1 // SPDX-License-Identifier: GPL-2.0-only
2 /* (C) 1999-2001 Paul `Rusty' Russell
3 * (C) 2002-2004 Netfilter Core Team <coreteam@netfilter.org>
4 * (C) 2002-2013 Jozsef Kadlecsik <kadlec@netfilter.org>
5 * (C) 2006-2012 Patrick McHardy <kaber@trash.net>
6 */
7
8 #include <linux/types.h>
9 #include <linux/timer.h>
10 #include <linux/module.h>
11 #include <linux/in.h>
12 #include <linux/tcp.h>
13 #include <linux/spinlock.h>
14 #include <linux/skbuff.h>
15 #include <linux/ipv6.h>
16 #include <net/ip6_checksum.h>
17 #include <asm/unaligned.h>
18
19 #include <net/tcp.h>
20
21 #include <linux/netfilter.h>
22 #include <linux/netfilter_ipv4.h>
23 #include <linux/netfilter_ipv6.h>
24 #include <net/netfilter/nf_conntrack.h>
25 #include <net/netfilter/nf_conntrack_l4proto.h>
26 #include <net/netfilter/nf_conntrack_ecache.h>
27 #include <net/netfilter/nf_conntrack_seqadj.h>
28 #include <net/netfilter/nf_conntrack_synproxy.h>
29 #include <net/netfilter/nf_conntrack_timeout.h>
30 #include <net/netfilter/nf_log.h>
31 #include <net/netfilter/ipv4/nf_conntrack_ipv4.h>
32 #include <net/netfilter/ipv6/nf_conntrack_ipv6.h>
33
34 /* "Be conservative in what you do,
35 be liberal in what you accept from others."
36 If it's non-zero, we mark only out of window RST segments as INVALID. */
37 static int nf_ct_tcp_be_liberal __read_mostly = 0;
38
39 /* If it is set to zero, we disable picking up already established
40 connections. */
41 static int nf_ct_tcp_loose __read_mostly = 1;
42
43 /* Max number of the retransmitted packets without receiving an (acceptable)
44 ACK from the destination. If this number is reached, a shorter timer
45 will be started. */
46 static int nf_ct_tcp_max_retrans __read_mostly = 3;
47
48 /* FIXME: Examine ipfilter's timeouts and conntrack transitions more
49 closely. They're more complex. --RR */
50
51 static const char *const tcp_conntrack_names[] = {
52 "NONE",
53 "SYN_SENT",
54 "SYN_RECV",
55 "ESTABLISHED",
56 "FIN_WAIT",
57 "CLOSE_WAIT",
58 "LAST_ACK",
59 "TIME_WAIT",
60 "CLOSE",
61 "SYN_SENT2",
62 };
63
64 #define SECS * HZ
65 #define MINS * 60 SECS
66 #define HOURS * 60 MINS
67 #define DAYS * 24 HOURS
68
69 static const unsigned int tcp_timeouts[TCP_CONNTRACK_TIMEOUT_MAX] = {
70 [TCP_CONNTRACK_SYN_SENT] = 2 MINS,
71 [TCP_CONNTRACK_SYN_RECV] = 60 SECS,
72 [TCP_CONNTRACK_ESTABLISHED] = 5 DAYS,
73 [TCP_CONNTRACK_FIN_WAIT] = 2 MINS,
74 [TCP_CONNTRACK_CLOSE_WAIT] = 60 SECS,
75 [TCP_CONNTRACK_LAST_ACK] = 30 SECS,
76 [TCP_CONNTRACK_TIME_WAIT] = 2 MINS,
77 [TCP_CONNTRACK_CLOSE] = 10 SECS,
78 [TCP_CONNTRACK_SYN_SENT2] = 2 MINS,
79 /* RFC1122 says the R2 limit should be at least 100 seconds.
80 Linux uses 15 packets as limit, which corresponds
81 to ~13-30min depending on RTO. */
82 [TCP_CONNTRACK_RETRANS] = 5 MINS,
83 [TCP_CONNTRACK_UNACK] = 5 MINS,
84 };
85
86 #define sNO TCP_CONNTRACK_NONE
87 #define sSS TCP_CONNTRACK_SYN_SENT
88 #define sSR TCP_CONNTRACK_SYN_RECV
89 #define sES TCP_CONNTRACK_ESTABLISHED
90 #define sFW TCP_CONNTRACK_FIN_WAIT
91 #define sCW TCP_CONNTRACK_CLOSE_WAIT
92 #define sLA TCP_CONNTRACK_LAST_ACK
93 #define sTW TCP_CONNTRACK_TIME_WAIT
94 #define sCL TCP_CONNTRACK_CLOSE
95 #define sS2 TCP_CONNTRACK_SYN_SENT2
96 #define sIV TCP_CONNTRACK_MAX
97 #define sIG TCP_CONNTRACK_IGNORE
98
99 /* What TCP flags are set from RST/SYN/FIN/ACK. */
100 enum tcp_bit_set {
101 TCP_SYN_SET,
102 TCP_SYNACK_SET,
103 TCP_FIN_SET,
104 TCP_ACK_SET,
105 TCP_RST_SET,
106 TCP_NONE_SET,
107 };
108
109 /*
110 * The TCP state transition table needs a few words...
111 *
112 * We are the man in the middle. All the packets go through us
113 * but might get lost in transit to the destination.
114 * It is assumed that the destinations can't receive segments
115 * we haven't seen.
116 *
117 * The checked segment is in window, but our windows are *not*
118 * equivalent with the ones of the sender/receiver. We always
119 * try to guess the state of the current sender.
120 *
121 * The meaning of the states are:
122 *
123 * NONE: initial state
124 * SYN_SENT: SYN-only packet seen
125 * SYN_SENT2: SYN-only packet seen from reply dir, simultaneous open
126 * SYN_RECV: SYN-ACK packet seen
127 * ESTABLISHED: ACK packet seen
128 * FIN_WAIT: FIN packet seen
129 * CLOSE_WAIT: ACK seen (after FIN)
130 * LAST_ACK: FIN seen (after FIN)
131 * TIME_WAIT: last ACK seen
132 * CLOSE: closed connection (RST)
133 *
134 * Packets marked as IGNORED (sIG):
135 * if they may be either invalid or valid
136 * and the receiver may send back a connection
137 * closing RST or a SYN/ACK.
138 *
139 * Packets marked as INVALID (sIV):
140 * if we regard them as truly invalid packets
141 */
142 static const u8 tcp_conntracks[2][6][TCP_CONNTRACK_MAX] = {
143 {
144 /* ORIGINAL */
145 /* sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sS2 */
146 /*syn*/ { sSS, sSS, sIG, sIG, sIG, sIG, sIG, sSS, sSS, sS2 },
147 /*
148 * sNO -> sSS Initialize a new connection
149 * sSS -> sSS Retransmitted SYN
150 * sS2 -> sS2 Late retransmitted SYN
151 * sSR -> sIG
152 * sES -> sIG Error: SYNs in window outside the SYN_SENT state
153 * are errors. Receiver will reply with RST
154 * and close the connection.
155 * Or we are not in sync and hold a dead connection.
156 * sFW -> sIG
157 * sCW -> sIG
158 * sLA -> sIG
159 * sTW -> sSS Reopened connection (RFC 1122).
160 * sCL -> sSS
161 */
162 /* sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sS2 */
163 /*synack*/ { sIV, sIV, sSR, sIV, sIV, sIV, sIV, sIV, sIV, sSR },
164 /*
165 * sNO -> sIV Too late and no reason to do anything
166 * sSS -> sIV Client can't send SYN and then SYN/ACK
167 * sS2 -> sSR SYN/ACK sent to SYN2 in simultaneous open
168 * sSR -> sSR Late retransmitted SYN/ACK in simultaneous open
169 * sES -> sIV Invalid SYN/ACK packets sent by the client
170 * sFW -> sIV
171 * sCW -> sIV
172 * sLA -> sIV
173 * sTW -> sIV
174 * sCL -> sIV
175 */
176 /* sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sS2 */
177 /*fin*/ { sIV, sIV, sFW, sFW, sLA, sLA, sLA, sTW, sCL, sIV },
178 /*
179 * sNO -> sIV Too late and no reason to do anything...
180 * sSS -> sIV Client migth not send FIN in this state:
181 * we enforce waiting for a SYN/ACK reply first.
182 * sS2 -> sIV
183 * sSR -> sFW Close started.
184 * sES -> sFW
185 * sFW -> sLA FIN seen in both directions, waiting for
186 * the last ACK.
187 * Migth be a retransmitted FIN as well...
188 * sCW -> sLA
189 * sLA -> sLA Retransmitted FIN. Remain in the same state.
190 * sTW -> sTW
191 * sCL -> sCL
192 */
193 /* sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sS2 */
194 /*ack*/ { sES, sIV, sES, sES, sCW, sCW, sTW, sTW, sCL, sIV },
195 /*
196 * sNO -> sES Assumed.
197 * sSS -> sIV ACK is invalid: we haven't seen a SYN/ACK yet.
