1 // SPDX-License-Identifier: GPL-2.0-only
2 /* (C) 1999-2001 Paul `Rusty' Russell
3 * (C) 2002-2004 Netfilter Core Team <coreteam@netfilter.org>
4 * (C) 2002-2013 Jozsef Kadlecsik <kadlec@netfilter.org>
5 * (C) 2006-2012 Patrick McHardy <kaber@trash.net>
8 #include <linux/types.h>
9 #include <linux/timer.h>
10 #include <linux/module.h>
12 #include <linux/tcp.h>
13 #include <linux/spinlock.h>
14 #include <linux/skbuff.h>
15 #include <linux/ipv6.h>
16 #include <net/ip6_checksum.h>
17 #include <asm/unaligned.h>
21 #include <linux/netfilter.h>
22 #include <linux/netfilter_ipv4.h>
23 #include <linux/netfilter_ipv6.h>
24 #include <net/netfilter/nf_conntrack.h>
25 #include <net/netfilter/nf_conntrack_l4proto.h>
26 #include <net/netfilter/nf_conntrack_ecache.h>
27 #include <net/netfilter/nf_conntrack_seqadj.h>
28 #include <net/netfilter/nf_conntrack_synproxy.h>
29 #include <net/netfilter/nf_conntrack_timeout.h>
30 #include <net/netfilter/nf_log.h>
31 #include <net/netfilter/ipv4/nf_conntrack_ipv4.h>
32 #include <net/netfilter/ipv6/nf_conntrack_ipv6.h>
34 /* FIXME: Examine ipfilter's timeouts and conntrack transitions more
35 closely. They're more complex. --RR */
37 static const char *const tcp_conntrack_names[] = {
50 enum nf_ct_tcp_action {
57 #define MINS * 60 SECS
58 #define HOURS * 60 MINS
59 #define DAYS * 24 HOURS
61 static const unsigned int tcp_timeouts[TCP_CONNTRACK_TIMEOUT_MAX] = {
62 [TCP_CONNTRACK_SYN_SENT] = 2 MINS,
63 [TCP_CONNTRACK_SYN_RECV] = 60 SECS,
64 [TCP_CONNTRACK_ESTABLISHED] = 5 DAYS,
65 [TCP_CONNTRACK_FIN_WAIT] = 2 MINS,
66 [TCP_CONNTRACK_CLOSE_WAIT] = 60 SECS,
67 [TCP_CONNTRACK_LAST_ACK] = 30 SECS,
68 [TCP_CONNTRACK_TIME_WAIT] = 2 MINS,
69 [TCP_CONNTRACK_CLOSE] = 10 SECS,
70 [TCP_CONNTRACK_SYN_SENT2] = 2 MINS,
71 /* RFC1122 says the R2 limit should be at least 100 seconds.
72 Linux uses 15 packets as limit, which corresponds
73 to ~13-30min depending on RTO. */
74 [TCP_CONNTRACK_RETRANS] = 5 MINS,
75 [TCP_CONNTRACK_UNACK] = 5 MINS,
78 #define sNO TCP_CONNTRACK_NONE
79 #define sSS TCP_CONNTRACK_SYN_SENT
80 #define sSR TCP_CONNTRACK_SYN_RECV
81 #define sES TCP_CONNTRACK_ESTABLISHED
82 #define sFW TCP_CONNTRACK_FIN_WAIT
83 #define sCW TCP_CONNTRACK_CLOSE_WAIT
84 #define sLA TCP_CONNTRACK_LAST_ACK
85 #define sTW TCP_CONNTRACK_TIME_WAIT
86 #define sCL TCP_CONNTRACK_CLOSE
87 #define sS2 TCP_CONNTRACK_SYN_SENT2
88 #define sIV TCP_CONNTRACK_MAX
89 #define sIG TCP_CONNTRACK_IGNORE
91 /* What TCP flags are set from RST/SYN/FIN/ACK. */
102 * The TCP state transition table needs a few words...
104 * We are the man in the middle. All the packets go through us
105 * but might get lost in transit to the destination.
106 * It is assumed that the destinations can't receive segments
109 * The checked segment is in window, but our windows are *not*
110 * equivalent with the ones of the sender/receiver. We always
111 * try to guess the state of the current sender.
113 * The meaning of the states are:
115 * NONE: initial state
116 * SYN_SENT: SYN-only packet seen
117 * SYN_SENT2: SYN-only packet seen from reply dir, simultaneous open
118 * SYN_RECV: SYN-ACK packet seen
119 * ESTABLISHED: ACK packet seen
120 * FIN_WAIT: FIN packet seen
121 * CLOSE_WAIT: ACK seen (after FIN)
122 * LAST_ACK: FIN seen (after FIN)
123 * TIME_WAIT: last ACK seen
124 * CLOSE: closed connection (RST)
126 * Packets marked as IGNORED (sIG):
127 * if they may be either invalid or valid
128 * and the receiver may send back a connection
129 * closing RST or a SYN/ACK.
131 * Packets marked as INVALID (sIV):
132 * if we regard them as truly invalid packets
134 static const u8 tcp_conntracks[2][6][TCP_CONNTRACK_MAX] = {
137 /* sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sS2 */
138 /*syn*/ { sSS, sSS, sIG, sIG, sIG, sIG, sIG, sSS, sSS, sS2 },
140 * sNO -> sSS Initialize a new connection
141 * sSS -> sSS Retransmitted SYN
142 * sS2 -> sS2 Late retransmitted SYN
144 * sES -> sIG Error: SYNs in window outside the SYN_SENT state
145 * are errors. Receiver will reply with RST
146 * and close the connection.
147 * Or we are not in sync and hold a dead connection.
151 * sTW -> sSS Reopened connection (RFC 1122).
154 /* sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sS2 */
155 /*synack*/ { sIV, sIV, sSR, sIV, sIV, sIV, sIV, sIV, sIV, sSR },
157 * sNO -> sIV Too late and no reason to do anything
158 * sSS -> sIV Client can't send SYN and then SYN/ACK
159 * sS2 -> sSR SYN/ACK sent to SYN2 in simultaneous open
160 * sSR -> sSR Late retransmitted SYN/ACK in simultaneous open
161 * sES -> sIV Invalid SYN/ACK packets sent by the client
168 /* sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sS2 */
169 /*fin*/ { sIV, sIV, sFW, sFW, sLA, sLA, sLA, sTW, sCL, sIV },
171 * sNO -> sIV Too late and no reason to do anything...
172 * sSS -> sIV Client migth not send FIN in this state:
173 * we enforce waiting for a SYN/ACK reply first.
175 * sSR -> sFW Close started.
177 * sFW -> sLA FIN seen in both directions, waiting for
179 * Migth be a retransmitted FIN as well...
181 * sLA -> sLA Retransmitted FIN. Remain in the same state.
185 /* sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sS2 */
186 /*ack*/ { sES, sIV, sES, sES, sCW, sCW, sTW, sTW, sCL, sIV },
188 * sNO -> sES Assumed.
189 * sSS -> sIV ACK is invalid: we haven't seen a SYN/ACK yet.
191 * sSR -> sES Established state is reached.
193 * sFW -> sCW Normal close request answered by ACK.
195 * sLA -> sTW Last ACK detected (RFC5961 challenged)
196 * sTW -> sTW Retransmitted last ACK. Remain in the same state.
199 /* sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sS2 */
200 /*rst*/ { sIV, sCL, sCL, sCL, sCL, sCL, sCL, sCL, sCL, sCL },
201 /*none*/ { sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV }
205 /* sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sS2 */
206 /*syn*/ { sIV, sS2, sIV, sIV, sIV, sIV, sIV, sSS, sIV, sS2 },
208 * sNO -> sIV Never reached.
209 * sSS -> sS2 Simultaneous open
210 * sS2 -> sS2 Retransmitted simultaneous SYN
211 * sSR -> sIV Invalid SYN packets sent by the server
216 * sTW -> sSS Reopened connection, but server may have switched role
219 /* sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sS2 */
220 /*synack*/ { sIV, sSR, sIG, sIG, sIG, sIG, sIG, sIG, sIG, sSR },
222 * sSS -> sSR Standard open.
223 * sS2 -> sSR Simultaneous open
224 * sSR -> sIG Retransmitted SYN/ACK, ignore it.
