net/npf/npf_handler.c

/*-
 * Copyright (c) 2020 Mindaugas Rasiukevicius <rmind at noxt eu>
 * Copyright (c) 2009-2025 The NetBSD Foundation, Inc.
 * All rights reserved.
 *
 * This material is based upon work partially supported by The
 * NetBSD Foundation under a contract with Mindaugas Rasiukevicius.
 *
 * Redistribution and use in source and binary forms, with or without
 * modification, are permitted provided that the following conditions
 * are met:
 * 1. Redistributions of source code must retain the above copyright
 *    notice, this list of conditions and the following disclaimer.
 * 2. Redistributions in binary form must reproduce the above copyright
 *    notice, this list of conditions and the following disclaimer in the
 *    documentation and/or other materials provided with the distribution.
 *
 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
 * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
 * POSSIBILITY OF SUCH DAMAGE.
 */

/*
 * NPF packet handler.
 *
 * This is the main entry point to the NPF where packet processing happens.
 * There are some important synchronization rules:
 *
 *	1) Lookups into the connection database and configuration (ruleset,
 *	tables, etc) are protected by Epoch-Based Reclamation (EBR);
 *
 *	2) The code in the critical path (protected by EBR) should generally
 *	not block (that includes adaptive mutex acquisitions);
 *
 *	3) Where it will blocks, references should be acquired atomically,
 *	while in the critical path, on the relevant objects.
 */

#ifdef _KERNEL
#include <sys/cdefs.h>
__KERNEL_RCSID(0, "$NetBSD: npf_handler.c,v 1.53 2025/07/01 18:42:37 joe Exp $");

#include <sys/types.h>
#include <sys/param.h>

#include <sys/mbuf.h>
#include <sys/mutex.h>
#include <net/if.h>
#include <net/pfil.h>
#include <sys/socketvar.h>

#include <netinet/in_systm.h>
#include <netinet/in.h>
#include <netinet/ip_var.h>
#include <netinet/ip6.h>
#include <netinet6/ip6_var.h>
#endif

#include "npf_impl.h"
#include "npf_conn.h"

#if defined(_NPF_STANDALONE)
#define	m_freem(m)		npf->mbufops->free(m)
#define	m_clear_flag(m,f)
#else
#define	m_clear_flag(m,f)	(m)->m_flags &= ~(f)
#endif

#ifndef INET6
#define ip6_reass_packet(x, y)	ENOTSUP
#endif

static int
npf_reassembly(npf_t *npf, npf_cache_t *npc, bool *mff)
{
	nbuf_t *nbuf = npc->npc_nbuf;
	int error = EINVAL;
	struct mbuf *m;

	*mff = false;
	m = nbuf_head_mbuf(nbuf);

	if (npf_iscached(npc, NPC_IP4) && npf->ip4_reassembly) {
		error = ip_reass_packet(&m);
	} else if (npf_iscached(npc, NPC_IP6) && npf->ip6_reassembly) {
		error = ip6_reass_packet(&m, npc->npc_hlen);
	} else {
		/*
		 * Reassembly is disabled: just pass the packet through
		 * the ruleset for inspection.
		 */
		return 0;
	}

	if (error) {
		/* Reassembly failed; free the mbuf, clear the nbuf. */
		npf_stats_inc(npf, NPF_STAT_REASSFAIL);
		m_freem(m);
		memset(nbuf, 0, sizeof(nbuf_t));
		return error;
	}
	if (m == NULL) {
		/* More fragments should come. */
		npf_stats_inc(npf, NPF_STAT_FRAGMENTS);
		*mff = true;
		return 0;
	}

	/*
	 * Reassembly is complete, we have the final packet.
	 * Cache again, since layer 4 data is accessible now.
	 */
	nbuf_init(npf, nbuf, m, nbuf->nb_ifp);
	npc->npc_info = 0;

	if (npf_cache_all(npc) & (NPC_IPFRAG|NPC_FMTERR)) {
		return EINVAL;
	}
	npf_stats_inc(npf, NPF_STAT_REASSEMBLY);
	return 0;
}

static inline bool
npf_packet_bypass_tag_p(nbuf_t *nbuf)
{
	uint32_t ntag;
	return nbuf_find_tag(nbuf, &ntag) == 0 && (ntag & NPF_NTAG_PASS) != 0;
}

