xref: /netbsd-src/sys/net/npf/npf_state_tcp.c (revision e61202360d5611414dd6f6115934a96aa1f50b1a)
1 /*	$NetBSD: npf_state_tcp.c,v 1.11 2012/10/06 23:50:17 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 1.11 2012/10/06 23:50:17 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 /*
54  * NPF TCP states.  Note: these states are different from the TCP FSM
55  * states of RFC 793.  The packet filter is a man-in-the-middle.
56  */
57 #define	NPF_TCPS_OK		(-1)
58 #define	NPF_TCPS_CLOSED		0
59 #define	NPF_TCPS_SYN_SENT	1
60 #define	NPF_TCPS_SIMSYN_SENT	2
61 #define	NPF_TCPS_SYN_RECEIVED	3
62 #define	NPF_TCPS_ESTABLISHED	4
63 #define	NPF_TCPS_FIN_SENT	5
64 #define	NPF_TCPS_FIN_RECEIVED	6
65 #define	NPF_TCPS_CLOSE_WAIT	7
66 #define	NPF_TCPS_FIN_WAIT	8
67 #define	NPF_TCPS_CLOSING	9
68 #define	NPF_TCPS_LAST_ACK	10
69 #define	NPF_TCPS_TIME_WAIT	11
70 
71 #define	NPF_TCP_NSTATES		12
72 
73 /*
74  * TCP connection timeout table (in seconds).
75  */
76 static u_int npf_tcp_timeouts[] __read_mostly = {
77 	/* Closed, timeout nearly immediately. */
78 	[NPF_TCPS_CLOSED]	= 10,
79 	/* Unsynchronised states. */
80 	[NPF_TCPS_SYN_SENT]	= 30,
81 	[NPF_TCPS_SIMSYN_SENT]	= 30,
82 	[NPF_TCPS_SYN_RECEIVED]	= 60,
83 	/* Established: 24 hours. */
84 	[NPF_TCPS_ESTABLISHED]	= 60 * 60 * 24,
85 	/* FIN seen: 4 minutes (2 * MSL). */
86 	[NPF_TCPS_FIN_SENT]	= 60 * 2 * 2,
87 	[NPF_TCPS_FIN_RECEIVED]	= 60 * 2 * 2,
88 	/* Half-closed cases: 6 hours. */
89 	[NPF_TCPS_CLOSE_WAIT]	= 60 * 60 * 6,
90 	[NPF_TCPS_FIN_WAIT]	= 60 * 60 * 6,
91 	/* Full close cases: 30 sec and 2 * MSL. */
92 	[NPF_TCPS_CLOSING]	= 30,
93 	[NPF_TCPS_LAST_ACK]	= 30,
94 	[NPF_TCPS_TIME_WAIT]	= 60 * 2 * 2,
95 };
96 
97 static bool npf_strict_order_rst __read_mostly = false;
98 
99 #define	NPF_TCP_MAXACKWIN	66000
100 
101 /*
102  * List of TCP flag cases and conversion of flags to a case (index).
103  */
104 
105 #define	TCPFC_INVALID		0
106 #define	TCPFC_SYN		1
107 #define	TCPFC_SYNACK		2
108 #define	TCPFC_ACK		3
109 #define	TCPFC_FIN		4
110 #define	TCPFC_COUNT		5
111 
112 static inline u_int
113 npf_tcpfl2case(const int tcpfl)
114 {
115 	u_int i, c;
116 
117 	CTASSERT(TH_FIN == 0x01);
118 	CTASSERT(TH_SYN == 0x02);
119 	CTASSERT(TH_ACK == 0x10);
120 
121 	/*
122 	 * Flags are shifted to use three least significant bits, thus each
123 	 * flag combination has a unique number ranging from 0 to 7, e.g.
124 	 * TH_SYN | TH_ACK has number 6, since (0x02 | (0x10 >> 2)) == 6.
125 	 * However, the requirement is to have number 0 for invalid cases,
126 	 * such as TH_SYN | TH_FIN, and to have the same number for TH_FIN
127 	 * and TH_FIN|TH_ACK cases.  Thus, we generate a mask assigning 3
128 	 * bits for each number, which contains the actual case numbers:
129 	 *
130 	 * TCPFC_SYNACK	<< (6 << 2) == 0x2000000 (6 - SYN,ACK)
131 	 * TCPFC_FIN	<< (5 << 2) == 0x0400000 (5 - FIN,ACK)
132 	 * ...
