Added epoll support on linux (#3798)

Added epoll event dispatching support on linux systems.
You can enable it by adding the configure option --enable-epoll

Credits to Hercules.

conf/packet_athena.conf

@@ -5,6 +5,19 @@
 // Display debug reports (When something goes wrong during the report, the report is saved.)
 debug: no
 
+// Linux/Epoll: Maximum Events per cycle
+// Default Value:
+//	(Maximum Supported Connections)/2
+// NOTE: this controls the maximum collected socket-events per-cycle (call to epoll_wait())
+//       for example settings this to 32 will allow up to 32 events (incoming data/new connections
+//       per server-cycle.
+// NOTE: Recommended Settings is at least half the maximum supported connections
+//       Settings this to a lower value, may cause lags/delays
+//       Depending on available CPU Time
+// NOTE: This Setting is only available on Linux when build using EPoll as event dispatcher!
+//
+//epoll_maxevents: 1024
+
 // How long can a socket stall before closing the connection (in seconds)
 stall_time: 60
 
@@ -22,7 +35,7 @@ enable_ip_rules: yes
 
 order: deny,allow
 // order: allow,deny
-// order: mutual-failture
+// order: mutual-failure
 
 // IP rules
 //   allow : Accepts connections from the ip range (even if flagged as DDoS)
@@ -45,7 +58,7 @@ order: deny,allow
 ddos_interval: 3000
 
 // Consecutive attempts trigger
-// (default is 5 attemps)
+// (default is 5 attempts)
 ddos_count: 5
 
 // The time interval after which the threat of DDoS is assumed to be gone. (msec)

configure

@@ -698,6 +698,7 @@ ac_user_opts='
 enable_option_checking
 enable_manager
 enable_packetver
+enable_epoll
 enable_debug
 enable_prere
 enable_vip
@@ -1339,6 +1340,7 @@ Optional Features:
   --enable-manager=ARG    memory managers: no, builtin, memwatch, dmalloc,
                           gcollect, bcheck (defaults to builtin)
   --enable-packetver=ARG  Sets the PACKETVER define. (see src/common/mmo.h)
+  --enable-epoll          use epoll(4) on Linux
   --enable-debug[=ARG]    Compiles extra debug code. (disabled by default)
                           (available options: yes, no, gdb)
   --enable-prere[=ARG]    Compiles serv in prere mode. (disabled by default)
@@ -3236,6 +3238,55 @@ fi
 
 
 
+#
+# Epoll
+#
+# Check whether --enable-epoll was given.
+if test "${enable_epoll+set}" = set; then :
+  enableval=$enable_epoll; enable_epoll=$enableval
+else
+  enable_epoll=no
+
+fi
+
+if test x$enable_epoll = xno; then
+	have_linux_epoll=no
+else
+	{ $as_echo "$as_me:${as_lineno-$LINENO}: checking for Linux epoll(4)" >&5
+$as_echo_n "checking for Linux epoll(4)... " >&6; }
+	cat confdefs.h - <<_ACEOF >conftest.$ac_ext
+/* end confdefs.h.  */
+
+		#ifndef __linux__
+		#error This is not Linux
+		#endif
+		#include <sys/epoll.h>
+
+int
+main ()
+{
+epoll_create1 (EPOLL_CLOEXEC);
+  ;
+  return 0;
+}
+_ACEOF
+if ac_fn_cxx_try_link "$LINENO"; then :
+  have_linux_epoll=yes
+else
+  have_linux_epoll=no
+
+fi
+rm -f core conftest.err conftest.$ac_objext \
+    conftest$ac_exeext conftest.$ac_ext
+	{ $as_echo "$as_me:${as_lineno-$LINENO}: result: $have_linux_epoll" >&5
+$as_echo "$have_linux_epoll" >&6; }
+fi
+if test x$enable_epoll,$have_linux_epoll = xyes,no; then
+    as_fn_error $? "epoll support explicitly enabled but not available" "$LINENO" 5
+fi
+
+
+
 #
 # debug
 #
@@ -5920,6 +5971,19 @@ if test -n "$enable_packetver" ; then
 fi
 
