xref: /netbsd-src/sys/net/npf/npf_state_tcp.c (revision 946379e7b37692fc43f68eb0d1c10daa0a7f3b6c)
1 /*	$NetBSD: npf_state_tcp.c,v 1.16 2014/07/25 20:07:32 rmind Exp $	*/
2 
3 /*-
4  * Copyright (c) 2010-2012 The NetBSD Foundation, Inc.
5  * All rights reserved.
6  *
7  * This material is based upon work partially supported by The
8  * NetBSD Foundation under a contract with Mindaugas Rasiukevicius.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  *
19  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
20  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
21  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
22  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
23  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
24  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
25  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
26  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
27  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
28  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
29  * POSSIBILITY OF SUCH DAMAGE.
30  */
31 
32 /*
33  * NPF TCP state engine for connection tracking.
34  */
35 
36 #include <sys/cdefs.h>
37 __KERNEL_RCSID(0, "$NetBSD: npf_state_tcp.c,v 1.16 2014/07/25 20:07:32 rmind Exp $");
38 
39 #include <sys/param.h>
40 #include <sys/types.h>
41 
42 #include <netinet/in.h>
43 #include <netinet/tcp.h>
44 #include <netinet/tcp_seq.h>
45 
46 #include "npf_impl.h"
47 
48 /*
49  * NPF TCP states.  Note: these states are different from the TCP FSM
50  * states of RFC 793.  The packet filter is a man-in-the-middle.
51  */
52 #define	NPF_TCPS_OK		255
53 #define	NPF_TCPS_CLOSED		0
54 #define	NPF_TCPS_SYN_SENT	1
55 #define	NPF_TCPS_SIMSYN_SENT	2
56 #define	NPF_TCPS_SYN_RECEIVED	3
57 #define	NPF_TCPS_ESTABLISHED	4
58 #define	NPF_TCPS_FIN_SENT	5
59 #define	NPF_TCPS_FIN_RECEIVED	6
60 #define	NPF_TCPS_CLOSE_WAIT	7
61 #define	NPF_TCPS_FIN_WAIT	8
62 #define	NPF_TCPS_CLOSING	9
63 #define	NPF_TCPS_LAST_ACK	10
64 #define	NPF_TCPS_TIME_WAIT	11
65 
66 #define	NPF_TCP_NSTATES		12
67 
68 /*
69  * TCP connection timeout table (in seconds).
70  */
71 static u_int npf_tcp_timeouts[] __read_mostly = {
72 	/* Closed, timeout nearly immediately. */
73 	[NPF_TCPS_CLOSED]	= 10,
74 	/* Unsynchronised states. */
75 	[NPF_TCPS_SYN_SENT]	= 30,
76 	[NPF_TCPS_SIMSYN_SENT]	= 30,
77 	[NPF_TCPS_SYN_RECEIVED]	= 60,
78 	/* Established: 24 hours. */
79 	[NPF_TCPS_ESTABLISHED]	= 60 * 60 * 24,
80 	/* FIN seen: 4 minutes (2 * MSL). */
81 	[NPF_TCPS_FIN_SENT]	= 60 * 2 * 2,
82 	[NPF_TCPS_FIN_RECEIVED]	= 60 * 2 * 2,
83 	/* Half-closed cases: 6 hours. */
84 	[NPF_TCPS_CLOSE_WAIT]	= 60 * 60 * 6,
85 	[NPF_TCPS_FIN_WAIT]	= 60 * 60 * 6,
86 	/* Full close cases: 30 sec and 2 * MSL. */
87 	[NPF_TCPS_CLOSING]	= 30,
88 	[NPF_TCPS_LAST_ACK]	= 30,
89 	[NPF_TCPS_TIME_WAIT]	= 60 * 2 * 2,
90 };
91 
92 static bool npf_strict_order_rst __read_mostly = true;
93 
94 #define	NPF_TCP_MAXACKWIN	66000
95 
96 /*
97  * List of TCP flag cases and conversion of flags to a case (index).
98  */
99 
100 #define	TCPFC_INVALID		0
101 #define	TCPFC_SYN		1
102 #define	TCPFC_SYNACK		2
103 #define	TCPFC_ACK		3
104 #define	TCPFC_FIN		4
105 #define	TCPFC_COUNT		5
106 
107 static inline u_int
108 npf_tcpfl2case(const u_int tcpfl)
109 {
110 	u_int i, c;
111 
112 	CTASSERT(TH_FIN == 0x01);
113 	CTASSERT(TH_SYN == 0x02);
114 	CTASSERT(TH_ACK == 0x10);
115 
116 	/*
117 	 * Flags are shifted to use three least significant bits, thus each
118 	 * flag combination has a unique number ranging from 0 to 7, e.g.