225 * sES -> sIG Late retransmitted SYN/ACK?
226 * sFW -> sIG Might be SYN/ACK answering ignored SYN
232 /* sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sS2 */
233 /*fin*/ { sIV, sIV, sFW, sFW, sLA, sLA, sLA, sTW, sCL, sIV },
235 * sSS -> sIV Server might not send FIN in this state.
237 * sSR -> sFW Close started.
239 * sFW -> sLA FIN seen in both directions.
241 * sLA -> sLA Retransmitted FIN.
245 /* sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sS2 */
246 /*ack*/ { sIV, sIG, sSR, sES, sCW, sCW, sTW, sTW, sCL, sIG },
248 * sSS -> sIG Might be a half-open connection.
250 * sSR -> sSR Might answer late resent SYN.
252 * sFW -> sCW Normal close request answered by ACK.
254 * sLA -> sTW Last ACK detected (RFC5961 challenged)
255 * sTW -> sTW Retransmitted last ACK.
258 /* sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sS2 */
259 /*rst*/ { sIV, sCL, sCL, sCL, sCL, sCL, sCL, sCL, sCL, sCL },
260 /*none*/ { sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV }
264 #ifdef CONFIG_NF_CONNTRACK_PROCFS
265 /* Print out the private part of the conntrack. */
266 static void tcp_print_conntrack(struct seq_file *s, struct nf_conn *ct)
268 if (test_bit(IPS_OFFLOAD_BIT, &ct->status))
271 seq_printf(s, "%s ", tcp_conntrack_names[ct->proto.tcp.state]);
275 static unsigned int get_conntrack_index(const struct tcphdr *tcph)
277 if (tcph->rst) return TCP_RST_SET;
278 else if (tcph->syn) return (tcph->ack ? TCP_SYNACK_SET : TCP_SYN_SET);
279 else if (tcph->fin) return TCP_FIN_SET;
280 else if (tcph->ack) return TCP_ACK_SET;
281 else return TCP_NONE_SET;
284 /* TCP connection tracking based on 'Real Stateful TCP Packet Filtering
285 in IP Filter' by Guido van Rooij.
287 http://www.sane.nl/events/sane2000/papers.html
288 http://www.darkart.com/mirrors/www.obfuscation.org/ipf/
290 The boundaries and the conditions are changed according to RFC793:
291 the packet must intersect the window (i.e. segments may be
292 after the right or before the left edge) and thus receivers may ACK
293 segments after the right edge of the window.
295 td_maxend = max(sack + max(win,1)) seen in reply packets
296 td_maxwin = max(max(win, 1)) + (sack - ack) seen in sent packets
297 td_maxwin += seq + len - sender.td_maxend
298 if seq + len > sender.td_maxend
299 td_end = max(seq + len) seen in sent packets
301 I. Upper bound for valid data: seq <= sender.td_maxend
302 II. Lower bound for valid data: seq + len >= sender.td_end - receiver.td_maxwin
303 III. Upper bound for valid (s)ack: sack <= receiver.td_end
304 IV. Lower bound for valid (s)ack: sack >= receiver.td_end - MAXACKWINDOW
306 where sack is the highest right edge of sack block found in the packet
307 or ack in the case of packet without SACK option.
309 The upper bound limit for a valid (s)ack is not ignored -
310 we doesn't have to deal with fragments.
313 static inline __u32 segment_seq_plus_len(__u32 seq,
315 unsigned int dataoff,
316 const struct tcphdr *tcph)
318 /* XXX Should I use payload length field in IP/IPv6 header ?
320 return (seq + len - dataoff - tcph->doff*4
321 + (tcph->syn ? 1 : 0) + (tcph->fin ? 1 : 0));
324 /* Fixme: what about big packets? */
325 #define MAXACKWINCONST 66000
326 #define MAXACKWINDOW(sender) \
327 ((sender)->td_maxwin > MAXACKWINCONST ? (sender)->td_maxwin \
331 * Simplified tcp_parse_options routine from tcp_input.c
333 static void tcp_options(const struct sk_buff *skb,
334 unsigned int dataoff,
335 const struct tcphdr *tcph,
336 struct ip_ct_tcp_state *state)
338 unsigned char buff[(15 * 4) - sizeof(struct tcphdr)];
339 const unsigned char *ptr;
340 int length = (tcph->doff*4) - sizeof(struct tcphdr);
345 ptr = skb_header_pointer(skb, dataoff + sizeof(struct tcphdr),
351 state->flags &= IP_CT_TCP_FLAG_BE_LIBERAL;
360 case TCPOPT_NOP: /* Ref: RFC 793 section 3.1 */
367 if (opsize < 2) /* "silly options" */
370 return; /* don't parse partial options */
372 if (opcode == TCPOPT_SACK_PERM
373 && opsize == TCPOLEN_SACK_PERM)
374 state->flags |= IP_CT_TCP_FLAG_SACK_PERM;
375 else if (opcode == TCPOPT_WINDOW
376 && opsize == TCPOLEN_WINDOW) {
377 state->td_scale = *(u_int8_t *)ptr;
379 if (state->td_scale > TCP_MAX_WSCALE)
380 state->td_scale = TCP_MAX_WSCALE;
383 IP_CT_TCP_FLAG_WINDOW_SCALE;
391 static void tcp_sack(const struct sk_buff *skb, unsigned int dataoff,
392 const struct tcphdr *tcph, __u32 *sack)
394 unsigned char buff[(15 * 4) - sizeof(struct tcphdr)];
395 const unsigned char *ptr;
396 int length = (tcph->doff*4) - sizeof(struct tcphdr);
402 ptr = skb_header_pointer(skb, dataoff + sizeof(struct tcphdr),
407 /* Fast path for timestamp-only option */
408 if (length == TCPOLEN_TSTAMP_ALIGNED
409 && *(__be32 *)ptr == htonl((TCPOPT_NOP << 24)
411 | (TCPOPT_TIMESTAMP << 8)
412 | TCPOLEN_TIMESTAMP))
422 case TCPOPT_NOP: /* Ref: RFC 793 section 3.1 */
429 if (opsize < 2) /* "silly options" */
432 return; /* don't parse partial options */
434 if (opcode == TCPOPT_SACK
435 && opsize >= (TCPOLEN_SACK_BASE
436 + TCPOLEN_SACK_PERBLOCK)
437 && !((opsize - TCPOLEN_SACK_BASE)
438 % TCPOLEN_SACK_PERBLOCK)) {
440 i < (opsize - TCPOLEN_SACK_BASE);
441 i += TCPOLEN_SACK_PERBLOCK) {
442 tmp = get_unaligned_be32((__be32 *)(ptr+i)+1);
444 if (after(tmp, *sack))
455 static void tcp_init_sender(struct ip_ct_tcp_state *sender,
456 struct ip_ct_tcp_state *receiver,
457 const struct sk_buff *skb,
458 unsigned int dataoff,
459 const struct tcphdr *tcph,
461 enum ip_conntrack_dir dir)
463 /* SYN-ACK in reply to a SYN
464 * or SYN from reply direction in simultaneous open.
467 sender->td_maxend = end;
468 sender->td_maxwin = (win == 0 ? 1 : win);
470 tcp_options(skb, dataoff, tcph, sender);
472 * Both sides must send the Window Scale option
473 * to enable window scaling in either direction.
475 if (dir == IP_CT_DIR_REPLY &&
476 !(sender->flags & IP_CT_TCP_FLAG_WINDOW_SCALE &&
477 receiver->flags & IP_CT_TCP_FLAG_WINDOW_SCALE)) {
478 sender->td_scale = 0;
479 receiver->td_scale = 0;
484 static enum nf_ct_tcp_action nf_tcp_log_invalid(const struct sk_buff *skb,
485 const struct nf_conn *ct,
486 const struct nf_hook_state *state,
487 const struct ip_ct_tcp_state *sender,
488 enum nf_ct_tcp_action ret,
489 const char *fmt, ...)
