url.c

           RFC1738 (section 3.1, page 5) says:

           The rest of the locator consists of data specific to the scheme,
           and is known as the "url-path". It supplies the details of how the
           specified resource can be accessed. Note that the "/" between the
           host (or port) and the url-path is NOT part of the url-path.

           As most agents use file://localhost/foo to get '/foo' although the
           slash preceding foo is a separator and not a slash for the path,
           a URL as file://localhost//foo must be valid as well, to refer to
           the same file with an absolute path.
        */

        if(ptr[1] && ('/' == ptr[1]))
          /* if there was two slashes, we skip the first one as that is then
             used truly as a separator */
          ptr++;

        /* This cannot be made with strcpy, as the memory chunks overlap! */
        memmove(path, ptr, strlen(ptr)+1);
      }
    }

    strcpy(conn->protostr, "file"); /* store protocol string lowercase */
  }
  else {
    /* clear path */
    path[0]=0;

    if(2 > sscanf(data->change.url,
                   "%15[^\n:]://%[^\n/]%[^\n]",
                   conn->protostr,
                   conn->host.name, path)) {

      /*
       * The URL was badly formatted, let's try the browser-style _without_
       * protocol specified like 'http://'.
       */
      if(1 > (rc = sscanf(data->change.url, "%[^\n/]%[^\n]",
                          conn->host.name, path)) ) {
        /*
         * We couldn't even get this format.
         * djgpp 2.04 has a sscanf() bug where 'conn->host.name' is
         * assigned, but the return value is EOF!
         */
#if defined(__DJGPP__) && (DJGPP_MINOR == 4)
        if (!(rc == -1 && *conn->host.name))
#endif
        {
          failf(data, "<url> malformed");
          return CURLE_URL_MALFORMAT;
        }
      }

      /*
       * Since there was no protocol part specified, we guess what protocol it
       * is based on the first letters of the server name.
       */

      /* Note: if you add a new protocol, please update the list in
       * lib/version.c too! */

      if(checkprefix("FTP.", conn->host.name))
        strcpy(conn->protostr, "ftp");
      else if(checkprefix("DICT.", conn->host.name))
        strcpy(conn->protostr, "DICT");
      else if(checkprefix("LDAP.", conn->host.name))
        strcpy(conn->protostr, "LDAP");
      else {
        strcpy(conn->protostr, "http");
      }

      conn->protocol |= PROT_MISSING; /* not given in URL */
    }
  }

  /* We search for '?' in the host name (but only on the right side of a
   * @-letter to allow ?-letters in username and password) to handle things
   * like http://example.com?param= (notice the missing '/').
   */
  at = strchr(conn->host.name, '@');
  if(at)
    tmp = strchr(at+1, '?');
  else
    tmp = strchr(conn->host.name, '?');

  if(tmp) {
    /* We must insert a slash before the '?'-letter in the URL. If the URL had
       a slash after the '?', that is where the path currently begins and the
       '?string' is still part of the host name.

       We must move the trailing part from the host name and put it first in
       the path. And have it all prefixed with a slash.
    */

    size_t hostlen = strlen(tmp);
    size_t pathlen = strlen(path);

    /* move the existing path plus the zero byte forward, to make room for
       the host-name part */
    memmove(path+hostlen+1, path, pathlen+1);

     /* now copy the trailing host part in front of the existing path */
    memcpy(path+1, tmp, hostlen);

    path[0]='/'; /* prepend the missing slash */

    *tmp=0; /* now cut off the hostname at the ? */
  }
  else if(!path[0]) {
    /* if there's no path set, use a single slash */
    strcpy(path, "/");
  }

  /* If the URL is malformatted (missing a '/' after hostname before path) we
   * insert a slash here. The only letter except '/' we accept to start a path
   * is '?'.
   */
  if(path[0] == '?') {
    /* We need this function to deal with overlapping memory areas. We know
       that the memory area 'path' points to is 'urllen' bytes big and that
       is bigger than the path. Use +1 to move the zero byte too. */
    memmove(&path[1], path, strlen(path)+1);
    path[0] = '/';
  }

  if (conn->host.name[0] == '[') {
    /* This looks like an IPv6 address literal.  See if there is an address
       scope.  */
    char *percent = strstr (conn->host.name, "%25");
    if (percent) {
      char *endp;
      unsigned int scope = strtoul (percent + 3, &endp, 10);
      if (*endp == ']') {
        /* The address scope was well formed.  Knock it out of the hostname.  */
        memmove(percent, endp, strlen(endp)+1);
        if (!data->state.this_is_a_follow)
          /* Don't honour a scope given in a Location: header */
          conn->scope = scope;
      } else
        infof(data, "Invalid IPv6 address format\n");
    }
  }

  if (data->set.scope)
    /* Override any scope that was set above.  */
    conn->scope = data->set.scope;

