tcp_main.c
5b532c7f
 /*
  * $Id$
  *
53c7e0f1
  * Copyright (C) 2001-2003 FhG Fokus
5b532c7f
  *
  * This file is part of ser, a free SIP server.
  *
  * ser is free software; you can redistribute it and/or modify
  * it under the terms of the GNU General Public License as published by
  * the Free Software Foundation; either version 2 of the License, or
  * (at your option) any later version
  *
  * For a license to use the ser software under conditions
  * other than those described here, or to purchase support for this
  * software, please contact iptel.org by e-mail at the following addresses:
  *    info@iptel.org
  *
  * ser is distributed in the hope that it will be useful,
  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  * GNU General Public License for more details.
  *
  * You should have received a copy of the GNU General Public License
  * along with this program; if not, write to the Free Software
  * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
  */
b33736bc
 /*
  * History:
  * --------
  *  2002-11-29  created by andrei
  *  2002-12-11  added tcp_send (andrei)
  *  2003-01-20  locking fixes, hashtables (andrei)
  *  2003-02-20  s/lock_t/gen_lock_t/ to avoid a conflict on solaris (andrei)
  *  2003-02-25  Nagle is disabled if -DDISABLE_NAGLE (andrei)
e6509c23
  *  2003-03-29  SO_REUSEADDR before calling bind to allow
  *              server restart, Nagle set on the (hopefuly) 
f15bede1
  *              correct socket (jiri)
  *  2003-03-31  always try to find the corresponding tcp listen socket for
  *               a temp. socket and store in in *->bind_address: added
  *               find_tcp_si, modified tcpconn_connect (andrei)
39546e5f
  *  2003-04-14  set sockopts to TOS low delay (andrei)
f6e50f08
  *  2003-06-30  moved tcp new connect checking & handling to
  *               handle_new_connect (andrei)
f535fd9a
  *  2003-07-09  tls_close called before closing the tcp connection (andrei)
9f4c52ce
  *  2003-10-24  converted to the new socket_info lists (andrei)
59653eb8
  *  2003-10-27  tcp port aliases support added (andrei)
06aaa54f
  *  2003-11-04  always lock before manipulating refcnt; sendchild
  *              does not inc refcnt by itself anymore (andrei)
  *  2003-11-07  different unix sockets are used for fd passing
  *              to/from readers/writers (andrei)
76b0ff49
  *  2003-11-17  handle_new_connect & tcp_connect will close the 
  *              new socket if tcpconn_new return 0 (e.g. out of mem) (andrei)
c55f3308
  *  2003-11-28  tcp_blocking_write & tcp_blocking_connect added (andrei)
a35ad52c
  *  2004-11-08  dropped find_tcp_si and replaced with find_si (andrei)
0ba367ec
  *  2005-06-07  new tcp optimized code, supports epoll (LT), sigio + real time
  *               signals, poll & select (andrei)
741a9937
  *  2005-06-26  *bsd kqueue support (andrei)
9eda5956
  *  2005-07-04  solaris /dev/poll support (andrei)
55d8155e
  *  2005-07-08  tcp_max_connections, tcp_connection_lifetime, don't accept
  *               more connections if tcp_max_connections is exceeded (andrei)
32f8b7ed
  *  2005-10-21  cleanup all the open connections on exit
98f3d5e2
  *              decrement the no. of open connections on timeout too    (andrei) *  2006-01-30  queue send_fd request and execute them at the end of the
  *              poll loop  (#ifdef) (andrei)
  *              process all children requests, before attempting to send
  *              them new stuff (fixes some deadlocks) (andrei)
1ce28841
  *  2006-02-03  timers are run only once per s (andrei)
746f7674
  *              tcp children fds can be non-blocking; send fds are queued on
  *              EAGAIN; lots of bug fixes (andrei)
d9515405
  *  2006-02-06  better tcp_max_connections checks, tcp_connections_no moved to
  *              shm (andrei)
e6a2b12e
  *  2006-04-12  tcp_send() changed to use struct dest_info (andrei)
be7401cc
  *  2006-11-02  switched to atomic ops for refcnt, locking improvements 
  *               (andrei)
d8b11bbc
  *  2006-11-04  switched to raw ticks (to fix conversion errors which could
  *               result in inf. lifetime) (andrei)
205fcb02
  *  2007-07-25  tcpconn_connect can now bind the socket on a specified
  *                source addr/port (andrei)
a288ee34
  *  2007-07-26   tcp_send() and tcpconn_get() can now use a specified source
  *                addr./port (andrei)
a6357e25
  *  2007-08-23   getsockname() for INADDR_ANY(SI_IS_ANY) sockets (andrei)
2aa5c1f7
  *  2007-08-27   split init_sock_opt into a lightweight init_sock_opt_accept() 
  *               used when accepting connections and init_sock_opt used for 
  *               connect/ new sockets (andrei)
5702d448
  *  2007-11-22  always add the connection & clear the coresponding flags before
  *               io_watch_add-ing its fd - it's safer this way (andrei)
ccb7fda2
  *  2007-11-26  improved tcp timers: switched to local_timer (andrei)
5c5cd736
  *  2007-11-27  added send fd cache and reader fd reuse (andrei)
20c64cc6
  *  2007-11-28  added support for TCP_DEFER_ACCEPT, KEEPALIVE, KEEPINTVL,
  *               KEEPCNT, QUICKACK, SYNCNT, LINGER2 (andrei)
885b9f62
  *  2007-12-04  support for queueing write requests (andrei)
20863813
  *  2007-12-12  destroy connection asap on wbuf. timeout (andrei)
19782e1c
  *  2007-12-13  changed the refcnt and destroy scheme, now refcnt is 1 if
db237b38
  *                linked into the hash tables (was 0) (andrei)
d22b82a0
  *  2007-12-21  support for pending connects (connections are added to the
  *               hash immediately and writes on them are buffered) (andrei)
7498b4dc
  *  2008-02-05  handle POLLRDHUP (if supported), POLLERR and
  *               POLLHUP (andrei)
  *              on write error check if there's still data in the socket 
  *               read buffer and process it first (andrei)
38429f23
  *  2009-02-26  direct blacklist support (andrei)
ffc72fcf
  *  2009-03-20  s/wq_timeout/send_timeout ; send_timeout is now in ticks
  *              (andrei)
7bb2b4ca
  * 2009-04-09  tcp ev and tcp stats macros added (andrei)
b33736bc
  */
5b532c7f
 
 
 #ifdef USE_TCP
 
6bc40dea
 
 #ifndef SHM_MEM
 #error "shared memory support needed (add -DSHM_MEM to Makefile.defs)"
 #endif
 
7498b4dc
 #define HANDLE_IO_INLINE
 #include "io_wait.h" /* include first to make sure the needed features are
 						turned on (e.g. _GNU_SOURCE for POLLRDHUP) */
 
5b532c7f
 #include <sys/time.h>
 #include <sys/types.h>
b988daef
 #include <sys/select.h>
5b532c7f
 #include <sys/socket.h>
0d2d4018
 #ifdef HAVE_FILIO_H
 #include <sys/filio.h> /* needed on solaris 2.x for FIONREAD */
 #elif defined __OS_solaris
 #define BSD_COMP  /* needed on older solaris for FIONREAD */
 #endif /* HAVE_FILIO_H / __OS_solaris */
7498b4dc
 #include <sys/ioctl.h>  /* ioctl() used on write error */
1f2c924e
 #include <netinet/in.h>
 #include <netinet/in_systm.h>
39546e5f
 #include <netinet/ip.h>
b33736bc
 #include <netinet/tcp.h>
ab130758
 #include <sys/uio.h>  /* writev*/
b33736bc
 #include <netdb.h>
0ba367ec
 #include <stdlib.h> /*exit() */
5b532c7f
 
 #include <unistd.h>
 
 #include <errno.h>
 #include <string.h>
 
0ba367ec
 #ifdef HAVE_SELECT
 #include <sys/select.h>
 #endif
 #include <sys/poll.h>
5b532c7f
 
 
 #include "ip_addr.h"
 #include "pass_fd.h"
6bc40dea
 #include "tcp_conn.h"
5b532c7f
 #include "globals.h"
6ee62314
 #include "pt.h"
8aeb47e2
 #include "locking.h"
5b532c7f
 #include "mem/mem.h"
6ee62314
 #include "mem/shm_mem.h"
 #include "timer.h"
7ec958f3
 #include "sr_module.h"
0c5da34b
 #include "tcp_server.h"
f2f969dd
 #include "tcp_init.h"
7bb2b4ca
 #include "tcp_stats.h"
 #include "tcp_ev.h"
518c9339
 #include "tsend.h"
746f7674
 #include "timer_ticks.h"
ccb7fda2
 #include "local_timer.h"
6c53d41a
 #ifdef CORE_TLS
28427aa4
 #include "tls/tls_server.h"
6c53d41a
 #define tls_loaded() 1
 #else
 #include "tls_hooks_init.h"
 #include "tls_hooks.h"
38429f23
 #endif /* CORE_TLS*/
 #ifdef USE_DST_BLACKLIST
 #include "dst_blacklist.h"
 #endif /* USE_DST_BLACKLIST */
6c53d41a
 
d9515405
 #include "tcp_info.h"
20c64cc6
 #include "tcp_options.h"
885b9f62
 #include "ut.h"
9188021a
 #include "cfg/cfg_struct.h"
8aeb47e2
 
5b532c7f
 #define local_malloc pkg_malloc
 #define local_free   pkg_free
 
0ba367ec
 #include <fcntl.h> /* must be included after io_wait.h if SIGIO_RT is used */
 
22f06258
 
92c0024c
 #ifdef NO_MSG_DONTWAIT
 #ifndef MSG_DONTWAIT
 /* should work inside tcp_main */
 #define MSG_DONTWAIT 0
 #endif
 #endif /*NO_MSG_DONTWAIT */
 
 
746f7674
 #define TCP_PASS_NEW_CONNECTION_ON_DATA /* don't pass a new connection
 										   immediately to a child, wait for
 										   some data on it first */
1ce28841
 #define TCP_LISTEN_BACKLOG 1024
28260509
 #define SEND_FD_QUEUE /* queue send fd requests on EAGAIN, instead of sending 
 							them immediately */
 #define TCP_CHILD_NON_BLOCKING 
98f3d5e2
 #ifdef SEND_FD_QUEUE
28260509
 #ifndef TCP_CHILD_NON_BLOCKING
 #define TCP_CHILD_NON_BLOCKING
 #endif
3e999281
 #define MAX_SEND_FD_QUEUE_SIZE	tcp_main_max_fd_no
98f3d5e2
 #define SEND_FD_QUEUE_SIZE		128  /* initial size */
746f7674
 #define MAX_SEND_FD_RETRIES		96	 /* FIXME: not used for now */
 #define SEND_FD_QUEUE_TIMEOUT	MS_TO_TICKS(2000)  /* 2 s */
98f3d5e2
 #endif
0ba367ec
 
ccb7fda2
 /* minimum interval local_timer_run() is allowed to run, in ticks */
 #define TCPCONN_TIMEOUT_MIN_RUN 1  /* once per tick */
 #define TCPCONN_WAIT_TIMEOUT 1 /* 1 tick */
0ba367ec
 
76cb799e
 #ifdef TCP_ASYNC
885b9f62
 static unsigned int* tcp_total_wq=0;
 #endif
 
 
0ba367ec
 enum fd_types { F_NONE, F_SOCKINFO /* a tcp_listen fd */,
 				F_TCPCONN, F_TCPCHILD, F_PROC };
 
5c5cd736
 
 #ifdef TCP_FD_CACHE
 
 #define TCP_FD_CACHE_SIZE 8
 
 struct fd_cache_entry{
 	struct tcp_connection* con;
 	int id;
 	int fd;
 };
 
 
 static struct fd_cache_entry fd_cache[TCP_FD_CACHE_SIZE];
 #endif /* TCP_FD_CACHE */
 
d8b11bbc
 static int is_tcp_main=0;
0ba367ec
 
3dc4f620
 
0ba367ec
 enum poll_types tcp_poll_method=0; /* by default choose the best method */
3e999281
 int tcp_main_max_fd_no=0;
3dc4f620
 int tcp_max_connections=DEFAULT_TCP_MAX_CONNECTIONS;
55d8155e
 
205fcb02
 static union sockaddr_union tcp_source_ipv4_addr; /* saved bind/srv v4 addr. */
 static union sockaddr_union* tcp_source_ipv4=0;
761bd309
 #ifdef USE_IPV6
205fcb02
 static union sockaddr_union tcp_source_ipv6_addr; /* saved bind/src v6 addr. */
 static union sockaddr_union* tcp_source_ipv6=0;
761bd309
 #endif
 
d9515405
 static int* tcp_connections_no=0; /* current open connections */
5b532c7f
 
59653eb8
 /* connection hash table (after ip&port) , includes also aliases */
 struct tcp_conn_alias** tcpconn_aliases_hash=0;
8aeb47e2
 /* connection hash table (after connection id) */
 struct tcp_connection** tcpconn_id_hash=0;
0bd53297
 gen_lock_t* tcpconn_lock=0;
8aeb47e2
 
37209e14
 struct tcp_child* tcp_children;
2719e69b
 static int* connection_id=0; /*  unique for each connection, used for 
0c5da34b
 								quickly finding the corresponding connection
 								for a reply */
 int unix_tcp_sock;
5b532c7f
 
0ba367ec
 static int tcp_proto_no=-1; /* tcp protocol number as returned by
 							   getprotobyname */
 
 static io_wait_h io_h;
5b532c7f
 
ccb7fda2
 static struct local_timer tcp_main_ltimer;
057063e6
 static ticks_t tcp_main_prev_ticks;
5b532c7f
 
bc977837
 
ccb7fda2
 static ticks_t tcpconn_main_timeout(ticks_t , struct timer_ln* , void* );
 
a288ee34
 inline static int _tcpconn_add_alias_unsafe(struct tcp_connection* c, int port,
22f06258
 										struct ip_addr* l_ip, int l_port,
 										int flags);
a288ee34
 
 
 
205fcb02
 /* sets source address used when opening new sockets and no source is specified
  *  (by default the address is choosen by the kernel)
  * Should be used only on init.
  * returns -1 on error */
 int tcp_set_src_addr(struct ip_addr* ip)
 {
 	switch (ip->af){
 		case AF_INET:
 			ip_addr2su(&tcp_source_ipv4_addr, ip, 0);
 			tcp_source_ipv4=&tcp_source_ipv4_addr;
 			break;
 		#ifdef USE_IPV6
 		case AF_INET6:
 			ip_addr2su(&tcp_source_ipv6_addr, ip, 0);
 			tcp_source_ipv6=&tcp_source_ipv6_addr;
 			break;
 		#endif
 		default:
 			return -1;
 	}
 	return 0;
 }
 
 
 
20c64cc6
 static inline int init_sock_keepalive(int s)
 {
 	int optval;
 	
 #ifdef HAVE_SO_KEEPALIVE
22db42e4
 	if (cfg_get(tcp, tcp_cfg, keepalive)){
20c64cc6
 		optval=1;
 		if (setsockopt(s, SOL_SOCKET, SO_KEEPALIVE, &optval,
 						sizeof(optval))<0){
 			LOG(L_WARN, "WARNING: init_sock_keepalive: failed to enable"
 						" SO_KEEPALIVE: %s\n", strerror(errno));
 			return -1;
 		}
 	}
 #endif
 #ifdef HAVE_TCP_KEEPINTVL
22db42e4
 	if ((optval=cfg_get(tcp, tcp_cfg, keepintvl))){
20c64cc6
 		if (setsockopt(s, IPPROTO_TCP, TCP_KEEPINTVL, &optval,
 						sizeof(optval))<0){
 			LOG(L_WARN, "WARNING: init_sock_keepalive: failed to set"
 						" keepalive probes interval: %s\n", strerror(errno));
 		}
 	}
 #endif
 #ifdef HAVE_TCP_KEEPIDLE
22db42e4
 	if ((optval=cfg_get(tcp, tcp_cfg, keepidle))){
20c64cc6
 		if (setsockopt(s, IPPROTO_TCP, TCP_KEEPIDLE, &optval,
 						sizeof(optval))<0){
 			LOG(L_WARN, "WARNING: init_sock_keepalive: failed to set"
 						" keepalive idle interval: %s\n", strerror(errno));
 		}
 	}
 #endif
 #ifdef HAVE_TCP_KEEPCNT
22db42e4
 	if ((optval=cfg_get(tcp, tcp_cfg, keepcnt))){
20c64cc6
 		if (setsockopt(s, IPPROTO_TCP, TCP_KEEPCNT, &optval,
 						sizeof(optval))<0){
 			LOG(L_WARN, "WARNING: init_sock_keepalive: failed to set"
 						" maximum keepalive count: %s\n", strerror(errno));
 		}
 	}
 #endif
 	return 0;
 }
 
 
 
2aa5c1f7
 /* set all socket/fd options for new sockets (e.g. before connect): 
  *  disable nagle, tos lowdelay, reuseaddr, non-blocking
  *
bc977837
  * return -1 on error */
 static int init_sock_opt(int s)
 {
 	int flags;
 	int optval;
 	
 #ifdef DISABLE_NAGLE
 	flags=1;
 	if ( (tcp_proto_no!=-1) && (setsockopt(s, tcp_proto_no , TCP_NODELAY,
 					&flags, sizeof(flags))<0) ){
 		LOG(L_WARN, "WARNING: init_sock_opt: could not disable Nagle: %s\n",
 				strerror(errno));
 	}
 #endif
 	/* tos*/
fe09f315
 	optval = tos;
bc977837
 	if (setsockopt(s, IPPROTO_IP, IP_TOS, (void*)&optval,sizeof(optval)) ==-1){
 		LOG(L_WARN, "WARNING: init_sock_opt: setsockopt tos: %s\n",
 				strerror(errno));
 		/* continue since this is not critical */
 	}
2aa5c1f7
 #if  !defined(TCP_DONT_REUSEADDR) 
 	optval=1;
 	if (setsockopt(s, SOL_SOCKET, SO_REUSEADDR,
 						(void*)&optval, sizeof(optval))==-1){
 		LOG(L_ERR, "ERROR: setsockopt SO_REUSEADDR %s\n",
 				strerror(errno));
 		/* continue, not critical */
 	}
 #endif /* !TCP_DONT_REUSEADDR */
20c64cc6
 #ifdef HAVE_TCP_SYNCNT
22db42e4
 	if ((optval=cfg_get(tcp, tcp_cfg, syncnt))){
20c64cc6
 		if (setsockopt(s, IPPROTO_TCP, TCP_SYNCNT, &optval,
 						sizeof(optval))<0){
 			LOG(L_WARN, "WARNING: init_sock_opt: failed to set"
 						" maximum SYN retr. count: %s\n", strerror(errno));
 		}
 	}
 #endif
 #ifdef HAVE_TCP_LINGER2
22db42e4
 	if ((optval=cfg_get(tcp, tcp_cfg, linger2))){
20c64cc6
 		if (setsockopt(s, IPPROTO_TCP, TCP_LINGER2, &optval,
 						sizeof(optval))<0){
 			LOG(L_WARN, "WARNING: init_sock_opt: failed to set"
 						" maximum LINGER2 timeout: %s\n", strerror(errno));
 		}
 	}
 #endif
 #ifdef HAVE_TCP_QUICKACK
22db42e4
 	if (cfg_get(tcp, tcp_cfg, delayed_ack)){
20c64cc6
 		optval=0; /* reset quick ack => delayed ack */
 		if (setsockopt(s, IPPROTO_TCP, TCP_QUICKACK, &optval,
 						sizeof(optval))<0){
 			LOG(L_WARN, "WARNING: init_sock_opt: failed to reset"
 						" TCP_QUICKACK: %s\n", strerror(errno));
 		}
 	}
 #endif /* HAVE_TCP_QUICKACK */
 	init_sock_keepalive(s);
2aa5c1f7
 	
bc977837
 	/* non-blocking */
 	flags=fcntl(s, F_GETFL);
 	if (flags==-1){
 		LOG(L_ERR, "ERROR: init_sock_opt: fnctl failed: (%d) %s\n",
 				errno, strerror(errno));
 		goto error;
 	}
 	if (fcntl(s, F_SETFL, flags|O_NONBLOCK)==-1){
 		LOG(L_ERR, "ERROR: init_sock_opt: fcntl: set non-blocking failed:"
 				" (%d) %s\n", errno, strerror(errno));
 		goto error;
 	}
 	return 0;
 error:
 	return -1;
 }
 
 
 
2aa5c1f7
 /* set all socket/fd options for "accepted" sockets 
  *  only nonblocking is set since the rest is inherited from the
  *  "parent" (listening) socket
  *  Note: setting O_NONBLOCK is required on linux but it's not needed on
  *        BSD and possibly solaris (where the flag is inherited from the 
  *        parent socket). However since there is no standard document 
  *        requiring a specific behaviour in this case it's safer to always set
  *        it (at least for now)  --andrei
  *  TODO: check on which OSes  O_NONBLOCK is inherited and make this 
  *        function a nop.
  *
  * return -1 on error */
 static int init_sock_opt_accept(int s)
 {
 	int flags;
 	
 	/* non-blocking */
 	flags=fcntl(s, F_GETFL);
 	if (flags==-1){
 		LOG(L_ERR, "ERROR: init_sock_opt_accept: fnctl failed: (%d) %s\n",
 				errno, strerror(errno));
 		goto error;
 	}
 	if (fcntl(s, F_SETFL, flags|O_NONBLOCK)==-1){
 		LOG(L_ERR, "ERROR: init_sock_opt_accept: "
 					"fcntl: set non-blocking failed: (%d) %s\n",
 					errno, strerror(errno));
 		goto error;
 	}
 	return 0;
 error:
 	return -1;
 }
 
 
 
0ba367ec
 /* blocking connect on a non-blocking fd; it will timeout after
  * tcp_connect_timeout 
  * if BLOCKING_USE_SELECT and HAVE_SELECT are defined it will internally
  * use select() instead of poll (bad if fd > FD_SET_SIZE, poll is preferred)
  */
38429f23
 static int tcp_blocking_connect(int fd, int type,
 								const struct sockaddr *servaddr,
bc977837
 								socklen_t addrlen)
 {
 	int n;
0ba367ec
 #if defined(HAVE_SELECT) && defined(BLOCKING_USE_SELECT)
bc977837
 	fd_set sel_set;
0ba367ec
 	fd_set orig_set;
bc977837
 	struct timeval timeout;
0ba367ec
 #else
 	struct pollfd pf;
 #endif
 	int elapsed;
 	int to;
bc977837
 	int ticks;
 	int err;
a9fae28b
 	unsigned int err_len;
a31a6e8c
 	int poll_err;
bc977837
 	
a31a6e8c
 	poll_err=0;
3dc4f620
 	to=cfg_get(tcp, tcp_cfg, connect_timeout_s);
0ba367ec
 	ticks=get_ticks();
bc977837
 again:
 	n=connect(fd, servaddr, addrlen);
 	if (n==-1){
0ba367ec
 		if (errno==EINTR){
 			elapsed=(get_ticks()-ticks)*TIMER_TICK;
 			if (elapsed<to)		goto again;
 			else goto error_timeout;
 		}
bc977837
 		if (errno!=EINPROGRESS && errno!=EALREADY){
30745c9d
 			goto error_errno;
bc977837
 		}
 	}else goto end;
 	
