3 files changed, 349 insertions, 117 deletions

diff --git a/sys/net/bridge.c b/sys/net/bridge.c
index fa9b504c4ebed..d0fa2d9758f3c 100644
--- a/sys/net/bridge.c
+++ b/sys/net/bridge.c
@@ -87,6 +87,7 @@
  *  - be very careful when bridging VLANs
  *  - loop detection is still not very robust.
  */
+#include "opt_pfil_hooks.h"
 
 #include <sys/param.h>
 #include <sys/mbuf.h>
@@ -98,7 +99,6 @@
 #include <sys/kernel.h>
 #include <sys/sysctl.h>
 
-#include <net/pfil.h>	/* for ipfilter */
 #include <net/if.h>
 #include <net/if_types.h>
 #include <net/if_var.h>
@@ -109,6 +109,11 @@
 #include <netinet/ip.h>
 #include <netinet/if_ether.h> /* for struct arpcom */
 
+#ifdef PFIL_HOOKS
+#include <net/pfil.h>
+#include <netinet/ip_var.h>
+#endif
+
 #include <net/route.h>
 #include <netinet/ip_fw.h>
 #include <netinet/ip_dummynet.h>
@@ -160,7 +165,6 @@ struct cluster_softc {
 
 
 extern struct protosw inetsw[];			/* from netinet/ip_input.c */
-extern u_char ip_protox[];			/* from netinet/ip_input.c */
 
 static int n_clusters;				/* number of clusters */
 static struct cluster_softc *clusters;
@@ -715,6 +719,7 @@ bridge_dst_lookup(struct ether_header *eh, struct cluster_softc *c)
      * Lookup local addresses in case one matches.  We optimize
      * for the common case of two interfaces.
      */
+    KASSERT(c->ports != 0, ("lookup with no ports!"));
     switch (c->ports) {
 	int i;
     default:
@@ -914,10 +919,6 @@ bdg_forward(struct mbuf *m0, struct ifnet *dst)
     int shared = bdg_copy;		/* someone else is using the mbuf */
     struct ifnet *real_dst = dst;	/* real dst from ether_output */
     struct ip_fw_args args;
-#ifdef PFIL_HOOKS
-    struct packet_filter_hook *pfh;
-    int rv;
-#endif /* PFIL_HOOKS */
     struct ether_header save_eh;
     struct mbuf *m;
 
@@ -969,7 +970,7 @@ bdg_forward(struct mbuf *m0, struct ifnet *dst)
      */
     if (src != NULL && (
 #ifdef PFIL_HOOKS
-	((pfh = pfil_hook_get(PFIL_IN, &inetsw[ip_protox[IPPROTO_IP]].pr_pfh)) != NULL && bdg_ipf !=0) ||
+	(inet_pfil_hook.ph_busy_count >= 0 && bdg_ipf != 0) ||
 #endif
 	(IPFW_LOADED && bdg_ipfw != 0))) {
 
@@ -1006,8 +1007,9 @@ bdg_forward(struct mbuf *m0, struct ifnet *dst)
 	 * NetBSD-style generic packet filter, pfil(9), hooks.
 	 * Enables ipf(8) in bridging.
 	 */
-	if (pfh != NULL && m0->m_pkthdr.len >= sizeof(struct ip) &&
-		ntohs(save_eh.ether_type) == ETHERTYPE_IP) {
+	if (inet_pfil_hook.ph_busy_count >= 0 &&
+	    m0->m_pkthdr.len >= sizeof(struct ip) &&
+	    ntohs(save_eh.ether_type) == ETHERTYPE_IP) {
 	    /*
 	     * before calling the firewall, swap fields the same as IP does.
 	     * here we assume the pkt is an IP one and the header is contiguous
@@ -1017,20 +1019,14 @@ bdg_forward(struct mbuf *m0, struct ifnet *dst)
 	    ip->ip_len = ntohs(ip->ip_len);
 	    ip->ip_off = ntohs(ip->ip_off);
 
