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> |
1da177e4 LT |
78 | |
79 | #include <linux/inet.h> | |
80 | #include <linux/ipv6.h> | |
81 | #include <linux/stddef.h> | |
82 | #include <linux/proc_fs.h> | |
83 | #include <linux/seq_file.h> | |
84 | ||
cfb6eeb4 YH |
85 | #include <linux/crypto.h> |
86 | #include <linux/scatterlist.h> | |
87 | ||
ab32ea5d BH |
88 | int sysctl_tcp_tw_reuse __read_mostly; |
89 | int sysctl_tcp_low_latency __read_mostly; | |
4bc2f18b | 90 | EXPORT_SYMBOL(sysctl_tcp_low_latency); |
1da177e4 | 91 | |
1da177e4 | 92 | |
cfb6eeb4 | 93 | #ifdef CONFIG_TCP_MD5SIG |
a915da9b | 94 | static int tcp_v4_md5_hash_hdr(char *md5_hash, const struct tcp_md5sig_key *key, |
318cf7aa | 95 | __be32 daddr, __be32 saddr, const struct tcphdr *th); |
cfb6eeb4 YH |
96 | #endif |
97 | ||
5caea4ea | 98 | struct inet_hashinfo tcp_hashinfo; |
4bc2f18b | 99 | EXPORT_SYMBOL(tcp_hashinfo); |
1da177e4 | 100 | |
cf533ea5 | 101 | static inline __u32 tcp_v4_init_sequence(const struct sk_buff *skb) |
1da177e4 | 102 | { |
eddc9ec5 ACM |
103 | return secure_tcp_sequence_number(ip_hdr(skb)->daddr, |
104 | ip_hdr(skb)->saddr, | |
aa8223c7 ACM |
105 | tcp_hdr(skb)->dest, |
106 | tcp_hdr(skb)->source); | |
1da177e4 LT |
107 | } |
108 | ||
6d6ee43e ACM |
109 | int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp) |
110 | { | |
111 | const struct tcp_timewait_sock *tcptw = tcp_twsk(sktw); | |
112 | struct tcp_sock *tp = tcp_sk(sk); | |
113 | ||
114 | /* With PAWS, it is safe from the viewpoint | |
115 | of data integrity. Even without PAWS it is safe provided sequence | |
116 | spaces do not overlap i.e. at data rates <= 80Mbit/sec. | |
117 | ||
118 | Actually, the idea is close to VJ's one, only timestamp cache is | |
119 | held not per host, but per port pair and TW bucket is used as state | |
120 | holder. | |
121 | ||
122 | If TW bucket has been already destroyed we fall back to VJ's scheme | |
123 | and use initial timestamp retrieved from peer table. | |
124 | */ | |
125 | if (tcptw->tw_ts_recent_stamp && | |
126 | (twp == NULL || (sysctl_tcp_tw_reuse && | |
9d729f72 | 127 | get_seconds() - tcptw->tw_ts_recent_stamp > 1))) { |
6d6ee43e ACM |
128 | tp->write_seq = tcptw->tw_snd_nxt + 65535 + 2; |
129 | if (tp->write_seq == 0) | |
130 | tp->write_seq = 1; | |
131 | tp->rx_opt.ts_recent = tcptw->tw_ts_recent; | |
132 | tp->rx_opt.ts_recent_stamp = tcptw->tw_ts_recent_stamp; | |
133 | sock_hold(sktw); | |
134 | return 1; | |
135 | } | |
136 | ||
137 | return 0; | |
138 | } | |
6d6ee43e ACM |
139 | EXPORT_SYMBOL_GPL(tcp_twsk_unique); |
140 | ||
ee995283 PE |
141 | static int tcp_repair_connect(struct sock *sk) |
142 | { | |
143 | tcp_connect_init(sk); | |
144 | tcp_finish_connect(sk, NULL); | |
145 | ||
146 | return 0; | |
147 | } | |
148 | ||
1da177e4 LT |
149 | /* This will initiate an outgoing connection. */ |
150 | int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len) | |
151 | { | |
2d7192d6 | 152 | struct sockaddr_in *usin = (struct sockaddr_in *)uaddr; |
1da177e4 LT |
153 | struct inet_sock *inet = inet_sk(sk); |
154 | struct tcp_sock *tp = tcp_sk(sk); | |
dca8b089 | 155 | __be16 orig_sport, orig_dport; |
bada8adc | 156 | __be32 daddr, nexthop; |
da905bd1 | 157 | struct flowi4 *fl4; |
2d7192d6 | 158 | struct rtable *rt; |
1da177e4 | 159 | int err; |
f6d8bd05 | 160 | struct ip_options_rcu *inet_opt; |
1da177e4 LT |
161 | |
162 | if (addr_len < sizeof(struct sockaddr_in)) | |
163 | return -EINVAL; | |
164 | ||
165 | if (usin->sin_family != AF_INET) | |
166 | return -EAFNOSUPPORT; | |
167 | ||
168 | nexthop = daddr = usin->sin_addr.s_addr; | |
f6d8bd05 ED |
169 | inet_opt = rcu_dereference_protected(inet->inet_opt, |
170 | sock_owned_by_user(sk)); | |
171 | if (inet_opt && inet_opt->opt.srr) { | |
1da177e4 LT |
172 | if (!daddr) |
173 | return -EINVAL; | |
f6d8bd05 | 174 | nexthop = inet_opt->opt.faddr; |
1da177e4 LT |
175 | } |
176 | ||
dca8b089 DM |
177 | orig_sport = inet->inet_sport; |
178 | orig_dport = usin->sin_port; | |
da905bd1 DM |
179 | fl4 = &inet->cork.fl.u.ip4; |
180 | rt = ip_route_connect(fl4, nexthop, inet->inet_saddr, | |
b23dd4fe DM |
181 | RT_CONN_FLAGS(sk), sk->sk_bound_dev_if, |
182 | IPPROTO_TCP, | |
183 | orig_sport, orig_dport, sk, true); | |
184 | if (IS_ERR(rt)) { | |
185 | err = PTR_ERR(rt); | |
186 | if (err == -ENETUNREACH) | |
7c73a6fa | 187 | IP_INC_STATS_BH(sock_net(sk), IPSTATS_MIB_OUTNOROUTES); |
b23dd4fe | 188 | return err; |
584bdf8c | 189 | } |
1da177e4 LT |
190 | |
191 | if (rt->rt_flags & (RTCF_MULTICAST | RTCF_BROADCAST)) { | |
192 | ip_rt_put(rt); | |
193 | return -ENETUNREACH; | |
194 | } | |
195 | ||
f6d8bd05 | 196 | if (!inet_opt || !inet_opt->opt.srr) |
da905bd1 | 197 | daddr = fl4->daddr; |
1da177e4 | 198 | |
c720c7e8 | 199 | if (!inet->inet_saddr) |
da905bd1 | 200 | inet->inet_saddr = fl4->saddr; |
c720c7e8 | 201 | inet->inet_rcv_saddr = inet->inet_saddr; |
1da177e4 | 202 | |
c720c7e8 | 203 | if (tp->rx_opt.ts_recent_stamp && inet->inet_daddr != daddr) { |
1da177e4 LT |
204 | /* Reset inherited state */ |
205 | tp->rx_opt.ts_recent = 0; | |
206 | tp->rx_opt.ts_recent_stamp = 0; | |
ee995283 PE |
207 | if (likely(!tp->repair)) |
208 | tp->write_seq = 0; | |
1da177e4 LT |
209 | } |
210 | ||
295ff7ed | 211 | if (tcp_death_row.sysctl_tw_recycle && |
81166dd6 DM |
212 | !tp->rx_opt.ts_recent_stamp && fl4->daddr == daddr) |
213 | tcp_fetch_timewait_stamp(sk, &rt->dst); | |
1da177e4 | 214 | |
c720c7e8 ED |
215 | inet->inet_dport = usin->sin_port; |
216 | inet->inet_daddr = daddr; | |
1da177e4 | 217 | |
d83d8461 | 218 | inet_csk(sk)->icsk_ext_hdr_len = 0; |
f6d8bd05 ED |
219 | if (inet_opt) |
220 | inet_csk(sk)->icsk_ext_hdr_len = inet_opt->opt.optlen; | |
1da177e4 | 221 | |
bee7ca9e | 222 | tp->rx_opt.mss_clamp = TCP_MSS_DEFAULT; |
1da177e4 LT |
223 | |
224 | /* Socket identity is still unknown (sport may be zero). | |
225 | * However we set state to SYN-SENT and not releasing socket | |
226 | * lock select source port, enter ourselves into the hash tables and | |
227 | * complete initialization after this. | |
228 | */ | |
229 | tcp_set_state(sk, TCP_SYN_SENT); | |
a7f5e7f1 | 230 | err = inet_hash_connect(&tcp_death_row, sk); |
1da177e4 LT |
231 | if (err) |
232 | goto failure; | |
233 | ||
da905bd1 | 234 | rt = ip_route_newports(fl4, rt, orig_sport, orig_dport, |
b23dd4fe DM |
235 | inet->inet_sport, inet->inet_dport, sk); |
236 | if (IS_ERR(rt)) { | |
237 | err = PTR_ERR(rt); | |
238 | rt = NULL; | |
1da177e4 | 239 | goto failure; |
b23dd4fe | 240 | } |
1da177e4 | 241 | /* OK, now commit destination to socket. */ |
bcd76111 | 242 | sk->sk_gso_type = SKB_GSO_TCPV4; |
d8d1f30b | 243 | sk_setup_caps(sk, &rt->dst); |
1da177e4 | 244 | |
ee995283 | 245 | if (!tp->write_seq && likely(!tp->repair)) |
c720c7e8 ED |
246 | tp->write_seq = secure_tcp_sequence_number(inet->inet_saddr, |
247 | inet->inet_daddr, | |
248 | inet->inet_sport, | |
1da177e4 LT |
249 | usin->sin_port); |
250 | ||
c720c7e8 | 251 | inet->inet_id = tp->write_seq ^ jiffies; |
1da177e4 | 252 | |
ee995283 PE |
253 | if (likely(!tp->repair)) |
254 | err = tcp_connect(sk); | |
255 | else | |
256 | err = tcp_repair_connect(sk); | |
257 | ||
1da177e4 LT |
258 | rt = NULL; |
259 | if (err) | |
260 | goto failure; | |
261 | ||
262 | return 0; | |
263 | ||
264 | failure: | |
7174259e ACM |
265 | /* |
266 | * This unhashes the socket and releases the local port, | |
267 | * if necessary. | |
268 | */ | |
1da177e4 LT |
269 | tcp_set_state(sk, TCP_CLOSE); |
270 | ip_rt_put(rt); | |
271 | sk->sk_route_caps = 0; | |
c720c7e8 | 272 | inet->inet_dport = 0; |
1da177e4 LT |
273 | return err; |
274 | } | |
4bc2f18b | 275 | EXPORT_SYMBOL(tcp_v4_connect); |
1da177e4 | 276 | |
1da177e4 LT |
277 | /* |
278 | * This routine does path mtu discovery as defined in RFC1191. | |
279 | */ | |
b71d1d42 | 280 | static void do_pmtu_discovery(struct sock *sk, const struct iphdr *iph, u32 mtu) |
1da177e4 LT |
281 | { |
282 | struct dst_entry *dst; | |
283 | struct inet_sock *inet = inet_sk(sk); | |
1da177e4 LT |
284 | |
285 | /* We are not interested in TCP_LISTEN and open_requests (SYN-ACKs | |
286 | * send out by Linux are always <576bytes so they should go through | |
287 | * unfragmented). | |
288 | */ | |
289 | if (sk->sk_state == TCP_LISTEN) | |
290 | return; | |
291 | ||
292 | /* We don't check in the destentry if pmtu discovery is forbidden | |
293 | * on this route. We just assume that no packet_to_big packets | |
294 | * are send back when pmtu discovery is not active. | |
e905a9ed | 295 | * There is a small race when the user changes this flag in the |
1da177e4 LT |
296 | * route, but I think that's acceptable. |
297 | */ | |
298 | if ((dst = __sk_dst_check(sk, 0)) == NULL) | |
299 | return; | |
300 | ||
301 | dst->ops->update_pmtu(dst, mtu); | |
302 | ||
303 | /* Something is about to be wrong... Remember soft error | |
304 | * for the case, if this connection will not able to recover. | |
305 | */ | |
306 | if (mtu < dst_mtu(dst) && ip_dont_fragment(sk, dst)) | |
307 | sk->sk_err_soft = EMSGSIZE; | |
308 | ||
309 | mtu = dst_mtu(dst); | |
310 | ||
311 | if (inet->pmtudisc != IP_PMTUDISC_DONT && | |
d83d8461 | 312 | inet_csk(sk)->icsk_pmtu_cookie > mtu) { |
1da177e4 LT |
313 | tcp_sync_mss(sk, mtu); |
314 | ||
315 | /* Resend the TCP packet because it's | |
316 | * clear that the old packet has been | |
317 | * dropped. This is the new "fast" path mtu | |
318 | * discovery. | |
319 | */ | |
320 | tcp_simple_retransmit(sk); | |
321 | } /* else let the usual retransmit timer handle it */ | |
322 | } | |
323 | ||
324 | /* | |
325 | * This routine is called by the ICMP module when it gets some | |
326 | * sort of error condition. If err < 0 then the socket should | |
327 | * be closed and the error returned to the user. If err > 0 | |
328 | * it's just the icmp type << 8 | icmp code. After adjustment | |
329 | * header points to the first 8 bytes of the tcp header. We need | |
330 | * to find the appropriate port. | |
331 | * | |
332 | * The locking strategy used here is very "optimistic". When | |
333 | * someone else accesses the socket the ICMP is just dropped | |
334 | * and for some paths there is no check at all. | |
335 | * A more general error queue to queue errors for later handling | |
336 | * is probably better. | |
337 | * | |
338 | */ | |
339 | ||
4d1a2d9e | 340 | void tcp_v4_err(struct sk_buff *icmp_skb, u32 info) |
1da177e4 | 341 | { |
b71d1d42 | 342 | const struct iphdr *iph = (const struct iphdr *)icmp_skb->data; |
4d1a2d9e | 343 | struct tcphdr *th = (struct tcphdr *)(icmp_skb->data + (iph->ihl << 2)); |
f1ecd5d9 | 344 | struct inet_connection_sock *icsk; |
1da177e4 LT |
345 | struct tcp_sock *tp; |
346 | struct inet_sock *inet; | |
4d1a2d9e DL |
347 | const int type = icmp_hdr(icmp_skb)->type; |
348 | const int code = icmp_hdr(icmp_skb)->code; | |
1da177e4 | 349 | struct sock *sk; |
f1ecd5d9 | 350 | struct sk_buff *skb; |
1da177e4 | 351 | __u32 seq; |
f1ecd5d9 | 352 | __u32 remaining; |
1da177e4 | 353 | int err; |
4d1a2d9e | 354 | struct net *net = dev_net(icmp_skb->dev); |
1da177e4 | 355 | |
4d1a2d9e | 356 | if (icmp_skb->len < (iph->ihl << 2) + 8) { |
dcfc23ca | 357 | ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS); |
1da177e4 LT |
358 | return; |
359 | } | |
360 | ||
fd54d716 | 361 | sk = inet_lookup(net, &tcp_hashinfo, iph->daddr, th->dest, |
4d1a2d9e | 362 | iph->saddr, th->source, inet_iif(icmp_skb)); |
1da177e4 | 363 | if (!sk) { |
dcfc23ca | 364 | ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS); |
1da177e4 LT |
365 | return; |
366 | } | |
367 | if (sk->sk_state == TCP_TIME_WAIT) { | |
