xref: /netbsd-src/sys/net/npf/npf_state_tcp.c (revision bdc22b2e01993381dcefeff2bc9b56ca75a4235c)
1 /*	$NetBSD: npf_state_tcp.c,v 1.18 2016/12/26 23:10:46 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 #ifdef _KERNEL
37 #include <sys/cdefs.h>
38 __KERNEL_RCSID(0, "$NetBSD: npf_state_tcp.c,v 1.18 2016/12/26 23:10:46 rmind Exp $");
39 
40 #include <sys/param.h>
41 #include <sys/types.h>
42 
43 #include <netinet/in.h>
44 #include <netinet/tcp.h>
45 #endif
46 
47 #include "npf_impl.h"
48 
49 /*
50  * NPF TCP states.  Note: these states are different from the TCP FSM
51  * states of RFC 793.  The packet filter is a man-in-the-middle.
52  */
53 #define	NPF_TCPS_OK		255
54 #define	NPF_TCPS_CLOSED		0
55 #define	NPF_TCPS_SYN_SENT	1
56 #define	NPF_TCPS_SIMSYN_SENT	2
57 #define	NPF_TCPS_SYN_RECEIVED	3
58 #define	NPF_TCPS_ESTABLISHED	4
59 #define	NPF_TCPS_FIN_SENT	5
60 #define	NPF_TCPS_FIN_RECEIVED	6
61 #define	NPF_TCPS_CLOSE_WAIT	7
62 #define	NPF_TCPS_FIN_WAIT	8
63 #define	NPF_TCPS_CLOSING	9
64 #define	NPF_TCPS_LAST_ACK	10
65 #define	NPF_TCPS_TIME_WAIT	11
66 
67 #define	NPF_TCP_NSTATES		12
68 
69 /*
70  * TCP connection timeout table (in seconds).
71  */
72 static u_int npf_tcp_timeouts[] __read_mostly = {
73 	/* Closed, timeout nearly immediately. */
74 	[NPF_TCPS_CLOSED]	= 10,
75 	/* Unsynchronised states. */
76 	[NPF_TCPS_SYN_SENT]	= 30,
77 	[NPF_TCPS_SIMSYN_SENT]	= 30,
78 	[NPF_TCPS_SYN_RECEIVED]	= 60,
79 	/* Established: 24 hours. */
80 	[NPF_TCPS_ESTABLISHED]	= 60 * 60 * 24,
81 	/* FIN seen: 4 minutes (2 * MSL). */
82 	[NPF_TCPS_FIN_SENT]	= 60 * 2 * 2,
83 	[NPF_TCPS_FIN_RECEIVED]	= 60 * 2 * 2,
84 	/* Half-closed cases: 6 hours. */
85 	[NPF_TCPS_CLOSE_WAIT]	= 60 * 60 * 6,
86 	[NPF_TCPS_FIN_WAIT]	= 60 * 60 * 6,
87 	/* Full close cases: 30 sec and 2 * MSL. */
88 	[NPF_TCPS_CLOSING]	= 30,
89 	[NPF_TCPS_LAST_ACK]	= 30,
90 	[NPF_TCPS_TIME_WAIT]	= 60 * 2 * 2,
91 };
92 
93 static bool npf_strict_order_rst __read_mostly = true;
94 
95 #define	NPF_TCP_MAXACKWIN	66000
96 
97 #define	SEQ_LT(a,b)		((int)((a)-(b)) < 0)
98 #define	SEQ_LEQ(a,b)		((int)((a)-(b)) <= 0)
99 #define	SEQ_GT(a,b)		((int)((a)-(b)) > 0)
100 #define	SEQ_GEQ(a,b)		((int)((a)-(b)) >= 0)
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 u_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 uint8_t 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 occur early. */
186 			[TCPFC_FIN]	= NPF_TCPS_FIN_RECEIVED,
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_SENT,
195 			/* Late SYN re-transmission. */
196 			[TCPFC_SYN]	= NPF_TCPS_OK,
197 		},
198 		[NPF_FLOW_BACK] = {
199 			/* SYN-ACK may be retransmitted. */
200 			[TCPFC_SYNACK]	= NPF_TCPS_OK,
201 			/* XXX: ACK of late SYN in simultaneous case? */
202 			[TCPFC_ACK]	= NPF_TCPS_OK,
203 			/* FIN may occur early. */
204 			[TCPFC_FIN]	= NPF_TCPS_FIN_RECEIVED,
205 		},
206 	},
207 	[NPF_TCPS_ESTABLISHED] = {
208 		/*
209 		 * Regular ACKs (data exchange) or FIN.
