]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * INET An implementation of the TCP/IP protocol suite for the LINUX | |
3 | * operating system. INET is implemented using the BSD Socket | |
4 | * interface as the means of communication with the user level. | |
5 | * | |
6 | * Implementation of the Transmission Control Protocol(TCP). | |
7 | * | |
02c30a84 | 8 | * Authors: Ross Biro |
1da177e4 LT |
9 | * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG> |
10 | * Mark Evans, <evansmp@uhura.aston.ac.uk> | |
11 | * Corey Minyard <wf-rch!minyard@relay.EU.net> | |
12 | * Florian La Roche, <flla@stud.uni-sb.de> | |
13 | * Charles Hedrick, <hedrick@klinzhai.rutgers.edu> | |
14 | * Linus Torvalds, <torvalds@cs.helsinki.fi> | |
15 | * Alan Cox, <gw4pts@gw4pts.ampr.org> | |
16 | * Matthew Dillon, <dillon@apollo.west.oic.com> | |
17 | * Arnt Gulbrandsen, <agulbra@nvg.unit.no> | |
18 | * Jorge Cwik, <jorge@laser.satlink.net> | |
19 | */ | |
20 | ||
1da177e4 LT |
21 | #include <linux/mm.h> |
22 | #include <linux/module.h> | |
5a0e3ad6 | 23 | #include <linux/slab.h> |
1da177e4 LT |
24 | #include <linux/sysctl.h> |
25 | #include <linux/workqueue.h> | |
60e2a778 | 26 | #include <linux/static_key.h> |
1da177e4 LT |
27 | #include <net/tcp.h> |
28 | #include <net/inet_common.h> | |
29 | #include <net/xfrm.h> | |
e5907459 | 30 | #include <net/busy_poll.h> |
1da177e4 | 31 | |
a2a385d6 | 32 | static bool tcp_in_window(u32 seq, u32 end_seq, u32 s_win, u32 e_win) |
1da177e4 LT |
33 | { |
34 | if (seq == s_win) | |
a2a385d6 | 35 | return true; |
1da177e4 | 36 | if (after(end_seq, s_win) && before(seq, e_win)) |
a2a385d6 | 37 | return true; |
a02cec21 | 38 | return seq == e_win && seq == end_seq; |
1da177e4 LT |
39 | } |
40 | ||
4fb17a60 NC |
41 | static enum tcp_tw_status |
42 | tcp_timewait_check_oow_rate_limit(struct inet_timewait_sock *tw, | |
43 | const struct sk_buff *skb, int mib_idx) | |
44 | { | |
45 | struct tcp_timewait_sock *tcptw = tcp_twsk((struct sock *)tw); | |
46 | ||
47 | if (!tcp_oow_rate_limited(twsk_net(tw), skb, mib_idx, | |
48 | &tcptw->tw_last_oow_ack_time)) { | |
49 | /* Send ACK. Note, we do not put the bucket, | |
50 | * it will be released by caller. | |
51 | */ | |
52 | return TCP_TW_ACK; | |
53 | } | |
54 | ||
55 | /* We are rate-limiting, so just release the tw sock and drop skb. */ | |
56 | inet_twsk_put(tw); | |
57 | return TCP_TW_SUCCESS; | |
58 | } | |
59 | ||
e905a9ed | 60 | /* |
1da177e4 LT |
61 | * * Main purpose of TIME-WAIT state is to close connection gracefully, |
62 | * when one of ends sits in LAST-ACK or CLOSING retransmitting FIN | |
63 | * (and, probably, tail of data) and one or more our ACKs are lost. | |
64 | * * What is TIME-WAIT timeout? It is associated with maximal packet | |
65 | * lifetime in the internet, which results in wrong conclusion, that | |
66 | * it is set to catch "old duplicate segments" wandering out of their path. | |
67 | * It is not quite correct. This timeout is calculated so that it exceeds | |
68 | * maximal retransmission timeout enough to allow to lose one (or more) | |
69 | * segments sent by peer and our ACKs. This time may be calculated from RTO. | |
70 | * * When TIME-WAIT socket receives RST, it means that another end | |
71 | * finally closed and we are allowed to kill TIME-WAIT too. | |
72 | * * Second purpose of TIME-WAIT is catching old duplicate segments. | |
73 | * Well, certainly it is pure paranoia, but if we load TIME-WAIT | |
74 | * with this semantics, we MUST NOT kill TIME-WAIT state with RSTs. | |
75 | * * If we invented some more clever way to catch duplicates | |
76 | * (f.e. based on PAWS), we could truncate TIME-WAIT to several RTOs. | |
77 | * | |
78 | * The algorithm below is based on FORMAL INTERPRETATION of RFCs. | |
79 | * When you compare it to RFCs, please, read section SEGMENT ARRIVES | |
80 | * from the very beginning. | |
81 | * | |
82 | * NOTE. With recycling (and later with fin-wait-2) TW bucket | |
83 | * is _not_ stateless. It means, that strictly speaking we must | |
84 | * spinlock it. I do not want! Well, probability of misbehaviour | |
85 | * is ridiculously low and, seems, we could use some mb() tricks | |
86 | * to avoid misread sequence numbers, states etc. --ANK | |
4308fc58 AC |
87 | * |
88 | * We don't need to initialize tmp_out.sack_ok as we don't use the results | |
1da177e4 LT |
89 | */ |
90 | enum tcp_tw_status | |
8feaf0c0 ACM |
91 | tcp_timewait_state_process(struct inet_timewait_sock *tw, struct sk_buff *skb, |
92 | const struct tcphdr *th) | |
1da177e4 LT |
93 | { |