/*
 * npfk_packet_handler: main packet handling routine for layer 3.
 *
 * Note: packet flow and inspection logic is in strict order.
 */
__dso_public int
npfk_packet_handler(npf_t *npf, struct mbuf **mp, ifnet_t *ifp, int di)
{
	nbuf_t nbuf;
	npf_cache_t npc;
	npf_conn_t *con;
	npf_rule_t *rl;
	npf_rproc_t *rp;
	int error, decision, flags, id_match;
	npf_match_info_t mi;
	bool mff;

	KASSERT(ifp != NULL);

	/*
	 * Initialize packet information cache.
	 * Note: it is enough to clear the info bits.
	 */
	nbuf_init(npf, &nbuf, *mp, ifp);
	memset(&npc, 0, sizeof(npf_cache_t));
	npc.npc_ctx = npf;
	npc.npc_nbuf = &nbuf;

	mi.mi_di = di;
	mi.mi_rid = 0;
	mi.mi_retfl = 0;

	*mp = NULL;
	decision = NPF_DECISION_BLOCK;
	error = 0;
	rp = NULL;
	con = NULL;

	/* Cache everything. */
	flags = npf_cache_all(&npc);

	/* Malformed packet, leave quickly. */
	if (flags & NPC_FMTERR) {
		error = EINVAL;
		goto out;
	}

	/* Determine whether it is an IP fragment. */
	if (__predict_false(flags & NPC_IPFRAG)) {
		/* Pass to IPv4/IPv6 reassembly mechanism. */
		error = npf_reassembly(npf, &npc, &mff);
		if (error) {
			goto out;
		}
		if (mff) {
			/* More fragments should come. */
			return 0;
		}
	}

	/* Just pass-through if specially tagged. */
	if (npf_packet_bypass_tag_p(&nbuf)) {
		goto pass;
	}

	/* Inspect the list of connections (if found, acquires a reference). */
	con = npf_conn_inspect(&npc, di, &error);

	/* If "passing" connection found - skip the ruleset inspection. */
	if (con && npf_conn_pass(con, &mi, &rp)) {
		npf_stats_inc(npf, NPF_STAT_PASS_CONN);
		KASSERT(error == 0);
		goto pass;
	}
	if (__predict_false(error)) {
		if (error == ENETUNREACH)
			goto block;
		goto out;
	}

	/* Acquire the lock, inspect the ruleset using this packet. */
	int slock = npf_config_read_enter(npf);
	npf_ruleset_t *rlset = npf_config_ruleset(npf);

	rl = npf_ruleset_inspect(&npc, rlset, di, NPF_RULE_LAYER_3);
	if (__predict_false(rl == NULL)) {
		const bool pass = npf_default_pass(npf);
		npf_config_read_exit(npf, slock);

		if (pass) {
			npf_stats_inc(npf, NPF_STAT_PASS_DEFAULT);
			goto pass;
		}
		npf_stats_inc(npf, NPF_STAT_BLOCK_DEFAULT);
		goto block;
	}

	/*
	 * Get the rule procedure (acquires a reference) for association
	 * with a connection (if any) and execution.
	 */
	KASSERT(rp == NULL);
	rp = npf_rule_getrproc(rl);

	/* check for matching process uid/gid before concluding */
	id_match = npf_rule_match_rid(rl, &npc, di);

	/* Conclude with the rule and release the lock. */
	error = npf_rule_conclude(rl, &mi);
	npf_config_read_exit(npf, slock);

	/* reverse between pass and block conditions */
	if (id_match != -1 && !id_match) {
		error = npf_rule_reverse(&npc, &mi, error);
	}

	/* reject packets whose addr-port pair matches no sockets  */
	if (id_match == ENOTCONN || error) {
		npf_stats_inc(npf, NPF_STAT_BLOCK_RULESET);
		goto block;
	}
	npf_stats_inc(npf, NPF_STAT_PASS_RULESET);

	/*
	 * Establish a "pass" connection, if required.  Just proceed if
	 * connection creation fails (e.g. due to unsupported protocol).
	 */
	if ((mi.mi_retfl & NPF_RULE_STATEFUL) != 0 && !con) {
		con = npf_conn_establish(&npc, di,
		    (mi.mi_retfl & NPF_RULE_GSTATEFUL) == 0);
		if (con) {
			/*
			 * Note: the reference on the rule procedure is
			 * transferred to the connection.  It will be
			 * released on connection destruction.
			 */
			npf_conn_setpass(con, &mi, rp);
		}
	}

pass:
	decision = NPF_DECISION_PASS;
	KASSERT(error == 0);