  /*
   * So if the URL was A://B/C,
   *   conn->protostr is A
   *   conn->host.name is B
   *   data->state.path is /C
   */
  (void)rc;
  return CURLE_OK;
}

static void llist_dtor(void *user, void *element)
{
  (void)user;
  (void)element;
  /* Do nothing */
}

/*
 * If we're doing a resumed transfer, we need to setup our stuff
 * properly.
 */
static CURLcode setup_range(struct SessionHandle *data)
{
  struct UrlState *s = &data->state;
  s->resume_from = data->set.set_resume_from;
  if(s->resume_from || data->set.str[STRING_SET_RANGE]) {
    if(s->rangestringalloc)
      free(s->range);

    if(s->resume_from)
      s->range = aprintf("%" FORMAT_OFF_TU "-", s->resume_from);
    else
      s->range = strdup(data->set.str[STRING_SET_RANGE]);

    s->rangestringalloc = (bool)(s->range?TRUE:FALSE);

    if(!s->range)
      return CURLE_OUT_OF_MEMORY;

    /* tell ourselves to fetch this range */
    s->use_range = TRUE;        /* enable range download */
  }
  else
    s->use_range = FALSE; /* disable range download */

  return CURLE_OK;
}


/***************************************************************
* Setup connection internals specific to the requested protocol
***************************************************************/
static CURLcode setup_connection_internals(struct SessionHandle *data,
                                           struct connectdata *conn)
{
  const struct Curl_handler * const * pp;
  const struct Curl_handler * p;
  CURLcode result;

  conn->socktype = SOCK_STREAM; /* most of them are TCP streams */

  /* Scan protocol handler table. */

  for (pp = protocols; (p = *pp) != NULL; pp++)
    if(strequal(p->scheme, conn->protostr)) {
      /* Protocol found in table. Perform setup complement if some. */
      conn->handler = p;

      if(p->setup_connection) {
        result = (*p->setup_connection)(conn);

        if(result != CURLE_OK)
          return result;

        p = conn->handler;              /* May have changed. */
      }

      conn->port = p->defport;
      conn->remote_port = (unsigned short)p->defport;
      conn->protocol |= p->protocol;
      return CURLE_OK;
    }

  /* The protocol was not found in the table, but we don't have to assign it
     to anything since it is already assigned to a dummy-struct in the
     create_conn() function when the connectdata struct is allocated. */
  failf(data, "Protocol %s not supported or disabled in " LIBCURL_NAME,
        conn->protostr);
  return CURLE_UNSUPPORTED_PROTOCOL;
}

/****************************************************************
* Detect what (if any) proxy to use. Remember that this selects a host
* name and is not limited to HTTP proxies only.
* The returned pointer must be freed by the caller (unless NULL)
****************************************************************/
static char *detect_proxy(struct connectdata *conn)
{
  char *proxy = NULL;

#ifndef CURL_DISABLE_HTTP
  /* If proxy was not specified, we check for default proxy environment
   * variables, to enable i.e Lynx compliance:
   *
   * http_proxy=http://some.server.dom:port/
   * https_proxy=http://some.server.dom:port/
   * ftp_proxy=http://some.server.dom:port/
   * no_proxy=domain1.dom,host.domain2.dom
   *   (a comma-separated list of hosts which should
   *   not be proxied, or an asterisk to override
   *   all proxy variables)
   * all_proxy=http://some.server.dom:port/
   *   (seems to exist for the CERN www lib. Probably
   *   the first to check for.)
   *
   * For compatibility, the all-uppercase versions of these variables are
   * checked if the lowercase versions don't exist.
   */
  char *no_proxy=NULL;
  char *no_proxy_tok_buf;
  char proxy_env[128];

