Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * INET An implementation of the TCP/IP protocol suite for the LINUX | |
3 | * operating system. INET is implemented using the BSD Socket | |
4 | * interface as the means of communication with the user level. | |
5 | * | |
6 | * Implementation of the Transmission Control Protocol(TCP). | |
7 | * | |
1da177e4 LT |
8 | * IPv4 specific functions |
9 | * | |
10 | * | |
11 | * code split from: | |
12 | * linux/ipv4/tcp.c | |
13 | * linux/ipv4/tcp_input.c | |
14 | * linux/ipv4/tcp_output.c | |
15 | * | |
16 | * See tcp.c for author information | |
17 | * | |
18 | * This program is free software; you can redistribute it and/or | |
19 | * modify it under the terms of the GNU General Public License | |
20 | * as published by the Free Software Foundation; either version | |
21 | * 2 of the License, or (at your option) any later version. | |
22 | */ | |
23 | ||
24 | /* | |
25 | * Changes: | |
26 | * David S. Miller : New socket lookup architecture. | |
27 | * This code is dedicated to John Dyson. | |
28 | * David S. Miller : Change semantics of established hash, | |
29 | * half is devoted to TIME_WAIT sockets | |
30 | * and the rest go in the other half. | |
31 | * Andi Kleen : Add support for syncookies and fixed | |
32 | * some bugs: ip options weren't passed to | |
33 | * the TCP layer, missed a check for an | |
34 | * ACK bit. | |
35 | * Andi Kleen : Implemented fast path mtu discovery. | |
36 | * Fixed many serious bugs in the | |
60236fdd | 37 | * request_sock handling and moved |
1da177e4 LT |
38 | * most of it into the af independent code. |
39 | * Added tail drop and some other bugfixes. | |
caa20d9a | 40 | * Added new listen semantics. |
1da177e4 LT |
41 | * Mike McLagan : Routing by source |
42 | * Juan Jose Ciarlante: ip_dynaddr bits | |
43 | * Andi Kleen: various fixes. | |
44 | * Vitaly E. Lavrov : Transparent proxy revived after year | |
45 | * coma. | |
46 | * Andi Kleen : Fix new listen. | |
47 | * Andi Kleen : Fix accept error reporting. | |
48 | * YOSHIFUJI Hideaki @USAGI and: Support IPV6_V6ONLY socket option, which | |
49 | * Alexey Kuznetsov allow both IPv4 and IPv6 sockets to bind | |
50 | * a single port at the same time. | |
51 | */ | |
52 | ||
afd46503 | 53 | #define pr_fmt(fmt) "TCP: " fmt |
1da177e4 | 54 | |
eb4dea58 | 55 | #include <linux/bottom_half.h> |
1da177e4 LT |
56 | #include <linux/types.h> |
57 | #include <linux/fcntl.h> | |
58 | #include <linux/module.h> | |
59 | #include <linux/random.h> | |
60 | #include <linux/cache.h> | |
61 | #include <linux/jhash.h> | |
62 | #include <linux/init.h> | |
63 | #include <linux/times.h> | |
5a0e3ad6 | 64 | #include <linux/slab.h> |
1da177e4 | 65 | |
457c4cbc | 66 | #include <net/net_namespace.h> |
1da177e4 | 67 | #include <net/icmp.h> |
304a1618 | 68 | #include <net/inet_hashtables.h> |
1da177e4 | 69 | #include <net/tcp.h> |
20380731 | 70 | #include <net/transp_v6.h> |
1da177e4 LT |
71 | #include <net/ipv6.h> |
72 | #include <net/inet_common.h> | |
6d6ee43e | 73 | #include <net/timewait_sock.h> |
1da177e4 | 74 | #include <net/xfrm.h> |
1a2449a8 | 75 | #include <net/netdma.h> |
6e5714ea | 76 | #include <net/secure_seq.h> |
d1a4c0b3 | 77 | #include <net/tcp_memcontrol.h> |
076bb0c8 | 78 | #include <net/busy_poll.h> |
1da177e4 LT |
79 | |
80 | #include <linux/inet.h> | |
81 | #include <linux/ipv6.h> | |
82 | #include <linux/stddef.h> | |
83 | #include <linux/proc_fs.h> | |
84 | #include <linux/seq_file.h> | |
85 | ||
cfb6eeb4 YH |
86 | #include <linux/crypto.h> |
87 | #include <linux/scatterlist.h> | |
88 | ||
ab32ea5d BH |
89 | int sysctl_tcp_tw_reuse __read_mostly; |
90 | int sysctl_tcp_low_latency __read_mostly; | |
4bc2f18b | 91 | EXPORT_SYMBOL(sysctl_tcp_low_latency); |
1da177e4 | 92 | |
1da177e4 | 93 | |
cfb6eeb4 | 94 | #ifdef CONFIG_TCP_MD5SIG |
a915da9b | 95 | static int tcp_v4_md5_hash_hdr(char *md5_hash, const struct tcp_md5sig_key *key, |
318cf7aa | 96 | __be32 daddr, __be32 saddr, const struct tcphdr *th); |
cfb6eeb4 YH |
97 | #endif |
98 | ||
5caea4ea | 99 | struct inet_hashinfo tcp_hashinfo; |
4bc2f18b | 100 | EXPORT_SYMBOL(tcp_hashinfo); |
1da177e4 | 101 | |
936b8bdb | 102 | static __u32 tcp_v4_init_sequence(const struct sk_buff *skb) |
1da177e4 | 103 | { |
eddc9ec5 ACM |
104 | return secure_tcp_sequence_number(ip_hdr(skb)->daddr, |
105 | ip_hdr(skb)->saddr, | |
aa8223c7 ACM |
106 | tcp_hdr(skb)->dest, |
107 | tcp_hdr(skb)->source); | |
1da177e4 LT |
108 | } |
109 | ||
6d6ee43e ACM |
110 | int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp) |
111 | { | |
112 | const struct tcp_timewait_sock *tcptw = tcp_twsk(sktw); | |
113 | struct tcp_sock *tp = tcp_sk(sk); | |
114 | ||
115 | /* With PAWS, it is safe from the viewpoint | |
116 | of data integrity. Even without PAWS it is safe provided sequence | |
117 | spaces do not overlap i.e. at data rates <= 80Mbit/sec. | |
118 | ||
119 | Actually, the idea is close to VJ's one, only timestamp cache is | |
120 | held not per host, but per port pair and TW bucket is used as state | |
121 | holder. | |
122 | ||
123 | If TW bucket has been already destroyed we fall back to VJ's scheme | |
124 | and use initial timestamp retrieved from peer table. | |
125 | */ | |
126 | if (tcptw->tw_ts_recent_stamp && | |
127 | (twp == NULL || (sysctl_tcp_tw_reuse && | |
9d729f72 | 128 | get_seconds() - tcptw->tw_ts_recent_stamp > 1))) { |
6d6ee43e ACM |
129 | tp->write_seq = tcptw->tw_snd_nxt + 65535 + 2; |
130 | if (tp->write_seq == 0) | |
131 | tp->write_seq = 1; | |
132 | tp->rx_opt.ts_recent = tcptw->tw_ts_recent; | |
133 | tp->rx_opt.ts_recent_stamp = tcptw->tw_ts_recent_stamp; | |
134 | sock_hold(sktw); | |
135 | return 1; | |
136 | } | |
137 | ||
138 | return 0; | |
139 | } | |
6d6ee43e ACM |
140 | EXPORT_SYMBOL_GPL(tcp_twsk_unique); |
141 | ||
1da177e4 LT |
142 | /* This will initiate an outgoing connection. */ |
143 | int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len) | |
144 | { | |
2d7192d6 | 145 | struct sockaddr_in *usin = (struct sockaddr_in *)uaddr; |
1da177e4 LT |
146 | struct inet_sock *inet = inet_sk(sk); |
147 | struct tcp_sock *tp = tcp_sk(sk); | |
dca8b089 | 148 | __be16 orig_sport, orig_dport; |
bada8adc | 149 | __be32 daddr, nexthop; |
da905bd1 | 150 | struct flowi4 *fl4; |
2d7192d6 | 151 | struct rtable *rt; |
1da177e4 | 152 | int err; |
f6d8bd05 | 153 | struct ip_options_rcu *inet_opt; |
1da177e4 LT |
154 | |
155 | if (addr_len < sizeof(struct sockaddr_in)) | |
156 | return -EINVAL; | |
157 | ||
158 | if (usin->sin_family != AF_INET) | |
159 | return -EAFNOSUPPORT; | |
160 | ||
161 | nexthop = daddr = usin->sin_addr.s_addr; | |
f6d8bd05 ED |
162 | inet_opt = rcu_dereference_protected(inet->inet_opt, |
163 | sock_owned_by_user(sk)); | |
164 | if (inet_opt && inet_opt->opt.srr) { | |
1da177e4 LT |
165 | if (!daddr) |
166 | return -EINVAL; | |
f6d8bd05 | 167 | nexthop = inet_opt->opt.faddr; |
1da177e4 LT |
168 | } |
169 | ||
dca8b089 DM |
170 | orig_sport = inet->inet_sport; |
171 | orig_dport = usin->sin_port; | |
da905bd1 DM |
172 | fl4 = &inet->cork.fl.u.ip4; |
173 | rt = ip_route_connect(fl4, nexthop, inet->inet_saddr, | |
b23dd4fe DM |
174 | RT_CONN_FLAGS(sk), sk->sk_bound_dev_if, |
175 | IPPROTO_TCP, | |
0e0d44ab | 176 | orig_sport, orig_dport, sk); |
b23dd4fe DM |
177 | if (IS_ERR(rt)) { |
178 | err = PTR_ERR(rt); | |
179 | if (err == -ENETUNREACH) | |
f1d8cba6 | 180 | IP_INC_STATS(sock_net(sk), IPSTATS_MIB_OUTNOROUTES); |
b23dd4fe | 181 | return err; |
584bdf8c | 182 | } |
1da177e4 LT |
183 | |
184 | if (rt->rt_flags & (RTCF_MULTICAST | RTCF_BROADCAST)) { | |
185 | ip_rt_put(rt); | |
186 | return -ENETUNREACH; | |
187 | } | |
188 | ||
f6d8bd05 | 189 | if (!inet_opt || !inet_opt->opt.srr) |
da905bd1 | 190 | daddr = fl4->daddr; |
1da177e4 | 191 | |
c720c7e8 | 192 | if (!inet->inet_saddr) |
da905bd1 | 193 | inet->inet_saddr = fl4->saddr; |
c720c7e8 | 194 | inet->inet_rcv_saddr = inet->inet_saddr; |
1da177e4 | 195 | |
c720c7e8 | 196 | if (tp->rx_opt.ts_recent_stamp && inet->inet_daddr != daddr) { |
1da177e4 LT |
197 | /* Reset inherited state */ |
198 | tp->rx_opt.ts_recent = 0; | |
199 | tp->rx_opt.ts_recent_stamp = 0; | |
ee995283 PE |
200 | if (likely(!tp->repair)) |
201 | tp->write_seq = 0; | |
1da177e4 LT |
202 | } |
203 | ||
295ff7ed | 204 | if (tcp_death_row.sysctl_tw_recycle && |
81166dd6 DM |
205 | !tp->rx_opt.ts_recent_stamp && fl4->daddr == daddr) |
206 | tcp_fetch_timewait_stamp(sk, &rt->dst); | |
1da177e4 | 207 | |
c720c7e8 ED |
208 | inet->inet_dport = usin->sin_port; |
209 | inet->inet_daddr = daddr; | |
1da177e4 | 210 | |
d83d8461 | 211 | inet_csk(sk)->icsk_ext_hdr_len = 0; |
f6d8bd05 ED |
212 | if (inet_opt) |
213 | inet_csk(sk)->icsk_ext_hdr_len = inet_opt->opt.optlen; | |
1da177e4 | 214 | |
bee7ca9e | 215 | tp->rx_opt.mss_clamp = TCP_MSS_DEFAULT; |
1da177e4 LT |
216 | |
217 | /* Socket identity is still unknown (sport may be zero). | |
218 | * However we set state to SYN-SENT and not releasing socket | |
219 | * lock select source port, enter ourselves into the hash tables and | |
220 | * complete initialization after this. | |
221 | */ | |
222 | tcp_set_state(sk, TCP_SYN_SENT); | |
a7f5e7f1 | 223 | err = inet_hash_connect(&tcp_death_row, sk); |
1da177e4 LT |
224 | if (err) |
225 | goto failure; | |
226 | ||
da905bd1 | 227 | rt = ip_route_newports(fl4, rt, orig_sport, orig_dport, |
b23dd4fe DM |
228 | inet->inet_sport, inet->inet_dport, sk); |
229 | if (IS_ERR(rt)) { | |
230 | err = PTR_ERR(rt); | |
231 | rt = NULL; | |
1da177e4 | 232 | goto failure; |
b23dd4fe | 233 | } |
1da177e4 | 234 | /* OK, now commit destination to socket. */ |
bcd76111 | 235 | sk->sk_gso_type = SKB_GSO_TCPV4; |
d8d1f30b | 236 | sk_setup_caps(sk, &rt->dst); |
1da177e4 | 237 | |
ee995283 | 238 | if (!tp->write_seq && likely(!tp->repair)) |
c720c7e8 ED |
239 | tp->write_seq = secure_tcp_sequence_number(inet->inet_saddr, |
240 | inet->inet_daddr, | |
241 | inet->inet_sport, | |
1da177e4 LT |
242 | usin->sin_port); |
243 | ||
c720c7e8 | 244 | inet->inet_id = tp->write_seq ^ jiffies; |
1da177e4 | 245 | |
2b916477 | 246 | err = tcp_connect(sk); |
ee995283 | 247 | |
1da177e4 LT |
248 | rt = NULL; |
249 | if (err) | |
250 | goto failure; | |
251 | ||
252 | return 0; | |
253 | ||
254 | failure: | |
7174259e ACM |
255 | /* |
256 | * This unhashes the socket and releases the local port, | |
257 | * if necessary. | |
258 | */ | |
1da177e4 LT |
259 | tcp_set_state(sk, TCP_CLOSE); |
260 | ip_rt_put(rt); | |
261 | sk->sk_route_caps = 0; | |
c720c7e8 | 262 | inet->inet_dport = 0; |
1da177e4 LT |
263 | return err; |
264 | } | |
4bc2f18b | 265 | EXPORT_SYMBOL(tcp_v4_connect); |
1da177e4 | 266 | |
1da177e4 | 267 | /* |
563d34d0 ED |
268 | * This routine reacts to ICMP_FRAG_NEEDED mtu indications as defined in RFC1191. |
269 | * It can be called through tcp_release_cb() if socket was owned by user | |
270 | * at the time tcp_v4_err() was called to handle ICMP message. | |
1da177e4 | 271 | */ |
563d34d0 | 272 | static void tcp_v4_mtu_reduced(struct sock *sk) |
1da177e4 LT |
273 | { |
274 | struct dst_entry *dst; | |
275 | struct inet_sock *inet = inet_sk(sk); | |
563d34d0 | 276 | u32 mtu = tcp_sk(sk)->mtu_info; |
1da177e4 | 277 | |
80d0a69f DM |
278 | dst = inet_csk_update_pmtu(sk, mtu); |
279 | if (!dst) | |
1da177e4 LT |
280 | return; |
281 | ||
1da177e4 LT |
282 | /* Something is about to be wrong... Remember soft error |
283 | * for the case, if this connection will not able to recover. | |
284 | */ | |
285 | if (mtu < dst_mtu(dst) && ip_dont_fragment(sk, dst)) | |
286 | sk->sk_err_soft = EMSGSIZE; | |
287 | ||
288 | mtu = dst_mtu(dst); | |
289 | ||
290 | if (inet->pmtudisc != IP_PMTUDISC_DONT && | |
482fc609 | 291 | ip_sk_accept_pmtu(sk) && |
d83d8461 | 292 | inet_csk(sk)->icsk_pmtu_cookie > mtu) { |
1da177e4 LT |
293 | tcp_sync_mss(sk, mtu); |
294 | ||
295 | /* Resend the TCP packet because it's | |
296 | * clear that the old packet has been | |
297 | * dropped. This is the new "fast" path mtu | |
298 | * discovery. | |
299 | */ | |
300 | tcp_simple_retransmit(sk); | |
301 | } /* else let the usual retransmit timer handle it */ | |
302 | } | |
303 | ||
55be7a9c DM |
304 | static void do_redirect(struct sk_buff *skb, struct sock *sk) |
305 | { | |
306 | struct dst_entry *dst = __sk_dst_check(sk, 0); | |
307 | ||
1ed5c48f | 308 | if (dst) |
6700c270 | 309 | dst->ops->redirect(dst, sk, skb); |
55be7a9c DM |
310 | } |
311 | ||
1da177e4 LT |
312 | /* |
313 | * This routine is called by the ICMP module when it gets some | |
314 | * sort of error condition. If err < 0 then the socket should | |
315 | * be closed and the error returned to the user. If err > 0 | |
316 | * it's just the icmp type << 8 | icmp code. After adjustment | |
317 | * header points to the first 8 bytes of the tcp header. We need | |
318 | * to find the appropriate port. | |
319 | * | |
