url.c

 *   ;options
 *   ;options:password
 *
 * Parameters:
 *
 * login    [in]     - The login string.
 * len      [in]     - The length of the login string.
 * userp    [in/out] - The address where a pointer to newly allocated memory
 *                     holding the user will be stored upon completion.
 * passdwp  [in/out] - The address where a pointer to newly allocated memory
 *                     holding the password will be stored upon completion.
 * optionsp [in/out] - The address where a pointer to newly allocated memory
 *                     holding the options will be stored upon completion.
 *
 * Returns CURLE_OK on success.
 */
static CURLcode parse_login_details(const char *login, const size_t len,
                                    char **userp, char **passwdp,
                                    char **optionsp)
{
  CURLcode result = CURLE_OK;
  char *ubuf = NULL;
  char *pbuf = NULL;
  char *obuf = NULL;
  const char *psep = NULL;
  const char *osep = NULL;
  size_t ulen;
  size_t plen;
  size_t olen;

  /* Attempt to find the password separator */
  if(passwdp) {
    psep = strchr(login, ':');

    /* Within the constraint of the login string */
    if(psep >= login + len)
      psep = NULL;
  }

  /* Attempt to find the options separator */
  if(optionsp) {
    osep = strchr(login, ';');

    /* Within the constraint of the login string */
    if(osep >= login + len)
      osep = NULL;
  }

  /* Calculate the portion lengths */
  ulen = (psep ?
          (size_t)(osep && psep > osep ? osep - login : psep - login) :
          (osep ? (size_t)(osep - login) : len));
  plen = (psep ?
          (osep && osep > psep ? (size_t)(osep - psep) :
                                 (size_t)(login + len - psep)) - 1 : 0);
  olen = (osep ?
          (psep && psep > osep ? (size_t)(psep - osep) :
                                 (size_t)(login + len - osep)) - 1 : 0);

  /* Allocate the user portion buffer */
  if(userp && ulen) {
    ubuf = malloc(ulen + 1);
    if(!ubuf)
      result = CURLE_OUT_OF_MEMORY;
  }

  /* Allocate the password portion buffer */
  if(!result && passwdp && plen) {
    pbuf = malloc(plen + 1);
    if(!pbuf) {
      free(ubuf);
      result = CURLE_OUT_OF_MEMORY;
    }
  }

  /* Allocate the options portion buffer */
  if(!result && optionsp && olen) {
    obuf = malloc(olen + 1);
    if(!obuf) {
      free(pbuf);
      free(ubuf);
      result = CURLE_OUT_OF_MEMORY;
    }
  }

  if(!result) {
    /* Store the user portion if necessary */
    if(ubuf) {
      memcpy(ubuf, login, ulen);
      ubuf[ulen] = '\0';
      Curl_safefree(*userp);
      *userp = ubuf;
    }

    /* Store the password portion if necessary */
    if(pbuf) {
      memcpy(pbuf, psep + 1, plen);
      pbuf[plen] = '\0';
      Curl_safefree(*passwdp);
      *passwdp = pbuf;
    }

    /* Store the options portion if necessary */
    if(obuf) {
      memcpy(obuf, osep + 1, olen);
      obuf[olen] = '\0';
      Curl_safefree(*optionsp);
      *optionsp = obuf;
    }
  }

  return result;
}

/*************************************************************
 * 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 Curl_easy *data,
                                  struct connectdata *conn)
{
  char *portptr;
  char endbracket;

  /* Note that at this point, the IPv6 address cannot contain any scope
     suffix as that has already been removed in the parseurlandfillconn()
     function */
  if((1 == sscanf(conn->host.name, "[%*45[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, ']');
    if(portptr) {
      *portptr++ = '\0'; /* zero terminate, killing the bracket */
      if(':' != *portptr)
        portptr = NULL; /* no port number available */
    }
  }
  else {
#ifdef ENABLE_IPV6
    struct in6_addr in6;
    if(Curl_inet_pton(AF_INET6, conn->host.name, &in6) > 0) {
      /* This is a numerical IPv6 address, meaning this is a wrongly formatted
         URL */
      failf(data, "IPv6 numerical address used in URL without brackets");
      return CURLE_URL_MALFORMAT;
    }
#endif

    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;
      char type[12]="";

      if(conn->bits.type_set)
        snprintf(type, sizeof(type), ";type=%c",
                 data->set.prefer_ascii?'A':
                 (data->set.ftp_list_only?'D':'I'));