119 	 * TH_SYN | TH_ACK has number 6, since (0x02 | (0x10 >> 2)) == 6.
120 	 * However, the requirement is to have number 0 for invalid cases,
121 	 * such as TH_SYN | TH_FIN, and to have the same number for TH_FIN
122 	 * and TH_FIN|TH_ACK cases.  Thus, we generate a mask assigning 3
123 	 * bits for each number, which contains the actual case numbers:
124 	 *
125 	 * TCPFC_SYNACK	<< (6 << 2) == 0x2000000 (6 - SYN,ACK)
126 	 * TCPFC_FIN	<< (5 << 2) == 0x0400000 (5 - FIN,ACK)
127 	 * ...
128 	 *
129 	 * Hence, OR'ed mask value is 0x2430140.
130 	 */
131 	i = (tcpfl & (TH_SYN | TH_FIN)) | ((tcpfl & TH_ACK) >> 2);
132 	c = (0x2430140 >> (i << 2)) & 7;
133 
134 	KASSERT(c < TCPFC_COUNT);
135 	return c;
136 }
137 
138 /*
139  * NPF transition table of a tracked TCP connection.
140  *
141  * There is a single state, which is changed in the following way:
142  *
143  * new_state = npf_tcp_fsm[old_state][direction][npf_tcpfl2case(tcp_flags)];
144  *
145  * Note that this state is different from the state in each end (host).
146  */
147 
148 static const uint8_t npf_tcp_fsm[NPF_TCP_NSTATES][2][TCPFC_COUNT] = {
149 	[NPF_TCPS_CLOSED] = {
150 		[NPF_FLOW_FORW] = {
151 			/* Handshake (1): initial SYN. */
152 			[TCPFC_SYN]	= NPF_TCPS_SYN_SENT,
153 		},
154 	},
155 	[NPF_TCPS_SYN_SENT] = {
156 		[NPF_FLOW_FORW] = {
157 			/* SYN may be retransmitted. */
158 			[TCPFC_SYN]	= NPF_TCPS_OK,
159 		},
160 		[NPF_FLOW_BACK] = {
161 			/* Handshake (2): SYN-ACK is expected. */
162 			[TCPFC_SYNACK]	= NPF_TCPS_SYN_RECEIVED,
163 			/* Simultaneous initiation - SYN. */
164 			[TCPFC_SYN]	= NPF_TCPS_SIMSYN_SENT,
165 		},
166 	},
167 	[NPF_TCPS_SIMSYN_SENT] = {
168 		[NPF_FLOW_FORW] = {
169 			/* Original SYN re-transmission. */
170 			[TCPFC_SYN]	= NPF_TCPS_OK,
171 			/* SYN-ACK response to simultaneous SYN. */
172 			[TCPFC_SYNACK]	= NPF_TCPS_SYN_RECEIVED,
173 		},
174 		[NPF_FLOW_BACK] = {
175 			/* Simultaneous SYN re-transmission.*/
176 			[TCPFC_SYN]	= NPF_TCPS_OK,
177 			/* SYN-ACK response to original SYN. */
178 			[TCPFC_SYNACK]	= NPF_TCPS_SYN_RECEIVED,
179 			/* FIN may occur early. */
180 			[TCPFC_FIN]	= NPF_TCPS_FIN_RECEIVED,
181 		},
182 	},
183 	[NPF_TCPS_SYN_RECEIVED] = {
184 		[NPF_FLOW_FORW] = {
185 			/* Handshake (3): ACK is expected. */
186 			[TCPFC_ACK]	= NPF_TCPS_ESTABLISHED,
187 			/* FIN may be sent early. */
188 			[TCPFC_FIN]	= NPF_TCPS_FIN_SENT,
189 		},
190 		[NPF_FLOW_BACK] = {
191 			/* SYN-ACK may be retransmitted. */
192 			[TCPFC_SYNACK]	= NPF_TCPS_OK,
193 			/* XXX: ACK of late SYN in simultaneous case? */
194 			[TCPFC_ACK]	= NPF_TCPS_OK,
195 			/* FIN may occur early. */
196 			[TCPFC_FIN]	= NPF_TCPS_FIN_RECEIVED,
197 		},
198 	},
199 	[NPF_TCPS_ESTABLISHED] = {
200 		/*
201 		 * Regular ACKs (data exchange) or FIN.