94 | struct tcp_options_received tmp_opt; | |
4957faad | 95 | struct tcp_timewait_sock *tcptw = tcp_twsk((struct sock *)tw); |
a2a385d6 | 96 | bool paws_reject = false; |
1da177e4 | 97 | |
bb5b7c11 | 98 | tmp_opt.saw_tstamp = 0; |
8feaf0c0 | 99 | if (th->doff > (sizeof(*th) >> 2) && tcptw->tw_ts_recent_stamp) { |
eed29f17 | 100 | tcp_parse_options(twsk_net(tw), skb, &tmp_opt, 0, NULL); |
1da177e4 LT |
101 | |
102 | if (tmp_opt.saw_tstamp) { | |
eee2faab AK |
103 | if (tmp_opt.rcv_tsecr) |
104 | tmp_opt.rcv_tsecr -= tcptw->tw_ts_offset; | |
8feaf0c0 ACM |
105 | tmp_opt.ts_recent = tcptw->tw_ts_recent; |
106 | tmp_opt.ts_recent_stamp = tcptw->tw_ts_recent_stamp; | |
c887e6d2 | 107 | paws_reject = tcp_paws_reject(&tmp_opt, th->rst); |
1da177e4 LT |
108 | } |
109 | } | |
110 | ||
111 | if (tw->tw_substate == TCP_FIN_WAIT2) { | |
112 | /* Just repeat all the checks of tcp_rcv_state_process() */ | |
113 | ||
114 | /* Out of window, send ACK */ | |
115 | if (paws_reject || | |
116 | !tcp_in_window(TCP_SKB_CB(skb)->seq, TCP_SKB_CB(skb)->end_seq, | |
8feaf0c0 ACM |
117 | tcptw->tw_rcv_nxt, |
118 | tcptw->tw_rcv_nxt + tcptw->tw_rcv_wnd)) | |
4fb17a60 NC |
119 | return tcp_timewait_check_oow_rate_limit( |
120 | tw, skb, LINUX_MIB_TCPACKSKIPPEDFINWAIT2); | |
1da177e4 LT |
121 | |
122 | if (th->rst) | |
123 | goto kill; | |
124 | ||
8feaf0c0 | 125 | if (th->syn && !before(TCP_SKB_CB(skb)->seq, tcptw->tw_rcv_nxt)) |
271c3b9b | 126 | return TCP_TW_RST; |
1da177e4 LT |
127 | |
128 | /* Dup ACK? */ | |
1ac530b3 WY |
129 | if (!th->ack || |
130 | !after(TCP_SKB_CB(skb)->end_seq, tcptw->tw_rcv_nxt) || | |
1da177e4 | 131 | TCP_SKB_CB(skb)->end_seq == TCP_SKB_CB(skb)->seq) { |
8feaf0c0 | 132 | inet_twsk_put(tw); |
1da177e4 LT |
133 | return TCP_TW_SUCCESS; |
134 | } | |
135 | ||
136 | /* New data or FIN. If new data arrive after half-duplex close, | |
137 | * reset. | |
138 | */ | |
139 | if (!th->fin || | |
271c3b9b | 140 | TCP_SKB_CB(skb)->end_seq != tcptw->tw_rcv_nxt + 1) |
1da177e4 | 141 | return TCP_TW_RST; |
1da177e4 LT |
142 | |
143 | /* FIN arrived, enter true time-wait state. */ | |
8feaf0c0 ACM |
144 | tw->tw_substate = TCP_TIME_WAIT; |
145 | tcptw->tw_rcv_nxt = TCP_SKB_CB(skb)->end_seq; | |
1da177e4 | 146 | if (tmp_opt.saw_tstamp) { |
cca9bab1 | 147 | tcptw->tw_ts_recent_stamp = ktime_get_seconds(); |
8feaf0c0 | 148 | tcptw->tw_ts_recent = tmp_opt.rcv_tsval; |
1da177e4 LT |
149 | } |
150 | ||
d82bae12 | 151 | inet_twsk_reschedule(tw, TCP_TIMEWAIT_LEN); |
1da177e4 LT |
152 | return TCP_TW_ACK; |
153 | } | |
154 | ||
155 | /* | |
156 | * Now real TIME-WAIT state. | |
157 | * | |
158 | * RFC 1122: | |
159 | * "When a connection is [...] on TIME-WAIT state [...] | |
160 | * [a TCP] MAY accept a new SYN from the remote TCP to | |
161 | * reopen the connection directly, if it: | |
e905a9ed | 162 | * |
1da177e4 LT |
163 | * (1) assigns its initial sequence number for the new |
164 | * connection to be larger than the largest sequence | |
165 | * number it used on the previous connection incarnation, | |
166 | * and | |
167 | * | |
e905a9ed | 168 | * (2) returns to TIME-WAIT state if the SYN turns out |
1da177e4 LT |
169 | * to be an old duplicate". |
170 | */ | |
171 | ||
172 | if (!paws_reject && | |
8feaf0c0 | 173 | (TCP_SKB_CB(skb)->seq == tcptw->tw_rcv_nxt && |
1da177e4 LT |
174 | (TCP_SKB_CB(skb)->seq == TCP_SKB_CB(skb)->end_seq || th->rst))) { |
175 | /* In window segment, it may be only reset or bare ack. */ | |
176 | ||
177 | if (th->rst) { | |
caa20d9a | 178 | /* This is TIME_WAIT assassination, in two flavors. |
1da177e4 LT |
179 | * Oh well... nobody has a sufficient solution to this |
180 | * protocol bug yet. | |
181 | */ | |
625357aa | 182 | if (twsk_net(tw)->ipv4.sysctl_tcp_rfc1337 == 0) { |
1da177e4 | 183 | kill: |
dbe7faa4 | 184 | inet_twsk_deschedule_put(tw); |
1da177e4 LT |
185 | return TCP_TW_SUCCESS; |
186 | } | |
63cc357f FW |
187 | } else { |
188 | inet_twsk_reschedule(tw, TCP_TIMEWAIT_LEN); | |
1da177e4 | 189 | } |
1da177e4 LT |
190 | |
191 | if (tmp_opt.saw_tstamp) { | |
8feaf0c0 | 192 | tcptw->tw_ts_recent = tmp_opt.rcv_tsval; |
cca9bab1 | 193 | tcptw->tw_ts_recent_stamp = ktime_get_seconds(); |
1da177e4 LT |
194 | } |
195 | ||
8feaf0c0 | 196 | inet_twsk_put(tw); |
1da177e4 LT |
197 | return TCP_TW_SUCCESS; |
198 | } | |
199 | ||
200 | /* Out of window segment. | |
201 | ||
202 | All the segments are ACKed immediately. | |
203 | ||
204 | The only exception is new SYN. We accept it, if it is | |
205 | not old duplicate and we are not in danger to be killed | |
