From: Brendan O'Dea <bod@optus.net>
Date: Sat, 4 Jun 2005 15:42:35 +0000 (+0000)
Subject: Include endpoint address in accounting dump files.
X-Git-Tag: 2.2.1-2fdn3.1~19^2^2~1^2~170
X-Git-Url: http://git.sameswireless.fr/l2tpns.git/commitdiff_plain/b830a899ca4587707123482c5dbb8c14edb8fd9c

Include endpoint address in accounting dump files.
Convert mainloop to use epoll rather than select.
---

diff --git a/Changes b/Changes
index a7e7740..22b2d2a 100644
--- a/Changes
+++ b/Changes
@@ -1,9 +1,7 @@
-* Thu Jun 2 2005 Brendan O'Dea <bod@optusnet.com.au> 2.1.0
+* Sun Jun 5 2005 Brendan O'Dea <bod@optusnet.com.au> 2.1.0
 - Add IPv6 support from Jonathan McDowell.
 - Add CHAP support from Jordan Hrycaj.
 - Add interim accounting support from Vladislav Bjelic.
-- Add Acct-Output-Gigawords, Acct-Input-Gigawords attributes to RADIUS
-  accounting packets.
 - Negotiate MRU, default 1458 to avoid fragmentation.
 - Sanity check that cluster_send_session is not called from a child
   process.
@@ -31,6 +29,13 @@
 - New config option: allow_duplicate_users which determines whether
   or not to kill older sessions with the same username.
 - Fix byte counters in accounting records.
+- Add Acct-Output-Gigawords, Acct-Input-Gigawords attributes to RADIUS
+  accounting packets.
+- Fix icmp host unreachable to use router address.
+- Include endpoint address in accounting dump files.
+- Convert mainloop to use epoll rather than select.
+- Add note about fragmentation in Docs/manual.html, and a sample
+  iptables rule for MSS clamping.
 - Merge 2.0.22:
   + Show session open time in "show session"/"show user" detailed output.
   + Have slaves with BGP configured drop BGP on receipt of a shutdown
diff --git a/Makefile b/Makefile
index 8ac1656..9cd6ff5 100644
--- a/Makefile
+++ b/Makefile
@@ -37,6 +37,7 @@ TESTS = generateload bounce
 DEFINES += -DSTATISTICS
 DEFINES += -DSTAT_CALLS
 DEFINES += -DRINGBUFFER
+DEFINES += -DHAVE_EPOLL
 
 DEFINES += -DBGP
 OBJS += bgp.o
diff --git a/bgp.c b/bgp.c
index e5af65c..96e2868 100644
--- a/bgp.c
+++ b/bgp.c
@@ -10,7 +10,7 @@
  *   nor RFC2385 (which requires a kernel patch on 2.4 kernels).
  */
 
-char const *cvs_id_bgp = "$Id: bgp.c,v 1.9 2004-12-16 08:49:52 bodea Exp $";
+char const *cvs_id_bgp = "$Id: bgp.c,v 1.10 2005-06-04 15:42:35 bodea Exp $";
 
 #include <stdlib.h>
 #include <unistd.h>
@@ -75,6 +75,10 @@ int bgp_setup(int as)
 
 	    return 0;
 	}
+
+	peer->edata.type = FD_TYPE_BGP;
+	peer->edata.index = i;
+	peer->events = 0;
     }
 
     if (as < 1)
@@ -270,6 +274,7 @@ static void bgp_clear(struct bgp_peer *peer)
     peer->inbuf->done = 0;
 
     peer->cli_flag = 0;
+    peer->events = 0;
 
     if (peer->state != peer->next_state)
     {
@@ -487,130 +492,160 @@ void bgp_enable_routing(int enable)
     LOG(4, 0, 0, "%s BGP routing\n", enable ? "Enabled" : "Suspended");
 }
 
-/* return a bitmask indicating if the socket should be added to the
-   read set (1) and or write set (2) for select */
-int bgp_select_state(struct bgp_peer *peer)
+#ifdef HAVE_EPOLL
+# include <sys/epoll.h>
+#else
+# include "fake_epoll.h"
+#endif
+
+/* return a bitmask of the events required to poll this peer's fd */
+int bgp_set_poll()
 {
-    int flags = 0;
+    int i;
 
     if (!bgp_configured)
     	return 0;
 
-    if (peer->state == Disabled || peer->state == Idle)
-    	return 0;
+    for (i = 0; i < BGP_NUM_PEERS; i++)
+    {
+	struct bgp_peer *peer = &bgp_peers[i];
+	int events = 0;
+
+	if (peer->state == Disabled || peer->state == Idle)
+	    continue;
+
+	if (peer->inbuf->done < BGP_MAX_PACKET_SIZE)
+	    events |= EPOLLIN;
+
+	if (peer->state == Connect ||		/* connection in progress */
+	    peer->update_routes ||		/* routing updates */
+	    peer->outbuf->packet.header.len)	/* pending output */
+	    events |= EPOLLOUT;
 
-    if (peer->inbuf->done < BGP_MAX_PACKET_SIZE)
-	flags |= 1;
+    	if (peer->events != events)
+	{
+	    struct epoll_event ev;
 
-    if (peer->state == Connect ||		/* connection in progress */
-	peer->update_routes ||			/* routing updates */
-	peer->outbuf->packet.header.len)	/* pending output */
-    	flags |= 2;
+	    ev.events = peer->events = events;
+	    ev.data.ptr = &peer->edata;
+	    epoll_ctl(epollfd, EPOLL_CTL_MOD, peer->sock, &ev);
+	}
+    }
 
-    return flags;
+    return 1;
 }
 
-/* process bgp peer */
-int bgp_process(struct bgp_peer *peer, int readable, int writable)
+/* process bgp events/timers */
+int bgp_process(uint32_t events[])
 {
+    int i;
+
     if (!bgp_configured)
     	return 0;
 
-    if (*peer->name && peer->cli_flag == BGP_CLI_RESTART)
-	return bgp_restart(peer);
+    for (i = 0; i < BGP_NUM_PEERS; i++)
+    {
+	struct bgp_peer *peer = &bgp_peers[i];
 
-    if (peer->state == Disabled)
-    	return 1;
+	if (*peer->name && peer->cli_flag == BGP_CLI_RESTART)
+	{
+	    bgp_restart(peer);
+	    continue;
+	}
 