320 | * The locking strategy used here is very "optimistic". When | |
321 | * someone else accesses the socket the ICMP is just dropped | |
322 | * and for some paths there is no check at all. | |
323 | * A more general error queue to queue errors for later handling | |
324 | * is probably better. | |
325 | * | |
326 | */ | |
327 | ||
4d1a2d9e | 328 | void tcp_v4_err(struct sk_buff *icmp_skb, u32 info) |
1da177e4 | 329 | { |
b71d1d42 | 330 | const struct iphdr *iph = (const struct iphdr *)icmp_skb->data; |
4d1a2d9e | 331 | struct tcphdr *th = (struct tcphdr *)(icmp_skb->data + (iph->ihl << 2)); |
f1ecd5d9 | 332 | struct inet_connection_sock *icsk; |
1da177e4 LT |
333 | struct tcp_sock *tp; |
334 | struct inet_sock *inet; | |
4d1a2d9e DL |
335 | const int type = icmp_hdr(icmp_skb)->type; |
336 | const int code = icmp_hdr(icmp_skb)->code; | |
1da177e4 | 337 | struct sock *sk; |
f1ecd5d9 | 338 | struct sk_buff *skb; |
0a672f74 YC |
339 | struct request_sock *fastopen; |
340 | __u32 seq, snd_una; | |
f1ecd5d9 | 341 | __u32 remaining; |
1da177e4 | 342 | int err; |
4d1a2d9e | 343 | struct net *net = dev_net(icmp_skb->dev); |
1da177e4 | 344 | |
4d1a2d9e | 345 | if (icmp_skb->len < (iph->ihl << 2) + 8) { |
dcfc23ca | 346 | ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS); |
1da177e4 LT |
347 | return; |
348 | } | |
349 | ||
fd54d716 | 350 | sk = inet_lookup(net, &tcp_hashinfo, iph->daddr, th->dest, |
4d1a2d9e | 351 | iph->saddr, th->source, inet_iif(icmp_skb)); |
1da177e4 | 352 | if (!sk) { |
dcfc23ca | 353 | ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS); |
1da177e4 LT |
354 | return; |
355 | } | |
356 | if (sk->sk_state == TCP_TIME_WAIT) { | |
9469c7b4 | 357 | inet_twsk_put(inet_twsk(sk)); |
1da177e4 LT |
358 | return; |
359 | } | |
360 | ||
361 | bh_lock_sock(sk); | |
362 | /* If too many ICMPs get dropped on busy | |
363 | * servers this needs to be solved differently. | |
563d34d0 ED |
364 | * We do take care of PMTU discovery (RFC1191) special case : |
365 | * we can receive locally generated ICMP messages while socket is held. | |
1da177e4 | 366 | */ |
b74aa930 ED |
367 | if (sock_owned_by_user(sk)) { |
368 | if (!(type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED)) | |
369 | NET_INC_STATS_BH(net, LINUX_MIB_LOCKDROPPEDICMPS); | |
370 | } | |
1da177e4 LT |
371 | if (sk->sk_state == TCP_CLOSE) |
372 | goto out; | |
373 | ||
97e3ecd1 | 374 | if (unlikely(iph->ttl < inet_sk(sk)->min_ttl)) { |
375 | NET_INC_STATS_BH(net, LINUX_MIB_TCPMINTTLDROP); | |
376 | goto out; | |
377 | } | |
378 | ||
f1ecd5d9 | 379 | icsk = inet_csk(sk); |
1da177e4 LT |
380 | tp = tcp_sk(sk); |
381 | seq = ntohl(th->seq); | |
0a672f74 YC |
382 | /* XXX (TFO) - tp->snd_una should be ISN (tcp_create_openreq_child() */ |
383 | fastopen = tp->fastopen_rsk; | |
384 | snd_una = fastopen ? tcp_rsk(fastopen)->snt_isn : tp->snd_una; | |
1da177e4 | 385 | if (sk->sk_state != TCP_LISTEN && |
0a672f74 | 386 | !between(seq, snd_una, tp->snd_nxt)) { |
de0744af | 387 | NET_INC_STATS_BH(net, LINUX_MIB_OUTOFWINDOWICMPS); |
1da177e4 LT |
388 | goto out; |
389 | } | |
390 | ||
391 | switch (type) { | |
55be7a9c DM |
392 | case ICMP_REDIRECT: |
393 | do_redirect(icmp_skb, sk); | |
394 | goto out; | |
1da177e4 LT |
395 | case ICMP_SOURCE_QUENCH: |
396 | /* Just silently ignore these. */ | |
397 | goto out; | |
398 | case ICMP_PARAMETERPROB: | |
399 | err = EPROTO; | |
400 | break; | |
401 | case ICMP_DEST_UNREACH: | |
402 | if (code > NR_ICMP_UNREACH) | |
403 | goto out; | |
404 | ||
405 | if (code == ICMP_FRAG_NEEDED) { /* PMTU discovery (RFC1191) */ | |
0d4f0608 ED |
406 | /* We are not interested in TCP_LISTEN and open_requests |
407 | * (SYN-ACKs send out by Linux are always <576bytes so | |
408 | * they should go through unfragmented). | |
409 | */ | |
410 | if (sk->sk_state == TCP_LISTEN) | |
411 | goto out; | |
412 | ||
563d34d0 | 413 | tp->mtu_info = info; |
144d56e9 | 414 | if (!sock_owned_by_user(sk)) { |
563d34d0 | 415 | tcp_v4_mtu_reduced(sk); |
144d56e9 ED |
416 | } else { |
417 | if (!test_and_set_bit(TCP_MTU_REDUCED_DEFERRED, &tp->tsq_flags)) | |
418 | sock_hold(sk); | |
419 | } | |
1da177e4 LT |
420 | goto out; |
421 | } | |
422 | ||
423 | err = icmp_err_convert[code].errno; | |
f1ecd5d9 DL |
424 | /* check if icmp_skb allows revert of backoff |
425 | * (see draft-zimmermann-tcp-lcd) */ | |
426 | if (code != ICMP_NET_UNREACH && code != ICMP_HOST_UNREACH) | |
427 | break; | |
428 | if (seq != tp->snd_una || !icsk->icsk_retransmits || | |
0a672f74 | 429 | !icsk->icsk_backoff || fastopen) |
f1ecd5d9 DL |
430 | break; |
431 | ||
8f49c270 DM |
432 | if (sock_owned_by_user(sk)) |
433 | break; | |
434 | ||
f1ecd5d9 | 435 | icsk->icsk_backoff--; |
740b0f18 | 436 | inet_csk(sk)->icsk_rto = (tp->srtt_us ? __tcp_set_rto(tp) : |
9ad7c049 | 437 | TCP_TIMEOUT_INIT) << icsk->icsk_backoff; |
f1ecd5d9 DL |
438 | tcp_bound_rto(sk); |
439 | ||
440 | skb = tcp_write_queue_head(sk); | |
441 | BUG_ON(!skb); | |
442 | ||
443 | remaining = icsk->icsk_rto - min(icsk->icsk_rto, | |
444 | tcp_time_stamp - TCP_SKB_CB(skb)->when); | |
445 | ||
446 | if (remaining) { | |
447 | inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS, | |
448 | remaining, TCP_RTO_MAX); | |
f1ecd5d9 DL |
449 | } else { |
450 | /* RTO revert clocked out retransmission. | |
451 | * Will retransmit now */ | |
452 | tcp_retransmit_timer(sk); | |
453 | } | |
454 | ||
1da177e4 LT |
455 | break; |
456 | case ICMP_TIME_EXCEEDED: | |
457 | err = EHOSTUNREACH; | |
458 | break; | |
459 | default: | |
460 | goto out; | |
461 | } | |
462 | ||
463 | switch (sk->sk_state) { | |
60236fdd | 464 | struct request_sock *req, **prev; |
1da177e4 LT |
465 | case TCP_LISTEN: |
466 | if (sock_owned_by_user(sk)) | |
467 | goto out; | |
468 | ||
463c84b9 ACM |
469 | req = inet_csk_search_req(sk, &prev, th->dest, |
470 | iph->daddr, iph->saddr); | |
1da177e4 LT |
471 | if (!req) |
472 | goto out; | |
473 | ||
474 | /* ICMPs are not backlogged, hence we cannot get | |
475 | an established socket here. | |
476 | */ | |
547b792c | 477 | WARN_ON(req->sk); |
1da177e4 | 478 | |
2e6599cb | 479 | if (seq != tcp_rsk(req)->snt_isn) { |
de0744af | 480 | NET_INC_STATS_BH(net, LINUX_MIB_OUTOFWINDOWICMPS); |
1da177e4 LT |
481 | goto out; |
482 | } | |
483 | ||
484 | /* | |
485 | * Still in SYN_RECV, just remove it silently. | |
486 | * There is no good way to pass the error to the newly | |
487 | * created socket, and POSIX does not want network | |
488 | * errors returned from accept(). | |
489 | */ | |
463c84b9 | 490 | inet_csk_reqsk_queue_drop(sk, req, prev); |
848bf15f | 491 | NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENDROPS); |
1da177e4 LT |
492 | goto out; |
493 | ||
494 | case TCP_SYN_SENT: | |
0a672f74 YC |
495 | case TCP_SYN_RECV: |
496 | /* Only in fast or simultaneous open. If a fast open socket is | |
497 | * is already accepted it is treated as a connected one below. | |
498 | */ | |
499 | if (fastopen && fastopen->sk == NULL) | |
500 | break; | |
501 | ||
1da177e4 | 502 | if (!sock_owned_by_user(sk)) { |
1da177e4 LT |
503 | sk->sk_err = err; |
504 | ||
505 | sk->sk_error_report(sk); | |
506 | ||
507 | tcp_done(sk); | |
508 | } else { | |
509 | sk->sk_err_soft = err; | |
510 | } | |
511 | goto out; | |
512 | } | |
513 | ||
514 | /* If we've already connected we will keep trying | |
515 | * until we time out, or the user gives up. | |
516 | * | |
517 | * rfc1122 4.2.3.9 allows to consider as hard errors | |
518 | * only PROTO_UNREACH and PORT_UNREACH (well, FRAG_FAILED too, | |
519 | * but it is obsoleted by pmtu discovery). | |
520 | * | |
521 | * Note, that in modern internet, where routing is unreliable | |
522 | * and in each dark corner broken firewalls sit, sending random | |
523 | * errors ordered by their masters even this two messages finally lose | |
524 | * their original sense (even Linux sends invalid PORT_UNREACHs) | |
525 | * | |
526 | * Now we are in compliance with RFCs. | |
527 | * --ANK (980905) | |
528 | */ | |
529 | ||
530 | inet = inet_sk(sk); | |
531 | if (!sock_owned_by_user(sk) && inet->recverr) { | |
532 | sk->sk_err = err; | |
533 | sk->sk_error_report(sk); | |
534 | } else { /* Only an error on timeout */ | |
535 | sk->sk_err_soft = err; | |
536 | } | |
537 | ||
538 | out: | |
539 | bh_unlock_sock(sk); | |
540 | sock_put(sk); | |
541 | } | |
542 | ||
28850dc7 | 543 | void __tcp_v4_send_check(struct sk_buff *skb, __be32 saddr, __be32 daddr) |
1da177e4 | 544 | { |
aa8223c7 | 545 | struct tcphdr *th = tcp_hdr(skb); |
1da177e4 | 546 | |
84fa7933 | 547 | if (skb->ip_summed == CHECKSUM_PARTIAL) { |
419f9f89 | 548 | th->check = ~tcp_v4_check(skb->len, saddr, daddr, 0); |
663ead3b | 549 | skb->csum_start = skb_transport_header(skb) - skb->head; |
ff1dcadb | 550 | skb->csum_offset = offsetof(struct tcphdr, check); |
1da177e4 | 551 | } else { |
419f9f89 | 552 | th->check = tcp_v4_check(skb->len, saddr, daddr, |
07f0757a | 553 | csum_partial(th, |
1da177e4 LT |
554 | th->doff << 2, |
555 | skb->csum)); | |
556 | } | |
557 | } | |
558 | ||
419f9f89 | 559 | /* This routine computes an IPv4 TCP checksum. */ |
bb296246 | 560 | void tcp_v4_send_check(struct sock *sk, struct sk_buff *skb) |
419f9f89 | 561 | { |
cf533ea5 | 562 | const struct inet_sock *inet = inet_sk(sk); |
419f9f89 HX |
563 | |
564 | __tcp_v4_send_check(skb, inet->inet_saddr, inet->inet_daddr); | |
565 | } | |
4bc2f18b | 566 | EXPORT_SYMBOL(tcp_v4_send_check); |
419f9f89 | 567 | |
1da177e4 LT |
568 | /* |
569 | * This routine will send an RST to the other tcp. | |
570 | * | |
571 | * Someone asks: why I NEVER use socket parameters (TOS, TTL etc.) | |
572 | * for reset. | |
573 | * Answer: if a packet caused RST, it is not for a socket | |
574 | * existing in our system, if it is matched to a socket, | |
575 | * it is just duplicate segment or bug in other side's TCP. | |
576 | * So that we build reply only basing on parameters | |
577 | * arrived with segment. | |
578 | * Exception: precedence violation. We do not implement it in any case. | |
579 | */ | |
580 | ||
cfb6eeb4 | 581 | static void tcp_v4_send_reset(struct sock *sk, struct sk_buff *skb) |
1da177e4 | 582 | { |
cf533ea5 | 583 | const struct tcphdr *th = tcp_hdr(skb); |
cfb6eeb4 YH |
584 | struct { |
585 | struct tcphdr th; | |
586 | #ifdef CONFIG_TCP_MD5SIG | |
714e85be | 587 | __be32 opt[(TCPOLEN_MD5SIG_ALIGNED >> 2)]; |
cfb6eeb4 YH |
588 | #endif |
589 | } rep; | |
1da177e4 | 590 | struct ip_reply_arg arg; |
cfb6eeb4 YH |
591 | #ifdef CONFIG_TCP_MD5SIG |
592 | struct tcp_md5sig_key *key; | |
658ddaaf SL |
593 | const __u8 *hash_location = NULL; |
594 | unsigned char newhash[16]; | |
595 | int genhash; | |
596 | struct sock *sk1 = NULL; | |
cfb6eeb4 | 597 | #endif |
a86b1e30 | 598 | struct net *net; |
1da177e4 LT |
599 | |
600 | /* Never send a reset in response to a reset. */ | |
601 | if (th->rst) | |
602 | return; | |
603 | ||
511c3f92 | 604 | if (skb_rtable(skb)->rt_type != RTN_LOCAL) |
1da177e4 LT |
605 | return; |
606 | ||
607 | /* Swap the send and the receive. */ | |
cfb6eeb4 YH |
608 | memset(&rep, 0, sizeof(rep)); |
609 | rep.th.dest = th->source; | |
610 | rep.th.source = th->dest; | |
611 | rep.th.doff = sizeof(struct tcphdr) / 4; | |
612 | rep.th.rst = 1; | |
1da177e4 LT |
613 | |
614 | if (th->ack) { | |
cfb6eeb4 | 615 | rep.th.seq = th->ack_seq; |
1da177e4 | 616 | } else { |
cfb6eeb4 YH |
617 | rep.th.ack = 1; |
618 | rep.th.ack_seq = htonl(ntohl(th->seq) + th->syn + th->fin + | |
619 | skb->len - (th->doff << 2)); | |
1da177e4 LT |
620 | } |
621 | ||
7174259e | 622 | memset(&arg, 0, sizeof(arg)); |
cfb6eeb4 YH |
623 | arg.iov[0].iov_base = (unsigned char *)&rep; |
624 | arg.iov[0].iov_len = sizeof(rep.th); | |
625 | ||
626 | #ifdef CONFIG_TCP_MD5SIG | |
658ddaaf SL |
627 | hash_location = tcp_parse_md5sig_option(th); |
628 | if (!sk && hash_location) { | |
629 | /* | |
630 | * active side is lost. Try to find listening socket through | |
631 | * source port, and then find md5 key through listening socket. | |
632 | * we are not loose security here: | |
633 | * Incoming packet is checked with md5 hash with finding key, | |
634 | * no RST generated if md5 hash doesn't match. | |
635 | */ | |
636 | sk1 = __inet_lookup_listener(dev_net(skb_dst(skb)->dev), | |
da5e3630 TH |
637 | &tcp_hashinfo, ip_hdr(skb)->saddr, |
638 | th->source, ip_hdr(skb)->daddr, | |
658ddaaf SL |
639 | ntohs(th->source), inet_iif(skb)); |
640 | /* don't send rst if it can't find key */ | |
641 | if (!sk1) | |
642 | return; | |
643 | rcu_read_lock(); | |
644 | key = tcp_md5_do_lookup(sk1, (union tcp_md5_addr *) | |
645 | &ip_hdr(skb)->saddr, AF_INET); | |
646 | if (!key) | |
647 | goto release_sk1; | |
648 | ||
649 | genhash = tcp_v4_md5_hash_skb(newhash, key, NULL, NULL, skb); | |
650 | if (genhash || memcmp(hash_location, newhash, 16) != 0) | |
651 | goto release_sk1; | |
652 | } else { | |
653 | key = sk ? tcp_md5_do_lookup(sk, (union tcp_md5_addr *) | |