198 * sS2 -> sIV
199 * sSR -> sES Established state is reached.
200 * sES -> sES :-)
201 * sFW -> sCW Normal close request answered by ACK.
202 * sCW -> sCW
203 * sLA -> sTW Last ACK detected (RFC5961 challenged)
204 * sTW -> sTW Retransmitted last ACK. Remain in the same state.
205 * sCL -> sCL
206 */
207 /* sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sS2 */
208 /*rst*/ { sIV, sCL, sCL, sCL, sCL, sCL, sCL, sCL, sCL, sCL },
209 /*none*/ { sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV }
210 },
211 {
212 /* REPLY */
213 /* sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sS2 */
214 /*syn*/ { sIV, sS2, sIV, sIV, sIV, sIV, sIV, sSS, sIV, sS2 },
215 /*
216 * sNO -> sIV Never reached.
217 * sSS -> sS2 Simultaneous open
218 * sS2 -> sS2 Retransmitted simultaneous SYN
219 * sSR -> sIV Invalid SYN packets sent by the server
220 * sES -> sIV
221 * sFW -> sIV
222 * sCW -> sIV
223 * sLA -> sIV
224 * sTW -> sSS Reopened connection, but server may have switched role
225 * sCL -> sIV
226 */
227 /* sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sS2 */
228 /*synack*/ { sIV, sSR, sIG, sIG, sIG, sIG, sIG, sIG, sIG, sSR },
229 /*
230 * sSS -> sSR Standard open.
231 * sS2 -> sSR Simultaneous open
232 * sSR -> sIG Retransmitted SYN/ACK, ignore it.
233 * sES -> sIG Late retransmitted SYN/ACK?
234 * sFW -> sIG Might be SYN/ACK answering ignored SYN
235 * sCW -> sIG
236 * sLA -> sIG
237 * sTW -> sIG
238 * sCL -> sIG
239 */
240 /* sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sS2 */
241 /*fin*/ { sIV, sIV, sFW, sFW, sLA, sLA, sLA, sTW, sCL, sIV },
242 /*
243 * sSS -> sIV Server might not send FIN in this state.
244 * sS2 -> sIV
245 * sSR -> sFW Close started.
246 * sES -> sFW
247 * sFW -> sLA FIN seen in both directions.
248 * sCW -> sLA
249 * sLA -> sLA Retransmitted FIN.
250 * sTW -> sTW
251 * sCL -> sCL
252 */
253 /* sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sS2 */
254 /*ack*/ { sIV, sIG, sSR, sES, sCW, sCW, sTW, sTW, sCL, sIG },
255 /*
256 * sSS -> sIG Might be a half-open connection.
257 * sS2 -> sIG
258 * sSR -> sSR Might answer late resent SYN.
259 * sES -> sES :-)
260 * sFW -> sCW Normal close request answered by ACK.
261 * sCW -> sCW
262 * sLA -> sTW Last ACK detected (RFC5961 challenged)
263 * sTW -> sTW Retransmitted last ACK.
264 * sCL -> sCL
265 */
266 /* sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sS2 */
267 /*rst*/ { sIV, sCL, sCL, sCL, sCL, sCL, sCL, sCL, sCL, sCL },
268 /*none*/ { sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV }
269 }
270 };
271
272 #ifdef CONFIG_NF_CONNTRACK_PROCFS
273 /* Print out the private part of the conntrack. */
tcp_print_conntrack(struct seq_file * s,struct nf_conn * ct)274 static void tcp_print_conntrack(struct seq_file *s, struct nf_conn *ct)
275 {
276 if (test_bit(IPS_OFFLOAD_BIT, &ct->status))
277 return;
278
279 seq_printf(s, "%s ", tcp_conntrack_names[ct->proto.tcp.state]);
280 }
281 #endif
282
get_conntrack_index(const struct tcphdr * tcph)283 static unsigned int get_conntrack_index(const struct tcphdr *tcph)
284 {
285 if (tcph->rst) return TCP_RST_SET;
286 else if (tcph->syn) return (tcph->ack ? TCP_SYNACK_SET : TCP_SYN_SET);
287 else if (tcph->fin) return TCP_FIN_SET;
288 else if (tcph->ack) return TCP_ACK_SET;
289 else return TCP_NONE_SET;
290 }
291
292 /* TCP connection tracking based on 'Real Stateful TCP Packet Filtering
293 in IP Filter' by Guido van Rooij.
294
295 http://www.sane.nl/events/sane2000/papers.html
296 http://www.darkart.com/mirrors/www.obfuscation.org/ipf/
297
298 The boundaries and the conditions are changed according to RFC793:
299 the packet must intersect the window (i.e. segments may be
300 after the right or before the left edge) and thus receivers may ACK
301 segments after the right edge of the window.
302
303 td_maxend = max(sack + max(win,1)) seen in reply packets
304 td_maxwin = max(max(win, 1)) + (sack - ack) seen in sent packets
305 td_maxwin += seq + len - sender.td_maxend
306 if seq + len > sender.td_maxend
307 td_end = max(seq + len) seen in sent packets
308
309 I. Upper bound for valid data: seq <= sender.td_maxend
310 II. Lower bound for valid data: seq + len >= sender.td_end - receiver.td_maxwin
311 III. Upper bound for valid (s)ack: sack <= receiver.td_end
312 IV. Lower bound for valid (s)ack: sack >= receiver.td_end - MAXACKWINDOW
313
314 where sack is the highest right edge of sack block found in the packet
315 or ack in the case of packet without SACK option.
316
317 The upper bound limit for a valid (s)ack is not ignored -
318 we doesn't have to deal with fragments.
319 */
320
segment_seq_plus_len(__u32 seq,size_t len,unsigned int dataoff,const struct tcphdr * tcph)321 static inline __u32 segment_seq_plus_len(__u32 seq,
322 size_t len,
323 unsigned int dataoff,
324 const struct tcphdr *tcph)
325 {
326 /* XXX Should I use payload length field in IP/IPv6 header ?
327 * - YK */
328 return (seq + len - dataoff - tcph->doff*4
329 + (tcph->syn ? 1 : 0) + (tcph->fin ? 1 : 0));
330 }
331
332 /* Fixme: what about big packets? */
333 #define MAXACKWINCONST 66000
334 #define MAXACKWINDOW(sender) \
335 ((sender)->td_maxwin > MAXACKWINCONST ? (sender)->td_maxwin \
336 : MAXACKWINCONST)
337
338 /*
339 * Simplified tcp_parse_options routine from tcp_input.c
340 */
tcp_options(const struct sk_buff * skb,unsigned int dataoff,const struct tcphdr * tcph,struct ip_ct_tcp_state * state)341 static void tcp_options(const struct sk_buff *skb,
342 unsigned int dataoff,
343 const struct tcphdr *tcph,
344 struct ip_ct_tcp_state *state)
345 {
346 unsigned char buff[(15 * 4) - sizeof(struct tcphdr)];
347 const unsigned char *ptr;
348 int length = (tcph->doff*4) - sizeof(struct tcphdr);
349
350 if (!length)
351 return;
352
353 ptr = skb_header_pointer(skb, dataoff + sizeof(struct tcphdr),
354 length, buff);
355 BUG_ON(ptr == NULL);
356
357 state->td_scale = 0;
358 state->flags &= IP_CT_TCP_FLAG_BE_LIBERAL;
359
360 while (length > 0) {
361 int opcode=*ptr++;
362 int opsize;
363
364 switch (opcode) {
365 case TCPOPT_EOL:
366 return;
367 case TCPOPT_NOP: /* Ref: RFC 793 section 3.1 */
368 length--;
369 continue;
370 default:
371 if (length < 2)
372 return;
373 opsize=*ptr++;
374 if (opsize < 2) /* "silly options" */
375 return;
376 if (opsize > length)
377 return; /* don't parse partial options */
378
379 if (opcode == TCPOPT_SACK_PERM
380 && opsize == TCPOLEN_SACK_PERM)
381 state->flags |= IP_CT_TCP_FLAG_SACK_PERM;
382 else if (opcode == TCPOPT_WINDOW
383 && opsize == TCPOLEN_WINDOW) {
384 state->td_scale = *(u_int8_t *)ptr;
385
386 if (state->td_scale > TCP_MAX_WSCALE)
387 state->td_scale = TCP_MAX_WSCALE;
388
389 state->flags |=
390 IP_CT_TCP_FLAG_WINDOW_SCALE;
391 }
392 ptr += opsize - 2;
393 length -= opsize;
394 }
395 }
396 }
397
tcp_sack(const struct sk_buff * skb,unsigned int dataoff,const struct tcphdr * tcph,__u32 * sack)398 static void tcp_sack(const struct sk_buff *skb, unsigned int dataoff,
399 const struct tcphdr *tcph, __u32 *sack)
400 {