133 	 *
134 	 * Hence, OR'ed mask value is 0x2430140.
135 	 */
136 	i = (tcpfl & (TH_SYN | TH_FIN)) | ((tcpfl & TH_ACK) >> 2);
137 	c = (0x2430140 >> (i << 2)) & 7;
138 
139 	KASSERT(c < TCPFC_COUNT);
140 	return c;
141 }
142 
143 /*
144  * NPF transition table of a tracked TCP connection.
145  *
146  * There is a single state, which is changed in the following way:
147  *
148  * new_state = npf_tcp_fsm[old_state][direction][npf_tcpfl2case(tcp_flags)];
149  *
150  * Note that this state is different from the state in each end (host).
151  */
152 
153 static const int npf_tcp_fsm[NPF_TCP_NSTATES][2][TCPFC_COUNT] = {
154 	[NPF_TCPS_CLOSED] = {
155 		[NPF_FLOW_FORW] = {
156 			/* Handshake (1): initial SYN. */
157 			[TCPFC_SYN]	= NPF_TCPS_SYN_SENT,
158 		},
159 	},
160 	[NPF_TCPS_SYN_SENT] = {
161 		[NPF_FLOW_FORW] = {
162 			/* SYN may be retransmitted. */
163 			[TCPFC_SYN]	= NPF_TCPS_OK,
164 		},
165 		[NPF_FLOW_BACK] = {
166 			/* Handshake (2): SYN-ACK is expected. */
167 			[TCPFC_SYNACK]	= NPF_TCPS_SYN_RECEIVED,
168 			/* Simultaneous initiation - SYN. */
169 			[TCPFC_SYN]	= NPF_TCPS_SIMSYN_SENT,
170 		},
171 	},
172 	[NPF_TCPS_SIMSYN_SENT] = {
173 		[NPF_FLOW_FORW] = {
174 			/* Original SYN re-transmission. */
175 			[TCPFC_SYN]	= NPF_TCPS_OK,
176 			/* SYN-ACK response to simultaneous SYN. */
177 			[TCPFC_SYNACK]	= NPF_TCPS_SYN_RECEIVED,
178 		},
179 		[NPF_FLOW_BACK] = {
180 			/* Simultaneous SYN re-transmission.*/
181 			[TCPFC_SYN]	= NPF_TCPS_OK,
182 			/* SYN-ACK response to original SYN. */
183 			[TCPFC_SYNACK]	= NPF_TCPS_SYN_RECEIVED,
184 			/* FIN may occur early. */
185 			[TCPFC_FIN]	= NPF_TCPS_FIN_RECEIVED,
186 		},
187 	},
188 	[NPF_TCPS_SYN_RECEIVED] = {
189 		[NPF_FLOW_FORW] = {
190 			/* Handshake (3): ACK is expected. */
191 			[TCPFC_ACK]	= NPF_TCPS_ESTABLISHED,
192 			/* FIN may be sent early. */
193 			[TCPFC_FIN]	= NPF_TCPS_FIN_SENT,
194 		},
195 		[NPF_FLOW_BACK] = {
196 			/* SYN-ACK may be retransmitted. */
197 			[TCPFC_SYNACK]	= NPF_TCPS_OK,
198 			/* XXX: ACK of late SYN in simultaneous case? */
199 			[TCPFC_ACK]	= NPF_TCPS_OK,
200 			/* FIN may occur early. */
201 			[TCPFC_FIN]	= NPF_TCPS_FIN_RECEIVED,
202 		},
203 	},
204 	[NPF_TCPS_ESTABLISHED] = {
205 		/*
206 		 * Regular ACKs (data exchange) or FIN.
207 		 * FIN packets may have ACK set.
