xref: /netbsd-src/sys/net/npf/npf_state_tcp.c (revision 9ddb6ab554e70fb9bbd90c3d96b812bc57755a14)
1 /*	$NetBSD: npf_state_tcp.c,v 1.3 2011/12/08 23:36:57 rmind Exp $	*/
2 
3 /*-
4  * Copyright (c) 2010-2011 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 1.3 2011/12/08 23:36:57 rmind Exp $");
38 
39 #include <sys/param.h>
40 #include <sys/types.h>
41 
42 #ifndef _KERNEL
43 #include <stdio.h>
44 #include <stdbool.h>
45 #include <inttypes.h>
46 #endif
47 #include <netinet/in.h>
48 #include <netinet/tcp.h>
49 #include <netinet/tcp_seq.h>
50 
51 #include "npf_impl.h"
52 
53 #if defined(_NPF_TESTING)
54 void	npf_state_sample(npf_state_t *);
55 #define	NPF_TCP_STATE_SAMPLE(nst)	npf_state_sample(nst)
56 #else
57 #define	NPF_TCP_STATE_SAMPLE(nst)
58 #endif
59 
60 /*
61  * NPF TCP states.  Note: these states are different from the TCP FSM
62  * states of RFC 793.  Mind that packet filter is a man-in-the-middle.
63  */
64 #define NPF_TCPS_OK		(-1)
65 #define	NPF_TCPS_CLOSED		0
66 #define	NPF_TCPS_SYN_SENT	1
67 #define	NPF_TCPS_SIMSYN_SENT	2
68 #define	NPF_TCPS_SYN_RECEIVED	3
69 #define	NPF_TCPS_ESTABLISHED	4
70 #define	NPF_TCPS_FIN_SEEN	5
71 #define	NPF_TCPS_CLOSE_WAIT	6
72 #define	NPF_TCPS_FIN_WAIT	7
73 #define	NPF_TCPS_CLOSING	8
74 #define	NPF_TCPS_LAST_ACK	9
75 #define	NPF_TCPS_TIME_WAIT	10
76 
77 #define	NPF_TCP_NSTATES		11
78 
79 /*
80  * TCP connection timeout table (in seconds).
81  */
82 static u_int npf_tcp_timeouts[] __read_mostly = {
83 	/* Closed, timeout nearly immediately. */
84 	[NPF_TCPS_CLOSED]	= 10,
85 	/* Unsynchronised states. */
86 	[NPF_TCPS_SYN_SENT]	= 30,
87 	[NPF_TCPS_SIMSYN_SENT]	= 30,
88 	[NPF_TCPS_SYN_RECEIVED]	= 60,
89 	/* Established, timeout: 24 hours. */
90 	[NPF_TCPS_ESTABLISHED]	= 60 * 60 * 24,
91 	/* Closure cases, timeout: 4 minutes (2 * MSL). */
92 	[NPF_TCPS_FIN_SEEN]	= 60 * 2 * 2,
93 	[NPF_TCPS_CLOSE_WAIT]	= 60 * 2 * 2,
94 	[NPF_TCPS_FIN_WAIT]	= 60 * 2 * 2,
95 	[NPF_TCPS_CLOSING]	= 30,
96 	[NPF_TCPS_LAST_ACK]	= 30,
97 	[NPF_TCPS_TIME_WAIT]	= 60 * 2 * 2,
98 };
99 
100 #define	NPF_TCP_MAXACKWIN	66000
101 
102 /*
103  * List of TCP flag cases and conversion of flags to a case (index).
104  */
105 
106 #define	TCPFC_INVALID		0
107 #define	TCPFC_SYN		1
108 #define	TCPFC_SYNACK		2
109 #define	TCPFC_ACK		3
110 #define	TCPFC_FIN		4
111 #define	TCPFC_COUNT		5
112 
113 static inline u_int
114 npf_tcpfl2case(const int tcpfl)
115 {
116 	u_int i, c;
117 
118 	CTASSERT(TH_FIN == 0x01);
119 	CTASSERT(TH_SYN == 0x02);
120 	CTASSERT(TH_ACK == 0x10);
121 
122 	/*
123 	 * Flags are shifted to use three least significant bits, thus each
124 	 * flag combination has a unique number ranging from 0 to 7, e.g.