      /*
       * 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:%hu%s%s%s", conn->given->scheme,
                    conn->bits.ipv6_ip?"[":"", conn->host.name,
                    conn->bits.ipv6_ip?"]":"", conn->remote_port,
                    data->state.slash_removed?"/":"", data->state.path,
                    type);
      if(!url)
        return CURLE_OUT_OF_MEMORY;

      if(data->change.url_alloc) {
        Curl_safefree(data->change.url);
        data->change.url_alloc = FALSE;
      }

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

    char *rest;
    long port;

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

    if((port < 0) || (port > 0xffff)) {
      /* Single unix standard says port numbers are 16 bits long */
      failf(data, "Port number out of range");
      return CURLE_URL_MALFORMAT;
    }

    else if(rest != &portptr[1]) {
      *portptr = '\0'; /* cut off the name there */
      conn->remote_port = curlx_ultous(port);
    }
    else
      /* Browser behavior adaptation. If there's a colon with no digits after,
         just cut off the name there which makes us ignore the colon and just
         use the default port. Firefox and Chrome both do that. */
      *portptr = '\0';
  }

  /* only if remote_port was not already parsed off the URL we use the
     default port number */
  if(conn->remote_port < 0)
    conn->remote_port = (unsigned short)conn->given->defport;

  return CURLE_OK;
}

/*
 * Override the login details from the URL with that in the CURLOPT_USERPWD
 * option or a .netrc file, if applicable.
 */
static CURLcode override_login(struct Curl_easy *data,
                               struct connectdata *conn,
                               char **userp, char **passwdp, char **optionsp)
{
  if(data->set.str[STRING_USERNAME]) {
    free(*userp);
    *userp = strdup(data->set.str[STRING_USERNAME]);
    if(!*userp)
      return CURLE_OUT_OF_MEMORY;
  }

  if(data->set.str[STRING_PASSWORD]) {
    free(*passwdp);
    *passwdp = strdup(data->set.str[STRING_PASSWORD]);
    if(!*passwdp)
      return CURLE_OUT_OF_MEMORY;
  }

  if(data->set.str[STRING_OPTIONS]) {
    free(*optionsp);
    *optionsp = strdup(data->set.str[STRING_OPTIONS]);
    if(!*optionsp)
      return CURLE_OUT_OF_MEMORY;
  }

  conn->bits.netrc = FALSE;
  if(data->set.use_netrc != CURL_NETRC_IGNORED) {
    int ret = Curl_parsenetrc(conn->host.name,
                              userp, passwdp,
                              data->set.str[STRING_NETRC_FILE]);
    if(ret > 0) {
      infof(data, "Couldn't find host %s in the "
            DOT_CHAR "netrc file; using defaults\n",
            conn->host.name);
    }
    else if(ret < 0) {
      return CURLE_OUT_OF_MEMORY;
    }
    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 = TRUE; /* enable user+password */
    }
  }

  return CURLE_OK;
}

/*
 * Set the login details so they're available in the connection
 */
static CURLcode set_login(struct connectdata *conn,
                          const char *user, const char *passwd,
                          const char *options)
{
  CURLcode result = CURLE_OK;

  /* If our protocol needs a password and we have none, use the defaults */
  if((conn->handler->flags & PROTOPT_NEEDSPWD) && !conn->bits.user_passwd) {
    /* Store the default user */
    conn->user = strdup(CURL_DEFAULT_USER);

    /* Store the default password */
    if(conn->user)
      conn->passwd = strdup(CURL_DEFAULT_PASSWORD);
    else
      conn->passwd = NULL;

    /* This is the default password, so DON'T set conn->bits.user_passwd */
  }
  else {
    /* Store the user, zero-length if not set */
    conn->user = strdup(user);

    /* Store the password (only if user is present), zero-length if not set */
    if(conn->user)
      conn->passwd = strdup(passwd);
    else
      conn->passwd = NULL;
  }

  if(!conn->user || !conn->passwd)
    result = CURLE_OUT_OF_MEMORY;

