/src/opensips/net/net_tcp.c

Source
/*
 * Copyright (C) 2014-2015 OpenSIPS Project
 * Copyright (C) 2001-2003 FhG Fokus
 *
 * This file is part of opensips, a free SIP server.
 *
 * opensips 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.
 *
 * opensips 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., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301, USA.
 *
 *
 * history:
 * ---------
 *  2015-01-xx  created (razvanc)
 */

#include <sys/types.h>
#include <sys/socket.h>
#include <netinet/in.h>
#include <netinet/in_systm.h>
#include <netinet/ip.h>
#include <netinet/tcp.h>
#include <sys/uio.h>  /* writev*/
#include <netdb.h>
#include <stdlib.h> /*exit() */
#include <time.h>   /*time() */
#include <unistd.h>
#include <errno.h>
#include <string.h>
#include <pthread.h>
#include <stdint.h>

#include "../mem/mem.h"
#include "../mem/shm_mem.h"
#include "../globals.h"
#include "../locking.h"
#include "../socket_info.h"
#include "../ut.h"
#include "../pt.h"
#include "../pt_load.h"
#include "../daemonize.h"
#include "../status_report.h"
#include "../reactor.h"
#include "../timer.h"
#include "../ipc.h"
#include "../receive.h"
#include "../lib/cond.h"
#include "../cfg_reload.h"

#include "tcp_passfd.h"
#include "net_tcp_proc.h"
#include "net_tcp_report.h"
#include "net_tcp.h"
#include "tcp_common.h"
#include "tcp_conn.h"
#include "tcp_conn_profile.h"
#include "trans.h"
#include "net_tcp_dbg.h"

struct struct_hist_list *con_hist;

enum tcp_worker_state { STATE_INACTIVE=0, STATE_ACTIVE, STATE_DRAINING};

static int tcpconn_prepare_write(struct tcp_connection *tcpconn);

/* definition of a TCP worker - the array of these TCP workers is
 * mainly intended to be used by the TCP main, to keep track of the
 * workers, about their load and so. Nevertheless, since the addition
 * of the process auto-scaling, other processes may need access to this
 * data, thus it's relocation in SHM (versus initial PKG). For example,
 * the attendant process is the one forking new TCP workers (scaling up),
 * so it must be able to set the ENABLE state for the TCP worker (and being
 * (seen by the TCP main proc). Similar, when a TCP worker shuts down, it has
 * to mark itself as DISABLED and the TCP main must see that.
 * Again, 99% this array is intended for TCP Main ops, it is not lock
 * protected, so be very careful with any ops from other procs.
 */
struct tcp_worker {
  pid_t pid;
  int pt_idx;     /*!< Index in the main Process Table */
  enum tcp_worker_state state;
};

/* definition of a TCP partition */
struct tcp_partition {
  /*! \brief connection hash table (after ip&port), includes also aliases */
  struct tcp_conn_alias** tcpconn_aliases_hash;
  /*! \brief connection hash table (after connection id) */
  struct tcp_connection** tcpconn_id_hash;
  gen_lock_t* tcpconn_lock;
};


/* array of TCP workers - to be used only by TCP MAIN */
struct tcp_worker *tcp_workers=0;
static int tcp_dispatch_sock[2] = { -1, -1 };

/* unique for each connection, used for
 * quickly finding the corresponding connection for a reply */
static unsigned int* connection_id=0;
static int *tcp_main_proc_no = 0;

/* connections are created both by TCP main (on accept) and by any other
 * process (on connect), so the counter above must be incremented under lock -
 * otherwise concurrent creations may end up sharing the same connection ID */
static gen_lock_t* connection_id_lock=0;

/* array of TCP partitions */
static struct tcp_partition tcp_parts[TCP_PARTITION_SIZE];

/*!< current number of open connections */
static unsigned int *tcp_connections_no = 0;
static gen_lock_t *tcp_connections_lock = 0;

/*!< by default don't accept aliases */
int tcp_accept_aliases=0;
int tcp_connect_timeout=DEFAULT_TCP_CONNECT_TIMEOUT;
int tcp_con_lifetime=DEFAULT_TCP_CONNECTION_LIFETIME;
int tcp_socket_backlog=DEFAULT_TCP_SOCKET_BACKLOG;
/*!< by default choose the best method */
enum poll_types tcp_poll_method=0;
int tcp_max_connections=DEFAULT_TCP_MAX_CONNECTIONS;
/* the configured/starting number of TCP workers */
int tcp_workers_no = UDP_WORKERS_NO;
/* the maximum numbers of TCP workers */
int tcp_workers_max_no;
/* the name of the auto-scaling profile (optional) */
char* tcp_auto_scaling_profile = NULL;
/* Max number of seconds that we expect a full SIP message
 * to arrive in. Anything above will close the connection. */
int tcp_max_msg_time = TCP_CHILD_MAX_MSG_TIME;
#ifdef HAVE_SO_KEEPALIVE
    int tcp_keepalive = 1;
#else
    int tcp_keepalive = 0;
#endif
int tcp_keepcount = 0;
int tcp_keepidle = 0;
int tcp_keepinterval = 0;

/*!< should we allow opening a new TCP conn when sending data 
 * over UAC branches? - branch flag to be set in the SIP messages */
int tcp_no_new_conn_bflag = 0;
/*!< should we allow opening a new TCP conn when sending data 
 * back to UAS (replies)? - msg flag to be set in the SIP messages */
int tcp_no_new_conn_rplflag = 0;
/*!< should a new TCP conn be open if needed? - variable used to used for
 * signalizing between SIP layer (branch flag) and TCP layer (tcp_send func)*/
int tcp_no_new_conn = 0;
int tcp_threads = 0;

/* if the TCP net layer is on or off (if no TCP based protos are loaded) */
static int tcp_disabled = 1;

/* is the process TCP MAIN ? */
int is_tcp_main = 0;

/* the ID of the TCP conn used for the last send operation in the
 * current process - attention, this is a really ugly HACK here */
unsigned int last_outgoing_tcp_id = 0;

static struct scaling_profile *s_profile = NULL;

/****************************** helper functions *****************************/
extern void handle_sigs(void);

static inline int init_sock_keepalive(int s, const struct tcp_conn_profile *prof)
{
  int ka;
#if defined(HAVE_TCP_KEEPINTVL) || defined(HAVE_TCP_KEEPIDLE) || defined(HAVE_TCP_KEEPCNT)
  int optval;
#endif

  if (prof->keepinterval || prof->keepidle || prof->keepcount)
    ka = 1; /* force on */
  else
    ka = prof->keepalive;

#ifdef HAVE_SO_KEEPALIVE
  if (setsockopt(s,SOL_SOCKET,SO_KEEPALIVE,&ka,sizeof(ka))<0){
    LM_WARN("setsockopt failed to enable SO_KEEPALIVE: %s\n",
      strerror(errno));
    return -1;
  }
  LM_DBG("TCP keepalive enabled on socket %d\n",s);
#endif
#ifdef HAVE_TCP_KEEPINTVL
  if ((optval = prof->keepinterval)) {
    if (setsockopt(s,IPPROTO_TCP,TCP_KEEPINTVL,&optval,sizeof(optval))<0){
      LM_WARN("setsockopt failed to set keepalive probes interval: %s\n",
        strerror(errno));
    }
  }
#endif
#ifdef HAVE_TCP_KEEPIDLE
  if ((optval = prof->keepidle)) {
    if (setsockopt(s,IPPROTO_TCP,TCP_KEEPIDLE,&optval,sizeof(optval))<0){
      LM_WARN("setsockopt failed to set keepalive idle interval: %s\n",
        strerror(errno));
    }
  }
#endif
#ifdef HAVE_TCP_KEEPCNT
  if ((optval = prof->keepcount)) {
    if (setsockopt(s,IPPROTO_TCP,TCP_KEEPCNT,&optval,sizeof(optval))<0){
      LM_WARN("setsockopt failed to set maximum keepalive count: %s\n",
        strerror(errno));
    }
  }
#endif
  return 0;
}

static inline void set_sock_reuseport(int s)
{
  int yes = 1;

  if (setsockopt(s,SOL_SOCKET,SO_REUSEPORT,&yes,sizeof(yes))<0){
    LM_WARN("setsockopt failed to set SO_REUSEPORT: %s\n",
      strerror(errno));
  }
  if (setsockopt(s,SOL_SOCKET,SO_REUSEADDR,&yes,sizeof(yes))<0){
    LM_WARN("setsockopt failed to set SO_REUSEADDR: %s\n",
      strerror(errno));
  }
}

/*! \brief Set all socket/fd options:  disable nagle, tos lowdelay,
 * non-blocking
 * \return -1 on error */
int tcp_init_sock_opt(int s, const struct tcp_conn_profile *prof, enum si_flags socketflags, int sock_tos)
{
  int flags;
  int optval;

#ifdef DISABLE_NAGLE
  flags=1;
  if (setsockopt(s, IPPROTO_TCP, TCP_NODELAY, &flags, sizeof(flags)) < 0){
    LM_WARN("could not disable Nagle: %s\n", strerror(errno));
  }
#endif
  /* tos*/
  optval = (sock_tos > 0) ? sock_tos : tos;
  if (optval > 0) {
    if (setsockopt(s, IPPROTO_IP, IP_TOS, (void*)&optval,sizeof(optval)) ==-1){
      LM_WARN("setsockopt tos: %s\n",  strerror(errno));
      /* continue since this is not critical */
    }
  }

  if (probe_max_sock_buff(s,1,MAX_SEND_BUFFER_SIZE,BUFFER_INCREMENT)) {
    LM_WARN("setsockopt tcp snd buff: %s\n", strerror(errno));
    /* continue since this is not critical */
  }

  init_sock_keepalive(s, prof);
  if (socketflags & SI_REUSEPORT)
    set_sock_reuseport(s);

  /* non-blocking */
  flags=fcntl(s, F_GETFL);
  if (flags==-1){
    LM_ERR("fcntl failed: (%d) %s\n", errno, strerror(errno));
    goto error;
  }
  if (fcntl(s, F_SETFL, flags|O_NONBLOCK)==-1){
    LM_ERR("set non-blocking failed: (%d) %s\n", errno, strerror(errno));
    goto error;
  }
  return 0;
error:
  return -1;
}

struct tcp_ipc_payload {
  struct receive_info rcv;
  struct tcp_connection *conn;
  int msg_len;
  int data_len;
  char msg_buf[0];
};

int tcp_dispatch_msg(char *msg, int len,
    struct receive_info *rcv, const void *data, int data_len)
{
  struct tcp_ipc_payload *payload;
  struct tcp_connection *conn = NULL;
  unsigned int alloc_len;
  int n;
  uintptr_t payload_ptr;

  if (len < 0) {
    LM_BUG("negative TCP message length: %d\n", len);
    return -1;
  }
  if (len && !msg) {
    LM_BUG("NULL TCP message buffer with non-zero length: %d\n", len);
    return -1;
  }
  if (data_len < 0) {
    LM_BUG("negative TCP dispatch data length: %d\n", data_len);
    return -1;
  }
  if (data_len && !data) {
    LM_BUG("missing TCP dispatch data buffer for %d bytes\n", data_len);
    return -1;
  }

  alloc_len = sizeof(*payload) + len + 1 + data_len;
  payload = shm_malloc(alloc_len);
  if (!payload) {
    LM_ERR("oom while allocating TCP IPC payload (%u bytes)\n", alloc_len);
    return -1;
  }

  memcpy(&payload->rcv, rcv, sizeof(payload->rcv));
  payload->conn = NULL;
  payload->msg_len = len;
  payload->data_len = data_len;
  memcpy(payload->msg_buf, msg, len);
  payload->msg_buf[len] = '\0';
  if (data_len)
    memcpy(payload->msg_buf + len + 1, data, data_len);

  if (rcv->proto_reserved1 &&
      tcp_conn_get(rcv->proto_reserved1, NULL, 0, PROTO_NONE,
        NULL, &conn, NULL) > 0) {
    payload->conn = conn;
  }