  no_proxy=curl_getenv("no_proxy");
  if(!no_proxy)
    no_proxy=curl_getenv("NO_PROXY");

  if(!no_proxy || !strequal("*", no_proxy)) {
    /* NO_PROXY wasn't specified or it wasn't just an asterisk */
    char *nope;

    nope=no_proxy?strtok_r(no_proxy, ", ", &no_proxy_tok_buf):NULL;
    while(nope) {
      size_t namelen;
      char *endptr = strchr(conn->host.name, ':');
      if(endptr)
        namelen=endptr-conn->host.name;
      else
        namelen=strlen(conn->host.name);

      if(strlen(nope) <= namelen) {
        char *checkn=
          conn->host.name + namelen - strlen(nope);
        if(checkprefix(nope, checkn)) {
          /* no proxy for this host! */
          break;
        }
      }
      nope=strtok_r(NULL, ", ", &no_proxy_tok_buf);
    }
    if(!nope) {
      /* It was not listed as without proxy */
      char *protop = conn->protostr;
      char *envp = proxy_env;
      char *prox;

      /* Now, build <protocol>_proxy and check for such a one to use */
      while(*protop)
        *envp++ = (char)tolower((int)*protop++);

      /* append _proxy */
      strcpy(envp, "_proxy");

      /* read the protocol proxy: */
      prox=curl_getenv(proxy_env);

      /*
       * We don't try the uppercase version of HTTP_PROXY because of
       * security reasons:
       *
       * When curl is used in a webserver application
       * environment (cgi or php), this environment variable can
       * be controlled by the web server user by setting the
       * http header 'Proxy:' to some value.
       *
       * This can cause 'internal' http/ftp requests to be
       * arbitrarily redirected by any external attacker.
       */
      if(!prox && !strequal("http_proxy", proxy_env)) {
        /* There was no lowercase variable, try the uppercase version: */
        for(envp = proxy_env; *envp; envp++)
          *envp = (char)toupper((int)*envp);
        prox=curl_getenv(proxy_env);
      }

      if(prox && *prox) { /* don't count "" strings */
        proxy = prox; /* use this */
      }
      else {
        proxy = curl_getenv("all_proxy"); /* default proxy to use */
        if(!proxy)
          proxy=curl_getenv("ALL_PROXY");
      }

      if(proxy && *proxy) {
        long bits = conn->protocol & (PROT_HTTPS|PROT_SSL|PROT_MISSING);

        if(conn->proxytype == CURLPROXY_HTTP) {
          /* force this connection's protocol to become HTTP */
          conn->protocol = PROT_HTTP | bits;
          conn->bits.proxy = conn->bits.httpproxy = TRUE;
        }
      }
    } /* if(!nope) - it wasn't specified non-proxy */
  } /* NO_PROXY wasn't specified or '*' */
  if(no_proxy)
    free(no_proxy);

#else /* !CURL_DISABLE_HTTP */

  (void)conn;
#endif /* CURL_DISABLE_HTTP */

  return proxy;
}

/*
 * If this is supposed to use a proxy, we need to figure out the proxy
 * host name, so that we can re-use an existing connection
 * that may exist registered to the same proxy host.
 * proxy will be freed before this function returns.
 */
static CURLcode parse_proxy(struct SessionHandle *data,
                            struct connectdata *conn, char *proxy)
{
  char *prox_portno;
  char *endofprot;

  /* We use 'proxyptr' to point to the proxy name from now on... */
  char *proxyptr=proxy;
  char *portptr;
  char *atsign;

  /* We do the proxy host string parsing here. We want the host name and the
   * port name. Accept a protocol:// prefix, even though it should just be
   * ignored.
   */

  /* Skip the protocol part if present */
  endofprot=strstr(proxyptr, "://");
  if(endofprot)
    proxyptr = endofprot+3;

  /* Is there a username and password given in this proxy url? */
  atsign = strchr(proxyptr, '@');
  if(atsign) {
    char proxyuser[MAX_CURL_USER_LENGTH];
    char proxypasswd[MAX_CURL_PASSWORD_LENGTH];
    proxypasswd[0] = 0;

    if(1 <= sscanf(proxyptr,
                   "%" MAX_CURL_USER_LENGTH_TXT"[^:@]:"
                   "%" MAX_CURL_PASSWORD_LENGTH_TXT "[^@]",
                   proxyuser, proxypasswd)) {
      CURLcode res = CURLE_OK;