  /* Store the options, null if not set */
  if(!result && options[0]) {
    conn->options = strdup(options);

    if(!conn->options)
      result = CURLE_OUT_OF_MEMORY;
  }

  return result;
}

/*
 * Parses a "host:port" string to connect to.
 * The hostname and the port may be empty; in this case, NULL is returned for
 * the hostname and -1 for the port.
 */
static CURLcode parse_connect_to_host_port(struct Curl_easy *data,
                                           const char *host,
                                           char **hostname_result,
                                           int *port_result)
{
  char *host_dup;
  char *hostptr;
  char *host_portno;
  char *portptr;
  int port = -1;

  *hostname_result = NULL;
  *port_result = -1;

  if(!host || !*host)
    return CURLE_OK;

  host_dup = strdup(host);
  if(!host_dup)
    return CURLE_OUT_OF_MEMORY;

  hostptr = host_dup;

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

  /* detect and extract RFC6874-style IPv6-addresses */
  if(*hostptr == '[') {
    char *ptr = ++hostptr; /* advance beyond the initial bracket */
    while(*ptr && (ISXDIGIT(*ptr) || (*ptr == ':') || (*ptr == '.')))
      ptr++;
    if(*ptr == '%') {
      /* There might be a zone identifier */
      if(strncmp("%25", ptr, 3))
        infof(data, "Please URL encode %% as %%25, see RFC 6874.\n");
      ptr++;
      /* Allow unreserved characters as defined in RFC 3986 */
      while(*ptr && (ISALPHA(*ptr) || ISXDIGIT(*ptr) || (*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
     * hostptr 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 server.com:1080 */
  host_portno = strchr(portptr, ':');
  if(host_portno) {
    char *endp = NULL;
    *host_portno = '\0'; /* cut off number from host name */
    host_portno++;
    if(*host_portno) {
      long portparse = strtol(host_portno, &endp, 10);
      if((endp && *endp) || (portparse < 0) || (portparse > 65535)) {
        infof(data, "No valid port number in connect to host string (%s)\n",
              host_portno);
        hostptr = NULL;
        port = -1;
      }
      else
        port = (int)portparse; /* we know it will fit */
    }
  }

  /* now, clone the cleaned host name */
  if(hostptr) {
    *hostname_result = strdup(hostptr);
    if(!*hostname_result) {
      free(host_dup);
      return CURLE_OUT_OF_MEMORY;
    }
  }

  *port_result = port;

  free(host_dup);
  return CURLE_OK;
}

/*
 * Parses one "connect to" string in the form:
 * "HOST:PORT:CONNECT-TO-HOST:CONNECT-TO-PORT".
 */
static CURLcode parse_connect_to_string(struct Curl_easy *data,
                                        struct connectdata *conn,
                                        const char *conn_to_host,
                                        char **host_result,
                                        int *port_result)
{
  CURLcode result = CURLE_OK;
  const char *ptr = conn_to_host;
  int host_match = FALSE;
  int port_match = FALSE;

  if(*ptr == ':') {
    /* an empty hostname always matches */
    host_match = TRUE;
    ptr++;
  }
  else {
    /* check whether the URL's hostname matches */
    size_t hostname_to_match_len;
    char *hostname_to_match = aprintf("%s%s%s",
                                      conn->bits.ipv6_ip ? "[" : "",
                                      conn->host.name,
                                      conn->bits.ipv6_ip ? "]" : "");
    if(!hostname_to_match)
      return CURLE_OUT_OF_MEMORY;
    hostname_to_match_len = strlen(hostname_to_match);
    host_match = curl_strnequal(ptr, hostname_to_match, hostname_to_match_len);
    free(hostname_to_match);
    ptr += hostname_to_match_len;

    host_match = host_match && *ptr == ':';
    ptr++;
  }

  if(host_match) {
    if(*ptr == ':') {
      /* an empty port always matches */
      port_match = TRUE;
      ptr++;
    }
    else {
      /* check whether the URL's port matches */
      char *ptr_next = strchr(ptr, ':');
      if(ptr_next) {
        char *endp = NULL;
        long port_to_match = strtol(ptr, &endp, 10);
        if((endp == ptr_next) && (port_to_match == conn->remote_port)) {
          port_match = TRUE;
          ptr = ptr_next + 1;
        }
      }
    }
  }

  if(host_match && port_match) {
    /* parse the hostname and port to connect to */
    result = parse_connect_to_host_port(data, ptr, host_result, port_result);
  }