208 		 */
209 		[NPF_FLOW_FORW] = {
210 			[TCPFC_ACK]	= NPF_TCPS_OK,
211 			/* FIN by the sender. */
212 			[TCPFC_FIN]	= NPF_TCPS_FIN_SENT,
213 		},
214 		[NPF_FLOW_BACK] = {
215 			[TCPFC_ACK]	= NPF_TCPS_OK,
216 			/* FIN by the receiver. */
217 			[TCPFC_FIN]	= NPF_TCPS_FIN_RECEIVED,
218 		},
219 	},
220 	[NPF_TCPS_FIN_SENT] = {
221 		[NPF_FLOW_FORW] = {
222 			/* FIN may be re-transmitted.  Late ACK as well. */
223 			[TCPFC_ACK]	= NPF_TCPS_OK,
224 			[TCPFC_FIN]	= NPF_TCPS_OK,
225 		},
226 		[NPF_FLOW_BACK] = {
227 			/* If ACK, connection is half-closed now. */
228 			[TCPFC_ACK]	= NPF_TCPS_FIN_WAIT,
229 			/* FIN or FIN-ACK race - immediate closing. */
230 			[TCPFC_FIN]	= NPF_TCPS_CLOSING,
231 		},
232 	},
233 	[NPF_TCPS_FIN_RECEIVED] = {
234 		/*
235 		 * FIN was received.  Equivalent scenario to sent FIN.
236 		 */
237 		[NPF_FLOW_FORW] = {
238 			[TCPFC_ACK]	= NPF_TCPS_CLOSE_WAIT,
239 			[TCPFC_FIN]	= NPF_TCPS_CLOSING,
240 		},
241 		[NPF_FLOW_BACK] = {
242 			[TCPFC_ACK]	= NPF_TCPS_OK,
243 			[TCPFC_FIN]	= NPF_TCPS_OK,
244 		},
245 	},
246 	[NPF_TCPS_CLOSE_WAIT] = {
247 		/* Sender has sent the FIN and closed its end. */
248 		[NPF_FLOW_FORW] = {
249 			[TCPFC_ACK]	= NPF_TCPS_OK,
250 			[TCPFC_FIN]	= NPF_TCPS_LAST_ACK,
251 		},
252 		[NPF_FLOW_BACK] = {
253 			[TCPFC_ACK]	= NPF_TCPS_OK,
254 			[TCPFC_FIN]	= NPF_TCPS_LAST_ACK,
255 		},
256 	},
257 	[NPF_TCPS_FIN_WAIT] = {
258 		/* Receiver has closed its end. */
259 		[NPF_FLOW_FORW] = {
260 			[TCPFC_ACK]	= NPF_TCPS_OK,
261 			[TCPFC_FIN]	= NPF_TCPS_LAST_ACK,
262 		},
263 		[NPF_FLOW_BACK] = {
264 			[TCPFC_ACK]	= NPF_TCPS_OK,
265 			[TCPFC_FIN]	= NPF_TCPS_LAST_ACK,
266 		},
267 	},
268 	[NPF_TCPS_CLOSING] = {
269 		/* Race of FINs - expecting ACK. */
270 		[NPF_FLOW_FORW] = {
271 			[TCPFC_ACK]	= NPF_TCPS_LAST_ACK,
272 		},
273 		[NPF_FLOW_BACK] = {
274 			[TCPFC_ACK]	= NPF_TCPS_LAST_ACK,
275 		},
276 	},
277 	[NPF_TCPS_LAST_ACK] = {
278 		/* FINs exchanged - expecting last ACK. */
279 		[NPF_FLOW_FORW] = {
280 			[TCPFC_ACK]	= NPF_TCPS_TIME_WAIT,
281 		},
282 		[NPF_FLOW_BACK] = {
283 			[TCPFC_ACK]	= NPF_TCPS_TIME_WAIT,
284 		},
285 	},
286 	[NPF_TCPS_TIME_WAIT] = {
287 		/* May re-open the connection as per RFC 1122. */
288 		[NPF_FLOW_FORW] = {
289 			[TCPFC_SYN]	= NPF_TCPS_SYN_SENT,
290 		},
291 	},
292 };
293 
294 /*
295  * npf_tcp_inwindow: determine whether the packet is in the TCP window
296  * and thus part of the connection we are tracking.
