url.c

/*
 * verboseconnect() displays verbose information after a connect
 */
#ifndef CURL_DISABLE_VERBOSE_STRINGS
static void verboseconnect(struct connectdata *conn)
{
  infof(conn->data, "Connected to %s (%s) port %d (#%d)\n",
        conn->bits.proxy ? conn->proxy.dispname : conn->host.dispname,
        conn->ip_addr_str, conn->port, conn->connectindex);
}
#endif

int Curl_protocol_getsock(struct connectdata *conn,
                          curl_socket_t *socks,
                          int numsocks)
{
  if(conn->handler->proto_getsock)
    return conn->handler->proto_getsock(conn, socks, numsocks);
  return GETSOCK_BLANK;
}

int Curl_doing_getsock(struct connectdata *conn,
                       curl_socket_t *socks,
                       int numsocks)
{
  if(conn && conn->handler->doing_getsock)
    return conn->handler->doing_getsock(conn, socks, numsocks);
  return GETSOCK_BLANK;
}

/*
 * We are doing protocol-specific connecting and this is being called over and
 * over from the multi interface until the connection phase is done on
 * protocol layer.
 */

CURLcode Curl_protocol_connecting(struct connectdata *conn,
                                  bool *done)
{
  CURLcode result=CURLE_OK;

  if(conn && conn->handler->connecting) {
    *done = FALSE;
    result = conn->handler->connecting(conn, done);
  }
  else
    *done = TRUE;

  return result;
}

/*
 * We are DOING this is being called over and over from the multi interface
 * until the DOING phase is done on protocol layer.
 */

CURLcode Curl_protocol_doing(struct connectdata *conn, bool *done)
{
  CURLcode result=CURLE_OK;

  if(conn && conn->handler->doing) {
    *done = FALSE;
    result = conn->handler->doing(conn, done);
  }
  else
    *done = TRUE;

  return result;
}

/*
 * We have discovered that the TCP connection has been successful, we can now
 * proceed with some action.
 *
 */
CURLcode Curl_protocol_connect(struct connectdata *conn,
                               bool *protocol_done)
{
  CURLcode result=CURLE_OK;
  struct SessionHandle *data = conn->data;

  *protocol_done = FALSE;

  if(conn->bits.tcpconnect && conn->bits.protoconnstart) {
    /* We already are connected, get back. This may happen when the connect
       worked fine in the first call, like when we connect to a local server
       or proxy. Note that we don't know if the protocol is actually done.

       Unless this protocol doesn't have any protocol-connect callback, as
       then we know we're done. */
    if(!conn->handler->connecting)
      *protocol_done = TRUE;

    return CURLE_OK;
  }

  if(!conn->bits.tcpconnect) {

    Curl_pgrsTime(data, TIMER_CONNECT); /* connect done */

    if(data->set.verbose)
      verboseconnect(conn);
  }

  if(!conn->bits.protoconnstart) {
    if(conn->handler->connect_it) {
      /* is there a protocol-specific connect() procedure? */

      /* Set start time here for timeout purposes in the connect procedure, it
         is later set again for the progress meter purpose */
      conn->now = Curl_tvnow();

      /* Call the protocol-specific connect function */
      result = conn->handler->connect_it(conn, protocol_done);
    }
    else
      *protocol_done = TRUE;

    /* it has started, possibly even completed but that knowledge isn't stored
       in this bit! */
    if(!result)
      conn->bits.protoconnstart = TRUE;
  }

  return result; /* pass back status */
}

/*
 * Helpers for IDNA convertions.
 */
#ifdef USE_LIBIDN
static bool is_ASCII_name(const char *hostname)
{
  const unsigned char *ch = (const unsigned char*)hostname;

  while(*ch) {
    if(*ch++ & 0x80)
      return FALSE;
  }
  return TRUE;
}

/*
 * Check if characters in hostname is allowed in Top Level Domain.
 */
static bool tld_check_name(struct SessionHandle *data,
                           const char *ace_hostname)
{
  size_t err_pos;
  char *uc_name = NULL;
  int rc;
#ifndef CURL_DISABLE_VERBOSE_STRINGS
  const char *tld_errmsg = "<no msg>";
#else
  (void)data;
#endif