      /* found user and password, rip them out.  note that we are
         unescaping them, as there is otherwise no way to have a
         username or password with reserved characters like ':' in
         them. */
      Curl_safefree(conn->proxyuser);
      conn->proxyuser = curl_easy_unescape(data, proxyuser, 0, NULL);

      if(!conn->proxyuser)
        res = CURLE_OUT_OF_MEMORY;
      else {
        Curl_safefree(conn->proxypasswd);
        conn->proxypasswd = curl_easy_unescape(data, proxypasswd, 0, NULL);

        if(!conn->proxypasswd)
          res = CURLE_OUT_OF_MEMORY;
      }

      if(CURLE_OK == res) {
        conn->bits.proxy_user_passwd = TRUE; /* enable it */
        atsign = strdup(atsign+1); /* the right side of the @-letter */

        if(atsign) {
          free(proxy); /* free the former proxy string */
          proxy = proxyptr = atsign; /* now use this instead */
        }
        else
          res = CURLE_OUT_OF_MEMORY;
      }

      if(res) {
        free(proxy); /* free the allocated proxy string */
        return res;
      }
    }
  }

  /* start scanning for port number at this point */
  portptr = proxyptr;

  /* detect and extract RFC2732-style IPv6-addresses */
  if(*proxyptr == '[') {
    char *ptr = ++proxyptr; /* advance beyond the initial bracket */
    while(*ptr && (ISXDIGIT(*ptr) || (*ptr == ':') || (*ptr == '%') || (*ptr == '.')))
      ptr++;
    if(*ptr == ']') {
      /* yeps, it ended nicely with a bracket as well */
      *ptr++ = 0;
    } else
      infof(data, "Invalid IPv6 address format\n");
    portptr = ptr;
    /* Note that if this didn't end with a bracket, we still advanced the
     * proxyptr first, but I can't see anything wrong with that as no host
     * name nor a numeric can legally start with a bracket.
     */
  }

  /* Get port number off proxy.server.com:1080 */
  prox_portno = strchr(portptr, ':');
  if(prox_portno) {
    *prox_portno = 0x0; /* cut off number from host name */
    prox_portno ++;
    /* now set the local port number */
    conn->port = atoi(prox_portno);
  }
  else {
    /* without a port number after the host name, some people seem to use
       a slash so we strip everything from the first slash */
    atsign = strchr(proxyptr, '/');
    if(atsign)
      *atsign = 0x0; /* cut off path part from host name */

    if(data->set.proxyport)
      /* None given in the proxy string, then get the default one if it is
         given */
      conn->port = data->set.proxyport;
  }

  /* now, clone the cleaned proxy host name */
  conn->proxy.rawalloc = strdup(proxyptr);
  conn->proxy.name = conn->proxy.rawalloc;

  free(proxy);
  if(!conn->proxy.rawalloc)
    return CURLE_OUT_OF_MEMORY;

  return CURLE_OK;
}

/*
 * Extract the user and password from the authentication string
 */
static CURLcode parse_proxy_auth(struct SessionHandle *data,
                                 struct connectdata *conn)
{
  char proxyuser[MAX_CURL_USER_LENGTH]="";
  char proxypasswd[MAX_CURL_PASSWORD_LENGTH]="";

  sscanf(data->set.str[STRING_PROXYUSERPWD],
         "%" MAX_CURL_USER_LENGTH_TXT "[^:]:"
         "%" MAX_CURL_PASSWORD_LENGTH_TXT "[^\n]",
         proxyuser, proxypasswd);

  conn->proxyuser = curl_easy_unescape(data, proxyuser, 0, NULL);
  if(!conn->proxyuser)
    return CURLE_OUT_OF_MEMORY;

  conn->proxypasswd = curl_easy_unescape(data, proxypasswd, 0, NULL);
  if(!conn->proxypasswd)
    return CURLE_OUT_OF_MEMORY;