210 		 * FIN packets may have ACK set.
211 		 */
212 		[NPF_FLOW_FORW] = {
213 			[TCPFC_ACK]	= NPF_TCPS_OK,
214 			/* FIN by the sender. */
215 			[TCPFC_FIN]	= NPF_TCPS_FIN_SENT,
216 		},
217 		[NPF_FLOW_BACK] = {
218 			[TCPFC_ACK]	= NPF_TCPS_OK,
219 			/* FIN by the receiver. */
220 			[TCPFC_FIN]	= NPF_TCPS_FIN_RECEIVED,
221 		},
222 	},
223 	[NPF_TCPS_FIN_SENT] = {
224 		[NPF_FLOW_FORW] = {
225 			/* FIN may be re-transmitted.  Late ACK as well. */
226 			[TCPFC_ACK]	= NPF_TCPS_OK,
227 			[TCPFC_FIN]	= NPF_TCPS_OK,
228 		},
229 		[NPF_FLOW_BACK] = {
230 			/* If ACK, connection is half-closed now. */
231 			[TCPFC_ACK]	= NPF_TCPS_FIN_WAIT,
232 			/* FIN or FIN-ACK race - immediate closing. */
233 			[TCPFC_FIN]	= NPF_TCPS_CLOSING,
234 		},
235 	},
236 	[NPF_TCPS_FIN_RECEIVED] = {
237 		/*
238 		 * FIN was received.  Equivalent scenario to sent FIN.
239 		 */
240 		[NPF_FLOW_FORW] = {
241 			[TCPFC_ACK]	= NPF_TCPS_CLOSE_WAIT,
242 			[TCPFC_FIN]	= NPF_TCPS_CLOSING,
243 		},
244 		[NPF_FLOW_BACK] = {
245 			[TCPFC_ACK]	= NPF_TCPS_OK,
246 			[TCPFC_FIN]	= NPF_TCPS_OK,
247 		},
248 	},
249 	[NPF_TCPS_CLOSE_WAIT] = {
250 		/* Sender has sent the FIN and closed its end. */
251 		[NPF_FLOW_FORW] = {
252 			[TCPFC_ACK]	= NPF_TCPS_OK,
253 			[TCPFC_FIN]	= NPF_TCPS_LAST_ACK,
254 		},
255 		[NPF_FLOW_BACK] = {
256 			[TCPFC_ACK]	= NPF_TCPS_OK,
257 			[TCPFC_FIN]	= NPF_TCPS_LAST_ACK,
258 		},
259 	},
260 	[NPF_TCPS_FIN_WAIT] = {
261 		/* Receiver has closed its end. */
262 		[NPF_FLOW_FORW] = {
263 			[TCPFC_ACK]	= NPF_TCPS_OK,
264 			[TCPFC_FIN]	= NPF_TCPS_LAST_ACK,
265 		},
266 		[NPF_FLOW_BACK] = {
267 			[TCPFC_ACK]	= NPF_TCPS_OK,
268 			[TCPFC_FIN]	= NPF_TCPS_LAST_ACK,
269 		},
270 	},
271 	[NPF_TCPS_CLOSING] = {
272 		/* Race of FINs - expecting ACK. */
273 		[NPF_FLOW_FORW] = {
274 			[TCPFC_ACK]	= NPF_TCPS_LAST_ACK,
275 		},
276 		[NPF_FLOW_BACK] = {
277 			[TCPFC_ACK]	= NPF_TCPS_LAST_ACK,
278 		},
279 	},
280 	[NPF_TCPS_LAST_ACK] = {
281 		/* FINs exchanged - expecting last ACK. */
282 		[NPF_FLOW_FORW] = {
283 			[TCPFC_ACK]	= NPF_TCPS_TIME_WAIT,
284 		},
285 		[NPF_FLOW_BACK] = {
286 			[TCPFC_ACK]	= NPF_TCPS_TIME_WAIT,
287 		},
288 	},
289 	[NPF_TCPS_TIME_WAIT] = {
290 		/* May re-open the connection as per RFC 1122. */
291 		[NPF_FLOW_FORW] = {
292 			[TCPFC_SYN]	= NPF_TCPS_SYN_SENT,
293 		},
294 	},
295 };
296 
297 /*
298  * npf_tcp_inwindow: determine whether the packet is in the TCP window
299  * and thus part of the connection we are tracking.