401 unsigned char buff[(15 * 4) - sizeof(struct tcphdr)];
402 const unsigned char *ptr;
403 int length = (tcph->doff*4) - sizeof(struct tcphdr);
404 __u32 tmp;
405
406 if (!length)
407 return;
408
409 ptr = skb_header_pointer(skb, dataoff + sizeof(struct tcphdr),
410 length, buff);
411 BUG_ON(ptr == NULL);
412
413 /* Fast path for timestamp-only option */
414 if (length == TCPOLEN_TSTAMP_ALIGNED
415 && *(__be32 *)ptr == htonl((TCPOPT_NOP << 24)
416 | (TCPOPT_NOP << 16)
417 | (TCPOPT_TIMESTAMP << 8)
418 | TCPOLEN_TIMESTAMP))
419 return;
420
421 while (length > 0) {
422 int opcode = *ptr++;
423 int opsize, i;
424
425 switch (opcode) {
426 case TCPOPT_EOL:
427 return;
428 case TCPOPT_NOP: /* Ref: RFC 793 section 3.1 */
429 length--;
430 continue;
431 default:
432 if (length < 2)
433 return;
434 opsize = *ptr++;
435 if (opsize < 2) /* "silly options" */
436 return;
437 if (opsize > length)
438 return; /* don't parse partial options */
439
440 if (opcode == TCPOPT_SACK
441 && opsize >= (TCPOLEN_SACK_BASE
442 + TCPOLEN_SACK_PERBLOCK)
443 && !((opsize - TCPOLEN_SACK_BASE)
444 % TCPOLEN_SACK_PERBLOCK)) {
445 for (i = 0;
446 i < (opsize - TCPOLEN_SACK_BASE);
447 i += TCPOLEN_SACK_PERBLOCK) {
448 tmp = get_unaligned_be32((__be32 *)(ptr+i)+1);
449
450 if (after(tmp, *sack))
451 *sack = tmp;
452 }
453 return;
454 }
455 ptr += opsize - 2;
456 length -= opsize;
457 }
458 }
459 }
460
tcp_in_window(const struct nf_conn * ct,struct ip_ct_tcp * state,enum ip_conntrack_dir dir,unsigned int index,const struct sk_buff * skb,unsigned int dataoff,const struct tcphdr * tcph)461 static bool tcp_in_window(const struct nf_conn *ct,
462 struct ip_ct_tcp *state,
463 enum ip_conntrack_dir dir,
464 unsigned int index,
465 const struct sk_buff *skb,
466 unsigned int dataoff,
467 const struct tcphdr *tcph)
468 {
469 struct net *net = nf_ct_net(ct);
470 struct nf_tcp_net *tn = nf_tcp_pernet(net);
471 struct ip_ct_tcp_state *sender = &state->seen[dir];
472 struct ip_ct_tcp_state *receiver = &state->seen[!dir];
473 const struct nf_conntrack_tuple *tuple = &ct->tuplehash[dir].tuple;
474 __u32 seq, ack, sack, end, win, swin;
475 u16 win_raw;
476 s32 receiver_offset;
477 bool res, in_recv_win;
478
479 /*
480 * Get the required data from the packet.
481 */
482 seq = ntohl(tcph->seq);
483 ack = sack = ntohl(tcph->ack_seq);
484 win_raw = ntohs(tcph->window);
485 win = win_raw;
486 end = segment_seq_plus_len(seq, skb->len, dataoff, tcph);
487
488 if (receiver->flags & IP_CT_TCP_FLAG_SACK_PERM)
489 tcp_sack(skb, dataoff, tcph, &sack);
490
491 /* Take into account NAT sequence number mangling */
492 receiver_offset = nf_ct_seq_offset(ct, !dir, ack - 1);
493 ack -= receiver_offset;
494 sack -= receiver_offset;
495
496 pr_debug("tcp_in_window: START\n");
497 pr_debug("tcp_in_window: ");
498 nf_ct_dump_tuple(tuple);
499 pr_debug("seq=%u ack=%u+(%d) sack=%u+(%d) win=%u end=%u\n",
500 seq, ack, receiver_offset, sack, receiver_offset, win, end);
501 pr_debug("tcp_in_window: sender end=%u maxend=%u maxwin=%u scale=%i "
502 "receiver end=%u maxend=%u maxwin=%u scale=%i\n",
503 sender->td_end, sender->td_maxend, sender->td_maxwin,
504 sender->td_scale,
505 receiver->td_end, receiver->td_maxend, receiver->td_maxwin,
506 receiver->td_scale);
507
508 if (sender->td_maxwin == 0) {
509 /*
510 * Initialize sender data.
511 */
512 if (tcph->syn) {
513 /*
514 * SYN-ACK in reply to a SYN
515 * or SYN from reply direction in simultaneous open.
516 */
517 sender->td_end =
518 sender->td_maxend = end;
519 sender->td_maxwin = (win == 0 ? 1 : win);
520
521 tcp_options(skb, dataoff, tcph, sender);
522 /*
523 * RFC 1323:
524 * Both sides must send the Window Scale option
525 * to enable window scaling in either direction.
526 */
527 if (!(sender->flags & IP_CT_TCP_FLAG_WINDOW_SCALE
528 && receiver->flags & IP_CT_TCP_FLAG_WINDOW_SCALE))
529 sender->td_scale =
530 receiver->td_scale = 0;
531 if (!tcph->ack)
532 /* Simultaneous open */
533 return true;
534 } else {
535 /*
536 * We are in the middle of a connection,
537 * its history is lost for us.
538 * Let's try to use the data from the packet.
539 */
540 sender->td_end = end;
541 swin = win << sender->td_scale;
542 sender->td_maxwin = (swin == 0 ? 1 : swin);
543 sender->td_maxend = end + sender->td_maxwin;
544 if (receiver->td_maxwin == 0) {
545 /* We haven't seen traffic in the other
546 * direction yet but we have to tweak window
547 * tracking to pass III and IV until that
548 * happens.
549 */
550 receiver->td_end = receiver->td_maxend = sack;
551 } else if (sack == receiver->td_end + 1) {
552 /* Likely a reply to a keepalive.
553 * Needed for III.
554 */
555 receiver->td_end++;
556 }
557
558 }
559 } else if (((state->state == TCP_CONNTRACK_SYN_SENT
560 && dir == IP_CT_DIR_ORIGINAL)
561 || (state->state == TCP_CONNTRACK_SYN_RECV
562 && dir == IP_CT_DIR_REPLY))
563 && after(end, sender->td_end)) {
564 /*
565 * RFC 793: "if a TCP is reinitialized ... then it need
566 * not wait at all; it must only be sure to use sequence
567 * numbers larger than those recently used."
568 */
569 sender->td_end =
570 sender->td_maxend = end;
571 sender->td_maxwin = (win == 0 ? 1 : win);
572
573 tcp_options(skb, dataoff, tcph, sender);
574 }
575
576 if (!(tcph->ack)) {
577 /*
578 * If there is no ACK, just pretend it was set and OK.
579 */
580 ack = sack = receiver->td_end;
581 } else if (((tcp_flag_word(tcph) & (TCP_FLAG_ACK|TCP_FLAG_RST)) ==
582 (TCP_FLAG_ACK|TCP_FLAG_RST))
583 && (ack == 0)) {
584 /*
585 * Broken TCP stacks, that set ACK in RST packets as well
586 * with zero ack value.
587 */
588 ack = sack = receiver->td_end;
589 }
590
591 if (tcph->rst && seq == 0 && state->state == TCP_CONNTRACK_SYN_SENT)
592 /*
593 * RST sent answering SYN.
594 */
595 seq = end = sender->td_end;
596
597 pr_debug("tcp_in_window: ");
598 nf_ct_dump_tuple(tuple);
599 pr_debug("seq=%u ack=%u+(%d) sack=%u+(%d) win=%u end=%u\n",
600 seq, ack, receiver_offset, sack, receiver_offset, win, end);
601 pr_debug("tcp_in_window: sender end=%u maxend=%u maxwin=%u scale=%i "
602 "receiver end=%u maxend=%u maxwin=%u scale=%i\n",
603 sender->td_end, sender->td_maxend, sender->td_maxwin,
604 sender->td_scale,
605 receiver->td_end, receiver->td_maxend, receiver->td_maxwin,
606 receiver->td_scale);
607
608 /* Is the ending sequence in the receive window (if available)? */
609 in_recv_win = !receiver->td_maxwin ||
610 after(end, sender->td_end - receiver->td_maxwin - 1);
611
612 pr_debug("tcp_in_window: I=%i II=%i III=%i IV=%i\n",
613 before(seq, sender->td_maxend + 1),
614 (in_recv_win ? 1 : 0),
615 before(sack, receiver->td_end + 1),
616 after(sack, receiver->td_end - MAXACKWINDOW(sender) - 1));
617
618 if (before(seq, sender->td_maxend + 1) &&
619 in_recv_win &&
620 before(sack, receiver->td_end + 1) &&
621 after(sack, receiver->td_end - MAXACKWINDOW(sender) - 1)) {
622 /*
623 * Take into account window scaling (RFC 1323).
624 */
625 if (!tcph->syn)
626 win <<= sender->td_scale;
627
628 /*
629 * Update sender data.