-	    do {
-		if (pfh->pfil_func) {
-		    rv = pfh->pfil_func(ip, ip->ip_hl << 2, src, 0, &m0);
-		    if (m0 == NULL) {
-			bdg_dropped++;
-			return NULL;
-		    }
-		    if (rv != 0) {
-			EH_RESTORE(m0);		/* restore Ethernet header */
-			return m0;
-		    }
-		    ip = mtod(m0, struct ip *);
-		}
-	    } while ((pfh = TAILQ_NEXT(pfh, pfil_link)) != NULL);
+	    if (pfil_run_hooks(&inet_pfil_hook, &m0, src, PFIL_IN) != 0) {
+		EH_RESTORE(m0);		/* restore Ethernet header */
+		return m0;
+	    }
+	    if (m0 == NULL) {
+		bdg_dropped++;
+		return NULL;
+	    }
 	    /*
 	     * If we get here, the firewall has passed the pkt, but the mbuf
 	     * pointer might have changed. Restore ip and the fields ntohs()'d.
diff --git a/sys/net/pfil.c b/sys/net/pfil.c
index 2ed40b2f0f4c1..c995ec27cf1e8 100644
--- a/sys/net/pfil.c
+++ b/sys/net/pfil.c
@@ -1,4 +1,5 @@
-/*	$FreeBSD$	*/
+/*	$FreeBSD$ */
+/*	$NetBSD: pfil.c,v 1.20 2001/11/12 23:49:46 lukem Exp $	*/
 
 /*
  * Copyright (c) 1996 Matthew R. Green
@@ -29,30 +30,198 @@
  */
 
 #include <sys/param.h>
+#include <sys/kernel.h>
 #include <sys/errno.h>
 #include <sys/malloc.h>
 #include <sys/socket.h>
 #include <sys/socketvar.h>
 #include <sys/systm.h>
+#include <sys/condvar.h>
+#include <sys/lock.h>
+#include <sys/mutex.h>
+#include <sys/proc.h>
 #include <sys/queue.h>
 
 #include <net/if.h>
 #include <net/pfil.h>
 
-static void pfil_init(struct pfil_head *);
-static int pfil_list_add(pfil_list_t *,
-    int (*)(void *, int, struct ifnet *, int, struct mbuf **), int);
+static struct mtx pfil_global_lock;
+
+MTX_SYSINIT(pfil_heads_lock, &pfil_global_lock, "pfil_head_list lock", MTX_DEF);
+
+static int pfil_list_add(pfil_list_t *, struct packet_filter_hook *, int);
+
 static int pfil_list_remove(pfil_list_t *,
-    int (*)(void *, int, struct ifnet *, int, struct mbuf **));
+    int (*)(void *, struct mbuf **, struct ifnet *, int), void *);
+
+LIST_HEAD(, pfil_head) pfil_head_list =
+    LIST_HEAD_INITIALIZER(&pfil_head_list);
+
+static __inline void
+PFIL_RLOCK(struct pfil_head *ph)
+{
+	mtx_lock(&ph->ph_mtx);
+	ph->ph_busy_count++;
+	mtx_unlock(&ph->ph_mtx);
+}
+
+static __inline void
+PFIL_RUNLOCK(struct pfil_head *ph)
+{
+	mtx_lock(&ph->ph_mtx);
+	ph->ph_busy_count--;
+	if (ph->ph_busy_count == 0 && ph->ph_want_write)
+		cv_signal(&ph->ph_cv);
+	mtx_unlock(&ph->ph_mtx);
+}
+
+static __inline void
+PFIL_WLOCK(struct pfil_head *ph)
+{
+	mtx_lock(&ph->ph_mtx);
+	ph->ph_want_write = 1;
+	while (ph->ph_busy_count > 0)
+		cv_wait(&ph->ph_cv, &ph->ph_mtx);
+}
+
+static __inline int
+PFIL_TRY_WLOCK(struct pfil_head *ph)
+{
+	mtx_lock(&ph->ph_mtx);
+	ph->ph_want_write = 1;
+	if (ph->ph_busy_count > 0) {
+		ph->ph_want_write = 0;
+		mtx_unlock(&ph->ph_mtx);
+		return EBUSY;
+	}
+	return 0;
+}
+
+static __inline void
+PFIL_WUNLOCK(struct pfil_head *ph)
+{
+	ph->ph_want_write = 0;				\
+	mtx_unlock(&ph->ph_mtx);
+	cv_signal(&ph->ph_cv);
+}
 