  payload_ptr = (uintptr_t)payload;
  n = send(tcp_dispatch_sock[1], &payload_ptr, sizeof(payload_ptr), 0);
  if (n != (int)sizeof(payload_ptr)) {
    LM_ERR("failed to dispatch TCP message to worker socket: %s\n",
      (n < 0) ? strerror(errno) : "short write");
    if (payload->conn)
      tcpconn_put(payload->conn);
    shm_free(payload);
    return -1;
  }

  return 0;
}

enum tcp_job_op {
  TCP_READ_JOB = 1,
  TCP_WRITE_JOB = 2,
  TCP_RUN_JOB = 3,
};

struct tcp_job {
  struct tcp_connection *conn;
  int op;
  tcp_thread_job_f run;
  void *data;
  long resp;
  int ret;
  struct tcp_job *next;
};

static struct tcp_pool {
  pthread_t *threads;
  int threads_no;
  int stop;
  struct tcp_job *task_head;
  struct tcp_job *task_tail;

  pthread_mutex_t done_lock;
  struct tcp_job *done_head;
  struct tcp_job *done_tail;

  int notify_pipe[2];
} tcp_pool = {
  .threads = NULL,
  .threads_no = 0,
  .stop = 0,
  .task_head = NULL,
  .task_tail = NULL,
  .done_lock = PTHREAD_MUTEX_INITIALIZER,
  .done_head = NULL,
  .done_tail = NULL,
  .notify_pipe = {-1, -1},
};

struct tcp_shared_write_queue {
  gen_cond_t cond;
  struct tcp_connection *head;
  struct tcp_connection *tail;
};

static struct tcp_shared_write_queue *tcp_write_queue = NULL;

static inline int tcp_threads_active(void)
{
  return tcp_pool.threads_no > 0;
}

int tcp_write_in_main(void)
{
  /* This is a process-independent policy: TCP writes are handled by the
   * dedicated TCP main process, regardless of the caller process. */
  return !tcp_disabled;
}



/********************** TCP conn management functions ************************/

/* initializes an already defined TCP listener */
int tcp_init_listener(struct socket_info *si)
{
  union sockaddr_union* addr = &si->su;

  if (init_su(addr, &si->address, si->port_no)<0){
    LM_ERR("could no init sockaddr_union\n");
    return -1;
  }

  return 0;
}

/* binding an defined TCP listener */
int tcp_bind_listener(struct socket_info *si)
{
  union sockaddr_union* addr;
  int optval;
#ifdef DISABLE_NAGLE
  int flag;
#endif

  addr = &si->su;
  si->socket = socket(AF2PF(addr->s.sa_family), SOCK_STREAM, 0);
  if (si->socket==-1){
    LM_ERR("socket failed with [%s]\n", strerror(errno));
    goto error;
  }
#ifdef DISABLE_NAGLE
  flag=1;
  if (setsockopt(si->socket, IPPROTO_TCP, TCP_NODELAY, &flag, sizeof(flag)) < 0){
    LM_ERR("could not disable Nagle: %s\n",strerror(errno));
  }
#endif

#if  !defined(TCP_DONT_REUSEADDR)
  /* 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
   */
  optval=1;
  if (setsockopt(si->socket, SOL_SOCKET, SO_REUSEADDR,
  (void*)&optval, sizeof(optval))==-1) {
    LM_ERR("setsockopt failed with [%s]\n", strerror(errno));
    goto error;
  }
#endif
  /* tos */
  optval = (si->tos > 0) ? si->tos : tos;
  if (optval > 0) {
    if (setsockopt(si->socket, IPPROTO_IP, IP_TOS, (void*)&optval,
    sizeof(optval)) ==-1){
      LM_WARN("setsockopt tos: %s\n", strerror(errno));
      /* continue since this is not critical */
    }
  }

  if (probe_max_sock_buff(si->socket,1,MAX_SEND_BUFFER_SIZE,
  BUFFER_INCREMENT)) {
    LM_WARN("setsockopt tcp snd buff: %s\n",strerror(errno));
    /* continue since this is not critical */
  }

  init_sock_keepalive(si->socket, &tcp_con_df_profile);
  if (si->flags & SI_REUSEPORT)
    set_sock_reuseport(si->socket);
  if (bind(si->socket, &addr->s, sockaddru_len(*addr))==-1){
    LM_ERR("bind(%x, %p, %d) on %s:%d : %s\n",
        si->socket, &addr->s,
        (unsigned)sockaddru_len(*addr),
        si->address_str.s,
        si->port_no,
        strerror(errno));
    goto error;
  }
  if (listen(si->socket, tcp_socket_backlog)==-1){
    LM_ERR("listen(%x, %p, %d) on %s: %s\n",
        si->socket, &addr->s,
        (unsigned)sockaddru_len(*addr),
        si->address_str.s,
        strerror(errno));
    goto error;
  }

  return 0;
error:
  if (si->socket!=-1){
    close(si->socket);
    si->socket=-1;
  }
  return -1;
}


/*! \brief finds a connection, if id=0 return NULL
 * \note WARNING: unprotected (locks) use tcpconn_get unless you really
 * know what you are doing */
static struct tcp_connection* _tcpconn_find(unsigned int id)
{
  struct tcp_connection *c;
  unsigned hash;

  if (id){
    hash=tcp_id_hash(id);
    for (c=TCP_PART(id).tcpconn_id_hash[hash]; c; c=c->id_next){
#ifdef EXTRA_DEBUG
      LM_DBG("c=%p, c->id=%u, port=%d\n",c, c->id, c->rcv.src_port);
      print_ip("ip=", &c->rcv.src_ip, "\n");
#endif
      if ((id==c->id) && c->state!=S_CONN_BAD &&
          !(c->flags & F_CONN_FORCE_CLOSED))
        return c;
    }
  }
  return 0;
}


/* returns the correlation ID of a TCP connection */
int tcp_get_correlation_id( unsigned int id, unsigned long long *cid)
{
  struct tcp_connection* c;

  TCPCONN_LOCK(id);
  if ( (c=_tcpconn_find(id))!=NULL ) {
    *cid = c->cid;
    TCPCONN_UNLOCK(id);
    return 0;
  }
  *cid = 0;
  TCPCONN_UNLOCK(id);
  return -1;
}

/* returns the correlation ID of a TCP connection */
int tcp_get_rcv( unsigned int id, struct receive_info *ri)
{
  struct tcp_connection* c;

  TCPCONN_LOCK(id);
  if ( (c=_tcpconn_find(id))!=NULL ) {
    memcpy(ri, &c->rcv, sizeof *ri);
    TCPCONN_UNLOCK(id);
    return 0;
  }
  TCPCONN_UNLOCK(id);
  return -1;
}

int tcp_get_main_proc_no(void)
{
  return tcp_main_proc_no ? *tcp_main_proc_no : -1;
}


/*! \brief _tcpconn_find with locks and acquire a shared connection reference */
int tcp_conn_get(unsigned int id, struct ip_addr* ip, int port,
    enum sip_protos proto, void *proto_extra_id,
    struct tcp_connection** conn, const struct socket_info* send_sock)
{
  struct tcp_connection* c;
  struct tcp_conn_alias* a;
  unsigned hash;
  unsigned int part;

  if (id) {
    part = id;
    TCPCONN_LOCK(part);
    if ( (c=_tcpconn_find(part))!=NULL )
      goto found;
    TCPCONN_UNLOCK(part);
  }

  /* continue search based on IP address + port + transport */
#ifdef EXTRA_DEBUG
  LM_DBG("%d  port %u\n",id, port);
  if (ip) print_ip("tcpconn_find: ip ", ip, "\n");
#endif
  if (ip){
    hash=tcp_addr_hash(ip, port);
    for( part=0 ; part<TCP_PARTITION_SIZE ; part++ ) {
      TCPCONN_LOCK(part);
      for (a=TCP_PART(part).tcpconn_aliases_hash[hash]; a; a=a->next) {
#ifdef EXTRA_DEBUG
        LM_DBG("a=%p, c=%p, c->id=%u, 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");
        if (send_sock && a->parent->rcv.bind_address) {
          print_ip("requested send_sock ip=", &send_sock->address,"\n");
          print_ip("found send_sock ip=", &a->parent->rcv.bind_address->address,"\n");
        }
#endif
        c = a->parent;
        if (c->state != S_CONN_BAD &&
            !(c->flags & F_CONN_FORCE_CLOSED) &&
            ((c->flags & F_CONN_INIT) ||
             (c->state == S_CONN_CONNECTING && c->fd == -1)) &&
            (send_sock==NULL || send_sock == a->parent->rcv.bind_address) &&
            port == a->port &&
            proto == c->type &&
            ip_addr_cmp(ip, &c->rcv.src_ip) &&
            (proto_extra_id == NULL ||
             ((c->flags & F_CONN_INIT) &&
              (protos[proto].net.stream.conn.match == NULL ||
               protos[proto].net.stream.conn.match(c, proto_extra_id)))) )
            goto found;
        }
      TCPCONN_UNLOCK(part);
    }
  }

  /* not found */
  *conn = NULL;
  return 0;

found:
  c->refcnt++;
  TCPCONN_UNLOCK(part);
  sh_log(c->hist, TCP_REF, "tcp_conn_get, (%d)", c->refcnt);

  LM_DBG("con found in state %d\n",c->state);

  *conn = c;
  return 1;
}


/* used to tune the tcp_connection attributes - not to be used inside the
   network layer, but onlu from the above layer (otherwise we may end up
   in strange deadlocks!) */
int tcp_conn_fcntl(struct receive_info *rcv, int attr, void *value)
{
  struct tcp_connection *con;

  switch (attr) {
  case DST_FCNTL_SET_LIFETIME:
    /* set connection timeout */
    TCPCONN_LOCK(rcv->proto_reserved1);
    con =_tcpconn_find(rcv->proto_reserved1);
    if (!con) {
      LM_ERR("Strange, tcp conn not found (id=%u)\n",
        rcv->proto_reserved1);
    } else {
      tcp_conn_set_lifetime( con, (int)(long)(value));
    }
    TCPCONN_UNLOCK(rcv->proto_reserved1);
    return 0;
  default:
    LM_ERR("unsupported operation %d on conn\n",attr);
    return -1;
  }
  return -1;
}


static struct tcp_connection* tcpconn_add(struct tcp_connection *c)
{
  unsigned hash;

  if (c){
    TCPCONN_LOCK(c->id);
    /* add it at the beginning of the list*/
    hash=tcp_id_hash(c->id);
    c->id_hash=hash;
    tcpconn_listadd(TCP_PART(c->id).tcpconn_id_hash[hash], c, id_next,
      id_prev);

    hash=tcp_addr_hash(&c->rcv.src_ip, c->rcv.src_port);
    /* set the first alias */
    c->con_aliases[0].port=c->rcv.src_port;
    c->con_aliases[0].hash=hash;
    c->con_aliases[0].parent=c;
    tcpconn_listadd(TCP_PART(c->id).tcpconn_aliases_hash[hash],
      &c->con_aliases[0], next, prev);
    c->aliases++;
    c->flags |= F_CONN_HASHED;
    TCPCONN_UNLOCK(c->id);
    LM_DBG("hashes: %d, %d\n", hash, c->id_hash);
    return c;
  }else{
    LM_CRIT("null connection pointer\n");
    return 0;
  }
}

static str e_tcp_src_ip = str_init("src_ip");
static str e_tcp_src_port = str_init("src_port");
static str e_tcp_dst_ip = str_init("dst_ip");
static str e_tcp_dst_port = str_init("dst_port");
static str e_tcp_c_proto = str_init("proto");

static void tcp_disconnect_event_raise(struct tcp_connection* c)
{
  evi_params_p list = 0;
  str src_ip,dst_ip, proto;
  int src_port,dst_port;
  char src_ip_buf[IP_ADDR_MAX_STR_SIZE],dst_ip_buf[IP_ADDR_MAX_STR_SIZE];