0ba367ec
 	/* poll/select loop */
 #if defined(HAVE_SELECT) && defined(BLOCKING_USE_SELECT)
 		FD_ZERO(&orig_set);
 		FD_SET(fd, &orig_set);
 #else
 		pf.fd=fd;
 		pf.events=POLLOUT;
 #endif
bc977837
 	while(1){
0ba367ec
 		elapsed=(get_ticks()-ticks)*TIMER_TICK;
98d282a6
 		if (elapsed>=to)
0ba367ec
 			goto error_timeout;
 #if defined(HAVE_SELECT) && defined(BLOCKING_USE_SELECT)
 		sel_set=orig_set;
98d282a6
 		timeout.tv_sec=to-elapsed;
bc977837
 		timeout.tv_usec=0;
 		n=select(fd+1, 0, &sel_set, 0, &timeout);
0ba367ec
 #else
98d282a6
 		n=poll(&pf, 1, (to-elapsed)*1000);
0ba367ec
 #endif
bc977837
 		if (n<0){
 			if (errno==EINTR) continue;
edcdea00
 			LOG(L_ERR, "ERROR: tcp_blocking_connect %s: poll/select failed:"
 					" (%d) %s\n",
 					su2a((union sockaddr_union*)servaddr, addrlen),
 					errno, strerror(errno));
bc977837
 			goto error;
0ba367ec
 		}else if (n==0) /* timeout */ continue;
 #if defined(HAVE_SELECT) && defined(BLOCKING_USE_SELECT)
a31a6e8c
 		if (FD_ISSET(fd, &sel_set))
0ba367ec
 #else
 		if (pf.revents&(POLLERR|POLLHUP|POLLNVAL)){ 
edcdea00
 			LOG(L_ERR, "ERROR: tcp_blocking_connect %s: poll error: "
 					"flags %x\n",
 					su2a((union sockaddr_union*)servaddr, addrlen),
0ba367ec
 					pf.revents);
a31a6e8c
 			poll_err=1;
 		}
0ba367ec
 #endif
a31a6e8c
 		{
bc977837
 			err_len=sizeof(err);
 			getsockopt(fd, SOL_SOCKET, SO_ERROR, &err, &err_len);
a31a6e8c
 			if ((err==0) && (poll_err==0)) goto end;
bc977837
 			if (err!=EINPROGRESS && err!=EALREADY){
edcdea00
 				LOG(L_ERR, "ERROR: tcp_blocking_connect %s: SO_ERROR (%d) "
 						"%s\n",
 						su2a((union sockaddr_union*)servaddr, addrlen),
bc977837
 						err, strerror(err));
30745c9d
 				errno=err;
 				goto error_errno;
bc977837
 			}
 		}
 	}
30745c9d
 error_errno:
 	switch(errno){
 		case ENETUNREACH:
 		case EHOSTUNREACH:
 #ifdef USE_DST_BLACKLIST
 			if (cfg_get(core, core_cfg, use_dst_blacklist))
 				dst_blacklist_su(BLST_ERR_CONNECT, type,
 							 (union sockaddr_union*)servaddr, 0);
 #endif /* USE_DST_BLACKLIST */
 			TCP_EV_CONNECT_UNREACHABLE(errno, 0, 0,
 							(union sockaddr_union*)servaddr, type);
 			break;
 		case ETIMEDOUT:
 #ifdef USE_DST_BLACKLIST
 			if (cfg_get(core, core_cfg, use_dst_blacklist))
 				dst_blacklist_su(BLST_ERR_CONNECT, type,
 								 (union sockaddr_union*)servaddr, 0);
 #endif /* USE_DST_BLACKLIST */
 			TCP_EV_CONNECT_TIMEOUT(errno, 0, 0,
 							(union sockaddr_union*)servaddr, type);
 			break;
 		case ECONNREFUSED:
 		case ECONNRESET:
 #ifdef USE_DST_BLACKLIST
 			if (cfg_get(core, core_cfg, use_dst_blacklist))
 				dst_blacklist_su(BLST_ERR_CONNECT, type,
 								 (union sockaddr_union*)servaddr, 0);
 #endif /* USE_DST_BLACKLIST */
 			TCP_EV_CONNECT_RST(errno, 0, 0,
 							(union sockaddr_union*)servaddr, type);
 			break;
 		case EAGAIN: /* not posix, but supported on linux and bsd */
 			TCP_EV_CONNECT_NO_MORE_PORTS(errno, 0, 0,
 							(union sockaddr_union*)servaddr, type);
 			break;
 		default:
 			TCP_EV_CONNECT_ERR(errno, 0, 0,
 								(union sockaddr_union*)servaddr, type);
 	}
 	LOG(L_ERR, "ERROR: tcp_blocking_connect %s: (%d) %s\n",
 			su2a((union sockaddr_union*)servaddr, addrlen),
 			errno, strerror(errno));
 	goto error;
0ba367ec
 error_timeout:
 	/* timeout */
38429f23
 #ifdef USE_DST_BLACKLIST
 	if (cfg_get(core, core_cfg, use_dst_blacklist))
 		dst_blacklist_su(BLST_ERR_CONNECT, type,
 							(union sockaddr_union*)servaddr, 0);
 #endif /* USE_DST_BLACKLIST */
7bb2b4ca
 	TCP_EV_CONNECT_TIMEOUT(0, 0, 0, (union sockaddr_union*)servaddr, type);
edcdea00
 	LOG(L_ERR, "ERROR: tcp_blocking_connect %s: timeout %d s elapsed "
 				"from %d s\n", su2a((union sockaddr_union*)servaddr, addrlen),
3dc4f620
 				elapsed, cfg_get(tcp, tcp_cfg, connect_timeout_s));
bc977837
 error:
30745c9d
 	TCP_STATS_CONNECT_FAILED();
bc977837
 	return -1;
 end:
 	return 0;
 }
 
 
 
885b9f62
 inline static int _tcpconn_write_nb(int fd, struct tcp_connection* c,
 									char* buf, int len);
 
 
76cb799e
 #ifdef TCP_ASYNC
885b9f62
 
 
20863813
 /* unsafe version */
 #define _wbufq_empty(con) ((con)->wbuf_q.first==0)
 /* unsafe version */
 #define _wbufq_non_empty(con) ((con)->wbuf_q.first!=0)
 
 
 /* unsafe version, call while holding the connection write lock */
 inline static int _wbufq_add(struct  tcp_connection* c, char* data, 
885b9f62
 							unsigned int size)
 {
 	struct tcp_wbuffer_queue* q;
 	struct tcp_wbuffer* wb;
 	unsigned int last_free;
 	unsigned int wb_size;
 	unsigned int crt_size;
 	ticks_t t;
 	
 	q=&c->wbuf_q;
 	t=get_ticks_raw();
22db42e4
 	if (unlikely(	((q->queued+size)>cfg_get(tcp, tcp_cfg, tcpconn_wq_max)) ||
 					((*tcp_total_wq+size)>cfg_get(tcp, tcp_cfg, tcp_wq_max)) ||
885b9f62
 					(q->first &&
20863813
 					TICKS_LT(q->wr_timeout, t)) )){
885b9f62
 		LOG(L_ERR, "ERROR: wbufq_add(%d bytes): write queue full or timeout "
 					" (%d, total %d, last write %d s ago)\n",
 					size, q->queued, *tcp_total_wq,
22db42e4
 					TICKS_TO_S(t-q->wr_timeout-
ffc72fcf
 						cfg_get(tcp, tcp_cfg, send_timeout)));
7bb2b4ca
 		if (q->first && TICKS_LT(q->wr_timeout, t)){
 			if (unlikely(c->state==S_CONN_CONNECT)){
38429f23
 #ifdef USE_DST_BLACKLIST
7bb2b4ca
 				if (likely(cfg_get(core, core_cfg, use_dst_blacklist))){
 					DBG("blacklisting, state=%d\n", c->state);
 					dst_blacklist_su( BLST_ERR_CONNECT, c->rcv.proto,
 										&c->rcv.src_su, 0);
 				}
 #endif /* USE_DST_BLACKLIST */
 				TCP_EV_CONNECT_TIMEOUT(0, TCP_LADDR(c), TCP_LPORT(c),
 											TCP_PSU(c), TCP_PROTO(c));
 				TCP_STATS_CONNECT_FAILED();
 			}else{
 #ifdef USE_DST_BLACKLIST
 				if (likely(cfg_get(core, core_cfg, use_dst_blacklist))){
 					DBG("blacklisting, state=%d\n", c->state);
 					dst_blacklist_su( BLST_ERR_SEND, c->rcv.proto,
 										&c->rcv.src_su, 0);
 				}
38429f23
 #endif /* USE_DST_BLACKLIST */
7bb2b4ca
 				TCP_EV_SEND_TIMEOUT(0, &c->rcv);
 				TCP_STATS_SEND_TIMEOUT();
 			}
 		}else{
 			/* if it's not a timeout => queue full */
 			TCP_EV_SENDQ_FULL(0, &c->rcv);
 			TCP_STATS_SENDQ_FULL();
 		}
885b9f62
 		goto error;
 	}
 	
 	if (unlikely(q->last==0)){
e655392a
 		wb_size=MAX_unsigned(cfg_get(tcp, tcp_cfg, wq_blk_size), size);
885b9f62
 		wb=shm_malloc(sizeof(*wb)+wb_size-1);
 		if (unlikely(wb==0))
 			goto error;
 		wb->b_size=wb_size;
 		wb->next=0;
 		q->last=wb;
 		q->first=wb;
 		q->last_used=0;
 		q->offset=0;
efc23dce
 		q->wr_timeout=get_ticks_raw()+
 			((c->state==S_CONN_CONNECT)?
 					S_TO_TICKS(cfg_get(tcp, tcp_cfg, connect_timeout_s)):
ffc72fcf
 					cfg_get(tcp, tcp_cfg, send_timeout));
885b9f62
 	}else{
 		wb=q->last;
 	}
 	
 	while(size){
 		last_free=wb->b_size-q->last_used;
 		if (last_free==0){
e655392a
 			wb_size=MAX_unsigned(cfg_get(tcp, tcp_cfg, wq_blk_size), size);
885b9f62
 			wb=shm_malloc(sizeof(*wb)+wb_size-1);
 			if (unlikely(wb==0))
 				goto error;
 			wb->b_size=wb_size;
 			wb->next=0;
 			q->last->next=wb;
 			q->last=wb;
 			q->last_used=0;
 			last_free=wb->b_size;
 		}
 		crt_size=MIN_unsigned(last_free, size);
74c6d280
 		memcpy(wb->buf+q->last_used, data, crt_size);
885b9f62
 		q->last_used+=crt_size;
 		size-=crt_size;
 		data+=crt_size;
 		q->queued+=crt_size;
 		atomic_add_int((int*)tcp_total_wq, crt_size);
 	}
 	return 0;
 error:
 	return -1;
 }
 
 
 
5b892e3e
 /* unsafe version, call while holding the connection write lock
  * inserts data at the beginning, it ignores the max queue size checks and
  * the timeout (use sparingly)
  * Note: it should never be called on a write buffer after wbufq_run() */
 inline static int _wbufq_insert(struct  tcp_connection* c, char* data, 
 							unsigned int size)
 {
 	struct tcp_wbuffer_queue* q;
 	struct tcp_wbuffer* wb;
 	
 	q=&c->wbuf_q;
 	if (likely(q->first==0)) /* if empty, use wbufq_add */
 		return _wbufq_add(c, data, size);
 	
22db42e4
 	if (unlikely((*tcp_total_wq+size)>cfg_get(tcp, tcp_cfg, tcp_wq_max))){
38429f23
 		LOG(L_ERR, "ERROR: wbufq_insert(%d bytes): write queue full"
5b892e3e
 					" (%d, total %d, last write %d s ago)\n",
 					size, q->queued, *tcp_total_wq,
 					TICKS_TO_S(get_ticks_raw()-q->wr_timeout-
ffc72fcf
 									cfg_get(tcp, tcp_cfg, send_timeout)));
5b892e3e
 		goto error;
 	}
 	if (unlikely(q->offset)){
 		LOG(L_CRIT, "BUG: wbufq_insert: non-null offset %d (bad call, should"
 				"never be called after the wbufq_run())\n", q->offset);
 		goto error;
 	}
 	if ((q->first==q->last) && ((q->last->b_size-q->last_used)>=size)){
 		/* one block with enough space in it for size bytes */
 		memmove(q->first->buf+size, q->first->buf, size);
 		memcpy(q->first->buf, data, size);
 		q->last_used+=size;
 	}else{
 		/* create a size bytes block directly */
 		wb=shm_malloc(sizeof(*wb)+size-1);
 		if (unlikely(wb==0))
 			goto error;
 		wb->b_size=size;
 		/* insert it */
 		wb->next=q->first;
 		q->first=wb;
 		memcpy(wb->buf, data, size);
 	}
 	
 	q->queued+=size;
 	atomic_add_int((int*)tcp_total_wq, size);
 	return 0;
 error:
 	return -1;
 }
 
 
 
20863813
 /* unsafe version, call while holding the connection write lock */
 inline static void _wbufq_destroy( struct  tcp_wbuffer_queue* q)
885b9f62
 {
 	struct tcp_wbuffer* wb;
 	struct tcp_wbuffer* next_wb;
 	int unqueued;
 	
 	unqueued=0;
 	if (likely(q->first)){
 		wb=q->first;
 		do{
 			next_wb=wb->next;
 			unqueued+=(wb==q->last)?q->last_used:wb->b_size;
 			if (wb==q->first)
 				unqueued-=q->offset;
 			shm_free(wb);
 			wb=next_wb;
 		}while(wb);
 	}
 	memset(q, 0, sizeof(*q));
 	atomic_add_int((int*)tcp_total_wq, -unqueued);
 }
 
 
 
20863813
 /* tries to empty the queue  (safe version, c->write_lock must not be hold)
885b9f62
  * returns -1 on error, bytes written on success (>=0) 
  * if the whole queue is emptied => sets *empty*/
 inline static int wbufq_run(int fd, struct tcp_connection* c, int* empty)
 {
 	struct tcp_wbuffer_queue* q;
 	struct tcp_wbuffer* wb;
 	int n;
 	int ret;
 	int block_size;
 	char* buf;
 	
 	*empty=0;
 	ret=0;
 	lock_get(&c->write_lock);
 	q=&c->wbuf_q;
 	while(q->first){
 		block_size=((q->first==q->last)?q->last_used:q->first->b_size)-
 						q->offset;
 		buf=q->first->buf+q->offset;
 		n=_tcpconn_write_nb(fd, c, buf, block_size);
 		if (likely(n>0)){
 			ret+=n;
 			if (likely(n==block_size)){
 				wb=q->first;
 				q->first=q->first->next; 
 				shm_free(wb);
 				q->offset=0;
 				q->queued-=block_size;
 				atomic_add_int((int*)tcp_total_wq, -block_size);
 			}else{
 				q->offset+=n;
 				q->queued-=n;
 				atomic_add_int((int*)tcp_total_wq, -n);
 				break;
 			}
 		}else{
 			if (n<0){
 				/* EINTR is handled inside _tcpconn_write_nb */
 				if (!(errno==EAGAIN || errno==EWOULDBLOCK)){
7bb2b4ca
 					if (unlikely(c->state==S_CONN_CONNECT)){
38429f23
 						switch(errno){
 							case ENETUNREACH:
7bb2b4ca
 							case EHOSTUNREACH: /* not posix for send() */
 #ifdef USE_DST_BLACKLIST
 								if (cfg_get(core, core_cfg, use_dst_blacklist))
 									dst_blacklist_su(BLST_ERR_CONNECT,
 															c->rcv.proto,
 															&c->rcv.src_su, 0);
 #endif /* USE_DST_BLACKLIST */
 								TCP_EV_CONNECT_UNREACHABLE(errno, TCP_LADDR(c),
 													TCP_LPORT(c), TCP_PSU(c),
 													TCP_PROTO(c));
 								break;
 							case ECONNREFUSED:
38429f23
 							case ECONNRESET:
7bb2b4ca
 #ifdef USE_DST_BLACKLIST
 								if (cfg_get(core, core_cfg, use_dst_blacklist))
 									dst_blacklist_su(BLST_ERR_CONNECT,
 															c->rcv.proto,
 															&c->rcv.src_su, 0);
 #endif /* USE_DST_BLACKLIST */
 								TCP_EV_CONNECT_RST(0, TCP_LADDR(c),
 													TCP_LPORT(c), TCP_PSU(c),
 													TCP_PROTO(c));
 								break;
 							default:
 								TCP_EV_CONNECT_ERR(errno, TCP_LADDR(c),
 													TCP_LPORT(c), TCP_PSU(c),
 													TCP_PROTO(c));
 						}
 						TCP_STATS_CONNECT_FAILED();
 					}else{
 						switch(errno){
 							case ECONNREFUSED:
 							case ECONNRESET:
 								TCP_STATS_CON_RESET();
 								/* no break */
 							case ENETUNREACH:
 							case EHOSTUNREACH: /* not posix for send() */
 #ifdef USE_DST_BLACKLIST
 								if (cfg_get(core, core_cfg, use_dst_blacklist))
 									dst_blacklist_su(BLST_ERR_SEND,
38429f23
 														c->rcv.proto,
 														&c->rcv.src_su, 0);
7bb2b4ca
 #endif /* USE_DST_BLACKLIST */
38429f23
 								break;
 						}
7bb2b4ca
 					}
885b9f62
 					ret=-1;
 					LOG(L_ERR, "ERROR: wbuf_runq: %s [%d]\n",
 						strerror(errno), errno);
 				}
 			}
 			break;
 		}
 	}
 	if (likely(q->first==0)){
 		q->last=0;
 		q->last_used=0;
 		q->offset=0;
 		*empty=1;
 	}
 	lock_release(&c->write_lock);
efc23dce
 	if (likely(ret>0)){
ffc72fcf
 		q->wr_timeout=get_ticks_raw()+cfg_get(tcp, tcp_cfg, send_timeout);
7bb2b4ca
 		if (unlikely(c->state==S_CONN_CONNECT || c->state==S_CONN_ACCEPT)){
 			TCP_STATS_ESTABLISHED(c->state);
efc23dce
 			c->state=S_CONN_OK;
7bb2b4ca
 		}
efc23dce
 	}
885b9f62
 	return ret;
 }
 
76cb799e
 #endif /* TCP_ASYNC */
885b9f62
 
 
 
518c9339
 #if 0
bc977837
 /* blocking write even on non-blocking sockets 
  * if TCP_TIMEOUT will return with error */
 static int tcp_blocking_write(struct tcp_connection* c, int fd, char* buf,
 								unsigned int len)
 {
 	int n;
 	fd_set sel_set;
 	struct timeval timeout;
 	int ticks;
 	int initial_len;
 	
 	initial_len=len;
 again:
 	
 	n=send(fd, buf, len,
 #ifdef HAVE_MSG_NOSIGNAL
 			MSG_NOSIGNAL
 #else
 			0
 #endif
 		);
 	if (n<0){
 		if (errno==EINTR)	goto again;
 		else if (errno!=EAGAIN && errno!=EWOULDBLOCK){
 			LOG(L_ERR, "tcp_blocking_write: failed to send: (%d) %s\n",
 					errno, strerror(errno));
30745c9d
 			TCP_EV_SEND_TIMEOUT(errno, &c->rcv);
 			TCP_STATS_SEND_TIMEOUT();
bc977837
 			goto error;
 		}
 	}else if (n<len){
 		/* partial write */
 		buf+=n;
 		len-=n;
 	}else{
 		/* success: full write */
 		goto end;
 	}
 	while(1){
 		FD_ZERO(&sel_set);
 		FD_SET(fd, &sel_set);
 		timeout.tv_sec=tcp_send_timeout;
 		timeout.tv_usec=0;
 		ticks=get_ticks();
 		n=select(fd+1, 0, &sel_set, 0, &timeout);
 		if (n<0){
 			if (errno==EINTR) continue; /* signal, ignore */
 			LOG(L_ERR, "ERROR: tcp_blocking_write: select failed: "
 					" (%d) %s\n", errno, strerror(errno));
 			goto error;
 		}else if (n==0){
 			/* timeout */
 			if (get_ticks()-ticks>=tcp_send_timeout){
 				LOG(L_ERR, "ERROR: tcp_blocking_write: send timeout (%d)\n",
 						tcp_send_timeout);
 				goto error;
 			}
 			continue;
 		}
 		if (FD_ISSET(fd, &sel_set)){
 			/* we can write again */
 			goto again;
 		}
 	}
 error:
 		return -1;
 end:
 		return initial_len;
 }
518c9339
 #endif
bc977837
 
 
 
f2f969dd
 struct tcp_connection* tcpconn_new(int sock, union sockaddr_union* su,
205fcb02
 									union sockaddr_union* local_addr,
f6e50f08
 									struct socket_info* ba, int type, 
 									int state)
5b532c7f
 {
 	struct tcp_connection *c;
e655392a
 	int rd_b_size;
5b532c7f
 	
e655392a
 	rd_b_size=cfg_get(tcp, tcp_cfg, rd_buf_size);
 	c=shm_malloc(sizeof(struct tcp_connection) + rd_b_size);
5b532c7f
 	if (c==0){
f6e50f08
 		LOG(L_ERR, "ERROR: tcpconn_new: mem. allocation failure\n");
5b532c7f
 		goto error;
 	}
e655392a
 	memset(c, 0, sizeof(struct tcp_connection)); /* zero init (skip rd buf)*/
5b532c7f
 	c->s=sock;
e29d027b
 	c->fd=-1; /* not initialized */
ab130758
 	if (lock_init(&c->write_lock)==0){
f6e50f08
 		LOG(L_ERR, "ERROR: tcpconn_new: init lock failed\n");
ab130758
 		goto error;
 	}
 	
f2f969dd
 	c->rcv.src_su=*su;
 	
be7401cc
 	atomic_set(&c->refcnt, 0);
8b0472d7
 	local_timer_init(&c->timer, tcpconn_main_timeout, c, 0);
f2f969dd
 	su2ip_addr(&c->rcv.src_ip, su);
 	c->rcv.src_port=su_getport(su);
 	c->rcv.bind_address=ba;
205fcb02
 	if (likely(local_addr)){
 		su2ip_addr(&c->rcv.dst_ip, local_addr);
 		c->rcv.dst_port=su_getport(local_addr);
 	}else if (ba){
f2f969dd
 		c->rcv.dst_ip=ba->address;
 		c->rcv.dst_port=ba->port_no;
 	}
9c01c860
 	print_ip("tcpconn_new: new tcp connection: ", &c->rcv.src_ip, "\n");
 	DBG(     "tcpconn_new: on port %d, type %d\n", c->rcv.src_port, type);
e655392a
 	init_tcp_req(&c->req, (char*)c+sizeof(struct tcp_connection), rd_b_size);
2719e69b
 	c->id=(*connection_id)++;
f2f969dd
 	c->rcv.proto_reserved1=0; /* this will be filled before receive_message*/
 	c->rcv.proto_reserved2=0;
f6e50f08
 	c->state=state;
 	c->extra_data=0;
 #ifdef USE_TLS
 	if (type==PROTO_TLS){
28427aa4
 		if (tls_tcpconn_init(c, sock)==-1) goto error;
f6e50f08
 	}else
 #endif /* USE_TLS*/
 	{
 		c->type=PROTO_TCP;
 		c->rcv.proto=PROTO_TCP;
3dc4f620
 		c->timeout=get_ticks_raw()+cfg_get(tcp, tcp_cfg, con_lifetime);
f6e50f08
 	}
7564dcab
 	