9469c7b4 | 368 | inet_twsk_put(inet_twsk(sk)); |
1da177e4 LT |
369 | return; |
370 | } | |
371 | ||
372 | bh_lock_sock(sk); | |
373 | /* If too many ICMPs get dropped on busy | |
374 | * servers this needs to be solved differently. | |
375 | */ | |
376 | if (sock_owned_by_user(sk)) | |
de0744af | 377 | NET_INC_STATS_BH(net, LINUX_MIB_LOCKDROPPEDICMPS); |
1da177e4 LT |
378 | |
379 | if (sk->sk_state == TCP_CLOSE) | |
380 | goto out; | |
381 | ||
97e3ecd1 | 382 | if (unlikely(iph->ttl < inet_sk(sk)->min_ttl)) { |
383 | NET_INC_STATS_BH(net, LINUX_MIB_TCPMINTTLDROP); | |
384 | goto out; | |
385 | } | |
386 | ||
f1ecd5d9 | 387 | icsk = inet_csk(sk); |
1da177e4 LT |
388 | tp = tcp_sk(sk); |
389 | seq = ntohl(th->seq); | |
390 | if (sk->sk_state != TCP_LISTEN && | |
391 | !between(seq, tp->snd_una, tp->snd_nxt)) { | |
de0744af | 392 | NET_INC_STATS_BH(net, LINUX_MIB_OUTOFWINDOWICMPS); |
1da177e4 LT |
393 | goto out; |
394 | } | |
395 | ||
396 | switch (type) { | |
397 | case ICMP_SOURCE_QUENCH: | |
398 | /* Just silently ignore these. */ | |
399 | goto out; | |
400 | case ICMP_PARAMETERPROB: | |
401 | err = EPROTO; | |
402 | break; | |
403 | case ICMP_DEST_UNREACH: | |
404 | if (code > NR_ICMP_UNREACH) | |
405 | goto out; | |
406 | ||
407 | if (code == ICMP_FRAG_NEEDED) { /* PMTU discovery (RFC1191) */ | |
408 | if (!sock_owned_by_user(sk)) | |
409 | do_pmtu_discovery(sk, iph, info); | |
410 | goto out; | |
411 | } | |
412 | ||
413 | err = icmp_err_convert[code].errno; | |
f1ecd5d9 DL |
414 | /* check if icmp_skb allows revert of backoff |
415 | * (see draft-zimmermann-tcp-lcd) */ | |
416 | if (code != ICMP_NET_UNREACH && code != ICMP_HOST_UNREACH) | |
417 | break; | |
418 | if (seq != tp->snd_una || !icsk->icsk_retransmits || | |
419 | !icsk->icsk_backoff) | |
420 | break; | |
421 | ||
8f49c270 DM |
422 | if (sock_owned_by_user(sk)) |
423 | break; | |
424 | ||
f1ecd5d9 | 425 | icsk->icsk_backoff--; |
9ad7c049 JC |
426 | inet_csk(sk)->icsk_rto = (tp->srtt ? __tcp_set_rto(tp) : |
427 | TCP_TIMEOUT_INIT) << icsk->icsk_backoff; | |
f1ecd5d9 DL |
428 | tcp_bound_rto(sk); |
429 | ||
430 | skb = tcp_write_queue_head(sk); | |
431 | BUG_ON(!skb); | |
432 | ||
433 | remaining = icsk->icsk_rto - min(icsk->icsk_rto, | |
434 | tcp_time_stamp - TCP_SKB_CB(skb)->when); | |
435 | ||
436 | if (remaining) { | |
437 | inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS, | |
438 | remaining, TCP_RTO_MAX); | |
f1ecd5d9 DL |
439 | } else { |
440 | /* RTO revert clocked out retransmission. | |
441 | * Will retransmit now */ | |
442 | tcp_retransmit_timer(sk); | |
443 | } | |
444 | ||
1da177e4 LT |
445 | break; |
446 | case ICMP_TIME_EXCEEDED: | |
447 | err = EHOSTUNREACH; | |
448 | break; | |
449 | default: | |
450 | goto out; | |
451 | } | |
452 | ||
453 | switch (sk->sk_state) { | |
60236fdd | 454 | struct request_sock *req, **prev; |
1da177e4 LT |
455 | case TCP_LISTEN: |
456 | if (sock_owned_by_user(sk)) | |
457 | goto out; | |
458 | ||
463c84b9 ACM |
459 | req = inet_csk_search_req(sk, &prev, th->dest, |
460 | iph->daddr, iph->saddr); | |
1da177e4 LT |
461 | if (!req) |
462 | goto out; | |
463 | ||
464 | /* ICMPs are not backlogged, hence we cannot get | |
465 | an established socket here. | |
466 | */ | |
547b792c | 467 | WARN_ON(req->sk); |
1da177e4 | 468 | |
2e6599cb | 469 | if (seq != tcp_rsk(req)->snt_isn) { |
de0744af | 470 | NET_INC_STATS_BH(net, LINUX_MIB_OUTOFWINDOWICMPS); |
1da177e4 LT |
471 | goto out; |
472 | } | |
473 | ||
474 | /* | |
475 | * Still in SYN_RECV, just remove it silently. | |
476 | * There is no good way to pass the error to the newly | |
477 | * created socket, and POSIX does not want network | |
478 | * errors returned from accept(). | |
479 | */ | |
463c84b9 | 480 | inet_csk_reqsk_queue_drop(sk, req, prev); |
1da177e4 LT |
481 | goto out; |
482 | ||
483 | case TCP_SYN_SENT: | |
484 | case TCP_SYN_RECV: /* Cannot happen. | |
485 | It can f.e. if SYNs crossed. | |
486 | */ | |
487 | if (!sock_owned_by_user(sk)) { | |
1da177e4 LT |
488 | sk->sk_err = err; |
489 | ||
490 | sk->sk_error_report(sk); | |
491 | ||
492 | tcp_done(sk); | |
493 | } else { | |
494 | sk->sk_err_soft = err; | |
495 | } | |
496 | goto out; | |
497 | } | |
498 | ||
499 | /* If we've already connected we will keep trying | |
500 | * until we time out, or the user gives up. | |
501 | * | |
502 | * rfc1122 4.2.3.9 allows to consider as hard errors | |
503 | * only PROTO_UNREACH and PORT_UNREACH (well, FRAG_FAILED too, | |
504 | * but it is obsoleted by pmtu discovery). | |
505 | * | |
506 | * Note, that in modern internet, where routing is unreliable | |
507 | * and in each dark corner broken firewalls sit, sending random | |
508 | * errors ordered by their masters even this two messages finally lose | |
509 | * their original sense (even Linux sends invalid PORT_UNREACHs) | |
510 | * | |
511 | * Now we are in compliance with RFCs. | |
512 | * --ANK (980905) | |
513 | */ | |
514 | ||
515 | inet = inet_sk(sk); | |
516 | if (!sock_owned_by_user(sk) && inet->recverr) { | |
517 | sk->sk_err = err; | |
518 | sk->sk_error_report(sk); | |
519 | } else { /* Only an error on timeout */ | |
520 | sk->sk_err_soft = err; | |
521 | } | |
522 | ||
523 | out: | |
524 | bh_unlock_sock(sk); | |
525 | sock_put(sk); | |
526 | } | |
527 | ||
419f9f89 HX |
528 | static void __tcp_v4_send_check(struct sk_buff *skb, |
529 | __be32 saddr, __be32 daddr) | |
1da177e4 | 530 | { |
aa8223c7 | 531 | struct tcphdr *th = tcp_hdr(skb); |
1da177e4 | 532 | |
84fa7933 | 533 | if (skb->ip_summed == CHECKSUM_PARTIAL) { |
419f9f89 | 534 | th->check = ~tcp_v4_check(skb->len, saddr, daddr, 0); |
663ead3b | 535 | skb->csum_start = skb_transport_header(skb) - skb->head; |
ff1dcadb | 536 | skb->csum_offset = offsetof(struct tcphdr, check); |
1da177e4 | 537 | } else { |
419f9f89 | 538 | th->check = tcp_v4_check(skb->len, saddr, daddr, |
07f0757a | 539 | csum_partial(th, |
1da177e4 LT |
540 | th->doff << 2, |
541 | skb->csum)); | |
542 | } | |
543 | } | |
544 | ||
419f9f89 | 545 | /* This routine computes an IPv4 TCP checksum. */ |
bb296246 | 546 | void tcp_v4_send_check(struct sock *sk, struct sk_buff *skb) |
419f9f89 | 547 | { |
cf533ea5 | 548 | const struct inet_sock *inet = inet_sk(sk); |
419f9f89 HX |
549 | |
550 | __tcp_v4_send_check(skb, inet->inet_saddr, inet->inet_daddr); | |
551 | } | |
4bc2f18b | 552 | EXPORT_SYMBOL(tcp_v4_send_check); |
419f9f89 | 553 | |
a430a43d HX |
554 | int tcp_v4_gso_send_check(struct sk_buff *skb) |
555 | { | |
eddc9ec5 | 556 | const struct iphdr *iph; |
a430a43d HX |
557 | struct tcphdr *th; |
558 | ||
559 | if (!pskb_may_pull(skb, sizeof(*th))) | |
560 | return -EINVAL; | |
561 | ||
eddc9ec5 | 562 | iph = ip_hdr(skb); |
aa8223c7 | 563 | th = tcp_hdr(skb); |
a430a43d HX |
564 | |
565 | th->check = 0; | |
84fa7933 | 566 | skb->ip_summed = CHECKSUM_PARTIAL; |
419f9f89 | 567 | __tcp_v4_send_check(skb, iph->saddr, iph->daddr); |
a430a43d HX |
568 | return 0; |
569 | } | |
570 | ||
1da177e4 LT |
571 | /* |
572 | * This routine will send an RST to the other tcp. | |
573 | * | |
574 | * Someone asks: why I NEVER use socket parameters (TOS, TTL etc.) | |
575 | * for reset. | |
576 | * Answer: if a packet caused RST, it is not for a socket | |
577 | * existing in our system, if it is matched to a socket, | |
578 | * it is just duplicate segment or bug in other side's TCP. | |
579 | * So that we build reply only basing on parameters | |
580 | * arrived with segment. | |
581 | * Exception: precedence violation. We do not implement it in any case. | |
582 | */ | |
583 | ||
cfb6eeb4 | 584 | static void tcp_v4_send_reset(struct sock *sk, struct sk_buff *skb) |
1da177e4 | 585 | { |
cf533ea5 | 586 | const struct tcphdr *th = tcp_hdr(skb); |
cfb6eeb4 YH |
587 | struct { |
588 | struct tcphdr th; | |
589 | #ifdef CONFIG_TCP_MD5SIG | |
714e85be | 590 | __be32 opt[(TCPOLEN_MD5SIG_ALIGNED >> 2)]; |
cfb6eeb4 YH |
591 | #endif |
592 | } rep; | |
1da177e4 | 593 | struct ip_reply_arg arg; |
cfb6eeb4 YH |
594 | #ifdef CONFIG_TCP_MD5SIG |
595 | struct tcp_md5sig_key *key; | |
658ddaaf SL |
596 | const __u8 *hash_location = NULL; |
597 | unsigned char newhash[16]; | |
598 | int genhash; | |
599 | struct sock *sk1 = NULL; | |
cfb6eeb4 | 600 | #endif |
a86b1e30 | 601 | struct net *net; |
1da177e4 LT |
602 | |
603 | /* Never send a reset in response to a reset. */ | |
604 | if (th->rst) | |
605 | return; | |
606 | ||
511c3f92 | 607 | if (skb_rtable(skb)->rt_type != RTN_LOCAL) |
1da177e4 LT |
608 | return; |
609 | ||
610 | /* Swap the send and the receive. */ | |
cfb6eeb4 YH |
611 | memset(&rep, 0, sizeof(rep)); |
612 | rep.th.dest = th->source; | |
613 | rep.th.source = th->dest; | |
614 | rep.th.doff = sizeof(struct tcphdr) / 4; | |
615 | rep.th.rst = 1; | |
1da177e4 LT |
616 | |
617 | if (th->ack) { | |
cfb6eeb4 | 618 | rep.th.seq = th->ack_seq; |
1da177e4 | 619 | } else { |
cfb6eeb4 YH |
620 | rep.th.ack = 1; |
621 | rep.th.ack_seq = htonl(ntohl(th->seq) + th->syn + th->fin + | |
622 | skb->len - (th->doff << 2)); | |
1da177e4 LT |
623 | } |
624 | ||
7174259e | 625 | memset(&arg, 0, sizeof(arg)); |
cfb6eeb4 YH |
626 | arg.iov[0].iov_base = (unsigned char *)&rep; |
627 | arg.iov[0].iov_len = sizeof(rep.th); | |
628 | ||
629 | #ifdef CONFIG_TCP_MD5SIG | |
658ddaaf SL |
630 | hash_location = tcp_parse_md5sig_option(th); |
631 | if (!sk && hash_location) { | |
632 | /* | |
633 | * active side is lost. Try to find listening socket through | |
634 | * source port, and then find md5 key through listening socket. | |
635 | * we are not loose security here: | |
636 | * Incoming packet is checked with md5 hash with finding key, | |
637 | * no RST generated if md5 hash doesn't match. | |
638 | */ | |
639 | sk1 = __inet_lookup_listener(dev_net(skb_dst(skb)->dev), | |
640 | &tcp_hashinfo, ip_hdr(skb)->daddr, | |
641 | ntohs(th->source), inet_iif(skb)); | |
642 | /* don't send rst if it can't find key */ | |
643 | if (!sk1) | |
644 | return; | |
645 | rcu_read_lock(); | |
646 | key = tcp_md5_do_lookup(sk1, (union tcp_md5_addr *) | |
647 | &ip_hdr(skb)->saddr, AF_INET); | |
648 | if (!key) | |
649 | goto release_sk1; | |
650 | ||
651 | genhash = tcp_v4_md5_hash_skb(newhash, key, NULL, NULL, skb); | |
652 | if (genhash || memcmp(hash_location, newhash, 16) != 0) | |
653 | goto release_sk1; | |
654 | } else { | |
655 | key = sk ? tcp_md5_do_lookup(sk, (union tcp_md5_addr *) | |
656 | &ip_hdr(skb)->saddr, | |
657 | AF_INET) : NULL; | |
658 | } | |
659 | ||
cfb6eeb4 YH |
660 | if (key) { |
661 | rep.opt[0] = htonl((TCPOPT_NOP << 24) | | |
662 | (TCPOPT_NOP << 16) | | |
663 | (TCPOPT_MD5SIG << 8) | | |
664 | TCPOLEN_MD5SIG); | |
665 | /* Update length and the length the header thinks exists */ | |
666 | arg.iov[0].iov_len += TCPOLEN_MD5SIG_ALIGNED; | |
667 | rep.th.doff = arg.iov[0].iov_len / 4; | |
668 | ||
49a72dfb | 669 | tcp_v4_md5_hash_hdr((__u8 *) &rep.opt[1], |
78e645cb IJ |
670 | key, ip_hdr(skb)->saddr, |
671 | ip_hdr(skb)->daddr, &rep.th); | |
cfb6eeb4 YH |
672 | } |
673 | #endif | |
eddc9ec5 ACM |
674 | arg.csum = csum_tcpudp_nofold(ip_hdr(skb)->daddr, |
675 | ip_hdr(skb)->saddr, /* XXX */ | |
52cd5750 | 676 | arg.iov[0].iov_len, IPPROTO_TCP, 0); |
1da177e4 | 677 | arg.csumoffset = offsetof(struct tcphdr, check) / 2; |
88ef4a5a | 678 | arg.flags = (sk && inet_sk(sk)->transparent) ? IP_REPLY_ARG_NOSRCCHECK : 0; |
e2446eaa SL |
679 | /* When socket is gone, all binding information is lost. |
680 | * routing might fail in this case. using iif for oif to | |
681 | * make sure we can deliver it | |
682 | */ | |
683 | arg.bound_dev_if = sk ? sk->sk_bound_dev_if : inet_iif(skb); | |