202 		 * FIN packets may have ACK set.
203 		 */
204 		[NPF_FLOW_FORW] = {
205 			[TCPFC_ACK]	= NPF_TCPS_OK,
206 			/* FIN by the sender. */
207 			[TCPFC_FIN]	= NPF_TCPS_FIN_SENT,
208 		},
209 		[NPF_FLOW_BACK] = {
210 			[TCPFC_ACK]	= NPF_TCPS_OK,
211 			/* FIN by the receiver. */
212 			[TCPFC_FIN]	= NPF_TCPS_FIN_RECEIVED,
213 		},
214 	},
215 	[NPF_TCPS_FIN_SENT] = {
216 		[NPF_FLOW_FORW] = {
217 			/* FIN may be re-transmitted.  Late ACK as well. */
218 			[TCPFC_ACK]	= NPF_TCPS_OK,
219 			[TCPFC_FIN]	= NPF_TCPS_OK,
220 		},
221 		[NPF_FLOW_BACK] = {
222 			/* If ACK, connection is half-closed now. */
223 			[TCPFC_ACK]	= NPF_TCPS_FIN_WAIT,
224 			/* FIN or FIN-ACK race - immediate closing. */
225 			[TCPFC_FIN]	= NPF_TCPS_CLOSING,
226 		},
227 	},
228 	[NPF_TCPS_FIN_RECEIVED] = {
229 		/*
230 		 * FIN was received.  Equivalent scenario to sent FIN.
231 		 */
232 		[NPF_FLOW_FORW] = {
233 			[TCPFC_ACK]	= NPF_TCPS_CLOSE_WAIT,
234 			[TCPFC_FIN]	= NPF_TCPS_CLOSING,
235 		},
236 		[NPF_FLOW_BACK] = {
237 			[TCPFC_ACK]	= NPF_TCPS_OK,
238 			[TCPFC_FIN]	= NPF_TCPS_OK,
239 		},
240 	},
241 	[NPF_TCPS_CLOSE_WAIT] = {
242 		/* Sender has sent the FIN and closed its end. */
243 		[NPF_FLOW_FORW] = {
244 			[TCPFC_ACK]	= NPF_TCPS_OK,
245 			[TCPFC_FIN]	= NPF_TCPS_LAST_ACK,
246 		},
247 		[NPF_FLOW_BACK] = {
248 			[TCPFC_ACK]	= NPF_TCPS_OK,
249 			[TCPFC_FIN]	= NPF_TCPS_LAST_ACK,
250 		},
251 	},
252 	[NPF_TCPS_FIN_WAIT] = {
253 		/* Receiver has closed its end. */
254 		[NPF_FLOW_FORW] = {
255 			[TCPFC_ACK]	= NPF_TCPS_OK,
256 			[TCPFC_FIN]	= NPF_TCPS_LAST_ACK,
257 		},
258 		[NPF_FLOW_BACK] = {
259 			[TCPFC_ACK]	= NPF_TCPS_OK,
260 			[TCPFC_FIN]	= NPF_TCPS_LAST_ACK,
261 		},
262 	},
263 	[NPF_TCPS_CLOSING] = {
264 		/* Race of FINs - expecting ACK. */
265 		[NPF_FLOW_FORW] = {
266 			[TCPFC_ACK]	= NPF_TCPS_LAST_ACK,
267 		},
268 		[NPF_FLOW_BACK] = {
269 			[TCPFC_ACK]	= NPF_TCPS_LAST_ACK,
270 		},
271 	},
272 	[NPF_TCPS_LAST_ACK] = {
273 		/* FINs exchanged - expecting last ACK. */
274 		[NPF_FLOW_FORW] = {
275 			[TCPFC_ACK]	= NPF_TCPS_TIME_WAIT,
276 		},
277 		[NPF_FLOW_BACK] = {
278 			[TCPFC_ACK]	= NPF_TCPS_TIME_WAIT,
279 		},
280 	},
281 	[NPF_TCPS_TIME_WAIT] = {
282 		/* May re-open the connection as per RFC 1122. */
283 		[NPF_FLOW_FORW] = {
284 			[TCPFC_SYN]	= NPF_TCPS_SYN_SENT,
285 		},
286 	},
287 };
288 
289 /*
290  * npf_tcp_inwindow: determine whether the packet is in the TCP window
291  * and thus part of the connection we are tracking.
