1 /* (C) 1999-2001 Paul `Rusty' Russell
2 * (C) 2002-2004 Netfilter Core Team <coreteam@netfilter.org>
3 * (C) 2002-2013 Jozsef Kadlecsik <kadlec@blackhole.kfki.hu>
4 * (C) 2006-2012 Patrick McHardy <kaber@trash.net>
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 as
8 * published by the Free Software Foundation.
11 #include <linux/types.h>
12 #include <linux/timer.h>
13 #include <linux/module.h>
15 #include <linux/tcp.h>
16 #include <linux/spinlock.h>
17 #include <linux/skbuff.h>
18 #include <linux/ipv6.h>
19 #include <net/ip6_checksum.h>
20 #include <asm/unaligned.h>
24 #include <linux/netfilter.h>
25 #include <linux/netfilter_ipv4.h>
26 #include <linux/netfilter_ipv6.h>
27 #include <net/netfilter/nf_conntrack.h>
28 #include <net/netfilter/nf_conntrack_l4proto.h>
29 #include <net/netfilter/nf_conntrack_ecache.h>
30 #include <net/netfilter/nf_conntrack_seqadj.h>
31 #include <net/netfilter/nf_conntrack_synproxy.h>
32 #include <net/netfilter/nf_conntrack_timeout.h>
33 #include <net/netfilter/nf_log.h>
34 #include <net/netfilter/ipv4/nf_conntrack_ipv4.h>
35 #include <net/netfilter/ipv6/nf_conntrack_ipv6.h>
37 /* "Be conservative in what you do,
38 be liberal in what you accept from others."
39 If it's non-zero, we mark only out of window RST segments as INVALID. */
40 static int nf_ct_tcp_be_liberal __read_mostly = 0;
42 /* If it is set to zero, we disable picking up already established
44 static int nf_ct_tcp_loose __read_mostly = 1;
46 /* Max number of the retransmitted packets without receiving an (acceptable)
47 ACK from the destination. If this number is reached, a shorter timer
49 static int nf_ct_tcp_max_retrans __read_mostly = 3;
51 /* FIXME: Examine ipfilter's timeouts and conntrack transitions more
52 closely. They're more complex. --RR */
54 static const char *const tcp_conntrack_names[] = {
68 #define MINS * 60 SECS
69 #define HOURS * 60 MINS
70 #define DAYS * 24 HOURS
72 static const unsigned int tcp_timeouts[TCP_CONNTRACK_TIMEOUT_MAX] = {
73 [TCP_CONNTRACK_SYN_SENT] = 2 MINS,
74 [TCP_CONNTRACK_SYN_RECV] = 60 SECS,
75 [TCP_CONNTRACK_ESTABLISHED] = 5 DAYS,
76 [TCP_CONNTRACK_FIN_WAIT] = 2 MINS,
77 [TCP_CONNTRACK_CLOSE_WAIT] = 60 SECS,
78 [TCP_CONNTRACK_LAST_ACK] = 30 SECS,
79 [TCP_CONNTRACK_TIME_WAIT] = 2 MINS,
80 [TCP_CONNTRACK_CLOSE] = 10 SECS,
81 [TCP_CONNTRACK_SYN_SENT2] = 2 MINS,
82 /* RFC1122 says the R2 limit should be at least 100 seconds.
83 Linux uses 15 packets as limit, which corresponds
84 to ~13-30min depending on RTO. */
85 [TCP_CONNTRACK_RETRANS] = 5 MINS,
86 [TCP_CONNTRACK_UNACK] = 5 MINS,
89 #define sNO TCP_CONNTRACK_NONE
90 #define sSS TCP_CONNTRACK_SYN_SENT
91 #define sSR TCP_CONNTRACK_SYN_RECV
92 #define sES TCP_CONNTRACK_ESTABLISHED
93 #define sFW TCP_CONNTRACK_FIN_WAIT
94 #define sCW TCP_CONNTRACK_CLOSE_WAIT
95 #define sLA TCP_CONNTRACK_LAST_ACK
96 #define sTW TCP_CONNTRACK_TIME_WAIT
97 #define sCL TCP_CONNTRACK_CLOSE
98 #define sS2 TCP_CONNTRACK_SYN_SENT2
99 #define sIV TCP_CONNTRACK_MAX
100 #define sIG TCP_CONNTRACK_IGNORE
102 /* What TCP flags are set from RST/SYN/FIN/ACK. */
113 * The TCP state transition table needs a few words...
115 * We are the man in the middle. All the packets go through us
116 * but might get lost in transit to the destination.
117 * It is assumed that the destinations can't receive segments
120 * The checked segment is in window, but our windows are *not*
121 * equivalent with the ones of the sender/receiver. We always
122 * try to guess the state of the current sender.
124 * The meaning of the states are:
126 * NONE: initial state
127 * SYN_SENT: SYN-only packet seen
128 * SYN_SENT2: SYN-only packet seen from reply dir, simultaneous open
129 * SYN_RECV: SYN-ACK packet seen
130 * ESTABLISHED: ACK packet seen
131 * FIN_WAIT: FIN packet seen
132 * CLOSE_WAIT: ACK seen (after FIN)
133 * LAST_ACK: FIN seen (after FIN)
134 * TIME_WAIT: last ACK seen
135 * CLOSE: closed connection (RST)
137 * Packets marked as IGNORED (sIG):
138 * if they may be either invalid or valid
139 * and the receiver may send back a connection
140 * closing RST or a SYN/ACK.
142 * Packets marked as INVALID (sIV):
143 * if we regard them as truly invalid packets
145 static const u8 tcp_conntracks[2][6][TCP_CONNTRACK_MAX] = {
148 /* sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sS2 */
149 /*syn*/ { sSS, sSS, sIG, sIG, sIG, sIG, sIG, sSS, sSS, sS2 },
151 * sNO -> sSS Initialize a new connection
152 * sSS -> sSS Retransmitted SYN
153 * sS2 -> sS2 Late retransmitted SYN
155 * sES -> sIG Error: SYNs in window outside the SYN_SENT state
156 * are errors. Receiver will reply with RST
157 * and close the connection.
158 * Or we are not in sync and hold a dead connection.
162 * sTW -> sSS Reopened connection (RFC 1122).
165 /* sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sS2 */
166 /*synack*/ { sIV, sIV, sSR, sIV, sIV, sIV, sIV, sIV, sIV, sSR },
168 * sNO -> sIV Too late and no reason to do anything
169 * sSS -> sIV Client can't send SYN and then SYN/ACK
170 * sS2 -> sSR SYN/ACK sent to SYN2 in simultaneous open
171 * sSR -> sSR Late retransmitted SYN/ACK in simultaneous open
172 * sES -> sIV Invalid SYN/ACK packets sent by the client
179 /* sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sS2 */
180 /*fin*/ { sIV, sIV, sFW, sFW, sLA, sLA, sLA, sTW, sCL, sIV },
182 * sNO -> sIV Too late and no reason to do anything...
183 * sSS -> sIV Client migth not send FIN in this state:
184 * we enforce waiting for a SYN/ACK reply first.
186 * sSR -> sFW Close started.
188 * sFW -> sLA FIN seen in both directions, waiting for
190 * Migth be a retransmitted FIN as well...
192 * sLA -> sLA Retransmitted FIN. Remain in the same state.
196 /* sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sS2 */
197 /*ack*/ { sES, sIV, sES, sES, sCW, sCW, sTW, sTW, sCL, sIV },
199 * sNO -> sES Assumed.
200 * sSS -> sIV ACK is invalid: we haven't seen a SYN/ACK yet.
202 * sSR -> sES Established state is reached.
204 * sFW -> sCW Normal close request answered by ACK.
206 * sLA -> sTW Last ACK detected (RFC5961 challenged)
207 * sTW -> sTW Retransmitted last ACK. Remain in the same state.
210 /* sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sS2 */
211 /*rst*/ { sIV, sCL, sCL, sCL, sCL, sCL, sCL, sCL, sCL, sCL },
212 /*none*/ { sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV }
216 /* sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sS2 */
217 /*syn*/ { sIV, sS2, sIV, sIV, sIV, sIV, sIV, sSS, sIV, sS2 },
219 * sNO -> sIV Never reached.
