]>
Commit | Line | Data |
---|---|---|
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 | * | |
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 | |
37 | * request_sock handling and moved | |
38 | * most of it into the af independent code. | |
39 | * Added tail drop and some other bugfixes. | |
40 | * Added new listen semantics. | |
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 | ||
53 | #define pr_fmt(fmt) "TCP: " fmt | |
54 | ||
55 | #include <linux/bottom_half.h> | |
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> | |
64 | #include <linux/slab.h> | |
65 | ||
66 | #include <net/net_namespace.h> | |
67 | #include <net/icmp.h> | |
68 | #include <net/inet_hashtables.h> | |
69 | #include <net/tcp.h> | |
70 | #include <net/transp_v6.h> | |
71 | #include <net/ipv6.h> | |
72 | #include <net/inet_common.h> | |
73 | #include <net/timewait_sock.h> | |
74 | #include <net/xfrm.h> | |
75 | #include <net/netdma.h> | |
76 | #include <net/secure_seq.h> | |
77 | #include <net/tcp_memcontrol.h> | |
78 | ||
79 | #include <linux/inet.h> | |
80 | #include <linux/ipv6.h> | |
81 | #include <linux/stddef.h> | |
82 | #include <linux/proc_fs.h> | |
83 | #include <linux/seq_file.h> | |
84 | ||
85 | #include <linux/crypto.h> | |
86 | #include <linux/scatterlist.h> | |
87 | ||
88 | int sysctl_tcp_tw_reuse __read_mostly; | |
89 | int sysctl_tcp_low_latency __read_mostly; | |
90 | EXPORT_SYMBOL(sysctl_tcp_low_latency); | |
91 | ||
92 | ||
93 | #ifdef CONFIG_TCP_MD5SIG | |
94 | static int tcp_v4_md5_hash_hdr(char *md5_hash, const struct tcp_md5sig_key *key, | |
95 | __be32 daddr, __be32 saddr, const struct tcphdr *th); | |
96 | #endif | |
97 | ||
98 | struct inet_hashinfo tcp_hashinfo; | |
99 | EXPORT_SYMBOL(tcp_hashinfo); | |
100 | ||
101 | static inline __u32 tcp_v4_init_sequence(const struct sk_buff *skb) | |
102 | { | |
103 | return secure_tcp_sequence_number(ip_hdr(skb)->daddr, | |
104 | ip_hdr(skb)->saddr, | |
105 | tcp_hdr(skb)->dest, | |
106 | tcp_hdr(skb)->source); | |
107 | } | |
108 | ||
109 | int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp) | |
110 | { | |
111 | const struct tcp_timewait_sock *tcptw = tcp_twsk(sktw); | |
112 | struct tcp_sock *tp = tcp_sk(sk); | |
113 | ||
114 | /* With PAWS, it is safe from the viewpoint | |
115 | of data integrity. Even without PAWS it is safe provided sequence | |
116 | spaces do not overlap i.e. at data rates <= 80Mbit/sec. | |
117 | ||
118 | Actually, the idea is close to VJ's one, only timestamp cache is | |
119 | held not per host, but per port pair and TW bucket is used as state | |
120 | holder. | |
121 | ||
122 | If TW bucket has been already destroyed we fall back to VJ's scheme | |
123 | and use initial timestamp retrieved from peer table. | |
124 | */ | |
125 | if (tcptw->tw_ts_recent_stamp && | |
126 | (twp == NULL || (sysctl_tcp_tw_reuse && | |
127 | get_seconds() - tcptw->tw_ts_recent_stamp > 1))) { | |
128 | tp->write_seq = tcptw->tw_snd_nxt + 65535 + 2; | |
129 | if (tp->write_seq == 0) | |
130 | tp->write_seq = 1; | |
131 | tp->rx_opt.ts_recent = tcptw->tw_ts_recent; | |
132 | tp->rx_opt.ts_recent_stamp = tcptw->tw_ts_recent_stamp; | |
133 | sock_hold(sktw); | |
134 | return 1; | |
135 | } | |
136 | ||
137 | return 0; | |
138 | } | |
139 | EXPORT_SYMBOL_GPL(tcp_twsk_unique); | |
140 | ||
141 | /* This will initiate an outgoing connection. */ | |
142 | int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len) | |
143 | { | |
144 | struct sockaddr_in *usin = (struct sockaddr_in *)uaddr; | |
145 | struct inet_sock *inet = inet_sk(sk); | |
146 | struct tcp_sock *tp = tcp_sk(sk); | |
147 | __be16 orig_sport, orig_dport; | |
148 | __be32 daddr, nexthop; | |
149 | struct flowi4 *fl4; | |
150 | struct rtable *rt; | |
151 | int err; | |
152 | struct ip_options_rcu *inet_opt; | |
153 | ||
154 | if (addr_len < sizeof(struct sockaddr_in)) | |
155 | return -EINVAL; | |
156 | ||
157 | if (usin->sin_family != AF_INET) | |
158 | return -EAFNOSUPPORT; | |
159 | ||
160 | nexthop = daddr = usin->sin_addr.s_addr; | |
161 | inet_opt = rcu_dereference_protected(inet->inet_opt, | |
162 | sock_owned_by_user(sk)); | |
163 | if (inet_opt && inet_opt->opt.srr) { | |
164 | if (!daddr) | |
165 | return -EINVAL; | |
166 | nexthop = inet_opt->opt.faddr; | |
167 | } | |
168 | ||
169 | orig_sport = inet->inet_sport; | |
170 | orig_dport = usin->sin_port; | |
171 | fl4 = &inet->cork.fl.u.ip4; | |
172 | rt = ip_route_connect(fl4, nexthop, inet->inet_saddr, | |
173 | RT_CONN_FLAGS(sk), sk->sk_bound_dev_if, | |
174 | IPPROTO_TCP, | |
175 | orig_sport, orig_dport, sk, true); | |
176 | if (IS_ERR(rt)) { | |
177 | err = PTR_ERR(rt); | |
178 | if (err == -ENETUNREACH) | |
179 | IP_INC_STATS_BH(sock_net(sk), IPSTATS_MIB_OUTNOROUTES); | |
180 | return err; | |
181 | } | |
182 | ||
183 | if (rt->rt_flags & (RTCF_MULTICAST | RTCF_BROADCAST)) { | |
184 | ip_rt_put(rt); | |
185 | return -ENETUNREACH; | |
186 | } | |
187 | ||
188 | if (!inet_opt || !inet_opt->opt.srr) | |
189 | daddr = fl4->daddr; | |
190 | ||
191 | if (!inet->inet_saddr) | |
192 | inet->inet_saddr = fl4->saddr; | |
193 | inet->inet_rcv_saddr = inet->inet_saddr; | |
194 | ||
195 | if (tp->rx_opt.ts_recent_stamp && inet->inet_daddr != daddr) { | |
196 | /* Reset inherited state */ | |
197 | tp->rx_opt.ts_recent = 0; | |
198 | tp->rx_opt.ts_recent_stamp = 0; | |
199 | if (likely(!tp->repair)) | |
200 | tp->write_seq = 0; | |
201 | } | |
202 | ||
203 | if (tcp_death_row.sysctl_tw_recycle && | |
204 | !tp->rx_opt.ts_recent_stamp && fl4->daddr == daddr) | |
205 | tcp_fetch_timewait_stamp(sk, &rt->dst); | |
206 | ||
207 | inet->inet_dport = usin->sin_port; | |
208 | inet->inet_daddr = daddr; | |
209 | ||
210 | inet_csk(sk)->icsk_ext_hdr_len = 0; | |
211 | if (inet_opt) | |
212 | inet_csk(sk)->icsk_ext_hdr_len = inet_opt->opt.optlen; | |
213 | ||
214 | tp->rx_opt.mss_clamp = TCP_MSS_DEFAULT; | |
215 | ||
216 | /* Socket identity is still unknown (sport may be zero). | |
217 | * However we set state to SYN-SENT and not releasing socket | |
218 | * lock select source port, enter ourselves into the hash tables and | |
219 | * complete initialization after this. | |
220 | */ | |
221 | tcp_set_state(sk, TCP_SYN_SENT); | |
222 | err = inet_hash_connect(&tcp_death_row, sk); | |
223 | if (err) | |
224 | goto failure; | |
225 | ||
226 | rt = ip_route_newports(fl4, rt, orig_sport, orig_dport, | |
227 | inet->inet_sport, inet->inet_dport, sk); | |
228 | if (IS_ERR(rt)) { | |
229 | err = PTR_ERR(rt); | |
230 | rt = NULL; | |
231 | goto failure; | |
232 | } | |
233 | /* OK, now commit destination to socket. */ | |
234 | sk->sk_gso_type = SKB_GSO_TCPV4; | |
235 | sk_setup_caps(sk, &rt->dst); | |
236 | ||
237 | if (!tp->write_seq && likely(!tp->repair)) | |
238 | tp->write_seq = secure_tcp_sequence_number(inet->inet_saddr, | |
239 | inet->inet_daddr, | |
240 | inet->inet_sport, | |
241 | usin->sin_port); | |
242 | ||
243 | inet->inet_id = tp->write_seq ^ jiffies; | |
244 | ||
245 | err = tcp_connect(sk); | |
246 | ||
247 | rt = NULL; | |
248 | if (err) | |
249 | goto failure; | |
250 | ||
251 | return 0; | |
252 | ||
253 | failure: | |
254 | /* | |
255 | * This unhashes the socket and releases the local port, | |
256 | * if necessary. | |
257 | */ | |
258 | tcp_set_state(sk, TCP_CLOSE); | |
259 | ip_rt_put(rt); | |
260 | sk->sk_route_caps = 0; | |
261 | inet->inet_dport = 0; | |
262 | return err; | |
263 | } | |
264 | EXPORT_SYMBOL(tcp_v4_connect); | |
265 | ||
266 | /* | |
267 | * This routine reacts to ICMP_FRAG_NEEDED mtu indications as defined in RFC1191. | |
268 | * It can be called through tcp_release_cb() if socket was owned by user | |
269 | * at the time tcp_v4_err() was called to handle ICMP message. | |
270 | */ | |
271 | static void tcp_v4_mtu_reduced(struct sock *sk) | |
272 | { | |
273 | struct dst_entry *dst; | |
274 | struct inet_sock *inet = inet_sk(sk); | |
275 | u32 mtu = tcp_sk(sk)->mtu_info; | |
276 | ||
277 | /* We are not interested in TCP_LISTEN and open_requests (SYN-ACKs | |
278 | * send out by Linux are always <576bytes so they should go through | |
279 | * unfragmented). | |
280 | */ | |
281 | if (sk->sk_state == TCP_LISTEN) | |
282 | return; | |
283 | ||
284 | dst = inet_csk_update_pmtu(sk, mtu); | |
285 | if (!dst) | |
286 | return; | |
287 | ||
288 | /* Something is about to be wrong... Remember soft error | |
289 | * for the case, if this connection will not able to recover. | |
290 | */ | |
291 | if (mtu < dst_mtu(dst) && ip_dont_fragment(sk, dst)) | |
292 | sk->sk_err_soft = EMSGSIZE; | |
293 | ||
294 | mtu = dst_mtu(dst); | |
295 | ||
296 | if (inet->pmtudisc != IP_PMTUDISC_DONT && | |
297 | inet_csk(sk)->icsk_pmtu_cookie > mtu) { | |
298 | tcp_sync_mss(sk, mtu); | |
299 | ||
300 | /* Resend the TCP packet because it's | |
301 | * clear that the old packet has been | |
302 | * dropped. This is the new "fast" path mtu | |
303 | * discovery. | |
304 | */ | |
305 | tcp_simple_retransmit(sk); | |
306 | } /* else let the usual retransmit timer handle it */ | |
307 | } | |
308 | ||
309 | static void do_redirect(struct sk_buff *skb, struct sock *sk) | |
310 | { | |
311 | struct dst_entry *dst = __sk_dst_check(sk, 0); | |
312 | ||
313 | if (dst) | |
314 | dst->ops->redirect(dst, sk, skb); | |
315 | } | |
316 | ||
317 | /* | |
318 | * This routine is called by the ICMP module when it gets some | |
319 | * sort of error condition. If err < 0 then the socket should | |
320 | * be closed and the error returned to the user. If err > 0 | |
321 | * it's just the icmp type << 8 | icmp code. After adjustment | |
322 | * header points to the first 8 bytes of the tcp header. We need | |
323 | * to find the appropriate port. | |
324 | * | |
325 | * The locking strategy used here is very "optimistic". When | |
326 | * someone else accesses the socket the ICMP is just dropped | |
327 | * and for some paths there is no check at all. | |
328 | * A more general error queue to queue errors for later handling | |
329 | * is probably better. | |
330 | * | |
331 | */ | |
332 | ||
333 | void tcp_v4_err(struct sk_buff *icmp_skb, u32 info) | |
334 | { | |
335 | const struct iphdr *iph = (const struct iphdr *)icmp_skb->data; | |
336 | struct tcphdr *th = (struct tcphdr *)(icmp_skb->data + (iph->ihl << 2)); | |
337 | struct inet_connection_sock *icsk; | |
338 | struct tcp_sock *tp; | |
339 | struct inet_sock *inet; | |
340 | const int type = icmp_hdr(icmp_skb)->type; | |
341 | const int code = icmp_hdr(icmp_skb)->code; | |
342 | struct sock *sk; | |
343 | struct sk_buff *skb; | |
344 | struct request_sock *req; | |
345 | __u32 seq; | |
346 | __u32 remaining; | |
347 | int err; | |
348 | struct net *net = dev_net(icmp_skb->dev); | |
349 | ||
350 | if (icmp_skb->len < (iph->ihl << 2) + 8) { | |
351 | ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS); | |
352 | return; | |
353 | } | |
354 | ||
355 | sk = inet_lookup(net, &tcp_hashinfo, iph->daddr, th->dest, | |
356 | iph->saddr, th->source, inet_iif(icmp_skb)); | |
357 | if (!sk) { | |
358 | ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS); | |
359 | return; | |
360 | } | |
361 | if (sk->sk_state == TCP_TIME_WAIT) { | |
362 | inet_twsk_put(inet_twsk(sk)); | |
363 | return; | |
364 | } | |
365 | ||
366 | bh_lock_sock(sk); | |
367 | /* If too many ICMPs get dropped on busy | |
368 | * servers this needs to be solved differently. | |
369 | * We do take care of PMTU discovery (RFC1191) special case : | |
370 | * we can receive locally generated ICMP messages while socket is held. | |
371 | */ | |
372 | if (sock_owned_by_user(sk)) { | |
373 | if (!(type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED)) | |
374 | NET_INC_STATS_BH(net, LINUX_MIB_LOCKDROPPEDICMPS); | |
375 | } | |
376 | if (sk->sk_state == TCP_CLOSE) | |
377 | goto out; | |
378 | ||
379 | if (unlikely(iph->ttl < inet_sk(sk)->min_ttl)) { | |
380 | NET_INC_STATS_BH(net, LINUX_MIB_TCPMINTTLDROP); | |
381 | goto out; | |
382 | } | |
383 | ||
384 | icsk = inet_csk(sk); | |
385 | tp = tcp_sk(sk); | |
386 | req = tp->fastopen_rsk; | |
387 | seq = ntohl(th->seq); | |
388 | if (sk->sk_state != TCP_LISTEN && | |
389 | !between(seq, tp->snd_una, tp->snd_nxt) && | |
390 | (req == NULL || seq != tcp_rsk(req)->snt_isn)) { | |
391 | /* For a Fast Open socket, allow seq to be snt_isn. */ | |
392 | NET_INC_STATS_BH(net, LINUX_MIB_OUTOFWINDOWICMPS); | |
393 | goto out; | |
394 | } | |
395 | ||
396 | switch (type) { | |
397 | case ICMP_REDIRECT: | |
398 | do_redirect(icmp_skb, sk); | |
399 | goto out; | |
400 | case ICMP_SOURCE_QUENCH: | |
401 | /* Just silently ignore these. */ | |
402 | goto out; | |
403 | case ICMP_PARAMETERPROB: | |
404 | err = EPROTO; | |
405 | break; | |