  return CURLE_OK;
}

/*
 *
 * Parse a user name and password in the URL and strip it out of the host name
 *
 * Inputs: data->set.use_netrc (CURLOPT_NETRC)
 *         conn->host.name
 *
 * Outputs: (almost :- all currently undefined)
 *          conn->bits.user_passwd  - non-zero if non-default passwords exist
 *          user                    - non-zero length if defined
 *          passwd                  -   ditto
 *          conn->host.name         - remove user name and password
 */
static CURLcode parse_url_userpass(struct SessionHandle *data,
                                   struct connectdata *conn,
                                   char *user, char *passwd)
{
  /* At this point, we're hoping all the other special cases have
   * been taken care of, so conn->host.name is at most
   *    [user[:password]]@]hostname
   *
   * We need somewhere to put the embedded details, so do that first.
   */

  user[0] =0;   /* to make everything well-defined */
  passwd[0]=0;

  if(conn->protocol & (PROT_FTP|PROT_HTTP|PROT_SCP|PROT_SFTP)) {
    /* This is a FTP, HTTP, SCP or SFTP URL, we will now try to extract the
     * possible user+password pair in a string like:
     * ftp://user:password@ftp.my.site:8021/README */
    char *ptr=strchr(conn->host.name, '@');
    char *userpass = conn->host.name;
    if(ptr != NULL) {
      /* there's a user+password given here, to the left of the @ */

      conn->host.name = ++ptr;

      /* So the hostname is sane.  Only bother interpreting the
       * results if we could care.  It could still be wasted
       * work because it might be overtaken by the programmatically
       * set user/passwd, but doing that first adds more cases here :-(
       */

      if(data->set.use_netrc != CURL_NETRC_REQUIRED) {
        /* We could use the one in the URL */

        conn->bits.user_passwd = 1; /* enable user+password */

        if(*userpass != ':') {
          /* the name is given, get user+password */
          sscanf(userpass, "%" MAX_CURL_USER_LENGTH_TXT "[^:@]:"
                 "%" MAX_CURL_PASSWORD_LENGTH_TXT "[^@]",
                 user, passwd);
        }
        else
          /* no name given, get the password only */
          sscanf(userpass, ":%" MAX_CURL_PASSWORD_LENGTH_TXT "[^@]", passwd);

        if(user[0]) {
          char *newname=curl_easy_unescape(data, user, 0, NULL);
          if(!newname)
            return CURLE_OUT_OF_MEMORY;
          if(strlen(newname) < MAX_CURL_USER_LENGTH)
            strcpy(user, newname);

          /* if the new name is longer than accepted, then just use
             the unconverted name, it'll be wrong but what the heck */
          free(newname);
        }
        if(passwd[0]) {
          /* we have a password found in the URL, decode it! */
          char *newpasswd=curl_easy_unescape(data, passwd, 0, NULL);
          if(!newpasswd)
            return CURLE_OUT_OF_MEMORY;
          if(strlen(newpasswd) < MAX_CURL_PASSWORD_LENGTH)
            strcpy(passwd, newpasswd);

          free(newpasswd);
        }
      }
    }
  }
  return CURLE_OK;
}

/*************************************************************
 * Figure out the remote port number and fix it in the URL
 *
 * No matter if we use a proxy or not, we have to figure out the remote
 * port number of various reasons.
 *
 * To be able to detect port number flawlessly, we must not confuse them
 * IPv6-specified addresses in the [0::1] style. (RFC2732)
 *
 * The conn->host.name is currently [user:passwd@]host[:port] where host
 * could be a hostname, IPv4 address or IPv6 address.
 *
 * The port number embedded in the URL is replaced, if necessary.
 *************************************************************/
static CURLcode parse_remote_port(struct SessionHandle *data,
                                  struct connectdata *conn)
{
  char *portptr;
  char endbracket;

  if((1 == sscanf(conn->host.name, "[%*39[0123456789abcdefABCDEF:.%]%c", &endbracket)) &&
     (']' == endbracket)) {
    /* this is a RFC2732-style specified IP-address */
    conn->bits.ipv6_ip = TRUE;

    conn->host.name++; /* skip over the starting bracket */
    portptr = strchr(conn->host.name, ']');
    *portptr++ = 0; /* zero terminate, killing the bracket */
    if(':' != *portptr)
      portptr = NULL; /* no port number available */
  }
  else
    portptr = strrchr(conn->host.name, ':');

  if(data->set.use_port && data->state.allow_port) {
    /* if set, we use this and ignore the port possibly given in the URL */
    conn->remote_port = (unsigned short)data->set.use_port;
    if(portptr)
      *portptr = '\0'; /* cut off the name there anyway - if there was a port
                      number - since the port number is to be ignored! */
    if(conn->bits.httpproxy) {
      /* we need to create new URL with the new port number */
      char *url;
      bool isftp = (bool)(strequal("ftp", conn->protostr) ||
                          strequal("ftps", conn->protostr));

