Commit | Line | Data |
---|---|---|
457c8996 | 1 | // SPDX-License-Identifier: GPL-2.0-only |
1da177e4 LT |
2 | /* |
3 | * INET An implementation of the TCP/IP protocol suite for the LINUX | |
4 | * operating system. INET is implemented using the BSD Socket | |
5 | * interface as the means of communication with the user level. | |
6 | * | |
7 | * Implementation of the Transmission Control Protocol(TCP). | |
8 | * | |
02c30a84 | 9 | * Authors: Ross Biro |
1da177e4 LT |
10 | * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG> |
11 | * Mark Evans, <evansmp@uhura.aston.ac.uk> | |
12 | * Corey Minyard <wf-rch!minyard@relay.EU.net> | |
13 | * Florian La Roche, <flla@stud.uni-sb.de> | |
14 | * Charles Hedrick, <hedrick@klinzhai.rutgers.edu> | |
15 | * Linus Torvalds, <torvalds@cs.helsinki.fi> | |
16 | * Alan Cox, <gw4pts@gw4pts.ampr.org> | |
17 | * Matthew Dillon, <dillon@apollo.west.oic.com> | |
18 | * Arnt Gulbrandsen, <agulbra@nvg.unit.no> | |
19 | * Jorge Cwik, <jorge@laser.satlink.net> | |
20 | */ | |
21 | ||
1da177e4 LT |
22 | #include <linux/mm.h> |
23 | #include <linux/module.h> | |
5a0e3ad6 | 24 | #include <linux/slab.h> |
1da177e4 LT |
25 | #include <linux/sysctl.h> |
26 | #include <linux/workqueue.h> | |
60e2a778 | 27 | #include <linux/static_key.h> |
1da177e4 LT |
28 | #include <net/tcp.h> |
29 | #include <net/inet_common.h> | |
30 | #include <net/xfrm.h> | |
e5907459 | 31 | #include <net/busy_poll.h> |
1da177e4 | 32 | |
a2a385d6 | 33 | static bool tcp_in_window(u32 seq, u32 end_seq, u32 s_win, u32 e_win) |
1da177e4 LT |
34 | { |
35 | if (seq == s_win) | |
a2a385d6 | 36 | return true; |
1da177e4 | 37 | if (after(end_seq, s_win) && before(seq, e_win)) |
a2a385d6 | 38 | return true; |
a02cec21 | 39 | return seq == e_win && seq == end_seq; |
1da177e4 LT |
40 | } |
41 | ||
4fb17a60 NC |
42 | static enum tcp_tw_status |
43 | tcp_timewait_check_oow_rate_limit(struct inet_timewait_sock *tw, | |
44 | const struct sk_buff *skb, int mib_idx) | |
45 | { | |
46 | struct tcp_timewait_sock *tcptw = tcp_twsk((struct sock *)tw); | |
47 | ||
48 | if (!tcp_oow_rate_limited(twsk_net(tw), skb, mib_idx, | |
49 | &tcptw->tw_last_oow_ack_time)) { | |
50 | /* Send ACK. Note, we do not put the bucket, | |
51 | * it will be released by caller. | |
52 | */ | |
53 | return TCP_TW_ACK; | |
54 | } | |
55 | ||
56 | /* We are rate-limiting, so just release the tw sock and drop skb. */ | |
57 | inet_twsk_put(tw); | |
58 | return TCP_TW_SUCCESS; | |
59 | } | |
60 | ||
e905a9ed | 61 | /* |
1da177e4 LT |
62 | * * Main purpose of TIME-WAIT state is to close connection gracefully, |
63 | * when one of ends sits in LAST-ACK or CLOSING retransmitting FIN | |
64 | * (and, probably, tail of data) and one or more our ACKs are lost. | |
65 | * * What is TIME-WAIT timeout? It is associated with maximal packet | |
66 | * lifetime in the internet, which results in wrong conclusion, that | |
67 | * it is set to catch "old duplicate segments" wandering out of their path. | |
68 | * It is not quite correct. This timeout is calculated so that it exceeds | |
69 | * maximal retransmission timeout enough to allow to lose one (or more) | |
70 | * segments sent by peer and our ACKs. This time may be calculated from RTO. | |
71 | * * When TIME-WAIT socket receives RST, it means that another end | |
72 | * finally closed and we are allowed to kill TIME-WAIT too. | |
73 | * * Second purpose of TIME-WAIT is catching old duplicate segments. | |
74 | * Well, certainly it is pure paranoia, but if we load TIME-WAIT | |
75 | * with this semantics, we MUST NOT kill TIME-WAIT state with RSTs. | |
76 | * * If we invented some more clever way to catch duplicates | |
77 | * (f.e. based on PAWS), we could truncate TIME-WAIT to several RTOs. | |
78 | * | |
79 | * The algorithm below is based on FORMAL INTERPRETATION of RFCs. | |
80 | * When you compare it to RFCs, please, read section SEGMENT ARRIVES | |
81 | * from the very beginning. | |
82 | * | |
83 | * NOTE. With recycling (and later with fin-wait-2) TW bucket | |
84 | * is _not_ stateless. It means, that strictly speaking we must | |
85 | * spinlock it. I do not want! Well, probability of misbehaviour | |
86 | * is ridiculously low and, seems, we could use some mb() tricks | |
87 | * to avoid misread sequence numbers, states etc. --ANK | |
4308fc58 AC |
88 | * |
89 | * We don't need to initialize tmp_out.sack_ok as we don't use the results | |
1da177e4 LT |
90 | */ |
91 | enum tcp_tw_status | |
8feaf0c0 ACM |
92 | tcp_timewait_state_process(struct inet_timewait_sock *tw, struct sk_buff *skb, |
93 | const struct tcphdr *th) | |