125 	 * TH_SYN | TH_ACK has number 6, since (0x02 | (0x10 >> 2)) == 6.
126 	 * However, the requirement is to have number 0 for invalid cases,
127 	 * such as TH_SYN | TH_FIN, and to have the same number for TH_FIN
128 	 * and TH_FIN|TH_ACK cases.  Thus, we generate a mask assigning 3
129 	 * bits for each number, which contains the actual case numbers:
130 	 *
131 	 * TCPFC_SYNACK	<< (6 << 2) == 0x2000000 (6 - SYN,ACK)
132 	 * TCPFC_FIN	<< (5 << 2) == 0x0400000 (5 - FIN,ACK)
133 	 * ...
134 	 *
135 	 * Hence, OR'ed mask value is 0x2430140.
136 	 */
137 	i = (tcpfl & (TH_SYN | TH_FIN)) | ((tcpfl & TH_ACK) >> 2);
138 	c = (0x2430140 >> (i << 2)) & 7;
139 
140 	KASSERT(c < TCPFC_COUNT);
141 	return c;
142 }
143 
144 /*
145  * NPF transition table of a tracked TCP connection.
146  *
147  * There is a single state, which is changed in the following way:
148  *
149  * new_state = npf_tcp_fsm[old_state][direction][npf_tcpfl2case(tcp_flags)];
150  *
151  * Note that this state is different from the state in each end (host).
152  */
153 
154 static const int npf_tcp_fsm[NPF_TCP_NSTATES][2][TCPFC_COUNT] = {
155 	[NPF_TCPS_CLOSED] = {
156 		[NPF_FLOW_FORW] = {
157 			/* Handshake (1): initial SYN. */
158 			[TCPFC_SYN]	= NPF_TCPS_SYN_SENT,
159 		},
160 	},
161 	[NPF_TCPS_SYN_SENT] = {
162 		[NPF_FLOW_FORW] = {
163 			/* SYN may be retransmitted. */
164 			[TCPFC_SYN]	= NPF_TCPS_OK,
165 		},
166 		[NPF_FLOW_BACK] = {
167 			/* Handshake (2): SYN-ACK is expected. */
168 			[TCPFC_SYNACK]	= NPF_TCPS_SYN_RECEIVED,
169 			/* Simultaneous initiation - SYN. */
170 			[TCPFC_SYN]	= NPF_TCPS_SIMSYN_SENT,
171 		},
172 	},
173 	[NPF_TCPS_SIMSYN_SENT] = {
174 		[NPF_FLOW_FORW] = {
175 			/* Original SYN re-transmission. */
176 			[TCPFC_SYN]	= NPF_TCPS_OK,
177 			/* SYN-ACK response to simultaneous SYN. */
178 			[TCPFC_SYNACK]	= NPF_TCPS_SYN_RECEIVED,
179 		},
180 		[NPF_FLOW_BACK] = {
181 			/* Simultaneous SYN re-transmission.*/
182 			[TCPFC_SYN]	= NPF_TCPS_OK,
183 			/* SYN-ACK response to original SYN. */
184 			[TCPFC_SYNACK]	= NPF_TCPS_SYN_RECEIVED,
185 			/* FIN may be sent early. */
186 			[TCPFC_FIN]	= NPF_TCPS_FIN_SEEN,
187 		},
188 	},
189 	[NPF_TCPS_SYN_RECEIVED] = {
190 		[NPF_FLOW_FORW] = {
191 			/* Handshake (3): ACK is expected. */
192 			[TCPFC_ACK]	= NPF_TCPS_ESTABLISHED,
193 			/* FIN may be sent early. */
194 			[TCPFC_FIN]	= NPF_TCPS_FIN_SEEN,
195 		},
196 		[NPF_FLOW_BACK] = {
197 			/* SYN-ACK may be retransmitted. */
198 			[TCPFC_SYNACK]	= NPF_TCPS_OK,
199 			/* XXX: ACK of late SYN in simultaneous case? */
200 			[TCPFC_ACK]	= NPF_TCPS_OK,
201 			/* FIN may be sent early. */
202 			[TCPFC_FIN]	= NPF_TCPS_FIN_SEEN,
203 		},
204 	},
205 	[NPF_TCPS_ESTABLISHED] = {
206 		/*
207 		 * Regular ACKs (data exchange) or FIN.
