig_pool(); | |
1084 | clear_hash_noput: | |
1085 | memset(md5_hash, 0, 16); | |
49a72dfb | 1086 | return 1; |
cfb6eeb4 YH |
1087 | } |
1088 | ||
39f8e58e ED |
1089 | int tcp_v4_md5_hash_skb(char *md5_hash, const struct tcp_md5sig_key *key, |
1090 | const struct sock *sk, | |
318cf7aa | 1091 | const struct sk_buff *skb) |
cfb6eeb4 | 1092 | { |
49a72dfb | 1093 | struct tcp_md5sig_pool *hp; |
cf80e0e4 | 1094 | struct ahash_request *req; |
318cf7aa | 1095 | const struct tcphdr *th = tcp_hdr(skb); |
cfb6eeb4 YH |
1096 | __be32 saddr, daddr; |
1097 | ||
39f8e58e ED |
1098 | if (sk) { /* valid for establish/request sockets */ |
1099 | saddr = sk->sk_rcv_saddr; | |
1100 | daddr = sk->sk_daddr; | |
cfb6eeb4 | 1101 | } else { |
49a72dfb AL |
1102 | const struct iphdr *iph = ip_hdr(skb); |
1103 | saddr = iph->saddr; | |
1104 | daddr = iph->daddr; | |
cfb6eeb4 | 1105 | } |
49a72dfb AL |
1106 | |
1107 | hp = tcp_get_md5sig_pool(); | |
1108 | if (!hp) | |
1109 | goto clear_hash_noput; | |
cf80e0e4 | 1110 | req = hp->md5_req; |
49a72dfb | 1111 | |
cf80e0e4 | 1112 | if (crypto_ahash_init(req)) |
49a72dfb AL |
1113 | goto clear_hash; |
1114 | ||
19689e38 | 1115 | if (tcp_v4_md5_hash_headers(hp, daddr, saddr, th, skb->len)) |
49a72dfb AL |
1116 | goto clear_hash; |
1117 | if (tcp_md5_hash_skb_data(hp, skb, th->doff << 2)) | |
1118 | goto clear_hash; | |
1119 | if (tcp_md5_hash_key(hp, key)) | |
1120 | goto clear_hash; | |
cf80e0e4 HX |
1121 | ahash_request_set_crypt(req, NULL, md5_hash, 0); |
1122 | if (crypto_ahash_final(req)) | |
49a72dfb AL |
1123 | goto clear_hash; |
1124 | ||
1125 | tcp_put_md5sig_pool(); | |
1126 | return 0; | |
1127 | ||
1128 | clear_hash: | |
1129 | tcp_put_md5sig_pool(); | |
1130 | clear_hash_noput: | |
1131 | memset(md5_hash, 0, 16); | |
1132 | return 1; | |
cfb6eeb4 | 1133 | } |
49a72dfb | 1134 | EXPORT_SYMBOL(tcp_v4_md5_hash_skb); |
cfb6eeb4 | 1135 | |
ba8e275a ED |
1136 | #endif |
1137 | ||
ff74e23f | 1138 | /* Called with rcu_read_lock() */ |
ba8e275a | 1139 | static bool tcp_v4_inbound_md5_hash(const struct sock *sk, |
ff74e23f | 1140 | const struct sk_buff *skb) |
cfb6eeb4 | 1141 | { |
ba8e275a | 1142 | #ifdef CONFIG_TCP_MD5SIG |
cfb6eeb4 YH |
1143 | /* |
1144 | * This gets called for each TCP segment that arrives | |
1145 | * so we want to be efficient. | |
1146 | * We have 3 drop cases: | |
1147 | * o No MD5 hash and one expected. | |
1148 | * o MD5 hash and we're not expecting one. | |
1149 | * o MD5 hash and its wrong. | |
1150 | */ | |
cf533ea5 | 1151 | const __u8 *hash_location = NULL; |
cfb6eeb4 | 1152 | struct tcp_md5sig_key *hash_expected; |
eddc9ec5 | 1153 | const struct iphdr *iph = ip_hdr(skb); |
cf533ea5 | 1154 | const struct tcphdr *th = tcp_hdr(skb); |
cfb6eeb4 | 1155 | int genhash; |
cfb6eeb4 YH |
1156 | unsigned char newhash[16]; |
1157 | ||
a915da9b ED |
1158 | hash_expected = tcp_md5_do_lookup(sk, (union tcp_md5_addr *)&iph->saddr, |
1159 | AF_INET); | |
7d5d5525 | 1160 | hash_location = tcp_parse_md5sig_option(th); |
cfb6eeb4 | 1161 | |
cfb6eeb4 YH |
1162 | /* We've parsed the options - do we have a hash? */ |
1163 | if (!hash_expected && !hash_location) | |
a2a385d6 | 1164 | return false; |
cfb6eeb4 YH |
1165 | |
1166 | if (hash_expected && !hash_location) { | |
c10d9310 | 1167 | NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPMD5NOTFOUND); |
a2a385d6 | 1168 | return true; |
cfb6eeb4 YH |
1169 | } |
1170 | ||
1171 | if (!hash_expected && hash_location) { | |
c10d9310 | 1172 | NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPMD5UNEXPECTED); |
a2a385d6 | 1173 | return true; |
cfb6eeb4 YH |
1174 | } |
1175 | ||
1176 | /* Okay, so this is hash_expected and hash_location - | |
1177 | * so we need to calculate the checksum. | |
1178 | */ | |
49a72dfb AL |
1179 | genhash = tcp_v4_md5_hash_skb(newhash, |
1180 | hash_expected, | |
39f8e58e | 1181 | NULL, skb); |
cfb6eeb4 YH |
1182 | |
1183 | if (genhash || memcmp(hash_location, newhash, 16) != 0) { | |
72145a68 | 1184 | NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPMD5FAILURE); |
e87cc472 JP |
1185 | net_info_ratelimited("MD5 Hash failed for (%pI4, %d)->(%pI4, %d)%s\n", |
1186 | &iph->saddr, ntohs(th->source), | |
1187 | &iph->daddr, ntohs(th->dest), | |
1188 | genhash ? " tcp_v4_calc_md5_hash failed" | |
1189 | : ""); | |
a2a385d6 | 1190 | return true; |
cfb6eeb4 | 1191 | } |
a2a385d6 | 1192 | return false; |
cfb6eeb4 | 1193 | #endif |
ba8e275a ED |
1194 | return false; |
1195 | } | |
cfb6eeb4 | 1196 | |
b40cf18e ED |
1197 | static void tcp_v4_init_req(struct request_sock *req, |
1198 | const struct sock *sk_listener, | |
16bea70a OP |
1199 | struct sk_buff *skb) |
1200 | { | |
1201 | struct inet_request_sock *ireq = inet_rsk(req); | |
1202 | ||
08d2cc3b ED |
1203 | sk_rcv_saddr_set(req_to_sk(req), ip_hdr(skb)->daddr); |
1204 | sk_daddr_set(req_to_sk(req), ip_hdr(skb)->saddr); | |
16bea70a OP |
1205 | ireq->opt = tcp_v4_save_options(skb); |
1206 | } | |
1207 | ||
f964629e ED |
1208 | static struct dst_entry *tcp_v4_route_req(const struct sock *sk, |
1209 | struct flowi *fl, | |
d94e0417 OP |
1210 | const struct request_sock *req, |
1211 | bool *strict) | |
1212 | { | |
1213 | struct dst_entry *dst = inet_csk_route_req(sk, &fl->u.ip4, req); | |
1214 | ||
1215 | if (strict) { | |
1216 | if (fl->u.ip4.daddr == inet_rsk(req)->ir_rmt_addr) | |
1217 | *strict = true; | |
1218 | else | |
1219 | *strict = false; | |
1220 | } | |
1221 | ||
1222 | return dst; | |
1223 | } | |
1224 | ||
72a3effa | 1225 | struct request_sock_ops tcp_request_sock_ops __read_mostly = { |
1da177e4 | 1226 | .family = PF_INET, |
2e6599cb | 1227 | .obj_size = sizeof(struct tcp_request_sock), |
5db92c99 | 1228 | .rtx_syn_ack = tcp_rtx_synack, |
60236fdd ACM |
1229 | .send_ack = tcp_v4_reqsk_send_ack, |
1230 | .destructor = tcp_v4_reqsk_destructor, | |
1da177e4 | 1231 | .send_reset = tcp_v4_send_reset, |
688d1945 | 1232 | .syn_ack_timeout = tcp_syn_ack_timeout, |
1da177e4 LT |
1233 | }; |
1234 | ||
b2e4b3de | 1235 | static const struct tcp_request_sock_ops tcp_request_sock_ipv4_ops = { |
2aec4a29 | 1236 | .mss_clamp = TCP_MSS_DEFAULT, |
16bea70a | 1237 | #ifdef CONFIG_TCP_MD5SIG |
fd3a154a | 1238 | .req_md5_lookup = tcp_v4_md5_lookup, |
e3afe7b7 | 1239 | .calc_md5_hash = tcp_v4_md5_hash_skb, |
b6332e6c | 1240 | #endif |
16bea70a | 1241 | .init_req = tcp_v4_init_req, |
fb7b37a7 OP |
1242 | #ifdef CONFIG_SYN_COOKIES |
1243 | .cookie_init_seq = cookie_v4_init_sequence, | |
1244 | #endif | |
d94e0417 | 1245 | .route_req = tcp_v4_route_req, |
936b8bdb | 1246 | .init_seq = tcp_v4_init_sequence, |
d6274bd8 | 1247 | .send_synack = tcp_v4_send_synack, |
16bea70a | 1248 | }; |
cfb6eeb4 | 1249 | |
1da177e4 LT |
1250 | int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb) |
1251 | { | |
1da177e4 | 1252 | /* Never answer to SYNs send to broadcast or multicast */ |
511c3f92 | 1253 | if (skb_rtable(skb)->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST)) |