5b532c7f
 	return c;
 	
 error:
f6e50f08
 	if (c) shm_free(c);
5b532c7f
 	return 0;
 }
 
 
 
d22b82a0
 /* do the actual connect, set sock. options a.s.o
  * returns socket on success, -1 on error
  * sets also *res_local_addr, res_si and state (S_CONN_CONNECT for an
  * unfinished connect and S_CONN_OK for a finished one)*/
 inline static int tcp_do_connect(	union sockaddr_union* server,
 									union sockaddr_union* from,
 									int type,
 									union sockaddr_union* res_local_addr,
 									struct socket_info** res_si,
 									enum tcp_conn_states *state
 									)
0c5da34b
 {
 	int s;
f15bede1
 	union sockaddr_union my_name;
56b80d50
 	socklen_t my_name_len;
a35ad52c
 	struct ip_addr ip;
76cb799e
 #ifdef TCP_ASYNC
885b9f62
 	int n;
76cb799e
 #endif /* TCP_ASYNC */
0c5da34b
 
 	s=socket(AF2PF(server->s.sa_family), SOCK_STREAM, 0);
d22b82a0
 	if (unlikely(s==-1)){
edcdea00
 		LOG(L_ERR, "ERROR: tcp_do_connect %s: socket: (%d) %s\n",
 				su2a(server, sizeof(*server)), errno, strerror(errno));
0c5da34b
 		goto error;
 	}
bc977837
 	if (init_sock_opt(s)<0){
edcdea00
 		LOG(L_ERR, "ERROR: tcp_do_connect %s: init_sock_opt failed\n",
 					su2a(server, sizeof(*server)));
bc977837
 		goto error;
39546e5f
 	}
205fcb02
 	
d22b82a0
 	if (unlikely(from && bind(s, &from->s, sockaddru_len(*from)) != 0)){
 		LOG(L_WARN, "WARNING: tcp_do_connect: binding to source address"
edcdea00
 					" %s failed: %s [%d]\n", su2a(from, sizeof(*from)),
 					strerror(errno), errno);
d22b82a0
 	}
 	*state=S_CONN_OK;
76cb799e
 #ifdef TCP_ASYNC
 	if (likely(cfg_get(tcp, tcp_cfg, async))){
885b9f62
 again:
 		n=connect(s, &server->s, sockaddru_len(*server));
575a0b47
 		if (likely(n==-1)){ /*non-blocking => most probable EINPROGRESS*/
d22b82a0
 			if (likely(errno==EINPROGRESS))
 				*state=S_CONN_CONNECT;
575a0b47
 			else if (errno==EINTR) goto again;
d22b82a0
 			else if (errno!=EALREADY){
7bb2b4ca
 				switch(errno){
 					case ENETUNREACH:
 					case EHOSTUNREACH:
38429f23
 #ifdef USE_DST_BLACKLIST
7bb2b4ca
 						if (cfg_get(core, core_cfg, use_dst_blacklist))
 							dst_blacklist_su(BLST_ERR_CONNECT, type, server,0);
 #endif /* USE_DST_BLACKLIST */
 						TCP_EV_CONNECT_UNREACHABLE(errno, 0, 0, server, type);
 						break;
 					case ETIMEDOUT:
 #ifdef USE_DST_BLACKLIST
 						if (cfg_get(core, core_cfg, use_dst_blacklist))
 							dst_blacklist_su(BLST_ERR_CONNECT, type, server,0);
38429f23
 #endif /* USE_DST_BLACKLIST */
7bb2b4ca
 						TCP_EV_CONNECT_TIMEOUT(errno, 0, 0, server, type);
 						break;
 					case ECONNREFUSED:
 					case ECONNRESET:
 #ifdef USE_DST_BLACKLIST
 						if (cfg_get(core, core_cfg, use_dst_blacklist))
 							dst_blacklist_su(BLST_ERR_CONNECT, type, server,0);
 #endif /* USE_DST_BLACKLIST */
 						TCP_EV_CONNECT_RST(errno, 0, 0, server, type);
 						break;
 					case EAGAIN:/* not posix, but supported on linux and bsd */
 						TCP_EV_CONNECT_NO_MORE_PORTS(errno, 0, 0, server,type);
 						break;
 					default:
 						TCP_EV_CONNECT_ERR(errno, 0, 0, server, type);
 				}
 				TCP_STATS_CONNECT_FAILED();
edcdea00
 				LOG(L_ERR, "ERROR: tcp_do_connect: connect %s: (%d) %s\n",
 							su2a(server, sizeof(*server)),
 							errno, strerror(errno));
885b9f62
 				goto error;
 			}
 		}
 	}else{
76cb799e
 #endif /* TCP_ASYNC */
38429f23
 		if (tcp_blocking_connect(s, type, &server->s,
 									sockaddru_len(*server))<0){
edcdea00
 			LOG(L_ERR, "ERROR: tcp_do_connect: tcp_blocking_connect %s"
 						" failed\n", su2a(server, sizeof(*server)));
885b9f62
 			goto error;
 		}
76cb799e
 #ifdef TCP_ASYNC
0c5da34b
 	}
76cb799e
 #endif /* TCP_ASYNC */
205fcb02
 	if (from){
 		su2ip_addr(&ip, from);
 		if (!ip_addr_any(&ip))
 			/* we already know the source ip, skip the sys. call */
 			goto find_socket;
 	}
f15bede1
 	my_name_len=sizeof(my_name);
d22b82a0
 	if (unlikely(getsockname(s, &my_name.s, &my_name_len)!=0)){
 		LOG(L_ERR, "ERROR: tcp_do_connect: getsockname failed: %s(%d)\n",
f15bede1
 				strerror(errno), errno);
d22b82a0
 		*res_si=0;
 		goto error;
f15bede1
 	}
205fcb02
 	from=&my_name; /* update from with the real "from" address */
a35ad52c
 	su2ip_addr(&ip, &my_name);
205fcb02
 find_socket:
06aaa54f
 #ifdef USE_TLS
d22b82a0
 	if (unlikely(type==PROTO_TLS))
 		*res_si=find_si(&ip, 0, PROTO_TLS);
06aaa54f
 	else
 #endif
d22b82a0
 		*res_si=find_si(&ip, 0, PROTO_TCP);
 	
 	if (unlikely(*res_si==0)){
edcdea00
 		LOG(L_WARN, "WARNING: tcp_do_connect %s: could not find corresponding"
 				" listening socket for %s, using default...\n",
 					su2a(server, sizeof(*server)), ip_addr2a(&ip));
d22b82a0
 		if (server->s.sa_family==AF_INET) *res_si=sendipv4_tcp;
f15bede1
 #ifdef USE_IPV6
d22b82a0
 		else *res_si=sendipv6_tcp;
f15bede1
 #endif
 	}
d22b82a0
 	*res_local_addr=*from;
 	return s;
 error:
 	if (s!=-1) close(s);
 	return -1;
 }
 
 
 
 struct tcp_connection* tcpconn_connect( union sockaddr_union* server, 
 										union sockaddr_union* from,
 										int type)
 {
 	int s;
 	struct socket_info* si;
 	union sockaddr_union my_name;
 	struct tcp_connection* con;
 	enum tcp_conn_states state;
 
 	s=-1;
 	
3dc4f620
 	if (*tcp_connections_no >= cfg_get(tcp, tcp_cfg, max_connections)){
d22b82a0
 		LOG(L_ERR, "ERROR: tcpconn_connect: maximum number of connections"
 					" exceeded (%d/%d)\n",
3dc4f620
 					*tcp_connections_no,
 					cfg_get(tcp, tcp_cfg, max_connections));
d22b82a0
 		goto error;
 	}
 	s=tcp_do_connect(server, from, type, &my_name, &si, &state);
 	if (s==-1){
edcdea00
 		LOG(L_ERR, "ERROR: tcp_do_connect %s: failed (%d) %s\n",
 				su2a(server, sizeof(*server)), errno, strerror(errno));
d22b82a0
 		goto error;
 	}
 	con=tcpconn_new(s, server, &my_name, si, type, state);
76b0ff49
 	if (con==0){
edcdea00
 		LOG(L_ERR, "ERROR: tcp_connect %s: tcpconn_new failed, closing the "
 				 " socket\n", su2a(server, sizeof(*server)));
76b0ff49
 		goto error;
 	}
 	return con;
f6e50f08
 	/*FIXME: set sock idx! */
0c5da34b
 error:
76b0ff49
 	if (s!=-1) close(s); /* close the opened socket */
0c5da34b
 	return 0;
 }
 
 
 
d22b82a0
 #ifdef TCP_CONNECT_WAIT
 int tcpconn_finish_connect( struct tcp_connection* c,
 												union sockaddr_union* from)
 {
 	int s;
 	int r;
 	union sockaddr_union local_addr;
 	struct socket_info* si;
 	enum tcp_conn_states state;
 	struct tcp_conn_alias* a;
3dc4f620
 	int new_conn_alias_flags;
d22b82a0
 	
 	s=tcp_do_connect(&c->rcv.src_su, from, c->type, &local_addr, &si, &state);
 	if (unlikely(s==-1)){
edcdea00
 		LOG(L_ERR, "ERROR: tcpconn_finish_connect %s: tcp_do_connect for %p"
 					" failed\n", su2a(&c->rcv.src_su, sizeof(c->rcv.src_su)),
 					c);
d22b82a0
 		return -1;
 	}
 	c->rcv.bind_address=si;
 	su2ip_addr(&c->rcv.dst_ip, &local_addr);
 	c->rcv.dst_port=su_getport(&local_addr);
 	/* update aliases if needed */
 	if (likely(from==0)){
3dc4f620
 		new_conn_alias_flags=cfg_get(tcp, tcp_cfg, new_conn_alias_flags);
d22b82a0
 		/* add aliases */
 		TCPCONN_LOCK;
 		_tcpconn_add_alias_unsafe(c, c->rcv.src_port, &c->rcv.dst_ip, 0,
3dc4f620
 													new_conn_alias_flags);
d22b82a0
 		_tcpconn_add_alias_unsafe(c, c->rcv.src_port, &c->rcv.dst_ip,
3dc4f620
 									c->rcv.dst_port, new_conn_alias_flags);
d22b82a0
 		TCPCONN_UNLOCK;
 	}else if (su_cmp(from, &local_addr)!=1){
3dc4f620
 		new_conn_alias_flags=cfg_get(tcp, tcp_cfg, new_conn_alias_flags);
d22b82a0
 		TCPCONN_LOCK;
 			/* remove all the aliases except the first one and re-add them
 			 * (there shouldn't be more then the 3 default aliases at this 
 			 * stage) */
 			for (r=1; r<c->aliases; r++){
 				a=&c->con_aliases[r];
 				tcpconn_listrm(tcpconn_aliases_hash[a->hash], a, next, prev);
 			}
 			c->aliases=1;
 			/* add the local_ip:0 and local_ip:local_port aliases */
 			_tcpconn_add_alias_unsafe(c, c->rcv.src_port, &c->rcv.dst_ip,
3dc4f620
 												0, new_conn_alias_flags);
d22b82a0
 			_tcpconn_add_alias_unsafe(c, c->rcv.src_port, &c->rcv.dst_ip,
3dc4f620
 									c->rcv.dst_port, new_conn_alias_flags);
d22b82a0
 		TCPCONN_UNLOCK;
 	}
 	
 	return s;
 }
 #endif /* TCP_CONNECT_WAIT */
 
 
 
d8b11bbc
 /* adds a tcp connection to the tcpconn hashes
  * Note: it's called _only_ from the tcp_main process */
a288ee34
 inline static struct tcp_connection*  tcpconn_add(struct tcp_connection *c)
0c5da34b
 {
a288ee34
 	struct ip_addr zero_ip;
3dc4f620
 	int new_conn_alias_flags;
8aeb47e2
 
a288ee34
 	if (likely(c)){
 		ip_addr_mk_any(c->rcv.src_ip.af, &zero_ip);
be7401cc
 		c->id_hash=tcp_id_hash(c->id);
a288ee34
 		c->aliases=0;
3dc4f620
 		new_conn_alias_flags=cfg_get(tcp, tcp_cfg, new_conn_alias_flags);
8aeb47e2
 		TCPCONN_LOCK;
19782e1c
 		c->flags|=F_CONN_HASHED;
8aeb47e2
 		/* add it at the begining of the list*/
be7401cc
 		tcpconn_listadd(tcpconn_id_hash[c->id_hash], c, id_next, id_prev);
a288ee34
 		/* set the aliases */
 		/* first alias is for (peer_ip, peer_port, 0 ,0) -- for finding
 		 *  any connection to peer_ip, peer_port
 		 * the second alias is for (peer_ip, peer_port, local_addr, 0) -- for
 		 *  finding any conenction to peer_ip, peer_port from local_addr 
 		 * the third alias is for (peer_ip, peer_port, local_addr, local_port) 
 		 *   -- for finding if a fully specified connection exists */
22f06258
 		_tcpconn_add_alias_unsafe(c, c->rcv.src_port, &zero_ip, 0,
3dc4f620
 													new_conn_alias_flags);
d22b82a0
 		if (likely(c->rcv.dst_ip.af && ! ip_addr_any(&c->rcv.dst_ip))){
 			_tcpconn_add_alias_unsafe(c, c->rcv.src_port, &c->rcv.dst_ip, 0,
3dc4f620
 													new_conn_alias_flags);
d22b82a0
 			_tcpconn_add_alias_unsafe(c, c->rcv.src_port, &c->rcv.dst_ip,
3dc4f620
 									c->rcv.dst_port, new_conn_alias_flags);
d22b82a0
 		}
a288ee34
 		/* ignore add_alias errors, there are some valid cases when one
 		 *  of the add_alias would fail (e.g. first add_alias for 2 connections
 		 *   with the same destination but different src. ip*/
8aeb47e2
 		TCPCONN_UNLOCK;
a288ee34
 		DBG("tcpconn_add: hashes: %d:%d:%d, %d\n",
 												c->con_aliases[0].hash,
 												c->con_aliases[1].hash,
 												c->con_aliases[2].hash,
be7401cc
 												c->id_hash);
8aeb47e2
 		return c;
 	}else{
 		LOG(L_CRIT, "tcpconn_add: BUG: null connection pointer\n");
 		return 0;
 	}
0c5da34b
 }
 
 
ccb7fda2
 static inline void _tcpconn_detach(struct tcp_connection *c)
ab130758
 {
59653eb8
 	int r;
ab130758
 	tcpconn_listrm(tcpconn_id_hash[c->id_hash], c, id_next, id_prev);
59653eb8
 	/* remove all the aliases */
 	for (r=0; r<c->aliases; r++)
 		tcpconn_listrm(tcpconn_aliases_hash[c->con_aliases[r].hash], 
 						&c->con_aliases[r], next, prev);
ccb7fda2
 }
 
 
 
 static inline void _tcpconn_free(struct tcp_connection* c)
 {
76cb799e
 #ifdef TCP_ASYNC
20863813
 	if (unlikely(_wbufq_non_empty(c)))
 		_wbufq_destroy(&c->wbuf_q);
885b9f62
 #endif
ab130758
 	lock_destroy(&c->write_lock);
f6e50f08
 #ifdef USE_TLS
ccb7fda2
 	if (unlikely(c->type==PROTO_TLS)) tls_tcpconn_clean(c);
f6e50f08
 #endif
ab130758
 	shm_free(c);
 }
 
 
0c5da34b
 
ccb7fda2
 /* unsafe tcpconn_rm version (nolocks) */
 void _tcpconn_rm(struct tcp_connection* c)
 {
 	_tcpconn_detach(c);
 	_tcpconn_free(c);
 }
 
 
 
5b532c7f
 void tcpconn_rm(struct tcp_connection* c)
 {
59653eb8
 	int r;
0c5da34b
 	TCPCONN_LOCK;
8aeb47e2
 	tcpconn_listrm(tcpconn_id_hash[c->id_hash], c, id_next, id_prev);
59653eb8
 	/* remove all the aliases */
 	for (r=0; r<c->aliases; r++)
 		tcpconn_listrm(tcpconn_aliases_hash[c->con_aliases[r].hash], 
 						&c->con_aliases[r], next, prev);
0c5da34b
 	TCPCONN_UNLOCK;
ab130758
 	lock_destroy(&c->write_lock);
f6e50f08
 #ifdef USE_TLS
 	if ((c->type==PROTO_TLS)&&(c->extra_data)) tls_tcpconn_clean(c);
 #endif
6bc40dea
 	shm_free(c);
 }
 
 
a288ee34
 /* finds a connection, if id=0 uses the ip addr, port, local_ip and local port
  *  (host byte order) and tries to find the connection that matches all of
  *   them. Wild cards can be used for local_ip and local_port (a 0 filled
  *   ip address and/or a 0 local port).
8aeb47e2
  * WARNING: unprotected (locks) use tcpconn_get unless you really
  * know what you are doing */
a288ee34
 struct tcp_connection* _tcpconn_find(int id, struct ip_addr* ip, int port,
 										struct ip_addr* l_ip, int l_port)
0c5da34b
 {
 
 	struct tcp_connection *c;
59653eb8
 	struct tcp_conn_alias* a;
8aeb47e2
 	unsigned hash;
a288ee34
 	int is_local_ip_any;
0c5da34b
 	
f2e456c3
 #ifdef EXTRA_DEBUG
9c01c860
 	DBG("tcpconn_find: %d  port %d\n",id, port);
59653eb8
 	if (ip) print_ip("tcpconn_find: ip ", ip, "\n");
f2e456c3
 #endif
00f45c49
 	if (likely(id)){
8aeb47e2
 		hash=tcp_id_hash(id);
 		for (c=tcpconn_id_hash[hash]; c; c=c->id_next){
f2e456c3
 #ifdef EXTRA_DEBUG
9c01c860
 			DBG("c=%p, c->id=%d, port=%d\n",c, c->id, c->rcv.src_port);
 			print_ip("ip=", &c->rcv.src_ip, "\n");
f2e456c3
 #endif
f6e50f08
 			if ((id==c->id)&&(c->state!=S_CONN_BAD)) return c;
8aeb47e2
 		}
00f45c49
 	}else if (likely(ip)){
a288ee34
 		hash=tcp_addr_hash(ip, port, l_ip, l_port);
 		is_local_ip_any=ip_addr_any(l_ip);
59653eb8
 		for (a=tcpconn_aliases_hash[hash]; a; a=a->next){
f2e456c3
 #ifdef EXTRA_DEBUG
59653eb8
 			DBG("a=%p, c=%p, c->id=%d, alias port= %d port=%d\n", a, a->parent,
 					a->parent->id, a->port, a->parent->rcv.src_port);
 			print_ip("ip=",&a->parent->rcv.src_ip,"\n");
f2e456c3
 #endif
59653eb8
 			if ( (a->parent->state!=S_CONN_BAD) && (port==a->port) &&
a288ee34
 					((l_port==0) || (l_port==a->parent->rcv.dst_port)) &&
 					(ip_addr_cmp(ip, &a->parent->rcv.src_ip)) &&
 					(is_local_ip_any ||
 						ip_addr_cmp(l_ip, &a->parent->rcv.dst_ip))
 				)
59653eb8
 				return a->parent;
8aeb47e2
 		}
0c5da34b
 	}
 	return 0;
 }
 
 
 
a288ee34
 /* _tcpconn_find with locks and timeout
  * local_addr contains the desired local ip:port. If null any local address 
  * will be used.  IN*ADDR_ANY or 0 port are wild cards.
  */
ab130758
 struct tcp_connection* tcpconn_get(int id, struct ip_addr* ip, int port,
a288ee34
 									union sockaddr_union* local_addr,
d8b11bbc
 									ticks_t timeout)
0c5da34b
 {
 	struct tcp_connection* c;
a288ee34
 	struct ip_addr local_ip;
 	int local_port;
 	
 	local_port=0;
00f45c49
 	if (likely(ip)){
 		if (unlikely(local_addr)){
a288ee34
 			su2ip_addr(&local_ip, local_addr);
 			local_port=su_getport(local_addr);
 		}else{
 			ip_addr_mk_any(ip->af, &local_ip);
 			local_port=0;
 		}
 	}
0c5da34b
 	TCPCONN_LOCK;
a288ee34
 	c=_tcpconn_find(id, ip, port, &local_ip, local_port);
00f45c49
 	if (likely(c)){ 
be7401cc
 			atomic_inc(&c->refcnt);
00f45c49
 			/* update the timeout only if the connection is not handled
 			 * by a tcp reader (the tcp reader process uses c->timeout for 
 			 * its own internal timeout and c->timeout will be overwritten
 			 * anyway on return to tcp_main) */
 			if (likely(c->reader_pid==0))
 				c->timeout=get_ticks_raw()+timeout;
ab130758
 	}
0c5da34b
 	TCPCONN_UNLOCK;
 	return c;
 }
 
 
 
a288ee34
 /* add c->dst:port, local_addr as an alias for the "id" connection, 
22f06258
  * flags: TCP_ALIAS_FORCE_ADD  - add an alias even if a previous one exists
  *        TCP_ALIAS_REPLACE    - if a prev. alias exists, replace it with the
  *                                new one
a288ee34
  * returns 0 on success, <0 on failure ( -1  - null c, -2 too many aliases,
  *  -3 alias already present and pointing to another connection)
  * WARNING: must be called with TCPCONN_LOCK held */
 inline static int _tcpconn_add_alias_unsafe(struct tcp_connection* c, int port,
22f06258
 										struct ip_addr* l_ip, int l_port,
 										int flags)
59653eb8
 {
 	unsigned hash;
 	struct tcp_conn_alias* a;
22f06258
 	struct tcp_conn_alias* nxt;
55cdf20a
 	struct tcp_connection* p;
a288ee34
 	int is_local_ip_any;
55cdf20a
 	int i;
 	int r;
59653eb8
 	
 	a=0;
a288ee34
 	is_local_ip_any=ip_addr_any(l_ip);
00f45c49
 	if (likely(c)){
a288ee34
 		hash=tcp_addr_hash(&c->rcv.src_ip, port, l_ip, l_port);
59653eb8
 		/* search the aliases for an already existing one */
f0302ef7
 		for (a=tcpconn_aliases_hash[hash], nxt=0; a; a=nxt){
 			nxt=a->next;
59653eb8
 			if ( (a->parent->state!=S_CONN_BAD) && (port==a->port) &&
a288ee34
 					( (l_port==0) || (l_port==a->parent->rcv.dst_port)) &&
 					(ip_addr_cmp(&c->rcv.src_ip, &a->parent->rcv.src_ip)) &&
 					( is_local_ip_any || 
 					  ip_addr_cmp(&a->parent->rcv.dst_ip, l_ip))
 					){
59653eb8
 				/* found */
00f45c49
 				if (unlikely(a->parent!=c)){
22f06258
 					if (flags & TCP_ALIAS_FORCE_ADD)
 						/* still have to walk the whole list to check if
 						 * the alias was not already added */
 						continue;
 					else if (flags & TCP_ALIAS_REPLACE){
55cdf20a
 						/* remove the alias =>
 						 * remove the current alias and all the following
 						 *  ones from the corresponding connection, shift the 
 						 *  connection aliases array and re-add the other 
 						 *  aliases (!= current one) */
 						p=a->parent;
 						for (i=0; (i<p->aliases) && (&(p->con_aliases[i])!=a);
 								i++);
 						if (unlikely(i==p->aliases)){
 							LOG(L_CRIT, "BUG: _tcpconn_add_alias_unsafe: "
 									" alias %p not found in con %p (id %d)\n",
 									a, p, p->id);
 							goto error_not_found;
 						}
 						for (r=i; r<p->aliases; r++){
 							tcpconn_listrm(
 								tcpconn_aliases_hash[p->con_aliases[r].hash],
 								&p->con_aliases[r], next, prev);
 						}
 						if (likely((i+1)<p->aliases)){
 							memmove(&p->con_aliases[i], &p->con_aliases[i+1],
 											(p->aliases-i-1)*
 												sizeof(p->con_aliases[0]));
 						}
 						p->aliases--;
 						/* re-add the remaining aliases */
 						for (r=i; r<p->aliases; r++){
 							tcpconn_listadd(
 								tcpconn_aliases_hash[p->con_aliases[r].hash], 
 								&p->con_aliases[r], next, prev);
 						}
22f06258
 					}else
 						goto error_sec;
 				}else goto ok;
59653eb8
 			}
 		}
00f45c49
 		if (unlikely(c->aliases>=TCP_CON_MAX_ALIASES)) goto error_aliases;
59653eb8
 		c->con_aliases[c->aliases].parent=c;
 		c->con_aliases[c->aliases].port=port;
 		c->con_aliases[c->aliases].hash=hash;
 		tcpconn_listadd(tcpconn_aliases_hash[hash], 
 								&c->con_aliases[c->aliases], next, prev);
 		c->aliases++;
 	}else goto error_not_found;
 ok:
 #ifdef EXTRA_DEBUG
a288ee34
 	if (a) DBG("_tcpconn_add_alias_unsafe: alias already present\n");
 	else   DBG("_tcpconn_add_alias_unsafe: alias port %d for hash %d, id %d\n",
59653eb8
 			port, hash, c->id);
 #endif
 	return 0;
 error_aliases:
a288ee34
 	/* too many aliases */
 	return -2;
 error_not_found:
 	/* null connection */
59653eb8
 	return -1;
a288ee34
 error_sec:
 	/* alias already present and pointing to a different connection
 	 * (hijack attempt?) */
 	return -3;
 }
 
 
 
 /* add port as an alias for the "id" connection, 
  * returns 0 on success,-1 on failure */
 int tcpconn_add_alias(int id, int port, int proto)
 {
 	struct tcp_connection* c;
 	int ret;
 	struct ip_addr zero_ip;
23dc5abf
 	int r;
3dc4f620
 	int alias_flags;
a288ee34
 	
 	/* fix the port */
 	port=port?port:((proto==PROTO_TLS)?SIPS_PORT:SIP_PORT);
 	TCPCONN_LOCK;
 	/* check if alias already exists */
 	c=_tcpconn_find(id, 0, 0, 0, 0);
00f45c49
 	if (likely(c)){
a288ee34
 		ip_addr_mk_any(c->rcv.src_ip.af, &zero_ip);
3dc4f620
 		alias_flags=cfg_get(tcp, tcp_cfg, alias_flags);
a288ee34
 		/* alias src_ip:port, 0, 0 */
22f06258
 		ret=_tcpconn_add_alias_unsafe(c, port,  &zero_ip, 0, 
3dc4f620
 										alias_flags);
a288ee34
 		if (ret<0 && ret!=-3) goto error;
 		/* alias src_ip:port, local_ip, 0 */
22f06258
 		ret=_tcpconn_add_alias_unsafe(c, port,  &c->rcv.dst_ip, 0, 
3dc4f620
 										alias_flags);
a288ee34
 		if (ret<0 && ret!=-3) goto error;
 		/* alias src_ip:port, local_ip, local_port */
22f06258
 		ret=_tcpconn_add_alias_unsafe(c, port, &c->rcv.dst_ip, c->rcv.dst_port,
3dc4f620
 										alias_flags);
00f45c49
 		if (unlikely(ret<0)) goto error;
a288ee34
 	}else goto error_not_found;
 	TCPCONN_UNLOCK;
 	return 0;
59653eb8
 error_not_found:
 	TCPCONN_UNLOCK;
 	LOG(L_ERR, "ERROR: tcpconn_add_alias: no connection found for id %d\n",id);
 	return -1;
a288ee34
 error:
59653eb8
 	TCPCONN_UNLOCK;
a288ee34
 	switch(ret){
 		case -2:
23dc5abf
 			LOG(L_ERR, "ERROR: tcpconn_add_alias: too many aliases (%d)"
 					" for connection %p (id %d) %s:%d <- %d\n",
 					c->aliases, c, c->id, ip_addr2a(&c->rcv.src_ip),
 					c->rcv.src_port, port);
 			for (r=0; r<c->aliases; r++){
 				LOG(L_ERR, "ERROR: tcpconn_add_alias: alias %d: for %p (%d)"
 						" %s:%d <-%d hash %x\n",  r, c, c->id, 
 						 ip_addr2a(&c->rcv.src_ip), c->rcv.src_port, 
 						c->con_aliases[r].port, c->con_aliases[r].hash);
 			}
a288ee34
 			break;
 		case -3:
 			LOG(L_ERR, "ERROR: tcpconn_add_alias: possible port"
 					" hijack attempt\n");
 			LOG(L_ERR, "ERROR: tcpconn_add_alias: alias for %d port %d already"
 						" present and points to another connection \n",
 						c->id, port);
 			break;
 		default:
 			LOG(L_ERR, "ERROR: tcpconn_add_alias: unkown error %d\n", ret);
 	}
59653eb8
 	return -1;
 }
 