220 * sSS -> sS2 Simultaneous open
221 * sS2 -> sS2 Retransmitted simultaneous SYN
222 * sSR -> sIV Invalid SYN packets sent by the server
227 * sTW -> sSS Reopened connection, but server may have switched role
230 /* sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sS2 */
231 /*synack*/ { sIV, sSR, sIG, sIG, sIG, sIG, sIG, sIG, sIG, sSR },
233 * sSS -> sSR Standard open.
234 * sS2 -> sSR Simultaneous open
235 * sSR -> sIG Retransmitted SYN/ACK, ignore it.
236 * sES -> sIG Late retransmitted SYN/ACK?
237 * sFW -> sIG Might be SYN/ACK answering ignored SYN
243 /* sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sS2 */
244 /*fin*/ { sIV, sIV, sFW, sFW, sLA, sLA, sLA, sTW, sCL, sIV },
246 * sSS -> sIV Server might not send FIN in this state.
248 * sSR -> sFW Close started.
250 * sFW -> sLA FIN seen in both directions.
252 * sLA -> sLA Retransmitted FIN.
256 /* sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sS2 */
257 /*ack*/ { sIV, sIG, sSR, sES, sCW, sCW, sTW, sTW, sCL, sIG },
259 * sSS -> sIG Might be a half-open connection.
261 * sSR -> sSR Might answer late resent SYN.
263 * sFW -> sCW Normal close request answered by ACK.
265 * sLA -> sTW Last ACK detected (RFC5961 challenged)
266 * sTW -> sTW Retransmitted last ACK.
269 /* sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sS2 */
270 /*rst*/ { sIV, sCL, sCL, sCL, sCL, sCL, sCL, sCL, sCL, sCL },
271 /*none*/ { sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV }
275 static inline struct nf_tcp_net *tcp_pernet(struct net *net)
277 return &net->ct.nf_ct_proto.tcp;
280 #ifdef CONFIG_NF_CONNTRACK_PROCFS
281 /* Print out the private part of the conntrack. */
282 static void tcp_print_conntrack(struct seq_file *s, struct nf_conn *ct)
284 if (test_bit(IPS_OFFLOAD_BIT, &ct->status))
287 seq_printf(s, "%s ", tcp_conntrack_names[ct->proto.tcp.state]);
291 static unsigned int get_conntrack_index(const struct tcphdr *tcph)
293 if (tcph->rst) return TCP_RST_SET;
294 else if (tcph->syn) return (tcph->ack ? TCP_SYNACK_SET : TCP_SYN_SET);
295 else if (tcph->fin) return TCP_FIN_SET;
296 else if (tcph->ack) return TCP_ACK_SET;
297 else return TCP_NONE_SET;
300 /* TCP connection tracking based on 'Real Stateful TCP Packet Filtering
301 in IP Filter' by Guido van Rooij.
303 http://www.sane.nl/events/sane2000/papers.html
304 http://www.darkart.com/mirrors/www.obfuscation.org/ipf/
306 The boundaries and the conditions are changed according to RFC793:
307 the packet must intersect the window (i.e. segments may be
308 after the right or before the left edge) and thus receivers may ACK
309 segments after the right edge of the window.
311 td_maxend = max(sack + max(win,1)) seen in reply packets
312 td_maxwin = max(max(win, 1)) + (sack - ack) seen in sent packets
313 td_maxwin += seq + len - sender.td_maxend
314 if seq + len > sender.td_maxend
315 td_end = max(seq + len) seen in sent packets
317 I. Upper bound for valid data: seq <= sender.td_maxend
318 II. Lower bound for valid data: seq + len >= sender.td_end - receiver.td_maxwin
319 III. Upper bound for valid (s)ack: sack <= receiver.td_end
320 IV. Lower bound for valid (s)ack: sack >= receiver.td_end - MAXACKWINDOW
322 where sack is the highest right edge of sack block found in the packet
323 or ack in the case of packet without SACK option.
325 The upper bound limit for a valid (s)ack is not ignored -
326 we doesn't have to deal with fragments.
329 static inline __u32 segment_seq_plus_len(__u32 seq,
331 unsigned int dataoff,
332 const struct tcphdr *tcph)
334 /* XXX Should I use payload length field in IP/IPv6 header ?
336 return (seq + len - dataoff - tcph->doff*4
337 + (tcph->syn ? 1 : 0) + (tcph->fin ? 1 : 0));
340 /* Fixme: what about big packets? */
341 #define MAXACKWINCONST 66000
342 #define MAXACKWINDOW(sender) \
343 ((sender)->td_maxwin > MAXACKWINCONST ? (sender)->td_maxwin \
347 * Simplified tcp_parse_options routine from tcp_input.c
349 static void tcp_options(const struct sk_buff *skb,
350 unsigned int dataoff,
351 const struct tcphdr *tcph,
352 struct ip_ct_tcp_state *state)
354 unsigned char buff[(15 * 4) - sizeof(struct tcphdr)];
355 const unsigned char *ptr;
356 int length = (tcph->doff*4) - sizeof(struct tcphdr);
361 ptr = skb_header_pointer(skb, dataoff + sizeof(struct tcphdr),
375 case TCPOPT_NOP: /* Ref: RFC 793 section 3.1 */
382 if (opsize < 2) /* "silly options" */
385 return; /* don't parse partial options */
387 if (opcode == TCPOPT_SACK_PERM
388 && opsize == TCPOLEN_SACK_PERM)
389 state->flags |= IP_CT_TCP_FLAG_SACK_PERM;
390 else if (opcode == TCPOPT_WINDOW
391 && opsize == TCPOLEN_WINDOW) {
392 state->td_scale = *(u_int8_t *)ptr;
394 if (state->td_scale > TCP_MAX_WSCALE)
395 state->td_scale = TCP_MAX_WSCALE;
398 IP_CT_TCP_FLAG_WINDOW_SCALE;
406 static void tcp_sack(const struct sk_buff *skb, unsigned int dataoff,
407 const struct tcphdr *tcph, __u32 *sack)
409 unsigned char buff[(15 * 4) - sizeof(struct tcphdr)];
410 const unsigned char *ptr;
411 int length = (tcph->doff*4) - sizeof(struct tcphdr);
417 ptr = skb_header_pointer(skb, dataoff + sizeof(struct tcphdr),
421 /* Fast path for timestamp-only option */
422 if (length == TCPOLEN_TSTAMP_ALIGNED
423 && *(__be32 *)ptr == htonl((TCPOPT_NOP << 24)
425 | (TCPOPT_TIMESTAMP << 8)
426 | TCPOLEN_TIMESTAMP))
436 case TCPOPT_NOP: /* Ref: RFC 793 section 3.1 */
443 if (opsize < 2) /* "silly options" */
446 return; /* don't parse partial options */
448 if (opcode == TCPOPT_SACK
449 && opsize >= (TCPOLEN_SACK_BASE
450 + TCPOLEN_SACK_PERBLOCK)
451 && !((opsize - TCPOLEN_SACK_BASE)
452 % TCPOLEN_SACK_PERBLOCK)) {
454 i < (opsize - TCPOLEN_SACK_BASE);
455 i += TCPOLEN_SACK_PERBLOCK) {
456 tmp = get_unaligned_be32((__be32 *)(ptr+i)+1);
458 if (after(tmp, *sack))
469 static bool tcp_in_window(const struct nf_conn *ct,
470 struct ip_ct_tcp *state,
471 enum ip_conntrack_dir dir,
473 const struct sk_buff *skb,
474 unsigned int dataoff,
475 const struct tcphdr *tcph)
477 struct net *net = nf_ct_net(ct);
478 struct nf_tcp_net *tn = tcp_pernet(net);
479 struct ip_ct_tcp_state *sender = &state->seen[dir];
480 struct ip_ct_tcp_state *receiver = &state->seen[!dir];
481 const struct nf_conntrack_tuple *tuple = &ct->tuplehash[dir].tuple;
482 __u32 seq, ack, sack, end, win, swin;
484 bool res, in_recv_win;
487 * Get the required data from the packet.