1da177e4 | 684 | |
adf30907 | 685 | net = dev_net(skb_dst(skb)->dev); |
66b13d99 | 686 | arg.tos = ip_hdr(skb)->tos; |
70e73416 DM |
687 | ip_send_unicast_reply(net->ipv4.tcp_sock, skb, ip_hdr(skb)->saddr, |
688 | ip_hdr(skb)->daddr, &arg, arg.iov[0].iov_len); | |
1da177e4 | 689 | |
63231bdd PE |
690 | TCP_INC_STATS_BH(net, TCP_MIB_OUTSEGS); |
691 | TCP_INC_STATS_BH(net, TCP_MIB_OUTRSTS); | |
658ddaaf SL |
692 | |
693 | #ifdef CONFIG_TCP_MD5SIG | |
694 | release_sk1: | |
695 | if (sk1) { | |
696 | rcu_read_unlock(); | |
697 | sock_put(sk1); | |
698 | } | |
699 | #endif | |
1da177e4 LT |
700 | } |
701 | ||
702 | /* The code following below sending ACKs in SYN-RECV and TIME-WAIT states | |
703 | outside socket context is ugly, certainly. What can I do? | |
704 | */ | |
705 | ||
9501f972 YH |
706 | static void tcp_v4_send_ack(struct sk_buff *skb, u32 seq, u32 ack, |
707 | u32 win, u32 ts, int oif, | |
88ef4a5a | 708 | struct tcp_md5sig_key *key, |
66b13d99 | 709 | int reply_flags, u8 tos) |
1da177e4 | 710 | { |
cf533ea5 | 711 | const struct tcphdr *th = tcp_hdr(skb); |
1da177e4 LT |
712 | struct { |
713 | struct tcphdr th; | |
714e85be | 714 | __be32 opt[(TCPOLEN_TSTAMP_ALIGNED >> 2) |
cfb6eeb4 | 715 | #ifdef CONFIG_TCP_MD5SIG |
714e85be | 716 | + (TCPOLEN_MD5SIG_ALIGNED >> 2) |
cfb6eeb4 YH |
717 | #endif |
718 | ]; | |
1da177e4 LT |
719 | } rep; |
720 | struct ip_reply_arg arg; | |
adf30907 | 721 | struct net *net = dev_net(skb_dst(skb)->dev); |
1da177e4 LT |
722 | |
723 | memset(&rep.th, 0, sizeof(struct tcphdr)); | |
7174259e | 724 | memset(&arg, 0, sizeof(arg)); |
1da177e4 LT |
725 | |
726 | arg.iov[0].iov_base = (unsigned char *)&rep; | |
727 | arg.iov[0].iov_len = sizeof(rep.th); | |
728 | if (ts) { | |
cfb6eeb4 YH |
729 | rep.opt[0] = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) | |
730 | (TCPOPT_TIMESTAMP << 8) | | |
731 | TCPOLEN_TIMESTAMP); | |
732 | rep.opt[1] = htonl(tcp_time_stamp); | |
733 | rep.opt[2] = htonl(ts); | |
cb48cfe8 | 734 | arg.iov[0].iov_len += TCPOLEN_TSTAMP_ALIGNED; |
1da177e4 LT |
735 | } |
736 | ||
737 | /* Swap the send and the receive. */ | |
738 | rep.th.dest = th->source; | |
739 | rep.th.source = th->dest; | |
740 | rep.th.doff = arg.iov[0].iov_len / 4; | |
741 | rep.th.seq = htonl(seq); | |
742 | rep.th.ack_seq = htonl(ack); | |
743 | rep.th.ack = 1; | |
744 | rep.th.window = htons(win); | |
745 | ||
cfb6eeb4 | 746 | #ifdef CONFIG_TCP_MD5SIG |
cfb6eeb4 YH |
747 | if (key) { |
748 | int offset = (ts) ? 3 : 0; | |
749 | ||
750 | rep.opt[offset++] = htonl((TCPOPT_NOP << 24) | | |
751 | (TCPOPT_NOP << 16) | | |
752 | (TCPOPT_MD5SIG << 8) | | |
753 | TCPOLEN_MD5SIG); | |
754 | arg.iov[0].iov_len += TCPOLEN_MD5SIG_ALIGNED; | |
755 | rep.th.doff = arg.iov[0].iov_len/4; | |
756 | ||
49a72dfb | 757 | tcp_v4_md5_hash_hdr((__u8 *) &rep.opt[offset], |
90b7e112 AL |
758 | key, ip_hdr(skb)->saddr, |
759 | ip_hdr(skb)->daddr, &rep.th); | |
cfb6eeb4 YH |
760 | } |
761 | #endif | |
88ef4a5a | 762 | arg.flags = reply_flags; |
eddc9ec5 ACM |
763 | arg.csum = csum_tcpudp_nofold(ip_hdr(skb)->daddr, |
764 | ip_hdr(skb)->saddr, /* XXX */ | |
1da177e4 LT |
765 | arg.iov[0].iov_len, IPPROTO_TCP, 0); |
766 | arg.csumoffset = offsetof(struct tcphdr, check) / 2; | |
9501f972 YH |
767 | if (oif) |
768 | arg.bound_dev_if = oif; | |
66b13d99 | 769 | arg.tos = tos; |
70e73416 DM |
770 | ip_send_unicast_reply(net->ipv4.tcp_sock, skb, ip_hdr(skb)->saddr, |
771 | ip_hdr(skb)->daddr, &arg, arg.iov[0].iov_len); | |
1da177e4 | 772 | |
63231bdd | 773 | TCP_INC_STATS_BH(net, TCP_MIB_OUTSEGS); |
1da177e4 LT |
774 | } |
775 | ||
776 | static void tcp_v4_timewait_ack(struct sock *sk, struct sk_buff *skb) | |
777 | { | |
8feaf0c0 | 778 | struct inet_timewait_sock *tw = inet_twsk(sk); |
cfb6eeb4 | 779 | struct tcp_timewait_sock *tcptw = tcp_twsk(sk); |
1da177e4 | 780 | |
9501f972 | 781 | tcp_v4_send_ack(skb, tcptw->tw_snd_nxt, tcptw->tw_rcv_nxt, |
7174259e | 782 | tcptw->tw_rcv_wnd >> tw->tw_rcv_wscale, |
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 | { |
9501f972 | 796 | tcp_v4_send_ack(skb, tcp_rsk(req)->snt_isn + 1, |
cfb6eeb4 | 797 | tcp_rsk(req)->rcv_isn + 1, req->rcv_wnd, |
9501f972 YH |
798 | req->ts_recent, |
799 | 0, | |
a915da9b ED |
800 | tcp_md5_do_lookup(sk, (union tcp_md5_addr *)&ip_hdr(skb)->daddr, |
801 | AF_INET), | |
66b13d99 ED |
802 | inet_rsk(req)->no_srccheck ? IP_REPLY_ARG_NOSRCCHECK : 0, |
803 | ip_hdr(skb)->tos); | |
1da177e4 LT |
804 | } |
805 | ||
1da177e4 | 806 | /* |
9bf1d83e | 807 | * Send a SYN-ACK after having received a SYN. |
60236fdd | 808 | * This still operates on a request_sock only, not on a big |
1da177e4 LT |
809 | * socket. |
810 | */ | |
72659ecc OP |
811 | static int tcp_v4_send_synack(struct sock *sk, struct dst_entry *dst, |
812 | struct request_sock *req, | |
fff32699 | 813 | struct request_values *rvp, |
7586eceb ED |
814 | u16 queue_mapping, |
815 | bool nocache) | |
1da177e4 | 816 | { |
2e6599cb | 817 | const struct inet_request_sock *ireq = inet_rsk(req); |
6bd023f3 | 818 | struct flowi4 fl4; |
1da177e4 LT |
819 | int err = -1; |
820 | struct sk_buff * skb; | |
821 | ||
822 | /* First, grab a route. */ | |
7586eceb | 823 | if (!dst && (dst = inet_csk_route_req(sk, &fl4, req, nocache)) == NULL) |
fd80eb94 | 824 | return -1; |
1da177e4 | 825 | |
e6b4d113 | 826 | skb = tcp_make_synack(sk, dst, req, rvp); |
1da177e4 LT |
827 | |
828 | if (skb) { | |
419f9f89 | 829 | __tcp_v4_send_check(skb, ireq->loc_addr, ireq->rmt_addr); |
1da177e4 | 830 | |
fff32699 | 831 | skb_set_queue_mapping(skb, queue_mapping); |
2e6599cb ACM |
832 | err = ip_build_and_send_pkt(skb, sk, ireq->loc_addr, |
833 | ireq->rmt_addr, | |
834 | ireq->opt); | |
b9df3cb8 | 835 | err = net_xmit_eval(err); |
1da177e4 LT |
836 | } |
837 | ||
1da177e4 LT |
838 | return err; |
839 | } | |
840 | ||
72659ecc | 841 | static int tcp_v4_rtx_synack(struct sock *sk, struct request_sock *req, |
e6b4d113 | 842 | struct request_values *rvp) |
fd80eb94 | 843 | { |
72659ecc | 844 | TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_RETRANSSEGS); |
7586eceb | 845 | return tcp_v4_send_synack(sk, NULL, req, rvp, 0, false); |
fd80eb94 DL |
846 | } |
847 | ||
1da177e4 | 848 | /* |
60236fdd | 849 | * IPv4 request_sock destructor. |
1da177e4 | 850 | */ |
60236fdd | 851 | static void tcp_v4_reqsk_destructor(struct request_sock *req) |
1da177e4 | 852 | { |
a51482bd | 853 | kfree(inet_rsk(req)->opt); |
1da177e4 LT |
854 | } |
855 | ||
946cedcc | 856 | /* |
a2a385d6 | 857 | * Return true if a syncookie should be sent |
946cedcc | 858 | */ |
a2a385d6 | 859 | bool tcp_syn_flood_action(struct sock *sk, |
946cedcc ED |
860 | const struct sk_buff *skb, |
861 | const char *proto) | |
1da177e4 | 862 | { |
946cedcc | 863 | const char *msg = "Dropping request"; |
a2a385d6 | 864 | bool want_cookie = false; |
946cedcc ED |
865 | struct listen_sock *lopt; |
866 | ||
867 | ||
1da177e4 | 868 | |
2a1d4bd4 | 869 | #ifdef CONFIG_SYN_COOKIES |
946cedcc | 870 | if (sysctl_tcp_syncookies) { |
2a1d4bd4 | 871 | msg = "Sending cookies"; |
a2a385d6 | 872 | want_cookie = true; |
946cedcc ED |
873 | NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPREQQFULLDOCOOKIES); |
874 | } else | |
80e40daa | 875 | #endif |
946cedcc ED |
876 | NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPREQQFULLDROP); |
877 | ||
878 | lopt = inet_csk(sk)->icsk_accept_queue.listen_opt; | |
879 | if (!lopt->synflood_warned) { | |
880 | lopt->synflood_warned = 1; | |
afd46503 | 881 | pr_info("%s: Possible SYN flooding on port %d. %s. Check SNMP counters.\n", |
946cedcc ED |
882 | proto, ntohs(tcp_hdr(skb)->dest), msg); |
883 | } | |
884 | return want_cookie; | |
2a1d4bd4 | 885 | } |
946cedcc | 886 | EXPORT_SYMBOL(tcp_syn_flood_action); |
1da177e4 LT |
887 | |
888 | /* | |
60236fdd | 889 | * Save and compile IPv4 options into the request_sock if needed. |
1da177e4 | 890 | */ |
f6d8bd05 ED |
891 | static struct ip_options_rcu *tcp_v4_save_options(struct sock *sk, |
892 | struct sk_buff *skb) | |
1da177e4 | 893 | { |
f6d8bd05 ED |
894 | const struct ip_options *opt = &(IPCB(skb)->opt); |
895 | struct ip_options_rcu *dopt = NULL; | |
1da177e4 LT |
896 | |
897 | if (opt && opt->optlen) { | |
f6d8bd05 ED |
898 | int opt_size = sizeof(*dopt) + opt->optlen; |
899 | ||
1da177e4 LT |
900 | dopt = kmalloc(opt_size, GFP_ATOMIC); |
901 | if (dopt) { | |
f6d8bd05 | 902 | if (ip_options_echo(&dopt->opt, skb)) { |
1da177e4 LT |
903 | kfree(dopt); |
904 | dopt = NULL; | |
905 | } | |
906 | } | |
907 | } | |
908 | return dopt; | |
909 | } | |
910 | ||
cfb6eeb4 YH |
911 | #ifdef CONFIG_TCP_MD5SIG |
912 | /* | |
913 | * RFC2385 MD5 checksumming requires a mapping of | |
914 | * IP address->MD5 Key. | |
915 | * We need to maintain these in the sk structure. | |
916 | */ | |
917 | ||
918 | /* Find the Key structure for an address. */ | |
a915da9b ED |
919 | struct tcp_md5sig_key *tcp_md5_do_lookup(struct sock *sk, |
920 | const union tcp_md5_addr *addr, | |
921 | int family) | |
cfb6eeb4 YH |
922 | { |
923 | struct tcp_sock *tp = tcp_sk(sk); | |
a915da9b ED |
924 | struct tcp_md5sig_key *key; |
925 | struct hlist_node *pos; | |
926 | unsigned int size = sizeof(struct in_addr); | |
a8afca03 | 927 | struct tcp_md5sig_info *md5sig; |
cfb6eeb4 | 928 | |
a8afca03 ED |
929 | /* caller either holds rcu_read_lock() or socket lock */ |
930 | md5sig = rcu_dereference_check(tp->md5sig_info, | |
b4fb05ea ED |
931 | sock_owned_by_user(sk) || |
932 | lockdep_is_held(&sk->sk_lock.slock)); | |
a8afca03 | 933 | if (!md5sig) |
cfb6eeb4 | 934 | return NULL; |
a915da9b ED |
935 | #if IS_ENABLED(CONFIG_IPV6) |
936 | if (family == AF_INET6) | |
937 | size = sizeof(struct in6_addr); | |
938 | #endif | |
a8afca03 | 939 | hlist_for_each_entry_rcu(key, pos, &md5sig->head, node) { |
a915da9b ED |
940 | if (key->family != family) |
941 | continue; | |
942 | if (!memcmp(&key->addr, addr, size)) | |
943 | return key; | |
cfb6eeb4 YH |
944 | } |
945 | return NULL; | |
946 | } | |
a915da9b | 947 | EXPORT_SYMBOL(tcp_md5_do_lookup); |
cfb6eeb4 YH |
948 | |
949 | struct tcp_md5sig_key *tcp_v4_md5_lookup(struct sock *sk, | |
950 | struct sock *addr_sk) | |
951 | { | |
a915da9b ED |
952 | union tcp_md5_addr *addr; |
953 | ||
954 | addr = (union tcp_md5_addr *)&inet_sk(addr_sk)->inet_daddr; | |
955 | return tcp_md5_do_lookup(sk, addr, AF_INET); | |
cfb6eeb4 | 956 | } |
cfb6eeb4 YH |
957 | EXPORT_SYMBOL(tcp_v4_md5_lookup); |
958 | ||
f5b99bcd AB |
959 | static struct tcp_md5sig_key *tcp_v4_reqsk_md5_lookup(struct sock *sk, |
960 | struct request_sock *req) | |
cfb6eeb4 | 961 | { |
a915da9b ED |
962 | union tcp_md5_addr *addr; |
963 | ||
964 | addr = (union tcp_md5_addr *)&inet_rsk(req)->rmt_addr; | |
965 | return tcp_md5_do_lookup(sk, addr, AF_INET); | |
cfb6eeb4 YH |
966 | } |
967 | ||
968 | /* This can be called on a newly created socket, from other files */ | |
a915da9b ED |
969 | int tcp_md5_do_add(struct sock *sk, const union tcp_md5_addr *addr, |
970 | int family, const u8 *newkey, u8 newkeylen, gfp_t gfp) | |
cfb6eeb4 YH |
971 | { |
972 | /* Add Key to the list */ | |
b0a713e9 | 973 | struct tcp_md5sig_key *key; |
cfb6eeb4 | 974 | struct tcp_sock *tp = tcp_sk(sk); |
a915da9b | 975 | struct tcp_md5sig_info *md5sig; |
cfb6eeb4 | 976 | |
a915da9b | 977 | key = tcp_md5_do_lookup(sk, (union tcp_md5_addr *)&addr, AF_INET); |
cfb6eeb4 YH |
978 | if (key) { |
979 | /* Pre-existing entry - just update that one. */ | |
a915da9b | 980 | memcpy(key->key, newkey, newkeylen); |
b0a713e9 | 981 | key->keylen = newkeylen; |
a915da9b ED |
982 | return 0; |
983 | } | |
260fcbeb | 984 | |
a8afca03 ED |