654 | &ip_hdr(skb)->saddr, | |
655 | AF_INET) : NULL; | |
656 | } | |
657 | ||
cfb6eeb4 YH |
658 | if (key) { |
659 | rep.opt[0] = htonl((TCPOPT_NOP << 24) | | |
660 | (TCPOPT_NOP << 16) | | |
661 | (TCPOPT_MD5SIG << 8) | | |
662 | TCPOLEN_MD5SIG); | |
663 | /* Update length and the length the header thinks exists */ | |
664 | arg.iov[0].iov_len += TCPOLEN_MD5SIG_ALIGNED; | |
665 | rep.th.doff = arg.iov[0].iov_len / 4; | |
666 | ||
49a72dfb | 667 | tcp_v4_md5_hash_hdr((__u8 *) &rep.opt[1], |
78e645cb IJ |
668 | key, ip_hdr(skb)->saddr, |
669 | ip_hdr(skb)->daddr, &rep.th); | |
cfb6eeb4 YH |
670 | } |
671 | #endif | |
eddc9ec5 ACM |
672 | arg.csum = csum_tcpudp_nofold(ip_hdr(skb)->daddr, |
673 | ip_hdr(skb)->saddr, /* XXX */ | |
52cd5750 | 674 | arg.iov[0].iov_len, IPPROTO_TCP, 0); |
1da177e4 | 675 | arg.csumoffset = offsetof(struct tcphdr, check) / 2; |
88ef4a5a | 676 | arg.flags = (sk && inet_sk(sk)->transparent) ? IP_REPLY_ARG_NOSRCCHECK : 0; |
e2446eaa | 677 | /* When socket is gone, all binding information is lost. |
4c675258 AK |
678 | * routing might fail in this case. No choice here, if we choose to force |
679 | * input interface, we will misroute in case of asymmetric route. | |
e2446eaa | 680 | */ |
4c675258 AK |
681 | if (sk) |
682 | arg.bound_dev_if = sk->sk_bound_dev_if; | |
1da177e4 | 683 | |
adf30907 | 684 | net = dev_net(skb_dst(skb)->dev); |
66b13d99 | 685 | arg.tos = ip_hdr(skb)->tos; |
be9f4a44 | 686 | ip_send_unicast_reply(net, skb, ip_hdr(skb)->saddr, |
70e73416 | 687 | ip_hdr(skb)->daddr, &arg, arg.iov[0].iov_len); |
1da177e4 | 688 | |
63231bdd PE |
689 | TCP_INC_STATS_BH(net, TCP_MIB_OUTSEGS); |
690 | TCP_INC_STATS_BH(net, TCP_MIB_OUTRSTS); | |
658ddaaf SL |
691 | |
692 | #ifdef CONFIG_TCP_MD5SIG | |
693 | release_sk1: | |
694 | if (sk1) { | |
695 | rcu_read_unlock(); | |
696 | sock_put(sk1); | |
697 | } | |
698 | #endif | |
1da177e4 LT |
699 | } |
700 | ||
701 | /* The code following below sending ACKs in SYN-RECV and TIME-WAIT states | |
702 | outside socket context is ugly, certainly. What can I do? | |
703 | */ | |
704 | ||
9501f972 | 705 | static void tcp_v4_send_ack(struct sk_buff *skb, u32 seq, u32 ack, |
ee684b6f | 706 | u32 win, u32 tsval, u32 tsecr, int oif, |
88ef4a5a | 707 | struct tcp_md5sig_key *key, |
66b13d99 | 708 | int reply_flags, u8 tos) |
1da177e4 | 709 | { |
cf533ea5 | 710 | const struct tcphdr *th = tcp_hdr(skb); |
1da177e4 LT |
711 | struct { |
712 | struct tcphdr th; | |
714e85be | 713 | __be32 opt[(TCPOLEN_TSTAMP_ALIGNED >> 2) |
cfb6eeb4 | 714 | #ifdef CONFIG_TCP_MD5SIG |
714e85be | 715 | + (TCPOLEN_MD5SIG_ALIGNED >> 2) |
cfb6eeb4 YH |
716 | #endif |
717 | ]; | |
1da177e4 LT |
718 | } rep; |
719 | struct ip_reply_arg arg; | |
adf30907 | 720 | struct net *net = dev_net(skb_dst(skb)->dev); |
1da177e4 LT |
721 | |
722 | memset(&rep.th, 0, sizeof(struct tcphdr)); | |
7174259e | 723 | memset(&arg, 0, sizeof(arg)); |
1da177e4 LT |
724 | |
725 | arg.iov[0].iov_base = (unsigned char *)&rep; | |
726 | arg.iov[0].iov_len = sizeof(rep.th); | |
ee684b6f | 727 | if (tsecr) { |
cfb6eeb4 YH |
728 | rep.opt[0] = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) | |
729 | (TCPOPT_TIMESTAMP << 8) | | |
730 | TCPOLEN_TIMESTAMP); | |
ee684b6f AV |
731 | rep.opt[1] = htonl(tsval); |
732 | rep.opt[2] = htonl(tsecr); | |
cb48cfe8 | 733 | arg.iov[0].iov_len += TCPOLEN_TSTAMP_ALIGNED; |
1da177e4 LT |
734 | } |
735 | ||
736 | /* Swap the send and the receive. */ | |
737 | rep.th.dest = th->source; | |
738 | rep.th.source = th->dest; | |
739 | rep.th.doff = arg.iov[0].iov_len / 4; | |
740 | rep.th.seq = htonl(seq); | |
741 | rep.th.ack_seq = htonl(ack); | |
742 | rep.th.ack = 1; | |
743 | rep.th.window = htons(win); | |
744 | ||
cfb6eeb4 | 745 | #ifdef CONFIG_TCP_MD5SIG |
cfb6eeb4 | 746 | if (key) { |
ee684b6f | 747 | int offset = (tsecr) ? 3 : 0; |
cfb6eeb4 YH |
748 | |
749 | rep.opt[offset++] = htonl((TCPOPT_NOP << 24) | | |
750 | (TCPOPT_NOP << 16) | | |
751 | (TCPOPT_MD5SIG << 8) | | |
752 | TCPOLEN_MD5SIG); | |
753 | arg.iov[0].iov_len += TCPOLEN_MD5SIG_ALIGNED; | |
754 | rep.th.doff = arg.iov[0].iov_len/4; | |
755 | ||
49a72dfb | 756 | tcp_v4_md5_hash_hdr((__u8 *) &rep.opt[offset], |
90b7e112 AL |
757 | key, ip_hdr(skb)->saddr, |
758 | ip_hdr(skb)->daddr, &rep.th); | |
cfb6eeb4 YH |
759 | } |
760 | #endif | |
88ef4a5a | 761 | arg.flags = reply_flags; |
eddc9ec5 ACM |
762 | arg.csum = csum_tcpudp_nofold(ip_hdr(skb)->daddr, |
763 | ip_hdr(skb)->saddr, /* XXX */ | |
1da177e4 LT |
764 | arg.iov[0].iov_len, IPPROTO_TCP, 0); |
765 | arg.csumoffset = offsetof(struct tcphdr, check) / 2; | |
9501f972 YH |
766 | if (oif) |
767 | arg.bound_dev_if = oif; | |
66b13d99 | 768 | arg.tos = tos; |
be9f4a44 | 769 | ip_send_unicast_reply(net, skb, ip_hdr(skb)->saddr, |
70e73416 | 770 | ip_hdr(skb)->daddr, &arg, arg.iov[0].iov_len); |
1da177e4 | 771 | |
63231bdd | 772 | TCP_INC_STATS_BH(net, TCP_MIB_OUTSEGS); |
1da177e4 LT |
773 | } |
774 | ||
775 | static void tcp_v4_timewait_ack(struct sock *sk, struct sk_buff *skb) | |
776 | { | |
8feaf0c0 | 777 | struct inet_timewait_sock *tw = inet_twsk(sk); |
cfb6eeb4 | 778 | struct tcp_timewait_sock *tcptw = tcp_twsk(sk); |
1da177e4 | 779 | |
9501f972 | 780 | tcp_v4_send_ack(skb, tcptw->tw_snd_nxt, tcptw->tw_rcv_nxt, |
7174259e | 781 | tcptw->tw_rcv_wnd >> tw->tw_rcv_wscale, |
ee684b6f | 782 | tcp_time_stamp + tcptw->tw_ts_offset, |
9501f972 YH |
783 | tcptw->tw_ts_recent, |
784 | tw->tw_bound_dev_if, | |
88ef4a5a | 785 | tcp_twsk_md5_key(tcptw), |
66b13d99 ED |
786 | tw->tw_transparent ? IP_REPLY_ARG_NOSRCCHECK : 0, |
787 | tw->tw_tos | |
9501f972 | 788 | ); |
1da177e4 | 789 | |
8feaf0c0 | 790 | inet_twsk_put(tw); |
1da177e4 LT |
791 | } |
792 | ||
6edafaaf | 793 | static void tcp_v4_reqsk_send_ack(struct sock *sk, struct sk_buff *skb, |
7174259e | 794 | struct request_sock *req) |
1da177e4 | 795 | { |
168a8f58 JC |
796 | /* sk->sk_state == TCP_LISTEN -> for regular TCP_SYN_RECV |
797 | * sk->sk_state == TCP_SYN_RECV -> for Fast Open. | |
798 | */ | |
799 | tcp_v4_send_ack(skb, (sk->sk_state == TCP_LISTEN) ? | |
800 | tcp_rsk(req)->snt_isn + 1 : tcp_sk(sk)->snd_nxt, | |
801 | tcp_rsk(req)->rcv_nxt, req->rcv_wnd, | |
ee684b6f | 802 | tcp_time_stamp, |
9501f972 YH |
803 | req->ts_recent, |
804 | 0, | |
a915da9b ED |
805 | tcp_md5_do_lookup(sk, (union tcp_md5_addr *)&ip_hdr(skb)->daddr, |
806 | AF_INET), | |
66b13d99 ED |
807 | inet_rsk(req)->no_srccheck ? IP_REPLY_ARG_NOSRCCHECK : 0, |
808 | ip_hdr(skb)->tos); | |
1da177e4 LT |
809 | } |
810 | ||
1da177e4 | 811 | /* |
9bf1d83e | 812 | * Send a SYN-ACK after having received a SYN. |
60236fdd | 813 | * This still operates on a request_sock only, not on a big |
1da177e4 LT |
814 | * socket. |
815 | */ | |
72659ecc | 816 | static int tcp_v4_send_synack(struct sock *sk, struct dst_entry *dst, |
d6274bd8 | 817 | struct flowi *fl, |
72659ecc | 818 | struct request_sock *req, |
843f4a55 YC |
819 | u16 queue_mapping, |
820 | struct tcp_fastopen_cookie *foc) | |
1da177e4 | 821 | { |
2e6599cb | 822 | const struct inet_request_sock *ireq = inet_rsk(req); |
6bd023f3 | 823 | struct flowi4 fl4; |
1da177e4 | 824 | int err = -1; |
d41db5af | 825 | struct sk_buff *skb; |
1da177e4 LT |
826 | |
827 | /* First, grab a route. */ | |
ba3f7f04 | 828 | if (!dst && (dst = inet_csk_route_req(sk, &fl4, req)) == NULL) |
fd80eb94 | 829 | return -1; |
1da177e4 | 830 | |
843f4a55 | 831 | skb = tcp_make_synack(sk, dst, req, foc); |
1da177e4 LT |
832 | |
833 | if (skb) { | |
634fb979 | 834 | __tcp_v4_send_check(skb, ireq->ir_loc_addr, ireq->ir_rmt_addr); |
1da177e4 | 835 | |
fff32699 | 836 | skb_set_queue_mapping(skb, queue_mapping); |
634fb979 ED |
837 | err = ip_build_and_send_pkt(skb, sk, ireq->ir_loc_addr, |
838 | ireq->ir_rmt_addr, | |
2e6599cb | 839 | ireq->opt); |
b9df3cb8 | 840 | err = net_xmit_eval(err); |
016818d0 NC |
841 | if (!tcp_rsk(req)->snt_synack && !err) |
842 | tcp_rsk(req)->snt_synack = tcp_time_stamp; | |
1da177e4 LT |
843 | } |
844 | ||
1da177e4 LT |
845 | return err; |
846 | } | |
847 | ||
1a2c6181 | 848 | static int tcp_v4_rtx_synack(struct sock *sk, struct request_sock *req) |
fd80eb94 | 849 | { |
d6274bd8 OP |
850 | const struct tcp_request_sock_ops *af_ops = tcp_rsk(req)->af_specific; |
851 | int res = af_ops->send_synack(sk, NULL, NULL, req, 0, NULL); | |
e6c022a4 | 852 | |
f19c29e3 | 853 | if (!res) { |
e6c022a4 | 854 | TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_RETRANSSEGS); |
f19c29e3 YC |
855 | NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPSYNRETRANS); |
856 | } | |
e6c022a4 | 857 | return res; |
fd80eb94 DL |
858 | } |
859 | ||
1da177e4 | 860 | /* |
60236fdd | 861 | * IPv4 request_sock destructor. |
1da177e4 | 862 | */ |
60236fdd | 863 | static void tcp_v4_reqsk_destructor(struct request_sock *req) |
1da177e4 | 864 | { |
a51482bd | 865 | kfree(inet_rsk(req)->opt); |
1da177e4 LT |
866 | } |
867 | ||
946cedcc | 868 | /* |
a2a385d6 | 869 | * Return true if a syncookie should be sent |
946cedcc | 870 | */ |
a2a385d6 | 871 | bool tcp_syn_flood_action(struct sock *sk, |
946cedcc ED |
872 | const struct sk_buff *skb, |
873 | const char *proto) | |
1da177e4 | 874 | { |
946cedcc | 875 | const char *msg = "Dropping request"; |
a2a385d6 | 876 | bool want_cookie = false; |
946cedcc ED |
877 | struct listen_sock *lopt; |
878 | ||
2a1d4bd4 | 879 | #ifdef CONFIG_SYN_COOKIES |
946cedcc | 880 | if (sysctl_tcp_syncookies) { |
2a1d4bd4 | 881 | msg = "Sending cookies"; |
a2a385d6 | 882 | want_cookie = true; |
946cedcc ED |
883 | NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPREQQFULLDOCOOKIES); |
884 | } else | |
80e40daa | 885 | #endif |
946cedcc ED |
886 | NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPREQQFULLDROP); |
887 | ||
888 | lopt = inet_csk(sk)->icsk_accept_queue.listen_opt; | |
5ad37d5d | 889 | if (!lopt->synflood_warned && sysctl_tcp_syncookies != 2) { |
946cedcc | 890 | lopt->synflood_warned = 1; |
afd46503 | 891 | pr_info("%s: Possible SYN flooding on port %d. %s. Check SNMP counters.\n", |
946cedcc ED |
892 | proto, ntohs(tcp_hdr(skb)->dest), msg); |
893 | } | |
894 | return want_cookie; | |
2a1d4bd4 | 895 | } |
946cedcc | 896 | EXPORT_SYMBOL(tcp_syn_flood_action); |
1da177e4 LT |
897 | |
898 | /* | |
60236fdd | 899 | * Save and compile IPv4 options into the request_sock if needed. |
1da177e4 | 900 | */ |
5dff747b | 901 | static struct ip_options_rcu *tcp_v4_save_options(struct sk_buff *skb) |
1da177e4 | 902 | { |
f6d8bd05 ED |
903 | const struct ip_options *opt = &(IPCB(skb)->opt); |
904 | struct ip_options_rcu *dopt = NULL; | |
1da177e4 LT |
905 | |
906 | if (opt && opt->optlen) { | |
f6d8bd05 ED |
907 | int opt_size = sizeof(*dopt) + opt->optlen; |
908 | ||
1da177e4 LT |
909 | dopt = kmalloc(opt_size, GFP_ATOMIC); |
910 | if (dopt) { | |
f6d8bd05 | 911 | if (ip_options_echo(&dopt->opt, skb)) { |
1da177e4 LT |
912 | kfree(dopt); |
913 | dopt = NULL; | |
914 | } | |
915 | } | |
916 | } | |
917 | return dopt; | |
918 | } | |
919 | ||
cfb6eeb4 YH |
920 | #ifdef CONFIG_TCP_MD5SIG |
921 | /* | |
922 | * RFC2385 MD5 checksumming requires a mapping of | |
923 | * IP address->MD5 Key. | |
924 | * We need to maintain these in the sk structure. | |
925 | */ | |
926 | ||
927 | /* Find the Key structure for an address. */ | |
a915da9b ED |
928 | struct tcp_md5sig_key *tcp_md5_do_lookup(struct sock *sk, |
929 | const union tcp_md5_addr *addr, | |
930 | int family) | |
cfb6eeb4 YH |
931 | { |
932 | struct tcp_sock *tp = tcp_sk(sk); | |
a915da9b | 933 | struct tcp_md5sig_key *key; |
a915da9b | 934 | unsigned int size = sizeof(struct in_addr); |
a8afca03 | 935 | struct tcp_md5sig_info *md5sig; |
cfb6eeb4 | 936 | |
a8afca03 ED |
937 | /* caller either holds rcu_read_lock() or socket lock */ |
938 | md5sig = rcu_dereference_check(tp->md5sig_info, | |
b4fb05ea ED |
939 | sock_owned_by_user(sk) || |
940 | lockdep_is_held(&sk->sk_lock.slock)); | |
a8afca03 | 941 | if (!md5sig) |
cfb6eeb4 | 942 | return NULL; |
a915da9b ED |
943 | #if IS_ENABLED(CONFIG_IPV6) |
944 | if (family == AF_INET6) | |
945 | size = sizeof(struct in6_addr); | |
946 | #endif | |
b67bfe0d | 947 | hlist_for_each_entry_rcu(key, &md5sig->head, node) { |
a915da9b ED |
948 | if (key->family != family) |
949 | continue; | |
950 | if (!memcmp(&key->addr, addr, size)) | |
951 | return key; | |
cfb6eeb4 YH |
952 | } |
953 | return NULL; | |
954 | } | |
a915da9b | 955 | EXPORT_SYMBOL(tcp_md5_do_lookup); |
cfb6eeb4 YH |
956 | |
957 | struct tcp_md5sig_key *tcp_v4_md5_lookup(struct sock *sk, | |
958 | struct sock *addr_sk) | |
959 | { | |
a915da9b ED |
960 | union tcp_md5_addr *addr; |
961 | ||
962 | addr = (union tcp_md5_addr *)&inet_sk(addr_sk)->inet_daddr; | |
963 | return tcp_md5_do_lookup(sk, addr, AF_INET); | |
cfb6eeb4 | 964 | } |
cfb6eeb4 YH |
965 | EXPORT_SYMBOL(tcp_v4_md5_lookup); |
966 | ||
f5b99bcd AB |