491 const struct nf_tcp_net *tn = nf_tcp_pernet(nf_ct_net(ct));
492 struct va_format vaf;
496 be_liberal = sender->flags & IP_CT_TCP_FLAG_BE_LIBERAL || tn->tcp_be_liberal;
498 return NFCT_TCP_ACCEPT;
503 nf_ct_l4proto_log_invalid(skb, ct, state, "%pV", &vaf);
509 static enum nf_ct_tcp_action
510 tcp_in_window(struct nf_conn *ct, enum ip_conntrack_dir dir,
511 unsigned int index, const struct sk_buff *skb,
512 unsigned int dataoff, const struct tcphdr *tcph,
513 const struct nf_hook_state *hook_state)
515 struct ip_ct_tcp *state = &ct->proto.tcp;
516 struct ip_ct_tcp_state *sender = &state->seen[dir];
517 struct ip_ct_tcp_state *receiver = &state->seen[!dir];
518 __u32 seq, ack, sack, end, win, swin;
519 bool in_recv_win, seq_ok;
524 * Get the required data from the packet.
526 seq = ntohl(tcph->seq);
527 ack = sack = ntohl(tcph->ack_seq);
528 win_raw = ntohs(tcph->window);
530 end = segment_seq_plus_len(seq, skb->len, dataoff, tcph);
532 if (receiver->flags & IP_CT_TCP_FLAG_SACK_PERM)
533 tcp_sack(skb, dataoff, tcph, &sack);
535 /* Take into account NAT sequence number mangling */
536 receiver_offset = nf_ct_seq_offset(ct, !dir, ack - 1);
537 ack -= receiver_offset;
538 sack -= receiver_offset;
540 if (sender->td_maxwin == 0) {
542 * Initialize sender data.
545 tcp_init_sender(sender, receiver,
549 /* Simultaneous open */
550 return NFCT_TCP_ACCEPT;
553 * We are in the middle of a connection,
554 * its history is lost for us.
555 * Let's try to use the data from the packet.
557 sender->td_end = end;
558 swin = win << sender->td_scale;
559 sender->td_maxwin = (swin == 0 ? 1 : swin);
560 sender->td_maxend = end + sender->td_maxwin;
561 if (receiver->td_maxwin == 0) {
562 /* We haven't seen traffic in the other
563 * direction yet but we have to tweak window
564 * tracking to pass III and IV until that
567 receiver->td_end = receiver->td_maxend = sack;
568 } else if (sack == receiver->td_end + 1) {
569 /* Likely a reply to a keepalive.
576 } else if (tcph->syn &&
577 after(end, sender->td_end) &&
578 (state->state == TCP_CONNTRACK_SYN_SENT ||
579 state->state == TCP_CONNTRACK_SYN_RECV)) {
581 * RFC 793: "if a TCP is reinitialized ... then it need
582 * not wait at all; it must only be sure to use sequence
583 * numbers larger than those recently used."
585 * Re-init state for this direction, just like for the first
586 * syn(-ack) reply, it might differ in seq, ack or tcp options.
588 tcp_init_sender(sender, receiver,
592 if (dir == IP_CT_DIR_REPLY && !tcph->ack)
593 return NFCT_TCP_ACCEPT;
598 * If there is no ACK, just pretend it was set and OK.
600 ack = sack = receiver->td_end;
601 } else if (((tcp_flag_word(tcph) & (TCP_FLAG_ACK|TCP_FLAG_RST)) ==
602 (TCP_FLAG_ACK|TCP_FLAG_RST))
605 * Broken TCP stacks, that set ACK in RST packets as well
606 * with zero ack value.
608 ack = sack = receiver->td_end;
611 if (tcph->rst && seq == 0 && state->state == TCP_CONNTRACK_SYN_SENT)
613 * RST sent answering SYN.
615 seq = end = sender->td_end;
617 seq_ok = before(seq, sender->td_maxend + 1);
619 u32 overshot = end - sender->td_maxend + 1;
622 ack_ok = after(sack, receiver->td_end - MAXACKWINDOW(sender) - 1);
623 in_recv_win = receiver->td_maxwin &&
624 after(end, sender->td_end - receiver->td_maxwin - 1);
628 overshot <= receiver->td_maxwin &&
629 before(sack, receiver->td_end + 1)) {
630 /* Work around TCPs that send more bytes than allowed by
631 * the receive window.
633 * If the (marked as invalid) packet is allowed to pass by
634 * the ruleset and the peer acks this data, then its possible
635 * all future packets will trigger 'ACK is over upper bound' check.
637 * Thus if only the sequence check fails then do update td_end so
638 * possible ACK for this data can update internal state.
640 sender->td_end = end;
641 sender->flags |= IP_CT_TCP_FLAG_DATA_UNACKNOWLEDGED;
643 return nf_tcp_log_invalid(skb, ct, hook_state, sender, NFCT_TCP_IGNORE,
644 "%u bytes more than expected", overshot);
647 return nf_tcp_log_invalid(skb, ct, hook_state, sender, NFCT_TCP_INVALID,
648 "SEQ is over upper bound %u (over the window of the receiver)",
649 sender->td_maxend + 1);
652 if (!before(sack, receiver->td_end + 1))
653 return nf_tcp_log_invalid(skb, ct, hook_state, sender, NFCT_TCP_INVALID,
654 "ACK is over upper bound %u (ACKed data not seen yet)",
655 receiver->td_end + 1);
657 /* Is the ending sequence in the receive window (if available)? */
658 in_recv_win = !receiver->td_maxwin ||
659 after(end, sender->td_end - receiver->td_maxwin - 1);
661 return nf_tcp_log_invalid(skb, ct, hook_state, sender, NFCT_TCP_IGNORE,
662 "SEQ is under lower bound %u (already ACKed data retransmitted)",
663 sender->td_end - receiver->td_maxwin - 1);
664 if (!after(sack, receiver->td_end - MAXACKWINDOW(sender) - 1))
665 return nf_tcp_log_invalid(skb, ct, hook_state, sender, NFCT_TCP_IGNORE,
666 "ignored ACK under lower bound %u (possible overly delayed)",
667 receiver->td_end - MAXACKWINDOW(sender) - 1);
669 /* Take into account window scaling (RFC 1323). */
671 win <<= sender->td_scale;
673 /* Update sender data. */
674 swin = win + (sack - ack);
675 if (sender->td_maxwin < swin)
676 sender->td_maxwin = swin;
677 if (after(end, sender->td_end)) {
678 sender->td_end = end;
679 sender->flags |= IP_CT_TCP_FLAG_DATA_UNACKNOWLEDGED;
682 if (!(sender->flags & IP_CT_TCP_FLAG_MAXACK_SET)) {
683 sender->td_maxack = ack;
684 sender->flags |= IP_CT_TCP_FLAG_MAXACK_SET;
685 } else if (after(ack, sender->td_maxack)) {
686 sender->td_maxack = ack;
690 /* Update receiver data. */
691 if (receiver->td_maxwin != 0 && after(end, sender->td_maxend))
692 receiver->td_maxwin += end - sender->td_maxend;
693 if (after(sack + win, receiver->td_maxend - 1)) {
694 receiver->td_maxend = sack + win;
696 receiver->td_maxend++;
698 if (ack == receiver->td_end)
699 receiver->flags &= ~IP_CT_TCP_FLAG_DATA_UNACKNOWLEDGED;
701 /* Check retransmissions. */
702 if (index == TCP_ACK_SET) {
703 if (state->last_dir == dir &&
704 state->last_seq == seq &&
705 state->last_ack == ack &&
706 state->last_end == end &&
707 state->last_win == win_raw) {
710 state->last_dir = dir;
711 state->last_seq = seq;
712 state->last_ack = ack;
713 state->last_end = end;
714 state->last_win = win_raw;
719 return NFCT_TCP_ACCEPT;
722 static void __cold nf_tcp_handle_invalid(struct nf_conn *ct,
723 enum ip_conntrack_dir dir,
725 const struct sk_buff *skb,
726 const struct nf_hook_state *hook_state)
728 const unsigned int *timeouts;
729 const struct nf_tcp_net *tn;
730 unsigned int timeout;
733 if (!test_bit(IPS_ASSURED_BIT, &ct->status) ||
734 test_bit(IPS_FIXED_TIMEOUT_BIT, &ct->status))
737 /* We don't want to have connections hanging around in ESTABLISHED
738 * state for long time 'just because' conntrack deemed a FIN/RST
741 * Shrink the timeout just like when there is unacked data.