985 | md5sig = rcu_dereference_protected(tp->md5sig_info, |
986 | sock_owned_by_user(sk)); | |
a915da9b ED |
987 | if (!md5sig) { |
988 | md5sig = kmalloc(sizeof(*md5sig), gfp); | |
989 | if (!md5sig) | |
cfb6eeb4 | 990 | return -ENOMEM; |
cfb6eeb4 | 991 | |
a915da9b ED |
992 | sk_nocaps_add(sk, NETIF_F_GSO_MASK); |
993 | INIT_HLIST_HEAD(&md5sig->head); | |
a8afca03 | 994 | rcu_assign_pointer(tp->md5sig_info, md5sig); |
a915da9b | 995 | } |
cfb6eeb4 | 996 | |
5f3d9cb2 | 997 | key = sock_kmalloc(sk, sizeof(*key), gfp); |
a915da9b ED |
998 | if (!key) |
999 | return -ENOMEM; | |
1000 | if (hlist_empty(&md5sig->head) && !tcp_alloc_md5sig_pool(sk)) { | |
5f3d9cb2 | 1001 | sock_kfree_s(sk, key, sizeof(*key)); |
a915da9b | 1002 | return -ENOMEM; |
cfb6eeb4 | 1003 | } |
a915da9b ED |
1004 | |
1005 | memcpy(key->key, newkey, newkeylen); | |
1006 | key->keylen = newkeylen; | |
1007 | key->family = family; | |
1008 | memcpy(&key->addr, addr, | |
1009 | (family == AF_INET6) ? sizeof(struct in6_addr) : | |
1010 | sizeof(struct in_addr)); | |
1011 | hlist_add_head_rcu(&key->node, &md5sig->head); | |
cfb6eeb4 YH |
1012 | return 0; |
1013 | } | |
a915da9b | 1014 | EXPORT_SYMBOL(tcp_md5_do_add); |
cfb6eeb4 | 1015 | |
a915da9b | 1016 | int tcp_md5_do_del(struct sock *sk, const union tcp_md5_addr *addr, int family) |
cfb6eeb4 YH |
1017 | { |
1018 | struct tcp_sock *tp = tcp_sk(sk); | |
a915da9b | 1019 | struct tcp_md5sig_key *key; |
a8afca03 | 1020 | struct tcp_md5sig_info *md5sig; |
a915da9b ED |
1021 | |
1022 | key = tcp_md5_do_lookup(sk, (union tcp_md5_addr *)&addr, AF_INET); | |
1023 | if (!key) | |
1024 | return -ENOENT; | |
1025 | hlist_del_rcu(&key->node); | |
5f3d9cb2 | 1026 | atomic_sub(sizeof(*key), &sk->sk_omem_alloc); |
a915da9b | 1027 | kfree_rcu(key, rcu); |
a8afca03 ED |
1028 | md5sig = rcu_dereference_protected(tp->md5sig_info, |
1029 | sock_owned_by_user(sk)); | |
1030 | if (hlist_empty(&md5sig->head)) | |
a915da9b ED |
1031 | tcp_free_md5sig_pool(); |
1032 | return 0; | |
cfb6eeb4 | 1033 | } |
a915da9b | 1034 | EXPORT_SYMBOL(tcp_md5_do_del); |
cfb6eeb4 | 1035 | |
a915da9b | 1036 | void tcp_clear_md5_list(struct sock *sk) |
cfb6eeb4 YH |
1037 | { |
1038 | struct tcp_sock *tp = tcp_sk(sk); | |
a915da9b ED |
1039 | struct tcp_md5sig_key *key; |
1040 | struct hlist_node *pos, *n; | |
a8afca03 | 1041 | struct tcp_md5sig_info *md5sig; |
cfb6eeb4 | 1042 | |
a8afca03 ED |
1043 | md5sig = rcu_dereference_protected(tp->md5sig_info, 1); |
1044 | ||
1045 | if (!hlist_empty(&md5sig->head)) | |
cfb6eeb4 | 1046 | tcp_free_md5sig_pool(); |
a8afca03 | 1047 | hlist_for_each_entry_safe(key, pos, 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 AL |
1148 | } else if (req) { |
1149 | saddr = inet_rsk(req)->loc_addr; | |
1150 | daddr = inet_rsk(req)->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 | ||
72a3effa | 1242 | struct request_sock_ops tcp_request_sock_ops __read_mostly = { |
1da177e4 | 1243 | .family = PF_INET, |
2e6599cb | 1244 | .obj_size = sizeof(struct tcp_request_sock), |
72659ecc | 1245 | .rtx_syn_ack = tcp_v4_rtx_synack, |
60236fdd ACM |
1246 | .send_ack = tcp_v4_reqsk_send_ack, |
1247 | .destructor = tcp_v4_reqsk_destructor, | |
1da177e4 | 1248 | .send_reset = tcp_v4_send_reset, |
72659ecc | 1249 | .syn_ack_timeout = tcp_syn_ack_timeout, |
1da177e4 LT |
1250 | }; |
1251 | ||
cfb6eeb4 | 1252 | #ifdef CONFIG_TCP_MD5SIG |
b2e4b3de | 1253 | static const struct tcp_request_sock_ops tcp_request_sock_ipv4_ops = { |
cfb6eeb4 | 1254 | .md5_lookup = tcp_v4_reqsk_md5_lookup, |
e3afe7b7 | 1255 | .calc_md5_hash = tcp_v4_md5_hash_skb, |
cfb6eeb4 | 1256 | }; |
b6332e6c | 1257 | #endif |
cfb6eeb4 | 1258 | |
1da177e4 LT |
1259 | int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb) |
1260 | { | |
4957faad | 1261 | struct tcp_extend_values tmp_ext; |
1da177e4 | 1262 | struct tcp_options_received tmp_opt; |
cf533ea5 | 1263 | const u8 *hash_location; |
60236fdd | 1264 | struct request_sock *req; |
e6b4d113 | 1265 | struct inet_request_sock *ireq; |
4957faad | 1266 | struct tcp_sock *tp = tcp_sk(sk); |
e6b4d113 | 1267 | struct dst_entry *dst = NULL; |
eddc9ec5 ACM |
1268 | __be32 saddr = ip_hdr(skb)->saddr; |
1269 | __be32 daddr = ip_hdr(skb)->daddr; | |
1da177e4 | 1270 | __u32 isn = TCP_SKB_CB(skb)->when; |
a2a385d6 | 1271 | bool want_cookie = false; |
1da177e4 LT |
1272 | |
1273 | /* Never answer to SYNs send to broadcast or multicast */ | |
511c3f92 | 1274 | if (skb_rtable(skb)->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST)) |
1da177e4 LT |
1275 | goto drop; |
1276 | ||
1277 | /* TW buckets are converted to open requests without | |
1278 | * limitations, they conserve resources and peer is | |
1279 | * evidently real one. | |
1280 | */ | |
463c84b9 | 1281 | if (inet_csk_reqsk_queue_is_full(sk) && !isn) { |
946cedcc ED |
1282 | want_cookie = tcp_syn_flood_action(sk, skb, "TCP"); |
1283 | if (!want_cookie) | |
1284 | goto drop; | |
1da177e4 LT |
1285 | } |
1286 | ||
1287 | /* Accept backlog is full. If we have already queued enough | |
1288 | * of warm entries in syn queue, drop request. It is better than | |
1289 | * clogging syn queue with openreqs with exponentially increasing | |
1290 | * timeout. | |
1291 | */ | |
463c84b9 | 1292 | if (sk_acceptq_is_full(sk) && inet_csk_reqsk_queue_young(sk) > 1) |
1da177e4 LT |
1293 | goto drop; |
1294 | ||
ce4a7d0d | 1295 | req = inet_reqsk_alloc(&tcp_request_sock_ops); |
1da177e4 LT |
1296 | if (!req) |
1297 | goto drop; | |
1298 | ||
cfb6eeb4 YH |
1299 | #ifdef CONFIG_TCP_MD5SIG |
1300 | tcp_rsk(req)->af_specific = &tcp_request_sock_ipv4_ops; | |
1301 | #endif | |
1302 | ||
1da177e4 | 1303 | tcp_clear_options(&tmp_opt); |
bee7ca9e | 1304 | tmp_opt.mss_clamp = TCP_MSS_DEFAULT; |
4957faad | 1305 | tmp_opt.user_mss = tp->rx_opt.user_mss; |
bb5b7c11 | 1306 | tcp_parse_options(skb, &tmp_opt, &hash_location, 0); |
4957faad WAS |
1307 | |
1308 | if (tmp_opt.cookie_plus > 0 && | |
1309 | tmp_opt.saw_tstamp && | |
1310 | !tp->rx_opt.cookie_out_never && | |
1311 | (sysctl_tcp_cookie_size > 0 || | |
1312 | (tp->cookie_values != NULL && | |
1313 | tp->cookie_values->cookie_desired > 0))) { | |
1314 | u8 *c; | |
1315 | u32 *mess = &tmp_ext.cookie_bakery[COOKIE_DIGEST_WORDS]; | |
1316 | int l = tmp_opt.cookie_plus - TCPOLEN_COOKIE_BASE; | |
1317 | ||
1318 | if (tcp_cookie_generator(&tmp_ext.cookie_bakery[0]) != 0) | |
1319 | goto drop_and_release; | |
1320 | ||
1321 | /* Secret recipe starts with IP addresses */ | |
0eae88f3 ED |
1322 | *mess++ ^= (__force u32)daddr; |
1323 | *mess++ ^= (__force u32)saddr; | |
1da177e4 | 1324 | |
4957faad WAS |
1325 | /* plus variable length Initiator Cookie */ |
1326 | c = (u8 *)mess; | |
1327 | while (l-- > 0) | |
1328 | *c++ ^= *hash_location++; | |
1329 | ||
a2a385d6 | 1330 | want_cookie = false; /* not our kind of cookie */ |
4957faad WAS |
1331 | tmp_ext.cookie_out_never = 0; /* false */ |
1332 | tmp_ext.cookie_plus = tmp_opt.cookie_plus; | |
1333 | } else if (!tp->rx_opt.cookie_in_always) { | |
1334 | /* redundant indications, but ensure initialization. */ | |
1335 | tmp_ext.cookie_out_never = 1; /* true */ | |
1336 | tmp_ext.cookie_plus = 0; | |
1337 | } else { | |
1338 | goto drop_and_release; | |
1339 | } | |
1340 | tmp_ext.cookie_in_always = tp->rx_opt.cookie_in_always; | |
1da177e4 | 1341 | |
4dfc2817 | 1342 | if (want_cookie && !tmp_opt.saw_tstamp) |
1da177e4 | 1343 | tcp_clear_options(&tmp_opt); |
1da177e4 | 1344 | |
1da177e4 | 1345 | tmp_opt.tstamp_ok = tmp_opt.saw_tstamp; |
1da177e4 LT |
1346 | tcp_openreq_init(req, &tmp_opt, skb); |
1347 | ||
bb5b7c11 DM |
1348 | ireq = inet_rsk(req); |
1349 | ireq->loc_addr = daddr; | |
1350 | ireq->rmt_addr = saddr; | |
1351 | ireq->no_srccheck = inet_sk(sk)->transparent; | |
1352 | ireq->opt = tcp_v4_save_options(sk, skb); | |
1353 | ||
284904aa | 1354 | if (security_inet_conn_request(sk, skb, req)) |
bb5b7c11 | 1355 | goto drop_and_free; |
284904aa | 1356 | |
172d69e6 | 1357 | if (!want_cookie || tmp_opt.tstamp_ok) |
bd14b1b2 | 1358 | TCP_ECN_create_request(req, skb); |
1da177e4 LT |
1359 | |
1360 | if (want_cookie) { | |
1da177e4 | 1361 | isn = cookie_v4_init_sequence(sk, skb, &req->mss); |
172d69e6 | 1362 | req->cookie_ts = tmp_opt.tstamp_ok; |
1da177e4 | 1363 | } else if (!isn) { |
6bd023f3 | 1364 | struct flowi4 fl4; |
1da177e4 LT |
1365 | |
1366 | /* VJ's idea. We save last timestamp seen | |
1367 | * from the destination in peer table, when entering | |
1368 | * state TIME-WAIT, and check against it before | |
1369 | * accepting new connection request. | |
1370 | * | |
1371 | * If "isn" is not zero, this request hit alive | |
1372 | * timewait bucket, so that all the necessary checks | |
1373 | * are made in the function processing timewait state. | |
1374 | */ | |
1375 | if (tmp_opt.saw_tstamp && | |
295ff7ed | 1376 | tcp_death_row.sysctl_tw_recycle && |
7586eceb | 1377 | (dst = inet_csk_route_req(sk, &fl4, req, want_cookie)) != NULL && |
81166dd6 DM |
1378 | fl4.daddr == saddr) { |
1379 | if (!tcp_peer_is_proven(req, dst, true)) { | |
de0744af | 1380 | NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_PAWSPASSIVEREJECTED); |
7cd04fa7 | 1381 | goto drop_and_release; |
1da177e4 LT |
1382 | } |
1383 | } | |
1384 | /* Kill the following clause, if you dislike this way. */ | |
1385 | else if (!sysctl_tcp_syncookies && | |
463c84b9 | 1386 | (sysctl_max_syn_backlog - inet_csk_reqsk_queue_len(sk) < |
1da177e4 | 1387 | (sysctl_max_syn_backlog >> 2)) && |
81166dd6 | 1388 | !tcp_peer_is_proven(req, dst, false)) { |
1da177e4 LT |
1389 | /* Without syncookies last quarter of |
1390 | * backlog is filled with destinations, | |
1391 | * proven to be alive. | |
1392 | * It means that we continue to communicate | |
1393 | * to destinations, already remembered | |
1394 | * to the moment of synflood. | |
1395 | */ | |
afd46503 | 1396 | LIMIT_NETDEBUG(KERN_DEBUG pr_fmt("drop open request from %pI4/%u\n"), |
673d57e7 | 1397 | &saddr, ntohs(tcp_hdr(skb)->source)); |
7cd04fa7 | 1398 | goto drop_and_release; |
1da177e4 LT |
1399 | } |
1400 | ||
a94f723d | 1401 | isn = tcp_v4_init_sequence(skb); |
1da177e4 | 1402 | } |
2e6599cb | 1403 | tcp_rsk(req)->snt_isn = isn; |
9ad7c049 | 1404 | tcp_rsk(req)->snt_synack = tcp_time_stamp; |
1da177e4 | 1405 | |
72659ecc | 1406 | if (tcp_v4_send_synack(sk, dst, req, |
fff32699 | 1407 | (struct request_values *)&tmp_ext, |
7586eceb ED |
1408 | skb_get_queue_mapping(skb), |
1409 | want_cookie) || | |
4957faad | 1410 | want_cookie) |
1da177e4 LT |
1411 | goto drop_and_free; |
1412 | ||
7cd04fa7 | 1413 | inet_csk_reqsk_queue_hash_add(sk, req, TCP_TIMEOUT_INIT); |
1da177e4 LT |
1414 | return 0; |
1415 | ||
7cd04fa7 DL |
1416 | drop_and_release: |
1417 | dst_release(dst); | |
1da177e4 | 1418 | drop_and_free: |
60236fdd | 1419 | reqsk_free(req); |
1da177e4 | 1420 | drop: |
1da177e4 LT |
1421 | return 0; |
1422 | } | |
4bc2f18b | 1423 | EXPORT_SYMBOL(tcp_v4_conn_request); |
1da177e4 LT |
1424 | |
1425 | ||
1426 | /* | |
1427 | * The three way handshake has completed - we got a valid synack - | |
1428 | * now create the new socket. | |
1429 | */ | |
1430 | struct sock *tcp_v4_syn_recv_sock(struct sock *sk, struct sk_buff *skb, | |
60236fdd | 1431 | struct request_sock *req, |
1da177e4 LT |
1432 | struct dst_entry *dst) |
1433 | { | |
2e6599cb | 1434 | struct inet_request_sock *ireq; |
1da177e4 LT |
1435 | struct inet_sock *newinet; |
1436 | struct tcp_sock *newtp; | |
1437 | struct sock *newsk; | |
cfb6eeb4 YH |
1438 | #ifdef CONFIG_TCP_MD5SIG |
1439 | struct tcp_md5sig_key *key; | |
1440 | #endif | |
f6d8bd05 | 1441 | struct ip_options_rcu *inet_opt; |
1da177e4 LT |
