Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * INET An implementation of the TCP/IP protocol suite for the LINUX | |
3 | * operating system. INET is implemented using the BSD Socket | |
4 | * interface as the means of communication with the user level. | |
5 | * | |
6 | * Implementation of the Transmission Control Protocol(TCP). | |
7 | * | |
1da177e4 LT |
8 | * IPv4 specific functions |
9 | * | |
10 | * | |
11 | * code split from: | |
12 | * linux/ipv4/tcp.c | |
13 | * linux/ipv4/tcp_input.c | |
14 | * linux/ipv4/tcp_output.c | |
15 | * | |
16 | * See tcp.c for author information | |
17 | * | |
18 | * This program is free software; you can redistribute it and/or | |
19 | * modify it under the terms of the GNU General Public License | |
20 | * as published by the Free Software Foundation; either version | |
21 | * 2 of the License, or (at your option) any later version. | |
22 | */ | |
23 | ||
24 | /* | |
25 | * Changes: | |
26 | * David S. Miller : New socket lookup architecture. | |
27 | * This code is dedicated to John Dyson. | |
28 | * David S. Miller : Change semantics of established hash, | |
29 | * half is devoted to TIME_WAIT sockets | |
30 | * and the rest go in the other half. | |
31 | * Andi Kleen : Add support for syncookies and fixed | |
32 | * some bugs: ip options weren't passed to | |
33 | * the TCP layer, missed a check for an | |
34 | * ACK bit. | |
35 | * Andi Kleen : Implemented fast path mtu discovery. | |
36 | * Fixed many serious bugs in the | |
60236fdd | 37 | * request_sock handling and moved |
1da177e4 LT |
38 | * most of it into the af independent code. |
39 | * Added tail drop and some other bugfixes. | |
caa20d9a | 40 | * Added new listen semantics. |
1da177e4 LT |
41 | * Mike McLagan : Routing by source |
42 | * Juan Jose Ciarlante: ip_dynaddr bits | |
43 | * Andi Kleen: various fixes. | |
44 | * Vitaly E. Lavrov : Transparent proxy revived after year | |
45 | * coma. | |
46 | * Andi Kleen : Fix new listen. | |
47 | * Andi Kleen : Fix accept error reporting. | |
48 | * YOSHIFUJI Hideaki @USAGI and: Support IPV6_V6ONLY socket option, which | |
49 | * Alexey Kuznetsov allow both IPv4 and IPv6 sockets to bind | |
50 | * a single port at the same time. | |
51 | */ | |
52 | ||
afd46503 | 53 | #define pr_fmt(fmt) "TCP: " fmt |
1da177e4 | 54 | |
eb4dea58 | 55 | #include <linux/bottom_half.h> |
1da177e4 LT |
56 | #include <linux/types.h> |
57 | #include <linux/fcntl.h> | |
58 | #include <linux/module.h> | |
59 | #include <linux/random.h> | |
60 | #include <linux/cache.h> | |
61 | #include <linux/jhash.h> | |
62 | #include <linux/init.h> | |
63 | #include <linux/times.h> | |
5a0e3ad6 | 64 | #include <linux/slab.h> |
1da177e4 | 65 | |
457c4cbc | 66 | #include <net/net_namespace.h> |
1da177e4 | 67 | #include <net/icmp.h> |
304a1618 | 68 | #include <net/inet_hashtables.h> |
1da177e4 | 69 | #include <net/tcp.h> |
20380731 | 70 | #include <net/transp_v6.h> |
1da177e4 LT |
71 | #include <net/ipv6.h> |
72 | #include <net/inet_common.h> | |
6d6ee43e | 73 | #include <net/timewait_sock.h> |
1da177e4 | 74 | #include <net/xfrm.h> |
6e5714ea | 75 | #include <net/secure_seq.h> |
076bb0c8 | 76 | #include <net/busy_poll.h> |
1da177e4 LT |
77 | |
78 | #include <linux/inet.h> | |
79 | #include <linux/ipv6.h> | |
80 | #include <linux/stddef.h> | |
81 | #include <linux/proc_fs.h> | |
82 | #include <linux/seq_file.h> | |
6797318e | 83 | #include <linux/inetdevice.h> |
1da177e4 | 84 | |
cf80e0e4 | 85 | #include <crypto/hash.h> |
cfb6eeb4 YH |
86 | #include <linux/scatterlist.h> |
87 | ||
c24b14c4 SL |
88 | #include <trace/events/tcp.h> |
89 | ||
cfb6eeb4 | 90 | #ifdef CONFIG_TCP_MD5SIG |
a915da9b | 91 | static int tcp_v4_md5_hash_hdr(char *md5_hash, const struct tcp_md5sig_key *key, |
318cf7aa | 92 | __be32 daddr, __be32 saddr, const struct tcphdr *th); |
cfb6eeb4 YH |
93 | #endif |
94 | ||
5caea4ea | 95 | struct inet_hashinfo tcp_hashinfo; |
4bc2f18b | 96 | EXPORT_SYMBOL(tcp_hashinfo); |
1da177e4 | 97 | |
84b114b9 | 98 | static u32 tcp_v4_init_seq(const struct sk_buff *skb) |
1da177e4 | 99 | { |
84b114b9 ED |
100 | return secure_tcp_seq(ip_hdr(skb)->daddr, |
101 | ip_hdr(skb)->saddr, | |
102 | tcp_hdr(skb)->dest, | |
103 | tcp_hdr(skb)->source); | |
104 | } | |
105 | ||
5d2ed052 | 106 | static u32 tcp_v4_init_ts_off(const struct net *net, const struct sk_buff *skb) |
84b114b9 | 107 | { |
5d2ed052 | 108 | return secure_tcp_ts_off(net, ip_hdr(skb)->daddr, ip_hdr(skb)->saddr); |
1da177e4 LT |
109 | } |
110 | ||
6d6ee43e ACM |
111 | int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp) |
112 | { | |
79e9fed4 | 113 | const struct inet_timewait_sock *tw = inet_twsk(sktw); |
6d6ee43e ACM |
114 | const struct tcp_timewait_sock *tcptw = tcp_twsk(sktw); |
115 | struct tcp_sock *tp = tcp_sk(sk); | |
79e9fed4 MŻ |
116 | int reuse = sock_net(sk)->ipv4.sysctl_tcp_tw_reuse; |
117 | ||
118 | if (reuse == 2) { | |
119 | /* Still does not detect *everything* that goes through | |
120 | * lo, since we require a loopback src or dst address | |
121 | * or direct binding to 'lo' interface. | |
122 | */ | |
123 | bool loopback = false; | |
124 | if (tw->tw_bound_dev_if == LOOPBACK_IFINDEX) | |
125 | loopback = true; | |
126 | #if IS_ENABLED(CONFIG_IPV6) | |
127 | if (tw->tw_family == AF_INET6) { | |
128 | if (ipv6_addr_loopback(&tw->tw_v6_daddr) || | |
129 | (ipv6_addr_v4mapped(&tw->tw_v6_daddr) && | |
130 | (tw->tw_v6_daddr.s6_addr[12] == 127)) || | |
131 | ipv6_addr_loopback(&tw->tw_v6_rcv_saddr) || | |
132 | (ipv6_addr_v4mapped(&tw->tw_v6_rcv_saddr) && | |
133 | (tw->tw_v6_rcv_saddr.s6_addr[12] == 127))) | |
134 | loopback = true; | |
135 | } else | |
136 | #endif | |
137 | { | |
138 | if (ipv4_is_loopback(tw->tw_daddr) || | |
139 | ipv4_is_loopback(tw->tw_rcv_saddr)) | |
140 | loopback = true; | |
141 | } | |
142 | if (!loopback) | |
143 | reuse = 0; | |
144 | } | |
6d6ee43e ACM |
145 | |
146 | /* With PAWS, it is safe from the viewpoint | |
147 | of data integrity. Even without PAWS it is safe provided sequence | |
148 | spaces do not overlap i.e. at data rates <= 80Mbit/sec. | |
149 | ||
150 | Actually, the idea is close to VJ's one, only timestamp cache is | |
151 | held not per host, but per port pair and TW bucket is used as state | |
152 | holder. | |
153 | ||
154 | If TW bucket has been already destroyed we fall back to VJ's scheme | |
155 | and use initial timestamp retrieved from peer table. | |
156 | */ | |
157 | if (tcptw->tw_ts_recent_stamp && | |
cca9bab1 AB |
158 | (!twp || (reuse && time_after32(ktime_get_seconds(), |
159 | tcptw->tw_ts_recent_stamp)))) { | |
21684dc4 SB |
160 | /* In case of repair and re-using TIME-WAIT sockets we still |
161 | * want to be sure that it is safe as above but honor the | |
162 | * sequence numbers and time stamps set as part of the repair | |
163 | * process. | |
164 | * | |
165 | * Without this check re-using a TIME-WAIT socket with TCP | |
166 | * repair would accumulate a -1 on the repair assigned | |
167 | * sequence number. The first time it is reused the sequence | |
168 | * is -1, the second time -2, etc. This fixes that issue | |
169 | * without appearing to create any others. | |
170 | */ | |
171 | if (likely(!tp->repair)) { | |
172 | tp->write_seq = tcptw->tw_snd_nxt + 65535 + 2; | |
173 | if (tp->write_seq == 0) | |
174 | tp->write_seq = 1; | |
175 | tp->rx_opt.ts_recent = tcptw->tw_ts_recent; | |
176 | tp->rx_opt.ts_recent_stamp = tcptw->tw_ts_recent_stamp; | |
177 | } | |
6d6ee43e ACM |
178 | sock_hold(sktw); |
179 | return 1; | |
180 | } | |
181 | ||
182 | return 0; | |
183 | } | |
6d6ee43e ACM |
184 | EXPORT_SYMBOL_GPL(tcp_twsk_unique); |
185 | ||
d74bad4e AI |
186 | static int tcp_v4_pre_connect(struct sock *sk, struct sockaddr *uaddr, |
187 | int addr_len) | |
188 | { | |
189 | /* This check is replicated from tcp_v4_connect() and intended to | |
190 | * prevent BPF program called below from accessing bytes that are out | |
191 | * of the bound specified by user in addr_len. | |
192 | */ | |
193 | if (addr_len < sizeof(struct sockaddr_in)) | |
194 | return -EINVAL; | |
195 | ||
196 | sock_owned_by_me(sk); | |
197 | ||
198 | return BPF_CGROUP_RUN_PROG_INET4_CONNECT(sk, uaddr); | |
199 | } | |
200 | ||
1da177e4 LT |
201 | /* This will initiate an outgoing connection. */ |
202 | int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len) | |
203 | { | |
2d7192d6 | 204 | struct sockaddr_in *usin = (struct sockaddr_in *)uaddr; |
1da177e4 LT |
205 | struct inet_sock *inet = inet_sk(sk); |
206 | struct tcp_sock *tp = tcp_sk(sk); | |
dca8b089 | 207 | __be16 orig_sport, orig_dport; |
bada8adc | 208 | __be32 daddr, nexthop; |
da905bd1 | 209 | struct flowi4 *fl4; |
2d7192d6 | 210 | struct rtable *rt; |
1da177e4 | 211 | int err; |
f6d8bd05 | 212 | struct ip_options_rcu *inet_opt; |
1946e672 | 213 | struct inet_timewait_death_row *tcp_death_row = &sock_net(sk)->ipv4.tcp_death_row; |
1da177e4 LT |
214 | |
215 | if (addr_len < sizeof(struct sockaddr_in)) | |
216 | return -EINVAL; | |
217 | ||
218 | if (usin->sin_family != AF_INET) | |
219 | return -EAFNOSUPPORT; | |
220 | ||
221 | nexthop = daddr = usin->sin_addr.s_addr; | |
f6d8bd05 | 222 | inet_opt = rcu_dereference_protected(inet->inet_opt, |
1e1d04e6 | 223 | lockdep_sock_is_held(sk)); |
f6d8bd05 | 224 | if (inet_opt && inet_opt->opt.srr) { |
1da177e4 LT |
225 | if (!daddr) |
226 | return -EINVAL; | |
f6d8bd05 | 227 | nexthop = inet_opt->opt.faddr; |
1da177e4 LT |
228 | } |
229 | ||
dca8b089 DM |
230 | orig_sport = inet->inet_sport; |
231 | orig_dport = usin->sin_port; | |
da905bd1 DM |
232 | fl4 = &inet->cork.fl.u.ip4; |
233 | rt = ip_route_connect(fl4, nexthop, inet->inet_saddr, | |
b23dd4fe DM |
234 | RT_CONN_FLAGS(sk), sk->sk_bound_dev_if, |
235 | IPPROTO_TCP, | |
0e0d44ab | 236 | orig_sport, orig_dport, sk); |
b23dd4fe DM |
237 | if (IS_ERR(rt)) { |
238 | err = PTR_ERR(rt); | |
239 | if (err == -ENETUNREACH) | |
f1d8cba6 | 240 | IP_INC_STATS(sock_net(sk), IPSTATS_MIB_OUTNOROUTES); |
b23dd4fe | 241 | return err; |
584bdf8c | 242 | } |
1da177e4 LT |
243 | |
244 | if (rt->rt_flags & (RTCF_MULTICAST | RTCF_BROADCAST)) { | |
245 | ip_rt_put(rt); | |
246 | return -ENETUNREACH; | |
247 | } | |
248 | ||
f6d8bd05 | 249 | if (!inet_opt || !inet_opt->opt.srr) |
da905bd1 | 250 | daddr = fl4->daddr; |
1da177e4 | 251 | |
c720c7e8 | 252 | if (!inet->inet_saddr) |
da905bd1 | 253 | inet->inet_saddr = fl4->saddr; |
d1e559d0 | 254 | sk_rcv_saddr_set(sk, inet->inet_saddr); |
1da177e4 | 255 | |
c720c7e8 | 256 | if (tp->rx_opt.ts_recent_stamp && inet->inet_daddr != daddr) { |
1da177e4 LT |
257 | /* Reset inherited state */ |
258 | tp->rx_opt.ts_recent = 0; | |
259 | tp->rx_opt.ts_recent_stamp = 0; | |
ee995283 PE |
260 | if (likely(!tp->repair)) |
261 | tp->write_seq = 0; | |
1da177e4 LT |
262 | } |
263 | ||
c720c7e8 | 264 | inet->inet_dport = usin->sin_port; |
d1e559d0 | 265 | sk_daddr_set(sk, daddr); |
1da177e4 | 266 | |
d83d8461 | 267 | inet_csk(sk)->icsk_ext_hdr_len = 0; |
f6d8bd05 ED |
268 | if (inet_opt) |
269 | inet_csk(sk)->icsk_ext_hdr_len = inet_opt->opt.optlen; | |
1da177e4 | 270 | |
bee7ca9e | 271 | tp->rx_opt.mss_clamp = TCP_MSS_DEFAULT; |
1da177e4 LT |
272 | |
273 | /* Socket identity is still unknown (sport may be zero). | |
274 | * However we set state to SYN-SENT and not releasing socket | |
275 | * lock select source port, enter ourselves into the hash tables and | |
276 | * complete initialization after this. | |
277 | */ | |
278 | tcp_set_state(sk, TCP_SYN_SENT); | |
1946e672 | 279 | err = inet_hash_connect(tcp_death_row, sk); |
1da177e4 LT |
280 | if (err) |
281 | goto failure; | |
282 | ||
877d1f62 | 283 | sk_set_txhash(sk); |
9e7ceb06 | 284 | |
da905bd1 | 285 | rt = ip_route_newports(fl4, rt, orig_sport, orig_dport, |
b23dd4fe DM |
286 | inet->inet_sport, inet->inet_dport, sk); |
287 | if (IS_ERR(rt)) { | |
288 | err = PTR_ERR(rt); | |
289 | rt = NULL; | |
1da177e4 | 290 | goto failure; |
b23dd4fe | 291 | } |
1da177e4 | 292 | /* OK, now commit destination to socket. */ |
bcd76111 | 293 | sk->sk_gso_type = SKB_GSO_TCPV4; |
d8d1f30b | 294 | sk_setup_caps(sk, &rt->dst); |
19f6d3f3 | 295 | rt = NULL; |
1da177e4 | 296 | |
00355fa5 | 297 | if (likely(!tp->repair)) { |
00355fa5 | 298 | if (!tp->write_seq) |
84b114b9 ED |
299 | tp->write_seq = secure_tcp_seq(inet->inet_saddr, |
300 | inet->inet_daddr, | |
301 | inet->inet_sport, | |
302 | usin->sin_port); | |
5d2ed052 ED |
303 | tp->tsoffset = secure_tcp_ts_off(sock_net(sk), |
304 | inet->inet_saddr, | |
84b114b9 | 305 | inet->inet_daddr); |
00355fa5 | 306 | } |
1da177e4 | 307 | |
c720c7e8 | 308 | inet->inet_id = tp->write_seq ^ jiffies; |
1da177e4 | 309 | |
19f6d3f3 WW |
310 | if (tcp_fastopen_defer_connect(sk, &err)) |
311 | return err; | |
312 | if (err) | |
313 | goto failure; | |
314 | ||
2b916477 | 315 | err = tcp_connect(sk); |
ee995283 | 316 | |
1da177e4 LT |
317 | if (err) |
318 | goto failure; | |
319 | ||
320 | return 0; | |
321 | ||
322 | failure: | |
7174259e ACM |
323 | /* |
324 | * This unhashes the socket and releases the local port, | |
325 | * if necessary. | |
326 | */ | |
1da177e4 LT |
327 | tcp_set_state(sk, TCP_CLOSE); |
328 | ip_rt_put(rt); | |
329 | sk->sk_route_caps = 0; | |
c720c7e8 | 330 | inet->inet_dport = 0; |
1da177e4 LT |
331 | return err; |
332 | } | |
4bc2f18b | 333 | EXPORT_SYMBOL(tcp_v4_connect); |
1da177e4 | 334 | |
1da177e4 | 335 | /* |
563d34d0 ED |
336 | * This routine reacts to ICMP_FRAG_NEEDED mtu indications as defined in RFC1191. |
337 | * It can be called through tcp_release_cb() if socket was owned by user | |
338 | * at the time tcp_v4_err() was called to handle ICMP message. | |
1da177e4 | 339 | */ |
4fab9071 | 340 | void tcp_v4_mtu_reduced(struct sock *sk) |
1da177e4 | 341 | { |
