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_log.h>
33 #include <net/netfilter/ipv4/nf_conntrack_ipv4.h>
34 #include <net/netfilter/ipv6/nf_conntrack_ipv6.h>
36 /* "Be conservative in what you do,
37 be liberal in what you accept from others."
38 If it's non-zero, we mark only out of window RST segments as INVALID. */
39 static int nf_ct_tcp_be_liberal __read_mostly = 0;
41 /* If it is set to zero, we disable picking up already established
43 static int nf_ct_tcp_loose __read_mostly = 1;
45 /* Max number of the retransmitted packets without receiving an (acceptable)
46 ACK from the destination. If this number is reached, a shorter timer
48 static int nf_ct_tcp_max_retrans __read_mostly = 3;
50 /* FIXME: Examine ipfilter's timeouts and conntrack transitions more
51 closely. They're more complex. --RR */
53 static const char *const tcp_conntrack_names[] = {
67 #define MINS * 60 SECS
68 #define HOURS * 60 MINS
69 #define DAYS * 24 HOURS
71 static unsigned int tcp_timeouts[TCP_CONNTRACK_TIMEOUT_MAX] __read_mostly = {
72 [TCP_CONNTRACK_SYN_SENT] = 2 MINS,
73 [TCP_CONNTRACK_SYN_RECV] = 60 SECS,
74 [TCP_CONNTRACK_ESTABLISHED] = 5 DAYS,
75 [TCP_CONNTRACK_FIN_WAIT] = 2 MINS,
76 [TCP_CONNTRACK_CLOSE_WAIT] = 60 SECS,
77 [TCP_CONNTRACK_LAST_ACK] = 30 SECS,
78 [TCP_CONNTRACK_TIME_WAIT] = 2 MINS,
79 [TCP_CONNTRACK_CLOSE] = 10 SECS,
80 [TCP_CONNTRACK_SYN_SENT2] = 2 MINS,
81 /* RFC1122 says the R2 limit should be at least 100 seconds.
82 Linux uses 15 packets as limit, which corresponds
83 to ~13-30min depending on RTO. */
84 [TCP_CONNTRACK_RETRANS] = 5 MINS,
85 [TCP_CONNTRACK_UNACK] = 5 MINS,
88 #define sNO TCP_CONNTRACK_NONE
89 #define sSS TCP_CONNTRACK_SYN_SENT
90 #define sSR TCP_CONNTRACK_SYN_RECV
91 #define sES TCP_CONNTRACK_ESTABLISHED
92 #define sFW TCP_CONNTRACK_FIN_WAIT
93 #define sCW TCP_CONNTRACK_CLOSE_WAIT
94 #define sLA TCP_CONNTRACK_LAST_ACK
95 #define sTW TCP_CONNTRACK_TIME_WAIT
96 #define sCL TCP_CONNTRACK_CLOSE
97 #define sS2 TCP_CONNTRACK_SYN_SENT2
98 #define sIV TCP_CONNTRACK_MAX
99 #define sIG TCP_CONNTRACK_IGNORE
101 /* What TCP flags are set from RST/SYN/FIN/ACK. */
112 * The TCP state transition table needs a few words...
114 * We are the man in the middle. All the packets go through us
115 * but might get lost in transit to the destination.
116 * It is assumed that the destinations can't receive segments
119 * The checked segment is in window, but our windows are *not*
120 * equivalent with the ones of the sender/receiver. We always
121 * try to guess the state of the current sender.
123 * The meaning of the states are:
125 * NONE: initial state
126 * SYN_SENT: SYN-only packet seen
127 * SYN_SENT2: SYN-only packet seen from reply dir, simultaneous open
128 * SYN_RECV: SYN-ACK packet seen
129 * ESTABLISHED: ACK packet seen
130 * FIN_WAIT: FIN packet seen
131 * CLOSE_WAIT: ACK seen (after FIN)
132 * LAST_ACK: FIN seen (after FIN)
133 * TIME_WAIT: last ACK seen
134 * CLOSE: closed connection (RST)
136 * Packets marked as IGNORED (sIG):
137 * if they may be either invalid or valid
138 * and the receiver may send back a connection
139 * closing RST or a SYN/ACK.
141 * Packets marked as INVALID (sIV):
142 * if we regard them as truly invalid packets
144 static const u8 tcp_conntracks[2][6][TCP_CONNTRACK_MAX] = {
147 /* sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sS2 */
148 /*syn*/ { sSS, sSS, sIG, sIG, sIG, sIG, sIG, sSS, sSS, sS2 },
150 * sNO -> sSS Initialize a new connection
151 * sSS -> sSS Retransmitted SYN
152 * sS2 -> sS2 Late retransmitted SYN
154 * sES -> sIG Error: SYNs in window outside the SYN_SENT state
155 * are errors. Receiver will reply with RST
156 * and close the connection.
157 * Or we are not in sync and hold a dead connection.
161 * sTW -> sSS Reopened connection (RFC 1122).
164 /* sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sS2 */
165 /*synack*/ { sIV, sIV, sSR, sIV, sIV, sIV, sIV, sIV, sIV, sSR },
167 * sNO -> sIV Too late and no reason to do anything
168 * sSS -> sIV Client can't send SYN and then SYN/ACK
169 * sS2 -> sSR SYN/ACK sent to SYN2 in simultaneous open
170 * sSR -> sSR Late retransmitted SYN/ACK in simultaneous open
171 * sES -> sIV Invalid SYN/ACK packets sent by the client
178 /* sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sS2 */
179 /*fin*/ { sIV, sIV, sFW, sFW, sLA, sLA, sLA, sTW, sCL, sIV },
181 * sNO -> sIV Too late and no reason to do anything...
182 * sSS -> sIV Client migth not send FIN in this state:
183 * we enforce waiting for a SYN/ACK reply first.
185 * sSR -> sFW Close started.
187 * sFW -> sLA FIN seen in both directions, waiting for
189 * Migth be a retransmitted FIN as well...
191 * sLA -> sLA Retransmitted FIN. Remain in the same state.
195 /* sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sS2 */
196 /*ack*/ { sES, sIV, sES, sES, sCW, sCW, sTW, sTW, sCL, sIV },
198 * sNO -> sES Assumed.
199 * sSS -> sIV ACK is invalid: we haven't seen a SYN/ACK yet.
201 * sSR -> sES Established state is reached.
203 * sFW -> sCW Normal close request answered by ACK.
205 * sLA -> sTW Last ACK detected.
206 * sTW -> sTW Retransmitted last ACK. Remain in the same state.
209 /* sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sS2 */
210 /*rst*/ { sIV, sCL, sCL, sCL, sCL, sCL, sCL, sCL, sCL, sCL },
211 /*none*/ { sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV }
215 /* sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sS2 */
216 /*syn*/ { sIV, sS2, sIV, sIV, sIV, sIV, sIV, sIV, sIV, sS2 },
218 * sNO -> sIV Never reached.
219 * sSS -> sS2 Simultaneous open
220 * sS2 -> sS2 Retransmitted simultaneous SYN
221 * sSR -> sIV Invalid SYN packets sent by the server
226 * sTW -> sIV Reopened connection, but server may not do it.
229 /* sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sS2 */
230 /*synack*/ { sIV, sSR, sIG, sIG, sIG, sIG, sIG, sIG, sIG, sSR },
232 * sSS -> sSR Standard open.
233 * sS2 -> sSR Simultaneous open
234 * sSR -> sIG Retransmitted SYN/ACK, ignore it.
235 * sES -> sIG Late retransmitted SYN/ACK?
236 * sFW -> sIG Might be SYN/ACK answering ignored SYN
242 /* sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sS2 */
243 /*fin*/ { sIV, sIV, sFW, sFW, sLA, sLA, sLA, sTW, sCL, sIV },
245 * sSS -> sIV Server might not send FIN in this state.
247 * sSR -> sFW Close started.
249 * sFW -> sLA FIN seen in both directions.
251 * sLA -> sLA Retransmitted FIN.
255 /* sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sS2 */
256 /*ack*/ { sIV, sIG, sSR, sES, sCW, sCW, sTW, sTW, sCL, sIG },
258 * sSS -> sIG Might be a half-open connection.
260 * sSR -> sSR Might answer late resent SYN.
262 * sFW -> sCW Normal close request answered by ACK.
264 * sLA -> sTW Last ACK detected.
265 * sTW -> sTW Retransmitted last ACK.