967 | static struct tcp_md5sig_key *tcp_v4_reqsk_md5_lookup(struct sock *sk, |
968 | struct request_sock *req) | |
cfb6eeb4 | 969 | { |
a915da9b ED |
970 | union tcp_md5_addr *addr; |
971 | ||
634fb979 | 972 | addr = (union tcp_md5_addr *)&inet_rsk(req)->ir_rmt_addr; |
a915da9b | 973 | return tcp_md5_do_lookup(sk, addr, AF_INET); |
cfb6eeb4 YH |
974 | } |
975 | ||
976 | /* This can be called on a newly created socket, from other files */ | |
a915da9b ED |
977 | int tcp_md5_do_add(struct sock *sk, const union tcp_md5_addr *addr, |
978 | int family, const u8 *newkey, u8 newkeylen, gfp_t gfp) | |
cfb6eeb4 YH |
979 | { |
980 | /* Add Key to the list */ | |
b0a713e9 | 981 | struct tcp_md5sig_key *key; |
cfb6eeb4 | 982 | struct tcp_sock *tp = tcp_sk(sk); |
a915da9b | 983 | struct tcp_md5sig_info *md5sig; |
cfb6eeb4 | 984 | |
c0353c7b | 985 | key = tcp_md5_do_lookup(sk, addr, family); |
cfb6eeb4 YH |
986 | if (key) { |
987 | /* Pre-existing entry - just update that one. */ | |
a915da9b | 988 | memcpy(key->key, newkey, newkeylen); |
b0a713e9 | 989 | key->keylen = newkeylen; |
a915da9b ED |
990 | return 0; |
991 | } | |
260fcbeb | 992 | |
a8afca03 ED |
993 | md5sig = rcu_dereference_protected(tp->md5sig_info, |
994 | sock_owned_by_user(sk)); | |
a915da9b ED |
995 | if (!md5sig) { |
996 | md5sig = kmalloc(sizeof(*md5sig), gfp); | |
997 | if (!md5sig) | |
cfb6eeb4 | 998 | return -ENOMEM; |
cfb6eeb4 | 999 | |
a915da9b ED |
1000 | sk_nocaps_add(sk, NETIF_F_GSO_MASK); |
1001 | INIT_HLIST_HEAD(&md5sig->head); | |
a8afca03 | 1002 | rcu_assign_pointer(tp->md5sig_info, md5sig); |
a915da9b | 1003 | } |
cfb6eeb4 | 1004 | |
5f3d9cb2 | 1005 | key = sock_kmalloc(sk, sizeof(*key), gfp); |
a915da9b ED |
1006 | if (!key) |
1007 | return -ENOMEM; | |
71cea17e | 1008 | if (!tcp_alloc_md5sig_pool()) { |
5f3d9cb2 | 1009 | sock_kfree_s(sk, key, sizeof(*key)); |
a915da9b | 1010 | return -ENOMEM; |
cfb6eeb4 | 1011 | } |
a915da9b ED |
1012 | |
1013 | memcpy(key->key, newkey, newkeylen); | |
1014 | key->keylen = newkeylen; | |
1015 | key->family = family; | |
1016 | memcpy(&key->addr, addr, | |
1017 | (family == AF_INET6) ? sizeof(struct in6_addr) : | |
1018 | sizeof(struct in_addr)); | |
1019 | hlist_add_head_rcu(&key->node, &md5sig->head); | |
cfb6eeb4 YH |
1020 | return 0; |
1021 | } | |
a915da9b | 1022 | EXPORT_SYMBOL(tcp_md5_do_add); |
cfb6eeb4 | 1023 | |
a915da9b | 1024 | int tcp_md5_do_del(struct sock *sk, const union tcp_md5_addr *addr, int family) |
cfb6eeb4 | 1025 | { |
a915da9b ED |
1026 | struct tcp_md5sig_key *key; |
1027 | ||
c0353c7b | 1028 | key = tcp_md5_do_lookup(sk, addr, family); |
a915da9b ED |
1029 | if (!key) |
1030 | return -ENOENT; | |
1031 | hlist_del_rcu(&key->node); | |
5f3d9cb2 | 1032 | atomic_sub(sizeof(*key), &sk->sk_omem_alloc); |
a915da9b | 1033 | kfree_rcu(key, rcu); |
a915da9b | 1034 | return 0; |
cfb6eeb4 | 1035 | } |
a915da9b | 1036 | EXPORT_SYMBOL(tcp_md5_do_del); |
cfb6eeb4 | 1037 | |
e0683e70 | 1038 | static void tcp_clear_md5_list(struct sock *sk) |
cfb6eeb4 YH |
1039 | { |
1040 | struct tcp_sock *tp = tcp_sk(sk); | |
a915da9b | 1041 | struct tcp_md5sig_key *key; |
b67bfe0d | 1042 | struct hlist_node *n; |
a8afca03 | 1043 | struct tcp_md5sig_info *md5sig; |
cfb6eeb4 | 1044 | |
a8afca03 ED |
1045 | md5sig = rcu_dereference_protected(tp->md5sig_info, 1); |
1046 | ||
b67bfe0d | 1047 | hlist_for_each_entry_safe(key, n, &md5sig->head, node) { |
a915da9b | 1048 | hlist_del_rcu(&key->node); |
5f3d9cb2 | 1049 | atomic_sub(sizeof(*key), &sk->sk_omem_alloc); |
a915da9b | 1050 | kfree_rcu(key, rcu); |
cfb6eeb4 YH |
1051 | } |
1052 | } | |
1053 | ||
7174259e ACM |
1054 | static int tcp_v4_parse_md5_keys(struct sock *sk, char __user *optval, |
1055 | int optlen) | |
cfb6eeb4 YH |
1056 | { |
1057 | struct tcp_md5sig cmd; | |
1058 | struct sockaddr_in *sin = (struct sockaddr_in *)&cmd.tcpm_addr; | |
cfb6eeb4 YH |
1059 | |
1060 | if (optlen < sizeof(cmd)) | |
1061 | return -EINVAL; | |
1062 | ||
7174259e | 1063 | if (copy_from_user(&cmd, optval, sizeof(cmd))) |
cfb6eeb4 YH |
1064 | return -EFAULT; |
1065 | ||
1066 | if (sin->sin_family != AF_INET) | |
1067 | return -EINVAL; | |
1068 | ||
a8afca03 | 1069 | if (!cmd.tcpm_key || !cmd.tcpm_keylen) |
a915da9b ED |
1070 | return tcp_md5_do_del(sk, (union tcp_md5_addr *)&sin->sin_addr.s_addr, |
1071 | AF_INET); | |
cfb6eeb4 YH |
1072 | |
1073 | if (cmd.tcpm_keylen > TCP_MD5SIG_MAXKEYLEN) | |
1074 | return -EINVAL; | |
1075 | ||
a915da9b ED |
1076 | return tcp_md5_do_add(sk, (union tcp_md5_addr *)&sin->sin_addr.s_addr, |
1077 | AF_INET, cmd.tcpm_key, cmd.tcpm_keylen, | |
1078 | GFP_KERNEL); | |
cfb6eeb4 YH |
1079 | } |
1080 | ||
49a72dfb AL |
1081 | static int tcp_v4_md5_hash_pseudoheader(struct tcp_md5sig_pool *hp, |
1082 | __be32 daddr, __be32 saddr, int nbytes) | |
cfb6eeb4 | 1083 | { |
cfb6eeb4 | 1084 | struct tcp4_pseudohdr *bp; |
49a72dfb | 1085 | struct scatterlist sg; |
cfb6eeb4 YH |
1086 | |
1087 | bp = &hp->md5_blk.ip4; | |
cfb6eeb4 YH |
1088 | |
1089 | /* | |
49a72dfb | 1090 | * 1. the TCP pseudo-header (in the order: source IP address, |
cfb6eeb4 YH |
1091 | * destination IP address, zero-padded protocol number, and |
1092 | * segment length) | |
1093 | */ | |
1094 | bp->saddr = saddr; | |
1095 | bp->daddr = daddr; | |
1096 | bp->pad = 0; | |
076fb722 | 1097 | bp->protocol = IPPROTO_TCP; |
49a72dfb | 1098 | bp->len = cpu_to_be16(nbytes); |
c7da57a1 | 1099 | |
49a72dfb AL |
1100 | sg_init_one(&sg, bp, sizeof(*bp)); |
1101 | return crypto_hash_update(&hp->md5_desc, &sg, sizeof(*bp)); | |
1102 | } | |
1103 | ||
a915da9b | 1104 | static int tcp_v4_md5_hash_hdr(char *md5_hash, const struct tcp_md5sig_key *key, |
318cf7aa | 1105 | __be32 daddr, __be32 saddr, const struct tcphdr *th) |
49a72dfb AL |
1106 | { |
1107 | struct tcp_md5sig_pool *hp; | |
1108 | struct hash_desc *desc; | |
1109 | ||
1110 | hp = tcp_get_md5sig_pool(); | |
1111 | if (!hp) | |
1112 | goto clear_hash_noput; | |
1113 | desc = &hp->md5_desc; | |
1114 | ||
1115 | if (crypto_hash_init(desc)) | |
1116 | goto clear_hash; | |
1117 | if (tcp_v4_md5_hash_pseudoheader(hp, daddr, saddr, th->doff << 2)) | |
1118 | goto clear_hash; | |
1119 | if (tcp_md5_hash_header(hp, th)) | |
1120 | goto clear_hash; | |
1121 | if (tcp_md5_hash_key(hp, key)) | |
1122 | goto clear_hash; | |
1123 | if (crypto_hash_final(desc, md5_hash)) | |
cfb6eeb4 YH |
1124 | goto clear_hash; |
1125 | ||
cfb6eeb4 | 1126 | tcp_put_md5sig_pool(); |
cfb6eeb4 | 1127 | return 0; |
49a72dfb | 1128 | |
cfb6eeb4 YH |
1129 | clear_hash: |
1130 | tcp_put_md5sig_pool(); | |
1131 | clear_hash_noput: | |
1132 | memset(md5_hash, 0, 16); | |
49a72dfb | 1133 | return 1; |
cfb6eeb4 YH |
1134 | } |
1135 | ||
49a72dfb | 1136 | int tcp_v4_md5_hash_skb(char *md5_hash, struct tcp_md5sig_key *key, |
318cf7aa ED |
1137 | const struct sock *sk, const struct request_sock *req, |
1138 | const struct sk_buff *skb) | |
cfb6eeb4 | 1139 | { |
49a72dfb AL |
1140 | struct tcp_md5sig_pool *hp; |
1141 | struct hash_desc *desc; | |
318cf7aa | 1142 | const struct tcphdr *th = tcp_hdr(skb); |
cfb6eeb4 YH |
1143 | __be32 saddr, daddr; |
1144 | ||
1145 | if (sk) { | |
c720c7e8 ED |
1146 | saddr = inet_sk(sk)->inet_saddr; |
1147 | daddr = inet_sk(sk)->inet_daddr; | |
49a72dfb | 1148 | } else if (req) { |
634fb979 ED |
1149 | saddr = inet_rsk(req)->ir_loc_addr; |
1150 | daddr = inet_rsk(req)->ir_rmt_addr; | |
cfb6eeb4 | 1151 | } else { |
49a72dfb AL |
1152 | const struct iphdr *iph = ip_hdr(skb); |
1153 | saddr = iph->saddr; | |
1154 | daddr = iph->daddr; | |
cfb6eeb4 | 1155 | } |
49a72dfb AL |
1156 | |
1157 | hp = tcp_get_md5sig_pool(); | |
1158 | if (!hp) | |
1159 | goto clear_hash_noput; | |
1160 | desc = &hp->md5_desc; | |
1161 | ||
1162 | if (crypto_hash_init(desc)) | |
1163 | goto clear_hash; | |
1164 | ||
1165 | if (tcp_v4_md5_hash_pseudoheader(hp, daddr, saddr, skb->len)) | |
1166 | goto clear_hash; | |
1167 | if (tcp_md5_hash_header(hp, th)) | |
1168 | goto clear_hash; | |
1169 | if (tcp_md5_hash_skb_data(hp, skb, th->doff << 2)) | |
1170 | goto clear_hash; | |
1171 | if (tcp_md5_hash_key(hp, key)) | |
1172 | goto clear_hash; | |
1173 | if (crypto_hash_final(desc, md5_hash)) | |
1174 | goto clear_hash; | |
1175 | ||
1176 | tcp_put_md5sig_pool(); | |
1177 | return 0; | |
1178 | ||
1179 | clear_hash: | |
1180 | tcp_put_md5sig_pool(); | |
1181 | clear_hash_noput: | |
1182 | memset(md5_hash, 0, 16); | |
1183 | return 1; | |
cfb6eeb4 | 1184 | } |
49a72dfb | 1185 | EXPORT_SYMBOL(tcp_v4_md5_hash_skb); |
cfb6eeb4 | 1186 | |
a2a385d6 | 1187 | static bool tcp_v4_inbound_md5_hash(struct sock *sk, const struct sk_buff *skb) |
cfb6eeb4 YH |
1188 | { |
1189 | /* | |
1190 | * This gets called for each TCP segment that arrives | |
1191 | * so we want to be efficient. | |
1192 | * We have 3 drop cases: | |
1193 | * o No MD5 hash and one expected. | |
1194 | * o MD5 hash and we're not expecting one. | |
1195 | * o MD5 hash and its wrong. | |
1196 | */ | |
cf533ea5 | 1197 | const __u8 *hash_location = NULL; |
cfb6eeb4 | 1198 | struct tcp_md5sig_key *hash_expected; |
eddc9ec5 | 1199 | const struct iphdr *iph = ip_hdr(skb); |
cf533ea5 | 1200 | const struct tcphdr *th = tcp_hdr(skb); |
cfb6eeb4 | 1201 | int genhash; |
cfb6eeb4 YH |
1202 | unsigned char newhash[16]; |
1203 | ||
a915da9b ED |
1204 | hash_expected = tcp_md5_do_lookup(sk, (union tcp_md5_addr *)&iph->saddr, |
1205 | AF_INET); | |
7d5d5525 | 1206 | hash_location = tcp_parse_md5sig_option(th); |
cfb6eeb4 | 1207 | |
cfb6eeb4 YH |
1208 | /* We've parsed the options - do we have a hash? */ |
1209 | if (!hash_expected && !hash_location) | |
a2a385d6 | 1210 | return false; |
cfb6eeb4 YH |
1211 | |
1212 | if (hash_expected && !hash_location) { | |
785957d3 | 1213 | NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPMD5NOTFOUND); |
a2a385d6 | 1214 | return true; |
cfb6eeb4 YH |
1215 | } |
1216 | ||
1217 | if (!hash_expected && hash_location) { | |
785957d3 | 1218 | NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPMD5UNEXPECTED); |
a2a385d6 | 1219 | return true; |
cfb6eeb4 YH |
1220 | } |
1221 | ||
1222 | /* Okay, so this is hash_expected and hash_location - | |
1223 | * so we need to calculate the checksum. | |
1224 | */ | |
49a72dfb AL |
1225 | genhash = tcp_v4_md5_hash_skb(newhash, |
1226 | hash_expected, | |
1227 | NULL, NULL, skb); | |
cfb6eeb4 YH |
1228 | |
1229 | if (genhash || memcmp(hash_location, newhash, 16) != 0) { | |
e87cc472 JP |
1230 | net_info_ratelimited("MD5 Hash failed for (%pI4, %d)->(%pI4, %d)%s\n", |
1231 | &iph->saddr, ntohs(th->source), | |
1232 | &iph->daddr, ntohs(th->dest), | |
1233 | genhash ? " tcp_v4_calc_md5_hash failed" | |
1234 | : ""); | |
a2a385d6 | 1235 | return true; |
cfb6eeb4 | 1236 | } |
a2a385d6 | 1237 | return false; |
cfb6eeb4 YH |
1238 | } |
1239 | ||
1240 | #endif | |
1241 | ||
16bea70a OP |
1242 | static void tcp_v4_init_req(struct request_sock *req, struct sock *sk, |
1243 | struct sk_buff *skb) | |
1244 | { | |
1245 | struct inet_request_sock *ireq = inet_rsk(req); | |
1246 | ||
1247 | ireq->ir_loc_addr = ip_hdr(skb)->daddr; | |
1248 | ireq->ir_rmt_addr = ip_hdr(skb)->saddr; | |
1249 | ireq->no_srccheck = inet_sk(sk)->transparent; | |
1250 | ireq->opt = tcp_v4_save_options(skb); | |
1251 | } | |
1252 | ||
d94e0417 OP |
1253 | static struct dst_entry *tcp_v4_route_req(struct sock *sk, struct flowi *fl, |
1254 | const struct request_sock *req, | |
1255 | bool *strict) | |
1256 | { | |
1257 | struct dst_entry *dst = inet_csk_route_req(sk, &fl->u.ip4, req); | |
1258 | ||
1259 | if (strict) { | |
1260 | if (fl->u.ip4.daddr == inet_rsk(req)->ir_rmt_addr) | |
1261 | *strict = true; | |
1262 | else | |
1263 | *strict = false; | |
1264 | } | |
1265 | ||
1266 | return dst; | |
1267 | } | |
1268 | ||
72a3effa | 1269 | struct request_sock_ops tcp_request_sock_ops __read_mostly = { |
1da177e4 | 1270 | .family = PF_INET, |
2e6599cb | 1271 | .obj_size = sizeof(struct tcp_request_sock), |
72659ecc | 1272 | .rtx_syn_ack = tcp_v4_rtx_synack, |
60236fdd ACM |
1273 | .send_ack = tcp_v4_reqsk_send_ack, |
1274 | .destructor = tcp_v4_reqsk_destructor, | |
1da177e4 | 1275 | .send_reset = tcp_v4_send_reset, |
72659ecc | 1276 | .syn_ack_timeout = tcp_syn_ack_timeout, |
1da177e4 LT |
1277 | }; |
1278 | ||
b2e4b3de | 1279 | static const struct tcp_request_sock_ops tcp_request_sock_ipv4_ops = { |
16bea70a | 1280 | #ifdef CONFIG_TCP_MD5SIG |
cfb6eeb4 | 1281 | .md5_lookup = tcp_v4_reqsk_md5_lookup, |
e3afe7b7 | 1282 | .calc_md5_hash = tcp_v4_md5_hash_skb, |
b6332e6c | 1283 | #endif |
16bea70a | 1284 | .init_req = tcp_v4_init_req, |
fb7b37a7 OP |
1285 | #ifdef CONFIG_SYN_COOKIES |
1286 | .cookie_init_seq = cookie_v4_init_sequence, | |
1287 | #endif | |
d94e0417 | 1288 | .route_req = tcp_v4_route_req, |
936b8bdb | 1289 | .init_seq = tcp_v4_init_sequence, |
d6274bd8 | 1290 | .send_synack = tcp_v4_send_synack, |
16bea70a | 1291 | }; |
cfb6eeb4 | 1292 | |
1da177e4 LT |