  /* Convert (and downcase) ACE-name back into locale's character set */
  rc = idna_to_unicode_lzlz(ace_hostname, &uc_name, 0);
  if(rc != IDNA_SUCCESS)
    return (FALSE);

  rc = tld_check_lz(uc_name, &err_pos, NULL);
#ifndef CURL_DISABLE_VERBOSE_STRINGS
#ifdef HAVE_TLD_STRERROR
  if(rc != TLD_SUCCESS)
    tld_errmsg = tld_strerror((Tld_rc)rc);
#endif
  if(rc == TLD_INVALID)
    infof(data, "WARNING: %s; pos %u = `%c'/0x%02X\n",
          tld_errmsg, err_pos, uc_name[err_pos],
          uc_name[err_pos] & 255);
  else if(rc != TLD_SUCCESS)
    infof(data, "WARNING: TLD check for %s failed; %s\n",
          uc_name, tld_errmsg);
#endif /* CURL_DISABLE_VERBOSE_STRINGS */
  if(uc_name)
     idn_free(uc_name);
  return (bool)(rc == TLD_SUCCESS);
}
#endif

/*
 * Perform any necessary IDN conversion of hostname
 */
static void fix_hostname(struct SessionHandle *data,
                         struct connectdata *conn, struct hostname *host)
{
#ifndef USE_LIBIDN
  (void)data;
  (void)conn;
#elif defined(CURL_DISABLE_VERBOSE_STRINGS)
  (void)conn;
#endif

  /* set the name we use to display the host name */
  host->dispname = host->name;

#ifdef USE_LIBIDN
  /*************************************************************
   * Check name for non-ASCII and convert hostname to ACE form.
   *************************************************************/
  if(!is_ASCII_name(host->name) &&
      stringprep_check_version(LIBIDN_REQUIRED_VERSION)) {
    char *ace_hostname = NULL;
    int rc = idna_to_ascii_lz(host->name, &ace_hostname, 0);
    infof (data, "Input domain encoded as `%s'\n",
           stringprep_locale_charset ());
    if(rc != IDNA_SUCCESS)
      infof(data, "Failed to convert %s to ACE; %s\n",
            host->name, Curl_idn_strerror(conn,rc));
    else {
      /* tld_check_name() displays a warning if the host name contains
         "illegal" characters for this TLD */
      (void)tld_check_name(data, ace_hostname);

      host->encalloc = ace_hostname;
      /* change the name pointer to point to the encoded hostname */
      host->name = host->encalloc;
    }
  }
#endif
}

/*
 * Allocate and initialize a new connectdata object.
 */
static struct connectdata *allocate_conn(void)
{
  struct connectdata *conn;

  conn = calloc(1, sizeof(struct connectdata));
  if(!conn)
    return NULL;

  conn->handler = &Curl_handler_dummy;  /* Be sure we have a handler defined
                                           already from start to avoid NULL
                                           situations and checks */

  /* and we setup a few fields in case we end up actually using this struct */

  conn->sock[FIRSTSOCKET] = CURL_SOCKET_BAD;     /* no file descriptor */
  conn->sock[SECONDARYSOCKET] = CURL_SOCKET_BAD; /* no file descriptor */
  conn->connectindex = -1;    /* no index */
  conn->port = -1; /* unknown at this point */

  /* Default protocol-independent behavior doesn't support persistent
     connections, so we set this to force-close. Protocols that support
     this need to set this to FALSE in their "curl_do" functions. */
  conn->bits.close = TRUE;

  /* Store creation time to help future close decision making */
  conn->created = Curl_tvnow();

  return conn;
}

static CURLcode findprotocol(struct SessionHandle *data,
                             struct connectdata *conn,
                             const char *protostr)
{
  const struct Curl_handler * const *pp;
  const struct Curl_handler *p;