1da177e4 | 342 | struct inet_sock *inet = inet_sk(sk); |
02b2faaf ED |
343 | struct dst_entry *dst; |
344 | u32 mtu; | |
1da177e4 | 345 | |
02b2faaf ED |
346 | if ((1 << sk->sk_state) & (TCPF_LISTEN | TCPF_CLOSE)) |
347 | return; | |
348 | mtu = tcp_sk(sk)->mtu_info; | |
80d0a69f DM |
349 | dst = inet_csk_update_pmtu(sk, mtu); |
350 | if (!dst) | |
1da177e4 LT |
351 | return; |
352 | ||
1da177e4 LT |
353 | /* Something is about to be wrong... Remember soft error |
354 | * for the case, if this connection will not able to recover. | |
355 | */ | |
356 | if (mtu < dst_mtu(dst) && ip_dont_fragment(sk, dst)) | |
357 | sk->sk_err_soft = EMSGSIZE; | |
358 | ||
359 | mtu = dst_mtu(dst); | |
360 | ||
361 | if (inet->pmtudisc != IP_PMTUDISC_DONT && | |
482fc609 | 362 | ip_sk_accept_pmtu(sk) && |
d83d8461 | 363 | inet_csk(sk)->icsk_pmtu_cookie > mtu) { |
1da177e4 LT |
364 | tcp_sync_mss(sk, mtu); |
365 | ||
366 | /* Resend the TCP packet because it's | |
367 | * clear that the old packet has been | |
368 | * dropped. This is the new "fast" path mtu | |
369 | * discovery. | |
370 | */ | |
371 | tcp_simple_retransmit(sk); | |
372 | } /* else let the usual retransmit timer handle it */ | |
373 | } | |
4fab9071 | 374 | EXPORT_SYMBOL(tcp_v4_mtu_reduced); |
1da177e4 | 375 | |
55be7a9c DM |
376 | static void do_redirect(struct sk_buff *skb, struct sock *sk) |
377 | { | |
378 | struct dst_entry *dst = __sk_dst_check(sk, 0); | |
379 | ||
1ed5c48f | 380 | if (dst) |
6700c270 | 381 | dst->ops->redirect(dst, sk, skb); |
55be7a9c DM |
382 | } |
383 | ||
26e37360 ED |
384 | |
385 | /* handle ICMP messages on TCP_NEW_SYN_RECV request sockets */ | |
9cf74903 | 386 | void tcp_req_err(struct sock *sk, u32 seq, bool abort) |
26e37360 ED |
387 | { |
388 | struct request_sock *req = inet_reqsk(sk); | |
389 | struct net *net = sock_net(sk); | |
390 | ||
391 | /* ICMPs are not backlogged, hence we cannot get | |
392 | * an established socket here. | |
393 | */ | |
26e37360 | 394 | if (seq != tcp_rsk(req)->snt_isn) { |
02a1d6e7 | 395 | __NET_INC_STATS(net, LINUX_MIB_OUTOFWINDOWICMPS); |
9cf74903 | 396 | } else if (abort) { |
26e37360 ED |
397 | /* |
398 | * Still in SYN_RECV, just remove it silently. | |
399 | * There is no good way to pass the error to the newly | |
400 | * created socket, and POSIX does not want network | |
401 | * errors returned from accept(). | |
402 | */ | |
c6973669 | 403 | inet_csk_reqsk_queue_drop(req->rsk_listener, req); |
9caad864 | 404 | tcp_listendrop(req->rsk_listener); |
26e37360 | 405 | } |
ef84d8ce | 406 | reqsk_put(req); |
26e37360 ED |
407 | } |
408 | EXPORT_SYMBOL(tcp_req_err); | |
409 | ||
1da177e4 LT |
410 | /* |
411 | * This routine is called by the ICMP module when it gets some | |
412 | * sort of error condition. If err < 0 then the socket should | |
413 | * be closed and the error returned to the user. If err > 0 | |
414 | * it's just the icmp type << 8 | icmp code. After adjustment | |
415 | * header points to the first 8 bytes of the tcp header. We need | |
416 | * to find the appropriate port. | |
417 | * | |
418 | * The locking strategy used here is very "optimistic". When | |
419 | * someone else accesses the socket the ICMP is just dropped | |
420 | * and for some paths there is no check at all. | |
421 | * A more general error queue to queue errors for later handling | |
422 | * is probably better. | |
423 | * | |
424 | */ | |
425 | ||
32bbd879 | 426 | int tcp_v4_err(struct sk_buff *icmp_skb, u32 info) |
1da177e4 | 427 | { |
b71d1d42 | 428 | const struct iphdr *iph = (const struct iphdr *)icmp_skb->data; |
4d1a2d9e | 429 | struct tcphdr *th = (struct tcphdr *)(icmp_skb->data + (iph->ihl << 2)); |
f1ecd5d9 | 430 | struct inet_connection_sock *icsk; |
1da177e4 LT |
431 | struct tcp_sock *tp; |
432 | struct inet_sock *inet; | |
4d1a2d9e DL |
433 | const int type = icmp_hdr(icmp_skb)->type; |
434 | const int code = icmp_hdr(icmp_skb)->code; | |
1da177e4 | 435 | struct sock *sk; |
f1ecd5d9 | 436 | struct sk_buff *skb; |
0a672f74 | 437 | struct request_sock *fastopen; |
9a568de4 ED |
438 | u32 seq, snd_una; |
439 | s32 remaining; | |
440 | u32 delta_us; | |
1da177e4 | 441 | int err; |
4d1a2d9e | 442 | struct net *net = dev_net(icmp_skb->dev); |
1da177e4 | 443 | |
26e37360 ED |
444 | sk = __inet_lookup_established(net, &tcp_hashinfo, iph->daddr, |
445 | th->dest, iph->saddr, ntohs(th->source), | |
3fa6f616 | 446 | inet_iif(icmp_skb), 0); |
1da177e4 | 447 | if (!sk) { |
5d3848bc | 448 | __ICMP_INC_STATS(net, ICMP_MIB_INERRORS); |
32bbd879 | 449 | return -ENOENT; |
1da177e4 LT |
450 | } |
451 | if (sk->sk_state == TCP_TIME_WAIT) { | |
9469c7b4 | 452 | inet_twsk_put(inet_twsk(sk)); |
32bbd879 | 453 | return 0; |
1da177e4 | 454 | } |
26e37360 | 455 | seq = ntohl(th->seq); |
32bbd879 SB |
456 | if (sk->sk_state == TCP_NEW_SYN_RECV) { |
457 | tcp_req_err(sk, seq, type == ICMP_PARAMETERPROB || | |
458 | type == ICMP_TIME_EXCEEDED || | |
459 | (type == ICMP_DEST_UNREACH && | |
460 | (code == ICMP_NET_UNREACH || | |
461 | code == ICMP_HOST_UNREACH))); | |
462 | return 0; | |
463 | } | |
1da177e4 LT |
464 | |
465 | bh_lock_sock(sk); | |
466 | /* If too many ICMPs get dropped on busy | |
467 | * servers this needs to be solved differently. | |
563d34d0 ED |
468 | * We do take care of PMTU discovery (RFC1191) special case : |
469 | * we can receive locally generated ICMP messages while socket is held. | |
1da177e4 | 470 | */ |
b74aa930 ED |
471 | if (sock_owned_by_user(sk)) { |
472 | if (!(type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED)) | |
02a1d6e7 | 473 | __NET_INC_STATS(net, LINUX_MIB_LOCKDROPPEDICMPS); |
b74aa930 | 474 | } |
1da177e4 LT |
475 | if (sk->sk_state == TCP_CLOSE) |
476 | goto out; | |
477 | ||
97e3ecd1 | 478 | if (unlikely(iph->ttl < inet_sk(sk)->min_ttl)) { |
02a1d6e7 | 479 | __NET_INC_STATS(net, LINUX_MIB_TCPMINTTLDROP); |
97e3ecd1 | 480 | goto out; |
481 | } | |
482 | ||
f1ecd5d9 | 483 | icsk = inet_csk(sk); |
1da177e4 | 484 | tp = tcp_sk(sk); |
0a672f74 YC |
485 | /* XXX (TFO) - tp->snd_una should be ISN (tcp_create_openreq_child() */ |
486 | fastopen = tp->fastopen_rsk; | |
487 | snd_una = fastopen ? tcp_rsk(fastopen)->snt_isn : tp->snd_una; | |
1da177e4 | 488 | if (sk->sk_state != TCP_LISTEN && |
0a672f74 | 489 | !between(seq, snd_una, tp->snd_nxt)) { |
02a1d6e7 | 490 | __NET_INC_STATS(net, LINUX_MIB_OUTOFWINDOWICMPS); |
1da177e4 LT |
491 | goto out; |
492 | } | |
493 | ||
494 | switch (type) { | |
55be7a9c | 495 | case ICMP_REDIRECT: |
45caeaa5 JM |
496 | if (!sock_owned_by_user(sk)) |
497 | do_redirect(icmp_skb, sk); | |
55be7a9c | 498 | goto out; |
1da177e4 LT |
499 | case ICMP_SOURCE_QUENCH: |
500 | /* Just silently ignore these. */ | |
501 | goto out; | |
502 | case ICMP_PARAMETERPROB: | |
503 | err = EPROTO; | |
504 | break; | |
505 | case ICMP_DEST_UNREACH: | |
506 | if (code > NR_ICMP_UNREACH) | |
507 | goto out; | |
508 | ||
509 | if (code == ICMP_FRAG_NEEDED) { /* PMTU discovery (RFC1191) */ | |
0d4f0608 ED |
510 | /* We are not interested in TCP_LISTEN and open_requests |
511 | * (SYN-ACKs send out by Linux are always <576bytes so | |
512 | * they should go through unfragmented). | |
513 | */ | |
514 | if (sk->sk_state == TCP_LISTEN) | |
515 | goto out; | |
516 | ||
563d34d0 | 517 | tp->mtu_info = info; |
144d56e9 | 518 | if (!sock_owned_by_user(sk)) { |
563d34d0 | 519 | tcp_v4_mtu_reduced(sk); |
144d56e9 | 520 | } else { |
7aa5470c | 521 | if (!test_and_set_bit(TCP_MTU_REDUCED_DEFERRED, &sk->sk_tsq_flags)) |
144d56e9 ED |
522 | sock_hold(sk); |
523 | } | |
1da177e4 LT |
524 | goto out; |
525 | } | |
526 | ||
527 | err = icmp_err_convert[code].errno; | |
f1ecd5d9 DL |
528 | /* check if icmp_skb allows revert of backoff |
529 | * (see draft-zimmermann-tcp-lcd) */ | |
530 | if (code != ICMP_NET_UNREACH && code != ICMP_HOST_UNREACH) | |
531 | break; | |
532 | if (seq != tp->snd_una || !icsk->icsk_retransmits || | |
0a672f74 | 533 | !icsk->icsk_backoff || fastopen) |
f1ecd5d9 DL |
534 | break; |
535 | ||
8f49c270 DM |
536 | if (sock_owned_by_user(sk)) |
537 | break; | |
538 | ||
f1ecd5d9 | 539 | icsk->icsk_backoff--; |
fcdd1cf4 ED |
540 | icsk->icsk_rto = tp->srtt_us ? __tcp_set_rto(tp) : |
541 | TCP_TIMEOUT_INIT; | |
542 | icsk->icsk_rto = inet_csk_rto_backoff(icsk, TCP_RTO_MAX); | |
f1ecd5d9 | 543 | |
75c119af | 544 | skb = tcp_rtx_queue_head(sk); |
f1ecd5d9 | 545 | |
9a568de4 | 546 | tcp_mstamp_refresh(tp); |
2fd66ffb | 547 | delta_us = (u32)(tp->tcp_mstamp - tcp_skb_timestamp_us(skb)); |
7faee5c0 | 548 | remaining = icsk->icsk_rto - |
9a568de4 | 549 | usecs_to_jiffies(delta_us); |
f1ecd5d9 | 550 | |
9a568de4 | 551 | if (remaining > 0) { |
f1ecd5d9 DL |
552 | inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS, |
553 | remaining, TCP_RTO_MAX); | |
f1ecd5d9 DL |
554 | } else { |
555 | /* RTO revert clocked out retransmission. | |
556 | * Will retransmit now */ | |
557 | tcp_retransmit_timer(sk); | |
558 | } | |
559 | ||
1da177e4 LT |
560 | break; |
561 | case ICMP_TIME_EXCEEDED: | |
562 | err = EHOSTUNREACH; | |
563 | break; | |
564 | default: | |
565 | goto out; | |
566 | } | |
567 | ||
568 | switch (sk->sk_state) { | |
1da177e4 | 569 | case TCP_SYN_SENT: |
0a672f74 YC |
570 | case TCP_SYN_RECV: |
571 | /* Only in fast or simultaneous open. If a fast open socket is | |
572 | * is already accepted it is treated as a connected one below. | |
573 | */ | |
51456b29 | 574 | if (fastopen && !fastopen->sk) |
0a672f74 YC |
575 | break; |
576 | ||
1da177e4 | 577 | if (!sock_owned_by_user(sk)) { |
1da177e4 LT |
578 | sk->sk_err = err; |
579 | ||
580 | sk->sk_error_report(sk); | |
581 | ||
582 | tcp_done(sk); | |
583 | } else { | |
584 | sk->sk_err_soft = err; | |
585 | } | |
586 | goto out; | |
587 | } | |
588 | ||
589 | /* If we've already connected we will keep trying | |
590 | * until we time out, or the user gives up. | |
591 | * | |
592 | * rfc1122 4.2.3.9 allows to consider as hard errors | |
593 | * only PROTO_UNREACH and PORT_UNREACH (well, FRAG_FAILED too, | |
594 | * but it is obsoleted by pmtu discovery). | |
595 | * | |
596 | * Note, that in modern internet, where routing is unreliable | |
597 | * and in each dark corner broken firewalls sit, sending random | |
598 | * errors ordered by their masters even this two messages finally lose | |
599 | * their original sense (even Linux sends invalid PORT_UNREACHs) | |
600 | * | |
601 | * Now we are in compliance with RFCs. | |
602 | * --ANK (980905) | |
603 | */ | |
604 | ||
605 | inet = inet_sk(sk); | |
606 | if (!sock_owned_by_user(sk) && inet->recverr) { | |
607 | sk->sk_err = err; | |
608 | sk->sk_error_report(sk); | |
609 | } else { /* Only an error on timeout */ | |
610 | sk->sk_err_soft = err; | |
611 | } | |
612 | ||
613 | out: | |
614 | bh_unlock_sock(sk); | |
615 | sock_put(sk); | |
32bbd879 | 616 | return 0; |
1da177e4 LT |
617 | } |
618 | ||
28850dc7 | 619 | void __tcp_v4_send_check(struct sk_buff *skb, __be32 saddr, __be32 daddr) |
1da177e4 | 620 | { |
aa8223c7 | 621 | struct tcphdr *th = tcp_hdr(skb); |
1da177e4 | 622 | |
98be9b12 ED |
623 | th->check = ~tcp_v4_check(skb->len, saddr, daddr, 0); |
624 | skb->csum_start = skb_transport_header(skb) - skb->head; | |
625 | skb->csum_offset = offsetof(struct tcphdr, check); | |
1da177e4 LT |
626 | } |
627 | ||
419f9f89 | 628 | /* This routine computes an IPv4 TCP checksum. */ |
bb296246 | 629 | void tcp_v4_send_check(struct sock *sk, struct sk_buff *skb) |
419f9f89 | 630 | { |
cf533ea5 | 631 | const struct inet_sock *inet = inet_sk(sk); |
419f9f89 HX |
632 | |
633 | __tcp_v4_send_check(skb, inet->inet_saddr, inet->inet_daddr); | |
634 | } | |
4bc2f18b | 635 | EXPORT_SYMBOL(tcp_v4_send_check); |
419f9f89 | 636 | |
1da177e4 LT |
637 | /* |
638 | * This routine will send an RST to the other tcp. | |
639 | * | |
640 | * Someone asks: why I NEVER use socket parameters (TOS, TTL etc.) | |
641 | * for reset. | |
642 | * Answer: if a packet caused RST, it is not for a socket | |
643 | * existing in our system, if it is matched to a socket, | |
644 | * it is just duplicate segment or bug in other side's TCP. | |
645 | * So that we build reply only basing on parameters | |
646 | * arrived with segment. | |
647 | * Exception: precedence violation. We do not implement it in any case. | |
648 | */ | |
649 | ||
a00e7444 | 650 | static void tcp_v4_send_reset(const struct sock *sk, struct sk_buff *skb) |
1da177e4 | 651 | { |
cf533ea5 | 652 | const struct tcphdr *th = tcp_hdr(skb); |
cfb6eeb4 YH |
653 | struct { |
654 | struct tcphdr th; | |
655 | #ifdef CONFIG_TCP_MD5SIG | |
714e85be | 656 | __be32 opt[(TCPOLEN_MD5SIG_ALIGNED >> 2)]; |
cfb6eeb4 YH |
657 | #endif |
658 | } rep; | |
1da177e4 | 659 | struct ip_reply_arg arg; |
cfb6eeb4 | 660 | #ifdef CONFIG_TCP_MD5SIG |
e46787f0 | 661 | struct tcp_md5sig_key *key = NULL; |
658ddaaf SL |
662 | const __u8 *hash_location = NULL; |
663 | unsigned char newhash[16]; | |
664 | int genhash; | |
665 | struct sock *sk1 = NULL; | |
cfb6eeb4 | 666 | #endif |
a86b1e30 | 667 | struct net *net; |
00483690 | 668 | struct sock *ctl_sk; |
1da177e4 LT |
669 | |
670 | /* Never send a reset in response to a reset. */ | |
671 | if (th->rst) | |
672 | return; | |
673 | ||
c3658e8d ED |
674 | /* If sk not NULL, it means we did a successful lookup and incoming |
675 | * route had to be correct. prequeue might have dropped our dst. | |
676 | */ | |
677 | if (!sk && skb_rtable(skb)->rt_type != RTN_LOCAL) | |
1da177e4 LT |
678 | return; |
679 | ||
680 | /* Swap the send and the receive. */ | |
cfb6eeb4 YH |
681 | memset(&rep, 0, sizeof(rep)); |
682 | rep.th.dest = th->source; | |
683 | rep.th.source = th->dest; | |
684 | rep.th.doff = sizeof(struct tcphdr) / 4; | |
685 | rep.th.rst = 1; | |
1da177e4 LT |
686 | |
687 | if (th->ack) { | |