      /*
       * This synthesized URL isn't always right--suffixes like ;type=A
       * are stripped off. It would be better to work directly from the
       * original URL and simply replace the port part of it.
       */
      url = aprintf("%s://%s%s%s:%d%s%s", conn->protostr,
             conn->bits.ipv6_ip?"[":"", conn->host.name,
             conn->bits.ipv6_ip?"]":"", conn->remote_port,
             isftp?"/":"", data->state.path);
      if(!url)
        return CURLE_OUT_OF_MEMORY;

      if(data->change.url_alloc)
        free(data->change.url);

      data->change.url = url;
      data->change.url_alloc = TRUE;
    }
  }
  else if(portptr) {
    /* no CURLOPT_PORT given, extract the one from the URL */

    char *rest;
    unsigned long port;

    port=strtoul(portptr+1, &rest, 10);  /* Port number must be decimal */

    if(rest != (portptr+1) && *rest == '\0') {
      /* The colon really did have only digits after it,
       * so it is either a port number or a mistake */

      if(port > 0xffff) {   /* Single unix standard says port numbers are
                              * 16 bits long */
        failf(data, "Port number too large: %lu", port);
        return CURLE_URL_MALFORMAT;
      }

      *portptr = '\0'; /* cut off the name there */
      conn->remote_port = (unsigned short)port;
    }
  }
  return CURLE_OK;
}

/*
 * Override a user name and password from the URL with that in the
 * CURLOPT_USERPWD option or a .netrc file, if applicable.
 */
static void override_userpass(struct SessionHandle *data,
                              struct connectdata *conn,
                              char *user, char *passwd)
{
  if(data->set.str[STRING_USERPWD] != NULL) {
    /* the name is given, get user+password */
    sscanf(data->set.str[STRING_USERPWD],
           "%" MAX_CURL_USER_LENGTH_TXT "[^:]:"
           "%" MAX_CURL_PASSWORD_LENGTH_TXT "[^\n]",
           user, passwd);
  }

  conn->bits.netrc = FALSE;
  if(data->set.use_netrc != CURL_NETRC_IGNORED) {
    if(Curl_parsenetrc(conn->host.name,
                       user, passwd,
                       data->set.str[STRING_NETRC_FILE])) {
      infof(data, "Couldn't find host %s in the "
            DOT_CHAR "netrc file; using defaults\n",
            conn->host.name);
    }
    else {
      /* set bits.netrc TRUE to remember that we got the name from a .netrc
         file, so that it is safe to use even if we followed a Location: to a
         different host or similar. */
      conn->bits.netrc = TRUE;

      conn->bits.user_passwd = 1; /* enable user+password */
    }
  }
}

/*
 * Set password so it's available in the connection.
 */
static CURLcode set_userpass(struct connectdata *conn,
                             const char *user, const char *passwd)
{
  /* If our protocol needs a password and we have none, use the defaults */
  if( (conn->protocol & PROT_FTP) &&
       !conn->bits.user_passwd) {

    conn->user = strdup(CURL_DEFAULT_USER);
    conn->passwd = strdup(CURL_DEFAULT_PASSWORD);
    /* This is the default password, so DON'T set conn->bits.user_passwd */
  }
  else {
    /* store user + password, zero-length if not set */
    conn->user = strdup(user);
    conn->passwd = strdup(passwd);
  }
  if(!conn->user || !conn->passwd)
    return CURLE_OUT_OF_MEMORY;

  return CURLE_OK;
}

/*************************************************************
 * Resolve the address of the server or proxy
 *************************************************************/
static CURLcode resolve_server(struct SessionHandle *data,
                               struct connectdata *conn,
                               struct Curl_dns_entry **addr,
                               bool *async)
{
  CURLcode result=CURLE_OK;
  long shortest = 0; /* default to no timeout */

  /*************************************************************
   * Set timeout if that is being used
   *************************************************************/
  if(data->set.timeout || data->set.connecttimeout) {