208 		 * FIN packets may have ACK set.
209 		 */
210 		[NPF_FLOW_FORW] = {
211 			[TCPFC_ACK]	= NPF_TCPS_OK,
212 			/* FIN by the sender. */
213 			[TCPFC_FIN]	= NPF_TCPS_FIN_SEEN,
214 		},
215 		[NPF_FLOW_BACK] = {
216 			[TCPFC_ACK]	= NPF_TCPS_OK,
217 			/* FIN by the receiver. */
218 			[TCPFC_FIN]	= NPF_TCPS_FIN_SEEN,
219 		},
220 	},
221 	[NPF_TCPS_FIN_SEEN] = {
222 		/*
223 		 * FIN was seen.  If ACK only, connection is half-closed now,
224 		 * need to determine which end is closed (sender or receiver).
225 		 * However, both FIN and FIN-ACK may race here - in which
226 		 * case we are closing immediately.
227 		 */
228 		[NPF_FLOW_FORW] = {
229 			[TCPFC_ACK]	= NPF_TCPS_CLOSE_WAIT,
230 			[TCPFC_FIN]	= NPF_TCPS_CLOSING,
231 		},
232 		[NPF_FLOW_BACK] = {
233 			[TCPFC_ACK]	= NPF_TCPS_FIN_WAIT,
234 			[TCPFC_FIN]	= NPF_TCPS_CLOSING,
235 		},
236 	},
237 	[NPF_TCPS_CLOSE_WAIT] = {
238 		/* Sender has sent the FIN and closed its end. */
239 		[NPF_FLOW_FORW] = {
240 			[TCPFC_ACK]	= NPF_TCPS_OK,
241 			[TCPFC_FIN]	= NPF_TCPS_LAST_ACK,
242 		},
243 		[NPF_FLOW_BACK] = {
244 			[TCPFC_ACK]	= NPF_TCPS_OK,
245 			[TCPFC_FIN]	= NPF_TCPS_LAST_ACK,
246 		},
247 	},
248 	[NPF_TCPS_FIN_WAIT] = {
249 		/* Receiver has closed its end. */
250 		[NPF_FLOW_FORW] = {
251 			[TCPFC_ACK]	= NPF_TCPS_OK,
252 			[TCPFC_FIN]	= NPF_TCPS_LAST_ACK,
253 		},
254 		[NPF_FLOW_BACK] = {
255 			[TCPFC_ACK]	= NPF_TCPS_OK,
256 			[TCPFC_FIN]	= NPF_TCPS_LAST_ACK,
257 		},
258 	},
259 	[NPF_TCPS_CLOSING] = {
260 		/* Race of FINs - expecting ACK. */
261 		[NPF_FLOW_FORW] = {
262 			[TCPFC_ACK]	= NPF_TCPS_LAST_ACK,
263 		},
264 		[NPF_FLOW_BACK] = {
265 			[TCPFC_ACK]	= NPF_TCPS_LAST_ACK,
266 		},
267 	},
268 	[NPF_TCPS_LAST_ACK] = {
269 		/* FINs exchanged - expecting last ACK. */
270 		[NPF_FLOW_FORW] = {
271 			[TCPFC_ACK]	= NPF_TCPS_TIME_WAIT,
272 		},
273 		[NPF_FLOW_BACK] = {
274 			[TCPFC_ACK]	= NPF_TCPS_TIME_WAIT,
275 		},
276 	},
277 	[NPF_TCPS_TIME_WAIT] = {
278 		/* May re-open the connection as per RFC 1122. */
279 		[NPF_FLOW_FORW] = {
280 			[TCPFC_SYN]	= NPF_TCPS_SYN_SENT,
281 		},
282 	},
283 };
284 
285 /*
286  * npf_tcp_inwindow: determine whether the packet is in the TCP window
287  * and thus part of the connection we are tracking.