268 /* sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sS2 */
269 /*rst*/ { sIV, sCL, sCL, sCL, sCL, sCL, sCL, sCL, sCL, sCL },
270 /*none*/ { sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV }
274 static inline struct nf_tcp_net *tcp_pernet(struct net *net)
276 return &net->ct.nf_ct_proto.tcp;
279 static bool tcp_pkt_to_tuple(const struct sk_buff *skb, unsigned int dataoff,
280 struct nf_conntrack_tuple *tuple)
282 const struct tcphdr *hp;
285 /* Actually only need first 8 bytes. */
286 hp = skb_header_pointer(skb, dataoff, 8, &_hdr);
290 tuple->src.u.tcp.port = hp->source;
291 tuple->dst.u.tcp.port = hp->dest;
296 static bool tcp_invert_tuple(struct nf_conntrack_tuple *tuple,
297 const struct nf_conntrack_tuple *orig)
299 tuple->src.u.tcp.port = orig->dst.u.tcp.port;
300 tuple->dst.u.tcp.port = orig->src.u.tcp.port;
304 /* Print out the per-protocol part of the tuple. */
305 static int tcp_print_tuple(struct seq_file *s,
306 const struct nf_conntrack_tuple *tuple)
308 return seq_printf(s, "sport=%hu dport=%hu ",
309 ntohs(tuple->src.u.tcp.port),
310 ntohs(tuple->dst.u.tcp.port));
313 /* Print out the private part of the conntrack. */
314 static int tcp_print_conntrack(struct seq_file *s, struct nf_conn *ct)
316 enum tcp_conntrack state;
318 spin_lock_bh(&ct->lock);
319 state = ct->proto.tcp.state;
320 spin_unlock_bh(&ct->lock);
322 return seq_printf(s, "%s ", tcp_conntrack_names[state]);
325 static unsigned int get_conntrack_index(const struct tcphdr *tcph)
327 if (tcph->rst) return TCP_RST_SET;
328 else if (tcph->syn) return (tcph->ack ? TCP_SYNACK_SET : TCP_SYN_SET);
329 else if (tcph->fin) return TCP_FIN_SET;
330 else if (tcph->ack) return TCP_ACK_SET;
331 else return TCP_NONE_SET;
334 /* TCP connection tracking based on 'Real Stateful TCP Packet Filtering
335 in IP Filter' by Guido van Rooij.
337 http://www.sane.nl/events/sane2000/papers.html
338 http://www.darkart.com/mirrors/www.obfuscation.org/ipf/
340 The boundaries and the conditions are changed according to RFC793:
341 the packet must intersect the window (i.e. segments may be
342 after the right or before the left edge) and thus receivers may ACK
343 segments after the right edge of the window.
345 td_maxend = max(sack + max(win,1)) seen in reply packets
346 td_maxwin = max(max(win, 1)) + (sack - ack) seen in sent packets
347 td_maxwin += seq + len - sender.td_maxend
348 if seq + len > sender.td_maxend
349 td_end = max(seq + len) seen in sent packets
351 I. Upper bound for valid data: seq <= sender.td_maxend
352 II. Lower bound for valid data: seq + len >= sender.td_end - receiver.td_maxwin
353 III. Upper bound for valid (s)ack: sack <= receiver.td_end
354 IV. Lower bound for valid (s)ack: sack >= receiver.td_end - MAXACKWINDOW
356 where sack is the highest right edge of sack block found in the packet
357 or ack in the case of packet without SACK option.
359 The upper bound limit for a valid (s)ack is not ignored -
360 we doesn't have to deal with fragments.
363 static inline __u32 segment_seq_plus_len(__u32 seq,
365 unsigned int dataoff,
366 const struct tcphdr *tcph)
368 /* XXX Should I use payload length field in IP/IPv6 header ?
370 return (seq + len - dataoff - tcph->doff*4
371 + (tcph->syn ? 1 : 0) + (tcph->fin ? 1 : 0));
374 /* Fixme: what about big packets? */
375 #define MAXACKWINCONST 66000
376 #define MAXACKWINDOW(sender) \
377 ((sender)->td_maxwin > MAXACKWINCONST ? (sender)->td_maxwin \
381 * Simplified tcp_parse_options routine from tcp_input.c
383 static void tcp_options(const struct sk_buff *skb,
384 unsigned int dataoff,
385 const struct tcphdr *tcph,
386 struct ip_ct_tcp_state *state)
388 unsigned char buff[(15 * 4) - sizeof(struct tcphdr)];
389 const unsigned char *ptr;
390 int length = (tcph->doff*4) - sizeof(struct tcphdr);
395 ptr = skb_header_pointer(skb, dataoff + sizeof(struct tcphdr),
409 case TCPOPT_NOP: /* Ref: RFC 793 section 3.1 */
414 if (opsize < 2) /* "silly options" */
417 return; /* don't parse partial options */
419 if (opcode == TCPOPT_SACK_PERM
420 && opsize == TCPOLEN_SACK_PERM)
421 state->flags |= IP_CT_TCP_FLAG_SACK_PERM;
422 else if (opcode == TCPOPT_WINDOW
423 && opsize == TCPOLEN_WINDOW) {
424 state->td_scale = *(u_int8_t *)ptr;
426 if (state->td_scale > 14) {
428 state->td_scale = 14;
431 IP_CT_TCP_FLAG_WINDOW_SCALE;
439 static void tcp_sack(const struct sk_buff *skb, unsigned int dataoff,
440 const struct tcphdr *tcph, __u32 *sack)
442 unsigned char buff[(15 * 4) - sizeof(struct tcphdr)];
443 const unsigned char *ptr;
444 int length = (tcph->doff*4) - sizeof(struct tcphdr);
450 ptr = skb_header_pointer(skb, dataoff + sizeof(struct tcphdr),
454 /* Fast path for timestamp-only option */
455 if (length == TCPOLEN_TSTAMP_ALIGNED
456 && *(__be32 *)ptr == htonl((TCPOPT_NOP << 24)
458 | (TCPOPT_TIMESTAMP << 8)
459 | TCPOLEN_TIMESTAMP))
469 case TCPOPT_NOP: /* Ref: RFC 793 section 3.1 */
474 if (opsize < 2) /* "silly options" */
477 return; /* don't parse partial options */
479 if (opcode == TCPOPT_SACK
480 && opsize >= (TCPOLEN_SACK_BASE
481 + TCPOLEN_SACK_PERBLOCK)
482 && !((opsize - TCPOLEN_SACK_BASE)
483 % TCPOLEN_SACK_PERBLOCK)) {
485 i < (opsize - TCPOLEN_SACK_BASE);
486 i += TCPOLEN_SACK_PERBLOCK) {
487 tmp = get_unaligned_be32((__be32 *)(ptr+i)+1);
489 if (after(tmp, *sack))
500 static bool tcp_in_window(const struct nf_conn *ct,
501 struct ip_ct_tcp *state,
502 enum ip_conntrack_dir dir,
504 const struct sk_buff *skb,
505 unsigned int dataoff,
506 const struct tcphdr *tcph,
509 struct net *net = nf_ct_net(ct);
510 struct nf_tcp_net *tn = tcp_pernet(net);
511 struct ip_ct_tcp_state *sender = &state->seen[dir];
512 struct ip_ct_tcp_state *receiver = &state->seen[!dir];
513 const struct nf_conntrack_tuple *tuple = &ct->tuplehash[dir].tuple;
514 __u32 seq, ack, sack, end, win, swin;
516 bool res, in_recv_win;
519 * Get the required data from the packet.
521 seq = ntohl(tcph->seq);
522 ack = sack = ntohl(tcph->ack_seq);
523 win = ntohs(tcph->window);
524 end = segment_seq_plus_len(seq, skb->len, dataoff, tcph);
526 if (receiver->flags & IP_CT_TCP_FLAG_SACK_PERM)
527 tcp_sack(skb, dataoff, tcph, &sack);
529 /* Take into account NAT sequence number mangling */
530 receiver_offset = nf_ct_seq_offset(ct, !dir, ack - 1);
531 ack -= receiver_offset;
532 sack -= receiver_offset;
534 pr_debug("tcp_in_window: START\n");
535 pr_debug("tcp_in_window: ");
536 nf_ct_dump_tuple(tuple);
537 pr_debug("seq=%u ack=%u+(%d) sack=%u+(%d) win=%u end=%u\n",
538 seq, ack, receiver_offset, sack, receiver_offset, win, end);
539 pr_debug("tcp_in_window: sender end=%u maxend=%u maxwin=%u scale=%i "
540 "receiver end=%u maxend=%u maxwin=%u scale=%i\n",
541 sender->td_end, sender->td_maxend, sender->td_maxwin,
543 receiver->td_end, receiver->td_maxend, receiver->td_maxwin,
546 if (sender->td_maxwin == 0) {
548 * Initialize sender data.