406 | case ICMP_DEST_UNREACH: | |
407 | if (code > NR_ICMP_UNREACH) | |
408 | goto out; | |
409 | ||
410 | if (code == ICMP_FRAG_NEEDED) { /* PMTU discovery (RFC1191) */ | |
411 | tp->mtu_info = info; | |
412 | if (!sock_owned_by_user(sk)) { | |
413 | tcp_v4_mtu_reduced(sk); | |
414 | } else { | |
415 | if (!test_and_set_bit(TCP_MTU_REDUCED_DEFERRED, &tp->tsq_flags)) | |
416 | sock_hold(sk); | |
417 | } | |
418 | goto out; | |
419 | } | |
420 | ||
421 | err = icmp_err_convert[code].errno; | |
422 | /* check if icmp_skb allows revert of backoff | |
423 | * (see draft-zimmermann-tcp-lcd) */ | |
424 | if (code != ICMP_NET_UNREACH && code != ICMP_HOST_UNREACH) | |
425 | break; | |
426 | if (seq != tp->snd_una || !icsk->icsk_retransmits || | |
427 | !icsk->icsk_backoff) | |
428 | break; | |
429 | ||
430 | /* XXX (TFO) - revisit the following logic for TFO */ | |
431 | ||
432 | if (sock_owned_by_user(sk)) | |
433 | break; | |
434 | ||
435 | icsk->icsk_backoff--; | |
436 | inet_csk(sk)->icsk_rto = (tp->srtt ? __tcp_set_rto(tp) : | |
437 | TCP_TIMEOUT_INIT) << icsk->icsk_backoff; | |
438 | tcp_bound_rto(sk); | |
439 | ||
440 | skb = tcp_write_queue_head(sk); | |
441 | BUG_ON(!skb); | |
442 | ||
443 | remaining = icsk->icsk_rto - min(icsk->icsk_rto, | |
444 | tcp_time_stamp - TCP_SKB_CB(skb)->when); | |
445 | ||
446 | if (remaining) { | |
447 | inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS, | |
448 | remaining, TCP_RTO_MAX); | |
449 | } else { | |
450 | /* RTO revert clocked out retransmission. | |
451 | * Will retransmit now */ | |
452 | tcp_retransmit_timer(sk); | |
453 | } | |
454 | ||
455 | break; | |
456 | case ICMP_TIME_EXCEEDED: | |
457 | err = EHOSTUNREACH; | |
458 | break; | |
459 | default: | |
460 | goto out; | |
461 | } | |
462 | ||
463 | /* XXX (TFO) - if it's a TFO socket and has been accepted, rather | |
464 | * than following the TCP_SYN_RECV case and closing the socket, | |
465 | * we ignore the ICMP error and keep trying like a fully established | |
466 | * socket. Is this the right thing to do? | |
467 | */ | |
468 | if (req && req->sk == NULL) | |
469 | goto out; | |
470 | ||
471 | switch (sk->sk_state) { | |
472 | struct request_sock *req, **prev; | |
473 | case TCP_LISTEN: | |
474 | if (sock_owned_by_user(sk)) | |
475 | goto out; | |
476 | ||
477 | req = inet_csk_search_req(sk, &prev, th->dest, | |
478 | iph->daddr, iph->saddr); | |
479 | if (!req) | |
480 | goto out; | |
481 | ||
482 | /* ICMPs are not backlogged, hence we cannot get | |
483 | an established socket here. | |
484 | */ | |
485 | WARN_ON(req->sk); | |
486 | ||
487 | if (seq != tcp_rsk(req)->snt_isn) { | |
488 | NET_INC_STATS_BH(net, LINUX_MIB_OUTOFWINDOWICMPS); | |
489 | goto out; | |
490 | } | |
491 | ||
492 | /* | |
493 | * Still in SYN_RECV, just remove it silently. | |
494 | * There is no good way to pass the error to the newly | |
495 | * created socket, and POSIX does not want network | |
496 | * errors returned from accept(). | |
497 | */ | |
498 | inet_csk_reqsk_queue_drop(sk, req, prev); | |
499 | NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENDROPS); | |
500 | goto out; | |
501 | ||
502 | case TCP_SYN_SENT: | |
503 | case TCP_SYN_RECV: /* Cannot happen. | |
504 | It can f.e. if SYNs crossed, | |
505 | or Fast Open. | |
506 | */ | |
507 | if (!sock_owned_by_user(sk)) { | |
508 | sk->sk_err = err; | |
509 | ||
510 | sk->sk_error_report(sk); | |
511 | ||
512 | tcp_done(sk); | |
513 | } else { | |
514 | sk->sk_err_soft = err; | |
515 | } | |
516 | goto out; | |
517 | } | |
518 | ||
519 | /* If we've already connected we will keep trying | |
520 | * until we time out, or the user gives up. | |
521 | * | |
522 | * rfc1122 4.2.3.9 allows to consider as hard errors | |
523 | * only PROTO_UNREACH and PORT_UNREACH (well, FRAG_FAILED too, | |
524 | * but it is obsoleted by pmtu discovery). | |
525 | * | |
526 | * Note, that in modern internet, where routing is unreliable | |
527 | * and in each dark corner broken firewalls sit, sending random | |
528 | * errors ordered by their masters even this two messages finally lose | |
529 | * their original sense (even Linux sends invalid PORT_UNREACHs) | |
530 | * | |
531 | * Now we are in compliance with RFCs. | |
532 | * --ANK (980905) | |
533 | */ | |
534 | ||
535 | inet = inet_sk(sk); | |
536 | if (!sock_owned_by_user(sk) && inet->recverr) { | |
537 | sk->sk_err = err; | |
538 | sk->sk_error_report(sk); | |
539 | } else { /* Only an error on timeout */ | |
540 | sk->sk_err_soft = err; | |
541 | } | |
542 | ||
543 | out: | |
544 | bh_unlock_sock(sk); | |
545 | sock_put(sk); | |
546 | } | |
547 | ||
548 | static void __tcp_v4_send_check(struct sk_buff *skb, | |
549 | __be32 saddr, __be32 daddr) | |
550 | { | |
551 | struct tcphdr *th = tcp_hdr(skb); | |
552 | ||
553 | if (skb->ip_summed == CHECKSUM_PARTIAL) { | |
554 | th->check = ~tcp_v4_check(skb->len, saddr, daddr, 0); | |
555 | skb->csum_start = skb_transport_header(skb) - skb->head; | |
556 | skb->csum_offset = offsetof(struct tcphdr, check); | |
557 | } else { | |
558 | th->check = tcp_v4_check(skb->len, saddr, daddr, | |
559 | csum_partial(th, | |
560 | th->doff << 2, | |
561 | skb->csum)); | |
562 | } | |
563 | } | |
564 | ||
565 | /* This routine computes an IPv4 TCP checksum. */ | |
566 | void tcp_v4_send_check(struct sock *sk, struct sk_buff *skb) | |
567 | { | |
568 | const struct inet_sock *inet = inet_sk(sk); | |
569 | ||
570 | __tcp_v4_send_check(skb, inet->inet_saddr, inet->inet_daddr); | |
571 | } | |
572 | EXPORT_SYMBOL(tcp_v4_send_check); | |
573 | ||
574 | int tcp_v4_gso_send_check(struct sk_buff *skb) | |
575 | { | |
576 | const struct iphdr *iph; | |
577 | struct tcphdr *th; | |
578 | ||
579 | if (!pskb_may_pull(skb, sizeof(*th))) | |
580 | return -EINVAL; | |
581 | ||
582 | iph = ip_hdr(skb); | |
583 | th = tcp_hdr(skb); | |
584 | ||
585 | th->check = 0; | |
586 | skb->ip_summed = CHECKSUM_PARTIAL; | |
587 | __tcp_v4_send_check(skb, iph->saddr, iph->daddr); | |
588 | return 0; | |
589 | } | |
590 | ||
591 | /* | |
592 | * This routine will send an RST to the other tcp. | |
593 | * | |
594 | * Someone asks: why I NEVER use socket parameters (TOS, TTL etc.) | |
595 | * for reset. | |
596 | * Answer: if a packet caused RST, it is not for a socket | |
597 | * existing in our system, if it is matched to a socket, | |
598 | * it is just duplicate segment or bug in other side's TCP. | |
599 | * So that we build reply only basing on parameters | |
600 | * arrived with segment. | |
601 | * Exception: precedence violation. We do not implement it in any case. | |
602 | */ | |
603 | ||
604 | static void tcp_v4_send_reset(struct sock *sk, struct sk_buff *skb) | |
605 | { | |
606 | const struct tcphdr *th = tcp_hdr(skb); | |
607 | struct { | |
608 | struct tcphdr th; | |
609 | #ifdef CONFIG_TCP_MD5SIG | |
610 | __be32 opt[(TCPOLEN_MD5SIG_ALIGNED >> 2)]; | |
611 | #endif | |
612 | } rep; | |
613 | struct ip_reply_arg arg; | |
614 | #ifdef CONFIG_TCP_MD5SIG | |
615 | struct tcp_md5sig_key *key; | |
616 | const __u8 *hash_location = NULL; | |
617 | unsigned char newhash[16]; | |
618 | int genhash; | |
619 | struct sock *sk1 = NULL; | |
620 | #endif | |
621 | struct net *net; | |
622 | ||
623 | /* Never send a reset in response to a reset. */ | |
624 | if (th->rst) | |
625 | return; | |
626 | ||
627 | if (skb_rtable(skb)->rt_type != RTN_LOCAL) | |
628 | return; | |
629 | ||
630 | /* Swap the send and the receive. */ | |
631 | memset(&rep, 0, sizeof(rep)); | |
632 | rep.th.dest = th->source; | |
633 | rep.th.source = th->dest; | |
634 | rep.th.doff = sizeof(struct tcphdr) / 4; | |
635 | rep.th.rst = 1; | |
636 | ||
637 | if (th->ack) { | |
638 | rep.th.seq = th->ack_seq; | |
639 | } else { | |
640 | rep.th.ack = 1; | |
641 | rep.th.ack_seq = htonl(ntohl(th->seq) + th->syn + th->fin + | |
642 | skb->len - (th->doff << 2)); | |
643 | } | |
644 | ||
645 | memset(&arg, 0, sizeof(arg)); | |
646 | arg.iov[0].iov_base = (unsigned char *)&rep; | |
647 | arg.iov[0].iov_len = sizeof(rep.th); | |
648 | ||
649 | #ifdef CONFIG_TCP_MD5SIG | |
650 | hash_location = tcp_parse_md5sig_option(th); | |
651 | if (!sk && hash_location) { | |
652 | /* | |
653 | * active side is lost. Try to find listening socket through | |
654 | * source port, and then find md5 key through listening socket. | |
655 | * we are not loose security here: | |
656 | * Incoming packet is checked with md5 hash with finding key, | |
657 | * no RST generated if md5 hash doesn't match. | |
658 | */ | |
659 | sk1 = __inet_lookup_listener(dev_net(skb_dst(skb)->dev), | |
660 | &tcp_hashinfo, ip_hdr(skb)->saddr, | |
661 | th->source, ip_hdr(skb)->daddr, | |
662 | ntohs(th->source), inet_iif(skb)); | |
663 | /* don't send rst if it can't find key */ | |
664 | if (!sk1) | |
665 | return; | |
666 | rcu_read_lock(); | |
667 | key = tcp_md5_do_lookup(sk1, (union tcp_md5_addr *) | |
668 | &ip_hdr(skb)->saddr, AF_INET); | |
669 | if (!key) | |
670 | goto release_sk1; | |
671 | ||
672 | genhash = tcp_v4_md5_hash_skb(newhash, key, NULL, NULL, skb); | |
673 | if (genhash || memcmp(hash_location, newhash, 16) != 0) | |
674 | goto release_sk1; | |
675 | } else { | |
676 | key = sk ? tcp_md5_do_lookup(sk, (union tcp_md5_addr *) | |
677 | &ip_hdr(skb)->saddr, | |
678 | AF_INET) : NULL; | |
679 | } | |
680 | ||
681 | if (key) { | |
682 | rep.opt[0] = htonl((TCPOPT_NOP << 24) | | |
683 | (TCPOPT_NOP << 16) | | |
684 | (TCPOPT_MD5SIG << 8) | | |
685 | TCPOLEN_MD5SIG); | |
686 | /* Update length and the length the header thinks exists */ | |
687 | arg.iov[0].iov_len += TCPOLEN_MD5SIG_ALIGNED; | |
688 | rep.th.doff = arg.iov[0].iov_len / 4; | |
689 | ||
690 | tcp_v4_md5_hash_hdr((__u8 *) &rep.opt[1], | |
691 | key, ip_hdr(skb)->saddr, | |
692 | ip_hdr(skb)->daddr, &rep.th); | |
693 | } | |
694 | #endif | |
695 | arg.csum = csum_tcpudp_nofold(ip_hdr(skb)->daddr, | |
696 | ip_hdr(skb)->saddr, /* XXX */ | |
697 | arg.iov[0].iov_len, IPPROTO_TCP, 0); | |
698 | arg.csumoffset = offsetof(struct tcphdr, check) / 2; | |
699 | arg.flags = (sk && inet_sk(sk)->transparent) ? IP_REPLY_ARG_NOSRCCHECK : 0; | |
700 | /* When socket is gone, all binding information is lost. | |
701 | * routing might fail in this case. No choice here, if we choose to force | |
702 | * input interface, we will misroute in case of asymmetric route. | |
703 | */ | |
704 | if (sk) | |
705 | arg.bound_dev_if = sk->sk_bound_dev_if; | |
706 | ||
707 | net = dev_net(skb_dst(skb)->dev); | |
708 | arg.tos = ip_hdr(skb)->tos; | |
709 | ip_send_unicast_reply(net, skb, ip_hdr(skb)->saddr, | |
710 | ip_hdr(skb)->daddr, &arg, arg.iov[0].iov_len); | |
711 | ||
712 | TCP_INC_STATS_BH(net, TCP_MIB_OUTSEGS); | |
713 | TCP_INC_STATS_BH(net, TCP_MIB_OUTRSTS); | |
714 | ||
715 | #ifdef CONFIG_TCP_MD5SIG | |
716 | release_sk1: | |
717 | if (sk1) { | |
718 | rcu_read_unlock(); | |
719 | sock_put(sk1); | |
720 | } | |
721 | #endif | |
722 | } | |
723 | ||
724 | /* The code following below sending ACKs in SYN-RECV and TIME-WAIT states | |
725 | outside socket context is ugly, certainly. What can I do? | |
726 | */ | |
727 | ||
728 | static void tcp_v4_send_ack(struct sk_buff *skb, u32 seq, u32 ack, | |
729 | u32 win, u32 tsval, u32 tsecr, int oif, | |
730 | struct tcp_md5sig_key *key, | |
731 | int reply_flags, u8 tos) | |
732 | { | |
733 | const struct tcphdr *th = tcp_hdr(skb); | |
734 | struct { | |
735 | struct tcphdr th; | |
736 | __be32 opt[(TCPOLEN_TSTAMP_ALIGNED >> 2) | |
737 | #ifdef CONFIG_TCP_MD5SIG | |
738 | + (TCPOLEN_MD5SIG_ALIGNED >> 2) | |
739 | #endif | |
740 | ]; | |
741 | } rep; | |
742 | struct ip_reply_arg arg; | |
743 | struct net *net = dev_net(skb_dst(skb)->dev); | |
744 | ||
745 | memset(&rep.th, 0, sizeof(struct tcphdr)); | |
746 | memset(&arg, 0, sizeof(arg)); | |
747 | ||
748 | arg.iov[0].iov_base = (unsigned char *)&rep; | |
749 | arg.iov[0].iov_len = sizeof(rep.th); | |
750 | if (tsecr) { | |
751 | rep.opt[0] = htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) | | |
752 | (TCPOPT_TIMESTAMP << 8) | | |
753 | TCPOLEN_TIMESTAMP); | |
754 | rep.opt[1] = htonl(tsval); | |
755 | rep.opt[2] = htonl(tsecr); | |
756 | arg.iov[0].iov_len += TCPOLEN_TSTAMP_ALIGNED; | |
757 | } | |
758 | ||
759 | /* Swap the send and the receive. */ | |
760 | rep.th.dest = th->source; | |
761 | rep.th.source = th->dest; | |
762 | rep.th.doff = arg.iov[0].iov_len / 4; | |
763 | rep.th.seq = htonl(seq); | |
764 | rep.th.ack_seq = htonl(ack); | |
765 | rep.th.ack = 1; | |
766 | rep.th.window = htons(win); | |
767 | ||
768 | #ifdef CONFIG_TCP_MD5SIG | |
769 | if (key) { | |
770 | int offset = (tsecr) ? 3 : 0; | |
771 | ||
772 | rep.opt[offset++] = htonl((TCPOPT_NOP << 24) | | |