  // event has to be triggered - check for subscribers
  if (!evi_probe_event(EVI_TCP_DISCONNECT)) {
    goto end;
  }

  if (!(list = evi_get_params()))
    goto end;

  src_ip.s = ip_addr2a( &c->rcv.src_ip );
  memcpy(src_ip_buf,src_ip.s,IP_ADDR_MAX_STR_SIZE);
  src_ip.s = src_ip_buf;
  src_ip.len = strlen(src_ip.s);

  if (evi_param_add_str(list, &e_tcp_src_ip, &src_ip)) {
    LM_ERR("unable to add parameter\n");
    goto end;
  }

  src_port = c->rcv.src_port;

  if (evi_param_add_int(list, &e_tcp_src_port, &src_port)) {
    LM_ERR("unable to add parameter\n");
    goto end;
  }

  dst_ip.s = ip_addr2a( &c->rcv.dst_ip );
  memcpy(dst_ip_buf,dst_ip.s,IP_ADDR_MAX_STR_SIZE);
  dst_ip.s = dst_ip_buf;
  dst_ip.len = strlen(dst_ip.s);

  if (evi_param_add_str(list, &e_tcp_dst_ip, &dst_ip)) {
    LM_ERR("unable to add parameter\n");
    goto end;
  }

  dst_port = c->rcv.dst_port;

  if (evi_param_add_int(list, &e_tcp_dst_port, &dst_port)) {
    LM_ERR("unable to add parameter\n");
    goto end;
  }

  proto.s = protos[c->rcv.proto].name;
  proto.len = strlen(proto.s);

  if (evi_param_add_str(list, &e_tcp_c_proto, &proto)) {
    LM_ERR("unable to add parameter\n");
    goto end;
  }

  if (is_tcp_main) {
    if (evi_dispatch_event(EVI_TCP_DISCONNECT, list)) {
      LM_ERR("unable to dispatch tcp disconnect event\n");
    }
  } else {
    if (evi_raise_event(EVI_TCP_DISCONNECT, list)) {
      LM_ERR("unable to send tcp disconnect event\n");
    }
  }
  list = 0;

end:
  if (list)
    evi_free_params(list);
}

/* convenience macro to aid in shm_free() debugging */
#define _tcpconn_rm(c, ne) \
  do {\
    __tcpconn_rm(c, ne);\
    shm_free(c);\
  } while (0)

struct tcp_req *tcp_conn_get_req(struct tcp_connection *c)
{
  if (!c)
    return NULL;

  if (c->con_req)
    return c->con_req;

  c->con_req = thread_malloc(sizeof(*c->con_req));
  if (!c->con_req) {
    LM_ERR("failed to allocate TCP request buffer for connection %u\n",
      c->id);
    return NULL;
  }
  memset(c->con_req, 0, sizeof(*c->con_req));

  return c->con_req;
}

void tcp_conn_destroy_req(struct tcp_connection *c)
{
  if (!c || !c->con_req)
    return;

  thread_free(c->con_req);
  c->con_req = NULL;
}

/*! \brief unsafe tcpconn_rm version (nolocks) */
static void __tcpconn_rm(struct tcp_connection* c, int no_event)
{
  int r;

  if (c->flags & F_CONN_HASHED) {
    tcpconn_listrm(TCP_PART(c->id).tcpconn_id_hash[c->id_hash], c,
      id_next, id_prev);
    /* remove all the aliases */
    for (r=0; r<c->aliases; r++)
      tcpconn_listrm(TCP_PART(c->id).tcpconn_aliases_hash[
        c->con_aliases[r].hash], &c->con_aliases[r], next, prev);
    c->flags &= ~F_CONN_HASHED;
  }
  lock_destroy(&c->write_lock);

  if (c->async) {
    for (r = 0; r<c->async->pending; r++)
      shm_free(c->async->chunks[r]);
    shm_free(c->async);
    c->async = NULL;
  }

  lock_get(tcp_connections_lock);
  (*tcp_connections_no)--;
  lock_release(tcp_connections_lock);

  /* Only TCP main has valid process-private connection state. */
  if (is_tcp_main) {
    if (c->proto_req)
      thread_free(c->proto_req);
    c->proto_req = NULL;
    tcp_conn_destroy_req(c);

    if (protos[c->type].net.stream.conn.clean)
      protos[c->type].net.stream.conn.clean(c);
  }

  if (!no_event) tcp_disconnect_event_raise(c);

#ifdef DBG_TCPCON
  sh_log(c->hist, TCP_DESTROY, "type=%d", c->type);
  sh_unref(c->hist);
  c->hist = NULL;
#endif

  /* shm_free(c); -- freed by _tcpconn_rm() */
}

/*! \brief add port as an alias for the "id" connection
 * \return 0 on success,-1 on failure */
int tcpconn_add_alias(struct sip_msg *msg, unsigned int id, int port, int proto)
{
  struct tcp_connection* c;
  unsigned hash;
  struct tcp_conn_alias* a;

  a=0;
  /* fix the port */
  port=port ? port : protos[proto].default_port ;
  TCPCONN_LOCK(id);
  /* check if alias already exists */
  c=_tcpconn_find(id);
  if (c) {
    if (msg && !(c->profile.alias_mode == TCP_ALIAS_ALWAYS
                   || (c->profile.alias_mode == TCP_ALIAS_RFC_5923
                       && msg->via1->alias))) {
      LM_DBG("refusing to add alias (alias_mode: %u, via 'alias': %u)\n",
              c->profile.alias_mode, !!msg->via1->alias);
      TCPCONN_UNLOCK(id);
      return 0;
    }

    hash=tcp_addr_hash(&c->rcv.src_ip, port);
    /* search the aliases for an already existing one */
    for (a=TCP_PART(id).tcpconn_aliases_hash[hash]; a; a=a->next) {
      if (a->parent->state != S_CONN_BAD &&
          port == a->port &&
          proto == a->parent->type &&
          ip_addr_cmp(&c->rcv.src_ip, &a->parent->rcv.src_ip)) {
        /* found */
        if (a->parent!=c) goto error_sec;
        else goto ok;
      }
    }
    if (c->aliases>=TCP_CON_MAX_ALIASES) goto error_aliases;
    c->con_aliases[c->aliases].parent=c;
    c->con_aliases[c->aliases].port=port;
    c->con_aliases[c->aliases].hash=hash;
    tcpconn_listadd(TCP_PART(id).tcpconn_aliases_hash[hash],
                &c->con_aliases[c->aliases], next, prev);
    c->aliases++;
  }else goto error_not_found;
ok:
  TCPCONN_UNLOCK(id);
#ifdef EXTRA_DEBUG
  if (a) LM_DBG("alias already present\n");
  else   LM_DBG("alias port %d for hash %d, id %u\n", port, hash, id);
#endif
  return 0;
error_aliases:
  TCPCONN_UNLOCK(id);
  LM_ERR("too many aliases for connection %p (%u)\n", c, id);
  return -1;
error_not_found:
  TCPCONN_UNLOCK(id);
  LM_ERR("no connection found for id %u\n",id);
  return -1;
error_sec:
  LM_WARN("possible port hijack attempt\n");
  LM_WARN("alias already present and points to another connection "
      "(%d : %d and %u : %d)\n", a->parent->id,  port, id, port);
  TCPCONN_UNLOCK(id);
  return -1;
}


static void tcpconn_put_rpc(int pid, void *param)
{
  tcpconn_put(param);
}


void tcpconn_put(struct tcp_connection* c)
{
  int destroy = 0;
  int release_in_main = 0;
  int tcp_main_proc;

  TCPCONN_LOCK(c->id);
  if ((c->flags & F_CONN_HASHED) == 0) {
    if (!is_tcp_main && c->refcnt == 1) {
      /* Keep the last reference until TCP main accepts ownership. */
      release_in_main = 1;
    } else {
      c->refcnt--;
      if (c->refcnt == 0)
        destroy = 1;
    }
  } else {
    /* Hashed connections are destroyed by TCP main lifetime handling. */
    c->refcnt--;
  }
  TCPCONN_UNLOCK(c->id);

  if (release_in_main) {
    tcp_main_proc = tcp_get_main_proc_no();
    if (tcp_main_proc < 0 ||
        ipc_send_rpc(tcp_main_proc, tcpconn_put_rpc, c) < 0)
      LM_ERR("failed to release connection %p (%u) in TCP main; "
        "leaving its final reference intact\n", c, c->id);
  } else if (destroy) {
    _tcpconn_rm(c, 1);
  }
}


static inline void tcpconn_ref(struct tcp_connection* c)
{
  TCPCONN_LOCK(c->id);
  c->refcnt++;
  TCPCONN_UNLOCK(c->id);
}


static inline unsigned int tcpconn_next_id(void)
{
  unsigned int id;

  lock_get(connection_id_lock);
  id = (*connection_id)++;
  lock_release(connection_id_lock);

  return id;
}


static struct tcp_connection* tcpconn_new(int sock, const union sockaddr_union* su,
                    const struct socket_info* si, const struct tcp_conn_profile *prof,
                    int state, int flags)
{
  struct tcp_connection *c;
  union sockaddr_union local_su;
  unsigned int su_size;
  int counted = 0;

  lock_get(tcp_connections_lock);
  if (*tcp_connections_no >= (unsigned int)tcp_max_connections) {
    lock_release(tcp_connections_lock);
    LM_ERR("maximum number of connections exceeded: %u/%d\n",
      *tcp_connections_no, tcp_max_connections);
    return 0;
  }
  (*tcp_connections_no)++;
  lock_release(tcp_connections_lock);
  counted = 1;

  c=(struct tcp_connection*)shm_malloc(sizeof(struct tcp_connection));
  if (c==0){
    LM_ERR("shared memory allocation failure\n");
    goto error_count;
  }
  memset(c, 0, sizeof(struct tcp_connection)); /* zero init */
  c->fd=sock;
  if (lock_init(&c->write_lock)==0){
    LM_ERR("init lock failed\n");
    goto error0;
  }

  c->rcv.src_su=*su;

  c->refcnt=0;
  su2ip_addr(&c->rcv.src_ip, su);
  c->rcv.src_port=su_getport(su);
  c->rcv.bind_address = si;
  c->rcv.dst_ip = si->address;
  if (sock >= 0) {
    su_size = sockaddru_len(*su);
    if (getsockname(sock, (struct sockaddr *)&local_su, &su_size)<0) {
      LM_ERR("failed to get info on received interface/IP %d/%s\n",
        errno, strerror(errno));
      goto error;
    }
    c->rcv.dst_port = su_getport(&local_su);
  } else {
    c->rcv.dst_port = (si->flags & SI_REUSEPORT) ? su_getport(&si->su) : 0;
  }
  print_ip("tcpconn_new: new tcp connection to: ", &c->rcv.src_ip, "\n");
  LM_DBG("on port %d, proto %d\n", c->rcv.src_port, si->proto);
  c->id=tcpconn_next_id();
  c->cid = (unsigned long long)c->id
        | ( (unsigned long long)(startup_time&0xFFFFFF) << 32 )
          | ( (unsigned long long)(rand()&0xFF) << 56 );

  c->rcv.proto_reserved1=0; /* this will be filled before receive_message*/
  c->rcv.proto_reserved2=0;
  c->state=state;
  c->extra_data=0;
  c->type = si->proto;
  c->rcv.proto = si->proto;
  /* start with the default conn lifetime */
  c->lifetime = get_ticks() + prof->con_lifetime;
  c->timeout = c->lifetime;
  c->profile = *prof;
  c->flags|=F_CONN_REMOVED|flags;
#ifdef DBG_TCPCON
  c->hist = sh_push(c, con_hist);
#endif

  if (protos[si->proto].net.stream.async_chunks) {
    c->async = shm_malloc(sizeof(struct tcp_async_data) +
        protos[si->proto].net.stream.async_chunks *
        sizeof(struct tcp_async_chunk));
    if (c->async) {
      c->async->allocated = protos[si->proto].net.stream.async_chunks;
      c->async->oldest = 0;
      c->async->pending = 0;
    } else {
      LM_ERR("could not allocate async data for con!\n");
      goto error;
    }
  }
  if (sock >= 0) {
    if (protos[si->proto].net.stream.conn.init &&
        protos[si->proto].net.stream.conn.init(c) < 0) {
      LM_ERR("failed to do proto %d specific init for conn %p\n",
          c->type, c);
      goto error;
    }
    c->flags |= F_CONN_INIT;
  }
  return c;

error:
  lock_destroy(&c->write_lock);
error0:
  shm_free(c);
error_count:
  if (counted) {
    lock_get(tcp_connections_lock);
    (*tcp_connections_no)--;
    lock_release(tcp_connections_lock);
  }
  return 0;
}