 
 
5c5cd736
 #ifdef TCP_FD_CACHE
 
 static void tcp_fd_cache_init()
 {
 	int r;
 	for (r=0; r<TCP_FD_CACHE_SIZE; r++)
 		fd_cache[r].fd=-1;
 }
 
 
 inline static struct fd_cache_entry* tcp_fd_cache_get(struct tcp_connection *c)
 {
 	int h;
 	
 	h=c->id%TCP_FD_CACHE_SIZE;
 	if ((fd_cache[h].fd>0) && (fd_cache[h].id==c->id) && (fd_cache[h].con==c))
 		return &fd_cache[h];
 	return 0;
 }
 
 
 inline static void tcp_fd_cache_rm(struct fd_cache_entry* e)
 {
 	e->fd=-1;
 }
 
 
 inline static void tcp_fd_cache_add(struct tcp_connection *c, int fd)
 {
 	int h;
 	
 	h=c->id%TCP_FD_CACHE_SIZE;
00f45c49
 	if (likely(fd_cache[h].fd>0))
5c5cd736
 		close(fd_cache[h].fd);
 	fd_cache[h].fd=fd;
 	fd_cache[h].id=c->id;
 	fd_cache[h].con=c;
 }
 
 #endif /* TCP_FD_CACHE */
 
 
19782e1c
 
 inline static int tcpconn_chld_put(struct tcp_connection* tcpconn);
 
 
 
bf458eea
 /* finds a tcpconn & sends on it
a288ee34
  * uses the dst members to, proto (TCP|TLS) and id and tries to send
  *  from the "from" address (if non null and id==0)
bf458eea
  * returns: number of bytes written (>=0) on success
  *          <0 on error */
a288ee34
 int tcp_send(struct dest_info* dst, union sockaddr_union* from,
 					char* buf, unsigned len)
0c5da34b
 {
 	struct tcp_connection *c;
06aaa54f
 	struct tcp_connection *tmp;
0c5da34b
 	struct ip_addr ip;
 	int port;
e29d027b
 	int fd;
0c5da34b
 	long response[2];
 	int n;
5c5cd736
 	int do_close_fd;
3dc4f620
 	ticks_t con_lifetime;
76cb799e
 #ifdef TCP_ASYNC
885b9f62
 	int enable_write_watch;
76cb799e
 #endif /* TCP_ASYNC */
5c5cd736
 #ifdef TCP_FD_CACHE
 	struct fd_cache_entry* fd_cache_e;
b4fa727e
 	int use_fd_cache;
0c5da34b
 	
22db42e4
 	use_fd_cache=cfg_get(tcp, tcp_cfg, fd_cache);
5c5cd736
 	fd_cache_e=0;
 #endif /* TCP_FD_CACHE */
 	do_close_fd=1; /* close the fd on exit */
e6a2b12e
 	port=su_getport(&dst->to);
3dc4f620
 	con_lifetime=cfg_get(tcp, tcp_cfg, con_lifetime);
5c5cd736
 	if (likely(port)){
e6a2b12e
 		su2ip_addr(&ip, &dst->to);
3dc4f620
 		c=tcpconn_get(dst->id, &ip, port, from, con_lifetime); 
5c5cd736
 	}else if (likely(dst->id)){
3dc4f620
 		c=tcpconn_get(dst->id, 0, 0, 0, con_lifetime);
f2f969dd
 	}else{
 		LOG(L_CRIT, "BUG: tcp_send called with null id & to\n");
 		return -1;
 	}
0c5da34b
 	
5c5cd736
 	if (likely(dst->id)){
 		if (unlikely(c==0)) {
 			if (likely(port)){
ab130758
 				/* try again w/o id */
3dc4f620
 				c=tcpconn_get(0, &ip, port, from, con_lifetime);
f2f969dd
 				goto no_id;
 			}else{
 				LOG(L_ERR, "ERROR: tcp_send: id %d not found, dropping\n",
e6a2b12e
 						dst->id);
f2f969dd
 				return -1;
 			}
 		}else goto get_fd;
 	}
 no_id:
5c5cd736
 		if (unlikely(c==0)){
de223f01
 			/* check if connect() is disabled */
 			if (cfg_get(tcp, tcp_cfg, no_connect))
 				return -1;
0c5da34b
 			DBG("tcp_send: no open tcp connection found, opening new one\n");
 			/* create tcp connection */
5c5cd736
 			if (likely(from==0)){
205fcb02
 				/* check to see if we have to use a specific source addr. */
 				switch (dst->to.s.sa_family) {
 					case AF_INET:
 							from = tcp_source_ipv4;
 						break;
 #ifdef USE_IPV6
 					case AF_INET6:
 							from = tcp_source_ipv6;
 						break;
 #endif
 					default:
 						/* error, bad af, ignore ... */
 						break;
 				}
a288ee34
 			}
76cb799e
 #if defined(TCP_CONNECT_WAIT) && defined(TCP_ASYNC)
22db42e4
 			if (likely(cfg_get(tcp, tcp_cfg, tcp_connect_wait) && 
76cb799e
 						cfg_get(tcp, tcp_cfg, async) )){
3dc4f620
 				if (unlikely(*tcp_connections_no >=
 								cfg_get(tcp, tcp_cfg, max_connections))){
edcdea00
 					LOG(L_ERR, "ERROR: tcp_send %s: maximum number of"
 								" connections exceeded (%d/%d)\n",
e88c2542
 								su2a(&dst->to, sizeof(dst->to)),
3dc4f620
 								*tcp_connections_no,
 								cfg_get(tcp, tcp_cfg, max_connections));
d22b82a0
 					return -1;
 				}
 				c=tcpconn_new(-1, &dst->to, from, 0, dst->proto,
efc23dce
 								S_CONN_CONNECT);
d22b82a0
 				if (unlikely(c==0)){
edcdea00
 					LOG(L_ERR, "ERROR: tcp_send %s: could not create new"
 							" connection\n",
e88c2542
 							su2a(&dst->to, sizeof(dst->to)));
d22b82a0
 					return -1;
 				}
 				c->flags|=F_CONN_PENDING|F_CONN_FD_CLOSED;
 				atomic_set(&c->refcnt, 2); /* ref from here and from main hash
 											 table */
 				/* add it to id hash and aliases */
 				if (unlikely(tcpconn_add(c)==0)){
edcdea00
 					LOG(L_ERR, "ERROR: tcp_send %s: could not add "
 								"connection %p\n",
e88c2542
 								su2a(&dst->to, sizeof(dst->to)),
edcdea00
 									c);
d22b82a0
 					_tcpconn_free(c);
 					n=-1;
 					goto end_no_conn;
 				}
 				/* do connect and if src ip or port changed, update the 
 				 * aliases */
 				if (unlikely((fd=tcpconn_finish_connect(c, from))<0)){
38429f23
 					/* tcpconn_finish_connect will automatically blacklist
 					   on error => no need to do it here */
edcdea00
 					LOG(L_ERR, "ERROR: tcp_send %s: tcpconn_finish_connect(%p)"
e88c2542
 							" failed\n", su2a(&dst->to, sizeof(dst->to)),
edcdea00
 								c);
d22b82a0
 					goto conn_wait_error;
 				}
 				/* ? TODO: it might be faster just to queue the write directly
 				 *  and send to main CONN_NEW_PENDING_WRITE */
 				/* delay sending the fd to main after the send */
 				
 				/* NOTE: no lock here, because the connection is marked as
 				 * pending and nobody else will try to write on it. However
 				 * this might produce out-of-order writes. If this is not
 				 * desired either lock before the write or use 
 				 * _wbufq_insert(...) */
 				n=_tcpconn_write_nb(fd, c, buf, len);
36ac6749
 				if (unlikely(n<(int)len)){
5b892e3e
 					if ((n>=0) || errno==EAGAIN || errno==EWOULDBLOCK){
 						DBG("tcp_send: pending write on new connection %p "
 								" (%d/%d bytes written)\n", c, n, len);
 						if (n<0) n=0;
7bb2b4ca
 						else{
 							TCP_STATS_ESTABLISHED(S_CONN_CONNECT);
efc23dce
 							c->state=S_CONN_OK; /* partial write => connect()
 													ended */
7bb2b4ca
 						}
d22b82a0
 						/* add to the write queue */
 						lock_get(&c->write_lock);
5b892e3e
 							if (unlikely(_wbufq_insert(c, buf+n, len-n)<0)){
d22b82a0
 								lock_release(&c->write_lock);
 								n=-1;
edcdea00
 								LOG(L_ERR, "ERROR: tcp_send %s: EAGAIN and"
d22b82a0
 										" write queue full or failed for %p\n",
e88c2542
 										su2a(&dst->to, sizeof(dst->to)),
d22b82a0
 										c);
 								goto conn_wait_error;
 							}
 						lock_release(&c->write_lock);
 						/* send to tcp_main */
 						response[0]=(long)c;
 						response[1]=CONN_NEW_PENDING_WRITE;
 						if (unlikely(send_fd(unix_tcp_sock, response, 
 												sizeof(response), fd) <= 0)){
edcdea00
 							LOG(L_ERR, "BUG: tcp_send %s: "
 										"CONN_NEW_PENDING_WRITE  for %p"
 										" failed:" " %s (%d)\n",
e88c2542
 										su2a(&dst->to, sizeof(dst->to)),
d22b82a0
 										c, strerror(errno), errno);
 							goto conn_wait_error;
 						}
316130a7
 						n=len;
d22b82a0
 						goto end;
 					}
7bb2b4ca
 					/* if first write failed it's most likely a
 					   connect error */
 					switch(errno){
 						case ENETUNREACH:
 						case EHOSTUNREACH:  /* not posix for send() */
38429f23
 #ifdef USE_DST_BLACKLIST
7bb2b4ca
 							if (cfg_get(core, core_cfg, use_dst_blacklist))
38429f23
 								dst_blacklist_add( BLST_ERR_CONNECT, dst, 0);
 #endif /* USE_DST_BLACKLIST */
7bb2b4ca
 							TCP_EV_CONNECT_UNREACHABLE(errno, TCP_LADDR(c),
 									TCP_LPORT(c), TCP_PSU(c), TCP_PROTO(c));
 							break;
 						case ECONNREFUSED:
 						case ECONNRESET:
 #ifdef USE_DST_BLACKLIST
 							if (cfg_get(core, core_cfg, use_dst_blacklist))
 								dst_blacklist_add( BLST_ERR_CONNECT, dst, 0);
 #endif /* USE_DST_BLACKLIST */
 							TCP_EV_CONNECT_RST(errno, TCP_LADDR(c),
 									TCP_LPORT(c), TCP_PSU(c), TCP_PROTO(c));
 							break;
 						default:
 							TCP_EV_CONNECT_ERR(errno, TCP_LADDR(c),
 									TCP_LPORT(c), TCP_PSU(c), TCP_PROTO(c));
 					}
d22b82a0
 					/* error: destroy it directly */
7bb2b4ca
 					TCP_STATS_CONNECT_FAILED();
edcdea00
 					LOG(L_ERR, "ERROR: tcp_send %s: connect & send "
d22b82a0
 										" for %p failed:" " %s (%d)\n",
e88c2542
 										su2a(&dst->to, sizeof(dst->to)),
d22b82a0
 										c, strerror(errno), errno);
 					goto conn_wait_error;
 				}
 				LOG(L_INFO, "tcp_send: quick connect for %p\n", c);
7bb2b4ca
 				TCP_STATS_ESTABLISHED(S_CONN_CONNECT);
efc23dce
 				c->state=S_CONN_OK;
d22b82a0
 				/* send to tcp_main */
 				response[0]=(long)c;
 				response[1]=CONN_NEW_COMPLETE;
 				if (unlikely(send_fd(unix_tcp_sock, response, 
 										sizeof(response), fd) <= 0)){
edcdea00
 					LOG(L_ERR, "BUG: tcp_send %s: CONN_NEW_COMPLETE  for %p"
d22b82a0
 								" failed:" " %s (%d)\n",
e88c2542
 								su2a(&dst->to, sizeof(dst->to)),
d22b82a0
 								c, strerror(errno), errno);
 					goto conn_wait_error;
 				}
 				goto end;
 			}
76cb799e
 #endif /* TCP_CONNECT_WAIT  && TCP_ASYNC */
5c5cd736
 			if (unlikely((c=tcpconn_connect(&dst->to, from, dst->proto))==0)){
edcdea00
 				LOG(L_ERR, "ERROR: tcp_send %s: connect failed\n",
e88c2542
 								su2a(&dst->to, sizeof(dst->to)));
534092ba
 				return -1;
0c5da34b
 			}
7bb2b4ca
 			if (likely(c->state==S_CONN_OK))
 				TCP_STATS_ESTABLISHED(S_CONN_CONNECT);
19782e1c
 			atomic_set(&c->refcnt, 2); /* ref. from here and it will also
 			                              be added in the tcp_main hash */
534092ba
 			fd=c->s;
d22b82a0
 			c->flags|=F_CONN_FD_CLOSED; /* not yet opened in main */
 			/* ? TODO: it might be faster just to queue the write and
 			 * send to main a CONN_NEW_PENDING_WRITE */
0c5da34b
 			
 			/* send the new tcpconn to "tcp main" */
 			response[0]=(long)c;
 			response[1]=CONN_NEW;
0ba367ec
 			n=send_fd(unix_tcp_sock, response, sizeof(response), c->s);
5c5cd736
 			if (unlikely(n<=0)){
edcdea00
 				LOG(L_ERR, "BUG: tcp_send %s: failed send_fd: %s (%d)\n",
e88c2542
 						su2a(&dst->to, sizeof(dst->to)),
534092ba
 						strerror(errno), errno);
19782e1c
 				/* we can safely delete it, it's not referenced by anybody */
 				_tcpconn_free(c);
06aaa54f
 				n=-1;
19782e1c
 				goto end_no_conn;
 			}
f2f969dd
 			goto send_it;
 		}
 get_fd:
76cb799e
 #ifdef TCP_ASYNC
885b9f62
 		/* if data is already queued, we don't need the fd any more */
76cb799e
 		if (unlikely(cfg_get(tcp, tcp_cfg, async) &&
22db42e4
 						(_wbufq_non_empty(c)
d22b82a0
 #ifdef TCP_CONNECT_WAIT
efc23dce
 											|| (c->flags&F_CONN_PENDING)
d22b82a0
 #endif /* TCP_CONNECT_WAIT */
 						) )){
885b9f62
 			lock_get(&c->write_lock);
d22b82a0
 				if (likely(_wbufq_non_empty(c)
 #ifdef TCP_CONNECT_WAIT
efc23dce
 							|| (c->flags&F_CONN_PENDING)
d22b82a0
 #endif /* TCP_CONNECT_WAIT */
 
 							)){
885b9f62
 					do_close_fd=0;
20863813
 					if (unlikely(_wbufq_add(c, buf, len)<0)){
885b9f62
 						lock_release(&c->write_lock);
 						n=-1;
 						goto error;
 					}
 					n=len;
 					lock_release(&c->write_lock);
 					goto release_c;
 				}
 			lock_release(&c->write_lock);
 		}
76cb799e
 #endif /* TCP_ASYNC */
5c5cd736
 		/* check if this is not the same reader process holding
 		 *  c  and if so send directly on c->fd */
 		if (c->reader_pid==my_pid()){
00f45c49
 			DBG("tcp_send: send from reader (%d (%d)), reusing fd\n",
5c5cd736
 					my_pid(), process_no);
 			fd=c->fd;
 			do_close_fd=0; /* don't close the fd on exit, it's in use */
 #ifdef TCP_FD_CACHE
b4fa727e
 			use_fd_cache=0; /* don't cache: problems would arise due to the
 							   close() on cache eviction (if the fd is still 
 							   used). If it has to be cached then dup() _must_ 
 							   be used */
 		}else if (likely(use_fd_cache && 
20c64cc6
 							((fd_cache_e=tcp_fd_cache_get(c))!=0))){
5c5cd736
 			fd=fd_cache_e->fd;
 			do_close_fd=0;
00f45c49
 			DBG("tcp_send: found fd in cache ( %d, %p, %d)\n",
5c5cd736
 					fd, c, fd_cache_e->id);
 #endif /* TCP_FD_CACHE */
 		}else{
06aaa54f
 			DBG("tcp_send: tcp connection found (%p), acquiring fd\n", c);
0c5da34b
 			/* get the fd */
 			response[0]=(long)c;
 			response[1]=CONN_GET_FD;
06aaa54f
 			n=send_all(unix_tcp_sock, response, sizeof(response));
5c5cd736
 			if (unlikely(n<=0)){
534092ba
 				LOG(L_ERR, "BUG: tcp_send: failed to get fd(write):%s (%d)\n",
 						strerror(errno), errno);
06aaa54f
 				n=-1;
534092ba
 				goto release_c;
 			}
f2f969dd
 			DBG("tcp_send, c= %p, n=%d\n", c, n);
b8116f04
 			n=receive_fd(unix_tcp_sock, &tmp, sizeof(tmp), &fd, MSG_WAITALL);
5c5cd736
 			if (unlikely(n<=0)){
534092ba
 				LOG(L_ERR, "BUG: tcp_send: failed to get fd(receive_fd):"
 							" %s (%d)\n", strerror(errno), errno);
06aaa54f
 				n=-1;
885b9f62
 				do_close_fd=0;
06aaa54f
 				goto release_c;
 			}
5c5cd736
 			if (unlikely(c!=tmp)){
06aaa54f
 				LOG(L_CRIT, "BUG: tcp_send: get_fd: got different connection:"
b8116f04
 						"  %p (id= %d, refcnt=%d state=%d) != "
 						"  %p (n=%d)\n",
be7401cc
 						  c,   c->id,   atomic_get(&c->refcnt),   c->state,
b8116f04
 						  tmp, n
06aaa54f
 				   );
 				n=-1; /* fail */
76b0ff49
 				goto end;
534092ba
 			}
e29d027b
 			DBG("tcp_send: after receive_fd: c= %p n=%d fd=%d\n",c, n, fd);
5c5cd736
 		}
0c5da34b
 	
f2f969dd
 	
 send_it:
0c5da34b
 	DBG("tcp_send: sending...\n");
ab130758
 	lock_get(&c->write_lock);
76cb799e
 #ifdef TCP_ASYNC
 	if (likely(cfg_get(tcp, tcp_cfg, async))){
d22b82a0
 		if (_wbufq_non_empty(c)
 #ifdef TCP_CONNECT_WAIT
efc23dce
 			|| (c->flags&F_CONN_PENDING) 
d22b82a0
 #endif /* TCP_CONNECT_WAIT */
 			){
20863813
 			if (unlikely(_wbufq_add(c, buf, len)<0)){
885b9f62
 				lock_release(&c->write_lock);
 				n=-1;
 				goto error;
 			}
 			lock_release(&c->write_lock);
 			n=len;
 			goto end;
 		}
 		n=_tcpconn_write_nb(fd, c, buf, len);
 	}else{
76cb799e
 #endif /* TCP_ASYNC */
e9b02e8e
 #ifdef USE_TLS
 	if (c->type==PROTO_TLS)
 		n=tls_blocking_write(c, fd, buf, len);
 	else
 #endif
518c9339
 		/* n=tcp_blocking_write(c, fd, buf, len); */
3dc4f620
 		n=tsend_stream(fd, buf, len,
ffc72fcf
 						TICKS_TO_S(cfg_get(tcp, tcp_cfg, send_timeout)) *
 						1000);
76cb799e
 #ifdef TCP_ASYNC
885b9f62
 	}
76cb799e
 #else /* ! TCP_ASYNC */
ab130758
 	lock_release(&c->write_lock);
76cb799e
 #endif /* TCP_ASYNC */
d22b82a0
 	
 	DBG("tcp_send: after real write: c= %p n=%d fd=%d\n",c, n, fd);
ab130758
 	DBG("tcp_send: buf=\n%.*s\n", (int)len, buf);
36ac6749
 	if (unlikely(n<(int)len)){
76cb799e
 #ifdef TCP_ASYNC
 		if (cfg_get(tcp, tcp_cfg, async) && 
5b892e3e
 				((n>=0) || errno==EAGAIN || errno==EWOULDBLOCK)){
20863813
 			enable_write_watch=_wbufq_empty(c);
5b892e3e
 			if (n<0) n=0;
efc23dce
 			else if (unlikely(c->state==S_CONN_CONNECT ||
7bb2b4ca
 						c->state==S_CONN_ACCEPT)){
 				TCP_STATS_ESTABLISHED(c->state);
efc23dce
 				c->state=S_CONN_OK; /* something was written */
7bb2b4ca
 			}
5b892e3e
 			if (unlikely(_wbufq_add(c, buf+n, len-n)<0)){
885b9f62
 				lock_release(&c->write_lock);
 				n=-1;
 				goto error;
 			}
 			lock_release(&c->write_lock);
 			n=len;
d22b82a0
 			if (likely(enable_write_watch)){
885b9f62
 				response[0]=(long)c;
 				response[1]=CONN_QUEUED_WRITE;
d22b82a0
 				if (send_all(unix_tcp_sock, response, sizeof(response)) <= 0){
885b9f62
 					LOG(L_ERR, "BUG: tcp_send: error return failed "
d22b82a0
 								"(write):%s (%d)\n", strerror(errno), errno);
885b9f62
 					n=-1;
 					goto error;
 				}
 			}
 			goto end;
19782e1c
 		}else{
 			lock_release(&c->write_lock);
885b9f62
 		}
76cb799e
 #endif /* TCP_ASYNC */
7bb2b4ca
 		if (unlikely(c->state==S_CONN_CONNECT)){
38429f23
 			switch(errno){
 				case ENETUNREACH:
7bb2b4ca
 				case EHOSTUNREACH: /* not posix for send() */
 #ifdef USE_DST_BLACKLIST
 					if (cfg_get(core, core_cfg, use_dst_blacklist))
 						dst_blacklist_su(BLST_ERR_CONNECT, c->rcv.proto,
 											&c->rcv.src_su, 0);
 #endif /* USE_DST_BLACKLIST */
 					TCP_EV_CONNECT_UNREACHABLE(errno, TCP_LADDR(c),
 									TCP_LPORT(c), TCP_PSU(c), TCP_PROTO(c));
 					break;
 				case ECONNREFUSED:
 				case ECONNRESET:
 #ifdef USE_DST_BLACKLIST
 					if (cfg_get(core, core_cfg, use_dst_blacklist))
 						dst_blacklist_su(BLST_ERR_CONNECT, c->rcv.proto,
 											&c->rcv.src_su, 0);
 #endif /* USE_DST_BLACKLIST */
 					TCP_EV_CONNECT_RST(errno, TCP_LADDR(c), TCP_LPORT(c),
 										TCP_PSU(c), TCP_PROTO(c));
 					break;
 				default:
 					TCP_EV_CONNECT_ERR(errno, TCP_LADDR(c), TCP_LPORT(c),
 										TCP_PSU(c), TCP_PROTO(c));
 				}
 			TCP_STATS_CONNECT_FAILED();
 		}else{
 			switch(errno){
 				case ECONNREFUSED:
38429f23
 				case ECONNRESET:
7bb2b4ca
 					TCP_STATS_CON_RESET();
 					/* no break */
 				case ENETUNREACH:
38429f23
 				/*case EHOSTUNREACH: -- not posix */
7bb2b4ca
 #ifdef USE_DST_BLACKLIST
 					if (cfg_get(core, core_cfg, use_dst_blacklist))
 						dst_blacklist_su(BLST_ERR_SEND, c->rcv.proto,
 												&c->rcv.src_su, 0);
 #endif /* USE_DST_BLACKLIST */
38429f23
 					break;
 			}
7bb2b4ca
 		}
edcdea00
 		LOG(L_ERR, "ERROR: tcp_send: failed to send on %p (%s:%d->%s): %s (%d)"
 					"\n", c, ip_addr2a(&c->rcv.dst_ip), c->rcv.dst_port,
e88c2542
 					su2a(&c->rcv.src_su, sizeof(c->rcv.src_su)),
edcdea00
 					strerror(errno), errno);
76cb799e
 #ifdef TCP_ASYNC
885b9f62
 error:
76cb799e
 #endif /* TCP_ASYNC */
ab130758
 		/* error on the connection , mark it as bad and set 0 timeout */
f6e50f08
 		c->state=S_CONN_BAD;
d8b11bbc
 		c->timeout=get_ticks_raw();
ab130758
 		/* tell "main" it should drop this (optional it will t/o anyway?)*/
 		response[0]=(long)c;
 		response[1]=CONN_ERROR;
bf458eea
 		if (send_all(unix_tcp_sock, response, sizeof(response))<=0){
d22b82a0
 			LOG(L_CRIT, "BUG: tcp_send: error return failed (write):%s (%d)\n",
ab130758
 					strerror(errno), errno);
d22b82a0
 			tcpconn_chld_put(c); /* deref. it manually */
06aaa54f
 			n=-1;
ab130758
 		}
bf458eea
 		/* CONN_ERROR will auto-dec refcnt => we must not call tcpconn_put 
 		 * if it succeeds */
5c5cd736
 #ifdef TCP_FD_CACHE
 		if (unlikely(fd_cache_e)){
edcdea00
 			LOG(L_ERR, "ERROR: tcp_send %s: error on cached fd, removing from"
 					" the cache (%d, %p, %d)\n", 
e88c2542
 					su2a(&c->rcv.src_su, sizeof(c->rcv.src_su)),
5c5cd736
 					fd, fd_cache_e->con, fd_cache_e->id);
 			tcp_fd_cache_rm(fd_cache_e);
 			close(fd);
 		}else
 #endif /* TCP_FD_CACHE */
 		if (do_close_fd) close(fd);
1c93f767
 		return n; /* error return, no tcpconn_put */
ab130758
 	}
19782e1c
 	