cfb6eeb4 | 688 | rep.th.seq = th->ack_seq; |
1da177e4 | 689 | } else { |
cfb6eeb4 YH |
690 | rep.th.ack = 1; |
691 | rep.th.ack_seq = htonl(ntohl(th->seq) + th->syn + th->fin + | |
692 | skb->len - (th->doff << 2)); | |
1da177e4 LT |
693 | } |
694 | ||
7174259e | 695 | memset(&arg, 0, sizeof(arg)); |
cfb6eeb4 YH |
696 | arg.iov[0].iov_base = (unsigned char *)&rep; |
697 | arg.iov[0].iov_len = sizeof(rep.th); | |
698 | ||
0f85feae | 699 | net = sk ? sock_net(sk) : dev_net(skb_dst(skb)->dev); |
cfb6eeb4 | 700 | #ifdef CONFIG_TCP_MD5SIG |
3b24d854 | 701 | rcu_read_lock(); |
658ddaaf | 702 | hash_location = tcp_parse_md5sig_option(th); |
271c3b9b | 703 | if (sk && sk_fullsock(sk)) { |
e46787f0 FW |
704 | key = tcp_md5_do_lookup(sk, (union tcp_md5_addr *) |
705 | &ip_hdr(skb)->saddr, AF_INET); | |
706 | } else if (hash_location) { | |
658ddaaf SL |
707 | /* |
708 | * active side is lost. Try to find listening socket through | |
709 | * source port, and then find md5 key through listening socket. | |
710 | * we are not loose security here: | |
711 | * Incoming packet is checked with md5 hash with finding key, | |
712 | * no RST generated if md5 hash doesn't match. | |
713 | */ | |
a583636a CG |
714 | sk1 = __inet_lookup_listener(net, &tcp_hashinfo, NULL, 0, |
715 | ip_hdr(skb)->saddr, | |
da5e3630 | 716 | th->source, ip_hdr(skb)->daddr, |
3fa6f616 DA |
717 | ntohs(th->source), inet_iif(skb), |
718 | tcp_v4_sdif(skb)); | |
658ddaaf SL |
719 | /* don't send rst if it can't find key */ |
720 | if (!sk1) | |
3b24d854 ED |
721 | goto out; |
722 | ||
658ddaaf SL |
723 | key = tcp_md5_do_lookup(sk1, (union tcp_md5_addr *) |
724 | &ip_hdr(skb)->saddr, AF_INET); | |
725 | if (!key) | |
3b24d854 ED |
726 | goto out; |
727 | ||
658ddaaf | 728 | |
39f8e58e | 729 | genhash = tcp_v4_md5_hash_skb(newhash, key, NULL, skb); |
658ddaaf | 730 | if (genhash || memcmp(hash_location, newhash, 16) != 0) |
3b24d854 ED |
731 | goto out; |
732 | ||
658ddaaf SL |
733 | } |
734 | ||
cfb6eeb4 YH |
735 | if (key) { |
736 | rep.opt[0] = htonl((TCPOPT_NOP << 24) | | |
737 | (TCPOPT_NOP << 16) | | |
738 | (TCPOPT_MD5SIG << 8) | | |
739 | TCPOLEN_MD5SIG); | |
740 | /* Update length and the length the header thinks exists */ | |
741 | arg.iov[0].iov_len += TCPOLEN_MD5SIG_ALIGNED; | |
742 | rep.th.doff = arg.iov[0].iov_len / 4; | |
743 | ||
49a72dfb | 744 | tcp_v4_md5_hash_hdr((__u8 *) &rep.opt[1], |
78e645cb IJ |
745 | key, ip_hdr(skb)->saddr, |
746 | ip_hdr(skb)->daddr, &rep.th); | |
cfb6eeb4 YH |
747 | } |
748 | #endif | |
eddc9ec5 ACM |
749 | arg.csum = csum_tcpudp_nofold(ip_hdr(skb)->daddr, |
750 | ip_hdr(skb)->saddr, /* XXX */ | |
52cd5750 | 751 | arg.iov[0].iov_len, IPPROTO_TCP, 0); |
1da177e4 | 752 | arg.csumoffset = offsetof(struct tcphdr, check) / 2; |
271c3b9b FW |
753 | arg.flags = (sk && inet_sk_transparent(sk)) ? IP_REPLY_ARG_NOSRCCHECK : 0; |
754 | ||
e2446eaa | 755 | /* When socket is gone, all binding information is lost. |
4c675258 AK |
756 | * routing might fail in this case. No choice here, if we choose to force |
757 | * input interface, we will misroute in case of asymmetric route. | |
e2446eaa | 758 | */ |
c24b14c4 | 759 | if (sk) { |
4c675258 | 760 | arg.bound_dev_if = sk->sk_bound_dev_if; |
5c487bb9 SL |
761 | if (sk_fullsock(sk)) |
762 | trace_tcp_send_reset(sk, skb); | |
c24b14c4 | 763 | } |
1da177e4 | 764 | |
271c3b9b FW |
765 | BUILD_BUG_ON(offsetof(struct sock, sk_bound_dev_if) != |
766 | offsetof(struct inet_timewait_sock, tw_bound_dev_if)); | |
767 | ||
66b13d99 | 768 | arg.tos = ip_hdr(skb)->tos; |
e2d118a1 | 769 | arg.uid = sock_net_uid(net, sk && sk_fullsock(sk) ? sk : NULL); |
47dcc20a | 770 | local_bh_disable(); |
00483690 JM |
771 | ctl_sk = *this_cpu_ptr(net->ipv4.tcp_sk); |
772 | if (sk) | |
773 | ctl_sk->sk_mark = (sk->sk_state == TCP_TIME_WAIT) ? | |
774 | inet_twsk(sk)->tw_mark : sk->sk_mark; | |
775 | ip_send_unicast_reply(ctl_sk, | |
bdbbb852 | 776 | skb, &TCP_SKB_CB(skb)->header.h4.opt, |
24a2d43d ED |
777 | ip_hdr(skb)->saddr, ip_hdr(skb)->daddr, |
778 | &arg, arg.iov[0].iov_len); | |
1da177e4 | 779 | |
00483690 | 780 | ctl_sk->sk_mark = 0; |
90bbcc60 ED |
781 | __TCP_INC_STATS(net, TCP_MIB_OUTSEGS); |
782 | __TCP_INC_STATS(net, TCP_MIB_OUTRSTS); | |
47dcc20a | 783 | local_bh_enable(); |
658ddaaf SL |
784 | |
785 | #ifdef CONFIG_TCP_MD5SIG | |
3b24d854 ED |
786 | out: |
787 | rcu_read_unlock(); | |
658ddaaf | 788 | #endif |
1da177e4 LT |
789 | } |
790 | ||
791 | /* The code following below sending ACKs in SYN-RECV and TIME-WAIT states | |
792 | outside socket context is ugly, certainly. What can I do? | |
793 | */ | |
794 | ||
e2d118a1 | 795 | static void tcp_v4_send_ack(const struct sock *sk, |
e62a123b | 796 | struct sk_buff *skb, u32 seq, u32 ack, |
ee684b6f | 797 | u32 win, u32 tsval, u32 tsecr, int oif, |
88ef4a5a | 798 | struct tcp_md5sig_key *key, |
66b13d99 | 799 | int reply_flags, u8 tos) |
1da177e4 | 800 | { |
cf533ea5 | 801 | const struct tcphdr *th = tcp_hdr(skb); |
1da177e4 LT |
802 | struct { |
803 | struct tcphdr th; | |
714e85be | 804 | __be32 opt[(TCPOLEN_TSTAMP_ALIGNED >> 2) |
cfb6eeb4 | 805 | #ifdef CONFIG_TCP_MD5SIG |
714e85be | 806 | + (TCPOLEN_MD5SIG_ALIGNED >> 2) |
cfb6eeb4 YH |
807 | #endif |
808 | ]; | |
1da177e4 | 809 | } rep; |
e2d118a1 | 810 | struct net *net = sock_net(sk); |
1da177e4 | 811 | struct ip_reply_arg arg; |
00483690 | 812 | struct sock *ctl_sk; |
1da177e4 LT |
813 | |
814 | memset(&rep.th, 0, sizeof(struct tcphdr)); | |
7174259e | 815 | memset(&arg, 0, sizeof(arg)); |
1da177e4 LT |
816 | |
817 | arg.iov[0].iov_base = (unsigned char *)&rep; | |
818 | arg.iov[0].iov_len = sizeof(rep.th); | |
ee684b6f | 819 | if (tsecr) { |
cfb6eeb4 YH |
820 | rep.opt[0] = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) | |
821 | (TCPOPT_TIMESTAMP << 8) | | |
822 | TCPOLEN_TIMESTAMP); | |
ee684b6f AV |
823 | rep.opt[1] = htonl(tsval); |
824 | rep.opt[2] = htonl(tsecr); | |
cb48cfe8 | 825 | arg.iov[0].iov_len += TCPOLEN_TSTAMP_ALIGNED; |
1da177e4 LT |
826 | } |
827 | ||
828 | /* Swap the send and the receive. */ | |
829 | rep.th.dest = th->source; | |
830 | rep.th.source = th->dest; | |
831 | rep.th.doff = arg.iov[0].iov_len / 4; | |
832 | rep.th.seq = htonl(seq); | |
833 | rep.th.ack_seq = htonl(ack); | |
834 | rep.th.ack = 1; | |
835 | rep.th.window = htons(win); | |
836 | ||
cfb6eeb4 | 837 | #ifdef CONFIG_TCP_MD5SIG |
cfb6eeb4 | 838 | if (key) { |
ee684b6f | 839 | int offset = (tsecr) ? 3 : 0; |
cfb6eeb4 YH |
840 | |
841 | rep.opt[offset++] = htonl((TCPOPT_NOP << 24) | | |
842 | (TCPOPT_NOP << 16) | | |
843 | (TCPOPT_MD5SIG << 8) | | |
844 | TCPOLEN_MD5SIG); | |
845 | arg.iov[0].iov_len += TCPOLEN_MD5SIG_ALIGNED; | |
846 | rep.th.doff = arg.iov[0].iov_len/4; | |
847 | ||
49a72dfb | 848 | tcp_v4_md5_hash_hdr((__u8 *) &rep.opt[offset], |
90b7e112 AL |
849 | key, ip_hdr(skb)->saddr, |
850 | ip_hdr(skb)->daddr, &rep.th); | |
cfb6eeb4 YH |
851 | } |
852 | #endif | |
88ef4a5a | 853 | arg.flags = reply_flags; |
eddc9ec5 ACM |
854 | arg.csum = csum_tcpudp_nofold(ip_hdr(skb)->daddr, |
855 | ip_hdr(skb)->saddr, /* XXX */ | |
1da177e4 LT |
856 | arg.iov[0].iov_len, IPPROTO_TCP, 0); |
857 | arg.csumoffset = offsetof(struct tcphdr, check) / 2; | |
9501f972 YH |
858 | if (oif) |
859 | arg.bound_dev_if = oif; | |
66b13d99 | 860 | arg.tos = tos; |
e2d118a1 | 861 | arg.uid = sock_net_uid(net, sk_fullsock(sk) ? sk : NULL); |
47dcc20a | 862 | local_bh_disable(); |
00483690 JM |
863 | ctl_sk = *this_cpu_ptr(net->ipv4.tcp_sk); |
864 | if (sk) | |
865 | ctl_sk->sk_mark = (sk->sk_state == TCP_TIME_WAIT) ? | |
866 | inet_twsk(sk)->tw_mark : sk->sk_mark; | |
867 | ip_send_unicast_reply(ctl_sk, | |
bdbbb852 | 868 | skb, &TCP_SKB_CB(skb)->header.h4.opt, |
24a2d43d ED |
869 | ip_hdr(skb)->saddr, ip_hdr(skb)->daddr, |
870 | &arg, arg.iov[0].iov_len); | |
1da177e4 | 871 | |
00483690 | 872 | ctl_sk->sk_mark = 0; |
90bbcc60 | 873 | __TCP_INC_STATS(net, TCP_MIB_OUTSEGS); |
47dcc20a | 874 | local_bh_enable(); |
1da177e4 LT |
875 | } |
876 | ||
877 | static void tcp_v4_timewait_ack(struct sock *sk, struct sk_buff *skb) | |
878 | { | |
8feaf0c0 | 879 | struct inet_timewait_sock *tw = inet_twsk(sk); |
cfb6eeb4 | 880 | struct tcp_timewait_sock *tcptw = tcp_twsk(sk); |
1da177e4 | 881 | |
e2d118a1 | 882 | tcp_v4_send_ack(sk, skb, |
e62a123b | 883 | tcptw->tw_snd_nxt, tcptw->tw_rcv_nxt, |
7174259e | 884 | tcptw->tw_rcv_wnd >> tw->tw_rcv_wscale, |
9a568de4 | 885 | tcp_time_stamp_raw() + tcptw->tw_ts_offset, |
9501f972 YH |
886 | tcptw->tw_ts_recent, |
887 | tw->tw_bound_dev_if, | |
88ef4a5a | 888 | tcp_twsk_md5_key(tcptw), |
66b13d99 ED |
889 | tw->tw_transparent ? IP_REPLY_ARG_NOSRCCHECK : 0, |
890 | tw->tw_tos | |
9501f972 | 891 | ); |
1da177e4 | 892 | |
8feaf0c0 | 893 | inet_twsk_put(tw); |
1da177e4 LT |
894 | } |
895 | ||
a00e7444 | 896 | static void tcp_v4_reqsk_send_ack(const struct sock *sk, struct sk_buff *skb, |
7174259e | 897 | struct request_sock *req) |
1da177e4 | 898 | { |
168a8f58 JC |
899 | /* sk->sk_state == TCP_LISTEN -> for regular TCP_SYN_RECV |
900 | * sk->sk_state == TCP_SYN_RECV -> for Fast Open. | |
901 | */ | |
e62a123b ED |
902 | u32 seq = (sk->sk_state == TCP_LISTEN) ? tcp_rsk(req)->snt_isn + 1 : |
903 | tcp_sk(sk)->snd_nxt; | |
904 | ||
20a2b49f ED |
905 | /* RFC 7323 2.3 |
906 | * The window field (SEG.WND) of every outgoing segment, with the | |
907 | * exception of <SYN> segments, MUST be right-shifted by | |
908 | * Rcv.Wind.Shift bits: | |
909 | */ | |
e2d118a1 | 910 | tcp_v4_send_ack(sk, skb, seq, |
20a2b49f ED |
911 | tcp_rsk(req)->rcv_nxt, |
912 | req->rsk_rcv_wnd >> inet_rsk(req)->rcv_wscale, | |
9a568de4 | 913 | tcp_time_stamp_raw() + tcp_rsk(req)->ts_off, |
9501f972 YH |
914 | req->ts_recent, |
915 | 0, | |
30791ac4 | 916 | tcp_md5_do_lookup(sk, (union tcp_md5_addr *)&ip_hdr(skb)->saddr, |
a915da9b | 917 | AF_INET), |
66b13d99 ED |
918 | inet_rsk(req)->no_srccheck ? IP_REPLY_ARG_NOSRCCHECK : 0, |
919 | ip_hdr(skb)->tos); | |
1da177e4 LT |
920 | } |
921 | ||
1da177e4 | 922 | /* |
9bf1d83e | 923 | * Send a SYN-ACK after having received a SYN. |
60236fdd | 924 | * This still operates on a request_sock only, not on a big |
1da177e4 LT |
925 | * socket. |
926 | */ | |
0f935dbe | 927 | static int tcp_v4_send_synack(const struct sock *sk, struct dst_entry *dst, |
d6274bd8 | 928 | struct flowi *fl, |
72659ecc | 929 | struct request_sock *req, |
ca6fb065 | 930 | struct tcp_fastopen_cookie *foc, |
b3d05147 | 931 | enum tcp_synack_type synack_type) |
1da177e4 | 932 | { |
2e6599cb | 933 | const struct inet_request_sock *ireq = inet_rsk(req); |
6bd023f3 | 934 | struct flowi4 fl4; |
1da177e4 | 935 | int err = -1; |
d41db5af | 936 | struct sk_buff *skb; |
1da177e4 LT |
937 | |
938 | /* First, grab a route. */ | |
ba3f7f04 | 939 | if (!dst && (dst = inet_csk_route_req(sk, &fl4, req)) == NULL) |
fd80eb94 | 940 | return -1; |
1da177e4 | 941 | |
b3d05147 | 942 | skb = tcp_make_synack(sk, dst, req, foc, synack_type); |
1da177e4 LT |
943 | |
944 | if (skb) { | |
634fb979 | 945 | __tcp_v4_send_check(skb, ireq->ir_loc_addr, ireq->ir_rmt_addr); |
1da177e4 | 946 | |
2ab2ddd3 | 947 | rcu_read_lock(); |
634fb979 ED |
948 | err = ip_build_and_send_pkt(skb, sk, ireq->ir_loc_addr, |
949 | ireq->ir_rmt_addr, | |
2ab2ddd3 ED |
950 | rcu_dereference(ireq->ireq_opt)); |
951 | rcu_read_unlock(); | |
b9df3cb8 | 952 | err = net_xmit_eval(err); |
1da177e4 LT |
953 | } |
954 | ||
1da177e4 LT |
955 | return err; |
956 | } | |
957 | ||
958 | /* | |
60236fdd | 959 | * IPv4 request_sock destructor. |
1da177e4 | 960 | */ |
60236fdd | 961 | static void tcp_v4_reqsk_destructor(struct request_sock *req) |
1da177e4 | 962 | { |
c92e8c02 | 963 | kfree(rcu_dereference_protected(inet_rsk(req)->ireq_opt, 1)); |
1da177e4 LT |
964 | } |
965 | ||
cfb6eeb4 YH |
966 | #ifdef CONFIG_TCP_MD5SIG |
967 | /* | |
968 | * RFC2385 MD5 checksumming requires a mapping of | |
969 | * IP address->MD5 Key. | |
970 | * We need to maintain these in the sk structure. | |
971 | */ | |
972 | ||
6015c71e ED |
973 | struct static_key tcp_md5_needed __read_mostly; |
974 | EXPORT_SYMBOL(tcp_md5_needed); | |
975 | ||
cfb6eeb4 | 976 | /* Find the Key structure for an address. */ |
6015c71e ED |
977 | struct tcp_md5sig_key *__tcp_md5_do_lookup(const struct sock *sk, |
978 | const union tcp_md5_addr *addr, | |
979 | int family) | |
cfb6eeb4 | 980 | { |
fd3a154a | 981 | const struct tcp_sock *tp = tcp_sk(sk); |
a915da9b | 982 | struct tcp_md5sig_key *key; |
fd3a154a | 983 | const struct tcp_md5sig_info *md5sig; |
6797318e ID |
984 | __be32 mask; |
985 | struct tcp_md5sig_key *best_match = NULL; | |
986 | bool match; | |
cfb6eeb4 | 987 | |
a8afca03 ED |
988 | /* caller either holds rcu_read_lock() or socket lock */ |
989 | md5sig = rcu_dereference_check(tp->md5sig_info, | |
1e1d04e6 | 990 | lockdep_sock_is_held(sk)); |
a8afca03 | 991 | if (!md5sig) |
cfb6eeb4 | 992 | return NULL; |
083a0326 | 993 | |
b67bfe0d | 994 | hlist_for_each_entry_rcu(key, &md5sig->head, node) { |
a915da9b ED |
995 | if (key->family != family) |
996 | continue; | |
6797318e ID |
997 | |
998 | if (family == AF_INET) { | |
999 | mask = inet_make_mask(key->prefixlen); | |
1000 | match = (key->addr.a4.s_addr & mask) == | |
1001 | (addr->a4.s_addr & mask); | |
1002 | #if IS_ENABLED(CONFIG_IPV6) | |
1003 | } else if (family == AF_INET6) { | |
1004 | match = ipv6_prefix_equal(&key->addr.a6, &addr->a6, | |
1005 | key->prefixlen); | |
1006 | #endif | |
1007 | } else { | |
1008 | match = false; | |
1009 | } | |
1010 | ||
1011 | if (match && (!best_match || | |
1012 | key->prefixlen > best_match->prefixlen)) | |
1013 | best_match = key; | |
1014 | } | |
1015 | return best_match; | |
1016 | } | |
6015c71e | 1017 | EXPORT_SYMBOL(__tcp_md5_do_lookup); |
6797318e | 1018 | |