773 | (TCPOPT_NOP << 16) | | |
774 | (TCPOPT_MD5SIG << 8) | | |
775 | TCPOLEN_MD5SIG); | |
776 | arg.iov[0].iov_len += TCPOLEN_MD5SIG_ALIGNED; | |
777 | rep.th.doff = arg.iov[0].iov_len/4; | |
778 | ||
779 | tcp_v4_md5_hash_hdr((__u8 *) &rep.opt[offset], | |
780 | key, ip_hdr(skb)->saddr, | |
781 | ip_hdr(skb)->daddr, &rep.th); | |
782 | } | |
783 | #endif | |
784 | arg.flags = reply_flags; | |
785 | arg.csum = csum_tcpudp_nofold(ip_hdr(skb)->daddr, | |
786 | ip_hdr(skb)->saddr, /* XXX */ | |
787 | arg.iov[0].iov_len, IPPROTO_TCP, 0); | |
788 | arg.csumoffset = offsetof(struct tcphdr, check) / 2; | |
789 | if (oif) | |
790 | arg.bound_dev_if = oif; | |
791 | arg.tos = tos; | |
792 | ip_send_unicast_reply(net, skb, ip_hdr(skb)->saddr, | |
793 | ip_hdr(skb)->daddr, &arg, arg.iov[0].iov_len); | |
794 | ||
795 | TCP_INC_STATS_BH(net, TCP_MIB_OUTSEGS); | |
796 | } | |
797 | ||
798 | static void tcp_v4_timewait_ack(struct sock *sk, struct sk_buff *skb) | |
799 | { | |
800 | struct inet_timewait_sock *tw = inet_twsk(sk); | |
801 | struct tcp_timewait_sock *tcptw = tcp_twsk(sk); | |
802 | ||
803 | tcp_v4_send_ack(skb, tcptw->tw_snd_nxt, tcptw->tw_rcv_nxt, | |
804 | tcptw->tw_rcv_wnd >> tw->tw_rcv_wscale, | |
805 | tcp_time_stamp + tcptw->tw_ts_offset, | |
806 | tcptw->tw_ts_recent, | |
807 | tw->tw_bound_dev_if, | |
808 | tcp_twsk_md5_key(tcptw), | |
809 | tw->tw_transparent ? IP_REPLY_ARG_NOSRCCHECK : 0, | |
810 | tw->tw_tos | |
811 | ); | |
812 | ||
813 | inet_twsk_put(tw); | |
814 | } | |
815 | ||
816 | static void tcp_v4_reqsk_send_ack(struct sock *sk, struct sk_buff *skb, | |
817 | struct request_sock *req) | |
818 | { | |
819 | /* sk->sk_state == TCP_LISTEN -> for regular TCP_SYN_RECV | |
820 | * sk->sk_state == TCP_SYN_RECV -> for Fast Open. | |
821 | */ | |
822 | tcp_v4_send_ack(skb, (sk->sk_state == TCP_LISTEN) ? | |
823 | tcp_rsk(req)->snt_isn + 1 : tcp_sk(sk)->snd_nxt, | |
824 | tcp_rsk(req)->rcv_nxt, req->rcv_wnd, | |
825 | tcp_time_stamp, | |
826 | req->ts_recent, | |
827 | 0, | |
828 | tcp_md5_do_lookup(sk, (union tcp_md5_addr *)&ip_hdr(skb)->daddr, | |
829 | AF_INET), | |
830 | inet_rsk(req)->no_srccheck ? IP_REPLY_ARG_NOSRCCHECK : 0, | |
831 | ip_hdr(skb)->tos); | |
832 | } | |
833 | ||
834 | /* | |
835 | * Send a SYN-ACK after having received a SYN. | |
836 | * This still operates on a request_sock only, not on a big | |
837 | * socket. | |
838 | */ | |
839 | static int tcp_v4_send_synack(struct sock *sk, struct dst_entry *dst, | |
840 | struct request_sock *req, | |
841 | struct request_values *rvp, | |
842 | u16 queue_mapping, | |
843 | bool nocache) | |
844 | { | |
845 | const struct inet_request_sock *ireq = inet_rsk(req); | |
846 | struct flowi4 fl4; | |
847 | int err = -1; | |
848 | struct sk_buff * skb; | |
849 | ||
850 | /* First, grab a route. */ | |
851 | if (!dst && (dst = inet_csk_route_req(sk, &fl4, req)) == NULL) | |
852 | return -1; | |
853 | ||
854 | skb = tcp_make_synack(sk, dst, req, rvp, NULL); | |
855 | ||
856 | if (skb) { | |
857 | __tcp_v4_send_check(skb, ireq->loc_addr, ireq->rmt_addr); | |
858 | ||
859 | skb_set_queue_mapping(skb, queue_mapping); | |
860 | err = ip_build_and_send_pkt(skb, sk, ireq->loc_addr, | |
861 | ireq->rmt_addr, | |
862 | ireq->opt); | |
863 | err = net_xmit_eval(err); | |
864 | if (!tcp_rsk(req)->snt_synack && !err) | |
865 | tcp_rsk(req)->snt_synack = tcp_time_stamp; | |
866 | } | |
867 | ||
868 | return err; | |
869 | } | |
870 | ||
871 | static int tcp_v4_rtx_synack(struct sock *sk, struct request_sock *req, | |
872 | struct request_values *rvp) | |
873 | { | |
874 | int res = tcp_v4_send_synack(sk, NULL, req, rvp, 0, false); | |
875 | ||
876 | if (!res) | |
877 | TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_RETRANSSEGS); | |
878 | return res; | |
879 | } | |
880 | ||
881 | /* | |
882 | * IPv4 request_sock destructor. | |
883 | */ | |
884 | static void tcp_v4_reqsk_destructor(struct request_sock *req) | |
885 | { | |
886 | kfree(inet_rsk(req)->opt); | |
887 | } | |
888 | ||
889 | /* | |
890 | * Return true if a syncookie should be sent | |
891 | */ | |
892 | bool tcp_syn_flood_action(struct sock *sk, | |
893 | const struct sk_buff *skb, | |
894 | const char *proto) | |
895 | { | |
896 | const char *msg = "Dropping request"; | |
897 | bool want_cookie = false; | |
898 | struct listen_sock *lopt; | |
899 | ||
900 | ||
901 | ||
902 | #ifdef CONFIG_SYN_COOKIES | |
903 | if (sysctl_tcp_syncookies) { | |
904 | msg = "Sending cookies"; | |
905 | want_cookie = true; | |
906 | NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPREQQFULLDOCOOKIES); | |
907 | } else | |
908 | #endif | |
909 | NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPREQQFULLDROP); | |
910 | ||
911 | lopt = inet_csk(sk)->icsk_accept_queue.listen_opt; | |
912 | if (!lopt->synflood_warned) { | |
913 | lopt->synflood_warned = 1; | |
914 | pr_info("%s: Possible SYN flooding on port %d. %s. Check SNMP counters.\n", | |
915 | proto, ntohs(tcp_hdr(skb)->dest), msg); | |
916 | } | |
917 | return want_cookie; | |
918 | } | |
919 | EXPORT_SYMBOL(tcp_syn_flood_action); | |
920 | ||
921 | /* | |
922 | * Save and compile IPv4 options into the request_sock if needed. | |
923 | */ | |
924 | static struct ip_options_rcu *tcp_v4_save_options(struct sk_buff *skb) | |
925 | { | |
926 | const struct ip_options *opt = &(IPCB(skb)->opt); | |
927 | struct ip_options_rcu *dopt = NULL; | |
928 | ||
929 | if (opt && opt->optlen) { | |
930 | int opt_size = sizeof(*dopt) + opt->optlen; | |
931 | ||
932 | dopt = kmalloc(opt_size, GFP_ATOMIC); | |
933 | if (dopt) { | |
934 | if (ip_options_echo(&dopt->opt, skb)) { | |
935 | kfree(dopt); | |
936 | dopt = NULL; | |
937 | } | |
938 | } | |
939 | } | |
940 | return dopt; | |
941 | } | |
942 | ||
943 | #ifdef CONFIG_TCP_MD5SIG | |
944 | /* | |
945 | * RFC2385 MD5 checksumming requires a mapping of | |
946 | * IP address->MD5 Key. | |
947 | * We need to maintain these in the sk structure. | |
948 | */ | |
949 | ||
950 | /* Find the Key structure for an address. */ | |
951 | struct tcp_md5sig_key *tcp_md5_do_lookup(struct sock *sk, | |
952 | const union tcp_md5_addr *addr, | |
953 | int family) | |
954 | { | |
955 | struct tcp_sock *tp = tcp_sk(sk); | |
956 | struct tcp_md5sig_key *key; | |
957 | struct hlist_node *pos; | |
958 | unsigned int size = sizeof(struct in_addr); | |
959 | struct tcp_md5sig_info *md5sig; | |
960 | ||
961 | /* caller either holds rcu_read_lock() or socket lock */ | |
962 | md5sig = rcu_dereference_check(tp->md5sig_info, | |
963 | sock_owned_by_user(sk) || | |
964 | lockdep_is_held(&sk->sk_lock.slock)); | |
965 | if (!md5sig) | |
966 | return NULL; | |
967 | #if IS_ENABLED(CONFIG_IPV6) | |
968 | if (family == AF_INET6) | |
969 | size = sizeof(struct in6_addr); | |
970 | #endif | |
971 | hlist_for_each_entry_rcu(key, pos, &md5sig->head, node) { | |
972 | if (key->family != family) | |
973 | continue; | |
974 | if (!memcmp(&key->addr, addr, size)) | |
975 | return key; | |
976 | } | |
977 | return NULL; | |
978 | } | |
979 | EXPORT_SYMBOL(tcp_md5_do_lookup); | |
980 | ||
981 | struct tcp_md5sig_key *tcp_v4_md5_lookup(struct sock *sk, | |
982 | struct sock *addr_sk) | |
983 | { | |
984 | union tcp_md5_addr *addr; | |
985 | ||
986 | addr = (union tcp_md5_addr *)&inet_sk(addr_sk)->inet_daddr; | |
987 | return tcp_md5_do_lookup(sk, addr, AF_INET); | |
988 | } | |
989 | EXPORT_SYMBOL(tcp_v4_md5_lookup); | |
990 | ||
991 | static struct tcp_md5sig_key *tcp_v4_reqsk_md5_lookup(struct sock *sk, | |
992 | struct request_sock *req) | |
993 | { | |
994 | union tcp_md5_addr *addr; | |
995 | ||
996 | addr = (union tcp_md5_addr *)&inet_rsk(req)->rmt_addr; | |
997 | return tcp_md5_do_lookup(sk, addr, AF_INET); | |
998 | } | |
999 | ||
1000 | /* This can be called on a newly created socket, from other files */ | |
1001 | int tcp_md5_do_add(struct sock *sk, const union tcp_md5_addr *addr, | |
1002 | int family, const u8 *newkey, u8 newkeylen, gfp_t gfp) | |
1003 | { | |
1004 | /* Add Key to the list */ | |
1005 | struct tcp_md5sig_key *key; | |
1006 | struct tcp_sock *tp = tcp_sk(sk); | |
1007 | struct tcp_md5sig_info *md5sig; | |
1008 | ||
1009 | key = tcp_md5_do_lookup(sk, (union tcp_md5_addr *)&addr, AF_INET); | |
1010 | if (key) { | |
1011 | /* Pre-existing entry - just update that one. */ | |
1012 | memcpy(key->key, newkey, newkeylen); | |
1013 | key->keylen = newkeylen; | |
1014 | return 0; | |
1015 | } | |
1016 | ||
1017 | md5sig = rcu_dereference_protected(tp->md5sig_info, | |
1018 | sock_owned_by_user(sk)); | |
1019 | if (!md5sig) { | |
1020 | md5sig = kmalloc(sizeof(*md5sig), gfp); | |
1021 | if (!md5sig) | |
1022 | return -ENOMEM; | |
1023 | ||
1024 | sk_nocaps_add(sk, NETIF_F_GSO_MASK); | |
1025 | INIT_HLIST_HEAD(&md5sig->head); | |
1026 | rcu_assign_pointer(tp->md5sig_info, md5sig); | |
1027 | } | |
1028 | ||
1029 | key = sock_kmalloc(sk, sizeof(*key), gfp); | |
1030 | if (!key) | |
1031 | return -ENOMEM; | |
1032 | if (hlist_empty(&md5sig->head) && !tcp_alloc_md5sig_pool(sk)) { | |
1033 | sock_kfree_s(sk, key, sizeof(*key)); | |
1034 | return -ENOMEM; | |
1035 | } | |
1036 | ||
1037 | memcpy(key->key, newkey, newkeylen); | |
1038 | key->keylen = newkeylen; | |
1039 | key->family = family; | |
1040 | memcpy(&key->addr, addr, | |
1041 | (family == AF_INET6) ? sizeof(struct in6_addr) : | |
1042 | sizeof(struct in_addr)); | |
1043 | hlist_add_head_rcu(&key->node, &md5sig->head); | |
1044 | return 0; | |
1045 | } | |
1046 | EXPORT_SYMBOL(tcp_md5_do_add); | |
1047 | ||
1048 | int tcp_md5_do_del(struct sock *sk, const union tcp_md5_addr *addr, int family) | |
1049 | { | |
1050 | struct tcp_sock *tp = tcp_sk(sk); | |
1051 | struct tcp_md5sig_key *key; | |
1052 | struct tcp_md5sig_info *md5sig; | |
1053 | ||
1054 | key = tcp_md5_do_lookup(sk, (union tcp_md5_addr *)&addr, AF_INET); | |
1055 | if (!key) | |
1056 | return -ENOENT; | |
1057 | hlist_del_rcu(&key->node); | |
1058 | atomic_sub(sizeof(*key), &sk->sk_omem_alloc); | |
1059 | kfree_rcu(key, rcu); | |
1060 | md5sig = rcu_dereference_protected(tp->md5sig_info, | |
1061 | sock_owned_by_user(sk)); | |
1062 | if (hlist_empty(&md5sig->head)) | |
1063 | tcp_free_md5sig_pool(); | |
1064 | return 0; | |
1065 | } | |
1066 | EXPORT_SYMBOL(tcp_md5_do_del); | |
1067 | ||
1068 | static void tcp_clear_md5_list(struct sock *sk) | |
1069 | { | |
1070 | struct tcp_sock *tp = tcp_sk(sk); | |
1071 | struct tcp_md5sig_key *key; | |
1072 | struct hlist_node *pos, *n; | |
1073 | struct tcp_md5sig_info *md5sig; | |
1074 | ||
1075 | md5sig = rcu_dereference_protected(tp->md5sig_info, 1); | |
1076 | ||
1077 | if (!hlist_empty(&md5sig->head)) | |
1078 | tcp_free_md5sig_pool(); | |
1079 | hlist_for_each_entry_safe(key, pos, n, &md5sig->head, node) { | |
1080 | hlist_del_rcu(&key->node); | |
1081 | atomic_sub(sizeof(*key), &sk->sk_omem_alloc); | |
1082 | kfree_rcu(key, rcu); | |
1083 | } | |
1084 | } | |
1085 | ||
1086 | static int tcp_v4_parse_md5_keys(struct sock *sk, char __user *optval, | |
1087 | int optlen) | |
1088 | { | |
1089 | struct tcp_md5sig cmd; | |
1090 | struct sockaddr_in *sin = (struct sockaddr_in *)&cmd.tcpm_addr; | |
1091 | ||
1092 | if (optlen < sizeof(cmd)) | |
1093 | return -EINVAL; | |
1094 | ||
1095 | if (copy_from_user(&cmd, optval, sizeof(cmd))) | |
1096 | return -EFAULT; | |
1097 | ||
1098 | if (sin->sin_family != AF_INET) | |
1099 | return -EINVAL; | |
1100 | ||
1101 | if (!cmd.tcpm_key || !cmd.tcpm_keylen) | |
1102 | return tcp_md5_do_del(sk, (union tcp_md5_addr *)&sin->sin_addr.s_addr, | |
1103 | AF_INET); | |
1104 | ||
1105 | if (cmd.tcpm_keylen > TCP_MD5SIG_MAXKEYLEN) | |
1106 | return -EINVAL; | |
1107 | ||
1108 | return tcp_md5_do_add(sk, (union tcp_md5_addr *)&sin->sin_addr.s_addr, | |
1109 | AF_INET, cmd.tcpm_key, cmd.tcpm_keylen, | |
1110 | GFP_KERNEL); | |
1111 | } | |
1112 | ||
1113 | static int tcp_v4_md5_hash_pseudoheader(struct tcp_md5sig_pool *hp, | |
1114 | __be32 daddr, __be32 saddr, int nbytes) | |
1115 | { | |
1116 | struct tcp4_pseudohdr *bp; | |
1117 | struct scatterlist sg; | |
1118 | ||
1119 | bp = &hp->md5_blk.ip4; | |
1120 | ||
1121 | /* | |
1122 | * 1. the TCP pseudo-header (in the order: source IP address, | |
1123 | * destination IP address, zero-padded protocol number, and | |
1124 | * segment length) | |
1125 | */ | |
1126 | bp->saddr = saddr; | |
1127 | bp->daddr = daddr; | |
1128 | bp->pad = 0; | |
1129 | bp->protocol = IPPROTO_TCP; | |
1130 | bp->len = cpu_to_be16(nbytes); | |
1131 | ||
1132 | sg_init_one(&sg, bp, sizeof(*bp)); | |
1133 | return crypto_hash_update(&hp->md5_desc, &sg, sizeof(*bp)); | |
1134 | } | |
1135 | ||
1136 | static int tcp_v4_md5_hash_hdr(char *md5_hash, const struct tcp_md5sig_key *key, | |