742 * This speeds up eviction of 'dead' connections where the
743 * connection and conntracks internal state are out of sync.
753 if (ct->proto.tcp.last_dir != dir &&
754 (ct->proto.tcp.last_index == TCP_FIN_SET ||
755 ct->proto.tcp.last_index == TCP_RST_SET)) {
756 expires = nf_ct_expires(ct);
757 if (expires < 120 * HZ)
760 tn = nf_tcp_pernet(nf_ct_net(ct));
761 timeouts = nf_ct_timeout_lookup(ct);
763 timeouts = tn->timeouts;
765 timeout = READ_ONCE(timeouts[TCP_CONNTRACK_UNACK]);
766 if (expires > timeout) {
767 nf_ct_l4proto_log_invalid(skb, ct, hook_state,
768 "packet (index %d, dir %d) response for index %d lower timeout to %u",
769 index, dir, ct->proto.tcp.last_index, timeout);
771 WRITE_ONCE(ct->timeout, timeout + nfct_time_stamp);
774 ct->proto.tcp.last_index = index;
775 ct->proto.tcp.last_dir = dir;
779 /* table of valid flag combinations - PUSH, ECE and CWR are always valid */
780 static const u8 tcp_valid_flags[(TCPHDR_FIN|TCPHDR_SYN|TCPHDR_RST|TCPHDR_ACK|
784 [TCPHDR_SYN|TCPHDR_URG] = 1,
785 [TCPHDR_SYN|TCPHDR_ACK] = 1,
787 [TCPHDR_RST|TCPHDR_ACK] = 1,
788 [TCPHDR_FIN|TCPHDR_ACK] = 1,
789 [TCPHDR_FIN|TCPHDR_ACK|TCPHDR_URG] = 1,
791 [TCPHDR_ACK|TCPHDR_URG] = 1,
794 static void tcp_error_log(const struct sk_buff *skb,
795 const struct nf_hook_state *state,
798 nf_l4proto_log_invalid(skb, state, IPPROTO_TCP, "%s", msg);
801 /* Protect conntrack agaist broken packets. Code taken from ipt_unclean.c. */
802 static bool tcp_error(const struct tcphdr *th,
804 unsigned int dataoff,
805 const struct nf_hook_state *state)
807 unsigned int tcplen = skb->len - dataoff;
810 /* Not whole TCP header or malformed packet */
811 if (th->doff*4 < sizeof(struct tcphdr) || tcplen < th->doff*4) {
812 tcp_error_log(skb, state, "truncated packet");
816 /* Checksum invalid? Ignore.
817 * We skip checking packets on the outgoing path
818 * because the checksum is assumed to be correct.
820 /* FIXME: Source route IP option packets --RR */
821 if (state->net->ct.sysctl_checksum &&
822 state->hook == NF_INET_PRE_ROUTING &&
823 nf_checksum(skb, state->hook, dataoff, IPPROTO_TCP, state->pf)) {
824 tcp_error_log(skb, state, "bad checksum");
828 /* Check TCP flags. */
829 tcpflags = (tcp_flag_byte(th) & ~(TCPHDR_ECE|TCPHDR_CWR|TCPHDR_PSH));
830 if (!tcp_valid_flags[tcpflags]) {
831 tcp_error_log(skb, state, "invalid tcp flag combination");
838 static noinline bool tcp_new(struct nf_conn *ct, const struct sk_buff *skb,
839 unsigned int dataoff,
840 const struct tcphdr *th)
842 enum tcp_conntrack new_state;
843 struct net *net = nf_ct_net(ct);
844 const struct nf_tcp_net *tn = nf_tcp_pernet(net);
846 /* Don't need lock here: this conntrack not in circulation yet */
847 new_state = tcp_conntracks[0][get_conntrack_index(th)][TCP_CONNTRACK_NONE];
849 /* Invalid: delete conntrack */
850 if (new_state >= TCP_CONNTRACK_MAX) {
851 pr_debug("nf_ct_tcp: invalid new deleting.\n");
855 if (new_state == TCP_CONNTRACK_SYN_SENT) {
856 memset(&ct->proto.tcp, 0, sizeof(ct->proto.tcp));
858 ct->proto.tcp.seen[0].td_end =
859 segment_seq_plus_len(ntohl(th->seq), skb->len,
861 ct->proto.tcp.seen[0].td_maxwin = ntohs(th->window);
862 if (ct->proto.tcp.seen[0].td_maxwin == 0)
863 ct->proto.tcp.seen[0].td_maxwin = 1;
864 ct->proto.tcp.seen[0].td_maxend =
865 ct->proto.tcp.seen[0].td_end;
867 tcp_options(skb, dataoff, th, &ct->proto.tcp.seen[0]);
868 } else if (tn->tcp_loose == 0) {
869 /* Don't try to pick up connections. */
872 memset(&ct->proto.tcp, 0, sizeof(ct->proto.tcp));
874 * We are in the middle of a connection,
875 * its history is lost for us.
876 * Let's try to use the data from the packet.
878 ct->proto.tcp.seen[0].td_end =
879 segment_seq_plus_len(ntohl(th->seq), skb->len,
881 ct->proto.tcp.seen[0].td_maxwin = ntohs(th->window);
882 if (ct->proto.tcp.seen[0].td_maxwin == 0)
883 ct->proto.tcp.seen[0].td_maxwin = 1;
884 ct->proto.tcp.seen[0].td_maxend =
885 ct->proto.tcp.seen[0].td_end +
886 ct->proto.tcp.seen[0].td_maxwin;
888 /* We assume SACK and liberal window checking to handle
890 ct->proto.tcp.seen[0].flags =
891 ct->proto.tcp.seen[1].flags = IP_CT_TCP_FLAG_SACK_PERM |
892 IP_CT_TCP_FLAG_BE_LIBERAL;
895 /* tcp_packet will set them */
896 ct->proto.tcp.last_index = TCP_NONE_SET;
900 static bool tcp_can_early_drop(const struct nf_conn *ct)
902 switch (ct->proto.tcp.state) {
903 case TCP_CONNTRACK_FIN_WAIT:
904 case TCP_CONNTRACK_LAST_ACK:
905 case TCP_CONNTRACK_TIME_WAIT:
906 case TCP_CONNTRACK_CLOSE:
907 case TCP_CONNTRACK_CLOSE_WAIT:
916 void nf_conntrack_tcp_set_closing(struct nf_conn *ct)
918 enum tcp_conntrack old_state;
919 const unsigned int *timeouts;
922 if (!nf_ct_is_confirmed(ct))
925 spin_lock_bh(&ct->lock);
926 old_state = ct->proto.tcp.state;
927 ct->proto.tcp.state = TCP_CONNTRACK_CLOSE;
929 if (old_state == TCP_CONNTRACK_CLOSE ||
930 test_bit(IPS_FIXED_TIMEOUT_BIT, &ct->status)) {
931 spin_unlock_bh(&ct->lock);
935 timeouts = nf_ct_timeout_lookup(ct);
937 const struct nf_tcp_net *tn;
939 tn = nf_tcp_pernet(nf_ct_net(ct));
940 timeouts = tn->timeouts;
943 timeout = timeouts[TCP_CONNTRACK_CLOSE];
944 WRITE_ONCE(ct->timeout, timeout + nfct_time_stamp);
946 spin_unlock_bh(&ct->lock);
948 nf_conntrack_event_cache(IPCT_PROTOINFO, ct);
951 static void nf_ct_tcp_state_reset(struct ip_ct_tcp_state *state)
954 state->td_maxend = 0;
955 state->td_maxwin = 0;
956 state->td_maxack = 0;
958 state->flags &= IP_CT_TCP_FLAG_BE_LIBERAL;
961 /* Returns verdict for packet, or -1 for invalid. */
962 int nf_conntrack_tcp_packet(struct nf_conn *ct,
964 unsigned int dataoff,
965 enum ip_conntrack_info ctinfo,
966 const struct nf_hook_state *state)
968 struct net *net = nf_ct_net(ct);
969 struct nf_tcp_net *tn = nf_tcp_pernet(net);
970 struct nf_conntrack_tuple *tuple;
971 enum tcp_conntrack new_state, old_state;
972 unsigned int index, *timeouts;
973 enum nf_ct_tcp_action res;
974 enum ip_conntrack_dir dir;
975 const struct tcphdr *th;
977 unsigned long timeout;
979 th = skb_header_pointer(skb, dataoff, sizeof(_tcph), &_tcph);
983 if (tcp_error(th, skb, dataoff, state))
986 if (!nf_ct_is_confirmed(ct) && !tcp_new(ct, skb, dataoff, th))
989 spin_lock_bh(&ct->lock);
990 old_state = ct->proto.tcp.state;
991 dir = CTINFO2DIR(ctinfo);
992 index = get_conntrack_index(th);
993 new_state = tcp_conntracks[dir][index][old_state];
994 tuple = &ct->tuplehash[dir].tuple;
997 case TCP_CONNTRACK_SYN_SENT:
998 if (old_state < TCP_CONNTRACK_TIME_WAIT)
1000 /* RFC 1122: "When a connection is closed actively,
1001 * it MUST linger in TIME-WAIT state for a time 2xMSL
1002 * (Maximum Segment Lifetime). However, it MAY accept
1003 * a new SYN from the remote TCP to reopen the connection
1004 * directly from TIME-WAIT state, if..."