206 | by delayed old duplicates. RFC check is that it has | |
207 | newer sequence number works at rates <40Mbit/sec. | |
208 | However, if paws works, it is reliable AND even more, | |
209 | we even may relax silly seq space cutoff. | |
210 | ||
211 | RED-PEN: we violate main RFC requirement, if this SYN will appear | |
212 | old duplicate (i.e. we receive RST in reply to SYN-ACK), | |
213 | we must return socket to time-wait state. It is not good, | |
214 | but not fatal yet. | |
215 | */ | |
216 | ||
217 | if (th->syn && !th->rst && !th->ack && !paws_reject && | |
8feaf0c0 ACM |
218 | (after(TCP_SKB_CB(skb)->seq, tcptw->tw_rcv_nxt) || |
219 | (tmp_opt.saw_tstamp && | |
220 | (s32)(tcptw->tw_ts_recent - tmp_opt.rcv_tsval) < 0))) { | |
221 | u32 isn = tcptw->tw_snd_nxt + 65535 + 2; | |
1da177e4 LT |
222 | if (isn == 0) |
223 | isn++; | |
04317daf | 224 | TCP_SKB_CB(skb)->tcp_tw_isn = isn; |
1da177e4 LT |
225 | return TCP_TW_SYN; |
226 | } | |
227 | ||
228 | if (paws_reject) | |
02a1d6e7 | 229 | __NET_INC_STATS(twsk_net(tw), LINUX_MIB_PAWSESTABREJECTED); |
1da177e4 | 230 | |
2de979bd | 231 | if (!th->rst) { |
1da177e4 LT |
232 | /* In this case we must reset the TIMEWAIT timer. |
233 | * | |
234 | * If it is ACKless SYN it may be both old duplicate | |
235 | * and new good SYN with random sequence number <rcv_nxt. | |
236 | * Do not reschedule in the last case. | |
237 | */ | |
238 | if (paws_reject || th->ack) | |
ed2e9239 | 239 | inet_twsk_reschedule(tw, TCP_TIMEWAIT_LEN); |
1da177e4 | 240 | |
4fb17a60 NC |
241 | return tcp_timewait_check_oow_rate_limit( |
242 | tw, skb, LINUX_MIB_TCPACKSKIPPEDTIMEWAIT); | |
1da177e4 | 243 | } |
8feaf0c0 | 244 | inet_twsk_put(tw); |
1da177e4 LT |
245 | return TCP_TW_SUCCESS; |
246 | } | |
4bc2f18b | 247 | EXPORT_SYMBOL(tcp_timewait_state_process); |
1da177e4 | 248 | |
e905a9ed | 249 | /* |
1da177e4 | 250 | * Move a socket to time-wait or dead fin-wait-2 state. |
e905a9ed | 251 | */ |
1da177e4 LT |
252 | void tcp_time_wait(struct sock *sk, int state, int timeo) |
253 | { | |
8292a17a | 254 | const struct inet_connection_sock *icsk = inet_csk(sk); |
8feaf0c0 | 255 | const struct tcp_sock *tp = tcp_sk(sk); |
789f558c | 256 | struct inet_timewait_sock *tw; |
1946e672 | 257 | struct inet_timewait_death_row *tcp_death_row = &sock_net(sk)->ipv4.tcp_death_row; |
1da177e4 | 258 | |
1946e672 | 259 | tw = inet_twsk_alloc(sk, tcp_death_row, state); |
1da177e4 | 260 | |
00db4124 | 261 | if (tw) { |
8feaf0c0 | 262 | struct tcp_timewait_sock *tcptw = tcp_twsk((struct sock *)tw); |
463c84b9 | 263 | const int rto = (icsk->icsk_rto << 2) - (icsk->icsk_rto >> 1); |
2397849b | 264 | struct inet_sock *inet = inet_sk(sk); |
8feaf0c0 | 265 | |
2397849b | 266 | tw->tw_transparent = inet->transparent; |
00483690 | 267 | tw->tw_mark = sk->sk_mark; |
1da177e4 | 268 | tw->tw_rcv_wscale = tp->rx_opt.rcv_wscale; |
8feaf0c0 ACM |
269 | tcptw->tw_rcv_nxt = tp->rcv_nxt; |
270 | tcptw->tw_snd_nxt = tp->snd_nxt; | |
271 | tcptw->tw_rcv_wnd = tcp_receive_window(tp); | |
272 | tcptw->tw_ts_recent = tp->rx_opt.ts_recent; | |
273 | tcptw->tw_ts_recent_stamp = tp->rx_opt.ts_recent_stamp; | |
ceaa1fef | 274 | tcptw->tw_ts_offset = tp->tsoffset; |
4fb17a60 | 275 | tcptw->tw_last_oow_ack_time = 0; |
1da177e4 | 276 | |
dfd56b8b | 277 | #if IS_ENABLED(CONFIG_IPV6) |
1da177e4 LT |
278 | if (tw->tw_family == PF_INET6) { |
279 | struct ipv6_pinfo *np = inet6_sk(sk); | |
280 | ||
efe4208f ED |
281 | tw->tw_v6_daddr = sk->sk_v6_daddr; |
282 | tw->tw_v6_rcv_saddr = sk->sk_v6_rcv_saddr; | |
b903d324 | 283 | tw->tw_tclass = np->tclass; |
21858cd0 | 284 | tw->tw_flowlabel = be32_to_cpu(np->flow_label & IPV6_FLOWLABEL_MASK); |
9fe516ba | 285 | tw->tw_ipv6only = sk->sk_ipv6only; |
c676270b | 286 | } |
1da177e4 | 287 | #endif |
cfb6eeb4 YH |
288 | |
289 | #ifdef CONFIG_TCP_MD5SIG | |
290 | /* | |
291 | * The timewait bucket does not have the key DB from the | |
292 | * sock structure. We just make a quick copy of the | |
293 | * md5 key being used (if indeed we are using one) | |
294 | * so the timewait ack generating code has the key. | |
295 | */ | |
296 | do { | |
297 | struct tcp_md5sig_key *key; | |
a915da9b | 298 | tcptw->tw_md5_key = NULL; |
cfb6eeb4 | 299 | key = tp->af_specific->md5_lookup(sk, sk); |
00db4124 | 300 | if (key) { |
a915da9b | 301 | tcptw->tw_md5_key = kmemdup(key, sizeof(*key), GFP_ATOMIC); |
49ca1943 | 302 | BUG_ON(tcptw->tw_md5_key && !tcp_alloc_md5sig_pool()); |
cfb6eeb4 | 303 | } |
2de979bd | 304 | } while (0); |
cfb6eeb4 YH |
305 | #endif |
306 | ||
1da177e4 LT |
307 | /* Get the TIME_WAIT timeout firing. */ |