-    if (peer->cli_flag)
-    {
-	switch (peer->cli_flag)
+	if (peer->state == Disabled)
+	    continue;
+
+	if (peer->cli_flag)
 	{
-	case BGP_CLI_SUSPEND:
-	    if (peer->routing)
+	    switch (peer->cli_flag)
 	    {
-		peer->routing = 0;
-		if (peer->state == Established)
-		    peer->update_routes = 1;
-	    }
+	    case BGP_CLI_SUSPEND:
+		if (peer->routing)
+		{
+		    peer->routing = 0;
+		    if (peer->state == Established)
+			peer->update_routes = 1;
+		}
 
-	    break;
+		break;
 
-	case BGP_CLI_ENABLE:
-	    if (!peer->routing)
-	    {
-		peer->routing = 1;
-		if (peer->state == Established)
-		    peer->update_routes = 1;
+	    case BGP_CLI_ENABLE:
+		if (!peer->routing)
+		{
+		    peer->routing = 1;
+		    if (peer->state == Established)
+			peer->update_routes = 1;
+		}
+
+		break;
 	    }
 
-	    break;
+	    peer->cli_flag = 0;
 	}
 
-	peer->cli_flag = 0;
-    }
-
-    /* handle empty/fill of buffers */
-    if (writable)
-    {
-	int r = 1;
-	if (peer->state == Connect)
-	    r = bgp_handle_connect(peer);
-	else if (peer->outbuf->packet.header.len)
-	    r = bgp_write(peer);
+	/* handle empty/fill of buffers */
+	if (events[i] & EPOLLOUT)
+	{
+	    int r = 1;
+	    if (peer->state == Connect)
+		r = bgp_handle_connect(peer);
+	    else if (peer->outbuf->packet.header.len)
+		r = bgp_write(peer);
 
-	if (!r)
-	    return 0;
-    }
+	    if (!r)
+		continue;
+	}
 
-    if (readable)
-    {
-	if (!bgp_read(peer))
-	    return 0;
-    }
+	if (events[i] & (EPOLLIN|EPOLLHUP))
+	{
+	    if (!bgp_read(peer))
+		continue;
+	}
 
-    /* process input buffer contents */
-    while (peer->inbuf->done >= sizeof(peer->inbuf->packet.header)
-	&& !peer->outbuf->packet.header.len) /* may need to queue a response */
-    {
-	if (bgp_handle_input(peer) < 0)
-	    return 0;
-    }
+	/* process input buffer contents */
+	while (peer->inbuf->done >= sizeof(peer->inbuf->packet.header)
+	    && !peer->outbuf->packet.header.len) /* may need to queue a response */
+	{
+	    if (bgp_handle_input(peer) < 0)
+		continue;
+	}
 
-    /* process pending updates */
-    if (peer->update_routes
-	&& !peer->outbuf->packet.header.len) /* ditto */
-    {
-	if (!bgp_send_update(peer))
-	    return 0;
-    }
+	/* process pending updates */
+	if (peer->update_routes
+	    && !peer->outbuf->packet.header.len) /* ditto */
+	{
+	    if (!bgp_send_update(peer))
+		continue;
+	}
 
-    /* process timers */
-    if (peer->state == Established)
-    {
-	if (time_now > peer->expire_time)
+	/* process timers */
+	if (peer->state == Established)
 	{
-	    LOG(1, 0, 0, "No message from BGP peer %s in %ds\n",
-		peer->name, peer->hold);
+	    if (time_now > peer->expire_time)
+	    {
+		LOG(1, 0, 0, "No message from BGP peer %s in %ds\n",
+		    peer->name, peer->hold);
 
-	    bgp_send_notification(peer, BGP_ERR_HOLD_TIMER_EXP, 0);
-	    return 0;
-	}
+		bgp_send_notification(peer, BGP_ERR_HOLD_TIMER_EXP, 0);
+		continue;
+	    }
 
-	if (time_now > peer->keepalive_time && !peer->outbuf->packet.header.len)
-	    bgp_send_keepalive(peer);
-    }
-    else if (peer->state == Idle)
-    {
-	if (time_now > peer->retry_time)
-	    return bgp_connect(peer);
-    }
-    else if (time_now > peer->state_time + BGP_STATE_TIME)
-    {
-	LOG(1, 0, 0, "%s timer expired for BGP peer %s\n",
-	    bgp_state_str(peer->state), peer->name);
+	    if (time_now > peer->keepalive_time && !peer->outbuf->packet.header.len)
+		bgp_send_keepalive(peer);
+	}
+	else if (peer->state == Idle)
+	{
+	    if (time_now > peer->retry_time)
+		bgp_connect(peer);
+	}
+	else if (time_now > peer->state_time + BGP_STATE_TIME)
+	{
+	    LOG(1, 0, 0, "%s timer expired for BGP peer %s\n",
+		bgp_state_str(peer->state), peer->name);
 
-	return bgp_restart(peer);
+	    bgp_restart(peer);
+	}
     }
 
     return 1;
@@ -661,6 +696,7 @@ static int bgp_connect(struct bgp_peer *peer)
 {
     static int bgp_port = 0;
     struct sockaddr_in addr;
+    struct epoll_event ev;
 
     if (!bgp_port)
     {
@@ -683,6 +719,11 @@ static int bgp_connect(struct bgp_peer *peer)
 	return 0;
     }
 
+    /* add to poll set */
+    ev.events = peer->events = EPOLLOUT;
+    ev.data.ptr = &peer->edata;
+    epoll_ctl(epollfd, EPOLL_CTL_ADD, peer->sock, &ev);
+
     /* set to non-blocking */
     fcntl(peer->sock, F_SETFL, fcntl(peer->sock, F_GETFL, 0) | O_NONBLOCK);
 
diff --git a/bgp.h b/bgp.h
index 4c14c7c..55b0ee3 100644
--- a/bgp.h
+++ b/bgp.h
@@ -1,5 +1,5 @@
 /* BGPv4 (RFC1771) */
-/* $Id: bgp.h,v 1.4 2004-12-16 08:49:52 bodea Exp $ */
+/* $Id: bgp.h,v 1.5 2005-06-04 15:42:35 bodea Exp $ */
 