630 */
631 swin = win + (sack - ack);
632 if (sender->td_maxwin < swin)
633 sender->td_maxwin = swin;
634 if (after(end, sender->td_end)) {
635 sender->td_end = end;
636 sender->flags |= IP_CT_TCP_FLAG_DATA_UNACKNOWLEDGED;
637 }
638 if (tcph->ack) {
639 if (!(sender->flags & IP_CT_TCP_FLAG_MAXACK_SET)) {
640 sender->td_maxack = ack;
641 sender->flags |= IP_CT_TCP_FLAG_MAXACK_SET;
642 } else if (after(ack, sender->td_maxack))
643 sender->td_maxack = ack;
644 }
645
646 /*
647 * Update receiver data.
648 */
649 if (receiver->td_maxwin != 0 && after(end, sender->td_maxend))
650 receiver->td_maxwin += end - sender->td_maxend;
651 if (after(sack + win, receiver->td_maxend - 1)) {
652 receiver->td_maxend = sack + win;
653 if (win == 0)
654 receiver->td_maxend++;
655 }
656 if (ack == receiver->td_end)
657 receiver->flags &= ~IP_CT_TCP_FLAG_DATA_UNACKNOWLEDGED;
658
659 /*
660 * Check retransmissions.
661 */
662 if (index == TCP_ACK_SET) {
663 if (state->last_dir == dir
664 && state->last_seq == seq
665 && state->last_ack == ack
666 && state->last_end == end
667 && state->last_win == win_raw)
668 state->retrans++;
669 else {
670 state->last_dir = dir;
671 state->last_seq = seq;
672 state->last_ack = ack;
673 state->last_end = end;
674 state->last_win = win_raw;
675 state->retrans = 0;
676 }
677 }
678 res = true;
679 } else {
680 res = false;
681 if (sender->flags & IP_CT_TCP_FLAG_BE_LIBERAL ||
682 tn->tcp_be_liberal)
683 res = true;
684 if (!res) {
685 nf_ct_l4proto_log_invalid(skb, ct,
686 "%s",
687 before(seq, sender->td_maxend + 1) ?
688 in_recv_win ?
689 before(sack, receiver->td_end + 1) ?
690 after(sack, receiver->td_end - MAXACKWINDOW(sender) - 1) ? "BUG"
691 : "ACK is under the lower bound (possible overly delayed ACK)"
692 : "ACK is over the upper bound (ACKed data not seen yet)"
693 : "SEQ is under the lower bound (already ACKed data retransmitted)"
694 : "SEQ is over the upper bound (over the window of the receiver)");
695 }
696 }
697
698 pr_debug("tcp_in_window: res=%u sender end=%u maxend=%u maxwin=%u "
699 "receiver end=%u maxend=%u maxwin=%u\n",
700 res, sender->td_end, sender->td_maxend, sender->td_maxwin,
701 receiver->td_end, receiver->td_maxend, receiver->td_maxwin);
702
703 return res;
704 }
705
706 /* table of valid flag combinations - PUSH, ECE and CWR are always valid */
707 static const u8 tcp_valid_flags[(TCPHDR_FIN|TCPHDR_SYN|TCPHDR_RST|TCPHDR_ACK|
708 TCPHDR_URG) + 1] =
709 {
710 [TCPHDR_SYN] = 1,
711 [TCPHDR_SYN|TCPHDR_URG] = 1,
712 [TCPHDR_SYN|TCPHDR_ACK] = 1,
713 [TCPHDR_RST] = 1,
714 [TCPHDR_RST|TCPHDR_ACK] = 1,
715 [TCPHDR_FIN|TCPHDR_ACK] = 1,
716 [TCPHDR_FIN|TCPHDR_ACK|TCPHDR_URG] = 1,
717 [TCPHDR_ACK] = 1,
718 [TCPHDR_ACK|TCPHDR_URG] = 1,
719 };
720
tcp_error_log(const struct sk_buff * skb,const struct nf_hook_state * state,const char * msg)721 static void tcp_error_log(const struct sk_buff *skb,
722 const struct nf_hook_state *state,
723 const char *msg)
724 {
725 nf_l4proto_log_invalid(skb, state->net, state->pf, IPPROTO_TCP, "%s", msg);
726 }
727
728 /* Protect conntrack agaist broken packets. Code taken from ipt_unclean.c. */
tcp_error(const struct tcphdr * th,struct sk_buff * skb,unsigned int dataoff,const struct nf_hook_state * state)729 static bool tcp_error(const struct tcphdr *th,
730 struct sk_buff *skb,
731 unsigned int dataoff,
732 const struct nf_hook_state *state)
733 {
734 unsigned int tcplen = skb->len - dataoff;
735 u8 tcpflags;
736
737 /* Not whole TCP header or malformed packet */
738 if (th->doff*4 < sizeof(struct tcphdr) || tcplen < th->doff*4) {
739 tcp_error_log(skb, state, "truncated packet");
740 return true;
741 }
742
743 /* Checksum invalid? Ignore.
744 * We skip checking packets on the outgoing path
745 * because the checksum is assumed to be correct.
746 */
747 /* FIXME: Source route IP option packets --RR */
748 if (state->net->ct.sysctl_checksum &&
749 state->hook == NF_INET_PRE_ROUTING &&
750 nf_checksum(skb, state->hook, dataoff, IPPROTO_TCP, state->pf)) {
751 tcp_error_log(skb, state, "bad checksum");
752 return true;
753 }
754
755 /* Check TCP flags. */
756 tcpflags = (tcp_flag_byte(th) & ~(TCPHDR_ECE|TCPHDR_CWR|TCPHDR_PSH));
757 if (!tcp_valid_flags[tcpflags]) {
758 tcp_error_log(skb, state, "invalid tcp flag combination");
759 return true;
760 }
761
762 return false;
763 }
764
tcp_new(struct nf_conn * ct,const struct sk_buff * skb,unsigned int dataoff,const struct tcphdr * th)765 static noinline bool tcp_new(struct nf_conn *ct, const struct sk_buff *skb,
766 unsigned int dataoff,
767 const struct tcphdr *th)
768 {
769 enum tcp_conntrack new_state;
770 struct net *net = nf_ct_net(ct);
771 const struct nf_tcp_net *tn = nf_tcp_pernet(net);
772 const struct ip_ct_tcp_state *sender = &ct->proto.tcp.seen[0];
773 const struct ip_ct_tcp_state *receiver = &ct->proto.tcp.seen[1];
774
775 /* Don't need lock here: this conntrack not in circulation yet */
776 new_state = tcp_conntracks[0][get_conntrack_index(th)][TCP_CONNTRACK_NONE];
777
778 /* Invalid: delete conntrack */
779 if (new_state >= TCP_CONNTRACK_MAX) {
780 pr_debug("nf_ct_tcp: invalid new deleting.\n");
781 return false;
782 }
783
784 if (new_state == TCP_CONNTRACK_SYN_SENT) {
785 memset(&ct->proto.tcp, 0, sizeof(ct->proto.tcp));
786 /* SYN packet */
787 ct->proto.tcp.seen[0].td_end =
788 segment_seq_plus_len(ntohl(th->seq), skb->len,
789 dataoff, th);
790 ct->proto.tcp.seen[0].td_maxwin = ntohs(th->window);
791 if (ct->proto.tcp.seen[0].td_maxwin == 0)
792 ct->proto.tcp.seen[0].td_maxwin = 1;
793 ct->proto.tcp.seen[0].td_maxend =
794 ct->proto.tcp.seen[0].td_end;
795
796 tcp_options(skb, dataoff, th, &ct->proto.tcp.seen[0]);
797 } else if (tn->tcp_loose == 0) {
798 /* Don't try to pick up connections. */
799 return false;
800 } else {
801 memset(&ct->proto.tcp, 0, sizeof(ct->proto.tcp));
802 /*
803 * We are in the middle of a connection,
804 * its history is lost for us.
805 * Let's try to use the data from the packet.