552 * SYN-ACK in reply to a SYN
553 * or SYN from reply direction in simultaneous open.
556 sender->td_maxend = end;
557 sender->td_maxwin = (win == 0 ? 1 : win);
559 tcp_options(skb, dataoff, tcph, sender);
562 * Both sides must send the Window Scale option
563 * to enable window scaling in either direction.
565 if (!(sender->flags & IP_CT_TCP_FLAG_WINDOW_SCALE
566 && receiver->flags & IP_CT_TCP_FLAG_WINDOW_SCALE))
568 receiver->td_scale = 0;
570 /* Simultaneous open */
574 * We are in the middle of a connection,
575 * its history is lost for us.
576 * Let's try to use the data from the packet.
578 sender->td_end = end;
579 swin = win << sender->td_scale;
580 sender->td_maxwin = (swin == 0 ? 1 : swin);
581 sender->td_maxend = end + sender->td_maxwin;
583 * We haven't seen traffic in the other direction yet
584 * but we have to tweak window tracking to pass III
585 * and IV until that happens.
587 if (receiver->td_maxwin == 0)
588 receiver->td_end = receiver->td_maxend = sack;
590 } else if (((state->state == TCP_CONNTRACK_SYN_SENT
591 && dir == IP_CT_DIR_ORIGINAL)
592 || (state->state == TCP_CONNTRACK_SYN_RECV
593 && dir == IP_CT_DIR_REPLY))
594 && after(end, sender->td_end)) {
596 * RFC 793: "if a TCP is reinitialized ... then it need
597 * not wait at all; it must only be sure to use sequence
598 * numbers larger than those recently used."
601 sender->td_maxend = end;
602 sender->td_maxwin = (win == 0 ? 1 : win);
604 tcp_options(skb, dataoff, tcph, sender);
609 * If there is no ACK, just pretend it was set and OK.
611 ack = sack = receiver->td_end;
612 } else if (((tcp_flag_word(tcph) & (TCP_FLAG_ACK|TCP_FLAG_RST)) ==
613 (TCP_FLAG_ACK|TCP_FLAG_RST))
616 * Broken TCP stacks, that set ACK in RST packets as well
617 * with zero ack value.
619 ack = sack = receiver->td_end;
622 if (tcph->rst && seq == 0 && state->state == TCP_CONNTRACK_SYN_SENT)
624 * RST sent answering SYN.
626 seq = end = sender->td_end;
628 pr_debug("tcp_in_window: ");
629 nf_ct_dump_tuple(tuple);
630 pr_debug("seq=%u ack=%u+(%d) sack=%u+(%d) win=%u end=%u\n",
631 seq, ack, receiver_offset, sack, receiver_offset, win, end);
632 pr_debug("tcp_in_window: sender end=%u maxend=%u maxwin=%u scale=%i "
633 "receiver end=%u maxend=%u maxwin=%u scale=%i\n",
634 sender->td_end, sender->td_maxend, sender->td_maxwin,
636 receiver->td_end, receiver->td_maxend, receiver->td_maxwin,
639 /* Is the ending sequence in the receive window (if available)? */
640 in_recv_win = !receiver->td_maxwin ||
641 after(end, sender->td_end - receiver->td_maxwin - 1);
643 pr_debug("tcp_in_window: I=%i II=%i III=%i IV=%i\n",
644 before(seq, sender->td_maxend + 1),
645 (in_recv_win ? 1 : 0),
646 before(sack, receiver->td_end + 1),
647 after(sack, receiver->td_end - MAXACKWINDOW(sender) - 1));
649 if (before(seq, sender->td_maxend + 1) &&
651 before(sack, receiver->td_end + 1) &&
652 after(sack, receiver->td_end - MAXACKWINDOW(sender) - 1)) {
654 * Take into account window scaling (RFC 1323).
657 win <<= sender->td_scale;
660 * Update sender data.
662 swin = win + (sack - ack);
663 if (sender->td_maxwin < swin)
664 sender->td_maxwin = swin;
665 if (after(end, sender->td_end)) {
666 sender->td_end = end;
667 sender->flags |= IP_CT_TCP_FLAG_DATA_UNACKNOWLEDGED;
670 if (!(sender->flags & IP_CT_TCP_FLAG_MAXACK_SET)) {
671 sender->td_maxack = ack;
672 sender->flags |= IP_CT_TCP_FLAG_MAXACK_SET;
673 } else if (after(ack, sender->td_maxack))
674 sender->td_maxack = ack;
678 * Update receiver data.
680 if (receiver->td_maxwin != 0 && after(end, sender->td_maxend))
681 receiver->td_maxwin += end - sender->td_maxend;
682 if (after(sack + win, receiver->td_maxend - 1)) {
683 receiver->td_maxend = sack + win;
685 receiver->td_maxend++;
687 if (ack == receiver->td_end)
688 receiver->flags &= ~IP_CT_TCP_FLAG_DATA_UNACKNOWLEDGED;
691 * Check retransmissions.
693 if (index == TCP_ACK_SET) {
694 if (state->last_dir == dir
695 && state->last_seq == seq
696 && state->last_ack == ack
697 && state->last_end == end
698 && state->last_win == win)
701 state->last_dir = dir;
702 state->last_seq = seq;
703 state->last_ack = ack;
704 state->last_end = end;
705 state->last_win = win;
712 if (sender->flags & IP_CT_TCP_FLAG_BE_LIBERAL ||
715 if (!res && LOG_INVALID(net, IPPROTO_TCP))
716 nf_log_packet(net, pf, 0, skb, NULL, NULL, NULL,
718 before(seq, sender->td_maxend + 1) ?
720 before(sack, receiver->td_end + 1) ?
721 after(sack, receiver->td_end - MAXACKWINDOW(sender) - 1) ? "BUG"
722 : "ACK is under the lower bound (possible overly delayed ACK)"
723 : "ACK is over the upper bound (ACKed data not seen yet)"
724 : "SEQ is under the lower bound (already ACKed data retransmitted)"
725 : "SEQ is over the upper bound (over the window of the receiver)");
728 pr_debug("tcp_in_window: res=%u sender end=%u maxend=%u maxwin=%u "
729 "receiver end=%u maxend=%u maxwin=%u\n",
730 res, sender->td_end, sender->td_maxend, sender->td_maxwin,
731 receiver->td_end, receiver->td_maxend, receiver->td_maxwin);
736 /* table of valid flag combinations - PUSH, ECE and CWR are always valid */
737 static const u8 tcp_valid_flags[(TCPHDR_FIN|TCPHDR_SYN|TCPHDR_RST|TCPHDR_ACK|
741 [TCPHDR_SYN|TCPHDR_URG] = 1,
742 [TCPHDR_SYN|TCPHDR_ACK] = 1,
744 [TCPHDR_RST|TCPHDR_ACK] = 1,
745 [TCPHDR_FIN|TCPHDR_ACK] = 1,
746 [TCPHDR_FIN|TCPHDR_ACK|TCPHDR_URG] = 1,
748 [TCPHDR_ACK|TCPHDR_URG] = 1,
751 /* Protect conntrack agaist broken packets. Code taken from ipt_unclean.c. */
752 static int tcp_error(struct net *net, struct nf_conn *tmpl,
754 unsigned int dataoff,
755 enum ip_conntrack_info *ctinfo,
757 unsigned int hooknum)
759 const struct tcphdr *th;
761 unsigned int tcplen = skb->len - dataoff;
764 /* Smaller that minimal TCP header? */
765 th = skb_header_pointer(skb, dataoff, sizeof(_tcph), &_tcph);
767 if (LOG_INVALID(net, IPPROTO_TCP))
768 nf_log_packet(net, pf, 0, skb, NULL, NULL, NULL,
769 "nf_ct_tcp: short packet ");
773 /* Not whole TCP header or malformed packet */
774 if (th->doff*4 < sizeof(struct tcphdr) || tcplen < th->doff*4) {
775 if (LOG_INVALID(net, IPPROTO_TCP))
776 nf_log_packet(net, pf, 0, skb, NULL, NULL, NULL,
777 "nf_ct_tcp: truncated/malformed packet ");
781 /* Checksum invalid? Ignore.
782 * We skip checking packets on the outgoing path
783 * because the checksum is assumed to be correct.