/* creates a new tcp connection structure
 * for an outgoing connection request; local private state is initialized later
 * a +1 ref is set for the new conn !
 * IMPORTANT - the function assumes you want to create a new TCP conn as
 * a result of a connect operation - the conn will be set as connect !!
 * Accepted connection are triggered internally only */
struct tcp_connection* tcp_conn_create(const union sockaddr_union* su,
    const struct socket_info* si, struct tcp_conn_profile *prof,
    int state)
{
  struct tcp_connection *c;
  struct tcp_conn_profile default_prof;

  if (!prof) {
    tcp_con_get_profile(su, &si->su, si->proto, &default_prof);
    prof = &default_prof;
  }

  /* create the connection structure */
  c = tcpconn_new(-1, su, si, prof, state, 0);
  if (c==NULL) {
    LM_ERR("tcpconn_new failed\n");
    return NULL;
  }

  c->refcnt++; /* safe to do it w/o locking, it's not yet
          available to the rest of the world */
  sh_log(c->hist, TCP_REF, "connect, (%d)", c->refcnt);
  return c;
}


static inline void tcpconn_destroy(struct tcp_connection* tcpconn)
{
  int fd;
  int unsigned id = tcpconn->id;
  int hashed;

  TCPCONN_LOCK(id); /*avoid races w/ tcp_send*/
  tcpconn->refcnt--;
  if (tcpconn->refcnt==0){
    LM_DBG("destroying connection %p, flags %04x\n",
        tcpconn, tcpconn->flags);
    fd=tcpconn->fd;
    /* no reporting here - the tcpconn_destroy() function is called
     * from the TCP_MAIN reactor when handling connectioned received
     * from a worker; and we generate the CLOSE reports from WORKERs */
    hashed = (tcpconn->flags & F_CONN_HASHED);
    _tcpconn_rm(tcpconn, hashed ? 0 : 1);
    if (fd >= 0)
      close(fd);
  }else{
    /* force timeout */
    tcpconn->lifetime=0;
    tcpconn->timeout=0;
    tcpconn->state=S_CONN_BAD;
    sh_log(tcpconn->hist, TCP_DEL_DELAY, "tcpconn_destroy delayed, (%d)",
      tcpconn->refcnt);
    LM_DBG("delaying (%p, flags %04x) ref = %d ...\n",
        tcpconn, tcpconn->flags, tcpconn->refcnt);

  }
  TCPCONN_UNLOCK(id);
}

static void tcpconn_destroy_rpc(int _, void *param)
{
  tcpconn_destroy(param);
}

/* wrapper to the internally used function */
void tcp_conn_destroy(struct tcp_connection* tcpconn)
{
  int tcp_main_proc;

  tcp_trigger_report(tcpconn, TCP_REPORT_CLOSE,
        "Closed by Proto layer");
  sh_log(tcpconn->hist, TCP_UNREF, "tcp_conn_destroy, (%d)", tcpconn->refcnt);

  if (!is_tcp_main) {
    tcp_main_proc = tcp_get_main_proc_no();
    if (tcp_main_proc < 0 ||
        ipc_send_rpc(tcp_main_proc, tcpconn_destroy_rpc, tcpconn) < 0)
      LM_ERR("failed to destroy connection %p (%u) in TCP main; "
        "leaving its reference intact\n", tcpconn, tcpconn->id);
    return;
  }

  tcpconn_destroy(tcpconn);
}

static inline int tcp_set_nonblock(int fd)
{
  int flags;

  flags = fcntl(fd, F_GETFL);
  if (flags == -1) {
    LM_ERR("fcntl(F_GETFL) failed for %d: %s\n", fd, strerror(errno));
    return -1;
  }

  if (fcntl(fd, F_SETFL, flags | O_NONBLOCK) == -1) {
    LM_ERR("fcntl(F_SETFL) failed for %d: %s\n", fd, strerror(errno));
    return -1;
  }

  return 0;
}

static void tcp_push_done_job(struct tcp_job *job)
{
  pthread_mutex_lock(&tcp_pool.done_lock);
  if (tcp_pool.done_tail)
    tcp_pool.done_tail->next = job;
  else
    tcp_pool.done_head = job;
  tcp_pool.done_tail = job;
  pthread_mutex_unlock(&tcp_pool.done_lock);
}

static struct tcp_job *tcp_pop_done_job(void)
{
  struct tcp_job *job;

  pthread_mutex_lock(&tcp_pool.done_lock);
  job = tcp_pool.done_head;
  if (job) {
    tcp_pool.done_head = job->next;
    if (tcp_pool.done_head == NULL)
      tcp_pool.done_tail = NULL;
  }
  pthread_mutex_unlock(&tcp_pool.done_lock);

  return job;
}

static inline struct tcp_connection *tcp_pop_shared_write_conn_locked(void)
{
  struct tcp_connection *conn;

  conn = tcp_write_queue->head;
  if (conn) {
    tcp_write_queue->head = conn->wq_next;
    if (tcp_write_queue->head == NULL)
      tcp_write_queue->tail = NULL;
    conn->wq_next = NULL;
  }

  return conn;
}

static void *tcp_thread_routine(void *arg)
{
  struct tcp_job *job;
  struct tcp_connection *conn;
  char wake = 'x';
  int rc;

  (void)arg;

  /* Reactor operations stay in TCP main; IO threads only run read/write
   * callbacks and notify completion back to the main thread. */
  while (1) {
    cond_lock(&tcp_write_queue->cond);
    while (!tcp_pool.stop && tcp_pool.task_head == NULL &&
        tcp_write_queue->head == NULL)
      cond_wait(&tcp_write_queue->cond);

    if (tcp_pool.stop && tcp_pool.task_head == NULL &&
        tcp_write_queue->head == NULL) {
      cond_unlock(&tcp_write_queue->cond);
      break;
    }

    job = tcp_pool.task_head;
    if (job) {
      tcp_pool.task_head = job->next;
      if (tcp_pool.task_head == NULL)
        tcp_pool.task_tail = NULL;
    } else if ((conn = tcp_pop_shared_write_conn_locked()) != NULL) {
      job = thread_malloc(sizeof(*job));
      if (!job) {
        LM_ERR("oom while building shared TCP write job\n");
        conn->flags &= ~F_CONN_WRITE_QUEUED;
        cond_unlock(&tcp_write_queue->cond);
        tcpconn_put(conn);
        continue;
      }
      job->conn = conn;
      job->op = TCP_WRITE_JOB;
      job->run = NULL;
      job->data = NULL;
      job->resp = 0;
      job->ret = 0;
      job->next = NULL;
    }
    cond_unlock(&tcp_write_queue->cond);

    conn = job->conn;
    if (job->op == TCP_READ_JOB) {
      if (conn->msg_attempts && get_ticks() > conn->timeout) {
        job->ret = -1;
        job->resp = -2;
      } else if (protos[conn->type].net.stream.read) {
        job->resp = protos[conn->type].net.stream.read(conn, &job->ret);
      } else {
        LM_ERR("missing stream.read callback for proto %d\n", conn->type);
        job->ret = -1;
        job->resp = -1;
      }
    } else if (job->op == TCP_WRITE_JOB) {
      if (protos[conn->type].net.stream.write) {
        if (tcpconn_prepare_write(conn) < 0) {
          job->ret = -1;
          job->resp = -1;
          goto done_job;
        }
        lock_get(&conn->write_lock);
        job->resp = protos[conn->type].net.stream.write(conn, conn->fd);
        lock_release(&conn->write_lock);
        job->ret = (job->resp < 0) ? -1 : 0;
      } else {
        LM_ERR("missing stream.write callback for proto %d\n", conn->type);
        job->ret = -1;
        job->resp = -1;
      }
    } else if (job->op == TCP_RUN_JOB) {
      job->ret = job->run ? job->run(job->data) : -1;
      thread_free(job);
      continue;
    } else {
      LM_ERR("unknown TCP job op %d\n", job->op);
      job->ret = -1;
      job->resp = -1;
    }

done_job:
    job->next = NULL;
    tcp_push_done_job(job);

    rc = write(tcp_pool.notify_pipe[1], &wake, 1);
    if (rc < 0 && errno != EAGAIN && errno != EWOULDBLOCK)
      LM_ERR("failed to notify TCP IO completion: %s\n", strerror(errno));
  }

  return NULL;
}

static int tcp_pool_init(void)
{
  int i;
  int started = 0;
  int threads_no;
  long cpu_no;

  if (tcp_threads > 0)
    threads_no = tcp_threads;
  else {
    cpu_no = sysconf(_SC_NPROCESSORS_ONLN);
    if (cpu_no > 0)
      threads_no = (int)cpu_no;
    else if (tcp_workers_no > 0)
      threads_no = tcp_workers_no;
    else
      threads_no = 1;
  }

  if (pipe(tcp_pool.notify_pipe) < 0) {
    LM_ERR("failed to create TCP IO notification pipe: %s\n", strerror(errno));
    goto error;
  }

  if (tcp_set_nonblock(tcp_pool.notify_pipe[0]) < 0 ||
      tcp_set_nonblock(tcp_pool.notify_pipe[1]) < 0)
    goto error;

  if (reactor_add_reader(tcp_pool.notify_pipe[0],
    F_TCP_NOTIFY, RCT_PRIO_PROC, NULL) < 0) {
    LM_ERR("failed to add TCP IO notify pipe to reactor\n");
    goto error;
  }

  tcp_pool.threads = pkg_malloc(sizeof(*tcp_pool.threads) * threads_no);
  if (!tcp_pool.threads) {
    LM_ERR("oom while allocating TCP IO threads array\n");
    goto error;
  }

  tcp_pool.stop = 0;
  tcp_pool.threads_no = threads_no;

  for (i = 0; i < threads_no; i++) {
    if (pthread_create(&tcp_pool.threads[i], NULL,
        tcp_thread_routine, NULL) != 0) {
      LM_ERR("failed to start TCP IO thread %d/%d\n", i + 1, threads_no);
      goto error;
    }
    started++;
  }