1293 | int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb) |
1294 | { | |
1295 | struct tcp_options_received tmp_opt; | |
60236fdd | 1296 | struct request_sock *req; |
4957faad | 1297 | struct tcp_sock *tp = tcp_sk(sk); |
e6b4d113 | 1298 | struct dst_entry *dst = NULL; |
eddc9ec5 | 1299 | __be32 saddr = ip_hdr(skb)->saddr; |
1da177e4 | 1300 | __u32 isn = TCP_SKB_CB(skb)->when; |
843f4a55 | 1301 | bool want_cookie = false, fastopen; |
168a8f58 JC |
1302 | struct flowi4 fl4; |
1303 | struct tcp_fastopen_cookie foc = { .len = -1 }; | |
16bea70a | 1304 | const struct tcp_request_sock_ops *af_ops; |
843f4a55 | 1305 | int err; |
1da177e4 LT |
1306 | |
1307 | /* Never answer to SYNs send to broadcast or multicast */ | |
511c3f92 | 1308 | if (skb_rtable(skb)->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST)) |
1da177e4 LT |
1309 | goto drop; |
1310 | ||
1311 | /* TW buckets are converted to open requests without | |
1312 | * limitations, they conserve resources and peer is | |
1313 | * evidently real one. | |
1314 | */ | |
5ad37d5d HFS |
1315 | if ((sysctl_tcp_syncookies == 2 || |
1316 | inet_csk_reqsk_queue_is_full(sk)) && !isn) { | |
946cedcc ED |
1317 | want_cookie = tcp_syn_flood_action(sk, skb, "TCP"); |
1318 | if (!want_cookie) | |
1319 | goto drop; | |
1da177e4 LT |
1320 | } |
1321 | ||
1322 | /* Accept backlog is full. If we have already queued enough | |
1323 | * of warm entries in syn queue, drop request. It is better than | |
1324 | * clogging syn queue with openreqs with exponentially increasing | |
1325 | * timeout. | |
1326 | */ | |
2aeef18d NS |
1327 | if (sk_acceptq_is_full(sk) && inet_csk_reqsk_queue_young(sk) > 1) { |
1328 | NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENOVERFLOWS); | |
1da177e4 | 1329 | goto drop; |
2aeef18d | 1330 | } |
1da177e4 | 1331 | |
ce4a7d0d | 1332 | req = inet_reqsk_alloc(&tcp_request_sock_ops); |
1da177e4 LT |
1333 | if (!req) |
1334 | goto drop; | |
1335 | ||
16bea70a | 1336 | af_ops = tcp_rsk(req)->af_specific = &tcp_request_sock_ipv4_ops; |
cfb6eeb4 | 1337 | |
1da177e4 | 1338 | tcp_clear_options(&tmp_opt); |
bee7ca9e | 1339 | tmp_opt.mss_clamp = TCP_MSS_DEFAULT; |
4957faad | 1340 | tmp_opt.user_mss = tp->rx_opt.user_mss; |
1a2c6181 | 1341 | tcp_parse_options(skb, &tmp_opt, 0, want_cookie ? NULL : &foc); |
1da177e4 | 1342 | |
4dfc2817 | 1343 | if (want_cookie && !tmp_opt.saw_tstamp) |
1da177e4 | 1344 | tcp_clear_options(&tmp_opt); |
1da177e4 | 1345 | |
1da177e4 | 1346 | tmp_opt.tstamp_ok = tmp_opt.saw_tstamp; |
e0f802fb | 1347 | tcp_openreq_init(req, &tmp_opt, skb, sk); |
1da177e4 | 1348 | |
16bea70a | 1349 | af_ops->init_req(req, sk, skb); |
bb5b7c11 | 1350 | |
284904aa | 1351 | if (security_inet_conn_request(sk, skb, req)) |
bb5b7c11 | 1352 | goto drop_and_free; |
284904aa | 1353 | |
172d69e6 | 1354 | if (!want_cookie || tmp_opt.tstamp_ok) |
5d134f1c | 1355 | TCP_ECN_create_request(req, skb, sock_net(sk)); |
1da177e4 LT |
1356 | |
1357 | if (want_cookie) { | |
fb7b37a7 | 1358 | isn = cookie_init_sequence(af_ops, sk, skb, &req->mss); |
172d69e6 | 1359 | req->cookie_ts = tmp_opt.tstamp_ok; |
1da177e4 | 1360 | } else if (!isn) { |
1da177e4 LT |
1361 | /* VJ's idea. We save last timestamp seen |
1362 | * from the destination in peer table, when entering | |
1363 | * state TIME-WAIT, and check against it before | |
1364 | * accepting new connection request. | |
1365 | * | |
1366 | * If "isn" is not zero, this request hit alive | |
1367 | * timewait bucket, so that all the necessary checks | |
1368 | * are made in the function processing timewait state. | |
1369 | */ | |
d94e0417 OP |
1370 | if (tmp_opt.saw_tstamp && tcp_death_row.sysctl_tw_recycle) { |
1371 | bool strict; | |
1372 | ||
1373 | dst = af_ops->route_req(sk, (struct flowi *)&fl4, req, | |
1374 | &strict); | |
1375 | if (dst && strict && | |
1376 | !tcp_peer_is_proven(req, dst, true)) { | |
de0744af | 1377 | NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_PAWSPASSIVEREJECTED); |
7cd04fa7 | 1378 | goto drop_and_release; |
1da177e4 LT |
1379 | } |
1380 | } | |
1381 | /* Kill the following clause, if you dislike this way. */ | |
1382 | else if (!sysctl_tcp_syncookies && | |
463c84b9 | 1383 | (sysctl_max_syn_backlog - inet_csk_reqsk_queue_len(sk) < |
1da177e4 | 1384 | (sysctl_max_syn_backlog >> 2)) && |
81166dd6 | 1385 | !tcp_peer_is_proven(req, dst, false)) { |
1da177e4 LT |
1386 | /* Without syncookies last quarter of |
1387 | * backlog is filled with destinations, | |
1388 | * proven to be alive. | |
1389 | * It means that we continue to communicate | |
1390 | * to destinations, already remembered | |
1391 | * to the moment of synflood. | |
1392 | */ | |
afd46503 | 1393 | LIMIT_NETDEBUG(KERN_DEBUG pr_fmt("drop open request from %pI4/%u\n"), |
673d57e7 | 1394 | &saddr, ntohs(tcp_hdr(skb)->source)); |
7cd04fa7 | 1395 | goto drop_and_release; |
1da177e4 LT |
1396 | } |
1397 | ||
936b8bdb | 1398 | isn = af_ops->init_seq(skb); |
1da177e4 | 1399 | } |
d94e0417 OP |
1400 | if (!dst) { |
1401 | dst = af_ops->route_req(sk, (struct flowi *)&fl4, req, NULL); | |
1402 | if (!dst) | |
1403 | goto drop_and_free; | |
1404 | } | |
168a8f58 | 1405 | |
843f4a55 | 1406 | tcp_rsk(req)->snt_isn = isn; |
843f4a55 YC |
1407 | tcp_openreq_init_rwin(req, sk, dst); |
1408 | fastopen = !want_cookie && | |
1409 | tcp_try_fastopen(sk, skb, req, &foc, dst); | |
d6274bd8 OP |
1410 | err = af_ops->send_synack(sk, dst, NULL, req, |
1411 | skb_get_queue_mapping(skb), &foc); | |
843f4a55 | 1412 | if (!fastopen) { |
168a8f58 JC |
1413 | if (err || want_cookie) |
1414 | goto drop_and_free; | |
1415 | ||
1416 | tcp_rsk(req)->listener = NULL; | |
168a8f58 | 1417 | inet_csk_reqsk_queue_hash_add(sk, req, TCP_TIMEOUT_INIT); |
843f4a55 | 1418 | } |
1da177e4 | 1419 | |
1da177e4 LT |
1420 | return 0; |
1421 | ||
7cd04fa7 DL |
1422 | drop_and_release: |
1423 | dst_release(dst); | |
1da177e4 | 1424 | drop_and_free: |
60236fdd | 1425 | reqsk_free(req); |
1da177e4 | 1426 | drop: |
848bf15f | 1427 | NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENDROPS); |
1da177e4 LT |
1428 | return 0; |
1429 | } | |
4bc2f18b | 1430 | EXPORT_SYMBOL(tcp_v4_conn_request); |
1da177e4 LT |
1431 | |
1432 | ||
1433 | /* | |
1434 | * The three way handshake has completed - we got a valid synack - | |
1435 | * now create the new socket. | |
1436 | */ | |
1437 | struct sock *tcp_v4_syn_recv_sock(struct sock *sk, struct sk_buff *skb, | |
60236fdd | 1438 | struct request_sock *req, |
1da177e4 LT |
1439 | struct dst_entry *dst) |
1440 | { | |
2e6599cb | 1441 | struct inet_request_sock *ireq; |
1da177e4 LT |
1442 | struct inet_sock *newinet; |
1443 | struct tcp_sock *newtp; | |
1444 | struct sock *newsk; | |
cfb6eeb4 YH |
1445 | #ifdef CONFIG_TCP_MD5SIG |
1446 | struct tcp_md5sig_key *key; | |
1447 | #endif | |
f6d8bd05 | 1448 | struct ip_options_rcu *inet_opt; |
1da177e4 LT |
1449 | |
1450 | if (sk_acceptq_is_full(sk)) | |
1451 | goto exit_overflow; | |
1452 | ||
1da177e4 LT |
1453 | newsk = tcp_create_openreq_child(sk, req, skb); |
1454 | if (!newsk) | |
093d2823 | 1455 | goto exit_nonewsk; |
1da177e4 | 1456 | |
bcd76111 | 1457 | newsk->sk_gso_type = SKB_GSO_TCPV4; |
fae6ef87 | 1458 | inet_sk_rx_dst_set(newsk, skb); |
1da177e4 LT |
1459 | |
1460 | newtp = tcp_sk(newsk); | |
1461 | newinet = inet_sk(newsk); | |
2e6599cb | 1462 | ireq = inet_rsk(req); |
634fb979 ED |
1463 | newinet->inet_daddr = ireq->ir_rmt_addr; |
1464 | newinet->inet_rcv_saddr = ireq->ir_loc_addr; | |
1465 | newinet->inet_saddr = ireq->ir_loc_addr; | |
f6d8bd05 ED |
1466 | inet_opt = ireq->opt; |
1467 | rcu_assign_pointer(newinet->inet_opt, inet_opt); | |
2e6599cb | 1468 | ireq->opt = NULL; |
463c84b9 | 1469 | newinet->mc_index = inet_iif(skb); |
eddc9ec5 | 1470 | newinet->mc_ttl = ip_hdr(skb)->ttl; |
4c507d28 | 1471 | newinet->rcv_tos = ip_hdr(skb)->tos; |
d83d8461 | 1472 | inet_csk(newsk)->icsk_ext_hdr_len = 0; |
f6d8bd05 ED |
1473 | if (inet_opt) |
1474 | inet_csk(newsk)->icsk_ext_hdr_len = inet_opt->opt.optlen; | |
c720c7e8 | 1475 | newinet->inet_id = newtp->write_seq ^ jiffies; |
1da177e4 | 1476 | |
dfd25fff ED |
1477 | if (!dst) { |
1478 | dst = inet_csk_route_child_sock(sk, newsk, req); | |
1479 | if (!dst) | |
1480 | goto put_and_exit; | |
1481 | } else { | |
1482 | /* syncookie case : see end of cookie_v4_check() */ | |
1483 | } | |
0e734419 DM |
1484 | sk_setup_caps(newsk, dst); |
1485 | ||
1da177e4 | 1486 | tcp_sync_mss(newsk, dst_mtu(dst)); |
0dbaee3b | 1487 | newtp->advmss = dst_metric_advmss(dst); |
f5fff5dc TQ |
1488 | if (tcp_sk(sk)->rx_opt.user_mss && |
1489 | tcp_sk(sk)->rx_opt.user_mss < newtp->advmss) | |
1490 | newtp->advmss = tcp_sk(sk)->rx_opt.user_mss; | |
1491 | ||
1da177e4 LT |
1492 | tcp_initialize_rcv_mss(newsk); |
1493 | ||
cfb6eeb4 YH |
1494 | #ifdef CONFIG_TCP_MD5SIG |
1495 | /* Copy over the MD5 key from the original socket */ | |
a915da9b ED |
1496 | key = tcp_md5_do_lookup(sk, (union tcp_md5_addr *)&newinet->inet_daddr, |
1497 | AF_INET); | |
c720c7e8 | 1498 | if (key != NULL) { |
cfb6eeb4 YH |
1499 | /* |
1500 | * We're using one, so create a matching key | |
1501 | * on the newsk structure. If we fail to get | |
1502 | * memory, then we end up not copying the key | |
1503 | * across. Shucks. | |
1504 | */ | |
a915da9b ED |
1505 | tcp_md5_do_add(newsk, (union tcp_md5_addr *)&newinet->inet_daddr, |
1506 | AF_INET, key->key, key->keylen, GFP_ATOMIC); | |
a465419b | 1507 | sk_nocaps_add(newsk, NETIF_F_GSO_MASK); |
cfb6eeb4 YH |
1508 | } |
1509 | #endif | |
1510 | ||
0e734419 DM |
1511 | if (__inet_inherit_port(sk, newsk) < 0) |
1512 | goto put_and_exit; | |
9327f705 | 1513 | __inet_hash_nolisten(newsk, NULL); |
1da177e4 LT |
1514 | |
1515 | return newsk; | |
1516 | ||
1517 | exit_overflow: | |
de0744af | 1518 | NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENOVERFLOWS); |
093d2823 BS |
1519 | exit_nonewsk: |
1520 | dst_release(dst); | |
1da177e4 | 1521 | exit: |
de0744af | 1522 | NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENDROPS); |
1da177e4 | 1523 | return NULL; |
0e734419 | 1524 | put_and_exit: |
e337e24d CP |
1525 | inet_csk_prepare_forced_close(newsk); |
1526 | tcp_done(newsk); | |
0e734419 | 1527 | goto exit; |
1da177e4 | 1528 | } |
4bc2f18b | 1529 | EXPORT_SYMBOL(tcp_v4_syn_recv_sock); |
1da177e4 LT |
1530 | |
1531 | static struct sock *tcp_v4_hnd_req(struct sock *sk, struct sk_buff *skb) | |
1532 | { | |
aa8223c7 | 1533 | struct tcphdr *th = tcp_hdr(skb); |
eddc9ec5 | 1534 | const struct iphdr *iph = ip_hdr(skb); |
1da177e4 | 1535 | struct sock *nsk; |
60236fdd | 1536 | struct request_sock **prev; |
1da177e4 | 1537 | /* Find possible connection requests. */ |
463c84b9 ACM |
1538 | struct request_sock *req = inet_csk_search_req(sk, &prev, th->source, |
1539 | iph->saddr, iph->daddr); | |
1da177e4 | 1540 | if (req) |
8336886f | 1541 | return tcp_check_req(sk, skb, req, prev, false); |
1da177e4 | 1542 | |
3b1e0a65 | 1543 | nsk = inet_lookup_established(sock_net(sk), &tcp_hashinfo, iph->saddr, |
c67499c0 | 1544 | th->source, iph->daddr, th->dest, inet_iif(skb)); |
1da177e4 LT |
1545 | |
1546 | if (nsk) { | |
1547 | if (nsk->sk_state != TCP_TIME_WAIT) { | |
1548 | bh_lock_sock(nsk); | |
1549 | return nsk; | |
1550 | } | |
9469c7b4 | 1551 | inet_twsk_put(inet_twsk(nsk)); |
1da177e4 LT |
1552 | return NULL; |
1553 | } | |
1554 | ||
1555 | #ifdef CONFIG_SYN_COOKIES | |
af9b4738 | 1556 | if (!th->syn) |
1da177e4 LT |
1557 | sk = cookie_v4_check(sk, skb, &(IPCB(skb)->opt)); |
1558 | #endif | |
1559 | return sk; | |
1560 | } | |
1561 | ||
1da177e4 LT |
1562 | /* The socket must have it's spinlock held when we get |
1563 | * here. | |
1564 | * | |
1565 | * We have a potential double-lock case here, so even when | |
1566 | * doing backlog processing we use the BH locking scheme. | |
1567 | * This is because we cannot sleep with the original spinlock | |
1568 | * held. | |
1569 | */ | |
1570 | int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb) | |
1571 | { | |
cfb6eeb4 YH |
1572 | struct sock *rsk; |
1573 | #ifdef CONFIG_TCP_MD5SIG | |
1574 | /* | |
1575 | * We really want to reject the packet as early as possible | |
1576 | * if: | |
1577 | * o We're expecting an MD5'd packet and this is no MD5 tcp option | |
1578 | * o There is an MD5 option and we're not expecting one | |
1579 | */ | |
7174259e | 1580 | if (tcp_v4_inbound_md5_hash(sk, skb)) |
cfb6eeb4 YH |
1581 | goto discard; |
1582 | #endif | |
1583 | ||
1da177e4 | 1584 | if (sk->sk_state == TCP_ESTABLISHED) { /* Fast path */ |
404e0a8b ED |
1585 | struct dst_entry *dst = sk->sk_rx_dst; |
1586 | ||
bdeab991 | 1587 | sock_rps_save_rxhash(sk, skb); |
404e0a8b | 1588 | if (dst) { |
505fbcf0 ED |
1589 | if (inet_sk(sk)->rx_dst_ifindex != skb->skb_iif || |
1590 | dst->ops->check(dst, 0) == NULL) { | |
92101b3b DM |
1591 | dst_release(dst); |
1592 | sk->sk_rx_dst = NULL; | |
1593 | } | |
1594 | } | |
