Home | History | Annotate | Line # | Download | only in npf
npf_ruleset.c revision 1.46.4.1
      1 /*-
      2  * Copyright (c) 2009-2015 The NetBSD Foundation, Inc.
      3  * All rights reserved.
      4  *
      5  * This material is based upon work partially supported by The
      6  * NetBSD Foundation under a contract with Mindaugas Rasiukevicius.
      7  *
      8  * Redistribution and use in source and binary forms, with or without
      9  * modification, are permitted provided that the following conditions
     10  * are met:
     11  * 1. Redistributions of source code must retain the above copyright
     12  *    notice, this list of conditions and the following disclaimer.
     13  * 2. Redistributions in binary form must reproduce the above copyright
     14  *    notice, this list of conditions and the following disclaimer in the
     15  *    documentation and/or other materials provided with the distribution.
     16  *
     17  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
     18  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
     19  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
     20  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
     21  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
     22  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
     23  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
     24  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
     25  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
     26  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
     27  * POSSIBILITY OF SUCH DAMAGE.
     28  */
     29 
     30 /*
     31  * NPF ruleset module.
     32  */
     33 
     34 #ifdef _KERNEL
     35 #include <sys/cdefs.h>
     36 __KERNEL_RCSID(0, "$NetBSD: npf_ruleset.c,v 1.46.4.1 2019/06/10 22:09:46 christos Exp $");
     37 
     38 #include <sys/param.h>
     39 #include <sys/types.h>
     40 
     41 #include <sys/atomic.h>
     42 #include <sys/kmem.h>
     43 #include <sys/queue.h>
     44 #include <sys/mbuf.h>
     45 #include <sys/types.h>
     46 
     47 #include <net/bpf.h>
     48 #include <net/bpfjit.h>
     49 #include <net/pfil.h>
     50 #include <net/if.h>
     51 #endif
     52 
     53 #include "npf_impl.h"
     54 
     55 struct npf_ruleset {
     56 	/*
     57 	 * - List of all rules.
     58 	 * - Dynamic (i.e. named) rules.
     59 	 * - G/C list for convenience.
     60 	 */
     61 	LIST_HEAD(, npf_rule)	rs_all;
     62 	LIST_HEAD(, npf_rule)	rs_dynamic;
     63 	LIST_HEAD(, npf_rule)	rs_gc;
     64 
     65 	/* Unique ID counter. */
     66 	uint64_t		rs_idcnt;
     67 
     68 	/* Number of array slots and active rules. */
     69 	u_int			rs_slots;
     70 	u_int			rs_nitems;
     71 
     72 	/* Array of ordered rules. */
     73 	npf_rule_t *		rs_rules[];
     74 };
     75 
     76 struct npf_rule {
     77 	/* Attributes, interface and skip slot. */
     78 	uint32_t		r_attr;
     79 	u_int			r_ifid;
     80 	u_int			r_skip_to;
     81 
     82 	/* Code to process, if any. */
     83 	int			r_type;
     84 	bpfjit_func_t		r_jcode;
     85 	void *			r_code;
     86 	u_int			r_clen;
     87 
     88 	/* NAT policy (optional), rule procedure and subset. */
     89 	npf_natpolicy_t *	r_natp;
     90 	npf_rproc_t *		r_rproc;
     91 
     92 	union {
     93 		/*
     94 		 * Dynamic group: rule subset and a group list entry.
     95 		 */
     96 		struct {
     97 			npf_rule_t *		r_subset;
     98 			LIST_ENTRY(npf_rule)	r_dentry;
     99 		};
    100 
    101 		/*
    102 		 * Dynamic rule: priority, parent group and next rule.
    103 		 */
    104 		struct {
    105 			int			r_priority;
    106 			npf_rule_t *		r_parent;
    107 			npf_rule_t *		r_next;
    108 		};
    109 	};
    110 
    111 	/* Rule ID, name and the optional key. */
    112 	uint64_t		r_id;
    113 	char			r_name[NPF_RULE_MAXNAMELEN];
    114 	uint8_t			r_key[NPF_RULE_MAXKEYLEN];
    115 
    116 	/* All-list entry and the auxiliary info. */
    117 	LIST_ENTRY(npf_rule)	r_aentry;
    118 	nvlist_t *		r_info;
    119 	size_t			r_info_len;
    120 };
    121 
    122 #define	SKIPTO_ADJ_FLAG		(1U << 31)
    123 #define	SKIPTO_MASK		(SKIPTO_ADJ_FLAG - 1)
    124 
    125 static nvlist_t *	npf_rule_export(npf_t *, const npf_rule_t *);
    126 
    127 /*
    128  * Private attributes - must be in the NPF_RULE_PRIVMASK range.