785 /* FIXME: Source route IP option packets --RR */
786 if (net->ct.sysctl_checksum && hooknum == NF_INET_PRE_ROUTING &&
787 nf_checksum(skb, hooknum, dataoff, IPPROTO_TCP, pf)) {
788 if (LOG_INVALID(net, IPPROTO_TCP))
789 nf_log_packet(net, pf, 0, skb, NULL, NULL, NULL,
790 "nf_ct_tcp: bad TCP checksum ");
794 /* Check TCP flags. */
795 tcpflags = (tcp_flag_byte(th) & ~(TCPHDR_ECE|TCPHDR_CWR|TCPHDR_PSH));
796 if (!tcp_valid_flags[tcpflags]) {
797 if (LOG_INVALID(net, IPPROTO_TCP))
798 nf_log_packet(net, pf, 0, skb, NULL, NULL, NULL,
799 "nf_ct_tcp: invalid TCP flag combination ");
806 static unsigned int *tcp_get_timeouts(struct net *net)
808 return tcp_pernet(net)->timeouts;
811 /* Returns verdict for packet, or -1 for invalid. */
812 static int tcp_packet(struct nf_conn *ct,
813 const struct sk_buff *skb,
814 unsigned int dataoff,
815 enum ip_conntrack_info ctinfo,
817 unsigned int hooknum,
818 unsigned int *timeouts)
820 struct net *net = nf_ct_net(ct);
821 struct nf_tcp_net *tn = tcp_pernet(net);
822 struct nf_conntrack_tuple *tuple;
823 enum tcp_conntrack new_state, old_state;
824 enum ip_conntrack_dir dir;
825 const struct tcphdr *th;
827 unsigned long timeout;
830 th = skb_header_pointer(skb, dataoff, sizeof(_tcph), &_tcph);
833 spin_lock_bh(&ct->lock);
834 old_state = ct->proto.tcp.state;
835 dir = CTINFO2DIR(ctinfo);
836 index = get_conntrack_index(th);
837 new_state = tcp_conntracks[dir][index][old_state];
838 tuple = &ct->tuplehash[dir].tuple;
841 case TCP_CONNTRACK_SYN_SENT:
842 if (old_state < TCP_CONNTRACK_TIME_WAIT)
844 /* RFC 1122: "When a connection is closed actively,
845 * it MUST linger in TIME-WAIT state for a time 2xMSL
846 * (Maximum Segment Lifetime). However, it MAY accept
847 * a new SYN from the remote TCP to reopen the connection
848 * directly from TIME-WAIT state, if..."
849 * We ignore the conditions because we are in the
850 * TIME-WAIT state anyway.
852 * Handle aborted connections: we and the server
853 * think there is an existing connection but the client
854 * aborts it and starts a new one.
856 if (((ct->proto.tcp.seen[dir].flags
857 | ct->proto.tcp.seen[!dir].flags)
858 & IP_CT_TCP_FLAG_CLOSE_INIT)
859 || (ct->proto.tcp.last_dir == dir
860 && ct->proto.tcp.last_index == TCP_RST_SET)) {
861 /* Attempt to reopen a closed/aborted connection.
862 * Delete this connection and look up again. */
863 spin_unlock_bh(&ct->lock);
865 /* Only repeat if we can actually remove the timer.
866 * Destruction may already be in progress in process
867 * context and we must give it a chance to terminate.
874 case TCP_CONNTRACK_IGNORE:
877 * Our connection entry may be out of sync, so ignore
878 * packets which may signal the real connection between
879 * the client and the server.
882 * b) SYN/ACK in REPLY
883 * c) ACK in reply direction after initial SYN in original.
885 * If the ignored packet is invalid, the receiver will send
886 * a RST we'll catch below.
888 if (index == TCP_SYNACK_SET
889 && ct->proto.tcp.last_index == TCP_SYN_SET
890 && ct->proto.tcp.last_dir != dir
891 && ntohl(th->ack_seq) == ct->proto.tcp.last_end) {
892 /* b) This SYN/ACK acknowledges a SYN that we earlier
893 * ignored as invalid. This means that the client and
894 * the server are both in sync, while the firewall is
895 * not. We get in sync from the previously annotated
898 old_state = TCP_CONNTRACK_SYN_SENT;
899 new_state = TCP_CONNTRACK_SYN_RECV;
900 ct->proto.tcp.seen[ct->proto.tcp.last_dir].td_end =
901 ct->proto.tcp.last_end;
902 ct->proto.tcp.seen[ct->proto.tcp.last_dir].td_maxend =
903 ct->proto.tcp.last_end;
904 ct->proto.tcp.seen[ct->proto.tcp.last_dir].td_maxwin =
905 ct->proto.tcp.last_win == 0 ?
906 1 : ct->proto.tcp.last_win;
907 ct->proto.tcp.seen[ct->proto.tcp.last_dir].td_scale =
908 ct->proto.tcp.last_wscale;
909 ct->proto.tcp.seen[ct->proto.tcp.last_dir].flags =
910 ct->proto.tcp.last_flags;
911 memset(&ct->proto.tcp.seen[dir], 0,
912 sizeof(struct ip_ct_tcp_state));
915 ct->proto.tcp.last_index = index;
916 ct->proto.tcp.last_dir = dir;
917 ct->proto.tcp.last_seq = ntohl(th->seq);
918 ct->proto.tcp.last_end =
919 segment_seq_plus_len(ntohl(th->seq), skb->len, dataoff, th);
920 ct->proto.tcp.last_win = ntohs(th->window);
922 /* a) This is a SYN in ORIGINAL. The client and the server
923 * may be in sync but we are not. In that case, we annotate
924 * the TCP options and let the packet go through. If it is a
925 * valid SYN packet, the server will reply with a SYN/ACK, and
926 * then we'll get in sync. Otherwise, the server ignores it. */
927 if (index == TCP_SYN_SET && dir == IP_CT_DIR_ORIGINAL) {
928 struct ip_ct_tcp_state seen = {};
930 ct->proto.tcp.last_flags =
931 ct->proto.tcp.last_wscale = 0;
932 tcp_options(skb, dataoff, th, &seen);
933 if (seen.flags & IP_CT_TCP_FLAG_WINDOW_SCALE) {
934 ct->proto.tcp.last_flags |=
935 IP_CT_TCP_FLAG_WINDOW_SCALE;
936 ct->proto.tcp.last_wscale = seen.td_scale;
938 if (seen.flags & IP_CT_TCP_FLAG_SACK_PERM) {
939 ct->proto.tcp.last_flags |=
940 IP_CT_TCP_FLAG_SACK_PERM;
943 spin_unlock_bh(&ct->lock);
944 if (LOG_INVALID(net, IPPROTO_TCP))
945 nf_log_packet(net, pf, 0, skb, NULL, NULL, NULL,
946 "nf_ct_tcp: invalid packet ignored in "
947 "state %s ", tcp_conntrack_names[old_state]);
949 case TCP_CONNTRACK_MAX:
950 /* Special case for SYN proxy: when the SYN to the server or
951 * the SYN/ACK from the server is lost, the client may transmit
952 * a keep-alive packet while in SYN_SENT state. This needs to
953 * be associated with the original conntrack entry in order to
954 * generate a new SYN with the correct sequence number.
956 if (nfct_synproxy(ct) && old_state == TCP_CONNTRACK_SYN_SENT &&
957 index == TCP_ACK_SET && dir == IP_CT_DIR_ORIGINAL &&
958 ct->proto.tcp.last_dir == IP_CT_DIR_ORIGINAL &&
959 ct->proto.tcp.seen[dir].td_end - 1 == ntohl(th->seq)) {
960 pr_debug("nf_ct_tcp: SYN proxy client keep alive\n");
961 spin_unlock_bh(&ct->lock);
966 pr_debug("nf_ct_tcp: Invalid dir=%i index=%u ostate=%u\n",
967 dir, get_conntrack_index(th), old_state);
968 spin_unlock_bh(&ct->lock);
969 if (LOG_INVALID(net, IPPROTO_TCP))
970 nf_log_packet(net, pf, 0, skb, NULL, NULL, NULL,
971 "nf_ct_tcp: invalid state ");
973 case TCP_CONNTRACK_CLOSE:
974 if (index == TCP_RST_SET
975 && (ct->proto.tcp.seen[!dir].flags & IP_CT_TCP_FLAG_MAXACK_SET)
976 && before(ntohl(th->seq), ct->proto.tcp.seen[!dir].td_maxack)) {
978 spin_unlock_bh(&ct->lock);
979 if (LOG_INVALID(net, IPPROTO_TCP))
980 nf_log_packet(net, pf, 0, skb, NULL, NULL,
981 NULL, "nf_ct_tcp: invalid RST ");
984 if (index == TCP_RST_SET
985 && ((test_bit(IPS_SEEN_REPLY_BIT, &ct->status)
986 && ct->proto.tcp.last_index == TCP_SYN_SET)
987 || (!test_bit(IPS_ASSURED_BIT, &ct->status)
988 && ct->proto.tcp.last_index == TCP_ACK_SET))
989 && ntohl(th->ack_seq) == ct->proto.tcp.last_end) {
990 /* RST sent to invalid SYN or ACK we had let through
991 * at a) and c) above:
993 * a) SYN was in window then
994 * c) we hold a half-open connection.