300  */
301 static bool
302 npf_tcp_inwindow(npf_cache_t *npc, npf_state_t *nst, const int di)
303 {
304 	const struct tcphdr * const th = npc->npc_l4.tcp;
305 	const int tcpfl = th->th_flags;
306 	npf_tcpstate_t *fstate, *tstate;
307 	int tcpdlen, ackskew;
308 	tcp_seq seq, ack, end;
309 	uint32_t win;
310 
311 	KASSERT(npf_iscached(npc, NPC_TCP));
312 	KASSERT(di == NPF_FLOW_FORW || di == NPF_FLOW_BACK);
313 
314 	/*
315 	 * Perform SEQ/ACK numbers check against boundaries.  Reference:
316 	 *
317 	 *	Rooij G., "Real stateful TCP packet filtering in IP Filter",
318 	 *	10th USENIX Security Symposium invited talk, Aug. 2001.
319 	 *
320 	 * There are four boundaries defined as following:
321 	 *	I)   SEQ + LEN	<= MAX { SND.ACK + MAX(SND.WIN, 1) }
322 	 *	II)  SEQ	>= MAX { SND.SEQ + SND.LEN - MAX(RCV.WIN, 1) }
323 	 *	III) ACK	<= MAX { RCV.SEQ + RCV.LEN }
324 	 *	IV)  ACK	>= MAX { RCV.SEQ + RCV.LEN } - MAXACKWIN
325 	 *
326 	 * Let these members of npf_tcpstate_t be the maximum seen values of:
327 	 *	nst_end		- SEQ + LEN
328 	 *	nst_maxend	- ACK + MAX(WIN, 1)
329 	 *	nst_maxwin	- MAX(WIN, 1)
330 	 */
331 
332 	tcpdlen = npf_tcpsaw(__UNCONST(npc), &seq, &ack, &win);
333 	end = seq + tcpdlen;
334 	if (tcpfl & TH_SYN) {
335 		end++;
336 	}
337 	if (tcpfl & TH_FIN) {
338 		end++;
339 	}
340 
341 	fstate = &nst->nst_tcpst[di];
342 	tstate = &nst->nst_tcpst[!di];
343 	win = win ? (win << fstate->nst_wscale) : 1;
344 
345 	/*
346 	 * Initialise if the first packet.
347 	 * Note: only case when nst_maxwin is zero.
348 	 */
349 	if (__predict_false(fstate->nst_maxwin == 0)) {
350 		/*
351 		 * Normally, it should be the first SYN or a re-transmission
352 		 * of SYN.  The state of the other side will get set with a
353 		 * SYN-ACK reply (see below).
354 		 */
355 		fstate->nst_end = end;
356 		fstate->nst_maxend = end;
357 		fstate->nst_maxwin = win;
358 		tstate->nst_end = 0;
359 		tstate->nst_maxend = 0;
360 		tstate->nst_maxwin = 1;
361 
362 		/*
363 		 * Handle TCP Window Scaling (RFC 1323).  Both sides may
364 		 * send this option in their SYN packets.
365 		 */
366 		fstate->nst_wscale = 0;
367 		(void)npf_fetch_tcpopts(npc, NULL, &fstate->nst_wscale);
368 
369 		tstate->nst_wscale = 0;
370 
371 		/* Done. */
372 		return true;
373 	}
374 
375 	if (fstate->nst_end == 0) {
376 		/*
377 		 * Should be a SYN-ACK reply to SYN.  If SYN is not set,
378 		 * then we are in the middle of connection and lost tracking.