    129  */
    130 #define	NPF_RULE_KEEPNAT	(0x01000000 & NPF_RULE_PRIVMASK)
    131 
    132 #define	NPF_DYNAMIC_GROUP_P(attr) \
    133     (((attr) & NPF_DYNAMIC_GROUP) == NPF_DYNAMIC_GROUP)
    134 
    135 #define	NPF_DYNAMIC_RULE_P(attr) \
    136     (((attr) & NPF_DYNAMIC_GROUP) == NPF_RULE_DYNAMIC)
    137 
    138 npf_ruleset_t *
    139 npf_ruleset_create(size_t slots)
    140 {
    141 	size_t len = offsetof(npf_ruleset_t, rs_rules[slots]);
    142 	npf_ruleset_t *rlset;
    143 
    144 	rlset = kmem_zalloc(len, KM_SLEEP);
    145 	LIST_INIT(&rlset->rs_dynamic);
    146 	LIST_INIT(&rlset->rs_all);
    147 	LIST_INIT(&rlset->rs_gc);
    148 	rlset->rs_slots = slots;
    149 
    150 	return rlset;
    151 }
    152 
    153 void
    154 npf_ruleset_destroy(npf_ruleset_t *rlset)
    155 {
    156 	size_t len = offsetof(npf_ruleset_t, rs_rules[rlset->rs_slots]);
    157 	npf_rule_t *rl;
    158 
    159 	while ((rl = LIST_FIRST(&rlset->rs_all)) != NULL) {
    160 		if (NPF_DYNAMIC_GROUP_P(rl->r_attr)) {
    161 			/*
    162 			 * Note: r_subset may point to the rules which
    163 			 * were inherited by a new ruleset.
    164 			 */
    165 			rl->r_subset = NULL;
    166 			LIST_REMOVE(rl, r_dentry);
    167 		}
    168 		if (NPF_DYNAMIC_RULE_P(rl->r_attr)) {
    169 			/* Not removing from r_subset, see above. */
    170 			KASSERT(rl->r_parent != NULL);
    171 		}
    172 		LIST_REMOVE(rl, r_aentry);
    173 		npf_rule_free(rl);
    174 	}
    175 	KASSERT(LIST_EMPTY(&rlset->rs_dynamic));
    176 
    177 	npf_ruleset_gc(rlset);
    178 	KASSERT(LIST_EMPTY(&rlset->rs_gc));
    179 	kmem_free(rlset, len);
    180 }
    181 
    182 /*
    183  * npf_ruleset_insert: insert the rule into the specified ruleset.
    184  */
    185 void
    186 npf_ruleset_insert(npf_ruleset_t *rlset, npf_rule_t *rl)
    187 {
    188 	u_int n = rlset->rs_nitems;
    189 
    190 	KASSERT(n < rlset->rs_slots);
    191 
    192 	LIST_INSERT_HEAD(&rlset->rs_all, rl, r_aentry);
    193 	if (NPF_DYNAMIC_GROUP_P(rl->r_attr)) {
    194 		LIST_INSERT_HEAD(&rlset->rs_dynamic, rl, r_dentry);
    195 	} else {
    196 		KASSERTMSG(rl->r_parent == NULL, "cannot be dynamic rule");
    197 		rl->r_attr &= ~NPF_RULE_DYNAMIC;
    198 	}
    199 
    200 	rlset->rs_rules[n] = rl;
    201 	rlset->rs_nitems++;
    202 	rl->r_id = ++rlset->rs_idcnt;
    203 
    204 	if (rl->r_skip_to < ++n) {
    205 		rl->r_skip_to = SKIPTO_ADJ_FLAG | n;
    206 	}
    207 }
    208 
    209 npf_rule_t *
    210 npf_ruleset_lookup(npf_ruleset_t *rlset, const char *name)
    211 {
    212 	npf_rule_t *rl;
    213 
    214 	LIST_FOREACH(rl, &rlset->rs_dynamic, r_dentry) {
    215 		KASSERT(NPF_DYNAMIC_GROUP_P(rl->r_attr));
    216 		if (strncmp(rl->r_name, name, NPF_RULE_MAXNAMELEN) == 0)
    217 			break;
    218 	}
    219 	return rl;
    220 }
    221 
    222 /*
    223  * npf_ruleset_add: insert dynamic rule into the (active) ruleset.
    224  */
    225 int
    226 npf_ruleset_add(npf_ruleset_t *rlset, const char *rname, npf_rule_t *rl)
    227 {
    228 	npf_rule_t *rg, *it, *target;
    229 	int priocmd;
    230 
    231 	if (!NPF_DYNAMIC_RULE_P(rl->r_attr)) {
    232 		return EINVAL;
    233 	}
    234 	rg = npf_ruleset_lookup(rlset, rname);
    235 	if (rg == NULL) {
    236 		return ESRCH;
    237 	}
    238 
    239 	/* Dynamic rule - assign a unique ID and save the parent. */
    240 	rl->r_id = ++rlset->rs_idcnt;
    241 	rl->r_parent = rg;
    242 
    243 	/*
    244 	 * Rule priority: (highest) 1, 2 ... n (lowest).
    245 	 * Negative priority indicates an operation and is reset to zero.