1137 | __be32 daddr, __be32 saddr, const struct tcphdr *th) | |
1138 | { | |
1139 | struct tcp_md5sig_pool *hp; | |
1140 | struct hash_desc *desc; | |
1141 | ||
1142 | hp = tcp_get_md5sig_pool(); | |
1143 | if (!hp) | |
1144 | goto clear_hash_noput; | |
1145 | desc = &hp->md5_desc; | |
1146 | ||
1147 | if (crypto_hash_init(desc)) | |
1148 | goto clear_hash; | |
1149 | if (tcp_v4_md5_hash_pseudoheader(hp, daddr, saddr, th->doff << 2)) | |
1150 | goto clear_hash; | |
1151 | if (tcp_md5_hash_header(hp, th)) | |
1152 | goto clear_hash; | |
1153 | if (tcp_md5_hash_key(hp, key)) | |
1154 | goto clear_hash; | |
1155 | if (crypto_hash_final(desc, md5_hash)) | |
1156 | goto clear_hash; | |
1157 | ||
1158 | tcp_put_md5sig_pool(); | |
1159 | return 0; | |
1160 | ||
1161 | clear_hash: | |
1162 | tcp_put_md5sig_pool(); | |
1163 | clear_hash_noput: | |
1164 | memset(md5_hash, 0, 16); | |
1165 | return 1; | |
1166 | } | |
1167 | ||
1168 | int tcp_v4_md5_hash_skb(char *md5_hash, struct tcp_md5sig_key *key, | |
1169 | const struct sock *sk, const struct request_sock *req, | |
1170 | const struct sk_buff *skb) | |
1171 | { | |
1172 | struct tcp_md5sig_pool *hp; | |
1173 | struct hash_desc *desc; | |
1174 | const struct tcphdr *th = tcp_hdr(skb); | |
1175 | __be32 saddr, daddr; | |
1176 | ||
1177 | if (sk) { | |
1178 | saddr = inet_sk(sk)->inet_saddr; | |
1179 | daddr = inet_sk(sk)->inet_daddr; | |
1180 | } else if (req) { | |
1181 | saddr = inet_rsk(req)->loc_addr; | |
1182 | daddr = inet_rsk(req)->rmt_addr; | |
1183 | } else { | |
1184 | const struct iphdr *iph = ip_hdr(skb); | |
1185 | saddr = iph->saddr; | |
1186 | daddr = iph->daddr; | |
1187 | } | |
1188 | ||
1189 | hp = tcp_get_md5sig_pool(); | |
1190 | if (!hp) | |
1191 | goto clear_hash_noput; | |
1192 | desc = &hp->md5_desc; | |
1193 | ||
1194 | if (crypto_hash_init(desc)) | |
1195 | goto clear_hash; | |
1196 | ||
1197 | if (tcp_v4_md5_hash_pseudoheader(hp, daddr, saddr, skb->len)) | |
1198 | goto clear_hash; | |
1199 | if (tcp_md5_hash_header(hp, th)) | |
1200 | goto clear_hash; | |
1201 | if (tcp_md5_hash_skb_data(hp, skb, th->doff << 2)) | |
1202 | goto clear_hash; | |
1203 | if (tcp_md5_hash_key(hp, key)) | |
1204 | goto clear_hash; | |
1205 | if (crypto_hash_final(desc, md5_hash)) | |
1206 | goto clear_hash; | |
1207 | ||
1208 | tcp_put_md5sig_pool(); | |
1209 | return 0; | |
1210 | ||
1211 | clear_hash: | |
1212 | tcp_put_md5sig_pool(); | |
1213 | clear_hash_noput: | |
1214 | memset(md5_hash, 0, 16); | |
1215 | return 1; | |
1216 | } | |
1217 | EXPORT_SYMBOL(tcp_v4_md5_hash_skb); | |
1218 | ||
1219 | static bool tcp_v4_inbound_md5_hash(struct sock *sk, const struct sk_buff *skb) | |
1220 | { | |
1221 | /* | |
1222 | * This gets called for each TCP segment that arrives | |
1223 | * so we want to be efficient. | |
1224 | * We have 3 drop cases: | |
1225 | * o No MD5 hash and one expected. | |
1226 | * o MD5 hash and we're not expecting one. | |
1227 | * o MD5 hash and its wrong. | |
1228 | */ | |
1229 | const __u8 *hash_location = NULL; | |
1230 | struct tcp_md5sig_key *hash_expected; | |
1231 | const struct iphdr *iph = ip_hdr(skb); | |
1232 | const struct tcphdr *th = tcp_hdr(skb); | |
1233 | int genhash; | |
1234 | unsigned char newhash[16]; | |
1235 | ||
1236 | hash_expected = tcp_md5_do_lookup(sk, (union tcp_md5_addr *)&iph->saddr, | |
1237 | AF_INET); | |
1238 | hash_location = tcp_parse_md5sig_option(th); | |
1239 | ||
1240 | /* We've parsed the options - do we have a hash? */ | |
1241 | if (!hash_expected && !hash_location) | |
1242 | return false; | |
1243 | ||
1244 | if (hash_expected && !hash_location) { | |
1245 | NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPMD5NOTFOUND); | |
1246 | return true; | |
1247 | } | |
1248 | ||
1249 | if (!hash_expected && hash_location) { | |
1250 | NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPMD5UNEXPECTED); | |
1251 | return true; | |
1252 | } | |
1253 | ||
1254 | /* Okay, so this is hash_expected and hash_location - | |
1255 | * so we need to calculate the checksum. | |
1256 | */ | |
1257 | genhash = tcp_v4_md5_hash_skb(newhash, | |
1258 | hash_expected, | |
1259 | NULL, NULL, skb); | |
1260 | ||
1261 | if (genhash || memcmp(hash_location, newhash, 16) != 0) { | |
1262 | net_info_ratelimited("MD5 Hash failed for (%pI4, %d)->(%pI4, %d)%s\n", | |
1263 | &iph->saddr, ntohs(th->source), | |
1264 | &iph->daddr, ntohs(th->dest), | |
1265 | genhash ? " tcp_v4_calc_md5_hash failed" | |
1266 | : ""); | |
1267 | return true; | |
1268 | } | |
1269 | return false; | |
1270 | } | |
1271 | ||
1272 | #endif | |
1273 | ||
1274 | struct request_sock_ops tcp_request_sock_ops __read_mostly = { | |
1275 | .family = PF_INET, | |
1276 | .obj_size = sizeof(struct tcp_request_sock), | |
1277 | .rtx_syn_ack = tcp_v4_rtx_synack, | |
1278 | .send_ack = tcp_v4_reqsk_send_ack, | |
1279 | .destructor = tcp_v4_reqsk_destructor, | |
1280 | .send_reset = tcp_v4_send_reset, | |
1281 | .syn_ack_timeout = tcp_syn_ack_timeout, | |
1282 | }; | |
1283 | ||
1284 | #ifdef CONFIG_TCP_MD5SIG | |
1285 | static const struct tcp_request_sock_ops tcp_request_sock_ipv4_ops = { | |
1286 | .md5_lookup = tcp_v4_reqsk_md5_lookup, | |
1287 | .calc_md5_hash = tcp_v4_md5_hash_skb, | |
1288 | }; | |
1289 | #endif | |
1290 | ||
1291 | static bool tcp_fastopen_check(struct sock *sk, struct sk_buff *skb, | |
1292 | struct request_sock *req, | |
1293 | struct tcp_fastopen_cookie *foc, | |
1294 | struct tcp_fastopen_cookie *valid_foc) | |
1295 | { | |
1296 | bool skip_cookie = false; | |
1297 | struct fastopen_queue *fastopenq; | |
1298 | ||
1299 | if (likely(!fastopen_cookie_present(foc))) { | |
1300 | /* See include/net/tcp.h for the meaning of these knobs */ | |
1301 | if ((sysctl_tcp_fastopen & TFO_SERVER_ALWAYS) || | |
1302 | ((sysctl_tcp_fastopen & TFO_SERVER_COOKIE_NOT_REQD) && | |
1303 | (TCP_SKB_CB(skb)->end_seq != TCP_SKB_CB(skb)->seq + 1))) | |
1304 | skip_cookie = true; /* no cookie to validate */ | |
1305 | else | |
1306 | return false; | |
1307 | } | |
1308 | fastopenq = inet_csk(sk)->icsk_accept_queue.fastopenq; | |
1309 | /* A FO option is present; bump the counter. */ | |
1310 | NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPFASTOPENPASSIVE); | |
1311 | ||
1312 | /* Make sure the listener has enabled fastopen, and we don't | |
1313 | * exceed the max # of pending TFO requests allowed before trying | |
1314 | * to validating the cookie in order to avoid burning CPU cycles | |
1315 | * unnecessarily. | |
1316 | * | |
1317 | * XXX (TFO) - The implication of checking the max_qlen before | |
1318 | * processing a cookie request is that clients can't differentiate | |
1319 | * between qlen overflow causing Fast Open to be disabled | |
1320 | * temporarily vs a server not supporting Fast Open at all. | |
1321 | */ | |
1322 | if ((sysctl_tcp_fastopen & TFO_SERVER_ENABLE) == 0 || | |
1323 | fastopenq == NULL || fastopenq->max_qlen == 0) | |
1324 | return false; | |
1325 | ||
1326 | if (fastopenq->qlen >= fastopenq->max_qlen) { | |
1327 | struct request_sock *req1; | |
1328 | spin_lock(&fastopenq->lock); | |
1329 | req1 = fastopenq->rskq_rst_head; | |
1330 | if ((req1 == NULL) || time_after(req1->expires, jiffies)) { | |
1331 | spin_unlock(&fastopenq->lock); | |
1332 | NET_INC_STATS_BH(sock_net(sk), | |
1333 | LINUX_MIB_TCPFASTOPENLISTENOVERFLOW); | |
1334 | /* Avoid bumping LINUX_MIB_TCPFASTOPENPASSIVEFAIL*/ | |
1335 | foc->len = -1; | |
1336 | return false; | |
1337 | } | |
1338 | fastopenq->rskq_rst_head = req1->dl_next; | |
1339 | fastopenq->qlen--; | |
1340 | spin_unlock(&fastopenq->lock); | |
1341 | reqsk_free(req1); | |
1342 | } | |
1343 | if (skip_cookie) { | |
1344 | tcp_rsk(req)->rcv_nxt = TCP_SKB_CB(skb)->end_seq; | |
1345 | return true; | |
1346 | } | |
1347 | if (foc->len == TCP_FASTOPEN_COOKIE_SIZE) { | |
1348 | if ((sysctl_tcp_fastopen & TFO_SERVER_COOKIE_NOT_CHKED) == 0) { | |
1349 | tcp_fastopen_cookie_gen(ip_hdr(skb)->saddr, valid_foc); | |
1350 | if ((valid_foc->len != TCP_FASTOPEN_COOKIE_SIZE) || | |
1351 | memcmp(&foc->val[0], &valid_foc->val[0], | |
1352 | TCP_FASTOPEN_COOKIE_SIZE) != 0) | |
1353 | return false; | |
1354 | valid_foc->len = -1; | |
1355 | } | |
1356 | /* Acknowledge the data received from the peer. */ | |
1357 | tcp_rsk(req)->rcv_nxt = TCP_SKB_CB(skb)->end_seq; | |
1358 | return true; | |
1359 | } else if (foc->len == 0) { /* Client requesting a cookie */ | |
1360 | tcp_fastopen_cookie_gen(ip_hdr(skb)->saddr, valid_foc); | |
1361 | NET_INC_STATS_BH(sock_net(sk), | |
1362 | LINUX_MIB_TCPFASTOPENCOOKIEREQD); | |
1363 | } else { | |
1364 | /* Client sent a cookie with wrong size. Treat it | |
1365 | * the same as invalid and return a valid one. | |
1366 | */ | |
1367 | tcp_fastopen_cookie_gen(ip_hdr(skb)->saddr, valid_foc); | |
1368 | } | |
1369 | return false; | |
1370 | } | |
1371 | ||
1372 | static int tcp_v4_conn_req_fastopen(struct sock *sk, | |
1373 | struct sk_buff *skb, | |
1374 | struct sk_buff *skb_synack, | |
1375 | struct request_sock *req, | |
1376 | struct request_values *rvp) | |
1377 | { | |
1378 | struct tcp_sock *tp = tcp_sk(sk); | |
1379 | struct request_sock_queue *queue = &inet_csk(sk)->icsk_accept_queue; | |
1380 | const struct inet_request_sock *ireq = inet_rsk(req); | |
1381 | struct sock *child; | |
1382 | int err; | |
1383 | ||
1384 | req->num_retrans = 0; | |
1385 | req->num_timeout = 0; | |
1386 | req->sk = NULL; | |
1387 | ||
1388 | child = inet_csk(sk)->icsk_af_ops->syn_recv_sock(sk, skb, req, NULL); | |
1389 | if (child == NULL) { | |
1390 | NET_INC_STATS_BH(sock_net(sk), | |
1391 | LINUX_MIB_TCPFASTOPENPASSIVEFAIL); | |
1392 | kfree_skb(skb_synack); | |
1393 | return -1; | |
1394 | } | |
1395 | err = ip_build_and_send_pkt(skb_synack, sk, ireq->loc_addr, | |
1396 | ireq->rmt_addr, ireq->opt); | |
1397 | err = net_xmit_eval(err); | |
1398 | if (!err) | |
1399 | tcp_rsk(req)->snt_synack = tcp_time_stamp; | |
1400 | /* XXX (TFO) - is it ok to ignore error and continue? */ | |
1401 | ||
1402 | spin_lock(&queue->fastopenq->lock); | |
1403 | queue->fastopenq->qlen++; | |
1404 | spin_unlock(&queue->fastopenq->lock); | |
1405 | ||
1406 | /* Initialize the child socket. Have to fix some values to take | |
1407 | * into account the child is a Fast Open socket and is created | |
1408 | * only out of the bits carried in the SYN packet. | |
1409 | */ | |
1410 | tp = tcp_sk(child); | |
1411 | ||
1412 | tp->fastopen_rsk = req; | |
1413 | /* Do a hold on the listner sk so that if the listener is being | |
1414 | * closed, the child that has been accepted can live on and still | |
1415 | * access listen_lock. | |
1416 | */ | |
1417 | sock_hold(sk); | |
1418 | tcp_rsk(req)->listener = sk; | |
1419 | ||
1420 | /* RFC1323: The window in SYN & SYN/ACK segments is never | |
1421 | * scaled. So correct it appropriately. | |
1422 | */ | |
1423 | tp->snd_wnd = ntohs(tcp_hdr(skb)->window); | |
1424 | ||
1425 | /* Activate the retrans timer so that SYNACK can be retransmitted. | |
1426 | * The request socket is not added to the SYN table of the parent | |
1427 | * because it's been added to the accept queue directly. | |
1428 | */ | |
1429 | inet_csk_reset_xmit_timer(child, ICSK_TIME_RETRANS, | |
1430 | TCP_TIMEOUT_INIT, TCP_RTO_MAX); | |
1431 | ||
1432 | /* Add the child socket directly into the accept queue */ | |
1433 | inet_csk_reqsk_queue_add(sk, req, child); | |
1434 | ||
1435 | /* Now finish processing the fastopen child socket. */ | |
1436 | inet_csk(child)->icsk_af_ops->rebuild_header(child); | |
1437 | tcp_init_congestion_control(child); | |
1438 | tcp_mtup_init(child); | |
1439 | tcp_init_buffer_space(child); | |
1440 | tcp_init_metrics(child); | |
1441 | ||
1442 | /* Queue the data carried in the SYN packet. We need to first | |
1443 | * bump skb's refcnt because the caller will attempt to free it. | |
1444 | * | |
1445 | * XXX (TFO) - we honor a zero-payload TFO request for now. | |
1446 | * (Any reason not to?) | |
1447 | */ | |
1448 | if (TCP_SKB_CB(skb)->end_seq == TCP_SKB_CB(skb)->seq + 1) { | |
1449 | /* Don't queue the skb if there is no payload in SYN. | |
1450 | * XXX (TFO) - How about SYN+FIN? | |
1451 | */ | |
1452 | tp->rcv_nxt = TCP_SKB_CB(skb)->end_seq; | |
1453 | } else { | |
1454 | skb = skb_get(skb); | |
1455 | skb_dst_drop(skb); | |
1456 | __skb_pull(skb, tcp_hdr(skb)->doff * 4); | |
1457 | skb_set_owner_r(skb, child); | |
1458 | __skb_queue_tail(&child->sk_receive_queue, skb); | |
1459 | tp->rcv_nxt = TCP_SKB_CB(skb)->end_seq; | |
1460 | tp->syn_data_acked = 1; | |
1461 | } | |
1462 | sk->sk_data_ready(sk, 0); | |
1463 | bh_unlock_sock(child); | |
1464 | sock_put(child); | |
1465 | WARN_ON(req->sk == NULL); | |