806 */
807 ct->proto.tcp.seen[0].td_end =
808 segment_seq_plus_len(ntohl(th->seq), skb->len,
809 dataoff, th);
810 ct->proto.tcp.seen[0].td_maxwin = ntohs(th->window);
811 if (ct->proto.tcp.seen[0].td_maxwin == 0)
812 ct->proto.tcp.seen[0].td_maxwin = 1;
813 ct->proto.tcp.seen[0].td_maxend =
814 ct->proto.tcp.seen[0].td_end +
815 ct->proto.tcp.seen[0].td_maxwin;
816
817 /* We assume SACK and liberal window checking to handle
818 * window scaling */
819 ct->proto.tcp.seen[0].flags =
820 ct->proto.tcp.seen[1].flags = IP_CT_TCP_FLAG_SACK_PERM |
821 IP_CT_TCP_FLAG_BE_LIBERAL;
822 }
823
824 /* tcp_packet will set them */
825 ct->proto.tcp.last_index = TCP_NONE_SET;
826
827 pr_debug("%s: sender end=%u maxend=%u maxwin=%u scale=%i "
828 "receiver end=%u maxend=%u maxwin=%u scale=%i\n",
829 __func__,
830 sender->td_end, sender->td_maxend, sender->td_maxwin,
831 sender->td_scale,
832 receiver->td_end, receiver->td_maxend, receiver->td_maxwin,
833 receiver->td_scale);
834 return true;
835 }
836
nf_conntrack_tcp_established(const struct nf_conn * ct)837 static bool nf_conntrack_tcp_established(const struct nf_conn *ct)
838 {
839 return ct->proto.tcp.state == TCP_CONNTRACK_ESTABLISHED &&
840 test_bit(IPS_ASSURED_BIT, &ct->status);
841 }
842
nf_ct_tcp_state_reset(struct ip_ct_tcp_state * state)843 static void nf_ct_tcp_state_reset(struct ip_ct_tcp_state *state)
844 {
845 state->td_end = 0;
846 state->td_maxend = 0;
847 state->td_maxwin = 0;
848 state->td_maxack = 0;
849 state->td_scale = 0;
850 state->flags &= IP_CT_TCP_FLAG_BE_LIBERAL;
851 }
852
853 /* Returns verdict for packet, or -1 for invalid. */
nf_conntrack_tcp_packet(struct nf_conn * ct,struct sk_buff * skb,unsigned int dataoff,enum ip_conntrack_info ctinfo,const struct nf_hook_state * state)854 int nf_conntrack_tcp_packet(struct nf_conn *ct,
855 struct sk_buff *skb,
856 unsigned int dataoff,
857 enum ip_conntrack_info ctinfo,
858 const struct nf_hook_state *state)
859 {
860 struct net *net = nf_ct_net(ct);
861 struct nf_tcp_net *tn = nf_tcp_pernet(net);
862 struct nf_conntrack_tuple *tuple;
863 enum tcp_conntrack new_state, old_state;
864 unsigned int index, *timeouts;
865 enum ip_conntrack_dir dir;
866 const struct tcphdr *th;
867 struct tcphdr _tcph;
868 unsigned long timeout;
869
870 th = skb_header_pointer(skb, dataoff, sizeof(_tcph), &_tcph);
871 if (th == NULL)
872 return -NF_ACCEPT;
873
874 if (tcp_error(th, skb, dataoff, state))
875 return -NF_ACCEPT;
876
877 if (!nf_ct_is_confirmed(ct) && !tcp_new(ct, skb, dataoff, th))
878 return -NF_ACCEPT;
879
880 spin_lock_bh(&ct->lock);
881 old_state = ct->proto.tcp.state;
882 dir = CTINFO2DIR(ctinfo);
883 index = get_conntrack_index(th);
884 new_state = tcp_conntracks[dir][index][old_state];
885 tuple = &ct->tuplehash[dir].tuple;
886
887 switch (new_state) {
888 case TCP_CONNTRACK_SYN_SENT:
889 if (old_state < TCP_CONNTRACK_TIME_WAIT)
890 break;
891 /* RFC 1122: "When a connection is closed actively,
892 * it MUST linger in TIME-WAIT state for a time 2xMSL
893 * (Maximum Segment Lifetime). However, it MAY accept
894 * a new SYN from the remote TCP to reopen the connection
895 * directly from TIME-WAIT state, if..."
896 * We ignore the conditions because we are in the
897 * TIME-WAIT state anyway.
898 *
899 * Handle aborted connections: we and the server
900 * think there is an existing connection but the client
901 * aborts it and starts a new one.
902 */
903 if (((ct->proto.tcp.seen[dir].flags
904 | ct->proto.tcp.seen[!dir].flags)
905 & IP_CT_TCP_FLAG_CLOSE_INIT)
906 || (ct->proto.tcp.last_dir == dir
907 && ct->proto.tcp.last_index == TCP_RST_SET)) {
908 /* Attempt to reopen a closed/aborted connection.
909 * Delete this connection and look up again. */
910 spin_unlock_bh(&ct->lock);
911
912 /* Only repeat if we can actually remove the timer.
913 * Destruction may already be in progress in process
914 * context and we must give it a chance to terminate.
915 */
916 if (nf_ct_kill(ct))
917 return -NF_REPEAT;
918 return NF_DROP;
919 }
920 fallthrough;
921 case TCP_CONNTRACK_IGNORE:
922 /* Ignored packets:
923 *
924 * Our connection entry may be out of sync, so ignore
925 * packets which may signal the real connection between
926 * the client and the server.
927 *
928 * a) SYN in ORIGINAL
929 * b) SYN/ACK in REPLY
930 * c) ACK in reply direction after initial SYN in original.
931 *
932 * If the ignored packet is invalid, the receiver will send
933 * a RST we'll catch below.
934 */
935 if (index == TCP_SYNACK_SET
936 && ct->proto.tcp.last_index == TCP_SYN_SET
937 && ct->proto.tcp.last_dir != dir
938 && ntohl(th->ack_seq) == ct->proto.tcp.last_end) {
939 /* b) This SYN/ACK acknowledges a SYN that we earlier
940 * ignored as invalid. This means that the client and
941 * the server are both in sync, while the firewall is
942 * not. We get in sync from the previously annotated
943 * values.
944 */
945 old_state = TCP_CONNTRACK_SYN_SENT;
946 new_state = TCP_CONNTRACK_SYN_RECV;
947 ct->proto.tcp.seen[ct->proto.tcp.last_dir].td_end =
948 ct->proto.tcp.last_end;
949 ct->proto.tcp.seen[ct->proto.tcp.last_dir].td_maxend =
950 ct->proto.tcp.last_end;
951 ct->proto.tcp.seen[ct->proto.tcp.last_dir].td_maxwin =
952 ct->proto.tcp.last_win == 0 ?
953 1 : ct->proto.tcp.last_win;
954 ct->proto.tcp.seen[ct->proto.tcp.last_dir].td_scale =
955 ct->proto.tcp.last_wscale;
956 ct->proto.tcp.last_flags &= ~IP_CT_EXP_CHALLENGE_ACK;
957 ct->proto.tcp.seen[ct->proto.tcp.last_dir].flags =
958 ct->proto.tcp.last_flags;
959 nf_ct_tcp_state_reset(&ct->proto.tcp.seen[dir]);
960 break;
961 }
962 ct->proto.tcp.last_index = index;
963 ct->proto.tcp.last_dir = dir;
964 ct->proto.tcp.last_seq = ntohl(th->seq);
965 ct->proto.tcp.last_end =
966 segment_seq_plus_len(ntohl(th->seq), skb->len, dataoff, th);
967 ct->proto.tcp.last_win = ntohs(th->window);
968
969 /* a) This is a SYN in ORIGINAL. The client and the server
970 * may be in sync but we are not. In that case, we annotate
971 * the TCP options and let the packet go through. If it is a
972 * valid SYN packet, the server will reply with a SYN/ACK, and
973 * then we'll get in sync. Otherwise, the server potentially
974 * responds with a challenge ACK if implementing RFC5961.
975 */
976 if (index == TCP_SYN_SET && dir == IP_CT_DIR_ORIGINAL) {
977 struct ip_ct_tcp_state seen = {};
978
979 ct->proto.tcp.last_flags =
980 ct->proto.tcp.last_wscale = 0;
981 tcp_options(skb, dataoff, th, &seen);
982 if (seen.flags & IP_CT_TCP_FLAG_WINDOW_SCALE) {
983 ct->proto.tcp.last_flags |=
984 IP_CT_TCP_FLAG_WINDOW_SCALE;
985 ct->proto.tcp.last_wscale = seen.td_scale;
986 }
987 if (seen.flags & IP_CT_TCP_FLAG_SACK_PERM) {
988 ct->proto.tcp.last_flags |=
989 IP_CT_TCP_FLAG_SACK_PERM;
990 }
991 /* Mark the potential for RFC5961 challenge ACK,
992 * this pose a special problem for LAST_ACK state
993 * as ACK is intrepretated as ACKing last FIN.
994 */
995 if (old_state == TCP_CONNTRACK_LAST_ACK)
996 ct->proto.tcp.last_flags |=
997 IP_CT_EXP_CHALLENGE_ACK;
998 }
999 spin_unlock_bh(&ct->lock);
1000 nf_ct_l4proto_log_invalid(skb, ct, "invalid packet ignored in "
1001 "state %s ", tcp_conntrack_names[old_state]);
1002 return NF_ACCEPT;
1003 case TCP_CONNTRACK_MAX:
1004 /* Special case for SYN proxy: when the SYN to the server or
1005 * the SYN/ACK from the server is lost, the client may transmit
1006 * a keep-alive packet while in SYN_SENT state. This needs to
1007 * be associated with the original conntrack entry in order to
1008 * generate a new SYN with the correct sequence number.