 #ifndef __BGP_H__
 #define __BGP_H__
@@ -173,6 +173,8 @@ struct bgp_peer {
     int cli_flag;			/* updates requested from CLI */
     char *path_attrs;			/* path attrs to send in UPDATE message */
     int path_attr_len;			/* length of path attrs */
+    uint32_t events;			/* events to poll */
+    struct event_data edata;		/* poll data */
 };
 
 /* bgp_peer.cli_flag */
@@ -194,8 +196,8 @@ int bgp_restart(struct bgp_peer *peer);
 int bgp_add_route(in_addr_t ip, in_addr_t mask);
 int bgp_del_route(in_addr_t ip, in_addr_t mask);
 void bgp_enable_routing(int enable);
-int bgp_select_state(struct bgp_peer *peer);
-int bgp_process(struct bgp_peer *peer, int readable, int writable);
+int bgp_set_poll(void);
+int bgp_process(uint32_t events[]);
 char const *bgp_state_str(enum bgp_state state);
 
 extern char const *cvs_id_bgp;
diff --git a/cli.c b/cli.c
index bf0987f..e6eb2c5 100644
--- a/cli.c
+++ b/cli.c
@@ -2,7 +2,7 @@
 // vim: sw=8 ts=8
 
 char const *cvs_name = "$Name:  $";
-char const *cvs_id_cli = "$Id: cli.c,v 1.60 2005-06-02 11:32:30 bodea Exp $";
+char const *cvs_id_cli = "$Id: cli.c,v 1.61 2005-06-04 15:42:35 bodea Exp $";
 
 #include <stdio.h>
 #include <stdarg.h>
@@ -49,15 +49,6 @@ extern struct cli_tunnel_actions *cli_tunnel_actions;
 extern tbft *filter_list;
 extern ip_filtert *ip_filters;
 
-static char *debug_levels[] = {
-	"CRIT",
-	"ERROR",
-	"WARN",
-	"INFO",
-	"CALL",
-	"DATA",
-};
-
 struct
 {
 	char critical;
@@ -68,15 +59,24 @@ struct
 	char data;
 } debug_flags;
 
-static int debug_session;
-static int debug_tunnel;
+#ifdef RINGBUFFER
+
 static int debug_rb_tail;
+static char *debug_levels[] = {
+	"CRIT",
+	"ERROR",
+	"WARN",
+	"INFO",
+	"CALL",
+	"DATA",
+};
+
+#endif
 
 static int cmd_show_session(struct cli_def *cli, char *command, char **argv, int argc);
 static int cmd_show_tunnels(struct cli_def *cli, char *command, char **argv, int argc);
 static int cmd_show_users(struct cli_def *cli, char *command, char **argv, int argc);
 static int cmd_show_radius(struct cli_def *cli, char *command, char **argv, int argc);
-static int cmd_show_counters(struct cli_def *cli, char *command, char **argv, int argc);
 static int cmd_show_version(struct cli_def *cli, char *command, char **argv, int argc);
 static int cmd_show_pool(struct cli_def *cli, char *command, char **argv, int argc);
 static int cmd_show_run(struct cli_def *cli, char *command, char **argv, int argc);
@@ -84,7 +84,6 @@ static int cmd_show_banana(struct cli_def *cli, char *command, char **argv, int
 static int cmd_show_plugins(struct cli_def *cli, char *command, char **argv, int argc);
 static int cmd_show_throttle(struct cli_def *cli, char *command, char **argv, int argc);
 static int cmd_write_memory(struct cli_def *cli, char *command, char **argv, int argc);
-static int cmd_clear_counters(struct cli_def *cli, char *command, char **argv, int argc);
 static int cmd_drop_user(struct cli_def *cli, char *command, char **argv, int argc);
 static int cmd_drop_tunnel(struct cli_def *cli, char *command, char **argv, int argc);
 static int cmd_drop_session(struct cli_def *cli, char *command, char **argv, int argc);
@@ -102,6 +101,11 @@ static int cmd_uptime(struct cli_def *cli, char *command, char **argv, int argc)
 static int regular_stuff(struct cli_def *cli);
 static void parsemac(char *string, char mac[6]);
 
+#ifdef STATISTICS
+static int cmd_show_counters(struct cli_def *cli, char *command, char **argv, int argc);
+static int cmd_clear_counters(struct cli_def *cli, char *command, char **argv, int argc);
+#endif /* STATISTICS */
+
 #ifdef BGP
 #define MODE_CONFIG_BGP 8
 static int cmd_router_bgp(struct cli_def *cli, char *command, char **argv, int argc);
@@ -312,8 +316,6 @@ void cli_do(int sockfd)
 		cli->users = 0;
 	}
 
-	debug_session = 0;
-	debug_tunnel = 0;
 #ifdef RINGBUFFER
 	debug_rb_tail = ringbuffer->tail;
 #endif
@@ -646,6 +648,7 @@ static int cmd_show_users(struct cli_def *cli, char *command, char **argv, int a
 	return CLI_OK;
 }
 
+#ifdef STATISTICS
 static int cmd_show_counters(struct cli_def *cli, char *command, char **argv, int argc)
 {
 	if (CLI_HELP_REQUESTED)
@@ -697,7 +700,7 @@ static int cmd_show_counters(struct cli_def *cli, char *command, char **argv, in
 	cli_print(cli, "%-30s%u", "multi_read_exceeded",	GET_STAT(multi_read_exceeded));
 
 
-#ifdef STATISTICS
+#ifdef STAT_CALLS
 	cli_print(cli, "\n%-30s%-10s", "Counter", "Value");
 	cli_print(cli, "-----------------------------------------");
 	cli_print(cli, "%-30s%u", "call_processtun",		GET_STAT(call_processtun));
@@ -731,7 +734,7 @@ static int cmd_show_counters(struct cli_def *cli, char *command, char **argv, in
 	cli_print(cli, "%-30s%u", "call_radiussend",		GET_STAT(call_radiussend));
 	cli_print(cli, "%-30s%u", "call_radiusretry",		GET_STAT(call_radiusretry));
 	cli_print(cli, "%-30s%u", "call_random_data",		GET_STAT(call_random_data));
-#endif
+#endif /* STAT_CALLS */
 
 	{
 		time_t l = GET_STAT(last_reset);
@@ -746,6 +749,19 @@ static int cmd_show_counters(struct cli_def *cli, char *command, char **argv, in
 	return CLI_OK;
 }
 