1442 | |
1443 | if (sk_acceptq_is_full(sk)) | |
1444 | goto exit_overflow; | |
1445 | ||
1da177e4 LT |
1446 | newsk = tcp_create_openreq_child(sk, req, skb); |
1447 | if (!newsk) | |
093d2823 | 1448 | goto exit_nonewsk; |
1da177e4 | 1449 | |
bcd76111 | 1450 | newsk->sk_gso_type = SKB_GSO_TCPV4; |
1da177e4 LT |
1451 | |
1452 | newtp = tcp_sk(newsk); | |
1453 | newinet = inet_sk(newsk); | |
2e6599cb | 1454 | ireq = inet_rsk(req); |
c720c7e8 ED |
1455 | newinet->inet_daddr = ireq->rmt_addr; |
1456 | newinet->inet_rcv_saddr = ireq->loc_addr; | |
1457 | newinet->inet_saddr = ireq->loc_addr; | |
f6d8bd05 ED |
1458 | inet_opt = ireq->opt; |
1459 | rcu_assign_pointer(newinet->inet_opt, inet_opt); | |
2e6599cb | 1460 | ireq->opt = NULL; |
463c84b9 | 1461 | newinet->mc_index = inet_iif(skb); |
eddc9ec5 | 1462 | newinet->mc_ttl = ip_hdr(skb)->ttl; |
4c507d28 | 1463 | newinet->rcv_tos = ip_hdr(skb)->tos; |
d83d8461 | 1464 | inet_csk(newsk)->icsk_ext_hdr_len = 0; |
f6d8bd05 ED |
1465 | if (inet_opt) |
1466 | inet_csk(newsk)->icsk_ext_hdr_len = inet_opt->opt.optlen; | |
c720c7e8 | 1467 | newinet->inet_id = newtp->write_seq ^ jiffies; |
1da177e4 | 1468 | |
dfd25fff ED |
1469 | if (!dst) { |
1470 | dst = inet_csk_route_child_sock(sk, newsk, req); | |
1471 | if (!dst) | |
1472 | goto put_and_exit; | |
1473 | } else { | |
1474 | /* syncookie case : see end of cookie_v4_check() */ | |
1475 | } | |
0e734419 DM |
1476 | sk_setup_caps(newsk, dst); |
1477 | ||
5d424d5a | 1478 | tcp_mtup_init(newsk); |
1da177e4 | 1479 | tcp_sync_mss(newsk, dst_mtu(dst)); |
0dbaee3b | 1480 | newtp->advmss = dst_metric_advmss(dst); |
f5fff5dc TQ |
1481 | if (tcp_sk(sk)->rx_opt.user_mss && |
1482 | tcp_sk(sk)->rx_opt.user_mss < newtp->advmss) | |
1483 | newtp->advmss = tcp_sk(sk)->rx_opt.user_mss; | |
1484 | ||
1da177e4 | 1485 | tcp_initialize_rcv_mss(newsk); |
9ad7c049 JC |
1486 | if (tcp_rsk(req)->snt_synack) |
1487 | tcp_valid_rtt_meas(newsk, | |
1488 | tcp_time_stamp - tcp_rsk(req)->snt_synack); | |
1489 | newtp->total_retrans = req->retrans; | |
1da177e4 | 1490 | |
cfb6eeb4 YH |
1491 | #ifdef CONFIG_TCP_MD5SIG |
1492 | /* Copy over the MD5 key from the original socket */ | |
a915da9b ED |
1493 | key = tcp_md5_do_lookup(sk, (union tcp_md5_addr *)&newinet->inet_daddr, |
1494 | AF_INET); | |
c720c7e8 | 1495 | if (key != NULL) { |
cfb6eeb4 YH |
1496 | /* |
1497 | * We're using one, so create a matching key | |
1498 | * on the newsk structure. If we fail to get | |
1499 | * memory, then we end up not copying the key | |
1500 | * across. Shucks. | |
1501 | */ | |
a915da9b ED |
1502 | tcp_md5_do_add(newsk, (union tcp_md5_addr *)&newinet->inet_daddr, |
1503 | AF_INET, key->key, key->keylen, GFP_ATOMIC); | |
a465419b | 1504 | sk_nocaps_add(newsk, NETIF_F_GSO_MASK); |
cfb6eeb4 YH |
1505 | } |
1506 | #endif | |
1507 | ||
0e734419 DM |
1508 | if (__inet_inherit_port(sk, newsk) < 0) |
1509 | goto put_and_exit; | |
9327f705 | 1510 | __inet_hash_nolisten(newsk, NULL); |
1da177e4 LT |
1511 | |
1512 | return newsk; | |
1513 | ||
1514 | exit_overflow: | |
de0744af | 1515 | NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENOVERFLOWS); |
093d2823 BS |
1516 | exit_nonewsk: |
1517 | dst_release(dst); | |
1da177e4 | 1518 | exit: |
de0744af | 1519 | NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENDROPS); |
1da177e4 | 1520 | return NULL; |
0e734419 | 1521 | put_and_exit: |
709e8697 | 1522 | tcp_clear_xmit_timers(newsk); |
d8a6e65f | 1523 | tcp_cleanup_congestion_control(newsk); |
918eb399 | 1524 | bh_unlock_sock(newsk); |
0e734419 DM |
1525 | sock_put(newsk); |
1526 | goto exit; | |
1da177e4 | 1527 | } |
4bc2f18b | 1528 | EXPORT_SYMBOL(tcp_v4_syn_recv_sock); |
1da177e4 LT |
1529 | |
1530 | static struct sock *tcp_v4_hnd_req(struct sock *sk, struct sk_buff *skb) | |
1531 | { | |
aa8223c7 | 1532 | struct tcphdr *th = tcp_hdr(skb); |
eddc9ec5 | 1533 | const struct iphdr *iph = ip_hdr(skb); |
1da177e4 | 1534 | struct sock *nsk; |
60236fdd | 1535 | struct request_sock **prev; |
1da177e4 | 1536 | /* Find possible connection requests. */ |
463c84b9 ACM |
1537 | struct request_sock *req = inet_csk_search_req(sk, &prev, th->source, |
1538 | iph->saddr, iph->daddr); | |
1da177e4 LT |
1539 | if (req) |
1540 | return tcp_check_req(sk, skb, req, prev); | |
1541 | ||
3b1e0a65 | 1542 | nsk = inet_lookup_established(sock_net(sk), &tcp_hashinfo, iph->saddr, |
c67499c0 | 1543 | th->source, iph->daddr, th->dest, inet_iif(skb)); |
1da177e4 LT |
1544 | |
1545 | if (nsk) { | |
1546 | if (nsk->sk_state != TCP_TIME_WAIT) { | |
1547 | bh_lock_sock(nsk); | |
1548 | return nsk; | |
1549 | } | |
9469c7b4 | 1550 | inet_twsk_put(inet_twsk(nsk)); |
1da177e4 LT |
1551 | return NULL; |
1552 | } | |
1553 | ||
1554 | #ifdef CONFIG_SYN_COOKIES | |
af9b4738 | 1555 | if (!th->syn) |
1da177e4 LT |
1556 | sk = cookie_v4_check(sk, skb, &(IPCB(skb)->opt)); |
1557 | #endif | |
1558 | return sk; | |
1559 | } | |
1560 | ||
b51655b9 | 1561 | static __sum16 tcp_v4_checksum_init(struct sk_buff *skb) |
1da177e4 | 1562 | { |
eddc9ec5 ACM |
1563 | const struct iphdr *iph = ip_hdr(skb); |
1564 | ||
84fa7933 | 1565 | if (skb->ip_summed == CHECKSUM_COMPLETE) { |
eddc9ec5 ACM |
1566 | if (!tcp_v4_check(skb->len, iph->saddr, |
1567 | iph->daddr, skb->csum)) { | |
fb286bb2 | 1568 | skb->ip_summed = CHECKSUM_UNNECESSARY; |
1da177e4 | 1569 | return 0; |
fb286bb2 | 1570 | } |
1da177e4 | 1571 | } |
fb286bb2 | 1572 | |
eddc9ec5 | 1573 | skb->csum = csum_tcpudp_nofold(iph->saddr, iph->daddr, |
fb286bb2 HX |
1574 | skb->len, IPPROTO_TCP, 0); |
1575 | ||
1da177e4 | 1576 | if (skb->len <= 76) { |
fb286bb2 | 1577 | return __skb_checksum_complete(skb); |
1da177e4 LT |
1578 | } |
1579 | return 0; | |
1580 | } | |
1581 | ||
1582 | ||
1583 | /* The socket must have it's spinlock held when we get | |
1584 | * here. | |
1585 | * | |
1586 | * We have a potential double-lock case here, so even when | |
1587 | * doing backlog processing we use the BH locking scheme. | |
1588 | * This is because we cannot sleep with the original spinlock | |
1589 | * held. | |
1590 | */ | |
1591 | int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb) | |
1592 | { | |
cfb6eeb4 YH |
1593 | struct sock *rsk; |
1594 | #ifdef CONFIG_TCP_MD5SIG | |
1595 | /* | |
1596 | * We really want to reject the packet as early as possible | |
1597 | * if: | |
1598 | * o We're expecting an MD5'd packet and this is no MD5 tcp option | |
1599 | * o There is an MD5 option and we're not expecting one | |
1600 | */ | |
7174259e | 1601 | if (tcp_v4_inbound_md5_hash(sk, skb)) |
cfb6eeb4 YH |
1602 | goto discard; |
1603 | #endif | |
1604 | ||
1da177e4 | 1605 | if (sk->sk_state == TCP_ESTABLISHED) { /* Fast path */ |
bdeab991 | 1606 | sock_rps_save_rxhash(sk, skb); |
aa8223c7 | 1607 | if (tcp_rcv_established(sk, skb, tcp_hdr(skb), skb->len)) { |
cfb6eeb4 | 1608 | rsk = sk; |
1da177e4 | 1609 | goto reset; |
cfb6eeb4 | 1610 | } |
1da177e4 LT |
1611 | return 0; |
1612 | } | |
1613 | ||
ab6a5bb6 | 1614 | if (skb->len < tcp_hdrlen(skb) || tcp_checksum_complete(skb)) |
1da177e4 LT |
1615 | goto csum_err; |
1616 | ||
1617 | if (sk->sk_state == TCP_LISTEN) { | |
1618 | struct sock *nsk = tcp_v4_hnd_req(sk, skb); | |
1619 | if (!nsk) | |
1620 | goto discard; | |
1621 | ||
1622 | if (nsk != sk) { | |
bdeab991 | 1623 | sock_rps_save_rxhash(nsk, skb); |
cfb6eeb4 YH |
1624 | if (tcp_child_process(sk, nsk, skb)) { |
1625 | rsk = nsk; | |
1da177e4 | 1626 | goto reset; |
cfb6eeb4 | 1627 | } |
1da177e4 LT |
1628 | return 0; |
1629 | } | |
ca55158c | 1630 | } else |
bdeab991 | 1631 | sock_rps_save_rxhash(sk, skb); |
ca55158c | 1632 | |
aa8223c7 | 1633 | if (tcp_rcv_state_process(sk, skb, tcp_hdr(skb), skb->len)) { |
cfb6eeb4 | 1634 | rsk = sk; |
1da177e4 | 1635 | goto reset; |
cfb6eeb4 | 1636 | } |
1da177e4 LT |
1637 | return 0; |
1638 | ||
1639 | reset: | |
cfb6eeb4 | 1640 | tcp_v4_send_reset(rsk, skb); |
1da177e4 LT |
1641 | discard: |
1642 | kfree_skb(skb); | |
1643 | /* Be careful here. If this function gets more complicated and | |
1644 | * gcc suffers from register pressure on the x86, sk (in %ebx) | |
1645 | * might be destroyed here. This current version compiles correctly, | |
1646 | * but you have been warned. | |
1647 | */ | |
1648 | return 0; | |
1649 | ||
1650 | csum_err: | |
63231bdd | 1651 | TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_INERRS); |
1da177e4 LT |
1652 | goto discard; |
1653 | } | |
4bc2f18b | 1654 | EXPORT_SYMBOL(tcp_v4_do_rcv); |
1da177e4 | 1655 | |
160eb5a6 | 1656 | void tcp_v4_early_demux(struct sk_buff *skb) |
41063e9d DM |
1657 | { |
1658 | struct net *net = dev_net(skb->dev); | |
1659 | const struct iphdr *iph; | |
1660 | const struct tcphdr *th; | |
fd62e09b | 1661 | struct net_device *dev; |
41063e9d | 1662 | struct sock *sk; |
41063e9d | 1663 | |
41063e9d | 1664 | if (skb->pkt_type != PACKET_HOST) |
160eb5a6 | 1665 | return; |
41063e9d DM |
1666 | |
1667 | if (!pskb_may_pull(skb, ip_hdrlen(skb) + sizeof(struct tcphdr))) | |
160eb5a6 | 1668 | return; |
41063e9d DM |
1669 | |
1670 | iph = ip_hdr(skb); | |
1671 | th = (struct tcphdr *) ((char *)iph + ip_hdrlen(skb)); | |
1672 | ||
1673 | if (th->doff < sizeof(struct tcphdr) / 4) | |
160eb5a6 | 1674 | return; |
41063e9d DM |
1675 | |
1676 | if (!pskb_may_pull(skb, ip_hdrlen(skb) + th->doff * 4)) | |
160eb5a6 | 1677 | return; |
41063e9d | 1678 | |
fd62e09b | 1679 | dev = skb->dev; |
41063e9d DM |
1680 | sk = __inet_lookup_established(net, &tcp_hashinfo, |
1681 | iph->saddr, th->source, | |
7011d085 | 1682 | iph->daddr, ntohs(th->dest), |
fd62e09b | 1683 | dev->ifindex); |
41063e9d DM |
1684 | if (sk) { |
1685 | skb->sk = sk; | |
1686 | skb->destructor = sock_edemux; | |
1687 | if (sk->sk_state != TCP_TIME_WAIT) { | |
1688 | struct dst_entry *dst = sk->sk_rx_dst; | |
1689 | if (dst) | |
1690 | dst = dst_check(dst, 0); | |
1691 | if (dst) { | |
fd62e09b DM |
1692 | struct rtable *rt = (struct rtable *) dst; |
1693 | ||
160eb5a6 | 1694 | if (rt->rt_iif == dev->ifindex) |
fd62e09b | 1695 | skb_dst_set_noref(skb, dst); |
41063e9d DM |
1696 | } |
1697 | } | |
1698 | } | |
41063e9d DM |
1699 | } |
1700 | ||
1da177e4 LT |
1701 | /* |
1702 | * From tcp_input.c | |
1703 | */ | |
1704 | ||
1705 | int tcp_v4_rcv(struct sk_buff *skb) | |
1706 | { | |
eddc9ec5 | 1707 | const struct iphdr *iph; |
cf533ea5 | 1708 | const struct tcphdr *th; |
1da177e4 LT |
1709 | struct sock *sk; |
1710 | int ret; | |
a86b1e30 | 1711 | struct net *net = dev_net(skb->dev); |
1da177e4 LT |
1712 | |
1713 | if (skb->pkt_type != PACKET_HOST) | |
1714 | goto discard_it; | |
1715 | ||
1716 | /* Count it even if it's bad */ | |
63231bdd | 1717 | TCP_INC_STATS_BH(net, TCP_MIB_INSEGS); |
1da177e4 LT |
1718 | |
1719 | if (!pskb_may_pull(skb, sizeof(struct tcphdr))) | |
1720 | goto discard_it; | |
1721 | ||
aa8223c7 | 1722 | th = tcp_hdr(skb); |
1da177e4 LT |
1723 | |
1724 | if (th->doff < sizeof(struct tcphdr) / 4) | |
1725 | goto bad_packet; | |
1726 | if (!pskb_may_pull(skb, th->doff * 4)) | |
1727 | goto discard_it; | |
1728 | ||
1729 | /* An explanation is required here, I think. | |
1730 | * Packet length and doff are validated by header prediction, | |
caa20d9a | 1731 | * provided case of th->doff==0 is eliminated. |
1da177e4 | 1732 | * So, we defer the checks. */ |
60476372 | 1733 | if (!skb_csum_unnecessary(skb) && tcp_v4_checksum_init(skb)) |
1da177e4 LT |
1734 | goto bad_packet; |
1735 | ||
aa8223c7 | 1736 | th = tcp_hdr(skb); |
eddc9ec5 | 1737 | iph = ip_hdr(skb); |
1da177e4 LT |
1738 | TCP_SKB_CB(skb)->seq = ntohl(th->seq); |
1739 | TCP_SKB_CB(skb)->end_seq = (TCP_SKB_CB(skb)->seq + th->syn + th->fin + | |
1740 | skb->len - th->doff * 4); | |