  LM_NOTICE("TCP single IO mode started with %d threads\n", threads_no);
  return 0;

error:
  cond_lock(&tcp_write_queue->cond);
  tcp_pool.stop = 1;
  cond_broadcast(&tcp_write_queue->cond);
  cond_unlock(&tcp_write_queue->cond);

  if (tcp_pool.threads) {
    for (i = 0; i < started; i++)
      pthread_join(tcp_pool.threads[i], NULL);
    pkg_free(tcp_pool.threads);
    tcp_pool.threads = NULL;
  }
  tcp_pool.threads_no = 0;

  if (tcp_pool.notify_pipe[0] >= 0) {
    reactor_del_reader(tcp_pool.notify_pipe[0], -1, 0);
    close(tcp_pool.notify_pipe[0]);
    tcp_pool.notify_pipe[0] = -1;
  }
  if (tcp_pool.notify_pipe[1] >= 0) {
    close(tcp_pool.notify_pipe[1]);
    tcp_pool.notify_pipe[1] = -1;
  }

  return -1;
}

static void tcp_pool_destroy(void)
{
  int i;
  struct tcp_job *job;
  struct tcp_job *next;
  struct tcp_connection *conn;

  if (!tcp_threads_active() && tcp_pool.notify_pipe[0] < 0)
    return;

  cond_lock(&tcp_write_queue->cond);
  tcp_pool.stop = 1;
  cond_broadcast(&tcp_write_queue->cond);
  cond_unlock(&tcp_write_queue->cond);

  for (i = 0; i < tcp_pool.threads_no; i++)
    pthread_join(tcp_pool.threads[i], NULL);

  if (tcp_pool.threads) {
    pkg_free(tcp_pool.threads);
    tcp_pool.threads = NULL;
  }
  tcp_pool.threads_no = 0;

  if (tcp_pool.notify_pipe[0] >= 0) {
    reactor_del_reader(tcp_pool.notify_pipe[0], -1, 0);
    close(tcp_pool.notify_pipe[0]);
    tcp_pool.notify_pipe[0] = -1;
  }
  if (tcp_pool.notify_pipe[1] >= 0) {
    close(tcp_pool.notify_pipe[1]);
    tcp_pool.notify_pipe[1] = -1;
  }

  cond_lock(&tcp_write_queue->cond);
  for (job = tcp_pool.task_head; job; job = next) {
    next = job->next;
    if (job->conn)
      tcpconn_put(job->conn);
    thread_free(job);
  }
  tcp_pool.task_head = tcp_pool.task_tail = NULL;
  while ((conn = tcp_pop_shared_write_conn_locked()) != NULL) {
    conn->flags &= ~F_CONN_WRITE_QUEUED;
    tcpconn_put(conn);
  }
  cond_unlock(&tcp_write_queue->cond);

  pthread_mutex_lock(&tcp_pool.done_lock);
  for (job = tcp_pool.done_head; job; job = next) {
    next = job->next;
    if (job->conn)
      tcpconn_put(job->conn);
    thread_free(job);
  }
  tcp_pool.done_head = tcp_pool.done_tail = NULL;
  pthread_mutex_unlock(&tcp_pool.done_lock);
}

static int tcp_queue_job(struct tcp_connection *tcpconn, int op)
{
  struct tcp_job *job;

  if (!tcp_threads_active())
    return -1;

  job = thread_malloc(sizeof(*job));
  if (!job) {
    LM_ERR("oom while queuing TCP IO job\n");
    return -1;
  }

  job->conn = tcpconn;
  job->op = op;
  job->run = NULL;
  job->data = NULL;
  job->resp = 0;
  job->ret = 0;
  job->next = NULL;

  cond_lock(&tcp_write_queue->cond);
  if (tcp_pool.task_tail)
    tcp_pool.task_tail->next = job;
  else
    tcp_pool.task_head = job;
  tcp_pool.task_tail = job;
  cond_signal(&tcp_write_queue->cond);
  cond_unlock(&tcp_write_queue->cond);

  return 0;
}

int tcp_async_write_job(struct tcp_connection *tcpconn)
{
  if (!tcp_write_queue)
    return -1;
  if ((tcpconn->flags & F_CONN_HASHED) == 0)
    tcpconn_add(tcpconn);

  cond_lock(&tcp_write_queue->cond);
  if (tcpconn->flags & F_CONN_WRITE_QUEUED) {
    cond_unlock(&tcp_write_queue->cond);
    /* a write job is already queued for this connection and it owns
     * exactly one reference, which it releases on completion; our
     * callers treat success as a transfer of their own reference, so
     * it has to be released here, otherwise it leaks and the
     * connection can never be destroyed */
    tcpconn_put(tcpconn);
    return 0;
  }

  tcpconn->flags |= F_CONN_WRITE_QUEUED;
  tcpconn->wq_next = NULL;
  if (tcp_write_queue->tail)
    tcp_write_queue->tail->wq_next = tcpconn;
  else
    tcp_write_queue->head = tcpconn;
  tcp_write_queue->tail = tcpconn;
  cond_signal(&tcp_write_queue->cond);
  cond_unlock(&tcp_write_queue->cond);
  return 0;
}

int tcp_run_task(tcp_thread_job_f run, void *data)
{
  struct tcp_job *job;

  if (!run)
    return -1;

  if (!tcp_threads_active()) {
    run(data);
    return 0;
  }

  job = thread_malloc(sizeof(*job));
  if (!job) {
    LM_ERR("oom while queuing TCP run job\n");
    return -1;
  }

  job->conn = NULL;
  job->op = TCP_RUN_JOB;
  job->run = run;
  job->data = data;
  job->resp = 0;
  job->ret = -1;
  job->next = NULL;

  cond_lock(&tcp_write_queue->cond);
  if (tcp_pool.task_tail)
    tcp_pool.task_tail->next = job;
  else
    tcp_pool.task_head = job;
  tcp_pool.task_tail = job;
  cond_signal(&tcp_write_queue->cond);
  cond_unlock(&tcp_write_queue->cond);

  return 0;
}

static inline int tcp_queue_write_job(struct tcp_connection *tcpconn)
{
  if (!(tcpconn->flags & F_CONN_REMOVED_READ) && tcpconn->fd != -1) {
    if (reactor_del_reader(tcpconn->fd, -1, 0) == -1)
      return -1;
    tcpconn->flags |= F_CONN_REMOVED_READ;
  }

  if (tcp_async_write_job(tcpconn) < 0)
    return -1;

  return 0;
}

static inline void tcp_fail_conn(struct tcp_connection *tcpconn,
    const char *reason, int report)
{
  if ((tcpconn->flags & F_CONN_REMOVED) != F_CONN_REMOVED &&
      tcpconn->fd != -1) {
    reactor_del_all(tcpconn->fd, -1, IO_FD_CLOSING);
    tcpconn->flags |= F_CONN_REMOVED;
  }

  if (report)
    tcp_trigger_report(tcpconn, TCP_REPORT_CLOSE, (void *)reason);

  tcpconn_destroy(tcpconn);
}

static inline void tcp_complete_read(struct tcp_job *job)
{
  struct tcp_connection *tcpconn;

  tcpconn = job->conn;

  if (job->resp == -2) {
    tcp_fail_conn(tcpconn, "Timeout waiting for a complete message", 1);
    return;
  }

  if (job->resp < 0 || tcpconn->state == S_CONN_BAD) {
    tcp_fail_conn(tcpconn, "Read error", 1);
    return;
  }

  if (tcpconn->state == S_CONN_EOF) {
    tcp_fail_conn(tcpconn, "EOF received", 1);
    return;
  }

  if (tcpconn->flags & F_CONN_REMOVED_READ) {
    if (reactor_add_reader(tcpconn->fd, F_TCPCONN, RCT_PRIO_NET, tcpconn) < 0) {
      LM_ERR("failed to re-add TCP conn %p for read events\n", tcpconn);
      tcp_fail_conn(tcpconn, "Failed to re-arm read", 0);
      return;
    }
    tcpconn->flags &= ~F_CONN_REMOVED_READ;
  }

  tcpconn_put(tcpconn);
}

static inline void tcp_complete_write(struct tcp_job *job)
{
  struct tcp_connection *tcpconn;
  int pending_chunks;

  tcpconn = job->conn;

  if (job->resp < 0 || tcpconn->state == S_CONN_BAD) {
    tcp_fail_conn(tcpconn, "Write error", 1);
    return;
  }

  lock_get(&tcpconn->write_lock);
  pending_chunks = (tcpconn->async && tcpconn->async->pending);
  lock_release(&tcpconn->write_lock);

  if ((tcpconn->flags & F_CONN_REMOVED_READ) && tcpconn->fd != -1) {
    if (reactor_add_reader(tcpconn->fd, F_TCPCONN, RCT_PRIO_NET,
        tcpconn) < 0) {
      LM_ERR("failed to add TCP conn %p for read events\n", tcpconn);
      tcp_fail_conn(tcpconn, "Failed to arm read", 0);
      return;
    }
    tcpconn->flags &= ~F_CONN_REMOVED_READ;
  }

  if (job->resp == 1) {
    if (reactor_add_writer(tcpconn->fd, F_TCPCONN, RCT_PRIO_NET, tcpconn) < 0) {
      LM_ERR("failed to re-add TCP conn %p for write events\n", tcpconn);
      tcp_fail_conn(tcpconn, "Failed to re-arm write", 0);
      return;
    }
    tcpconn->flags &= ~F_CONN_REMOVED_WRITE;
    tcpconn->flags &= ~F_CONN_WRITE_QUEUED;
    tcpconn_put(tcpconn);
    return;
  }

  if (pending_chunks) {
    tcpconn->flags &= ~F_CONN_WRITE_QUEUED;
    if (tcp_async_write_job(tcpconn) < 0) {
      LM_ERR("failed queuing follow-up TCP write job\n");
      tcpconn->flags &= ~F_CONN_WRITE_QUEUED;
      if (reactor_add_writer(tcpconn->fd, F_TCPCONN, RCT_PRIO_NET, tcpconn) < 0) {
        tcp_fail_conn(tcpconn, "Failed queueing follow-up write", 0);
        return;
      }
      tcpconn->flags &= ~F_CONN_REMOVED_WRITE;
      tcpconn_put(tcpconn);
    }
    return;
  }

  tcpconn->flags &= ~F_CONN_WRITE_QUEUED;
  tcpconn_put(tcpconn);
}

static inline int handle_tcp_notify(int fd)
{
  char buf[64];
  int n;
  struct tcp_job *job;

  while ((n = read(fd, buf, sizeof(buf))) > 0)
    ;
  if (n < 0 && errno != EAGAIN && errno != EWOULDBLOCK)
    LM_ERR("failed to read TCP IO notify fd: %s\n", strerror(errno));

  while ((job = tcp_pop_done_job()) != NULL) {
    if (job->op == TCP_READ_JOB)
      tcp_complete_read(job);
    else
      tcp_complete_write(job);
    thread_free(job);
  }

  return 0;
}


/************************ TCP MAIN process functions ************************/

/*! \brief
 * handles a new connection, called internally by tcp_main_loop/handle_io.
 * \param si - pointer to one of the tcp socket_info structures on which
 *              an io event was detected (connection attempt)
 * \return  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(const struct socket_info* si)
{
  union sockaddr_union su;
  struct tcp_connection* tcpconn;
  struct tcp_conn_profile prof;
  socklen_t su_len = sizeof(su);
  int new_sock;
  unsigned int id;

  /* coverity[overrun-buffer-arg: FALSE] - union has 28 bytes, CID #200070 */
  new_sock = accept(si->socket, &(su.s), &su_len);
  if (new_sock == -1) {
    if ((errno == EAGAIN) || (errno == EWOULDBLOCK))
      return 0;
    LM_ERR("failed to accept connection(%d): %s\n", errno, strerror(errno));
    return -1;
  }

  tcp_con_get_profile(&su, &si->su, si->proto, &prof);
  if (tcp_init_sock_opt(new_sock, &prof, si->flags, si->tos) < 0) {
    LM_ERR("tcp_init_sock_opt failed\n");
    close(new_sock);
    return 1; /* success, because the accept was successful */
  }

  /* add socket to list */
  tcpconn = tcpconn_new(new_sock, &su, si, &prof, S_CONN_OK,
    F_CONN_ACCEPTED);
  if (tcpconn) {
    /* Safe: the connection is not yet visible outside TCP main. */
    tcpconn->refcnt++;
    sh_log(tcpconn->hist, TCP_REF, "accept, (%d)", tcpconn->refcnt);
    tcpconn_add(tcpconn);
    LM_DBG("new connection: %p %d flags: %04x\n",
      tcpconn, tcpconn->fd, tcpconn->flags);
    if (reactor_add_reader(tcpconn->fd, F_TCPCONN, RCT_PRIO_NET,
        tcpconn) < 0) {
      LM_ERR("failed to add accepted TCP conn to reactor\n");
      id = tcpconn->id;
      TCPCONN_LOCK(id);
      tcpconn->refcnt--;
      if (tcpconn->refcnt == 0) {
        _tcpconn_rm(tcpconn, 1);
        close(new_sock);
      } else {
        tcpconn->lifetime = 0;
        tcpconn->timeout = 0;
      }
      TCPCONN_UNLOCK(id);
    } else {
      tcpconn->flags &= ~F_CONN_REMOVED_READ;
      tcpconn_put(tcpconn);
    }
  } else {
    LM_ERR("tcpconn_new failed, closing socket\n");
    close(new_sock);
  }
  return 1; /* accept() was successful */
}