489 seq = ntohl(tcph->seq);
490 ack = sack = ntohl(tcph->ack_seq);
491 win = ntohs(tcph->window);
492 end = segment_seq_plus_len(seq, skb->len, dataoff, tcph);
494 if (receiver->flags & IP_CT_TCP_FLAG_SACK_PERM)
495 tcp_sack(skb, dataoff, tcph, &sack);
497 /* Take into account NAT sequence number mangling */
498 receiver_offset = nf_ct_seq_offset(ct, !dir, ack - 1);
499 ack -= receiver_offset;
500 sack -= receiver_offset;
502 pr_debug("tcp_in_window: START\n");
503 pr_debug("tcp_in_window: ");
504 nf_ct_dump_tuple(tuple);
505 pr_debug("seq=%u ack=%u+(%d) sack=%u+(%d) win=%u end=%u\n",
506 seq, ack, receiver_offset, sack, receiver_offset, win, end);
507 pr_debug("tcp_in_window: sender end=%u maxend=%u maxwin=%u scale=%i "
508 "receiver end=%u maxend=%u maxwin=%u scale=%i\n",
509 sender->td_end, sender->td_maxend, sender->td_maxwin,
511 receiver->td_end, receiver->td_maxend, receiver->td_maxwin,
514 if (sender->td_maxwin == 0) {
516 * Initialize sender data.
520 * SYN-ACK in reply to a SYN
521 * or SYN from reply direction in simultaneous open.
524 sender->td_maxend = end;
525 sender->td_maxwin = (win == 0 ? 1 : win);
527 tcp_options(skb, dataoff, tcph, sender);
530 * Both sides must send the Window Scale option
531 * to enable window scaling in either direction.
533 if (!(sender->flags & IP_CT_TCP_FLAG_WINDOW_SCALE
534 && receiver->flags & IP_CT_TCP_FLAG_WINDOW_SCALE))
536 receiver->td_scale = 0;
538 /* Simultaneous open */
542 * We are in the middle of a connection,
543 * its history is lost for us.
544 * Let's try to use the data from the packet.
546 sender->td_end = end;
547 swin = win << sender->td_scale;
548 sender->td_maxwin = (swin == 0 ? 1 : swin);
549 sender->td_maxend = end + sender->td_maxwin;
551 * We haven't seen traffic in the other direction yet
552 * but we have to tweak window tracking to pass III
553 * and IV until that happens.
555 if (receiver->td_maxwin == 0)
556 receiver->td_end = receiver->td_maxend = sack;
558 } else if (((state->state == TCP_CONNTRACK_SYN_SENT
559 && dir == IP_CT_DIR_ORIGINAL)
560 || (state->state == TCP_CONNTRACK_SYN_RECV
561 && dir == IP_CT_DIR_REPLY))
562 && after(end, sender->td_end)) {
564 * RFC 793: "if a TCP is reinitialized ... then it need
565 * not wait at all; it must only be sure to use sequence
566 * numbers larger than those recently used."
569 sender->td_maxend = end;
570 sender->td_maxwin = (win == 0 ? 1 : win);
572 tcp_options(skb, dataoff, tcph, sender);
577 * If there is no ACK, just pretend it was set and OK.
579 ack = sack = receiver->td_end;
580 } else if (((tcp_flag_word(tcph) & (TCP_FLAG_ACK|TCP_FLAG_RST)) ==
581 (TCP_FLAG_ACK|TCP_FLAG_RST))
584 * Broken TCP stacks, that set ACK in RST packets as well
585 * with zero ack value.
587 ack = sack = receiver->td_end;
590 if (tcph->rst && seq == 0 && state->state == TCP_CONNTRACK_SYN_SENT)
592 * RST sent answering SYN.
594 seq = end = sender->td_end;
596 pr_debug("tcp_in_window: ");
597 nf_ct_dump_tuple(tuple);
598 pr_debug("seq=%u ack=%u+(%d) sack=%u+(%d) win=%u end=%u\n",
599 seq, ack, receiver_offset, sack, receiver_offset, win, end);
600 pr_debug("tcp_in_window: sender end=%u maxend=%u maxwin=%u scale=%i "
601 "receiver end=%u maxend=%u maxwin=%u scale=%i\n",
602 sender->td_end, sender->td_maxend, sender->td_maxwin,
604 receiver->td_end, receiver->td_maxend, receiver->td_maxwin,
607 /* Is the ending sequence in the receive window (if available)? */
608 in_recv_win = !receiver->td_maxwin ||
609 after(end, sender->td_end - receiver->td_maxwin - 1);
611 pr_debug("tcp_in_window: I=%i II=%i III=%i IV=%i\n",
612 before(seq, sender->td_maxend + 1),
613 (in_recv_win ? 1 : 0),
614 before(sack, receiver->td_end + 1),
615 after(sack, receiver->td_end - MAXACKWINDOW(sender) - 1));
617 if (before(seq, sender->td_maxend + 1) &&
619 before(sack, receiver->td_end + 1) &&
620 after(sack, receiver->td_end - MAXACKWINDOW(sender) - 1)) {
622 * Take into account window scaling (RFC 1323).
625 win <<= sender->td_scale;
628 * Update sender data.
630 swin = win + (sack - ack);
631 if (sender->td_maxwin < swin)
632 sender->td_maxwin = swin;
633 if (after(end, sender->td_end)) {
634 sender->td_end = end;
635 sender->flags |= IP_CT_TCP_FLAG_DATA_UNACKNOWLEDGED;
638 if (!(sender->flags & IP_CT_TCP_FLAG_MAXACK_SET)) {
639 sender->td_maxack = ack;
640 sender->flags |= IP_CT_TCP_FLAG_MAXACK_SET;
641 } else if (after(ack, sender->td_maxack))
642 sender->td_maxack = ack;
646 * Update receiver data.
648 if (receiver->td_maxwin != 0 && after(end, sender->td_maxend))
649 receiver->td_maxwin += end - sender->td_maxend;
650 if (after(sack + win, receiver->td_maxend - 1)) {
651 receiver->td_maxend = sack + win;
653 receiver->td_maxend++;
655 if (ack == receiver->td_end)
656 receiver->flags &= ~IP_CT_TCP_FLAG_DATA_UNACKNOWLEDGED;
659 * Check retransmissions.
661 if (index == TCP_ACK_SET) {
662 if (state->last_dir == dir
663 && state->last_seq == seq
664 && state->last_ack == ack
665 && state->last_end == end
666 && state->last_win == win)
669 state->last_dir = dir;
670 state->last_seq = seq;
671 state->last_ack = ack;
672 state->last_end = end;
673 state->last_win = win;
680 if (sender->flags & IP_CT_TCP_FLAG_BE_LIBERAL ||
684 nf_ct_l4proto_log_invalid(skb, ct,
686 before(seq, sender->td_maxend + 1) ?
688 before(sack, receiver->td_end + 1) ?
689 after(sack, receiver->td_end - MAXACKWINDOW(sender) - 1) ? "BUG"
690 : "ACK is under the lower bound (possible overly delayed ACK)"
691 : "ACK is over the upper bound (ACKed data not seen yet)"
692 : "SEQ is under the lower bound (already ACKed data retransmitted)"
693 : "SEQ is over the upper bound (over the window of the receiver)");
697 pr_debug("tcp_in_window: res=%u sender end=%u maxend=%u maxwin=%u "
698 "receiver end=%u maxend=%u maxwin=%u\n",
699 res, sender->td_end, sender->td_maxend, sender->td_maxwin,
700 receiver->td_end, receiver->td_maxend, receiver->td_maxwin);
705 /* table of valid flag combinations - PUSH, ECE and CWR are always valid */
706 static const u8 tcp_valid_flags[(TCPHDR_FIN|TCPHDR_SYN|TCPHDR_RST|TCPHDR_ACK|
710 [TCPHDR_SYN|TCPHDR_URG] = 1,
711 [TCPHDR_SYN|TCPHDR_ACK] = 1,
713 [TCPHDR_RST|TCPHDR_ACK] = 1,
714 [TCPHDR_FIN|TCPHDR_ACK] = 1,
715 [TCPHDR_FIN|TCPHDR_ACK|TCPHDR_URG] = 1,
717 [TCPHDR_ACK|TCPHDR_URG] = 1,
720 static void tcp_error_log(const struct sk_buff *skb, struct net *net,
721 u8 pf, const char *msg)
723 nf_l4proto_log_invalid(skb, net, pf, IPPROTO_TCP, "%s", msg);
726 /* Protect conntrack agaist broken packets. Code taken from ipt_unclean.c. */
727 static int tcp_error(struct net *net, struct nf_conn *tmpl,
729 unsigned int dataoff,
731 unsigned int hooknum)
733 const struct tcphdr *th;
735 unsigned int tcplen = skb->len - dataoff;
738 /* Smaller that minimal TCP header? */
739 th = skb_header_pointer(skb, dataoff, sizeof(_tcph), &_tcph);
741 tcp_error_log(skb, net, pf, "short packet");
745 /* Not whole TCP header or malformed packet */
746 if (th->doff*4 < sizeof(struct tcphdr) || tcplen < th->doff*4) {
747 tcp_error_log(skb, net, pf, "truncated packet");
751 /* Checksum invalid? Ignore.