996 * Delete our connection entry.
997 * We skip window checking, because packet might ACK
998 * segments we ignored. */
1001 /* Just fall through */
1003 /* Keep compilers happy. */
1007 if (!tcp_in_window(ct, &ct->proto.tcp, dir, index,
1008 skb, dataoff, th, pf)) {
1009 spin_unlock_bh(&ct->lock);
1013 /* From now on we have got in-window packets */
1014 ct->proto.tcp.last_index = index;
1015 ct->proto.tcp.last_dir = dir;
1017 pr_debug("tcp_conntracks: ");
1018 nf_ct_dump_tuple(tuple);
1019 pr_debug("syn=%i ack=%i fin=%i rst=%i old=%i new=%i\n",
1020 (th->syn ? 1 : 0), (th->ack ? 1 : 0),
1021 (th->fin ? 1 : 0), (th->rst ? 1 : 0),
1022 old_state, new_state);
1024 ct->proto.tcp.state = new_state;
1025 if (old_state != new_state
1026 && new_state == TCP_CONNTRACK_FIN_WAIT)
1027 ct->proto.tcp.seen[dir].flags |= IP_CT_TCP_FLAG_CLOSE_INIT;
1029 if (ct->proto.tcp.retrans >= tn->tcp_max_retrans &&
1030 timeouts[new_state] > timeouts[TCP_CONNTRACK_RETRANS])
1031 timeout = timeouts[TCP_CONNTRACK_RETRANS];
1032 else if ((ct->proto.tcp.seen[0].flags | ct->proto.tcp.seen[1].flags) &
1033 IP_CT_TCP_FLAG_DATA_UNACKNOWLEDGED &&
1034 timeouts[new_state] > timeouts[TCP_CONNTRACK_UNACK])
1035 timeout = timeouts[TCP_CONNTRACK_UNACK];
1037 timeout = timeouts[new_state];
1038 spin_unlock_bh(&ct->lock);
1040 if (new_state != old_state)
1041 nf_conntrack_event_cache(IPCT_PROTOINFO, ct);
1043 if (!test_bit(IPS_SEEN_REPLY_BIT, &ct->status)) {
1044 /* If only reply is a RST, we can consider ourselves not to
1045 have an established connection: this is a fairly common
1046 problem case, so we can delete the conntrack
1047 immediately. --RR */
1049 nf_ct_kill_acct(ct, ctinfo, skb);
1052 /* ESTABLISHED without SEEN_REPLY, i.e. mid-connection
1053 * pickup with loose=1. Avoid large ESTABLISHED timeout.
1055 if (new_state == TCP_CONNTRACK_ESTABLISHED &&
1056 timeout > timeouts[TCP_CONNTRACK_UNACK])
1057 timeout = timeouts[TCP_CONNTRACK_UNACK];
1058 } else if (!test_bit(IPS_ASSURED_BIT, &ct->status)
1059 && (old_state == TCP_CONNTRACK_SYN_RECV
1060 || old_state == TCP_CONNTRACK_ESTABLISHED)
1061 && new_state == TCP_CONNTRACK_ESTABLISHED) {
1062 /* Set ASSURED if we see see valid ack in ESTABLISHED
1063 after SYN_RECV or a valid answer for a picked up
1065 set_bit(IPS_ASSURED_BIT, &ct->status);
1066 nf_conntrack_event_cache(IPCT_ASSURED, ct);
1068 nf_ct_refresh_acct(ct, ctinfo, skb, timeout);
1073 /* Called when a new connection for this protocol found. */
1074 static bool tcp_new(struct nf_conn *ct, const struct sk_buff *skb,
1075 unsigned int dataoff, unsigned int *timeouts)
1077 enum tcp_conntrack new_state;
1078 const struct tcphdr *th;
1079 struct tcphdr _tcph;
1080 struct net *net = nf_ct_net(ct);
1081 struct nf_tcp_net *tn = tcp_pernet(net);
1082 const struct ip_ct_tcp_state *sender = &ct->proto.tcp.seen[0];
1083 const struct ip_ct_tcp_state *receiver = &ct->proto.tcp.seen[1];
1085 th = skb_header_pointer(skb, dataoff, sizeof(_tcph), &_tcph);
1088 /* Don't need lock here: this conntrack not in circulation yet */
1089 new_state = tcp_conntracks[0][get_conntrack_index(th)][TCP_CONNTRACK_NONE];
1091 /* Invalid: delete conntrack */
1092 if (new_state >= TCP_CONNTRACK_MAX) {
1093 pr_debug("nf_ct_tcp: invalid new deleting.\n");
1097 if (new_state == TCP_CONNTRACK_SYN_SENT) {
1098 memset(&ct->proto.tcp, 0, sizeof(ct->proto.tcp));
1100 ct->proto.tcp.seen[0].td_end =
1101 segment_seq_plus_len(ntohl(th->seq), skb->len,
1103 ct->proto.tcp.seen[0].td_maxwin = ntohs(th->window);
1104 if (ct->proto.tcp.seen[0].td_maxwin == 0)
1105 ct->proto.tcp.seen[0].td_maxwin = 1;
1106 ct->proto.tcp.seen[0].td_maxend =
1107 ct->proto.tcp.seen[0].td_end;
1109 tcp_options(skb, dataoff, th, &ct->proto.tcp.seen[0]);
1110 } else if (tn->tcp_loose == 0) {
1111 /* Don't try to pick up connections. */
1114 memset(&ct->proto.tcp, 0, sizeof(ct->proto.tcp));
1116 * We are in the middle of a connection,
1117 * its history is lost for us.
1118 * Let's try to use the data from the packet.