1005 * We ignore the conditions because we are in the
1006 * TIME-WAIT state anyway.
1008 * Handle aborted connections: we and the server
1009 * think there is an existing connection but the client
1010 * aborts it and starts a new one.
1012 if (((ct->proto.tcp.seen[dir].flags
1013 | ct->proto.tcp.seen[!dir].flags)
1014 & IP_CT_TCP_FLAG_CLOSE_INIT)
1015 || (ct->proto.tcp.last_dir == dir
1016 && ct->proto.tcp.last_index == TCP_RST_SET)) {
1017 /* Attempt to reopen a closed/aborted connection.
1018 * Delete this connection and look up again. */
1019 spin_unlock_bh(&ct->lock);
1021 /* Only repeat if we can actually remove the timer.
1022 * Destruction may already be in progress in process
1023 * context and we must give it a chance to terminate.
1030 case TCP_CONNTRACK_IGNORE:
1033 * Our connection entry may be out of sync, so ignore
1034 * packets which may signal the real connection between
1035 * the client and the server.
1037 * a) SYN in ORIGINAL
1038 * b) SYN/ACK in REPLY
1039 * c) ACK in reply direction after initial SYN in original.
1041 * If the ignored packet is invalid, the receiver will send
1042 * a RST we'll catch below.
1044 if (index == TCP_SYNACK_SET
1045 && ct->proto.tcp.last_index == TCP_SYN_SET
1046 && ct->proto.tcp.last_dir != dir
1047 && ntohl(th->ack_seq) == ct->proto.tcp.last_end) {
1048 /* b) This SYN/ACK acknowledges a SYN that we earlier
1049 * ignored as invalid. This means that the client and
1050 * the server are both in sync, while the firewall is
1051 * not. We get in sync from the previously annotated
1054 old_state = TCP_CONNTRACK_SYN_SENT;
1055 new_state = TCP_CONNTRACK_SYN_RECV;
1056 ct->proto.tcp.seen[ct->proto.tcp.last_dir].td_end =
1057 ct->proto.tcp.last_end;
1058 ct->proto.tcp.seen[ct->proto.tcp.last_dir].td_maxend =
1059 ct->proto.tcp.last_end;
1060 ct->proto.tcp.seen[ct->proto.tcp.last_dir].td_maxwin =
1061 ct->proto.tcp.last_win == 0 ?
1062 1 : ct->proto.tcp.last_win;
1063 ct->proto.tcp.seen[ct->proto.tcp.last_dir].td_scale =
1064 ct->proto.tcp.last_wscale;
1065 ct->proto.tcp.last_flags &= ~IP_CT_EXP_CHALLENGE_ACK;
1066 ct->proto.tcp.seen[ct->proto.tcp.last_dir].flags =
1067 ct->proto.tcp.last_flags;
1068 nf_ct_tcp_state_reset(&ct->proto.tcp.seen[dir]);
1071 ct->proto.tcp.last_index = index;
1072 ct->proto.tcp.last_dir = dir;
1073 ct->proto.tcp.last_seq = ntohl(th->seq);
1074 ct->proto.tcp.last_end =
1075 segment_seq_plus_len(ntohl(th->seq), skb->len, dataoff, th);
1076 ct->proto.tcp.last_win = ntohs(th->window);
1078 /* a) This is a SYN in ORIGINAL. The client and the server
1079 * may be in sync but we are not. In that case, we annotate
1080 * the TCP options and let the packet go through. If it is a
1081 * valid SYN packet, the server will reply with a SYN/ACK, and
1082 * then we'll get in sync. Otherwise, the server potentially
1083 * responds with a challenge ACK if implementing RFC5961.
1085 if (index == TCP_SYN_SET && dir == IP_CT_DIR_ORIGINAL) {
1086 struct ip_ct_tcp_state seen = {};
1088 ct->proto.tcp.last_flags =
1089 ct->proto.tcp.last_wscale = 0;
1090 tcp_options(skb, dataoff, th, &seen);
1091 if (seen.flags & IP_CT_TCP_FLAG_WINDOW_SCALE) {
1092 ct->proto.tcp.last_flags |=
1093 IP_CT_TCP_FLAG_WINDOW_SCALE;
1094 ct->proto.tcp.last_wscale = seen.td_scale;
1096 if (seen.flags & IP_CT_TCP_FLAG_SACK_PERM) {
1097 ct->proto.tcp.last_flags |=
1098 IP_CT_TCP_FLAG_SACK_PERM;
1100 /* Mark the potential for RFC5961 challenge ACK,
1101 * this pose a special problem for LAST_ACK state
1102 * as ACK is intrepretated as ACKing last FIN.
1104 if (old_state == TCP_CONNTRACK_LAST_ACK)
1105 ct->proto.tcp.last_flags |=
1106 IP_CT_EXP_CHALLENGE_ACK;
1109 /* possible challenge ack reply to syn */
1110 if (old_state == TCP_CONNTRACK_SYN_SENT &&
1111 index == TCP_ACK_SET &&
1112 dir == IP_CT_DIR_REPLY)
1113 ct->proto.tcp.last_ack = ntohl(th->ack_seq);
1115 spin_unlock_bh(&ct->lock);
1116 nf_ct_l4proto_log_invalid(skb, ct, state,
1117 "packet (index %d) in dir %d ignored, state %s",
1119 tcp_conntrack_names[old_state]);
1121 case TCP_CONNTRACK_MAX:
1122 /* Special case for SYN proxy: when the SYN to the server or
1123 * the SYN/ACK from the server is lost, the client may transmit
1124 * a keep-alive packet while in SYN_SENT state. This needs to
1125 * be associated with the original conntrack entry in order to
1126 * generate a new SYN with the correct sequence number.
1128 if (nfct_synproxy(ct) && old_state == TCP_CONNTRACK_SYN_SENT &&
1129 index == TCP_ACK_SET && dir == IP_CT_DIR_ORIGINAL &&
1130 ct->proto.tcp.last_dir == IP_CT_DIR_ORIGINAL &&
1131 ct->proto.tcp.seen[dir].td_end - 1 == ntohl(th->seq)) {
1132 pr_debug("nf_ct_tcp: SYN proxy client keep alive\n");
1133 spin_unlock_bh(&ct->lock);
1137 /* Invalid packet */
1138 spin_unlock_bh(&ct->lock);
1139 nf_ct_l4proto_log_invalid(skb, ct, state,
1140 "packet (index %d) in dir %d invalid, state %s",
1142 tcp_conntrack_names[old_state]);
1144 case TCP_CONNTRACK_TIME_WAIT:
1145 /* RFC5961 compliance cause stack to send "challenge-ACK"
1146 * e.g. in response to spurious SYNs. Conntrack MUST
1147 * not believe this ACK is acking last FIN.