  /* Scan protocol handler table and match against 'protostr' to set a few
     variables based on the URL. Now that the handler may be changed later
     when the protocol specific setup function is called. */
  for (pp = protocols; (p = *pp) != NULL; pp++) {
    if(Curl_raw_equal(p->scheme, protostr)) {
      /* Protocol found in table. Check if allowed */
      if(!(data->set.allowed_protocols & p->protocol))
        /* nope, get out */
        break;

      /* it is allowed for "normal" request, now do an extra check if this is
         the result of a redirect */
      if(data->state.this_is_a_follow &&
         !(data->set.redir_protocols & p->protocol))
        /* nope, get out */
        break;

      /* Perform setup complement if some. */
      conn->handler = p;
      conn->protocol |= p->protocol;

      /* 'port' and 'remote_port' are set in setup_connection_internals() */
      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,
        protostr);

  return CURLE_UNSUPPORTED_PROTOCOL;
}

/*
 * Parse URL and fill in the relevant members of the connection struct.
 */
static CURLcode parseurlandfillconn(struct SessionHandle *data,
                                    struct connectdata *conn)
{
  char *at;
  char *tmp;
  char *path = data->state.path;
  int rc;
  char protobuf[16];
  const char *protop;

  /*************************************************************
   * Parse the URL.
   *
   * We need to parse the url even when using the proxy, because we will need
   * the hostname and port in case we are trying to SSL connect through the
   * proxy -- and we don't know if we will need to use SSL until we parse the
   * url ...
   ************************************************************/
  if((2 == sscanf(data->change.url, "%15[^:]:%[^\n]",
                  protobuf, path)) &&
     Curl_raw_equal(protobuf, "file")) {
    if(path[0] == '/' && path[1] == '/') {
      /* Allow omitted hostname (e.g. file:/<path>).  This is not strictly
       * speaking a valid file: URL by RFC 1738, but treating file:/<path> as
       * file://localhost/<path> is similar to how other schemes treat missing
       * hostnames.  See RFC 1808. */

      /* This cannot be done with strcpy() in a portable manner, since the
         memory areas overlap! */
      memmove(path, path + 2, strlen(path + 2)+1);
    }
    /*
     * we deal with file://<host>/<path> differently since it supports no
     * hostname other than "localhost" and "127.0.0.1", which is unique among
     * the URL protocols specified in RFC 1738
     */
    if(path[0] != '/') {
      /* the URL included a host name, we ignore host names in file:// URLs
         as the standards don't define what to do with them */
      char *ptr=strchr(path, '/');
      if(ptr) {
        /* there was a slash present

           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);
      }
    }

    protop = "file"; /* protocol string */
  }
  else {
    /* clear path */
    path[0]=0;

    if(2 > sscanf(data->change.url,
                   "%15[^\n:]://%[^\n/]%[^\n]",
                   protobuf,
                   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))
        protop = "ftp";
      else if(checkprefix("DICT.", conn->host.name))
        protop = "DICT";
      else if(checkprefix("LDAP.", conn->host.name))
        protop = "LDAP";
      else if(checkprefix("IMAP.", conn->host.name))
        protop = "IMAP";
      else {
        protop = "http";
      }

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

  /* 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 long 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 = (unsigned int)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,
   *   protop is A
   *   conn->host.name is B
   *   data->state.path is /C
   */

  return findprotocol(data, conn, protop);
}

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.
* This MUST get called after proxy magic has been figured out.
***************************************************************/
static CURLcode setup_connection_internals(struct connectdata *conn)
{
  const struct Curl_handler * p;
  CURLcode result;

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

  /* Scan protocol handler table. */

  /* Perform setup complement if some. */
  p = conn->handler;

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

    if(result != CURLE_OK)
      return result;