308 | if (timeo < rto) | |
309 | timeo = rto; | |
310 | ||
d82bae12 SHY |
311 | if (state == TCP_TIME_WAIT) |
312 | timeo = TCP_TIMEWAIT_LEN; | |
1da177e4 | 313 | |
cfac7f83 ED |
314 | /* tw_timer is pinned, so we need to make sure BH are disabled |
315 | * in following section, otherwise timer handler could run before | |
316 | * we complete the initialization. | |
317 | */ | |
318 | local_bh_disable(); | |
789f558c | 319 | inet_twsk_schedule(tw, timeo); |
ec94c269 ED |
320 | /* Linkage updates. |
321 | * Note that access to tw after this point is illegal. | |
322 | */ | |
323 | inet_twsk_hashdance(tw, sk, &tcp_hashinfo); | |
cfac7f83 | 324 | local_bh_enable(); |
1da177e4 LT |
325 | } else { |
326 | /* Sorry, if we're out of memory, just CLOSE this | |
327 | * socket up. We've got bigger problems than | |
328 | * non-graceful socket closings. | |
329 | */ | |
c10d9310 | 330 | NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPTIMEWAITOVERFLOW); |
1da177e4 LT |
331 | } |
332 | ||
333 | tcp_update_metrics(sk); | |
334 | tcp_done(sk); | |
335 | } | |
cc35c88a | 336 | EXPORT_SYMBOL(tcp_time_wait); |
1da177e4 | 337 | |
cfb6eeb4 YH |
338 | void tcp_twsk_destructor(struct sock *sk) |
339 | { | |
b6242b9b | 340 | #ifdef CONFIG_TCP_MD5SIG |
a928630a | 341 | struct tcp_timewait_sock *twsk = tcp_twsk(sk); |
2397849b | 342 | |
71cea17e | 343 | if (twsk->tw_md5_key) |
a915da9b | 344 | kfree_rcu(twsk->tw_md5_key, rcu); |
cfb6eeb4 YH |
345 | #endif |
346 | } | |
cfb6eeb4 YH |
347 | EXPORT_SYMBOL_GPL(tcp_twsk_destructor); |
348 | ||
b1964b5f ED |
349 | /* Warning : This function is called without sk_listener being locked. |
350 | * Be sure to read socket fields once, as their value could change under us. | |
351 | */ | |
843f4a55 | 352 | void tcp_openreq_init_rwin(struct request_sock *req, |
b1964b5f ED |
353 | const struct sock *sk_listener, |
354 | const struct dst_entry *dst) | |
843f4a55 YC |
355 | { |
356 | struct inet_request_sock *ireq = inet_rsk(req); | |
b1964b5f | 357 | const struct tcp_sock *tp = tcp_sk(sk_listener); |
b1964b5f | 358 | int full_space = tcp_full_space(sk_listener); |
b1964b5f ED |
359 | u32 window_clamp; |
360 | __u8 rcv_wscale; | |
13d3b1eb | 361 | u32 rcv_wnd; |
3541f9e8 | 362 | int mss; |
843f4a55 | 363 | |
3541f9e8 | 364 | mss = tcp_mss_clamp(tp, dst_metric_advmss(dst)); |
b1964b5f | 365 | window_clamp = READ_ONCE(tp->window_clamp); |
843f4a55 | 366 | /* Set this up on the first call only */ |
ed53d0ab | 367 | req->rsk_window_clamp = window_clamp ? : dst_metric(dst, RTAX_WINDOW); |
843f4a55 YC |
368 | |
369 | /* limit the window selection if the user enforce a smaller rx buffer */ | |
b1964b5f | 370 | if (sk_listener->sk_userlocks & SOCK_RCVBUF_LOCK && |
ed53d0ab ED |
371 | (req->rsk_window_clamp > full_space || req->rsk_window_clamp == 0)) |
372 | req->rsk_window_clamp = full_space; | |
843f4a55 | 373 | |
13d3b1eb LB |
374 | rcv_wnd = tcp_rwnd_init_bpf((struct sock *)req); |
375 | if (rcv_wnd == 0) | |
376 | rcv_wnd = dst_metric(dst, RTAX_INITRWND); | |
377 | else if (full_space < rcv_wnd * mss) | |
378 | full_space = rcv_wnd * mss; | |
379 | ||
843f4a55 | 380 | /* tcp_full_space because it is guaranteed to be the first packet */ |
ceef9ab6 | 381 | tcp_select_initial_window(sk_listener, full_space, |
843f4a55 | 382 | mss - (ireq->tstamp_ok ? TCPOLEN_TSTAMP_ALIGNED : 0), |
ed53d0ab ED |
383 | &req->rsk_rcv_wnd, |
384 | &req->rsk_window_clamp, | |
843f4a55 YC |
385 | ireq->wscale_ok, |
386 | &rcv_wscale, | |
13d3b1eb | 387 | rcv_wnd); |
843f4a55 YC |
388 | ireq->rcv_wscale = rcv_wscale; |
389 | } | |
390 | EXPORT_SYMBOL(tcp_openreq_init_rwin); | |
391 | ||
735d3831 FW |
392 | static void tcp_ecn_openreq_child(struct tcp_sock *tp, |
393 | const struct request_sock *req) | |
bdf1ee5d IJ |
394 | { |
395 | tp->ecn_flags = inet_rsk(req)->ecn_ok ? TCP_ECN_OK : 0; | |
396 | } | |
397 | ||
81164413 DB |
398 | void tcp_ca_openreq_child(struct sock *sk, const struct dst_entry *dst) |
399 | { | |
400 | struct inet_connection_sock *icsk = inet_csk(sk); | |
401 | u32 ca_key = dst_metric(dst, RTAX_CC_ALGO); | |
402 | bool ca_got_dst = false; | |
403 | ||
404 | if (ca_key != TCP_CA_UNSPEC) { | |
405 | const struct tcp_congestion_ops *ca; | |
406 | ||
407 | rcu_read_lock(); | |
408 | ca = tcp_ca_find_key(ca_key); | |
409 | if (likely(ca && try_module_get(ca->owner))) { | |
410 | icsk->icsk_ca_dst_locked = tcp_ca_dst_locked(dst); | |
411 | icsk->icsk_ca_ops = ca; | |
412 | ca_got_dst = true; | |
413 | } | |
414 | rcu_read_unlock(); | |
415 | } | |
416 | ||
9f950415 NC |