1da177e4 LT |
1254 | goto drop; |
1255 | ||
1fb6f159 OP |
1256 | return tcp_conn_request(&tcp_request_sock_ops, |
1257 | &tcp_request_sock_ipv4_ops, sk, skb); | |
1da177e4 | 1258 | |
1da177e4 | 1259 | drop: |
9caad864 | 1260 | tcp_listendrop(sk); |
1da177e4 LT |
1261 | return 0; |
1262 | } | |
4bc2f18b | 1263 | EXPORT_SYMBOL(tcp_v4_conn_request); |
1da177e4 LT |
1264 | |
1265 | ||
1266 | /* | |
1267 | * The three way handshake has completed - we got a valid synack - | |
1268 | * now create the new socket. | |
1269 | */ | |
0c27171e | 1270 | struct sock *tcp_v4_syn_recv_sock(const struct sock *sk, struct sk_buff *skb, |
60236fdd | 1271 | struct request_sock *req, |
5e0724d0 ED |
1272 | struct dst_entry *dst, |
1273 | struct request_sock *req_unhash, | |
1274 | bool *own_req) | |
1da177e4 | 1275 | { |
2e6599cb | 1276 | struct inet_request_sock *ireq; |
1da177e4 LT |
1277 | struct inet_sock *newinet; |
1278 | struct tcp_sock *newtp; | |
1279 | struct sock *newsk; | |
cfb6eeb4 YH |
1280 | #ifdef CONFIG_TCP_MD5SIG |
1281 | struct tcp_md5sig_key *key; | |
1282 | #endif | |
f6d8bd05 | 1283 | struct ip_options_rcu *inet_opt; |
1da177e4 LT |
1284 | |
1285 | if (sk_acceptq_is_full(sk)) | |
1286 | goto exit_overflow; | |
1287 | ||
1da177e4 LT |
1288 | newsk = tcp_create_openreq_child(sk, req, skb); |
1289 | if (!newsk) | |
093d2823 | 1290 | goto exit_nonewsk; |
1da177e4 | 1291 | |
bcd76111 | 1292 | newsk->sk_gso_type = SKB_GSO_TCPV4; |
fae6ef87 | 1293 | inet_sk_rx_dst_set(newsk, skb); |
1da177e4 LT |
1294 | |
1295 | newtp = tcp_sk(newsk); | |
1296 | newinet = inet_sk(newsk); | |
2e6599cb | 1297 | ireq = inet_rsk(req); |
d1e559d0 ED |
1298 | sk_daddr_set(newsk, ireq->ir_rmt_addr); |
1299 | sk_rcv_saddr_set(newsk, ireq->ir_loc_addr); | |
6dd9a14e | 1300 | newsk->sk_bound_dev_if = ireq->ir_iif; |
634fb979 | 1301 | newinet->inet_saddr = ireq->ir_loc_addr; |
f6d8bd05 ED |
1302 | inet_opt = ireq->opt; |
1303 | rcu_assign_pointer(newinet->inet_opt, inet_opt); | |
2e6599cb | 1304 | ireq->opt = NULL; |
463c84b9 | 1305 | newinet->mc_index = inet_iif(skb); |
eddc9ec5 | 1306 | newinet->mc_ttl = ip_hdr(skb)->ttl; |
4c507d28 | 1307 | newinet->rcv_tos = ip_hdr(skb)->tos; |
d83d8461 | 1308 | inet_csk(newsk)->icsk_ext_hdr_len = 0; |
f6d8bd05 ED |
1309 | if (inet_opt) |
1310 | inet_csk(newsk)->icsk_ext_hdr_len = inet_opt->opt.optlen; | |
c720c7e8 | 1311 | newinet->inet_id = newtp->write_seq ^ jiffies; |
1da177e4 | 1312 | |
dfd25fff ED |
1313 | if (!dst) { |
1314 | dst = inet_csk_route_child_sock(sk, newsk, req); | |
1315 | if (!dst) | |
1316 | goto put_and_exit; | |
1317 | } else { | |
1318 | /* syncookie case : see end of cookie_v4_check() */ | |
1319 | } | |
0e734419 DM |
1320 | sk_setup_caps(newsk, dst); |
1321 | ||
81164413 DB |
1322 | tcp_ca_openreq_child(newsk, dst); |
1323 | ||
1da177e4 | 1324 | tcp_sync_mss(newsk, dst_mtu(dst)); |
0dbaee3b | 1325 | newtp->advmss = dst_metric_advmss(dst); |
f5fff5dc TQ |
1326 | if (tcp_sk(sk)->rx_opt.user_mss && |
1327 | tcp_sk(sk)->rx_opt.user_mss < newtp->advmss) | |
1328 | newtp->advmss = tcp_sk(sk)->rx_opt.user_mss; | |
1329 | ||
1da177e4 LT |
1330 | tcp_initialize_rcv_mss(newsk); |
1331 | ||
cfb6eeb4 YH |
1332 | #ifdef CONFIG_TCP_MD5SIG |
1333 | /* Copy over the MD5 key from the original socket */ | |
a915da9b ED |
1334 | key = tcp_md5_do_lookup(sk, (union tcp_md5_addr *)&newinet->inet_daddr, |
1335 | AF_INET); | |
00db4124 | 1336 | if (key) { |
cfb6eeb4 YH |
1337 | /* |
1338 | * We're using one, so create a matching key | |
1339 | * on the newsk structure. If we fail to get | |
1340 | * memory, then we end up not copying the key | |
1341 | * across. Shucks. | |
1342 | */ | |
a915da9b ED |
1343 | tcp_md5_do_add(newsk, (union tcp_md5_addr *)&newinet->inet_daddr, |
1344 | AF_INET, key->key, key->keylen, GFP_ATOMIC); | |
a465419b | 1345 | sk_nocaps_add(newsk, NETIF_F_GSO_MASK); |
cfb6eeb4 YH |
1346 | } |
1347 | #endif | |
1348 | ||
0e734419 DM |
1349 | if (__inet_inherit_port(sk, newsk) < 0) |
1350 | goto put_and_exit; | |
5e0724d0 | 1351 | *own_req = inet_ehash_nolisten(newsk, req_to_sk(req_unhash)); |
805c4bc0 | 1352 | if (*own_req) |
49a496c9 | 1353 | tcp_move_syn(newtp, req); |
1da177e4 LT |
1354 | |
1355 | return newsk; | |
1356 | ||
1357 | exit_overflow: | |
c10d9310 | 1358 | NET_INC_STATS(sock_net(sk), LINUX_MIB_LISTENOVERFLOWS); |
093d2823 BS |
1359 | exit_nonewsk: |
1360 | dst_release(dst); | |
1da177e4 | 1361 | exit: |
9caad864 | 1362 | tcp_listendrop(sk); |
1da177e4 | 1363 | return NULL; |
0e734419 | 1364 | put_and_exit: |
e337e24d CP |
1365 | inet_csk_prepare_forced_close(newsk); |
1366 | tcp_done(newsk); | |
0e734419 | 1367 | goto exit; |
1da177e4 | 1368 | } |
4bc2f18b | 1369 | EXPORT_SYMBOL(tcp_v4_syn_recv_sock); |
1da177e4 | 1370 | |
079096f1 | 1371 | static struct sock *tcp_v4_cookie_check(struct sock *sk, struct sk_buff *skb) |
1da177e4 | 1372 | { |
079096f1 | 1373 | #ifdef CONFIG_SYN_COOKIES |
52452c54 | 1374 | const struct tcphdr *th = tcp_hdr(skb); |
1da177e4 | 1375 | |
af9b4738 | 1376 | if (!th->syn) |
461b74c3 | 1377 | sk = cookie_v4_check(sk, skb); |
1da177e4 LT |
1378 | #endif |
1379 | return sk; | |
1380 | } | |
1381 | ||
1da177e4 | 1382 | /* The socket must have it's spinlock held when we get |
e994b2f0 | 1383 | * here, unless it is a TCP_LISTEN socket. |
1da177e4 LT |
1384 | * |
1385 | * We have a potential double-lock case here, so even when | |
1386 | * doing backlog processing we use the BH locking scheme. | |
1387 | * This is because we cannot sleep with the original spinlock | |
1388 | * held. | |
1389 | */ | |
1390 | int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb) | |
1391 | { | |
cfb6eeb4 | 1392 | struct sock *rsk; |
cfb6eeb4 | 1393 | |
1da177e4 | 1394 | if (sk->sk_state == TCP_ESTABLISHED) { /* Fast path */ |
404e0a8b ED |
1395 | struct dst_entry *dst = sk->sk_rx_dst; |
1396 | ||
bdeab991 | 1397 | sock_rps_save_rxhash(sk, skb); |
3d97379a | 1398 | sk_mark_napi_id(sk, skb); |
404e0a8b | 1399 | if (dst) { |
505fbcf0 | 1400 | if (inet_sk(sk)->rx_dst_ifindex != skb->skb_iif || |
51456b29 | 1401 | !dst->ops->check(dst, 0)) { |
92101b3b DM |
1402 | dst_release(dst); |
1403 | sk->sk_rx_dst = NULL; | |
1404 | } | |
1405 | } | |
c995ae22 | 1406 | tcp_rcv_established(sk, skb, tcp_hdr(skb), skb->len); |
1da177e4 LT |
1407 | return 0; |
1408 | } | |
1409 | ||
12e25e10 | 1410 | if (tcp_checksum_complete(skb)) |
1da177e4 LT |
1411 | goto csum_err; |
1412 | ||
1413 | if (sk->sk_state == TCP_LISTEN) { | |
079096f1 ED |
1414 | struct sock *nsk = tcp_v4_cookie_check(sk, skb); |
1415 | ||
1da177e4 LT |
1416 | if (!nsk) |
1417 | goto discard; | |
1da177e4 | 1418 | if (nsk != sk) { |
bdeab991 | 1419 | sock_rps_save_rxhash(nsk, skb); |
38cb5245 | 1420 | sk_mark_napi_id(nsk, skb); |
cfb6eeb4 YH |
1421 | if (tcp_child_process(sk, nsk, skb)) { |
1422 | rsk = nsk; | |
1da177e4 | 1423 | goto reset; |
cfb6eeb4 | 1424 | } |
1da177e4 LT |
1425 | return 0; |
1426 | } | |