1009 */
1010 if (nfct_synproxy(ct) && old_state == TCP_CONNTRACK_SYN_SENT &&
1011 index == TCP_ACK_SET && dir == IP_CT_DIR_ORIGINAL &&
1012 ct->proto.tcp.last_dir == IP_CT_DIR_ORIGINAL &&
1013 ct->proto.tcp.seen[dir].td_end - 1 == ntohl(th->seq)) {
1014 pr_debug("nf_ct_tcp: SYN proxy client keep alive\n");
1015 spin_unlock_bh(&ct->lock);
1016 return NF_ACCEPT;
1017 }
1018
1019 /* Invalid packet */
1020 pr_debug("nf_ct_tcp: Invalid dir=%i index=%u ostate=%u\n",
1021 dir, get_conntrack_index(th), old_state);
1022 spin_unlock_bh(&ct->lock);
1023 nf_ct_l4proto_log_invalid(skb, ct, "invalid state");
1024 return -NF_ACCEPT;
1025 case TCP_CONNTRACK_TIME_WAIT:
1026 /* RFC5961 compliance cause stack to send "challenge-ACK"
1027 * e.g. in response to spurious SYNs. Conntrack MUST
1028 * not believe this ACK is acking last FIN.
1029 */
1030 if (old_state == TCP_CONNTRACK_LAST_ACK &&
1031 index == TCP_ACK_SET &&
1032 ct->proto.tcp.last_dir != dir &&
1033 ct->proto.tcp.last_index == TCP_SYN_SET &&
1034 (ct->proto.tcp.last_flags & IP_CT_EXP_CHALLENGE_ACK)) {
1035 /* Detected RFC5961 challenge ACK */
1036 ct->proto.tcp.last_flags &= ~IP_CT_EXP_CHALLENGE_ACK;
1037 spin_unlock_bh(&ct->lock);
1038 nf_ct_l4proto_log_invalid(skb, ct, "challenge-ack ignored");
1039 return NF_ACCEPT; /* Don't change state */
1040 }
1041 break;
1042 case TCP_CONNTRACK_SYN_SENT2:
1043 /* tcp_conntracks table is not smart enough to handle
1044 * simultaneous open.
1045 */
1046 ct->proto.tcp.last_flags |= IP_CT_TCP_SIMULTANEOUS_OPEN;
1047 break;
1048 case TCP_CONNTRACK_SYN_RECV:
1049 if (dir == IP_CT_DIR_REPLY && index == TCP_ACK_SET &&
1050 ct->proto.tcp.last_flags & IP_CT_TCP_SIMULTANEOUS_OPEN)
1051 new_state = TCP_CONNTRACK_ESTABLISHED;
1052 break;
1053 case TCP_CONNTRACK_CLOSE:
1054 if (index != TCP_RST_SET)
1055 break;
1056
1057 if (ct->proto.tcp.seen[!dir].flags & IP_CT_TCP_FLAG_MAXACK_SET) {
1058 u32 seq = ntohl(th->seq);
1059
1060 if (before(seq, ct->proto.tcp.seen[!dir].td_maxack)) {
1061 /* Invalid RST */
1062 spin_unlock_bh(&ct->lock);
1063 nf_ct_l4proto_log_invalid(skb, ct, "invalid rst");
1064 return -NF_ACCEPT;
1065 }
1066
1067 if (!nf_conntrack_tcp_established(ct) ||
1068 seq == ct->proto.tcp.seen[!dir].td_maxack)
1069 break;
1070
1071 /* Check if rst is part of train, such as
1072 * foo:80 > bar:4379: P, 235946583:235946602(19) ack 42
1073 * foo:80 > bar:4379: R, 235946602:235946602(0) ack 42
1074 */
1075 if (ct->proto.tcp.last_index == TCP_ACK_SET &&
1076 ct->proto.tcp.last_dir == dir &&
1077 seq == ct->proto.tcp.last_end)
1078 break;
1079
1080 /* ... RST sequence number doesn't match exactly, keep
1081 * established state to allow a possible challenge ACK.
1082 */
1083 new_state = old_state;
1084 }
1085 if (((test_bit(IPS_SEEN_REPLY_BIT, &ct->status)
1086 && ct->proto.tcp.last_index == TCP_SYN_SET)
1087 || (!test_bit(IPS_ASSURED_BIT, &ct->status)
1088 && ct->proto.tcp.last_index == TCP_ACK_SET))
1089 && ntohl(th->ack_seq) == ct->proto.tcp.last_end) {
1090 /* RST sent to invalid SYN or ACK we had let through
1091 * at a) and c) above:
1092 *
1093 * a) SYN was in window then
1094 * c) we hold a half-open connection.
1095 *
1096 * Delete our connection entry.
1097 * We skip window checking, because packet might ACK
1098 * segments we ignored. */
1099 goto in_window;
1100 }
1101 break;
1102 default:
1103 /* Keep compilers happy. */
1104 break;
1105 }
1106
1107 if (!tcp_in_window(ct, &ct->proto.tcp, dir, index,
1108 skb, dataoff, th)) {
1109 spin_unlock_bh(&ct->lock);
1110 return -NF_ACCEPT;
1111 }
1112 in_window:
1113 /* From now on we have got in-window packets */
1114 ct->proto.tcp.last_index = index;
1115 ct->proto.tcp.last_dir = dir;
1116
1117 pr_debug("tcp_conntracks: ");
1118 nf_ct_dump_tuple(tuple);
1119 pr_debug("syn=%i ack=%i fin=%i rst=%i old=%i new=%i\n",
1120 (th->syn ? 1 : 0), (th->ack ? 1 : 0),
1121 (th->fin ? 1 : 0), (th->rst ? 1 : 0),
1122 old_state, new_state);
1123
1124 ct->proto.tcp.state = new_state;
1125 if (old_state != new_state
1126 && new_state == TCP_CONNTRACK_FIN_WAIT)
1127 ct->proto.tcp.seen[dir].flags |= IP_CT_TCP_FLAG_CLOSE_INIT;
1128
1129 timeouts = nf_ct_timeout_lookup(ct);
1130 if (!timeouts)
1131 timeouts = tn->timeouts;
1132
1133 if (ct->proto.tcp.retrans >= tn->tcp_max_retrans &&
1134 timeouts[new_state] > timeouts[TCP_CONNTRACK_RETRANS])
1135 timeout = timeouts[TCP_CONNTRACK_RETRANS];
1136 else if (unlikely(index == TCP_RST_SET))
1137 timeout = timeouts[TCP_CONNTRACK_CLOSE];
1138 else if ((ct->proto.tcp.seen[0].flags | ct->proto.tcp.seen[1].flags) &
1139 IP_CT_TCP_FLAG_DATA_UNACKNOWLEDGED &&
1140 timeouts[new_state] > timeouts[TCP_CONNTRACK_UNACK])
1141 timeout = timeouts[TCP_CONNTRACK_UNACK];
1142 else if (ct->proto.tcp.last_win == 0 &&
1143 timeouts[new_state] > timeouts[TCP_CONNTRACK_RETRANS])
1144 timeout = timeouts[TCP_CONNTRACK_RETRANS];
1145 else
1146 timeout = timeouts[new_state];
1147 spin_unlock_bh(&ct->lock);
1148
1149 if (new_state != old_state)
1150 nf_conntrack_event_cache(IPCT_PROTOINFO, ct);
1151
1152 if (!test_bit(IPS_SEEN_REPLY_BIT, &ct->status)) {
1153 /* If only reply is a RST, we can consider ourselves not to
1154 have an established connection: this is a fairly common
1155 problem case, so we can delete the conntrack
1156 immediately. --RR */
1157 if (th->rst) {
1158 nf_ct_kill_acct(ct, ctinfo, skb);
1159 return NF_ACCEPT;
1160 }
1161
1162 if (index == TCP_SYN_SET && old_state == TCP_CONNTRACK_SYN_SENT) {
1163 /* do not renew timeout on SYN retransmit.
1164 *
1165 * Else port reuse by client or NAT middlebox can keep
1166 * entry alive indefinitely (including nat info).
1167 */
1168 return NF_ACCEPT;
1169 }
1170
1171 /* ESTABLISHED without SEEN_REPLY, i.e. mid-connection
1172 * pickup with loose=1. Avoid large ESTABLISHED timeout.
