xref: /netbsd-src/sys/net/npf/npf_state_tcp.c (revision a4ddc2c8fb9af816efe3b1c375a5530aef0e89e9)
1 /*	$NetBSD: npf_state_tcp.c,v 1.12 2012/12/24 19:05:45 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 1.12 2012/12/24 19:05:45 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(npf_cache_t *npc, nbuf_t *nbuf, npf_state_t *nst, const int di)
300 {
301 	const struct tcphdr * const th = npc->npc_l4.tcp;
302 	const int tcpfl = th->th_flags;
303 	npf_tcpstate_t *fstate, *tstate;
304 	int tcpdlen, ackskew;
305 	tcp_seq seq, ack, end;
306 	uint32_t win;
307 
308 	KASSERT(npf_iscached(npc, NPC_TCP));
309 	KASSERT(di == NPF_FLOW_FORW || di == NPF_FLOW_BACK);
310 
311 	/*
312 	 * Perform SEQ/ACK numbers check against boundaries.  Reference:
313 	 *
314 	 *	Rooij G., "Real stateful TCP packet filtering in IP Filter",
315 	 *	10th USENIX Security Symposium invited talk, Aug. 2001.
316 	 *
317 	 * There are four boundaries defined as following:
318 	 *	I)   SEQ + LEN	<= MAX { SND.ACK + MAX(SND.WIN, 1) }
319 	 *	II)  SEQ	>= MAX { SND.SEQ + SND.LEN - MAX(RCV.WIN, 1) }
320 	 *	III) ACK	<= MAX { RCV.SEQ + RCV.LEN }
321 	 *	IV)  ACK	>= MAX { RCV.SEQ + RCV.LEN } - MAXACKWIN
322 	 *
323 	 * Let these members of npf_tcpstate_t be the maximum seen values of:
324 	 *	nst_end		- SEQ + LEN
325 	 *	nst_maxend	- ACK + MAX(WIN, 1)
326 	 *	nst_maxwin	- MAX(WIN, 1)
327 	 */
328 
329 	tcpdlen = npf_tcpsaw(__UNCONST(npc), &seq, &ack, &win);
330 	end = seq + tcpdlen;
331 	if (tcpfl & TH_SYN) {
332 		end++;
333 	}
334 	if (tcpfl & TH_FIN) {
335 		end++;
336 	}
337 
338 	fstate = &nst->nst_tcpst[di];
339 	tstate = &nst->nst_tcpst[!di];
340 	win = win ? (win << fstate->nst_wscale) : 1;
341 
342 	/*
343 	 * Initialise if the first packet.
344 	 * Note: only case when nst_maxwin is zero.
345 	 */
346 	if (__predict_false(fstate->nst_maxwin == 0)) {
347 		/*
348 		 * Normally, it should be the first SYN or a re-transmission
349 		 * of SYN.  The state of the other side will get set with a
350 		 * SYN-ACK reply (see below).
351 		 */
352 		fstate->nst_end = end;
353 		fstate->nst_maxend = end;
354 		fstate->nst_maxwin = win;
355 		tstate->nst_end = 0;
356 		tstate->nst_maxend = 0;
357 		tstate->nst_maxwin = 1;
358 
359 		/*
360 		 * Handle TCP Window Scaling (RFC 1323).  Both sides may
361 		 * send this option in their SYN packets.
362 		 */
363 		fstate->nst_wscale = 0;
364 		(void)npf_fetch_tcpopts(npc, nbuf, NULL, &fstate->nst_wscale);
365 
366 		tstate->nst_wscale = 0;
367 
368 		/* Done. */
369 		return true;
370 	}
371 	if (fstate->nst_end == 0) {
372 		/*
373 		 * Should be a SYN-ACK reply to SYN.  If SYN is not set,
374 		 * then we are in the middle of connection and lost tracking.