ca55158c | 1427 | } else |
bdeab991 | 1428 | sock_rps_save_rxhash(sk, skb); |
ca55158c | 1429 | |
72ab4a86 | 1430 | if (tcp_rcv_state_process(sk, skb)) { |
cfb6eeb4 | 1431 | rsk = sk; |
1da177e4 | 1432 | goto reset; |
cfb6eeb4 | 1433 | } |
1da177e4 LT |
1434 | return 0; |
1435 | ||
1436 | reset: | |
cfb6eeb4 | 1437 | tcp_v4_send_reset(rsk, skb); |
1da177e4 LT |
1438 | discard: |
1439 | kfree_skb(skb); | |
1440 | /* Be careful here. If this function gets more complicated and | |
1441 | * gcc suffers from register pressure on the x86, sk (in %ebx) | |
1442 | * might be destroyed here. This current version compiles correctly, | |
1443 | * but you have been warned. | |
1444 | */ | |
1445 | return 0; | |
1446 | ||
1447 | csum_err: | |
c10d9310 ED |
1448 | TCP_INC_STATS(sock_net(sk), TCP_MIB_CSUMERRORS); |
1449 | TCP_INC_STATS(sock_net(sk), TCP_MIB_INERRS); | |
1da177e4 LT |
1450 | goto discard; |
1451 | } | |
4bc2f18b | 1452 | EXPORT_SYMBOL(tcp_v4_do_rcv); |
1da177e4 | 1453 | |
160eb5a6 | 1454 | void tcp_v4_early_demux(struct sk_buff *skb) |
41063e9d | 1455 | { |
41063e9d DM |
1456 | const struct iphdr *iph; |
1457 | const struct tcphdr *th; | |
1458 | struct sock *sk; | |
41063e9d | 1459 | |
41063e9d | 1460 | if (skb->pkt_type != PACKET_HOST) |
160eb5a6 | 1461 | return; |
41063e9d | 1462 | |
45f00f99 | 1463 | if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct tcphdr))) |
160eb5a6 | 1464 | return; |
41063e9d DM |
1465 | |
1466 | iph = ip_hdr(skb); | |
45f00f99 | 1467 | th = tcp_hdr(skb); |
41063e9d DM |
1468 | |
1469 | if (th->doff < sizeof(struct tcphdr) / 4) | |
160eb5a6 | 1470 | return; |
41063e9d | 1471 | |
45f00f99 | 1472 | sk = __inet_lookup_established(dev_net(skb->dev), &tcp_hashinfo, |
41063e9d | 1473 | iph->saddr, th->source, |
7011d085 | 1474 | iph->daddr, ntohs(th->dest), |
9cb429d6 | 1475 | skb->skb_iif); |
41063e9d DM |
1476 | if (sk) { |
1477 | skb->sk = sk; | |
1478 | skb->destructor = sock_edemux; | |
f7e4eb03 | 1479 | if (sk_fullsock(sk)) { |
d0c294c5 | 1480 | struct dst_entry *dst = READ_ONCE(sk->sk_rx_dst); |
505fbcf0 | 1481 | |
41063e9d DM |
1482 | if (dst) |
1483 | dst = dst_check(dst, 0); | |
92101b3b | 1484 | if (dst && |
505fbcf0 | 1485 | inet_sk(sk)->rx_dst_ifindex == skb->skb_iif) |
92101b3b | 1486 | skb_dst_set_noref(skb, dst); |
41063e9d DM |
1487 | } |
1488 | } | |
41063e9d DM |
1489 | } |
1490 | ||
b2fb4f54 ED |
1491 | /* Packet is added to VJ-style prequeue for processing in process |
1492 | * context, if a reader task is waiting. Apparently, this exciting | |
1493 | * idea (VJ's mail "Re: query about TCP header on tcp-ip" of 07 Sep 93) | |
1494 | * failed somewhere. Latency? Burstiness? Well, at least now we will | |
1495 | * see, why it failed. 8)8) --ANK | |
1496 | * | |
1497 | */ | |
1498 | bool tcp_prequeue(struct sock *sk, struct sk_buff *skb) | |
1499 | { | |
1500 | struct tcp_sock *tp = tcp_sk(sk); | |
1501 | ||
1502 | if (sysctl_tcp_low_latency || !tp->ucopy.task) | |
1503 | return false; | |
1504 | ||
1505 | if (skb->len <= tcp_hdrlen(skb) && | |
1506 | skb_queue_len(&tp->ucopy.prequeue) == 0) | |
1507 | return false; | |
1508 | ||
ca777eff ED |
1509 | /* Before escaping RCU protected region, we need to take care of skb |
1510 | * dst. Prequeue is only enabled for established sockets. | |
1511 | * For such sockets, we might need the skb dst only to set sk->sk_rx_dst | |
1512 | * Instead of doing full sk_rx_dst validity here, let's perform | |
1513 | * an optimistic check. | |
1514 | */ | |
1515 | if (likely(sk->sk_rx_dst)) | |
1516 | skb_dst_drop(skb); | |
1517 | else | |
5037e9ef | 1518 | skb_dst_force_safe(skb); |
ca777eff | 1519 | |
b2fb4f54 ED |
1520 | __skb_queue_tail(&tp->ucopy.prequeue, skb); |
1521 | tp->ucopy.memory += skb->truesize; | |
0cef6a4c ED |
1522 | if (skb_queue_len(&tp->ucopy.prequeue) >= 32 || |
1523 | tp->ucopy.memory + atomic_read(&sk->sk_rmem_alloc) > sk->sk_rcvbuf) { | |
b2fb4f54 ED |
1524 | struct sk_buff *skb1; |
1525 | ||
1526 | BUG_ON(sock_owned_by_user(sk)); | |
0cef6a4c ED |
1527 | __NET_ADD_STATS(sock_net(sk), LINUX_MIB_TCPPREQUEUEDROPPED, |
1528 | skb_queue_len(&tp->ucopy.prequeue)); | |
b2fb4f54 | 1529 | |
0cef6a4c | 1530 | while ((skb1 = __skb_dequeue(&tp->ucopy.prequeue)) != NULL) |
b2fb4f54 | 1531 | sk_backlog_rcv(sk, skb1); |
b2fb4f54 ED |
1532 | |
1533 | tp->ucopy.memory = 0; | |
1534 | } else if (skb_queue_len(&tp->ucopy.prequeue) == 1) { | |
1535 | wake_up_interruptible_sync_poll(sk_sleep(sk), | |
1536 | POLLIN | POLLRDNORM | POLLRDBAND); | |
1537 | if (!inet_csk_ack_scheduled(sk)) | |
1538 | inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK, | |
1539 | (3 * tcp_rto_min(sk)) / 4, | |
1540 | TCP_RTO_MAX); | |
1541 | } | |
1542 | return true; | |
1543 | } | |
1544 | EXPORT_SYMBOL(tcp_prequeue); | |
1545 | ||
c9c33212 ED |
1546 | bool tcp_add_backlog(struct sock *sk, struct sk_buff *skb) |
1547 | { | |
1548 | u32 limit = sk->sk_rcvbuf + sk->sk_sndbuf; | |
1549 | ||
1550 | /* Only socket owner can try to collapse/prune rx queues | |
1551 | * to reduce memory overhead, so add a little headroom here. | |
1552 | * Few sockets backlog are possibly concurrently non empty. | |
1553 | */ | |
1554 | limit += 64*1024; | |
1555 | ||
1556 | /* In case all data was pulled from skb frags (in __pskb_pull_tail()), | |
1557 | * we can fix skb->truesize to its real value to avoid future drops. | |
1558 | * This is valid because skb is not yet charged to the socket. | |
1559 | * It has been noticed pure SACK packets were sometimes dropped | |
1560 | * (if cooked by drivers without copybreak feature). | |
1561 | */ | |
1562 | if (!skb->data_len) | |
1563 | skb->truesize = SKB_TRUESIZE(skb_end_offset(skb)); | |
1564 | ||
1565 | if (unlikely(sk_add_backlog(sk, skb, limit))) { | |
1566 | bh_unlock_sock(sk); | |
1567 | __NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPBACKLOGDROP); | |
1568 | return true; | |
1569 | } | |
1570 | return false; | |
1571 | } | |
1572 | EXPORT_SYMBOL(tcp_add_backlog); | |
1573 | ||
ac6e7800 ED |
1574 | int tcp_filter(struct sock *sk, struct sk_buff *skb) |
1575 | { | |
1576 | struct tcphdr *th = (struct tcphdr *)skb->data; | |
1577 | unsigned int eaten = skb->len; | |
1578 | int err; | |
1579 | ||
1580 | err = sk_filter_trim_cap(sk, skb, th->doff * 4); | |
1581 | if (!err) { | |
1582 | eaten -= skb->len; | |
1583 | TCP_SKB_CB(skb)->end_seq -= eaten; | |
1584 | } | |
1585 | return err; | |
1586 | } | |
1587 | EXPORT_SYMBOL(tcp_filter); | |
1588 | ||
1da177e4 LT |
1589 | /* |
1590 | * From tcp_input.c | |
1591 | */ | |
1592 | ||
1593 | int tcp_v4_rcv(struct sk_buff *skb) | |
1594 | { | |
3b24d854 | 1595 | struct net *net = dev_net(skb->dev); |
eddc9ec5 | 1596 | const struct iphdr *iph; |
cf533ea5 | 1597 | const struct tcphdr *th; |
3b24d854 | 1598 | bool refcounted; |
1da177e4 LT |
1599 | struct sock *sk; |
1600 | int ret; | |
1601 | ||
1602 | if (skb->pkt_type != PACKET_HOST) | |
1603 | goto discard_it; | |
1604 | ||