  return result;
}

/*
 * Processes all strings in the "connect to" slist, and uses the "connect
 * to host" and "connect to port" of the first string that matches.
 */
static CURLcode parse_connect_to_slist(struct Curl_easy *data,
                                       struct connectdata *conn,
                                       struct curl_slist *conn_to_host)
{
  CURLcode result = CURLE_OK;
  char *host = NULL;
  int port = 0;

  while(conn_to_host && !host) {
    result = parse_connect_to_string(data, conn, conn_to_host->data,
                                     &host, &port);
    if(result)
      return result;

    if(host && *host) {
      bool ipv6host;
      conn->conn_to_host.rawalloc = host;
      conn->conn_to_host.name = host;
      conn->bits.conn_to_host = TRUE;

      ipv6host = strchr(host, ':') != NULL;
      infof(data, "Connecting to hostname: %s%s%s\n",
            ipv6host ? "[" : "", host, ipv6host ? "]" : "");
    }
    else {
      /* no "connect to host" */
      conn->bits.conn_to_host = FALSE;
      free(host);
    }

    if(port >= 0) {
      conn->conn_to_port = port;
      conn->bits.conn_to_port = TRUE;
      infof(data, "Connecting to port: %d\n", port);
    }
    else {
      /* no "connect to port" */
      conn->bits.conn_to_port = FALSE;
    }

    conn_to_host = conn_to_host->next;
  }

  return result;
}

/*************************************************************
 * Resolve the address of the server or proxy
 *************************************************************/
static CURLcode resolve_server(struct Curl_easy *data,
                               struct connectdata *conn,
                               bool *async)
{
  CURLcode result=CURLE_OK;
  long timeout_ms = Curl_timeleft(data, NULL, TRUE);

  /*************************************************************
   * Resolve the name of the server or proxy
   *************************************************************/
  if(conn->bits.reuse)
    /* We're reusing the connection - no need to resolve anything, and
       fix_hostname() was called already in create_conn() for the re-use
       case. */
    *async = FALSE;

  else {
    /* this is a fresh connect */
    int rc;
    struct Curl_dns_entry *hostaddr;

#ifdef USE_UNIX_SOCKETS
    if(data->set.str[STRING_UNIX_SOCKET_PATH]) {
      /* Unix domain sockets are local. The host gets ignored, just use the
       * specified domain socket address. Do not cache "DNS entries". There is
       * no DNS involved and we already have the filesystem path available */
      const char *path = data->set.str[STRING_UNIX_SOCKET_PATH];

      hostaddr = calloc(1, sizeof(struct Curl_dns_entry));
      if(!hostaddr)
        result = CURLE_OUT_OF_MEMORY;
      else if((hostaddr->addr = Curl_unix2addr(path)) != NULL)
        hostaddr->inuse++;
      else {
        /* Long paths are not supported for now */
        if(strlen(path) >= sizeof(((struct sockaddr_un *)0)->sun_path)) {
          failf(data, "Unix socket path too long: '%s'", path);
          result = CURLE_COULDNT_RESOLVE_HOST;
        }
        else
          result = CURLE_OUT_OF_MEMORY;
        free(hostaddr);
        hostaddr = NULL;
      }
    }
    else
#endif
    if(!conn->proxy.name || !*conn->proxy.name) {
      struct hostname *connhost;
      if(conn->bits.conn_to_host)
        connhost = &conn->conn_to_host;
      else
        connhost = &conn->host;

      /* 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. */
      if(conn->bits.conn_to_port)
        conn->port = conn->conn_to_port;
      else
        conn->port = conn->remote_port; /* it is the same port */

      /* Resolve target host right on */
      rc = Curl_resolv_timeout(conn, connhost->name, (int)conn->port,
                               &hostaddr, timeout_ms);
      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'", connhost->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. */

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

      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 */
      }
    }
    DEBUGASSERT(conn->dns_entry == NULL);
    conn->dns_entry = 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)
{
  free_fixed_hostname(&old_conn->proxy);
  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 or if the case is
     different this time etc */
  free_fixed_hostname(&conn->host);
  free_fixed_hostname(&conn->conn_to_host);
  Curl_safefree(conn->host.rawalloc);
  Curl_safefree(conn->conn_to_host.rawalloc);
  conn->host=old_conn->host;
  conn->bits.conn_to_host = old_conn->bits.conn_to_host;
  conn->conn_to_host = old_conn->conn_to_host;
  conn->bits.conn_to_port = old_conn->bits.conn_to_port;
  conn->conn_to_port = old_conn->conn_to_port;