/*! \brief
 * handles an io event on one of the watched tcp connections
 *
 * \param    tcpconn - pointer to the tcp_connection for which we have an io ev.
 * \param    fd_i    - index in the fd_array table (needed for delete)
 * \return   handle_* return convention, but on success it always returns 0
 */
inline static int handle_tcpconn_ev(struct tcp_connection* tcpconn, int fd_i,
    int event_type)
{
  int err;
  unsigned int err_len;

  if (event_type == IO_WATCH_READ) {
    LM_DBG("data available on %p %d\n", tcpconn, tcpconn->fd);
    if (reactor_del_reader(tcpconn->fd, fd_i, 0) == -1)
      return -1;
    tcpconn->flags |= F_CONN_REMOVED_READ;
    tcpconn_ref(tcpconn); /* refcnt ++ */
    sh_log(tcpconn->hist, TCP_REF, "tcp-main read queued, (%d)",
      tcpconn->refcnt);
    if (tcp_queue_job(tcpconn, TCP_READ_JOB) < 0) {
      LM_ERR("failed queuing TCP read job\n");
      if (reactor_add_reader(tcpconn->fd, F_TCPCONN, RCT_PRIO_NET,
          tcpconn) < 0) {
        tcp_fail_conn(tcpconn, "Failed queueing read", 0);
        return 0;
      }
      tcpconn->flags &= ~F_CONN_REMOVED_READ;
      tcpconn_put(tcpconn);
    }
    return 0;
  } else {
    LM_DBG("connection %p fd %d is now writable\n", tcpconn, tcpconn->fd);
    /* we received a write event */
    if (tcpconn->state == S_CONN_CONNECTING) {
      /* we're coming from an async connect & write
       * let's see if we connected successfully */
      err_len = sizeof(err);
      if (getsockopt(tcpconn->fd, SOL_SOCKET, SO_ERROR, &err, &err_len) < 0 ||
          err != 0) {
        LM_DBG("Failed connection attempt\n");
        tcpconn_ref(tcpconn);
        sh_log(tcpconn->hist, TCP_REF, "tcpconn connect, (%d)", tcpconn->refcnt);
        reactor_del_all(tcpconn->fd, fd_i, IO_FD_CLOSING);
        tcpconn->flags|=F_CONN_REMOVED;
        tcp_trigger_report(tcpconn, TCP_REPORT_CLOSE,
          "Async connect failed");
        sh_log(tcpconn->hist, TCP_UNREF, "tcpconn connect, (%d)", tcpconn->refcnt);
        tcpconn_destroy(tcpconn);
        return 0;
      }

      /* we successfully connected - further treat this case as if we
       * were coming from an async write */
      tcpconn->state = S_CONN_OK;
      LM_DBG("Successfully completed previous async connect\n");

      /* now that we completed the async connection, we also need to
       * listen for READ events, otherwise these will get lost */
      if (tcpconn->flags & F_CONN_REMOVED_READ) {
        if (reactor_add_reader(tcpconn->fd, F_TCPCONN, RCT_PRIO_NET,
            tcpconn) < 0) {
          LM_ERR("failed to re-arm TCP conn %p for read events\n",
            tcpconn);
          tcp_fail_conn(tcpconn, "Failed to re-arm read", 0);
          return 0;
        }
        tcpconn->flags &= ~F_CONN_REMOVED_READ;
      }

      goto async_write;
    } else {
async_write:
      /* no more write events for now */
        if (reactor_del_writer(tcpconn->fd, fd_i, 0) == -1)
          return -1;
      tcpconn->flags |= F_CONN_REMOVED_WRITE;
      tcpconn_ref(tcpconn); /* refcnt ++ */
      sh_log(tcpconn->hist, TCP_REF, "tcpconn write, (%d)",
        tcpconn->refcnt);
      if (tcp_queue_write_job(tcpconn) < 0) {
        LM_ERR("failed queuing TCP write job\n");
        if (reactor_add_writer(tcpconn->fd, F_TCPCONN, RCT_PRIO_NET,
            tcpconn) < 0) {
          tcp_fail_conn(tcpconn, "Failed queueing write", 0);
          return 0;
        }
        tcpconn->flags &= ~F_CONN_REMOVED_WRITE;
        tcpconn_put(tcpconn);
      }
      return 0;
    }
  }
}


/*! \brief generic handle io routine, it will call the appropiate
 *  handle_xxx() based on the fd_map type
 *
 * \param  fm  - pointer to a fd hash entry
 * \param  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 successful read from the fd (when there might be more io
 *            queued -- the receive buffer might still be non-empty)
 */
inline static int handle_io(struct fd_map* fm, int idx,int event_type)
{
  int ret = 0;

  pt_become_active();
  /* for now we do not do any profiling here as all ops here are
     only internals related to TCP passing beetween processing, no real
     processing
   */
  switch(fm->type){
    case F_TCP_LISTENER:
      ret = handle_new_connect((const struct socket_info*)fm->data);
      break;
    case F_TCPCONN:
      ret = handle_tcpconn_ev((struct tcp_connection*)fm->data, idx,
        event_type);
      break;
    case F_TCP_NOTIFY:
      ret = handle_tcp_notify(fm->fd);
      break;
    case F_IPC:
      ipc_handle_job(fm->fd);
      break;
    case F_NONE:
      LM_CRIT("empty fd map\n");
      goto error;
    default:
      LM_CRIT("unknown fd type %d\n", fm->type);
      goto error;
  }
  pt_become_idle();
  return ret;
error:
  pt_become_idle();
  return -1;
}


/*
 * iterates through all TCP connections and closes expired ones
 * Note: runs once per second at most
 */
#define tcpconn_lifetime(last_sec) \
  do { \
    int now; \
    now = get_ticks(); \
    if (last_sec != now) { \
      last_sec = now; \
      __tcpconn_lifetime(0); \
    } \
  } while (0)


/*! \brief very inefficient for now - FIXME
 * keep in sync with tcpconn_destroy, the "delete" part should be
 * the same except for io_watch_del..
 * \todo FIXME (very inefficient for now)
 */
static inline void __tcpconn_lifetime(int shutdown)
{
  struct tcp_connection *c, *next;
  unsigned int ticks,part;
  unsigned h;
  int fd;
  void *reason;

  if (have_ticks())
    ticks=get_ticks();
  else
    ticks=0;

  for( part=0 ; part<TCP_PARTITION_SIZE ; part++ ) {
    if (!shutdown) TCPCONN_LOCK(part); /* fixme: we can lock only on delete IMO */
    for(h=0; h<TCP_ID_HASH_SIZE; h++){
      c=TCP_PART(part).tcpconn_id_hash[h];
      while(c){
        next=c->id_next;
        if (shutdown || ((c->refcnt == 0) &&
        ((ticks > c->lifetime) ||
        (c->msg_attempts && ticks > c->timeout)))) {
          if (!shutdown)
            LM_DBG("timeout for hash=%d - %p"
                " (%d > %d)\n", h, c, ticks, c->lifetime);
          fd=c->fd;
          /* report the closing of the connection . Note that
           * there are connectioned that use an foced expire to 0
           * as a way to be deleted - we are not interested in */
          /* Also, do not trigger reporting when shutdown
           * is done */
          if (c->lifetime>0 && !shutdown) {
            reason = (c->msg_attempts && ticks > c->timeout) ?
              "Timeout waiting for a complete message" :
              "Timeout on no traffic";
            tcp_trigger_report(c, TCP_REPORT_CLOSE, reason);
          }
          if ((!shutdown)&&(fd>0)&&(c->refcnt==0)) {
            /* if any of read or write are set, we need to remove
             * the fd from the reactor */
            if ((c->flags & F_CONN_REMOVED) != F_CONN_REMOVED){
              reactor_del_all( fd, -1, IO_FD_CLOSING);
              c->flags|=F_CONN_REMOVED;
            }
            close(fd);
            c->fd = -1;
            }
            _tcpconn_rm(c, shutdown?1:0);
          }
          c=next;
        }
    }
    if (!shutdown) TCPCONN_UNLOCK(part);
  }
}


static void tcp_main_server(void)
{
  static unsigned int last_sec = 0;
  struct socket_info_full* sif;
  struct sr_module *m;
  int n;

  /* instruct tls_mgm to initialize all TLS domains */
  for (m=modules; m; m = m->next) {
    if (strcmp(m->exports->name, "tls_mgm") == 0)
      if (init_child(PROC_TCP_MAIN) < 0) {
        LM_ERR("error in init_child for PROC_TCP_MAIN\n");
        goto error;
      }
  }

  /* we run in a separate, dedicated process, with its own reactor
   * (reactors are per process) */
  if (init_worker_reactor("TCP_main", RCT_PRIO_MAX)<0)
    goto error;

  /* now start watching all the fds */

  /* add all the sockets we listens on for connections */
  for (n = PROTO_FIRST; n < PROTO_LAST; n++)
    if (is_tcp_based_proto(n))
      for (sif = protos[n].listeners; sif; sif = sif->next) {
        struct socket_info* si = &sif->socket_info;
        if (protos[n].tran.bind_listener &&
            protos[n].tran.bind_listener(si) < 0) {
          LM_ERR("failed to bind listener [%.*s], proto %s\n",
            si->name.len, si->name.s, protos[n].name);
          goto error;
        }
        if (si->socket != -1 &&
            reactor_add_reader(si->socket, F_TCP_LISTENER,
              RCT_PRIO_NET, si) < 0) {
          LM_ERR("failed to add listen socket to reactor\n");
          goto error;
        }
      }
  /* init: start watching for the IPC jobs */
  if (reactor_add_reader(IPC_FD_READ_SELF, F_IPC, RCT_PRIO_ASYNC, NULL)<0){
    LM_CRIT("failed to add IPC pipe to reactor\n");
    goto error;
  }

  if (tcp_pool_init() < 0)
    goto error;

  is_tcp_main = 1;

  /* main loop (requires "handle_io()" implementation) */
  reactor_main_loop( TCP_MAIN_SELECT_TIMEOUT, error,
      tcpconn_lifetime(last_sec) );

error:
  tcp_pool_destroy();
  destroy_worker_reactor();
  LM_CRIT("exiting...");
  exit(-1);
}



/**************************** Control functions ******************************/

/* initializes the TCP network level in terms of data structures */
int tcp_init(void)
{
  unsigned int i;

  /* first we do auto-detection to see if there are any TCP based
   * protocols loaded */
  for ( i=PROTO_FIRST ; i<PROTO_LAST ; i++ )
    if (is_tcp_based_proto(i) && proto_has_listeners(i)) {
      tcp_disabled=0;
      break;
    }

  tcp_init_con_profiles();

  if (tcp_disabled)
    return 0;

#ifdef DBG_TCPCON
  con_hist = shl_init("TCP con", 10000, 0);
  if (!con_hist) {
    LM_ERR("oom con hist\n");
    goto error;
  }
#endif

  if (tcp_auto_scaling_profile) {
    s_profile = get_scaling_profile(tcp_auto_scaling_profile);
    if (s_profile==NULL) {
      LM_WARN("TCP scaling profile <%s> not defined "
        "-> ignoring it...\n", tcp_auto_scaling_profile);
    } else {
      auto_scaling_enabled = 1;
    }
  }

  tcp_workers_max_no = (s_profile && (tcp_workers_no<s_profile->max_procs)) ?
    s_profile->max_procs : tcp_workers_no ;