1741 | TCP_SKB_CB(skb)->ack_seq = ntohl(th->ack_seq); | |
1742 | TCP_SKB_CB(skb)->when = 0; | |
b82d1bb4 | 1743 | TCP_SKB_CB(skb)->ip_dsfield = ipv4_get_dsfield(iph); |
1da177e4 LT |
1744 | TCP_SKB_CB(skb)->sacked = 0; |
1745 | ||
9a1f27c4 | 1746 | sk = __inet_lookup_skb(&tcp_hashinfo, skb, th->source, th->dest); |
1da177e4 LT |
1747 | if (!sk) |
1748 | goto no_tcp_socket; | |
1749 | ||
bb134d5d ED |
1750 | process: |
1751 | if (sk->sk_state == TCP_TIME_WAIT) | |
1752 | goto do_time_wait; | |
1753 | ||
6cce09f8 ED |
1754 | if (unlikely(iph->ttl < inet_sk(sk)->min_ttl)) { |
1755 | NET_INC_STATS_BH(net, LINUX_MIB_TCPMINTTLDROP); | |
d218d111 | 1756 | goto discard_and_relse; |
6cce09f8 | 1757 | } |
d218d111 | 1758 | |
1da177e4 LT |
1759 | if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb)) |
1760 | goto discard_and_relse; | |
b59c2701 | 1761 | nf_reset(skb); |
1da177e4 | 1762 | |
fda9ef5d | 1763 | if (sk_filter(sk, skb)) |
1da177e4 LT |
1764 | goto discard_and_relse; |
1765 | ||
1766 | skb->dev = NULL; | |
1767 | ||
c6366184 | 1768 | bh_lock_sock_nested(sk); |
1da177e4 LT |
1769 | ret = 0; |
1770 | if (!sock_owned_by_user(sk)) { | |
1a2449a8 CL |
1771 | #ifdef CONFIG_NET_DMA |
1772 | struct tcp_sock *tp = tcp_sk(sk); | |
1773 | if (!tp->ucopy.dma_chan && tp->ucopy.pinned_list) | |
a2bd1140 | 1774 | tp->ucopy.dma_chan = net_dma_find_channel(); |
1a2449a8 | 1775 | if (tp->ucopy.dma_chan) |
1da177e4 | 1776 | ret = tcp_v4_do_rcv(sk, skb); |
1a2449a8 CL |
1777 | else |
1778 | #endif | |
1779 | { | |
1780 | if (!tcp_prequeue(sk, skb)) | |
ae8d7f88 | 1781 | ret = tcp_v4_do_rcv(sk, skb); |
1a2449a8 | 1782 | } |
da882c1f ED |
1783 | } else if (unlikely(sk_add_backlog(sk, skb, |
1784 | sk->sk_rcvbuf + sk->sk_sndbuf))) { | |
6b03a53a | 1785 | bh_unlock_sock(sk); |
6cce09f8 | 1786 | NET_INC_STATS_BH(net, LINUX_MIB_TCPBACKLOGDROP); |
6b03a53a ZY |
1787 | goto discard_and_relse; |
1788 | } | |
1da177e4 LT |
1789 | bh_unlock_sock(sk); |
1790 | ||
1791 | sock_put(sk); | |
1792 | ||
1793 | return ret; | |
1794 | ||
1795 | no_tcp_socket: | |
1796 | if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) | |
1797 | goto discard_it; | |
1798 | ||
1799 | if (skb->len < (th->doff << 2) || tcp_checksum_complete(skb)) { | |
1800 | bad_packet: | |
63231bdd | 1801 | TCP_INC_STATS_BH(net, TCP_MIB_INERRS); |
1da177e4 | 1802 | } else { |
cfb6eeb4 | 1803 | tcp_v4_send_reset(NULL, skb); |
1da177e4 LT |
1804 | } |
1805 | ||
1806 | discard_it: | |
1807 | /* Discard frame. */ | |
1808 | kfree_skb(skb); | |
e905a9ed | 1809 | return 0; |
1da177e4 LT |
1810 | |
1811 | discard_and_relse: | |
1812 | sock_put(sk); | |
1813 | goto discard_it; | |
1814 | ||
1815 | do_time_wait: | |
1816 | if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) { | |
9469c7b4 | 1817 | inet_twsk_put(inet_twsk(sk)); |
1da177e4 LT |
1818 | goto discard_it; |
1819 | } | |
1820 | ||
1821 | if (skb->len < (th->doff << 2) || tcp_checksum_complete(skb)) { | |
63231bdd | 1822 | TCP_INC_STATS_BH(net, TCP_MIB_INERRS); |
9469c7b4 | 1823 | inet_twsk_put(inet_twsk(sk)); |
1da177e4 LT |
1824 | goto discard_it; |
1825 | } | |
9469c7b4 | 1826 | switch (tcp_timewait_state_process(inet_twsk(sk), skb, th)) { |
1da177e4 | 1827 | case TCP_TW_SYN: { |
c346dca1 | 1828 | struct sock *sk2 = inet_lookup_listener(dev_net(skb->dev), |
c67499c0 | 1829 | &tcp_hashinfo, |
eddc9ec5 | 1830 | iph->daddr, th->dest, |
463c84b9 | 1831 | inet_iif(skb)); |
1da177e4 | 1832 | if (sk2) { |
9469c7b4 YH |
1833 | inet_twsk_deschedule(inet_twsk(sk), &tcp_death_row); |
1834 | inet_twsk_put(inet_twsk(sk)); | |
1da177e4 LT |
1835 | sk = sk2; |
1836 | goto process; | |
1837 | } | |
1838 | /* Fall through to ACK */ | |
1839 | } | |
1840 | case TCP_TW_ACK: | |
1841 | tcp_v4_timewait_ack(sk, skb); | |
1842 | break; | |
1843 | case TCP_TW_RST: | |
1844 | goto no_tcp_socket; | |
1845 | case TCP_TW_SUCCESS:; | |
1846 | } | |
1847 | goto discard_it; | |
1848 | } | |
1849 | ||
ccb7c410 DM |
1850 | static struct timewait_sock_ops tcp_timewait_sock_ops = { |
1851 | .twsk_obj_size = sizeof(struct tcp_timewait_sock), | |
1852 | .twsk_unique = tcp_twsk_unique, | |
1853 | .twsk_destructor= tcp_twsk_destructor, | |
ccb7c410 | 1854 | }; |
1da177e4 | 1855 | |
3b401a81 | 1856 | const struct inet_connection_sock_af_ops ipv4_specific = { |
543d9cfe ACM |
1857 | .queue_xmit = ip_queue_xmit, |
1858 | .send_check = tcp_v4_send_check, | |
1859 | .rebuild_header = inet_sk_rebuild_header, | |
1860 | .conn_request = tcp_v4_conn_request, | |
1861 | .syn_recv_sock = tcp_v4_syn_recv_sock, | |
543d9cfe ACM |
1862 | .net_header_len = sizeof(struct iphdr), |
1863 | .setsockopt = ip_setsockopt, | |
1864 | .getsockopt = ip_getsockopt, | |
1865 | .addr2sockaddr = inet_csk_addr2sockaddr, | |
1866 | .sockaddr_len = sizeof(struct sockaddr_in), | |
ab1e0a13 | 1867 | .bind_conflict = inet_csk_bind_conflict, |
3fdadf7d | 1868 | #ifdef CONFIG_COMPAT |
543d9cfe ACM |
1869 | .compat_setsockopt = compat_ip_setsockopt, |
1870 | .compat_getsockopt = compat_ip_getsockopt, | |
3fdadf7d | 1871 | #endif |
1da177e4 | 1872 | }; |
4bc2f18b | 1873 | EXPORT_SYMBOL(ipv4_specific); |
1da177e4 | 1874 | |
cfb6eeb4 | 1875 | #ifdef CONFIG_TCP_MD5SIG |
b2e4b3de | 1876 | static const struct tcp_sock_af_ops tcp_sock_ipv4_specific = { |
cfb6eeb4 | 1877 | .md5_lookup = tcp_v4_md5_lookup, |
49a72dfb | 1878 | .calc_md5_hash = tcp_v4_md5_hash_skb, |
cfb6eeb4 | 1879 | .md5_parse = tcp_v4_parse_md5_keys, |
cfb6eeb4 | 1880 | }; |
b6332e6c | 1881 | #endif |
cfb6eeb4 | 1882 | |
1da177e4 LT |
1883 | /* NOTE: A lot of things set to zero explicitly by call to |
1884 | * sk_alloc() so need not be done here. | |
1885 | */ | |
1886 | static int tcp_v4_init_sock(struct sock *sk) | |
1887 | { | |
6687e988 | 1888 | struct inet_connection_sock *icsk = inet_csk(sk); |
1da177e4 | 1889 | |
900f65d3 | 1890 | tcp_init_sock(sk); |
1da177e4 | 1891 | |
8292a17a | 1892 | icsk->icsk_af_ops = &ipv4_specific; |
900f65d3 | 1893 | |
cfb6eeb4 | 1894 | #ifdef CONFIG_TCP_MD5SIG |
ac807fa8 | 1895 | tcp_sk(sk)->af_specific = &tcp_sock_ipv4_specific; |
cfb6eeb4 | 1896 | #endif |
1da177e4 | 1897 | |
1da177e4 LT |
1898 | return 0; |
1899 | } | |
1900 | ||
7d06b2e0 | 1901 | void tcp_v4_destroy_sock(struct sock *sk) |
1da177e4 LT |
1902 | { |
1903 | struct tcp_sock *tp = tcp_sk(sk); | |
1904 | ||
1905 | tcp_clear_xmit_timers(sk); | |
1906 | ||
6687e988 | 1907 | tcp_cleanup_congestion_control(sk); |
317a76f9 | 1908 | |
1da177e4 | 1909 | /* Cleanup up the write buffer. */ |
fe067e8a | 1910 | tcp_write_queue_purge(sk); |
1da177e4 LT |
1911 | |
1912 | /* Cleans up our, hopefully empty, out_of_order_queue. */ | |
e905a9ed | 1913 | __skb_queue_purge(&tp->out_of_order_queue); |
1da177e4 | 1914 | |
cfb6eeb4 YH |
1915 | #ifdef CONFIG_TCP_MD5SIG |
1916 | /* Clean up the MD5 key list, if any */ | |
1917 | if (tp->md5sig_info) { | |
a915da9b | 1918 | tcp_clear_md5_list(sk); |
a8afca03 | 1919 | kfree_rcu(tp->md5sig_info, rcu); |
cfb6eeb4 YH |
1920 | tp->md5sig_info = NULL; |
1921 | } | |
1922 | #endif | |
1923 | ||
1a2449a8 CL |
1924 | #ifdef CONFIG_NET_DMA |
1925 | /* Cleans up our sk_async_wait_queue */ | |
e905a9ed | 1926 | __skb_queue_purge(&sk->sk_async_wait_queue); |
1a2449a8 CL |
1927 | #endif |
1928 | ||
1da177e4 LT |
1929 | /* Clean prequeue, it must be empty really */ |
1930 | __skb_queue_purge(&tp->ucopy.prequeue); | |
1931 | ||
1932 | /* Clean up a referenced TCP bind bucket. */ | |
463c84b9 | 1933 | if (inet_csk(sk)->icsk_bind_hash) |
ab1e0a13 | 1934 | inet_put_port(sk); |
1da177e4 LT |
1935 | |
1936 | /* | |
1937 | * If sendmsg cached page exists, toss it. | |
1938 | */ | |
1939 | if (sk->sk_sndmsg_page) { | |
1940 | __free_page(sk->sk_sndmsg_page); | |
1941 | sk->sk_sndmsg_page = NULL; | |
1942 | } | |
1943 | ||
435cf559 WAS |
1944 | /* TCP Cookie Transactions */ |
1945 | if (tp->cookie_values != NULL) { | |
1946 | kref_put(&tp->cookie_values->kref, | |
1947 | tcp_cookie_values_release); | |
1948 | tp->cookie_values = NULL; | |
1949 | } | |
1950 | ||
180d8cd9 | 1951 | sk_sockets_allocated_dec(sk); |
d1a4c0b3 | 1952 | sock_release_memcg(sk); |
1da177e4 | 1953 | } |
1da177e4 LT |
1954 | EXPORT_SYMBOL(tcp_v4_destroy_sock); |
1955 | ||
1956 | #ifdef CONFIG_PROC_FS | |
1957 | /* Proc filesystem TCP sock list dumping. */ | |
1958 | ||
3ab5aee7 | 1959 | static inline struct inet_timewait_sock *tw_head(struct hlist_nulls_head *head) |
1da177e4 | 1960 | { |
3ab5aee7 | 1961 | return hlist_nulls_empty(head) ? NULL : |
8feaf0c0 | 1962 | list_entry(head->first, struct inet_timewait_sock, tw_node); |
1da177e4 LT |
1963 | } |
1964 | ||
8feaf0c0 | 1965 | static inline struct inet_timewait_sock *tw_next(struct inet_timewait_sock *tw) |
1da177e4 | 1966 | { |
3ab5aee7 ED |
1967 | return !is_a_nulls(tw->tw_node.next) ? |
1968 | hlist_nulls_entry(tw->tw_node.next, typeof(*tw), tw_node) : NULL; | |
1da177e4 LT |
1969 | } |
1970 | ||
a8b690f9 TH |
1971 | /* |
1972 | * Get next listener socket follow cur. If cur is NULL, get first socket | |
1973 | * starting from bucket given in st->bucket; when st->bucket is zero the | |
1974 | * very first socket in the hash table is returned. | |
1975 | */ | |
1da177e4 LT |
1976 | static void *listening_get_next(struct seq_file *seq, void *cur) |
1977 | { | |
463c84b9 | 1978 | struct inet_connection_sock *icsk; |
c25eb3bf | 1979 | struct hlist_nulls_node *node; |
1da177e4 | 1980 | struct sock *sk = cur; |
5caea4ea | 1981 | struct inet_listen_hashbucket *ilb; |
5799de0b | 1982 | struct tcp_iter_state *st = seq->private; |
a4146b1b | 1983 | struct net *net = seq_file_net(seq); |
1da177e4 LT |
1984 | |
1985 | if (!sk) { | |
a8b690f9 | 1986 | ilb = &tcp_hashinfo.listening_hash[st->bucket]; |
5caea4ea | 1987 | spin_lock_bh(&ilb->lock); |
c25eb3bf | 1988 | sk = sk_nulls_head(&ilb->head); |
a8b690f9 | 1989 | st->offset = 0; |
1da177e4 LT |
1990 | goto get_sk; |
1991 | } | |
5caea4ea | 1992 | ilb = &tcp_hashinfo.listening_hash[st->bucket]; |
1da177e4 | 1993 | ++st->num; |
a8b690f9 | 1994 | ++st->offset; |
1da177e4 LT |
1995 | |
1996 | if (st->state == TCP_SEQ_STATE_OPENREQ) { | |
60236fdd | 1997 | struct request_sock *req = cur; |
1da177e4 | 1998 | |
72a3effa | 1999 | icsk = inet_csk(st->syn_wait_sk); |
1da177e4 LT |
2000 | req = req->dl_next; |
2001 | while (1) { | |
2002 | while (req) { | |
bdccc4ca | 2003 | if (req->rsk_ops->family == st->family) { |
1da177e4 LT |
2004 | cur = req; |
2005 | goto out; | |
2006 | } | |
2007 | req = req->dl_next; | |
2008 | } | |
72a3effa | 2009 | if (++st->sbucket >= icsk->icsk_accept_queue.listen_opt->nr_table_entries) |
1da177e4 LT |
2010 | break; |
2011 | get_req: | |
463c84b9 | 2012 | req = icsk->icsk_accept_queue.listen_opt->syn_table[st->sbucket]; |
1da177e4 | 2013 | } |
1bde5ac4 | 2014 | sk = sk_nulls_next(st->syn_wait_sk); |
1da177e4 | 2015 | st->state = TCP_SEQ_STATE_LISTENING; |
463c84b9 | 2016 | read_unlock_bh(&icsk->icsk_accept_queue.syn_wait_lock); |
1da177e4 | 2017 | } else { |
e905a9ed | 2018 | icsk = inet_csk(sk); |
463c84b9 ACM |
2019 | read_lock_bh(&icsk->icsk_accept_queue.syn_wait_lock); |
2020 | if (reqsk_queue_len(&icsk->icsk_accept_queue)) | |
1da177e4 | 2021 | goto start_req; |
463c84b9 | 2022 | read_unlock_bh(&icsk->icsk_accept_queue.syn_wait_lock); |
1bde5ac4 | 2023 | sk = sk_nulls_next(sk); |
1da177e4 LT |
2024 | } |
2025 | get_sk: | |
c25eb3bf | 2026 | sk_nulls_for_each_from(sk, node) { |
8475ef9f PE |
2027 | if (!net_eq(sock_net(sk), net)) |
2028 | continue; | |
2029 | if (sk->sk_family == st->family) { | |
1da177e4 LT |
2030 | cur = sk; |
2031 | goto out; | |
2032 | } | |
e905a9ed | 2033 | icsk = inet_csk(sk); |
463c84b9 ACM |