1466 | return 0; | |
1467 | } | |
1468 | ||
1469 | int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb) | |
1470 | { | |
1471 | struct tcp_extend_values tmp_ext; | |
1472 | struct tcp_options_received tmp_opt; | |
1473 | const u8 *hash_location; | |
1474 | struct request_sock *req; | |
1475 | struct inet_request_sock *ireq; | |
1476 | struct tcp_sock *tp = tcp_sk(sk); | |
1477 | struct dst_entry *dst = NULL; | |
1478 | __be32 saddr = ip_hdr(skb)->saddr; | |
1479 | __be32 daddr = ip_hdr(skb)->daddr; | |
1480 | __u32 isn = TCP_SKB_CB(skb)->when; | |
1481 | bool want_cookie = false; | |
1482 | struct flowi4 fl4; | |
1483 | struct tcp_fastopen_cookie foc = { .len = -1 }; | |
1484 | struct tcp_fastopen_cookie valid_foc = { .len = -1 }; | |
1485 | struct sk_buff *skb_synack; | |
1486 | int do_fastopen; | |
1487 | ||
1488 | /* Never answer to SYNs send to broadcast or multicast */ | |
1489 | if (skb_rtable(skb)->rt_flags & (RTCF_BROADCAST | RTCF_MULTICAST)) | |
1490 | goto drop; | |
1491 | ||
1492 | /* TW buckets are converted to open requests without | |
1493 | * limitations, they conserve resources and peer is | |
1494 | * evidently real one. | |
1495 | */ | |
1496 | if (inet_csk_reqsk_queue_is_full(sk) && !isn) { | |
1497 | want_cookie = tcp_syn_flood_action(sk, skb, "TCP"); | |
1498 | if (!want_cookie) | |
1499 | goto drop; | |
1500 | } | |
1501 | ||
1502 | /* Accept backlog is full. If we have already queued enough | |
1503 | * of warm entries in syn queue, drop request. It is better than | |
1504 | * clogging syn queue with openreqs with exponentially increasing | |
1505 | * timeout. | |
1506 | */ | |
1507 | if (sk_acceptq_is_full(sk) && inet_csk_reqsk_queue_young(sk) > 1) { | |
1508 | NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENOVERFLOWS); | |
1509 | goto drop; | |
1510 | } | |
1511 | ||
1512 | req = inet_reqsk_alloc(&tcp_request_sock_ops); | |
1513 | if (!req) | |
1514 | goto drop; | |
1515 | ||
1516 | #ifdef CONFIG_TCP_MD5SIG | |
1517 | tcp_rsk(req)->af_specific = &tcp_request_sock_ipv4_ops; | |
1518 | #endif | |
1519 | ||
1520 | tcp_clear_options(&tmp_opt); | |
1521 | tmp_opt.mss_clamp = TCP_MSS_DEFAULT; | |
1522 | tmp_opt.user_mss = tp->rx_opt.user_mss; | |
1523 | tcp_parse_options(skb, &tmp_opt, &hash_location, 0, | |
1524 | want_cookie ? NULL : &foc); | |
1525 | ||
1526 | if (tmp_opt.cookie_plus > 0 && | |
1527 | tmp_opt.saw_tstamp && | |
1528 | !tp->rx_opt.cookie_out_never && | |
1529 | (sysctl_tcp_cookie_size > 0 || | |
1530 | (tp->cookie_values != NULL && | |
1531 | tp->cookie_values->cookie_desired > 0))) { | |
1532 | u8 *c; | |
1533 | u32 *mess = &tmp_ext.cookie_bakery[COOKIE_DIGEST_WORDS]; | |
1534 | int l = tmp_opt.cookie_plus - TCPOLEN_COOKIE_BASE; | |
1535 | ||
1536 | if (tcp_cookie_generator(&tmp_ext.cookie_bakery[0]) != 0) | |
1537 | goto drop_and_release; | |
1538 | ||
1539 | /* Secret recipe starts with IP addresses */ | |
1540 | *mess++ ^= (__force u32)daddr; | |
1541 | *mess++ ^= (__force u32)saddr; | |
1542 | ||
1543 | /* plus variable length Initiator Cookie */ | |
1544 | c = (u8 *)mess; | |
1545 | while (l-- > 0) | |
1546 | *c++ ^= *hash_location++; | |
1547 | ||
1548 | want_cookie = false; /* not our kind of cookie */ | |
1549 | tmp_ext.cookie_out_never = 0; /* false */ | |
1550 | tmp_ext.cookie_plus = tmp_opt.cookie_plus; | |
1551 | } else if (!tp->rx_opt.cookie_in_always) { | |
1552 | /* redundant indications, but ensure initialization. */ | |
1553 | tmp_ext.cookie_out_never = 1; /* true */ | |
1554 | tmp_ext.cookie_plus = 0; | |
1555 | } else { | |
1556 | goto drop_and_release; | |
1557 | } | |
1558 | tmp_ext.cookie_in_always = tp->rx_opt.cookie_in_always; | |
1559 | ||
1560 | if (want_cookie && !tmp_opt.saw_tstamp) | |
1561 | tcp_clear_options(&tmp_opt); | |
1562 | ||
1563 | tmp_opt.tstamp_ok = tmp_opt.saw_tstamp; | |
1564 | tcp_openreq_init(req, &tmp_opt, skb); | |
1565 | ||
1566 | ireq = inet_rsk(req); | |
1567 | ireq->loc_addr = daddr; | |
1568 | ireq->rmt_addr = saddr; | |
1569 | ireq->no_srccheck = inet_sk(sk)->transparent; | |
1570 | ireq->opt = tcp_v4_save_options(skb); | |
1571 | ||
1572 | if (security_inet_conn_request(sk, skb, req)) | |
1573 | goto drop_and_free; | |
1574 | ||
1575 | if (!want_cookie || tmp_opt.tstamp_ok) | |
1576 | TCP_ECN_create_request(req, skb, sock_net(sk)); | |
1577 | ||
1578 | if (want_cookie) { | |
1579 | isn = cookie_v4_init_sequence(sk, skb, &req->mss); | |
1580 | req->cookie_ts = tmp_opt.tstamp_ok; | |
1581 | } else if (!isn) { | |
1582 | /* VJ's idea. We save last timestamp seen | |
1583 | * from the destination in peer table, when entering | |
1584 | * state TIME-WAIT, and check against it before | |
1585 | * accepting new connection request. | |
1586 | * | |
1587 | * If "isn" is not zero, this request hit alive | |
1588 | * timewait bucket, so that all the necessary checks | |
1589 | * are made in the function processing timewait state. | |
1590 | */ | |
1591 | if (tmp_opt.saw_tstamp && | |
1592 | tcp_death_row.sysctl_tw_recycle && | |
1593 | (dst = inet_csk_route_req(sk, &fl4, req)) != NULL && | |
1594 | fl4.daddr == saddr) { | |
1595 | if (!tcp_peer_is_proven(req, dst, true)) { | |
1596 | NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_PAWSPASSIVEREJECTED); | |
1597 | goto drop_and_release; | |
1598 | } | |
1599 | } | |
1600 | /* Kill the following clause, if you dislike this way. */ | |
1601 | else if (!sysctl_tcp_syncookies && | |
1602 | (sysctl_max_syn_backlog - inet_csk_reqsk_queue_len(sk) < | |
1603 | (sysctl_max_syn_backlog >> 2)) && | |
1604 | !tcp_peer_is_proven(req, dst, false)) { | |
1605 | /* Without syncookies last quarter of | |
1606 | * backlog is filled with destinations, | |
1607 | * proven to be alive. | |
1608 | * It means that we continue to communicate | |
1609 | * to destinations, already remembered | |
1610 | * to the moment of synflood. | |
1611 | */ | |
1612 | LIMIT_NETDEBUG(KERN_DEBUG pr_fmt("drop open request from %pI4/%u\n"), | |
1613 | &saddr, ntohs(tcp_hdr(skb)->source)); | |
1614 | goto drop_and_release; | |
1615 | } | |
1616 | ||
1617 | isn = tcp_v4_init_sequence(skb); | |
1618 | } | |
1619 | tcp_rsk(req)->snt_isn = isn; | |
1620 | ||
1621 | if (dst == NULL) { | |
1622 | dst = inet_csk_route_req(sk, &fl4, req); | |
1623 | if (dst == NULL) | |
1624 | goto drop_and_free; | |
1625 | } | |
1626 | do_fastopen = tcp_fastopen_check(sk, skb, req, &foc, &valid_foc); | |
1627 | ||
1628 | /* We don't call tcp_v4_send_synack() directly because we need | |
1629 | * to make sure a child socket can be created successfully before | |
1630 | * sending back synack! | |
1631 | * | |
1632 | * XXX (TFO) - Ideally one would simply call tcp_v4_send_synack() | |
1633 | * (or better yet, call tcp_send_synack() in the child context | |
1634 | * directly, but will have to fix bunch of other code first) | |
1635 | * after syn_recv_sock() except one will need to first fix the | |
1636 | * latter to remove its dependency on the current implementation | |
1637 | * of tcp_v4_send_synack()->tcp_select_initial_window(). | |
1638 | */ | |
1639 | skb_synack = tcp_make_synack(sk, dst, req, | |
1640 | (struct request_values *)&tmp_ext, | |
1641 | fastopen_cookie_present(&valid_foc) ? &valid_foc : NULL); | |
1642 | ||
1643 | if (skb_synack) { | |
1644 | __tcp_v4_send_check(skb_synack, ireq->loc_addr, ireq->rmt_addr); | |
1645 | skb_set_queue_mapping(skb_synack, skb_get_queue_mapping(skb)); | |
1646 | } else | |
1647 | goto drop_and_free; | |
1648 | ||
1649 | if (likely(!do_fastopen)) { | |
1650 | int err; | |
1651 | err = ip_build_and_send_pkt(skb_synack, sk, ireq->loc_addr, | |
1652 | ireq->rmt_addr, ireq->opt); | |
1653 | err = net_xmit_eval(err); | |
1654 | if (err || want_cookie) | |
1655 | goto drop_and_free; | |
1656 | ||
1657 | tcp_rsk(req)->snt_synack = tcp_time_stamp; | |
1658 | tcp_rsk(req)->listener = NULL; | |
1659 | /* Add the request_sock to the SYN table */ | |
1660 | inet_csk_reqsk_queue_hash_add(sk, req, TCP_TIMEOUT_INIT); | |
1661 | if (fastopen_cookie_present(&foc) && foc.len != 0) | |
1662 | NET_INC_STATS_BH(sock_net(sk), | |
1663 | LINUX_MIB_TCPFASTOPENPASSIVEFAIL); | |
1664 | } else if (tcp_v4_conn_req_fastopen(sk, skb, skb_synack, req, | |
1665 | (struct request_values *)&tmp_ext)) | |
1666 | goto drop_and_free; | |
1667 | ||
1668 | return 0; | |
1669 | ||
1670 | drop_and_release: | |
1671 | dst_release(dst); | |
1672 | drop_and_free: | |
1673 | reqsk_free(req); | |
1674 | drop: | |
1675 | NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENDROPS); | |
1676 | return 0; | |
1677 | } | |
1678 | EXPORT_SYMBOL(tcp_v4_conn_request); | |
1679 | ||
1680 | ||
1681 | /* | |
1682 | * The three way handshake has completed - we got a valid synack - | |
1683 | * now create the new socket. | |
1684 | */ | |
1685 | struct sock *tcp_v4_syn_recv_sock(struct sock *sk, struct sk_buff *skb, | |
1686 | struct request_sock *req, | |
1687 | struct dst_entry *dst) | |
1688 | { | |
1689 | struct inet_request_sock *ireq; | |
1690 | struct inet_sock *newinet; | |
1691 | struct tcp_sock *newtp; | |
1692 | struct sock *newsk; | |
1693 | #ifdef CONFIG_TCP_MD5SIG | |
1694 | struct tcp_md5sig_key *key; | |
1695 | #endif | |
1696 | struct ip_options_rcu *inet_opt; | |
1697 | ||
1698 | if (sk_acceptq_is_full(sk)) | |
1699 | goto exit_overflow; | |
1700 | ||
1701 | newsk = tcp_create_openreq_child(sk, req, skb); | |
1702 | if (!newsk) | |
1703 | goto exit_nonewsk; | |
1704 | ||
1705 | newsk->sk_gso_type = SKB_GSO_TCPV4; | |
1706 | inet_sk_rx_dst_set(newsk, skb); | |
1707 | ||
1708 | newtp = tcp_sk(newsk); | |
1709 | newinet = inet_sk(newsk); | |
1710 | ireq = inet_rsk(req); | |
1711 | newinet->inet_daddr = ireq->rmt_addr; | |
1712 | newinet->inet_rcv_saddr = ireq->loc_addr; | |
1713 | newinet->inet_saddr = ireq->loc_addr; | |
1714 | inet_opt = ireq->opt; | |
1715 | rcu_assign_pointer(newinet->inet_opt, inet_opt); | |
1716 | ireq->opt = NULL; | |
1717 | newinet->mc_index = inet_iif(skb); | |
1718 | newinet->mc_ttl = ip_hdr(skb)->ttl; | |
1719 | newinet->rcv_tos = ip_hdr(skb)->tos; | |
1720 | inet_csk(newsk)->icsk_ext_hdr_len = 0; | |
1721 | if (inet_opt) | |
1722 | inet_csk(newsk)->icsk_ext_hdr_len = inet_opt->opt.optlen; | |
1723 | newinet->inet_id = newtp->write_seq ^ jiffies; | |
1724 | ||
1725 | if (!dst) { | |
1726 | dst = inet_csk_route_child_sock(sk, newsk, req); | |
1727 | if (!dst) | |
1728 | goto put_and_exit; | |
1729 | } else { | |
1730 | /* syncookie case : see end of cookie_v4_check() */ | |
1731 | } | |
1732 | sk_setup_caps(newsk, dst); | |
1733 | ||
1734 | tcp_mtup_init(newsk); | |
1735 | tcp_sync_mss(newsk, dst_mtu(dst)); | |
1736 | newtp->advmss = dst_metric_advmss(dst); | |
1737 | if (tcp_sk(sk)->rx_opt.user_mss && | |
1738 | tcp_sk(sk)->rx_opt.user_mss < newtp->advmss) | |
1739 | newtp->advmss = tcp_sk(sk)->rx_opt.user_mss; | |
1740 | ||
1741 | tcp_initialize_rcv_mss(newsk); | |
1742 | tcp_synack_rtt_meas(newsk, req); | |
1743 | newtp->total_retrans = req->num_retrans; | |
1744 | ||
1745 | #ifdef CONFIG_TCP_MD5SIG | |
1746 | /* Copy over the MD5 key from the original socket */ | |
1747 | key = tcp_md5_do_lookup(sk, (union tcp_md5_addr *)&newinet->inet_daddr, | |
1748 | AF_INET); | |
1749 | if (key != NULL) { | |
1750 | /* | |
1751 | * We're using one, so create a matching key | |
1752 | * on the newsk structure. If we fail to get | |
1753 | * memory, then we end up not copying the key | |
1754 | * across. Shucks. | |
1755 | */ | |
1756 | tcp_md5_do_add(newsk, (union tcp_md5_addr *)&newinet->inet_daddr, | |
1757 | AF_INET, key->key, key->keylen, GFP_ATOMIC); | |
1758 | sk_nocaps_add(newsk, NETIF_F_GSO_MASK); | |
1759 | } | |
1760 | #endif | |
1761 | ||
1762 | if (__inet_inherit_port(sk, newsk) < 0) | |
1763 | goto put_and_exit; | |
1764 | __inet_hash_nolisten(newsk, NULL); | |
1765 | ||
1766 | return newsk; | |
1767 | ||
1768 | exit_overflow: | |
1769 | NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENOVERFLOWS); | |
1770 | exit_nonewsk: | |
1771 | dst_release(dst); | |
1772 | exit: | |
1773 | NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_LISTENDROPS); | |
1774 | return NULL; | |
1775 | put_and_exit: | |
1776 | inet_csk_prepare_forced_close(newsk); | |
1777 | tcp_done(newsk); | |
1778 | goto exit; | |
1779 | } | |
1780 | EXPORT_SYMBOL(tcp_v4_syn_recv_sock); | |
1781 | ||
1782 | static struct sock *tcp_v4_hnd_req(struct sock *sk, struct sk_buff *skb) | |
1783 | { | |
1784 | struct tcphdr *th = tcp_hdr(skb); | |
1785 | const struct iphdr *iph = ip_hdr(skb); | |
1786 | struct sock *nsk; | |
1787 | struct request_sock **prev; | |
1788 | /* Find possible connection requests. */ | |
1789 | struct request_sock *req = inet_csk_search_req(sk, &prev, th->source, | |