1173 */
1174 if (new_state == TCP_CONNTRACK_ESTABLISHED &&
1175 timeout > timeouts[TCP_CONNTRACK_UNACK])
1176 timeout = timeouts[TCP_CONNTRACK_UNACK];
1177 } else if (!test_bit(IPS_ASSURED_BIT, &ct->status)
1178 && (old_state == TCP_CONNTRACK_SYN_RECV
1179 || old_state == TCP_CONNTRACK_ESTABLISHED)
1180 && new_state == TCP_CONNTRACK_ESTABLISHED) {
1181 /* Set ASSURED if we see valid ack in ESTABLISHED
1182 after SYN_RECV or a valid answer for a picked up
1183 connection. */
1184 set_bit(IPS_ASSURED_BIT, &ct->status);
1185 nf_conntrack_event_cache(IPCT_ASSURED, ct);
1186 }
1187 nf_ct_refresh_acct(ct, ctinfo, skb, timeout);
1188
1189 return NF_ACCEPT;
1190 }
1191
tcp_can_early_drop(const struct nf_conn * ct)1192 static bool tcp_can_early_drop(const struct nf_conn *ct)
1193 {
1194 switch (ct->proto.tcp.state) {
1195 case TCP_CONNTRACK_FIN_WAIT:
1196 case TCP_CONNTRACK_LAST_ACK:
1197 case TCP_CONNTRACK_TIME_WAIT:
1198 case TCP_CONNTRACK_CLOSE:
1199 case TCP_CONNTRACK_CLOSE_WAIT:
1200 return true;
1201 default:
1202 break;
1203 }
1204
1205 return false;
1206 }
1207
1208 #if IS_ENABLED(CONFIG_NF_CT_NETLINK)
1209
1210 #include <linux/netfilter/nfnetlink.h>
1211 #include <linux/netfilter/nfnetlink_conntrack.h>
1212
tcp_to_nlattr(struct sk_buff * skb,struct nlattr * nla,struct nf_conn * ct)1213 static int tcp_to_nlattr(struct sk_buff *skb, struct nlattr *nla,
1214 struct nf_conn *ct)
1215 {
1216 struct nlattr *nest_parms;
1217 struct nf_ct_tcp_flags tmp = {};
1218
1219 spin_lock_bh(&ct->lock);
1220 nest_parms = nla_nest_start(skb, CTA_PROTOINFO_TCP);
1221 if (!nest_parms)
1222 goto nla_put_failure;
1223
1224 if (nla_put_u8(skb, CTA_PROTOINFO_TCP_STATE, ct->proto.tcp.state) ||
1225 nla_put_u8(skb, CTA_PROTOINFO_TCP_WSCALE_ORIGINAL,
1226 ct->proto.tcp.seen[0].td_scale) ||
1227 nla_put_u8(skb, CTA_PROTOINFO_TCP_WSCALE_REPLY,
1228 ct->proto.tcp.seen[1].td_scale))
1229 goto nla_put_failure;
1230
1231 tmp.flags = ct->proto.tcp.seen[0].flags;
1232 if (nla_put(skb, CTA_PROTOINFO_TCP_FLAGS_ORIGINAL,
1233 sizeof(struct nf_ct_tcp_flags), &tmp))
1234 goto nla_put_failure;
1235
1236 tmp.flags = ct->proto.tcp.seen[1].flags;
1237 if (nla_put(skb, CTA_PROTOINFO_TCP_FLAGS_REPLY,
1238 sizeof(struct nf_ct_tcp_flags), &tmp))
1239 goto nla_put_failure;
1240 spin_unlock_bh(&ct->lock);
1241
1242 nla_nest_end(skb, nest_parms);
1243
1244 return 0;
1245
1246 nla_put_failure:
1247 spin_unlock_bh(&ct->lock);
1248 return -1;
1249 }
1250
1251 static const struct nla_policy tcp_nla_policy[CTA_PROTOINFO_TCP_MAX+1] = {
1252 [CTA_PROTOINFO_TCP_STATE] = { .type = NLA_U8 },
1253 [CTA_PROTOINFO_TCP_WSCALE_ORIGINAL] = { .type = NLA_U8 },
1254 [CTA_PROTOINFO_TCP_WSCALE_REPLY] = { .type = NLA_U8 },
1255 [CTA_PROTOINFO_TCP_FLAGS_ORIGINAL] = { .len = sizeof(struct nf_ct_tcp_flags) },
1256 [CTA_PROTOINFO_TCP_FLAGS_REPLY] = { .len = sizeof(struct nf_ct_tcp_flags) },
1257 };
1258
1259 #define TCP_NLATTR_SIZE ( \
1260 NLA_ALIGN(NLA_HDRLEN + 1) + \
1261 NLA_ALIGN(NLA_HDRLEN + 1) + \
1262 NLA_ALIGN(NLA_HDRLEN + sizeof(struct nf_ct_tcp_flags)) + \
1263 NLA_ALIGN(NLA_HDRLEN + sizeof(struct nf_ct_tcp_flags)))
1264
nlattr_to_tcp(struct nlattr * cda[],struct nf_conn * ct)1265 static int nlattr_to_tcp(struct nlattr *cda[], struct nf_conn *ct)
1266 {
1267 struct nlattr *pattr = cda[CTA_PROTOINFO_TCP];
1268 struct nlattr *tb[CTA_PROTOINFO_TCP_MAX+1];
1269 int err;
1270
1271 /* updates could not contain anything about the private
1272 * protocol info, in that case skip the parsing */
1273 if (!pattr)
1274 return 0;
1275
1276 err = nla_parse_nested_deprecated(tb, CTA_PROTOINFO_TCP_MAX, pattr,
1277 tcp_nla_policy, NULL);
1278 if (err < 0)
1279 return err;
1280
1281 if (tb[CTA_PROTOINFO_TCP_STATE] &&
1282 nla_get_u8(tb[CTA_PROTOINFO_TCP_STATE]) >= TCP_CONNTRACK_MAX)
1283 return -EINVAL;
1284
1285 spin_lock_bh(&ct->lock);
1286 if (tb[CTA_PROTOINFO_TCP_STATE])
1287 ct->proto.tcp.state = nla_get_u8(tb[CTA_PROTOINFO_TCP_STATE]);
1288
1289 if (tb[CTA_PROTOINFO_TCP_FLAGS_ORIGINAL]) {
1290 struct nf_ct_tcp_flags *attr =
1291 nla_data(tb[CTA_PROTOINFO_TCP_FLAGS_ORIGINAL]);
1292 ct->proto.tcp.seen[0].flags &= ~attr->mask;
1293 ct->proto.tcp.seen[0].flags |= attr->flags & attr->mask;
1294 }
1295
1296 if (tb[CTA_PROTOINFO_TCP_FLAGS_REPLY]) {
1297 struct nf_ct_tcp_flags *attr =
1298 nla_data(tb[CTA_PROTOINFO_TCP_FLAGS_REPLY]);
1299 ct->proto.tcp.seen[1].flags &= ~attr->mask;
1300 ct->proto.tcp.seen[1].flags |= attr->flags & attr->mask;
1301 }
1302
1303 if (tb[CTA_PROTOINFO_TCP_WSCALE_ORIGINAL] &&
1304 tb[CTA_PROTOINFO_TCP_WSCALE_REPLY] &&
1305 ct->proto.tcp.seen[0].flags & IP_CT_TCP_FLAG_WINDOW_SCALE &&
1306 ct->proto.tcp.seen[1].flags & IP_CT_TCP_FLAG_WINDOW_SCALE) {
1307 ct->proto.tcp.seen[0].td_scale =
1308 nla_get_u8(tb[CTA_PROTOINFO_TCP_WSCALE_ORIGINAL]);
1309 ct->proto.tcp.seen[1].td_scale =
1310 nla_get_u8(tb[CTA_PROTOINFO_TCP_WSCALE_REPLY]);
1311 }
1312 spin_unlock_bh(&ct->lock);
1313
1314 return 0;
1315 }
1316
tcp_nlattr_tuple_size(void)1317 static unsigned int tcp_nlattr_tuple_size(void)
1318 {
1319 static unsigned int size __read_mostly;
1320
1321 if (!size)
1322 size = nla_policy_len(nf_ct_port_nla_policy, CTA_PROTO_MAX + 1);
1323
1324 return size;
1325 }
1326 #endif
1327
1328 #ifdef CONFIG_NF_CONNTRACK_TIMEOUT
1329
1330 #include <linux/netfilter/nfnetlink.h>
1331 #include <linux/netfilter/nfnetlink_cttimeout.h>
1332
tcp_timeout_nlattr_to_obj(struct nlattr * tb[],struct net * net,void * data)1333 static int tcp_timeout_nlattr_to_obj(struct nlattr *tb[],
1334 struct net *net, void *data)
1335 {
1336 struct nf_tcp_net *tn = nf_tcp_pernet(net);
1337 unsigned int *timeouts = data;
1338 int i;
1339
1340 if (!timeouts)