1120 ct->proto.tcp.seen[0].td_end =
1121 segment_seq_plus_len(ntohl(th->seq), skb->len,
1123 ct->proto.tcp.seen[0].td_maxwin = ntohs(th->window);
1124 if (ct->proto.tcp.seen[0].td_maxwin == 0)
1125 ct->proto.tcp.seen[0].td_maxwin = 1;
1126 ct->proto.tcp.seen[0].td_maxend =
1127 ct->proto.tcp.seen[0].td_end +
1128 ct->proto.tcp.seen[0].td_maxwin;
1130 /* We assume SACK and liberal window checking to handle
1132 ct->proto.tcp.seen[0].flags =
1133 ct->proto.tcp.seen[1].flags = IP_CT_TCP_FLAG_SACK_PERM |
1134 IP_CT_TCP_FLAG_BE_LIBERAL;
1137 /* tcp_packet will set them */
1138 ct->proto.tcp.last_index = TCP_NONE_SET;
1140 pr_debug("tcp_new: sender end=%u maxend=%u maxwin=%u scale=%i "
1141 "receiver end=%u maxend=%u maxwin=%u scale=%i\n",
1142 sender->td_end, sender->td_maxend, sender->td_maxwin,
1144 receiver->td_end, receiver->td_maxend, receiver->td_maxwin,
1145 receiver->td_scale);
1149 #if IS_ENABLED(CONFIG_NF_CT_NETLINK)
1151 #include <linux/netfilter/nfnetlink.h>
1152 #include <linux/netfilter/nfnetlink_conntrack.h>
1154 static int tcp_to_nlattr(struct sk_buff *skb, struct nlattr *nla,
1157 struct nlattr *nest_parms;
1158 struct nf_ct_tcp_flags tmp = {};
1160 spin_lock_bh(&ct->lock);
1161 nest_parms = nla_nest_start(skb, CTA_PROTOINFO_TCP | NLA_F_NESTED);
1163 goto nla_put_failure;
1165 if (nla_put_u8(skb, CTA_PROTOINFO_TCP_STATE, ct->proto.tcp.state) ||
1166 nla_put_u8(skb, CTA_PROTOINFO_TCP_WSCALE_ORIGINAL,
1167 ct->proto.tcp.seen[0].td_scale) ||
1168 nla_put_u8(skb, CTA_PROTOINFO_TCP_WSCALE_REPLY,
1169 ct->proto.tcp.seen[1].td_scale))
1170 goto nla_put_failure;
1172 tmp.flags = ct->proto.tcp.seen[0].flags;
1173 if (nla_put(skb, CTA_PROTOINFO_TCP_FLAGS_ORIGINAL,
1174 sizeof(struct nf_ct_tcp_flags), &tmp))
1175 goto nla_put_failure;
1177 tmp.flags = ct->proto.tcp.seen[1].flags;
1178 if (nla_put(skb, CTA_PROTOINFO_TCP_FLAGS_REPLY,
1179 sizeof(struct nf_ct_tcp_flags), &tmp))
1180 goto nla_put_failure;
1181 spin_unlock_bh(&ct->lock);
1183 nla_nest_end(skb, nest_parms);
1188 spin_unlock_bh(&ct->lock);
1192 static const struct nla_policy tcp_nla_policy[CTA_PROTOINFO_TCP_MAX+1] = {
1193 [CTA_PROTOINFO_TCP_STATE] = { .type = NLA_U8 },
1194 [CTA_PROTOINFO_TCP_WSCALE_ORIGINAL] = { .type = NLA_U8 },
1195 [CTA_PROTOINFO_TCP_WSCALE_REPLY] = { .type = NLA_U8 },
1196 [CTA_PROTOINFO_TCP_FLAGS_ORIGINAL] = { .len = sizeof(struct nf_ct_tcp_flags) },
1197 [CTA_PROTOINFO_TCP_FLAGS_REPLY] = { .len = sizeof(struct nf_ct_tcp_flags) },
1200 static int nlattr_to_tcp(struct nlattr *cda[], struct nf_conn *ct)
1202 struct nlattr *pattr = cda[CTA_PROTOINFO_TCP];
1203 struct nlattr *tb[CTA_PROTOINFO_TCP_MAX+1];
1206 /* updates could not contain anything about the private
1207 * protocol info, in that case skip the parsing */
1211 err = nla_parse_nested(tb, CTA_PROTOINFO_TCP_MAX, pattr, tcp_nla_policy);
1215 if (tb[CTA_PROTOINFO_TCP_STATE] &&
1216 nla_get_u8(tb[CTA_PROTOINFO_TCP_STATE]) >= TCP_CONNTRACK_MAX)
1219 spin_lock_bh(&ct->lock);
1220 if (tb[CTA_PROTOINFO_TCP_STATE])
1221 ct->proto.tcp.state = nla_get_u8(tb[CTA_PROTOINFO_TCP_STATE]);
1223 if (tb[CTA_PROTOINFO_TCP_FLAGS_ORIGINAL]) {
1224 struct nf_ct_tcp_flags *attr =
1225 nla_data(tb[CTA_PROTOINFO_TCP_FLAGS_ORIGINAL]);
1226 ct->proto.tcp.seen[0].flags &= ~attr->mask;
1227 ct->proto.tcp.seen[0].flags |= attr->flags & attr->mask;
1230 if (tb[CTA_PROTOINFO_TCP_FLAGS_REPLY]) {
1231 struct nf_ct_tcp_flags *attr =
1232 nla_data(tb[CTA_PROTOINFO_TCP_FLAGS_REPLY]);
1233 ct->proto.tcp.seen[1].flags &= ~attr->mask;
1234 ct->proto.tcp.seen[1].flags |= attr->flags & attr->mask;
1237 if (tb[CTA_PROTOINFO_TCP_WSCALE_ORIGINAL] &&
1238 tb[CTA_PROTOINFO_TCP_WSCALE_REPLY] &&
1239 ct->proto.tcp.seen[0].flags & IP_CT_TCP_FLAG_WINDOW_SCALE &&
1240 ct->proto.tcp.seen[1].flags & IP_CT_TCP_FLAG_WINDOW_SCALE) {
1241 ct->proto.tcp.seen[0].td_scale =
1242 nla_get_u8(tb[CTA_PROTOINFO_TCP_WSCALE_ORIGINAL]);
1243 ct->proto.tcp.seen[1].td_scale =
1244 nla_get_u8(tb[CTA_PROTOINFO_TCP_WSCALE_REPLY]);
1246 spin_unlock_bh(&ct->lock);
1251 static int tcp_nlattr_size(void)
1253 return nla_total_size(0) /* CTA_PROTOINFO_TCP */
1254 + nla_policy_len(tcp_nla_policy, CTA_PROTOINFO_TCP_MAX + 1);
1257 static int tcp_nlattr_tuple_size(void)
1259 return nla_policy_len(nf_ct_port_nla_policy, CTA_PROTO_MAX + 1);
1263 #if IS_ENABLED(CONFIG_NF_CT_NETLINK_TIMEOUT)
1265 #include <linux/netfilter/nfnetlink.h>
1266 #include <linux/netfilter/nfnetlink_cttimeout.h>
1268 static int tcp_timeout_nlattr_to_obj(struct nlattr *tb[],
1269 struct net *net, void *data)
1271 unsigned int *timeouts = data;
1272 struct nf_tcp_net *tn = tcp_pernet(net);
1275 /* set default TCP timeouts. */
1276 for (i=0; i<TCP_CONNTRACK_TIMEOUT_MAX; i++)
1277 timeouts[i] = tn->timeouts[i];
1279 if (tb[CTA_TIMEOUT_TCP_SYN_SENT]) {
1280 timeouts[TCP_CONNTRACK_SYN_SENT] =
1281 ntohl(nla_get_be32(tb[CTA_TIMEOUT_TCP_SYN_SENT]))*HZ;
1283 if (tb[CTA_TIMEOUT_TCP_SYN_RECV]) {
1284 timeouts[TCP_CONNTRACK_SYN_RECV] =
1285 ntohl(nla_get_be32(tb[CTA_TIMEOUT_TCP_SYN_RECV]))*HZ;
1287 if (tb[CTA_TIMEOUT_TCP_ESTABLISHED]) {
1288 timeouts[TCP_CONNTRACK_ESTABLISHED] =
1289 ntohl(nla_get_be32(tb[CTA_TIMEOUT_TCP_ESTABLISHED]))*HZ;
1291 if (tb[CTA_TIMEOUT_TCP_FIN_WAIT]) {
1292 timeouts[TCP_CONNTRACK_FIN_WAIT] =
1293 ntohl(nla_get_be32(tb[CTA_TIMEOUT_TCP_FIN_WAIT]))*HZ;
1295 if (tb[CTA_TIMEOUT_TCP_CLOSE_WAIT]) {
1296 timeouts[TCP_CONNTRACK_CLOSE_WAIT] =
1297 ntohl(nla_get_be32(tb[CTA_TIMEOUT_TCP_CLOSE_WAIT]))*HZ;
1299 if (tb[CTA_TIMEOUT_TCP_LAST_ACK]) {
1300 timeouts[TCP_CONNTRACK_LAST_ACK] =
1301 ntohl(nla_get_be32(tb[CTA_TIMEOUT_TCP_LAST_ACK]))*HZ;
1303 if (tb[CTA_TIMEOUT_TCP_TIME_WAIT]) {
1304 timeouts[TCP_CONNTRACK_TIME_WAIT] =
1305 ntohl(nla_get_be32(tb[CTA_TIMEOUT_TCP_TIME_WAIT]))*HZ;
1307 if (tb[CTA_TIMEOUT_TCP_CLOSE]) {