-static void
-pfil_init(ph)
-	 struct pfil_head *ph;
+#define PFIL_LIST_LOCK() mtx_lock(&pfil_global_lock)
+#define PFIL_LIST_UNLOCK() mtx_unlock(&pfil_global_lock)
+
+/*
+ * pfil_run_hooks() runs the specified packet filter hooks.
+ */
+int
+pfil_run_hooks(struct pfil_head *ph, struct mbuf **mp, struct ifnet *ifp,
+    int dir)
 {
+	struct packet_filter_hook *pfh;
+	struct mbuf *m = *mp;
+	int rv = 0;
+
+	if (ph->ph_busy_count == -1 || ph->ph_want_write)
+		return (0);
+
+	PFIL_RLOCK(ph);
+	for (pfh = pfil_hook_get(dir, ph); pfh != NULL;
+	     pfh = TAILQ_NEXT(pfh, pfil_link)) {
+		if (pfh->pfil_func != NULL) {
+			rv = (*pfh->pfil_func)(pfh->pfil_arg, &m, ifp, dir);
+			if (rv != 0 || m == NULL)
+				break;
+		}
+	}
+	PFIL_RUNLOCK(ph);
+	
+	*mp = m;
+	return (rv);
+}
+
+/*
+ * pfil_head_register() registers a pfil_head with the packet filter
+ * hook mechanism.
+ */
+int
+pfil_head_register(struct pfil_head *ph)
+{
+	struct pfil_head *lph;
+
+	PFIL_LIST_LOCK();
+	LIST_FOREACH(lph, &pfil_head_list, ph_list)
+		if (ph->ph_type == lph->ph_type &&
+		    ph->ph_un.phu_val == lph->ph_un.phu_val) {
+			PFIL_LIST_UNLOCK();
+			return EEXIST;
+		}
+	PFIL_LIST_UNLOCK();
+
+	if (mtx_initialized(&ph->ph_mtx)) {	/* should not happen */
+		KASSERT((0), ("%s: allready initialized!", __func__));
+		return EBUSY;
+	} else {
+		ph->ph_busy_count = -1;
+		ph->ph_want_write = 1;
+		mtx_init(&ph->ph_mtx, "pfil_head_mtx", NULL, MTX_DEF);
+		cv_init(&ph->ph_cv, "pfil_head_cv");
+		mtx_lock(&ph->ph_mtx);			/* XXX: race? */
+	}
 
 	TAILQ_INIT(&ph->ph_in);
 	TAILQ_INIT(&ph->ph_out);
-	ph->ph_init = 1;
+
+	PFIL_LIST_LOCK();
+	LIST_INSERT_HEAD(&pfil_head_list, ph, ph_list);
+	PFIL_LIST_UNLOCK();
+	
+	PFIL_WUNLOCK(ph);
+	
+	return (0);
+}
+
+/*
+ * pfil_head_unregister() removes a pfil_head from the packet filter
+ * hook mechanism.
+ */
+int
+pfil_head_unregister(struct pfil_head *ph)
+{
+	struct packet_filter_hook *pfh, *pfnext;
+		
+	PFIL_LIST_LOCK();
+	/* 
+	 * LIST_REMOVE is safe for unlocked pfil_heads in ph_list.
+	 * No need to WLOCK all of them.
+	 */
+	LIST_REMOVE(ph, ph_list);
+	PFIL_LIST_UNLOCK();
+
+	PFIL_WLOCK(ph);			/* XXX: may sleep (cv_wait)! */
+	
+	TAILQ_FOREACH_SAFE(pfh, &ph->ph_in, pfil_link, pfnext)
+		free(pfh, M_IFADDR);
+	TAILQ_FOREACH_SAFE(pfh, &ph->ph_out, pfil_link, pfnext)
+		free(pfh, M_IFADDR);
+	cv_destroy(&ph->ph_cv);
+	mtx_destroy(&ph->ph_mtx);
+	
+	return (0);
+}
+
+/*
+ * pfil_head_get() returns the pfil_head for a given key/dlt.
+ */
+struct pfil_head *
+pfil_head_get(int type, u_long val)
+{
+	struct pfil_head *ph;
+
+	PFIL_LIST_LOCK();
+	LIST_FOREACH(ph, &pfil_head_list, ph_list)
+		if (ph->ph_type == type && ph->ph_un.phu_val == val)
+			break;
+	PFIL_LIST_UNLOCK();
+	
+	return (ph);
 }
 