417 | /* If no valid choice made yet, assign current system default ca. */ |
418 | if (!ca_got_dst && | |
419 | (!icsk->icsk_ca_setsockopt || | |
420 | !try_module_get(icsk->icsk_ca_ops->owner))) | |
81164413 DB |
421 | tcp_assign_congestion_control(sk); |
422 | ||
423 | tcp_set_ca_state(sk, TCP_CA_Open); | |
424 | } | |
425 | EXPORT_SYMBOL_GPL(tcp_ca_openreq_child); | |
426 | ||
60e2a778 UB |
427 | static void smc_check_reset_syn_req(struct tcp_sock *oldtp, |
428 | struct request_sock *req, | |
429 | struct tcp_sock *newtp) | |
430 | { | |
431 | #if IS_ENABLED(CONFIG_SMC) | |
432 | struct inet_request_sock *ireq; | |
433 | ||
434 | if (static_branch_unlikely(&tcp_have_smc)) { | |
435 | ireq = inet_rsk(req); | |
436 | if (oldtp->syn_smc && !ireq->smc_ok) | |
437 | newtp->syn_smc = 0; | |
438 | } | |
439 | #endif | |
440 | } | |
441 | ||
1da177e4 LT |
442 | /* This is not only more efficient than what we used to do, it eliminates |
443 | * a lot of code duplication between IPv4/IPv6 SYN recv processing. -DaveM | |
444 | * | |
445 | * Actually, we could lots of memory writes here. tp of listening | |
446 | * socket contains all necessary default parameters. | |
447 | */ | |
c28c6f04 ED |
448 | struct sock *tcp_create_openreq_child(const struct sock *sk, |
449 | struct request_sock *req, | |
450 | struct sk_buff *skb) | |
1da177e4 | 451 | { |
e56c57d0 | 452 | struct sock *newsk = inet_csk_clone_lock(sk, req, GFP_ATOMIC); |
242b1bbe ED |
453 | const struct inet_request_sock *ireq = inet_rsk(req); |
454 | struct tcp_request_sock *treq = tcp_rsk(req); | |
455 | struct inet_connection_sock *newicsk; | |
456 | struct tcp_sock *oldtp, *newtp; | |
457 | ||
458 | if (!newsk) | |
459 | return NULL; | |
460 | ||
461 | newicsk = inet_csk(newsk); | |
462 | newtp = tcp_sk(newsk); | |
463 | oldtp = tcp_sk(sk); | |
464 | ||
465 | smc_check_reset_syn_req(oldtp, req, newtp); | |
466 | ||
467 | /* Now setup tcp_sock */ | |
468 | newtp->pred_flags = 0; | |
469 | ||
470 | newtp->rcv_wup = newtp->copied_seq = | |
471 | newtp->rcv_nxt = treq->rcv_isn + 1; | |
472 | newtp->segs_in = 1; | |
473 | ||
474 | newtp->snd_sml = newtp->snd_una = | |
475 | newtp->snd_nxt = newtp->snd_up = treq->snt_isn + 1; | |
476 | ||
477 | INIT_LIST_HEAD(&newtp->tsq_node); | |
478 | INIT_LIST_HEAD(&newtp->tsorted_sent_queue); | |
479 | ||
480 | tcp_init_wl(newtp, treq->rcv_isn); | |
481 | ||
242b1bbe | 482 | minmax_reset(&newtp->rtt_min, tcp_jiffies32, ~0U); |
242b1bbe ED |
483 | newicsk->icsk_ack.lrcvtime = tcp_jiffies32; |
484 | ||
242b1bbe ED |
485 | newtp->lsndtime = tcp_jiffies32; |
486 | newsk->sk_txhash = treq->txhash; | |
242b1bbe ED |
487 | newtp->total_retrans = req->num_retrans; |
488 | ||
242b1bbe ED |
489 | /* There's a bubble in the pipe until at least the first ACK. */ |
490 | newtp->app_limited = ~0U; | |
491 | ||
492 | tcp_init_xmit_timers(newsk); | |
493 | newtp->write_seq = newtp->pushed_seq = treq->snt_isn + 1; | |
494 | ||
495 | newtp->rx_opt.saw_tstamp = 0; | |
496 | ||
497 | newtp->rx_opt.dsack = 0; | |
498 | newtp->rx_opt.num_sacks = 0; | |
499 | ||
242b1bbe ED |
500 | if (sock_flag(newsk, SOCK_KEEPOPEN)) |
501 | inet_csk_reset_keepalive_timer(newsk, | |
502 | keepalive_time_when(newtp)); | |
503 | ||
504 | newtp->rx_opt.tstamp_ok = ireq->tstamp_ok; | |
505 | newtp->rx_opt.sack_ok = ireq->sack_ok; | |
506 | newtp->window_clamp = req->rsk_window_clamp; | |
507 | newtp->rcv_ssthresh = req->rsk_rcv_wnd; | |
508 | newtp->rcv_wnd = req->rsk_rcv_wnd; | |
509 | newtp->rx_opt.wscale_ok = ireq->wscale_ok; | |
510 | if (newtp->rx_opt.wscale_ok) { | |
511 | newtp->rx_opt.snd_wscale = ireq->snd_wscale; | |
512 | newtp->rx_opt.rcv_wscale = ireq->rcv_wscale; | |
513 | } else { | |
514 | newtp->rx_opt.snd_wscale = newtp->rx_opt.rcv_wscale = 0; | |
515 | newtp->window_clamp = min(newtp->window_clamp, 65535U); | |
516 | } | |
517 | newtp->snd_wnd = ntohs(tcp_hdr(skb)->window) << newtp->rx_opt.snd_wscale; | |
518 | newtp->max_window = newtp->snd_wnd; | |
519 | ||
520 | if (newtp->rx_opt.tstamp_ok) { | |
521 | newtp->rx_opt.ts_recent = req->ts_recent; | |
cca9bab1 | 522 | newtp->rx_opt.ts_recent_stamp = ktime_get_seconds(); |
242b1bbe ED |
523 | newtp->tcp_header_len = sizeof(struct tcphdr) + TCPOLEN_TSTAMP_ALIGNED; |
524 | } else { | |
525 | newtp->rx_opt.ts_recent_stamp = 0; | |
526 | newtp->tcp_header_len = sizeof(struct tcphdr); | |
527 | } | |
528 | newtp->tsoffset = treq->ts_off; | |
cfb6eeb4 | 529 | #ifdef CONFIG_TCP_MD5SIG |
242b1bbe ED |
530 | newtp->md5sig_info = NULL; /*XXX*/ |
531 | if (newtp->af_specific->md5_lookup(sk, newsk)) | |