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

  if(conn->port < 0)
    /* we check for -1 here since if proxy was detected already, this
       was very likely already set to the proxy port */
    conn->port = p->defport;
  conn->remote_port = (unsigned short)p->defport;
  conn->protocol |= p->protocol;

  return CURLE_OK;
}

#ifndef CURL_DISABLE_PROXY
/****************************************************************
* Checks if the host is in the noproxy list. returns true if it matches
* and therefore the proxy should NOT be used.
****************************************************************/
static bool check_noproxy(const char* name, const char* no_proxy)
{
  /* 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)
   */
  size_t tok_start;
  size_t tok_end;
  const char* separator = ", ";
  size_t no_proxy_len;
  size_t namelen;
  char *endptr;

  if(no_proxy && no_proxy[0]) {
    if(Curl_raw_equal("*", no_proxy)) {
      return TRUE;
    }

    /* NO_PROXY was specified and it wasn't just an asterisk */

    no_proxy_len = strlen(no_proxy);
    endptr = strchr(name, ':');
    if(endptr)
      namelen = endptr - name;
    else
      namelen = strlen(name);

    for (tok_start = 0; tok_start < no_proxy_len; tok_start = tok_end + 1) {
      while (tok_start < no_proxy_len &&
             strchr(separator, no_proxy[tok_start]) != NULL) {
        /* Look for the beginning of the token. */
        ++tok_start;
      }

      if(tok_start == no_proxy_len)
        break; /* It was all trailing separator chars, no more tokens. */

      for (tok_end = tok_start; tok_end < no_proxy_len &&
             strchr(separator, no_proxy[tok_end]) == NULL; ++tok_end) {
        /* Look for the end of the token. */
      }

      /* To match previous behaviour, where it was necessary to specify
       * ".local.com" to prevent matching "notlocal.com", we will leave
       * the '.' off.
       */
      if(no_proxy[tok_start] == '.')
        ++tok_start;

      if((tok_end - tok_start) <= namelen) {
        /* Match the last part of the name to the domain we are checking. */
        const char *checkn = name + namelen - (tok_end - tok_start);
        if(Curl_raw_nequal(no_proxy + tok_start, checkn, tok_end - tok_start)) {
          if((tok_end - tok_start) == namelen || *(checkn - 1) == '.') {
            /* We either have an exact match, or the previous character is a .
             * so it is within the same domain, so no proxy for this host.
             */
            return TRUE;
          }
        }
      } /* if((tok_end - tok_start) <= namelen) */
    } /* for (tok_start = 0; tok_start < no_proxy_len;
         tok_start = tok_end + 1) */
  } /* NO_PROXY was specified and it wasn't just an asterisk */

  return FALSE;
}

/****************************************************************
* 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 proxy_env[128];

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

  if(!check_noproxy(conn->host.name, no_proxy)) {
    /* It was not listed as without proxy */
    const char *protop = conn->handler->scheme;
    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 && !Curl_raw_equal("http_proxy", proxy_env)) {
      /* There was no lowercase variable, try the uppercase version: */
      Curl_strntoupper(proxy_env, proxy_env, sizeof(proxy_env));
      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(!check_noproxy(conn->host.name, no_proxy)) - it wasn't specified
       non-proxy */
  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;
  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(proxy, "://");
  if(endofprot)
    proxyptr = endofprot+3;
  else
    proxyptr = proxy;

  /* 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]="";

  if(data->set.str[STRING_PROXYUSERNAME] != NULL) {
    strncpy(proxyuser, data->set.str[STRING_PROXYUSERNAME],
            MAX_CURL_USER_LENGTH);
    proxyuser[MAX_CURL_USER_LENGTH-1] = '\0';   /*To be on safe side*/
  }
  if(data->set.str[STRING_PROXYPASSWORD] != NULL) {
    strncpy(proxypasswd, data->set.str[STRING_PROXYPASSWORD],
            MAX_CURL_PASSWORD_LENGTH);
    proxypasswd[MAX_CURL_PASSWORD_LENGTH-1] = '\0'; /*To be on safe side*/
  }

  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;
}
#endif /* CURL_DISABLE_PROXY */

/*
 *
 * 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 :-(
       */

      conn->bits.userpwd_in_url = 1;
      if(data->set.use_netrc != CURL_NETRC_REQUIRED) {
        /* We could use the one in the URL */

        conn->bits.user_passwd = TRUE; /* 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.
 *