1605 | /* Count it even if it's bad */ | |
90bbcc60 | 1606 | __TCP_INC_STATS(net, TCP_MIB_INSEGS); |
1da177e4 LT |
1607 | |
1608 | if (!pskb_may_pull(skb, sizeof(struct tcphdr))) | |
1609 | goto discard_it; | |
1610 | ||
ea1627c2 | 1611 | th = (const struct tcphdr *)skb->data; |
1da177e4 | 1612 | |
ea1627c2 | 1613 | if (unlikely(th->doff < sizeof(struct tcphdr) / 4)) |
1da177e4 LT |
1614 | goto bad_packet; |
1615 | if (!pskb_may_pull(skb, th->doff * 4)) | |
1616 | goto discard_it; | |
1617 | ||
1618 | /* An explanation is required here, I think. | |
1619 | * Packet length and doff are validated by header prediction, | |
caa20d9a | 1620 | * provided case of th->doff==0 is eliminated. |
1da177e4 | 1621 | * So, we defer the checks. */ |
ed70fcfc TH |
1622 | |
1623 | if (skb_checksum_init(skb, IPPROTO_TCP, inet_compute_pseudo)) | |
6a5dc9e5 | 1624 | goto csum_error; |
1da177e4 | 1625 | |
ea1627c2 | 1626 | th = (const struct tcphdr *)skb->data; |
eddc9ec5 | 1627 | iph = ip_hdr(skb); |
971f10ec ED |
1628 | /* This is tricky : We move IPCB at its correct location into TCP_SKB_CB() |
1629 | * barrier() makes sure compiler wont play fool^Waliasing games. | |
1630 | */ | |
1631 | memmove(&TCP_SKB_CB(skb)->header.h4, IPCB(skb), | |
1632 | sizeof(struct inet_skb_parm)); | |
1633 | barrier(); | |
1634 | ||
1da177e4 LT |
1635 | TCP_SKB_CB(skb)->seq = ntohl(th->seq); |
1636 | TCP_SKB_CB(skb)->end_seq = (TCP_SKB_CB(skb)->seq + th->syn + th->fin + | |
1637 | skb->len - th->doff * 4); | |
1638 | TCP_SKB_CB(skb)->ack_seq = ntohl(th->ack_seq); | |
e11ecddf | 1639 | TCP_SKB_CB(skb)->tcp_flags = tcp_flag_byte(th); |
04317daf | 1640 | TCP_SKB_CB(skb)->tcp_tw_isn = 0; |
b82d1bb4 | 1641 | TCP_SKB_CB(skb)->ip_dsfield = ipv4_get_dsfield(iph); |
1da177e4 LT |
1642 | TCP_SKB_CB(skb)->sacked = 0; |
1643 | ||
4bdc3d66 | 1644 | lookup: |
a583636a | 1645 | sk = __inet_lookup_skb(&tcp_hashinfo, skb, __tcp_hdrlen(th), th->source, |
3b24d854 | 1646 | th->dest, &refcounted); |
1da177e4 LT |
1647 | if (!sk) |
1648 | goto no_tcp_socket; | |
1649 | ||
bb134d5d ED |
1650 | process: |
1651 | if (sk->sk_state == TCP_TIME_WAIT) | |
1652 | goto do_time_wait; | |
1653 | ||
079096f1 ED |
1654 | if (sk->sk_state == TCP_NEW_SYN_RECV) { |
1655 | struct request_sock *req = inet_reqsk(sk); | |
7716682c | 1656 | struct sock *nsk; |
079096f1 ED |
1657 | |
1658 | sk = req->rsk_listener; | |
72923555 | 1659 | if (unlikely(tcp_v4_inbound_md5_hash(sk, skb))) { |
e65c332d | 1660 | sk_drops_add(sk, skb); |
72923555 ED |
1661 | reqsk_put(req); |
1662 | goto discard_it; | |
1663 | } | |
7716682c | 1664 | if (unlikely(sk->sk_state != TCP_LISTEN)) { |
f03f2e15 | 1665 | inet_csk_reqsk_queue_drop_and_put(sk, req); |
4bdc3d66 ED |
1666 | goto lookup; |
1667 | } | |
3b24d854 ED |
1668 | /* We own a reference on the listener, increase it again |
1669 | * as we might lose it too soon. | |
1670 | */ | |
7716682c | 1671 | sock_hold(sk); |
3b24d854 | 1672 | refcounted = true; |
7716682c | 1673 | nsk = tcp_check_req(sk, skb, req, false); |
079096f1 ED |
1674 | if (!nsk) { |
1675 | reqsk_put(req); | |
7716682c | 1676 | goto discard_and_relse; |
079096f1 ED |
1677 | } |
1678 | if (nsk == sk) { | |
079096f1 ED |
1679 | reqsk_put(req); |
1680 | } else if (tcp_child_process(sk, nsk, skb)) { | |
1681 | tcp_v4_send_reset(nsk, skb); | |
7716682c | 1682 | goto discard_and_relse; |
079096f1 | 1683 | } else { |
7716682c | 1684 | sock_put(sk); |
079096f1 ED |
1685 | return 0; |
1686 | } | |
1687 | } | |
6cce09f8 | 1688 | if (unlikely(iph->ttl < inet_sk(sk)->min_ttl)) { |
02a1d6e7 | 1689 | __NET_INC_STATS(net, LINUX_MIB_TCPMINTTLDROP); |
d218d111 | 1690 | goto discard_and_relse; |
6cce09f8 | 1691 | } |
d218d111 | 1692 | |
1da177e4 LT |
1693 | if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb)) |
1694 | goto discard_and_relse; | |
9ea88a15 | 1695 | |
9ea88a15 DP |
1696 | if (tcp_v4_inbound_md5_hash(sk, skb)) |
1697 | goto discard_and_relse; | |
9ea88a15 | 1698 | |
b59c2701 | 1699 | nf_reset(skb); |
1da177e4 | 1700 | |
ac6e7800 | 1701 | if (tcp_filter(sk, skb)) |
1da177e4 | 1702 | goto discard_and_relse; |
ac6e7800 ED |
1703 | th = (const struct tcphdr *)skb->data; |
1704 | iph = ip_hdr(skb); | |
1da177e4 LT |
1705 | |
1706 | skb->dev = NULL; | |
1707 | ||
e994b2f0 ED |
1708 | if (sk->sk_state == TCP_LISTEN) { |
1709 | ret = tcp_v4_do_rcv(sk, skb); | |
1710 | goto put_and_return; | |
1711 | } | |
1712 | ||
1713 | sk_incoming_cpu_update(sk); | |
1714 | ||
c6366184 | 1715 | bh_lock_sock_nested(sk); |
a44d6eac | 1716 | tcp_segs_in(tcp_sk(sk), skb); |
1da177e4 LT |
1717 | ret = 0; |
1718 | if (!sock_owned_by_user(sk)) { | |
7bced397 | 1719 | if (!tcp_prequeue(sk, skb)) |
1da177e4 | 1720 | ret = tcp_v4_do_rcv(sk, skb); |
c9c33212 | 1721 | } else if (tcp_add_backlog(sk, skb)) { |
6b03a53a ZY |
1722 | goto discard_and_relse; |
1723 | } | |
1da177e4 LT |
1724 | bh_unlock_sock(sk); |
1725 | ||
e994b2f0 | 1726 | put_and_return: |
3b24d854 ED |
1727 | if (refcounted) |
1728 | sock_put(sk); | |
1da177e4 LT |
1729 | |
1730 | return ret; | |
1731 | ||
1732 | no_tcp_socket: | |
1733 | if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) | |
1734 | goto discard_it; | |
1735 | ||
12e25e10 | 1736 | if (tcp_checksum_complete(skb)) { |
6a5dc9e5 | 1737 | csum_error: |
90bbcc60 | 1738 | __TCP_INC_STATS(net, TCP_MIB_CSUMERRORS); |
1da177e4 | 1739 | bad_packet: |
90bbcc60 | 1740 | __TCP_INC_STATS(net, TCP_MIB_INERRS); |
1da177e4 | 1741 | } else { |
cfb6eeb4 | 1742 | tcp_v4_send_reset(NULL, skb); |
1da177e4 LT |
1743 | } |
1744 | ||
1745 | discard_it: | |
1746 | /* Discard frame. */ | |
1747 | kfree_skb(skb); | |
e905a9ed | 1748 | return 0; |
1da177e4 LT |
1749 | |
1750 | discard_and_relse: | |
532182cd | 1751 | sk_drops_add(sk, skb); |
3b24d854 ED |
1752 | if (refcounted) |
1753 | sock_put(sk); | |
1da177e4 LT |
1754 | goto discard_it; |
1755 | ||
1756 | do_time_wait: | |
1757 | if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) { | |
9469c7b4 | 1758 | inet_twsk_put(inet_twsk(sk)); |
1da177e4 LT |
1759 | goto discard_it; |
1760 | } | |
1761 | ||
6a5dc9e5 ED |
1762 | if (tcp_checksum_complete(skb)) { |
1763 | inet_twsk_put(inet_twsk(sk)); | |
1764 | goto csum_error; | |
1da177e4 | 1765 | } |
9469c7b4 | 1766 | switch (tcp_timewait_state_process(inet_twsk(sk), skb, th)) { |
1da177e4 | 1767 | case TCP_TW_SYN: { |
c346dca1 | 1768 | struct sock *sk2 = inet_lookup_listener(dev_net(skb->dev), |
a583636a CG |
1769 | &tcp_hashinfo, skb, |
1770 | __tcp_hdrlen(th), | |
da5e3630 | 1771 | iph->saddr, th->source, |
eddc9ec5 | 1772 | iph->daddr, th->dest, |
463c84b9 | 1773 | inet_iif(skb)); |
1da177e4 | 1774 | if (sk2) { |
dbe7faa4 | 1775 | inet_twsk_deschedule_put(inet_twsk(sk)); |
1da177e4 | 1776 | sk = sk2; |
3b24d854 | 1777 | refcounted = false; |
1da177e4 LT |
1778 | goto process; |
1779 | } | |
1780 | /* Fall through to ACK */ | |
1781 | } | |
1782 | case TCP_TW_ACK: | |
1783 | tcp_v4_timewait_ack(sk, skb); | |