    246 	 */
    247 	if ((priocmd = rl->r_priority) < 0) {
    248 		rl->r_priority = 0;
    249 	}
    250 
    251 	/*
    252 	 * WARNING: once rg->subset or target->r_next of an *active*
    253 	 * rule is set, then our rule becomes globally visible and active.
    254 	 * Must issue a load fence to ensure rl->r_next visibility first.
    255 	 */
    256 	switch (priocmd) {
    257 	case NPF_PRI_LAST:
    258 	default:
    259 		target = NULL;
    260 		it = rg->r_subset;
    261 		while (it && it->r_priority <= rl->r_priority) {
    262 			target = it;
    263 			it = it->r_next;
    264 		}
    265 		if (target) {
    266 			rl->r_next = target->r_next;
    267 			membar_producer();
    268 			target->r_next = rl;
    269 			break;
    270 		}
    271 		/* FALLTHROUGH */
    272 
    273 	case NPF_PRI_FIRST:
    274 		rl->r_next = rg->r_subset;
    275 		membar_producer();
    276 		rg->r_subset = rl;
    277 		break;
    278 	}
    279 
    280 	/* Finally, add into the all-list. */
    281 	LIST_INSERT_HEAD(&rlset->rs_all, rl, r_aentry);
    282 	return 0;
    283 }
    284 
    285 static void
    286 npf_ruleset_unlink(npf_rule_t *rl, npf_rule_t *prev)
    287 {
    288 	KASSERT(NPF_DYNAMIC_RULE_P(rl->r_attr));
    289 	if (prev) {
    290 		prev->r_next = rl->r_next;
    291 	} else {
    292 		npf_rule_t *rg = rl->r_parent;
    293 		rg->r_subset = rl->r_next;
    294 	}
    295 	LIST_REMOVE(rl, r_aentry);
    296 }
    297 
    298 /*
    299  * npf_ruleset_remove: remove the dynamic rule given the rule ID.
    300  */
    301 int
    302 npf_ruleset_remove(npf_ruleset_t *rlset, const char *rname, uint64_t id)
    303 {
    304 	npf_rule_t *rg, *prev = NULL;
    305 
    306 	if ((rg = npf_ruleset_lookup(rlset, rname)) == NULL) {
    307 		return ESRCH;
    308 	}
    309 	for (npf_rule_t *rl = rg->r_subset; rl; rl = rl->r_next) {
    310 		KASSERT(rl->r_parent == rg);
    311 		KASSERT(NPF_DYNAMIC_RULE_P(rl->r_attr));
    312 
    313 		/* Compare ID.  On match, remove and return. */
    314 		if (rl->r_id == id) {
    315 			npf_ruleset_unlink(rl, prev);
    316 			LIST_INSERT_HEAD(&rlset->rs_gc, rl, r_aentry);
    317 			return 0;
    318 		}
    319 		prev = rl;
    320 	}
    321 	return ENOENT;
    322 }
    323 
    324 /*
    325  * npf_ruleset_remkey: remove the dynamic rule given the rule key.
    326  */
    327 int
    328 npf_ruleset_remkey(npf_ruleset_t *rlset, const char *rname,
    329     const void *key, size_t len)
    330 {
    331 	npf_rule_t *rg, *rlast = NULL, *prev = NULL, *lastprev = NULL;
    332 
    333 	KASSERT(len && len <= NPF_RULE_MAXKEYLEN);
    334 
    335 	if ((rg = npf_ruleset_lookup(rlset, rname)) == NULL) {
    336 		return ESRCH;
    337 	}
    338 
    339 	/* Compare the key and find the last in the list. */
    340 	for (npf_rule_t *rl = rg->r_subset; rl; rl = rl->r_next) {
    341 		KASSERT(rl->r_parent == rg);
    342 		KASSERT(NPF_DYNAMIC_RULE_P(rl->r_attr));
    343 		if (memcmp(rl->r_key, key, len) == 0) {
    344 			lastprev = prev;
    345 			rlast = rl;
    346 		}
    347 		prev = rl;
    348 	}
    349 	if (!rlast) {
    350 		return ENOENT;
    351 	}
    352 	npf_ruleset_unlink(rlast, lastprev);
    353 	LIST_INSERT_HEAD(&rlset->rs_gc, rlast, r_aentry);
    354 	return 0;
    355 }
    356 
    357 /*
    358  * npf_ruleset_list: serialise and return the dynamic rules.
    359  */
    360 nvlist_t *
    361 npf_ruleset_list(npf_t *npf, npf_ruleset_t *rlset, const char *rname)
    362 {
    363 	nvlist_t *rgroup;
    364 	npf_rule_t *rg;
    365 
    366 	KASSERT(npf_config_locked_p(npf));
    367 
    368 	if ((rg = npf_ruleset_lookup(rlset, rname)) == NULL) {
    369 		return NULL;
    370 	}
    371 	if ((rgroup = nvlist_create(0)) == NULL) {
    372 		return NULL;
    373 	}
    374 	for (npf_rule_t *rl = rg->r_subset; rl; rl = rl->r_next) {
    375 		nvlist_t *rule;
    376 
    377 		KASSERT(rl->r_parent == rg);
    378 		KASSERT(NPF_DYNAMIC_RULE_P(rl->r_attr));
    379 
    380 		rule = npf_rule_export(npf, rl);
    381 		if (!rule) {
    382 			nvlist_destroy(rgroup);
    383 			return NULL;
    384 		}
    385 		nvlist_append_nvlist_array(rgroup, "rules", rule);
    386 		nvlist_destroy(rule);
    387 	}
    388 	return rgroup;
    389 }
    390 
    391 /*
    392  * npf_ruleset_flush: flush the dynamic rules in the ruleset by inserting
    393  * them into the G/C list.