e8f37d57 WF |
1019 | static struct tcp_md5sig_key *tcp_md5_do_lookup_exact(const struct sock *sk, |
1020 | const union tcp_md5_addr *addr, | |
1021 | int family, u8 prefixlen) | |
6797318e ID |
1022 | { |
1023 | const struct tcp_sock *tp = tcp_sk(sk); | |
1024 | struct tcp_md5sig_key *key; | |
1025 | unsigned int size = sizeof(struct in_addr); | |
1026 | const struct tcp_md5sig_info *md5sig; | |
1027 | ||
1028 | /* caller either holds rcu_read_lock() or socket lock */ | |
1029 | md5sig = rcu_dereference_check(tp->md5sig_info, | |
1030 | lockdep_sock_is_held(sk)); | |
1031 | if (!md5sig) | |
1032 | return NULL; | |
1033 | #if IS_ENABLED(CONFIG_IPV6) | |
1034 | if (family == AF_INET6) | |
1035 | size = sizeof(struct in6_addr); | |
1036 | #endif | |
1037 | hlist_for_each_entry_rcu(key, &md5sig->head, node) { | |
1038 | if (key->family != family) | |
1039 | continue; | |
1040 | if (!memcmp(&key->addr, addr, size) && | |
1041 | key->prefixlen == prefixlen) | |
a915da9b | 1042 | return key; |
cfb6eeb4 YH |
1043 | } |
1044 | return NULL; | |
1045 | } | |
1046 | ||
b83e3deb | 1047 | struct tcp_md5sig_key *tcp_v4_md5_lookup(const struct sock *sk, |
fd3a154a | 1048 | const struct sock *addr_sk) |
cfb6eeb4 | 1049 | { |
b52e6921 | 1050 | const union tcp_md5_addr *addr; |
a915da9b | 1051 | |
b52e6921 | 1052 | addr = (const union tcp_md5_addr *)&addr_sk->sk_daddr; |
a915da9b | 1053 | return tcp_md5_do_lookup(sk, addr, AF_INET); |
cfb6eeb4 | 1054 | } |
cfb6eeb4 YH |
1055 | EXPORT_SYMBOL(tcp_v4_md5_lookup); |
1056 | ||
cfb6eeb4 | 1057 | /* This can be called on a newly created socket, from other files */ |
a915da9b | 1058 | int tcp_md5_do_add(struct sock *sk, const union tcp_md5_addr *addr, |
6797318e ID |
1059 | int family, u8 prefixlen, const u8 *newkey, u8 newkeylen, |
1060 | gfp_t gfp) | |
cfb6eeb4 YH |
1061 | { |
1062 | /* Add Key to the list */ | |
b0a713e9 | 1063 | struct tcp_md5sig_key *key; |
cfb6eeb4 | 1064 | struct tcp_sock *tp = tcp_sk(sk); |
a915da9b | 1065 | struct tcp_md5sig_info *md5sig; |
cfb6eeb4 | 1066 | |
6797318e | 1067 | key = tcp_md5_do_lookup_exact(sk, addr, family, prefixlen); |
cfb6eeb4 YH |
1068 | if (key) { |
1069 | /* Pre-existing entry - just update that one. */ | |
a915da9b | 1070 | memcpy(key->key, newkey, newkeylen); |
b0a713e9 | 1071 | key->keylen = newkeylen; |
a915da9b ED |
1072 | return 0; |
1073 | } | |
260fcbeb | 1074 | |
a8afca03 | 1075 | md5sig = rcu_dereference_protected(tp->md5sig_info, |
1e1d04e6 | 1076 | lockdep_sock_is_held(sk)); |
a915da9b ED |
1077 | if (!md5sig) { |
1078 | md5sig = kmalloc(sizeof(*md5sig), gfp); | |
1079 | if (!md5sig) | |
cfb6eeb4 | 1080 | return -ENOMEM; |
cfb6eeb4 | 1081 | |
a915da9b ED |
1082 | sk_nocaps_add(sk, NETIF_F_GSO_MASK); |
1083 | INIT_HLIST_HEAD(&md5sig->head); | |
a8afca03 | 1084 | rcu_assign_pointer(tp->md5sig_info, md5sig); |
a915da9b | 1085 | } |
cfb6eeb4 | 1086 | |
5f3d9cb2 | 1087 | key = sock_kmalloc(sk, sizeof(*key), gfp); |
a915da9b ED |
1088 | if (!key) |
1089 | return -ENOMEM; | |
71cea17e | 1090 | if (!tcp_alloc_md5sig_pool()) { |
5f3d9cb2 | 1091 | sock_kfree_s(sk, key, sizeof(*key)); |
a915da9b | 1092 | return -ENOMEM; |
cfb6eeb4 | 1093 | } |
a915da9b ED |
1094 | |
1095 | memcpy(key->key, newkey, newkeylen); | |
1096 | key->keylen = newkeylen; | |
1097 | key->family = family; | |
6797318e | 1098 | key->prefixlen = prefixlen; |
a915da9b ED |
1099 | memcpy(&key->addr, addr, |
1100 | (family == AF_INET6) ? sizeof(struct in6_addr) : | |
1101 | sizeof(struct in_addr)); | |
1102 | hlist_add_head_rcu(&key->node, &md5sig->head); | |
cfb6eeb4 YH |
1103 | return 0; |
1104 | } | |
a915da9b | 1105 | EXPORT_SYMBOL(tcp_md5_do_add); |
cfb6eeb4 | 1106 | |
6797318e ID |
1107 | int tcp_md5_do_del(struct sock *sk, const union tcp_md5_addr *addr, int family, |
1108 | u8 prefixlen) | |
cfb6eeb4 | 1109 | { |
a915da9b ED |
1110 | struct tcp_md5sig_key *key; |
1111 | ||
6797318e | 1112 | key = tcp_md5_do_lookup_exact(sk, addr, family, prefixlen); |
a915da9b ED |
1113 | if (!key) |
1114 | return -ENOENT; | |
1115 | hlist_del_rcu(&key->node); | |
5f3d9cb2 | 1116 | atomic_sub(sizeof(*key), &sk->sk_omem_alloc); |
a915da9b | 1117 | kfree_rcu(key, rcu); |
a915da9b | 1118 | return 0; |
cfb6eeb4 | 1119 | } |
a915da9b | 1120 | EXPORT_SYMBOL(tcp_md5_do_del); |
cfb6eeb4 | 1121 | |
e0683e70 | 1122 | static void tcp_clear_md5_list(struct sock *sk) |
cfb6eeb4 YH |
1123 | { |
1124 | struct tcp_sock *tp = tcp_sk(sk); | |
a915da9b | 1125 | struct tcp_md5sig_key *key; |
b67bfe0d | 1126 | struct hlist_node *n; |
a8afca03 | 1127 | struct tcp_md5sig_info *md5sig; |
cfb6eeb4 | 1128 | |
a8afca03 ED |
1129 | md5sig = rcu_dereference_protected(tp->md5sig_info, 1); |
1130 | ||
b67bfe0d | 1131 | hlist_for_each_entry_safe(key, n, &md5sig->head, node) { |
a915da9b | 1132 | hlist_del_rcu(&key->node); |
5f3d9cb2 | 1133 | atomic_sub(sizeof(*key), &sk->sk_omem_alloc); |
a915da9b | 1134 | kfree_rcu(key, rcu); |
cfb6eeb4 YH |
1135 | } |
1136 | } | |
1137 | ||
8917a777 ID |
1138 | static int tcp_v4_parse_md5_keys(struct sock *sk, int optname, |
1139 | char __user *optval, int optlen) | |
cfb6eeb4 YH |
1140 | { |
1141 | struct tcp_md5sig cmd; | |
1142 | struct sockaddr_in *sin = (struct sockaddr_in *)&cmd.tcpm_addr; | |
8917a777 | 1143 | u8 prefixlen = 32; |
cfb6eeb4 YH |
1144 | |
1145 | if (optlen < sizeof(cmd)) | |
1146 | return -EINVAL; | |
1147 | ||
7174259e | 1148 | if (copy_from_user(&cmd, optval, sizeof(cmd))) |
cfb6eeb4 YH |
1149 | return -EFAULT; |
1150 | ||
1151 | if (sin->sin_family != AF_INET) | |
1152 | return -EINVAL; | |
1153 | ||
8917a777 ID |
1154 | if (optname == TCP_MD5SIG_EXT && |
1155 | cmd.tcpm_flags & TCP_MD5SIG_FLAG_PREFIX) { | |
1156 | prefixlen = cmd.tcpm_prefixlen; | |
1157 | if (prefixlen > 32) | |
1158 | return -EINVAL; | |
1159 | } | |
1160 | ||
64a124ed | 1161 | if (!cmd.tcpm_keylen) |
a915da9b | 1162 | return tcp_md5_do_del(sk, (union tcp_md5_addr *)&sin->sin_addr.s_addr, |
8917a777 | 1163 | AF_INET, prefixlen); |
cfb6eeb4 YH |
1164 | |
1165 | if (cmd.tcpm_keylen > TCP_MD5SIG_MAXKEYLEN) | |
1166 | return -EINVAL; | |
1167 | ||
a915da9b | 1168 | return tcp_md5_do_add(sk, (union tcp_md5_addr *)&sin->sin_addr.s_addr, |
8917a777 | 1169 | AF_INET, prefixlen, cmd.tcpm_key, cmd.tcpm_keylen, |
a915da9b | 1170 | GFP_KERNEL); |
cfb6eeb4 YH |
1171 | } |
1172 | ||
19689e38 ED |
1173 | static int tcp_v4_md5_hash_headers(struct tcp_md5sig_pool *hp, |
1174 | __be32 daddr, __be32 saddr, | |
1175 | const struct tcphdr *th, int nbytes) | |
cfb6eeb4 | 1176 | { |
cfb6eeb4 | 1177 | struct tcp4_pseudohdr *bp; |
49a72dfb | 1178 | struct scatterlist sg; |
19689e38 | 1179 | struct tcphdr *_th; |
cfb6eeb4 | 1180 | |
19689e38 | 1181 | bp = hp->scratch; |
cfb6eeb4 YH |
1182 | bp->saddr = saddr; |
1183 | bp->daddr = daddr; | |
1184 | bp->pad = 0; | |
076fb722 | 1185 | bp->protocol = IPPROTO_TCP; |
49a72dfb | 1186 | bp->len = cpu_to_be16(nbytes); |
c7da57a1 | 1187 | |
19689e38 ED |
1188 | _th = (struct tcphdr *)(bp + 1); |
1189 | memcpy(_th, th, sizeof(*th)); | |
1190 | _th->check = 0; | |
1191 | ||
1192 | sg_init_one(&sg, bp, sizeof(*bp) + sizeof(*th)); | |
1193 | ahash_request_set_crypt(hp->md5_req, &sg, NULL, | |
1194 | sizeof(*bp) + sizeof(*th)); | |
cf80e0e4 | 1195 | return crypto_ahash_update(hp->md5_req); |
49a72dfb AL |
1196 | } |
1197 | ||
a915da9b | 1198 | static int tcp_v4_md5_hash_hdr(char *md5_hash, const struct tcp_md5sig_key *key, |
318cf7aa | 1199 | __be32 daddr, __be32 saddr, const struct tcphdr *th) |
49a72dfb AL |
1200 | { |
1201 | struct tcp_md5sig_pool *hp; | |
cf80e0e4 | 1202 | struct ahash_request *req; |
49a72dfb AL |
1203 | |
1204 | hp = tcp_get_md5sig_pool(); | |
1205 | if (!hp) | |
1206 | goto clear_hash_noput; | |
cf80e0e4 | 1207 | req = hp->md5_req; |
49a72dfb | 1208 | |
cf80e0e4 | 1209 | if (crypto_ahash_init(req)) |
49a72dfb | 1210 | goto clear_hash; |
19689e38 | 1211 | if (tcp_v4_md5_hash_headers(hp, daddr, saddr, th, th->doff << 2)) |
49a72dfb AL |
1212 | goto clear_hash; |
1213 | if (tcp_md5_hash_key(hp, key)) | |
1214 | goto clear_hash; | |
cf80e0e4 HX |
1215 | ahash_request_set_crypt(req, NULL, md5_hash, 0); |
1216 | if (crypto_ahash_final(req)) | |
cfb6eeb4 YH |
1217 | goto clear_hash; |
1218 | ||
cfb6eeb4 | 1219 | tcp_put_md5sig_pool(); |
cfb6eeb4 | 1220 | return 0; |
49a72dfb | 1221 | |
cfb6eeb4 YH |
1222 | clear_hash: |
1223 | tcp_put_md5sig_pool(); | |
1224 | clear_hash_noput: | |
1225 | memset(md5_hash, 0, 16); | |
49a72dfb | 1226 | return 1; |
cfb6eeb4 YH |
1227 | } |
1228 | ||
39f8e58e ED |
1229 | int tcp_v4_md5_hash_skb(char *md5_hash, const struct tcp_md5sig_key *key, |
1230 | const struct sock *sk, | |
318cf7aa | 1231 | const struct sk_buff *skb) |
cfb6eeb4 | 1232 | { |
49a72dfb | 1233 | struct tcp_md5sig_pool *hp; |
cf80e0e4 | 1234 | struct ahash_request *req; |
318cf7aa | 1235 | const struct tcphdr *th = tcp_hdr(skb); |
cfb6eeb4 YH |
1236 | __be32 saddr, daddr; |
1237 | ||
39f8e58e ED |
1238 | if (sk) { /* valid for establish/request sockets */ |
1239 | saddr = sk->sk_rcv_saddr; | |
1240 | daddr = sk->sk_daddr; | |
cfb6eeb4 | 1241 | } else { |
49a72dfb AL |
1242 | const struct iphdr *iph = ip_hdr(skb); |
1243 | saddr = iph->saddr; | |
1244 | daddr = iph->daddr; | |
cfb6eeb4 | 1245 | } |
49a72dfb AL |
1246 | |
1247 | hp = tcp_get_md5sig_pool(); | |
1248 | if (!hp) | |
1249 | goto clear_hash_noput; | |
cf80e0e4 | 1250 | req = hp->md5_req; |
49a72dfb | 1251 | |
cf80e0e4 | 1252 | if (crypto_ahash_init(req)) |
49a72dfb AL |
1253 | goto clear_hash; |
1254 | ||
19689e38 | 1255 | if (tcp_v4_md5_hash_headers(hp, daddr, saddr, th, skb->len)) |
49a72dfb AL |
1256 | goto clear_hash; |
1257 | if (tcp_md5_hash_skb_data(hp, skb, th->doff << 2)) | |
1258 | goto clear_hash; | |
1259 | if (tcp_md5_hash_key(hp, key)) | |
1260 | goto clear_hash; | |
cf80e0e4 HX |
1261 | ahash_request_set_crypt(req, NULL, md5_hash, 0); |
1262 | if (crypto_ahash_final(req)) | |
49a72dfb AL |
1263 | goto clear_hash; |
1264 | ||
1265 | tcp_put_md5sig_pool(); | |
1266 | return 0; | |
1267 | ||
1268 | clear_hash: | |
1269 | tcp_put_md5sig_pool(); | |
1270 | clear_hash_noput: | |
1271 | memset(md5_hash, 0, 16); | |
1272 | return 1; | |
cfb6eeb4 | 1273 | } |
49a72dfb | 1274 | EXPORT_SYMBOL(tcp_v4_md5_hash_skb); |
cfb6eeb4 | 1275 | |
ba8e275a ED |
1276 | #endif |
1277 | ||
ff74e23f | 1278 | /* Called with rcu_read_lock() */ |
ba8e275a | 1279 | static bool tcp_v4_inbound_md5_hash(const struct sock *sk, |
ff74e23f | 1280 | const struct sk_buff *skb) |
cfb6eeb4 | 1281 | { |
ba8e275a | 1282 | #ifdef CONFIG_TCP_MD5SIG |
cfb6eeb4 YH |
1283 | /* |
1284 | * This gets called for each TCP segment that arrives | |
1285 | * so we want to be efficient. | |
1286 | * We have 3 drop cases: | |
1287 | * o No MD5 hash and one expected. | |
1288 | * o MD5 hash and we're not expecting one. | |
1289 | * o MD5 hash and its wrong. | |
1290 | */ | |
cf533ea5 | 1291 | const __u8 *hash_location = NULL; |
cfb6eeb4 | 1292 | struct tcp_md5sig_key *hash_expected; |
eddc9ec5 | 1293 | const struct iphdr *iph = ip_hdr(skb); |
cf533ea5 | 1294 | const struct tcphdr *th = tcp_hdr(skb); |
cfb6eeb4 | 1295 | int genhash; |
cfb6eeb4 YH |
1296 | unsigned char newhash[16]; |
1297 | ||
a915da9b ED |
1298 | hash_expected = tcp_md5_do_lookup(sk, (union tcp_md5_addr *)&iph->saddr, |
1299 | AF_INET); | |
7d5d5525 | 1300 | hash_location = tcp_parse_md5sig_option(th); |
cfb6eeb4 | 1301 | |
cfb6eeb4 YH |
1302 | /* We've parsed the options - do we have a hash? */ |
1303 | if (!hash_expected && !hash_location) | |
a2a385d6 | 1304 | return false; |
cfb6eeb4 YH |
1305 | |
1306 | if (hash_expected && !hash_location) { | |
c10d9310 | 1307 | NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPMD5NOTFOUND); |
a2a385d6 | 1308 | return true; |
cfb6eeb4 YH |
1309 | } |
1310 | ||
1311 | if (!hash_expected && hash_location) { | |
c10d9310 | 1312 | NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPMD5UNEXPECTED); |
a2a385d6 | 1313 | return true; |
cfb6eeb4 YH |
1314 | } |
1315 | ||
1316 | /* Okay, so this is hash_expected and hash_location - | |
1317 | * so we need to calculate the checksum. | |
1318 | */ | |
49a72dfb AL |
1319 | genhash = tcp_v4_md5_hash_skb(newhash, |
1320 | hash_expected, | |
39f8e58e | 1321 | NULL, skb); |
cfb6eeb4 YH |
1322 | |
1323 | if (genhash || memcmp(hash_location, newhash, 16) != 0) { | |
72145a68 | 1324 | NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPMD5FAILURE); |
e87cc472 JP |
1325 | net_info_ratelimited("MD5 Hash failed for (%pI4, %d)->(%pI4, %d)%s\n", |
1326 | &iph->saddr, ntohs(th->source), | |
1327 | &iph->daddr, ntohs(th->dest), | |
1328 | genhash ? " tcp_v4_calc_md5_hash failed" | |
1329 | : ""); | |
a2a385d6 | 1330 | return true; |
cfb6eeb4 | 1331 | } |
a2a385d6 | 1332 | return false; |
cfb6eeb4 | 1333 | #endif |
ba8e275a ED |
1334 | return false; |
1335 | } | |
cfb6eeb4 | 1336 | |
b40cf18e ED |
1337 | static void tcp_v4_init_req(struct request_sock *req, |
1338 | const struct sock *sk_listener, | |
16bea70a OP |
1339 | struct sk_buff *skb) |
1340 | { | |
1341 | struct inet_request_sock *ireq = inet_rsk(req); | |
c92e8c02 | 1342 | struct net *net = sock_net(sk_listener); |
16bea70a | 1343 | |
08d2cc3b ED |
1344 | sk_rcv_saddr_set(req_to_sk(req), ip_hdr(skb)->daddr); |
1345 | sk_daddr_set(req_to_sk(req), ip_hdr(skb)->saddr); | |
c92e8c02 | 1346 | RCU_INIT_POINTER(ireq->ireq_opt, tcp_v4_save_options(net, skb)); |
16bea70a OP |
1347 | } |
1348 | ||
f964629e ED |
1349 | static struct dst_entry *tcp_v4_route_req(const struct sock *sk, |
1350 | struct flowi *fl, | |
4396e461 | 1351 | const struct request_sock *req) |
d94e0417 | 1352 | { |
4396e461 | 1353 | return inet_csk_route_req(sk, &fl->u.ip4, req); |
d94e0417 OP |
1354 | } |
1355 | ||