  /* persist connection info in session handle */
  Curl_persistconninfo(conn);

  conn_reset_all_postponed_data(old_conn); /* free buffers */
  conn_reset_all_postponed_data(conn);     /* reset unprocessed data */

  /* 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_safefree(old_conn->localdev);

  Curl_llist_destroy(old_conn->send_pipe, NULL);
  Curl_llist_destroy(old_conn->recv_pipe, NULL);

  old_conn->send_pipe = NULL;
  old_conn->recv_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 async is set TRUE when an async DNS resolution is pending
 * @see Curl_setup_conn()
 *
 * *NOTE* this function assigns the conn->data pointer!
 */

static CURLcode create_conn(struct Curl_easy *data,
                            struct connectdata **in_connect,
                            bool *async)
{
  CURLcode result = CURLE_OK;
  struct connectdata *conn;
  struct connectdata *conn_temp = NULL;
  size_t urllen;
  char *user = NULL;
  char *passwd = NULL;
  char *options = NULL;
  bool reuse;
  char *proxy = NULL;
  bool prot_missing = FALSE;
  bool connections_available = TRUE;
  bool force_reuse = FALSE;
  bool waitpipe = FALSE;
  size_t max_host_connections = Curl_multi_max_host_connections(data->multi);
  size_t max_total_connections = Curl_multi_max_total_connections(data->multi);

  *async = FALSE;

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

  if(!data->change.url) {
    result = CURLE_URL_MALFORMAT;
    goto out;
  }

  /* 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(data);

  if(!conn) {
    result = CURLE_OUT_OF_MEMORY;
    goto out;
  }

  /* 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;

  /* This initing continues below, see the comment "Continue connectdata
   * initialization here" */

  /***********************************************************
   * We need to allocate memory to store the path in. We get the size of the
   * full URL to be sure, and we need to make it at least 256 bytes since
   * other parts of the code will rely on this fact
   ***********************************************************/
#define LEAST_PATH_ALLOC 256
  urllen=strlen(data->change.url);
  if(urllen < LEAST_PATH_ALLOC)
    urllen=LEAST_PATH_ALLOC;

  /*
   * We malloc() the buffers below urllen+2 to make room for 2 possibilities:
   * 1 - an extra terminating zero
   * 2 - an extra slash (in case a syntax like "www.host.com?moo" is used)
   */

  Curl_safefree(data->state.pathbuffer);
  data->state.path = NULL;

  data->state.pathbuffer = malloc(urllen+2);
  if(NULL == data->state.pathbuffer) {
    result = CURLE_OUT_OF_MEMORY; /* really bad error */
    goto out;
  }
  data->state.path = data->state.pathbuffer;

  conn->host.rawalloc = malloc(urllen+2);
  if(NULL == conn->host.rawalloc) {
    Curl_safefree(data->state.pathbuffer);
    data->state.path = NULL;
    result = CURLE_OUT_OF_MEMORY;
    goto out;
  }

  conn->host.name = conn->host.rawalloc;
  conn->host.name[0] = 0;

  user = strdup("");
  passwd = strdup("");
  options = strdup("");
  if(!user || !passwd || !options) {
    result = CURLE_OUT_OF_MEMORY;
    goto out;
  }

  result = parseurlandfillconn(data, conn, &prot_missing, &user, &passwd,
                               &options);
  if(result)
    goto out;

  /*************************************************************
   * No protocol part in URL was used, add it!
   *************************************************************/
  if(prot_missing) {
    /* We're guessing prefixes here and if we're told to use a proxy or if
       we're gonna follow a Location: later or... then we need the protocol
       part added so that we have a valid URL. */
    char *reurl;
    char *ch_lower;

    reurl = aprintf("%s://%s", conn->handler->scheme, data->change.url);

    if(!reurl) {
      result = CURLE_OUT_OF_MEMORY;
      goto out;
    }

    /* Change protocol prefix to lower-case */
    for(ch_lower = reurl; *ch_lower != ':'; ch_lower++)
      *ch_lower = (char)TOLOWER(*ch_lower);

    if(data->change.url_alloc) {
      Curl_safefree(data->change.url);
      data->change.url_alloc = FALSE;
    }

    data->change.url = reurl;
    data->change.url_alloc = TRUE; /* free this later */
  }