    /* We set the timeout on the name resolving phase first, separately from
     * the download/upload part to allow a maximum time on everything. This is
     * a signal-based timeout, why it won't work and shouldn't be used in
     * multi-threaded environments. */

    shortest = data->set.timeout; /* default to this timeout value */
    if(shortest && data->set.connecttimeout &&
       (data->set.connecttimeout < shortest))
      /* if both are set, pick the shortest */
      shortest = data->set.connecttimeout;
    else if(!shortest)
      /* if timeout is not set, use the connect timeout */
      shortest = data->set.connecttimeout;
  /* We can expect the conn->created time to be "now", as that was just
     recently set in the beginning of this function and nothing slow
     has been done since then until now. */
  }

  /*************************************************************
   * Resolve the name of the server or proxy
   *************************************************************/
  if(conn->bits.reuse) {
    /* re-used connection, no resolving is necessary */
    *addr = NULL;
    /* we'll need to clear conn->dns_entry later in Curl_disconnect() */

    if(conn->bits.proxy)
      fix_hostname(data, conn, &conn->host);
  }
  else {
    /* this is a fresh connect */
    int rc;
    struct Curl_dns_entry *hostaddr;

    /* set a pointer to the hostname we display */
    fix_hostname(data, conn, &conn->host);

    if(!conn->proxy.name || !*conn->proxy.name) {
      /* If not connecting via a proxy, extract the port from the URL, if it is
       * there, thus overriding any defaults that might have been set above. */
      conn->port =  conn->remote_port; /* it is the same port */

      /* Resolve target host right on */
      rc = Curl_resolv_timeout(conn, conn->host.name, (int)conn->port,
                               &hostaddr, shortest);
      if(rc == CURLRESOLV_PENDING)
        *async = TRUE;

      else if (rc == CURLRESOLV_TIMEDOUT)
        result = CURLE_OPERATION_TIMEDOUT;

      else if(!hostaddr) {
        failf(data, "Couldn't resolve host '%s'", conn->host.dispname);
        result =  CURLE_COULDNT_RESOLVE_HOST;
        /* don't return yet, we need to clean up the timeout first */
      }
    }
    else {
      /* This is a proxy that hasn't been resolved yet. */

      /* IDN-fix the proxy name */
      fix_hostname(data, conn, &conn->proxy);

      /* resolve proxy */
      rc = Curl_resolv_timeout(conn, conn->proxy.name, (int)conn->port,
                               &hostaddr, shortest);

      if(rc == CURLRESOLV_PENDING)
        *async = TRUE;

      else if (rc == CURLRESOLV_TIMEDOUT)
        result = CURLE_OPERATION_TIMEDOUT;

      else if(!hostaddr) {
        failf(data, "Couldn't resolve proxy '%s'", conn->proxy.dispname);
        result = CURLE_COULDNT_RESOLVE_PROXY;
        /* don't return yet, we need to clean up the timeout first */
      }
    }
    *addr = hostaddr;
  }

  return result;
}

/*
 * Cleanup the connection just allocated before we can move along and use the
 * previously existing one.  All relevant data is copied over and old_conn is
 * ready for freeing once this function returns.
 */
static void reuse_conn(struct connectdata *old_conn,
                       struct connectdata *conn)
{
  if(old_conn->proxy.rawalloc)
    free(old_conn->proxy.rawalloc);

  /* free the SSL config struct from this connection struct as this was
     allocated in vain and is targeted for destruction */
  Curl_free_ssl_config(&old_conn->ssl_config);

  conn->data = old_conn->data;

  /* get the user+password information from the old_conn struct since it may
   * be new for this request even when we re-use an existing connection */
  conn->bits.user_passwd = old_conn->bits.user_passwd;
  if(conn->bits.user_passwd) {
    /* use the new user name and password though */
    Curl_safefree(conn->user);
    Curl_safefree(conn->passwd);
    conn->user = old_conn->user;
    conn->passwd = old_conn->passwd;
    old_conn->user = NULL;
    old_conn->passwd = NULL;
  }

  conn->bits.proxy_user_passwd = old_conn->bits.proxy_user_passwd;
  if(conn->bits.proxy_user_passwd) {
    /* use the new proxy user name and proxy password though */
    Curl_safefree(conn->proxyuser);
    Curl_safefree(conn->proxypasswd);
    conn->proxyuser = old_conn->proxyuser;
    conn->proxypasswd = old_conn->proxypasswd;
    old_conn->proxyuser = NULL;
    old_conn->proxypasswd = NULL;
  }