1da177e4 LT |
94 | { |
95 | struct tcp_options_received tmp_opt; | |
4957faad | 96 | struct tcp_timewait_sock *tcptw = tcp_twsk((struct sock *)tw); |
a2a385d6 | 97 | bool paws_reject = false; |
1da177e4 | 98 | |
bb5b7c11 | 99 | tmp_opt.saw_tstamp = 0; |
8feaf0c0 | 100 | if (th->doff > (sizeof(*th) >> 2) && tcptw->tw_ts_recent_stamp) { |
eed29f17 | 101 | tcp_parse_options(twsk_net(tw), skb, &tmp_opt, 0, NULL); |
1da177e4 LT |
102 | |
103 | if (tmp_opt.saw_tstamp) { | |
eee2faab AK |
104 | if (tmp_opt.rcv_tsecr) |
105 | tmp_opt.rcv_tsecr -= tcptw->tw_ts_offset; | |
8feaf0c0 ACM |
106 | tmp_opt.ts_recent = tcptw->tw_ts_recent; |
107 | tmp_opt.ts_recent_stamp = tcptw->tw_ts_recent_stamp; | |
c887e6d2 | 108 | paws_reject = tcp_paws_reject(&tmp_opt, th->rst); |
1da177e4 LT |
109 | } |
110 | } | |
111 | ||
112 | if (tw->tw_substate == TCP_FIN_WAIT2) { | |
113 | /* Just repeat all the checks of tcp_rcv_state_process() */ | |
114 | ||
115 | /* Out of window, send ACK */ | |
116 | if (paws_reject || | |
117 | !tcp_in_window(TCP_SKB_CB(skb)->seq, TCP_SKB_CB(skb)->end_seq, | |
8feaf0c0 ACM |
118 | tcptw->tw_rcv_nxt, |
119 | tcptw->tw_rcv_nxt + tcptw->tw_rcv_wnd)) | |
4fb17a60 NC |
120 | return tcp_timewait_check_oow_rate_limit( |
121 | tw, skb, LINUX_MIB_TCPACKSKIPPEDFINWAIT2); | |
1da177e4 LT |
122 | |
123 | if (th->rst) | |
124 | goto kill; | |
125 | ||
8feaf0c0 | 126 | if (th->syn && !before(TCP_SKB_CB(skb)->seq, tcptw->tw_rcv_nxt)) |
271c3b9b | 127 | return TCP_TW_RST; |
1da177e4 LT |
128 | |
129 | /* Dup ACK? */ | |
1ac530b3 WY |
130 | if (!th->ack || |
131 | !after(TCP_SKB_CB(skb)->end_seq, tcptw->tw_rcv_nxt) || | |
1da177e4 | 132 | TCP_SKB_CB(skb)->end_seq == TCP_SKB_CB(skb)->seq) { |
8feaf0c0 | 133 | inet_twsk_put(tw); |
1da177e4 LT |
134 | return TCP_TW_SUCCESS; |
135 | } | |
136 | ||
137 | /* New data or FIN. If new data arrive after half-duplex close, | |
138 | * reset. | |
139 | */ | |
140 | if (!th->fin || | |
271c3b9b | 141 | TCP_SKB_CB(skb)->end_seq != tcptw->tw_rcv_nxt + 1) |
1da177e4 | 142 | return TCP_TW_RST; |
1da177e4 LT |
143 | |
144 | /* FIN arrived, enter true time-wait state. */ | |
8feaf0c0 ACM |
145 | tw->tw_substate = TCP_TIME_WAIT; |
146 | tcptw->tw_rcv_nxt = TCP_SKB_CB(skb)->end_seq; | |
1da177e4 | 147 | if (tmp_opt.saw_tstamp) { |
cca9bab1 | 148 | tcptw->tw_ts_recent_stamp = ktime_get_seconds(); |
8feaf0c0 | 149 | tcptw->tw_ts_recent = tmp_opt.rcv_tsval; |
1da177e4 LT |
150 | } |
151 | ||
d82bae12 | 152 | inet_twsk_reschedule(tw, TCP_TIMEWAIT_LEN); |
1da177e4 LT |
153 | return TCP_TW_ACK; |
154 | } | |
155 | ||
156 | /* | |
157 | * Now real TIME-WAIT state. | |
158 | * | |
159 | * RFC 1122: | |
160 | * "When a connection is [...] on TIME-WAIT state [...] | |
161 | * [a TCP] MAY accept a new SYN from the remote TCP to | |
162 | * reopen the connection directly, if it: | |
e905a9ed | 163 | * |
1da177e4 LT |
164 | * (1) assigns its initial sequence number for the new |
165 | * connection to be larger than the largest sequence | |
166 | * number it used on the previous connection incarnation, | |
167 | * and | |
168 | * | |
e905a9ed | 169 | * (2) returns to TIME-WAIT state if the SYN turns out |
1da177e4 LT |
170 | * to be an old duplicate". |
171 | */ | |
172 | ||
173 | if (!paws_reject && | |
8feaf0c0 | 174 | (TCP_SKB_CB(skb)->seq == tcptw->tw_rcv_nxt && |
1da177e4 LT |
175 | (TCP_SKB_CB(skb)->seq == TCP_SKB_CB(skb)->end_seq || th->rst))) { |
176 | /* In window segment, it may be only reset or bare ack. */ | |
177 | ||
178 | if (th->rst) { | |
caa20d9a | 179 | /* This is TIME_WAIT assassination, in two flavors. |
1da177e4 LT |
180 | * Oh well... nobody has a sufficient solution to this |
181 | * protocol bug yet. | |
182 | */ | |
625357aa | 183 | if (twsk_net(tw)->ipv4.sysctl_tcp_rfc1337 == 0) { |
1da177e4 | 184 | kill: |
dbe7faa4 | 185 | inet_twsk_deschedule_put(tw); |
1da177e4 LT |
186 | return TCP_TW_SUCCESS; |
187 | } | |
63cc357f FW |
188 | } else { |
189 | inet_twsk_reschedule(tw, TCP_TIMEWAIT_LEN); | |
1da177e4 | 190 | } |
1da177e4 LT |
191 | |
192 | if (tmp_opt.saw_tstamp) { | |
8feaf0c0 | 193 | tcptw->tw_ts_recent = tmp_opt.rcv_tsval; |
cca9bab1 | 194 | tcptw->tw_ts_recent_stamp = ktime_get_seconds(); |
1da177e4 LT |
195 | } |
196 | ||
8feaf0c0 | 197 | inet_twsk_put(tw); |
1da177e4 LT |
198 | return TCP_TW_SUCCESS; |
199 | } | |
200 | ||
201 | /* Out of window segment. | |
202 | ||
203 | All the segments are ACKed immediately. | |
204 | ||
205 | The only exception is new SYN. We accept it, if it is | |
206 | not old duplicate and we are not in danger to be killed | |