    394  */
    395 int
    396 npf_ruleset_flush(npf_ruleset_t *rlset, const char *rname)
    397 {
    398 	npf_rule_t *rg, *rl;
    399 
    400 	if ((rg = npf_ruleset_lookup(rlset, rname)) == NULL) {
    401 		return ESRCH;
    402 	}
    403 
    404 	rl = atomic_swap_ptr(&rg->r_subset, NULL);
    405 	membar_producer();
    406 
    407 	while (rl) {
    408 		KASSERT(NPF_DYNAMIC_RULE_P(rl->r_attr));
    409 		KASSERT(rl->r_parent == rg);
    410 
    411 		LIST_REMOVE(rl, r_aentry);
    412 		LIST_INSERT_HEAD(&rlset->rs_gc, rl, r_aentry);
    413 		rl = rl->r_next;
    414 	}
    415 	rlset->rs_idcnt = 0;
    416 	return 0;
    417 }
    418 
    419 /*
    420  * npf_ruleset_gc: destroy the rules in G/C list.
    421  */
    422 void
    423 npf_ruleset_gc(npf_ruleset_t *rlset)
    424 {
    425 	npf_rule_t *rl;
    426 
    427 	while ((rl = LIST_FIRST(&rlset->rs_gc)) != NULL) {
    428 		LIST_REMOVE(rl, r_aentry);
    429 		npf_rule_free(rl);
    430 	}
    431 }
    432 
    433 /*
    434  * npf_ruleset_export: serialise and return the static rules.
    435  */
    436 int
    437 npf_ruleset_export(npf_t *npf, const npf_ruleset_t *rlset,
    438     const char *key, nvlist_t *npf_dict)
    439 {
    440 	const unsigned nitems = rlset->rs_nitems;
    441 	unsigned n = 0;
    442 	int error = 0;
    443 
    444 	KASSERT(npf_config_locked_p(npf));
    445 
    446 	while (n < nitems) {
    447 		const npf_rule_t *rl = rlset->rs_rules[n];
    448 		const npf_natpolicy_t *natp = rl->r_natp;
    449 		nvlist_t *rule;
    450 
    451 		rule = npf_rule_export(npf, rl);
    452 		if (!rule) {
    453 			error = ENOMEM;
    454 			break;
    455 		}
    456 		if (natp && (error = npf_nat_policyexport(natp, rule)) != 0) {
    457 			nvlist_destroy(rule);
    458 			break;
    459 		}
    460 		nvlist_append_nvlist_array(npf_dict, key, rule);
    461 		nvlist_destroy(rule);
    462 		n++;
    463 	}
    464 	return error;
    465 }
    466 
    467 /*
    468  * npf_ruleset_reload: prepare the new ruleset by scanning the active
    469  * ruleset and: 1) sharing the dynamic rules 2) sharing NAT policies.
    470  *
    471  * => The active (old) ruleset should be exclusively locked.
    472  */
    473 void
    474 npf_ruleset_reload(npf_t *npf, npf_ruleset_t *newset,
    475     npf_ruleset_t *oldset, bool load)
    476 {
    477 	npf_rule_t *rg, *rl;
    478 	uint64_t nid = 0;
    479 
    480 	KASSERT(npf_config_locked_p(npf));
    481 
    482 	/*
    483 	 * Scan the dynamic rules and share (migrate) if needed.
    484 	 */
    485 	LIST_FOREACH(rg, &newset->rs_dynamic, r_dentry) {
    486 		npf_rule_t *active_rgroup;
    487 
    488 		/* Look for a dynamic ruleset group with such name. */
    489 		active_rgroup = npf_ruleset_lookup(oldset, rg->r_name);
    490 		if (active_rgroup == NULL) {
    491 			continue;
    492 		}
    493 
    494 		/*
    495 		 * ATOMICITY: Copy the head pointer of the linked-list,
    496 		 * but do not remove the rules from the active r_subset.
    497 		 * This is necessary because the rules are still active
    498 		 * and therefore are accessible for inspection via the
    499 		 * old ruleset.
    500 		 */
    501 		rg->r_subset = active_rgroup->r_subset;
    502 
    503 		/*
    504 		 * We can safely migrate to the new all-rule list and
    505 		 * reset the parent rule, though.
