+/*********************************************************************
+ *
+ * Function : client_protocol_is_unsupported
+ *
+ * Description : Checks if the client used a known unsupported
+ * protocol and deals with it by sending an error
+ * response.
+ *
+ * Parameters :
+ * 1 : csp = Current client state (buffers, headers, etc...)
+ * 2 : req = the first request line send by the client
+ *
+ * Returns : TRUE if an error response has been generated, or
+ * FALSE if the request doesn't look invalid.
+ *
+ *********************************************************************/
+int client_protocol_is_unsupported(const struct client_state *csp, char *req)
+{
+ char buf[BUFFER_SIZE];
+
+ /*
+ * If it's a FTP or gopher request, we don't support it.
+ *
+ * These checks are better than nothing, but they might
+ * not work in all configurations and some clients might
+ * have problems digesting the answer.
+ *
+ * They should, however, never cause more problems than
+ * Privoxy's old behaviour (returning the misleading HTML
+ * error message:
+ *
+ * "Could not resolve http://(ftp|gopher)://example.org").
+ */
+ if (!strncmpic(req, "GET ftp://", 10) || !strncmpic(req, "GET gopher://", 13))
+ {
+ if (!strncmpic(req, "GET ftp://", 10))
+ {
+ strlcpy(buf, FTP_RESPONSE, sizeof(buf));
+ log_error(LOG_LEVEL_ERROR, "%s tried to use Privoxy as FTP proxy: %s",
+ csp->ip_addr_str, req);
+ }
+ else
+ {
+ strlcpy(buf, GOPHER_RESPONSE, sizeof(buf));
+ log_error(LOG_LEVEL_ERROR, "%s tried to use Privoxy as gopher proxy: %s",
+ csp->ip_addr_str, req);
+ }
+ log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" 400 0", csp->ip_addr_str, req);
+ freez(req);
+ write_socket(csp->cfd, buf, strlen(buf));
+
+ return TRUE;
+ }
+
+ return FALSE;
+}
+
+
+/*********************************************************************
+ *
+ * Function : get_request_destination_elsewhere
+ *
+ * Description : If the client's request was redirected into
+ * Privoxy without the client's knowledge,
+ * the request line lacks the destination host.
+ *
+ * This function tries to get it elsewhere,
+ * provided accept-intercepted-requests is enabled.
+ *
+ * "Elsewhere" currently only means "Host: header",
+ * but in the future we may ask the redirecting
+ * packet filter to look the destination up.
+ *
+ * If the destination stays unknown, an error
+ * response is send to the client and headers
+ * are freed so that chat() can return directly.
+ *
+ * Parameters :
+ * 1 : csp = Current client state (buffers, headers, etc...)
+ * 2 : headers = a header list
+ *
+ * Returns : JB_ERR_OK if the destination is now known, or
+ * JB_ERR_PARSE if it isn't.
+ *
+ *********************************************************************/
+jb_err get_request_destination_elsewhere(struct client_state *csp, struct list *headers)
+{
+ char *req;
+
+ if (!(csp->config->feature_flags & RUNTIME_FEATURE_ACCEPT_INTERCEPTED_REQUESTS))
+ {
+ log_error(LOG_LEVEL_ERROR, "%s's request: \'%s\' is invalid."