76cb799e
 #ifdef TCP_ASYNC
19782e1c
 	lock_release(&c->write_lock);
76cb799e
 #endif /* TCP_ASYNC */
efc23dce
 	/* in non-async mode here we're either in S_CONN_OK or S_CONN_ACCEPT*/
7bb2b4ca
 	if (unlikely(c->state==S_CONN_CONNECT || c->state==S_CONN_ACCEPT)){
 			TCP_STATS_ESTABLISHED(c->state);
efc23dce
 			c->state=S_CONN_OK;
7bb2b4ca
 	}
534092ba
 end:
5c5cd736
 #ifdef TCP_FD_CACHE
b4fa727e
 	if (unlikely((fd_cache_e==0) && use_fd_cache)){
5c5cd736
 		tcp_fd_cache_add(c, fd);
 	}else
 #endif /* TCP_FD_CACHE */
 	if (do_close_fd) close(fd);
534092ba
 release_c:
19782e1c
 	tcpconn_chld_put(c); /* release c (dec refcnt & free on 0) */
 end_no_conn:
0c5da34b
 	return n;
d22b82a0
 #ifdef TCP_CONNECT_WAIT
 conn_wait_error:
 	/* connect or send failed on newly created connection which was not
 	 * yet sent to tcp_main (but was already hashed) => don't send to main,
 	 * unhash and destroy directly (if refcnt>2 it will be destroyed when the 
 	 * last sender releases the connection (tcpconn_chld_put(c))) or when
 	 * tcp_main receives a CONN_ERROR it*/
 	c->state=S_CONN_BAD;
 	TCPCONN_LOCK;
 		if (c->flags & F_CONN_HASHED){
 			/* if some other parallel tcp_send did send CONN_ERROR to
 			 * tcp_main, the connection might be already detached */
 			_tcpconn_detach(c);
 			c->flags&=~F_CONN_HASHED;
 			TCPCONN_UNLOCK;
 			tcpconn_put(c);
 		}else
 			TCPCONN_UNLOCK;
 	/* dec refcnt -> mark it for destruction */
 	tcpconn_chld_put(c);
 	return -1;
 #endif /* TCP_CONNET_WAIT */
0c5da34b
 }
 
 
 
f2f969dd
 int tcp_init(struct socket_info* sock_info)
5b532c7f
 {
 	union sockaddr_union* addr;
e6509c23
 	int optval;
20c64cc6
 #ifdef HAVE_TCP_ACCEPT_FILTER
 	struct accept_filter_arg afa;
 #endif /* HAVE_TCP_ACCEPT_FILTER */
b33736bc
 #ifdef DISABLE_NAGLE
 	int flag;
 	struct protoent* pe;
7ec958f3
 
 	if (tcp_proto_no==-1){ /* if not already set */
 		pe=getprotobyname("tcp");
 		if (pe==0){
 			LOG(L_ERR, "ERROR: tcp_init: could not get TCP protocol number\n");
 			tcp_proto_no=-1;
 		}else{
 			tcp_proto_no=pe->p_proto;
 		}
b33736bc
 	}
 #endif
5b532c7f
 	
 	addr=&sock_info->su;
faa66933
 	/* sock_info->proto=PROTO_TCP; */
6eacb2bc
 	if (init_su(addr, &sock_info->address, sock_info->port_no)<0){
5b532c7f
 		LOG(L_ERR, "ERROR: tcp_init: could no init sockaddr_union\n");
 		goto error;
 	}
edcdea00
 	DBG("tcp_init: added %s\n", su2a(addr, sizeof(*addr)));
5b532c7f
 	sock_info->socket=socket(AF2PF(addr->s.sa_family), SOCK_STREAM, 0);
 	if (sock_info->socket==-1){
 		LOG(L_ERR, "ERROR: tcp_init: socket: %s\n", strerror(errno));
 		goto error;
 	}
b33736bc
 #ifdef DISABLE_NAGLE
 	flag=1;
 	if ( (tcp_proto_no!=-1) &&
 		 (setsockopt(sock_info->socket, tcp_proto_no , TCP_NODELAY,
 					 &flag, sizeof(flag))<0) ){
 		LOG(L_ERR, "ERROR: tcp_init: could not disable Nagle: %s\n",
 				strerror(errno));
 	}
 #endif
e6509c23
 
 
39546e5f
 #if  !defined(TCP_DONT_REUSEADDR) 
e6509c23
 	/* Stevens, "Network Programming", Section 7.5, "Generic Socket
      * Options": "...server started,..a child continues..on existing
 	 * connection..listening server is restarted...call to bind fails
 	 * ... ALL TCP servers should specify the SO_REUSEADDRE option 
 	 * to allow the server to be restarted in this situation
 	 *
 	 * Indeed, without this option, the server can't restart.
 	 *   -jiri
 	 */
 	optval=1;
 	if (setsockopt(sock_info->socket, SOL_SOCKET, SO_REUSEADDR,
 				(void*)&optval, sizeof(optval))==-1) {
 		LOG(L_ERR, "ERROR: tcp_init: setsockopt %s\n",
 			strerror(errno));
 		goto error;
 	}
 #endif
39546e5f
 	/* tos */
fe09f315
 	optval = tos;
39546e5f
 	if (setsockopt(sock_info->socket, IPPROTO_IP, IP_TOS, (void*)&optval, 
 				sizeof(optval)) ==-1){
 		LOG(L_WARN, "WARNING: tcp_init: setsockopt tos: %s\n", strerror(errno));
 		/* continue since this is not critical */
 	}
20c64cc6
 #ifdef HAVE_TCP_DEFER_ACCEPT
 	/* linux only */
22db42e4
 	if ((optval=cfg_get(tcp, tcp_cfg, defer_accept))){
20c64cc6
 		if (setsockopt(sock_info->socket, IPPROTO_TCP, TCP_DEFER_ACCEPT,
 					(void*)&optval, sizeof(optval)) ==-1){
 			LOG(L_WARN, "WARNING: tcp_init: setsockopt TCP_DEFER_ACCEPT %s\n",
 						strerror(errno));
 		/* continue since this is not critical */
 		}
 	}
 #endif /* HAVE_TCP_DEFFER_ACCEPT */
 #ifdef HAVE_TCP_SYNCNT
22db42e4
 	if ((optval=cfg_get(tcp, tcp_cfg, syncnt))){
20c64cc6
 		if (setsockopt(sock_info->socket, IPPROTO_TCP, TCP_SYNCNT, &optval,
 						sizeof(optval))<0){
 			LOG(L_WARN, "WARNING: tcp_init: failed to set"
 						" maximum SYN retr. count: %s\n", strerror(errno));
 		}
 	}
 #endif
 #ifdef HAVE_TCP_LINGER2
22db42e4
 	if ((optval=cfg_get(tcp, tcp_cfg, linger2))){
20c64cc6
 		if (setsockopt(sock_info->socket, IPPROTO_TCP, TCP_LINGER2, &optval,
 						sizeof(optval))<0){
 			LOG(L_WARN, "WARNING: tcp_init: failed to set"
 						" maximum LINGER2 timeout: %s\n", strerror(errno));
 		}
 	}
 #endif
 	init_sock_keepalive(sock_info->socket);
5b532c7f
 	if (bind(sock_info->socket, &addr->s, sockaddru_len(*addr))==-1){
0ba367ec
 		LOG(L_ERR, "ERROR: tcp_init: bind(%x, %p, %d) on %s:%d : %s\n",
 				sock_info->socket,  &addr->s, 
4d080f49
 				(unsigned)sockaddru_len(*addr),
5b532c7f
 				sock_info->address_str.s,
0ba367ec
 				sock_info->port_no,
5b532c7f
 				strerror(errno));
 		goto error;
 	}
1ce28841
 	if (listen(sock_info->socket, TCP_LISTEN_BACKLOG)==-1){
5b532c7f
 		LOG(L_ERR, "ERROR: tcp_init: listen(%x, %p, %d) on %s: %s\n",
 				sock_info->socket, &addr->s, 
4d080f49
 				(unsigned)sockaddru_len(*addr),
5b532c7f
 				sock_info->address_str.s,
 				strerror(errno));
 		goto error;
 	}
a0b4a4b9
 #ifdef HAVE_TCP_ACCEPT_FILTER
 	/* freebsd */
22db42e4
 	if (cfg_get(tcp, tcp_cfg, defer_accept)){
a0b4a4b9
 		memset(&afa, 0, sizeof(afa));
 		strcpy(afa.af_name, "dataready");
 		if (setsockopt(sock_info->socket, SOL_SOCKET, SO_ACCEPTFILTER,
 					(void*)&afa, sizeof(afa)) ==-1){
 			LOG(L_WARN, "WARNING: tcp_init: setsockopt SO_ACCEPTFILTER %s\n",
 						strerror(errno));
 		/* continue since this is not critical */
 		}
 	}
 #endif /* HAVE_TCP_ACCEPT_FILTER */
5b532c7f
 	
 	return 0;
 error:
 	if (sock_info->socket!=-1){
 		close(sock_info->socket);
 		sock_info->socket=-1;
 	}
 	return -1;
 }
 
 
 
19782e1c
 /* close tcp_main's fd from a tcpconn
db237b38
  * WARNING: call only in tcp_main context */
19782e1c
 inline static void tcpconn_close_main_fd(struct tcp_connection* tcpconn)
 {
 	int fd;
 	
 	
 	fd=tcpconn->s;
 #ifdef USE_TLS
 	/*FIXME: lock ->writelock ? */
 	if (tcpconn->type==PROTO_TLS)
 		tls_close(tcpconn, fd);
 #endif
 #ifdef TCP_FD_CACHE
22db42e4
 	if (likely(cfg_get(tcp, tcp_cfg, fd_cache))) shutdown(fd, SHUT_RDWR);
19782e1c
 #endif /* TCP_FD_CACHE */
 close_again:
 	if (unlikely(close(fd)<0)){
 		if (errno==EINTR)
 			goto close_again;
 		LOG(L_ERR, "ERROR: tcpconn_put_destroy; close() failed: %s (%d)\n",
 				strerror(errno), errno);
 	}
 }
 
 
 
 /* dec refcnt & frees the connection if refcnt==0
  * returns 1 if the connection is freed, 0 otherwise
  *
  * WARNING: use only from child processes */
 inline static int tcpconn_chld_put(struct tcp_connection* tcpconn)
 {
 	if (unlikely(atomic_dec_and_test(&tcpconn->refcnt))){
d22b82a0
 		DBG("tcpconn_chld_put: destroying connection %p (%d, %d) "
19782e1c
 				"flags %04x\n", tcpconn, tcpconn->id,
 				tcpconn->s, tcpconn->flags);
 		/* sanity checks */
23dc5abf
 		membar_read_atomic_op(); /* make sure we see the current flags */
efc23dce
 		if (unlikely(!(tcpconn->flags & F_CONN_FD_CLOSED) ||
 			(tcpconn->flags &
 				(F_CONN_HASHED|F_CONN_MAIN_TIMER|
 				 F_CONN_READ_W|F_CONN_WRITE_W)) )){
d22b82a0
 			LOG(L_CRIT, "BUG: tcpconn_chld_put: %p bad flags = %0x\n",
19782e1c
 					tcpconn, tcpconn->flags);
 			abort();
 		}
 		_tcpconn_free(tcpconn); /* destroys also the wbuf_q if still present*/
 		return 1;
 	}
 	return 0;
 }
 
 
 
 /* simple destroy function (the connection should be already removed
  * from the hashes and the fds should not be watched anymore for IO)
  */
 inline static void tcpconn_destroy(struct tcp_connection* tcpconn)
 {
 		DBG("tcpconn_destroy: destroying connection %p (%d, %d) "
 				"flags %04x\n", tcpconn, tcpconn->id,
 				tcpconn->s, tcpconn->flags);
 		if (unlikely(tcpconn->flags & F_CONN_HASHED)){
 			LOG(L_CRIT, "BUG: tcpconn_destroy: called with hashed"
 						" connection (%p)\n", tcpconn);
 			/* try to continue */
d22b82a0
 			if (likely(tcpconn->flags & F_CONN_MAIN_TIMER))
 				local_timer_del(&tcp_main_ltimer, &tcpconn->timer);
19782e1c
 			TCPCONN_LOCK;
 				_tcpconn_detach(tcpconn);
 			TCPCONN_UNLOCK;
 		}
 		if (likely(!(tcpconn->flags & F_CONN_FD_CLOSED))){
 			tcpconn_close_main_fd(tcpconn);
 			(*tcp_connections_no)--;
 		}
 		_tcpconn_free(tcpconn); /* destroys also the wbuf_q if still present*/
 }
 
 
 
 /* tries to destroy the connection: dec. refcnt and if 0 destroys the
  *  connection, else it will mark it as BAD and close the main fds
  *
  * returns 1 if the connection was destroyed, 0 otherwise
  *
  * WARNING: - the connection _has_ to be removed from the hash and timer
  *  first (use tcpconn_try_unhash() for this )
  *         - the fd should not be watched anymore (io_watch_del()...)
  *         - must be called _only_ from the tcp_main process context
  *          (or else the fd will remain open)
  */
 inline static int tcpconn_put_destroy(struct tcp_connection* tcpconn)
 {
23dc5abf
 	if (unlikely((tcpconn->flags & 
fc665070
 			(F_CONN_WRITE_W|F_CONN_HASHED|F_CONN_MAIN_TIMER|F_CONN_READ_W)) )){
19782e1c
 		/* sanity check */
23dc5abf
 		if (unlikely(tcpconn->flags & F_CONN_HASHED)){
 			LOG(L_CRIT, "BUG: tcpconn_destroy: called with hashed and/or"
 						"on timer connection (%p), flags = %0x\n",
 						tcpconn, tcpconn->flags);
 			/* try to continue */
 			if (likely(tcpconn->flags & F_CONN_MAIN_TIMER))
 				local_timer_del(&tcp_main_ltimer, &tcpconn->timer);
 			TCPCONN_LOCK;
 				_tcpconn_detach(tcpconn);
 			TCPCONN_UNLOCK;
 		}else{
 			LOG(L_CRIT, "BUG: tcpconn_put_destroy: %p flags = %0x\n",
19782e1c
 					tcpconn, tcpconn->flags);
23dc5abf
 		}
19782e1c
 	}
23dc5abf
 	tcpconn->state=S_CONN_BAD;
 	/* in case it's still in a reader timer */
 	tcpconn->timeout=get_ticks_raw();
 	/* fast close: close fds now */
 	if (likely(!(tcpconn->flags & F_CONN_FD_CLOSED))){
 		tcpconn_close_main_fd(tcpconn);
 		tcpconn->flags|=F_CONN_FD_CLOSED;
 		(*tcp_connections_no)--;
 	}
 	/* all the flags / ops on the tcpconn must be done prior to decrementing
 	 * the refcnt. and at least a membar_write_atomic_op() mem. barrier or
 	 *  a mb_atomic_* op must * be used to make sure all the changed flags are
 	 *  written into memory prior to the new refcnt value */
 	if (unlikely(mb_atomic_dec_and_test(&tcpconn->refcnt))){
 		_tcpconn_free(tcpconn);
19782e1c
 		return 1;
 	}
 	return 0;
 }
 
 
 
 /* try to remove a connection from the hashes and timer.
  * returns 1 if the connection was removed, 0 if not (connection not in
  *  hash)
  *
  * WARNING: call it only in the  tcp_main process context or else the
  *  timer removal won't work.
  */
 inline static int tcpconn_try_unhash(struct tcp_connection* tcpconn)
 {
 	if (likely(tcpconn->flags & F_CONN_HASHED)){
 		tcpconn->state=S_CONN_BAD;
d22b82a0
 		if (likely(tcpconn->flags & F_CONN_MAIN_TIMER)){
19782e1c
 			local_timer_del(&tcp_main_ltimer, &tcpconn->timer);
d22b82a0
 			tcpconn->flags&=~F_CONN_MAIN_TIMER;
 		}else
19782e1c
 			/* in case it's still in a reader timer */
 			tcpconn->timeout=get_ticks_raw();
 		TCPCONN_LOCK;
d22b82a0
 			if (tcpconn->flags & F_CONN_HASHED){
 				tcpconn->flags&=~F_CONN_HASHED;
 				_tcpconn_detach(tcpconn);
 				TCPCONN_UNLOCK;
 			}else{
 				/* tcp_send was faster and did unhash it itself */
 				TCPCONN_UNLOCK;
 				return 0;
 			}
76cb799e
 #ifdef TCP_ASYNC
19782e1c
 		/* empty possible write buffers (optional) */
 		if (unlikely(_wbufq_non_empty(tcpconn))){
 			lock_get(&tcpconn->write_lock);
 				/* check again, while holding the lock */
 				if (likely(_wbufq_non_empty(tcpconn)))
 					_wbufq_destroy(&tcpconn->wbuf_q);
 			lock_release(&tcpconn->write_lock);
 		}
76cb799e
 #endif /* TCP_ASYNC */
19782e1c
 		return 1;
 	}
 	return 0;
 }
746f7674
 
 
 
98f3d5e2
 #ifdef SEND_FD_QUEUE
 struct send_fd_info{
 	struct tcp_connection* tcp_conn;
746f7674
 	ticks_t expire;
98f3d5e2
 	int unix_sock;
746f7674
 	unsigned int retries; /* debugging */
98f3d5e2
 };
 
 struct tcp_send_fd_q{
 	struct send_fd_info* data; /* buffer */
 	struct send_fd_info* crt;  /* pointer inside the buffer */
 	struct send_fd_info* end;  /* points after the last valid position */
 };
 
 
 static struct tcp_send_fd_q send2child_q;
 
 
 
 static int send_fd_queue_init(struct tcp_send_fd_q *q, unsigned int size)
5b532c7f
 {
98f3d5e2
 	q->data=pkg_malloc(size*sizeof(struct send_fd_info));
 	if (q->data==0){
 		LOG(L_ERR, "ERROR: send_fd_queue_init: out of memory\n");
06aaa54f
 		return -1;
 	}
98f3d5e2
 	q->crt=&q->data[0];
 	q->end=&q->data[size];
06aaa54f
 	return 0;
5b532c7f
 }
 
98f3d5e2
 static void send_fd_queue_destroy(struct tcp_send_fd_q *q)
 {
 	if (q->data){
 		pkg_free(q->data);
 		q->data=0;
 		q->crt=q->end=0;
 	}
 }
5b532c7f
 
98f3d5e2
 
 
 static int init_send_fd_queues()
f6e50f08
 {
98f3d5e2
 	if (send_fd_queue_init(&send2child_q, SEND_FD_QUEUE_SIZE)!=0)
 		goto error;
 	return 0;
 error:
 	LOG(L_ERR, "ERROR: init_send_fd_queues: init failed\n");
 	return -1;
 }
 
 
 
 static void destroy_send_fd_queues()
 {
 	send_fd_queue_destroy(&send2child_q);
 }
 
 
 
 
 inline static int send_fd_queue_add(	struct tcp_send_fd_q* q, 
 										int unix_sock,
 										struct tcp_connection *t)
 {
 	struct send_fd_info* tmp;
 	unsigned long new_size;
f6e50f08
 	
98f3d5e2
 	if (q->crt>=q->end){
 		new_size=q->end-&q->data[0];
 		if (new_size< MAX_SEND_FD_QUEUE_SIZE/2){
 			new_size*=2;
 		}else new_size=MAX_SEND_FD_QUEUE_SIZE;
ccb7fda2
 		if (unlikely(q->crt>=&q->data[new_size])){
98f3d5e2
 			LOG(L_ERR, "ERROR: send_fd_queue_add: queue full: %ld/%ld\n",
d9515405
 					(long)(q->crt-&q->data[0]-1), new_size);
98f3d5e2
 			goto error;
 		}
 		LOG(L_CRIT, "INFO: send_fd_queue: queue full: %ld, extending to %ld\n",
d9515405
 				(long)(q->end-&q->data[0]), new_size);
98f3d5e2
 		tmp=pkg_realloc(q->data, new_size*sizeof(struct send_fd_info));
ccb7fda2
 		if (unlikely(tmp==0)){
98f3d5e2
 			LOG(L_ERR, "ERROR: send_fd_queue_add: out of memory\n");
 			goto error;
 		}
 		q->crt=(q->crt-&q->data[0])+tmp;
 		q->data=tmp;
 		q->end=&q->data[new_size];
0ba367ec
 	}
98f3d5e2
 	q->crt->tcp_conn=t;
 	q->crt->unix_sock=unix_sock;
746f7674
 	q->crt->expire=get_ticks_raw()+SEND_FD_QUEUE_TIMEOUT;
98f3d5e2
 	q->crt->retries=0;
 	q->crt++;
 	return 0;
 error:
 	return -1;
 }
 
 
 
 inline static void send_fd_queue_run(struct tcp_send_fd_q* q)
 {
 	struct send_fd_info* p;
 	struct send_fd_info* t;
0ba367ec
 	
98f3d5e2
 	for (p=t=&q->data[0]; p<q->crt; p++){
ccb7fda2
 		if (unlikely(send_fd(p->unix_sock, &(p->tcp_conn),
 					sizeof(struct tcp_connection*), p->tcp_conn->s)<=0)){
28260509
 			if ( ((errno==EAGAIN)||(errno==EWOULDBLOCK)) && 
746f7674
 							((s_ticks_t)(p->expire-get_ticks_raw())>0)){
98f3d5e2
 				/* leave in queue for a future try */
 				*t=*p;
 				t->retries++;
 				t++;
 			}else{
28260509
 				LOG(L_ERR, "ERROR: run_send_fd_queue: send_fd failed"
746f7674
 						   " on socket %d , queue entry %ld, retries %d,"
 						   " connection %p, tcp socket %d, errno=%d (%s) \n",
d9515405
 						   p->unix_sock, (long)(p-&q->data[0]), p->retries,
746f7674
 						   p->tcp_conn, p->tcp_conn->s, errno,
 						   strerror(errno));
76cb799e
 #ifdef TCP_ASYNC
885b9f62
 				if (p->tcp_conn->flags & F_CONN_WRITE_W){
 					io_watch_del(&io_h, p->tcp_conn->s, -1, IO_FD_CLOSING);
 					p->tcp_conn->flags &=~F_CONN_WRITE_W;
 				}
 #endif
 				p->tcp_conn->flags &= ~F_CONN_READER;
19782e1c
 				if (likely(tcpconn_try_unhash(p->tcp_conn)))
 					tcpconn_put(p->tcp_conn);
 				tcpconn_put_destroy(p->tcp_conn); /* dec refcnt & destroy */
98f3d5e2
 			}
bc977837
 		}
f6e50f08
 	}
98f3d5e2
 	q->crt=t;
f6e50f08
 }
98f3d5e2
 #else
 #define send_fd_queue_run(q)
 #endif
f6e50f08
 