532 | newtp->tcp_header_len += TCPOLEN_MD5SIG_ALIGNED; | |
cfb6eeb4 | 533 | #endif |
242b1bbe ED |
534 | if (skb->len >= TCP_MSS_DEFAULT + newtp->tcp_header_len) |
535 | newicsk->icsk_ack.last_seg_size = skb->len - newtp->tcp_header_len; | |
536 | newtp->rx_opt.mss_clamp = req->mss; | |
537 | tcp_ecn_openreq_child(newtp, req); | |
538 | newtp->fastopen_req = NULL; | |
539 | newtp->fastopen_rsk = NULL; | |
540 | newtp->syn_data_acked = 0; | |
541 | newtp->rack.mstamp = 0; | |
542 | newtp->rack.advanced = 0; | |
543 | newtp->rack.reo_wnd_steps = 1; | |
544 | newtp->rack.last_delivered = 0; | |
545 | newtp->rack.reo_wnd_persist = 0; | |
546 | newtp->rack.dsack_seen = 0; | |
547 | ||
548 | __TCP_INC_STATS(sock_net(sk), TCP_MIB_PASSIVEOPENS); | |
549 | ||
1da177e4 LT |
550 | return newsk; |
551 | } | |
4bc2f18b | 552 | EXPORT_SYMBOL(tcp_create_openreq_child); |
1da177e4 | 553 | |
e905a9ed | 554 | /* |
8336886f JC |
555 | * Process an incoming packet for SYN_RECV sockets represented as a |
556 | * request_sock. Normally sk is the listener socket but for TFO it | |
557 | * points to the child socket. | |
558 | * | |
559 | * XXX (TFO) - The current impl contains a special check for ack | |
560 | * validation and inside tcp_v4_reqsk_send_ack(). Can we do better? | |
4308fc58 AC |
561 | * |
562 | * We don't need to initialize tmp_opt.sack_ok as we don't use the results | |
1da177e4 LT |
563 | */ |
564 | ||
5a5f3a8d | 565 | struct sock *tcp_check_req(struct sock *sk, struct sk_buff *skb, |
60236fdd | 566 | struct request_sock *req, |
e0f9759f | 567 | bool fastopen, bool *req_stolen) |
1da177e4 | 568 | { |
4957faad | 569 | struct tcp_options_received tmp_opt; |
4957faad | 570 | struct sock *child; |
aa8223c7 | 571 | const struct tcphdr *th = tcp_hdr(skb); |
714e85be | 572 | __be32 flg = tcp_flag_word(th) & (TCP_FLAG_RST|TCP_FLAG_SYN|TCP_FLAG_ACK); |
a2a385d6 | 573 | bool paws_reject = false; |
5e0724d0 | 574 | bool own_req; |
1da177e4 | 575 | |
bb5b7c11 DM |
576 | tmp_opt.saw_tstamp = 0; |
577 | if (th->doff > (sizeof(struct tcphdr)>>2)) { | |
eed29f17 | 578 | tcp_parse_options(sock_net(sk), skb, &tmp_opt, 0, NULL); |
1da177e4 LT |
579 | |
580 | if (tmp_opt.saw_tstamp) { | |
581 | tmp_opt.ts_recent = req->ts_recent; | |
95a22cae FW |
582 | if (tmp_opt.rcv_tsecr) |
583 | tmp_opt.rcv_tsecr -= tcp_rsk(req)->ts_off; | |
1da177e4 LT |
584 | /* We do not store true stamp, but it is not required, |
585 | * it can be estimated (approximately) | |
586 | * from another data. | |
587 | */ | |
cca9bab1 | 588 | tmp_opt.ts_recent_stamp = ktime_get_seconds() - ((TCP_TIMEOUT_INIT/HZ)<<req->num_timeout); |
c887e6d2 | 589 | paws_reject = tcp_paws_reject(&tmp_opt, th->rst); |
1da177e4 LT |
590 | } |
591 | } | |
592 | ||
593 | /* Check for pure retransmitted SYN. */ | |
2e6599cb | 594 | if (TCP_SKB_CB(skb)->seq == tcp_rsk(req)->rcv_isn && |
1da177e4 LT |
595 | flg == TCP_FLAG_SYN && |
596 | !paws_reject) { | |
597 | /* | |
598 | * RFC793 draws (Incorrectly! It was fixed in RFC1122) | |
599 | * this case on figure 6 and figure 8, but formal | |
600 | * protocol description says NOTHING. | |
601 | * To be more exact, it says that we should send ACK, | |
602 | * because this segment (at least, if it has no data) | |
603 | * is out of window. | |
604 | * | |
605 | * CONCLUSION: RFC793 (even with RFC1122) DOES NOT | |
606 | * describe SYN-RECV state. All the description | |
607 | * is wrong, we cannot believe to it and should | |
608 | * rely only on common sense and implementation | |
609 | * experience. | |
610 | * | |
611 | * Enforce "SYN-ACK" according to figure 8, figure 6 | |
612 | * of RFC793, fixed by RFC1122. | |
8336886f JC |
613 | * |
614 | * Note that even if there is new data in the SYN packet | |
615 | * they will be thrown away too. | |
cd75eff6 YC |
616 | * |
617 | * Reset timer after retransmitting SYNACK, similar to | |
618 | * the idea of fast retransmit in recovery. | |
1da177e4 | 619 | */ |
a9b2c06d NC |
620 | if (!tcp_oow_rate_limited(sock_net(sk), skb, |
621 | LINUX_MIB_TCPACKSKIPPEDSYNRECV, | |
622 | &tcp_rsk(req)->last_oow_ack_time) && | |
623 | ||
dd929c1b ED |
624 | !inet_rtx_syn_ack(sk, req)) { |
625 | unsigned long expires = jiffies; | |
626 | ||
627 | expires += min(TCP_TIMEOUT_INIT << req->num_timeout, | |
628 | TCP_RTO_MAX); | |
629 | if (!fastopen) | |
630 | mod_timer_pending(&req->rsk_timer, expires); | |
631 | else | |
632 | req->rsk_timer.expires = expires; | |
633 | } | |
1da177e4 LT |
634 | return NULL; |
635 | } | |
636 | ||
637 | /* Further reproduces section "SEGMENT ARRIVES" | |