+ " Privoxy isn't configured to accept intercepted requests.",
+ csp->ip_addr_str, csp->http->cmd);
+ /* XXX: Use correct size */
+ log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" 400 0",
+ csp->ip_addr_str, csp->http->cmd);
+
+ write_socket(csp->cfd, CHEADER, strlen(CHEADER));
+ destroy_list(headers);
+
+ return JB_ERR_PARSE;
+ }
+ else if (JB_ERR_OK == get_destination_from_headers(headers, csp->http))
+ {
+ /* Split the domain we just got for pattern matching */
+ init_domain_components(csp->http);
+
+ return JB_ERR_OK;
+ }
+ else
+ {
+ /* We can't work without destination. Go spread the news.*/
+
+ req = list_to_text(headers);
+ chomp(req);
+ /* XXX: Use correct size */
+ log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" 400 0",
+ csp->ip_addr_str, csp->http->cmd);
+ log_error(LOG_LEVEL_ERROR,
+ "Privoxy was unable to get the destination for %s's request:\n%s\n%s",
+ csp->ip_addr_str, csp->http->cmd, req);
+ freez(req);
+
+ write_socket(csp->cfd, MISSING_DESTINATION_RESPONSE, strlen(MISSING_DESTINATION_RESPONSE));
+ destroy_list(headers);
+
+ return JB_ERR_PARSE;
+ }
+ /*
+ * TODO: If available, use PF's ioctl DIOCNATLOOK as last resort
+ * to get the destination IP address, use it as host directly
+ * or do a reverse DNS lookup first.
+ */
+}
+
+
+/*********************************************************************
+ *
+ * Function : get_server_headers
+ *
+ * Description : Parses server headers in iob and fills them
+ * into csp->headers so that they can later be
+ * handled by sed().
+ *
+ * Parameters :
+ * 1 : csp = Current client state (buffers, headers, etc...)
+ *
+ * Returns : JB_ERR_OK if everything went fine, or
+ * JB_ERR_PARSE if the headers were incomplete.
+ *
+ *********************************************************************/
+jb_err get_server_headers(struct client_state *csp)
+{
+ int continue_hack_in_da_house = 0;
+ char * header;
+
+ while (((header = get_header(csp)) != NULL) || continue_hack_in_da_house)
+ {
+ if (header == NULL)
+ {
+ /*
+ * continue hack in da house. Ignore the ending of
+ * this head and continue enlisting header lines.
+ * The reason is described below.
+ */
+ enlist(csp->headers, "");
+ continue_hack_in_da_house = 0;
+ continue;
+ }
+ else if (0 == strncmpic(header, "HTTP/1.1 100", 12))
+ {
+ /*
+ * It's a bodyless continue response, don't
+ * stop header parsing after reaching it's end.
+ *
+ * As a result Privoxy will concatenate the
+ * next response's head and parse and deliver
+ * the headers as if they belonged to one request.
+ *
+ * The client will separate them because of the
+ * empty line between them.
+ *
+ * XXX: What we're doing here is clearly against
+ * the intended purpose of the continue header,
+ * and under some conditions (HTTP/1.0 client request)
+ * it's a standard violation.
+ *
+ * Anyway, "sort of against the spec" is preferable
+ * to "always getting confused by Continue responses"
+ * (Privoxy's behaviour before this hack was added)
+ */
+ log_error(LOG_LEVEL_HEADER, "Continue hack in da house.");
+ continue_hack_in_da_house = 1;
+ }
+ else if (*header == '\0')
+ {
+ /*
+ * If the header is empty, but the Continue hack
+ * isn't active, we can assume that we reached the
+ * end of the buffer before we hit the end of the
+ * head.
+ *
+ * Inform the caller an let it decide how to handle it.
+ */
+ return JB_ERR_PARSE;
+ }
+
+ /* Enlist header */
+ if (JB_ERR_MEMORY == enlist(csp->headers, header))
+ {
+ /*
+ * XXX: Should we quit the request and return a
+ * out of memory error page instead?
+ */
+ log_error(LOG_LEVEL_ERROR,
+ "Out of memory while enlisting server headers. %s lost.",
+ header);
+ }
+ freez(header);
+ }
+
+ return JB_ERR_OK;
+}
+
+
+/*********************************************************************
+ *
+ * Function : crunch_reason
+ *
+ * Description : Translates the crunch reason code into a string.
+ *
+ * Parameters :
+ * 1 : rsp = a http_response
+ *
+ * Returns : A string with the crunch reason or an error description.