1149 if (old_state == TCP_CONNTRACK_LAST_ACK &&
1150 index == TCP_ACK_SET &&
1151 ct->proto.tcp.last_dir != dir &&
1152 ct->proto.tcp.last_index == TCP_SYN_SET &&
1153 (ct->proto.tcp.last_flags & IP_CT_EXP_CHALLENGE_ACK)) {
1154 /* Detected RFC5961 challenge ACK */
1155 ct->proto.tcp.last_flags &= ~IP_CT_EXP_CHALLENGE_ACK;
1156 spin_unlock_bh(&ct->lock);
1157 nf_ct_l4proto_log_invalid(skb, ct, state, "challenge-ack ignored");
1158 return NF_ACCEPT; /* Don't change state */
1161 case TCP_CONNTRACK_SYN_SENT2:
1162 /* tcp_conntracks table is not smart enough to handle
1163 * simultaneous open.
1165 ct->proto.tcp.last_flags |= IP_CT_TCP_SIMULTANEOUS_OPEN;
1167 case TCP_CONNTRACK_SYN_RECV:
1168 if (dir == IP_CT_DIR_REPLY && index == TCP_ACK_SET &&
1169 ct->proto.tcp.last_flags & IP_CT_TCP_SIMULTANEOUS_OPEN)
1170 new_state = TCP_CONNTRACK_ESTABLISHED;
1172 case TCP_CONNTRACK_CLOSE:
1173 if (index != TCP_RST_SET)
1176 /* If we are closing, tuple might have been re-used already.
1177 * last_index, last_ack, and all other ct fields used for
1178 * sequence/window validation are outdated in that case.
1180 * As the conntrack can already be expired by GC under pressure,
1181 * just skip validation checks.
1183 if (tcp_can_early_drop(ct))
1186 /* td_maxack might be outdated if we let a SYN through earlier */
1187 if ((ct->proto.tcp.seen[!dir].flags & IP_CT_TCP_FLAG_MAXACK_SET) &&
1188 ct->proto.tcp.last_index != TCP_SYN_SET) {
1189 u32 seq = ntohl(th->seq);
1191 /* If we are not in established state and SEQ=0 this is most
1192 * likely an answer to a SYN we let go through above (last_index
1193 * can be updated due to out-of-order ACKs).
1195 if (seq == 0 && !nf_conntrack_tcp_established(ct))
1198 if (before(seq, ct->proto.tcp.seen[!dir].td_maxack) &&
1199 !tn->tcp_ignore_invalid_rst) {
1201 spin_unlock_bh(&ct->lock);
1202 nf_ct_l4proto_log_invalid(skb, ct, state, "invalid rst");
1206 if (!nf_conntrack_tcp_established(ct) ||
1207 seq == ct->proto.tcp.seen[!dir].td_maxack)
1210 /* Check if rst is part of train, such as
1211 * foo:80 > bar:4379: P, 235946583:235946602(19) ack 42
1212 * foo:80 > bar:4379: R, 235946602:235946602(0) ack 42
1214 if (ct->proto.tcp.last_index == TCP_ACK_SET &&
1215 ct->proto.tcp.last_dir == dir &&
1216 seq == ct->proto.tcp.last_end)
1219 /* ... RST sequence number doesn't match exactly, keep
1220 * established state to allow a possible challenge ACK.
1222 new_state = old_state;
1224 if (((test_bit(IPS_SEEN_REPLY_BIT, &ct->status)
1225 && ct->proto.tcp.last_index == TCP_SYN_SET)
1226 || (!test_bit(IPS_ASSURED_BIT, &ct->status)
1227 && ct->proto.tcp.last_index == TCP_ACK_SET))
1228 && ntohl(th->ack_seq) == ct->proto.tcp.last_end) {
1229 /* RST sent to invalid SYN or ACK we had let through
1230 * at a) and c) above:
1232 * a) SYN was in window then
1233 * c) we hold a half-open connection.
1235 * Delete our connection entry.
1236 * We skip window checking, because packet might ACK
1237 * segments we ignored. */
1241 /* Reset in response to a challenge-ack we let through earlier */
1242 if (old_state == TCP_CONNTRACK_SYN_SENT &&
1243 ct->proto.tcp.last_index == TCP_ACK_SET &&
1244 ct->proto.tcp.last_dir == IP_CT_DIR_REPLY &&
1245 ntohl(th->seq) == ct->proto.tcp.last_ack)
1250 /* Keep compilers happy. */
1254 res = tcp_in_window(ct, dir, index,
1255 skb, dataoff, th, state);
1257 case NFCT_TCP_IGNORE:
1258 spin_unlock_bh(&ct->lock);
1260 case NFCT_TCP_INVALID:
1261 nf_tcp_handle_invalid(ct, dir, index, skb, state);
1262 spin_unlock_bh(&ct->lock);
1264 case NFCT_TCP_ACCEPT:
1268 /* From now on we have got in-window packets */
1269 ct->proto.tcp.last_index = index;
1270 ct->proto.tcp.last_dir = dir;
1272 pr_debug("tcp_conntracks: ");
1273 nf_ct_dump_tuple(tuple);
1274 pr_debug("syn=%i ack=%i fin=%i rst=%i old=%i new=%i\n",
1275 (th->syn ? 1 : 0), (th->ack ? 1 : 0),
1276 (th->fin ? 1 : 0), (th->rst ? 1 : 0),
1277 old_state, new_state);
1279 ct->proto.tcp.state = new_state;
1280 if (old_state != new_state
1281 && new_state == TCP_CONNTRACK_FIN_WAIT)
1282 ct->proto.tcp.seen[dir].flags |= IP_CT_TCP_FLAG_CLOSE_INIT;
1284 timeouts = nf_ct_timeout_lookup(ct);
1286 timeouts = tn->timeouts;
1288 if (ct->proto.tcp.retrans >= tn->tcp_max_retrans &&
1289 timeouts[new_state] > timeouts[TCP_CONNTRACK_RETRANS])
1290 timeout = timeouts[TCP_CONNTRACK_RETRANS];
1291 else if (unlikely(index == TCP_RST_SET))
1292 timeout = timeouts[TCP_CONNTRACK_CLOSE];
1293 else if ((ct->proto.tcp.seen[0].flags | ct->proto.tcp.seen[1].flags) &
1294 IP_CT_TCP_FLAG_DATA_UNACKNOWLEDGED &&
1295 timeouts[new_state] > timeouts[TCP_CONNTRACK_UNACK])
1296 timeout = timeouts[TCP_CONNTRACK_UNACK];
1297 else if (ct->proto.tcp.last_win == 0 &&
1298 timeouts[new_state] > timeouts[TCP_CONNTRACK_RETRANS])
1299 timeout = timeouts[TCP_CONNTRACK_RETRANS];
1301 timeout = timeouts[new_state];
1302 spin_unlock_bh(&ct->lock);
1304 if (new_state != old_state)
1305 nf_conntrack_event_cache(IPCT_PROTOINFO, ct);
1307 if (!test_bit(IPS_SEEN_REPLY_BIT, &ct->status)) {
1308 /* If only reply is a RST, we can consider ourselves not to
1309 have an established connection: this is a fairly common
1310 problem case, so we can delete the conntrack
1311 immediately. --RR */
1313 nf_ct_kill_acct(ct, ctinfo, skb);
1317 if (index == TCP_SYN_SET && old_state == TCP_CONNTRACK_SYN_SENT) {
1318 /* do not renew timeout on SYN retransmit.
1320 * Else port reuse by client or NAT middlebox can keep
1321 * entry alive indefinitely (including nat info).
1326 /* ESTABLISHED without SEEN_REPLY, i.e. mid-connection
1327 * pickup with loose=1. Avoid large ESTABLISHED timeout.