  /* init tcp workers array */
  tcp_workers = (struct tcp_worker*)shm_malloc
    ( tcp_workers_max_no*sizeof(struct tcp_worker) );
  if (tcp_workers==0) {
    LM_CRIT("could not alloc tcp_workers array in shm memory\n");
    goto error;
  }
  memset( tcp_workers, 0, tcp_workers_max_no*sizeof(struct tcp_worker));
  /* init globals */
  connection_id=(unsigned int*)shm_malloc(sizeof(unsigned int));
  if (connection_id==0){
    LM_CRIT("could not alloc globals in shm memory\n");
    goto error;
  }
  // The  rand()  function returns a pseudo-random integer in the range 0 to
  // RAND_MAX inclusive (i.e., the mathematical range [0, RAND_MAX]).
  *connection_id=(unsigned int)rand();
  connection_id_lock=lock_alloc();
  if (connection_id_lock==0){
    LM_CRIT("could not alloc connection ID lock\n");
    goto error;
  }
  if (lock_init(connection_id_lock)==0){
    LM_CRIT("could not init connection ID lock\n");
    lock_dealloc((void*)connection_id_lock);
    connection_id_lock=0;
    goto error;
  }
  tcp_connections_no = (unsigned int *)shm_malloc(sizeof(*tcp_connections_no));
  if (tcp_connections_no == 0) {
    LM_CRIT("could not alloc tcp connection counter in shm memory\n");
    goto error;
  }
  *tcp_connections_no = 0;
  tcp_main_proc_no = (int *)shm_malloc(sizeof(*tcp_main_proc_no));
  if (tcp_main_proc_no == 0) {
    LM_CRIT("could not alloc tcp main proc slot in shm memory\n");
    goto error;
  }
  *tcp_main_proc_no = -1;
  tcp_write_queue = shm_malloc(sizeof(*tcp_write_queue));
  if (tcp_write_queue == 0) {
    LM_CRIT("could not alloc tcp shared write queue in shm memory\n");
    goto error;
  }
  memset(tcp_write_queue, 0, sizeof(*tcp_write_queue));
  if (cond_init(&tcp_write_queue->cond) != 0) {
    LM_CRIT("could not init tcp shared write queue cond\n");
    shm_free(tcp_write_queue);
    tcp_write_queue = 0;
    goto error;
  }
  tcp_connections_lock = lock_alloc();
  if (tcp_connections_lock == 0) {
    LM_CRIT("could not alloc tcp connection counter lock\n");
    goto error;
  }
  if (lock_init(tcp_connections_lock) == 0) {
    LM_CRIT("could not init tcp connection counter lock\n");
    lock_dealloc((void *)tcp_connections_lock);
    tcp_connections_lock = 0;
    goto error;
  }
  memset( &tcp_parts, 0, TCP_PARTITION_SIZE*sizeof(struct tcp_partition));
  /* init partitions */
  for( i=0 ; i<TCP_PARTITION_SIZE ; i++ ) {
    /* init lock */
    tcp_parts[i].tcpconn_lock=lock_alloc();
    if (tcp_parts[i].tcpconn_lock==0){
      LM_CRIT("could not alloc lock\n");
      goto error;
    }
    if (lock_init(tcp_parts[i].tcpconn_lock)==0){
      LM_CRIT("could not init lock\n");
      lock_dealloc((void*)tcp_parts[i].tcpconn_lock);
      tcp_parts[i].tcpconn_lock=0;
      goto error;
    }
    /* alloc hashtables*/
    tcp_parts[i].tcpconn_aliases_hash=(struct tcp_conn_alias**)
      shm_malloc(TCP_ALIAS_HASH_SIZE* sizeof(struct tcp_conn_alias*));
    if (tcp_parts[i].tcpconn_aliases_hash==0){
      LM_CRIT("could not alloc address hashtable in shm memory\n");
      goto error;
    }
    tcp_parts[i].tcpconn_id_hash=(struct tcp_connection**)
      shm_malloc(TCP_ID_HASH_SIZE*sizeof(struct tcp_connection*));
    if (tcp_parts[i].tcpconn_id_hash==0){
      LM_CRIT("could not alloc id hashtable in shm memory\n");
      goto error;
    }
    /* init hashtables*/
    memset((void*)tcp_parts[i].tcpconn_aliases_hash, 0,
      TCP_ALIAS_HASH_SIZE * sizeof(struct tcp_conn_alias*));
    memset((void*)tcp_parts[i].tcpconn_id_hash, 0,
      TCP_ID_HASH_SIZE * sizeof(struct tcp_connection*));
  }

  return 0;
error:
  /* clean-up */
  tcp_destroy();
  return -1;
}


/* destroys the TCP data */
void tcp_destroy(void)
{
  int part;

  if (tcp_parts[0].tcpconn_id_hash)
      /* force close/expire for all active tcpconns*/
      __tcpconn_lifetime(1);

  if (connection_id){
    shm_free(connection_id);
    connection_id=0;
  }

  if (tcp_connections_no) {
    shm_free(tcp_connections_no);
    tcp_connections_no = 0;
  }

  if (tcp_main_proc_no) {
    shm_free(tcp_main_proc_no);
    tcp_main_proc_no = 0;
  }

  if (tcp_dispatch_sock[0] >= 0) {
    close(tcp_dispatch_sock[0]);
    tcp_dispatch_sock[0] = -1;
  }
  if (tcp_dispatch_sock[1] >= 0) {
    close(tcp_dispatch_sock[1]);
    tcp_dispatch_sock[1] = -1;
  }

  if (tcp_write_queue) {
    /* Skip cond teardown during attendant shutdown. */
    /* cond_destroy(&tcp_write_queue->cond); */
    shm_free(tcp_write_queue);
    tcp_write_queue = 0;
  }

  if (tcp_connections_lock) {
    lock_destroy(tcp_connections_lock);
    lock_dealloc((void *)tcp_connections_lock);
    tcp_connections_lock = 0;
  }

  if (connection_id_lock){
    lock_destroy(connection_id_lock);
    lock_dealloc((void*)connection_id_lock);
    connection_id_lock=0;
  }

  for ( part=0 ; part<TCP_PARTITION_SIZE ; part++ ) {
    if (tcp_parts[part].tcpconn_id_hash){
      shm_free(tcp_parts[part].tcpconn_id_hash);
      tcp_parts[part].tcpconn_id_hash=0;
    }
    if (tcp_parts[part].tcpconn_aliases_hash){
      shm_free(tcp_parts[part].tcpconn_aliases_hash);
      tcp_parts[part].tcpconn_aliases_hash=0;
    }
    if (tcp_parts[part].tcpconn_lock){
      lock_destroy(tcp_parts[part].tcpconn_lock);
      lock_dealloc((void*)tcp_parts[part].tcpconn_lock);
      tcp_parts[part].tcpconn_lock=0;
    }
  }
}


static int _get_own_tcp_worker_id(void)
{
  pid_t pid;
  int i;

  pid = getpid();
  for( i=0 ; i<tcp_workers_max_no ; i++)
    if(tcp_workers[i].pid==pid)
      return i;

  return -1;
}


void tcp_reset_worker_slot(void)
{
  int i;

  if ((i=_get_own_tcp_worker_id())>=0) {
    tcp_workers[i].state=STATE_INACTIVE;
    tcp_workers[i].pid=0;
    tcp_workers[i].pt_idx=0;
  }
}


static int fork_dynamic_tcp_process(void *foo)
{
  int p_id;
  int r;
  const struct internal_fork_params ifp_sr_tcp = {
    .proc_desc = "SIP receiver TCP",
    .flags = OSS_PROC_DYNAMIC|OSS_PROC_NEEDS_SCRIPT,
    .type = TYPE_TCP,
  };

  /* search for a free slot in the TCP workers table */
  for (r = 0; r < tcp_workers_max_no; r++)
    if (tcp_workers[r].state == STATE_INACTIVE)
      break;

  if (r == tcp_workers_max_no) {
    LM_BUG("trying to fork one more TCP worker but no free slot in "
      "the TCP table (size=%d)\n", tcp_workers_max_no);
    return -1;
  }

  if ((p_id = internal_fork(&ifp_sr_tcp)) < 0) {
    LM_ERR("cannot fork dynamic TCP worker process\n");
    return -1;
  } else if (p_id == 0) {
    /* new TCP worker process */
    if (tcp_dispatch_sock[1] >= 0)
      close(tcp_dispatch_sock[1]);
    tcp_dispatch_sock[1] = -1;

    set_proc_attrs("TCP receiver");
    tcp_workers[r].pid = getpid();

    if (tcp_worker_proc_reactor_init(tcp_dispatch_sock[0]) < 0 ||
    init_child(20000) ||
    self_update_routing_script() < 0)
      goto error;

    report_conditional_status(1, 0);
    clean_read_pipeend();

    tcp_worker_proc_loop();
    destroy_worker_reactor();

error:
    report_failure_status();
    LM_ERR("initializing new TCP worker failed, exiting with error\n");
    pt[process_no].flags |= OSS_PROC_SELFEXIT;
    exit(-1);
  } else {
    /* parent/attendant */
    tcp_workers[r].state = STATE_ACTIVE;
    tcp_workers[r].pt_idx = p_id;
    return p_id;
  }

  return 0;
}


static void tcp_process_graceful_terminate(int sender, void *param)
{
  int i;

  /* accept this only from the attendant process */
  if (sender != 0) {
    LM_BUG("graceful terminate received from a non-main process\n");
    return;
  }
  LM_NOTICE("process %d received RPC to terminate from Main\n", process_no);

  /* reserve this slot until tcp_terminate_worker() detaches the process
   * from the shared dispatch queue and completes its pending async work */
  if ((i = _get_own_tcp_worker_id()) >= 0)
    tcp_workers[i].state = STATE_DRAINING;

  tcp_terminate_worker();
}


/* counts the number of TCP processes to start with; this number may
 * change during runtime due auto-scaling */
int tcp_count_processes(unsigned int *extra)
{
  if (extra) *extra = 0;

  if (tcp_disabled)
    return 0;

  if (s_profile && extra && s_profile->max_procs > tcp_workers_no)
    *extra = s_profile->max_procs - tcp_workers_no;

  return 1 /* tcp main / IO process */ + tcp_workers_no /* dispatch workers */;
}


int tcp_start_processes(int *chd_rank, int *startup_done)
{
  int r, p_id, flags;
  const struct internal_fork_params ifp_sr_tcp = {
    .proc_desc = "SIP receiver TCP",
    .flags = OSS_PROC_NEEDS_SCRIPT,
    .type = TYPE_TCP,
  };

  if (tcp_disabled)
    return 0;

  /* create the shared dispatch socket from TCP main to TCP workers */
  if (socketpair(AF_UNIX, SOCK_DGRAM, 0, tcp_dispatch_sock) < 0) {
    LM_ERR("socketpair failed for TCP worker dispatch: %s\n",
      strerror(errno));
    goto error;
  }
  flags = fcntl(tcp_dispatch_sock[0], F_GETFL);
  if (flags == -1) {
    LM_ERR("fcntl failed for TCP worker dispatch socket: %s\n",
      strerror(errno));
    goto error;
  }
  if (fcntl(tcp_dispatch_sock[0], F_SETFL, flags | O_NONBLOCK) == -1) {
    LM_ERR("failed to set non-blocking on TCP worker dispatch socket: %s\n",
      strerror(errno));
    goto error;
  }
  flags = fcntl(tcp_dispatch_sock[1], F_GETFL);
  if (flags == -1) {
    LM_ERR("fcntl failed for TCP worker dispatch socket: %s\n",
      strerror(errno));
    goto error;
  }
  if (fcntl(tcp_dispatch_sock[1], F_SETFL, flags | O_NONBLOCK) == -1) {
    LM_ERR("failed to set non-blocking on TCP worker dispatch socket: %s\n",
      strerror(errno));
    goto error;
  }

  if (auto_scaling_enabled && s_profile &&
  create_process_group(TYPE_TCP, NULL, s_profile,
    fork_dynamic_tcp_process, tcp_process_graceful_terminate) != 0)
    LM_ERR("failed to create TCP auto-scaling process group; "
      "automatic scaling will not be possible\n");

  for (r = 0; r < tcp_workers_no; r++) {
    (*chd_rank)++;
    p_id = internal_fork(&ifp_sr_tcp);
    if (p_id < 0) {
      LM_ERR("cannot fork TCP worker process %d\n", r);
      goto error;
    } else if (p_id > 0) {
      /* parent */
      tcp_workers[r].state = STATE_ACTIVE;
      tcp_workers[r].pt_idx = p_id;
      continue;
    }

    /* child */
    if (tcp_dispatch_sock[1] >= 0)
      close(tcp_dispatch_sock[1]);
    tcp_dispatch_sock[1] = -1;

    set_proc_attrs("TCP receiver");
    tcp_workers[r].pid = getpid();

    if (tcp_worker_proc_reactor_init(tcp_dispatch_sock[0]) < 0 ||
        init_child(*chd_rank) < 0) {
      LM_ERR("init_child failed for TCP worker %d\n", r);
      report_failure_status();
      if (startup_done)
        *startup_done = -1;
      exit(-1);
    }