207 | by delayed old duplicates. RFC check is that it has | |
208 | newer sequence number works at rates <40Mbit/sec. | |
209 | However, if paws works, it is reliable AND even more, | |
210 | we even may relax silly seq space cutoff. | |
211 | ||
212 | RED-PEN: we violate main RFC requirement, if this SYN will appear | |
213 | old duplicate (i.e. we receive RST in reply to SYN-ACK), | |
214 | we must return socket to time-wait state. It is not good, | |
215 | but not fatal yet. | |
216 | */ | |
217 | ||
218 | if (th->syn && !th->rst && !th->ack && !paws_reject && | |
8feaf0c0 ACM |
219 | (after(TCP_SKB_CB(skb)->seq, tcptw->tw_rcv_nxt) || |
220 | (tmp_opt.saw_tstamp && | |
221 | (s32)(tcptw->tw_ts_recent - tmp_opt.rcv_tsval) < 0))) { | |
222 | u32 isn = tcptw->tw_snd_nxt + 65535 + 2; | |
1da177e4 LT |
223 | if (isn == 0) |
224 | isn++; | |
04317daf | 225 | TCP_SKB_CB(skb)->tcp_tw_isn = isn; |
1da177e4 LT |
226 | return TCP_TW_SYN; |
227 | } | |
228 | ||
229 | if (paws_reject) | |
02a1d6e7 | 230 | __NET_INC_STATS(twsk_net(tw), LINUX_MIB_PAWSESTABREJECTED); |
1da177e4 | 231 | |
2de979bd | 232 | if (!th->rst) { |
1da177e4 LT |
233 | /* In this case we must reset the TIMEWAIT timer. |
234 | * | |
235 | * If it is ACKless SYN it may be both old duplicate | |
236 | * and new good SYN with random sequence number <rcv_nxt. | |
237 | * Do not reschedule in the last case. | |
238 | */ | |
239 | if (paws_reject || th->ack) | |
ed2e9239 | 240 | inet_twsk_reschedule(tw, TCP_TIMEWAIT_LEN); |
1da177e4 | 241 | |
4fb17a60 NC |
242 | return tcp_timewait_check_oow_rate_limit( |
243 | tw, skb, LINUX_MIB_TCPACKSKIPPEDTIMEWAIT); | |
1da177e4 | 244 | } |
8feaf0c0 | 245 | inet_twsk_put(tw); |
1da177e4 LT |
246 | return TCP_TW_SUCCESS; |
247 | } | |
4bc2f18b | 248 | EXPORT_SYMBOL(tcp_timewait_state_process); |
1da177e4 | 249 | |
e905a9ed | 250 | /* |
1da177e4 | 251 | * Move a socket to time-wait or dead fin-wait-2 state. |
e905a9ed | 252 | */ |
1da177e4 LT |
253 | void tcp_time_wait(struct sock *sk, int state, int timeo) |
254 | { | |
8292a17a | 255 | const struct inet_connection_sock *icsk = inet_csk(sk); |
8feaf0c0 | 256 | const struct tcp_sock *tp = tcp_sk(sk); |
789f558c | 257 | struct inet_timewait_sock *tw; |
1946e672 | 258 | struct inet_timewait_death_row *tcp_death_row = &sock_net(sk)->ipv4.tcp_death_row; |
1da177e4 | 259 | |
1946e672 | 260 | tw = inet_twsk_alloc(sk, tcp_death_row, state); |
1da177e4 | 261 | |
00db4124 | 262 | if (tw) { |
8feaf0c0 | 263 | struct tcp_timewait_sock *tcptw = tcp_twsk((struct sock *)tw); |
463c84b9 | 264 | const int rto = (icsk->icsk_rto << 2) - (icsk->icsk_rto >> 1); |
2397849b | 265 | struct inet_sock *inet = inet_sk(sk); |
8feaf0c0 | 266 | |
2397849b | 267 | tw->tw_transparent = inet->transparent; |
00483690 | 268 | tw->tw_mark = sk->sk_mark; |
f6c0f5d2 | 269 | tw->tw_priority = sk->sk_priority; |
1da177e4 | 270 | tw->tw_rcv_wscale = tp->rx_opt.rcv_wscale; |
8feaf0c0 ACM |
271 | tcptw->tw_rcv_nxt = tp->rcv_nxt; |
272 | tcptw->tw_snd_nxt = tp->snd_nxt; | |
273 | tcptw->tw_rcv_wnd = tcp_receive_window(tp); | |
274 | tcptw->tw_ts_recent = tp->rx_opt.ts_recent; | |
275 | tcptw->tw_ts_recent_stamp = tp->rx_opt.ts_recent_stamp; | |
ceaa1fef | 276 | tcptw->tw_ts_offset = tp->tsoffset; |
4fb17a60 | 277 | tcptw->tw_last_oow_ack_time = 0; |
a842fe14 | 278 | tcptw->tw_tx_delay = tp->tcp_tx_delay; |
dfd56b8b | 279 | #if IS_ENABLED(CONFIG_IPV6) |
1da177e4 LT |
280 | if (tw->tw_family == PF_INET6) { |
281 | struct ipv6_pinfo *np = inet6_sk(sk); | |
282 | ||
efe4208f ED |
283 | tw->tw_v6_daddr = sk->sk_v6_daddr; |
284 | tw->tw_v6_rcv_saddr = sk->sk_v6_rcv_saddr; | |
b903d324 | 285 | tw->tw_tclass = np->tclass; |
21858cd0 | 286 | tw->tw_flowlabel = be32_to_cpu(np->flow_label & IPV6_FLOWLABEL_MASK); |
c67b8555 | 287 | tw->tw_txhash = sk->sk_txhash; |
9fe516ba | 288 | tw->tw_ipv6only = sk->sk_ipv6only; |
c676270b | 289 | } |
1da177e4 | 290 | #endif |
cfb6eeb4 YH |
291 | |
292 | #ifdef CONFIG_TCP_MD5SIG | |
293 | /* | |
294 | * The timewait bucket does not have the key DB from the | |
295 | * sock structure. We just make a quick copy of the | |
296 | * md5 key being used (if indeed we are using one) | |
297 | * so the timewait ack generating code has the key. | |
298 | */ | |
299 | do { | |
a915da9b | 300 | tcptw->tw_md5_key = NULL; |
6aedbf98 ED |
301 | if (static_branch_unlikely(&tcp_md5_needed)) { |
302 | struct tcp_md5sig_key *key; | |
303 | ||
304 | key = tp->af_specific->md5_lookup(sk, sk); | |
305 | if (key) { | |
306 | tcptw->tw_md5_key = kmemdup(key, sizeof(*key), GFP_ATOMIC); | |
307 | BUG_ON(tcptw->tw_md5_key && !tcp_alloc_md5sig_pool()); | |