638 | for state SYN-RECEIVED of RFC793. | |
639 | It is broken, however, it does not work only | |
640 | when SYNs are crossed. | |
641 | ||
642 | You would think that SYN crossing is impossible here, since | |
643 | we should have a SYN_SENT socket (from connect()) on our end, | |
644 | but this is not true if the crossed SYNs were sent to both | |
645 | ends by a malicious third party. We must defend against this, | |
646 | and to do that we first verify the ACK (as per RFC793, page | |
647 | 36) and reset if it is invalid. Is this a true full defense? | |
648 | To convince ourselves, let us consider a way in which the ACK | |
649 | test can still pass in this 'malicious crossed SYNs' case. | |
650 | Malicious sender sends identical SYNs (and thus identical sequence | |
651 | numbers) to both A and B: | |
652 | ||
653 | A: gets SYN, seq=7 | |
654 | B: gets SYN, seq=7 | |
655 | ||
656 | By our good fortune, both A and B select the same initial | |
657 | send sequence number of seven :-) | |
658 | ||
659 | A: sends SYN|ACK, seq=7, ack_seq=8 | |
660 | B: sends SYN|ACK, seq=7, ack_seq=8 | |
661 | ||
662 | So we are now A eating this SYN|ACK, ACK test passes. So | |
663 | does sequence test, SYN is truncated, and thus we consider | |
664 | it a bare ACK. | |
665 | ||
ec0a1966 DM |
666 | If icsk->icsk_accept_queue.rskq_defer_accept, we silently drop this |
667 | bare ACK. Otherwise, we create an established connection. Both | |
668 | ends (listening sockets) accept the new incoming connection and try | |
669 | to talk to each other. 8-) | |
1da177e4 LT |
670 | |
671 | Note: This case is both harmless, and rare. Possibility is about the | |
672 | same as us discovering intelligent life on another plant tomorrow. | |
673 | ||
674 | But generally, we should (RFC lies!) to accept ACK | |
675 | from SYNACK both here and in tcp_rcv_state_process(). | |
676 | tcp_rcv_state_process() does not, hence, we do not too. | |
677 | ||
678 | Note that the case is absolutely generic: | |
679 | we cannot optimize anything here without | |
680 | violating protocol. All the checks must be made | |
681 | before attempt to create socket. | |
682 | */ | |
683 | ||
684 | /* RFC793 page 36: "If the connection is in any non-synchronized state ... | |
685 | * and the incoming segment acknowledges something not yet | |
caa20d9a | 686 | * sent (the segment carries an unacceptable ACK) ... |
1da177e4 LT |
687 | * a reset is sent." |
688 | * | |
8336886f JC |
689 | * Invalid ACK: reset will be sent by listening socket. |
690 | * Note that the ACK validity check for a Fast Open socket is done | |
691 | * elsewhere and is checked directly against the child socket rather | |
692 | * than req because user data may have been sent out. | |
1da177e4 | 693 | */ |
8336886f | 694 | if ((flg & TCP_FLAG_ACK) && !fastopen && |
435cf559 | 695 | (TCP_SKB_CB(skb)->ack_seq != |
1a2c6181 | 696 | tcp_rsk(req)->snt_isn + 1)) |
1da177e4 LT |
697 | return sk; |
698 | ||
699 | /* Also, it would be not so bad idea to check rcv_tsecr, which | |
700 | * is essentially ACK extension and too early or too late values | |
701 | * should cause reset in unsynchronized states. | |
702 | */ | |
703 | ||
704 | /* RFC793: "first check sequence number". */ | |
705 | ||
706 | if (paws_reject || !tcp_in_window(TCP_SKB_CB(skb)->seq, TCP_SKB_CB(skb)->end_seq, | |
ed53d0ab | 707 | tcp_rsk(req)->rcv_nxt, tcp_rsk(req)->rcv_nxt + req->rsk_rcv_wnd)) { |
1da177e4 | 708 | /* Out of window: send ACK and drop. */ |
4ce7e93c ED |
709 | if (!(flg & TCP_FLAG_RST) && |
710 | !tcp_oow_rate_limited(sock_net(sk), skb, | |
711 | LINUX_MIB_TCPACKSKIPPEDSYNRECV, | |
712 | &tcp_rsk(req)->last_oow_ack_time)) | |
6edafaaf | 713 | req->rsk_ops->send_ack(sk, skb, req); |
1da177e4 | 714 | if (paws_reject) |
02a1d6e7 | 715 | __NET_INC_STATS(sock_net(sk), LINUX_MIB_PAWSESTABREJECTED); |
1da177e4 LT |
716 | return NULL; |
717 | } | |
718 | ||
719 | /* In sequence, PAWS is OK. */ | |
720 | ||
8336886f | 721 | if (tmp_opt.saw_tstamp && !after(TCP_SKB_CB(skb)->seq, tcp_rsk(req)->rcv_nxt)) |
2aaab9a0 | 722 | req->ts_recent = tmp_opt.rcv_tsval; |
1da177e4 | 723 | |
2aaab9a0 AL |
724 | if (TCP_SKB_CB(skb)->seq == tcp_rsk(req)->rcv_isn) { |
725 | /* Truncate SYN, it is out of window starting | |
726 | at tcp_rsk(req)->rcv_isn + 1. */ | |
727 | flg &= ~TCP_FLAG_SYN; | |
728 | } | |
1da177e4 | 729 | |
2aaab9a0 AL |
730 | /* RFC793: "second check the RST bit" and |
731 | * "fourth, check the SYN bit" | |
732 | */ | |
733 | if (flg & (TCP_FLAG_RST|TCP_FLAG_SYN)) { | |