379 		 */
380 		fstate->nst_end = end;
381 		fstate->nst_maxend = end + 1;
382 		fstate->nst_maxwin = win;
383 		fstate->nst_wscale = 0;
384 
385 		/* Handle TCP Window Scaling (must be ignored if no SYN). */
386 		if (tcpfl & TH_SYN) {
387 			(void)npf_fetch_tcpopts(npc, NULL, &fstate->nst_wscale);
388 		}
389 	}
390 
391 	if ((tcpfl & TH_ACK) == 0) {
392 		/* Pretend that an ACK was sent. */
393 		ack = tstate->nst_end;
394 	} else if ((tcpfl & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST) && ack == 0) {
395 		/* Workaround for some TCP stacks. */
396 		ack = tstate->nst_end;
397 	}
398 
399 	if (__predict_false(tcpfl & TH_RST)) {
400 		/* RST to the initial SYN may have zero SEQ - fix it up. */
401 		if (seq == 0 && nst->nst_state == NPF_TCPS_SYN_SENT) {
402 			end = fstate->nst_end;
403 			seq = end;
404 		}
405 
406 		/* Strict in-order sequence for RST packets (RFC 5961). */
407 		if (npf_strict_order_rst && (fstate->nst_end - seq) > 1) {
408 			return false;
409 		}
410 	}
411 
412 	/*
413 	 * Determine whether the data is within previously noted window,
414 	 * that is, upper boundary for valid data (I).
415 	 */
416 	if (!SEQ_LEQ(end, fstate->nst_maxend)) {
417 		npf_stats_inc(npc->npc_ctx, NPF_STAT_INVALID_STATE_TCP1);
418 		return false;
419 	}
420 
421 	/* Lower boundary (II), which is no more than one window back. */
422 	if (!SEQ_GEQ(seq, fstate->nst_end - tstate->nst_maxwin)) {
423 		npf_stats_inc(npc->npc_ctx, NPF_STAT_INVALID_STATE_TCP2);
424 		return false;
425 	}
426 
427 	/*
428 	 * Boundaries for valid acknowledgments (III, IV) - one predicted
429 	 * window up or down, since packets may be fragmented.
430 	 */
431 	ackskew = tstate->nst_end - ack;
432 	if (ackskew < -NPF_TCP_MAXACKWIN ||
433 	    ackskew > (NPF_TCP_MAXACKWIN << fstate->nst_wscale)) {
434 		npf_stats_inc(npc->npc_ctx, NPF_STAT_INVALID_STATE_TCP3);
435 		return false;
436 	}
437 
438 	/*
439 	 * Packet has been passed.
440 	 *
441 	 * Negative ackskew might be due to fragmented packets.  Since the
442 	 * total length of the packet is unknown - bump the boundary.
443 	 */
444 
445 	if (ackskew < 0) {
446 		tstate->nst_end = ack;
447 	}
448 	/* Keep track of the maximum window seen. */
449 	if (fstate->nst_maxwin < win) {
450 		fstate->nst_maxwin = win;
451 	}
452 	if (SEQ_GT(end, fstate->nst_end)) {
453 		fstate->nst_end = end;
454 	}
455 	/* Note the window for upper boundary. */
456 	if (SEQ_GEQ(ack + win, tstate->nst_maxend)) {
457 		tstate->nst_maxend = ack + win;
458 	}
459 	return true;
460 }
461 
462 /*
463  * npf_state_tcp: inspect TCP segment, determine whether it belongs to
464  * the connection and track its state.
465  */
466 bool
467 npf_state_tcp(npf_cache_t *npc, npf_state_t *nst, int di)
468 {
469 	const struct tcphdr * const th = npc->npc_l4.tcp;
470 	const u_int tcpfl = th->th_flags, state = nst->nst_state;
471 	u_int nstate;
472 
473 	KASSERT(nst->nst_state < NPF_TCP_NSTATES);
474 
475 	/* Look for a transition to a new state. */
476 	if (__predict_true((tcpfl & TH_RST) == 0)) {
477 		const u_int flagcase = npf_tcpfl2case(tcpfl);
478 		nstate = npf_tcp_fsm[state][di][flagcase];
479 	} else if (state == NPF_TCPS_TIME_WAIT) {
480 		/* Prevent TIME-WAIT assassination (RFC 1337). */
481 		nstate = NPF_TCPS_OK;
482 	} else {
483 		nstate = NPF_TCPS_CLOSED;
484 	}
485 
486 	/* Determine whether TCP packet really belongs to this connection. */
487 	if (!npf_tcp_inwindow(npc, nst, di)) {
488 		return false;
489 	}
490 	if (__predict_true(nstate == NPF_TCPS_OK)) {
491 		return true;
492 	}
493 
494 	nst->nst_state = nstate;
495 	return true;
496 }
497 
498 int
499 npf_state_tcp_timeout(const npf_state_t *nst)
500 {
501 	const u_int state = nst->nst_state;
502 
503 	KASSERT(state < NPF_TCP_NSTATES);
504 	return npf_tcp_timeouts[state];
505 }
506