    506 		 */
    507 		for (rl = rg->r_subset; rl; rl = rl->r_next) {
    508 			KASSERT(NPF_DYNAMIC_RULE_P(rl->r_attr));
    509 			LIST_REMOVE(rl, r_aentry);
    510 			LIST_INSERT_HEAD(&newset->rs_all, rl, r_aentry);
    511 
    512 			KASSERT(rl->r_parent == active_rgroup);
    513 			rl->r_parent = rg;
    514 		}
    515 	}
    516 
    517 	/*
    518 	 * If performing the load of connections then NAT policies may
    519 	 * already have translated connections associated with them and
    520 	 * we should not share or inherit anything.
    521 	 */
    522 	if (load)
    523 		return;
    524 
    525 	/*
    526 	 * Scan all rules in the new ruleset and share NAT policies.
    527 	 * Also, assign a unique ID for each policy here.
    528 	 */
    529 	LIST_FOREACH(rl, &newset->rs_all, r_aentry) {
    530 		npf_natpolicy_t *np;
    531 		npf_rule_t *actrl;
    532 
    533 		/* Does the rule have a NAT policy associated? */
    534 		if ((np = rl->r_natp) == NULL) {
    535 			continue;
    536 		}
    537 
    538 		/*
    539 		 * First, try to share the active port map.  If this
    540 		 * policy will be unused, npf_nat_freepolicy() will
    541 		 * drop the reference.
    542 		 */
    543 		npf_ruleset_sharepm(oldset, np);
    544 
    545 		/* Does it match with any policy in the active ruleset? */
    546 		LIST_FOREACH(actrl, &oldset->rs_all, r_aentry) {
    547 			if (!actrl->r_natp)
    548 				continue;
    549 			if ((actrl->r_attr & NPF_RULE_KEEPNAT) != 0)
    550 				continue;
    551 			if (npf_nat_cmppolicy(actrl->r_natp, np))
    552 				break;
    553 		}
    554 		if (!actrl) {
    555 			/* No: just set the ID and continue. */
    556 			npf_nat_setid(np, ++nid);
    557 			continue;
    558 		}
    559 
    560 		/* Yes: inherit the matching NAT policy. */
    561 		rl->r_natp = actrl->r_natp;
    562 		npf_nat_setid(rl->r_natp, ++nid);
    563 
    564 		/*
    565 		 * Finally, mark the active rule to not destroy its NAT
    566 		 * policy later as we inherited it (but the rule must be
    567 		 * kept active for now).  Destroy the new/unused policy.
    568 		 */
    569 		actrl->r_attr |= NPF_RULE_KEEPNAT;
    570 		npf_nat_freepolicy(np);
    571 	}
    572 
    573 	/* Inherit the ID counter. */
    574 	newset->rs_idcnt = oldset->rs_idcnt;
    575 }
    576 
    577 /*
    578  * npf_ruleset_sharepm: attempt to share the active NAT portmap.
    579  */
    580 npf_rule_t *
    581 npf_ruleset_sharepm(npf_ruleset_t *rlset, npf_natpolicy_t *mnp)
    582 {
    583 	npf_natpolicy_t *np;
    584 	npf_rule_t *rl;
    585 
    586 	/*
    587 	 * Scan the NAT policies in the ruleset and match with the
    588 	 * given policy based on the translation IP address.  If they
    589 	 * match - adjust the given NAT policy to use the active NAT
    590 	 * portmap.  In such case the reference on the old portmap is
    591 	 * dropped and acquired on the active one.
    592 	 */
    593 	LIST_FOREACH(rl, &rlset->rs_all, r_aentry) {
    594 		np = rl->r_natp;
    595 		if (np == NULL || np == mnp)
    596 			continue;
    597 		if (npf_nat_sharepm(np, mnp))
    598 			break;
    599 	}
    600 	return rl;
    601 }
    602 
    603 npf_natpolicy_t *
    604 npf_ruleset_findnat(npf_ruleset_t *rlset, uint64_t id)
    605 {
    606 	npf_rule_t *rl;
    607 
    608 	LIST_FOREACH(rl, &rlset->rs_all, r_aentry) {
    609 		npf_natpolicy_t *np = rl->r_natp;
    610 		if (np && npf_nat_getid(np) == id) {
    611 			return np;
    612 		}
    613 	}
    614 	return NULL;
    615 }
    616 
    617 /*
    618  * npf_ruleset_freealg: inspect the ruleset and disassociate specified
    619  * ALG from all NAT entries using it.
    620  */
    621 void
    622 npf_ruleset_freealg(npf_ruleset_t *rlset, npf_alg_t *alg)
    623 {
    624 	npf_rule_t *rl;
    625 	npf_natpolicy_t *np;
    626 
    627 	LIST_FOREACH(rl, &rlset->rs_all, r_aentry) {
    628 		if ((np = rl->r_natp) != NULL) {
    629 			npf_nat_freealg(np, alg);
    630 		}
    631 	}
    632 }
    633 
    634 /*
    635  * npf_rule_alloc: allocate a rule and initialise it.