288  */
289 static bool
290 npf_tcp_inwindow(const npf_cache_t *npc, nbuf_t *nbuf, npf_state_t *nst,
291     const int di)
292 {
293 	const struct tcphdr * const th = &npc->npc_l4.tcp;
294 	const int tcpfl = th->th_flags;
295 	npf_tcpstate_t *fstate, *tstate;
296 	int tcpdlen, wscale, ackskew;
297 	tcp_seq seq, ack, end;
298 	uint32_t win;
299 
300 	KASSERT(npf_iscached(npc, NPC_TCP));
301 	KASSERT(di == NPF_FLOW_FORW || di == NPF_FLOW_BACK);
302 
303 	/*
304 	 * Perform SEQ/ACK numbers check against boundaries.  Reference:
305 	 *
306 	 *	Rooij G., "Real stateful TCP packet filtering in IP Filter",
307 	 *	10th USENIX Security Symposium invited talk, Aug. 2001.
308 	 *
309 	 * There are four boundaries defined as following:
310 	 *	I)   SEQ + LEN	<= MAX { SND.ACK + MAX(SND.WIN, 1) }
311 	 *	II)  SEQ	>= MAX { SND.SEQ + SND.LEN - MAX(RCV.WIN, 1) }
312 	 *	III) ACK	<= MAX { RCV.SEQ + RCV.LEN }
313 	 *	IV)  ACK	>= MAX { RCV.SEQ + RCV.LEN } - MAXACKWIN
314 	 *
315 	 * Let these members of npf_tcpstate_t be the maximum seen values of:
316 	 *	nst_end		- SEQ + LEN
317 	 *	nst_maxend	- ACK + MAX(WIN, 1)
318 	 *	nst_maxwin	- MAX(WIN, 1)
319 	 */
320 
321 	tcpdlen = npf_tcpsaw(__UNCONST(npc), &seq, &ack, &win);
322 	end = seq + tcpdlen;
323 	if (tcpfl & TH_SYN) {
324 		end++;
325 	}
326 	if (tcpfl & TH_FIN) {
327 		end++;
328 	}
329 
330 	fstate = &nst->nst_tcpst[di];
331 	tstate = &nst->nst_tcpst[!di];
332 	win = win ? (win << fstate->nst_wscale) : 1;
333 
334 	/*
335 	 * Initialise if the first packet.
336 	 * Note: only case when nst_maxwin is zero.
337 	 */
338 	if (__predict_false(fstate->nst_maxwin == 0)) {
339 		/*
340 		 * Should be first SYN or re-transmission of SYN.  State of
341 		 * other side will get set with a SYN-ACK reply (see below).
342 		 */
343 		fstate->nst_end = end;
344 		fstate->nst_maxend = end;
345 		fstate->nst_maxwin = win;
346 		tstate->nst_end = 0;
347 		tstate->nst_maxend = 0;
348 		tstate->nst_maxwin = 1;
349 
350 		/*
351 		 * Handle TCP Window Scaling (RFC 1323).  Both sides may
352 		 * send this option in their SYN packets.
353 		 */
354 		if (npf_fetch_tcpopts(npc, nbuf, NULL, &wscale)) {
355 			fstate->nst_wscale = wscale;
356 		} else {
357 			fstate->nst_wscale = 0;
358 		}
359 		tstate->nst_wscale = 0;
360 
361 		/* Done. */
362 		return true;
363 	}
364 	if (fstate->nst_end == 0) {
365 		/*
366 		 * Should be a SYN-ACK reply to SYN.  If SYN is not set,
367 		 * then we are in the middle of connection and lost tracking.
368 		 */
369 		fstate->nst_end = end;
370 		fstate->nst_maxend = end + 1;
371 		fstate->nst_maxwin = win;
372 
373 		/* Handle TCP Window Scaling (must be ignored if no SYN). */
374 		if (tcpfl & TH_SYN) {
375 			fstate->nst_wscale =
376 			    npf_fetch_tcpopts(npc, nbuf, NULL, &wscale) ?