1790 | iph->saddr, iph->daddr); | |
1791 | if (req) | |
1792 | return tcp_check_req(sk, skb, req, prev, false); | |
1793 | ||
1794 | nsk = inet_lookup_established(sock_net(sk), &tcp_hashinfo, iph->saddr, | |
1795 | th->source, iph->daddr, th->dest, inet_iif(skb)); | |
1796 | ||
1797 | if (nsk) { | |
1798 | if (nsk->sk_state != TCP_TIME_WAIT) { | |
1799 | bh_lock_sock(nsk); | |
1800 | return nsk; | |
1801 | } | |
1802 | inet_twsk_put(inet_twsk(nsk)); | |
1803 | return NULL; | |
1804 | } | |
1805 | ||
1806 | #ifdef CONFIG_SYN_COOKIES | |
1807 | if (!th->syn) | |
1808 | sk = cookie_v4_check(sk, skb, &(IPCB(skb)->opt)); | |
1809 | #endif | |
1810 | return sk; | |
1811 | } | |
1812 | ||
1813 | static __sum16 tcp_v4_checksum_init(struct sk_buff *skb) | |
1814 | { | |
1815 | const struct iphdr *iph = ip_hdr(skb); | |
1816 | ||
1817 | if (skb->ip_summed == CHECKSUM_COMPLETE) { | |
1818 | if (!tcp_v4_check(skb->len, iph->saddr, | |
1819 | iph->daddr, skb->csum)) { | |
1820 | skb->ip_summed = CHECKSUM_UNNECESSARY; | |
1821 | return 0; | |
1822 | } | |
1823 | } | |
1824 | ||
1825 | skb->csum = csum_tcpudp_nofold(iph->saddr, iph->daddr, | |
1826 | skb->len, IPPROTO_TCP, 0); | |
1827 | ||
1828 | if (skb->len <= 76) { | |
1829 | return __skb_checksum_complete(skb); | |
1830 | } | |
1831 | return 0; | |
1832 | } | |
1833 | ||
1834 | ||
1835 | /* The socket must have it's spinlock held when we get | |
1836 | * here. | |
1837 | * | |
1838 | * We have a potential double-lock case here, so even when | |
1839 | * doing backlog processing we use the BH locking scheme. | |
1840 | * This is because we cannot sleep with the original spinlock | |
1841 | * held. | |
1842 | */ | |
1843 | int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb) | |
1844 | { | |
1845 | struct sock *rsk; | |
1846 | #ifdef CONFIG_TCP_MD5SIG | |
1847 | /* | |
1848 | * We really want to reject the packet as early as possible | |
1849 | * if: | |
1850 | * o We're expecting an MD5'd packet and this is no MD5 tcp option | |
1851 | * o There is an MD5 option and we're not expecting one | |
1852 | */ | |
1853 | if (tcp_v4_inbound_md5_hash(sk, skb)) | |
1854 | goto discard; | |
1855 | #endif | |
1856 | ||
1857 | if (sk->sk_state == TCP_ESTABLISHED) { /* Fast path */ | |
1858 | struct dst_entry *dst = sk->sk_rx_dst; | |
1859 | ||
1860 | sock_rps_save_rxhash(sk, skb); | |
1861 | if (dst) { | |
1862 | if (inet_sk(sk)->rx_dst_ifindex != skb->skb_iif || | |
1863 | dst->ops->check(dst, 0) == NULL) { | |
1864 | dst_release(dst); | |
1865 | sk->sk_rx_dst = NULL; | |
1866 | } | |
1867 | } | |
1868 | if (tcp_rcv_established(sk, skb, tcp_hdr(skb), skb->len)) { | |
1869 | rsk = sk; | |
1870 | goto reset; | |
1871 | } | |
1872 | return 0; | |
1873 | } | |
1874 | ||
1875 | if (skb->len < tcp_hdrlen(skb) || tcp_checksum_complete(skb)) | |
1876 | goto csum_err; | |
1877 | ||
1878 | if (sk->sk_state == TCP_LISTEN) { | |
1879 | struct sock *nsk = tcp_v4_hnd_req(sk, skb); | |
1880 | if (!nsk) | |
1881 | goto discard; | |
1882 | ||
1883 | if (nsk != sk) { | |
1884 | sock_rps_save_rxhash(nsk, skb); | |
1885 | if (tcp_child_process(sk, nsk, skb)) { | |
1886 | rsk = nsk; | |
1887 | goto reset; | |
1888 | } | |
1889 | return 0; | |
1890 | } | |
1891 | } else | |
1892 | sock_rps_save_rxhash(sk, skb); | |
1893 | ||
1894 | if (tcp_rcv_state_process(sk, skb, tcp_hdr(skb), skb->len)) { | |
1895 | rsk = sk; | |
1896 | goto reset; | |
1897 | } | |
1898 | return 0; | |
1899 | ||
1900 | reset: | |
1901 | tcp_v4_send_reset(rsk, skb); | |
1902 | discard: | |
1903 | kfree_skb(skb); | |
1904 | /* Be careful here. If this function gets more complicated and | |
1905 | * gcc suffers from register pressure on the x86, sk (in %ebx) | |
1906 | * might be destroyed here. This current version compiles correctly, | |
1907 | * but you have been warned. | |
1908 | */ | |
1909 | return 0; | |
1910 | ||
1911 | csum_err: | |
1912 | TCP_INC_STATS_BH(sock_net(sk), TCP_MIB_INERRS); | |
1913 | goto discard; | |
1914 | } | |
1915 | EXPORT_SYMBOL(tcp_v4_do_rcv); | |
1916 | ||
1917 | void tcp_v4_early_demux(struct sk_buff *skb) | |
1918 | { | |
1919 | const struct iphdr *iph; | |
1920 | const struct tcphdr *th; | |
1921 | struct sock *sk; | |
1922 | ||
1923 | if (skb->pkt_type != PACKET_HOST) | |
1924 | return; | |
1925 | ||
1926 | if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct tcphdr))) | |
1927 | return; | |
1928 | ||
1929 | iph = ip_hdr(skb); | |
1930 | th = tcp_hdr(skb); | |
1931 | ||
1932 | if (th->doff < sizeof(struct tcphdr) / 4) | |
1933 | return; | |
1934 | ||
1935 | sk = __inet_lookup_established(dev_net(skb->dev), &tcp_hashinfo, | |
1936 | iph->saddr, th->source, | |
1937 | iph->daddr, ntohs(th->dest), | |
1938 | skb->skb_iif); | |
1939 | if (sk) { | |
1940 | skb->sk = sk; | |
1941 | skb->destructor = sock_edemux; | |
1942 | if (sk->sk_state != TCP_TIME_WAIT) { | |
1943 | struct dst_entry *dst = sk->sk_rx_dst; | |
1944 | ||
1945 | if (dst) | |
1946 | dst = dst_check(dst, 0); | |
1947 | if (dst && | |
1948 | inet_sk(sk)->rx_dst_ifindex == skb->skb_iif) | |
1949 | skb_dst_set_noref(skb, dst); | |
1950 | } | |
1951 | } | |
1952 | } | |
1953 | ||
1954 | /* | |
1955 | * From tcp_input.c | |
1956 | */ | |
1957 | ||
1958 | int tcp_v4_rcv(struct sk_buff *skb) | |
1959 | { | |
1960 | const struct iphdr *iph; | |
1961 | const struct tcphdr *th; | |
1962 | struct sock *sk; | |
1963 | int ret; | |
1964 | struct net *net = dev_net(skb->dev); | |
1965 | ||
1966 | if (skb->pkt_type != PACKET_HOST) | |
1967 | goto discard_it; | |
1968 | ||
1969 | /* Count it even if it's bad */ | |
1970 | TCP_INC_STATS_BH(net, TCP_MIB_INSEGS); | |
1971 | ||
1972 | if (!pskb_may_pull(skb, sizeof(struct tcphdr))) | |
1973 | goto discard_it; | |
1974 | ||
1975 | th = tcp_hdr(skb); | |
1976 | ||
1977 | if (th->doff < sizeof(struct tcphdr) / 4) | |
1978 | goto bad_packet; | |
1979 | if (!pskb_may_pull(skb, th->doff * 4)) | |
1980 | goto discard_it; | |
1981 | ||
1982 | /* An explanation is required here, I think. | |
1983 | * Packet length and doff are validated by header prediction, | |
1984 | * provided case of th->doff==0 is eliminated. | |
1985 | * So, we defer the checks. */ | |
1986 | if (!skb_csum_unnecessary(skb) && tcp_v4_checksum_init(skb)) | |
1987 | goto bad_packet; | |
1988 | ||
1989 | th = tcp_hdr(skb); | |
1990 | iph = ip_hdr(skb); | |
1991 | TCP_SKB_CB(skb)->seq = ntohl(th->seq); | |
1992 | TCP_SKB_CB(skb)->end_seq = (TCP_SKB_CB(skb)->seq + th->syn + th->fin + | |
1993 | skb->len - th->doff * 4); | |
1994 | TCP_SKB_CB(skb)->ack_seq = ntohl(th->ack_seq); | |
1995 | TCP_SKB_CB(skb)->when = 0; | |
1996 | TCP_SKB_CB(skb)->ip_dsfield = ipv4_get_dsfield(iph); | |
1997 | TCP_SKB_CB(skb)->sacked = 0; | |
1998 | ||
1999 | sk = __inet_lookup_skb(&tcp_hashinfo, skb, th->source, th->dest); | |
2000 | if (!sk) | |
2001 | goto no_tcp_socket; | |
2002 | ||
2003 | process: | |
2004 | if (sk->sk_state == TCP_TIME_WAIT) | |
2005 | goto do_time_wait; | |
2006 | ||
2007 | if (unlikely(iph->ttl < inet_sk(sk)->min_ttl)) { | |
2008 | NET_INC_STATS_BH(net, LINUX_MIB_TCPMINTTLDROP); | |
2009 | goto discard_and_relse; | |
2010 | } | |
2011 | ||
2012 | if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb)) | |
2013 | goto discard_and_relse; | |
2014 | nf_reset(skb); | |
2015 | ||
2016 | if (sk_filter(sk, skb)) | |
2017 | goto discard_and_relse; | |
2018 | ||
2019 | skb->dev = NULL; | |
2020 | ||
2021 | bh_lock_sock_nested(sk); | |
2022 | ret = 0; | |
2023 | if (!sock_owned_by_user(sk)) { | |
2024 | #ifdef CONFIG_NET_DMA | |
2025 | struct tcp_sock *tp = tcp_sk(sk); | |
2026 | if (!tp->ucopy.dma_chan && tp->ucopy.pinned_list) | |
2027 | tp->ucopy.dma_chan = net_dma_find_channel(); | |
2028 | if (tp->ucopy.dma_chan) | |
2029 | ret = tcp_v4_do_rcv(sk, skb); | |
2030 | else | |
2031 | #endif | |
2032 | { | |
2033 | if (!tcp_prequeue(sk, skb)) | |
2034 | ret = tcp_v4_do_rcv(sk, skb); | |
2035 | } | |
2036 | } else if (unlikely(sk_add_backlog(sk, skb, | |
2037 | sk->sk_rcvbuf + sk->sk_sndbuf))) { | |
2038 | bh_unlock_sock(sk); | |
2039 | NET_INC_STATS_BH(net, LINUX_MIB_TCPBACKLOGDROP); | |
2040 | goto discard_and_relse; | |
2041 | } | |
2042 | bh_unlock_sock(sk); | |
2043 | ||
2044 | sock_put(sk); | |
2045 | ||
2046 | return ret; | |
2047 | ||
2048 | no_tcp_socket: | |
2049 | if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) | |
2050 | goto discard_it; | |
2051 | ||
2052 | if (skb->len < (th->doff << 2) || tcp_checksum_complete(skb)) { | |
2053 | bad_packet: | |
2054 | TCP_INC_STATS_BH(net, TCP_MIB_INERRS); | |
2055 | } else { | |
2056 | tcp_v4_send_reset(NULL, skb); | |
2057 | } | |
2058 | ||
2059 | discard_it: | |
2060 | /* Discard frame. */ | |
2061 | kfree_skb(skb); | |
2062 | return 0; | |
2063 | ||
2064 | discard_and_relse: | |
2065 | sock_put(sk); | |
2066 | goto discard_it; | |
2067 | ||
2068 | do_time_wait: | |
2069 | if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) { | |
2070 | inet_twsk_put(inet_twsk(sk)); | |
2071 | goto discard_it; | |
2072 | } | |
2073 | ||
2074 | if (skb->len < (th->doff << 2) || tcp_checksum_complete(skb)) { | |
2075 | TCP_INC_STATS_BH(net, TCP_MIB_INERRS); | |
2076 | inet_twsk_put(inet_twsk(sk)); | |
2077 | goto discard_it; | |
2078 | } | |
2079 | switch (tcp_timewait_state_process(inet_twsk(sk), skb, th)) { | |
2080 | case TCP_TW_SYN: { | |
2081 | struct sock *sk2 = inet_lookup_listener(dev_net(skb->dev), | |
2082 | &tcp_hashinfo, | |
2083 | iph->saddr, th->source, | |
2084 | iph->daddr, th->dest, | |
2085 | inet_iif(skb)); | |
2086 | if (sk2) { | |
2087 | inet_twsk_deschedule(inet_twsk(sk), &tcp_death_row); | |
2088 | inet_twsk_put(inet_twsk(sk)); | |
2089 | sk = sk2; | |
2090 | goto process; | |
2091 | } | |
2092 | /* Fall through to ACK */ | |
2093 | } | |
2094 | case TCP_TW_ACK: | |
2095 | tcp_v4_timewait_ack(sk, skb); | |
2096 | break; | |
2097 | case TCP_TW_RST: | |
2098 | goto no_tcp_socket; | |
2099 | case TCP_TW_SUCCESS:; | |
2100 | } | |
2101 | goto discard_it; | |
2102 | } | |
2103 | ||
2104 | static struct timewait_sock_ops tcp_timewait_sock_ops = { | |
2105 | .twsk_obj_size = sizeof(struct tcp_timewait_sock), | |
2106 | .twsk_unique = tcp_twsk_unique, | |
2107 | .twsk_destructor= tcp_twsk_destructor, | |
2108 | }; | |
2109 | ||
2110 | void inet_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb) | |
2111 | { | |
2112 | struct dst_entry *dst = skb_dst(skb); | |
2113 | ||
2114 | dst_hold(dst); | |
2115 | sk->sk_rx_dst = dst; | |
2116 | inet_sk(sk)->rx_dst_ifindex = skb->skb_iif; | |
2117 | } | |
2118 | EXPORT_SYMBOL(inet_sk_rx_dst_set); | |
2119 | ||
2120 | const struct inet_connection_sock_af_ops ipv4_specific = { | |
2121 | .queue_xmit = ip_queue_xmit, | |
2122 | .send_check = tcp_v4_send_check, | |
2123 | .rebuild_header = inet_sk_rebuild_header, | |
2124 | .sk_rx_dst_set = inet_sk_rx_dst_set, | |
2125 | .conn_request = tcp_v4_conn_request, | |
2126 | .syn_recv_sock = tcp_v4_syn_recv_sock, | |
2127 | .net_header_len = sizeof(struct iphdr), | |
2128 | .setsockopt = ip_setsockopt, | |
2129 | .getsockopt = ip_getsockopt, | |
2130 | .addr2sockaddr = inet_csk_addr2sockaddr, | |
2131 | .sockaddr_len = sizeof(struct sockaddr_in), | |
2132 | .bind_conflict = inet_csk_bind_conflict, | |
2133 | #ifdef CONFIG_COMPAT | |
2134 | .compat_setsockopt = compat_ip_setsockopt, | |
2135 | .compat_getsockopt = compat_ip_getsockopt, | |
2136 | #endif | |
2137 | }; | |
2138 | EXPORT_SYMBOL(ipv4_specific); | |
2139 | ||
2140 | #ifdef CONFIG_TCP_MD5SIG | |
2141 | static const struct tcp_sock_af_ops tcp_sock_ipv4_specific = { | |
2142 | .md5_lookup = tcp_v4_md5_lookup, | |
2143 | .calc_md5_hash = tcp_v4_md5_hash_skb, | |
2144 | .md5_parse = tcp_v4_parse_md5_keys, | |
2145 | }; | |
2146 | #endif | |
2147 | ||
2148 | /* NOTE: A lot of things set to zero explicitly by call to | |
2149 | * sk_alloc() so need not be done here. | |
2150 | */ | |
2151 | static int tcp_v4_init_sock(struct sock *sk) | |
2152 | { | |
2153 | struct inet_connection_sock *icsk = inet_csk(sk); | |
2154 | ||
2155 | tcp_init_sock(sk); | |
2156 | ||
2157 | icsk->icsk_af_ops = &ipv4_specific; | |
2158 | ||
2159 | #ifdef CONFIG_TCP_MD5SIG | |
2160 | tcp_sk(sk)->af_specific = &tcp_sock_ipv4_specific; | |
2161 | #endif | |
2162 | ||
2163 | return 0; | |
2164 | } | |
2165 | ||
2166 | void tcp_v4_destroy_sock(struct sock *sk) | |
2167 | { | |
2168 | struct tcp_sock *tp = tcp_sk(sk); | |
2169 | ||
2170 | tcp_clear_xmit_timers(sk); | |
2171 | ||
2172 | tcp_cleanup_congestion_control(sk); | |
2173 | ||
2174 | /* Cleanup up the write buffer. */ | |
2175 | tcp_write_queue_purge(sk); | |
2176 | ||
2177 | /* Cleans up our, hopefully empty, out_of_order_queue. */ | |