    636  */
    637 npf_rule_t *
    638 npf_rule_alloc(npf_t *npf, const nvlist_t *rule)
    639 {
    640 	npf_rule_t *rl;
    641 	const char *rname;
    642 	const void *key, *info;
    643 	size_t len;
    644 
    645 	/* Allocate a rule structure and keep the information. */
    646 	rl = kmem_zalloc(sizeof(npf_rule_t), KM_SLEEP);
    647 	info = dnvlist_get_binary(rule, "info", &rl->r_info_len, NULL, 0);
    648 	if (info) {
    649 		rl->r_info = kmem_alloc(rl->r_info_len, KM_SLEEP);
    650 		memcpy(rl->r_info, info, rl->r_info_len);
    651 	}
    652 	rl->r_natp = NULL;
    653 
    654 	/* Name (optional) */
    655 	if ((rname = dnvlist_get_string(rule, "name", NULL)) != NULL) {
    656 		strlcpy(rl->r_name, rname, NPF_RULE_MAXNAMELEN);
    657 	} else {
    658 		rl->r_name[0] = '\0';
    659 	}
    660 
    661 	/* Attributes, priority and interface ID (optional). */
    662 	rl->r_attr = dnvlist_get_number(rule, "attr", 0);
    663 	rl->r_attr &= ~NPF_RULE_PRIVMASK;
    664 
    665 	if (NPF_DYNAMIC_RULE_P(rl->r_attr)) {
    666 		/* Priority of the dynamic rule. */
    667 		rl->r_priority = dnvlist_get_number(rule, "prio", 0);
    668 	} else {
    669 		/* The skip-to index.  No need to validate it. */
    670 		rl->r_skip_to = dnvlist_get_number(rule, "skip-to", 0);
    671 	}
    672 
    673 	/* Interface name; register and get the npf-if-id. */
    674 	if ((rname = dnvlist_get_string(rule, "ifname", NULL)) != NULL) {
    675 		if ((rl->r_ifid = npf_ifmap_register(npf, rname)) == 0) {
    676 			kmem_free(rl, sizeof(npf_rule_t));
    677 			return NULL;
    678 		}
    679 	} else {
    680 		rl->r_ifid = 0;
    681 	}
    682 
    683 	/* Key (optional). */
    684 	if ((key = dnvlist_get_binary(rule, "key", &len, NULL, 0)) != NULL) {
    685 		if (len > NPF_RULE_MAXKEYLEN) {
    686 			kmem_free(rl, sizeof(npf_rule_t));
    687 			return NULL;
    688 		}
    689 		memcpy(rl->r_key, key, len);
    690 	}
    691 	return rl;
    692 }
    693 
    694 static nvlist_t *
    695 npf_rule_export(npf_t *npf, const npf_rule_t *rl)
    696 {
    697 	nvlist_t *rule = nvlist_create(0);
    698 	unsigned skip_to = 0;
    699 	npf_rproc_t *rp;
    700 
    701 	nvlist_add_number(rule, "attr", rl->r_attr);
    702 	nvlist_add_number(rule, "prio", rl->r_priority);
    703 	if ((rl->r_skip_to & SKIPTO_ADJ_FLAG) == 0) {
    704 		skip_to = rl->r_skip_to & SKIPTO_MASK;
    705 	}
    706 	nvlist_add_number(rule, "skip-to", skip_to);
    707 	nvlist_add_number(rule, "code-type", rl->r_type);
    708 	if (rl->r_code) {
    709 		nvlist_add_binary(rule, "code", rl->r_code, rl->r_clen);
    710 	}
    711 	if (rl->r_ifid) {
    712 		const char *ifname = npf_ifmap_getname(npf, rl->r_ifid);
    713 		nvlist_add_string(rule, "ifname", ifname);
    714 	}
    715 	nvlist_add_number(rule, "id", rl->r_id);
    716 
    717 	if (rl->r_name[0]) {
    718 		nvlist_add_string(rule, "name", rl->r_name);
    719 	}
    720 	if (NPF_DYNAMIC_RULE_P(rl->r_attr)) {
    721 		nvlist_add_binary(rule, "key", rl->r_key, NPF_RULE_MAXKEYLEN);
    722 	}
    723 	if (rl->r_info) {
    724 		nvlist_add_binary(rule, "info", rl->r_info, rl->r_info_len);
    725 	}
    726 	if ((rp = npf_rule_getrproc(rl)) != NULL) {
    727 		const char *rname = npf_rproc_getname(rp);
    728 		nvlist_add_string(rule, "rproc", rname);
    729 		npf_rproc_release(rp);
    730 	}
    731 	return rule;
    732 }
    733 
    734 /*
    735  * npf_rule_setcode: assign filter code to the rule.
    736  *
    737  * => The code must be validated by the caller.
    738  * => JIT compilation may be performed here.