  /* host can change, when doing keepalive with a proxy ! */
  if(conn->bits.proxy) {
    free(conn->host.rawalloc);
    conn->host=old_conn->host;
  }
  else
    free(old_conn->host.rawalloc); /* free the newly allocated name buffer */

  /* re-use init */
  conn->bits.reuse = TRUE; /* yes, we're re-using here */

  Curl_safefree(old_conn->user);
  Curl_safefree(old_conn->passwd);
  Curl_safefree(old_conn->proxyuser);
  Curl_safefree(old_conn->proxypasswd);
  Curl_llist_destroy(old_conn->send_pipe, NULL);
  Curl_llist_destroy(old_conn->recv_pipe, NULL);
  Curl_llist_destroy(old_conn->pend_pipe, NULL);
  Curl_safefree(old_conn->master_buffer);
}

/**
 * create_conn() sets up a new connectdata struct, or re-uses an already
 * existing one, and resolves host name.
 *
 * if this function returns CURLE_OK and *async is set to TRUE, the resolve
 * response will be coming asynchronously. If *async is FALSE, the name is
 * already resolved.
 *
 * @param data The sessionhandle pointer
 * @param in_connect is set to the next connection data pointer
 * @param addr is set to the new dns entry for this connection. If this
 *        connection is re-used it will be NULL.
 * @param async is set TRUE/FALSE depending on the nature of this lookup
 * @return CURLcode
 * @see setup_conn()
 *
 * *NOTE* this function assigns the conn->data pointer!
 */

static CURLcode create_conn(struct SessionHandle *data,
                            struct connectdata **in_connect,
                            struct Curl_dns_entry **addr,
                            bool *async)
{
  CURLcode result=CURLE_OK;
  struct connectdata *conn;
  struct connectdata *conn_temp = NULL;
  size_t urllen;
  char user[MAX_CURL_USER_LENGTH];
  char passwd[MAX_CURL_PASSWORD_LENGTH];
  bool reuse;
  char *proxy = NULL;

  *addr = NULL; /* nothing yet */
  *async = FALSE;

  /*************************************************************
   * Check input data
   *************************************************************/

  if(!data->change.url)
    return CURLE_URL_MALFORMAT;

  /* First, split up the current URL in parts so that we can use the
     parts for checking against the already present connections. In order
     to not have to modify everything at once, we allocate a temporary
     connection data struct and fill in for comparison purposes. */

  conn = allocate_conn();

  /* We must set the return variable as soon as possible, so that our
     parent can cleanup any possible allocs we may have done before
     any failure */
  *in_connect = conn;

  if(!conn)
    return CURLE_OUT_OF_MEMORY;

  conn->data = data; /* Setup the association between this connection
                        and the SessionHandle */

  conn->proxytype = data->set.proxytype; /* type */
  conn->bits.proxy = (bool)(data->set.str[STRING_PROXY] &&
                            *data->set.str[STRING_PROXY]);
  conn->bits.httpproxy = (bool)(conn->bits.proxy
                                && (conn->proxytype == CURLPROXY_HTTP));


  conn->bits.user_passwd = (bool)(NULL != data->set.str[STRING_USERPWD]);
  conn->bits.proxy_user_passwd = (bool)(NULL != data->set.str[STRING_PROXYUSERPWD]);
  conn->bits.tunnel_proxy = data->set.tunnel_thru_httpproxy;
  conn->bits.ftp_use_epsv = data->set.ftp_use_epsv;
  conn->bits.ftp_use_eprt = data->set.ftp_use_eprt;

  if(data->multi && Curl_multi_canPipeline(data->multi) &&
      !conn->master_buffer) {
    /* Allocate master_buffer to be used for pipelining */
    conn->master_buffer = calloc(BUFSIZE, sizeof (char));
    if(!conn->master_buffer)
      return CURLE_OUT_OF_MEMORY;
  }

  /* Initialize the pipeline lists */
  conn->send_pipe = Curl_llist_alloc((curl_llist_dtor) llist_dtor);