  /*************************************************************
   * If the protocol can't handle url query strings, then cut
   * off the unhandable part
   *************************************************************/
  if((conn->given->flags&PROTOPT_NOURLQUERY)) {
    char *path_q_sep = strchr(conn->data->state.path, '?');
    if(path_q_sep) {
      /* according to rfc3986, allow the query (?foo=bar)
         also on protocols that can't handle it.

         cut the string-part after '?'
      */

      /* terminate the string */
      path_q_sep[0] = 0;
    }
  }

  if(data->set.str[STRING_BEARER]) {
    conn->oauth_bearer = strdup(data->set.str[STRING_BEARER]);
    if(!conn->oauth_bearer) {
      result = CURLE_OUT_OF_MEMORY;
      goto out;
    }
  }

#ifndef CURL_DISABLE_PROXY
  /*************************************************************
   * Extract the user and password from the authentication string
   *************************************************************/
  if(conn->bits.proxy_user_passwd) {
    result = parse_proxy_auth(data, conn);
    if(result)
      goto out;
  }

  /*************************************************************
   * Detect what (if any) proxy to use
   *************************************************************/
  if(data->set.str[STRING_PROXY]) {
    proxy = strdup(data->set.str[STRING_PROXY]);
    /* if global proxy is set, this is it */
    if(NULL == proxy) {
      failf(data, "memory shortage");
      result = CURLE_OUT_OF_MEMORY;
      goto out;
    }
  }

  if(data->set.str[STRING_NOPROXY] &&
     check_noproxy(conn->host.name, data->set.str[STRING_NOPROXY])) {
    free(proxy);  /* proxy is in exception list */
    proxy = NULL;
  }
  else if(!proxy)
    proxy = detect_proxy(conn);

#ifdef USE_UNIX_SOCKETS
  if(proxy && data->set.str[STRING_UNIX_SOCKET_PATH]) {
    free(proxy);  /* Unix domain sockets cannot be proxied, so disable it */
    proxy = NULL;
  }
#endif

  if(proxy && (!*proxy || (conn->handler->flags & PROTOPT_NONETWORK))) {
    free(proxy);  /* Don't bother with an empty proxy string or if the
                     protocol doesn't work with network */
    proxy = NULL;
  }

  /***********************************************************************
   * If this is supposed to use a proxy, we need to figure out the proxy host
   * name, proxy type and port number, so that we can re-use an existing
   * connection that may exist registered to the same proxy host.
   ***********************************************************************/
  if(proxy) {
    result = parse_proxy(data, conn, proxy);

    free(proxy); /* parse_proxy copies the proxy string */
    proxy = NULL;

    if(result)
      goto out;

    if((conn->proxytype == CURLPROXY_HTTP) ||
       (conn->proxytype == CURLPROXY_HTTP_1_0)) {
#ifdef CURL_DISABLE_HTTP
      /* asking for a HTTP proxy is a bit funny when HTTP is disabled... */
      result = CURLE_UNSUPPORTED_PROTOCOL;
      goto out;
#else
      /* force this connection's protocol to become HTTP if not already
         compatible - if it isn't tunneling through */
      if(!(conn->handler->protocol & PROTO_FAMILY_HTTP) &&
         !conn->bits.tunnel_proxy)
        conn->handler = &Curl_handler_http;

      conn->bits.httpproxy = TRUE;
#endif
    }
    else {
      conn->bits.httpproxy = FALSE; /* not a HTTP proxy */
      conn->bits.tunnel_proxy = FALSE; /* no tunneling if not HTTP */
    }
    conn->bits.proxy = TRUE;
  }
  else {
    /* we aren't using the proxy after all... */
    conn->bits.proxy = FALSE;
    conn->bits.httpproxy = FALSE;
    conn->bits.proxy_user_passwd = FALSE;
    conn->bits.tunnel_proxy = FALSE;
  }

#endif /* CURL_DISABLE_PROXY */

  /*************************************************************
   * If the protocol is using SSL and HTTP proxy is used, we set
   * the tunnel_proxy bit.
   *************************************************************/
  if((conn->given->flags&PROTOPT_SSL) && conn->bits.httpproxy)
    conn->bits.tunnel_proxy = TRUE;

  /*************************************************************
   * Figure out the remote port number and fix it in the URL
   *************************************************************/