752 * We skip checking packets on the outgoing path
753 * because the checksum is assumed to be correct.
755 /* FIXME: Source route IP option packets --RR */
756 if (net->ct.sysctl_checksum && hooknum == NF_INET_PRE_ROUTING &&
757 nf_checksum(skb, hooknum, dataoff, IPPROTO_TCP, pf)) {
758 tcp_error_log(skb, net, pf, "bad checksum");
762 /* Check TCP flags. */
763 tcpflags = (tcp_flag_byte(th) & ~(TCPHDR_ECE|TCPHDR_CWR|TCPHDR_PSH));
764 if (!tcp_valid_flags[tcpflags]) {
765 tcp_error_log(skb, net, pf, "invalid tcp flag combination");
772 /* Returns verdict for packet, or -1 for invalid. */
773 static int tcp_packet(struct nf_conn *ct,
774 const struct sk_buff *skb,
775 unsigned int dataoff,
776 enum ip_conntrack_info ctinfo)
778 struct net *net = nf_ct_net(ct);
779 struct nf_tcp_net *tn = tcp_pernet(net);
780 struct nf_conntrack_tuple *tuple;
781 enum tcp_conntrack new_state, old_state;
782 unsigned int index, *timeouts;
783 enum ip_conntrack_dir dir;
784 const struct tcphdr *th;
786 unsigned long timeout;
788 th = skb_header_pointer(skb, dataoff, sizeof(_tcph), &_tcph);
791 spin_lock_bh(&ct->lock);
792 old_state = ct->proto.tcp.state;
793 dir = CTINFO2DIR(ctinfo);
794 index = get_conntrack_index(th);
795 new_state = tcp_conntracks[dir][index][old_state];
796 tuple = &ct->tuplehash[dir].tuple;
799 case TCP_CONNTRACK_SYN_SENT:
800 if (old_state < TCP_CONNTRACK_TIME_WAIT)
802 /* RFC 1122: "When a connection is closed actively,
803 * it MUST linger in TIME-WAIT state for a time 2xMSL
804 * (Maximum Segment Lifetime). However, it MAY accept
805 * a new SYN from the remote TCP to reopen the connection
806 * directly from TIME-WAIT state, if..."
807 * We ignore the conditions because we are in the
808 * TIME-WAIT state anyway.
810 * Handle aborted connections: we and the server
811 * think there is an existing connection but the client
812 * aborts it and starts a new one.
814 if (((ct->proto.tcp.seen[dir].flags
815 | ct->proto.tcp.seen[!dir].flags)
816 & IP_CT_TCP_FLAG_CLOSE_INIT)
817 || (ct->proto.tcp.last_dir == dir
818 && ct->proto.tcp.last_index == TCP_RST_SET)) {
819 /* Attempt to reopen a closed/aborted connection.
820 * Delete this connection and look up again. */
821 spin_unlock_bh(&ct->lock);
823 /* Only repeat if we can actually remove the timer.
824 * Destruction may already be in progress in process
825 * context and we must give it a chance to terminate.
832 case TCP_CONNTRACK_IGNORE:
835 * Our connection entry may be out of sync, so ignore
836 * packets which may signal the real connection between
837 * the client and the server.
840 * b) SYN/ACK in REPLY
841 * c) ACK in reply direction after initial SYN in original.
843 * If the ignored packet is invalid, the receiver will send
844 * a RST we'll catch below.
846 if (index == TCP_SYNACK_SET
847 && ct->proto.tcp.last_index == TCP_SYN_SET
848 && ct->proto.tcp.last_dir != dir
849 && ntohl(th->ack_seq) == ct->proto.tcp.last_end) {
850 /* b) This SYN/ACK acknowledges a SYN that we earlier
851 * ignored as invalid. This means that the client and
852 * the server are both in sync, while the firewall is
853 * not. We get in sync from the previously annotated
856 old_state = TCP_CONNTRACK_SYN_SENT;
857 new_state = TCP_CONNTRACK_SYN_RECV;
858 ct->proto.tcp.seen[ct->proto.tcp.last_dir].td_end =
859 ct->proto.tcp.last_end;
860 ct->proto.tcp.seen[ct->proto.tcp.last_dir].td_maxend =
861 ct->proto.tcp.last_end;
862 ct->proto.tcp.seen[ct->proto.tcp.last_dir].td_maxwin =
863 ct->proto.tcp.last_win == 0 ?
864 1 : ct->proto.tcp.last_win;
865 ct->proto.tcp.seen[ct->proto.tcp.last_dir].td_scale =
866 ct->proto.tcp.last_wscale;
867 ct->proto.tcp.last_flags &= ~IP_CT_EXP_CHALLENGE_ACK;
868 ct->proto.tcp.seen[ct->proto.tcp.last_dir].flags =
869 ct->proto.tcp.last_flags;
870 memset(&ct->proto.tcp.seen[dir], 0,
871 sizeof(struct ip_ct_tcp_state));
874 ct->proto.tcp.last_index = index;
875 ct->proto.tcp.last_dir = dir;
876 ct->proto.tcp.last_seq = ntohl(th->seq);
877 ct->proto.tcp.last_end =
878 segment_seq_plus_len(ntohl(th->seq), skb->len, dataoff, th);
879 ct->proto.tcp.last_win = ntohs(th->window);
881 /* a) This is a SYN in ORIGINAL. The client and the server
882 * may be in sync but we are not. In that case, we annotate
883 * the TCP options and let the packet go through. If it is a
884 * valid SYN packet, the server will reply with a SYN/ACK, and
885 * then we'll get in sync. Otherwise, the server potentially
886 * responds with a challenge ACK if implementing RFC5961.
888 if (index == TCP_SYN_SET && dir == IP_CT_DIR_ORIGINAL) {
889 struct ip_ct_tcp_state seen = {};
891 ct->proto.tcp.last_flags =
892 ct->proto.tcp.last_wscale = 0;
893 tcp_options(skb, dataoff, th, &seen);
894 if (seen.flags & IP_CT_TCP_FLAG_WINDOW_SCALE) {
895 ct->proto.tcp.last_flags |=
896 IP_CT_TCP_FLAG_WINDOW_SCALE;
897 ct->proto.tcp.last_wscale = seen.td_scale;
899 if (seen.flags & IP_CT_TCP_FLAG_SACK_PERM) {
900 ct->proto.tcp.last_flags |=
901 IP_CT_TCP_FLAG_SACK_PERM;
903 /* Mark the potential for RFC5961 challenge ACK,
904 * this pose a special problem for LAST_ACK state
905 * as ACK is intrepretated as ACKing last FIN.
907 if (old_state == TCP_CONNTRACK_LAST_ACK)
908 ct->proto.tcp.last_flags |=
909 IP_CT_EXP_CHALLENGE_ACK;
911 spin_unlock_bh(&ct->lock);
912 nf_ct_l4proto_log_invalid(skb, ct, "invalid packet ignored in "
913 "state %s ", tcp_conntrack_names[old_state]);
915 case TCP_CONNTRACK_MAX:
916 /* Special case for SYN proxy: when the SYN to the server or
917 * the SYN/ACK from the server is lost, the client may transmit
918 * a keep-alive packet while in SYN_SENT state. This needs to
919 * be associated with the original conntrack entry in order to
920 * generate a new SYN with the correct sequence number.