377 			    wscale : 0;
378 		}
379 	}
380 	if ((tcpfl & TH_ACK) == 0) {
381 		/* Pretend that an ACK was sent. */
382 		ack = tstate->nst_end;
383 	} else if ((tcpfl & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST) && ack == 0) {
384 		/* Workaround for some TCP stacks. */
385 		ack = tstate->nst_end;
386 	}
387 	if (seq == end) {
388 		/* If packet contains no data - assume it is valid. */
389 		end = fstate->nst_end;
390 		seq = end;
391 	}
392 
393 	NPF_TCP_STATE_SAMPLE(nst);
394 #if 0
395 	/* Strict in-order sequence for RST packets. */
396 	if (((tcpfl & TH_RST) != 0) && (fstate->nst_end - seq) > 1) {
397 		return false;
398 	}
399 #endif
400 	/*
401 	 * Determine whether the data is within previously noted window,
402 	 * that is, upper boundary for valid data (I).
403 	 */
404 	if (!SEQ_LEQ(end, fstate->nst_maxend)) {
405 		npf_stats_inc(NPF_STAT_INVALID_STATE_TCP1);
406 		return false;
407 	}
408 
409 	/* Lower boundary (II), which is no more than one window back. */
410 	if (!SEQ_GEQ(seq, fstate->nst_end - tstate->nst_maxwin)) {
411 		npf_stats_inc(NPF_STAT_INVALID_STATE_TCP2);
412 		return false;
413 	}
414 
415 	/*
416 	 * Boundaries for valid acknowledgments (III, IV) - on predicted
417 	 * window up or down, since packets may be fragmented.
418 	 */
419 	ackskew = tstate->nst_end - ack;
420 	if (ackskew < -NPF_TCP_MAXACKWIN ||
421 	    ackskew > (NPF_TCP_MAXACKWIN << fstate->nst_wscale)) {
422 		npf_stats_inc(NPF_STAT_INVALID_STATE_TCP3);
423 		return false;
424 	}
425 
426 	/*
427 	 * Packet has been passed.
428 	 *
429 	 * Negative ackskew might be due to fragmented packets.  Since the
430 	 * total length of the packet is unknown - bump the boundary.
431 	 */
432 	if (ackskew < 0) {
433 		tstate->nst_end = end;
434 	}
435 	/* Keep track of the maximum window seen. */
436 	if (fstate->nst_maxwin < win) {
437 		fstate->nst_maxwin = win;
438 	}
439 	if (SEQ_GT(end, fstate->nst_end)) {
440 		fstate->nst_end = end;
441 	}
442 	/* Note the window for upper boundary. */
443 	if (SEQ_GEQ(ack + win, tstate->nst_maxend)) {
444 		tstate->nst_maxend = ack + win;
445 	}
446 	return true;
447 }
448 
449 bool
450 npf_state_tcp(const npf_cache_t *npc, nbuf_t *nbuf, npf_state_t *nst, int di)
451 {
452 	const struct tcphdr * const th = &npc->npc_l4.tcp;
453 	const int tcpfl = th->th_flags, state = nst->nst_state;
454 	int nstate;
455 
456 	/* Look for a transition to a new state. */
457 	if (__predict_true((tcpfl & TH_RST) == 0)) {
458 		const int flagcase = npf_tcpfl2case(tcpfl);
459 		nstate = npf_tcp_fsm[state][di][flagcase];
460 	} else if (state == NPF_TCPS_TIME_WAIT) {
461 		/* Prevent TIME-WAIT assassination (RFC 1337). */
462 		nstate = NPF_TCPS_OK;
463 	} else {
464 		nstate = NPF_TCPS_CLOSED;
465 	}
466 	/* Determine whether TCP packet really belongs to this connection. */
467 	if (!npf_tcp_inwindow(npc, nbuf, nst, di)) {
468 		return false;
469 	}
470 	if (__predict_true(nstate == NPF_TCPS_OK)) {
471 		return true;
472 	}
473 	nst->nst_state = nstate;
474 	return true;
475 }
476 
477 int
478 npf_state_tcp_timeout(const npf_state_t *nst)
479 {
480 	const u_int state = nst->nst_state;
481 
482 	KASSERT(state < NPF_TCP_NSTATES);
483 	return npf_tcp_timeouts[state];
484 }
485