 
+#
+# Epoll
+#
+case $have_linux_epoll in
+	"yes")
+		CPPFLAGS="$CPPFLAGS -DSOCKET_EPOLL"
+		;;
+	"no")
+		# default value
+		;;
+esac
+
+
 #
 # Debug
 #

configure.in

@@ -53,6 +53,40 @@ AC_ARG_ENABLE(
 )
 
 
+#
+# Epoll
+#
+AC_ARG_ENABLE(
+	[epoll],
+	AC_HELP_STRING(
+		[--enable-epoll],
+		[use epoll(4) on Linux]
+	),
+	[enable_epoll=$enableval],
+	[enable_epoll=no]
+)
+if test x$enable_epoll = xno; then
+	have_linux_epoll=no
+else
+	AC_MSG_CHECKING([for Linux epoll(4)])
+	AC_LINK_IFELSE([AC_LANG_PROGRAM(
+		[
+		#ifndef __linux__
+		#error This is not Linux
+		#endif
+		#include <sys/epoll.h>
+		],
+		[epoll_create1 (EPOLL_CLOEXEC);])],
+		[have_linux_epoll=yes],
+		[have_linux_epoll=no]
+	)
+	AC_MSG_RESULT([$have_linux_epoll])
+fi
+if test x$enable_epoll,$have_linux_epoll = xyes,no; then
+	AC_MSG_ERROR([epoll support explicitly enabled but not available])
+fi
+
+
 #
 # debug
 #
@@ -985,6 +1019,19 @@ if test -n "$enable_packetver" ; then
 fi
 
 
+#
+# Epoll
+#
+case $have_linux_epoll in
+	"yes")
+		CPPFLAGS="$CPPFLAGS -DSOCKET_EPOLL"
+		;;
+	"no")
+		# default value
+		;;
+esac
+
+
 #
 # Debug
 #

src/common/socket.cpp

@@ -9,22 +9,34 @@
 	#include "winapi.hpp"
 #else
 	#include <errno.h>
-#include <netinet/tcp.h>
+	#include <arpa/inet.h>
 	#include <net/if.h>
-	#include <unistd.h>
-#include <sys/ioctl.h>
 	#include <netdb.h>
-	#include <arpa/inet.h>
+	#include <sys/ioctl.h>
+	#include <sys/socket.h>
+	#include <sys/time.h>
+	#include <unistd.h>
+
+	#if defined(__linux__) || defined(__linux)
+		#include <linux/tcp.h>
+
+		#ifdef SOCKET_EPOLL
+			#include <sys/epoll.h>
+		#endif
+	#else 
+		#include <netinet/in.h>
+		#include <netinet/tcp.h>
+	#endif
 
 	#ifndef SIOCGIFCONF
-	#include <sys/sockio.h> // SIOCGIFCONF on Solaris, maybe others? [Shinomori]
+		#include <sys/sockio.h> // SIOCGIFCONF on Solaris, maybe others? [Shinomori]
 	#endif
 	#ifndef FIONBIO
-	#include <sys/filio.h> // FIONBIO on Solaris [FlavioJS]
+		#include <sys/filio.h> // FIONBIO on Solaris [FlavioJS]
 	#endif
 
 	#ifdef HAVE_SETRLIMIT
-	#include <sys/resource.h>
+		#include <sys/resource.h>
 	#endif
 #endif
 
@@ -203,7 +215,17 @@ char* sErr(int code)
 	#define MSG_NOSIGNAL 0
 #endif
 
-fd_set readfds;
+#ifndef SOCKET_EPOLL
+	// Select based Event Dispatcher
+	fd_set readfds;
+#else
+	// Epoll based Event Dispatcher
+	static int epoll_maxevents = (FD_SETSIZE / 2);
+	static int epfd = SOCKET_ERROR;
+	static struct epoll_event epevent;
+	static struct epoll_event *epevents = nullptr;
+#endif
+
 int fd_max;
 time_t last_tick;
 time_t stall_time = 60;
@@ -478,8 +500,22 @@ int connect_client(int listen_fd)
 	}
 #endif
 
