xref: /netbsd-src/sys/net/npf/npf_handler.c (revision 796c32c94f6e154afc9de0f63da35c91bb739b45)
1 /*	$NetBSD: npf_handler.c,v 1.37 2017/02/19 20:27:22 christos Exp $	*/
2 
3 /*-
4  * Copyright (c) 2009-2013 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 packet handler.
34  *
35  * Note: pfil(9) hooks are currently locked by softnet_lock and kernel-lock.
36  */
37 
38 #ifdef _KERNEL
39 #include <sys/cdefs.h>
40 __KERNEL_RCSID(0, "$NetBSD: npf_handler.c,v 1.37 2017/02/19 20:27:22 christos Exp $");
41 
42 #include <sys/types.h>
43 #include <sys/param.h>
44 
45 #include <sys/mbuf.h>
46 #include <sys/mutex.h>
47 #include <net/if.h>
48 #include <net/pfil.h>
49 #include <sys/socketvar.h>
50 
51 #include <netinet/in_systm.h>
52 #include <netinet/in.h>
53 #include <netinet/ip_var.h>
54 #include <netinet/ip6.h>
55 #include <netinet6/ip6_var.h>
56 #endif
57 
58 #include "npf_impl.h"
59 #include "npf_conn.h"
60 
61 #if defined(_NPF_STANDALONE)
62 #define	m_freem(m)		npf->mbufops->free(m)
63 #define	m_clear_flag(m,f)
64 #else
65 #define	m_clear_flag(m,f)	(m)->m_flags &= ~(f)
66 #endif
67 
68 #ifndef INET6
69 #define ip6_reass_packet(x, y)	ENOTSUP
70 #endif
71 
72 static int
73 npf_reassembly(npf_t *npf, npf_cache_t *npc, struct mbuf **mp)
74 {
75 	nbuf_t *nbuf = npc->npc_nbuf;
76 	int error = EINVAL;
77 
78 	/* Reset the mbuf as it may have changed. */
79 	*mp = nbuf_head_mbuf(nbuf);
80 	nbuf_reset(nbuf);
81 
82 	if (npf_iscached(npc, NPC_IP4)) {
83 		struct ip *ip = nbuf_dataptr(nbuf);
84 		error = ip_reass_packet(mp, ip);
85 	} else if (npf_iscached(npc, NPC_IP6)) {
86 		/*
87 		 * Note: ip6_reass_packet() offset is the start of
88 		 * the fragment header.
89 		 */
90 		error = ip6_reass_packet(mp, npc->npc_hlen);
91 		if (error && *mp == NULL) {
92 			memset(nbuf, 0, sizeof(nbuf_t));
93 		}
94 	}
95 	if (error) {
96 		npf_stats_inc(npf, NPF_STAT_REASSFAIL);
97 		return error;
98 	}
99 	if (*mp == NULL) {
100 		/* More fragments should come. */
101 		npf_stats_inc(npf, NPF_STAT_FRAGMENTS);
102 		return 0;
103 	}
104 
105 	/*
106 	 * Reassembly is complete, we have the final packet.
107 	 * Cache again, since layer 4 data is accessible now.
108 	 */
109 	nbuf_init(npf, nbuf, *mp, nbuf->nb_ifp);
110 	npc->npc_info = 0;
111 
112 	if (npf_cache_all(npc) & NPC_IPFRAG) {
113 		return EINVAL;
114 	}
115 	npf_stats_inc(npf, NPF_STAT_REASSEMBLY);
116 	return 0;
117 }
118 
119 /*
120  * npf_packet_handler: main packet handling routine for layer 3.
121  *
122  * Note: packet flow and inspection logic is in strict order.
123  */
124 __dso_public int
125 npf_packet_handler(npf_t *npf, struct mbuf **mp, ifnet_t *ifp, int di)
126 {
127 	nbuf_t nbuf;
128 	npf_cache_t npc;
129 	npf_conn_t *con;
130 	npf_rule_t *rl;
131 	npf_rproc_t *rp;
132 	int error, decision, flags;
133 	uint32_t ntag;
134 	npf_match_info_t mi;
135 
136 	/* QSBR checkpoint. */
137 	pserialize_checkpoint(npf->qsbr);
138 	KASSERT(ifp != NULL);
139 
140 	/*
141 	 * Initialise packet information cache.
142 	 * Note: it is enough to clear the info bits.
143 	 */
144 	npc.npc_ctx = npf;
145 	nbuf_init(npf, &nbuf, *mp, ifp);
146 	npc.npc_nbuf = &nbuf;
147 	npc.npc_info = 0;
148 
149 	mi.mi_di = di;
150 	mi.mi_rid = 0;
151 	mi.mi_retfl = 0;
152 
153 	decision = NPF_DECISION_BLOCK;
154 	error = 0;
155 	rp = NULL;
156 
157 	/* Cache everything.  Determine whether it is an IP fragment. */
158 	flags = npf_cache_all(&npc);
159 	if (__predict_false(flags & NPC_IPFRAG)) {
160 		/*
161 		 * We pass IPv6 fragments unconditionally
162 		 * The first IPv6 fragment is not marked as such
163 		 * and passes through the filter
164 		 */
165 		if (flags & NPC_IP6)
166 			return 0;
167 		/*
168 		 * Pass to IPv4 reassembly mechanism.