 
885b9f62
 /* non blocking write() on a tcpconnection, unsafe version (should be called
  * while holding  c->write_lock). The fd should be non-blocking.
  *  returns number of bytes written on success, -1 on error (and sets errno)
  */
 inline static int _tcpconn_write_nb(int fd, struct tcp_connection* c,
 									char* buf, int len)
 {
 	int n;
 	
 again:
 #ifdef USE_TLS
 	if (unlikely(c->type==PROTO_TLS))
 		/* FIXME: tls_nonblocking_write !! */
 		n=tls_blocking_write(c, fd, buf, len);
 	else
 #endif /* USE_TLS */
 		n=send(fd, buf, len,
 #ifdef HAVE_MSG_NOSIGNAL
 					MSG_NOSIGNAL
 #else
 					0
 #endif /* HAVE_MSG_NOSIGNAL */
 			  );
 	if (unlikely(n<0)){
 		if (errno==EINTR) goto again;
 	}
 	return n;
 }
 
 
0ba367ec
 
 /* handles io from a tcp child process
  * params: tcp_c - pointer in the tcp_children array, to the entry for
  *                 which an io event was detected 
  *         fd_i  - fd index in the fd_array (usefull for optimizing
  *                 io_watch_deletes)
  * returns:  handle_* return convention: -1 on error, 0 on EAGAIN (no more
  *           io events queued), >0 on success. success/error refer only to
  *           the reads from the fd.
  */
 inline static int handle_tcp_child(struct tcp_child* tcp_c, int fd_i)
5b532c7f
 {
 	struct tcp_connection* tcpconn;
 	long response[2];
0c5da34b
 	int cmd;
6a157851
 	int bytes;
885b9f62
 	int n;
ccb7fda2
 	ticks_t t;
20863813
 	ticks_t crt_timeout;
3dc4f620
 	ticks_t con_lifetime;
0ba367ec
 	
ccb7fda2
 	if (unlikely(tcp_c->unix_sock<=0)){
0ba367ec
 		/* (we can't have a fd==0, 0 is never closed )*/
 		LOG(L_CRIT, "BUG: handle_tcp_child: fd %d for %d "
 				"(pid %d, ser no %d)\n", tcp_c->unix_sock,
 				(int)(tcp_c-&tcp_children[0]), tcp_c->pid, tcp_c->proc_no);
 		goto error;
 	}
 	/* read until sizeof(response)
 	 * (this is a SOCK_STREAM so read is not atomic) */
 	bytes=recv_all(tcp_c->unix_sock, response, sizeof(response), MSG_DONTWAIT);
ccb7fda2
 	if (unlikely(bytes<(int)sizeof(response))){
0ba367ec
 		if (bytes==0){
 			/* EOF -> bad, child has died */
 			DBG("DBG: handle_tcp_child: dead tcp child %d (pid %d, no %d)"
 					" (shutting down?)\n", (int)(tcp_c-&tcp_children[0]), 
 					tcp_c->pid, tcp_c->proc_no );
741a9937
 			/* don't listen on it any more */
 			io_watch_del(&io_h, tcp_c->unix_sock, fd_i, 0); 
0ba367ec
 			goto error; /* eof. so no more io here, it's ok to return error */
 		}else if (bytes<0){
 			/* EAGAIN is ok if we try to empty the buffer
 			 * e.g.: SIGIO_RT overflow mode or EPOLL ET */
 			if ((errno!=EAGAIN) && (errno!=EWOULDBLOCK)){
2b04f2a8
 				LOG(L_CRIT, "ERROR: handle_tcp_child: read from tcp child %ld "
0ba367ec
 						" (pid %d, no %d) %s [%d]\n",
2b04f2a8
 						(long)(tcp_c-&tcp_children[0]), tcp_c->pid,
 						tcp_c->proc_no, strerror(errno), errno );
0ba367ec
 			}else{
 				bytes=0;
 			}
 			/* try to ignore ? */
 			goto end;
 		}else{
 			/* should never happen */
 			LOG(L_CRIT, "BUG: handle_tcp_child: too few bytes received (%d)\n",
 					bytes );
 			bytes=0; /* something was read so there is no error; otoh if
 					  receive_fd returned less then requested => the receive
 					  buffer is empty => no more io queued on this fd */
 			goto end;
 		}
 	}
 	
 	DBG("handle_tcp_child: reader response= %lx, %ld from %d \n",
 					response[0], response[1], (int)(tcp_c-&tcp_children[0]));
 	cmd=response[1];
 	tcpconn=(struct tcp_connection*)response[0];
ccb7fda2
 	if (unlikely(tcpconn==0)){
0ba367ec
 		/* should never happen */
 		LOG(L_CRIT, "BUG: handle_tcp_child: null tcpconn pointer received"
 				 " from tcp child %d (pid %d): %lx, %lx\n",
 				 	(int)(tcp_c-&tcp_children[0]), tcp_c->pid,
 					response[0], response[1]) ;
 		goto end;
 	}
 	switch(cmd){
 		case CONN_RELEASE:
 			tcp_c->busy--;
19782e1c
 			if (unlikely(tcpconn_put(tcpconn))){
 				tcpconn_destroy(tcpconn);
 				break;
 			}
ccb7fda2
 			if (unlikely(tcpconn->state==S_CONN_BAD)){ 
76cb799e
 #ifdef TCP_ASYNC
885b9f62
 				if (unlikely(tcpconn->flags & F_CONN_WRITE_W)){
 					io_watch_del(&io_h, tcpconn->s, -1, IO_FD_CLOSING);
 					tcpconn->flags &= ~F_CONN_WRITE_W;
 				}
76cb799e
 #endif /* TCP_ASYNC */
19782e1c
 				if (tcpconn_try_unhash(tcpconn))
 					tcpconn_put_destroy(tcpconn);
0ba367ec
 				break;
 			}
 			/* update the timeout*/
ccb7fda2
 			t=get_ticks_raw();
3dc4f620
 			con_lifetime=cfg_get(tcp, tcp_cfg, con_lifetime);
 			tcpconn->timeout=t+con_lifetime;
 			crt_timeout=con_lifetime;
76cb799e
 #ifdef TCP_ASYNC
 			if (unlikely(cfg_get(tcp, tcp_cfg, async) && 
20863813
 							_wbufq_non_empty(tcpconn) )){
dda6ba60
 				if (unlikely(TICKS_GE(t, tcpconn->wbuf_q.wr_timeout))){
20863813
 					DBG("handle_tcp_child: wr. timeout on CONN_RELEASE for %p "
 							"refcnt= %d\n", tcpconn,
 							atomic_get(&tcpconn->refcnt));
 					/* timeout */
7bb2b4ca
 					if (unlikely(tcpconn->state==S_CONN_CONNECT)){
38429f23
 #ifdef USE_DST_BLACKLIST
7bb2b4ca
 						if (cfg_get(core, core_cfg, use_dst_blacklist))
 							dst_blacklist_su( BLST_ERR_CONNECT,
 													tcpconn->rcv.proto,
 													&tcpconn->rcv.src_su, 0);
 #endif /* USE_DST_BLACKLIST */
 						TCP_EV_CONNECT_TIMEOUT(0, TCP_LADDR(tcpconn),
 										TCP_LPORT(tcpconn), TCP_PSU(tcpconn),
 										TCP_PROTO(tcpconn));
 						TCP_STATS_CONNECT_FAILED();
 					}else{
 #ifdef USE_DST_BLACKLIST
 						if (cfg_get(core, core_cfg, use_dst_blacklist))
 							dst_blacklist_su( BLST_ERR_SEND,
38429f23
 													tcpconn->rcv.proto,
 													&tcpconn->rcv.src_su, 0);
 #endif /* USE_DST_BLACKLIST */
7bb2b4ca
 						TCP_EV_SEND_TIMEOUT(0, &tcpconn->rcv);
 						TCP_STATS_SEND_TIMEOUT();
 					}
20863813
 					if (unlikely(tcpconn->flags & F_CONN_WRITE_W)){
 						io_watch_del(&io_h, tcpconn->s, -1, IO_FD_CLOSING);
 						tcpconn->flags&=~F_CONN_WRITE_W;
 					}
19782e1c
 					if (tcpconn_try_unhash(tcpconn))
 						tcpconn_put_destroy(tcpconn);
20863813
 					break;
 				}else{
3dc4f620
 					crt_timeout=MIN_unsigned(con_lifetime,
20863813
 											tcpconn->wbuf_q.wr_timeout-t);
 				}
 			}
76cb799e
 #endif /* TCP_ASYNC */
ccb7fda2
 			/* re-activate the timer */
 			tcpconn->timer.f=tcpconn_main_timeout;
8b0472d7
 			local_timer_reinit(&tcpconn->timer);
20863813
 			local_timer_add(&tcp_main_ltimer, &tcpconn->timer, crt_timeout, t);
3135b4bb
 			/* must be after the de-ref*/
92c0024c
 			tcpconn->flags|=(F_CONN_MAIN_TIMER|F_CONN_READ_W|F_CONN_WANTS_RD);
fc665070
 			tcpconn->flags&=~(F_CONN_READER|F_CONN_OOB_DATA);
76cb799e
 #ifdef TCP_ASYNC
885b9f62
 			if (unlikely(tcpconn->flags & F_CONN_WRITE_W))
 				n=io_watch_chg(&io_h, tcpconn->s, POLLIN| POLLOUT, -1);
 			else
76cb799e
 #endif /* TCP_ASYNC */
885b9f62
 				n=io_watch_add(&io_h, tcpconn->s, POLLIN, F_TCPCONN, tcpconn);
 			if (unlikely(n<0)){
ccb7fda2
 				LOG(L_CRIT, "ERROR: tcp_main: handle_tcp_child: failed to add"
 						" new socket to the fd list\n");
fc665070
 				tcpconn->flags&=~F_CONN_READ_W;
76cb799e
 #ifdef TCP_ASYNC
885b9f62
 				if (unlikely(tcpconn->flags & F_CONN_WRITE_W)){
 					io_watch_del(&io_h, tcpconn->s, -1, IO_FD_CLOSING);
 					tcpconn->flags&=~F_CONN_WRITE_W;
 				}
76cb799e
 #endif /* TCP_ASYNC */
575a0b47
 				if (tcpconn_try_unhash(tcpconn))
19782e1c
 					tcpconn_put_destroy(tcpconn);
 				break;
ccb7fda2
 			}
0ba367ec
 			DBG("handle_tcp_child: CONN_RELEASE  %p refcnt= %d\n", 
be7401cc
 							tcpconn, atomic_get(&tcpconn->refcnt));
0ba367ec
 			break;
 		case CONN_ERROR:
 		case CONN_DESTROY:
 		case CONN_EOF:
 			/* WARNING: this will auto-dec. refcnt! */
 				tcp_c->busy--;
 				/* main doesn't listen on it => we don't have to delete it
 				 if (tcpconn->s!=-1)
741a9937
 					io_watch_del(&io_h, tcpconn->s, -1, IO_FD_CLOSING);
0ba367ec
 				*/
76cb799e
 #ifdef TCP_ASYNC
885b9f62
 				if ((tcpconn->flags & F_CONN_WRITE_W) && (tcpconn->s!=-1)){
 					io_watch_del(&io_h, tcpconn->s, -1, IO_FD_CLOSING);
 					tcpconn->flags&=~F_CONN_WRITE_W;
 				}
76cb799e
 #endif /* TCP_ASYNC */
19782e1c
 				if (tcpconn_try_unhash(tcpconn))
 					tcpconn_put(tcpconn);
 				tcpconn_put_destroy(tcpconn); /* deref & delete if refcnt==0 */
0ba367ec
 				break;
 		default:
 				LOG(L_CRIT, "BUG: handle_tcp_child:  unknown cmd %d"
 									" from tcp reader %d\n",
 									cmd, (int)(tcp_c-&tcp_children[0]));
 	}
 end:
 	return bytes;
 error:
 	return -1;
 }
 
 
 
 /* handles io from a "generic" ser process (get fd or new_fd from a tcp_send)
  * 
  * params: p     - pointer in the ser processes array (pt[]), to the entry for
  *                 which an io event was detected
  *         fd_i  - fd index in the fd_array (usefull for optimizing
  *                 io_watch_deletes)
  * returns:  handle_* return convention:
  *          -1 on error reading from the fd,
  *           0 on EAGAIN  or when no  more io events are queued 
  *             (receive buffer empty),
  *           >0 on successfull reads from the fd (the receive buffer might
  *             be non-empty).
  */
 inline static int handle_ser_child(struct process_table* p, int fd_i)
 {
 	struct tcp_connection* tcpconn;
 	long response[2];
 	int cmd;
 	int bytes;
 	int ret;
744a2341
 	int fd;
885b9f62
 	int flags;
ccb7fda2
 	ticks_t t;
3dc4f620
 	ticks_t con_lifetime;
76cb799e
 #ifdef TCP_ASYNC
dda6ba60
 	ticks_t nxt_timeout;
76cb799e
 #endif /* TCP_ASYNC */
0ba367ec
 	
 	ret=-1;
ccb7fda2
 	if (unlikely(p->unix_sock<=0)){
0ba367ec
 		/* (we can't have a fd==0, 0 is never closed )*/
 		LOG(L_CRIT, "BUG: handle_ser_child: fd %d for %d "
 				"(pid %d)\n", p->unix_sock, (int)(p-&pt[0]), p->pid);
 		goto error;
 	}
 			
 	/* get all bytes and the fd (if transmitted)
 	 * (this is a SOCK_STREAM so read is not atomic) */
 	bytes=receive_fd(p->unix_sock, response, sizeof(response), &fd,
 						MSG_DONTWAIT);
ccb7fda2
 	if (unlikely(bytes<(int)sizeof(response))){
0ba367ec
 		/* too few bytes read */
 		if (bytes==0){
 			/* EOF -> bad, child has died */
 			DBG("DBG: handle_ser_child: dead child %d, pid %d"
 					" (shutting down?)\n", (int)(p-&pt[0]), p->pid);
 			/* don't listen on it any more */
741a9937
 			io_watch_del(&io_h, p->unix_sock, fd_i, 0);
0ba367ec
 			goto error; /* child dead => no further io events from it */
 		}else if (bytes<0){
 			/* EAGAIN is ok if we try to empty the buffer
 			 * e.g: SIGIO_RT overflow mode or EPOLL ET */
 			if ((errno!=EAGAIN) && (errno!=EWOULDBLOCK)){
 				LOG(L_CRIT, "ERROR: handle_ser_child: read from child %d  "
 						"(pid %d):  %s [%d]\n", (int)(p-&pt[0]), p->pid,
 						strerror(errno), errno);
 				ret=-1;
 			}else{
 				ret=0;
 			}
 			/* try to ignore ? */
 			goto end;
 		}else{
 			/* should never happen */
 			LOG(L_CRIT, "BUG: handle_ser_child: too few bytes received (%d)\n",
 					bytes );
 			ret=0; /* something was read so there is no error; otoh if
 					  receive_fd returned less then requested => the receive
 					  buffer is empty => no more io queued on this fd */
 			goto end;
 		}
 	}
 	ret=1; /* something was received, there might be more queued */
 	DBG("handle_ser_child: read response= %lx, %ld, fd %d from %d (%d)\n",
 					response[0], response[1], fd, (int)(p-&pt[0]), p->pid);
 	cmd=response[1];
 	tcpconn=(struct tcp_connection*)response[0];
ccb7fda2
 	if (unlikely(tcpconn==0)){
0ba367ec
 		LOG(L_CRIT, "BUG: handle_ser_child: null tcpconn pointer received"
 				 " from child %d (pid %d): %lx, %lx\n",
 				 	(int)(p-&pt[0]), p->pid, response[0], response[1]) ;
 		goto end;
 	}
 	switch(cmd){
 		case CONN_ERROR:
d22b82a0
 			LOG(L_ERR, "handle_ser_child: ERROR: received CON_ERROR for %p"
 					" (id %d), refcnt %d\n", 
 					tcpconn, tcpconn->id, atomic_get(&tcpconn->refcnt));
 #ifdef TCP_CONNECT_WAIT
 			/* if the connection is pending => it might be on the way of
 			 * reaching tcp_main (e.g. CONN_NEW_COMPLETE or 
 			 *  CONN_NEW_PENDING_WRITE) =>  it cannot be destroyed here */
 			if ( !(tcpconn->flags & F_CONN_PENDING) && 
 					tcpconn_try_unhash(tcpconn) )
 				tcpconn_put(tcpconn);
 #else /* ! TCP_CONNECT_WAIT */
 			if ( tcpconn_try_unhash(tcpconn) )
 				tcpconn_put(tcpconn);
 #endif /* TCP_CONNECT_WAIT */
fc665070
 			if ( ((tcpconn->flags & (F_CONN_WRITE_W|F_CONN_READ_W)) ) &&
 					(tcpconn->s!=-1)){
741a9937
 				io_watch_del(&io_h, tcpconn->s, -1, IO_FD_CLOSING);
fc665070
 				tcpconn->flags&=~(F_CONN_WRITE_W|F_CONN_READ_W);
06c04bc2
 			}
19782e1c
 			tcpconn_put_destroy(tcpconn); /* dec refcnt & destroy on 0 */
0ba367ec
 			break;
 		case CONN_GET_FD:
 			/* send the requested FD  */
 			/* WARNING: take care of setting refcnt properly to
19782e1c
 			 * avoid race conditions */
ccb7fda2
 			if (unlikely(send_fd(p->unix_sock, &tcpconn, sizeof(tcpconn),
 								tcpconn->s)<=0)){
0ba367ec
 				LOG(L_ERR, "ERROR: handle_ser_child: send_fd failed\n");
 			}
 			break;
 		case CONN_NEW:
 			/* update the fd in the requested tcpconn*/
 			/* WARNING: take care of setting refcnt properly to
19782e1c
 			 * avoid race conditions */
ccb7fda2
 			if (unlikely(fd==-1)){
0ba367ec
 				LOG(L_CRIT, "BUG: handle_ser_child: CONN_NEW:"
 							" no fd received\n");
19782e1c
 				tcpconn->flags|=F_CONN_FD_CLOSED;
 				tcpconn_put_destroy(tcpconn);
0ba367ec
 				break;
 			}
d9515405
 			(*tcp_connections_no)++;
0ba367ec
 			tcpconn->s=fd;
 			/* add tcpconn to the list*/
 			tcpconn_add(tcpconn);
 			/* update the timeout*/
ccb7fda2
 			t=get_ticks_raw();
3dc4f620
 			con_lifetime=cfg_get(tcp, tcp_cfg, con_lifetime);
 			tcpconn->timeout=t+con_lifetime;
20863813
 			/* activate the timer (already properly init. in tcpconn_new())
19782e1c
 			 * no need for reinit */
ccb7fda2
 			local_timer_add(&tcp_main_ltimer, &tcpconn->timer, 
3dc4f620
 								con_lifetime, t);
92c0024c
 			tcpconn->flags|=(F_CONN_MAIN_TIMER|F_CONN_READ_W|F_CONN_WANTS_RD)
76cb799e
 #ifdef TCP_ASYNC
92c0024c
 					/* not used for now, the connection is sent to tcp_main
 					 * before knowing whether we can write on it or we should 
 					 * wait */
 							| (((int)!(tcpconn->flags & F_CONN_WANTS_WR)-1)& 
 								F_CONN_WRITE_W)
76cb799e
 #endif /* TCP_ASYNC */
92c0024c
 				;
fc665070
 			tcpconn->flags&=~F_CONN_FD_CLOSED;
885b9f62
 			flags=POLLIN 
76cb799e
 #ifdef TCP_ASYNC
885b9f62
 					/* not used for now, the connection is sent to tcp_main
 					 * before knowing if we can write on it or we should 
 					 * wait */
92c0024c
 					| (((int)!(tcpconn->flags & F_CONN_WANTS_WR)-1) & POLLOUT)
76cb799e
 #endif /* TCP_ASYNC */
885b9f62
 					;
ccb7fda2
 			if (unlikely(
885b9f62
 					io_watch_add(&io_h, tcpconn->s, flags,
a0553f4e
 												F_TCPCONN, tcpconn)<0)){
ccb7fda2
 				LOG(L_CRIT, "ERROR: tcp_main: handle_ser_child: failed to add"
 						" new socket to the fd list\n");
fc665070
 				tcpconn->flags&=~(F_CONN_WRITE_W|F_CONN_READ_W);
19782e1c
 				tcpconn_try_unhash(tcpconn); /*  unhash & dec refcnt */
 				tcpconn_put_destroy(tcpconn);
ccb7fda2
 			}
0ba367ec
 			break;
76cb799e
 #ifdef TCP_ASYNC
885b9f62
 		case CONN_QUEUED_WRITE:
d22b82a0
 			/* received only if the wr. queue is empty and a write finishes
 			 * with EAGAIN (common after connect())
 			 * it should only enable write watching on the fd. The connection
 			 * should be  already in the hash. The refcnt is not changed.
 			 */
19782e1c
 			if (unlikely((tcpconn->state==S_CONN_BAD) || 
 							!(tcpconn->flags & F_CONN_HASHED) ))
 				break;
92c0024c
 			if (!(tcpconn->flags & F_CONN_WANTS_WR)){
 				tcpconn->flags|=F_CONN_WANTS_WR;
2a45890d
 				t=get_ticks_raw();
 				if (likely((tcpconn->flags & F_CONN_MAIN_TIMER) && 
 					(TICKS_LT(tcpconn->wbuf_q.wr_timeout, tcpconn->timeout)) &&
 						TICKS_LT(t, tcpconn->wbuf_q.wr_timeout) )){
 					/* _wbufq_nonempty() is guaranteed here */
 					/* update the timer */
 					local_timer_del(&tcp_main_ltimer, &tcpconn->timer);
 					local_timer_reinit(&tcpconn->timer);
 					local_timer_add(&tcp_main_ltimer, &tcpconn->timer,
 										tcpconn->wbuf_q.wr_timeout-t, t);
 					DBG("tcp_main: handle_ser_child: CONN_QUEUED_WRITE; %p "
 							"timeout adjusted to %d s\n", tcpconn, 
 							TICKS_TO_S(tcpconn->wbuf_q.wr_timeout-t));
 				}
92c0024c
 				if (!(tcpconn->flags & F_CONN_WRITE_W)){
 					tcpconn->flags|=F_CONN_WRITE_W;
 					if (!(tcpconn->flags & F_CONN_READ_W)){
 						if (unlikely(io_watch_add(&io_h, tcpconn->s, POLLOUT,
885b9f62
 												F_TCPCONN, tcpconn)<0)){
92c0024c
 							LOG(L_CRIT, "ERROR: tcp_main: handle_ser_child:"
 										" failed to enable write watch on"
 										" socket\n");
 							if (tcpconn_try_unhash(tcpconn))
 								tcpconn_put_destroy(tcpconn);
 							break;
 						}
 					}else{
 						if (unlikely(io_watch_chg(&io_h, tcpconn->s,
 													POLLIN|POLLOUT, -1)<0)){
 							LOG(L_CRIT, "ERROR: tcp_main: handle_ser_child:"
 									" failed to change socket watch events\n");
 							io_watch_del(&io_h, tcpconn->s, -1, IO_FD_CLOSING);
 							tcpconn->flags&=~F_CONN_READ_W;
 							if (tcpconn_try_unhash(tcpconn))
 								tcpconn_put_destroy(tcpconn);
 							break;
 						}
885b9f62
 					}
 				}
 			}else{
 				LOG(L_WARN, "tcp_main: hanlder_ser_child: connection %p"
 							" already watched for write\n", tcpconn);
 			}
 			break;
d22b82a0
 #ifdef TCP_CONNECT_WAIT
 		case CONN_NEW_COMPLETE:
 		case CONN_NEW_PENDING_WRITE:
 				/* received when a pending connect completes in the same
 				 * tcp_send() that initiated it
efc23dce
 				 * the connection is already in the hash with F_CONN_PENDING
 				 * flag (added by tcp_send()) and refcnt at least 1 (for the
d22b82a0
 				 *  hash)*/
fc665070
 			tcpconn->flags&=~(F_CONN_PENDING|F_CONN_FD_CLOSED);
d22b82a0
 			if (unlikely((tcpconn->state==S_CONN_BAD) || (fd==-1))){
 				if (unlikely(fd==-1))
 					LOG(L_CRIT, "BUG: handle_ser_child: CONN_NEW_COMPLETE:"
 								" no fd received\n");
 				else
 					LOG(L_WARN, "WARNING: handle_ser_child: CONN_NEW_COMPLETE:"
 							" received connection with error\n");
 				tcpconn->flags|=F_CONN_FD_CLOSED;
 				tcpconn->state=S_CONN_BAD;
 				tcpconn_try_unhash(tcpconn);
 				tcpconn_put_destroy(tcpconn);
 				break;
 			}
 			(*tcp_connections_no)++;
 			tcpconn->s=fd;
 			/* update the timeout*/
 			t=get_ticks_raw();
3dc4f620
 			con_lifetime=cfg_get(tcp, tcp_cfg, con_lifetime);
 			tcpconn->timeout=t+con_lifetime;
 			nxt_timeout=con_lifetime;
d22b82a0
 			if (unlikely(cmd==CONN_NEW_COMPLETE)){
 				/* check if needs to be watched for write */
 				lock_get(&tcpconn->write_lock);
 					/* if queue non empty watch it for write */
 					flags=(_wbufq_empty(tcpconn)-1)&POLLOUT;
 				lock_release(&tcpconn->write_lock);
dda6ba60
 				if (flags){
 					if (TICKS_LT(tcpconn->wbuf_q.wr_timeout, tcpconn->timeout)
 							&& TICKS_LT(t, tcpconn->wbuf_q.wr_timeout))
 						nxt_timeout=tcpconn->wbuf_q.wr_timeout-t;
 					tcpconn->flags|=F_CONN_WRITE_W|F_CONN_WANTS_WR;
 				}
 				/* activate the timer (already properly init. in 
 				   tcpconn_new())  no need for reinit */
 				local_timer_add(&tcp_main_ltimer, &tcpconn->timer, nxt_timeout,
 									t);
 				tcpconn->flags|=F_CONN_MAIN_TIMER|F_CONN_READ_W| 
 								F_CONN_WANTS_RD;
d22b82a0
 			}else{
 				/* CONN_NEW_PENDING_WRITE */
 				/* no need to check, we have something queued for write */
 				flags=POLLOUT;
dda6ba60
 				if (TICKS_LT(tcpconn->wbuf_q.wr_timeout, tcpconn->timeout)
 						&& TICKS_LT(t, tcpconn->wbuf_q.wr_timeout))
 					nxt_timeout=tcpconn->wbuf_q.wr_timeout-t;
 				/* activate the timer (already properly init. in 
 				   tcpconn_new())  no need for reinit */
 				local_timer_add(&tcp_main_ltimer, &tcpconn->timer, nxt_timeout,
 									t);
 				tcpconn->flags|=F_CONN_MAIN_TIMER|F_CONN_READ_W| 
 								F_CONN_WANTS_RD |
 								F_CONN_WRITE_W|F_CONN_WANTS_WR;
d22b82a0
 			}
 			flags|=POLLIN;
 			if (unlikely(
 					io_watch_add(&io_h, tcpconn->s, flags,
 												F_TCPCONN, tcpconn)<0)){
 				LOG(L_CRIT, "ERROR: tcp_main: handle_ser_child: failed to add"
 						" new socket to the fd list\n");
fc665070
 				tcpconn->flags&=~(F_CONN_WRITE_W|F_CONN_READ_W);
d22b82a0
 				tcpconn_try_unhash(tcpconn); /*  unhash & dec refcnt */
 				tcpconn_put_destroy(tcpconn);
 			}
 			break;
 #endif /* TCP_CONNECT_WAIT */
76cb799e
 #endif /* TCP_ASYNC */
0ba367ec
 		default:
 			LOG(L_CRIT, "BUG: handle_ser_child: unknown cmd %d\n", cmd);
 	}
 end:
 	return ret;
 error:
 	return -1;
 }
 