+static int cmd_clear_counters(struct cli_def *cli, char *command, char **argv, int argc)
+{
+	if (CLI_HELP_REQUESTED)
+		return CLI_HELP_NO_ARGS;
+
+	memset(_statistics, 0, sizeof(struct Tstats));
+	SET_STAT(last_reset, time(NULL));
+
+	cli_print(cli, "Counters cleared");
+	return CLI_OK;
+}
+#endif /* STATISTICS */
+
 static int cmd_show_version(struct cli_def *cli, char *command, char **argv, int argc)
 {
 	int tag = 0;
@@ -1134,18 +1150,6 @@ static int cmd_show_banana(struct cli_def *cli, char *command, char **argv, int
 	return CLI_OK;
 }
 
-static int cmd_clear_counters(struct cli_def *cli, char *command, char **argv, int argc)
-{
-	if (CLI_HELP_REQUESTED)
-		return CLI_HELP_NO_ARGS;
-
-	memset(_statistics, 0, sizeof(struct Tstats));
-	SET_STAT(last_reset, time(NULL));
-
-	cli_print(cli, "Counters cleared");
-	return CLI_OK;
-}
-
 static int cmd_drop_user(struct cli_def *cli, char *command, char **argv, int argc)
 {
 	int i;
@@ -1896,10 +1900,10 @@ static int cmd_set(struct cli_def *cli, char *command, char **argv, int argc)
 
 int regular_stuff(struct cli_def *cli)
 {
+#ifdef RINGBUFFER
 	int out = 0;
 	int i;
 
-#ifdef RINGBUFFER
 	for (i = debug_rb_tail; i != ringbuffer->tail; i = (i + 1) % RINGBUFFER_SIZE)
 	{
 		char *m = ringbuffer->buffer[i].message;
diff --git a/cluster.c b/cluster.c
index 8444b78..83d9501 100644
--- a/cluster.c
+++ b/cluster.c
@@ -1,6 +1,6 @@
 // L2TPNS Clustering Stuff
 
-char const *cvs_id_cluster = "$Id: cluster.c,v 1.40 2005-06-02 11:32:30 bodea Exp $";
+char const *cvs_id_cluster = "$Id: cluster.c,v 1.41 2005-06-04 15:42:35 bodea Exp $";
 
 #include <stdio.h>
 #include <stdlib.h>
@@ -38,7 +38,7 @@ char const *cvs_id_cluster = "$Id: cluster.c,v 1.40 2005-06-02 11:32:30 bodea Ex
  */
 
 // Module variables.
-int cluster_sockfd = 0;			// The filedescriptor for the cluster communications port.
+extern int cluster_sockfd;		// The filedescriptor for the cluster communications port.
 
 in_addr_t my_address = 0;		// The network address of my ethernet port.
 static int walk_session_number = 0;	// The next session to send when doing the slow table walk.
diff --git a/fake_epoll.h b/fake_epoll.h
new file mode 100644
index 0000000..b44f846
--- /dev/null
+++ b/fake_epoll.h
@@ -0,0 +1,179 @@
+/* kludge up some limited epoll semantics using select for 2.4 kernels */
+/* $Id: fake_epoll.h,v 1.1 2005-06-04 15:42:35 bodea Exp $ */
+
+#ifndef __FAKE_EPOLL_H__
+#define __FAKE_EPOLL_H__
+
+#define EPOLLIN		0x01
+#define EPOLLOUT	0x04
+#define EPOLLERR	0x08
+#define EPOLLHUP	0x10
+
+#define EPOLL_CTL_ADD	1
+#define EPOLL_CTL_DEL	2
+#define EPOLL_CTL_MOD	3
+
+struct epoll_event {
+    uint32_t events;
+    union epoll_data {
+	void *ptr;
+	int fd;
+	uint32_t u32;
+	uint64_t u64;
+    } data;
+};
+
+int epoll_ctl(int epfd, int op, int fd, struct epoll_event *event);
+
+#ifdef FAKE_EPOLL_IMPLEMENTATION
+
+#include <sys/select.h>
+
+static fd_set _epoll_read_set;
+static fd_set _epoll_write_set;
+static int _epoll_fds;
+static struct epoll_event *_epoll_data[128];
+
+static int epoll_create(int size __attribute__ ((unused)))
+{
+    static int once = 0;
+    if (once++)
+    {
+	errno = ENFILE; /* only support one instance */
+	return -1;
+    }
+
+    FD_ZERO(&_epoll_read_set);
+    FD_ZERO(&_epoll_write_set);
+    _epoll_fds = 0;
+
+    memset(_epoll_data, 0, sizeof(_epoll_data));
+
+    return 1; /* "descriptor" */
+}
+
+int epoll_ctl(int epfd __attribute__ ((unused)), int op, int fd,
+    struct epoll_event *event)
+{
+    if (fd > (sizeof(_epoll_data)/sizeof(*_epoll_data)) - 1)
+    {
+	errno = EINVAL;
+	return -1;
+    }
+
+    switch (op)
+    {
+    case EPOLL_CTL_ADD:
+	if (event->events & EPOLLIN)
+	    FD_SET(fd, &_epoll_read_set);
+
+	if (event->events & EPOLLOUT)
+	    FD_SET(fd, &_epoll_write_set);
+
+	if (fd >= _epoll_fds)
+	    _epoll_fds = fd + 1;
+
+	if (_epoll_data[fd])
+	    free(_epoll_data[fd]);
+
+	if (!(_epoll_data[fd] = malloc(sizeof(*_epoll_data))))
+	{
+	    errno = ENOMEM;
+	    return -1;
+	}
+
+	memcpy(_epoll_data[fd], &event->data, sizeof(*_epoll_data));
+	break;
+
+    case EPOLL_CTL_MOD:
+	if (event->events & EPOLLIN)
+	    FD_SET(fd, &_epoll_read_set);
+	else
+	    FD_CLR(fd, &_epoll_read_set);
+
+	if (event->events & EPOLLOUT)
+	    FD_SET(fd, &_epoll_write_set);
+	else
+	    FD_CLR(fd, &_epoll_write_set);
+
+	memcpy(_epoll_data[fd], &event->data, sizeof(*_epoll_data));
+	break;
+
+    case EPOLL_CTL_DEL:
+	FD_CLR(fd, &_epoll_read_set);
+	FD_CLR(fd, &_epoll_write_set);
+
+	free(_epoll_data[fd]);
+	_epoll_data[fd] = 0;
+
+	if (fd == _epoll_fds - 1)
+	{
+	    _epoll_fds = 0;
+	    while (fd-- > 0)
+	    {
+		if (FD_ISSET(fd, &_epoll_read_set) ||
+		    FD_ISSET(fd, &_epoll_write_set))
+		{
+		    _epoll_fds = fd + 1;
+		    break;
+		}
+	    }
+	}
+
+	break;
+    }
+
+    return 0;
+}
+
+static int epoll_wait(int epfd __attribute__ ((unused)),
+    struct epoll_event *events, int maxevents, int timout)
+{
+    fd_set r;
+    fd_set w;
+    struct timeval t;
+    struct timeval *tp;
+    int n;
+    int e;
+    int i;
+
+    memcpy(&r, &_epoll_read_set, sizeof(r));
+    memcpy(&w, &_epoll_write_set, sizeof(w));
+
+    if (timout >= 0)
+    {
+	t.tv_sec = 0;
+	t.tv_usec = timout * 1000;
+    	tp = &t;
+    }
+    else
+	tp = 0;
+
+    n = select(_epoll_fds, &r, &w, 0, tp);
+    if (n > maxevents)
+    	n = maxevents;
+
+    for (i = e = 0; n > 0 && i < _epoll_fds; i++)
+    {
+	if (!_epoll_data[i])
+	    continue;
+
+	events[e].events = 0;
+	if (FD_ISSET(i, &r))
+	    events[e].events |= EPOLLIN;
+
+	if (FD_ISSET(i, &w))
+	    events[e].events |= EPOLLOUT;
+
+	if (events[e].events)
+	{
+	    memcpy(&events[e++].data, _epoll_data[i], sizeof(events[0].data));
+	    n--;
+	}
+    }
+
+    return e;
+}
+
+#endif /* FAKE_EPOLL_IMPLEMENTATION */
+#endif /* __FAKE_EPOLL_H__ */
diff --git a/l2tpns.c b/l2tpns.c
index 9942cf9..b0b51a2 100644
--- a/l2tpns.c
+++ b/l2tpns.c
@@ -4,7 +4,7 @@
 // Copyright (c) 2002 FireBrick (Andrews & Arnold Ltd / Watchfront Ltd) - GPL licenced
 // vim: sw=8 ts=8
 