1784 | break; | |
1785 | case TCP_TW_RST: | |
271c3b9b FW |
1786 | tcp_v4_send_reset(sk, skb); |
1787 | inet_twsk_deschedule_put(inet_twsk(sk)); | |
1788 | goto discard_it; | |
1da177e4 LT |
1789 | case TCP_TW_SUCCESS:; |
1790 | } | |
1791 | goto discard_it; | |
1792 | } | |
1793 | ||
ccb7c410 DM |
1794 | static struct timewait_sock_ops tcp_timewait_sock_ops = { |
1795 | .twsk_obj_size = sizeof(struct tcp_timewait_sock), | |
1796 | .twsk_unique = tcp_twsk_unique, | |
1797 | .twsk_destructor= tcp_twsk_destructor, | |
ccb7c410 | 1798 | }; |
1da177e4 | 1799 | |
63d02d15 | 1800 | void inet_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb) |
5d299f3d ED |
1801 | { |
1802 | struct dst_entry *dst = skb_dst(skb); | |
1803 | ||
5037e9ef | 1804 | if (dst && dst_hold_safe(dst)) { |
ca777eff ED |
1805 | sk->sk_rx_dst = dst; |
1806 | inet_sk(sk)->rx_dst_ifindex = skb->skb_iif; | |
1807 | } | |
5d299f3d | 1808 | } |
63d02d15 | 1809 | EXPORT_SYMBOL(inet_sk_rx_dst_set); |
5d299f3d | 1810 | |
3b401a81 | 1811 | const struct inet_connection_sock_af_ops ipv4_specific = { |
543d9cfe ACM |
1812 | .queue_xmit = ip_queue_xmit, |
1813 | .send_check = tcp_v4_send_check, | |
1814 | .rebuild_header = inet_sk_rebuild_header, | |
5d299f3d | 1815 | .sk_rx_dst_set = inet_sk_rx_dst_set, |
543d9cfe ACM |
1816 | .conn_request = tcp_v4_conn_request, |
1817 | .syn_recv_sock = tcp_v4_syn_recv_sock, | |
543d9cfe ACM |
1818 | .net_header_len = sizeof(struct iphdr), |
1819 | .setsockopt = ip_setsockopt, | |
1820 | .getsockopt = ip_getsockopt, | |
1821 | .addr2sockaddr = inet_csk_addr2sockaddr, | |
1822 | .sockaddr_len = sizeof(struct sockaddr_in), | |
ab1e0a13 | 1823 | .bind_conflict = inet_csk_bind_conflict, |
3fdadf7d | 1824 | #ifdef CONFIG_COMPAT |
543d9cfe ACM |
1825 | .compat_setsockopt = compat_ip_setsockopt, |
1826 | .compat_getsockopt = compat_ip_getsockopt, | |
3fdadf7d | 1827 | #endif |
4fab9071 | 1828 | .mtu_reduced = tcp_v4_mtu_reduced, |
1da177e4 | 1829 | }; |
4bc2f18b | 1830 | EXPORT_SYMBOL(ipv4_specific); |
1da177e4 | 1831 | |
cfb6eeb4 | 1832 | #ifdef CONFIG_TCP_MD5SIG |
b2e4b3de | 1833 | static const struct tcp_sock_af_ops tcp_sock_ipv4_specific = { |
cfb6eeb4 | 1834 | .md5_lookup = tcp_v4_md5_lookup, |
49a72dfb | 1835 | .calc_md5_hash = tcp_v4_md5_hash_skb, |
cfb6eeb4 | 1836 | .md5_parse = tcp_v4_parse_md5_keys, |
cfb6eeb4 | 1837 | }; |
b6332e6c | 1838 | #endif |
cfb6eeb4 | 1839 | |
1da177e4 LT |
1840 | /* NOTE: A lot of things set to zero explicitly by call to |
1841 | * sk_alloc() so need not be done here. | |
1842 | */ | |
1843 | static int tcp_v4_init_sock(struct sock *sk) | |
1844 | { | |
6687e988 | 1845 | struct inet_connection_sock *icsk = inet_csk(sk); |
1da177e4 | 1846 | |
900f65d3 | 1847 | tcp_init_sock(sk); |
1da177e4 | 1848 | |
8292a17a | 1849 | icsk->icsk_af_ops = &ipv4_specific; |
900f65d3 | 1850 | |
cfb6eeb4 | 1851 | #ifdef CONFIG_TCP_MD5SIG |
ac807fa8 | 1852 | tcp_sk(sk)->af_specific = &tcp_sock_ipv4_specific; |
cfb6eeb4 | 1853 | #endif |
1da177e4 | 1854 | |
1da177e4 LT |
1855 | return 0; |
1856 | } | |
1857 | ||
7d06b2e0 | 1858 | void tcp_v4_destroy_sock(struct sock *sk) |
1da177e4 LT |
1859 | { |
1860 | struct tcp_sock *tp = tcp_sk(sk); | |
1861 | ||
1862 | tcp_clear_xmit_timers(sk); | |
1863 | ||
6687e988 | 1864 | tcp_cleanup_congestion_control(sk); |
317a76f9 | 1865 | |
1da177e4 | 1866 | /* Cleanup up the write buffer. */ |
fe067e8a | 1867 | tcp_write_queue_purge(sk); |
1da177e4 LT |
1868 | |
1869 | /* Cleans up our, hopefully empty, out_of_order_queue. */ | |
9f5afeae | 1870 | skb_rbtree_purge(&tp->out_of_order_queue); |
1da177e4 | 1871 | |
cfb6eeb4 YH |
1872 | #ifdef CONFIG_TCP_MD5SIG |
1873 | /* Clean up the MD5 key list, if any */ | |
1874 | if (tp->md5sig_info) { | |
a915da9b | 1875 | tcp_clear_md5_list(sk); |
a8afca03 | 1876 | kfree_rcu(tp->md5sig_info, rcu); |
cfb6eeb4 YH |
1877 | tp->md5sig_info = NULL; |
1878 | } | |
1879 | #endif | |
1a2449a8 | 1880 | |
1da177e4 LT |
1881 | /* Clean prequeue, it must be empty really */ |
1882 | __skb_queue_purge(&tp->ucopy.prequeue); | |
1883 | ||
1884 | /* Clean up a referenced TCP bind bucket. */ | |
463c84b9 | 1885 | if (inet_csk(sk)->icsk_bind_hash) |
ab1e0a13 | 1886 | inet_put_port(sk); |
1da177e4 | 1887 | |
00db4124 | 1888 | BUG_ON(tp->fastopen_rsk); |
435cf559 | 1889 | |
cf60af03 YC |
1890 | /* If socket is aborted during connect operation */ |
1891 | tcp_free_fastopen_req(tp); | |
cd8ae852 | 1892 | tcp_saved_syn_free(tp); |
cf60af03 | 1893 | |
777c6ae5 | 1894 | local_bh_disable(); |
180d8cd9 | 1895 | sk_sockets_allocated_dec(sk); |
777c6ae5 | 1896 | local_bh_enable(); |
1da177e4 | 1897 | } |
1da177e4 LT |
1898 | EXPORT_SYMBOL(tcp_v4_destroy_sock); |
1899 | ||
1900 | #ifdef CONFIG_PROC_FS | |
1901 | /* Proc filesystem TCP sock list dumping. */ | |
1902 | ||
a8b690f9 TH |
1903 | /* |
1904 | * Get next listener socket follow cur. If cur is NULL, get first socket | |
1905 | * starting from bucket given in st->bucket; when st->bucket is zero the | |
1906 | * very first socket in the hash table is returned. | |
1907 | */ | |
1da177e4 LT |
1908 | static void *listening_get_next(struct seq_file *seq, void *cur) |
1909 | { | |
5799de0b | 1910 | struct tcp_iter_state *st = seq->private; |
a4146b1b | 1911 | struct net *net = seq_file_net(seq); |
3b24d854 | 1912 | struct inet_listen_hashbucket *ilb; |
3b24d854 | 1913 | struct sock *sk = cur; |
1da177e4 LT |
1914 | |
1915 | if (!sk) { | |
3b24d854 | 1916 | get_head: |
a8b690f9 | 1917 | ilb = &tcp_hashinfo.listening_hash[st->bucket]; |
9652dc2e | 1918 | spin_lock(&ilb->lock); |
3b24d854 | 1919 | sk = sk_head(&ilb->head); |
a8b690f9 | 1920 | st->offset = 0; |
1da177e4 LT |
1921 | goto get_sk; |
1922 | } | |
5caea4ea | 1923 | ilb = &tcp_hashinfo.listening_hash[st->bucket]; |
1da177e4 | 1924 | ++st->num; |
a8b690f9 | 1925 | ++st->offset; |
1da177e4 | 1926 | |
3b24d854 | 1927 | sk = sk_next(sk); |
1da177e4 | 1928 | get_sk: |
3b24d854 | 1929 | sk_for_each_from(sk) { |
8475ef9f PE |
1930 | if (!net_eq(sock_net(sk), net)) |
1931 | continue; | |
3b24d854 ED |
1932 | if (sk->sk_family == st->family) |
1933 | return sk; | |
1da177e4 | 1934 | } |
9652dc2e | 1935 | spin_unlock(&ilb->lock); |
a8b690f9 | 1936 | st->offset = 0; |
3b24d854 ED |
1937 | if (++st->bucket < INET_LHTABLE_SIZE) |
1938 | goto get_head; | |
1939 | return NULL; | |
1da177e4 LT |
1940 | } |
1941 | ||
1942 | static void *listening_get_idx(struct seq_file *seq, loff_t *pos) | |
1943 | { | |
a8b690f9 TH |
1944 | struct tcp_iter_state *st = seq->private; |
1945 | void *rc; | |
1946 | ||
1947 | st->bucket = 0; | |
1948 | st->offset = 0; | |
1949 | rc = listening_get_next(seq, NULL); | |
1da177e4 LT |
1950 | |
1951 | while (rc && *pos) { | |
1952 | rc = listening_get_next(seq, rc); | |
1953 | --*pos; | |
1954 | } | |
1955 | return rc; | |
1956 | } | |
1957 | ||
05dbc7b5 | 1958 | static inline bool empty_bucket(const struct tcp_iter_state *st) |
6eac5604 | 1959 | { |