308 | } | |
cfb6eeb4 | 309 | } |
2de979bd | 310 | } while (0); |
cfb6eeb4 YH |
311 | #endif |
312 | ||
1da177e4 LT |
313 | /* Get the TIME_WAIT timeout firing. */ |
314 | if (timeo < rto) | |
315 | timeo = rto; | |
316 | ||
d82bae12 SHY |
317 | if (state == TCP_TIME_WAIT) |
318 | timeo = TCP_TIMEWAIT_LEN; | |
1da177e4 | 319 | |
cfac7f83 ED |
320 | /* tw_timer is pinned, so we need to make sure BH are disabled |
321 | * in following section, otherwise timer handler could run before | |
322 | * we complete the initialization. | |
323 | */ | |
324 | local_bh_disable(); | |
789f558c | 325 | inet_twsk_schedule(tw, timeo); |
ec94c269 ED |
326 | /* Linkage updates. |
327 | * Note that access to tw after this point is illegal. | |
328 | */ | |
329 | inet_twsk_hashdance(tw, sk, &tcp_hashinfo); | |
cfac7f83 | 330 | local_bh_enable(); |
1da177e4 LT |
331 | } else { |
332 | /* Sorry, if we're out of memory, just CLOSE this | |
333 | * socket up. We've got bigger problems than | |
334 | * non-graceful socket closings. | |
335 | */ | |
c10d9310 | 336 | NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPTIMEWAITOVERFLOW); |
1da177e4 LT |
337 | } |
338 | ||
339 | tcp_update_metrics(sk); | |
340 | tcp_done(sk); | |
341 | } | |
cc35c88a | 342 | EXPORT_SYMBOL(tcp_time_wait); |
1da177e4 | 343 | |
cfb6eeb4 YH |
344 | void tcp_twsk_destructor(struct sock *sk) |
345 | { | |
b6242b9b | 346 | #ifdef CONFIG_TCP_MD5SIG |
6aedbf98 ED |
347 | if (static_branch_unlikely(&tcp_md5_needed)) { |
348 | struct tcp_timewait_sock *twsk = tcp_twsk(sk); | |
2397849b | 349 | |
6aedbf98 ED |
350 | if (twsk->tw_md5_key) |
351 | kfree_rcu(twsk->tw_md5_key, rcu); | |
352 | } | |
cfb6eeb4 YH |
353 | #endif |
354 | } | |
cfb6eeb4 YH |
355 | EXPORT_SYMBOL_GPL(tcp_twsk_destructor); |
356 | ||
b1964b5f ED |
357 | /* Warning : This function is called without sk_listener being locked. |
358 | * Be sure to read socket fields once, as their value could change under us. | |
359 | */ | |
843f4a55 | 360 | void tcp_openreq_init_rwin(struct request_sock *req, |
b1964b5f ED |
361 | const struct sock *sk_listener, |
362 | const struct dst_entry *dst) | |
843f4a55 YC |
363 | { |
364 | struct inet_request_sock *ireq = inet_rsk(req); | |
b1964b5f | 365 | const struct tcp_sock *tp = tcp_sk(sk_listener); |
b1964b5f | 366 | int full_space = tcp_full_space(sk_listener); |
b1964b5f ED |
367 | u32 window_clamp; |
368 | __u8 rcv_wscale; | |
13d3b1eb | 369 | u32 rcv_wnd; |
3541f9e8 | 370 | int mss; |
843f4a55 | 371 | |
3541f9e8 | 372 | mss = tcp_mss_clamp(tp, dst_metric_advmss(dst)); |
b1964b5f | 373 | window_clamp = READ_ONCE(tp->window_clamp); |
843f4a55 | 374 | /* Set this up on the first call only */ |
ed53d0ab | 375 | req->rsk_window_clamp = window_clamp ? : dst_metric(dst, RTAX_WINDOW); |
843f4a55 YC |
376 | |
377 | /* limit the window selection if the user enforce a smaller rx buffer */ | |
b1964b5f | 378 | if (sk_listener->sk_userlocks & SOCK_RCVBUF_LOCK && |
ed53d0ab ED |
379 | (req->rsk_window_clamp > full_space || req->rsk_window_clamp == 0)) |
380 | req->rsk_window_clamp = full_space; | |
843f4a55 | 381 | |
13d3b1eb LB |
382 | rcv_wnd = tcp_rwnd_init_bpf((struct sock *)req); |
383 | if (rcv_wnd == 0) | |
384 | rcv_wnd = dst_metric(dst, RTAX_INITRWND); | |
385 | else if (full_space < rcv_wnd * mss) | |
386 | full_space = rcv_wnd * mss; | |
387 | ||
843f4a55 | 388 | /* tcp_full_space because it is guaranteed to be the first packet */ |
ceef9ab6 | 389 | tcp_select_initial_window(sk_listener, full_space, |
843f4a55 | 390 | mss - (ireq->tstamp_ok ? TCPOLEN_TSTAMP_ALIGNED : 0), |
ed53d0ab ED |
391 | &req->rsk_rcv_wnd, |
392 | &req->rsk_window_clamp, | |
843f4a55 YC |
393 | ireq->wscale_ok, |
394 | &rcv_wscale, | |
13d3b1eb | 395 | rcv_wnd); |
843f4a55 YC |
396 | ireq->rcv_wscale = rcv_wscale; |
397 | } | |
398 | EXPORT_SYMBOL(tcp_openreq_init_rwin); | |
399 | ||
735d3831 FW |
400 | static void tcp_ecn_openreq_child(struct tcp_sock *tp, |
401 | const struct request_sock *req) | |
bdf1ee5d IJ |
402 | { |
403 | tp->ecn_flags = inet_rsk(req)->ecn_ok ? TCP_ECN_OK : 0; | |
404 | } | |
405 | ||
81164413 DB |
406 | void tcp_ca_openreq_child(struct sock *sk, const struct dst_entry *dst) |
407 | { | |
408 | struct inet_connection_sock *icsk = inet_csk(sk); | |
409 | u32 ca_key = dst_metric(dst, RTAX_CC_ALGO); | |
410 | bool ca_got_dst = false; | |
411 | ||
412 | if (ca_key != TCP_CA_UNSPEC) { | |
413 | const struct tcp_congestion_ops *ca; | |
414 | ||
415 | rcu_read_lock(); | |
416 | ca = tcp_ca_find_key(ca_key); | |
417 | if (likely(ca && try_module_get(ca->owner))) { | |