	/*
	 * Perform NAT.
	 */
	error = npf_do_nat(&npc, con, di);

block:
	/*
	 * Execute the rule procedure, if any is associated.
	 * It may reverse the decision from pass to block.
	 */
	if (rp && !npf_rproc_run(&npc, rp, &mi, &decision)) {
		if (con) {
			npf_conn_release(con);
		}
		npf_rproc_release(rp);
		/* mbuf already freed */
		return 0;
	}

out:
	/*
	 * Release the reference on a connection.  Release the reference
	 * on a rule procedure only if there was no association.
	 */
	if (con) {
		npf_conn_release(con);
	} else if (rp) {
		npf_rproc_release(rp);
	}

	/* Get the new mbuf pointer. */
	if ((*mp = nbuf_head_mbuf(&nbuf)) == NULL) {
		return error ? error : ENOMEM;
	}

	/* Pass the packet if decided and there is no error. */
	if (decision == NPF_DECISION_PASS && !error) {
		/*
		 * XXX: Disable for now, it will be set accordingly later,
		 * for optimisations (to reduce inspection).
		 */
		m_clear_flag(*mp, M_CANFASTFWD);
		return 0;
	}

	/*
	 * Block the packet.  ENETUNREACH is used to indicate blocking.
	 * Depending on the flags and protocol, return TCP reset (RST) or
	 * ICMP destination unreachable.
	 */
	if (mi.mi_retfl && npf_return_block(&npc, mi.mi_retfl)) {
		*mp = NULL;
	}

	if (!error) {
		error = ENETUNREACH;
	}

	/* Free the mbuf chain. */
	m_freem(*mp);
	*mp = NULL;
	return error;
}

__dso_public int
npfk_layer2_handler(npf_t *npf, struct mbuf **mp, ifnet_t *ifp, int di)
{
	nbuf_t nbuf;
	npf_cache_t npc;
	npf_rule_t *rl;
	int error, decision, flags;
	npf_match_info_t mi;

	KASSERT(ifp != NULL);

	/*
	 * as usual, get packet info
	 * including the interface the frame is traveling on
	 */
	nbuf_init(npf, &nbuf, *mp, ifp);
	memset(&npc, 0, sizeof(npc));
	npc.npc_ctx = npf;
	npc.npc_nbuf = &nbuf;

	mi.mi_di = di;
	mi.mi_rid = 0;
	mi.mi_retfl = 0;

	*mp = NULL;
	decision = NPF_DECISION_BLOCK;
	error = 0;

	/* Cache only ether header. */
	flags = npf_cache_ether(&npc);

	/* Malformed packet, leave quickly. */
	if (flags & NPC_FMTERR) {
		error = EINVAL;
		goto out;
	}

	/* Just pass-through if specially tagged. */
	if (npf_packet_bypass_tag_p(&nbuf)) {
		goto pass;
	}

	/* Acquire the lock, inspect the ruleset using this packet. */
	int slock = npf_config_read_enter(npf);
	npf_ruleset_t *rlset = npf_config_ruleset(npf);

	rl = npf_ruleset_inspect(&npc, rlset, di, NPF_RULE_LAYER_2);
	if (__predict_false(rl == NULL)) {
		const bool pass = npf_default_pass(npf);
		npf_config_read_exit(npf, slock);

		if (pass) {
			npf_stats_inc(npf, NPF_STAT_PASS_DEFAULT);
			goto pass;
		}
		npf_stats_inc(npf, NPF_STAT_BLOCK_DEFAULT);
		goto out;
	}

	/* Conclude with the rule and release the lock. */
	error = npf_rule_conclude(rl, &mi);
	npf_config_read_exit(npf, slock);

	if (error) {
		npf_stats_inc(npf, NPF_ETHER_STAT_BLOCK);
		goto out;
	}
	npf_stats_inc(npf, NPF_ETHER_STAT_PASS);

pass:
	decision = NPF_DECISION_PASS;
	KASSERT(error == 0);

out:

	/* Get the new mbuf pointer. */
	if ((*mp = nbuf_head_mbuf(&nbuf)) == NULL) {
		return error ? error : ENOMEM;
	}

	/* Pass the packet if decided and there is no error. */
	if (decision == NPF_DECISION_PASS && !error) {
		return 0;
	}

	if (!error) {
		error = ENETUNREACH;
	}

	if (*mp) {
		/* Free the mbuf chain. */
		m_freem(*mp);
		*mp = NULL;
	}
	return error;
}