922 if (nfct_synproxy(ct) && old_state == TCP_CONNTRACK_SYN_SENT &&
923 index == TCP_ACK_SET && dir == IP_CT_DIR_ORIGINAL &&
924 ct->proto.tcp.last_dir == IP_CT_DIR_ORIGINAL &&
925 ct->proto.tcp.seen[dir].td_end - 1 == ntohl(th->seq)) {
926 pr_debug("nf_ct_tcp: SYN proxy client keep alive\n");
927 spin_unlock_bh(&ct->lock);
932 pr_debug("nf_ct_tcp: Invalid dir=%i index=%u ostate=%u\n",
933 dir, get_conntrack_index(th), old_state);
934 spin_unlock_bh(&ct->lock);
935 nf_ct_l4proto_log_invalid(skb, ct, "invalid state");
937 case TCP_CONNTRACK_TIME_WAIT:
938 /* RFC5961 compliance cause stack to send "challenge-ACK"
939 * e.g. in response to spurious SYNs. Conntrack MUST
940 * not believe this ACK is acking last FIN.
942 if (old_state == TCP_CONNTRACK_LAST_ACK &&
943 index == TCP_ACK_SET &&
944 ct->proto.tcp.last_dir != dir &&
945 ct->proto.tcp.last_index == TCP_SYN_SET &&
946 (ct->proto.tcp.last_flags & IP_CT_EXP_CHALLENGE_ACK)) {
947 /* Detected RFC5961 challenge ACK */
948 ct->proto.tcp.last_flags &= ~IP_CT_EXP_CHALLENGE_ACK;
949 spin_unlock_bh(&ct->lock);
950 nf_ct_l4proto_log_invalid(skb, ct, "challenge-ack ignored");
951 return NF_ACCEPT; /* Don't change state */
954 case TCP_CONNTRACK_SYN_SENT2:
955 /* tcp_conntracks table is not smart enough to handle
958 ct->proto.tcp.last_flags |= IP_CT_TCP_SIMULTANEOUS_OPEN;
960 case TCP_CONNTRACK_SYN_RECV:
961 if (dir == IP_CT_DIR_REPLY && index == TCP_ACK_SET &&
962 ct->proto.tcp.last_flags & IP_CT_TCP_SIMULTANEOUS_OPEN)
963 new_state = TCP_CONNTRACK_ESTABLISHED;
965 case TCP_CONNTRACK_CLOSE:
966 if (index == TCP_RST_SET
967 && (ct->proto.tcp.seen[!dir].flags & IP_CT_TCP_FLAG_MAXACK_SET)
968 && before(ntohl(th->seq), ct->proto.tcp.seen[!dir].td_maxack)) {
970 spin_unlock_bh(&ct->lock);
971 nf_ct_l4proto_log_invalid(skb, ct, "invalid rst");
974 if (index == TCP_RST_SET
975 && ((test_bit(IPS_SEEN_REPLY_BIT, &ct->status)
976 && ct->proto.tcp.last_index == TCP_SYN_SET)
977 || (!test_bit(IPS_ASSURED_BIT, &ct->status)
978 && ct->proto.tcp.last_index == TCP_ACK_SET))
979 && ntohl(th->ack_seq) == ct->proto.tcp.last_end) {
980 /* RST sent to invalid SYN or ACK we had let through
981 * at a) and c) above:
983 * a) SYN was in window then
984 * c) we hold a half-open connection.
986 * Delete our connection entry.
987 * We skip window checking, because packet might ACK
988 * segments we ignored. */
991 /* Just fall through */
993 /* Keep compilers happy. */
997 if (!tcp_in_window(ct, &ct->proto.tcp, dir, index,
999 spin_unlock_bh(&ct->lock);
1003 /* From now on we have got in-window packets */
1004 ct->proto.tcp.last_index = index;
1005 ct->proto.tcp.last_dir = dir;
1007 pr_debug("tcp_conntracks: ");
1008 nf_ct_dump_tuple(tuple);
1009 pr_debug("syn=%i ack=%i fin=%i rst=%i old=%i new=%i\n",
1010 (th->syn ? 1 : 0), (th->ack ? 1 : 0),
1011 (th->fin ? 1 : 0), (th->rst ? 1 : 0),
1012 old_state, new_state);
1014 ct->proto.tcp.state = new_state;
1015 if (old_state != new_state
1016 && new_state == TCP_CONNTRACK_FIN_WAIT)
1017 ct->proto.tcp.seen[dir].flags |= IP_CT_TCP_FLAG_CLOSE_INIT;
1019 timeouts = nf_ct_timeout_lookup(ct);
1021 timeouts = tn->timeouts;
1023 if (ct->proto.tcp.retrans >= tn->tcp_max_retrans &&
1024 timeouts[new_state] > timeouts[TCP_CONNTRACK_RETRANS])
1025 timeout = timeouts[TCP_CONNTRACK_RETRANS];
1026 else if ((ct->proto.tcp.seen[0].flags | ct->proto.tcp.seen[1].flags) &
1027 IP_CT_TCP_FLAG_DATA_UNACKNOWLEDGED &&
1028 timeouts[new_state] > timeouts[TCP_CONNTRACK_UNACK])
1029 timeout = timeouts[TCP_CONNTRACK_UNACK];
1030 else if (ct->proto.tcp.last_win == 0 &&
1031 timeouts[new_state] > timeouts[TCP_CONNTRACK_RETRANS])
1032 timeout = timeouts[TCP_CONNTRACK_RETRANS];
1034 timeout = timeouts[new_state];
1035 spin_unlock_bh(&ct->lock);
1037 if (new_state != old_state)
1038 nf_conntrack_event_cache(IPCT_PROTOINFO, ct);
1040 if (!test_bit(IPS_SEEN_REPLY_BIT, &ct->status)) {
1041 /* If only reply is a RST, we can consider ourselves not to
1042 have an established connection: this is a fairly common
1043 problem case, so we can delete the conntrack
1044 immediately. --RR */
1046 nf_ct_kill_acct(ct, ctinfo, skb);
1049 /* ESTABLISHED without SEEN_REPLY, i.e. mid-connection
1050 * pickup with loose=1. Avoid large ESTABLISHED timeout.
1052 if (new_state == TCP_CONNTRACK_ESTABLISHED &&
1053 timeout > timeouts[TCP_CONNTRACK_UNACK])
1054 timeout = timeouts[TCP_CONNTRACK_UNACK];
1055 } else if (!test_bit(IPS_ASSURED_BIT, &ct->status)
1056 && (old_state == TCP_CONNTRACK_SYN_RECV
1057 || old_state == TCP_CONNTRACK_ESTABLISHED)
1058 && new_state == TCP_CONNTRACK_ESTABLISHED) {
1059 /* Set ASSURED if we see see valid ack in ESTABLISHED
1060 after SYN_RECV or a valid answer for a picked up
1062 set_bit(IPS_ASSURED_BIT, &ct->status);
1063 nf_conntrack_event_cache(IPCT_ASSURED, ct);
1065 nf_ct_refresh_acct(ct, ctinfo, skb, timeout);
1070 /* Called when a new connection for this protocol found. */
1071 static bool tcp_new(struct nf_conn *ct, const struct sk_buff *skb,
1072 unsigned int dataoff)
1074 enum tcp_conntrack new_state;
1075 const struct tcphdr *th;
1076 struct tcphdr _tcph;
1077 struct net *net = nf_ct_net(ct);
1078 struct nf_tcp_net *tn = tcp_pernet(net);
1079 const struct ip_ct_tcp_state *sender = &ct->proto.tcp.seen[0];
1080 const struct ip_ct_tcp_state *receiver = &ct->proto.tcp.seen[1];
1082 th = skb_header_pointer(skb, dataoff, sizeof(_tcph), &_tcph);
1085 /* Don't need lock here: this conntrack not in circulation yet */
1086 new_state = tcp_conntracks[0][get_conntrack_index(th)][TCP_CONNTRACK_NONE];
1088 /* Invalid: delete conntrack */
1089 if (new_state >= TCP_CONNTRACK_MAX) {
1090 pr_debug("nf_ct_tcp: invalid new deleting.\n");
1094 if (new_state == TCP_CONNTRACK_SYN_SENT) {
1095 memset(&ct->proto.tcp, 0, sizeof(ct->proto.tcp));
1097 ct->proto.tcp.seen[0].td_end =
1098 segment_seq_plus_len(ntohl(th->seq), skb->len,
1100 ct->proto.tcp.seen[0].td_maxwin = ntohs(th->window);
1101 if (ct->proto.tcp.seen[0].td_maxwin == 0)
1102 ct->proto.tcp.seen[0].td_maxwin = 1;
1103 ct->proto.tcp.seen[0].td_maxend =
1104 ct->proto.tcp.seen[0].td_end;
1106 tcp_options(skb, dataoff, th, &ct->proto.tcp.seen[0]);
1107 } else if (tn->tcp_loose == 0) {
1108 /* Don't try to pick up connections. */
1111 memset(&ct->proto.tcp, 0, sizeof(ct->proto.tcp));
1113 * We are in the middle of a connection,
1114 * its history is lost for us.