-	if( fd_max <= fd ) fd_max = fd + 1;
+#ifndef SOCKET_EPOLL
+	// Select Based Event Dispatcher
 	sFD_SET(fd,&readfds);
+#else
+	// Epoll based Event Dispatcher
+	epevent.data.fd = fd;
+	epevent.events = EPOLLIN;
+
+	if( epoll_ctl( epfd, EPOLL_CTL_ADD, fd, &epevent ) == SOCKET_ERROR ){
+		ShowError( "connect_client: Failed to add to epoll event dispatcher for new socket #%d: %s\n", fd, error_msg() );
+		sClose( fd );
+		return -1;
+	}
+#endif
+
+	if( fd_max <= fd ) fd_max = fd + 1;
 
 	create_session(fd, recv_to_fifo, send_from_fifo, default_func_parse);
 	session[fd]->client_addr = ntohl(client_address.sin_addr.s_addr);
@@ -531,8 +567,22 @@ int make_listen_bind(uint32 ip, uint16 port)
 		exit(EXIT_FAILURE);
 	}
 
-	if(fd_max <= fd) fd_max = fd + 1;
+#ifndef SOCKET_EPOLL
+	// Select Based Event Dispatcher
 	sFD_SET(fd, &readfds);
+#else
+	// Epoll based Event Dispatcher
+	epevent.data.fd = fd;
+	epevent.events = EPOLLIN;
+
+	if( epoll_ctl( epfd, EPOLL_CTL_ADD, fd, &epevent ) == SOCKET_ERROR ){
+		ShowError( "make_listen_bind: failed to add listener socket #%d to epoll event dispatcher: %s\n", fd, error_msg() );
+		sClose(fd);
+		exit(EXIT_FAILURE);
+	}
+#endif
+
+	if(fd_max <= fd) fd_max = fd + 1;
 
 	create_session(fd, connect_client, null_send, null_parse);
 	session[fd]->client_addr = 0; // just listens
@@ -643,8 +693,22 @@ int make_connection(uint32 ip, uint16 port, bool silent,int timeout) {
 	set_nonblocking(fd, 1);
 #endif
 
-	if (fd_max <= fd) fd_max = fd + 1;
+#ifndef SOCKET_EPOLL
+	// Select Based Event Dispatcher
 	sFD_SET(fd,&readfds);
+#else
+	// Epoll based Event Dispatcher
+	epevent.data.fd = fd;
+	epevent.events = EPOLLIN;
+
+	if( epoll_ctl( epfd, EPOLL_CTL_ADD, fd, &epevent ) == SOCKET_ERROR ){
+		ShowError( "make_connection: failed to add socket #%d to epoll event dispatcher: %s\n", fd, error_msg() );
+		sClose(fd);
+		return -1;
+	}
+#endif
+
+	if (fd_max <= fd) fd_max = fd + 1;
 
 	create_session(fd, recv_to_fifo, send_from_fifo, default_func_parse);
 	session[fd]->client_addr = ntohl(remote_address.sin_addr.s_addr);
@@ -821,8 +885,10 @@ int WFIFOSET(int fd, size_t len)
 
 int do_sockets(t_tick next)
 {
+#ifndef SOCKET_EPOLL
 	fd_set rfd;
 	struct timeval timeout;
+#endif
 	int ret,i;
 
 	// PRESEND Timers are executed before do_sendrecv and can send packets and/or set sessions to eof.
@@ -840,6 +906,9 @@ int do_sockets(t_tick next)
 	}
 #endif
 
+#ifndef SOCKET_EPOLL
+	// Select based Event Dispatcher
+
 	// can timeout until the next tick
 	timeout.tv_sec  = (long)(next/1000);
 	timeout.tv_usec = (long)(next%1000*1000);
@@ -856,6 +925,20 @@ int do_sockets(t_tick next)
 		}
 		return 0; // interrupted by a signal, just loop and try again
 	}
+#else
+	// Epoll based Event Dispatcher
+
+	ret = epoll_wait( epfd, epevents, epoll_maxevents, next );
+
+	if( ret == SOCKET_ERROR ){
+		if( sErrno != S_EINTR ){
+			ShowFatalError( "do_sockets: epoll_wait() failed, %s!\n", error_msg() );
+			exit( EXIT_FAILURE );
+		}
+
+		return 0; // interrupted by a signal, just loop and try again
+	}
+#endif
 
 	last_tick = time(NULL);
 