+ *
+ *********************************************************************/
+const char *crunch_reason(const struct http_response *rsp)
+{
+ char * reason = NULL;
+
+ assert(rsp != NULL);
+ if (rsp == NULL)
+ {
+ return "Internal error while searching for crunch reason";
+ }
+
+ switch (rsp->reason)
+ {
+ case RSP_REASON_UNSUPPORTED:
+ reason = "Unsupported HTTP feature";
+ break;
+ case RSP_REASON_BLOCKED:
+ reason = "Blocked";
+ break;
+ case RSP_REASON_UNTRUSTED:
+ reason = "Untrusted";
+ break;
+ case RSP_REASON_REDIRECTED:
+ reason = "Redirected";
+ break;
+ case RSP_REASON_CGI_CALL:
+ reason = "CGI Call";
+ break;
+ case RSP_REASON_NO_SUCH_DOMAIN:
+ reason = "DNS failure";
+ break;
+ case RSP_REASON_FORWARDING_FAILED:
+ reason = "Forwarding failed";
+ break;
+ case RSP_REASON_CONNECT_FAILED:
+ reason = "Connection failure";
+ break;
+ case RSP_REASON_OUT_OF_MEMORY:
+ reason = "Out of memory (may mask other reasons)";
+ break;
+ default:
+ reason = "No reason recorded";
+ break;
+ }
+
+ return reason;
+}
+
+
+/*********************************************************************
+ *
+ * Function : send_crunch_response
+ *
+ * Description : Delivers already prepared response for
+ * intercepted requests, logs the interception
+ * and frees the response.
+ *
+ * Parameters :
+ * 1 : csp = Current client state (buffers, headers, etc...)
+ * 1 : rsp = Fully prepared response. Will be freed on exit.
+ *
+ * Returns : Nothing.
+ *
+ *********************************************************************/
+void send_crunch_response(struct client_state *csp, struct http_response *rsp)
+{
+ const struct http_request *http = csp->http;
+ char status_code[4];
+
+ assert(rsp != NULL);
+ assert(rsp->head != NULL);
+
+ if (rsp == NULL)
+ {
+ /*
+ * Not supposed to happen. If it does
+ * anyway, treat it as an unknown error.
+ */
+ cgi_error_unknown(csp, rsp, RSP_REASON_INTERNAL_ERROR);
+ /* return code doesn't matter */
+ }
+
+ if (rsp == NULL)
+ {
+ /* If rsp is still NULL, we have serious internal problems. */
+ log_error(LOG_LEVEL_FATAL,
+ "NULL response in send_crunch_response and cgi_error_unknown failed as well.");
+ }
+
+ /*
+ * Extract the status code from the actual head
+ * that was send to the client. It is the only
+ * way to get it right for all requests, including
+ * the fixed ones for out-of-memory problems.
+ *
+ * A head starts like this: 'HTTP/1.1 200...'
+ * 0123456789|11
+ * 10
+ */
+ status_code[0] = rsp->head[9];
+ status_code[1] = rsp->head[10];
+ status_code[2] = rsp->head[11];
+ status_code[3] = '\0';
+
+ /* Write the answer to the client */
+ if (write_socket(csp->cfd, rsp->head, rsp->head_length)
+ || write_socket(csp->cfd, rsp->body, rsp->content_length))
+ {
+ /* There is nothing we can do about it. */
+ log_error(LOG_LEVEL_ERROR, "write to: %s failed: %E", csp->http->host);
+ }
+
+ /* Log that the request was crunched and why. */
+ log_error(LOG_LEVEL_GPC, "%s%s crunch! (%s)",
+ http->hostport, http->path, crunch_reason(rsp));
+ log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" %s %d",
+ csp->ip_addr_str, http->ocmd, status_code, rsp->content_length);
+
+ /* Clean up and return */
+ if (cgi_error_memory() != rsp)
+ {
+ free_http_response(rsp);
+ }
+ return;
+}
+
+
+/*********************************************************************
+ *
+ * Function : request_contains_null_bytes
+ *
+ * Description : Checks for NULL bytes in the request and sends
+ * an error message to the client if any were found.