c995ae22 | 1595 | tcp_rcv_established(sk, skb, tcp_hdr(skb), skb->len); |
1da177e4 LT |
1596 | return 0; |
1597 | } | |
1598 | ||
ab6a5bb6 | 1599 | if (skb->len < tcp_hdrlen(skb) || tcp_checksum_complete(skb)) |
1da177e4 LT |
1600 | goto csum_err; |
1601 | ||
1602 | if (sk->sk_state == TCP_LISTEN) { | |
1603 | struct sock *nsk = tcp_v4_hnd_req(sk, skb); | |
1604 | if (!nsk) | |
1605 | goto discard; | |
1606 | ||
1607 | if (nsk != sk) { | |
bdeab991 | 1608 | sock_rps_save_rxhash(nsk, skb); |
cfb6eeb4 YH |
1609 | if (tcp_child_process(sk, nsk, skb)) { |
1610 | rsk = nsk; | |
1da177e4 | 1611 | goto reset; |
cfb6eeb4 | 1612 | } |
1da177e4 LT |
1613 | return 0; |
1614 | } | |
ca55158c | 1615 | } else |
bdeab991 | 1616 | sock_rps_save_rxhash(sk, skb); |
ca55158c | 1617 | |
aa8223c7 | 1618 | if (tcp_rcv_state_process(sk, skb, tcp_hdr(skb), skb->len)) { |
cfb6eeb4 | 1619 | rsk = sk; |
1da177e4 | 1620 | goto reset; |
cfb6eeb4 | 1621 | } |
1da177e4 LT |
1622 | return 0; |
1623 | ||
1624 | reset: | |
cfb6eeb4 | 1625 | tcp_v4_send_reset(rsk, skb); |
1da177e4 LT |
1626 | discard: |
1627 | kfree_skb(skb); | |
1628 | /* Be careful here. If this function gets more complicated and | |
1629 | * gcc suffers from register pressure on the x86, sk (in %ebx) | |
1630 | * might be destroyed here. This current version compiles correctly, | |
1631 | * but you have been warned. | |
1632 | */ | |
1633 | return 0; | |
1634 | ||
1635 | csum_err: | |
6a5dc9e5 | 1636 | TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_CSUMERRORS); |
63231bdd | 1637 | TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_INERRS); |
1da177e4 LT |
1638 | goto discard; |
1639 | } | |
4bc2f18b | 1640 | EXPORT_SYMBOL(tcp_v4_do_rcv); |
1da177e4 | 1641 | |
160eb5a6 | 1642 | void tcp_v4_early_demux(struct sk_buff *skb) |
41063e9d | 1643 | { |
41063e9d DM |
1644 | const struct iphdr *iph; |
1645 | const struct tcphdr *th; | |
1646 | struct sock *sk; | |
41063e9d | 1647 | |
41063e9d | 1648 | if (skb->pkt_type != PACKET_HOST) |
160eb5a6 | 1649 | return; |
41063e9d | 1650 | |
45f00f99 | 1651 | if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct tcphdr))) |
160eb5a6 | 1652 | return; |
41063e9d DM |
1653 | |
1654 | iph = ip_hdr(skb); | |
45f00f99 | 1655 | th = tcp_hdr(skb); |
41063e9d DM |
1656 | |
1657 | if (th->doff < sizeof(struct tcphdr) / 4) | |
160eb5a6 | 1658 | return; |
41063e9d | 1659 | |
45f00f99 | 1660 | sk = __inet_lookup_established(dev_net(skb->dev), &tcp_hashinfo, |
41063e9d | 1661 | iph->saddr, th->source, |
7011d085 | 1662 | iph->daddr, ntohs(th->dest), |
9cb429d6 | 1663 | skb->skb_iif); |
41063e9d DM |
1664 | if (sk) { |
1665 | skb->sk = sk; | |
1666 | skb->destructor = sock_edemux; | |
1667 | if (sk->sk_state != TCP_TIME_WAIT) { | |
1668 | struct dst_entry *dst = sk->sk_rx_dst; | |
505fbcf0 | 1669 | |
41063e9d DM |
1670 | if (dst) |
1671 | dst = dst_check(dst, 0); | |
92101b3b | 1672 | if (dst && |
505fbcf0 | 1673 | inet_sk(sk)->rx_dst_ifindex == skb->skb_iif) |
92101b3b | 1674 | skb_dst_set_noref(skb, dst); |
41063e9d DM |
1675 | } |
1676 | } | |
41063e9d DM |
1677 | } |
1678 | ||
b2fb4f54 ED |
1679 | /* Packet is added to VJ-style prequeue for processing in process |
1680 | * context, if a reader task is waiting. Apparently, this exciting | |
1681 | * idea (VJ's mail "Re: query about TCP header on tcp-ip" of 07 Sep 93) | |
1682 | * failed somewhere. Latency? Burstiness? Well, at least now we will | |
1683 | * see, why it failed. 8)8) --ANK | |
1684 | * | |
1685 | */ | |
1686 | bool tcp_prequeue(struct sock *sk, struct sk_buff *skb) | |
1687 | { | |
1688 | struct tcp_sock *tp = tcp_sk(sk); | |
1689 | ||
1690 | if (sysctl_tcp_low_latency || !tp->ucopy.task) | |
1691 | return false; | |
1692 | ||
1693 | if (skb->len <= tcp_hdrlen(skb) && | |
1694 | skb_queue_len(&tp->ucopy.prequeue) == 0) | |
1695 | return false; | |
1696 | ||
58717686 | 1697 | skb_dst_force(skb); |
b2fb4f54 ED |
1698 | __skb_queue_tail(&tp->ucopy.prequeue, skb); |
1699 | tp->ucopy.memory += skb->truesize; | |
1700 | if (tp->ucopy.memory > sk->sk_rcvbuf) { | |
1701 | struct sk_buff *skb1; | |
1702 | ||
1703 | BUG_ON(sock_owned_by_user(sk)); | |
1704 | ||
1705 | while ((skb1 = __skb_dequeue(&tp->ucopy.prequeue)) != NULL) { | |
1706 | sk_backlog_rcv(sk, skb1); | |
1707 | NET_INC_STATS_BH(sock_net(sk), | |
1708 | LINUX_MIB_TCPPREQUEUEDROPPED); | |
1709 | } | |
1710 | ||
1711 | tp->ucopy.memory = 0; | |
1712 | } else if (skb_queue_len(&tp->ucopy.prequeue) == 1) { | |
1713 | wake_up_interruptible_sync_poll(sk_sleep(sk), | |
1714 | POLLIN | POLLRDNORM | POLLRDBAND); | |
1715 | if (!inet_csk_ack_scheduled(sk)) | |
1716 | inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK, | |
1717 | (3 * tcp_rto_min(sk)) / 4, | |
1718 | TCP_RTO_MAX); | |
1719 | } | |
1720 | return true; | |
1721 | } | |
1722 | EXPORT_SYMBOL(tcp_prequeue); | |
1723 | ||
1da177e4 LT |
1724 | /* |
1725 | * From tcp_input.c | |
1726 | */ | |
1727 | ||
1728 | int tcp_v4_rcv(struct sk_buff *skb) | |
1729 | { | |
eddc9ec5 | 1730 | const struct iphdr *iph; |
cf533ea5 | 1731 | const struct tcphdr *th; |
1da177e4 LT |
1732 | struct sock *sk; |
1733 | int ret; | |
a86b1e30 | 1734 | struct net *net = dev_net(skb->dev); |
1da177e4 LT |
1735 | |
1736 | if (skb->pkt_type != PACKET_HOST) | |
1737 | goto discard_it; | |
1738 | ||
1739 | /* Count it even if it's bad */ | |
63231bdd | 1740 | TCP_INC_STATS_BH(net, TCP_MIB_INSEGS); |
1da177e4 LT |
1741 | |
1742 | if (!pskb_may_pull(skb, sizeof(struct tcphdr))) | |
1743 | goto discard_it; | |
1744 | ||
aa8223c7 | 1745 | th = tcp_hdr(skb); |
1da177e4 LT |
1746 | |
1747 | if (th->doff < sizeof(struct tcphdr) / 4) | |
1748 | goto bad_packet; | |
1749 | if (!pskb_may_pull(skb, th->doff * 4)) | |
1750 | goto discard_it; | |
1751 | ||
1752 | /* An explanation is required here, I think. | |
1753 | * Packet length and doff are validated by header prediction, | |
caa20d9a | 1754 | * provided case of th->doff==0 is eliminated. |
1da177e4 | 1755 | * So, we defer the checks. */ |
ed70fcfc TH |
1756 | |
1757 | if (skb_checksum_init(skb, IPPROTO_TCP, inet_compute_pseudo)) | |
6a5dc9e5 | 1758 | goto csum_error; |
1da177e4 | 1759 | |
aa8223c7 | 1760 | th = tcp_hdr(skb); |
eddc9ec5 | 1761 | iph = ip_hdr(skb); |
1da177e4 LT |
1762 | TCP_SKB_CB(skb)->seq = ntohl(th->seq); |
1763 | TCP_SKB_CB(skb)->end_seq = (TCP_SKB_CB(skb)->seq + th->syn + th->fin + | |
1764 | skb->len - th->doff * 4); | |
1765 | TCP_SKB_CB(skb)->ack_seq = ntohl(th->ack_seq); | |
1766 | TCP_SKB_CB(skb)->when = 0; | |
b82d1bb4 | 1767 | TCP_SKB_CB(skb)->ip_dsfield = ipv4_get_dsfield(iph); |
1da177e4 LT |
1768 | TCP_SKB_CB(skb)->sacked = 0; |
1769 | ||
9a1f27c4 | 1770 | sk = __inet_lookup_skb(&tcp_hashinfo, skb, th->source, th->dest); |
1da177e4 LT |
1771 | if (!sk) |
1772 | goto no_tcp_socket; | |
1773 | ||
bb134d5d ED |
1774 | process: |
1775 | if (sk->sk_state == TCP_TIME_WAIT) | |
1776 | goto do_time_wait; | |
1777 | ||
6cce09f8 ED |
1778 | if (unlikely(iph->ttl < inet_sk(sk)->min_ttl)) { |
1779 | NET_INC_STATS_BH(net, LINUX_MIB_TCPMINTTLDROP); | |
d218d111 | 1780 | goto discard_and_relse; |
6cce09f8 | 1781 | } |
d218d111 | 1782 | |
1da177e4 LT |
1783 | if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb)) |
1784 | goto discard_and_relse; | |
b59c2701 | 1785 | nf_reset(skb); |
1da177e4 | 1786 | |
fda9ef5d | 1787 | if (sk_filter(sk, skb)) |
1da177e4 LT |
1788 | goto discard_and_relse; |
1789 | ||
8b80cda5 | 1790 | sk_mark_napi_id(sk, skb); |
1da177e4 LT |
1791 | skb->dev = NULL; |
1792 | ||
c6366184 | 1793 | bh_lock_sock_nested(sk); |
1da177e4 LT |
1794 | ret = 0; |
1795 | if (!sock_owned_by_user(sk)) { | |
1a2449a8 CL |
1796 | #ifdef CONFIG_NET_DMA |
1797 | struct tcp_sock *tp = tcp_sk(sk); | |
1798 | if (!tp->ucopy.dma_chan && tp->ucopy.pinned_list) | |
a2bd1140 | 1799 | tp->ucopy.dma_chan = net_dma_find_channel(); |
1a2449a8 | 1800 | if (tp->ucopy.dma_chan) |
1da177e4 | 1801 | ret = tcp_v4_do_rcv(sk, skb); |
1a2449a8 CL |
1802 | else |
1803 | #endif | |
1804 | { | |
1805 | if (!tcp_prequeue(sk, skb)) | |
ae8d7f88 | 1806 | ret = tcp_v4_do_rcv(sk, skb); |
1a2449a8 | 1807 | } |
da882c1f ED |
1808 | } else if (unlikely(sk_add_backlog(sk, skb, |
1809 | sk->sk_rcvbuf + sk->sk_sndbuf))) { | |
6b03a53a | 1810 | bh_unlock_sock(sk); |
6cce09f8 | 1811 | NET_INC_STATS_BH(net, LINUX_MIB_TCPBACKLOGDROP); |
6b03a53a ZY |
1812 | goto discard_and_relse; |
1813 | } | |
1da177e4 LT |
1814 | bh_unlock_sock(sk); |
1815 | ||
1816 | sock_put(sk); | |
1817 | ||
1818 | return ret; | |
1819 | ||
1820 | no_tcp_socket: | |
1821 | if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) | |
1822 | goto discard_it; | |
1823 | ||
1824 | if (skb->len < (th->doff << 2) || tcp_checksum_complete(skb)) { | |
6a5dc9e5 ED |
1825 | csum_error: |
1826 | TCP_INC_STATS_BH(net, TCP_MIB_CSUMERRORS); | |
1da177e4 | 1827 | bad_packet: |
63231bdd | 1828 | TCP_INC_STATS_BH(net, TCP_MIB_INERRS); |
1da177e4 | 1829 | } else { |
cfb6eeb4 | 1830 | tcp_v4_send_reset(NULL, skb); |
1da177e4 LT |
1831 | } |
1832 | ||
1833 | discard_it: | |
1834 | /* Discard frame. */ | |
1835 | kfree_skb(skb); | |
e905a9ed | 1836 | return 0; |
1da177e4 LT |
1837 | |
1838 | discard_and_relse: | |
1839 | sock_put(sk); | |
1840 | goto discard_it; | |
1841 | ||
1842 | do_time_wait: | |
1843 | if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) { | |
9469c7b4 | 1844 | inet_twsk_put(inet_twsk(sk)); |
1da177e4 LT |
1845 | goto discard_it; |
1846 | } | |
1847 | ||
6a5dc9e5 | 1848 | if (skb->len < (th->doff << 2)) { |
9469c7b4 | 1849 | inet_twsk_put(inet_twsk(sk)); |
6a5dc9e5 ED |
1850 | goto bad_packet; |
1851 | } | |
1852 | if (tcp_checksum_complete(skb)) { | |
1853 | inet_twsk_put(inet_twsk(sk)); | |
1854 | goto csum_error; | |
1da177e4 | 1855 | } |
9469c7b4 | 1856 | switch (tcp_timewait_state_process(inet_twsk(sk), skb, th)) { |
1da177e4 | 1857 | case TCP_TW_SYN: { |
c346dca1 | 1858 | struct sock *sk2 = inet_lookup_listener(dev_net(skb->dev), |
c67499c0 | 1859 | &tcp_hashinfo, |
da5e3630 | 1860 | iph->saddr, th->source, |
eddc9ec5 | 1861 | iph->daddr, th->dest, |
463c84b9 | 1862 | inet_iif(skb)); |
1da177e4 | 1863 | if (sk2) { |
9469c7b4 YH |
1864 | inet_twsk_deschedule(inet_twsk(sk), &tcp_death_row); |
1865 | inet_twsk_put(inet_twsk(sk)); | |
1da177e4 LT |
1866 | sk = sk2; |
1867 | goto process; | |
1868 | } | |
1869 | /* Fall through to ACK */ | |
1870 | } | |
1871 | case TCP_TW_ACK: | |
1872 | tcp_v4_timewait_ack(sk, skb); | |
1873 | break; | |
1874 | case TCP_TW_RST: | |
1875 | goto no_tcp_socket; | |
1876 | case TCP_TW_SUCCESS:; | |
1877 | } | |
1878 | goto discard_it; | |
1879 | } | |
1880 | ||
ccb7c410 DM |
1881 | static struct timewait_sock_ops tcp_timewait_sock_ops = { |
1882 | .twsk_obj_size = sizeof(struct tcp_timewait_sock), | |
1883 | .twsk_unique = tcp_twsk_unique, | |
1884 | .twsk_destructor= tcp_twsk_destructor, | |
ccb7c410 | 1885 | }; |
1da177e4 | 1886 | |
63d02d15 | 1887 | void inet_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb) |
5d299f3d ED |
1888 | { |
1889 | struct dst_entry *dst = skb_dst(skb); | |
1890 | ||
1891 | dst_hold(dst); | |
1892 | sk->sk_rx_dst = dst; | |
1893 | inet_sk(sk)->rx_dst_ifindex = skb->skb_iif; | |
1894 | } | |
63d02d15 | 1895 | EXPORT_SYMBOL(inet_sk_rx_dst_set); |
5d299f3d | 1896 | |
3b401a81 | 1897 | const struct inet_connection_sock_af_ops ipv4_specific = { |
543d9cfe ACM |
1898 | .queue_xmit = ip_queue_xmit, |
1899 | .send_check = tcp_v4_send_check, | |
1900 | .rebuild_header = inet_sk_rebuild_header, | |
5d299f3d | 1901 | .sk_rx_dst_set = inet_sk_rx_dst_set, |
543d9cfe ACM |
1902 | .conn_request = tcp_v4_conn_request, |
1903 | .syn_recv_sock = tcp_v4_syn_recv_sock, | |
543d9cfe ACM |
1904 | .net_header_len = sizeof(struct iphdr), |
1905 | .setsockopt = ip_setsockopt, | |
1906 | .getsockopt = ip_getsockopt, | |
1907 | .addr2sockaddr = inet_csk_addr2sockaddr, | |
1908 | .sockaddr_len = sizeof(struct sockaddr_in), | |
ab1e0a13 | 1909 | .bind_conflict = inet_csk_bind_conflict, |
3fdadf7d | 1910 | #ifdef CONFIG_COMPAT |
543d9cfe ACM |
1911 | .compat_setsockopt = compat_ip_setsockopt, |
1912 | .compat_getsockopt = compat_ip_getsockopt, | |
3fdadf7d | 1913 | #endif |
1da177e4 | 1914 | }; |
4bc2f18b | 1915 | EXPORT_SYMBOL(ipv4_specific); |
1da177e4 | 1916 | |
cfb6eeb4 | 1917 | #ifdef CONFIG_TCP_MD5SIG |
b2e4b3de | 1918 | static const struct tcp_sock_af_ops tcp_sock_ipv4_specific = { |
cfb6eeb4 | 1919 | .md5_lookup = tcp_v4_md5_lookup, |
49a72dfb | 1920 | .calc_md5_hash = tcp_v4_md5_hash_skb, |
cfb6eeb4 | 1921 | .md5_parse = tcp_v4_parse_md5_keys, |
cfb6eeb4 | 1922 | }; |
b6332e6c | 1923 | #endif |
cfb6eeb4 | 1924 | |
1da177e4 LT |
1925 | /* NOTE: A lot of things set to zero explicitly by call to |
1926 | * sk_alloc() so need not be done here. | |
1927 | */ | |
1928 | static int tcp_v4_init_sock(struct sock *sk) | |
1929 | { | |
6687e988 | 1930 | struct inet_connection_sock *icsk = inet_csk(sk); |
1da177e4 | 1931 | |