2178 | __skb_queue_purge(&tp->out_of_order_queue); | |
2179 | ||
2180 | #ifdef CONFIG_TCP_MD5SIG | |
2181 | /* Clean up the MD5 key list, if any */ | |
2182 | if (tp->md5sig_info) { | |
2183 | tcp_clear_md5_list(sk); | |
2184 | kfree_rcu(tp->md5sig_info, rcu); | |
2185 | tp->md5sig_info = NULL; | |
2186 | } | |
2187 | #endif | |
2188 | ||
2189 | #ifdef CONFIG_NET_DMA | |
2190 | /* Cleans up our sk_async_wait_queue */ | |
2191 | __skb_queue_purge(&sk->sk_async_wait_queue); | |
2192 | #endif | |
2193 | ||
2194 | /* Clean prequeue, it must be empty really */ | |
2195 | __skb_queue_purge(&tp->ucopy.prequeue); | |
2196 | ||
2197 | /* Clean up a referenced TCP bind bucket. */ | |
2198 | if (inet_csk(sk)->icsk_bind_hash) | |
2199 | inet_put_port(sk); | |
2200 | ||
2201 | /* TCP Cookie Transactions */ | |
2202 | if (tp->cookie_values != NULL) { | |
2203 | kref_put(&tp->cookie_values->kref, | |
2204 | tcp_cookie_values_release); | |
2205 | tp->cookie_values = NULL; | |
2206 | } | |
2207 | BUG_ON(tp->fastopen_rsk != NULL); | |
2208 | ||
2209 | /* If socket is aborted during connect operation */ | |
2210 | tcp_free_fastopen_req(tp); | |
2211 | ||
2212 | sk_sockets_allocated_dec(sk); | |
2213 | sock_release_memcg(sk); | |
2214 | } | |
2215 | EXPORT_SYMBOL(tcp_v4_destroy_sock); | |
2216 | ||
2217 | #ifdef CONFIG_PROC_FS | |
2218 | /* Proc filesystem TCP sock list dumping. */ | |
2219 | ||
2220 | static inline struct inet_timewait_sock *tw_head(struct hlist_nulls_head *head) | |
2221 | { | |
2222 | return hlist_nulls_empty(head) ? NULL : | |
2223 | list_entry(head->first, struct inet_timewait_sock, tw_node); | |
2224 | } | |
2225 | ||
2226 | static inline struct inet_timewait_sock *tw_next(struct inet_timewait_sock *tw) | |
2227 | { | |
2228 | return !is_a_nulls(tw->tw_node.next) ? | |
2229 | hlist_nulls_entry(tw->tw_node.next, typeof(*tw), tw_node) : NULL; | |
2230 | } | |
2231 | ||
2232 | /* | |
2233 | * Get next listener socket follow cur. If cur is NULL, get first socket | |
2234 | * starting from bucket given in st->bucket; when st->bucket is zero the | |
2235 | * very first socket in the hash table is returned. | |
2236 | */ | |
2237 | static void *listening_get_next(struct seq_file *seq, void *cur) | |
2238 | { | |
2239 | struct inet_connection_sock *icsk; | |
2240 | struct hlist_nulls_node *node; | |
2241 | struct sock *sk = cur; | |
2242 | struct inet_listen_hashbucket *ilb; | |
2243 | struct tcp_iter_state *st = seq->private; | |
2244 | struct net *net = seq_file_net(seq); | |
2245 | ||
2246 | if (!sk) { | |
2247 | ilb = &tcp_hashinfo.listening_hash[st->bucket]; | |
2248 | spin_lock_bh(&ilb->lock); | |
2249 | sk = sk_nulls_head(&ilb->head); | |
2250 | st->offset = 0; | |
2251 | goto get_sk; | |
2252 | } | |
2253 | ilb = &tcp_hashinfo.listening_hash[st->bucket]; | |
2254 | ++st->num; | |
2255 | ++st->offset; | |
2256 | ||
2257 | if (st->state == TCP_SEQ_STATE_OPENREQ) { | |
2258 | struct request_sock *req = cur; | |
2259 | ||
2260 | icsk = inet_csk(st->syn_wait_sk); | |
2261 | req = req->dl_next; | |
2262 | while (1) { | |
2263 | while (req) { | |
2264 | if (req->rsk_ops->family == st->family) { | |
2265 | cur = req; | |
2266 | goto out; | |
2267 | } | |
2268 | req = req->dl_next; | |
2269 | } | |
2270 | if (++st->sbucket >= icsk->icsk_accept_queue.listen_opt->nr_table_entries) | |
2271 | break; | |
2272 | get_req: | |
2273 | req = icsk->icsk_accept_queue.listen_opt->syn_table[st->sbucket]; | |
2274 | } | |
2275 | sk = sk_nulls_next(st->syn_wait_sk); | |
2276 | st->state = TCP_SEQ_STATE_LISTENING; | |
2277 | read_unlock_bh(&icsk->icsk_accept_queue.syn_wait_lock); | |
2278 | } else { | |
2279 | icsk = inet_csk(sk); | |
2280 | read_lock_bh(&icsk->icsk_accept_queue.syn_wait_lock); | |
2281 | if (reqsk_queue_len(&icsk->icsk_accept_queue)) | |
2282 | goto start_req; | |
2283 | read_unlock_bh(&icsk->icsk_accept_queue.syn_wait_lock); | |
2284 | sk = sk_nulls_next(sk); | |
2285 | } | |
2286 | get_sk: | |
2287 | sk_nulls_for_each_from(sk, node) { | |
2288 | if (!net_eq(sock_net(sk), net)) | |
2289 | continue; | |
2290 | if (sk->sk_family == st->family) { | |
2291 | cur = sk; | |
2292 | goto out; | |
2293 | } | |
2294 | icsk = inet_csk(sk); | |
2295 | read_lock_bh(&icsk->icsk_accept_queue.syn_wait_lock); | |
2296 | if (reqsk_queue_len(&icsk->icsk_accept_queue)) { | |
2297 | start_req: | |
2298 | st->uid = sock_i_uid(sk); | |
2299 | st->syn_wait_sk = sk; | |
2300 | st->state = TCP_SEQ_STATE_OPENREQ; | |
2301 | st->sbucket = 0; | |
2302 | goto get_req; | |
2303 | } | |
2304 | read_unlock_bh(&icsk->icsk_accept_queue.syn_wait_lock); | |
2305 | } | |
2306 | spin_unlock_bh(&ilb->lock); | |
2307 | st->offset = 0; | |
2308 | if (++st->bucket < INET_LHTABLE_SIZE) { | |
2309 | ilb = &tcp_hashinfo.listening_hash[st->bucket]; | |
2310 | spin_lock_bh(&ilb->lock); | |
2311 | sk = sk_nulls_head(&ilb->head); | |
2312 | goto get_sk; | |
2313 | } | |
2314 | cur = NULL; | |
2315 | out: | |
2316 | return cur; | |
2317 | } | |
2318 | ||
2319 | static void *listening_get_idx(struct seq_file *seq, loff_t *pos) | |
2320 | { | |
2321 | struct tcp_iter_state *st = seq->private; | |
2322 | void *rc; | |
2323 | ||
2324 | st->bucket = 0; | |
2325 | st->offset = 0; | |
2326 | rc = listening_get_next(seq, NULL); | |
2327 | ||
2328 | while (rc && *pos) { | |
2329 | rc = listening_get_next(seq, rc); | |
2330 | --*pos; | |
2331 | } | |
2332 | return rc; | |
2333 | } | |
2334 | ||
2335 | static inline bool empty_bucket(struct tcp_iter_state *st) | |
2336 | { | |
2337 | return hlist_nulls_empty(&tcp_hashinfo.ehash[st->bucket].chain) && | |
2338 | hlist_nulls_empty(&tcp_hashinfo.ehash[st->bucket].twchain); | |
2339 | } | |
2340 | ||
2341 | /* | |
2342 | * Get first established socket starting from bucket given in st->bucket. | |
2343 | * If st->bucket is zero, the very first socket in the hash is returned. | |
2344 | */ | |
2345 | static void *established_get_first(struct seq_file *seq) | |
2346 | { | |
2347 | struct tcp_iter_state *st = seq->private; | |
2348 | struct net *net = seq_file_net(seq); | |
2349 | void *rc = NULL; | |
2350 | ||
2351 | st->offset = 0; | |
2352 | for (; st->bucket <= tcp_hashinfo.ehash_mask; ++st->bucket) { | |
2353 | struct sock *sk; | |
2354 | struct hlist_nulls_node *node; | |
2355 | struct inet_timewait_sock *tw; | |
2356 | spinlock_t *lock = inet_ehash_lockp(&tcp_hashinfo, st->bucket); | |
2357 | ||
2358 | /* Lockless fast path for the common case of empty buckets */ | |
2359 | if (empty_bucket(st)) | |
2360 | continue; | |
2361 | ||
2362 | spin_lock_bh(lock); | |
2363 | sk_nulls_for_each(sk, node, &tcp_hashinfo.ehash[st->bucket].chain) { | |
2364 | if (sk->sk_family != st->family || | |
2365 | !net_eq(sock_net(sk), net)) { | |
2366 | continue; | |
2367 | } | |
2368 | rc = sk; | |
2369 | goto out; | |
2370 | } | |
2371 | st->state = TCP_SEQ_STATE_TIME_WAIT; | |
2372 | inet_twsk_for_each(tw, node, | |
2373 | &tcp_hashinfo.ehash[st->bucket].twchain) { | |
2374 | if (tw->tw_family != st->family || | |
2375 | !net_eq(twsk_net(tw), net)) { | |
2376 | continue; | |
2377 | } | |
2378 | rc = tw; | |
2379 | goto out; | |
2380 | } | |
2381 | spin_unlock_bh(lock); | |
2382 | st->state = TCP_SEQ_STATE_ESTABLISHED; | |
2383 | } | |
2384 | out: | |
2385 | return rc; | |
2386 | } | |
2387 | ||
2388 | static void *established_get_next(struct seq_file *seq, void *cur) | |
2389 | { | |
2390 | struct sock *sk = cur; | |
2391 | struct inet_timewait_sock *tw; | |
2392 | struct hlist_nulls_node *node; | |
2393 | struct tcp_iter_state *st = seq->private; | |
2394 | struct net *net = seq_file_net(seq); | |
2395 | ||
2396 | ++st->num; | |
2397 | ++st->offset; | |
2398 | ||
2399 | if (st->state == TCP_SEQ_STATE_TIME_WAIT) { | |
2400 | tw = cur; | |
2401 | tw = tw_next(tw); | |
2402 | get_tw: | |
2403 | while (tw && (tw->tw_family != st->family || !net_eq(twsk_net(tw), net))) { | |
2404 | tw = tw_next(tw); | |
2405 | } | |
2406 | if (tw) { | |
2407 | cur = tw; | |
2408 | goto out; | |
2409 | } | |
2410 | spin_unlock_bh(inet_ehash_lockp(&tcp_hashinfo, st->bucket)); | |
2411 | st->state = TCP_SEQ_STATE_ESTABLISHED; | |
2412 | ||
2413 | /* Look for next non empty bucket */ | |
2414 | st->offset = 0; | |
2415 | while (++st->bucket <= tcp_hashinfo.ehash_mask && | |
2416 | empty_bucket(st)) | |
2417 | ; | |
2418 | if (st->bucket > tcp_hashinfo.ehash_mask) | |
2419 | return NULL; | |
2420 | ||
2421 | spin_lock_bh(inet_ehash_lockp(&tcp_hashinfo, st->bucket)); | |
2422 | sk = sk_nulls_head(&tcp_hashinfo.ehash[st->bucket].chain); | |
2423 | } else | |
2424 | sk = sk_nulls_next(sk); | |
2425 | ||
2426 | sk_nulls_for_each_from(sk, node) { | |
2427 | if (sk->sk_family == st->family && net_eq(sock_net(sk), net)) | |
2428 | goto found; | |
2429 | } | |
2430 | ||
2431 | st->state = TCP_SEQ_STATE_TIME_WAIT; | |
2432 | tw = tw_head(&tcp_hashinfo.ehash[st->bucket].twchain); | |
2433 | goto get_tw; | |
2434 | found: | |
2435 | cur = sk; | |
2436 | out: | |
2437 | return cur; | |
2438 | } | |
2439 | ||
2440 | static void *established_get_idx(struct seq_file *seq, loff_t pos) | |
2441 | { | |
2442 | struct tcp_iter_state *st = seq->private; | |
2443 | void *rc; | |
2444 | ||
2445 | st->bucket = 0; | |
2446 | rc = established_get_first(seq); | |
2447 | ||
2448 | while (rc && pos) { | |
2449 | rc = established_get_next(seq, rc); | |
2450 | --pos; | |
2451 | } | |
2452 | return rc; | |
2453 | } | |
2454 | ||
2455 | static void *tcp_get_idx(struct seq_file *seq, loff_t pos) | |
2456 | { | |
2457 | void *rc; | |
2458 | struct tcp_iter_state *st = seq->private; | |
2459 | ||
2460 | st->state = TCP_SEQ_STATE_LISTENING; | |
2461 | rc = listening_get_idx(seq, &pos); | |
2462 | ||
2463 | if (!rc) { | |
2464 | st->state = TCP_SEQ_STATE_ESTABLISHED; | |
2465 | rc = established_get_idx(seq, pos); | |
2466 | } | |
2467 | ||
2468 | return rc; | |
2469 | } | |
2470 | ||
2471 | static void *tcp_seek_last_pos(struct seq_file *seq) | |
2472 | { | |
2473 | struct tcp_iter_state *st = seq->private; | |
2474 | int offset = st->offset; | |
2475 | int orig_num = st->num; | |
2476 | void *rc = NULL; | |
2477 | ||
2478 | switch (st->state) { | |
2479 | case TCP_SEQ_STATE_OPENREQ: | |
2480 | case TCP_SEQ_STATE_LISTENING: | |
2481 | if (st->bucket >= INET_LHTABLE_SIZE) | |
2482 | break; | |
2483 | st->state = TCP_SEQ_STATE_LISTENING; | |
2484 | rc = listening_get_next(seq, NULL); | |
2485 | while (offset-- && rc) | |
2486 | rc = listening_get_next(seq, rc); | |
2487 | if (rc) | |
2488 | break; | |
2489 | st->bucket = 0; | |
2490 | /* Fallthrough */ | |
2491 | case TCP_SEQ_STATE_ESTABLISHED: | |
2492 | case TCP_SEQ_STATE_TIME_WAIT: | |
2493 | st->state = TCP_SEQ_STATE_ESTABLISHED; | |
2494 | if (st->bucket > tcp_hashinfo.ehash_mask) | |
2495 | break; | |
2496 | rc = established_get_first(seq); | |
2497 | while (offset-- && rc) | |
2498 | rc = established_get_next(seq, rc); | |
2499 | } | |
2500 | ||
2501 | st->num = orig_num; | |
2502 | ||
2503 | return rc; | |
2504 | } | |
2505 | ||
2506 | static void *tcp_seq_start(struct seq_file *seq, loff_t *pos) | |
2507 | { | |
2508 | struct tcp_iter_state *st = seq->private; | |
2509 | void *rc; | |
2510 | ||
2511 | if (*pos && *pos == st->last_pos) { | |
2512 | rc = tcp_seek_last_pos(seq); | |
2513 | if (rc) | |
2514 | goto out; | |
2515 | } | |
2516 | ||
2517 | st->state = TCP_SEQ_STATE_LISTENING; | |
2518 | st->num = 0; | |
2519 | st->bucket = 0; | |
2520 | st->offset = 0; | |
2521 | rc = *pos ? tcp_get_idx(seq, *pos - 1) : SEQ_START_TOKEN; | |
2522 | ||
2523 | out: | |
2524 | st->last_pos = *pos; | |
2525 | return rc; | |
2526 | } | |
2527 | ||
2528 | static void *tcp_seq_next(struct seq_file *seq, void *v, loff_t *pos) | |
2529 | { | |
2530 | struct tcp_iter_state *st = seq->private; | |
2531 | void *rc = NULL; | |
2532 | ||
2533 | if (v == SEQ_START_TOKEN) { | |
2534 | rc = tcp_get_idx(seq, 0); | |
2535 | goto out; | |
2536 | } | |
2537 | ||
2538 | switch (st->state) { | |
2539 | case TCP_SEQ_STATE_OPENREQ: | |
2540 | case TCP_SEQ_STATE_LISTENING: | |
2541 | rc = listening_get_next(seq, v); | |
2542 | if (!rc) { | |
2543 | st->state = TCP_SEQ_STATE_ESTABLISHED; | |
2544 | st->bucket = 0; | |
2545 | st->offset = 0; | |
2546 | rc = established_get_first(seq); | |
2547 | } | |
2548 | break; | |
2549 | case TCP_SEQ_STATE_ESTABLISHED: | |
2550 | case TCP_SEQ_STATE_TIME_WAIT: | |
2551 | rc = established_get_next(seq, v); | |
2552 | break; | |
2553 | } | |
2554 | out: | |
2555 | ++*pos; | |
2556 | st->last_pos = *pos; | |
2557 | return rc; | |
2558 | } | |
2559 | ||
2560 | static void tcp_seq_stop(struct seq_file *seq, void *v) | |
2561 | { | |
2562 | struct tcp_iter_state *st = seq->private; | |
2563 | ||