    /* first TCP worker runs startup_route if not already run */
    if (startup_done && *startup_done == 0 && r == 0) {
      LM_DBG("running startup route for first TCP worker\n");
      if (run_startup_route() < 0) {
        LM_ERR("startup route processing failed in TCP worker\n");
        report_failure_status();
        *startup_done = -1;
        exit(-1);
      }
      *startup_done = 1;
    }

    report_conditional_status((!no_daemon_mode), 0);
    tcp_worker_proc_loop();
  }

  /* Keep the worker endpoint in the attendant when auto-scaling is enabled,
   * so dynamically forked TCP workers can inherit the shared queue. */
  if (!s_profile && tcp_dispatch_sock[0] >= 0) {
    close(tcp_dispatch_sock[0]);
    tcp_dispatch_sock[0] = -1;
  }

  if (startup_done && tcp_workers_no > 0)
    while (!(*startup_done)) {
      usleep(5);
      handle_sigs();
    }

  return 0;
error:
  if (tcp_dispatch_sock[0] >= 0) {
    close(tcp_dispatch_sock[0]);
    tcp_dispatch_sock[0] = -1;
  }
  if (tcp_dispatch_sock[1] >= 0) {
    close(tcp_dispatch_sock[1]);
    tcp_dispatch_sock[1] = -1;
  }
  return -1;
}

static int tcpconn_update_local_port(struct tcp_connection *tcpconn)
{
  union sockaddr_union local_su;
  socklen_t su_size;

  su_size = sockaddru_len(tcpconn->rcv.src_su);
  if (getsockname(tcpconn->fd, (struct sockaddr *)&local_su, &su_size) < 0) {
    LM_ERR("failed to get local socket info on conn %u: %s\n",
      tcpconn->id, strerror(errno));
    return -1;
  }

  tcpconn->rcv.dst_port = su_getport(&local_su);
  return 0;
}

static int tcpconn_prepare_write(struct tcp_connection *tcpconn)
{
  int fd;
  int connected = 0;

  if ((tcpconn->flags & F_CONN_INIT) == 0) {
    if (protos[tcpconn->type].net.stream.conn.init &&
        protos[tcpconn->type].net.stream.conn.init(tcpconn) < 0) {
      LM_ERR("failed to init proto %d conn %p in TCP main\n",
        tcpconn->type, tcpconn);
      return -1;
    }
    tcpconn->flags |= F_CONN_INIT;
  }

  if (tcpconn->fd < 0) {
    if (!tcpconn->rcv.bind_address) {
      LM_ERR("missing bind_address for outbound conn %u\n", tcpconn->id);
      return -1;
    }

    fd = tcp_sync_connect_fd(&tcpconn->rcv.bind_address->su,
        &tcpconn->rcv.src_su, tcpconn->type, &tcpconn->profile,
        tcpconn->rcv.bind_address->flags,
        tcpconn->rcv.bind_address->tos);
    if (fd < 0)
      return -1;

    tcpconn->fd = fd;
    if (tcpconn_update_local_port(tcpconn) < 0) {
      close(fd);
      tcpconn->fd = -1;
      return -1;
    }

    tcpconn->state = S_CONN_OK;
    connected = 1;
  }

  if (connected && protos[tcpconn->type].net.stream.conn.connect) {
    if (protos[tcpconn->type].net.stream.conn.connect(tcpconn) < 0) {
      LM_ERR("failed to finish proto %d connect on conn %u\n",
        tcpconn->type, tcpconn->id);
      return -1;
    }
  }

  return 0;
}


int tcp_start_listener(void)
{
  int p_id;
  const struct internal_fork_params ifp_tcp_main = {
    .proc_desc = "TCP main",
    .flags = 0,
    .type = TYPE_NONE,
  };

  if (tcp_disabled)
    return 0;

  /* start the TCP manager process */
  if ( (p_id=internal_fork(&ifp_tcp_main))<0 ) {
    LM_CRIT("cannot fork tcp main process\n");
    goto error;
  }else if (p_id==0){
      /* child */
    report_conditional_status( (!no_daemon_mode), 0);

    tcp_main_server();
    exit(-1);
  }
  *tcp_main_proc_no = p_id;

  return 0;
error:
  return -1;
}

int tcp_has_async_write(void)
{
  return reactor_has_async();
}

static int tcp_close_conn_run(void *data)
{
  struct tcp_connection *conn = data;

  if (!conn)
    return -1;

  tcp_conn_destroy(conn);
  return 0;
}

static void tcp_close_conn_rpc(int pid, void *param)
{
  tcp_close_conn_run(param);
}

int tcp_close_connection(str *ipport)
{
  struct tcp_connection *conn = NULL;
  struct ip_addr *ip;
  str host;
  unsigned int id;
  int port, proto, rc, tcp_main_proc;
  int start_proto, end_proto;
  unsigned int p;
  char *sep;

  if (tcp_disabled)
    return 0;

  sep = q_memchr(ipport->s, ':', ipport->len);
  if (!sep) {
    if (str2int(ipport, &id) < 0) {
      LM_ERR("failed to parse tcp connection [%.*s]\n",
        ipport->len, ipport->s);
      return -1;
    }

    switch (tcp_conn_get(id, NULL, 0, PROTO_NONE, NULL, &conn, NULL)) {
    case 1:
      goto found;
    case -1:
      return -1;
    default:
      return 0;
    }
  }

  if (parse_phostport(ipport->s, ipport->len, &host.s, &host.len,
  &port, &proto) != 0 || port <= 0) {
    LM_ERR("failed to parse tcp connection [%.*s]\n",
      ipport->len, ipport->s);
    return -1;
  }

  ip = str2ip(&host);
  if (!ip)
    ip = str2ip6(&host);
  if (!ip) {
    LM_ERR("invalid IP in tcp connection [%.*s]\n",
      ipport->len, ipport->s);
    return -1;
  }

  if (proto != PROTO_NONE) {
    if (!is_tcp_based_proto(proto)) {
      LM_ERR("protocol %d is not TCP based for [%.*s]\n",
        proto, ipport->len, ipport->s);
      return -1;
    }
    start_proto = proto;
    end_proto = proto + 1;
  } else {
    start_proto = 0;
    end_proto = PROTO_LAST;
  }

  for (p = start_proto; p < (unsigned int)end_proto; p++) {
    if (!is_tcp_based_proto(p))
      continue;

    switch (tcp_conn_get(0, ip, port, p, NULL, &conn, NULL)) {
    case 1:
      goto found;
    case -1:
      return -1;
    }
  }

  return 0;

found:
  TCPCONN_LOCK(conn->id);
  conn->flags |= F_CONN_FORCE_CLOSED;
  TCPCONN_UNLOCK(conn->id);

  tcp_main_proc = tcp_get_main_proc_no();
  if (tcp_main_proc < 0)
    rc = -1;
  else if (process_no == tcp_main_proc)
    rc = tcp_close_conn_run(conn);
  else
    rc = ipc_send_rpc(tcp_main_proc, tcp_close_conn_rpc, conn);

  if (rc < 0) {
    TCPCONN_LOCK(conn->id);
    conn->flags &= ~F_CONN_FORCE_CLOSED;
    TCPCONN_UNLOCK(conn->id);
    tcp_conn_release(conn, 0);
    return -1;
  }

  return 1;
}


/***************************** MI functions **********************************/

mi_response_t *mi_tcp_list_conns(const mi_params_t *params,
            struct mi_handler *async_hdl)
{
  mi_response_t *resp;
  mi_item_t *resp_obj;
  mi_item_t *conns_arr, *conn_item;
  struct tcp_connection *conn;
  time_t _ts;
  char date_buf[MI_DATE_BUF_LEN];
  int date_buf_len;
  unsigned int i,j,part;
  char proto_buf[PROTO_NAME_MAX_SIZE];
  char *proto_s;
  int proto_len;
  int proto_filter = PROTO_NONE;
  int filter_by_proto = 0;
  str proto_name;
  struct tm ltime;
  char *p;

  if (tcp_disabled)
    return init_mi_result_null();

  switch (try_get_mi_string_param(params, "proto", &proto_s, &proto_len)) {
  case 0:
    proto_name.s = proto_s;
    proto_name.len = proto_len;

    if (proto_len == 3 && str_strcasecmp(&proto_name, _str("any")) == 0)
      ;
    else if (parse_proto((unsigned char *)proto_s, proto_len,
        (int *)&proto_filter) < 0)
      return init_mi_error(400, MI_SSTR("Bad protocol"));
    else
      filter_by_proto = 1;
    break;
  case -1:
    break;
  default:
    return init_mi_param_error();
  }

  resp = init_mi_result_object(&resp_obj);
  if (!resp)
    return 0;

  conns_arr = add_mi_array(resp_obj, MI_SSTR("Connections"));
  if (!conns_arr) {
    free_mi_response(resp);
    return 0;
  }

  for( part=0 ; part<TCP_PARTITION_SIZE ; part++) {
    TCPCONN_LOCK(part);
    for( i=0; i<TCP_ID_HASH_SIZE ; i++ ) {
      for(conn=TCP_PART(part).tcpconn_id_hash[i];conn;conn=conn->id_next){
        if (filter_by_proto && conn->type != proto_filter)
          continue;

        /* add one object fo each conn */
        conn_item = add_mi_object(conns_arr, 0, 0);
        if (!conn_item)
          goto error;

        /* add ID */
        if (add_mi_number(conn_item, MI_SSTR("ID"), conn->id) < 0)
          goto error;

        /* add type/proto */
        p = proto2str(conn->type, proto_buf);
        if (add_mi_string(conn_item, MI_SSTR("Type"), proto_buf,
          (int)(long)(p-proto_buf)) < 0)
          goto error;

        /* add state */
        if (add_mi_number(conn_item, MI_SSTR("State"), conn->state) < 0)
          goto error;

        /* add Remote IP:Port */
        if (add_mi_string_fmt(conn_item, MI_SSTR("Remote"), "%s:%d",
          ip_addr2a(&conn->rcv.src_ip), conn->rcv.src_port) < 0)
          goto error;

        /* add Local IP:Port */
        if (add_mi_string_fmt(conn_item, MI_SSTR("Local"), "%s:%d",
          ip_addr2a(&conn->rcv.dst_ip), conn->rcv.dst_port) < 0)
          goto error;

        if (protos[conn->type].net.stream.conn.dump &&
            protos[conn->type].net.stream.conn.dump(conn,
              conn_item) < 0)
          goto error;

        /* add lifetime */
        _ts = (time_t)conn->lifetime + startup_time;
        localtime_r(&_ts, &ltime);
        date_buf_len = strftime(date_buf, MI_DATE_BUF_LEN - 1,
                    "%Y-%m-%d %H:%M:%S", &ltime);
        if (date_buf_len != 0) {
          if (add_mi_string(conn_item, MI_SSTR("Lifetime"),
            date_buf, date_buf_len) < 0)
            goto error;
        } else {
          if (add_mi_number(conn_item, MI_SSTR("Lifetime"), _ts) < 0)
            goto error;
        }

        /* add the port-aliases */
        for( j=0 ; j<conn->aliases ; j++ )
          /* add one node for each conn */
          add_mi_number( conn_item, MI_SSTR("Alias port"),
            conn->con_aliases[j].port );

      }
    }

    TCPCONN_UNLOCK(part);
  }

  return resp;

error:
  TCPCONN_UNLOCK(part);
  LM_ERR("failed to add MI item\n");
  free_mi_response(resp);
  return 0;
}


mi_response_t *mi_tcp_close_conn(const mi_params_t *params,
            struct mi_handler *_)
{
  str ipport;
  int rc;

  if (get_mi_string_param(params, "ipport", &ipport.s, &ipport.len) < 0)
    return init_mi_param_error();

  rc = tcp_close_connection(&ipport);
  if (rc < 0)
    return init_mi_error(400, MI_SSTR("Bad tcp connection"));
  if (rc == 0)
    return init_mi_result_null();

  return init_mi_result_ok();
}

Coverage Report

Created: 2026-08-31 06:36