900f65d3 | 1932 | tcp_init_sock(sk); |
1da177e4 | 1933 | |
8292a17a | 1934 | icsk->icsk_af_ops = &ipv4_specific; |
900f65d3 | 1935 | |
cfb6eeb4 | 1936 | #ifdef CONFIG_TCP_MD5SIG |
ac807fa8 | 1937 | tcp_sk(sk)->af_specific = &tcp_sock_ipv4_specific; |
cfb6eeb4 | 1938 | #endif |
1da177e4 | 1939 | |
1da177e4 LT |
1940 | return 0; |
1941 | } | |
1942 | ||
7d06b2e0 | 1943 | void tcp_v4_destroy_sock(struct sock *sk) |
1da177e4 LT |
1944 | { |
1945 | struct tcp_sock *tp = tcp_sk(sk); | |
1946 | ||
1947 | tcp_clear_xmit_timers(sk); | |
1948 | ||
6687e988 | 1949 | tcp_cleanup_congestion_control(sk); |
317a76f9 | 1950 | |
1da177e4 | 1951 | /* Cleanup up the write buffer. */ |
fe067e8a | 1952 | tcp_write_queue_purge(sk); |
1da177e4 LT |
1953 | |
1954 | /* Cleans up our, hopefully empty, out_of_order_queue. */ | |
e905a9ed | 1955 | __skb_queue_purge(&tp->out_of_order_queue); |
1da177e4 | 1956 | |
cfb6eeb4 YH |
1957 | #ifdef CONFIG_TCP_MD5SIG |
1958 | /* Clean up the MD5 key list, if any */ | |
1959 | if (tp->md5sig_info) { | |
a915da9b | 1960 | tcp_clear_md5_list(sk); |
a8afca03 | 1961 | kfree_rcu(tp->md5sig_info, rcu); |
cfb6eeb4 YH |
1962 | tp->md5sig_info = NULL; |
1963 | } | |
1964 | #endif | |
1965 | ||
1a2449a8 CL |
1966 | #ifdef CONFIG_NET_DMA |
1967 | /* Cleans up our sk_async_wait_queue */ | |
e905a9ed | 1968 | __skb_queue_purge(&sk->sk_async_wait_queue); |
1a2449a8 CL |
1969 | #endif |
1970 | ||
1da177e4 LT |
1971 | /* Clean prequeue, it must be empty really */ |
1972 | __skb_queue_purge(&tp->ucopy.prequeue); | |
1973 | ||
1974 | /* Clean up a referenced TCP bind bucket. */ | |
463c84b9 | 1975 | if (inet_csk(sk)->icsk_bind_hash) |
ab1e0a13 | 1976 | inet_put_port(sk); |
1da177e4 | 1977 | |
168a8f58 | 1978 | BUG_ON(tp->fastopen_rsk != NULL); |
435cf559 | 1979 | |
cf60af03 YC |
1980 | /* If socket is aborted during connect operation */ |
1981 | tcp_free_fastopen_req(tp); | |
1982 | ||
180d8cd9 | 1983 | sk_sockets_allocated_dec(sk); |
d1a4c0b3 | 1984 | sock_release_memcg(sk); |
1da177e4 | 1985 | } |
1da177e4 LT |
1986 | EXPORT_SYMBOL(tcp_v4_destroy_sock); |
1987 | ||
1988 | #ifdef CONFIG_PROC_FS | |
1989 | /* Proc filesystem TCP sock list dumping. */ | |
1990 | ||
a8b690f9 TH |
1991 | /* |
1992 | * Get next listener socket follow cur. If cur is NULL, get first socket | |
1993 | * starting from bucket given in st->bucket; when st->bucket is zero the | |
1994 | * very first socket in the hash table is returned. | |
1995 | */ | |
1da177e4 LT |
1996 | static void *listening_get_next(struct seq_file *seq, void *cur) |
1997 | { | |
463c84b9 | 1998 | struct inet_connection_sock *icsk; |
c25eb3bf | 1999 | struct hlist_nulls_node *node; |
1da177e4 | 2000 | struct sock *sk = cur; |
5caea4ea | 2001 | struct inet_listen_hashbucket *ilb; |
5799de0b | 2002 | struct tcp_iter_state *st = seq->private; |
a4146b1b | 2003 | struct net *net = seq_file_net(seq); |
1da177e4 LT |
2004 | |
2005 | if (!sk) { | |
a8b690f9 | 2006 | ilb = &tcp_hashinfo.listening_hash[st->bucket]; |
5caea4ea | 2007 | spin_lock_bh(&ilb->lock); |
c25eb3bf | 2008 | sk = sk_nulls_head(&ilb->head); |
a8b690f9 | 2009 | st->offset = 0; |
1da177e4 LT |
2010 | goto get_sk; |
2011 | } | |
5caea4ea | 2012 | ilb = &tcp_hashinfo.listening_hash[st->bucket]; |
1da177e4 | 2013 | ++st->num; |
a8b690f9 | 2014 | ++st->offset; |
1da177e4 LT |
2015 | |
2016 | if (st->state == TCP_SEQ_STATE_OPENREQ) { | |
60236fdd | 2017 | struct request_sock *req = cur; |
1da177e4 | 2018 | |
72a3effa | 2019 | icsk = inet_csk(st->syn_wait_sk); |
1da177e4 LT |
2020 | req = req->dl_next; |
2021 | while (1) { | |
2022 | while (req) { | |
bdccc4ca | 2023 | if (req->rsk_ops->family == st->family) { |
1da177e4 LT |
2024 | cur = req; |
2025 | goto out; | |
2026 | } | |
2027 | req = req->dl_next; | |
2028 | } | |
72a3effa | 2029 | if (++st->sbucket >= icsk->icsk_accept_queue.listen_opt->nr_table_entries) |
1da177e4 LT |
2030 | break; |
2031 | get_req: | |
463c84b9 | 2032 | req = icsk->icsk_accept_queue.listen_opt->syn_table[st->sbucket]; |
1da177e4 | 2033 | } |
1bde5ac4 | 2034 | sk = sk_nulls_next(st->syn_wait_sk); |
1da177e4 | 2035 | st->state = TCP_SEQ_STATE_LISTENING; |
463c84b9 | 2036 | read_unlock_bh(&icsk->icsk_accept_queue.syn_wait_lock); |
1da177e4 | 2037 | } else { |
e905a9ed | 2038 | icsk = inet_csk(sk); |
463c84b9 ACM |
2039 | read_lock_bh(&icsk->icsk_accept_queue.syn_wait_lock); |
2040 | if (reqsk_queue_len(&icsk->icsk_accept_queue)) | |
1da177e4 | 2041 | goto start_req; |
463c84b9 | 2042 | read_unlock_bh(&icsk->icsk_accept_queue.syn_wait_lock); |
1bde5ac4 | 2043 | sk = sk_nulls_next(sk); |
1da177e4 LT |
2044 | } |
2045 | get_sk: | |
c25eb3bf | 2046 | sk_nulls_for_each_from(sk, node) { |
8475ef9f PE |
2047 | if (!net_eq(sock_net(sk), net)) |
2048 | continue; | |
2049 | if (sk->sk_family == st->family) { | |
1da177e4 LT |
2050 | cur = sk; |
2051 | goto out; | |
2052 | } | |
e905a9ed | 2053 | icsk = inet_csk(sk); |
463c84b9 ACM |
2054 | read_lock_bh(&icsk->icsk_accept_queue.syn_wait_lock); |
2055 | if (reqsk_queue_len(&icsk->icsk_accept_queue)) { | |
1da177e4 LT |
2056 | start_req: |
2057 | st->uid = sock_i_uid(sk); | |
2058 | st->syn_wait_sk = sk; | |
2059 | st->state = TCP_SEQ_STATE_OPENREQ; | |
2060 | st->sbucket = 0; | |
2061 | goto get_req; | |
2062 | } | |
463c84b9 | 2063 | read_unlock_bh(&icsk->icsk_accept_queue.syn_wait_lock); |
1da177e4 | 2064 | } |
5caea4ea | 2065 | spin_unlock_bh(&ilb->lock); |
a8b690f9 | 2066 | st->offset = 0; |
0f7ff927 | 2067 | if (++st->bucket < INET_LHTABLE_SIZE) { |
5caea4ea ED |
2068 | ilb = &tcp_hashinfo.listening_hash[st->bucket]; |
2069 | spin_lock_bh(&ilb->lock); | |
c25eb3bf | 2070 | sk = sk_nulls_head(&ilb->head); |
1da177e4 LT |
2071 | goto get_sk; |
2072 | } | |
2073 | cur = NULL; | |
2074 | out: | |
2075 | return cur; | |
2076 | } | |
2077 | ||
2078 | static void *listening_get_idx(struct seq_file *seq, loff_t *pos) | |
2079 | { | |
a8b690f9 TH |
2080 | struct tcp_iter_state *st = seq->private; |
2081 | void *rc; | |
2082 | ||
2083 | st->bucket = 0; | |
2084 | st->offset = 0; | |
2085 | rc = listening_get_next(seq, NULL); | |
1da177e4 LT |
2086 | |
2087 | while (rc && *pos) { | |
2088 | rc = listening_get_next(seq, rc); | |
2089 | --*pos; | |
2090 | } | |
2091 | return rc; | |
2092 | } | |
2093 | ||
05dbc7b5 | 2094 | static inline bool empty_bucket(const struct tcp_iter_state *st) |
6eac5604 | 2095 | { |
05dbc7b5 | 2096 | return hlist_nulls_empty(&tcp_hashinfo.ehash[st->bucket].chain); |
6eac5604 AK |
2097 | } |
2098 | ||
a8b690f9 TH |
2099 | /* |
2100 | * Get first established socket starting from bucket given in st->bucket. | |
2101 | * If st->bucket is zero, the very first socket in the hash is returned. | |
2102 | */ | |
1da177e4 LT |
2103 | static void *established_get_first(struct seq_file *seq) |
2104 | { | |
5799de0b | 2105 | struct tcp_iter_state *st = seq->private; |
a4146b1b | 2106 | struct net *net = seq_file_net(seq); |
1da177e4 LT |
2107 | void *rc = NULL; |
2108 | ||
a8b690f9 TH |
2109 | st->offset = 0; |
2110 | for (; st->bucket <= tcp_hashinfo.ehash_mask; ++st->bucket) { | |
1da177e4 | 2111 | struct sock *sk; |
3ab5aee7 | 2112 | struct hlist_nulls_node *node; |
9db66bdc | 2113 | spinlock_t *lock = inet_ehash_lockp(&tcp_hashinfo, st->bucket); |
1da177e4 | 2114 | |
6eac5604 AK |
2115 | /* Lockless fast path for the common case of empty buckets */ |
2116 | if (empty_bucket(st)) | |
2117 | continue; | |
2118 | ||
9db66bdc | 2119 | spin_lock_bh(lock); |
3ab5aee7 | 2120 | sk_nulls_for_each(sk, node, &tcp_hashinfo.ehash[st->bucket].chain) { |
f40c8174 | 2121 | if (sk->sk_family != st->family || |
878628fb | 2122 | !net_eq(sock_net(sk), net)) { |
1da177e4 LT |
2123 | continue; |
2124 | } | |
2125 | rc = sk; | |
2126 | goto out; | |
2127 | } | |
9db66bdc | 2128 | spin_unlock_bh(lock); |
1da177e4 LT |
2129 | } |
2130 | out: | |
2131 | return rc; | |
2132 | } | |
2133 | ||
2134 | static void *established_get_next(struct seq_file *seq, void *cur) | |
2135 | { | |
2136 | struct sock *sk = cur; | |
3ab5aee7 | 2137 | struct hlist_nulls_node *node; |
5799de0b | 2138 | struct tcp_iter_state *st = seq->private; |
a4146b1b | 2139 | struct net *net = seq_file_net(seq); |
1da177e4 LT |
2140 | |
2141 | ++st->num; | |
a8b690f9 | 2142 | ++st->offset; |
1da177e4 | 2143 | |
05dbc7b5 | 2144 | sk = sk_nulls_next(sk); |
1da177e4 | 2145 | |
3ab5aee7 | 2146 | sk_nulls_for_each_from(sk, node) { |
878628fb | 2147 | if (sk->sk_family == st->family && net_eq(sock_net(sk), net)) |
05dbc7b5 | 2148 | return sk; |
1da177e4 LT |
2149 | } |
2150 | ||
05dbc7b5 ED |
2151 | spin_unlock_bh(inet_ehash_lockp(&tcp_hashinfo, st->bucket)); |
2152 | ++st->bucket; | |
2153 | return established_get_first(seq); | |
1da177e4 LT |
2154 | } |
2155 | ||
2156 | static void *established_get_idx(struct seq_file *seq, loff_t pos) | |
2157 | { | |
a8b690f9 TH |
2158 | struct tcp_iter_state *st = seq->private; |
2159 | void *rc; | |
2160 | ||
2161 | st->bucket = 0; | |
2162 | rc = established_get_first(seq); | |
1da177e4 LT |
2163 | |
2164 | while (rc && pos) { | |
2165 | rc = established_get_next(seq, rc); | |
2166 | --pos; | |
7174259e | 2167 | } |
1da177e4 LT |
2168 | return rc; |
2169 | } | |
2170 | ||
2171 | static void *tcp_get_idx(struct seq_file *seq, loff_t pos) | |
2172 | { | |
2173 | void *rc; | |
5799de0b | 2174 | struct tcp_iter_state *st = seq->private; |
1da177e4 | 2175 | |
1da177e4 LT |
2176 | st->state = TCP_SEQ_STATE_LISTENING; |
2177 | rc = listening_get_idx(seq, &pos); | |
2178 | ||
2179 | if (!rc) { | |
1da177e4 LT |
2180 | st->state = TCP_SEQ_STATE_ESTABLISHED; |
2181 | rc = established_get_idx(seq, pos); | |
2182 | } | |
2183 | ||
2184 | return rc; | |
2185 | } | |
2186 | ||
a8b690f9 TH |
2187 | static void *tcp_seek_last_pos(struct seq_file *seq) |
2188 | { | |
2189 | struct tcp_iter_state *st = seq->private; | |
2190 | int offset = st->offset; | |
2191 | int orig_num = st->num; | |
2192 | void *rc = NULL; | |
2193 | ||
2194 | switch (st->state) { | |
2195 | case TCP_SEQ_STATE_OPENREQ: | |
2196 | case TCP_SEQ_STATE_LISTENING: | |
2197 | if (st->bucket >= INET_LHTABLE_SIZE) | |
2198 | break; | |
2199 | st->state = TCP_SEQ_STATE_LISTENING; | |
2200 | rc = listening_get_next(seq, NULL); | |
2201 | while (offset-- && rc) | |
2202 | rc = listening_get_next(seq, rc); | |
2203 | if (rc) | |
2204 | break; | |
2205 | st->bucket = 0; | |
05dbc7b5 | 2206 | st->state = TCP_SEQ_STATE_ESTABLISHED; |
a8b690f9 TH |
2207 | /* Fallthrough */ |
2208 | case TCP_SEQ_STATE_ESTABLISHED: | |
a8b690f9 TH |
2209 | if (st->bucket > tcp_hashinfo.ehash_mask) |
2210 | break; | |
2211 | rc = established_get_first(seq); | |
2212 | while (offset-- && rc) | |
2213 | rc = established_get_next(seq, rc); | |
2214 | } | |
2215 | ||
2216 | st->num = orig_num; | |
2217 | ||
2218 | return rc; | |
2219 | } | |
2220 | ||
1da177e4 LT |
2221 | static void *tcp_seq_start(struct seq_file *seq, loff_t *pos) |
2222 | { | |
5799de0b | 2223 | struct tcp_iter_state *st = seq->private; |
a8b690f9 TH |
2224 | void *rc; |
2225 | ||
2226 | if (*pos && *pos == st->last_pos) { | |
2227 | rc = tcp_seek_last_pos(seq); | |
2228 | if (rc) | |
2229 | goto out; | |
2230 | } | |
2231 | ||
1da177e4 LT |
2232 | st->state = TCP_SEQ_STATE_LISTENING; |
2233 | st->num = 0; | |
a8b690f9 TH |
2234 | st->bucket = 0; |
2235 | st->offset = 0; | |
2236 | rc = *pos ? tcp_get_idx(seq, *pos - 1) : SEQ_START_TOKEN; | |
2237 | ||
2238 | out: | |
2239 | st->last_pos = *pos; | |
2240 | return rc; | |
1da177e4 LT |
2241 | } |
2242 | ||
2243 | static void *tcp_seq_next(struct seq_file *seq, void *v, loff_t *pos) | |
2244 | { | |
a8b690f9 | 2245 | struct tcp_iter_state *st = seq->private; |
1da177e4 | 2246 | void *rc = NULL; |
1da177e4 LT |
2247 | |
2248 | if (v == SEQ_START_TOKEN) { | |
2249 | rc = tcp_get_idx(seq, 0); | |
2250 | goto out; | |
2251 | } | |
1da177e4 LT |
2252 | |
2253 | switch (st->state) { | |
2254 | case TCP_SEQ_STATE_OPENREQ: | |
2255 | case TCP_SEQ_STATE_LISTENING: | |
2256 | rc = listening_get_next(seq, v); | |
2257 | if (!rc) { | |
1da177e4 | 2258 | st->state = TCP_SEQ_STATE_ESTABLISHED; |
a8b690f9 TH |
2259 | st->bucket = 0; |
2260 | st->offset = 0; | |
1da177e4 LT |
2261 | rc = established_get_first(seq); |
2262 | } | |
2263 | break; | |
2264 | case TCP_SEQ_STATE_ESTABLISHED: | |
1da177e4 LT |
2265 | rc = established_get_next(seq, v); |
2266 | break; | |
2267 | } | |
2268 | out: | |
2269 | ++*pos; | |
a8b690f9 | 2270 | st->last_pos = *pos; |
1da177e4 LT |
2271 | return rc; |
2272 | } | |
2273 | ||
2274 | static void tcp_seq_stop(struct seq_file *seq, void *v) | |
2275 | { | |
5799de0b | 2276 | struct tcp_iter_state *st = seq->private; |
1da177e4 LT |
2277 | |
2278 | switch (st->state) { | |
2279 | case TCP_SEQ_STATE_OPENREQ: | |
2280 | if (v) { | |
463c84b9 ACM |
2281 | struct inet_connection_sock *icsk = inet_csk(st->syn_wait_sk); |