2034 | read_lock_bh(&icsk->icsk_accept_queue.syn_wait_lock); |
2035 | if (reqsk_queue_len(&icsk->icsk_accept_queue)) { | |
1da177e4 LT |
2036 | start_req: |
2037 | st->uid = sock_i_uid(sk); | |
2038 | st->syn_wait_sk = sk; | |
2039 | st->state = TCP_SEQ_STATE_OPENREQ; | |
2040 | st->sbucket = 0; | |
2041 | goto get_req; | |
2042 | } | |
463c84b9 | 2043 | read_unlock_bh(&icsk->icsk_accept_queue.syn_wait_lock); |
1da177e4 | 2044 | } |
5caea4ea | 2045 | spin_unlock_bh(&ilb->lock); |
a8b690f9 | 2046 | st->offset = 0; |
0f7ff927 | 2047 | if (++st->bucket < INET_LHTABLE_SIZE) { |
5caea4ea ED |
2048 | ilb = &tcp_hashinfo.listening_hash[st->bucket]; |
2049 | spin_lock_bh(&ilb->lock); | |
c25eb3bf | 2050 | sk = sk_nulls_head(&ilb->head); |
1da177e4 LT |
2051 | goto get_sk; |
2052 | } | |
2053 | cur = NULL; | |
2054 | out: | |
2055 | return cur; | |
2056 | } | |
2057 | ||
2058 | static void *listening_get_idx(struct seq_file *seq, loff_t *pos) | |
2059 | { | |
a8b690f9 TH |
2060 | struct tcp_iter_state *st = seq->private; |
2061 | void *rc; | |
2062 | ||
2063 | st->bucket = 0; | |
2064 | st->offset = 0; | |
2065 | rc = listening_get_next(seq, NULL); | |
1da177e4 LT |
2066 | |
2067 | while (rc && *pos) { | |
2068 | rc = listening_get_next(seq, rc); | |
2069 | --*pos; | |
2070 | } | |
2071 | return rc; | |
2072 | } | |
2073 | ||
a2a385d6 | 2074 | static inline bool empty_bucket(struct tcp_iter_state *st) |
6eac5604 | 2075 | { |
3ab5aee7 ED |
2076 | return hlist_nulls_empty(&tcp_hashinfo.ehash[st->bucket].chain) && |
2077 | hlist_nulls_empty(&tcp_hashinfo.ehash[st->bucket].twchain); | |
6eac5604 AK |
2078 | } |
2079 | ||
a8b690f9 TH |
2080 | /* |
2081 | * Get first established socket starting from bucket given in st->bucket. | |
2082 | * If st->bucket is zero, the very first socket in the hash is returned. | |
2083 | */ | |
1da177e4 LT |
2084 | static void *established_get_first(struct seq_file *seq) |
2085 | { | |
5799de0b | 2086 | struct tcp_iter_state *st = seq->private; |
a4146b1b | 2087 | struct net *net = seq_file_net(seq); |
1da177e4 LT |
2088 | void *rc = NULL; |
2089 | ||
a8b690f9 TH |
2090 | st->offset = 0; |
2091 | for (; st->bucket <= tcp_hashinfo.ehash_mask; ++st->bucket) { | |
1da177e4 | 2092 | struct sock *sk; |
3ab5aee7 | 2093 | struct hlist_nulls_node *node; |
8feaf0c0 | 2094 | struct inet_timewait_sock *tw; |
9db66bdc | 2095 | spinlock_t *lock = inet_ehash_lockp(&tcp_hashinfo, st->bucket); |
1da177e4 | 2096 | |
6eac5604 AK |
2097 | /* Lockless fast path for the common case of empty buckets */ |
2098 | if (empty_bucket(st)) | |
2099 | continue; | |
2100 | ||
9db66bdc | 2101 | spin_lock_bh(lock); |
3ab5aee7 | 2102 | sk_nulls_for_each(sk, node, &tcp_hashinfo.ehash[st->bucket].chain) { |
f40c8174 | 2103 | if (sk->sk_family != st->family || |
878628fb | 2104 | !net_eq(sock_net(sk), net)) { |
1da177e4 LT |
2105 | continue; |
2106 | } | |
2107 | rc = sk; | |
2108 | goto out; | |
2109 | } | |
2110 | st->state = TCP_SEQ_STATE_TIME_WAIT; | |
8feaf0c0 | 2111 | inet_twsk_for_each(tw, node, |
dbca9b27 | 2112 | &tcp_hashinfo.ehash[st->bucket].twchain) { |
28518fc1 | 2113 | if (tw->tw_family != st->family || |
878628fb | 2114 | !net_eq(twsk_net(tw), net)) { |
1da177e4 LT |
2115 | continue; |
2116 | } | |
2117 | rc = tw; | |
2118 | goto out; | |
2119 | } | |
9db66bdc | 2120 | spin_unlock_bh(lock); |
1da177e4 LT |
2121 | st->state = TCP_SEQ_STATE_ESTABLISHED; |
2122 | } | |
2123 | out: | |
2124 | return rc; | |
2125 | } | |
2126 | ||
2127 | static void *established_get_next(struct seq_file *seq, void *cur) | |
2128 | { | |
2129 | struct sock *sk = cur; | |
8feaf0c0 | 2130 | struct inet_timewait_sock *tw; |
3ab5aee7 | 2131 | struct hlist_nulls_node *node; |
5799de0b | 2132 | struct tcp_iter_state *st = seq->private; |
a4146b1b | 2133 | struct net *net = seq_file_net(seq); |
1da177e4 LT |
2134 | |
2135 | ++st->num; | |
a8b690f9 | 2136 | ++st->offset; |
1da177e4 LT |
2137 | |
2138 | if (st->state == TCP_SEQ_STATE_TIME_WAIT) { | |
2139 | tw = cur; | |
2140 | tw = tw_next(tw); | |
2141 | get_tw: | |
878628fb | 2142 | while (tw && (tw->tw_family != st->family || !net_eq(twsk_net(tw), net))) { |
1da177e4 LT |
2143 | tw = tw_next(tw); |
2144 | } | |
2145 | if (tw) { | |
2146 | cur = tw; | |
2147 | goto out; | |
2148 | } | |
9db66bdc | 2149 | spin_unlock_bh(inet_ehash_lockp(&tcp_hashinfo, st->bucket)); |
1da177e4 LT |
2150 | st->state = TCP_SEQ_STATE_ESTABLISHED; |
2151 | ||
6eac5604 | 2152 | /* Look for next non empty bucket */ |
a8b690f9 | 2153 | st->offset = 0; |
f373b53b | 2154 | while (++st->bucket <= tcp_hashinfo.ehash_mask && |
6eac5604 AK |
2155 | empty_bucket(st)) |
2156 | ; | |
f373b53b | 2157 | if (st->bucket > tcp_hashinfo.ehash_mask) |
6eac5604 AK |
2158 | return NULL; |
2159 | ||
9db66bdc | 2160 | spin_lock_bh(inet_ehash_lockp(&tcp_hashinfo, st->bucket)); |
3ab5aee7 | 2161 | sk = sk_nulls_head(&tcp_hashinfo.ehash[st->bucket].chain); |
1da177e4 | 2162 | } else |
3ab5aee7 | 2163 | sk = sk_nulls_next(sk); |
1da177e4 | 2164 | |
3ab5aee7 | 2165 | sk_nulls_for_each_from(sk, node) { |
878628fb | 2166 | if (sk->sk_family == st->family && net_eq(sock_net(sk), net)) |
1da177e4 LT |
2167 | goto found; |
2168 | } | |
2169 | ||
2170 | st->state = TCP_SEQ_STATE_TIME_WAIT; | |
dbca9b27 | 2171 | tw = tw_head(&tcp_hashinfo.ehash[st->bucket].twchain); |
1da177e4 LT |
2172 | goto get_tw; |
2173 | found: | |
2174 | cur = sk; | |
2175 | out: | |
2176 | return cur; | |
2177 | } | |
2178 | ||
2179 | static void *established_get_idx(struct seq_file *seq, loff_t pos) | |
2180 | { | |
a8b690f9 TH |
2181 | struct tcp_iter_state *st = seq->private; |
2182 | void *rc; | |
2183 | ||
2184 | st->bucket = 0; | |
2185 | rc = established_get_first(seq); | |
1da177e4 LT |
2186 | |
2187 | while (rc && pos) { | |
2188 | rc = established_get_next(seq, rc); | |
2189 | --pos; | |
7174259e | 2190 | } |
1da177e4 LT |
2191 | return rc; |
2192 | } | |
2193 | ||
2194 | static void *tcp_get_idx(struct seq_file *seq, loff_t pos) | |
2195 | { | |
2196 | void *rc; | |
5799de0b | 2197 | struct tcp_iter_state *st = seq->private; |
1da177e4 | 2198 | |
1da177e4 LT |
2199 | st->state = TCP_SEQ_STATE_LISTENING; |
2200 | rc = listening_get_idx(seq, &pos); | |
2201 | ||
2202 | if (!rc) { | |
1da177e4 LT |
2203 | st->state = TCP_SEQ_STATE_ESTABLISHED; |
2204 | rc = established_get_idx(seq, pos); | |
2205 | } | |
2206 | ||
2207 | return rc; | |
2208 | } | |
2209 | ||
a8b690f9 TH |
2210 | static void *tcp_seek_last_pos(struct seq_file *seq) |
2211 | { | |
2212 | struct tcp_iter_state *st = seq->private; | |
2213 | int offset = st->offset; | |
2214 | int orig_num = st->num; | |
2215 | void *rc = NULL; | |
2216 | ||
2217 | switch (st->state) { | |
2218 | case TCP_SEQ_STATE_OPENREQ: | |
2219 | case TCP_SEQ_STATE_LISTENING: | |
2220 | if (st->bucket >= INET_LHTABLE_SIZE) | |
2221 | break; | |
2222 | st->state = TCP_SEQ_STATE_LISTENING; | |
2223 | rc = listening_get_next(seq, NULL); | |
2224 | while (offset-- && rc) | |
2225 | rc = listening_get_next(seq, rc); | |
2226 | if (rc) | |
2227 | break; | |
2228 | st->bucket = 0; | |
2229 | /* Fallthrough */ | |
2230 | case TCP_SEQ_STATE_ESTABLISHED: | |
2231 | case TCP_SEQ_STATE_TIME_WAIT: | |
2232 | st->state = TCP_SEQ_STATE_ESTABLISHED; | |
2233 | if (st->bucket > tcp_hashinfo.ehash_mask) | |
2234 | break; | |
2235 | rc = established_get_first(seq); | |
2236 | while (offset-- && rc) | |
2237 | rc = established_get_next(seq, rc); | |
2238 | } | |
2239 | ||
2240 | st->num = orig_num; | |
2241 | ||
2242 | return rc; | |
2243 | } | |
2244 | ||
1da177e4 LT |
2245 | static void *tcp_seq_start(struct seq_file *seq, loff_t *pos) |
2246 | { | |
5799de0b | 2247 | struct tcp_iter_state *st = seq->private; |
a8b690f9 TH |
2248 | void *rc; |
2249 | ||
2250 | if (*pos && *pos == st->last_pos) { | |
2251 | rc = tcp_seek_last_pos(seq); | |
2252 | if (rc) | |
2253 | goto out; | |
2254 | } | |
2255 | ||
1da177e4 LT |
2256 | st->state = TCP_SEQ_STATE_LISTENING; |
2257 | st->num = 0; | |
a8b690f9 TH |
2258 | st->bucket = 0; |
2259 | st->offset = 0; | |
2260 | rc = *pos ? tcp_get_idx(seq, *pos - 1) : SEQ_START_TOKEN; | |
2261 | ||
2262 | out: | |
2263 | st->last_pos = *pos; | |
2264 | return rc; | |
1da177e4 LT |
2265 | } |
2266 | ||
2267 | static void *tcp_seq_next(struct seq_file *seq, void *v, loff_t *pos) | |
2268 | { | |
a8b690f9 | 2269 | struct tcp_iter_state *st = seq->private; |
1da177e4 | 2270 | void *rc = NULL; |
1da177e4 LT |
2271 | |
2272 | if (v == SEQ_START_TOKEN) { | |
2273 | rc = tcp_get_idx(seq, 0); | |
2274 | goto out; | |
2275 | } | |
1da177e4 LT |
2276 | |
2277 | switch (st->state) { | |
2278 | case TCP_SEQ_STATE_OPENREQ: | |
2279 | case TCP_SEQ_STATE_LISTENING: | |
2280 | rc = listening_get_next(seq, v); | |
2281 | if (!rc) { | |
1da177e4 | 2282 | st->state = TCP_SEQ_STATE_ESTABLISHED; |
a8b690f9 TH |
2283 | st->bucket = 0; |
2284 | st->offset = 0; | |
1da177e4 LT |
2285 | rc = established_get_first(seq); |
2286 | } | |
2287 | break; | |
2288 | case TCP_SEQ_STATE_ESTABLISHED: | |
2289 | case TCP_SEQ_STATE_TIME_WAIT: | |
2290 | rc = established_get_next(seq, v); | |
2291 | break; | |
2292 | } | |
2293 | out: | |
2294 | ++*pos; | |
a8b690f9 | 2295 | st->last_pos = *pos; |
1da177e4 LT |
2296 | return rc; |
2297 | } | |
2298 | ||
2299 | static void tcp_seq_stop(struct seq_file *seq, void *v) | |
2300 | { | |
5799de0b | 2301 | struct tcp_iter_state *st = seq->private; |
1da177e4 LT |
2302 | |
2303 | switch (st->state) { | |
2304 | case TCP_SEQ_STATE_OPENREQ: | |
2305 | if (v) { | |
463c84b9 ACM |
2306 | struct inet_connection_sock *icsk = inet_csk(st->syn_wait_sk); |
2307 | read_unlock_bh(&icsk->icsk_accept_queue.syn_wait_lock); | |
1da177e4 LT |
2308 | } |
2309 | case TCP_SEQ_STATE_LISTENING: | |
2310 | if (v != SEQ_START_TOKEN) | |
5caea4ea | 2311 | spin_unlock_bh(&tcp_hashinfo.listening_hash[st->bucket].lock); |
1da177e4 LT |
2312 | break; |
2313 | case TCP_SEQ_STATE_TIME_WAIT: | |
2314 | case TCP_SEQ_STATE_ESTABLISHED: | |
2315 | if (v) | |
9db66bdc | 2316 | spin_unlock_bh(inet_ehash_lockp(&tcp_hashinfo, st->bucket)); |
1da177e4 LT |
2317 | break; |
2318 | } | |
2319 | } | |
2320 | ||
73cb88ec | 2321 | int tcp_seq_open(struct inode *inode, struct file *file) |
1da177e4 LT |
2322 | { |
2323 | struct tcp_seq_afinfo *afinfo = PDE(inode)->data; | |
1da177e4 | 2324 | struct tcp_iter_state *s; |
52d6f3f1 | 2325 | int err; |
1da177e4 | 2326 | |
52d6f3f1 DL |
2327 | err = seq_open_net(inode, file, &afinfo->seq_ops, |
2328 | sizeof(struct tcp_iter_state)); | |
2329 | if (err < 0) | |
2330 | return err; | |
f40c8174 | 2331 | |
52d6f3f1 | 2332 | s = ((struct seq_file *)file->private_data)->private; |
1da177e4 | 2333 | s->family = afinfo->family; |
a8b690f9 | 2334 | s->last_pos = 0; |
f40c8174 DL |
2335 | return 0; |
2336 | } | |
73cb88ec | 2337 | EXPORT_SYMBOL(tcp_seq_open); |
f40c8174 | 2338 | |
6f8b13bc | 2339 | int tcp_proc_register(struct net *net, struct tcp_seq_afinfo *afinfo) |
1da177e4 LT |
2340 | { |
2341 | int rc = 0; | |
2342 | struct proc_dir_entry *p; | |
2343 | ||
9427c4b3 DL |
2344 | afinfo->seq_ops.start = tcp_seq_start; |
2345 | afinfo->seq_ops.next = tcp_seq_next; | |
2346 | afinfo->seq_ops.stop = tcp_seq_stop; | |
2347 | ||
84841c3c | 2348 | p = proc_create_data(afinfo->name, S_IRUGO, net->proc_net, |
73cb88ec | 2349 | afinfo->seq_fops, afinfo); |
84841c3c | 2350 | if (!p) |
1da177e4 LT |
2351 | rc = -ENOMEM; |
2352 | return rc; | |
2353 | } | |
4bc2f18b | 2354 | EXPORT_SYMBOL(tcp_proc_register); |
1da177e4 | 2355 | |
6f8b13bc | 2356 | void tcp_proc_unregister(struct net *net, struct tcp_seq_afinfo *afinfo) |