1308 timeouts[TCP_CONNTRACK_CLOSE] =
1309 ntohl(nla_get_be32(tb[CTA_TIMEOUT_TCP_CLOSE]))*HZ;
1311 if (tb[CTA_TIMEOUT_TCP_SYN_SENT2]) {
1312 timeouts[TCP_CONNTRACK_SYN_SENT2] =
1313 ntohl(nla_get_be32(tb[CTA_TIMEOUT_TCP_SYN_SENT2]))*HZ;
1315 if (tb[CTA_TIMEOUT_TCP_RETRANS]) {
1316 timeouts[TCP_CONNTRACK_RETRANS] =
1317 ntohl(nla_get_be32(tb[CTA_TIMEOUT_TCP_RETRANS]))*HZ;
1319 if (tb[CTA_TIMEOUT_TCP_UNACK]) {
1320 timeouts[TCP_CONNTRACK_UNACK] =
1321 ntohl(nla_get_be32(tb[CTA_TIMEOUT_TCP_UNACK]))*HZ;
1327 tcp_timeout_obj_to_nlattr(struct sk_buff *skb, const void *data)
1329 const unsigned int *timeouts = data;
1331 if (nla_put_be32(skb, CTA_TIMEOUT_TCP_SYN_SENT,
1332 htonl(timeouts[TCP_CONNTRACK_SYN_SENT] / HZ)) ||
1333 nla_put_be32(skb, CTA_TIMEOUT_TCP_SYN_RECV,
1334 htonl(timeouts[TCP_CONNTRACK_SYN_RECV] / HZ)) ||
1335 nla_put_be32(skb, CTA_TIMEOUT_TCP_ESTABLISHED,
1336 htonl(timeouts[TCP_CONNTRACK_ESTABLISHED] / HZ)) ||
1337 nla_put_be32(skb, CTA_TIMEOUT_TCP_FIN_WAIT,
1338 htonl(timeouts[TCP_CONNTRACK_FIN_WAIT] / HZ)) ||
1339 nla_put_be32(skb, CTA_TIMEOUT_TCP_CLOSE_WAIT,
1340 htonl(timeouts[TCP_CONNTRACK_CLOSE_WAIT] / HZ)) ||
1341 nla_put_be32(skb, CTA_TIMEOUT_TCP_LAST_ACK,
1342 htonl(timeouts[TCP_CONNTRACK_LAST_ACK] / HZ)) ||
1343 nla_put_be32(skb, CTA_TIMEOUT_TCP_TIME_WAIT,
1344 htonl(timeouts[TCP_CONNTRACK_TIME_WAIT] / HZ)) ||
1345 nla_put_be32(skb, CTA_TIMEOUT_TCP_CLOSE,
1346 htonl(timeouts[TCP_CONNTRACK_CLOSE] / HZ)) ||
1347 nla_put_be32(skb, CTA_TIMEOUT_TCP_SYN_SENT2,
1348 htonl(timeouts[TCP_CONNTRACK_SYN_SENT2] / HZ)) ||
1349 nla_put_be32(skb, CTA_TIMEOUT_TCP_RETRANS,
1350 htonl(timeouts[TCP_CONNTRACK_RETRANS] / HZ)) ||
1351 nla_put_be32(skb, CTA_TIMEOUT_TCP_UNACK,
1352 htonl(timeouts[TCP_CONNTRACK_UNACK] / HZ)))
1353 goto nla_put_failure;
1360 static const struct nla_policy tcp_timeout_nla_policy[CTA_TIMEOUT_TCP_MAX+1] = {
1361 [CTA_TIMEOUT_TCP_SYN_SENT] = { .type = NLA_U32 },
1362 [CTA_TIMEOUT_TCP_SYN_RECV] = { .type = NLA_U32 },
1363 [CTA_TIMEOUT_TCP_ESTABLISHED] = { .type = NLA_U32 },
1364 [CTA_TIMEOUT_TCP_FIN_WAIT] = { .type = NLA_U32 },
1365 [CTA_TIMEOUT_TCP_CLOSE_WAIT] = { .type = NLA_U32 },
1366 [CTA_TIMEOUT_TCP_LAST_ACK] = { .type = NLA_U32 },
1367 [CTA_TIMEOUT_TCP_TIME_WAIT] = { .type = NLA_U32 },
1368 [CTA_TIMEOUT_TCP_CLOSE] = { .type = NLA_U32 },
1369 [CTA_TIMEOUT_TCP_SYN_SENT2] = { .type = NLA_U32 },
1370 [CTA_TIMEOUT_TCP_RETRANS] = { .type = NLA_U32 },
1371 [CTA_TIMEOUT_TCP_UNACK] = { .type = NLA_U32 },
1373 #endif /* CONFIG_NF_CT_NETLINK_TIMEOUT */
1375 #ifdef CONFIG_SYSCTL
1376 static struct ctl_table tcp_sysctl_table[] = {
1378 .procname = "nf_conntrack_tcp_timeout_syn_sent",
1379 .maxlen = sizeof(unsigned int),
1381 .proc_handler = proc_dointvec_jiffies,
1384 .procname = "nf_conntrack_tcp_timeout_syn_recv",
1385 .maxlen = sizeof(unsigned int),
1387 .proc_handler = proc_dointvec_jiffies,
1390 .procname = "nf_conntrack_tcp_timeout_established",
1391 .maxlen = sizeof(unsigned int),
1393 .proc_handler = proc_dointvec_jiffies,
1396 .procname = "nf_conntrack_tcp_timeout_fin_wait",
1397 .maxlen = sizeof(unsigned int),
1399 .proc_handler = proc_dointvec_jiffies,
1402 .procname = "nf_conntrack_tcp_timeout_close_wait",
1403 .maxlen = sizeof(unsigned int),
1405 .proc_handler = proc_dointvec_jiffies,
1408 .procname = "nf_conntrack_tcp_timeout_last_ack",
1409 .maxlen = sizeof(unsigned int),
1411 .proc_handler = proc_dointvec_jiffies,
1414 .procname = "nf_conntrack_tcp_timeout_time_wait",
1415 .maxlen = sizeof(unsigned int),
1417 .proc_handler = proc_dointvec_jiffies,
1420 .procname = "nf_conntrack_tcp_timeout_close",
1421 .maxlen = sizeof(unsigned int),
1423 .proc_handler = proc_dointvec_jiffies,
1426 .procname = "nf_conntrack_tcp_timeout_max_retrans",
1427 .maxlen = sizeof(unsigned int),
1429 .proc_handler = proc_dointvec_jiffies,
1432 .procname = "nf_conntrack_tcp_timeout_unacknowledged",
1433 .maxlen = sizeof(unsigned int),
1435 .proc_handler = proc_dointvec_jiffies,
1438 .procname = "nf_conntrack_tcp_loose",
1439 .maxlen = sizeof(unsigned int),
1441 .proc_handler = proc_dointvec,
1444 .procname = "nf_conntrack_tcp_be_liberal",
1445 .maxlen = sizeof(unsigned int),
1447 .proc_handler = proc_dointvec,
1450 .procname = "nf_conntrack_tcp_max_retrans",
1451 .maxlen = sizeof(unsigned int),
1453 .proc_handler = proc_dointvec,
1458 #ifdef CONFIG_NF_CONNTRACK_PROC_COMPAT
1459 static struct ctl_table tcp_compat_sysctl_table[] = {
1461 .procname = "ip_conntrack_tcp_timeout_syn_sent",
1462 .maxlen = sizeof(unsigned int),
1464 .proc_handler = proc_dointvec_jiffies,
1467 .procname = "ip_conntrack_tcp_timeout_syn_sent2",
1468 .maxlen = sizeof(unsigned int),
1470 .proc_handler = proc_dointvec_jiffies,
1473 .procname = "ip_conntrack_tcp_timeout_syn_recv",
1474 .maxlen = sizeof(unsigned int),
1476 .proc_handler = proc_dointvec_jiffies,
1479 .procname = "ip_conntrack_tcp_timeout_established",
1480 .maxlen = sizeof(unsigned int),
1482 .proc_handler = proc_dointvec_jiffies,
1485 .procname = "ip_conntrack_tcp_timeout_fin_wait",
1486 .maxlen = sizeof(unsigned int),
1488 .proc_handler = proc_dointvec_jiffies,
1491 .procname = "ip_conntrack_tcp_timeout_close_wait",
1492 .maxlen = sizeof(unsigned int),
1494 .proc_handler = proc_dointvec_jiffies,
1497 .procname = "ip_conntrack_tcp_timeout_last_ack",
1498 .maxlen = sizeof(unsigned int),
1500 .proc_handler = proc_dointvec_jiffies,
1503 .procname = "ip_conntrack_tcp_timeout_time_wait",
1504 .maxlen = sizeof(unsigned int),
1506 .proc_handler = proc_dointvec_jiffies,
1509 .procname = "ip_conntrack_tcp_timeout_close",
1510 .maxlen = sizeof(unsigned int),
1512 .proc_handler = proc_dointvec_jiffies,
1515 .procname = "ip_conntrack_tcp_timeout_max_retrans",
1516 .maxlen = sizeof(unsigned int),