292  */
293 static bool
294 npf_tcp_inwindow(npf_cache_t *npc, npf_state_t *nst, const int di)
295 {
296 	const struct tcphdr * const th = npc->npc_l4.tcp;
297 	const int tcpfl = th->th_flags;
298 	npf_tcpstate_t *fstate, *tstate;
299 	int tcpdlen, ackskew;
300 	tcp_seq seq, ack, end;
301 	uint32_t win;
302 
303 	KASSERT(npf_iscached(npc, NPC_TCP));
304 	KASSERT(di == NPF_FLOW_FORW || di == NPF_FLOW_BACK);
305 
306 	/*
307 	 * Perform SEQ/ACK numbers check against boundaries.  Reference:
308 	 *
309 	 *	Rooij G., "Real stateful TCP packet filtering in IP Filter",
310 	 *	10th USENIX Security Symposium invited talk, Aug. 2001.
311 	 *
312 	 * There are four boundaries defined as following:
313 	 *	I)   SEQ + LEN	<= MAX { SND.ACK + MAX(SND.WIN, 1) }
314 	 *	II)  SEQ	>= MAX { SND.SEQ + SND.LEN - MAX(RCV.WIN, 1) }
315 	 *	III) ACK	<= MAX { RCV.SEQ + RCV.LEN }
316 	 *	IV)  ACK	>= MAX { RCV.SEQ + RCV.LEN } - MAXACKWIN
317 	 *
318 	 * Let these members of npf_tcpstate_t be the maximum seen values of:
319 	 *	nst_end		- SEQ + LEN
320 	 *	nst_maxend	- ACK + MAX(WIN, 1)
321 	 *	nst_maxwin	- MAX(WIN, 1)
322 	 */
323 
324 	tcpdlen = npf_tcpsaw(__UNCONST(npc), &seq, &ack, &win);
325 	end = seq + tcpdlen;
326 	if (tcpfl & TH_SYN) {
327 		end++;
328 	}
329 	if (tcpfl & TH_FIN) {
330 		end++;
331 	}
332 
333 	fstate = &nst->nst_tcpst[di];
334 	tstate = &nst->nst_tcpst[!di];
335 	win = win ? (win << fstate->nst_wscale) : 1;
336 
337 	/*
338 	 * Initialise if the first packet.
339 	 * Note: only case when nst_maxwin is zero.
340 	 */
341 	if (__predict_false(fstate->nst_maxwin == 0)) {
342 		/*
343 		 * Normally, it should be the first SYN or a re-transmission
344 		 * of SYN.  The state of the other side will get set with a
345 		 * SYN-ACK reply (see below).
346 		 */
347 		fstate->nst_end = end;
348 		fstate->nst_maxend = end;
349 		fstate->nst_maxwin = win;
350 		tstate->nst_end = 0;
351 		tstate->nst_maxend = 0;
352 		tstate->nst_maxwin = 1;
353 
354 		/*
355 		 * Handle TCP Window Scaling (RFC 1323).  Both sides may
356 		 * send this option in their SYN packets.
357 		 */
358 		fstate->nst_wscale = 0;
359 		(void)npf_fetch_tcpopts(npc, NULL, &fstate->nst_wscale);
360 
361 		tstate->nst_wscale = 0;
362 
363 		/* Done. */
364 		return true;
365 	}
366 
367 	if (fstate->nst_end == 0) {
368 		/*
369 		 * Should be a SYN-ACK reply to SYN.  If SYN is not set,
370 		 * then we are in the middle of connection and lost tracking.