2282 | read_unlock_bh(&icsk->icsk_accept_queue.syn_wait_lock); | |
1da177e4 LT |
2283 | } |
2284 | case TCP_SEQ_STATE_LISTENING: | |
2285 | if (v != SEQ_START_TOKEN) | |
5caea4ea | 2286 | spin_unlock_bh(&tcp_hashinfo.listening_hash[st->bucket].lock); |
1da177e4 | 2287 | break; |
1da177e4 LT |
2288 | case TCP_SEQ_STATE_ESTABLISHED: |
2289 | if (v) | |
9db66bdc | 2290 | spin_unlock_bh(inet_ehash_lockp(&tcp_hashinfo, st->bucket)); |
1da177e4 LT |
2291 | break; |
2292 | } | |
2293 | } | |
2294 | ||
73cb88ec | 2295 | int tcp_seq_open(struct inode *inode, struct file *file) |
1da177e4 | 2296 | { |
d9dda78b | 2297 | struct tcp_seq_afinfo *afinfo = PDE_DATA(inode); |
1da177e4 | 2298 | struct tcp_iter_state *s; |
52d6f3f1 | 2299 | int err; |
1da177e4 | 2300 | |
52d6f3f1 DL |
2301 | err = seq_open_net(inode, file, &afinfo->seq_ops, |
2302 | sizeof(struct tcp_iter_state)); | |
2303 | if (err < 0) | |
2304 | return err; | |
f40c8174 | 2305 | |
52d6f3f1 | 2306 | s = ((struct seq_file *)file->private_data)->private; |
1da177e4 | 2307 | s->family = afinfo->family; |
a8b690f9 | 2308 | s->last_pos = 0; |
f40c8174 DL |
2309 | return 0; |
2310 | } | |
73cb88ec | 2311 | EXPORT_SYMBOL(tcp_seq_open); |
f40c8174 | 2312 | |
6f8b13bc | 2313 | int tcp_proc_register(struct net *net, struct tcp_seq_afinfo *afinfo) |
1da177e4 LT |
2314 | { |
2315 | int rc = 0; | |
2316 | struct proc_dir_entry *p; | |
2317 | ||
9427c4b3 DL |
2318 | afinfo->seq_ops.start = tcp_seq_start; |
2319 | afinfo->seq_ops.next = tcp_seq_next; | |
2320 | afinfo->seq_ops.stop = tcp_seq_stop; | |
2321 | ||
84841c3c | 2322 | p = proc_create_data(afinfo->name, S_IRUGO, net->proc_net, |
73cb88ec | 2323 | afinfo->seq_fops, afinfo); |
84841c3c | 2324 | if (!p) |
1da177e4 LT |
2325 | rc = -ENOMEM; |
2326 | return rc; | |
2327 | } | |
4bc2f18b | 2328 | EXPORT_SYMBOL(tcp_proc_register); |
1da177e4 | 2329 | |
6f8b13bc | 2330 | void tcp_proc_unregister(struct net *net, struct tcp_seq_afinfo *afinfo) |
1da177e4 | 2331 | { |
ece31ffd | 2332 | remove_proc_entry(afinfo->name, net->proc_net); |
1da177e4 | 2333 | } |
4bc2f18b | 2334 | EXPORT_SYMBOL(tcp_proc_unregister); |
1da177e4 | 2335 | |
cf533ea5 | 2336 | static void get_openreq4(const struct sock *sk, const struct request_sock *req, |
652586df | 2337 | struct seq_file *f, int i, kuid_t uid) |
1da177e4 | 2338 | { |
2e6599cb | 2339 | const struct inet_request_sock *ireq = inet_rsk(req); |
a399a805 | 2340 | long delta = req->expires - jiffies; |
1da177e4 | 2341 | |
5e659e4c | 2342 | seq_printf(f, "%4d: %08X:%04X %08X:%04X" |
652586df | 2343 | " %02X %08X:%08X %02X:%08lX %08X %5u %8d %u %d %pK", |
1da177e4 | 2344 | i, |
634fb979 | 2345 | ireq->ir_loc_addr, |
c720c7e8 | 2346 | ntohs(inet_sk(sk)->inet_sport), |
634fb979 ED |
2347 | ireq->ir_rmt_addr, |
2348 | ntohs(ireq->ir_rmt_port), | |
1da177e4 LT |
2349 | TCP_SYN_RECV, |
2350 | 0, 0, /* could print option size, but that is af dependent. */ | |
2351 | 1, /* timers active (only the expire timer) */ | |
a399a805 | 2352 | jiffies_delta_to_clock_t(delta), |
e6c022a4 | 2353 | req->num_timeout, |
a7cb5a49 | 2354 | from_kuid_munged(seq_user_ns(f), uid), |
1da177e4 LT |
2355 | 0, /* non standard timer */ |
2356 | 0, /* open_requests have no inode */ | |
2357 | atomic_read(&sk->sk_refcnt), | |
652586df | 2358 | req); |
1da177e4 LT |
2359 | } |
2360 | ||
652586df | 2361 | static void get_tcp4_sock(struct sock *sk, struct seq_file *f, int i) |
1da177e4 LT |
2362 | { |
2363 | int timer_active; | |
2364 | unsigned long timer_expires; | |
cf533ea5 | 2365 | const struct tcp_sock *tp = tcp_sk(sk); |
cf4c6bf8 | 2366 | const struct inet_connection_sock *icsk = inet_csk(sk); |
cf533ea5 | 2367 | const struct inet_sock *inet = inet_sk(sk); |
168a8f58 | 2368 | struct fastopen_queue *fastopenq = icsk->icsk_accept_queue.fastopenq; |
c720c7e8 ED |
2369 | __be32 dest = inet->inet_daddr; |
2370 | __be32 src = inet->inet_rcv_saddr; | |
2371 | __u16 destp = ntohs(inet->inet_dport); | |
2372 | __u16 srcp = ntohs(inet->inet_sport); | |
49d09007 | 2373 | int rx_queue; |
1da177e4 | 2374 | |
6ba8a3b1 ND |
2375 | if (icsk->icsk_pending == ICSK_TIME_RETRANS || |
2376 | icsk->icsk_pending == ICSK_TIME_EARLY_RETRANS || | |
2377 | icsk->icsk_pending == ICSK_TIME_LOSS_PROBE) { | |
1da177e4 | 2378 | timer_active = 1; |
463c84b9 ACM |
2379 | timer_expires = icsk->icsk_timeout; |
2380 | } else if (icsk->icsk_pending == ICSK_TIME_PROBE0) { | |
1da177e4 | 2381 | timer_active = 4; |
463c84b9 | 2382 | timer_expires = icsk->icsk_timeout; |
cf4c6bf8 | 2383 | } else if (timer_pending(&sk->sk_timer)) { |
1da177e4 | 2384 | timer_active = 2; |
cf4c6bf8 | 2385 | timer_expires = sk->sk_timer.expires; |
1da177e4 LT |
2386 | } else { |
2387 | timer_active = 0; | |
2388 | timer_expires = jiffies; | |
2389 | } | |
2390 | ||
49d09007 ED |
2391 | if (sk->sk_state == TCP_LISTEN) |
2392 | rx_queue = sk->sk_ack_backlog; | |
2393 | else | |
2394 | /* | |
2395 | * because we dont lock socket, we might find a transient negative value | |
2396 | */ | |
2397 | rx_queue = max_t(int, tp->rcv_nxt - tp->copied_seq, 0); | |
2398 | ||
5e659e4c | 2399 | seq_printf(f, "%4d: %08X:%04X %08X:%04X %02X %08X:%08X %02X:%08lX " |
652586df | 2400 | "%08X %5u %8d %lu %d %pK %lu %lu %u %u %d", |
cf4c6bf8 | 2401 | i, src, srcp, dest, destp, sk->sk_state, |
47da8ee6 | 2402 | tp->write_seq - tp->snd_una, |
49d09007 | 2403 | rx_queue, |
1da177e4 | 2404 | timer_active, |
a399a805 | 2405 | jiffies_delta_to_clock_t(timer_expires - jiffies), |
463c84b9 | 2406 | icsk->icsk_retransmits, |
a7cb5a49 | 2407 | from_kuid_munged(seq_user_ns(f), sock_i_uid(sk)), |
6687e988 | 2408 | icsk->icsk_probes_out, |
cf4c6bf8 IJ |
2409 | sock_i_ino(sk), |
2410 | atomic_read(&sk->sk_refcnt), sk, | |
7be87351 SH |
2411 | jiffies_to_clock_t(icsk->icsk_rto), |
2412 | jiffies_to_clock_t(icsk->icsk_ack.ato), | |
463c84b9 | 2413 | (icsk->icsk_ack.quick << 1) | icsk->icsk_ack.pingpong, |
1da177e4 | 2414 | tp->snd_cwnd, |
168a8f58 JC |
2415 | sk->sk_state == TCP_LISTEN ? |
2416 | (fastopenq ? fastopenq->max_qlen : 0) : | |
652586df | 2417 | (tcp_in_initial_slowstart(tp) ? -1 : tp->snd_ssthresh)); |
1da177e4 LT |
2418 | } |
2419 | ||
cf533ea5 | 2420 | static void get_timewait4_sock(const struct inet_timewait_sock *tw, |
652586df | 2421 | struct seq_file *f, int i) |
1da177e4 | 2422 | { |
23f33c2d | 2423 | __be32 dest, src; |
1da177e4 | 2424 | __u16 destp, srcp; |
e2a1d3e4 | 2425 | s32 delta = tw->tw_ttd - inet_tw_time_stamp(); |
1da177e4 LT |
2426 | |
2427 | dest = tw->tw_daddr; | |
2428 | src = tw->tw_rcv_saddr; | |
2429 | destp = ntohs(tw->tw_dport); | |
2430 | srcp = ntohs(tw->tw_sport); | |
2431 | ||
5e659e4c | 2432 | seq_printf(f, "%4d: %08X:%04X %08X:%04X" |
652586df | 2433 | " %02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %pK", |
1da177e4 | 2434 | i, src, srcp, dest, destp, tw->tw_substate, 0, 0, |
a399a805 | 2435 | 3, jiffies_delta_to_clock_t(delta), 0, 0, 0, 0, |
652586df | 2436 | atomic_read(&tw->tw_refcnt), tw); |
1da177e4 LT |
2437 | } |
2438 | ||
2439 | #define TMPSZ 150 | |
2440 | ||
2441 | static int tcp4_seq_show(struct seq_file *seq, void *v) | |
2442 | { | |
5799de0b | 2443 | struct tcp_iter_state *st; |
05dbc7b5 | 2444 | struct sock *sk = v; |
1da177e4 | 2445 | |
652586df | 2446 | seq_setwidth(seq, TMPSZ - 1); |
1da177e4 | 2447 | if (v == SEQ_START_TOKEN) { |
652586df | 2448 | seq_puts(seq, " sl local_address rem_address st tx_queue " |
1da177e4 LT |
2449 | "rx_queue tr tm->when retrnsmt uid timeout " |
2450 | "inode"); | |
2451 | goto out; | |
2452 | } | |
2453 | st = seq->private; | |
2454 | ||
2455 | switch (st->state) { | |
2456 | case TCP_SEQ_STATE_LISTENING: | |
2457 | case TCP_SEQ_STATE_ESTABLISHED: | |
05dbc7b5 | 2458 | if (sk->sk_state == TCP_TIME_WAIT) |
652586df | 2459 | get_timewait4_sock(v, seq, st->num); |
05dbc7b5 | 2460 | else |
652586df | 2461 | get_tcp4_sock(v, seq, st->num); |
1da177e4 LT |
2462 | break; |
2463 | case TCP_SEQ_STATE_OPENREQ: | |
652586df | 2464 | get_openreq4(st->syn_wait_sk, v, seq, st->num, st->uid); |
1da177e4 LT |
2465 | break; |
2466 | } | |
1da177e4 | 2467 | out: |
652586df | 2468 | seq_pad(seq, '\n'); |
1da177e4 LT |
2469 | return 0; |
2470 | } | |
2471 | ||
73cb88ec AV |
2472 | static const struct file_operations tcp_afinfo_seq_fops = { |
2473 | .owner = THIS_MODULE, | |
2474 | .open = tcp_seq_open, | |
2475 | .read = seq_read, | |
2476 | .llseek = seq_lseek, | |
2477 | .release = seq_release_net | |
2478 | }; | |
2479 | ||
1da177e4 | 2480 | static struct tcp_seq_afinfo tcp4_seq_afinfo = { |
1da177e4 LT |
2481 | .name = "tcp", |
2482 | .family = AF_INET, | |
73cb88ec | 2483 | .seq_fops = &tcp_afinfo_seq_fops, |
9427c4b3 DL |
2484 | .seq_ops = { |
2485 | .show = tcp4_seq_show, | |
2486 | }, | |
1da177e4 LT |
2487 | }; |
2488 | ||
2c8c1e72 | 2489 | static int __net_init tcp4_proc_init_net(struct net *net) |
757764f6 PE |
2490 | { |
2491 | return tcp_proc_register(net, &tcp4_seq_afinfo); | |
2492 | } | |
2493 | ||
2c8c1e72 | 2494 | static void __net_exit tcp4_proc_exit_net(struct net *net) |
757764f6 PE |
2495 | { |
2496 | tcp_proc_unregister(net, &tcp4_seq_afinfo); | |
2497 | } | |
2498 | ||
2499 | static struct pernet_operations tcp4_net_ops = { | |
2500 | .init = tcp4_proc_init_net, | |
2501 | .exit = tcp4_proc_exit_net, | |
2502 | }; | |
2503 | ||
1da177e4 LT |
2504 | int __init tcp4_proc_init(void) |
2505 | { | |
757764f6 | 2506 | return register_pernet_subsys(&tcp4_net_ops); |
1da177e4 LT |
2507 | } |
2508 | ||
2509 | void tcp4_proc_exit(void) | |
2510 | { | |
757764f6 | 2511 | unregister_pernet_subsys(&tcp4_net_ops); |
1da177e4 LT |
2512 | } |
2513 | #endif /* CONFIG_PROC_FS */ | |
2514 | ||
2515 | struct proto tcp_prot = { | |
2516 | .name = "TCP", | |
2517 | .owner = THIS_MODULE, | |
2518 | .close = tcp_close, | |
2519 | .connect = tcp_v4_connect, | |
2520 | .disconnect = tcp_disconnect, | |
463c84b9 | 2521 | .accept = inet_csk_accept, |
1da177e4 LT |
2522 | .ioctl = tcp_ioctl, |
2523 | .init = tcp_v4_init_sock, | |
2524 | .destroy = tcp_v4_destroy_sock, | |
2525 | .shutdown = tcp_shutdown, | |
2526 | .setsockopt = tcp_setsockopt, | |
2527 | .getsockopt = tcp_getsockopt, | |
1da177e4 | 2528 | .recvmsg = tcp_recvmsg, |
7ba42910 CG |
2529 | .sendmsg = tcp_sendmsg, |
2530 | .sendpage = tcp_sendpage, | |
1da177e4 | 2531 | .backlog_rcv = tcp_v4_do_rcv, |
46d3ceab | 2532 | .release_cb = tcp_release_cb, |
563d34d0 | 2533 | .mtu_reduced = tcp_v4_mtu_reduced, |
ab1e0a13 ACM |
2534 | .hash = inet_hash, |
2535 | .unhash = inet_unhash, | |
2536 | .get_port = inet_csk_get_port, | |
1da177e4 | 2537 | .enter_memory_pressure = tcp_enter_memory_pressure, |
c9bee3b7 | 2538 | .stream_memory_free = tcp_stream_memory_free, |
1da177e4 | 2539 | .sockets_allocated = &tcp_sockets_allocated, |
0a5578cf | 2540 | .orphan_count = &tcp_orphan_count, |
1da177e4 LT |
2541 | .memory_allocated = &tcp_memory_allocated, |
2542 | .memory_pressure = &tcp_memory_pressure, | |
a4fe34bf | 2543 | .sysctl_mem = sysctl_tcp_mem, |
1da177e4 LT |
2544 | .sysctl_wmem = sysctl_tcp_wmem, |
2545 | .sysctl_rmem = sysctl_tcp_rmem, | |
2546 | .max_header = MAX_TCP_HEADER, | |
2547 | .obj_size = sizeof(struct tcp_sock), | |
3ab5aee7 | 2548 | .slab_flags = SLAB_DESTROY_BY_RCU, |
6d6ee43e | 2549 | .twsk_prot = &tcp_timewait_sock_ops, |
60236fdd | 2550 | .rsk_prot = &tcp_request_sock_ops, |
39d8cda7 | 2551 | .h.hashinfo = &tcp_hashinfo, |
7ba42910 | 2552 | .no_autobind = true, |
543d9cfe ACM |
2553 | #ifdef CONFIG_COMPAT |
2554 | .compat_setsockopt = compat_tcp_setsockopt, | |
2555 | .compat_getsockopt = compat_tcp_getsockopt, | |
2556 | #endif | |
c255a458 | 2557 | #ifdef CONFIG_MEMCG_KMEM |
d1a4c0b3 GC |
2558 | .init_cgroup = tcp_init_cgroup, |
2559 | .destroy_cgroup = tcp_destroy_cgroup, | |
2560 | .proto_cgroup = tcp_proto_cgroup, | |
2561 | #endif | |
1da177e4 | 2562 | }; |
4bc2f18b | 2563 | EXPORT_SYMBOL(tcp_prot); |
1da177e4 | 2564 | |
046ee902 DL |
2565 | static int __net_init tcp_sk_init(struct net *net) |
2566 | { | |
5d134f1c | 2567 | net->ipv4.sysctl_tcp_ecn = 2; |
be9f4a44 | 2568 | return 0; |
046ee902 DL |
2569 | } |
2570 | ||
2571 | static void __net_exit tcp_sk_exit(struct net *net) | |
2572 | { | |
b099ce26 EB |
2573 | } |
2574 | ||
2575 | static void __net_exit tcp_sk_exit_batch(struct list_head *net_exit_list) | |
2576 | { | |
2577 | inet_twsk_purge(&tcp_hashinfo, &tcp_death_row, AF_INET); | |
046ee902 DL |
2578 | } |
2579 | ||
2580 | static struct pernet_operations __net_initdata tcp_sk_ops = { | |
b099ce26 EB |
2581 | .init = tcp_sk_init, |
2582 | .exit = tcp_sk_exit, | |
2583 | .exit_batch = tcp_sk_exit_batch, | |
046ee902 DL |
2584 | }; |
2585 | ||
9b0f976f | 2586 | void __init tcp_v4_init(void) |
1da177e4 | 2587 | { |
5caea4ea | 2588 | inet_hashinfo_init(&tcp_hashinfo); |
6a1b3054 | 2589 | if (register_pernet_subsys(&tcp_sk_ops)) |
1da177e4 | 2590 | panic("Failed to create the TCP control socket.\n"); |
1da177e4 | 2591 | } |