418 | icsk->icsk_ca_dst_locked = tcp_ca_dst_locked(dst); | |
419 | icsk->icsk_ca_ops = ca; | |
420 | ca_got_dst = true; | |
421 | } | |
422 | rcu_read_unlock(); | |
423 | } | |
424 | ||
9f950415 NC |
425 | /* If no valid choice made yet, assign current system default ca. */ |
426 | if (!ca_got_dst && | |
427 | (!icsk->icsk_ca_setsockopt || | |
428 | !try_module_get(icsk->icsk_ca_ops->owner))) | |
81164413 DB |
429 | tcp_assign_congestion_control(sk); |
430 | ||
431 | tcp_set_ca_state(sk, TCP_CA_Open); | |
432 | } | |
433 | EXPORT_SYMBOL_GPL(tcp_ca_openreq_child); | |
434 | ||
60e2a778 UB |
435 | static void smc_check_reset_syn_req(struct tcp_sock *oldtp, |
436 | struct request_sock *req, | |
437 | struct tcp_sock *newtp) | |
438 | { | |
439 | #if IS_ENABLED(CONFIG_SMC) | |
440 | struct inet_request_sock *ireq; | |
441 | ||
442 | if (static_branch_unlikely(&tcp_have_smc)) { | |
443 | ireq = inet_rsk(req); | |
444 | if (oldtp->syn_smc && !ireq->smc_ok) | |
445 | newtp->syn_smc = 0; | |
446 | } | |
447 | #endif | |
448 | } | |
449 | ||
1da177e4 LT |
450 | /* This is not only more efficient than what we used to do, it eliminates |
451 | * a lot of code duplication between IPv4/IPv6 SYN recv processing. -DaveM | |
452 | * | |
453 | * Actually, we could lots of memory writes here. tp of listening | |
454 | * socket contains all necessary default parameters. | |
455 | */ | |
c28c6f04 ED |
456 | struct sock *tcp_create_openreq_child(const struct sock *sk, |
457 | struct request_sock *req, | |
458 | struct sk_buff *skb) | |
1da177e4 | 459 | { |
e56c57d0 | 460 | struct sock *newsk = inet_csk_clone_lock(sk, req, GFP_ATOMIC); |
242b1bbe ED |
461 | const struct inet_request_sock *ireq = inet_rsk(req); |
462 | struct tcp_request_sock *treq = tcp_rsk(req); | |
463 | struct inet_connection_sock *newicsk; | |
464 | struct tcp_sock *oldtp, *newtp; | |
dba7d9b8 | 465 | u32 seq; |
242b1bbe ED |
466 | |
467 | if (!newsk) | |
468 | return NULL; | |
469 | ||
470 | newicsk = inet_csk(newsk); | |
471 | newtp = tcp_sk(newsk); | |
472 | oldtp = tcp_sk(sk); | |
473 | ||
474 | smc_check_reset_syn_req(oldtp, req, newtp); | |
475 | ||
476 | /* Now setup tcp_sock */ | |
477 | newtp->pred_flags = 0; | |
478 | ||
dba7d9b8 ED |
479 | seq = treq->rcv_isn + 1; |
480 | newtp->rcv_wup = seq; | |
7db48e98 | 481 | WRITE_ONCE(newtp->copied_seq, seq); |
dba7d9b8 | 482 | WRITE_ONCE(newtp->rcv_nxt, seq); |
242b1bbe ED |
483 | newtp->segs_in = 1; |
484 | ||
485 | newtp->snd_sml = newtp->snd_una = | |
486 | newtp->snd_nxt = newtp->snd_up = treq->snt_isn + 1; | |
487 | ||
488 | INIT_LIST_HEAD(&newtp->tsq_node); | |
489 | INIT_LIST_HEAD(&newtp->tsorted_sent_queue); | |
490 | ||
491 | tcp_init_wl(newtp, treq->rcv_isn); | |
492 | ||
242b1bbe | 493 | minmax_reset(&newtp->rtt_min, tcp_jiffies32, ~0U); |
242b1bbe ED |
494 | newicsk->icsk_ack.lrcvtime = tcp_jiffies32; |
495 | ||
242b1bbe ED |
496 | newtp->lsndtime = tcp_jiffies32; |
497 | newsk->sk_txhash = treq->txhash; | |
242b1bbe ED |
498 | newtp->total_retrans = req->num_retrans; |
499 | ||
242b1bbe | 500 | tcp_init_xmit_timers(newsk); |
0f317464 | 501 | WRITE_ONCE(newtp->write_seq, newtp->pushed_seq = treq->snt_isn + 1); |
242b1bbe | 502 | |
242b1bbe ED |
503 | if (sock_flag(newsk, SOCK_KEEPOPEN)) |
504 | inet_csk_reset_keepalive_timer(newsk, | |
505 | keepalive_time_when(newtp)); | |
506 | ||
507 | newtp->rx_opt.tstamp_ok = ireq->tstamp_ok; | |
508 | newtp->rx_opt.sack_ok = ireq->sack_ok; | |
509 | newtp->window_clamp = req->rsk_window_clamp; | |
510 | newtp->rcv_ssthresh = req->rsk_rcv_wnd; | |
511 | newtp->rcv_wnd = req->rsk_rcv_wnd; | |
512 | newtp->rx_opt.wscale_ok = ireq->wscale_ok; | |
513 | if (newtp->rx_opt.wscale_ok) { | |
514 | newtp->rx_opt.snd_wscale = ireq->snd_wscale; | |
515 | newtp->rx_opt.rcv_wscale = ireq->rcv_wscale; | |
516 | } else { | |
517 | newtp->rx_opt.snd_wscale = newtp->rx_opt.rcv_wscale = 0; | |
518 | newtp->window_clamp = min(newtp->window_clamp, 65535U); | |
519 | } | |
520 | newtp->snd_wnd = ntohs(tcp_hdr(skb)->window) << newtp->rx_opt.snd_wscale; | |
521 | newtp->max_window = newtp->snd_wnd; | |
522 | ||
523 | if (newtp->rx_opt.tstamp_ok) { | |
524 | newtp->rx_opt.ts_recent = req->ts_recent; | |
cca9bab1 | 525 | newtp->rx_opt.ts_recent_stamp = ktime_get_seconds(); |
242b1bbe ED |
526 | newtp->tcp_header_len = sizeof(struct tcphdr) + TCPOLEN_TSTAMP_ALIGNED; |
527 | } else { | |
528 | newtp->rx_opt.ts_recent_stamp = 0; | |
529 | newtp->tcp_header_len = sizeof(struct tcphdr); | |
530 | } | |
336c39a0 YC |
531 | if (req->num_timeout) { |
532 | newtp->undo_marker = treq->snt_isn; | |