05dbc7b5 | 1960 | return hlist_nulls_empty(&tcp_hashinfo.ehash[st->bucket].chain); |
6eac5604 AK |
1961 | } |
1962 | ||
a8b690f9 TH |
1963 | /* |
1964 | * Get first established socket starting from bucket given in st->bucket. | |
1965 | * If st->bucket is zero, the very first socket in the hash is returned. | |
1966 | */ | |
1da177e4 LT |
1967 | static void *established_get_first(struct seq_file *seq) |
1968 | { | |
5799de0b | 1969 | struct tcp_iter_state *st = seq->private; |
a4146b1b | 1970 | struct net *net = seq_file_net(seq); |
1da177e4 LT |
1971 | void *rc = NULL; |
1972 | ||
a8b690f9 TH |
1973 | st->offset = 0; |
1974 | for (; st->bucket <= tcp_hashinfo.ehash_mask; ++st->bucket) { | |
1da177e4 | 1975 | struct sock *sk; |
3ab5aee7 | 1976 | struct hlist_nulls_node *node; |
9db66bdc | 1977 | spinlock_t *lock = inet_ehash_lockp(&tcp_hashinfo, st->bucket); |
1da177e4 | 1978 | |
6eac5604 AK |
1979 | /* Lockless fast path for the common case of empty buckets */ |
1980 | if (empty_bucket(st)) | |
1981 | continue; | |
1982 | ||
9db66bdc | 1983 | spin_lock_bh(lock); |
3ab5aee7 | 1984 | sk_nulls_for_each(sk, node, &tcp_hashinfo.ehash[st->bucket].chain) { |
f40c8174 | 1985 | if (sk->sk_family != st->family || |
878628fb | 1986 | !net_eq(sock_net(sk), net)) { |
1da177e4 LT |
1987 | continue; |
1988 | } | |
1989 | rc = sk; | |
1990 | goto out; | |
1991 | } | |
9db66bdc | 1992 | spin_unlock_bh(lock); |
1da177e4 LT |
1993 | } |
1994 | out: | |
1995 | return rc; | |
1996 | } | |
1997 | ||
1998 | static void *established_get_next(struct seq_file *seq, void *cur) | |
1999 | { | |
2000 | struct sock *sk = cur; | |
3ab5aee7 | 2001 | struct hlist_nulls_node *node; |
5799de0b | 2002 | struct tcp_iter_state *st = seq->private; |
a4146b1b | 2003 | struct net *net = seq_file_net(seq); |
1da177e4 LT |
2004 | |
2005 | ++st->num; | |
a8b690f9 | 2006 | ++st->offset; |
1da177e4 | 2007 | |
05dbc7b5 | 2008 | sk = sk_nulls_next(sk); |
1da177e4 | 2009 | |
3ab5aee7 | 2010 | sk_nulls_for_each_from(sk, node) { |
878628fb | 2011 | if (sk->sk_family == st->family && net_eq(sock_net(sk), net)) |
05dbc7b5 | 2012 | return sk; |
1da177e4 LT |
2013 | } |
2014 | ||
05dbc7b5 ED |
2015 | spin_unlock_bh(inet_ehash_lockp(&tcp_hashinfo, st->bucket)); |
2016 | ++st->bucket; | |
2017 | return established_get_first(seq); | |
1da177e4 LT |
2018 | } |
2019 | ||
2020 | static void *established_get_idx(struct seq_file *seq, loff_t pos) | |
2021 | { | |
a8b690f9 TH |
2022 | struct tcp_iter_state *st = seq->private; |
2023 | void *rc; | |
2024 | ||
2025 | st->bucket = 0; | |
2026 | rc = established_get_first(seq); | |
1da177e4 LT |
2027 | |
2028 | while (rc && pos) { | |
2029 | rc = established_get_next(seq, rc); | |
2030 | --pos; | |
7174259e | 2031 | } |
1da177e4 LT |
2032 | return rc; |
2033 | } | |
2034 | ||
2035 | static void *tcp_get_idx(struct seq_file *seq, loff_t pos) | |
2036 | { | |
2037 | void *rc; | |
5799de0b | 2038 | struct tcp_iter_state *st = seq->private; |
1da177e4 | 2039 | |
1da177e4 LT |
2040 | st->state = TCP_SEQ_STATE_LISTENING; |
2041 | rc = listening_get_idx(seq, &pos); | |
2042 | ||
2043 | if (!rc) { | |
1da177e4 LT |
2044 | st->state = TCP_SEQ_STATE_ESTABLISHED; |
2045 | rc = established_get_idx(seq, pos); | |
2046 | } | |
2047 | ||
2048 | return rc; | |
2049 | } | |
2050 | ||
a8b690f9 TH |
2051 | static void *tcp_seek_last_pos(struct seq_file *seq) |
2052 | { | |
2053 | struct tcp_iter_state *st = seq->private; | |
2054 | int offset = st->offset; | |
2055 | int orig_num = st->num; | |
2056 | void *rc = NULL; | |
2057 | ||
2058 | switch (st->state) { | |
a8b690f9 TH |
2059 | case TCP_SEQ_STATE_LISTENING: |
2060 | if (st->bucket >= INET_LHTABLE_SIZE) | |
2061 | break; | |
2062 | st->state = TCP_SEQ_STATE_LISTENING; | |
2063 | rc = listening_get_next(seq, NULL); | |
2064 | while (offset-- && rc) | |
2065 | rc = listening_get_next(seq, rc); | |
2066 | if (rc) | |
2067 | break; | |
2068 | st->bucket = 0; | |
05dbc7b5 | 2069 | st->state = TCP_SEQ_STATE_ESTABLISHED; |
a8b690f9 TH |
2070 | /* Fallthrough */ |
2071 | case TCP_SEQ_STATE_ESTABLISHED: | |
a8b690f9 TH |
2072 | if (st->bucket > tcp_hashinfo.ehash_mask) |
2073 | break; | |
2074 | rc = established_get_first(seq); | |
2075 | while (offset-- && rc) | |
2076 | rc = established_get_next(seq, rc); | |
2077 | } | |
2078 | ||
2079 | st->num = orig_num; | |
2080 | ||
2081 | return rc; | |
2082 | } | |
2083 | ||
1da177e4 LT |
2084 | static void *tcp_seq_start(struct seq_file *seq, loff_t *pos) |
2085 | { | |
5799de0b | 2086 | struct tcp_iter_state *st = seq->private; |
a8b690f9 TH |
2087 | void *rc; |
2088 | ||
2089 | if (*pos && *pos == st->last_pos) { | |
2090 | rc = tcp_seek_last_pos(seq); | |
2091 | if (rc) | |
2092 | goto out; | |
2093 | } | |
2094 | ||
1da177e4 LT |
2095 | st->state = TCP_SEQ_STATE_LISTENING; |
2096 | st->num = 0; | |
a8b690f9 TH |
2097 | st->bucket = 0; |
2098 | st->offset = 0; | |
2099 | rc = *pos ? tcp_get_idx(seq, *pos - 1) : SEQ_START_TOKEN; | |
2100 | ||
2101 | out: | |
2102 | st->last_pos = *pos; | |
2103 | return rc; | |
1da177e4 LT |
2104 | } |
2105 | ||
2106 | static void *tcp_seq_next(struct seq_file *seq, void *v, loff_t *pos) | |
2107 | { | |
a8b690f9 | 2108 | struct tcp_iter_state *st = seq->private; |
1da177e4 | 2109 | void *rc = NULL; |
1da177e4 LT |
2110 | |
2111 | if (v == SEQ_START_TOKEN) { | |
2112 | rc = tcp_get_idx(seq, 0); | |
2113 | goto out; | |
2114 | } | |
1da177e4 LT |
2115 | |
2116 | switch (st->state) { | |
1da177e4 LT |
2117 | case TCP_SEQ_STATE_LISTENING: |
2118 | rc = listening_get_next(seq, v); | |
2119 | if (!rc) { | |
1da177e4 | 2120 | st->state = TCP_SEQ_STATE_ESTABLISHED; |
a8b690f9 TH |
2121 | st->bucket = 0; |
2122 | st->offset = 0; | |
1da177e4 LT |
2123 | rc = established_get_first(seq); |
2124 | } | |
2125 | break; | |
2126 | case TCP_SEQ_STATE_ESTABLISHED: | |
1da177e4 LT |
2127 | rc = established_get_next(seq, v); |
2128 | break; | |
2129 | } | |
2130 | out: | |
2131 | ++*pos; | |
a8b690f9 | 2132 | st->last_pos = *pos; |
1da177e4 LT |
2133 | return rc; |
2134 | } | |
2135 | ||
2136 | static void tcp_seq_stop(struct seq_file *seq, void *v) | |
2137 | { | |
5799de0b | 2138 | struct tcp_iter_state *st = seq->private; |
1da177e4 LT |
2139 | |
2140 | switch (st->state) { | |
1da177e4 LT |
2141 | case TCP_SEQ_STATE_LISTENING: |
2142 | if (v != SEQ_START_TOKEN) | |
9652dc2e | 2143 | spin_unlock(&tcp_hashinfo.listening_hash[st->bucket].lock); |
1da177e4 | 2144 | break; |
1da177e4 LT |
2145 | case TCP_SEQ_STATE_ESTABLISHED: |
2146 | if (v) | |
9db66bdc | 2147 | spin_unlock_bh(inet_ehash_lockp(&tcp_hashinfo, st->bucket)); |
1da177e4 LT |
2148 | break; |
2149 | } | |
2150 | } | |
2151 | ||
73cb88ec | 2152 | int tcp_seq_open(struct inode *inode, struct file *file) |
1da177e4 | 2153 | { |
d9dda78b | 2154 | struct tcp_seq_afinfo *afinfo = PDE_DATA(inode); |
1da177e4 | 2155 | struct tcp_iter_state *s; |
52d6f3f1 | 2156 | int err; |
1da177e4 | 2157 | |