297  */
298 static bool
299 npf_tcp_inwindow(const npf_cache_t *npc, nbuf_t *nbuf, npf_state_t *nst,
300     const int di)
301 {
302 	const struct tcphdr * const th = &npc->npc_l4.tcp;
303 	const int tcpfl = th->th_flags;
304 	npf_tcpstate_t *fstate, *tstate;
305 	int tcpdlen, ackskew;
306 	tcp_seq seq, ack, end;
307 	uint32_t win;
308 
309 	KASSERT(npf_iscached(npc, NPC_TCP));
310 	KASSERT(di == NPF_FLOW_FORW || di == NPF_FLOW_BACK);
311 
312 	/*
313 	 * Perform SEQ/ACK numbers check against boundaries.  Reference:
314 	 *
315 	 *	Rooij G., "Real stateful TCP packet filtering in IP Filter",
316 	 *	10th USENIX Security Symposium invited talk, Aug. 2001.
317 	 *
318 	 * There are four boundaries defined as following:
319 	 *	I)   SEQ + LEN	<= MAX { SND.ACK + MAX(SND.WIN, 1) }
320 	 *	II)  SEQ	>= MAX { SND.SEQ + SND.LEN - MAX(RCV.WIN, 1) }
321 	 *	III) ACK	<= MAX { RCV.SEQ + RCV.LEN }
322 	 *	IV)  ACK	>= MAX { RCV.SEQ + RCV.LEN } - MAXACKWIN
323 	 *
324 	 * Let these members of npf_tcpstate_t be the maximum seen values of:
325 	 *	nst_end		- SEQ + LEN
326 	 *	nst_maxend	- ACK + MAX(WIN, 1)
327 	 *	nst_maxwin	- MAX(WIN, 1)
328 	 */
329 
330 	tcpdlen = npf_tcpsaw(__UNCONST(npc), &seq, &ack, &win);
331 	end = seq + tcpdlen;
332 	if (tcpfl & TH_SYN) {
333 		end++;
334 	}
335 	if (tcpfl & TH_FIN) {
336 		end++;
337 	}
338 
339 	fstate = &nst->nst_tcpst[di];
340 	tstate = &nst->nst_tcpst[!di];
341 	win = win ? (win << fstate->nst_wscale) : 1;
342 
343 	/*
344 	 * Initialise if the first packet.
345 	 * Note: only case when nst_maxwin is zero.
346 	 */
347 	if (__predict_false(fstate->nst_maxwin == 0)) {
348 		/*
349 		 * Normally, it should be the first SYN or a re-transmission
350 		 * of SYN.  The state of the other side will get set with a
351 		 * SYN-ACK reply (see below).
352 		 */
353 		fstate->nst_end = end;
354 		fstate->nst_maxend = end;
355 		fstate->nst_maxwin = win;
356 		tstate->nst_end = 0;
357 		tstate->nst_maxend = 0;
358 		tstate->nst_maxwin = 1;
359 
360 		/*
361 		 * Handle TCP Window Scaling (RFC 1323).  Both sides may
362 		 * send this option in their SYN packets.
363 		 */
364 		fstate->nst_wscale = 0;
365 		(void)npf_fetch_tcpopts(npc, nbuf, NULL, &fstate->nst_wscale);
366 
367 		tstate->nst_wscale = 0;
368 
369 		/* Done. */
370 		return true;
371 	}
372 	if (fstate->nst_end == 0) {
373 		/*
374 		 * Should be a SYN-ACK reply to SYN.  If SYN is not set,
375 		 * then we are in the middle of connection and lost tracking.
376 		 */
377 		fstate->nst_end = end;
378 		fstate->nst_maxend = end + 1;
379 		fstate->nst_maxwin = win;
380 		fstate->nst_wscale = 0;
381 
382 		/* Handle TCP Window Scaling (must be ignored if no SYN). */
383 		if (tcpfl & TH_SYN) {
384 			(void)npf_fetch_tcpopts(npc, nbuf, NULL,
385 			    &fstate->nst_wscale);
386 		}
387 	}
388 
389 	if ((tcpfl & TH_ACK) == 0) {
390 		/* Pretend that an ACK was sent. */
391 		ack = tstate->nst_end;
392 	} else if ((tcpfl & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST) && ack == 0) {
393 		/* Workaround for some TCP stacks. */
394 		ack = tstate->nst_end;
395 	}
396 
397 	if (__predict_false(tcpfl & TH_RST)) {
398 		/* RST to the initial SYN may have zero SEQ - fix it up. */
399 		if (seq == 0 && nst->nst_state == NPF_TCPS_SYN_SENT) {
400 			end = fstate->nst_end;
401 			seq = end;
402 		}
403 
404 		/* Strict in-order sequence for RST packets. */
405 		if (npf_strict_order_rst && (fstate->nst_end - seq) > 1) {
406 			return false;
407 		}
408 	}
409 
410 	/*
411 	 * Determine whether the data is within previously noted window,
412 	 * that is, upper boundary for valid data (I).