 
 
98f3d5e2
 /* sends a tcpconn + fd to a choosen child */
 inline static int send2child(struct tcp_connection* tcpconn)
 {
 	int i;
 	int min_busy;
 	int idx;
746f7674
 	static int crt=0; /* current child */
 	int last;
98f3d5e2
 	
 	min_busy=tcp_children[0].busy;
 	idx=0;
746f7674
 	last=crt+tcp_children_no;
 	for (; crt<last; crt++){
 		i=crt%tcp_children_no;
98f3d5e2
 		if (!tcp_children[i].busy){
 			idx=i;
 			min_busy=0;
 			break;
 		}else if (min_busy>tcp_children[i].busy){
 			min_busy=tcp_children[i].busy;
 			idx=i;
 		}
 	}
746f7674
 	crt=idx+1; /* next time we start with crt%tcp_children_no */
98f3d5e2
 	
 	tcp_children[idx].busy++;
 	tcp_children[idx].n_reqs++;
ccb7fda2
 	if (unlikely(min_busy)){
98f3d5e2
 		DBG("WARNING: send2child: no free tcp receiver, "
 				" connection passed to the least busy one (%d)\n",
 				min_busy);
 	}
 	DBG("send2child: to tcp child %d %d(%d), %p\n", idx, 
 					tcp_children[idx].proc_no,
 					tcp_children[idx].pid, tcpconn);
 	/* first make sure this child doesn't have pending request for
 	 * tcp_main (to avoid a possible deadlock: e.g. child wants to
 	 * send a release command, but the master fills its socket buffer
 	 * with new connection commands => deadlock) */
 	/* answer tcp_send requests first */
 	while(handle_ser_child(&pt[tcp_children[idx].proc_no], -1)>0);
 	/* process tcp readers requests */
 	while(handle_tcp_child(&tcp_children[idx], -1)>0);
 		
 #ifdef SEND_FD_QUEUE
28260509
 	/* if queue full, try to queue the io */
ccb7fda2
 	if (unlikely(send_fd(tcp_children[idx].unix_sock, &tcpconn,
 							sizeof(tcpconn), tcpconn->s)<=0)){
28260509
 		if ((errno==EAGAIN)||(errno==EWOULDBLOCK)){
 			/* FIXME: remove after debugging */
746f7674
 			 LOG(L_CRIT, "INFO: tcp child %d, socket %d: queue full,"
 					 	" %d requests queued (total handled %d)\n",
 					idx, tcp_children[idx].unix_sock, min_busy,
 					tcp_children[idx].n_reqs-1);
28260509
 			if (send_fd_queue_add(&send2child_q, tcp_children[idx].unix_sock, 
98f3d5e2
 						tcpconn)!=0){
28260509
 				LOG(L_ERR, "ERROR: send2child: queue send op. failed\n");
 				return -1;
 			}
 		}else{
 			LOG(L_ERR, "ERROR: send2child: send_fd failed\n");
746f7674
 			return -1;
28260509
 		}
98f3d5e2
 	}
 #else
ccb7fda2
 	if (unlikely(send_fd(tcp_children[idx].unix_sock, &tcpconn,
 						sizeof(tcpconn), tcpconn->s)<=0)){
98f3d5e2
 		LOG(L_ERR, "ERROR: send2child: send_fd failed\n");
 		return -1;
 	}
 #endif
 	
 	return 0;
 }
 
 
 
 /* handles a new connection, called internally by tcp_main_loop/handle_io.
  * params: si - pointer to one of the tcp socket_info structures on which
  *              an io event was detected (connection attempt)
  * returns:  handle_* return convention: -1 on error, 0 on EAGAIN (no more
  *           io events queued), >0 on success. success/error refer only to
  *           the accept.
  */
 static inline int handle_new_connect(struct socket_info* si)
 {
 	union sockaddr_union su;
a6357e25
 	union sockaddr_union sock_name;
 	unsigned sock_name_len;
 	union sockaddr_union* dst_su;
98f3d5e2
 	struct tcp_connection* tcpconn;
 	socklen_t su_len;
 	int new_sock;
 	
 	/* got a connection on r */
 	su_len=sizeof(su);
 	new_sock=accept(si->socket, &(su.s), &su_len);
ccb7fda2
 	if (unlikely(new_sock==-1)){
98f3d5e2
 		if ((errno==EAGAIN)||(errno==EWOULDBLOCK))
 			return 0;
 		LOG(L_ERR,  "WARNING: handle_new_connect: error while accepting"
 				" connection(%d): %s\n", errno, strerror(errno));
 		return -1;
 	}
3dc4f620
 	if (unlikely(*tcp_connections_no>=cfg_get(tcp, tcp_cfg, max_connections))){
98f3d5e2
 		LOG(L_ERR, "ERROR: maximum number of connections exceeded: %d/%d\n",
3dc4f620
 					*tcp_connections_no,
 					cfg_get(tcp, tcp_cfg, max_connections));
98f3d5e2
 		close(new_sock);
7bb2b4ca
 		TCP_STATS_LOCAL_REJECT();
98f3d5e2
 		return 1; /* success, because the accept was succesfull */
 	}
ccb7fda2
 	if (unlikely(init_sock_opt_accept(new_sock)<0)){
98f3d5e2
 		LOG(L_ERR, "ERROR: handle_new_connect: init_sock_opt failed\n");
 		close(new_sock);
 		return 1; /* success, because the accept was succesfull */
 	}
8f0e1af3
 	(*tcp_connections_no)++;
7bb2b4ca
 	TCP_STATS_ESTABLISHED(S_CONN_ACCEPT);
98f3d5e2
 	
a6357e25
 	dst_su=&si->su;
ccb7fda2
 	if (unlikely(si->flags & SI_IS_ANY)){
a6357e25
 		/* INADDR_ANY => get local dst */
 		sock_name_len=sizeof(sock_name);
 		if (getsockname(new_sock, &sock_name.s, &sock_name_len)!=0){
 			LOG(L_ERR, "ERROR: handle_new_connect:"
 						" getsockname failed: %s(%d)\n",
 						strerror(errno), errno);
 			/* go on with the 0.0.0.0 dst from the sock_info */
 		}else{
 			dst_su=&sock_name;
 		}
 	}
98f3d5e2
 	/* add socket to list */
a6357e25
 	tcpconn=tcpconn_new(new_sock, &su, dst_su, si, si->proto, S_CONN_ACCEPT);
ccb7fda2
 	if (likely(tcpconn)){
746f7674
 #ifdef TCP_PASS_NEW_CONNECTION_ON_DATA
19782e1c
 		atomic_set(&tcpconn->refcnt, 1); /* safe, not yet available to the
 											outside world */
746f7674
 		tcpconn_add(tcpconn);
ccb7fda2
 		/* activate the timer */
 		local_timer_add(&tcp_main_ltimer, &tcpconn->timer, 
3dc4f620
 								cfg_get(tcp, tcp_cfg, con_lifetime),
 								get_ticks_raw());
92c0024c
 		tcpconn->flags|=(F_CONN_MAIN_TIMER|F_CONN_READ_W|F_CONN_WANTS_RD);
a0553f4e
 		if (unlikely(io_watch_add(&io_h, tcpconn->s, POLLIN, 
 													F_TCPCONN, tcpconn)<0)){
ccb7fda2
 			LOG(L_CRIT, "ERROR: tcp_main: handle_new_connect: failed to add"
 						" new socket to the fd list\n");
fc665070
 			tcpconn->flags&=~F_CONN_READ_W;
19782e1c
 			if (tcpconn_try_unhash(tcpconn))
 				tcpconn_put_destroy(tcpconn);
ccb7fda2
 		}
746f7674
 #else
19782e1c
 		atomic_set(&tcpconn->refcnt, 2); /* safe, not yet available to the
be7401cc
 											outside world */
19782e1c
 		/* prepare it for passing to a child */
 		tcpconn->flags|=F_CONN_READER;
98f3d5e2
 		tcpconn_add(tcpconn);
edcdea00
 		DBG("handle_new_connect: new connection from %s: %p %d flags: %04x\n",
 			su2a(&su, sizeof(su)), tcpconn, tcpconn->s, tcpconn->flags);
ccb7fda2
 		if(unlikely(send2child(tcpconn)<0)){
98f3d5e2
 			LOG(L_ERR,"ERROR: handle_new_connect: no children "
 					"available\n");
885b9f62
 			tcpconn->flags&=~F_CONN_READER;
19782e1c
 			tcpconn_put(tcpconn);
 			tcpconn_try_unhash(tcpconn);
 			tcpconn_put_destroy(tcpconn);
98f3d5e2
 		}
746f7674
 #endif
98f3d5e2
 	}else{ /*tcpconn==0 */
 		LOG(L_ERR, "ERROR: handle_new_connect: tcpconn_new failed, "
 				"closing socket\n");
 		close(new_sock);
8f0e1af3
 		(*tcp_connections_no)--;
98f3d5e2
 	}
 	return 1; /* accept() was succesfull */
 }
 
 
 
 /* handles an io event on one of the watched tcp connections
  * 
  * params: tcpconn - pointer to the tcp_connection for which we have an io ev.
  *         fd_i    - index in the fd_array table (needed for delete)
  * returns:  handle_* return convention, but on success it always returns 0
885b9f62
  *           (because it's one-shot, after a succesful execution the fd is
98f3d5e2
  *            removed from tcp_main's watch fd list and passed to a child =>
  *            tcp_main is not interested in further io events that might be
  *            queued for this fd)
  */
885b9f62
 inline static int handle_tcpconn_ev(struct tcp_connection* tcpconn, short ev, 
 										int fd_i)
98f3d5e2
 {
76cb799e
 #ifdef TCP_ASYNC
885b9f62
 	int empty_q;
7498b4dc
 	int bytes;
76cb799e
 #endif /* TCP_ASYNC */
98f3d5e2
 	/*  is refcnt!=0 really necessary? 
 	 *  No, in fact it's a bug: I can have the following situation: a send only
 	 *   tcp connection used by n processes simultaneously => refcnt = n. In 
 	 *   the same time I can have a read event and this situation is perfectly
 	 *   valid. -- andrei
 	 */
 #if 0
 	if ((tcpconn->refcnt!=0)){
 		/* FIXME: might be valid for sigio_rt iff fd flags are not cleared
 		 *        (there is a short window in which it could generate a sig
 		 *         that would be catched by tcp_main) */
 		LOG(L_CRIT, "BUG: handle_tcpconn_ev: io event on referenced"
 					" tcpconn (%p), refcnt=%d, fd=%d\n",
 					tcpconn, tcpconn->refcnt, tcpconn->s);
 		return -1;
 	}
 #endif
ccb7fda2
 	/* pass it to child, so remove it from the io watch list  and the local
 	 *  timer */
76cb799e
 #ifdef TCP_ASYNC
fcc935bb
 	empty_q=0; /* warning fix */
7498b4dc
 	if (unlikely((ev & (POLLOUT|POLLERR|POLLHUP)) &&
 					(tcpconn->flags & F_CONN_WRITE_W))){
 		if (unlikely((ev & (POLLERR|POLLHUP)) || 
 					(wbufq_run(tcpconn->s, tcpconn, &empty_q)<0))){
 			if (unlikely(io_watch_del(&io_h, tcpconn->s, fd_i, 0)<0)){
 				LOG(L_ERR, "ERROR: handle_tcpconn_ev: io_watch_del(1) failed:"
 							" for %p, fd %d\n", tcpconn, tcpconn->s);
 			}
fc665070
 			if ((tcpconn->flags & F_CONN_READ_W) && (ev & POLLIN)){
7498b4dc
 				/* connection is watched for read and there is a read event
 				 * (unfortunately if we have POLLIN here we don't know if 
 				 * there's really any data in the read buffer or the POLLIN
 				 * was generated by the error or EOF => to avoid loosing
 				 *  data it's safer to either directly check the read buffer 
92c0024c
 				 *  or try a read)*/
7498b4dc
 				/* in most cases the read buffer will be empty, so in general
 				 * is cheaper to check it here and then send the 
 				 * conn.  to a a child only if needed (another syscall + at 
 				 * least 2 * syscalls in the reader + ...) */
 				if ((ioctl(tcpconn->s, FIONREAD, &bytes)>=0) && (bytes>0)){
92c0024c
 					tcpconn->flags&=~(F_CONN_WRITE_W|F_CONN_READ_W|
 										F_CONN_WANTS_RD|F_CONN_WANTS_WR);
7498b4dc
 					tcpconn->flags|=F_CONN_FORCE_EOF|F_CONN_WR_ERROR;
 					goto send_to_child;
 				}
 				/* if bytes==0 or ioctl failed, destroy the connection now */
 			}
92c0024c
 			tcpconn->flags&=~(F_CONN_WRITE_W|F_CONN_READ_W|
 								F_CONN_WANTS_RD|F_CONN_WANTS_WR);
19782e1c
 			if (unlikely(!tcpconn_try_unhash(tcpconn))){
 				LOG(L_CRIT, "BUG: tcpconn_ev: unhashed connection %p\n",
 							tcpconn);
 			}
7bb2b4ca
 			if (unlikely(ev & POLLERR)){
 				if (unlikely(tcpconn->state=S_CONN_CONNECT)){
fb751cbf
 #ifdef USE_DST_BLACKLIST
 					if (cfg_get(core, core_cfg, use_dst_blacklist))
 						dst_blacklist_su(BLST_ERR_CONNECT, tcpconn->rcv.proto,
 											&tcpconn->rcv.src_su, 0);
 #endif /* USE_DST_BLACKLIST */
7bb2b4ca
 					TCP_EV_CONNECT_ERR(0, TCP_LADDR(tcpconn),
 										TCP_LPORT(tcpconn), TCP_PSU(tcpconn),
 										TCP_PROTO(tcpconn));
 					TCP_STATS_CONNECT_FAILED();
 				}else{
fb751cbf
 #ifdef USE_DST_BLACKLIST
 					if (cfg_get(core, core_cfg, use_dst_blacklist))
 						dst_blacklist_su(BLST_ERR_SEND, tcpconn->rcv.proto,
 											&tcpconn->rcv.src_su, 0);
 #endif /* USE_DST_BLACKLIST */
7bb2b4ca
 					TCP_STATS_CON_RESET(); /* FIXME: it could != RST */
 				}
 			}
19782e1c
 			tcpconn_put_destroy(tcpconn);
885b9f62
 			goto error;
 		}
 		if (empty_q){
92c0024c
 			tcpconn->flags&=~F_CONN_WANTS_WR;
fc665070
 			if (!(tcpconn->flags & F_CONN_READ_W)){
7498b4dc
 				if (unlikely(io_watch_del(&io_h, tcpconn->s, fd_i, 0)==-1)){
 					LOG(L_ERR, "ERROR: handle_tcpconn_ev: io_watch_del(2)"
 								" failed:" " for %p, fd %d\n",
 								tcpconn, tcpconn->s);
885b9f62
 					goto error;
7498b4dc
 				}
885b9f62
 			}else{
 				if (unlikely(io_watch_chg(&io_h, tcpconn->s,
7498b4dc
 											POLLIN, fd_i)==-1)){
 					LOG(L_ERR, "ERROR: handle_tcpconn_ev: io_watch_chg(1)"
 								" failed:" " for %p, fd %d\n",
 								tcpconn, tcpconn->s);
885b9f62
 					goto error;
7498b4dc
 				}
885b9f62
 			}
8b0472d7
 			tcpconn->flags&=~F_CONN_WRITE_W;
885b9f62
 		}
503bc281
 		ev&=~POLLOUT; /* clear POLLOUT */
885b9f62
 	}
fc665070
 	if (likely(ev && (tcpconn->flags & F_CONN_READ_W))){
503bc281
 		/* if still some other IO event (POLLIN|POLLHUP|POLLERR) and
 		 * connection is still watched in tcp_main for reads, send it to a
 		 * child and stop watching it for input (but continue watching for
 		 *  writes if needed): */
885b9f62
 		if (unlikely(tcpconn->flags & F_CONN_WRITE_W)){
7498b4dc
 			if (unlikely(io_watch_chg(&io_h, tcpconn->s, POLLOUT, fd_i)==-1)){
 				LOG(L_ERR, "ERROR: handle_tcpconn_ev: io_watch_chg(2)"
 							" failed:" " for %p, fd %d\n",
 							tcpconn, tcpconn->s);
885b9f62
 				goto error;
7498b4dc
 			}
885b9f62
 		}else
 #else
 	{
76cb799e
 #endif /* TCP_ASYNC */
7498b4dc
 			if (unlikely(io_watch_del(&io_h, tcpconn->s, fd_i, 0)==-1)){
 				LOG(L_ERR, "ERROR: handle_tcpconn_ev: io_watch_del(3)"
 							" failed:" " for %p, fd %d\n",
 							tcpconn, tcpconn->s);
885b9f62
 				goto error;
7498b4dc
 			}
76cb799e
 #ifdef TCP_ASYNC
7498b4dc
 send_to_child:
 #endif
2a45890d
 		DBG("tcp: DBG: sendig to child, events %x\n", ev);
7498b4dc
 #ifdef POLLRDHUP
 		tcpconn->flags|=((int)!(ev & (POLLRDHUP|POLLHUP|POLLERR)) -1) & 
 							F_CONN_EOF_SEEN;
 #else /* POLLRDHUP */
 		tcpconn->flags|=((int)!(ev & (POLLHUP|POLLERR)) -1) & F_CONN_EOF_SEEN;
 #endif /* POLLRDHUP */
 		tcpconn->flags|= ((int)!(ev & POLLPRI) -1)  & F_CONN_OOB_DATA;
fc665070
 		tcpconn->flags|=F_CONN_READER;
885b9f62
 		local_timer_del(&tcp_main_ltimer, &tcpconn->timer);
92c0024c
 		tcpconn->flags&=~(F_CONN_MAIN_TIMER|F_CONN_READ_W|F_CONN_WANTS_RD);
885b9f62
 		tcpconn_ref(tcpconn); /* refcnt ++ */
 		if (unlikely(send2child(tcpconn)<0)){
 			LOG(L_ERR,"ERROR: handle_tcpconn_ev: no children available\n");
 			tcpconn->flags&=~F_CONN_READER;
76cb799e
 #ifdef TCP_ASYNC
885b9f62
 			if (tcpconn->flags & F_CONN_WRITE_W){
7498b4dc
 				if (unlikely(io_watch_del(&io_h, tcpconn->s, fd_i, 0)<0)){
 					LOG(L_ERR, "ERROR: handle_tcpconn_ev: io_watch_del(4)"
 							" failed:" " for %p, fd %d\n",
 							tcpconn, tcpconn->s);
 				}
885b9f62
 				tcpconn->flags&=~F_CONN_WRITE_W;
 			}
76cb799e
 #endif /* TCP_ASYNC */
19782e1c
 			tcpconn_put(tcpconn);
 			tcpconn_try_unhash(tcpconn); 
 			tcpconn_put_destroy(tcpconn); /* because of the tcpconn_ref() */
885b9f62
 		}
98f3d5e2
 	}
 	return 0; /* we are not interested in possibly queued io events, 
885b9f62
 				 the fd was either passed to a child, closed, or for writes,
 				 everything possible was already written */
98f3d5e2
 error:
 	return -1;
 }
 
 
 
0ba367ec
 /* generic handle io routine, it will call the appropiate
  *  handle_xxx() based on the fd_map type
  *
  * params:  fm  - pointer to a fd hash entry
  *          idx - index in the fd_array (or -1 if not known)
  * return: -1 on error
  *          0 on EAGAIN or when by some other way it is known that no more 
  *            io events are queued on the fd (the receive buffer is empty).
  *            Usefull to detect when there are no more io events queued for
  *            sigio_rt, epoll_et, kqueue.
  *         >0 on successfull read from the fd (when there might be more io
  *            queued -- the receive buffer might still be non-empty)
  */
885b9f62
 inline static int handle_io(struct fd_map* fm, short ev, int idx)
0ba367ec
 {	
 	int ret;
9188021a
 
 	/* update the local config */
 	cfg_update();
0ba367ec
 	
 	switch(fm->type){
 		case F_SOCKINFO:
 			ret=handle_new_connect((struct socket_info*)fm->data);
 			break;
 		case F_TCPCONN:
885b9f62
 			ret=handle_tcpconn_ev((struct tcp_connection*)fm->data, ev, idx);
0ba367ec
 			break;
 		case F_TCPCHILD:
 			ret=handle_tcp_child((struct tcp_child*)fm->data, idx);
 			break;
 		case F_PROC:
 			ret=handle_ser_child((struct process_table*)fm->data, idx);
 			break;
 		case F_NONE:
895cb2af
 			LOG(L_CRIT, "BUG: handle_io: empty fd map: %p {%d, %d, %p},"
 						" idx %d\n", fm, fm->fd, fm->type, fm->data, idx);
0ba367ec
 			goto error;
 		default:
 			LOG(L_CRIT, "BUG: handle_io: uknown fd type %d\n", fm->type); 
 			goto error;
 	}
 	return ret;
 error:
 	return -1;
 }
 
 
 
ccb7fda2
 /* timer handler for tcpconnection handled by tcp_main */
 static ticks_t tcpconn_main_timeout(ticks_t t, struct timer_ln* tl, void* data)
 {
 	struct tcp_connection *c;
 	int fd;
22db42e4
 	int tcp_async;
ccb7fda2
 	
 	c=(struct tcp_connection*)data; 
 	/* or (struct tcp...*)(tl-offset(c->timer)) */
 	