1329 if (new_state == TCP_CONNTRACK_ESTABLISHED &&
1330 timeout > timeouts[TCP_CONNTRACK_UNACK])
1331 timeout = timeouts[TCP_CONNTRACK_UNACK];
1332 } else if (!test_bit(IPS_ASSURED_BIT, &ct->status)
1333 && (old_state == TCP_CONNTRACK_SYN_RECV
1334 || old_state == TCP_CONNTRACK_ESTABLISHED)
1335 && new_state == TCP_CONNTRACK_ESTABLISHED) {
1336 /* Set ASSURED if we see valid ack in ESTABLISHED
1337 after SYN_RECV or a valid answer for a picked up
1339 set_bit(IPS_ASSURED_BIT, &ct->status);
1340 nf_conntrack_event_cache(IPCT_ASSURED, ct);
1342 nf_ct_refresh_acct(ct, ctinfo, skb, timeout);
1347 #if IS_ENABLED(CONFIG_NF_CT_NETLINK)
1349 #include <linux/netfilter/nfnetlink.h>
1350 #include <linux/netfilter/nfnetlink_conntrack.h>
1352 static int tcp_to_nlattr(struct sk_buff *skb, struct nlattr *nla,
1353 struct nf_conn *ct, bool destroy)
1355 struct nlattr *nest_parms;
1356 struct nf_ct_tcp_flags tmp = {};
1358 spin_lock_bh(&ct->lock);
1359 nest_parms = nla_nest_start(skb, CTA_PROTOINFO_TCP);
1361 goto nla_put_failure;
1363 if (nla_put_u8(skb, CTA_PROTOINFO_TCP_STATE, ct->proto.tcp.state))
1364 goto nla_put_failure;
1369 if (nla_put_u8(skb, CTA_PROTOINFO_TCP_WSCALE_ORIGINAL,
1370 ct->proto.tcp.seen[0].td_scale) ||
1371 nla_put_u8(skb, CTA_PROTOINFO_TCP_WSCALE_REPLY,
1372 ct->proto.tcp.seen[1].td_scale))
1373 goto nla_put_failure;
1375 tmp.flags = ct->proto.tcp.seen[0].flags;
1376 if (nla_put(skb, CTA_PROTOINFO_TCP_FLAGS_ORIGINAL,
1377 sizeof(struct nf_ct_tcp_flags), &tmp))
1378 goto nla_put_failure;
1380 tmp.flags = ct->proto.tcp.seen[1].flags;
1381 if (nla_put(skb, CTA_PROTOINFO_TCP_FLAGS_REPLY,
1382 sizeof(struct nf_ct_tcp_flags), &tmp))
1383 goto nla_put_failure;
1385 spin_unlock_bh(&ct->lock);
1386 nla_nest_end(skb, nest_parms);
1391 spin_unlock_bh(&ct->lock);
1395 static const struct nla_policy tcp_nla_policy[CTA_PROTOINFO_TCP_MAX+1] = {
1396 [CTA_PROTOINFO_TCP_STATE] = { .type = NLA_U8 },
1397 [CTA_PROTOINFO_TCP_WSCALE_ORIGINAL] = { .type = NLA_U8 },
1398 [CTA_PROTOINFO_TCP_WSCALE_REPLY] = { .type = NLA_U8 },
1399 [CTA_PROTOINFO_TCP_FLAGS_ORIGINAL] = { .len = sizeof(struct nf_ct_tcp_flags) },
1400 [CTA_PROTOINFO_TCP_FLAGS_REPLY] = { .len = sizeof(struct nf_ct_tcp_flags) },
1403 #define TCP_NLATTR_SIZE ( \
1404 NLA_ALIGN(NLA_HDRLEN + 1) + \
1405 NLA_ALIGN(NLA_HDRLEN + 1) + \
1406 NLA_ALIGN(NLA_HDRLEN + sizeof(struct nf_ct_tcp_flags)) + \
1407 NLA_ALIGN(NLA_HDRLEN + sizeof(struct nf_ct_tcp_flags)))
1409 static int nlattr_to_tcp(struct nlattr *cda[], struct nf_conn *ct)
1411 struct nlattr *pattr = cda[CTA_PROTOINFO_TCP];
1412 struct nlattr *tb[CTA_PROTOINFO_TCP_MAX+1];
1415 /* updates could not contain anything about the private
1416 * protocol info, in that case skip the parsing */
1420 err = nla_parse_nested_deprecated(tb, CTA_PROTOINFO_TCP_MAX, pattr,
1421 tcp_nla_policy, NULL);
1425 if (tb[CTA_PROTOINFO_TCP_STATE] &&
1426 nla_get_u8(tb[CTA_PROTOINFO_TCP_STATE]) >= TCP_CONNTRACK_MAX)
1429 spin_lock_bh(&ct->lock);
1430 if (tb[CTA_PROTOINFO_TCP_STATE])
1431 ct->proto.tcp.state = nla_get_u8(tb[CTA_PROTOINFO_TCP_STATE]);
1433 if (tb[CTA_PROTOINFO_TCP_FLAGS_ORIGINAL]) {
1434 struct nf_ct_tcp_flags *attr =
1435 nla_data(tb[CTA_PROTOINFO_TCP_FLAGS_ORIGINAL]);
1436 ct->proto.tcp.seen[0].flags &= ~attr->mask;
1437 ct->proto.tcp.seen[0].flags |= attr->flags & attr->mask;
1440 if (tb[CTA_PROTOINFO_TCP_FLAGS_REPLY]) {
1441 struct nf_ct_tcp_flags *attr =
1442 nla_data(tb[CTA_PROTOINFO_TCP_FLAGS_REPLY]);
1443 ct->proto.tcp.seen[1].flags &= ~attr->mask;
1444 ct->proto.tcp.seen[1].flags |= attr->flags & attr->mask;
1447 if (tb[CTA_PROTOINFO_TCP_WSCALE_ORIGINAL] &&
1448 tb[CTA_PROTOINFO_TCP_WSCALE_REPLY] &&
1449 ct->proto.tcp.seen[0].flags & IP_CT_TCP_FLAG_WINDOW_SCALE &&
1450 ct->proto.tcp.seen[1].flags & IP_CT_TCP_FLAG_WINDOW_SCALE) {
1451 ct->proto.tcp.seen[0].td_scale =
1452 nla_get_u8(tb[CTA_PROTOINFO_TCP_WSCALE_ORIGINAL]);
1453 ct->proto.tcp.seen[1].td_scale =
1454 nla_get_u8(tb[CTA_PROTOINFO_TCP_WSCALE_REPLY]);
1456 spin_unlock_bh(&ct->lock);
1461 static unsigned int tcp_nlattr_tuple_size(void)
1463 static unsigned int size __read_mostly;
1466 size = nla_policy_len(nf_ct_port_nla_policy, CTA_PROTO_MAX + 1);
1472 #ifdef CONFIG_NF_CONNTRACK_TIMEOUT
1474 #include <linux/netfilter/nfnetlink.h>
1475 #include <linux/netfilter/nfnetlink_cttimeout.h>
1477 static int tcp_timeout_nlattr_to_obj(struct nlattr *tb[],
1478 struct net *net, void *data)
1480 struct nf_tcp_net *tn = nf_tcp_pernet(net);
1481 unsigned int *timeouts = data;
1485 timeouts = tn->timeouts;
1486 /* set default TCP timeouts. */
1487 for (i=0; i<TCP_CONNTRACK_TIMEOUT_MAX; i++)
1488 timeouts[i] = tn->timeouts[i];
1490 if (tb[CTA_TIMEOUT_TCP_SYN_SENT]) {
1491 timeouts[TCP_CONNTRACK_SYN_SENT] =
1492 ntohl(nla_get_be32(tb[CTA_TIMEOUT_TCP_SYN_SENT]))*HZ;
1495 if (tb[CTA_TIMEOUT_TCP_SYN_RECV]) {
1496 timeouts[TCP_CONNTRACK_SYN_RECV] =
1497 ntohl(nla_get_be32(tb[CTA_TIMEOUT_TCP_SYN_RECV]))*HZ;
1499 if (tb[CTA_TIMEOUT_TCP_ESTABLISHED]) {
1500 timeouts[TCP_CONNTRACK_ESTABLISHED] =
1501 ntohl(nla_get_be32(tb[CTA_TIMEOUT_TCP_ESTABLISHED]))*HZ;