72a3effa | 1356 | struct request_sock_ops tcp_request_sock_ops __read_mostly = { |
1da177e4 | 1357 | .family = PF_INET, |
2e6599cb | 1358 | .obj_size = sizeof(struct tcp_request_sock), |
5db92c99 | 1359 | .rtx_syn_ack = tcp_rtx_synack, |
60236fdd ACM |
1360 | .send_ack = tcp_v4_reqsk_send_ack, |
1361 | .destructor = tcp_v4_reqsk_destructor, | |
1da177e4 | 1362 | .send_reset = tcp_v4_send_reset, |
688d1945 | 1363 | .syn_ack_timeout = tcp_syn_ack_timeout, |
1da177e4 LT |
1364 | }; |
1365 | ||
b2e4b3de | 1366 | static const struct tcp_request_sock_ops tcp_request_sock_ipv4_ops = { |
2aec4a29 | 1367 | .mss_clamp = TCP_MSS_DEFAULT, |
16bea70a | 1368 | #ifdef CONFIG_TCP_MD5SIG |
fd3a154a | 1369 | .req_md5_lookup = tcp_v4_md5_lookup, |
e3afe7b7 | 1370 | .calc_md5_hash = tcp_v4_md5_hash_skb, |
b6332e6c | 1371 | #endif |
16bea70a | 1372 | .init_req = tcp_v4_init_req, |
fb7b37a7 OP |
1373 | #ifdef CONFIG_SYN_COOKIES |
1374 | .cookie_init_seq = cookie_v4_init_sequence, | |
1375 | #endif | |
d94e0417 | 1376 | .route_req = tcp_v4_route_req, |
84b114b9 ED |
1377 | .init_seq = tcp_v4_init_seq, |
1378 | .init_ts_off = tcp_v4_init_ts_off, | |
d6274bd8 | 1379 | .send_synack = tcp_v4_send_synack, |
16bea70a | 1380 | }; |
cfb6eeb4 | 1381 | |
1da177e4 LT |
1382 | int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb) |
1383 | { | |
1da177e4 | 1384 | /* Never answer to SYNs send to broadcast or multicast */ |
511c3f92 | 1385 | if (skb_rtable(skb)->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST)) |
1da177e4 LT |
1386 | goto drop; |
1387 | ||
1fb6f159 OP |
1388 | return tcp_conn_request(&tcp_request_sock_ops, |
1389 | &tcp_request_sock_ipv4_ops, sk, skb); | |
1da177e4 | 1390 | |
1da177e4 | 1391 | drop: |
9caad864 | 1392 | tcp_listendrop(sk); |
1da177e4 LT |
1393 | return 0; |
1394 | } | |
4bc2f18b | 1395 | EXPORT_SYMBOL(tcp_v4_conn_request); |
1da177e4 LT |
1396 | |
1397 | ||
1398 | /* | |
1399 | * The three way handshake has completed - we got a valid synack - | |
1400 | * now create the new socket. | |
1401 | */ | |
0c27171e | 1402 | struct sock *tcp_v4_syn_recv_sock(const struct sock *sk, struct sk_buff *skb, |
60236fdd | 1403 | struct request_sock *req, |
5e0724d0 ED |
1404 | struct dst_entry *dst, |
1405 | struct request_sock *req_unhash, | |
1406 | bool *own_req) | |
1da177e4 | 1407 | { |
2e6599cb | 1408 | struct inet_request_sock *ireq; |
1da177e4 LT |
1409 | struct inet_sock *newinet; |
1410 | struct tcp_sock *newtp; | |
1411 | struct sock *newsk; | |
cfb6eeb4 YH |
1412 | #ifdef CONFIG_TCP_MD5SIG |
1413 | struct tcp_md5sig_key *key; | |
1414 | #endif | |
f6d8bd05 | 1415 | struct ip_options_rcu *inet_opt; |
1da177e4 LT |
1416 | |
1417 | if (sk_acceptq_is_full(sk)) | |
1418 | goto exit_overflow; | |
1419 | ||
1da177e4 LT |
1420 | newsk = tcp_create_openreq_child(sk, req, skb); |
1421 | if (!newsk) | |
093d2823 | 1422 | goto exit_nonewsk; |
1da177e4 | 1423 | |
bcd76111 | 1424 | newsk->sk_gso_type = SKB_GSO_TCPV4; |
fae6ef87 | 1425 | inet_sk_rx_dst_set(newsk, skb); |
1da177e4 LT |
1426 | |
1427 | newtp = tcp_sk(newsk); | |
1428 | newinet = inet_sk(newsk); | |
2e6599cb | 1429 | ireq = inet_rsk(req); |
d1e559d0 ED |
1430 | sk_daddr_set(newsk, ireq->ir_rmt_addr); |
1431 | sk_rcv_saddr_set(newsk, ireq->ir_loc_addr); | |
6dd9a14e | 1432 | newsk->sk_bound_dev_if = ireq->ir_iif; |
c92e8c02 ED |
1433 | newinet->inet_saddr = ireq->ir_loc_addr; |
1434 | inet_opt = rcu_dereference(ireq->ireq_opt); | |
1435 | RCU_INIT_POINTER(newinet->inet_opt, inet_opt); | |
463c84b9 | 1436 | newinet->mc_index = inet_iif(skb); |
eddc9ec5 | 1437 | newinet->mc_ttl = ip_hdr(skb)->ttl; |
4c507d28 | 1438 | newinet->rcv_tos = ip_hdr(skb)->tos; |
d83d8461 | 1439 | inet_csk(newsk)->icsk_ext_hdr_len = 0; |
f6d8bd05 ED |
1440 | if (inet_opt) |
1441 | inet_csk(newsk)->icsk_ext_hdr_len = inet_opt->opt.optlen; | |
c720c7e8 | 1442 | newinet->inet_id = newtp->write_seq ^ jiffies; |
1da177e4 | 1443 | |
dfd25fff ED |
1444 | if (!dst) { |
1445 | dst = inet_csk_route_child_sock(sk, newsk, req); | |
1446 | if (!dst) | |
1447 | goto put_and_exit; | |
1448 | } else { | |
1449 | /* syncookie case : see end of cookie_v4_check() */ | |
1450 | } | |
0e734419 DM |
1451 | sk_setup_caps(newsk, dst); |
1452 | ||
81164413 DB |
1453 | tcp_ca_openreq_child(newsk, dst); |
1454 | ||
1da177e4 | 1455 | tcp_sync_mss(newsk, dst_mtu(dst)); |
3541f9e8 | 1456 | newtp->advmss = tcp_mss_clamp(tcp_sk(sk), dst_metric_advmss(dst)); |
f5fff5dc | 1457 | |
1da177e4 LT |
1458 | tcp_initialize_rcv_mss(newsk); |
1459 | ||
cfb6eeb4 YH |
1460 | #ifdef CONFIG_TCP_MD5SIG |
1461 | /* Copy over the MD5 key from the original socket */ | |
a915da9b ED |
1462 | key = tcp_md5_do_lookup(sk, (union tcp_md5_addr *)&newinet->inet_daddr, |
1463 | AF_INET); | |
00db4124 | 1464 | if (key) { |
cfb6eeb4 YH |
1465 | /* |
1466 | * We're using one, so create a matching key | |
1467 | * on the newsk structure. If we fail to get | |
1468 | * memory, then we end up not copying the key | |
1469 | * across. Shucks. | |
1470 | */ | |
a915da9b | 1471 | tcp_md5_do_add(newsk, (union tcp_md5_addr *)&newinet->inet_daddr, |
6797318e | 1472 | AF_INET, 32, key->key, key->keylen, GFP_ATOMIC); |
a465419b | 1473 | sk_nocaps_add(newsk, NETIF_F_GSO_MASK); |
cfb6eeb4 YH |
1474 | } |
1475 | #endif | |
1476 | ||
0e734419 DM |
1477 | if (__inet_inherit_port(sk, newsk) < 0) |
1478 | goto put_and_exit; | |
5e0724d0 | 1479 | *own_req = inet_ehash_nolisten(newsk, req_to_sk(req_unhash)); |
c92e8c02 | 1480 | if (likely(*own_req)) { |
49a496c9 | 1481 | tcp_move_syn(newtp, req); |
c92e8c02 ED |
1482 | ireq->ireq_opt = NULL; |
1483 | } else { | |
1484 | newinet->inet_opt = NULL; | |
1485 | } | |
1da177e4 LT |
1486 | return newsk; |
1487 | ||
1488 | exit_overflow: | |
c10d9310 | 1489 | NET_INC_STATS(sock_net(sk), LINUX_MIB_LISTENOVERFLOWS); |
093d2823 BS |
1490 | exit_nonewsk: |
1491 | dst_release(dst); | |
1da177e4 | 1492 | exit: |
9caad864 | 1493 | tcp_listendrop(sk); |
1da177e4 | 1494 | return NULL; |
0e734419 | 1495 | put_and_exit: |
c92e8c02 | 1496 | newinet->inet_opt = NULL; |
e337e24d CP |
1497 | inet_csk_prepare_forced_close(newsk); |
1498 | tcp_done(newsk); | |
0e734419 | 1499 | goto exit; |
1da177e4 | 1500 | } |
4bc2f18b | 1501 | EXPORT_SYMBOL(tcp_v4_syn_recv_sock); |
1da177e4 | 1502 | |
079096f1 | 1503 | static struct sock *tcp_v4_cookie_check(struct sock *sk, struct sk_buff *skb) |
1da177e4 | 1504 | { |
079096f1 | 1505 | #ifdef CONFIG_SYN_COOKIES |
52452c54 | 1506 | const struct tcphdr *th = tcp_hdr(skb); |
1da177e4 | 1507 | |
af9b4738 | 1508 | if (!th->syn) |
461b74c3 | 1509 | sk = cookie_v4_check(sk, skb); |
1da177e4 LT |
1510 | #endif |
1511 | return sk; | |
1512 | } | |
1513 | ||
1da177e4 | 1514 | /* The socket must have it's spinlock held when we get |
e994b2f0 | 1515 | * here, unless it is a TCP_LISTEN socket. |
1da177e4 LT |
1516 | * |
1517 | * We have a potential double-lock case here, so even when | |
1518 | * doing backlog processing we use the BH locking scheme. | |
1519 | * This is because we cannot sleep with the original spinlock | |
1520 | * held. | |
1521 | */ | |
1522 | int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb) | |
1523 | { | |
cfb6eeb4 | 1524 | struct sock *rsk; |
cfb6eeb4 | 1525 | |
1da177e4 | 1526 | if (sk->sk_state == TCP_ESTABLISHED) { /* Fast path */ |
404e0a8b ED |
1527 | struct dst_entry *dst = sk->sk_rx_dst; |
1528 | ||
bdeab991 | 1529 | sock_rps_save_rxhash(sk, skb); |
3d97379a | 1530 | sk_mark_napi_id(sk, skb); |
404e0a8b | 1531 | if (dst) { |
505fbcf0 | 1532 | if (inet_sk(sk)->rx_dst_ifindex != skb->skb_iif || |
51456b29 | 1533 | !dst->ops->check(dst, 0)) { |
92101b3b DM |
1534 | dst_release(dst); |
1535 | sk->sk_rx_dst = NULL; | |
1536 | } | |
1537 | } | |
3d97d88e | 1538 | tcp_rcv_established(sk, skb); |
1da177e4 LT |
1539 | return 0; |
1540 | } | |
1541 | ||
12e25e10 | 1542 | if (tcp_checksum_complete(skb)) |
1da177e4 LT |
1543 | goto csum_err; |
1544 | ||
1545 | if (sk->sk_state == TCP_LISTEN) { | |
079096f1 ED |
1546 | struct sock *nsk = tcp_v4_cookie_check(sk, skb); |
1547 | ||
1da177e4 LT |
1548 | if (!nsk) |
1549 | goto discard; | |
1da177e4 | 1550 | if (nsk != sk) { |
cfb6eeb4 YH |
1551 | if (tcp_child_process(sk, nsk, skb)) { |
1552 | rsk = nsk; | |
1da177e4 | 1553 | goto reset; |
cfb6eeb4 | 1554 | } |
1da177e4 LT |
1555 | return 0; |
1556 | } | |
ca55158c | 1557 | } else |
bdeab991 | 1558 | sock_rps_save_rxhash(sk, skb); |
ca55158c | 1559 | |
72ab4a86 | 1560 | if (tcp_rcv_state_process(sk, skb)) { |
cfb6eeb4 | 1561 | rsk = sk; |
1da177e4 | 1562 | goto reset; |
cfb6eeb4 | 1563 | } |
1da177e4 LT |
1564 | return 0; |
1565 | ||
1566 | reset: | |
cfb6eeb4 | 1567 | tcp_v4_send_reset(rsk, skb); |
1da177e4 LT |
1568 | discard: |
1569 | kfree_skb(skb); | |
1570 | /* Be careful here. If this function gets more complicated and | |
1571 | * gcc suffers from register pressure on the x86, sk (in %ebx) | |
1572 | * might be destroyed here. This current version compiles correctly, | |
1573 | * but you have been warned. | |
1574 | */ | |
1575 | return 0; | |
1576 | ||
1577 | csum_err: | |
c10d9310 ED |
1578 | TCP_INC_STATS(sock_net(sk), TCP_MIB_CSUMERRORS); |
1579 | TCP_INC_STATS(sock_net(sk), TCP_MIB_INERRS); | |
1da177e4 LT |
1580 | goto discard; |
1581 | } | |
4bc2f18b | 1582 | EXPORT_SYMBOL(tcp_v4_do_rcv); |
1da177e4 | 1583 | |
7487449c | 1584 | int tcp_v4_early_demux(struct sk_buff *skb) |
41063e9d | 1585 | { |
41063e9d DM |
1586 | const struct iphdr *iph; |
1587 | const struct tcphdr *th; | |
1588 | struct sock *sk; | |
41063e9d | 1589 | |
41063e9d | 1590 | if (skb->pkt_type != PACKET_HOST) |
7487449c | 1591 | return 0; |
41063e9d | 1592 | |
45f00f99 | 1593 | if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct tcphdr))) |
7487449c | 1594 | return 0; |
41063e9d DM |
1595 | |
1596 | iph = ip_hdr(skb); | |
45f00f99 | 1597 | th = tcp_hdr(skb); |
41063e9d DM |
1598 | |
1599 | if (th->doff < sizeof(struct tcphdr) / 4) | |
7487449c | 1600 | return 0; |
41063e9d | 1601 | |
45f00f99 | 1602 | sk = __inet_lookup_established(dev_net(skb->dev), &tcp_hashinfo, |
41063e9d | 1603 | iph->saddr, th->source, |
7011d085 | 1604 | iph->daddr, ntohs(th->dest), |
3fa6f616 | 1605 | skb->skb_iif, inet_sdif(skb)); |
41063e9d DM |
1606 | if (sk) { |
1607 | skb->sk = sk; | |
1608 | skb->destructor = sock_edemux; | |
f7e4eb03 | 1609 | if (sk_fullsock(sk)) { |
d0c294c5 | 1610 | struct dst_entry *dst = READ_ONCE(sk->sk_rx_dst); |
505fbcf0 | 1611 | |
41063e9d DM |
1612 | if (dst) |
1613 | dst = dst_check(dst, 0); | |
92101b3b | 1614 | if (dst && |
505fbcf0 | 1615 | inet_sk(sk)->rx_dst_ifindex == skb->skb_iif) |
92101b3b | 1616 | skb_dst_set_noref(skb, dst); |
41063e9d DM |
1617 | } |
1618 | } | |
7487449c | 1619 | return 0; |
41063e9d DM |
1620 | } |
1621 | ||
c9c33212 ED |
1622 | bool tcp_add_backlog(struct sock *sk, struct sk_buff *skb) |
1623 | { | |
1624 | u32 limit = sk->sk_rcvbuf + sk->sk_sndbuf; | |
4f693b55 ED |
1625 | struct skb_shared_info *shinfo; |
1626 | const struct tcphdr *th; | |
1627 | struct tcphdr *thtail; | |
1628 | struct sk_buff *tail; | |
1629 | unsigned int hdrlen; | |
1630 | bool fragstolen; | |
1631 | u32 gso_segs; | |
1632 | int delta; | |
c9c33212 ED |
1633 | |
1634 | /* In case all data was pulled from skb frags (in __pskb_pull_tail()), | |
1635 | * we can fix skb->truesize to its real value to avoid future drops. | |
1636 | * This is valid because skb is not yet charged to the socket. | |
1637 | * It has been noticed pure SACK packets were sometimes dropped | |
1638 | * (if cooked by drivers without copybreak feature). | |
1639 | */ | |
60b1af33 | 1640 | skb_condense(skb); |
c9c33212 | 1641 | |
ade9628e ED |
1642 | skb_dst_drop(skb); |
1643 | ||
4f693b55 ED |
1644 | if (unlikely(tcp_checksum_complete(skb))) { |
1645 | bh_unlock_sock(sk); | |
1646 | __TCP_INC_STATS(sock_net(sk), TCP_MIB_CSUMERRORS); | |
1647 | __TCP_INC_STATS(sock_net(sk), TCP_MIB_INERRS); | |
1648 | return true; | |
1649 | } | |
1650 | ||
1651 | /* Attempt coalescing to last skb in backlog, even if we are | |
1652 | * above the limits. | |
1653 | * This is okay because skb capacity is limited to MAX_SKB_FRAGS. | |
1654 | */ | |
1655 | th = (const struct tcphdr *)skb->data; | |
1656 | hdrlen = th->doff * 4; | |
1657 | shinfo = skb_shinfo(skb); | |
1658 | ||
1659 | if (!shinfo->gso_size) | |
1660 | shinfo->gso_size = skb->len - hdrlen; | |
1661 | ||
1662 | if (!shinfo->gso_segs) | |
1663 | shinfo->gso_segs = 1; | |
1664 | ||
1665 | tail = sk->sk_backlog.tail; | |
1666 | if (!tail) | |
1667 | goto no_coalesce; | |
1668 | thtail = (struct tcphdr *)tail->data; | |
1669 | ||
1670 | if (TCP_SKB_CB(tail)->end_seq != TCP_SKB_CB(skb)->seq || | |
1671 | TCP_SKB_CB(tail)->ip_dsfield != TCP_SKB_CB(skb)->ip_dsfield || | |
1672 | ((TCP_SKB_CB(tail)->tcp_flags | | |
1673 | TCP_SKB_CB(skb)->tcp_flags) & TCPHDR_URG) || | |
1674 | ((TCP_SKB_CB(tail)->tcp_flags ^ | |
1675 | TCP_SKB_CB(skb)->tcp_flags) & (TCPHDR_ECE | TCPHDR_CWR)) || | |
1676 | #ifdef CONFIG_TLS_DEVICE | |
1677 | tail->decrypted != skb->decrypted || | |
1678 | #endif | |
1679 | thtail->doff != th->doff || | |
1680 | memcmp(thtail + 1, th + 1, hdrlen - sizeof(*th))) | |
1681 | goto no_coalesce; | |
1682 | ||
1683 | __skb_pull(skb, hdrlen); | |
1684 | if (skb_try_coalesce(tail, skb, &fragstolen, &delta)) { | |
1685 | thtail->window = th->window; | |
1686 | ||
1687 | TCP_SKB_CB(tail)->end_seq = TCP_SKB_CB(skb)->end_seq; | |
1688 | ||
1689 | if (after(TCP_SKB_CB(skb)->ack_seq, TCP_SKB_CB(tail)->ack_seq)) | |
1690 | TCP_SKB_CB(tail)->ack_seq = TCP_SKB_CB(skb)->ack_seq; | |
1691 | ||
1692 | TCP_SKB_CB(tail)->tcp_flags |= TCP_SKB_CB(skb)->tcp_flags; | |
1693 | ||
1694 | if (TCP_SKB_CB(skb)->has_rxtstamp) { | |
1695 | TCP_SKB_CB(tail)->has_rxtstamp = true; | |
1696 | tail->tstamp = skb->tstamp; | |
