xref: /netbsd-src/sys/net/npf/npf_handler.c (revision bdc22b2e01993381dcefeff2bc9b56ca75a4235c)
1 /*	$NetBSD: npf_handler.c,v 1.44 2018/07/10 16:49:09 maxv 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 1.44 2018/07/10 16:49:09 maxv 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, bool *mff)
74 {
75 	nbuf_t *nbuf = npc->npc_nbuf;
76 	int error = EINVAL;
77 	struct mbuf *m;
78 
79 	*mff = false;
80 	m = nbuf_head_mbuf(nbuf);
81 
82 	if (npf_iscached(npc, NPC_IP4)) {
83 		error = ip_reass_packet(&m);
84 	} else if (npf_iscached(npc, NPC_IP6)) {
85 		error = ip6_reass_packet(&m, npc->npc_hlen);
86 	}
87 
88 	if (error) {
89 		/* Reass failed. Free the mbuf, clear the nbuf. */
90 		npf_stats_inc(npf, NPF_STAT_REASSFAIL);
91 		m_freem(m);
92 		memset(nbuf, 0, sizeof(nbuf_t));
93 		return error;
94 	}
95 	if (m == NULL) {
96 		/* More fragments should come. */
97 		npf_stats_inc(npf, NPF_STAT_FRAGMENTS);
98 		*mff = true;
99 		return 0;
100 	}
101 
102 	/*
103 	 * Reassembly is complete, we have the final packet.
104 	 * Cache again, since layer 4 data is accessible now.
105 	 */
106 	nbuf_init(npf, nbuf, m, nbuf->nb_ifp);
107 	npc->npc_info = 0;
108 
109 	if (npf_cache_all(npc) & (NPC_IPFRAG|NPC_FMTERR)) {
110 		return EINVAL;
111 	}
112 	npf_stats_inc(npf, NPF_STAT_REASSEMBLY);
113 	return 0;
114 }
115 
116 /*
117  * npf_packet_handler: main packet handling routine for layer 3.
118  *
119  * Note: packet flow and inspection logic is in strict order.
120  */
121 __dso_public int
122 npf_packet_handler(npf_t *npf, struct mbuf **mp, ifnet_t *ifp, int di)
123 {
124 	nbuf_t nbuf;
125 	npf_cache_t npc;
126 	npf_conn_t *con;
127 	npf_rule_t *rl;
128 	npf_rproc_t *rp;
129 	int error, decision, flags;
130 	uint32_t ntag;
131 	npf_match_info_t mi;
132 	bool mff;
133 
134 	/* QSBR checkpoint. */
135 	pserialize_checkpoint(npf->qsbr);
136 	KASSERT(ifp != NULL);
137 
138 	/*
139 	 * Initialise packet information cache.
140 	 * Note: it is enough to clear the info bits.
141 	 */
142 	npc.npc_ctx = npf;
143 	nbuf_init(npf, &nbuf, *mp, ifp);
144 	npc.npc_nbuf = &nbuf;
145 	npc.npc_info = 0;
146 
147 	mi.mi_di = di;
148 	mi.mi_rid = 0;
149 	mi.mi_retfl = 0;
150 
151 	*mp = NULL;
152 	decision = NPF_DECISION_BLOCK;
153 	error = 0;
154 	rp = NULL;
155 	con = NULL;
156 
157 	/* Cache everything. */
158 	flags = npf_cache_all(&npc);
159 
160 	/* If error on the format, leave quickly. */
161 	if (flags & NPC_FMTERR) {
162 		error = EINVAL;
163 		goto out;
164 	}
165 
166 	/* Determine whether it is an IP fragment. */
167 	if (__predict_false(flags & NPC_IPFRAG)) {
168 		/* Pass to IPv4/IPv6 reassembly mechanism. */
169 		error = npf_reassembly(npf, &npc, &mff);
170 		if (error) {
171 			goto out;
172 		}
173 		if (mff) {
174 			/* More fragments should come. */
175 			return 0;
176 		}
177 	}
178 
179 	/* Just pass-through if specially tagged. */
180 	if (nbuf_find_tag(&nbuf, &ntag) == 0 && (ntag & NPF_NTAG_PASS) != 0) {
181 		goto pass;
182 	}
183 
184 	/* Inspect the list of connections (if found, acquires a reference). */
185 	con = npf_conn_inspect(&npc, di, &error);
186 
187 	/* If "passing" connection found - skip the ruleset inspection. */
188 	if (con && npf_conn_pass(con, &mi, &rp)) {
189 		npf_stats_inc(npf, NPF_STAT_PASS_CONN);
190 		KASSERT(error == 0);
191 		goto pass;
192 	}
193 	if (__predict_false(error)) {
194 		if (error == ENETUNREACH)
195 			goto block;
196 		goto out;
197 	}
198 
199 	/* Acquire the lock, inspect the ruleset using this packet. */
200 	int slock = npf_config_read_enter();
201 	npf_ruleset_t *rlset = npf_config_ruleset(npf);
202 
203 	rl = npf_ruleset_inspect(&npc, rlset, di, NPF_LAYER_3);
204 	if (__predict_false(rl == NULL)) {
205 		const bool pass = npf_default_pass(npf);
206 		npf_config_read_exit(slock);
207 
208 		if (pass) {
209 			npf_stats_inc(npf, NPF_STAT_PASS_DEFAULT);
210 			goto pass;
211 		}
212 		npf_stats_inc(npf, NPF_STAT_BLOCK_DEFAULT);
213 		goto block;
214 	}
215 
216 	/*
217 	 * Get the rule procedure (acquires a reference) for association
218 	 * with a connection (if any) and execution.
219 	 */
220 	KASSERT(rp == NULL);
221 	rp = npf_rule_getrproc(rl);
222 
223 	/* Conclude with the rule and release the lock. */
224 	error = npf_rule_conclude(rl, &mi);
225 	npf_config_read_exit(slock);
226 
227 	if (error) {
228 		npf_stats_inc(npf, NPF_STAT_BLOCK_RULESET);
229 		goto block;
230 	}
231 	npf_stats_inc(npf, NPF_STAT_PASS_RULESET);
232 
233 	/*
234 	 * Establish a "pass" connection, if required.  Just proceed if
235 	 * connection creation fails (e.g. due to unsupported protocol).
236 	 */
237 	if ((mi.mi_retfl & NPF_RULE_STATEFUL) != 0 && !con) {
238 		con = npf_conn_establish(&npc, di,
239 		    (mi.mi_retfl & NPF_RULE_MULTIENDS) == 0);
240 		if (con) {
241 			/*
242 			 * Note: the reference on the rule procedure is
243 			 * transfered to the connection.  It will be
244 			 * released on connection destruction.
245 			 */
246 			npf_conn_setpass(con, &mi, rp);
247 		}
248 	}
249 
250 pass:
251 	decision = NPF_DECISION_PASS;
252 	KASSERT(error == 0);
253 	/*
254 	 * Perform NAT.
255 	 */
256 	error = npf_do_nat(&npc, con, di);
257 
258 block:
259 	/*
260 	 * Execute the rule procedure, if any is associated.
261 	 * It may reverse the decision from pass to block.
262 	 */
263 	if (rp && !npf_rproc_run(&npc, rp, &mi, &decision)) {
264 		if (con) {
265 			npf_conn_release(con);
266 		}
267 		npf_rproc_release(rp);
268 		/* mbuf already freed */
269 		return 0;
270 	}
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 	/* Get the new mbuf pointer. */
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