pf: Initialize pf_kpool mutexes earlier

There are some error paths in ioctl handlers that will call
pf_krule_free() before the rule's rpool.mtx field is initialized,
causing a panic with INVARIANTS enabled.

Fix the problem by introducing pf_krule_alloc() and initializing the
mutex there.  This does mean that the rule->krule and pool->kpool
conversion functions need to stop zeroing the input structure, but I
don't see a nicer way to handle this except perhaps by guarding the
mtx_destroy() with a mtx_initialized() check.

Constify some related functions while here and add a regression test
based on a syzkaller reproducer.

Reported by:	syzbot+77cd12872691d219c158@syzkaller.appspotmail.com
Reviewed by:	kp
Sponsored by:	The FreeBSD Foundation

(cherry picked from commit 773e3a7)

sys/net/pfvar.h

@@ -167,7 +167,7 @@ pf_counter_u64_periodic(struct pf_counter_u64 *pfcu64)
 }
 
 static inline u_int64_t
-pf_counter_u64_fetch(struct pf_counter_u64 *pfcu64)
+pf_counter_u64_fetch(const struct pf_counter_u64 *pfcu64)
 {
 	struct pf_counter_u64_pcpu *pcpu;
 	u_int64_t sum;
@@ -261,7 +261,7 @@ pf_counter_u64_add(struct pf_counter_u64 *pfcu64, uint32_t n)
 }
 
 static inline u_int64_t
-pf_counter_u64_fetch(struct pf_counter_u64 *pfcu64)
+pf_counter_u64_fetch(const struct pf_counter_u64 *pfcu64)
 {
 
 	return (counter_u64_fetch(pfcu64->counter));
@@ -2145,6 +2145,7 @@ struct pf_kruleset	*pf_find_kruleset(const char *);
 struct pf_kruleset	*pf_find_or_create_kruleset(const char *);
 void			 pf_rs_initialize(void);
 
+struct pf_krule		*pf_krule_alloc(void);
 void			 pf_krule_free(struct pf_krule *);
 #endif
 

sys/netpfil/pf/pf_ioctl.c

@@ -1508,6 +1508,16 @@ pf_altq_get_nth_active(u_int32_t n)
 }
 #endif /* ALTQ */
 
+struct pf_krule *
+pf_krule_alloc(void)
+{
+	struct pf_krule *rule;
+
+	rule = malloc(sizeof(struct pf_krule), M_PFRULE, M_WAITOK | M_ZERO);
+	mtx_init(&rule->rpool.mtx, "pf_krule_pool", NULL, MTX_DEF);
+	return (rule);
+}
+
 void
 pf_krule_free(struct pf_krule *rule)
 {
@@ -1577,27 +1587,23 @@ pf_kpool_to_pool(const struct pf_kpool *kpool, struct pf_pool *pool)
 	pool->opts = kpool->opts;
 }
 
-static int
+static void
 pf_pool_to_kpool(const struct pf_pool *pool, struct pf_kpool *kpool)
 {
 	_Static_assert(sizeof(pool->key) == sizeof(kpool->key), "");
 	_Static_assert(sizeof(pool->counter) == sizeof(kpool->counter), "");
 
-	bzero(kpool, sizeof(*kpool));
-
 	bcopy(&pool->key, &kpool->key, sizeof(kpool->key));
 	bcopy(&pool->counter, &kpool->counter, sizeof(kpool->counter));
 
 	kpool->tblidx = pool->tblidx;
 	kpool->proxy_port[0] = pool->proxy_port[0];
 	kpool->proxy_port[1] = pool->proxy_port[1];
 	kpool->opts = pool->opts;
-
-	return (0);
 }
 
 static void
-pf_krule_to_rule(struct pf_krule *krule, struct pf_rule *rule)
+pf_krule_to_rule(const struct pf_krule *krule, struct pf_rule *rule)
 {
 
 	bzero(rule, sizeof(*rule));
@@ -1720,8 +1726,6 @@ pf_rule_to_krule(const struct pf_rule *rule, struct pf_krule *krule)
 	if (ret != 0)
 		return (ret);
 
-	bzero(krule, sizeof(*krule));
-
 	bcopy(&rule->src, &krule->src, sizeof(rule->src));
 	bcopy(&rule->dst, &krule->dst, sizeof(rule->dst));
 