-char const *cvs_id_l2tpns = "$Id: l2tpns.c,v 1.107 2005-06-02 11:32:30 bodea Exp $";
+char const *cvs_id_l2tpns = "$Id: l2tpns.c,v 1.108 2005-06-04 15:42:35 bodea Exp $";
 
 #include <arpa/inet.h>
 #include <assert.h>
@@ -52,7 +52,7 @@ char const *cvs_id_l2tpns = "$Id: l2tpns.c,v 1.107 2005-06-02 11:32:30 bodea Exp
 
 #ifdef BGP
 #include "bgp.h"
-#endif /* BGP */
+#endif
 
 // Globals
 configt *config = NULL;		// all configuration
@@ -64,13 +64,14 @@ int snoopfd = -1;		// UDP file handle for sending out intercept data
 int *radfds = NULL;		// RADIUS requests file handles
 int ifrfd = -1;			// File descriptor for routing, etc
 int ifr6fd = -1;		// File descriptor for IPv6 routing, etc
-static int rand_fd = -1;	// Random data source
+int rand_fd = -1;		// Random data source
+int cluster_sockfd = -1;	// Intra-cluster communications socket.
+int epollfd = -1;		// event polling
 time_t basetime = 0;		// base clock
 char hostname[1000] = "";	// us.
 static int tunidx;		// ifr_ifindex of tun device
 static int syslog_log = 0;	// are we logging to syslog
 static FILE *log_stream = 0;	// file handle for direct logging (i.e. direct into file, not via syslog).
-extern int cluster_sockfd;	// Intra-cluster communications socket.
 uint32_t last_id = 0;		// Unique ID for radius accounting
 
 struct cli_session_actions *cli_session_actions = NULL;	// Pending session changes requested by CLI
@@ -1012,7 +1013,8 @@ static void processipout(uint8_t * buf, int len)
 		if (rate++ < config->icmp_rate) // Only send a max of icmp_rate per second.
 		{
 			LOG(4, 0, 0, "IP: Sending ICMP host unreachable to %s\n", fmtaddr(*(in_addr_t *)(buf + 12), 0));
-			host_unreachable(*(in_addr_t *)(buf + 12), *(uint16_t *)(buf + 4), ip, buf, (len < 64) ? 64 : len);
+			host_unreachable(*(in_addr_t *)(buf + 12), *(uint16_t *)(buf + 4),
+				config->bind_address ? config->bind_address : my_address, buf, len);
 		}
 		return;
 	}
@@ -2850,13 +2852,13 @@ static void regular_cleanups(double period)
 static int still_busy(void)
 {
 	int i;
-	static time_t stopped_bgp = 0;
 	static clockt last_talked = 0;
 	static clockt start_busy_wait = 0;
 
 	if (!config->cluster_iam_master)
 	{
 #ifdef BGP
+		static time_t stopped_bgp = 0;
 	    	if (bgp_configured)
 		{
 			if (!stopped_bgp)
@@ -2924,41 +2926,89 @@ static int still_busy(void)
 	return 0;
 }
 
-static fd_set readset;
-static int readset_n = 0;
+#ifdef HAVE_EPOLL
+# include <sys/epoll.h>
+#else
+# define FAKE_EPOLL_IMPLEMENTATION /* include the functions */
+# include "fake_epoll.h"
+#endif
+
+// the base set of fds polled: control, cli, udp, tun, cluster
+#define BASE_FDS	5
+
+// additional polled fds
+#ifdef BGP
+# define EXTRA_FDS	BGP_NUM_PEERS
+#else
+# define EXTRA_FDS	0
+#endif
 
 // main loop - gets packets on tun or udp and processes them
 static void mainloop(void)
 {
 	int i;
 	uint8_t buf[65536];
-	struct timeval to;
 	clockt next_cluster_ping = 0;	// send initial ping immediately
+	struct epoll_event events[BASE_FDS + RADIUS_FDS + EXTRA_FDS];
+	int maxevent = sizeof(events)/sizeof(*events);
+
+	if ((epollfd = epoll_create(maxevent)) < 0)
+	{
+	    	LOG(0, 0, 0, "epoll_create failed: %s\n", strerror(errno));
+		exit(1);
+	}
 