533 | newtp->retrans_stamp = div_u64(treq->snt_synack, | |
534 | USEC_PER_SEC / TCP_TS_HZ); | |
535 | } | |
242b1bbe | 536 | newtp->tsoffset = treq->ts_off; |
cfb6eeb4 | 537 | #ifdef CONFIG_TCP_MD5SIG |
242b1bbe ED |
538 | newtp->md5sig_info = NULL; /*XXX*/ |
539 | if (newtp->af_specific->md5_lookup(sk, newsk)) | |
540 | newtp->tcp_header_len += TCPOLEN_MD5SIG_ALIGNED; | |
cfb6eeb4 | 541 | #endif |
242b1bbe ED |
542 | if (skb->len >= TCP_MSS_DEFAULT + newtp->tcp_header_len) |
543 | newicsk->icsk_ack.last_seg_size = skb->len - newtp->tcp_header_len; | |
544 | newtp->rx_opt.mss_clamp = req->mss; | |
545 | tcp_ecn_openreq_child(newtp, req); | |
546 | newtp->fastopen_req = NULL; | |
d983ea6f | 547 | RCU_INIT_POINTER(newtp->fastopen_rsk, NULL); |
242b1bbe ED |
548 | |
549 | __TCP_INC_STATS(sock_net(sk), TCP_MIB_PASSIVEOPENS); | |
550 | ||
1da177e4 LT |
551 | return newsk; |
552 | } | |
4bc2f18b | 553 | EXPORT_SYMBOL(tcp_create_openreq_child); |
1da177e4 | 554 | |
e905a9ed | 555 | /* |
8336886f JC |
556 | * Process an incoming packet for SYN_RECV sockets represented as a |
557 | * request_sock. Normally sk is the listener socket but for TFO it | |
558 | * points to the child socket. | |
559 | * | |
560 | * XXX (TFO) - The current impl contains a special check for ack | |
561 | * validation and inside tcp_v4_reqsk_send_ack(). Can we do better? | |
4308fc58 AC |
562 | * |
563 | * We don't need to initialize tmp_opt.sack_ok as we don't use the results | |
1da177e4 LT |
564 | */ |
565 | ||
5a5f3a8d | 566 | struct sock *tcp_check_req(struct sock *sk, struct sk_buff *skb, |
60236fdd | 567 | struct request_sock *req, |
e0f9759f | 568 | bool fastopen, bool *req_stolen) |
1da177e4 | 569 | { |
4957faad | 570 | struct tcp_options_received tmp_opt; |
4957faad | 571 | struct sock *child; |
aa8223c7 | 572 | const struct tcphdr *th = tcp_hdr(skb); |
714e85be | 573 | __be32 flg = tcp_flag_word(th) & (TCP_FLAG_RST|TCP_FLAG_SYN|TCP_FLAG_ACK); |
a2a385d6 | 574 | bool paws_reject = false; |
5e0724d0 | 575 | bool own_req; |
1da177e4 | 576 | |
bb5b7c11 DM |
577 | tmp_opt.saw_tstamp = 0; |
578 | if (th->doff > (sizeof(struct tcphdr)>>2)) { | |
eed29f17 | 579 | tcp_parse_options(sock_net(sk), skb, &tmp_opt, 0, NULL); |
1da177e4 LT |
580 | |
581 | if (tmp_opt.saw_tstamp) { | |
582 | tmp_opt.ts_recent = req->ts_recent; | |
95a22cae FW |
583 | if (tmp_opt.rcv_tsecr) |
584 | tmp_opt.rcv_tsecr -= tcp_rsk(req)->ts_off; | |
1da177e4 LT |
585 | /* We do not store true stamp, but it is not required, |
586 | * it can be estimated (approximately) | |
587 | * from another data. | |
588 | */ | |
cca9bab1 | 589 | tmp_opt.ts_recent_stamp = ktime_get_seconds() - ((TCP_TIMEOUT_INIT/HZ)<<req->num_timeout); |
c887e6d2 | 590 | paws_reject = tcp_paws_reject(&tmp_opt, th->rst); |
1da177e4 LT |
591 | } |
592 | } | |
593 | ||
594 | /* Check for pure retransmitted SYN. */ | |
2e6599cb | 595 | if (TCP_SKB_CB(skb)->seq == tcp_rsk(req)->rcv_isn && |
1da177e4 LT |
596 | flg == TCP_FLAG_SYN && |
597 | !paws_reject) { | |
598 | /* | |
599 | * RFC793 draws (Incorrectly! It was fixed in RFC1122) | |
600 | * this case on figure 6 and figure 8, but formal | |
601 | * protocol description says NOTHING. | |
602 | * To be more exact, it says that we should send ACK, | |
603 | * because this segment (at least, if it has no data) | |
604 | * is out of window. | |
605 | * | |
606 | * CONCLUSION: RFC793 (even with RFC1122) DOES NOT | |
607 | * describe SYN-RECV state. All the description | |
608 | * is wrong, we cannot believe to it and should | |
609 | * rely only on common sense and implementation | |
610 | * experience. | |
611 | * | |
612 | * Enforce "SYN-ACK" according to figure 8, figure 6 | |
613 | * of RFC793, fixed by RFC1122. | |
8336886f JC |
614 | * |
615 | * Note that even if there is new data in the SYN packet | |
616 | * they will be thrown away too. | |
cd75eff6 YC |
617 | * |
618 | * Reset timer after retransmitting SYNACK, similar to | |
619 | * the idea of fast retransmit in recovery. | |
1da177e4 | 620 | */ |
a9b2c06d NC |
621 | if (!tcp_oow_rate_limited(sock_net(sk), skb, |
622 | LINUX_MIB_TCPACKSKIPPEDSYNRECV, | |
623 | &tcp_rsk(req)->last_oow_ack_time) && | |
624 | ||
dd929c1b ED |
625 | !inet_rtx_syn_ack(sk, req)) { |
626 | unsigned long expires = jiffies; | |
627 | ||
628 | expires += min(TCP_TIMEOUT_INIT << req->num_timeout, | |
629 | TCP_RTO_MAX); | |
630 | if (!fastopen) | |
631 | mod_timer_pending(&req->rsk_timer, expires); | |
632 | else | |
633 | req->rsk_timer.expires = expires; | |
634 | } | |
1da177e4 LT |
635 | return NULL; |
636 | } | |