90bbcc60 | 734 | __TCP_INC_STATS(sock_net(sk), TCP_MIB_ATTEMPTFAILS); |
2aaab9a0 AL |
735 | goto embryonic_reset; |
736 | } | |
1da177e4 | 737 | |
2aaab9a0 AL |
738 | /* ACK sequence verified above, just make sure ACK is |
739 | * set. If ACK not set, just silently drop the packet. | |
8336886f JC |
740 | * |
741 | * XXX (TFO) - if we ever allow "data after SYN", the | |
742 | * following check needs to be removed. | |
2aaab9a0 AL |
743 | */ |
744 | if (!(flg & TCP_FLAG_ACK)) | |
745 | return NULL; | |
ec0a1966 | 746 | |
8336886f JC |
747 | /* For Fast Open no more processing is needed (sk is the |
748 | * child socket). | |
749 | */ | |
750 | if (fastopen) | |
751 | return sk; | |
752 | ||
d1b99ba4 | 753 | /* While TCP_DEFER_ACCEPT is active, drop bare ACK. */ |
e6c022a4 | 754 | if (req->num_timeout < inet_csk(sk)->icsk_accept_queue.rskq_defer_accept && |
2aaab9a0 AL |
755 | TCP_SKB_CB(skb)->end_seq == tcp_rsk(req)->rcv_isn + 1) { |
756 | inet_rsk(req)->acked = 1; | |
02a1d6e7 | 757 | __NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPDEFERACCEPTDROP); |
2aaab9a0 AL |
758 | return NULL; |
759 | } | |
760 | ||
761 | /* OK, ACK is valid, create big socket and | |
762 | * feed this segment to it. It will repeat all | |
763 | * the tests. THIS SEGMENT MUST MOVE SOCKET TO | |
764 | * ESTABLISHED STATE. If it will be dropped after | |
765 | * socket is created, wait for troubles. | |
766 | */ | |
5e0724d0 ED |
767 | child = inet_csk(sk)->icsk_af_ops->syn_recv_sock(sk, skb, req, NULL, |
768 | req, &own_req); | |
51456b29 | 769 | if (!child) |
2aaab9a0 | 770 | goto listen_overflow; |
1da177e4 | 771 | |
6bcfd7f8 | 772 | sock_rps_save_rxhash(child, skb); |
0f1c28ae | 773 | tcp_synack_rtt_meas(child, req); |
e0f9759f | 774 | *req_stolen = !own_req; |
5e0724d0 | 775 | return inet_csk_complete_hashdance(sk, child, req, own_req); |
1da177e4 | 776 | |
2aaab9a0 | 777 | listen_overflow: |
65c9410c | 778 | if (!sock_net(sk)->ipv4.sysctl_tcp_abort_on_overflow) { |
2aaab9a0 AL |
779 | inet_rsk(req)->acked = 1; |
780 | return NULL; | |
781 | } | |
1da177e4 | 782 | |
2aaab9a0 | 783 | embryonic_reset: |
8336886f JC |
784 | if (!(flg & TCP_FLAG_RST)) { |
785 | /* Received a bad SYN pkt - for TFO We try not to reset | |
786 | * the local connection unless it's really necessary to | |
787 | * avoid becoming vulnerable to outside attack aiming at | |
788 | * resetting legit local connections. | |
789 | */ | |
2aaab9a0 | 790 | req->rsk_ops->send_reset(sk, skb); |
8336886f JC |
791 | } else if (fastopen) { /* received a valid RST pkt */ |
792 | reqsk_fastopen_remove(sk, req, true); | |
793 | tcp_reset(sk); | |
794 | } | |
795 | if (!fastopen) { | |
52452c54 | 796 | inet_csk_reqsk_queue_drop(sk, req); |
02a1d6e7 | 797 | __NET_INC_STATS(sock_net(sk), LINUX_MIB_EMBRYONICRSTS); |
8336886f | 798 | } |
2aaab9a0 | 799 | return NULL; |
1da177e4 | 800 | } |
4bc2f18b | 801 | EXPORT_SYMBOL(tcp_check_req); |
1da177e4 LT |
802 | |
803 | /* | |
804 | * Queue segment on the new socket if the new socket is active, | |
805 | * otherwise we just shortcircuit this and continue with | |
806 | * the new socket. | |
8336886f JC |
807 | * |
808 | * For the vast majority of cases child->sk_state will be TCP_SYN_RECV | |
809 | * when entering. But other states are possible due to a race condition | |
810 | * where after __inet_lookup_established() fails but before the listener | |
811 | * locked is obtained, other packets cause the same connection to | |
812 | * be created. | |
1da177e4 LT |
813 | */ |
814 | ||
815 | int tcp_child_process(struct sock *parent, struct sock *child, | |
816 | struct sk_buff *skb) | |
817 | { | |
818 | int ret = 0; | |
819 | int state = child->sk_state; | |
820 | ||
e5907459 AD |
821 | /* record NAPI ID of child */ |
822 | sk_mark_napi_id(child, skb); | |
823 | ||
a44d6eac | 824 | tcp_segs_in(tcp_sk(child), skb); |
1da177e4 | 825 | if (!sock_owned_by_user(child)) { |
72ab4a86 | 826 | ret = tcp_rcv_state_process(child, skb); |
1da177e4 LT |
827 | /* Wakeup parent, send SIGIO */ |
828 | if (state == TCP_SYN_RECV && child->sk_state != state) | |
676d2369 | 829 | parent->sk_data_ready(parent); |
1da177e4 LT |
830 | } else { |
831 | /* Alas, it is possible again, because we do lookup | |
832 | * in main socket hash table and lock on listening | |
833 | * socket does not protect us more. | |
834 | */ | |
a3a858ff | 835 | __sk_add_backlog(child, skb); |
1da177e4 LT |
836 | } |
837 | ||
838 | bh_unlock_sock(child); | |
839 | sock_put(child); | |
840 | return ret; | |
841 | } | |
1da177e4 | 842 | EXPORT_SYMBOL(tcp_child_process); |