1115 * Let's try to use the data from the packet.
1117 ct->proto.tcp.seen[0].td_end =
1118 segment_seq_plus_len(ntohl(th->seq), skb->len,
1120 ct->proto.tcp.seen[0].td_maxwin = ntohs(th->window);
1121 if (ct->proto.tcp.seen[0].td_maxwin == 0)
1122 ct->proto.tcp.seen[0].td_maxwin = 1;
1123 ct->proto.tcp.seen[0].td_maxend =
1124 ct->proto.tcp.seen[0].td_end +
1125 ct->proto.tcp.seen[0].td_maxwin;
1127 /* We assume SACK and liberal window checking to handle
1129 ct->proto.tcp.seen[0].flags =
1130 ct->proto.tcp.seen[1].flags = IP_CT_TCP_FLAG_SACK_PERM |
1131 IP_CT_TCP_FLAG_BE_LIBERAL;
1134 /* tcp_packet will set them */
1135 ct->proto.tcp.last_index = TCP_NONE_SET;
1137 pr_debug("tcp_new: sender end=%u maxend=%u maxwin=%u scale=%i "
1138 "receiver end=%u maxend=%u maxwin=%u scale=%i\n",
1139 sender->td_end, sender->td_maxend, sender->td_maxwin,
1141 receiver->td_end, receiver->td_maxend, receiver->td_maxwin,
1142 receiver->td_scale);
1146 static bool tcp_can_early_drop(const struct nf_conn *ct)
1148 switch (ct->proto.tcp.state) {
1149 case TCP_CONNTRACK_FIN_WAIT:
1150 case TCP_CONNTRACK_LAST_ACK:
1151 case TCP_CONNTRACK_TIME_WAIT:
1152 case TCP_CONNTRACK_CLOSE:
1153 case TCP_CONNTRACK_CLOSE_WAIT:
1162 #if IS_ENABLED(CONFIG_NF_CT_NETLINK)
1164 #include <linux/netfilter/nfnetlink.h>
1165 #include <linux/netfilter/nfnetlink_conntrack.h>
1167 static int tcp_to_nlattr(struct sk_buff *skb, struct nlattr *nla,
1170 struct nlattr *nest_parms;
1171 struct nf_ct_tcp_flags tmp = {};
1173 spin_lock_bh(&ct->lock);
1174 nest_parms = nla_nest_start(skb, CTA_PROTOINFO_TCP | NLA_F_NESTED);
1176 goto nla_put_failure;
1178 if (nla_put_u8(skb, CTA_PROTOINFO_TCP_STATE, ct->proto.tcp.state) ||
1179 nla_put_u8(skb, CTA_PROTOINFO_TCP_WSCALE_ORIGINAL,
1180 ct->proto.tcp.seen[0].td_scale) ||
1181 nla_put_u8(skb, CTA_PROTOINFO_TCP_WSCALE_REPLY,
1182 ct->proto.tcp.seen[1].td_scale))
1183 goto nla_put_failure;
1185 tmp.flags = ct->proto.tcp.seen[0].flags;
1186 if (nla_put(skb, CTA_PROTOINFO_TCP_FLAGS_ORIGINAL,
1187 sizeof(struct nf_ct_tcp_flags), &tmp))
1188 goto nla_put_failure;
1190 tmp.flags = ct->proto.tcp.seen[1].flags;
1191 if (nla_put(skb, CTA_PROTOINFO_TCP_FLAGS_REPLY,
1192 sizeof(struct nf_ct_tcp_flags), &tmp))
1193 goto nla_put_failure;
1194 spin_unlock_bh(&ct->lock);
1196 nla_nest_end(skb, nest_parms);
1201 spin_unlock_bh(&ct->lock);
1205 static const struct nla_policy tcp_nla_policy[CTA_PROTOINFO_TCP_MAX+1] = {
1206 [CTA_PROTOINFO_TCP_STATE] = { .type = NLA_U8 },
1207 [CTA_PROTOINFO_TCP_WSCALE_ORIGINAL] = { .type = NLA_U8 },
1208 [CTA_PROTOINFO_TCP_WSCALE_REPLY] = { .type = NLA_U8 },
1209 [CTA_PROTOINFO_TCP_FLAGS_ORIGINAL] = { .len = sizeof(struct nf_ct_tcp_flags) },
1210 [CTA_PROTOINFO_TCP_FLAGS_REPLY] = { .len = sizeof(struct nf_ct_tcp_flags) },
1213 #define TCP_NLATTR_SIZE ( \
1214 NLA_ALIGN(NLA_HDRLEN + 1) + \
1215 NLA_ALIGN(NLA_HDRLEN + 1) + \
1216 NLA_ALIGN(NLA_HDRLEN + sizeof(sizeof(struct nf_ct_tcp_flags))) + \
1217 NLA_ALIGN(NLA_HDRLEN + sizeof(sizeof(struct nf_ct_tcp_flags))))
1219 static int nlattr_to_tcp(struct nlattr *cda[], struct nf_conn *ct)
1221 struct nlattr *pattr = cda[CTA_PROTOINFO_TCP];
1222 struct nlattr *tb[CTA_PROTOINFO_TCP_MAX+1];
1225 /* updates could not contain anything about the private
1226 * protocol info, in that case skip the parsing */
1230 err = nla_parse_nested(tb, CTA_PROTOINFO_TCP_MAX, pattr,
1231 tcp_nla_policy, NULL);
1235 if (tb[CTA_PROTOINFO_TCP_STATE] &&
1236 nla_get_u8(tb[CTA_PROTOINFO_TCP_STATE]) >= TCP_CONNTRACK_MAX)
1239 spin_lock_bh(&ct->lock);
1240 if (tb[CTA_PROTOINFO_TCP_STATE])
1241 ct->proto.tcp.state = nla_get_u8(tb[CTA_PROTOINFO_TCP_STATE]);
1243 if (tb[CTA_PROTOINFO_TCP_FLAGS_ORIGINAL]) {
1244 struct nf_ct_tcp_flags *attr =
1245 nla_data(tb[CTA_PROTOINFO_TCP_FLAGS_ORIGINAL]);
1246 ct->proto.tcp.seen[0].flags &= ~attr->mask;
1247 ct->proto.tcp.seen[0].flags |= attr->flags & attr->mask;
1250 if (tb[CTA_PROTOINFO_TCP_FLAGS_REPLY]) {
1251 struct nf_ct_tcp_flags *attr =
1252 nla_data(tb[CTA_PROTOINFO_TCP_FLAGS_REPLY]);
1253 ct->proto.tcp.seen[1].flags &= ~attr->mask;
1254 ct->proto.tcp.seen[1].flags |= attr->flags & attr->mask;
1257 if (tb[CTA_PROTOINFO_TCP_WSCALE_ORIGINAL] &&
1258 tb[CTA_PROTOINFO_TCP_WSCALE_REPLY] &&
1259 ct->proto.tcp.seen[0].flags & IP_CT_TCP_FLAG_WINDOW_SCALE &&
1260 ct->proto.tcp.seen[1].flags & IP_CT_TCP_FLAG_WINDOW_SCALE) {
1261 ct->proto.tcp.seen[0].td_scale =
1262 nla_get_u8(tb[CTA_PROTOINFO_TCP_WSCALE_ORIGINAL]);
1263 ct->proto.tcp.seen[1].td_scale =
1264 nla_get_u8(tb[CTA_PROTOINFO_TCP_WSCALE_REPLY]);
1266 spin_unlock_bh(&ct->lock);
1271 static unsigned int tcp_nlattr_tuple_size(void)
1273 static unsigned int size __read_mostly;
1276 size = nla_policy_len(nf_ct_port_nla_policy, CTA_PROTO_MAX + 1);
1282 #if IS_ENABLED(CONFIG_NF_CT_NETLINK_TIMEOUT)