169 		 */
170 		error = npf_reassembly(npf, &npc, mp);
171 		if (error) {
172 			con = NULL;
173 			goto out;
174 		}
175 		if (*mp == NULL) {
176 			/* More fragments should come; return. */
177 			return 0;
178 		}
179 	}
180 
181 	/* Just pass-through if specially tagged. */
182 	if (nbuf_find_tag(&nbuf, &ntag) == 0 && (ntag & NPF_NTAG_PASS) != 0) {
183 		con = NULL;
184 		goto pass;
185 	}
186 
187 	/* Inspect the list of connections (if found, acquires a reference). */
188 	con = npf_conn_inspect(&npc, di, &error);
189 
190 	/* If "passing" connection found - skip the ruleset inspection. */
191 	if (con && npf_conn_pass(con, &mi, &rp)) {
192 		npf_stats_inc(npf, NPF_STAT_PASS_CONN);
193 		KASSERT(error == 0);
194 		goto pass;
195 	}
196 	if (__predict_false(error)) {
197 		if (error == ENETUNREACH)
198 			goto block;
199 		goto out;
200 	}
201 
202 	/* Acquire the lock, inspect the ruleset using this packet. */
203 	int slock = npf_config_read_enter();
204 	npf_ruleset_t *rlset = npf_config_ruleset(npf);
205 
206 	rl = npf_ruleset_inspect(&npc, rlset, di, NPF_LAYER_3);
207 	if (__predict_false(rl == NULL)) {
208 		const bool pass = npf_default_pass(npf);
209 		npf_config_read_exit(slock);
210 
211 		if (pass) {
212 			npf_stats_inc(npf, NPF_STAT_PASS_DEFAULT);
213 			goto pass;
214 		}
215 		npf_stats_inc(npf, NPF_STAT_BLOCK_DEFAULT);
216 		goto block;
217 	}
218 
219 	/*
220 	 * Get the rule procedure (acquires a reference) for association
221 	 * with a connection (if any) and execution.
222 	 */
223 	KASSERT(rp == NULL);
224 	rp = npf_rule_getrproc(rl);
225 
226 	/* Conclude with the rule and release the lock. */
227 	error = npf_rule_conclude(rl, &mi);
228 	npf_config_read_exit(slock);
229 
230 	if (error) {
231 		npf_stats_inc(npf, NPF_STAT_BLOCK_RULESET);
232 		goto block;
233 	}
234 	npf_stats_inc(npf, NPF_STAT_PASS_RULESET);
235 
236 	/*
237 	 * Establish a "pass" connection, if required.  Just proceed if
238 	 * connection creation fails (e.g. due to unsupported protocol).
239 	 */
240 	if ((mi.mi_retfl & NPF_RULE_STATEFUL) != 0 && !con) {
241 		con = npf_conn_establish(&npc, di,
242 		    (mi.mi_retfl & NPF_RULE_MULTIENDS) == 0);
243 		if (con) {
244 			/*
245 			 * Note: the reference on the rule procedure is
246 			 * transfered to the connection.  It will be
247 			 * released on connection destruction.
248 			 */
249 			npf_conn_setpass(con, &mi, rp);
250 		}
251 	}
252 pass:
253 	decision = NPF_DECISION_PASS;
254 	KASSERT(error == 0);
255 	/*
256 	 * Perform NAT.
257 	 */
258 	error = npf_do_nat(&npc, con, di);
259 block:
260 	/*
261 	 * Execute the rule procedure, if any is associated.
262 	 * It may reverse the decision from pass to block.
263 	 */
264 	if (rp && !npf_rproc_run(&npc, rp, &mi, &decision)) {
265 		if (con) {
266 			npf_conn_release(con);
267 		}
268 		npf_rproc_release(rp);
269 		*mp = NULL;
270 		return 0;
271 	}
272 out:
273 	/*
274 	 * Release the reference on a connection.  Release the reference
275 	 * on a rule procedure only if there was no association.
276 	 */
277 	if (con) {
278 		npf_conn_release(con);
279 	} else if (rp) {
280 		npf_rproc_release(rp);
281 	}
282 
283 	/* Reset mbuf pointer before returning to the caller. */
284 	if ((*mp = nbuf_head_mbuf(&nbuf)) == NULL) {
285 		return error ? error : ENOMEM;
286 	}
287 
288 	/* Pass the packet if decided and there is no error. */
289 	if (decision == NPF_DECISION_PASS && !error) {
290 		/*
291 		 * XXX: Disable for now, it will be set accordingly later,
292 		 * for optimisations (to reduce inspection).
293 		 */
294 		m_clear_flag(*mp, M_CANFASTFWD);
295 		return 0;
296 	}
297 
298 	/*
299 	 * Block the packet.  ENETUNREACH is used to indicate blocking.
300 	 * Depending on the flags and protocol, return TCP reset (RST) or
301 	 * ICMP destination unreachable.
302 	 */
303 	if (mi.mi_retfl && npf_return_block(&npc, mi.mi_retfl)) {
304 		*mp = NULL;
305 	}
306 
307 	if (!error) {
308 		error = ENETUNREACH;
309 	}
310 
311 	if (*mp) {
312 		/* Free the mbuf chain. */
313 		m_freem(*mp);
314 		*mp = NULL;
315 	}
316 	return error;
317 }
318