637 | ||
638 | /* Further reproduces section "SEGMENT ARRIVES" | |
639 | for state SYN-RECEIVED of RFC793. | |
640 | It is broken, however, it does not work only | |
641 | when SYNs are crossed. | |
642 | ||
643 | You would think that SYN crossing is impossible here, since | |
644 | we should have a SYN_SENT socket (from connect()) on our end, | |
645 | but this is not true if the crossed SYNs were sent to both | |
646 | ends by a malicious third party. We must defend against this, | |
647 | and to do that we first verify the ACK (as per RFC793, page | |
648 | 36) and reset if it is invalid. Is this a true full defense? | |
649 | To convince ourselves, let us consider a way in which the ACK | |
650 | test can still pass in this 'malicious crossed SYNs' case. | |
651 | Malicious sender sends identical SYNs (and thus identical sequence | |
652 | numbers) to both A and B: | |
653 | ||
654 | A: gets SYN, seq=7 | |
655 | B: gets SYN, seq=7 | |
656 | ||
657 | By our good fortune, both A and B select the same initial | |
658 | send sequence number of seven :-) | |
659 | ||
660 | A: sends SYN|ACK, seq=7, ack_seq=8 | |
661 | B: sends SYN|ACK, seq=7, ack_seq=8 | |
662 | ||
663 | So we are now A eating this SYN|ACK, ACK test passes. So | |
664 | does sequence test, SYN is truncated, and thus we consider | |
665 | it a bare ACK. | |
666 | ||
ec0a1966 DM |
667 | If icsk->icsk_accept_queue.rskq_defer_accept, we silently drop this |
668 | bare ACK. Otherwise, we create an established connection. Both | |
669 | ends (listening sockets) accept the new incoming connection and try | |
670 | to talk to each other. 8-) | |
1da177e4 LT |
671 | |
672 | Note: This case is both harmless, and rare. Possibility is about the | |
673 | same as us discovering intelligent life on another plant tomorrow. | |
674 | ||
675 | But generally, we should (RFC lies!) to accept ACK | |
676 | from SYNACK both here and in tcp_rcv_state_process(). | |
677 | tcp_rcv_state_process() does not, hence, we do not too. | |
678 | ||
679 | Note that the case is absolutely generic: | |
680 | we cannot optimize anything here without | |
681 | violating protocol. All the checks must be made | |
682 | before attempt to create socket. | |
683 | */ | |
684 | ||
685 | /* RFC793 page 36: "If the connection is in any non-synchronized state ... | |
686 | * and the incoming segment acknowledges something not yet | |
caa20d9a | 687 | * sent (the segment carries an unacceptable ACK) ... |
1da177e4 LT |
688 | * a reset is sent." |
689 | * | |
8336886f JC |
690 | * Invalid ACK: reset will be sent by listening socket. |
691 | * Note that the ACK validity check for a Fast Open socket is done | |
692 | * elsewhere and is checked directly against the child socket rather | |
693 | * than req because user data may have been sent out. | |
1da177e4 | 694 | */ |
8336886f | 695 | if ((flg & TCP_FLAG_ACK) && !fastopen && |
435cf559 | 696 | (TCP_SKB_CB(skb)->ack_seq != |
1a2c6181 | 697 | tcp_rsk(req)->snt_isn + 1)) |
1da177e4 LT |
698 | return sk; |
699 | ||
700 | /* Also, it would be not so bad idea to check rcv_tsecr, which | |
701 | * is essentially ACK extension and too early or too late values | |
702 | * should cause reset in unsynchronized states. | |
703 | */ | |
704 | ||
705 | /* RFC793: "first check sequence number". */ | |
706 | ||
707 | if (paws_reject || !tcp_in_window(TCP_SKB_CB(skb)->seq, TCP_SKB_CB(skb)->end_seq, | |
ed53d0ab | 708 | tcp_rsk(req)->rcv_nxt, tcp_rsk(req)->rcv_nxt + req->rsk_rcv_wnd)) { |
1da177e4 | 709 | /* Out of window: send ACK and drop. */ |
4ce7e93c ED |
710 | if (!(flg & TCP_FLAG_RST) && |
711 | !tcp_oow_rate_limited(sock_net(sk), skb, | |
712 | LINUX_MIB_TCPACKSKIPPEDSYNRECV, | |
713 | &tcp_rsk(req)->last_oow_ack_time)) | |
6edafaaf | 714 | req->rsk_ops->send_ack(sk, skb, req); |
1da177e4 | 715 | if (paws_reject) |
02a1d6e7 | 716 | __NET_INC_STATS(sock_net(sk), LINUX_MIB_PAWSESTABREJECTED); |
1da177e4 LT |
717 | return NULL; |
718 | } | |
719 | ||
720 | /* In sequence, PAWS is OK. */ | |
721 | ||
8336886f | 722 | if (tmp_opt.saw_tstamp && !after(TCP_SKB_CB(skb)->seq, tcp_rsk(req)->rcv_nxt)) |
2aaab9a0 | 723 | req->ts_recent = tmp_opt.rcv_tsval; |
1da177e4 | 724 | |
2aaab9a0 AL |
725 | if (TCP_SKB_CB(skb)->seq == tcp_rsk(req)->rcv_isn) { |
726 | /* Truncate SYN, it is out of window starting | |
727 | at tcp_rsk(req)->rcv_isn + 1. */ | |
728 | flg &= ~TCP_FLAG_SYN; | |
729 | } | |
1da177e4 | 730 | |
2aaab9a0 AL |
731 | /* RFC793: "second check the RST bit" and |
732 | * "fourth, check the SYN bit" | |
733 | */ | |
734 | if (flg & (TCP_FLAG_RST|TCP_FLAG_SYN)) { | |