 	LOG(4, 0, 0, "Beginning of main loop.  udpfd=%d, tunfd=%d, cluster_sockfd=%d, controlfd=%d\n",
 		udpfd, tunfd, cluster_sockfd, controlfd);
 
-	FD_ZERO(&readset);
-	FD_SET(udpfd, &readset);
-	FD_SET(tunfd, &readset);
-	FD_SET(controlfd, &readset);
-	FD_SET(clifd, &readset);
-	if (cluster_sockfd) FD_SET(cluster_sockfd, &readset);
-	readset_n = udpfd;
-	if (tunfd > readset_n)          readset_n = tunfd;
-	if (controlfd > readset_n)      readset_n = controlfd;
-	if (clifd > readset_n)          readset_n = clifd;
-	if (cluster_sockfd > readset_n) readset_n = cluster_sockfd;
-
-	while (!main_quit || still_busy())
+	/* setup our fds to poll for input */
 	{
-		fd_set r;
-		int n = readset_n;
+		static struct event_data d[BASE_FDS];
+		struct epoll_event e;
+
+		e.events = EPOLLIN;
+		i = 0;
+
+		d[i].type = FD_TYPE_CONTROL;
+		e.data.ptr = &d[i++];
+		epoll_ctl(epollfd, EPOLL_CTL_ADD, controlfd, &e);
+
+		d[i].type = FD_TYPE_CLI;
+		e.data.ptr = &d[i++];
+		epoll_ctl(epollfd, EPOLL_CTL_ADD, clifd, &e);
+
+		d[i].type = FD_TYPE_UDP;
+		e.data.ptr = &d[i++];
+		epoll_ctl(epollfd, EPOLL_CTL_ADD, udpfd, &e);
+
+		d[i].type = FD_TYPE_TUN;
+		e.data.ptr = &d[i++];
+		epoll_ctl(epollfd, EPOLL_CTL_ADD, tunfd, &e);
+
+		d[i].type = FD_TYPE_CLUSTER;
+		e.data.ptr = &d[i++];
+		epoll_ctl(epollfd, EPOLL_CTL_ADD, cluster_sockfd, &e);
+	}
+
 #ifdef BGP
-		fd_set w;
-		int bgp_set[BGP_NUM_PEERS];
+	signal(SIGPIPE, SIG_IGN);
+	bgp_setup(config->as_number);
+	if (config->bind_address)
+		bgp_add_route(config->bind_address, 0xffffffff);
+
+	for (i = 0; i < BGP_NUM_PEERS; i++)
+	{
+		if (config->neighbour[i].name[0])
+			bgp_start(&bgp_peers[i], config->neighbour[i].name,
+				config->neighbour[i].as, config->neighbour[i].keepalive,
+				config->neighbour[i].hold, 0); /* 0 = routing disabled */
+	}
 #endif /* BGP */
+
+	while (!main_quit || still_busy())
+	{
 		int more = 0;
+		int n;
 
 		if (config->reload_config)
 		{
@@ -2966,35 +3016,11 @@ static void mainloop(void)
 			update_config();
 		}
 
-		memcpy(&r, &readset, sizeof(fd_set));
-		to.tv_sec = 0;
-		to.tv_usec = 100000; // 1/10th of a second.
-
 #ifdef BGP
-		FD_ZERO(&w);
-		for (i = 0; i < BGP_NUM_PEERS; i++)
-		{
-			bgp_set[i] = bgp_select_state(&bgp_peers[i]);
-			if (bgp_set[i] & 1)
-			{
-				FD_SET(bgp_peers[i].sock, &r);
-				if (bgp_peers[i].sock > n)
-					n = bgp_peers[i].sock;
-			}
-
-			if (bgp_set[i] & 2)
-			{
-				FD_SET(bgp_peers[i].sock, &w);
-				if (bgp_peers[i].sock > n)
-					n = bgp_peers[i].sock;
-			}
-		}
-
-		n = select(n + 1, &r, &w, 0, &to);
-#else /* BGP */
-		n = select(n + 1, &r, 0, 0, &to);
+		bgp_set_poll();
 #endif /* BGP */
 
+		n = epoll_wait(epollfd, events, maxevent, 100); // timeout 100ms (1/10th sec)
 		STAT(select_called);
 
 		TIME = now(NULL);
@@ -3008,67 +3034,83 @@ static void mainloop(void)
 			main_quit++;
 			break;
 		}
-		else if (n)
+
+		if (n)
 		{
 			struct sockaddr_in addr;
 			int alen, c, s;
+			int udp_ready = 0;
+			int tun_ready = 0;
+			int cluster_ready = 0;
 			int udp_pkts = 0;
 			int tun_pkts = 0;
 			int cluster_pkts = 0;
+#ifdef BGP
+			uint32_t bgp_events[BGP_NUM_PEERS];
+			memset(bgp_events, 0, sizeof(bgp_events));
+#endif /* BGP */
 
-			// nsctl commands
-			if (FD_ISSET(controlfd, &r))
+			for (c = n, i = 0; i < c; i++)
 			{
-				alen = sizeof(addr);
-				processcontrol(buf, recvfrom(controlfd, buf, sizeof(buf), MSG_WAITALL, (void *) &addr, &alen), &addr, alen);
-				n--;
-			}
+				struct event_data *d = events[i].data.ptr;
+				switch (d->type)
+				{
+				case FD_TYPE_CONTROL: // nsctl commands
+					alen = sizeof(addr);
+					processcontrol(buf, recvfrom(controlfd, buf, sizeof(buf), MSG_WAITALL, (void *) &addr, &alen), &addr, alen);
+					n--;
+					break;
 
-			// RADIUS responses
-			if (config->cluster_iam_master)
-			{
-				for (i = 0; i < config->num_radfds; i++)
+				case FD_TYPE_CLI: // CLI connections
 				{
-					if (FD_ISSET(radfds[i], &r))
+					int cli;
+					
+					alen = sizeof(addr);
+					if ((cli = accept(clifd, (struct sockaddr *)&addr, &alen)) >= 0)
 					{
-						processrad(buf, recv(radfds[i], buf, sizeof(buf), 0), i);
-						n--;
+						cli_do(cli);
+						close(cli);
 					}
-				}
-			}
+					else
+						LOG(0, 0, 0, "accept error: %s\n", strerror(errno));
 