    739  */
    740 void
    741 npf_rule_setcode(npf_rule_t *rl, const int type, void *code, size_t size)
    742 {
    743 	KASSERT(type == NPF_CODE_BPF);
    744 
    745 	rl->r_type = type;
    746 	rl->r_code = code;
    747 	rl->r_clen = size;
    748 	rl->r_jcode = npf_bpf_compile(code, size);
    749 }
    750 
    751 /*
    752  * npf_rule_setrproc: assign a rule procedure and hold a reference on it.
    753  */
    754 void
    755 npf_rule_setrproc(npf_rule_t *rl, npf_rproc_t *rp)
    756 {
    757 	npf_rproc_acquire(rp);
    758 	rl->r_rproc = rp;
    759 }
    760 
    761 /*
    762  * npf_rule_free: free the specified rule.
    763  */
    764 void
    765 npf_rule_free(npf_rule_t *rl)
    766 {
    767 	npf_natpolicy_t *np = rl->r_natp;
    768 	npf_rproc_t *rp = rl->r_rproc;
    769 
    770 	if (np && (rl->r_attr & NPF_RULE_KEEPNAT) == 0) {
    771 		/* Free NAT policy. */
    772 		npf_nat_freepolicy(np);
    773 	}
    774 	if (rp) {
    775 		/* Release rule procedure. */
    776 		npf_rproc_release(rp);
    777 	}
    778 	if (rl->r_code) {
    779 		/* Free byte-code. */
    780 		kmem_free(rl->r_code, rl->r_clen);
    781 	}
    782 	if (rl->r_jcode) {
    783 		/* Free JIT code. */
    784 		bpf_jit_freecode(rl->r_jcode);
    785 	}
    786 	if (rl->r_info) {
    787 		kmem_free(rl->r_info, rl->r_info_len);
    788 	}
    789 	kmem_free(rl, sizeof(npf_rule_t));
    790 }
    791 
    792 /*
    793  * npf_rule_getid: return the unique ID of a rule.
    794  * npf_rule_getrproc: acquire a reference and return rule procedure, if any.
    795  * npf_rule_getnat: get NAT policy assigned to the rule.
    796  */
    797 
    798 uint64_t
    799 npf_rule_getid(const npf_rule_t *rl)
    800 {
    801 	KASSERT(NPF_DYNAMIC_RULE_P(rl->r_attr));
    802 	return rl->r_id;
    803 }
    804 
    805 npf_rproc_t *
    806 npf_rule_getrproc(const npf_rule_t *rl)
    807 {
    808 	npf_rproc_t *rp = rl->r_rproc;
    809 
    810 	if (rp) {
    811 		npf_rproc_acquire(rp);
    812 	}
    813 	return rp;
    814 }
    815 
    816 npf_natpolicy_t *
    817 npf_rule_getnat(const npf_rule_t *rl)
    818 {
    819 	return rl->r_natp;
    820 }
    821 
    822 /*
    823  * npf_rule_setnat: assign NAT policy to the rule and insert into the
    824  * NAT policy list in the ruleset.
    825  */
    826 void
    827 npf_rule_setnat(npf_rule_t *rl, npf_natpolicy_t *np)
    828 {
    829 	KASSERT(rl->r_natp == NULL);
    830 	rl->r_natp = np;
    831 }
    832 
    833 /*
    834  * npf_rule_inspect: match the interface, direction and run the filter code.
    835  * Returns true if rule matches and false otherwise.
    836  */
    837 static inline bool
    838 npf_rule_inspect(const npf_rule_t *rl, bpf_args_t *bc_args,
    839     const int di_mask, const u_int ifid)
    840 {
    841 	/* Match the interface. */
    842 	if (rl->r_ifid && rl->r_ifid != ifid) {
    843 		return false;
    844 	}
    845 
    846 	/* Match the direction. */
    847 	if ((rl->r_attr & NPF_RULE_DIMASK) != NPF_RULE_DIMASK) {
    848 		if ((rl->r_attr & di_mask) == 0)
    849 			return false;
    850 	}
    851 
    852 	/* Any code? */
    853 	if (!rl->r_code) {
    854 		KASSERT(rl->r_jcode == NULL);
    855 		return true;
    856 	}
    857 	KASSERT(rl->r_type == NPF_CODE_BPF);
    858 	return npf_bpf_filter(bc_args, rl->r_code, rl->r_jcode) != 0;
    859 }
    860 
    861 /*
    862  * npf_rule_reinspect: re-inspect the dynamic rule by iterating its list.
    863  * This is only for the dynamic rules.  Subrules cannot have nested rules.
    864  */
    865 static inline npf_rule_t *
    866 npf_rule_reinspect(const npf_rule_t *rg, bpf_args_t *bc_args,
    867     const int di_mask, const u_int ifid)
    868 {
    869 	npf_rule_t *final_rl = NULL, *rl;
    870 
    871 	KASSERT(NPF_DYNAMIC_GROUP_P(rg->r_attr));
    872 
    873 	for (rl = rg->r_subset; rl; rl = rl->r_next) {
    874 		KASSERT(!final_rl || rl->r_priority >= final_rl->r_priority);
    875 		if (!npf_rule_inspect(rl, bc_args, di_mask, ifid)) {
    876 			continue;
    877 		}
    878 		if (rl->r_attr & NPF_RULE_FINAL) {
    879 			return rl;
    880 		}
    881 		final_rl = rl;
    882 	}
    883 	return final_rl;
    884 }
    885 
    886 /*
    887  * npf_ruleset_inspect: inspect the packet against the given ruleset.