1518 .proc_handler = proc_dointvec_jiffies,
1521 .procname = "ip_conntrack_tcp_loose",
1522 .maxlen = sizeof(unsigned int),
1524 .proc_handler = proc_dointvec,
1527 .procname = "ip_conntrack_tcp_be_liberal",
1528 .maxlen = sizeof(unsigned int),
1530 .proc_handler = proc_dointvec,
1533 .procname = "ip_conntrack_tcp_max_retrans",
1534 .maxlen = sizeof(unsigned int),
1536 .proc_handler = proc_dointvec,
1540 #endif /* CONFIG_NF_CONNTRACK_PROC_COMPAT */
1541 #endif /* CONFIG_SYSCTL */
1543 static int tcp_kmemdup_sysctl_table(struct nf_proto_net *pn,
1544 struct nf_tcp_net *tn)
1546 #ifdef CONFIG_SYSCTL
1550 pn->ctl_table = kmemdup(tcp_sysctl_table,
1551 sizeof(tcp_sysctl_table),
1556 pn->ctl_table[0].data = &tn->timeouts[TCP_CONNTRACK_SYN_SENT];
1557 pn->ctl_table[1].data = &tn->timeouts[TCP_CONNTRACK_SYN_RECV];
1558 pn->ctl_table[2].data = &tn->timeouts[TCP_CONNTRACK_ESTABLISHED];
1559 pn->ctl_table[3].data = &tn->timeouts[TCP_CONNTRACK_FIN_WAIT];
1560 pn->ctl_table[4].data = &tn->timeouts[TCP_CONNTRACK_CLOSE_WAIT];
1561 pn->ctl_table[5].data = &tn->timeouts[TCP_CONNTRACK_LAST_ACK];
1562 pn->ctl_table[6].data = &tn->timeouts[TCP_CONNTRACK_TIME_WAIT];
1563 pn->ctl_table[7].data = &tn->timeouts[TCP_CONNTRACK_CLOSE];
1564 pn->ctl_table[8].data = &tn->timeouts[TCP_CONNTRACK_RETRANS];
1565 pn->ctl_table[9].data = &tn->timeouts[TCP_CONNTRACK_UNACK];
1566 pn->ctl_table[10].data = &tn->tcp_loose;
1567 pn->ctl_table[11].data = &tn->tcp_be_liberal;
1568 pn->ctl_table[12].data = &tn->tcp_max_retrans;
1573 static int tcp_kmemdup_compat_sysctl_table(struct nf_proto_net *pn,
1574 struct nf_tcp_net *tn)
1576 #ifdef CONFIG_SYSCTL
1577 #ifdef CONFIG_NF_CONNTRACK_PROC_COMPAT
1578 pn->ctl_compat_table = kmemdup(tcp_compat_sysctl_table,
1579 sizeof(tcp_compat_sysctl_table),
1581 if (!pn->ctl_compat_table)
1584 pn->ctl_compat_table[0].data = &tn->timeouts[TCP_CONNTRACK_SYN_SENT];
1585 pn->ctl_compat_table[1].data = &tn->timeouts[TCP_CONNTRACK_SYN_SENT2];
1586 pn->ctl_compat_table[2].data = &tn->timeouts[TCP_CONNTRACK_SYN_RECV];
1587 pn->ctl_compat_table[3].data = &tn->timeouts[TCP_CONNTRACK_ESTABLISHED];
1588 pn->ctl_compat_table[4].data = &tn->timeouts[TCP_CONNTRACK_FIN_WAIT];
1589 pn->ctl_compat_table[5].data = &tn->timeouts[TCP_CONNTRACK_CLOSE_WAIT];
1590 pn->ctl_compat_table[6].data = &tn->timeouts[TCP_CONNTRACK_LAST_ACK];
1591 pn->ctl_compat_table[7].data = &tn->timeouts[TCP_CONNTRACK_TIME_WAIT];
1592 pn->ctl_compat_table[8].data = &tn->timeouts[TCP_CONNTRACK_CLOSE];
1593 pn->ctl_compat_table[9].data = &tn->timeouts[TCP_CONNTRACK_RETRANS];
1594 pn->ctl_compat_table[10].data = &tn->tcp_loose;
1595 pn->ctl_compat_table[11].data = &tn->tcp_be_liberal;
1596 pn->ctl_compat_table[12].data = &tn->tcp_max_retrans;
1602 static int tcp_init_net(struct net *net, u_int16_t proto)
1605 struct nf_tcp_net *tn = tcp_pernet(net);
1606 struct nf_proto_net *pn = &tn->pn;
1611 for (i = 0; i < TCP_CONNTRACK_TIMEOUT_MAX; i++)
1612 tn->timeouts[i] = tcp_timeouts[i];
1614 tn->tcp_loose = nf_ct_tcp_loose;
1615 tn->tcp_be_liberal = nf_ct_tcp_be_liberal;
1616 tn->tcp_max_retrans = nf_ct_tcp_max_retrans;
1619 if (proto == AF_INET) {
1620 ret = tcp_kmemdup_compat_sysctl_table(pn, tn);
1624 ret = tcp_kmemdup_sysctl_table(pn, tn);
1626 nf_ct_kfree_compat_sysctl_table(pn);
1628 ret = tcp_kmemdup_sysctl_table(pn, tn);
1633 static struct nf_proto_net *tcp_get_net_proto(struct net *net)
1635 return &net->ct.nf_ct_proto.tcp.pn;
1638 struct nf_conntrack_l4proto nf_conntrack_l4proto_tcp4 __read_mostly =
1641 .l4proto = IPPROTO_TCP,
1643 .pkt_to_tuple = tcp_pkt_to_tuple,
1644 .invert_tuple = tcp_invert_tuple,
1645 .print_tuple = tcp_print_tuple,
1646 .print_conntrack = tcp_print_conntrack,
1647 .packet = tcp_packet,
1648 .get_timeouts = tcp_get_timeouts,
1651 #if IS_ENABLED(CONFIG_NF_CT_NETLINK)
1652 .to_nlattr = tcp_to_nlattr,
1653 .nlattr_size = tcp_nlattr_size,
1654 .from_nlattr = nlattr_to_tcp,
1655 .tuple_to_nlattr = nf_ct_port_tuple_to_nlattr,
1656 .nlattr_to_tuple = nf_ct_port_nlattr_to_tuple,
1657 .nlattr_tuple_size = tcp_nlattr_tuple_size,
1658 .nla_policy = nf_ct_port_nla_policy,
1660 #if IS_ENABLED(CONFIG_NF_CT_NETLINK_TIMEOUT)
1662 .nlattr_to_obj = tcp_timeout_nlattr_to_obj,
1663 .obj_to_nlattr = tcp_timeout_obj_to_nlattr,
1664 .nlattr_max = CTA_TIMEOUT_TCP_MAX,
1665 .obj_size = sizeof(unsigned int) *
1666 TCP_CONNTRACK_TIMEOUT_MAX,
1667 .nla_policy = tcp_timeout_nla_policy,
1669 #endif /* CONFIG_NF_CT_NETLINK_TIMEOUT */
1670 .init_net = tcp_init_net,
1671 .get_net_proto = tcp_get_net_proto,
1673 EXPORT_SYMBOL_GPL(nf_conntrack_l4proto_tcp4);
1675 struct nf_conntrack_l4proto nf_conntrack_l4proto_tcp6 __read_mostly =
1677 .l3proto = PF_INET6,
1678 .l4proto = IPPROTO_TCP,
1680 .pkt_to_tuple = tcp_pkt_to_tuple,
1681 .invert_tuple = tcp_invert_tuple,
1682 .print_tuple = tcp_print_tuple,
1683 .print_conntrack = tcp_print_conntrack,
1684 .packet = tcp_packet,
1685 .get_timeouts = tcp_get_timeouts,
1688 #if IS_ENABLED(CONFIG_NF_CT_NETLINK)
1689 .to_nlattr = tcp_to_nlattr,
1690 .nlattr_size = tcp_nlattr_size,
1691 .from_nlattr = nlattr_to_tcp,
1692 .tuple_to_nlattr = nf_ct_port_tuple_to_nlattr,
1693 .nlattr_to_tuple = nf_ct_port_nlattr_to_tuple,
1694 .nlattr_tuple_size = tcp_nlattr_tuple_size,
1695 .nla_policy = nf_ct_port_nla_policy,
1697 #if IS_ENABLED(CONFIG_NF_CT_NETLINK_TIMEOUT)
1699 .nlattr_to_obj = tcp_timeout_nlattr_to_obj,
1700 .obj_to_nlattr = tcp_timeout_obj_to_nlattr,
1701 .nlattr_max = CTA_TIMEOUT_TCP_MAX,
1702 .obj_size = sizeof(unsigned int) *
1703 TCP_CONNTRACK_TIMEOUT_MAX,
1704 .nla_policy = tcp_timeout_nla_policy,
1706 #endif /* CONFIG_NF_CT_NETLINK_TIMEOUT */
1707 .init_net = tcp_init_net,
1708 .get_net_proto = tcp_get_net_proto,
1710 EXPORT_SYMBOL_GPL(nf_conntrack_l4proto_tcp6);