 /*
@@ -64,53 +233,71 @@ pfil_init(ph)
  *	PFIL_WAITOK	OK to call malloc with M_WAITOK.
  */
 int
-pfil_add_hook(func, flags, ph)
-	int	(*func)(void *, int, struct ifnet *, int, struct mbuf **);
-	int	flags;
-	struct	pfil_head	*ph;
+pfil_add_hook(int (*func)(void *, struct mbuf **, struct ifnet *, int),
+    void *arg, int flags, struct pfil_head *ph)
 {
-	int err = 0;
+	struct packet_filter_hook *pfh1 = NULL;
+	struct packet_filter_hook *pfh2 = NULL;
+	int err;
 
-	if (ph->ph_init == 0)
-		pfil_init(ph);
+	/* Get memory */
+	if (flags & PFIL_IN) {
+		pfh1 = (struct packet_filter_hook *)malloc(sizeof(*pfh1), 
+		    M_IFADDR, (flags & PFIL_WAITOK) ? M_WAITOK : M_NOWAIT);
+		if (pfh1 == NULL) {
+			err = ENOMEM;
+			goto error;
+		}
+	}
+	if (flags & PFIL_OUT) {
+		pfh2 = (struct packet_filter_hook *)malloc(sizeof(*pfh1),
+		    M_IFADDR, (flags & PFIL_WAITOK) ? M_WAITOK : M_NOWAIT);
+		if (pfh2 == NULL) {
+			err = ENOMEM;
+			goto error;
+		}
+	}
 
-	if (flags & PFIL_IN)
-		err = pfil_list_add(&ph->ph_in, func, flags & ~PFIL_OUT);
-	if (err)
-		return err;
-	if (flags & PFIL_OUT)
-		err = pfil_list_add(&ph->ph_out, func, flags & ~PFIL_IN);
-	if (err) {
-		if (flags & PFIL_IN)
-			pfil_list_remove(&ph->ph_in, func);
-		return err;
+	/* Lock */
+	if (flags & PFIL_WAITOK)
+		PFIL_WLOCK(ph);
+	else {
+		err = PFIL_TRY_WLOCK(ph);
+		if (err)
+			goto error;
 	}
-	return 0;
-}
 
-static int
-pfil_list_add(list, func, flags)
-	pfil_list_t *list;
-	int	(*func)(void *, int, struct ifnet *, int, struct mbuf **);
-	int flags;
-{
-	struct packet_filter_hook *pfh;
+	/* Add */
+	if (flags & PFIL_IN) {
+		pfh1->pfil_func = func;
+		pfh1->pfil_arg = arg;
+		err = pfil_list_add(&ph->ph_in, pfh1, flags & ~PFIL_OUT);
+		if (err)
+			goto done;
+	}
+	if (flags & PFIL_OUT) {
+		pfh2->pfil_func = func;
+		pfh2->pfil_arg = arg;
+		err = pfil_list_add(&ph->ph_out, pfh2, flags & ~PFIL_IN);
+		if (err) {
+			if (flags & PFIL_IN)
+				pfil_list_remove(&ph->ph_in, func, arg);
+			goto done;
+		}
+	}
+
+	ph->ph_busy_count = 0;
+	PFIL_WUNLOCK(ph);
 
-	pfh = (struct packet_filter_hook *)malloc(sizeof(*pfh), M_IFADDR,
-	    flags & PFIL_WAITOK ? M_WAITOK : M_NOWAIT);
-	if (pfh == NULL)
-		return ENOMEM;
-	pfh->pfil_func = func;
-	/*
-	 * insert the input list in reverse order of the output list
-	 * so that the same path is followed in or out of the kernel.
-	 */
-	
-	if (flags & PFIL_IN)
-		TAILQ_INSERT_HEAD(list, pfh, pfil_link);
-	else
-		TAILQ_INSERT_TAIL(list, pfh, pfil_link);
 	return 0;
+done:
+	PFIL_WUNLOCK(ph);
+error:
+	if (pfh1 != NULL)
+		free(pfh1, M_IFADDR);
+	if (pfh2 != NULL)
+		free(pfh2, M_IFADDR);
+	return err;
 }
 