1da177e4 | 2357 | { |
6f8b13bc | 2358 | proc_net_remove(net, afinfo->name); |
1da177e4 | 2359 | } |
4bc2f18b | 2360 | EXPORT_SYMBOL(tcp_proc_unregister); |
1da177e4 | 2361 | |
cf533ea5 | 2362 | static void get_openreq4(const struct sock *sk, const struct request_sock *req, |
5e659e4c | 2363 | struct seq_file *f, int i, int uid, int *len) |
1da177e4 | 2364 | { |
2e6599cb | 2365 | const struct inet_request_sock *ireq = inet_rsk(req); |
1da177e4 LT |
2366 | int ttd = req->expires - jiffies; |
2367 | ||
5e659e4c | 2368 | seq_printf(f, "%4d: %08X:%04X %08X:%04X" |
71338aa7 | 2369 | " %02X %08X:%08X %02X:%08lX %08X %5d %8d %u %d %pK%n", |
1da177e4 | 2370 | i, |
2e6599cb | 2371 | ireq->loc_addr, |
c720c7e8 | 2372 | ntohs(inet_sk(sk)->inet_sport), |
2e6599cb ACM |
2373 | ireq->rmt_addr, |
2374 | ntohs(ireq->rmt_port), | |
1da177e4 LT |
2375 | TCP_SYN_RECV, |
2376 | 0, 0, /* could print option size, but that is af dependent. */ | |
2377 | 1, /* timers active (only the expire timer) */ | |
2378 | jiffies_to_clock_t(ttd), | |
2379 | req->retrans, | |
2380 | uid, | |
2381 | 0, /* non standard timer */ | |
2382 | 0, /* open_requests have no inode */ | |
2383 | atomic_read(&sk->sk_refcnt), | |
5e659e4c PE |
2384 | req, |
2385 | len); | |
1da177e4 LT |
2386 | } |
2387 | ||
5e659e4c | 2388 | static void get_tcp4_sock(struct sock *sk, struct seq_file *f, int i, int *len) |
1da177e4 LT |
2389 | { |
2390 | int timer_active; | |
2391 | unsigned long timer_expires; | |
cf533ea5 | 2392 | const struct tcp_sock *tp = tcp_sk(sk); |
cf4c6bf8 | 2393 | const struct inet_connection_sock *icsk = inet_csk(sk); |
cf533ea5 | 2394 | const struct inet_sock *inet = inet_sk(sk); |
c720c7e8 ED |
2395 | __be32 dest = inet->inet_daddr; |
2396 | __be32 src = inet->inet_rcv_saddr; | |
2397 | __u16 destp = ntohs(inet->inet_dport); | |
2398 | __u16 srcp = ntohs(inet->inet_sport); | |
49d09007 | 2399 | int rx_queue; |
1da177e4 | 2400 | |
463c84b9 | 2401 | if (icsk->icsk_pending == ICSK_TIME_RETRANS) { |
1da177e4 | 2402 | timer_active = 1; |
463c84b9 ACM |
2403 | timer_expires = icsk->icsk_timeout; |
2404 | } else if (icsk->icsk_pending == ICSK_TIME_PROBE0) { | |
1da177e4 | 2405 | timer_active = 4; |
463c84b9 | 2406 | timer_expires = icsk->icsk_timeout; |
cf4c6bf8 | 2407 | } else if (timer_pending(&sk->sk_timer)) { |
1da177e4 | 2408 | timer_active = 2; |
cf4c6bf8 | 2409 | timer_expires = sk->sk_timer.expires; |
1da177e4 LT |
2410 | } else { |
2411 | timer_active = 0; | |
2412 | timer_expires = jiffies; | |
2413 | } | |
2414 | ||
49d09007 ED |
2415 | if (sk->sk_state == TCP_LISTEN) |
2416 | rx_queue = sk->sk_ack_backlog; | |
2417 | else | |
2418 | /* | |
2419 | * because we dont lock socket, we might find a transient negative value | |
2420 | */ | |
2421 | rx_queue = max_t(int, tp->rcv_nxt - tp->copied_seq, 0); | |
2422 | ||
5e659e4c | 2423 | seq_printf(f, "%4d: %08X:%04X %08X:%04X %02X %08X:%08X %02X:%08lX " |
71338aa7 | 2424 | "%08X %5d %8d %lu %d %pK %lu %lu %u %u %d%n", |
cf4c6bf8 | 2425 | i, src, srcp, dest, destp, sk->sk_state, |
47da8ee6 | 2426 | tp->write_seq - tp->snd_una, |
49d09007 | 2427 | rx_queue, |
1da177e4 LT |
2428 | timer_active, |
2429 | jiffies_to_clock_t(timer_expires - jiffies), | |
463c84b9 | 2430 | icsk->icsk_retransmits, |
cf4c6bf8 | 2431 | sock_i_uid(sk), |
6687e988 | 2432 | icsk->icsk_probes_out, |
cf4c6bf8 IJ |
2433 | sock_i_ino(sk), |
2434 | atomic_read(&sk->sk_refcnt), sk, | |
7be87351 SH |
2435 | jiffies_to_clock_t(icsk->icsk_rto), |
2436 | jiffies_to_clock_t(icsk->icsk_ack.ato), | |
463c84b9 | 2437 | (icsk->icsk_ack.quick << 1) | icsk->icsk_ack.pingpong, |
1da177e4 | 2438 | tp->snd_cwnd, |
0b6a05c1 | 2439 | tcp_in_initial_slowstart(tp) ? -1 : tp->snd_ssthresh, |
5e659e4c | 2440 | len); |
1da177e4 LT |
2441 | } |
2442 | ||
cf533ea5 | 2443 | static void get_timewait4_sock(const struct inet_timewait_sock *tw, |
5e659e4c | 2444 | struct seq_file *f, int i, int *len) |
1da177e4 | 2445 | { |
23f33c2d | 2446 | __be32 dest, src; |
1da177e4 LT |
2447 | __u16 destp, srcp; |
2448 | int ttd = tw->tw_ttd - jiffies; | |
2449 | ||
2450 | if (ttd < 0) | |
2451 | ttd = 0; | |
2452 | ||
2453 | dest = tw->tw_daddr; | |
2454 | src = tw->tw_rcv_saddr; | |
2455 | destp = ntohs(tw->tw_dport); | |
2456 | srcp = ntohs(tw->tw_sport); | |
2457 | ||
5e659e4c | 2458 | seq_printf(f, "%4d: %08X:%04X %08X:%04X" |
71338aa7 | 2459 | " %02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %pK%n", |
1da177e4 LT |
2460 | i, src, srcp, dest, destp, tw->tw_substate, 0, 0, |
2461 | 3, jiffies_to_clock_t(ttd), 0, 0, 0, 0, | |
5e659e4c | 2462 | atomic_read(&tw->tw_refcnt), tw, len); |
1da177e4 LT |
2463 | } |
2464 | ||
2465 | #define TMPSZ 150 | |
2466 | ||
2467 | static int tcp4_seq_show(struct seq_file *seq, void *v) | |
2468 | { | |
5799de0b | 2469 | struct tcp_iter_state *st; |
5e659e4c | 2470 | int len; |
1da177e4 LT |
2471 | |
2472 | if (v == SEQ_START_TOKEN) { | |
2473 | seq_printf(seq, "%-*s\n", TMPSZ - 1, | |
2474 | " sl local_address rem_address st tx_queue " | |
2475 | "rx_queue tr tm->when retrnsmt uid timeout " | |
2476 | "inode"); | |
2477 | goto out; | |
2478 | } | |
2479 | st = seq->private; | |
2480 | ||
2481 | switch (st->state) { | |
2482 | case TCP_SEQ_STATE_LISTENING: | |
2483 | case TCP_SEQ_STATE_ESTABLISHED: | |
5e659e4c | 2484 | get_tcp4_sock(v, seq, st->num, &len); |
1da177e4 LT |
2485 | break; |
2486 | case TCP_SEQ_STATE_OPENREQ: | |
5e659e4c | 2487 | get_openreq4(st->syn_wait_sk, v, seq, st->num, st->uid, &len); |
1da177e4 LT |
2488 | break; |
2489 | case TCP_SEQ_STATE_TIME_WAIT: | |
5e659e4c | 2490 | get_timewait4_sock(v, seq, st->num, &len); |
1da177e4 LT |
2491 | break; |
2492 | } | |
5e659e4c | 2493 | seq_printf(seq, "%*s\n", TMPSZ - 1 - len, ""); |
1da177e4 LT |
2494 | out: |
2495 | return 0; | |
2496 | } | |
2497 | ||
73cb88ec AV |
2498 | static const struct file_operations tcp_afinfo_seq_fops = { |
2499 | .owner = THIS_MODULE, | |
2500 | .open = tcp_seq_open, | |
2501 | .read = seq_read, | |
2502 | .llseek = seq_lseek, | |
2503 | .release = seq_release_net | |
2504 | }; | |
2505 | ||
1da177e4 | 2506 | static struct tcp_seq_afinfo tcp4_seq_afinfo = { |
1da177e4 LT |
2507 | .name = "tcp", |
2508 | .family = AF_INET, | |
73cb88ec | 2509 | .seq_fops = &tcp_afinfo_seq_fops, |
9427c4b3 DL |
2510 | .seq_ops = { |
2511 | .show = tcp4_seq_show, | |
2512 | }, | |
1da177e4 LT |
2513 | }; |
2514 | ||
2c8c1e72 | 2515 | static int __net_init tcp4_proc_init_net(struct net *net) |
757764f6 PE |
2516 | { |
2517 | return tcp_proc_register(net, &tcp4_seq_afinfo); | |
2518 | } | |
2519 | ||
2c8c1e72 | 2520 | static void __net_exit tcp4_proc_exit_net(struct net *net) |
757764f6 PE |
2521 | { |
2522 | tcp_proc_unregister(net, &tcp4_seq_afinfo); | |
2523 | } | |
2524 | ||
2525 | static struct pernet_operations tcp4_net_ops = { | |
2526 | .init = tcp4_proc_init_net, | |
2527 | .exit = tcp4_proc_exit_net, | |
2528 | }; | |
2529 | ||
1da177e4 LT |
2530 | int __init tcp4_proc_init(void) |
2531 | { | |
757764f6 | 2532 | return register_pernet_subsys(&tcp4_net_ops); |
1da177e4 LT |
2533 | } |
2534 | ||
2535 | void tcp4_proc_exit(void) | |
2536 | { | |
757764f6 | 2537 | unregister_pernet_subsys(&tcp4_net_ops); |
1da177e4 LT |
2538 | } |
2539 | #endif /* CONFIG_PROC_FS */ | |
2540 | ||
bf296b12 HX |
2541 | struct sk_buff **tcp4_gro_receive(struct sk_buff **head, struct sk_buff *skb) |
2542 | { | |
b71d1d42 | 2543 | const struct iphdr *iph = skb_gro_network_header(skb); |
bf296b12 HX |
2544 | |
2545 | switch (skb->ip_summed) { | |
2546 | case CHECKSUM_COMPLETE: | |
86911732 | 2547 | if (!tcp_v4_check(skb_gro_len(skb), iph->saddr, iph->daddr, |
bf296b12 HX |
2548 | skb->csum)) { |
2549 | skb->ip_summed = CHECKSUM_UNNECESSARY; | |
2550 | break; | |
2551 | } | |
2552 | ||
2553 | /* fall through */ | |
2554 | case CHECKSUM_NONE: | |
2555 | NAPI_GRO_CB(skb)->flush = 1; | |
2556 | return NULL; | |
2557 | } | |
2558 | ||
2559 | return tcp_gro_receive(head, skb); | |
2560 | } | |
bf296b12 HX |
2561 | |
2562 | int tcp4_gro_complete(struct sk_buff *skb) | |
2563 | { | |
b71d1d42 | 2564 | const struct iphdr *iph = ip_hdr(skb); |
bf296b12 HX |
2565 | struct tcphdr *th = tcp_hdr(skb); |
2566 | ||
2567 | th->check = ~tcp_v4_check(skb->len - skb_transport_offset(skb), | |
2568 | iph->saddr, iph->daddr, 0); | |
2569 | skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4; | |
2570 | ||
2571 | return tcp_gro_complete(skb); | |
2572 | } | |
bf296b12 | 2573 | |
1da177e4 LT |
2574 | struct proto tcp_prot = { |
2575 | .name = "TCP", | |
2576 | .owner = THIS_MODULE, | |
2577 | .close = tcp_close, | |
2578 | .connect = tcp_v4_connect, | |
2579 | .disconnect = tcp_disconnect, | |
463c84b9 | 2580 | .accept = inet_csk_accept, |
1da177e4 LT |
2581 | .ioctl = tcp_ioctl, |
2582 | .init = tcp_v4_init_sock, | |
2583 | .destroy = tcp_v4_destroy_sock, | |
2584 | .shutdown = tcp_shutdown, | |
2585 | .setsockopt = tcp_setsockopt, | |
2586 | .getsockopt = tcp_getsockopt, | |
1da177e4 | 2587 | .recvmsg = tcp_recvmsg, |
7ba42910 CG |
2588 | .sendmsg = tcp_sendmsg, |
2589 | .sendpage = tcp_sendpage, | |
1da177e4 | 2590 | .backlog_rcv = tcp_v4_do_rcv, |
46d3ceab | 2591 | .release_cb = tcp_release_cb, |
ab1e0a13 ACM |
2592 | .hash = inet_hash, |
2593 | .unhash = inet_unhash, | |
2594 | .get_port = inet_csk_get_port, | |
1da177e4 LT |
2595 | .enter_memory_pressure = tcp_enter_memory_pressure, |
2596 | .sockets_allocated = &tcp_sockets_allocated, | |
0a5578cf | 2597 | .orphan_count = &tcp_orphan_count, |
1da177e4 LT |
2598 | .memory_allocated = &tcp_memory_allocated, |
2599 | .memory_pressure = &tcp_memory_pressure, | |
1da177e4 LT |
2600 | .sysctl_wmem = sysctl_tcp_wmem, |
2601 | .sysctl_rmem = sysctl_tcp_rmem, | |
2602 | .max_header = MAX_TCP_HEADER, | |
2603 | .obj_size = sizeof(struct tcp_sock), | |
3ab5aee7 | 2604 | .slab_flags = SLAB_DESTROY_BY_RCU, |
6d6ee43e | 2605 | .twsk_prot = &tcp_timewait_sock_ops, |
60236fdd | 2606 | .rsk_prot = &tcp_request_sock_ops, |
39d8cda7 | 2607 | .h.hashinfo = &tcp_hashinfo, |
7ba42910 | 2608 | .no_autobind = true, |
543d9cfe ACM |
2609 | #ifdef CONFIG_COMPAT |
2610 | .compat_setsockopt = compat_tcp_setsockopt, | |
2611 | .compat_getsockopt = compat_tcp_getsockopt, | |
2612 | #endif | |
d1a4c0b3 GC |
2613 | #ifdef CONFIG_CGROUP_MEM_RES_CTLR_KMEM |
2614 | .init_cgroup = tcp_init_cgroup, | |
2615 | .destroy_cgroup = tcp_destroy_cgroup, | |
2616 | .proto_cgroup = tcp_proto_cgroup, | |
2617 | #endif | |
1da177e4 | 2618 | }; |
4bc2f18b | 2619 | EXPORT_SYMBOL(tcp_prot); |
1da177e4 | 2620 | |
046ee902 DL |
2621 | static int __net_init tcp_sk_init(struct net *net) |
2622 | { | |
2623 | return inet_ctl_sock_create(&net->ipv4.tcp_sock, | |
2624 | PF_INET, SOCK_RAW, IPPROTO_TCP, net); | |
2625 | } | |
2626 | ||
2627 | static void __net_exit tcp_sk_exit(struct net *net) | |
2628 | { | |
2629 | inet_ctl_sock_destroy(net->ipv4.tcp_sock); | |
b099ce26 EB |
2630 | } |
2631 | ||
2632 | static void __net_exit tcp_sk_exit_batch(struct list_head *net_exit_list) | |
2633 | { | |
2634 | inet_twsk_purge(&tcp_hashinfo, &tcp_death_row, AF_INET); | |
046ee902 DL |
2635 | } |
2636 | ||
2637 | static struct pernet_operations __net_initdata tcp_sk_ops = { | |
b099ce26 EB |
2638 | .init = tcp_sk_init, |
2639 | .exit = tcp_sk_exit, | |
2640 | .exit_batch = tcp_sk_exit_batch, | |
046ee902 DL |
2641 | }; |
2642 | ||
9b0f976f | 2643 | void __init tcp_v4_init(void) |
1da177e4 | 2644 | { |
5caea4ea | 2645 | inet_hashinfo_init(&tcp_hashinfo); |
6a1b3054 | 2646 | if (register_pernet_subsys(&tcp_sk_ops)) |
1da177e4 | 2647 | panic("Failed to create the TCP control socket.\n"); |
1da177e4 | 2648 | } |