@@ -867,6 +950,26 @@ int do_sockets(t_tick next)
 		if( session[fd] )
 			session[fd]->func_recv(fd);
 	}
+#elif defined(SOCKET_EPOLL)
+	// epoll based selection
+
+	for( i = 0; i < ret; i++ ){
+		struct epoll_event *it = &epevents[i];
+		int fd = it->data.fd;
+		struct socket_data *sock = session[fd];
+
+		if( !sock ){
+			continue;
+		}
+
+		if( ( it->events & (EPOLLERR|EPOLLHUP) ) || !( it->events & EPOLLIN ) ){
+			// Got Error on this connection
+			set_eof( fd );
+		}else if( it->events & EPOLLIN ){
+			// data waiting
+			sock->func_recv( fd );
+		}
+	}
 #else
 	// otherwise assume that the fd_set is a bit-array and enumerate it in a standard way
 	for( i = 1; ret && i < fd_max; ++i )
@@ -1238,6 +1341,17 @@ int socket_config_read(const char* cfgName)
 			ddos_autoreset = atoi(w2);
 		else if (!strcmpi(w1,"debug"))
 			access_debug = config_switch(w2);
+#ifdef SOCKET_EPOLL
+		else if( !strcmpi( w1, "epoll_maxevents" ) ){
+			epoll_maxevents = atoi(w2);
+
+			// minimum that seems to be useful
+			if( epoll_maxevents < 16 ){
+				ShowWarning( "socket_config_read: epoll_maxevents is set too low. Defaulting to 16...\n" );
+				epoll_maxevents = 16;
+			}
+		}
+#endif
 #endif
 		else if (!strcmpi(w1, "import"))
 			socket_config_read(w2);
@@ -1287,6 +1401,16 @@ void socket_final(void)
 	if( WSACleanup() != 0 ){
 		ShowError("socket_final: WinSock could not be cleaned up! %s\n", error_msg() );
 	}
+#elif defined(SOCKET_EPOLL)
+	if( epfd != SOCKET_ERROR ){
+		sClose(epfd);
+		epfd = SOCKET_ERROR;
+	}
+
+	if( epevents != nullptr ){
+		aFree( epevents );
+		epevents = nullptr;
+	}
 #endif
 }
 
@@ -1297,7 +1421,17 @@ void do_close(int fd)
 		return;// invalid
 
 	flush_fifo(fd); // Try to send what's left (although it might not succeed since it's a nonblocking socket)
+
+#ifndef SOCKET_EPOLL
+	// Select based Event Dispatcher
 	sFD_CLR(fd, &readfds);// this needs to be done before closing the socket
+#else
+	// Epoll based Event Dispatcher
+	epevent.data.fd = fd;
+	epevent.events = EPOLLIN;
+	epoll_ctl( epfd, EPOLL_CTL_DEL, fd, &epevent ); // removing the socket from epoll when it's being closed is not required but recommended
+#endif
+
 	sShutdown(fd, SHUT_RDWR); // Disallow further reads/writes
 	sClose(fd); // We don't really care if these closing functions return an error, we are just shutting down and not reusing this socket.
 	if (session[fd]) delete_session(fd);
@@ -1442,7 +1576,25 @@ void socket_init(void)
 	// Get initial local ips
 	naddr_ = socket_getips(addr_,16);
 
+#ifndef SOCKET_EPOLL
+	// Select based Event Dispatcher:
 	sFD_ZERO(&readfds);
+	ShowInfo( "Server uses '" CL_WHITE "select" CL_RESET "' as event dispatcher\n" );
+#else
+	// Epoll based Event Dispatcher
+	epfd = epoll_create( FD_SETSIZE ); // 2.6.8 or newer ignores the expected socket amount argument
+
+	if( epfd == SOCKET_ERROR ){
+		ShowError( "Failed to create epoll event dispatcher: %s\n", error_msg() );
+		exit( EXIT_FAILURE );
+	}
+
+	memset( &epevent, 0x00, sizeof( struct epoll_event ) );
+	epevents = (struct epoll_event *)aCalloc( epoll_maxevents, sizeof( struct epoll_event ) );
+
+	ShowInfo( "Server uses '" CL_WHITE "epoll" CL_RESET "' with up to " CL_WHITE "%d" CL_RESET " events per cycle as event dispatcher\n", epoll_maxevents );
+#endif
+
 #if defined(SEND_SHORTLIST)
 	memset(send_shortlist_set, 0, sizeof(send_shortlist_set));
 #endif