-			// CLI connections
-			if (FD_ISSET(clifd, &r))
-			{
-				int cli;
-				
-				alen = sizeof(addr);
-				if ((cli = accept(clifd, (struct sockaddr *)&addr, &alen)) >= 0)
-				{
-					cli_do(cli);
-					close(cli);
+					n--;
+					break;
 				}
-				else
-					LOG(0, 0, 0, "accept error: %s\n", strerror(errno));
 
-				n--;
-			}
+				// these are handled below, with multiple interleaved reads
+				case FD_TYPE_UDP:	udp_ready++; break;
+				case FD_TYPE_TUN:	tun_ready++; break;
+				case FD_TYPE_CLUSTER:	cluster_ready++; break;
+
+				case FD_TYPE_RADIUS: // RADIUS response
+					s = recv(radfds[d->index], buf, sizeof(buf), 0);
+					if (s >= 0 && config->cluster_iam_master)
+						processrad(buf, s, d->index);
+
+					n--;
+					break;
 
 #ifdef BGP
-			for (i = 0; i < BGP_NUM_PEERS; i++)
-			{
-				int isr = bgp_set[i] ? FD_ISSET(bgp_peers[i].sock, &r) : 0;
-				int isw = bgp_set[i] ? FD_ISSET(bgp_peers[i].sock, &w) : 0;
-				bgp_process(&bgp_peers[i], isr, isw);
-				if (isr) n--;
-				if (isw) n--;
+				case FD_TYPE_BGP:
+				    	bgp_events[d->index] = events[i].events;
+				    	n--;
+					break;
+#endif /* BGP */
+
+				default:
+				    	LOG(0, 0, 0, "Unexpected fd type returned from epoll_wait: %d\n", d->type);
+				}
 			}
+
+#ifdef BGP
+			bgp_process(bgp_events);
 #endif /* BGP */
 
 			for (c = 0; n && c < config->multi_read_count; c++)
 			{
 				// L2TP
-				if (FD_ISSET(udpfd, &r))
+				if (udp_ready)
 				{
 					alen = sizeof(addr);
 					if ((s = recvfrom(udpfd, buf, sizeof(buf), 0, (void *) &addr, &alen)) > 0)
@@ -3078,13 +3120,13 @@ static void mainloop(void)
 					}
 					else
 					{
-						FD_CLR(udpfd, &r);
+						udp_ready = 0;
 						n--;
 					}
 				}
 
 				// incoming IP
-				if (FD_ISSET(tunfd, &r))
+				if (tun_ready)
 				{
 					if ((s = read(tunfd, buf, sizeof(buf))) > 0)
 					{
@@ -3093,13 +3135,13 @@ static void mainloop(void)
 					}
 					else
 					{
-						FD_CLR(tunfd, &r);
+						tun_ready = 0;
 						n--;
 					}
 				}
 
 				// cluster
-				if (FD_ISSET(cluster_sockfd, &r))
+				if (cluster_ready)
 				{
 					alen = sizeof(addr);
 					if ((s = recvfrom(cluster_sockfd, buf, sizeof(buf), MSG_WAITALL, (void *) &addr, &alen)) > 0)
@@ -3109,7 +3151,7 @@ static void mainloop(void)
 					}
 					else
 					{
-						FD_CLR(cluster_sockfd, &r);
+						cluster_ready = 0;
 						n--;
 					}
 				}
@@ -3129,7 +3171,7 @@ static void mainloop(void)
 		}
 
 			// Runs on every machine (master and slaves).
-		if (cluster_sockfd && next_cluster_ping <= TIME)
+		if (next_cluster_ping <= TIME)
 		{
 			// Check to see which of the cluster is still alive..
 
@@ -3707,10 +3749,12 @@ static int dump_session(FILE **f, sessiont *s)
 		LOG(3, 0, 0, "Dumping accounting information to %s\n", filename);
 		fprintf(*f, "# dslwatch.pl dump file V1.01\n"
 			"# host: %s\n"
+			"# endpoint: %s\n"
 			"# time: %ld\n"
 			"# uptime: %ld\n"
 			"# format: username ip qos uptxoctets downrxoctets\n",
 			hostname,
+			fmtaddr(config->bind_address ? config->bind_address : my_address, 0),
 			now,
 			now - basetime);
 	}
@@ -3851,19 +3895,6 @@ int main(int argc, char *argv[])
 	if (cluster_init() < 0)
 		exit(1);
 
-#ifdef BGP
-	signal(SIGPIPE, SIG_IGN);
-	bgp_setup(config->as_number);
-	bgp_add_route(config->bind_address, 0xffffffff);
-	for (i = 0; i < BGP_NUM_PEERS; i++)
-	{
-		if (config->neighbour[i].name[0])
-			bgp_start(&bgp_peers[i], config->neighbour[i].name,
-				config->neighbour[i].as, config->neighbour[i].keepalive,
-				config->neighbour[i].hold, 0); /* 0 = routing disabled */
-	}
-#endif /* BGP */
-
 	inittun();
 	LOG(1, 0, 0, "Set up on interface %s\n", config->tundevice);
 
@@ -4094,8 +4125,6 @@ static void update_config()
 	if (!config->numradiusservers)
 		LOG(0, 0, 0, "No RADIUS servers defined!\n");
 
-	config->num_radfds = 1 << RADIUS_SHIFT;
-
 	// parse radius_authtypes_s
 	config->radius_authtypes = config->radius_authprefer = 0;
 	p = config->radius_authtypes_s;
@@ -4853,6 +4882,9 @@ static tunnelidt new_tunnel()
 void become_master(void)
 {
 	int s, i;
+	static struct event_data d[RADIUS_FDS];
+	struct epoll_event e;
+
 	run_plugins(PLUGIN_BECOME_MASTER, NULL);
 
 	// running a bunch of iptables commands is slow and can cause
@@ -4871,11 +4903,14 @@ void become_master(void)
 	}
 
 	// add radius fds
-	for (i = 0; i < config->num_radfds; i++)
+	e.events = EPOLLIN;
+	for (i = 0; i < RADIUS_FDS; i++)
 	{
-		FD_SET(radfds[i], &readset);
-		if (radfds[i] > readset_n)
-			readset_n = radfds[i];
+	    	d[i].type = FD_TYPE_RADIUS;
+		d[i].index = i;
+		e.data.ptr = &d[i];
+
+		epoll_ctl(epollfd, EPOLL_CTL_ADD, radfds[i], &e);
 	}
 }
 
diff --git a/l2tpns.h b/l2tpns.h
index 040861c..6d820b5 100644
--- a/l2tpns.h
+++ b/l2tpns.h
@@ -1,5 +1,5 @@
 // L2TPNS Global Stuff
-// $Id: l2tpns.h,v 1.75 2005-06-02 11:32:31 bodea Exp $
+// $Id: l2tpns.h,v 1.76 2005-06-04 15:42:36 bodea Exp $
 
 #ifndef __L2TPNS_H__
 #define __L2TPNS_H__
@@ -23,6 +23,7 @@
 #define MAXTBFS		6000		// Maximum token bucket filters. Might need up to 2 * session.
 