371 		 */
372 		fstate->nst_end = end;
373 		fstate->nst_maxend = end + 1;
374 		fstate->nst_maxwin = win;
375 		fstate->nst_wscale = 0;
376 
377 		/* Handle TCP Window Scaling (must be ignored if no SYN). */
378 		if (tcpfl & TH_SYN) {
379 			(void)npf_fetch_tcpopts(npc, NULL, &fstate->nst_wscale);
380 		}
381 	}
382 
383 	if ((tcpfl & TH_ACK) == 0) {
384 		/* Pretend that an ACK was sent. */
385 		ack = tstate->nst_end;
386 	} else if ((tcpfl & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST) && ack == 0) {
387 		/* Workaround for some TCP stacks. */
388 		ack = tstate->nst_end;
389 	}
390 
391 	if (__predict_false(tcpfl & TH_RST)) {
392 		/* RST to the initial SYN may have zero SEQ - fix it up. */
393 		if (seq == 0 && nst->nst_state == NPF_TCPS_SYN_SENT) {
394 			end = fstate->nst_end;
395 			seq = end;
396 		}
397 
398 		/* Strict in-order sequence for RST packets (RFC 5961). */
399 		if (npf_strict_order_rst && (fstate->nst_end - seq) > 1) {
400 			return false;
401 		}
402 	}
403 
404 	/*
405 	 * Determine whether the data is within previously noted window,
406 	 * that is, upper boundary for valid data (I).
407 	 */
408 	if (!SEQ_LEQ(end, fstate->nst_maxend)) {
409 		npf_stats_inc(NPF_STAT_INVALID_STATE_TCP1);
410 		return false;
411 	}
412 
413 	/* Lower boundary (II), which is no more than one window back. */
414 	if (!SEQ_GEQ(seq, fstate->nst_end - tstate->nst_maxwin)) {
415 		npf_stats_inc(NPF_STAT_INVALID_STATE_TCP2);
416 		return false;
417 	}
418 
419 	/*
420 	 * Boundaries for valid acknowledgments (III, IV) - one predicted
421 	 * window up or down, since packets may be fragmented.
422 	 */
423 	ackskew = tstate->nst_end - ack;
424 	if (ackskew < -NPF_TCP_MAXACKWIN ||
425 	    ackskew > (NPF_TCP_MAXACKWIN << fstate->nst_wscale)) {
426 		npf_stats_inc(NPF_STAT_INVALID_STATE_TCP3);
427 		return false;
428 	}
429 
430 	/*
431 	 * Packet has been passed.
432 	 *
433 	 * Negative ackskew might be due to fragmented packets.  Since the
434 	 * total length of the packet is unknown - bump the boundary.
435 	 */
436 
437 	if (ackskew < 0) {
438 		tstate->nst_end = ack;
439 	}
440 	/* Keep track of the maximum window seen. */
441 	if (fstate->nst_maxwin < win) {
442 		fstate->nst_maxwin = win;
443 	}
444 	if (SEQ_GT(end, fstate->nst_end)) {
445 		fstate->nst_end = end;
446 	}
447 	/* Note the window for upper boundary. */
448 	if (SEQ_GEQ(ack + win, tstate->nst_maxend)) {
449 		tstate->nst_maxend = ack + win;
450 	}
451 	return true;
452 }
453 
454 /*
455  * npf_state_tcp: inspect TCP segment, determine whether it belongs to
456  * the connection and track its state.
457  */
458 bool
459 npf_state_tcp(npf_cache_t *npc, npf_state_t *nst, int di)
460 {
461 	const struct tcphdr * const th = npc->npc_l4.tcp;
462 	const u_int tcpfl = th->th_flags, state = nst->nst_state;
463 	u_int nstate;
464 
465 	KASSERT(nst->nst_state < NPF_TCP_NSTATES);
466 
467 	/* Look for a transition to a new state. */
468 	if (__predict_true((tcpfl & TH_RST) == 0)) {
469 		const u_int flagcase = npf_tcpfl2case(tcpfl);
470 		nstate = npf_tcp_fsm[state][di][flagcase];
471 	} else if (state == NPF_TCPS_TIME_WAIT) {
472 		/* Prevent TIME-WAIT assassination (RFC 1337). */
473 		nstate = NPF_TCPS_OK;
474 	} else {
475 		nstate = NPF_TCPS_CLOSED;
476 	}
477 
478 	/* Determine whether TCP packet really belongs to this connection. */
479 	if (!npf_tcp_inwindow(npc, nst, di)) {
480 		return false;
481 	}
482 	if (__predict_true(nstate == NPF_TCPS_OK)) {
483 		return true;
484 	}
485 
486 	nst->nst_state = nstate;
487 	return true;
488 }
489 
490 int
491 npf_state_tcp_timeout(const npf_state_t *nst)
492 {
493 	const u_int state = nst->nst_state;
494 
495 	KASSERT(state < NPF_TCP_NSTATES);
496 	return npf_tcp_timeouts[state];
497 }
498