76cb799e
 #ifdef TCP_ASYNC
20863813
 	DBG( "tcp_main: entering timer for %p (ticks=%d, timeout=%d (%d s), "
 			"wr_timeout=%d (%d s)), write queue: %d bytes\n",
 			c, t, c->timeout, TICKS_TO_S(c->timeout-t),
 			c->wbuf_q.wr_timeout, TICKS_TO_S(c->wbuf_q.wr_timeout-t),
 			c->wbuf_q.queued);
 	
76cb799e
 	tcp_async=cfg_get(tcp, tcp_cfg, async);
22db42e4
 	if (likely(TICKS_LT(t, c->timeout) && ( !tcp_async | _wbufq_empty(c) |
 					TICKS_LT(t, c->wbuf_q.wr_timeout)) )){
 		if (unlikely(tcp_async && _wbufq_non_empty(c)))
20863813
 			return (ticks_t)MIN_unsigned(c->timeout-t, c->wbuf_q.wr_timeout-t);
 		else
 			return (ticks_t)(c->timeout - t);
 	}
38429f23
 	/* if time out due to write, add it to the blacklist */
7bb2b4ca
 	if (tcp_async && _wbufq_non_empty(c) && TICKS_GE(t, c->wbuf_q.wr_timeout)){
 		if (unlikely(c->state==S_CONN_CONNECT)){
 #ifdef USE_DST_BLACKLIST
 			if (cfg_get(core, core_cfg, use_dst_blacklist))
 				dst_blacklist_su(BLST_ERR_CONNECT, c->rcv.proto,
 									&c->rcv.src_su, 0);
 #endif /* USE_DST_BLACKLIST */
 			TCP_EV_CONNECT_TIMEOUT(0, TCP_LADDR(c), TCP_LPORT(c), TCP_PSU(c),
 									TCP_PROTO(c));
 			TCP_STATS_CONNECT_FAILED();
 		}else{
 #ifdef USE_DST_BLACKLIST
 			if (cfg_get(core, core_cfg, use_dst_blacklist))
 				dst_blacklist_su(BLST_ERR_SEND, c->rcv.proto,
 									&c->rcv.src_su, 0);
38429f23
 #endif /* USE_DST_BLACKLIST */
7bb2b4ca
 			TCP_EV_SEND_TIMEOUT(0, &c->rcv);
 			TCP_STATS_SEND_TIMEOUT();
 		}
 	}
76cb799e
 #else /* ! TCP_ASYNC */
ccb7fda2
 	if (TICKS_LT(t, c->timeout)){
 		/* timeout extended, exit */
 		return (ticks_t)(c->timeout - t);
 	}
76cb799e
 #endif /* TCP_ASYNC */
20863813
 	DBG("tcp_main: timeout for %p\n", c);
19782e1c
 	if (likely(c->flags & F_CONN_HASHED)){
55cdf20a
 		c->flags&=~(F_CONN_HASHED|F_CONN_MAIN_TIMER);
19782e1c
 		c->state=S_CONN_BAD;
ccb7fda2
 		TCPCONN_LOCK;
19782e1c
 			_tcpconn_detach(c);
 		TCPCONN_UNLOCK;
 	}else{
55cdf20a
 		c->flags&=~F_CONN_MAIN_TIMER;
19782e1c
 		LOG(L_CRIT, "BUG: tcp_main: timer: called with unhashed connection %p"
 				"\n", c);
 		tcpconn_ref(c); /* ugly hack to try to go on */
 	}
 	fd=c->s;
 	if (likely(fd>0)){
fc665070
 		if (likely(c->flags & (F_CONN_READ_W|F_CONN_WRITE_W))){
19782e1c
 			io_watch_del(&io_h, fd, -1, IO_FD_CLOSING);
fc665070
 			c->flags&=~(F_CONN_READ_W|F_CONN_WRITE_W);
19782e1c
 		}
ccb7fda2
 	}
7bb2b4ca
 	TCP_EV_IDLE_CONN_CLOSED(0, &c->rcv);
 	TCP_STATS_CON_TIMEOUT();
19782e1c
 	tcpconn_put_destroy(c);
 	return 0;
ccb7fda2
 }
 
 
 
 static inline void tcp_timer_run()
 {
 	ticks_t ticks;
 	
 	ticks=get_ticks_raw();
057063e6
 	if (unlikely((ticks-tcp_main_prev_ticks)<TCPCONN_TIMEOUT_MIN_RUN)) return;
 	tcp_main_prev_ticks=ticks;
ccb7fda2
 	local_timer_run(&tcp_main_ltimer, ticks);
 }
 
 
 
 /* keep in sync with tcpconn_destroy, the "delete" part should be
d8b11bbc
  * the same except for io_watch_del..
ccb7fda2
  * Note: this function is called only on shutdown by the main ser process via
  * cleanup(). However it's also safe to call it from the tcp_main process.
  * => with the ser shutdown exception, it cannot execute in parallel
d8b11bbc
  * with tcpconn_add() or tcpconn_destroy()*/
ccb7fda2
 static inline void tcpconn_destroy_all()
0ba367ec
 {
 	struct tcp_connection *c, *next;
 	unsigned h;
 	int fd;
 	
 	
ccb7fda2
 	TCPCONN_LOCK; 
0ba367ec
 	for(h=0; h<TCP_ID_HASH_SIZE; h++){
 		c=tcpconn_id_hash[h];
 		while(c){
 			next=c->id_next;
ccb7fda2
 				if (is_tcp_main){
d8b11bbc
 					/* we cannot close or remove the fd if we are not in the
 					 * tcp main proc.*/
d22b82a0
 					if ((c->flags & F_CONN_MAIN_TIMER)){
ccb7fda2
 						local_timer_del(&tcp_main_ltimer, &c->timer);
d22b82a0
 						c->flags&=~F_CONN_MAIN_TIMER;
 					} /* else still in some reader */
d8b11bbc
 					fd=c->s;
fc665070
 					if (fd>0 && (c->flags & (F_CONN_READ_W|F_CONN_WRITE_W))){
d8b11bbc
 						io_watch_del(&io_h, fd, -1, IO_FD_CLOSING);
fc665070
 						c->flags&=~(F_CONN_READ_W|F_CONN_WRITE_W);
d8b11bbc
 					}
 				}else{
 					fd=-1;
 				}
0ba367ec
 #ifdef USE_TLS
ccb7fda2
 				if (fd>0 && c->type==PROTO_TLS)
0ba367ec
 					tls_close(c, fd);
 #endif
 				_tcpconn_rm(c);
d8b11bbc
 				if (fd>0) {
5c5cd736
 #ifdef TCP_FD_CACHE
22db42e4
 					if (likely(cfg_get(tcp, tcp_cfg, fd_cache)))
 						shutdown(fd, SHUT_RDWR);
5c5cd736
 #endif /* TCP_FD_CACHE */
0ba367ec
 					close(fd);
 				}
d9515405
 				(*tcp_connections_no)--;
0ba367ec
 			c=next;
 		}
 	}
 	TCPCONN_UNLOCK;
 }
 
5b532c7f
 
0ba367ec
 
 /* tcp main loop */
 void tcp_main_loop()
 {
 
 	struct socket_info* si;
 	int r;
 	
d8b11bbc
 	is_tcp_main=1; /* mark this process as tcp main */
 	
3e999281
 	tcp_main_max_fd_no=get_max_open_fds();
746f7674
 	/* init send fd queues (here because we want mem. alloc only in the tcp
 	 *  process */
 #ifdef SEND_FD_QUEUE
 	if (init_send_fd_queues()<0){
 		LOG(L_CRIT, "ERROR: init_tcp: could not init send fd queues\n");
 		goto error;
 	}
 #endif
0ba367ec
 	/* init io_wait (here because we want the memory allocated only in
 	 * the tcp_main process) */
3e999281
 	if  (init_io_wait(&io_h, tcp_main_max_fd_no, tcp_poll_method)<0)
0ba367ec
 		goto error;
 	/* init: start watching all the fds*/
 	
5c5cd736
 	/* init local timer */
057063e6
 	tcp_main_prev_ticks=get_ticks_raw();
5c5cd736
 	if (init_local_timer(&tcp_main_ltimer, get_ticks_raw())!=0){
 		LOG(L_ERR, "ERROR: init_tcp: failed to init local timer\n");
 		goto error;
 	}
 #ifdef TCP_FD_CACHE
22db42e4
 	if (cfg_get(tcp, tcp_cfg, fd_cache)) tcp_fd_cache_init();
5c5cd736
 #endif /* TCP_FD_CACHE */
 	
d8b11bbc
 	/* add all the sockets we listen on for connections */
9f4c52ce
 	for (si=tcp_listen; si; si=si->next){
 		if ((si->proto==PROTO_TCP) &&(si->socket!=-1)){
a0553f4e
 			if (io_watch_add(&io_h, si->socket, POLLIN, F_SOCKINFO, si)<0){
0ba367ec
 				LOG(L_CRIT, "ERROR: tcp_main_loop: init: failed to add "
 							"listen socket to the fd list\n");
 				goto error;
 			}
9f4c52ce
 		}else{
 			LOG(L_CRIT, "BUG: tcp_main_loop: non tcp address in tcp_listen\n");
5b532c7f
 		}
9f4c52ce
 	}
f6e50f08
 #ifdef USE_TLS
6c53d41a
 	if (!tls_disable && tls_loaded()){
9f4c52ce
 		for (si=tls_listen; si; si=si->next){
 			if ((si->proto==PROTO_TLS) && (si->socket!=-1)){
a0553f4e
 				if (io_watch_add(&io_h, si->socket, POLLIN, F_SOCKINFO, si)<0){
0ba367ec
 					LOG(L_CRIT, "ERROR: tcp_main_loop: init: failed to add "
 							"tls listen socket to the fd list\n");
 					goto error;
 				}
9f4c52ce
 			}else{
faa66933
 				LOG(L_CRIT, "BUG: tcp_main_loop: non tls address"
 						" in tls_listen\n");
 			}
f6e50f08
 		}
1d9e67a8
 	}
9f4c52ce
 #endif
0ba367ec
 	/* add all the unix sockets used for communcation with other ser processes
 	 *  (get fd, new connection a.s.o) */
f2f969dd
 	for (r=1; r<process_no; r++){
0ba367ec
 		if (pt[r].unix_sock>0) /* we can't have 0, we never close it!*/
a0553f4e
 			if (io_watch_add(&io_h, pt[r].unix_sock, POLLIN,F_PROC, &pt[r])<0){
0ba367ec
 					LOG(L_CRIT, "ERROR: tcp_main_loop: init: failed to add "
 							"process %d unix socket to the fd list\n", r);
 					goto error;
 			}
5b532c7f
 	}
0ba367ec
 	/* add all the unix sokets used for communication with the tcp childs */
06aaa54f
 	for (r=0; r<tcp_children_no; r++){
0ba367ec
 		if (tcp_children[r].unix_sock>0)/*we can't have 0, we never close it!*/
a0553f4e
 			if (io_watch_add(&io_h, tcp_children[r].unix_sock, POLLIN,
 									F_TCPCHILD, &tcp_children[r]) <0){
0ba367ec
 				LOG(L_CRIT, "ERROR: tcp_main_loop: init: failed to add "
 						"tcp child %d unix socket to the fd list\n", r);
 				goto error;
 			}
06aaa54f
 	}
9188021a
 
 
 	/* initialize the cfg framework */
 	if (cfg_child_init()) goto error;
 
0ba367ec
 	/* main loop */
 	switch(io_h.poll_method){
 		case POLL_POLL:
 			while(1){
 				/* wait and process IO */
 				io_wait_loop_poll(&io_h, TCP_MAIN_SELECT_TIMEOUT, 0); 
98f3d5e2
 				send_fd_queue_run(&send2child_q); /* then new io */
0ba367ec
 				/* remove old connections */
ccb7fda2
 				tcp_timer_run();
0ba367ec
 			}
 			break;
 #ifdef HAVE_SELECT
 		case POLL_SELECT:
 			while(1){
 				io_wait_loop_select(&io_h, TCP_MAIN_SELECT_TIMEOUT, 0);
98f3d5e2
 				send_fd_queue_run(&send2child_q); /* then new io */
ccb7fda2
 				tcp_timer_run();
0ba367ec
 			}
 			break;
f6e50f08
 #endif
0ba367ec
 #ifdef HAVE_SIGIO_RT
 		case POLL_SIGIO_RT:
 			while(1){
 				io_wait_loop_sigio_rt(&io_h, TCP_MAIN_SELECT_TIMEOUT);
98f3d5e2
 				send_fd_queue_run(&send2child_q); /* then new io */
ccb7fda2
 				tcp_timer_run();
5b532c7f
 			}
0ba367ec
 			break;
 #endif
 #ifdef HAVE_EPOLL
 		case POLL_EPOLL_LT:
 			while(1){
 				io_wait_loop_epoll(&io_h, TCP_MAIN_SELECT_TIMEOUT, 0);
98f3d5e2
 				send_fd_queue_run(&send2child_q); /* then new io */
ccb7fda2
 				tcp_timer_run();
06aaa54f
 			}
0ba367ec
 			break;
 		case POLL_EPOLL_ET:
 			while(1){
 				io_wait_loop_epoll(&io_h, TCP_MAIN_SELECT_TIMEOUT, 1);
98f3d5e2
 				send_fd_queue_run(&send2child_q); /* then new io */
ccb7fda2
 				tcp_timer_run();
5b532c7f
 			}
0ba367ec
 			break;
 #endif
741a9937
 #ifdef HAVE_KQUEUE
 		case POLL_KQUEUE:
 			while(1){
 				io_wait_loop_kqueue(&io_h, TCP_MAIN_SELECT_TIMEOUT, 0);
98f3d5e2
 				send_fd_queue_run(&send2child_q); /* then new io */
ccb7fda2
 				tcp_timer_run();
741a9937
 			}
 			break;
 #endif
9eda5956
 #ifdef HAVE_DEVPOLL
 		case POLL_DEVPOLL:
 			while(1){
 				io_wait_loop_devpoll(&io_h, TCP_MAIN_SELECT_TIMEOUT, 0);
98f3d5e2
 				send_fd_queue_run(&send2child_q); /* then new io */
ccb7fda2
 				tcp_timer_run();
9eda5956
 			}
 			break;
 #endif
0ba367ec
 		default:
 			LOG(L_CRIT, "BUG: tcp_main_loop: no support for poll method "
 					" %s (%d)\n", 
 					poll_method_name(io_h.poll_method), io_h.poll_method);
 			goto error;
5b532c7f
 	}
0ba367ec
 error:
746f7674
 #ifdef SEND_FD_QUEUE
 	destroy_send_fd_queues();
 #endif
0ba367ec
 	destroy_io_wait(&io_h);
 	LOG(L_CRIT, "ERROR: tcp_main_loop: exiting...");
 	exit(-1);
 }
 
 
 
 /* cleanup before exit */
 void destroy_tcp()
 {
 		if (tcpconn_id_hash){
c2518f54
 			if (tcpconn_lock)
 				TCPCONN_UNLOCK; /* hack: force-unlock the tcp lock in case
 								   some process was terminated while holding 
 								   it; this will allow an almost gracious 
 								   shutdown */
ccb7fda2
 			tcpconn_destroy_all(); 
0ba367ec
 			shm_free(tcpconn_id_hash);
 			tcpconn_id_hash=0;
 		}
0dc023ec
 		DESTROY_TCP_STATS();
d9515405
 		if (tcp_connections_no){
 			shm_free(tcp_connections_no);
 			tcp_connections_no=0;
 		}
76cb799e
 #ifdef TCP_ASYNC
885b9f62
 		if (tcp_total_wq){
 			shm_free(tcp_total_wq);
 			tcp_total_wq=0;
 		}
76cb799e
 #endif /* TCP_ASYNC */
0ba367ec
 		if (connection_id){
 			shm_free(connection_id);
 			connection_id=0;
 		}
 		if (tcpconn_aliases_hash){
 			shm_free(tcpconn_aliases_hash);
 			tcpconn_aliases_hash=0;
 		}
 		if (tcpconn_lock){
 			lock_destroy(tcpconn_lock);
 			lock_dealloc((void*)tcpconn_lock);
 			tcpconn_lock=0;
 		}
746f7674
 		if (tcp_children){
 			pkg_free(tcp_children);
 			tcp_children=0;
 		}
ccb7fda2
 		destroy_local_timer(&tcp_main_ltimer);
5b532c7f
 }
 
 
 
0c5da34b
 int init_tcp()
 {
55d8155e
 	char* poll_err;
 	
20c64cc6
 	tcp_options_check();
22db42e4
 	if (tcp_cfg==0){
 		BUG("tcp_cfg not initialized\n");
 		goto error;
 	}
8aeb47e2
 	/* init lock */
 	tcpconn_lock=lock_alloc();
 	if (tcpconn_lock==0){
 		LOG(L_CRIT, "ERROR: init_tcp: could not alloc lock\n");
 		goto error;
 	}
 	if (lock_init(tcpconn_lock)==0){
 		LOG(L_CRIT, "ERROR: init_tcp: could not init lock\n");
 		lock_dealloc((void*)tcpconn_lock);
 		tcpconn_lock=0;
 		goto error;
 	}
2719e69b
 	/* init globals */
d9515405
 	tcp_connections_no=shm_malloc(sizeof(int));
 	if (tcp_connections_no==0){
 		LOG(L_CRIT, "ERROR: init_tcp: could not alloc globals\n");
 		goto error;
 	}
 	*tcp_connections_no=0;
0dc023ec
 	if (INIT_TCP_STATS()!=0) goto error;
d9515405
 	connection_id=shm_malloc(sizeof(int));
2719e69b
 	if (connection_id==0){
 		LOG(L_CRIT, "ERROR: init_tcp: could not alloc globals\n");
 		goto error;
 	}
 	*connection_id=1;
76cb799e
 #ifdef TCP_ASYNC
885b9f62
 	tcp_total_wq=shm_malloc(sizeof(*tcp_total_wq));
 	if (tcp_total_wq==0){
 		LOG(L_CRIT, "ERROR: init_tcp: could not alloc globals\n");
 		goto error;
 	}
76cb799e
 #endif /* TCP_ASYNC */
8aeb47e2
 	/* alloc hashtables*/
59653eb8
 	tcpconn_aliases_hash=(struct tcp_conn_alias**)
 			shm_malloc(TCP_ALIAS_HASH_SIZE* sizeof(struct tcp_conn_alias*));
 	if (tcpconn_aliases_hash==0){
8aeb47e2
 		LOG(L_CRIT, "ERROR: init_tcp: could not alloc address hashtable\n");
 		goto error;
 	}
 	tcpconn_id_hash=(struct tcp_connection**)shm_malloc(TCP_ID_HASH_SIZE*
 								sizeof(struct tcp_connection*));
 	if (tcpconn_id_hash==0){
 		LOG(L_CRIT, "ERROR: init_tcp: could not alloc id hashtable\n");
0c5da34b
 		goto error;
 	}
8aeb47e2
 	/* init hashtables*/
59653eb8
 	memset((void*)tcpconn_aliases_hash, 0, 
 			TCP_ALIAS_HASH_SIZE * sizeof(struct tcp_conn_alias*));
8aeb47e2
 	memset((void*)tcpconn_id_hash, 0, 
 			TCP_ID_HASH_SIZE * sizeof(struct tcp_connection*));
0ba367ec
 	
55d8155e
 	/* fix config variables */
3dc4f620
 	poll_err=check_poll_method(tcp_poll_method);
55d8155e
 	
d8b11bbc
 	/* set an appropriate poll method */
55d8155e
 	if (poll_err || (tcp_poll_method==0)){
 		tcp_poll_method=choose_poll_method();
 		if (poll_err){
 			LOG(L_ERR, "ERROR: init_tcp: %s, using %s instead\n",
 					poll_err, poll_method_name(tcp_poll_method));
 		}else{
 			LOG(L_INFO, "init_tcp: using %s as the io watch method"
 					" (auto detected)\n", poll_method_name(tcp_poll_method));
 		}
 	}else{
 			LOG(L_INFO, "init_tcp: using %s io watch method (config)\n",
 					poll_method_name(tcp_poll_method));
 	}
0ba367ec
 	
0c5da34b
 	return 0;
 error:
0ba367ec
 	/* clean-up */
 	destroy_tcp();
 	return -1;
8aeb47e2
 }
 
 
37209e14
 #ifdef TCP_CHILD_NON_BLOCKING
28260509
 /* returns -1 on error */
 static int set_non_blocking(int s)
 {
 	int flags;
 	/* non-blocking */
 	flags=fcntl(s, F_GETFL);
 	if (flags==-1){
 		LOG(L_ERR, "ERROR: set_non_blocking: fnctl failed: (%d) %s\n",
 				errno, strerror(errno));
 		goto error;
 	}
 	if (fcntl(s, F_SETFL, flags|O_NONBLOCK)==-1){
 		LOG(L_ERR, "ERROR: set_non_blocking: fcntl: set non-blocking failed:"
 				" (%d) %s\n", errno, strerror(errno));
 		goto error;
 	}
 	return 0;
 error:
 	return -1;
 }
 
37209e14
 #endif
 
 
 /*  returns -1 on error, 0 on success */
 int tcp_fix_child_sockets(int* fd)
 {
 #ifdef TCP_CHILD_NON_BLOCKING
 	if ((set_non_blocking(fd[0])<0) ||
 		(set_non_blocking(fd[1])<0)){
 		return -1;
 	}
 #endif
 	return 0;
 }
 
28260509
 
 
5b532c7f
 /* starts the tcp processes */
6ee62314
 int tcp_init_children()
5b532c7f
 {
 	int r;
37209e14
 	int reader_fd_1; /* for comm. with the tcp children read  */
5b532c7f
 	pid_t pid;
55d8155e
 	struct socket_info *si;
 	
 	/* estimate max fd. no:
 	 * 1 tcp send unix socket/all_proc, 
 	 *  + 1 udp sock/udp proc + 1 tcp_child sock/tcp child*
 	 *  + no_listen_tcp */
 	for(r=0, si=tcp_listen; si; si=si->next, r++);
 #ifdef USE_TLS
 	if (! tls_disable)
d3ab3df2
 		for (si=tls_listen; si; si=si->next, r++);
55d8155e
 #endif
5b532c7f
 	
3e999281
 	register_fds(r+tcp_max_connections+get_max_procs()-1 /* tcp main */);
 #if 0
37209e14
 	tcp_max_fd_no=get_max_procs()*2 +r-1 /* timer */ +3; /* stdin/out/err*/
 	/* max connections can be temporarily exceeded with estimated_process_count
d9515405
 	 * - tcp_main (tcpconn_connect called simultaneously in all all the 
 	 *  processes) */
37209e14
 	tcp_max_fd_no+=tcp_max_connections+get_max_procs()-1 /* tcp main */;
3e999281
 #endif
746f7674
 	/* alloc the children array */
 	tcp_children=pkg_malloc(sizeof(struct tcp_child)*tcp_children_no);
 	if (tcp_children==0){
 			LOG(L_ERR, "ERROR: tcp_init_children: out of memory\n");
 			goto error;
 	}
5b532c7f
 	/* create the tcp sock_info structures */
f2f969dd
 	/* copy the sockets --moved to main_loop*/
5b532c7f
 	
 	/* fork children & create the socket pairs*/
 	for(r=0; r<tcp_children_no; r++){
3167c744
 		child_rank++;
144c394f
 		pid=fork_tcp_process(child_rank, "tcp receiver", r, &reader_fd_1);
5b532c7f
 		if (pid<0){
 			LOG(L_ERR, "ERROR: tcp_main: fork failed: %s\n",
 					strerror(errno));
 			goto error;
 		}else if (pid>0){
 			/* parent */
 		}else{
 			/* child */
7ec958f3
 			bind_address=0; /* force a SEGFAULT if someone uses a non-init.
 							   bind address on tcp */
37209e14
 			tcp_receive_loop(reader_fd_1);
5b532c7f
 		}
 	}
6ee62314
 	return 0;
5b532c7f
 error:
 	return -1;
 }
 
d9515405
 
 
 void tcp_get_info(struct tcp_gen_info *ti)
 {
 	ti->tcp_readers=tcp_children_no;
 	ti->tcp_max_connections=tcp_max_connections;
 	ti->tcp_connections_no=*tcp_connections_no;
76cb799e
 #ifdef TCP_ASYNC
885b9f62
 	ti->tcp_write_queued=*tcp_total_wq;
 #else
 	ti->tcp_write_queued=0;
76cb799e
 #endif /* TCP_ASYNC */
d9515405
 }
 
5b532c7f
 #endif