52d6f3f1 DL |
2158 | err = seq_open_net(inode, file, &afinfo->seq_ops, |
2159 | sizeof(struct tcp_iter_state)); | |
2160 | if (err < 0) | |
2161 | return err; | |
f40c8174 | 2162 | |
52d6f3f1 | 2163 | s = ((struct seq_file *)file->private_data)->private; |
1da177e4 | 2164 | s->family = afinfo->family; |
688d1945 | 2165 | s->last_pos = 0; |
f40c8174 DL |
2166 | return 0; |
2167 | } | |
73cb88ec | 2168 | EXPORT_SYMBOL(tcp_seq_open); |
f40c8174 | 2169 | |
6f8b13bc | 2170 | int tcp_proc_register(struct net *net, struct tcp_seq_afinfo *afinfo) |
1da177e4 LT |
2171 | { |
2172 | int rc = 0; | |
2173 | struct proc_dir_entry *p; | |
2174 | ||
9427c4b3 DL |
2175 | afinfo->seq_ops.start = tcp_seq_start; |
2176 | afinfo->seq_ops.next = tcp_seq_next; | |
2177 | afinfo->seq_ops.stop = tcp_seq_stop; | |
2178 | ||
84841c3c | 2179 | p = proc_create_data(afinfo->name, S_IRUGO, net->proc_net, |
73cb88ec | 2180 | afinfo->seq_fops, afinfo); |
84841c3c | 2181 | if (!p) |
1da177e4 LT |
2182 | rc = -ENOMEM; |
2183 | return rc; | |
2184 | } | |
4bc2f18b | 2185 | EXPORT_SYMBOL(tcp_proc_register); |
1da177e4 | 2186 | |
6f8b13bc | 2187 | void tcp_proc_unregister(struct net *net, struct tcp_seq_afinfo *afinfo) |
1da177e4 | 2188 | { |
ece31ffd | 2189 | remove_proc_entry(afinfo->name, net->proc_net); |
1da177e4 | 2190 | } |
4bc2f18b | 2191 | EXPORT_SYMBOL(tcp_proc_unregister); |
1da177e4 | 2192 | |
d4f06873 | 2193 | static void get_openreq4(const struct request_sock *req, |
aa3a0c8c | 2194 | struct seq_file *f, int i) |
1da177e4 | 2195 | { |
2e6599cb | 2196 | const struct inet_request_sock *ireq = inet_rsk(req); |
fa76ce73 | 2197 | long delta = req->rsk_timer.expires - jiffies; |
1da177e4 | 2198 | |
5e659e4c | 2199 | seq_printf(f, "%4d: %08X:%04X %08X:%04X" |
652586df | 2200 | " %02X %08X:%08X %02X:%08lX %08X %5u %8d %u %d %pK", |
1da177e4 | 2201 | i, |
634fb979 | 2202 | ireq->ir_loc_addr, |
d4f06873 | 2203 | ireq->ir_num, |
634fb979 ED |
2204 | ireq->ir_rmt_addr, |
2205 | ntohs(ireq->ir_rmt_port), | |
1da177e4 LT |
2206 | TCP_SYN_RECV, |
2207 | 0, 0, /* could print option size, but that is af dependent. */ | |
2208 | 1, /* timers active (only the expire timer) */ | |
a399a805 | 2209 | jiffies_delta_to_clock_t(delta), |
e6c022a4 | 2210 | req->num_timeout, |
aa3a0c8c ED |
2211 | from_kuid_munged(seq_user_ns(f), |
2212 | sock_i_uid(req->rsk_listener)), | |
1da177e4 LT |
2213 | 0, /* non standard timer */ |
2214 | 0, /* open_requests have no inode */ | |
d4f06873 | 2215 | 0, |
652586df | 2216 | req); |
1da177e4 LT |
2217 | } |
2218 | ||
652586df | 2219 | static void get_tcp4_sock(struct sock *sk, struct seq_file *f, int i) |
1da177e4 LT |
2220 | { |
2221 | int timer_active; | |
2222 | unsigned long timer_expires; | |
cf533ea5 | 2223 | const struct tcp_sock *tp = tcp_sk(sk); |
cf4c6bf8 | 2224 | const struct inet_connection_sock *icsk = inet_csk(sk); |
cf533ea5 | 2225 | const struct inet_sock *inet = inet_sk(sk); |
0536fcc0 | 2226 | const struct fastopen_queue *fastopenq = &icsk->icsk_accept_queue.fastopenq; |
c720c7e8 ED |
2227 | __be32 dest = inet->inet_daddr; |
2228 | __be32 src = inet->inet_rcv_saddr; | |
2229 | __u16 destp = ntohs(inet->inet_dport); | |
2230 | __u16 srcp = ntohs(inet->inet_sport); | |
49d09007 | 2231 | int rx_queue; |
00fd38d9 | 2232 | int state; |
1da177e4 | 2233 | |
6ba8a3b1 ND |
2234 | if (icsk->icsk_pending == ICSK_TIME_RETRANS || |
2235 | icsk->icsk_pending == ICSK_TIME_EARLY_RETRANS || | |
2236 | icsk->icsk_pending == ICSK_TIME_LOSS_PROBE) { | |
1da177e4 | 2237 | timer_active = 1; |
463c84b9 ACM |
2238 | timer_expires = icsk->icsk_timeout; |
2239 | } else if (icsk->icsk_pending == ICSK_TIME_PROBE0) { | |
1da177e4 | 2240 | timer_active = 4; |
463c84b9 | 2241 | timer_expires = icsk->icsk_timeout; |
cf4c6bf8 | 2242 | } else if (timer_pending(&sk->sk_timer)) { |
1da177e4 | 2243 | timer_active = 2; |
cf4c6bf8 | 2244 | timer_expires = sk->sk_timer.expires; |
1da177e4 LT |
2245 | } else { |
2246 | timer_active = 0; | |
2247 | timer_expires = jiffies; | |
2248 | } | |
2249 | ||
00fd38d9 ED |
2250 | state = sk_state_load(sk); |
2251 | if (state == TCP_LISTEN) | |
49d09007 ED |
2252 | rx_queue = sk->sk_ack_backlog; |
2253 | else | |
00fd38d9 ED |
2254 | /* Because we don't lock the socket, |
2255 | * we might find a transient negative value. | |
49d09007 ED |
2256 | */ |
2257 | rx_queue = max_t(int, tp->rcv_nxt - tp->copied_seq, 0); | |
2258 | ||
5e659e4c | 2259 | seq_printf(f, "%4d: %08X:%04X %08X:%04X %02X %08X:%08X %02X:%08lX " |
652586df | 2260 | "%08X %5u %8d %lu %d %pK %lu %lu %u %u %d", |
00fd38d9 | 2261 | i, src, srcp, dest, destp, state, |
47da8ee6 | 2262 | tp->write_seq - tp->snd_una, |
49d09007 | 2263 | rx_queue, |
1da177e4 | 2264 | timer_active, |
a399a805 | 2265 | jiffies_delta_to_clock_t(timer_expires - jiffies), |
463c84b9 | 2266 | icsk->icsk_retransmits, |
a7cb5a49 | 2267 | from_kuid_munged(seq_user_ns(f), sock_i_uid(sk)), |
6687e988 | 2268 | icsk->icsk_probes_out, |
cf4c6bf8 IJ |
2269 | sock_i_ino(sk), |
2270 | atomic_read(&sk->sk_refcnt), sk, | |
7be87351 SH |
2271 | jiffies_to_clock_t(icsk->icsk_rto), |
2272 | jiffies_to_clock_t(icsk->icsk_ack.ato), | |
463c84b9 | 2273 | (icsk->icsk_ack.quick << 1) | icsk->icsk_ack.pingpong, |
1da177e4 | 2274 | tp->snd_cwnd, |
00fd38d9 ED |
2275 | state == TCP_LISTEN ? |
2276 | fastopenq->max_qlen : | |
652586df | 2277 | (tcp_in_initial_slowstart(tp) ? -1 : tp->snd_ssthresh)); |
1da177e4 LT |
2278 | } |
2279 | ||
cf533ea5 | 2280 | static void get_timewait4_sock(const struct inet_timewait_sock *tw, |
652586df | 2281 | struct seq_file *f, int i) |
1da177e4 | 2282 | { |
789f558c | 2283 | long delta = tw->tw_timer.expires - jiffies; |
23f33c2d | 2284 | __be32 dest, src; |
1da177e4 | 2285 | __u16 destp, srcp; |
1da177e4 LT |
2286 | |
2287 | dest = tw->tw_daddr; | |
2288 | src = tw->tw_rcv_saddr; | |
2289 | destp = ntohs(tw->tw_dport); | |
2290 | srcp = ntohs(tw->tw_sport); | |
2291 | ||
5e659e4c | 2292 | seq_printf(f, "%4d: %08X:%04X %08X:%04X" |
652586df | 2293 | " %02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %pK", |
1da177e4 | 2294 | i, src, srcp, dest, destp, tw->tw_substate, 0, 0, |
a399a805 | 2295 | 3, jiffies_delta_to_clock_t(delta), 0, 0, 0, 0, |
652586df | 2296 | atomic_read(&tw->tw_refcnt), tw); |
1da177e4 LT |
2297 | } |
2298 | ||
2299 | #define TMPSZ 150 | |
2300 | ||
2301 | static int tcp4_seq_show(struct seq_file *seq, void *v) | |
2302 | { | |
5799de0b | 2303 | struct tcp_iter_state *st; |
05dbc7b5 | 2304 | struct sock *sk = v; |
1da177e4 | 2305 | |
652586df | 2306 | seq_setwidth(seq, TMPSZ - 1); |
1da177e4 | 2307 | if (v == SEQ_START_TOKEN) { |
652586df | 2308 | seq_puts(seq, " sl local_address rem_address st tx_queue " |
1da177e4 LT |
2309 | "rx_queue tr tm->when retrnsmt uid timeout " |
2310 | "inode"); | |
2311 | goto out; | |
2312 | } | |
2313 | st = seq->private; | |
2314 | ||
079096f1 ED |
2315 | if (sk->sk_state == TCP_TIME_WAIT) |