2564 | switch (st->state) { | |
2565 | case TCP_SEQ_STATE_OPENREQ: | |
2566 | if (v) { | |
2567 | struct inet_connection_sock *icsk = inet_csk(st->syn_wait_sk); | |
2568 | read_unlock_bh(&icsk->icsk_accept_queue.syn_wait_lock); | |
2569 | } | |
2570 | case TCP_SEQ_STATE_LISTENING: | |
2571 | if (v != SEQ_START_TOKEN) | |
2572 | spin_unlock_bh(&tcp_hashinfo.listening_hash[st->bucket].lock); | |
2573 | break; | |
2574 | case TCP_SEQ_STATE_TIME_WAIT: | |
2575 | case TCP_SEQ_STATE_ESTABLISHED: | |
2576 | if (v) | |
2577 | spin_unlock_bh(inet_ehash_lockp(&tcp_hashinfo, st->bucket)); | |
2578 | break; | |
2579 | } | |
2580 | } | |
2581 | ||
2582 | int tcp_seq_open(struct inode *inode, struct file *file) | |
2583 | { | |
2584 | struct tcp_seq_afinfo *afinfo = PDE(inode)->data; | |
2585 | struct tcp_iter_state *s; | |
2586 | int err; | |
2587 | ||
2588 | err = seq_open_net(inode, file, &afinfo->seq_ops, | |
2589 | sizeof(struct tcp_iter_state)); | |
2590 | if (err < 0) | |
2591 | return err; | |
2592 | ||
2593 | s = ((struct seq_file *)file->private_data)->private; | |
2594 | s->family = afinfo->family; | |
2595 | s->last_pos = 0; | |
2596 | return 0; | |
2597 | } | |
2598 | EXPORT_SYMBOL(tcp_seq_open); | |
2599 | ||
2600 | int tcp_proc_register(struct net *net, struct tcp_seq_afinfo *afinfo) | |
2601 | { | |
2602 | int rc = 0; | |
2603 | struct proc_dir_entry *p; | |
2604 | ||
2605 | afinfo->seq_ops.start = tcp_seq_start; | |
2606 | afinfo->seq_ops.next = tcp_seq_next; | |
2607 | afinfo->seq_ops.stop = tcp_seq_stop; | |
2608 | ||
2609 | p = proc_create_data(afinfo->name, S_IRUGO, net->proc_net, | |
2610 | afinfo->seq_fops, afinfo); | |
2611 | if (!p) | |
2612 | rc = -ENOMEM; | |
2613 | return rc; | |
2614 | } | |
2615 | EXPORT_SYMBOL(tcp_proc_register); | |
2616 | ||
2617 | void tcp_proc_unregister(struct net *net, struct tcp_seq_afinfo *afinfo) | |
2618 | { | |
2619 | remove_proc_entry(afinfo->name, net->proc_net); | |
2620 | } | |
2621 | EXPORT_SYMBOL(tcp_proc_unregister); | |
2622 | ||
2623 | static void get_openreq4(const struct sock *sk, const struct request_sock *req, | |
2624 | struct seq_file *f, int i, kuid_t uid, int *len) | |
2625 | { | |
2626 | const struct inet_request_sock *ireq = inet_rsk(req); | |
2627 | long delta = req->expires - jiffies; | |
2628 | ||
2629 | seq_printf(f, "%4d: %08X:%04X %08X:%04X" | |
2630 | " %02X %08X:%08X %02X:%08lX %08X %5d %8d %u %d %pK%n", | |
2631 | i, | |
2632 | ireq->loc_addr, | |
2633 | ntohs(inet_sk(sk)->inet_sport), | |
2634 | ireq->rmt_addr, | |
2635 | ntohs(ireq->rmt_port), | |
2636 | TCP_SYN_RECV, | |
2637 | 0, 0, /* could print option size, but that is af dependent. */ | |
2638 | 1, /* timers active (only the expire timer) */ | |
2639 | jiffies_delta_to_clock_t(delta), | |
2640 | req->num_timeout, | |
2641 | from_kuid_munged(seq_user_ns(f), uid), | |
2642 | 0, /* non standard timer */ | |
2643 | 0, /* open_requests have no inode */ | |
2644 | atomic_read(&sk->sk_refcnt), | |
2645 | req, | |
2646 | len); | |
2647 | } | |
2648 | ||
2649 | static void get_tcp4_sock(struct sock *sk, struct seq_file *f, int i, int *len) | |
2650 | { | |
2651 | int timer_active; | |
2652 | unsigned long timer_expires; | |
2653 | const struct tcp_sock *tp = tcp_sk(sk); | |
2654 | const struct inet_connection_sock *icsk = inet_csk(sk); | |
2655 | const struct inet_sock *inet = inet_sk(sk); | |
2656 | struct fastopen_queue *fastopenq = icsk->icsk_accept_queue.fastopenq; | |
2657 | __be32 dest = inet->inet_daddr; | |
2658 | __be32 src = inet->inet_rcv_saddr; | |
2659 | __u16 destp = ntohs(inet->inet_dport); | |
2660 | __u16 srcp = ntohs(inet->inet_sport); | |
2661 | int rx_queue; | |
2662 | ||
2663 | if (icsk->icsk_pending == ICSK_TIME_RETRANS) { | |
2664 | timer_active = 1; | |
2665 | timer_expires = icsk->icsk_timeout; | |
2666 | } else if (icsk->icsk_pending == ICSK_TIME_PROBE0) { | |
2667 | timer_active = 4; | |
2668 | timer_expires = icsk->icsk_timeout; | |
2669 | } else if (timer_pending(&sk->sk_timer)) { | |
2670 | timer_active = 2; | |
2671 | timer_expires = sk->sk_timer.expires; | |
2672 | } else { | |
2673 | timer_active = 0; | |
2674 | timer_expires = jiffies; | |
2675 | } | |
2676 | ||
2677 | if (sk->sk_state == TCP_LISTEN) | |
2678 | rx_queue = sk->sk_ack_backlog; | |
2679 | else | |
2680 | /* | |
2681 | * because we dont lock socket, we might find a transient negative value | |
2682 | */ | |
2683 | rx_queue = max_t(int, tp->rcv_nxt - tp->copied_seq, 0); | |
2684 | ||
2685 | seq_printf(f, "%4d: %08X:%04X %08X:%04X %02X %08X:%08X %02X:%08lX " | |
2686 | "%08X %5d %8d %lu %d %pK %lu %lu %u %u %d%n", | |
2687 | i, src, srcp, dest, destp, sk->sk_state, | |
2688 | tp->write_seq - tp->snd_una, | |
2689 | rx_queue, | |
2690 | timer_active, | |
2691 | jiffies_delta_to_clock_t(timer_expires - jiffies), | |
2692 | icsk->icsk_retransmits, | |
2693 | from_kuid_munged(seq_user_ns(f), sock_i_uid(sk)), | |
2694 | icsk->icsk_probes_out, | |
2695 | sock_i_ino(sk), | |
2696 | atomic_read(&sk->sk_refcnt), sk, | |
2697 | jiffies_to_clock_t(icsk->icsk_rto), | |
2698 | jiffies_to_clock_t(icsk->icsk_ack.ato), | |
2699 | (icsk->icsk_ack.quick << 1) | icsk->icsk_ack.pingpong, | |
2700 | tp->snd_cwnd, | |
2701 | sk->sk_state == TCP_LISTEN ? | |
2702 | (fastopenq ? fastopenq->max_qlen : 0) : | |
2703 | (tcp_in_initial_slowstart(tp) ? -1 : tp->snd_ssthresh), | |
2704 | len); | |
2705 | } | |
2706 | ||
2707 | static void get_timewait4_sock(const struct inet_timewait_sock *tw, | |
2708 | struct seq_file *f, int i, int *len) | |
2709 | { | |
2710 | __be32 dest, src; | |
2711 | __u16 destp, srcp; | |
2712 | long delta = tw->tw_ttd - jiffies; | |
2713 | ||
2714 | dest = tw->tw_daddr; | |
2715 | src = tw->tw_rcv_saddr; | |
2716 | destp = ntohs(tw->tw_dport); | |
2717 | srcp = ntohs(tw->tw_sport); | |
2718 | ||
2719 | seq_printf(f, "%4d: %08X:%04X %08X:%04X" | |
2720 | " %02X %08X:%08X %02X:%08lX %08X %5d %8d %d %d %pK%n", | |
2721 | i, src, srcp, dest, destp, tw->tw_substate, 0, 0, | |
2722 | 3, jiffies_delta_to_clock_t(delta), 0, 0, 0, 0, | |
2723 | atomic_read(&tw->tw_refcnt), tw, len); | |
2724 | } | |
2725 | ||
2726 | #define TMPSZ 150 | |
2727 | ||
2728 | static int tcp4_seq_show(struct seq_file *seq, void *v) | |
2729 | { | |
2730 | struct tcp_iter_state *st; | |
2731 | int len; | |
2732 | ||
2733 | if (v == SEQ_START_TOKEN) { | |
2734 | seq_printf(seq, "%-*s\n", TMPSZ - 1, | |
2735 | " sl local_address rem_address st tx_queue " | |
2736 | "rx_queue tr tm->when retrnsmt uid timeout " | |
2737 | "inode"); | |
2738 | goto out; | |
2739 | } | |
2740 | st = seq->private; | |
2741 | ||
2742 | switch (st->state) { | |
2743 | case TCP_SEQ_STATE_LISTENING: | |
2744 | case TCP_SEQ_STATE_ESTABLISHED: | |
2745 | get_tcp4_sock(v, seq, st->num, &len); | |
2746 | break; | |
2747 | case TCP_SEQ_STATE_OPENREQ: | |
2748 | get_openreq4(st->syn_wait_sk, v, seq, st->num, st->uid, &len); | |
2749 | break; | |
2750 | case TCP_SEQ_STATE_TIME_WAIT: | |
2751 | get_timewait4_sock(v, seq, st->num, &len); | |
2752 | break; | |
2753 | } | |
2754 | seq_printf(seq, "%*s\n", TMPSZ - 1 - len, ""); | |
2755 | out: | |
2756 | return 0; | |
2757 | } | |
2758 | ||
2759 | static const struct file_operations tcp_afinfo_seq_fops = { | |
2760 | .owner = THIS_MODULE, | |
2761 | .open = tcp_seq_open, | |
2762 | .read = seq_read, | |
2763 | .llseek = seq_lseek, | |
2764 | .release = seq_release_net | |
2765 | }; | |
2766 | ||
2767 | static struct tcp_seq_afinfo tcp4_seq_afinfo = { | |
2768 | .name = "tcp", | |
2769 | .family = AF_INET, | |
2770 | .seq_fops = &tcp_afinfo_seq_fops, | |
2771 | .seq_ops = { | |
2772 | .show = tcp4_seq_show, | |
2773 | }, | |
2774 | }; | |
2775 | ||
2776 | static int __net_init tcp4_proc_init_net(struct net *net) | |
2777 | { | |
2778 | return tcp_proc_register(net, &tcp4_seq_afinfo); | |
2779 | } | |
2780 | ||
2781 | static void __net_exit tcp4_proc_exit_net(struct net *net) | |
2782 | { | |
2783 | tcp_proc_unregister(net, &tcp4_seq_afinfo); | |
2784 | } | |
2785 | ||
2786 | static struct pernet_operations tcp4_net_ops = { | |
2787 | .init = tcp4_proc_init_net, | |
2788 | .exit = tcp4_proc_exit_net, | |
2789 | }; | |
2790 | ||
2791 | int __init tcp4_proc_init(void) | |
2792 | { | |
2793 | return register_pernet_subsys(&tcp4_net_ops); | |
2794 | } | |
2795 | ||
2796 | void tcp4_proc_exit(void) | |
2797 | { | |
2798 | unregister_pernet_subsys(&tcp4_net_ops); | |
2799 | } | |
2800 | #endif /* CONFIG_PROC_FS */ | |
2801 | ||
2802 | struct sk_buff **tcp4_gro_receive(struct sk_buff **head, struct sk_buff *skb) | |
2803 | { | |
2804 | const struct iphdr *iph = skb_gro_network_header(skb); | |
2805 | __wsum wsum; | |
2806 | __sum16 sum; | |
2807 | ||
2808 | switch (skb->ip_summed) { | |
2809 | case CHECKSUM_COMPLETE: | |
2810 | if (!tcp_v4_check(skb_gro_len(skb), iph->saddr, iph->daddr, | |
2811 | skb->csum)) { | |
2812 | skb->ip_summed = CHECKSUM_UNNECESSARY; | |
2813 | break; | |
2814 | } | |
2815 | flush: | |
2816 | NAPI_GRO_CB(skb)->flush = 1; | |
2817 | return NULL; | |
2818 | ||
2819 | case CHECKSUM_NONE: | |
2820 | wsum = csum_tcpudp_nofold(iph->saddr, iph->daddr, | |
2821 | skb_gro_len(skb), IPPROTO_TCP, 0); | |
2822 | sum = csum_fold(skb_checksum(skb, | |
2823 | skb_gro_offset(skb), | |
2824 | skb_gro_len(skb), | |
2825 | wsum)); | |
2826 | if (sum) | |
2827 | goto flush; | |
2828 | ||
2829 | skb->ip_summed = CHECKSUM_UNNECESSARY; | |
2830 | break; | |
2831 | } | |
2832 | ||
2833 | return tcp_gro_receive(head, skb); | |
2834 | } | |
2835 | ||
2836 | int tcp4_gro_complete(struct sk_buff *skb) | |
2837 | { | |
2838 | const struct iphdr *iph = ip_hdr(skb); | |
2839 | struct tcphdr *th = tcp_hdr(skb); | |
2840 | ||
2841 | th->check = ~tcp_v4_check(skb->len - skb_transport_offset(skb), | |
2842 | iph->saddr, iph->daddr, 0); | |
2843 | skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4; | |
2844 | ||
2845 | return tcp_gro_complete(skb); | |
2846 | } | |
2847 | ||
2848 | struct proto tcp_prot = { | |
2849 | .name = "TCP", | |
2850 | .owner = THIS_MODULE, | |
2851 | .close = tcp_close, | |
2852 | .connect = tcp_v4_connect, | |
2853 | .disconnect = tcp_disconnect, | |
2854 | .accept = inet_csk_accept, | |
2855 | .ioctl = tcp_ioctl, | |
2856 | .init = tcp_v4_init_sock, | |
2857 | .destroy = tcp_v4_destroy_sock, | |
2858 | .shutdown = tcp_shutdown, | |
2859 | .setsockopt = tcp_setsockopt, | |
2860 | .getsockopt = tcp_getsockopt, | |
2861 | .recvmsg = tcp_recvmsg, | |
2862 | .sendmsg = tcp_sendmsg, | |
2863 | .sendpage = tcp_sendpage, | |
2864 | .backlog_rcv = tcp_v4_do_rcv, | |
2865 | .release_cb = tcp_release_cb, | |
2866 | .mtu_reduced = tcp_v4_mtu_reduced, | |
2867 | .hash = inet_hash, | |
2868 | .unhash = inet_unhash, | |
2869 | .get_port = inet_csk_get_port, | |
2870 | .enter_memory_pressure = tcp_enter_memory_pressure, | |
2871 | .sockets_allocated = &tcp_sockets_allocated, | |
2872 | .orphan_count = &tcp_orphan_count, | |
2873 | .memory_allocated = &tcp_memory_allocated, | |
2874 | .memory_pressure = &tcp_memory_pressure, | |
2875 | .sysctl_wmem = sysctl_tcp_wmem, | |
2876 | .sysctl_rmem = sysctl_tcp_rmem, | |
2877 | .max_header = MAX_TCP_HEADER, | |
2878 | .obj_size = sizeof(struct tcp_sock), | |
2879 | .slab_flags = SLAB_DESTROY_BY_RCU, | |
2880 | .twsk_prot = &tcp_timewait_sock_ops, | |
2881 | .rsk_prot = &tcp_request_sock_ops, | |
2882 | .h.hashinfo = &tcp_hashinfo, | |
2883 | .no_autobind = true, | |
2884 | #ifdef CONFIG_COMPAT | |
2885 | .compat_setsockopt = compat_tcp_setsockopt, | |
2886 | .compat_getsockopt = compat_tcp_getsockopt, | |
2887 | #endif | |
2888 | #ifdef CONFIG_MEMCG_KMEM | |
2889 | .init_cgroup = tcp_init_cgroup, | |
2890 | .destroy_cgroup = tcp_destroy_cgroup, | |
2891 | .proto_cgroup = tcp_proto_cgroup, | |
2892 | #endif | |
2893 | }; | |
2894 | EXPORT_SYMBOL(tcp_prot); | |
2895 | ||
2896 | static int __net_init tcp_sk_init(struct net *net) | |
2897 | { | |
2898 | net->ipv4.sysctl_tcp_ecn = 2; | |
2899 | return 0; | |
2900 | } | |
2901 | ||
2902 | static void __net_exit tcp_sk_exit(struct net *net) | |
2903 | { | |
2904 | } | |
2905 | ||
2906 | static void __net_exit tcp_sk_exit_batch(struct list_head *net_exit_list) | |
2907 | { | |
2908 | inet_twsk_purge(&tcp_hashinfo, &tcp_death_row, AF_INET); | |
2909 | } | |
2910 | ||
2911 | static struct pernet_operations __net_initdata tcp_sk_ops = { | |
2912 | .init = tcp_sk_init, | |
2913 | .exit = tcp_sk_exit, | |
2914 | .exit_batch = tcp_sk_exit_batch, | |
2915 | }; | |
2916 | ||
2917 | void __init tcp_v4_init(void) | |
2918 | { | |
2919 | inet_hashinfo_init(&tcp_hashinfo); | |
2920 | if (register_pernet_subsys(&tcp_sk_ops)) | |
2921 | panic("Failed to create the TCP control socket.\n"); | |
2922 | } |