1284 #include <linux/netfilter/nfnetlink.h>
1285 #include <linux/netfilter/nfnetlink_cttimeout.h>
1287 static int tcp_timeout_nlattr_to_obj(struct nlattr *tb[],
1288 struct net *net, void *data)
1290 struct nf_tcp_net *tn = tcp_pernet(net);
1291 unsigned int *timeouts = data;
1295 timeouts = tn->timeouts;
1296 /* set default TCP timeouts. */
1297 for (i=0; i<TCP_CONNTRACK_TIMEOUT_MAX; i++)
1298 timeouts[i] = tn->timeouts[i];
1300 if (tb[CTA_TIMEOUT_TCP_SYN_SENT]) {
1301 timeouts[TCP_CONNTRACK_SYN_SENT] =
1302 ntohl(nla_get_be32(tb[CTA_TIMEOUT_TCP_SYN_SENT]))*HZ;
1304 if (tb[CTA_TIMEOUT_TCP_SYN_RECV]) {
1305 timeouts[TCP_CONNTRACK_SYN_RECV] =
1306 ntohl(nla_get_be32(tb[CTA_TIMEOUT_TCP_SYN_RECV]))*HZ;
1308 if (tb[CTA_TIMEOUT_TCP_ESTABLISHED]) {
1309 timeouts[TCP_CONNTRACK_ESTABLISHED] =
1310 ntohl(nla_get_be32(tb[CTA_TIMEOUT_TCP_ESTABLISHED]))*HZ;
1312 if (tb[CTA_TIMEOUT_TCP_FIN_WAIT]) {
1313 timeouts[TCP_CONNTRACK_FIN_WAIT] =
1314 ntohl(nla_get_be32(tb[CTA_TIMEOUT_TCP_FIN_WAIT]))*HZ;
1316 if (tb[CTA_TIMEOUT_TCP_CLOSE_WAIT]) {
1317 timeouts[TCP_CONNTRACK_CLOSE_WAIT] =
1318 ntohl(nla_get_be32(tb[CTA_TIMEOUT_TCP_CLOSE_WAIT]))*HZ;
1320 if (tb[CTA_TIMEOUT_TCP_LAST_ACK]) {
1321 timeouts[TCP_CONNTRACK_LAST_ACK] =
1322 ntohl(nla_get_be32(tb[CTA_TIMEOUT_TCP_LAST_ACK]))*HZ;
1324 if (tb[CTA_TIMEOUT_TCP_TIME_WAIT]) {
1325 timeouts[TCP_CONNTRACK_TIME_WAIT] =
1326 ntohl(nla_get_be32(tb[CTA_TIMEOUT_TCP_TIME_WAIT]))*HZ;
1328 if (tb[CTA_TIMEOUT_TCP_CLOSE]) {
1329 timeouts[TCP_CONNTRACK_CLOSE] =
1330 ntohl(nla_get_be32(tb[CTA_TIMEOUT_TCP_CLOSE]))*HZ;
1332 if (tb[CTA_TIMEOUT_TCP_SYN_SENT2]) {
1333 timeouts[TCP_CONNTRACK_SYN_SENT2] =
1334 ntohl(nla_get_be32(tb[CTA_TIMEOUT_TCP_SYN_SENT2]))*HZ;
1336 if (tb[CTA_TIMEOUT_TCP_RETRANS]) {
1337 timeouts[TCP_CONNTRACK_RETRANS] =
1338 ntohl(nla_get_be32(tb[CTA_TIMEOUT_TCP_RETRANS]))*HZ;
1340 if (tb[CTA_TIMEOUT_TCP_UNACK]) {
1341 timeouts[TCP_CONNTRACK_UNACK] =
1342 ntohl(nla_get_be32(tb[CTA_TIMEOUT_TCP_UNACK]))*HZ;
1348 tcp_timeout_obj_to_nlattr(struct sk_buff *skb, const void *data)
1350 const unsigned int *timeouts = data;
1352 if (nla_put_be32(skb, CTA_TIMEOUT_TCP_SYN_SENT,
1353 htonl(timeouts[TCP_CONNTRACK_SYN_SENT] / HZ)) ||
1354 nla_put_be32(skb, CTA_TIMEOUT_TCP_SYN_RECV,
1355 htonl(timeouts[TCP_CONNTRACK_SYN_RECV] / HZ)) ||
1356 nla_put_be32(skb, CTA_TIMEOUT_TCP_ESTABLISHED,
1357 htonl(timeouts[TCP_CONNTRACK_ESTABLISHED] / HZ)) ||
1358 nla_put_be32(skb, CTA_TIMEOUT_TCP_FIN_WAIT,
1359 htonl(timeouts[TCP_CONNTRACK_FIN_WAIT] / HZ)) ||
1360 nla_put_be32(skb, CTA_TIMEOUT_TCP_CLOSE_WAIT,
1361 htonl(timeouts[TCP_CONNTRACK_CLOSE_WAIT] / HZ)) ||
1362 nla_put_be32(skb, CTA_TIMEOUT_TCP_LAST_ACK,
1363 htonl(timeouts[TCP_CONNTRACK_LAST_ACK] / HZ)) ||
1364 nla_put_be32(skb, CTA_TIMEOUT_TCP_TIME_WAIT,
1365 htonl(timeouts[TCP_CONNTRACK_TIME_WAIT] / HZ)) ||
1366 nla_put_be32(skb, CTA_TIMEOUT_TCP_CLOSE,
1367 htonl(timeouts[TCP_CONNTRACK_CLOSE] / HZ)) ||
1368 nla_put_be32(skb, CTA_TIMEOUT_TCP_SYN_SENT2,
1369 htonl(timeouts[TCP_CONNTRACK_SYN_SENT2] / HZ)) ||
1370 nla_put_be32(skb, CTA_TIMEOUT_TCP_RETRANS,
1371 htonl(timeouts[TCP_CONNTRACK_RETRANS] / HZ)) ||
1372 nla_put_be32(skb, CTA_TIMEOUT_TCP_UNACK,
1373 htonl(timeouts[TCP_CONNTRACK_UNACK] / HZ)))
1374 goto nla_put_failure;
1381 static const struct nla_policy tcp_timeout_nla_policy[CTA_TIMEOUT_TCP_MAX+1] = {
1382 [CTA_TIMEOUT_TCP_SYN_SENT] = { .type = NLA_U32 },
1383 [CTA_TIMEOUT_TCP_SYN_RECV] = { .type = NLA_U32 },
1384 [CTA_TIMEOUT_TCP_ESTABLISHED] = { .type = NLA_U32 },
1385 [CTA_TIMEOUT_TCP_FIN_WAIT] = { .type = NLA_U32 },
1386 [CTA_TIMEOUT_TCP_CLOSE_WAIT] = { .type = NLA_U32 },
1387 [CTA_TIMEOUT_TCP_LAST_ACK] = { .type = NLA_U32 },
1388 [CTA_TIMEOUT_TCP_TIME_WAIT] = { .type = NLA_U32 },
1389 [CTA_TIMEOUT_TCP_CLOSE] = { .type = NLA_U32 },
1390 [CTA_TIMEOUT_TCP_SYN_SENT2] = { .type = NLA_U32 },
1391 [CTA_TIMEOUT_TCP_RETRANS] = { .type = NLA_U32 },
1392 [CTA_TIMEOUT_TCP_UNACK] = { .type = NLA_U32 },
1394 #endif /* CONFIG_NF_CT_NETLINK_TIMEOUT */
1396 #ifdef CONFIG_SYSCTL
1397 static struct ctl_table tcp_sysctl_table[] = {
1399 .procname = "nf_conntrack_tcp_timeout_syn_sent",
1400 .maxlen = sizeof(unsigned int),
1402 .proc_handler = proc_dointvec_jiffies,
1405 .procname = "nf_conntrack_tcp_timeout_syn_recv",
1406 .maxlen = sizeof(unsigned int),
1408 .proc_handler = proc_dointvec_jiffies,
1411 .procname = "nf_conntrack_tcp_timeout_established",
1412 .maxlen = sizeof(unsigned int),
1414 .proc_handler = proc_dointvec_jiffies,
1417 .procname = "nf_conntrack_tcp_timeout_fin_wait",
1418 .maxlen = sizeof(unsigned int),
1420 .proc_handler = proc_dointvec_jiffies,
1423 .procname = "nf_conntrack_tcp_timeout_close_wait",
1424 .maxlen = sizeof(unsigned int),
1426 .proc_handler = proc_dointvec_jiffies,