1697 | skb_hwtstamps(tail)->hwtstamp = skb_hwtstamps(skb)->hwtstamp; | |
1698 | } | |
1699 | ||
1700 | /* Not as strict as GRO. We only need to carry mss max value */ | |
1701 | skb_shinfo(tail)->gso_size = max(shinfo->gso_size, | |
1702 | skb_shinfo(tail)->gso_size); | |
1703 | ||
1704 | gso_segs = skb_shinfo(tail)->gso_segs + shinfo->gso_segs; | |
1705 | skb_shinfo(tail)->gso_segs = min_t(u32, gso_segs, 0xFFFF); | |
1706 | ||
1707 | sk->sk_backlog.len += delta; | |
1708 | __NET_INC_STATS(sock_net(sk), | |
1709 | LINUX_MIB_TCPBACKLOGCOALESCE); | |
1710 | kfree_skb_partial(skb, fragstolen); | |
1711 | return false; | |
1712 | } | |
1713 | __skb_push(skb, hdrlen); | |
1714 | ||
1715 | no_coalesce: | |
1716 | /* Only socket owner can try to collapse/prune rx queues | |
1717 | * to reduce memory overhead, so add a little headroom here. | |
1718 | * Few sockets backlog are possibly concurrently non empty. | |
1719 | */ | |
1720 | limit += 64*1024; | |
1721 | ||
c9c33212 ED |
1722 | if (unlikely(sk_add_backlog(sk, skb, limit))) { |
1723 | bh_unlock_sock(sk); | |
1724 | __NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPBACKLOGDROP); | |
1725 | return true; | |
1726 | } | |
1727 | return false; | |
1728 | } | |
1729 | EXPORT_SYMBOL(tcp_add_backlog); | |
1730 | ||
ac6e7800 ED |
1731 | int tcp_filter(struct sock *sk, struct sk_buff *skb) |
1732 | { | |
1733 | struct tcphdr *th = (struct tcphdr *)skb->data; | |
1734 | unsigned int eaten = skb->len; | |
1735 | int err; | |
1736 | ||
1737 | err = sk_filter_trim_cap(sk, skb, th->doff * 4); | |
1738 | if (!err) { | |
1739 | eaten -= skb->len; | |
1740 | TCP_SKB_CB(skb)->end_seq -= eaten; | |
1741 | } | |
1742 | return err; | |
1743 | } | |
1744 | EXPORT_SYMBOL(tcp_filter); | |
1745 | ||
eeea10b8 ED |
1746 | static void tcp_v4_restore_cb(struct sk_buff *skb) |
1747 | { | |
1748 | memmove(IPCB(skb), &TCP_SKB_CB(skb)->header.h4, | |
1749 | sizeof(struct inet_skb_parm)); | |
1750 | } | |
1751 | ||
1752 | static void tcp_v4_fill_cb(struct sk_buff *skb, const struct iphdr *iph, | |
1753 | const struct tcphdr *th) | |
1754 | { | |
1755 | /* This is tricky : We move IPCB at its correct location into TCP_SKB_CB() | |
1756 | * barrier() makes sure compiler wont play fool^Waliasing games. | |
1757 | */ | |
1758 | memmove(&TCP_SKB_CB(skb)->header.h4, IPCB(skb), | |
1759 | sizeof(struct inet_skb_parm)); | |
1760 | barrier(); | |
1761 | ||
1762 | TCP_SKB_CB(skb)->seq = ntohl(th->seq); | |
1763 | TCP_SKB_CB(skb)->end_seq = (TCP_SKB_CB(skb)->seq + th->syn + th->fin + | |
1764 | skb->len - th->doff * 4); | |
1765 | TCP_SKB_CB(skb)->ack_seq = ntohl(th->ack_seq); | |
1766 | TCP_SKB_CB(skb)->tcp_flags = tcp_flag_byte(th); | |
1767 | TCP_SKB_CB(skb)->tcp_tw_isn = 0; | |
1768 | TCP_SKB_CB(skb)->ip_dsfield = ipv4_get_dsfield(iph); | |
1769 | TCP_SKB_CB(skb)->sacked = 0; | |
1770 | TCP_SKB_CB(skb)->has_rxtstamp = | |
1771 | skb->tstamp || skb_hwtstamps(skb)->hwtstamp; | |
1772 | } | |
1773 | ||
1da177e4 LT |
1774 | /* |
1775 | * From tcp_input.c | |
1776 | */ | |
1777 | ||
1778 | int tcp_v4_rcv(struct sk_buff *skb) | |
1779 | { | |
3b24d854 | 1780 | struct net *net = dev_net(skb->dev); |
3fa6f616 | 1781 | int sdif = inet_sdif(skb); |
eddc9ec5 | 1782 | const struct iphdr *iph; |
cf533ea5 | 1783 | const struct tcphdr *th; |
3b24d854 | 1784 | bool refcounted; |
1da177e4 LT |
1785 | struct sock *sk; |
1786 | int ret; | |
1787 | ||
1788 | if (skb->pkt_type != PACKET_HOST) | |
1789 | goto discard_it; | |
1790 | ||
1791 | /* Count it even if it's bad */ | |
90bbcc60 | 1792 | __TCP_INC_STATS(net, TCP_MIB_INSEGS); |
1da177e4 LT |
1793 | |
1794 | if (!pskb_may_pull(skb, sizeof(struct tcphdr))) | |
1795 | goto discard_it; | |
1796 | ||
ea1627c2 | 1797 | th = (const struct tcphdr *)skb->data; |
1da177e4 | 1798 | |
ea1627c2 | 1799 | if (unlikely(th->doff < sizeof(struct tcphdr) / 4)) |
1da177e4 LT |
1800 | goto bad_packet; |
1801 | if (!pskb_may_pull(skb, th->doff * 4)) | |
1802 | goto discard_it; | |
1803 | ||
1804 | /* An explanation is required here, I think. | |
1805 | * Packet length and doff are validated by header prediction, | |
caa20d9a | 1806 | * provided case of th->doff==0 is eliminated. |
1da177e4 | 1807 | * So, we defer the checks. */ |
ed70fcfc TH |
1808 | |
1809 | if (skb_checksum_init(skb, IPPROTO_TCP, inet_compute_pseudo)) | |
6a5dc9e5 | 1810 | goto csum_error; |
1da177e4 | 1811 | |
ea1627c2 | 1812 | th = (const struct tcphdr *)skb->data; |
eddc9ec5 | 1813 | iph = ip_hdr(skb); |
4bdc3d66 | 1814 | lookup: |
a583636a | 1815 | sk = __inet_lookup_skb(&tcp_hashinfo, skb, __tcp_hdrlen(th), th->source, |
3fa6f616 | 1816 | th->dest, sdif, &refcounted); |
1da177e4 LT |
1817 | if (!sk) |
1818 | goto no_tcp_socket; | |
1819 | ||
bb134d5d ED |
1820 | process: |
1821 | if (sk->sk_state == TCP_TIME_WAIT) | |
1822 | goto do_time_wait; | |
1823 | ||
079096f1 ED |
1824 | if (sk->sk_state == TCP_NEW_SYN_RECV) { |
1825 | struct request_sock *req = inet_reqsk(sk); | |
e0f9759f | 1826 | bool req_stolen = false; |
7716682c | 1827 | struct sock *nsk; |
079096f1 ED |
1828 | |
1829 | sk = req->rsk_listener; | |
72923555 | 1830 | if (unlikely(tcp_v4_inbound_md5_hash(sk, skb))) { |
e65c332d | 1831 | sk_drops_add(sk, skb); |
72923555 ED |
1832 | reqsk_put(req); |
1833 | goto discard_it; | |
1834 | } | |
4fd44a98 FL |
1835 | if (tcp_checksum_complete(skb)) { |
1836 | reqsk_put(req); | |
1837 | goto csum_error; | |
1838 | } | |
7716682c | 1839 | if (unlikely(sk->sk_state != TCP_LISTEN)) { |
f03f2e15 | 1840 | inet_csk_reqsk_queue_drop_and_put(sk, req); |
4bdc3d66 ED |
1841 | goto lookup; |
1842 | } | |
3b24d854 ED |
1843 | /* We own a reference on the listener, increase it again |
1844 | * as we might lose it too soon. | |
1845 | */ | |
7716682c | 1846 | sock_hold(sk); |
3b24d854 | 1847 | refcounted = true; |
1f3b359f | 1848 | nsk = NULL; |
eeea10b8 ED |
1849 | if (!tcp_filter(sk, skb)) { |
1850 | th = (const struct tcphdr *)skb->data; | |
1851 | iph = ip_hdr(skb); | |
1852 | tcp_v4_fill_cb(skb, iph, th); | |
e0f9759f | 1853 | nsk = tcp_check_req(sk, skb, req, false, &req_stolen); |
eeea10b8 | 1854 | } |
079096f1 ED |
1855 | if (!nsk) { |
1856 | reqsk_put(req); | |
e0f9759f ED |
1857 | if (req_stolen) { |
1858 | /* Another cpu got exclusive access to req | |
1859 | * and created a full blown socket. | |
1860 | * Try to feed this packet to this socket | |
1861 | * instead of discarding it. | |
1862 | */ | |
1863 | tcp_v4_restore_cb(skb); | |
1864 | sock_put(sk); | |
1865 | goto lookup; | |
1866 | } | |
7716682c | 1867 | goto discard_and_relse; |
079096f1 ED |
1868 | } |
1869 | if (nsk == sk) { | |
079096f1 | 1870 | reqsk_put(req); |
eeea10b8 | 1871 | tcp_v4_restore_cb(skb); |
079096f1 ED |
1872 | } else if (tcp_child_process(sk, nsk, skb)) { |
1873 | tcp_v4_send_reset(nsk, skb); | |
7716682c | 1874 | goto discard_and_relse; |
079096f1 | 1875 | } else { |
7716682c | 1876 | sock_put(sk); |
079096f1 ED |
1877 | return 0; |
1878 | } | |
1879 | } | |
6cce09f8 | 1880 | if (unlikely(iph->ttl < inet_sk(sk)->min_ttl)) { |
02a1d6e7 | 1881 | __NET_INC_STATS(net, LINUX_MIB_TCPMINTTLDROP); |
d218d111 | 1882 | goto discard_and_relse; |
6cce09f8 | 1883 | } |
d218d111 | 1884 | |
1da177e4 LT |
1885 | if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb)) |
1886 | goto discard_and_relse; | |
9ea88a15 | 1887 | |
9ea88a15 DP |
1888 | if (tcp_v4_inbound_md5_hash(sk, skb)) |
1889 | goto discard_and_relse; | |
9ea88a15 | 1890 | |
b59c2701 | 1891 | nf_reset(skb); |
1da177e4 | 1892 | |
ac6e7800 | 1893 | if (tcp_filter(sk, skb)) |
1da177e4 | 1894 | goto discard_and_relse; |
ac6e7800 ED |
1895 | th = (const struct tcphdr *)skb->data; |
1896 | iph = ip_hdr(skb); | |
eeea10b8 | 1897 | tcp_v4_fill_cb(skb, iph, th); |
1da177e4 LT |
1898 | |
1899 | skb->dev = NULL; | |
1900 | ||
e994b2f0 ED |
1901 | if (sk->sk_state == TCP_LISTEN) { |
1902 | ret = tcp_v4_do_rcv(sk, skb); | |
1903 | goto put_and_return; | |
1904 | } | |
1905 | ||
1906 | sk_incoming_cpu_update(sk); | |
1907 | ||
c6366184 | 1908 | bh_lock_sock_nested(sk); |
a44d6eac | 1909 | tcp_segs_in(tcp_sk(sk), skb); |
1da177e4 LT |
1910 | ret = 0; |
1911 | if (!sock_owned_by_user(sk)) { | |
e7942d06 | 1912 | ret = tcp_v4_do_rcv(sk, skb); |
c9c33212 | 1913 | } else if (tcp_add_backlog(sk, skb)) { |
6b03a53a ZY |
1914 | goto discard_and_relse; |
1915 | } | |
1da177e4 LT |
1916 | bh_unlock_sock(sk); |
1917 | ||
e994b2f0 | 1918 | put_and_return: |
3b24d854 ED |
1919 | if (refcounted) |
1920 | sock_put(sk); | |
1da177e4 LT |
1921 | |
1922 | return ret; | |
1923 | ||
1924 | no_tcp_socket: | |
1925 | if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) | |
1926 | goto discard_it; | |
1927 | ||
eeea10b8 ED |
1928 | tcp_v4_fill_cb(skb, iph, th); |
1929 | ||
12e25e10 | 1930 | if (tcp_checksum_complete(skb)) { |
6a5dc9e5 | 1931 | csum_error: |
90bbcc60 | 1932 | __TCP_INC_STATS(net, TCP_MIB_CSUMERRORS); |
1da177e4 | 1933 | bad_packet: |
90bbcc60 | 1934 | __TCP_INC_STATS(net, TCP_MIB_INERRS); |
1da177e4 | 1935 | } else { |
cfb6eeb4 | 1936 | tcp_v4_send_reset(NULL, skb); |
1da177e4 LT |
1937 | } |
1938 | ||
1939 | discard_it: | |
1940 | /* Discard frame. */ | |
1941 | kfree_skb(skb); | |
e905a9ed | 1942 | return 0; |
1da177e4 LT |
1943 | |
1944 | discard_and_relse: | |
532182cd | 1945 | sk_drops_add(sk, skb); |
3b24d854 ED |
1946 | if (refcounted) |
1947 | sock_put(sk); | |
1da177e4 LT |
1948 | goto discard_it; |
1949 | ||
1950 | do_time_wait: | |
1951 | if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) { | |
9469c7b4 | 1952 | inet_twsk_put(inet_twsk(sk)); |
1da177e4 LT |
1953 | goto discard_it; |
1954 | } | |
1955 | ||
eeea10b8 ED |
1956 | tcp_v4_fill_cb(skb, iph, th); |
1957 | ||
6a5dc9e5 ED |
1958 | if (tcp_checksum_complete(skb)) { |
1959 | inet_twsk_put(inet_twsk(sk)); | |
1960 | goto csum_error; | |
1da177e4 | 1961 | } |
9469c7b4 | 1962 | switch (tcp_timewait_state_process(inet_twsk(sk), skb, th)) { |
1da177e4 | 1963 | case TCP_TW_SYN: { |
c346dca1 | 1964 | struct sock *sk2 = inet_lookup_listener(dev_net(skb->dev), |
a583636a CG |
1965 | &tcp_hashinfo, skb, |
1966 | __tcp_hdrlen(th), | |
da5e3630 | 1967 | iph->saddr, th->source, |
eddc9ec5 | 1968 | iph->daddr, th->dest, |
3fa6f616 DA |
1969 | inet_iif(skb), |
1970 | sdif); | |
1da177e4 | 1971 | if (sk2) { |
dbe7faa4 | 1972 | inet_twsk_deschedule_put(inet_twsk(sk)); |
1da177e4 | 1973 | sk = sk2; |
eeea10b8 | 1974 | tcp_v4_restore_cb(skb); |
3b24d854 | 1975 | refcounted = false; |
1da177e4 LT |
1976 | goto process; |
1977 | } | |
1da177e4 | 1978 | } |
fcfd6dfa GS |
1979 | /* to ACK */ |
1980 | /* fall through */ | |
1da177e4 LT |
1981 | case TCP_TW_ACK: |
1982 | tcp_v4_timewait_ack(sk, skb); | |
1983 | break; | |
1984 | case TCP_TW_RST: | |
271c3b9b FW |
1985 | tcp_v4_send_reset(sk, skb); |
1986 | inet_twsk_deschedule_put(inet_twsk(sk)); | |
1987 | goto discard_it; | |
1da177e4 LT |
1988 | case TCP_TW_SUCCESS:; |
1989 | } | |
1990 | goto discard_it; | |
1991 | } | |
1992 | ||
ccb7c410 DM |
1993 | static struct timewait_sock_ops tcp_timewait_sock_ops = { |
1994 | .twsk_obj_size = sizeof(struct tcp_timewait_sock), | |
1995 | .twsk_unique = tcp_twsk_unique, | |
1996 | .twsk_destructor= tcp_twsk_destructor, | |
ccb7c410 | 1997 | }; |
1da177e4 | 1998 | |
63d02d15 | 1999 | void inet_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb) |
5d299f3d ED |
2000 | { |
2001 | struct dst_entry *dst = skb_dst(skb); | |
2002 | ||
5037e9ef | 2003 | if (dst && dst_hold_safe(dst)) { |
ca777eff ED |
2004 | sk->sk_rx_dst = dst; |
2005 | inet_sk(sk)->rx_dst_ifindex = skb->skb_iif; | |
2006 | } | |
5d299f3d | 2007 | } |
63d02d15 | 2008 | EXPORT_SYMBOL(inet_sk_rx_dst_set); |
5d299f3d | 2009 | |
3b401a81 | 2010 | const struct inet_connection_sock_af_ops ipv4_specific = { |
543d9cfe ACM |
2011 | .queue_xmit = ip_queue_xmit, |
2012 | .send_check = tcp_v4_send_check, | |
2013 | .rebuild_header = inet_sk_rebuild_header, | |
5d299f3d | 2014 | .sk_rx_dst_set = inet_sk_rx_dst_set, |
543d9cfe ACM |
2015 | .conn_request = tcp_v4_conn_request, |
2016 | .syn_recv_sock = tcp_v4_syn_recv_sock, | |
543d9cfe ACM |
2017 | .net_header_len = sizeof(struct iphdr), |
2018 | .setsockopt = ip_setsockopt, | |
2019 | .getsockopt = ip_getsockopt, | |
2020 | .addr2sockaddr = inet_csk_addr2sockaddr, | |
2021 | .sockaddr_len = sizeof(struct sockaddr_in), | |
3fdadf7d | 2022 | #ifdef CONFIG_COMPAT |
543d9cfe ACM |
2023 | .compat_setsockopt = compat_ip_setsockopt, |
2024 | .compat_getsockopt = compat_ip_getsockopt, | |
3fdadf7d | 2025 | #endif |
4fab9071 | 2026 | .mtu_reduced = tcp_v4_mtu_reduced, |
1da177e4 | 2027 | }; |
4bc2f18b | 2028 | EXPORT_SYMBOL(ipv4_specific); |
1da177e4 | 2029 | |
cfb6eeb4 | 2030 | #ifdef CONFIG_TCP_MD5SIG |
b2e4b3de | 2031 | static const struct tcp_sock_af_ops tcp_sock_ipv4_specific = { |
cfb6eeb4 | 2032 | .md5_lookup = tcp_v4_md5_lookup, |
49a72dfb | 2033 | .calc_md5_hash = tcp_v4_md5_hash_skb, |
cfb6eeb4 | 2034 | .md5_parse = tcp_v4_parse_md5_keys, |
cfb6eeb4 | 2035 | }; |
b6332e6c | 2036 | #endif |
cfb6eeb4 | 2037 | |
1da177e4 LT |
2038 | /* NOTE: A lot of things set to zero explicitly by call to |
2039 | * sk_alloc() so need not be done here. | |
2040 | */ | |
2041 | static int tcp_v4_init_sock(struct sock *sk) | |
2042 | { | |
6687e988 | 2043 | struct inet_connection_sock *icsk = inet_csk(sk); |
1da177e4 | 2044 | |
900f65d3 | 2045 | tcp_init_sock(sk); |
1da177e4 | 2046 | |
8292a17a | 2047 | icsk->icsk_af_ops = &ipv4_specific; |
900f65d3 | 2048 | |
cfb6eeb4 | 2049 | #ifdef CONFIG_TCP_MD5SIG |
ac807fa8 | 2050 | tcp_sk(sk)->af_specific = &tcp_sock_ipv4_specific; |
cfb6eeb4 | 2051 | #endif |
1da177e4 | 2052 | |
1da177e4 LT |
2053 | return 0; |
2054 | } | |
2055 | ||
7d06b2e0 | 2056 | void tcp_v4_destroy_sock(struct sock *sk) |