    888  *
    889  * Loop through the rules in the set and run the byte-code of each rule
    890  * against the packet (nbuf chain).  If sub-ruleset is found, inspect it.
    891  */
    892 npf_rule_t *
    893 npf_ruleset_inspect(npf_cache_t *npc, const npf_ruleset_t *rlset,
    894     const int di, const int layer)
    895 {
    896 	nbuf_t *nbuf = npc->npc_nbuf;
    897 	const int di_mask = (di & PFIL_IN) ? NPF_RULE_IN : NPF_RULE_OUT;
    898 	const u_int nitems = rlset->rs_nitems;
    899 	const u_int ifid = nbuf->nb_ifid;
    900 	npf_rule_t *final_rl = NULL;
    901 	bpf_args_t bc_args;
    902 	u_int n = 0;
    903 
    904 	KASSERT(((di & PFIL_IN) != 0) ^ ((di & PFIL_OUT) != 0));
    905 
    906 	/*
    907 	 * Prepare the external memory store and the arguments for
    908 	 * the BPF programs to be executed.  Reset mbuf before taking
    909 	 * any pointers for the BPF.
    910 	 */
    911 	uint32_t bc_words[NPF_BPF_NWORDS];
    912 
    913 	nbuf_reset(nbuf);
    914 	npf_bpf_prepare(npc, &bc_args, bc_words);
    915 
    916 	while (n < nitems) {
    917 		npf_rule_t *rl = rlset->rs_rules[n];
    918 		const u_int skip_to = rl->r_skip_to & SKIPTO_MASK;
    919 		const uint32_t attr = rl->r_attr;
    920 
    921 		KASSERT(!nbuf_flag_p(nbuf, NBUF_DATAREF_RESET));
    922 		KASSERT(n < skip_to);
    923 
    924 		/* Group is a barrier: return a matching if found any. */
    925 		if ((attr & NPF_RULE_GROUP) != 0 && final_rl) {
    926 			break;
    927 		}
    928 
    929 		/* Main inspection of the rule. */
    930 		if (!npf_rule_inspect(rl, &bc_args, di_mask, ifid)) {
    931 			n = skip_to;
    932 			continue;
    933 		}
    934 
    935 		if (NPF_DYNAMIC_GROUP_P(attr)) {
    936 			/*
    937 			 * If this is a dynamic rule, re-inspect the subrules.
    938 			 * If it has any matching rule, then it is final.
    939 			 */
    940 			rl = npf_rule_reinspect(rl, &bc_args, di_mask, ifid);
    941 			if (rl != NULL) {
    942 				final_rl = rl;
    943 				break;
    944 			}
    945 		} else if ((attr & NPF_RULE_GROUP) == 0) {
    946 			/*
    947 			 * Groups themselves are not matching.
    948 			 */
    949 			final_rl = rl;
    950 		}
    951 
    952 		/* Set the matching rule and check for "final". */
    953 		if (attr & NPF_RULE_FINAL) {
    954 			break;
    955 		}
    956 		n++;
    957 	}
    958 
    959 	KASSERT(!nbuf_flag_p(nbuf, NBUF_DATAREF_RESET));
    960 	return final_rl;
    961 }
    962 
    963 /*
    964  * npf_rule_conclude: return decision and the flags for conclusion.
    965  *
    966  * => Returns ENETUNREACH if "block" and 0 if "pass".
    967  */
    968 int
    969 npf_rule_conclude(const npf_rule_t *rl, npf_match_info_t *mi)
    970 {
    971 	/* If not passing - drop the packet. */
    972 	mi->mi_retfl = rl->r_attr;
    973 	mi->mi_rid = rl->r_id;
    974 	return (rl->r_attr & NPF_RULE_PASS) ? 0 : ENETUNREACH;
    975 }
    976 
    977 
    978 #if defined(DDB) || defined(_NPF_TESTING)
    979 
    980 void
    981 npf_ruleset_dump(npf_t *npf, const char *name)
    982 {
    983 	npf_ruleset_t *rlset = npf_config_ruleset(npf);
    984 	npf_rule_t *rg, *rl;
    985 
    986 	LIST_FOREACH(rg, &rlset->rs_dynamic, r_dentry) {
    987 		printf("ruleset '%s':\n", rg->r_name);
    988 		for (rl = rg->r_subset; rl; rl = rl->r_next) {
    989 			printf("\tid %"PRIu64", key: ", rl->r_id);
    990 			for (u_int i = 0; i < NPF_RULE_MAXKEYLEN; i++)
    991 				printf("%x", rl->r_key[i]);
    992 			printf("\n");
    993 		}
    994 	}
    995 }
    996 
    997 #endif
    998