413 	 */
414 	if (!SEQ_LEQ(end, fstate->nst_maxend)) {
415 		npf_stats_inc(NPF_STAT_INVALID_STATE_TCP1);
416 		return false;
417 	}
418 
419 	/* Lower boundary (II), which is no more than one window back. */
420 	if (!SEQ_GEQ(seq, fstate->nst_end - tstate->nst_maxwin)) {
421 		npf_stats_inc(NPF_STAT_INVALID_STATE_TCP2);
422 		return false;
423 	}
424 
425 	/*
426 	 * Boundaries for valid acknowledgments (III, IV) - one predicted
427 	 * window up or down, since packets may be fragmented.
428 	 */
429 	ackskew = tstate->nst_end - ack;
430 	if (ackskew < -NPF_TCP_MAXACKWIN ||
431 	    ackskew > (NPF_TCP_MAXACKWIN << fstate->nst_wscale)) {
432 		npf_stats_inc(NPF_STAT_INVALID_STATE_TCP3);
433 		return false;
434 	}
435 
436 	/*
437 	 * Packet has been passed.
438 	 *
439 	 * Negative ackskew might be due to fragmented packets.  Since the
440 	 * total length of the packet is unknown - bump the boundary.
441 	 */
442 
443 	if (ackskew < 0) {
444 		tstate->nst_end = ack;
445 	}
446 	/* Keep track of the maximum window seen. */
447 	if (fstate->nst_maxwin < win) {
448 		fstate->nst_maxwin = win;
449 	}
450 	if (SEQ_GT(end, fstate->nst_end)) {
451 		fstate->nst_end = end;
452 	}
453 	/* Note the window for upper boundary. */
454 	if (SEQ_GEQ(ack + win, tstate->nst_maxend)) {
455 		tstate->nst_maxend = ack + win;
456 	}
457 	return true;
458 }
459 
460 /*
461  * npf_state_tcp: inspect TCP segment, determine whether it belongs to
462  * the connection and track its state.
463  */
464 bool
465 npf_state_tcp(const npf_cache_t *npc, nbuf_t *nbuf, npf_state_t *nst, int di)
466 {
467 	const struct tcphdr * const th = &npc->npc_l4.tcp;
468 	const int tcpfl = th->th_flags, state = nst->nst_state;
469 	int nstate;
470 
471 	KASSERT(nst->nst_state == 0 || mutex_owned(&nst->nst_lock));
472 
473 	/* Look for a transition to a new state. */
474 	if (__predict_true((tcpfl & TH_RST) == 0)) {
475 		const int flagcase = npf_tcpfl2case(tcpfl);
476 		nstate = npf_tcp_fsm[state][di][flagcase];
477 	} else if (state == NPF_TCPS_TIME_WAIT) {
478 		/* Prevent TIME-WAIT assassination (RFC 1337). */
479 		nstate = NPF_TCPS_OK;
480 	} else {
481 		nstate = NPF_TCPS_CLOSED;
482 	}
483 
484 	/* Determine whether TCP packet really belongs to this connection. */
485 	if (!npf_tcp_inwindow(npc, nbuf, nst, di)) {
486 		return false;
487 	}
488 	if (__predict_true(nstate == NPF_TCPS_OK)) {
489 		return true;
490 	}
491 
492 	nst->nst_state = nstate;
493 	return true;
494 }
495 
496 int
497 npf_state_tcp_timeout(const npf_state_t *nst)
498 {
499 	const u_int state = nst->nst_state;
500 
501 	KASSERT(state < NPF_TCP_NSTATES);
502 	return npf_tcp_timeouts[state];
503 }
504