 /*
@@ -118,54 +305,71 @@ pfil_list_add(list, func, flags)
  * hook list.
  */
 int
-pfil_remove_hook(func, flags, ph)
-	int	(*func)(void *, int, struct ifnet *, int, struct mbuf **);
-	int	flags;
-	struct	pfil_head	*ph;
+pfil_remove_hook(int (*func)(void *, struct mbuf **, struct ifnet *, int),
+    void *arg, int flags, struct pfil_head *ph)
 {
 	int err = 0;
 
-	if (ph->ph_init == 0)
-		pfil_init(ph);
+	if (flags & PFIL_WAITOK)
+		PFIL_WLOCK(ph);
+	else {
+		err = PFIL_TRY_WLOCK(ph);
+		if (err)
+			return err;
+	}
 
 	if (flags & PFIL_IN)
-		err = pfil_list_remove(&ph->ph_in, func);
+		err = pfil_list_remove(&ph->ph_in, func, arg);
 	if ((err == 0) && (flags & PFIL_OUT))
-		err = pfil_list_remove(&ph->ph_out, func);
+		err = pfil_list_remove(&ph->ph_out, func, arg);
+
+	if (TAILQ_EMPTY(&ph->ph_in) && TAILQ_EMPTY(&ph->ph_out))
+		ph->ph_busy_count = -1;
+
+	PFIL_WUNLOCK(ph);
+	
 	return err;
 }
 
+static int
+pfil_list_add(pfil_list_t *list, struct packet_filter_hook *pfh1, int flags)
+{
+	struct packet_filter_hook *pfh;
+
+	/*
+	 * First make sure the hook is not already there.
+	 */
+	TAILQ_FOREACH(pfh, list, pfil_link)
+		if (pfh->pfil_func == pfh1->pfil_func &&
+		    pfh->pfil_arg == pfh1->pfil_arg)
+			return EEXIST;
+	/*
+	 * insert the input list in reverse order of the output list
+	 * so that the same path is followed in or out of the kernel.
+	 */
+	if (flags & PFIL_IN)
+		TAILQ_INSERT_HEAD(list, pfh1, pfil_link);
+	else
+		TAILQ_INSERT_TAIL(list, pfh1, pfil_link);
+
+	return 0;
+}
+
 /*
  * pfil_list_remove is an internal function that takes a function off the
  * specified list.
  */
 static int
-pfil_list_remove(list, func)
-	pfil_list_t *list;
-	int	(*func)(void *, int, struct ifnet *, int, struct mbuf **);
+pfil_list_remove(pfil_list_t *list,
+    int (*func)(void *, struct mbuf **, struct ifnet *, int), void *arg)
 {
 	struct packet_filter_hook *pfh;
 
 	TAILQ_FOREACH(pfh, list, pfil_link)
-		if (pfh->pfil_func == func) {
+		if (pfh->pfil_func == func && pfh->pfil_arg == arg) {
 			TAILQ_REMOVE(list, pfh, pfil_link);
 			free(pfh, M_IFADDR);
 			return 0;
 		}
 	return ENOENT;
 }
-
-struct packet_filter_hook *
-pfil_hook_get(flag, ph)
-	int flag;
-	struct	pfil_head	*ph;
-{
-	if (ph->ph_init != 0)
-		switch (flag) {
-		case PFIL_IN:
-			return (TAILQ_FIRST(&ph->ph_in));
-		case PFIL_OUT:
-			return (TAILQ_FIRST(&ph->ph_out));
-		}
-	return NULL;
-}
diff --git a/sys/net/pfil.h b/sys/net/pfil.h
index 057a531848803..b1f29aecc2687 100644
--- a/sys/net/pfil.h
+++ b/sys/net/pfil.h
@@ -1,4 +1,5 @@
-/*	$FreeBSD$	*/
+/*	$FreeBSD$ */
+/*	$NetBSD: pfil.h,v 1.22 2003/06/23 12:57:08 martin Exp $	*/
 