90bbcc60 | 735 | __TCP_INC_STATS(sock_net(sk), TCP_MIB_ATTEMPTFAILS); |
2aaab9a0 AL |
736 | goto embryonic_reset; |
737 | } | |
1da177e4 | 738 | |
2aaab9a0 AL |
739 | /* ACK sequence verified above, just make sure ACK is |
740 | * set. If ACK not set, just silently drop the packet. | |
8336886f JC |
741 | * |
742 | * XXX (TFO) - if we ever allow "data after SYN", the | |
743 | * following check needs to be removed. | |
2aaab9a0 AL |
744 | */ |
745 | if (!(flg & TCP_FLAG_ACK)) | |
746 | return NULL; | |
ec0a1966 | 747 | |
8336886f JC |
748 | /* For Fast Open no more processing is needed (sk is the |
749 | * child socket). | |
750 | */ | |
751 | if (fastopen) | |
752 | return sk; | |
753 | ||
d1b99ba4 | 754 | /* While TCP_DEFER_ACCEPT is active, drop bare ACK. */ |
e6c022a4 | 755 | if (req->num_timeout < inet_csk(sk)->icsk_accept_queue.rskq_defer_accept && |
2aaab9a0 AL |
756 | TCP_SKB_CB(skb)->end_seq == tcp_rsk(req)->rcv_isn + 1) { |
757 | inet_rsk(req)->acked = 1; | |
02a1d6e7 | 758 | __NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPDEFERACCEPTDROP); |
2aaab9a0 AL |
759 | return NULL; |
760 | } | |
761 | ||
762 | /* OK, ACK is valid, create big socket and | |
763 | * feed this segment to it. It will repeat all | |
764 | * the tests. THIS SEGMENT MUST MOVE SOCKET TO | |
765 | * ESTABLISHED STATE. If it will be dropped after | |
766 | * socket is created, wait for troubles. | |
767 | */ | |
5e0724d0 ED |
768 | child = inet_csk(sk)->icsk_af_ops->syn_recv_sock(sk, skb, req, NULL, |
769 | req, &own_req); | |
51456b29 | 770 | if (!child) |
2aaab9a0 | 771 | goto listen_overflow; |
1da177e4 | 772 | |
6bcfd7f8 | 773 | sock_rps_save_rxhash(child, skb); |
0f1c28ae | 774 | tcp_synack_rtt_meas(child, req); |
e0f9759f | 775 | *req_stolen = !own_req; |
5e0724d0 | 776 | return inet_csk_complete_hashdance(sk, child, req, own_req); |
1da177e4 | 777 | |
2aaab9a0 | 778 | listen_overflow: |
65c9410c | 779 | if (!sock_net(sk)->ipv4.sysctl_tcp_abort_on_overflow) { |
2aaab9a0 AL |
780 | inet_rsk(req)->acked = 1; |
781 | return NULL; | |
782 | } | |
1da177e4 | 783 | |
2aaab9a0 | 784 | embryonic_reset: |
8336886f JC |
785 | if (!(flg & TCP_FLAG_RST)) { |
786 | /* Received a bad SYN pkt - for TFO We try not to reset | |
787 | * the local connection unless it's really necessary to | |
788 | * avoid becoming vulnerable to outside attack aiming at | |
789 | * resetting legit local connections. | |
790 | */ | |
2aaab9a0 | 791 | req->rsk_ops->send_reset(sk, skb); |
8336886f JC |
792 | } else if (fastopen) { /* received a valid RST pkt */ |
793 | reqsk_fastopen_remove(sk, req, true); | |
794 | tcp_reset(sk); | |
795 | } | |
796 | if (!fastopen) { | |
52452c54 | 797 | inet_csk_reqsk_queue_drop(sk, req); |
02a1d6e7 | 798 | __NET_INC_STATS(sock_net(sk), LINUX_MIB_EMBRYONICRSTS); |
8336886f | 799 | } |
2aaab9a0 | 800 | return NULL; |
1da177e4 | 801 | } |
4bc2f18b | 802 | EXPORT_SYMBOL(tcp_check_req); |
1da177e4 LT |
803 | |
804 | /* | |
805 | * Queue segment on the new socket if the new socket is active, | |
806 | * otherwise we just shortcircuit this and continue with | |
807 | * the new socket. | |
8336886f JC |
808 | * |
809 | * For the vast majority of cases child->sk_state will be TCP_SYN_RECV | |
810 | * when entering. But other states are possible due to a race condition | |
811 | * where after __inet_lookup_established() fails but before the listener | |
812 | * locked is obtained, other packets cause the same connection to | |
813 | * be created. | |
1da177e4 LT |
814 | */ |
815 | ||
816 | int tcp_child_process(struct sock *parent, struct sock *child, | |
817 | struct sk_buff *skb) | |
818 | { | |
819 | int ret = 0; | |
820 | int state = child->sk_state; | |
821 | ||
e5907459 AD |
822 | /* record NAPI ID of child */ |
823 | sk_mark_napi_id(child, skb); | |
824 | ||
a44d6eac | 825 | tcp_segs_in(tcp_sk(child), skb); |
1da177e4 | 826 | if (!sock_owned_by_user(child)) { |
72ab4a86 | 827 | ret = tcp_rcv_state_process(child, skb); |
1da177e4 LT |
828 | /* Wakeup parent, send SIGIO */ |
829 | if (state == TCP_SYN_RECV && child->sk_state != state) | |
676d2369 | 830 | parent->sk_data_ready(parent); |
1da177e4 LT |
831 | } else { |
832 | /* Alas, it is possible again, because we do lookup | |
833 | * in main socket hash table and lock on listening | |
834 | * socket does not protect us more. | |
835 | */ | |
a3a858ff | 836 | __sk_add_backlog(child, skb); |
1da177e4 LT |
837 | } |
838 | ||
839 | bh_unlock_sock(child); | |
840 | sock_put(child); | |
841 | return ret; | |
842 | } | |
1da177e4 | 843 | EXPORT_SYMBOL(tcp_child_process); |