2316 | get_timewait4_sock(v, seq, st->num); | |
2317 | else if (sk->sk_state == TCP_NEW_SYN_RECV) | |
aa3a0c8c | 2318 | get_openreq4(v, seq, st->num); |
079096f1 ED |
2319 | else |
2320 | get_tcp4_sock(v, seq, st->num); | |
1da177e4 | 2321 | out: |
652586df | 2322 | seq_pad(seq, '\n'); |
1da177e4 LT |
2323 | return 0; |
2324 | } | |
2325 | ||
73cb88ec AV |
2326 | static const struct file_operations tcp_afinfo_seq_fops = { |
2327 | .owner = THIS_MODULE, | |
2328 | .open = tcp_seq_open, | |
2329 | .read = seq_read, | |
2330 | .llseek = seq_lseek, | |
2331 | .release = seq_release_net | |
2332 | }; | |
2333 | ||
1da177e4 | 2334 | static struct tcp_seq_afinfo tcp4_seq_afinfo = { |
1da177e4 LT |
2335 | .name = "tcp", |
2336 | .family = AF_INET, | |
73cb88ec | 2337 | .seq_fops = &tcp_afinfo_seq_fops, |
9427c4b3 DL |
2338 | .seq_ops = { |
2339 | .show = tcp4_seq_show, | |
2340 | }, | |
1da177e4 LT |
2341 | }; |
2342 | ||
2c8c1e72 | 2343 | static int __net_init tcp4_proc_init_net(struct net *net) |
757764f6 PE |
2344 | { |
2345 | return tcp_proc_register(net, &tcp4_seq_afinfo); | |
2346 | } | |
2347 | ||
2c8c1e72 | 2348 | static void __net_exit tcp4_proc_exit_net(struct net *net) |
757764f6 PE |
2349 | { |
2350 | tcp_proc_unregister(net, &tcp4_seq_afinfo); | |
2351 | } | |
2352 | ||
2353 | static struct pernet_operations tcp4_net_ops = { | |
2354 | .init = tcp4_proc_init_net, | |
2355 | .exit = tcp4_proc_exit_net, | |
2356 | }; | |
2357 | ||
1da177e4 LT |
2358 | int __init tcp4_proc_init(void) |
2359 | { | |
757764f6 | 2360 | return register_pernet_subsys(&tcp4_net_ops); |
1da177e4 LT |
2361 | } |
2362 | ||
2363 | void tcp4_proc_exit(void) | |
2364 | { | |
757764f6 | 2365 | unregister_pernet_subsys(&tcp4_net_ops); |
1da177e4 LT |
2366 | } |
2367 | #endif /* CONFIG_PROC_FS */ | |
2368 | ||
2369 | struct proto tcp_prot = { | |
2370 | .name = "TCP", | |
2371 | .owner = THIS_MODULE, | |
2372 | .close = tcp_close, | |
2373 | .connect = tcp_v4_connect, | |
2374 | .disconnect = tcp_disconnect, | |
463c84b9 | 2375 | .accept = inet_csk_accept, |
1da177e4 LT |
2376 | .ioctl = tcp_ioctl, |
2377 | .init = tcp_v4_init_sock, | |
2378 | .destroy = tcp_v4_destroy_sock, | |
2379 | .shutdown = tcp_shutdown, | |
2380 | .setsockopt = tcp_setsockopt, | |
2381 | .getsockopt = tcp_getsockopt, | |
1da177e4 | 2382 | .recvmsg = tcp_recvmsg, |
7ba42910 CG |
2383 | .sendmsg = tcp_sendmsg, |
2384 | .sendpage = tcp_sendpage, | |
1da177e4 | 2385 | .backlog_rcv = tcp_v4_do_rcv, |
46d3ceab | 2386 | .release_cb = tcp_release_cb, |
ab1e0a13 ACM |
2387 | .hash = inet_hash, |
2388 | .unhash = inet_unhash, | |
2389 | .get_port = inet_csk_get_port, | |
1da177e4 | 2390 | .enter_memory_pressure = tcp_enter_memory_pressure, |
c9bee3b7 | 2391 | .stream_memory_free = tcp_stream_memory_free, |
1da177e4 | 2392 | .sockets_allocated = &tcp_sockets_allocated, |
0a5578cf | 2393 | .orphan_count = &tcp_orphan_count, |
1da177e4 LT |
2394 | .memory_allocated = &tcp_memory_allocated, |
2395 | .memory_pressure = &tcp_memory_pressure, | |
a4fe34bf | 2396 | .sysctl_mem = sysctl_tcp_mem, |
1da177e4 LT |
2397 | .sysctl_wmem = sysctl_tcp_wmem, |
2398 | .sysctl_rmem = sysctl_tcp_rmem, | |
2399 | .max_header = MAX_TCP_HEADER, | |
2400 | .obj_size = sizeof(struct tcp_sock), | |
3ab5aee7 | 2401 | .slab_flags = SLAB_DESTROY_BY_RCU, |
6d6ee43e | 2402 | .twsk_prot = &tcp_timewait_sock_ops, |
60236fdd | 2403 | .rsk_prot = &tcp_request_sock_ops, |
39d8cda7 | 2404 | .h.hashinfo = &tcp_hashinfo, |
7ba42910 | 2405 | .no_autobind = true, |
543d9cfe ACM |
2406 | #ifdef CONFIG_COMPAT |
2407 | .compat_setsockopt = compat_tcp_setsockopt, | |
2408 | .compat_getsockopt = compat_tcp_getsockopt, | |
d1a4c0b3 | 2409 | #endif |
c1e64e29 | 2410 | .diag_destroy = tcp_abort, |
1da177e4 | 2411 | }; |
4bc2f18b | 2412 | EXPORT_SYMBOL(tcp_prot); |
1da177e4 | 2413 | |
bdbbb852 ED |
2414 | static void __net_exit tcp_sk_exit(struct net *net) |
2415 | { | |
2416 | int cpu; | |
2417 | ||
2418 | for_each_possible_cpu(cpu) | |
2419 | inet_ctl_sock_destroy(*per_cpu_ptr(net->ipv4.tcp_sk, cpu)); | |
2420 | free_percpu(net->ipv4.tcp_sk); | |
2421 | } | |
2422 | ||
046ee902 DL |
2423 | static int __net_init tcp_sk_init(struct net *net) |
2424 | { | |
bdbbb852 ED |
2425 | int res, cpu; |
2426 | ||
2427 | net->ipv4.tcp_sk = alloc_percpu(struct sock *); | |
2428 | if (!net->ipv4.tcp_sk) | |
2429 | return -ENOMEM; | |
2430 | ||
2431 | for_each_possible_cpu(cpu) { | |
2432 | struct sock *sk; | |
2433 | ||
2434 | res = inet_ctl_sock_create(&sk, PF_INET, SOCK_RAW, | |
2435 | IPPROTO_TCP, net); | |
2436 | if (res) | |
2437 | goto fail; | |
a9d6532b | 2438 | sock_set_flag(sk, SOCK_USE_WRITE_QUEUE); |
bdbbb852 ED |
2439 | *per_cpu_ptr(net->ipv4.tcp_sk, cpu) = sk; |
2440 | } | |
49213555 | 2441 | |
5d134f1c | 2442 | net->ipv4.sysctl_tcp_ecn = 2; |
49213555 DB |
2443 | net->ipv4.sysctl_tcp_ecn_fallback = 1; |
2444 | ||
b0f9ca53 | 2445 | net->ipv4.sysctl_tcp_base_mss = TCP_BASE_MSS; |
6b58e0a5 | 2446 | net->ipv4.sysctl_tcp_probe_threshold = TCP_PROBE_THRESHOLD; |
05cbc0db | 2447 | net->ipv4.sysctl_tcp_probe_interval = TCP_PROBE_INTERVAL; |
046ee902 | 2448 | |
13b287e8 | 2449 | net->ipv4.sysctl_tcp_keepalive_time = TCP_KEEPALIVE_TIME; |
9bd6861b | 2450 | net->ipv4.sysctl_tcp_keepalive_probes = TCP_KEEPALIVE_PROBES; |
b840d15d | 2451 | net->ipv4.sysctl_tcp_keepalive_intvl = TCP_KEEPALIVE_INTVL; |
13b287e8 | 2452 | |
6fa25166 | 2453 | net->ipv4.sysctl_tcp_syn_retries = TCP_SYN_RETRIES; |
7c083ecb | 2454 | net->ipv4.sysctl_tcp_synack_retries = TCP_SYNACK_RETRIES; |
0aca737d | 2455 | net->ipv4.sysctl_tcp_syncookies = 1; |
1043e25f | 2456 | net->ipv4.sysctl_tcp_reordering = TCP_FASTRETRANS_THRESH; |
ae5c3f40 | 2457 | net->ipv4.sysctl_tcp_retries1 = TCP_RETR1; |
c6214a97 | 2458 | net->ipv4.sysctl_tcp_retries2 = TCP_RETR2; |
c402d9be | 2459 | net->ipv4.sysctl_tcp_orphan_retries = 0; |
1e579caa | 2460 | net->ipv4.sysctl_tcp_fin_timeout = TCP_FIN_TIMEOUT; |
4979f2d9 | 2461 | net->ipv4.sysctl_tcp_notsent_lowat = UINT_MAX; |
56ab6b93 | 2462 | net->ipv4.sysctl_tcp_tw_reuse = 0; |
12ed8244 | 2463 | |
49213555 | 2464 | return 0; |
bdbbb852 ED |
2465 | fail: |
2466 | tcp_sk_exit(net); | |
2467 | ||
2468 | return res; | |
b099ce26 EB |
2469 | } |
2470 | ||
2471 | static void __net_exit tcp_sk_exit_batch(struct list_head *net_exit_list) | |
2472 | { | |
2473 | inet_twsk_purge(&tcp_hashinfo, &tcp_death_row, AF_INET); | |
046ee902 DL |
2474 | } |
2475 | ||
2476 | static struct pernet_operations __net_initdata tcp_sk_ops = { | |
b099ce26 EB |
2477 | .init = tcp_sk_init, |
2478 | .exit = tcp_sk_exit, | |
2479 | .exit_batch = tcp_sk_exit_batch, | |
046ee902 DL |
2480 | }; |
2481 | ||
9b0f976f | 2482 | void __init tcp_v4_init(void) |
1da177e4 | 2483 | { |
5caea4ea | 2484 | inet_hashinfo_init(&tcp_hashinfo); |
6a1b3054 | 2485 | if (register_pernet_subsys(&tcp_sk_ops)) |
1da177e4 | 2486 | panic("Failed to create the TCP control socket.\n"); |
1da177e4 | 2487 | } |