1da177e4 LT |
2057 | { |
2058 | struct tcp_sock *tp = tcp_sk(sk); | |
2059 | ||
e1a4aa50 SL |
2060 | trace_tcp_destroy_sock(sk); |
2061 | ||
1da177e4 LT |
2062 | tcp_clear_xmit_timers(sk); |
2063 | ||
6687e988 | 2064 | tcp_cleanup_congestion_control(sk); |
317a76f9 | 2065 | |
734942cc DW |
2066 | tcp_cleanup_ulp(sk); |
2067 | ||
1da177e4 | 2068 | /* Cleanup up the write buffer. */ |
fe067e8a | 2069 | tcp_write_queue_purge(sk); |
1da177e4 | 2070 | |
cf1ef3f0 WW |
2071 | /* Check if we want to disable active TFO */ |
2072 | tcp_fastopen_active_disable_ofo_check(sk); | |
2073 | ||
1da177e4 | 2074 | /* Cleans up our, hopefully empty, out_of_order_queue. */ |
9f5afeae | 2075 | skb_rbtree_purge(&tp->out_of_order_queue); |
1da177e4 | 2076 | |
cfb6eeb4 YH |
2077 | #ifdef CONFIG_TCP_MD5SIG |
2078 | /* Clean up the MD5 key list, if any */ | |
2079 | if (tp->md5sig_info) { | |
a915da9b | 2080 | tcp_clear_md5_list(sk); |
fb7df5e4 | 2081 | kfree_rcu(rcu_dereference_protected(tp->md5sig_info, 1), rcu); |
cfb6eeb4 YH |
2082 | tp->md5sig_info = NULL; |
2083 | } | |
2084 | #endif | |
1a2449a8 | 2085 | |
1da177e4 | 2086 | /* Clean up a referenced TCP bind bucket. */ |
463c84b9 | 2087 | if (inet_csk(sk)->icsk_bind_hash) |
ab1e0a13 | 2088 | inet_put_port(sk); |
1da177e4 | 2089 | |
00db4124 | 2090 | BUG_ON(tp->fastopen_rsk); |
435cf559 | 2091 | |
cf60af03 YC |
2092 | /* If socket is aborted during connect operation */ |
2093 | tcp_free_fastopen_req(tp); | |
1fba70e5 | 2094 | tcp_fastopen_destroy_cipher(sk); |
cd8ae852 | 2095 | tcp_saved_syn_free(tp); |
cf60af03 | 2096 | |
180d8cd9 | 2097 | sk_sockets_allocated_dec(sk); |
1da177e4 | 2098 | } |
1da177e4 LT |
2099 | EXPORT_SYMBOL(tcp_v4_destroy_sock); |
2100 | ||
2101 | #ifdef CONFIG_PROC_FS | |
2102 | /* Proc filesystem TCP sock list dumping. */ | |
2103 | ||
a8b690f9 TH |
2104 | /* |
2105 | * Get next listener socket follow cur. If cur is NULL, get first socket | |
2106 | * starting from bucket given in st->bucket; when st->bucket is zero the | |
2107 | * very first socket in the hash table is returned. | |
2108 | */ | |
1da177e4 LT |
2109 | static void *listening_get_next(struct seq_file *seq, void *cur) |
2110 | { | |
37d849bb | 2111 | struct tcp_seq_afinfo *afinfo = PDE_DATA(file_inode(seq->file)); |
5799de0b | 2112 | struct tcp_iter_state *st = seq->private; |
a4146b1b | 2113 | struct net *net = seq_file_net(seq); |
3b24d854 | 2114 | struct inet_listen_hashbucket *ilb; |
3b24d854 | 2115 | struct sock *sk = cur; |
1da177e4 LT |
2116 | |
2117 | if (!sk) { | |
3b24d854 | 2118 | get_head: |
a8b690f9 | 2119 | ilb = &tcp_hashinfo.listening_hash[st->bucket]; |
9652dc2e | 2120 | spin_lock(&ilb->lock); |
3b24d854 | 2121 | sk = sk_head(&ilb->head); |
a8b690f9 | 2122 | st->offset = 0; |
1da177e4 LT |
2123 | goto get_sk; |
2124 | } | |
5caea4ea | 2125 | ilb = &tcp_hashinfo.listening_hash[st->bucket]; |
1da177e4 | 2126 | ++st->num; |
a8b690f9 | 2127 | ++st->offset; |
1da177e4 | 2128 | |
3b24d854 | 2129 | sk = sk_next(sk); |
1da177e4 | 2130 | get_sk: |
3b24d854 | 2131 | sk_for_each_from(sk) { |
8475ef9f PE |
2132 | if (!net_eq(sock_net(sk), net)) |
2133 | continue; | |
37d849bb | 2134 | if (sk->sk_family == afinfo->family) |
3b24d854 | 2135 | return sk; |
1da177e4 | 2136 | } |
9652dc2e | 2137 | spin_unlock(&ilb->lock); |
a8b690f9 | 2138 | st->offset = 0; |
3b24d854 ED |
2139 | if (++st->bucket < INET_LHTABLE_SIZE) |
2140 | goto get_head; | |
2141 | return NULL; | |
1da177e4 LT |
2142 | } |
2143 | ||
2144 | static void *listening_get_idx(struct seq_file *seq, loff_t *pos) | |
2145 | { | |
a8b690f9 TH |
2146 | struct tcp_iter_state *st = seq->private; |
2147 | void *rc; | |
2148 | ||
2149 | st->bucket = 0; | |
2150 | st->offset = 0; | |
2151 | rc = listening_get_next(seq, NULL); | |
1da177e4 LT |
2152 | |
2153 | while (rc && *pos) { | |
2154 | rc = listening_get_next(seq, rc); | |
2155 | --*pos; | |
2156 | } | |
2157 | return rc; | |
2158 | } | |
2159 | ||
05dbc7b5 | 2160 | static inline bool empty_bucket(const struct tcp_iter_state *st) |
6eac5604 | 2161 | { |
05dbc7b5 | 2162 | return hlist_nulls_empty(&tcp_hashinfo.ehash[st->bucket].chain); |
6eac5604 AK |
2163 | } |
2164 | ||
a8b690f9 TH |
2165 | /* |
2166 | * Get first established socket starting from bucket given in st->bucket. | |
2167 | * If st->bucket is zero, the very first socket in the hash is returned. | |
2168 | */ | |
1da177e4 LT |
2169 | static void *established_get_first(struct seq_file *seq) |
2170 | { | |
37d849bb | 2171 | struct tcp_seq_afinfo *afinfo = PDE_DATA(file_inode(seq->file)); |
5799de0b | 2172 | struct tcp_iter_state *st = seq->private; |
a4146b1b | 2173 | struct net *net = seq_file_net(seq); |
1da177e4 LT |
2174 | void *rc = NULL; |
2175 | ||
a8b690f9 TH |
2176 | st->offset = 0; |
2177 | for (; st->bucket <= tcp_hashinfo.ehash_mask; ++st->bucket) { | |
1da177e4 | 2178 | struct sock *sk; |
3ab5aee7 | 2179 | struct hlist_nulls_node *node; |
9db66bdc | 2180 | spinlock_t *lock = inet_ehash_lockp(&tcp_hashinfo, st->bucket); |
1da177e4 | 2181 | |
6eac5604 AK |
2182 | /* Lockless fast path for the common case of empty buckets */ |
2183 | if (empty_bucket(st)) | |
2184 | continue; | |
2185 | ||
9db66bdc | 2186 | spin_lock_bh(lock); |
3ab5aee7 | 2187 | sk_nulls_for_each(sk, node, &tcp_hashinfo.ehash[st->bucket].chain) { |
37d849bb | 2188 | if (sk->sk_family != afinfo->family || |
878628fb | 2189 | !net_eq(sock_net(sk), net)) { |
1da177e4 LT |
2190 | continue; |
2191 | } | |
2192 | rc = sk; | |
2193 | goto out; | |
2194 | } | |
9db66bdc | 2195 | spin_unlock_bh(lock); |
1da177e4 LT |
2196 | } |
2197 | out: | |
2198 | return rc; | |
2199 | } | |
2200 | ||
2201 | static void *established_get_next(struct seq_file *seq, void *cur) | |
2202 | { | |
37d849bb | 2203 | struct tcp_seq_afinfo *afinfo = PDE_DATA(file_inode(seq->file)); |
1da177e4 | 2204 | struct sock *sk = cur; |
3ab5aee7 | 2205 | struct hlist_nulls_node *node; |
5799de0b | 2206 | struct tcp_iter_state *st = seq->private; |
a4146b1b | 2207 | struct net *net = seq_file_net(seq); |
1da177e4 LT |
2208 | |
2209 | ++st->num; | |
a8b690f9 | 2210 | ++st->offset; |
1da177e4 | 2211 | |
05dbc7b5 | 2212 | sk = sk_nulls_next(sk); |
1da177e4 | 2213 | |
3ab5aee7 | 2214 | sk_nulls_for_each_from(sk, node) { |
37d849bb CH |
2215 | if (sk->sk_family == afinfo->family && |
2216 | net_eq(sock_net(sk), net)) | |
05dbc7b5 | 2217 | return sk; |
1da177e4 LT |
2218 | } |
2219 | ||
05dbc7b5 ED |
2220 | spin_unlock_bh(inet_ehash_lockp(&tcp_hashinfo, st->bucket)); |
2221 | ++st->bucket; | |
2222 | return established_get_first(seq); | |
1da177e4 LT |
2223 | } |
2224 | ||
2225 | static void *established_get_idx(struct seq_file *seq, loff_t pos) | |
2226 | { | |
a8b690f9 TH |
2227 | struct tcp_iter_state *st = seq->private; |
2228 | void *rc; | |
2229 | ||
2230 | st->bucket = 0; | |
2231 | rc = established_get_first(seq); | |
1da177e4 LT |
2232 | |
2233 | while (rc && pos) { | |
2234 | rc = established_get_next(seq, rc); | |
2235 | --pos; | |
7174259e | 2236 | } |
1da177e4 LT |
2237 | return rc; |
2238 | } | |
2239 | ||
2240 | static void *tcp_get_idx(struct seq_file *seq, loff_t pos) | |
2241 | { | |
2242 | void *rc; | |
5799de0b | 2243 | struct tcp_iter_state *st = seq->private; |
1da177e4 | 2244 | |
1da177e4 LT |
2245 | st->state = TCP_SEQ_STATE_LISTENING; |
2246 | rc = listening_get_idx(seq, &pos); | |
2247 | ||
2248 | if (!rc) { | |
1da177e4 LT |
2249 | st->state = TCP_SEQ_STATE_ESTABLISHED; |
2250 | rc = established_get_idx(seq, pos); | |
2251 | } | |
2252 | ||
2253 | return rc; | |
2254 | } | |
2255 | ||
a8b690f9 TH |
2256 | static void *tcp_seek_last_pos(struct seq_file *seq) |
2257 | { | |
2258 | struct tcp_iter_state *st = seq->private; | |
2259 | int offset = st->offset; | |
2260 | int orig_num = st->num; | |
2261 | void *rc = NULL; | |
2262 | ||
2263 | switch (st->state) { | |
a8b690f9 TH |
2264 | case TCP_SEQ_STATE_LISTENING: |
2265 | if (st->bucket >= INET_LHTABLE_SIZE) | |
2266 | break; | |
2267 | st->state = TCP_SEQ_STATE_LISTENING; | |
2268 | rc = listening_get_next(seq, NULL); | |
2269 | while (offset-- && rc) | |
2270 | rc = listening_get_next(seq, rc); | |
2271 | if (rc) | |
2272 | break; | |
2273 | st->bucket = 0; | |
05dbc7b5 | 2274 | st->state = TCP_SEQ_STATE_ESTABLISHED; |
a8b690f9 TH |
2275 | /* Fallthrough */ |
2276 | case TCP_SEQ_STATE_ESTABLISHED: | |
a8b690f9 TH |
2277 | if (st->bucket > tcp_hashinfo.ehash_mask) |
2278 | break; | |
2279 | rc = established_get_first(seq); | |
2280 | while (offset-- && rc) | |
2281 | rc = established_get_next(seq, rc); | |
2282 | } | |
2283 | ||
2284 | st->num = orig_num; | |
2285 | ||
2286 | return rc; | |
2287 | } | |
2288 | ||
37d849bb | 2289 | void *tcp_seq_start(struct seq_file *seq, loff_t *pos) |
1da177e4 | 2290 | { |
5799de0b | 2291 | struct tcp_iter_state *st = seq->private; |
a8b690f9 TH |
2292 | void *rc; |
2293 | ||
2294 | if (*pos && *pos == st->last_pos) { | |
2295 | rc = tcp_seek_last_pos(seq); | |
2296 | if (rc) | |
2297 | goto out; | |
2298 | } | |
2299 | ||
1da177e4 LT |
2300 | st->state = TCP_SEQ_STATE_LISTENING; |
2301 | st->num = 0; | |
a8b690f9 TH |
2302 | st->bucket = 0; |
2303 | st->offset = 0; | |
2304 | rc = *pos ? tcp_get_idx(seq, *pos - 1) : SEQ_START_TOKEN; | |
2305 | ||
2306 | out: | |
2307 | st->last_pos = *pos; | |
2308 | return rc; | |
1da177e4 | 2309 | } |
37d849bb | 2310 | EXPORT_SYMBOL(tcp_seq_start); |
1da177e4 | 2311 | |
37d849bb | 2312 | void *tcp_seq_next(struct seq_file *seq, void *v, loff_t *pos) |
1da177e4 | 2313 | { |
a8b690f9 | 2314 | struct tcp_iter_state *st = seq->private; |
1da177e4 | 2315 | void *rc = NULL; |
1da177e4 LT |
2316 | |
2317 | if (v == SEQ_START_TOKEN) { | |
2318 | rc = tcp_get_idx(seq, 0); | |
2319 | goto out; | |
2320 | } | |
1da177e4 LT |
2321 | |
2322 | switch (st->state) { | |
1da177e4 LT |
2323 | case TCP_SEQ_STATE_LISTENING: |
2324 | rc = listening_get_next(seq, v); | |
2325 | if (!rc) { | |
1da177e4 | 2326 | st->state = TCP_SEQ_STATE_ESTABLISHED; |
a8b690f9 TH |
2327 | st->bucket = 0; |
2328 | st->offset = 0; | |
1da177e4 LT |
2329 | rc = established_get_first(seq); |
2330 | } | |
2331 | break; | |
2332 | case TCP_SEQ_STATE_ESTABLISHED: | |
1da177e4 LT |
2333 | rc = established_get_next(seq, v); |
2334 | break; | |
2335 | } | |
2336 | out: | |
2337 | ++*pos; | |
a8b690f9 | 2338 | st->last_pos = *pos; |
1da177e4 LT |
2339 | return rc; |
2340 | } | |
37d849bb | 2341 | EXPORT_SYMBOL(tcp_seq_next); |
1da177e4 | 2342 | |
37d849bb | 2343 | void tcp_seq_stop(struct seq_file *seq, void *v) |
1da177e4 | 2344 | { |
5799de0b | 2345 | struct tcp_iter_state *st = seq->private; |
1da177e4 LT |
2346 | |
2347 | switch (st->state) { | |
1da177e4 LT |
2348 | case TCP_SEQ_STATE_LISTENING: |
2349 | if (v != SEQ_START_TOKEN) | |
9652dc2e | 2350 | spin_unlock(&tcp_hashinfo.listening_hash[st->bucket].lock); |
1da177e4 | 2351 | break; |
1da177e4 LT |
2352 | case TCP_SEQ_STATE_ESTABLISHED: |
2353 | if (v) | |
9db66bdc | 2354 | spin_unlock_bh(inet_ehash_lockp(&tcp_hashinfo, st->bucket)); |
1da177e4 LT |
2355 | break; |
2356 | } | |
2357 | } | |
37d849bb | 2358 | EXPORT_SYMBOL(tcp_seq_stop); |
1da177e4 | 2359 | |
d4f06873 | 2360 | static void get_openreq4(const struct request_sock *req, |
aa3a0c8c | 2361 | struct seq_file *f, int i) |
1da177e4 | 2362 | { |
2e6599cb | 2363 | const struct inet_request_sock *ireq = inet_rsk(req); |
fa76ce73 | 2364 | long delta = req->rsk_timer.expires - jiffies; |
1da177e4 | 2365 | |
5e659e4c | 2366 | seq_printf(f, "%4d: %08X:%04X %08X:%04X" |
652586df | 2367 | " %02X %08X:%08X %02X:%08lX %08X %5u %8d %u %d %pK", |
1da177e4 | 2368 | i, |
634fb979 | 2369 | ireq->ir_loc_addr, |
d4f06873 | 2370 | ireq->ir_num, |
634fb979 ED |
2371 | ireq->ir_rmt_addr, |
2372 | ntohs(ireq->ir_rmt_port), | |
1da177e4 LT |
2373 | TCP_SYN_RECV, |
2374 | 0, 0, /* could print option size, but that is af dependent. */ | |
2375 | 1, /* timers active (only the expire timer) */ | |
a399a805 | 2376 | jiffies_delta_to_clock_t(delta), |
e6c022a4 | 2377 | req->num_timeout, |
aa3a0c8c ED |
2378 | from_kuid_munged(seq_user_ns(f), |
2379 | sock_i_uid(req->rsk_listener)), | |
1da177e4 LT |
2380 | 0, /* non standard timer */ |
2381 | 0, /* open_requests have no inode */ | |
d4f06873 | 2382 | 0, |
652586df | 2383 | req); |
1da177e4 LT |
2384 | } |
2385 | ||
652586df | 2386 | static void get_tcp4_sock(struct sock *sk, struct seq_file *f, int i) |
1da177e4 LT |
2387 | { |
2388 | int timer_active; | |
2389 | unsigned long timer_expires; | |
cf533ea5 | 2390 | const struct tcp_sock *tp = tcp_sk(sk); |
cf4c6bf8 | 2391 | const struct inet_connection_sock *icsk = inet_csk(sk); |
cf533ea5 | 2392 | const struct inet_sock *inet = inet_sk(sk); |
0536fcc0 | 2393 | const struct fastopen_queue *fastopenq = &icsk->icsk_accept_queue.fastopenq; |
c720c7e8 ED |
2394 | __be32 dest = inet->inet_daddr; |
2395 | __be32 src = inet->inet_rcv_saddr; | |
2396 | __u16 destp = ntohs(inet->inet_dport); | |
2397 | __u16 srcp = ntohs(inet->inet_sport); | |
49d09007 | 2398 | int rx_queue; |
00fd38d9 | 2399 | int state; |
1da177e4 | 2400 | |
6ba8a3b1 | 2401 | if (icsk->icsk_pending == ICSK_TIME_RETRANS || |
57dde7f7 | 2402 | icsk->icsk_pending == ICSK_TIME_REO_TIMEOUT || |
6ba8a3b1 | 2403 | icsk->icsk_pending == ICSK_TIME_LOSS_PROBE) { |
1da177e4 | 2404 | timer_active = 1; |
463c84b9 ACM |
2405 | timer_expires = icsk->icsk_timeout; |
2406 | } else if (icsk->icsk_pending == ICSK_TIME_PROBE0) { | |
1da177e4 | 2407 | timer_active = 4; |
463c84b9 | 2408 | timer_expires = icsk->icsk_timeout; |
cf4c6bf8 | 2409 | } else if (timer_pending(&sk->sk_timer)) { |
1da177e4 | 2410 | timer_active = 2; |
cf4c6bf8 | 2411 | timer_expires = sk->sk_timer.expires; |
1da177e4 LT |
2412 | } else { |
2413 | timer_active = 0; | |