1429 .procname = "nf_conntrack_tcp_timeout_last_ack",
1430 .maxlen = sizeof(unsigned int),
1432 .proc_handler = proc_dointvec_jiffies,
1435 .procname = "nf_conntrack_tcp_timeout_time_wait",
1436 .maxlen = sizeof(unsigned int),
1438 .proc_handler = proc_dointvec_jiffies,
1441 .procname = "nf_conntrack_tcp_timeout_close",
1442 .maxlen = sizeof(unsigned int),
1444 .proc_handler = proc_dointvec_jiffies,
1447 .procname = "nf_conntrack_tcp_timeout_max_retrans",
1448 .maxlen = sizeof(unsigned int),
1450 .proc_handler = proc_dointvec_jiffies,
1453 .procname = "nf_conntrack_tcp_timeout_unacknowledged",
1454 .maxlen = sizeof(unsigned int),
1456 .proc_handler = proc_dointvec_jiffies,
1459 .procname = "nf_conntrack_tcp_loose",
1460 .maxlen = sizeof(unsigned int),
1462 .proc_handler = proc_dointvec,
1465 .procname = "nf_conntrack_tcp_be_liberal",
1466 .maxlen = sizeof(unsigned int),
1468 .proc_handler = proc_dointvec,
1471 .procname = "nf_conntrack_tcp_max_retrans",
1472 .maxlen = sizeof(unsigned int),
1474 .proc_handler = proc_dointvec,
1478 #endif /* CONFIG_SYSCTL */
1480 static int tcp_kmemdup_sysctl_table(struct nf_proto_net *pn,
1481 struct nf_tcp_net *tn)
1483 #ifdef CONFIG_SYSCTL
1487 pn->ctl_table = kmemdup(tcp_sysctl_table,
1488 sizeof(tcp_sysctl_table),
1493 pn->ctl_table[0].data = &tn->timeouts[TCP_CONNTRACK_SYN_SENT];
1494 pn->ctl_table[1].data = &tn->timeouts[TCP_CONNTRACK_SYN_RECV];
1495 pn->ctl_table[2].data = &tn->timeouts[TCP_CONNTRACK_ESTABLISHED];
1496 pn->ctl_table[3].data = &tn->timeouts[TCP_CONNTRACK_FIN_WAIT];
1497 pn->ctl_table[4].data = &tn->timeouts[TCP_CONNTRACK_CLOSE_WAIT];
1498 pn->ctl_table[5].data = &tn->timeouts[TCP_CONNTRACK_LAST_ACK];
1499 pn->ctl_table[6].data = &tn->timeouts[TCP_CONNTRACK_TIME_WAIT];
1500 pn->ctl_table[7].data = &tn->timeouts[TCP_CONNTRACK_CLOSE];
1501 pn->ctl_table[8].data = &tn->timeouts[TCP_CONNTRACK_RETRANS];
1502 pn->ctl_table[9].data = &tn->timeouts[TCP_CONNTRACK_UNACK];
1503 pn->ctl_table[10].data = &tn->tcp_loose;
1504 pn->ctl_table[11].data = &tn->tcp_be_liberal;
1505 pn->ctl_table[12].data = &tn->tcp_max_retrans;
1510 static int tcp_init_net(struct net *net, u_int16_t proto)
1512 struct nf_tcp_net *tn = tcp_pernet(net);
1513 struct nf_proto_net *pn = &tn->pn;
1518 for (i = 0; i < TCP_CONNTRACK_TIMEOUT_MAX; i++)
1519 tn->timeouts[i] = tcp_timeouts[i];
1521 tn->tcp_loose = nf_ct_tcp_loose;
1522 tn->tcp_be_liberal = nf_ct_tcp_be_liberal;
1523 tn->tcp_max_retrans = nf_ct_tcp_max_retrans;
1526 return tcp_kmemdup_sysctl_table(pn, tn);
1529 static struct nf_proto_net *tcp_get_net_proto(struct net *net)
1531 return &net->ct.nf_ct_proto.tcp.pn;
1534 const struct nf_conntrack_l4proto nf_conntrack_l4proto_tcp4 =
1537 .l4proto = IPPROTO_TCP,
1538 #ifdef CONFIG_NF_CONNTRACK_PROCFS
1539 .print_conntrack = tcp_print_conntrack,
1541 .packet = tcp_packet,
1544 .can_early_drop = tcp_can_early_drop,
1545 #if IS_ENABLED(CONFIG_NF_CT_NETLINK)
1546 .to_nlattr = tcp_to_nlattr,
1547 .from_nlattr = nlattr_to_tcp,
1548 .tuple_to_nlattr = nf_ct_port_tuple_to_nlattr,
1549 .nlattr_to_tuple = nf_ct_port_nlattr_to_tuple,
1550 .nlattr_tuple_size = tcp_nlattr_tuple_size,
1551 .nlattr_size = TCP_NLATTR_SIZE,
1552 .nla_policy = nf_ct_port_nla_policy,
1554 #if IS_ENABLED(CONFIG_NF_CT_NETLINK_TIMEOUT)
1556 .nlattr_to_obj = tcp_timeout_nlattr_to_obj,
1557 .obj_to_nlattr = tcp_timeout_obj_to_nlattr,
1558 .nlattr_max = CTA_TIMEOUT_TCP_MAX,
1559 .obj_size = sizeof(unsigned int) *
1560 TCP_CONNTRACK_TIMEOUT_MAX,
1561 .nla_policy = tcp_timeout_nla_policy,
1563 #endif /* CONFIG_NF_CT_NETLINK_TIMEOUT */
1564 .init_net = tcp_init_net,
1565 .get_net_proto = tcp_get_net_proto,
1567 EXPORT_SYMBOL_GPL(nf_conntrack_l4proto_tcp4);
1569 const struct nf_conntrack_l4proto nf_conntrack_l4proto_tcp6 =
1571 .l3proto = PF_INET6,
1572 .l4proto = IPPROTO_TCP,
1573 #ifdef CONFIG_NF_CONNTRACK_PROCFS
1574 .print_conntrack = tcp_print_conntrack,
1576 .packet = tcp_packet,
1579 .can_early_drop = tcp_can_early_drop,
1580 #if IS_ENABLED(CONFIG_NF_CT_NETLINK)
1581 .nlattr_size = TCP_NLATTR_SIZE,
1582 .to_nlattr = tcp_to_nlattr,
1583 .from_nlattr = nlattr_to_tcp,
1584 .tuple_to_nlattr = nf_ct_port_tuple_to_nlattr,
1585 .nlattr_to_tuple = nf_ct_port_nlattr_to_tuple,
1586 .nlattr_tuple_size = tcp_nlattr_tuple_size,
1587 .nla_policy = nf_ct_port_nla_policy,
1589 #if IS_ENABLED(CONFIG_NF_CT_NETLINK_TIMEOUT)
1591 .nlattr_to_obj = tcp_timeout_nlattr_to_obj,
1592 .obj_to_nlattr = tcp_timeout_obj_to_nlattr,
1593 .nlattr_max = CTA_TIMEOUT_TCP_MAX,
1594 .obj_size = sizeof(unsigned int) *
1595 TCP_CONNTRACK_TIMEOUT_MAX,
1596 .nla_policy = tcp_timeout_nla_policy,
1598 #endif /* CONFIG_NF_CT_NETLINK_TIMEOUT */
1599 .init_net = tcp_init_net,
1600 .get_net_proto = tcp_get_net_proto,
1602 EXPORT_SYMBOL_GPL(nf_conntrack_l4proto_tcp6);