 /*
  * Copyright (c) 1996 Matthew R. Green
@@ -31,6 +32,10 @@
 #ifndef _NET_PFIL_H_
 #define _NET_PFIL_H_
 
+#include <sys/systm.h>
+#include <sys/_lock.h>
+#include <sys/_mutex.h>
+#include <sys/condvar.h>	/* XXX */
 #include <sys/queue.h>
 
 struct mbuf;
@@ -42,7 +47,8 @@ struct ifnet;
  */
 struct packet_filter_hook {
         TAILQ_ENTRY(packet_filter_hook) pfil_link;
-	int	(*pfil_func)(void *, int, struct ifnet *, int, struct mbuf **);
+	int	(*pfil_func)(void *, struct mbuf **, struct ifnet *, int);
+	void	*pfil_arg;
 	int	pfil_flags;
 };
 
@@ -53,28 +59,54 @@ struct packet_filter_hook {
 
 typedef	TAILQ_HEAD(pfil_list, packet_filter_hook) pfil_list_t;
 
+#define	PFIL_TYPE_AF		1	/* key is AF_* type */
+#define	PFIL_TYPE_IFNET		2	/* key is ifnet pointer */
+
 struct pfil_head {
 	pfil_list_t	ph_in;
 	pfil_list_t	ph_out;
-	int		ph_init;
+	int		ph_type;
+	/*
+	 * Locking: use a busycounter per pfil_head.
+	 * Use ph_busy_count = -1 to indicate pfil_head is empty.
+	 */
+	int		ph_busy_count;	/* count of threads with read lock */
+	int		ph_want_write;	/* want write lock flag */
+	struct cv	ph_cv;		/* for waking up writers */
+	struct mtx	ph_mtx;		/* mutex on locking state */
+	union {
+		u_long		phu_val;
+		void		*phu_ptr;
+	} ph_un;
+#define	ph_af		ph_un.phu_val
+#define	ph_ifnet	ph_un.phu_ptr
+	LIST_ENTRY(pfil_head) ph_list;
 };
 
-struct packet_filter_hook *pfil_hook_get(int, struct pfil_head *);
-int	pfil_add_hook(int (*func)(void *, int,
-	    struct ifnet *, int, struct mbuf **), int, struct pfil_head *);
-int	pfil_remove_hook(int (*func)(void *, int,
-	    struct ifnet *, int, struct mbuf **), int, struct pfil_head *);
+int	pfil_run_hooks(struct pfil_head *, struct mbuf **, struct ifnet *,
+	    int);
+
+int	pfil_add_hook(int (*func)(void *, struct mbuf **,
+	    struct ifnet *, int), void *, int, struct pfil_head *);
+int	pfil_remove_hook(int (*func)(void *, struct mbuf **,
+	    struct ifnet *, int), void *, int, struct pfil_head *);
+
+int	pfil_head_register(struct pfil_head *);
+int	pfil_head_unregister(struct pfil_head *);
 
-/* XXX */
-#if defined(_KERNEL) && !defined(KLD_MODULE)
-#include "opt_ipfilter.h"
-#endif
+struct pfil_head *pfil_head_get(int, u_long);
 
-#if IPFILTER > 0
-#ifdef PFIL_HOOKS
-#undef PFIL_HOOKS
-#endif
-#define PFIL_HOOKS
-#endif /* IPFILTER */
+static __inline struct packet_filter_hook *
+pfil_hook_get(int dir, struct pfil_head *ph)
+{
+	KASSERT(ph->ph_busy_count > 0, 
+	    ("pfil_hook_get: called on unbusy pfil_head"));
+	if (dir == PFIL_IN)
+		return (TAILQ_FIRST(&ph->ph_in));
+	else if (dir == PFIL_OUT)
+		return (TAILQ_FIRST(&ph->ph_out));
+	else
+		return (NULL);
+}
 
 #endif /* _NET_PFIL_H_ */