2414 | timer_expires = jiffies; | |
2415 | } | |
2416 | ||
986ffdfd | 2417 | state = inet_sk_state_load(sk); |
00fd38d9 | 2418 | if (state == TCP_LISTEN) |
49d09007 ED |
2419 | rx_queue = sk->sk_ack_backlog; |
2420 | else | |
00fd38d9 ED |
2421 | /* Because we don't lock the socket, |
2422 | * we might find a transient negative value. | |
49d09007 ED |
2423 | */ |
2424 | rx_queue = max_t(int, tp->rcv_nxt - tp->copied_seq, 0); | |
2425 | ||
5e659e4c | 2426 | seq_printf(f, "%4d: %08X:%04X %08X:%04X %02X %08X:%08X %02X:%08lX " |
652586df | 2427 | "%08X %5u %8d %lu %d %pK %lu %lu %u %u %d", |
00fd38d9 | 2428 | i, src, srcp, dest, destp, state, |
47da8ee6 | 2429 | tp->write_seq - tp->snd_una, |
49d09007 | 2430 | rx_queue, |
1da177e4 | 2431 | timer_active, |
a399a805 | 2432 | jiffies_delta_to_clock_t(timer_expires - jiffies), |
463c84b9 | 2433 | icsk->icsk_retransmits, |
a7cb5a49 | 2434 | from_kuid_munged(seq_user_ns(f), sock_i_uid(sk)), |
6687e988 | 2435 | icsk->icsk_probes_out, |
cf4c6bf8 | 2436 | sock_i_ino(sk), |
41c6d650 | 2437 | refcount_read(&sk->sk_refcnt), sk, |
7be87351 SH |
2438 | jiffies_to_clock_t(icsk->icsk_rto), |
2439 | jiffies_to_clock_t(icsk->icsk_ack.ato), | |
463c84b9 | 2440 | (icsk->icsk_ack.quick << 1) | icsk->icsk_ack.pingpong, |
1da177e4 | 2441 | tp->snd_cwnd, |
00fd38d9 ED |
2442 | state == TCP_LISTEN ? |
2443 | fastopenq->max_qlen : | |
652586df | 2444 | (tcp_in_initial_slowstart(tp) ? -1 : tp->snd_ssthresh)); |
1da177e4 LT |
2445 | } |
2446 | ||
cf533ea5 | 2447 | static void get_timewait4_sock(const struct inet_timewait_sock *tw, |
652586df | 2448 | struct seq_file *f, int i) |
1da177e4 | 2449 | { |
789f558c | 2450 | long delta = tw->tw_timer.expires - jiffies; |
23f33c2d | 2451 | __be32 dest, src; |
1da177e4 | 2452 | __u16 destp, srcp; |
1da177e4 LT |
2453 | |
2454 | dest = tw->tw_daddr; | |
2455 | src = tw->tw_rcv_saddr; | |
2456 | destp = ntohs(tw->tw_dport); | |
2457 | srcp = ntohs(tw->tw_sport); | |
2458 | ||
5e659e4c | 2459 | seq_printf(f, "%4d: %08X:%04X %08X:%04X" |
652586df | 2460 | " %02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %pK", |
1da177e4 | 2461 | i, src, srcp, dest, destp, tw->tw_substate, 0, 0, |
a399a805 | 2462 | 3, jiffies_delta_to_clock_t(delta), 0, 0, 0, 0, |
41c6d650 | 2463 | refcount_read(&tw->tw_refcnt), tw); |
1da177e4 LT |
2464 | } |
2465 | ||
2466 | #define TMPSZ 150 | |
2467 | ||
2468 | static int tcp4_seq_show(struct seq_file *seq, void *v) | |
2469 | { | |
5799de0b | 2470 | struct tcp_iter_state *st; |
05dbc7b5 | 2471 | struct sock *sk = v; |
1da177e4 | 2472 | |
652586df | 2473 | seq_setwidth(seq, TMPSZ - 1); |
1da177e4 | 2474 | if (v == SEQ_START_TOKEN) { |
652586df | 2475 | seq_puts(seq, " sl local_address rem_address st tx_queue " |
1da177e4 LT |
2476 | "rx_queue tr tm->when retrnsmt uid timeout " |
2477 | "inode"); | |
2478 | goto out; | |
2479 | } | |
2480 | st = seq->private; | |
2481 | ||
079096f1 ED |
2482 | if (sk->sk_state == TCP_TIME_WAIT) |
2483 | get_timewait4_sock(v, seq, st->num); | |
2484 | else if (sk->sk_state == TCP_NEW_SYN_RECV) | |
aa3a0c8c | 2485 | get_openreq4(v, seq, st->num); |
079096f1 ED |
2486 | else |
2487 | get_tcp4_sock(v, seq, st->num); | |
1da177e4 | 2488 | out: |
652586df | 2489 | seq_pad(seq, '\n'); |
1da177e4 LT |
2490 | return 0; |
2491 | } | |
2492 | ||
37d849bb CH |
2493 | static const struct seq_operations tcp4_seq_ops = { |
2494 | .show = tcp4_seq_show, | |
2495 | .start = tcp_seq_start, | |
2496 | .next = tcp_seq_next, | |
2497 | .stop = tcp_seq_stop, | |
2498 | }; | |
2499 | ||
1da177e4 | 2500 | static struct tcp_seq_afinfo tcp4_seq_afinfo = { |
1da177e4 | 2501 | .family = AF_INET, |
1da177e4 LT |
2502 | }; |
2503 | ||
2c8c1e72 | 2504 | static int __net_init tcp4_proc_init_net(struct net *net) |
757764f6 | 2505 | { |
c3506372 CH |
2506 | if (!proc_create_net_data("tcp", 0444, net->proc_net, &tcp4_seq_ops, |
2507 | sizeof(struct tcp_iter_state), &tcp4_seq_afinfo)) | |
37d849bb CH |
2508 | return -ENOMEM; |
2509 | return 0; | |
757764f6 PE |
2510 | } |
2511 | ||
2c8c1e72 | 2512 | static void __net_exit tcp4_proc_exit_net(struct net *net) |
757764f6 | 2513 | { |
37d849bb | 2514 | remove_proc_entry("tcp", net->proc_net); |
757764f6 PE |
2515 | } |
2516 | ||
2517 | static struct pernet_operations tcp4_net_ops = { | |
2518 | .init = tcp4_proc_init_net, | |
2519 | .exit = tcp4_proc_exit_net, | |
2520 | }; | |
2521 | ||
1da177e4 LT |
2522 | int __init tcp4_proc_init(void) |
2523 | { | |
757764f6 | 2524 | return register_pernet_subsys(&tcp4_net_ops); |
1da177e4 LT |
2525 | } |
2526 | ||
2527 | void tcp4_proc_exit(void) | |
2528 | { | |
757764f6 | 2529 | unregister_pernet_subsys(&tcp4_net_ops); |
1da177e4 LT |
2530 | } |
2531 | #endif /* CONFIG_PROC_FS */ | |
2532 | ||
2533 | struct proto tcp_prot = { | |
2534 | .name = "TCP", | |
2535 | .owner = THIS_MODULE, | |
2536 | .close = tcp_close, | |
d74bad4e | 2537 | .pre_connect = tcp_v4_pre_connect, |
1da177e4 LT |
2538 | .connect = tcp_v4_connect, |
2539 | .disconnect = tcp_disconnect, | |
463c84b9 | 2540 | .accept = inet_csk_accept, |
1da177e4 LT |
2541 | .ioctl = tcp_ioctl, |
2542 | .init = tcp_v4_init_sock, | |
2543 | .destroy = tcp_v4_destroy_sock, | |
2544 | .shutdown = tcp_shutdown, | |
2545 | .setsockopt = tcp_setsockopt, | |
2546 | .getsockopt = tcp_getsockopt, | |
4b9d07a4 | 2547 | .keepalive = tcp_set_keepalive, |
1da177e4 | 2548 | .recvmsg = tcp_recvmsg, |
7ba42910 CG |
2549 | .sendmsg = tcp_sendmsg, |
2550 | .sendpage = tcp_sendpage, | |
1da177e4 | 2551 | .backlog_rcv = tcp_v4_do_rcv, |
46d3ceab | 2552 | .release_cb = tcp_release_cb, |
ab1e0a13 ACM |
2553 | .hash = inet_hash, |
2554 | .unhash = inet_unhash, | |
2555 | .get_port = inet_csk_get_port, | |
1da177e4 | 2556 | .enter_memory_pressure = tcp_enter_memory_pressure, |
06044751 | 2557 | .leave_memory_pressure = tcp_leave_memory_pressure, |
c9bee3b7 | 2558 | .stream_memory_free = tcp_stream_memory_free, |
1da177e4 | 2559 | .sockets_allocated = &tcp_sockets_allocated, |
0a5578cf | 2560 | .orphan_count = &tcp_orphan_count, |
1da177e4 LT |
2561 | .memory_allocated = &tcp_memory_allocated, |
2562 | .memory_pressure = &tcp_memory_pressure, | |
a4fe34bf | 2563 | .sysctl_mem = sysctl_tcp_mem, |
356d1833 ED |
2564 | .sysctl_wmem_offset = offsetof(struct net, ipv4.sysctl_tcp_wmem), |
2565 | .sysctl_rmem_offset = offsetof(struct net, ipv4.sysctl_tcp_rmem), | |
1da177e4 LT |
2566 | .max_header = MAX_TCP_HEADER, |
2567 | .obj_size = sizeof(struct tcp_sock), | |
5f0d5a3a | 2568 | .slab_flags = SLAB_TYPESAFE_BY_RCU, |
6d6ee43e | 2569 | .twsk_prot = &tcp_timewait_sock_ops, |
60236fdd | 2570 | .rsk_prot = &tcp_request_sock_ops, |
39d8cda7 | 2571 | .h.hashinfo = &tcp_hashinfo, |
7ba42910 | 2572 | .no_autobind = true, |
543d9cfe ACM |
2573 | #ifdef CONFIG_COMPAT |
2574 | .compat_setsockopt = compat_tcp_setsockopt, | |
2575 | .compat_getsockopt = compat_tcp_getsockopt, | |
d1a4c0b3 | 2576 | #endif |
c1e64e29 | 2577 | .diag_destroy = tcp_abort, |
1da177e4 | 2578 | }; |
4bc2f18b | 2579 | EXPORT_SYMBOL(tcp_prot); |
1da177e4 | 2580 | |
bdbbb852 ED |
2581 | static void __net_exit tcp_sk_exit(struct net *net) |
2582 | { | |
2583 | int cpu; | |
2584 | ||
6670e152 SH |
2585 | module_put(net->ipv4.tcp_congestion_control->owner); |
2586 | ||
bdbbb852 ED |
2587 | for_each_possible_cpu(cpu) |
2588 | inet_ctl_sock_destroy(*per_cpu_ptr(net->ipv4.tcp_sk, cpu)); | |
2589 | free_percpu(net->ipv4.tcp_sk); | |
2590 | } | |
2591 | ||
046ee902 DL |
2592 | static int __net_init tcp_sk_init(struct net *net) |
2593 | { | |
fee83d09 | 2594 | int res, cpu, cnt; |
bdbbb852 ED |
2595 | |
2596 | net->ipv4.tcp_sk = alloc_percpu(struct sock *); | |
2597 | if (!net->ipv4.tcp_sk) | |
2598 | return -ENOMEM; | |
2599 | ||
2600 | for_each_possible_cpu(cpu) { | |
2601 | struct sock *sk; | |
2602 | ||
2603 | res = inet_ctl_sock_create(&sk, PF_INET, SOCK_RAW, | |
2604 | IPPROTO_TCP, net); | |
2605 | if (res) | |
2606 | goto fail; | |
a9d6532b | 2607 | sock_set_flag(sk, SOCK_USE_WRITE_QUEUE); |
431280ee ED |
2608 | |
2609 | /* Please enforce IP_DF and IPID==0 for RST and | |
2610 | * ACK sent in SYN-RECV and TIME-WAIT state. | |
2611 | */ | |
2612 | inet_sk(sk)->pmtudisc = IP_PMTUDISC_DO; | |
2613 | ||
bdbbb852 ED |
2614 | *per_cpu_ptr(net->ipv4.tcp_sk, cpu) = sk; |
2615 | } | |
49213555 | 2616 | |
5d134f1c | 2617 | net->ipv4.sysctl_tcp_ecn = 2; |
49213555 DB |
2618 | net->ipv4.sysctl_tcp_ecn_fallback = 1; |
2619 | ||
b0f9ca53 | 2620 | net->ipv4.sysctl_tcp_base_mss = TCP_BASE_MSS; |
6b58e0a5 | 2621 | net->ipv4.sysctl_tcp_probe_threshold = TCP_PROBE_THRESHOLD; |
05cbc0db | 2622 | net->ipv4.sysctl_tcp_probe_interval = TCP_PROBE_INTERVAL; |
046ee902 | 2623 | |
13b287e8 | 2624 | net->ipv4.sysctl_tcp_keepalive_time = TCP_KEEPALIVE_TIME; |
9bd6861b | 2625 | net->ipv4.sysctl_tcp_keepalive_probes = TCP_KEEPALIVE_PROBES; |
b840d15d | 2626 | net->ipv4.sysctl_tcp_keepalive_intvl = TCP_KEEPALIVE_INTVL; |
13b287e8 | 2627 | |
6fa25166 | 2628 | net->ipv4.sysctl_tcp_syn_retries = TCP_SYN_RETRIES; |
7c083ecb | 2629 | net->ipv4.sysctl_tcp_synack_retries = TCP_SYNACK_RETRIES; |
0aca737d | 2630 | net->ipv4.sysctl_tcp_syncookies = 1; |
1043e25f | 2631 | net->ipv4.sysctl_tcp_reordering = TCP_FASTRETRANS_THRESH; |
ae5c3f40 | 2632 | net->ipv4.sysctl_tcp_retries1 = TCP_RETR1; |
c6214a97 | 2633 | net->ipv4.sysctl_tcp_retries2 = TCP_RETR2; |
c402d9be | 2634 | net->ipv4.sysctl_tcp_orphan_retries = 0; |
1e579caa | 2635 | net->ipv4.sysctl_tcp_fin_timeout = TCP_FIN_TIMEOUT; |
4979f2d9 | 2636 | net->ipv4.sysctl_tcp_notsent_lowat = UINT_MAX; |
79e9fed4 | 2637 | net->ipv4.sysctl_tcp_tw_reuse = 2; |
12ed8244 | 2638 | |
fee83d09 | 2639 | cnt = tcp_hashinfo.ehash_mask + 1; |
743e4815 | 2640 | net->ipv4.tcp_death_row.sysctl_max_tw_buckets = cnt / 2; |
1946e672 HY |
2641 | net->ipv4.tcp_death_row.hashinfo = &tcp_hashinfo; |
2642 | ||
fee83d09 | 2643 | net->ipv4.sysctl_max_syn_backlog = max(128, cnt / 256); |
f9301034 | 2644 | net->ipv4.sysctl_tcp_sack = 1; |
9bb37ef0 | 2645 | net->ipv4.sysctl_tcp_window_scaling = 1; |
5d2ed052 | 2646 | net->ipv4.sysctl_tcp_timestamps = 1; |
2ae21cf5 | 2647 | net->ipv4.sysctl_tcp_early_retrans = 3; |
e20223f1 | 2648 | net->ipv4.sysctl_tcp_recovery = TCP_RACK_LOSS_DETECTION; |
b510f0d2 | 2649 | net->ipv4.sysctl_tcp_slow_start_after_idle = 1; /* By default, RFC2861 behavior. */ |
e0a1e5b5 | 2650 | net->ipv4.sysctl_tcp_retrans_collapse = 1; |
c6e21803 | 2651 | net->ipv4.sysctl_tcp_max_reordering = 300; |
6496f6bd | 2652 | net->ipv4.sysctl_tcp_dsack = 1; |
0c12654a | 2653 | net->ipv4.sysctl_tcp_app_win = 31; |
94f0893e | 2654 | net->ipv4.sysctl_tcp_adv_win_scale = 1; |
af9b69a7 | 2655 | net->ipv4.sysctl_tcp_frto = 2; |
4540c0cf | 2656 | net->ipv4.sysctl_tcp_moderate_rcvbuf = 1; |
d06a9904 ED |
2657 | /* This limits the percentage of the congestion window which we |
2658 | * will allow a single TSO frame to consume. Building TSO frames | |
2659 | * which are too large can cause TCP streams to be bursty. | |
2660 | */ | |
2661 | net->ipv4.sysctl_tcp_tso_win_divisor = 3; | |
c73e5807 ED |
2662 | /* Default TSQ limit of 16 TSO segments */ |
2663 | net->ipv4.sysctl_tcp_limit_output_bytes = 16 * 65536; | |
b530b681 ED |
2664 | /* rfc5961 challenge ack rate limiting */ |
2665 | net->ipv4.sysctl_tcp_challenge_ack_limit = 1000; | |
26e9596e | 2666 | net->ipv4.sysctl_tcp_min_tso_segs = 2; |
bd239704 | 2667 | net->ipv4.sysctl_tcp_min_rtt_wlen = 300; |
790f00e1 | 2668 | net->ipv4.sysctl_tcp_autocorking = 1; |
4170ba6b | 2669 | net->ipv4.sysctl_tcp_invalid_ratelimit = HZ/2; |
23a7102a | 2670 | net->ipv4.sysctl_tcp_pacing_ss_ratio = 200; |
c26e91f8 | 2671 | net->ipv4.sysctl_tcp_pacing_ca_ratio = 120; |
356d1833 ED |
2672 | if (net != &init_net) { |
2673 | memcpy(net->ipv4.sysctl_tcp_rmem, | |
2674 | init_net.ipv4.sysctl_tcp_rmem, | |
2675 | sizeof(init_net.ipv4.sysctl_tcp_rmem)); | |
2676 | memcpy(net->ipv4.sysctl_tcp_wmem, | |
2677 | init_net.ipv4.sysctl_tcp_wmem, | |
2678 | sizeof(init_net.ipv4.sysctl_tcp_wmem)); | |
2679 | } | |
6d82aa24 | 2680 | net->ipv4.sysctl_tcp_comp_sack_delay_ns = NSEC_PER_MSEC; |
9c21d2fc | 2681 | net->ipv4.sysctl_tcp_comp_sack_nr = 44; |
e1cfcbe8 | 2682 | net->ipv4.sysctl_tcp_fastopen = TFO_CLIENT_ENABLE; |
43713848 | 2683 | spin_lock_init(&net->ipv4.tcp_fastopen_ctx_lock); |
3733be14 HY |
2684 | net->ipv4.sysctl_tcp_fastopen_blackhole_timeout = 60 * 60; |
2685 | atomic_set(&net->ipv4.tfo_active_disable_times, 0); | |
e1cfcbe8 | 2686 | |
6670e152 SH |
2687 | /* Reno is always built in */ |
2688 | if (!net_eq(net, &init_net) && | |
2689 | try_module_get(init_net.ipv4.tcp_congestion_control->owner)) | |
2690 | net->ipv4.tcp_congestion_control = init_net.ipv4.tcp_congestion_control; | |
2691 | else | |
2692 | net->ipv4.tcp_congestion_control = &tcp_reno; | |
2693 | ||
49213555 | 2694 | return 0; |
bdbbb852 ED |
2695 | fail: |
2696 | tcp_sk_exit(net); | |
2697 | ||
2698 | return res; | |
b099ce26 EB |
2699 | } |
2700 | ||
2701 | static void __net_exit tcp_sk_exit_batch(struct list_head *net_exit_list) | |
2702 | { | |
43713848 HY |
2703 | struct net *net; |
2704 | ||
1946e672 | 2705 | inet_twsk_purge(&tcp_hashinfo, AF_INET); |
43713848 HY |
2706 | |
2707 | list_for_each_entry(net, net_exit_list, exit_list) | |
2708 | tcp_fastopen_ctx_destroy(net); | |
046ee902 DL |
2709 | } |
2710 | ||
2711 | static struct pernet_operations __net_initdata tcp_sk_ops = { | |
b099ce26 EB |
2712 | .init = tcp_sk_init, |
2713 | .exit = tcp_sk_exit, | |
2714 | .exit_batch = tcp_sk_exit_batch, | |
046ee902 DL |
2715 | }; |
2716 | ||
9b0f976f | 2717 | void __init tcp_v4_init(void) |
1da177e4 | 2718 | { |
6a1b3054 | 2719 | if (register_pernet_subsys(&tcp_sk_ops)) |
1da177e4 | 2720 | panic("Failed to create the TCP control socket.\n"); |
1da177e4 | 2721 | } |