@@ -1735,9 +1739,7 @@ pf_rule_to_krule(const struct pf_rule *rule, struct pf_krule *krule)
 	strlcpy(krule->overload_tblname, rule->overload_tblname,
 	    sizeof(rule->overload_tblname));
 
-	ret = pf_pool_to_kpool(&rule->rpool, &krule->rpool);
-	if (ret != 0)
-		return (ret);
+	pf_pool_to_kpool(&rule->rpool, &krule->rpool);
 
 	/* Don't allow userspace to set evaulations, packets or bytes. */
 	/* kif, anchor, overload_tbl are not copied over. */
@@ -1970,8 +1972,6 @@ pf_ioctl_addrule(struct pf_krule *rule, uint32_t ticket,
 	int			 rs_num;
 	int			 error = 0;
 
-	mtx_init(&rule->rpool.mtx, "pf_krule_pool", NULL, MTX_DEF);
-
 	if ((rule->return_icmp >> 8) > ICMP_MAXTYPE) {
 		error = EINVAL;
 		goto errout_unlocked;
@@ -2336,7 +2336,7 @@ pfioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flags, struct thread *td
 		if (! nvlist_exists_nvlist(nvl, "rule"))
 			ERROUT(EINVAL);
 
-		rule = malloc(sizeof(*rule), M_PFRULE, M_WAITOK | M_ZERO);
+		rule = pf_krule_alloc();
 		error = pf_nvrule_to_krule(nvlist_get_nvlist(nvl, "rule"),
 		    rule);
 		if (error)
@@ -2369,10 +2369,10 @@ pfioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flags, struct thread *td
 		struct pfioc_rule	*pr = (struct pfioc_rule *)addr;
 		struct pf_krule		*rule;
 
-		rule = malloc(sizeof(*rule), M_PFRULE, M_WAITOK | M_ZERO);
+		rule = pf_krule_alloc();
 		error = pf_rule_to_krule(&pr->rule, rule);
 		if (error != 0) {
-			free(rule, M_PFRULE);
+			pf_krule_free(rule);
 			break;
 		}
 
@@ -2622,7 +2622,7 @@ pfioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flags, struct thread *td
 		}
 
 		if (pcr->action != PF_CHANGE_REMOVE) {
-			newrule = malloc(sizeof(*newrule), M_PFRULE, M_WAITOK | M_ZERO);
+			newrule = pf_krule_alloc();
 			error = pf_rule_to_krule(&pcr->rule, newrule);
 			if (error != 0) {
 				free(newrule, M_PFRULE);

sys/netpfil/pf/pf_nv.c

@@ -223,8 +223,6 @@ pf_nvpool_to_pool(const nvlist_t *nvl, struct pf_kpool *kpool)
 {
 	int error = 0;
 
-	bzero(kpool, sizeof(*kpool));
-
 	PFNV_CHK(pf_nvbinary(nvl, "key", &kpool->key, sizeof(kpool->key)));
 
 	if (nvlist_exists_nvlist(nvl, "counter")) {

tests/sys/netpfil/pf/ioctl/validation.c

@@ -869,6 +869,32 @@ ATF_TC_CLEANUP(rpool_mtx, tc)
 	COMMON_CLEANUP();
 }
 
+ATF_TC_WITH_CLEANUP(rpool_mtx2);
+ATF_TC_HEAD(rpool_mtx2, tc)
+{
+	atf_tc_set_md_var(tc, "require.user", "root");
+}
+
+ATF_TC_BODY(rpool_mtx2, tc)
+{
+	struct pfioc_rule rule;
+
+	COMMON_HEAD();
+
+	memset(&rule, 0, sizeof(rule));
+
+	rule.pool_ticket = 1000000;
+	rule.action = PF_CHANGE_ADD_HEAD;
+	rule.rule.af = AF_INET;
+
+	ioctl(dev, DIOCCHANGERULE, &rule);
+}
+
+ATF_TC_CLEANUP(rpool_mtx2, tc)
+{
+	COMMON_CLEANUP();
+}
+
 
 ATF_TP_ADD_TCS(tp)
 {
@@ -893,6 +919,7 @@ ATF_TP_ADD_TCS(tp)
 	ATF_TP_ADD_TC(tp, getsrcnodes);
 	ATF_TP_ADD_TC(tp, tag);
 	ATF_TP_ADD_TC(tp, rpool_mtx);
+	ATF_TP_ADD_TC(tp, rpool_mtx2);
 
 	return (atf_no_error());
 }