1341 timeouts = tn->timeouts;
1342 /* set default TCP timeouts. */
1343 for (i=0; i<TCP_CONNTRACK_TIMEOUT_MAX; i++)
1344 timeouts[i] = tn->timeouts[i];
1345
1346 if (tb[CTA_TIMEOUT_TCP_SYN_SENT]) {
1347 timeouts[TCP_CONNTRACK_SYN_SENT] =
1348 ntohl(nla_get_be32(tb[CTA_TIMEOUT_TCP_SYN_SENT]))*HZ;
1349 }
1350
1351 if (tb[CTA_TIMEOUT_TCP_SYN_RECV]) {
1352 timeouts[TCP_CONNTRACK_SYN_RECV] =
1353 ntohl(nla_get_be32(tb[CTA_TIMEOUT_TCP_SYN_RECV]))*HZ;
1354 }
1355 if (tb[CTA_TIMEOUT_TCP_ESTABLISHED]) {
1356 timeouts[TCP_CONNTRACK_ESTABLISHED] =
1357 ntohl(nla_get_be32(tb[CTA_TIMEOUT_TCP_ESTABLISHED]))*HZ;
1358 }
1359 if (tb[CTA_TIMEOUT_TCP_FIN_WAIT]) {
1360 timeouts[TCP_CONNTRACK_FIN_WAIT] =
1361 ntohl(nla_get_be32(tb[CTA_TIMEOUT_TCP_FIN_WAIT]))*HZ;
1362 }
1363 if (tb[CTA_TIMEOUT_TCP_CLOSE_WAIT]) {
1364 timeouts[TCP_CONNTRACK_CLOSE_WAIT] =
1365 ntohl(nla_get_be32(tb[CTA_TIMEOUT_TCP_CLOSE_WAIT]))*HZ;
1366 }
1367 if (tb[CTA_TIMEOUT_TCP_LAST_ACK]) {
1368 timeouts[TCP_CONNTRACK_LAST_ACK] =
1369 ntohl(nla_get_be32(tb[CTA_TIMEOUT_TCP_LAST_ACK]))*HZ;
1370 }
1371 if (tb[CTA_TIMEOUT_TCP_TIME_WAIT]) {
1372 timeouts[TCP_CONNTRACK_TIME_WAIT] =
1373 ntohl(nla_get_be32(tb[CTA_TIMEOUT_TCP_TIME_WAIT]))*HZ;
1374 }
1375 if (tb[CTA_TIMEOUT_TCP_CLOSE]) {
1376 timeouts[TCP_CONNTRACK_CLOSE] =
1377 ntohl(nla_get_be32(tb[CTA_TIMEOUT_TCP_CLOSE]))*HZ;
1378 }
1379 if (tb[CTA_TIMEOUT_TCP_SYN_SENT2]) {
1380 timeouts[TCP_CONNTRACK_SYN_SENT2] =
1381 ntohl(nla_get_be32(tb[CTA_TIMEOUT_TCP_SYN_SENT2]))*HZ;
1382 }
1383 if (tb[CTA_TIMEOUT_TCP_RETRANS]) {
1384 timeouts[TCP_CONNTRACK_RETRANS] =
1385 ntohl(nla_get_be32(tb[CTA_TIMEOUT_TCP_RETRANS]))*HZ;
1386 }
1387 if (tb[CTA_TIMEOUT_TCP_UNACK]) {
1388 timeouts[TCP_CONNTRACK_UNACK] =
1389 ntohl(nla_get_be32(tb[CTA_TIMEOUT_TCP_UNACK]))*HZ;
1390 }
1391
1392 timeouts[CTA_TIMEOUT_TCP_UNSPEC] = timeouts[CTA_TIMEOUT_TCP_SYN_SENT];
1393 return 0;
1394 }
1395
1396 static int
tcp_timeout_obj_to_nlattr(struct sk_buff * skb,const void * data)1397 tcp_timeout_obj_to_nlattr(struct sk_buff *skb, const void *data)
1398 {
1399 const unsigned int *timeouts = data;
1400
1401 if (nla_put_be32(skb, CTA_TIMEOUT_TCP_SYN_SENT,
1402 htonl(timeouts[TCP_CONNTRACK_SYN_SENT] / HZ)) ||
1403 nla_put_be32(skb, CTA_TIMEOUT_TCP_SYN_RECV,
1404 htonl(timeouts[TCP_CONNTRACK_SYN_RECV] / HZ)) ||
1405 nla_put_be32(skb, CTA_TIMEOUT_TCP_ESTABLISHED,
1406 htonl(timeouts[TCP_CONNTRACK_ESTABLISHED] / HZ)) ||
1407 nla_put_be32(skb, CTA_TIMEOUT_TCP_FIN_WAIT,
1408 htonl(timeouts[TCP_CONNTRACK_FIN_WAIT] / HZ)) ||
1409 nla_put_be32(skb, CTA_TIMEOUT_TCP_CLOSE_WAIT,
1410 htonl(timeouts[TCP_CONNTRACK_CLOSE_WAIT] / HZ)) ||
1411 nla_put_be32(skb, CTA_TIMEOUT_TCP_LAST_ACK,
1412 htonl(timeouts[TCP_CONNTRACK_LAST_ACK] / HZ)) ||
1413 nla_put_be32(skb, CTA_TIMEOUT_TCP_TIME_WAIT,
1414 htonl(timeouts[TCP_CONNTRACK_TIME_WAIT] / HZ)) ||
1415 nla_put_be32(skb, CTA_TIMEOUT_TCP_CLOSE,
1416 htonl(timeouts[TCP_CONNTRACK_CLOSE] / HZ)) ||
1417 nla_put_be32(skb, CTA_TIMEOUT_TCP_SYN_SENT2,
1418 htonl(timeouts[TCP_CONNTRACK_SYN_SENT2] / HZ)) ||
1419 nla_put_be32(skb, CTA_TIMEOUT_TCP_RETRANS,
1420 htonl(timeouts[TCP_CONNTRACK_RETRANS] / HZ)) ||
1421 nla_put_be32(skb, CTA_TIMEOUT_TCP_UNACK,
1422 htonl(timeouts[TCP_CONNTRACK_UNACK] / HZ)))
1423 goto nla_put_failure;
1424 return 0;
1425
1426 nla_put_failure:
1427 return -ENOSPC;
1428 }
1429
1430 static const struct nla_policy tcp_timeout_nla_policy[CTA_TIMEOUT_TCP_MAX+1] = {
1431 [CTA_TIMEOUT_TCP_SYN_SENT] = { .type = NLA_U32 },
1432 [CTA_TIMEOUT_TCP_SYN_RECV] = { .type = NLA_U32 },
1433 [CTA_TIMEOUT_TCP_ESTABLISHED] = { .type = NLA_U32 },
1434 [CTA_TIMEOUT_TCP_FIN_WAIT] = { .type = NLA_U32 },
1435 [CTA_TIMEOUT_TCP_CLOSE_WAIT] = { .type = NLA_U32 },
1436 [CTA_TIMEOUT_TCP_LAST_ACK] = { .type = NLA_U32 },
1437 [CTA_TIMEOUT_TCP_TIME_WAIT] = { .type = NLA_U32 },
1438 [CTA_TIMEOUT_TCP_CLOSE] = { .type = NLA_U32 },
1439 [CTA_TIMEOUT_TCP_SYN_SENT2] = { .type = NLA_U32 },
1440 [CTA_TIMEOUT_TCP_RETRANS] = { .type = NLA_U32 },
1441 [CTA_TIMEOUT_TCP_UNACK] = { .type = NLA_U32 },
1442 };
1443 #endif /* CONFIG_NF_CONNTRACK_TIMEOUT */
1444
nf_conntrack_tcp_init_net(struct net * net)1445 void nf_conntrack_tcp_init_net(struct net *net)
1446 {
1447 struct nf_tcp_net *tn = nf_tcp_pernet(net);
1448 int i;
1449
1450 for (i = 0; i < TCP_CONNTRACK_TIMEOUT_MAX; i++)
1451 tn->timeouts[i] = tcp_timeouts[i];
1452
1453 /* timeouts[0] is unused, make it same as SYN_SENT so
1454 * ->timeouts[0] contains 'new' timeout, like udp or icmp.
1455 */
1456 tn->timeouts[0] = tcp_timeouts[TCP_CONNTRACK_SYN_SENT];
1457 tn->tcp_loose = nf_ct_tcp_loose;
1458 tn->tcp_be_liberal = nf_ct_tcp_be_liberal;
1459 tn->tcp_max_retrans = nf_ct_tcp_max_retrans;
1460 }
1461
1462 const struct nf_conntrack_l4proto nf_conntrack_l4proto_tcp =
1463 {
1464 .l4proto = IPPROTO_TCP,
1465 #ifdef CONFIG_NF_CONNTRACK_PROCFS
1466 .print_conntrack = tcp_print_conntrack,
1467 #endif
1468 .can_early_drop = tcp_can_early_drop,
1469 #if IS_ENABLED(CONFIG_NF_CT_NETLINK)
1470 .to_nlattr = tcp_to_nlattr,
1471 .from_nlattr = nlattr_to_tcp,
1472 .tuple_to_nlattr = nf_ct_port_tuple_to_nlattr,
1473 .nlattr_to_tuple = nf_ct_port_nlattr_to_tuple,
1474 .nlattr_tuple_size = tcp_nlattr_tuple_size,
1475 .nlattr_size = TCP_NLATTR_SIZE,
1476 .nla_policy = nf_ct_port_nla_policy,
1477 #endif
1478 #ifdef CONFIG_NF_CONNTRACK_TIMEOUT
1479 .ctnl_timeout = {
1480 .nlattr_to_obj = tcp_timeout_nlattr_to_obj,
1481 .obj_to_nlattr = tcp_timeout_obj_to_nlattr,
1482 .nlattr_max = CTA_TIMEOUT_TCP_MAX,
1483 .obj_size = sizeof(unsigned int) *
1484 TCP_CONNTRACK_TIMEOUT_MAX,
1485 .nla_policy = tcp_timeout_nla_policy,
1486 },
1487 #endif /* CONFIG_NF_CONNTRACK_TIMEOUT */
1488 };
1489