375 		 */
376 		fstate->nst_end = end;
377 		fstate->nst_maxend = end + 1;
378 		fstate->nst_maxwin = win;
379 		fstate->nst_wscale = 0;
380 
381 		/* Handle TCP Window Scaling (must be ignored if no SYN). */
382 		if (tcpfl & TH_SYN) {
383 			(void)npf_fetch_tcpopts(npc, nbuf, NULL,
384 			    &fstate->nst_wscale);
385 		}
386 	}
387 
388 	if ((tcpfl & TH_ACK) == 0) {
389 		/* Pretend that an ACK was sent. */
390 		ack = tstate->nst_end;
391 	} else if ((tcpfl & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST) && ack == 0) {
392 		/* Workaround for some TCP stacks. */
393 		ack = tstate->nst_end;
394 	}
395 
396 	if (__predict_false(tcpfl & TH_RST)) {
397 		/* RST to the initial SYN may have zero SEQ - fix it up. */
398 		if (seq == 0 && nst->nst_state == NPF_TCPS_SYN_SENT) {
399 			end = fstate->nst_end;
400 			seq = end;
401 		}
402 
403 		/* Strict in-order sequence for RST packets. */
404 		if (npf_strict_order_rst && (fstate->nst_end - seq) > 1) {
405 			return false;
406 		}
407 	}
408 
409 	/*
410 	 * Determine whether the data is within previously noted window,
411 	 * that is, upper boundary for valid data (I).
412 	 */
413 	if (!SEQ_LEQ(end, fstate->nst_maxend)) {
414 		npf_stats_inc(NPF_STAT_INVALID_STATE_TCP1);
415 		return false;
416 	}
417 
418 	/* Lower boundary (II), which is no more than one window back. */
419 	if (!SEQ_GEQ(seq, fstate->nst_end - tstate->nst_maxwin)) {
420 		npf_stats_inc(NPF_STAT_INVALID_STATE_TCP2);
421 		return false;
422 	}
423 
424 	/*
425 	 * Boundaries for valid acknowledgments (III, IV) - one predicted
426 	 * window up or down, since packets may be fragmented.
427 	 */
428 	ackskew = tstate->nst_end - ack;
429 	if (ackskew < -NPF_TCP_MAXACKWIN ||
430 	    ackskew > (NPF_TCP_MAXACKWIN << fstate->nst_wscale)) {
431 		npf_stats_inc(NPF_STAT_INVALID_STATE_TCP3);
432 		return false;
433 	}
434 
435 	/*
436 	 * Packet has been passed.
437 	 *
438 	 * Negative ackskew might be due to fragmented packets.  Since the
439 	 * total length of the packet is unknown - bump the boundary.
440 	 */
441 
442 	if (ackskew < 0) {
443 		tstate->nst_end = ack;
444 	}
445 	/* Keep track of the maximum window seen. */
446 	if (fstate->nst_maxwin < win) {
447 		fstate->nst_maxwin = win;
448 	}
449 	if (SEQ_GT(end, fstate->nst_end)) {
450 		fstate->nst_end = end;
451 	}
452 	/* Note the window for upper boundary. */
453 	if (SEQ_GEQ(ack + win, tstate->nst_maxend)) {
454 		tstate->nst_maxend = ack + win;
455 	}
456 	return true;
457 }
458 
459 /*
460  * npf_state_tcp: inspect TCP segment, determine whether it belongs to
461  * the connection and track its state.
462  */
463 bool
464 npf_state_tcp(npf_cache_t *npc, nbuf_t *nbuf, npf_state_t *nst, int di)
465 {
466 	const struct tcphdr * const th = npc->npc_l4.tcp;
467 	const int tcpfl = th->th_flags, state = nst->nst_state;
468 	int nstate;
469 
470 	KASSERT(nst->nst_state == 0 || mutex_owned(&nst->nst_lock));
471 
472 	/* Look for a transition to a new state. */
473 	if (__predict_true((tcpfl & TH_RST) == 0)) {
474 		const int flagcase = npf_tcpfl2case(tcpfl);
475 		nstate = npf_tcp_fsm[state][di][flagcase];
476 	} else if (state == NPF_TCPS_TIME_WAIT) {
477 		/* Prevent TIME-WAIT assassination (RFC 1337). */
478 		nstate = NPF_TCPS_OK;
479 	} else {
480 		nstate = NPF_TCPS_CLOSED;
481 	}
482 
483 	/* Determine whether TCP packet really belongs to this connection. */
484 	if (!npf_tcp_inwindow(npc, nbuf, nst, di)) {
485 		return false;
486 	}
487 	if (__predict_true(nstate == NPF_TCPS_OK)) {
488 		return true;
489 	}
490 
491 	nst->nst_state = nstate;
492 	return true;
493 }
494 
495 int
496 npf_state_tcp_timeout(const npf_state_t *nst)
497 {
498 	const u_int state = nst->nst_state;
499 
500 	KASSERT(state < NPF_TCP_NSTATES);
501 	return npf_tcp_timeouts[state];
502 }
503