1503 if (tb[CTA_TIMEOUT_TCP_FIN_WAIT]) {
1504 timeouts[TCP_CONNTRACK_FIN_WAIT] =
1505 ntohl(nla_get_be32(tb[CTA_TIMEOUT_TCP_FIN_WAIT]))*HZ;
1507 if (tb[CTA_TIMEOUT_TCP_CLOSE_WAIT]) {
1508 timeouts[TCP_CONNTRACK_CLOSE_WAIT] =
1509 ntohl(nla_get_be32(tb[CTA_TIMEOUT_TCP_CLOSE_WAIT]))*HZ;
1511 if (tb[CTA_TIMEOUT_TCP_LAST_ACK]) {
1512 timeouts[TCP_CONNTRACK_LAST_ACK] =
1513 ntohl(nla_get_be32(tb[CTA_TIMEOUT_TCP_LAST_ACK]))*HZ;
1515 if (tb[CTA_TIMEOUT_TCP_TIME_WAIT]) {
1516 timeouts[TCP_CONNTRACK_TIME_WAIT] =
1517 ntohl(nla_get_be32(tb[CTA_TIMEOUT_TCP_TIME_WAIT]))*HZ;
1519 if (tb[CTA_TIMEOUT_TCP_CLOSE]) {
1520 timeouts[TCP_CONNTRACK_CLOSE] =
1521 ntohl(nla_get_be32(tb[CTA_TIMEOUT_TCP_CLOSE]))*HZ;
1523 if (tb[CTA_TIMEOUT_TCP_SYN_SENT2]) {
1524 timeouts[TCP_CONNTRACK_SYN_SENT2] =
1525 ntohl(nla_get_be32(tb[CTA_TIMEOUT_TCP_SYN_SENT2]))*HZ;
1527 if (tb[CTA_TIMEOUT_TCP_RETRANS]) {
1528 timeouts[TCP_CONNTRACK_RETRANS] =
1529 ntohl(nla_get_be32(tb[CTA_TIMEOUT_TCP_RETRANS]))*HZ;
1531 if (tb[CTA_TIMEOUT_TCP_UNACK]) {
1532 timeouts[TCP_CONNTRACK_UNACK] =
1533 ntohl(nla_get_be32(tb[CTA_TIMEOUT_TCP_UNACK]))*HZ;
1536 timeouts[CTA_TIMEOUT_TCP_UNSPEC] = timeouts[CTA_TIMEOUT_TCP_SYN_SENT];
1541 tcp_timeout_obj_to_nlattr(struct sk_buff *skb, const void *data)
1543 const unsigned int *timeouts = data;
1545 if (nla_put_be32(skb, CTA_TIMEOUT_TCP_SYN_SENT,
1546 htonl(timeouts[TCP_CONNTRACK_SYN_SENT] / HZ)) ||
1547 nla_put_be32(skb, CTA_TIMEOUT_TCP_SYN_RECV,
1548 htonl(timeouts[TCP_CONNTRACK_SYN_RECV] / HZ)) ||
1549 nla_put_be32(skb, CTA_TIMEOUT_TCP_ESTABLISHED,
1550 htonl(timeouts[TCP_CONNTRACK_ESTABLISHED] / HZ)) ||
1551 nla_put_be32(skb, CTA_TIMEOUT_TCP_FIN_WAIT,
1552 htonl(timeouts[TCP_CONNTRACK_FIN_WAIT] / HZ)) ||
1553 nla_put_be32(skb, CTA_TIMEOUT_TCP_CLOSE_WAIT,
1554 htonl(timeouts[TCP_CONNTRACK_CLOSE_WAIT] / HZ)) ||
1555 nla_put_be32(skb, CTA_TIMEOUT_TCP_LAST_ACK,
1556 htonl(timeouts[TCP_CONNTRACK_LAST_ACK] / HZ)) ||
1557 nla_put_be32(skb, CTA_TIMEOUT_TCP_TIME_WAIT,
1558 htonl(timeouts[TCP_CONNTRACK_TIME_WAIT] / HZ)) ||
1559 nla_put_be32(skb, CTA_TIMEOUT_TCP_CLOSE,
1560 htonl(timeouts[TCP_CONNTRACK_CLOSE] / HZ)) ||
1561 nla_put_be32(skb, CTA_TIMEOUT_TCP_SYN_SENT2,
1562 htonl(timeouts[TCP_CONNTRACK_SYN_SENT2] / HZ)) ||
1563 nla_put_be32(skb, CTA_TIMEOUT_TCP_RETRANS,
1564 htonl(timeouts[TCP_CONNTRACK_RETRANS] / HZ)) ||
1565 nla_put_be32(skb, CTA_TIMEOUT_TCP_UNACK,
1566 htonl(timeouts[TCP_CONNTRACK_UNACK] / HZ)))
1567 goto nla_put_failure;
1574 static const struct nla_policy tcp_timeout_nla_policy[CTA_TIMEOUT_TCP_MAX+1] = {
1575 [CTA_TIMEOUT_TCP_SYN_SENT] = { .type = NLA_U32 },
1576 [CTA_TIMEOUT_TCP_SYN_RECV] = { .type = NLA_U32 },
1577 [CTA_TIMEOUT_TCP_ESTABLISHED] = { .type = NLA_U32 },
1578 [CTA_TIMEOUT_TCP_FIN_WAIT] = { .type = NLA_U32 },
1579 [CTA_TIMEOUT_TCP_CLOSE_WAIT] = { .type = NLA_U32 },
1580 [CTA_TIMEOUT_TCP_LAST_ACK] = { .type = NLA_U32 },
1581 [CTA_TIMEOUT_TCP_TIME_WAIT] = { .type = NLA_U32 },
1582 [CTA_TIMEOUT_TCP_CLOSE] = { .type = NLA_U32 },
1583 [CTA_TIMEOUT_TCP_SYN_SENT2] = { .type = NLA_U32 },
1584 [CTA_TIMEOUT_TCP_RETRANS] = { .type = NLA_U32 },
1585 [CTA_TIMEOUT_TCP_UNACK] = { .type = NLA_U32 },
1587 #endif /* CONFIG_NF_CONNTRACK_TIMEOUT */
1589 void nf_conntrack_tcp_init_net(struct net *net)
1591 struct nf_tcp_net *tn = nf_tcp_pernet(net);
1594 for (i = 0; i < TCP_CONNTRACK_TIMEOUT_MAX; i++)
1595 tn->timeouts[i] = tcp_timeouts[i];
1597 /* timeouts[0] is unused, make it same as SYN_SENT so
1598 * ->timeouts[0] contains 'new' timeout, like udp or icmp.
1600 tn->timeouts[0] = tcp_timeouts[TCP_CONNTRACK_SYN_SENT];
1602 /* If it is set to zero, we disable picking up already established
1607 /* "Be conservative in what you do,
1608 * be liberal in what you accept from others."
1609 * If it's non-zero, we mark only out of window RST segments as INVALID.
1611 tn->tcp_be_liberal = 0;
1613 /* If it's non-zero, we turn off RST sequence number check */
1614 tn->tcp_ignore_invalid_rst = 0;
1616 /* Max number of the retransmitted packets without receiving an (acceptable)
1617 * ACK from the destination. If this number is reached, a shorter timer
1620 tn->tcp_max_retrans = 3;
1622 #if IS_ENABLED(CONFIG_NF_FLOW_TABLE)
1623 tn->offload_timeout = 30 * HZ;
1627 const struct nf_conntrack_l4proto nf_conntrack_l4proto_tcp =
1629 .l4proto = IPPROTO_TCP,
1630 #ifdef CONFIG_NF_CONNTRACK_PROCFS
1631 .print_conntrack = tcp_print_conntrack,
1633 .can_early_drop = tcp_can_early_drop,
1634 #if IS_ENABLED(CONFIG_NF_CT_NETLINK)
1635 .to_nlattr = tcp_to_nlattr,
1636 .from_nlattr = nlattr_to_tcp,
1637 .tuple_to_nlattr = nf_ct_port_tuple_to_nlattr,
1638 .nlattr_to_tuple = nf_ct_port_nlattr_to_tuple,
1639 .nlattr_tuple_size = tcp_nlattr_tuple_size,
1640 .nlattr_size = TCP_NLATTR_SIZE,
1641 .nla_policy = nf_ct_port_nla_policy,
1643 #ifdef CONFIG_NF_CONNTRACK_TIMEOUT
1645 .nlattr_to_obj = tcp_timeout_nlattr_to_obj,
1646 .obj_to_nlattr = tcp_timeout_obj_to_nlattr,
1647 .nlattr_max = CTA_TIMEOUT_TCP_MAX,
1648 .obj_size = sizeof(unsigned int) *
1649 TCP_CONNTRACK_TIMEOUT_MAX,
1650 .nla_policy = tcp_timeout_nla_policy,
1652 #endif /* CONFIG_NF_CONNTRACK_TIMEOUT */