 #define RADIUS_SHIFT	6
+#define	RADIUS_FDS	(1 << RADIUS_SHIFT)
 #define RADIUS_MASK	((1 << RADIUS_SHIFT) - 1)
 #define MAXRADIUS	(1 << (8 + RADIUS_SHIFT))
 
@@ -455,7 +456,6 @@ typedef struct
 	in_addr_t	radiusserver[MAXRADSERVER];	// radius servers
 	uint16_t	radiusport[MAXRADSERVER];	// radius base ports
 	uint8_t		numradiusservers;		// radius server count
-	short		num_radfds;			// Number of radius filehandles allocated
 
 	char		radius_authtypes_s[32];		// list of valid authentication types (chap, pap) in order of preference
 	int		radius_authtypes;
@@ -692,7 +692,20 @@ extern struct Tstats *_statistics;
 extern in_addr_t my_address;
 extern int tun_write(uint8_t *data, int size);
 extern int clifd;
-
+extern int epollfd;
+
+struct event_data {
+	enum {
+	    	FD_TYPE_CONTROL,
+	    	FD_TYPE_CLI,
+	    	FD_TYPE_UDP,
+	    	FD_TYPE_TUN,
+	    	FD_TYPE_CLUSTER,
+		FD_TYPE_RADIUS,
+		FD_TYPE_BGP,
+	} type;
+	int index; // for RADIUS, BGP
+};
 
 #define TIME (config->current_time)
 
diff --git a/l2tpns.spec b/l2tpns.spec
index 2503604..4265858 100644
--- a/l2tpns.spec
+++ b/l2tpns.spec
@@ -43,5 +43,5 @@ rm -rf %{buildroot}
 %attr(644,root,root) /usr/share/man/man[58]/*
 
 %changelog
-* Thu Jun 2 2005 Brendan O'Dea <bod@optusnet.com.au> 2.1.0-1
+* Sun Jun 5 2005 Brendan O'Dea <bod@optusnet.com.au> 2.1.0-1
 - 2.1.0 release, see /usr/share/doc/l2tpns-2.1.0/Changes
diff --git a/radius.c b/radius.c
index 0569c20..f0500b8 100644
--- a/radius.c
+++ b/radius.c
@@ -1,6 +1,6 @@
 // L2TPNS Radius Stuff
 
-char const *cvs_id_radius = "$Id: radius.c,v 1.32 2005-06-02 11:32:32 bodea Exp $";
+char const *cvs_id_radius = "$Id: radius.c,v 1.33 2005-06-04 15:42:36 bodea Exp $";
 
 #include <time.h>
 #include <stdio.h>
@@ -29,9 +29,9 @@ extern ip_filtert *ip_filters;
 void initrad(void)
 {
 	int i;
-	LOG(3, 0, 0, "Creating %d sockets for RADIUS queries\n", config->num_radfds);
-	radfds = calloc(sizeof(int), config->num_radfds);
-	for (i = 0; i < config->num_radfds; i++)
+	LOG(3, 0, 0, "Creating %d sockets for RADIUS queries\n", RADIUS_FDS);
+	radfds = calloc(sizeof(int), RADIUS_FDS);
+	for (i = 0; i < RADIUS_FDS; i++)
 	{
 		int flags;
 		radfds[i] = socket(AF_INET, SOCK_DGRAM, IPPROTO_UDP);
diff --git a/util.c b/util.c
index 205fe03..0ba92ec 100644
--- a/util.c
+++ b/util.c
@@ -1,6 +1,6 @@
 /* Misc util functions */
 
-char const *cvs_id_util = "$Id: util.c,v 1.10 2005-01-25 04:19:07 bodea Exp $";
+char const *cvs_id_util = "$Id: util.c,v 1.11 2005-06-04 15:42:36 bodea Exp $";
 
 #include <unistd.h>
 #include <errno.h>
@@ -40,7 +40,7 @@ void *shared_malloc(unsigned int size)
 }
 
 extern int forked;
-extern int udpfd, controlfd, tunfd, snoopfd, ifrfd, ifr6fd, cluster_sockfd;
+extern int udpfd, controlfd, tunfd, snoopfd, ifrfd, ifr6fd, rand_fd, cluster_sockfd;
 extern int *radfds;
 
 pid_t fork_and_close()
@@ -65,7 +65,7 @@ pid_t fork_and_close()
 
 	signal(SIGPIPE, SIG_DFL);
 	signal(SIGCHLD, SIG_DFL);
-	signal(SIGHUP, SIG_DFL);
+	signal(SIGHUP,  SIG_DFL);
 	signal(SIGUSR1, SIG_DFL);
 	signal(SIGQUIT, SIG_DFL);
 	signal(SIGKILL, SIG_DFL);
@@ -79,10 +79,12 @@ pid_t fork_and_close()
 	if (snoopfd != -1)        close(snoopfd);
 	if (ifrfd != -1)          close(ifrfd);
 	if (ifr6fd != -1)         close(ifr6fd);
+	if (rand_fd != -1)        close(rand_fd);
 	if (cluster_sockfd != -1) close(cluster_sockfd);
 	if (clifd != -1)          close(clifd);
+	if (epollfd != -1)        close(epollfd);
 
-	for (i = 0; radfds && i < config->num_radfds; i++)
+	for (i = 0; radfds && i < RADIUS_FDS; i++)
 		close(radfds[i]);
 #ifdef BGP
 	for (i = 0; i < BGP_NUM_PEERS; i++)