+ *
+ * Parameters :
+ * 1 : csp = Current client state (buffers, headers, etc...)
+ * 2 : buf = Data from the client's request to check.
+ * 3 : len = The data length.
+ *
+ * Returns : TRUE if the request contained one or more NULL bytes, or
+ * FALSE otherwise.
+ *
+ *********************************************************************/
+int request_contains_null_bytes(const struct client_state *csp, char *buf, int len)
+{
+ size_t c_len; /* Request lenght when treated as C string */
+
+ c_len = strlen(buf);
+
+ if (c_len < len)
+ {
+ /*
+ * Null byte(s) found. Log the request,
+ * return an error response and hang up.
+ */
+ size_t tmp_len = c_len;
+
+ do
+ {
+ /*
+ * Replace NULL byte(s) with '°' characters
+ * so the request can be logged as string.
+ * XXX: Is there a better replacement character?
+ */
+ buf[tmp_len]='°';
+ tmp_len += strlen(buf+tmp_len);
+ } while (tmp_len < len);
+
+ log_error(LOG_LEVEL_ERROR, "%s\'s request contains at least one NULL byte "
+ "(length=%d, strlen=%d).", csp->ip_addr_str, len, c_len);
+ log_error(LOG_LEVEL_HEADER,
+ "Offending request data with NULL bytes turned into \'°\' characters: %s", buf);
+
+ write_socket(csp->cfd, NULL_BYTE_RESPONSE, strlen(NULL_BYTE_RESPONSE));
+
+ /* XXX: Log correct size */
+ log_error(LOG_LEVEL_CLF, "%s - - [%T] \"Invalid request\" 400 0", csp->ip_addr_str);
+
+ return TRUE;
+ }
+
+ return FALSE;
+}
+
+
+/*********************************************************************
+ *
+ * Function : crunch_response_triggered
+ *
+ * Description : Checks if the request has to be crunched,
+ * and delivers the crunch response if necessary.
+ *
+ * Parameters :
+ * 1 : csp = Current client state (buffers, headers, etc...)
+ * 2 : crunchers = list of cruncher functions to run
+ *
+ * Returns : TRUE if the request was answered with a crunch response
+ * FALSE otherwise.
+ *
+ *********************************************************************/
+int crunch_response_triggered(struct client_state *csp, const struct cruncher crunchers[])
+{
+ struct http_response *rsp = NULL;
+ const struct cruncher *c;
+
+ for (c = crunchers; c->cruncher != NULL; c++)
+ {
+ /*
+ * Check the cruncher if either Privoxy is toggled
+ * on and the request isn't forced, or if the cruncher
+ * applies to forced requests as well.
+ */
+ if (((csp->flags & CSP_FLAG_TOGGLED_ON) &&
+ !(csp->flags & CSP_FLAG_FORCED)) ||
+ (c->flags & CF_IGNORE_FORCE))
+ {
+ rsp = c->cruncher(csp);
+ if (NULL != rsp)
+ {
+ /* Deliver, log and free the interception response. */
+ send_crunch_response(csp, rsp);
+#ifdef FEATURE_STATISTICS
+ if (c->flags & CF_COUNT_AS_REJECT)
+ {
+ csp->flags |= CSP_FLAG_REJECTED;
+ }
+#endif /* def FEATURE_STATISTICS */
+
+ return TRUE;
+ }
+ }
+ }
+
+ return FALSE;
+}
+
+
+/*********************************************************************
+ *
+ * Function : build_request_line
+ *
+ * Description : Builds the HTTP request line.
+ *
+ * If a HTTP forwarder is used it expects the whole URL,
+ * web servers only get the path.
+ *
+ * Parameters :
+ * 1 : csp = Current client state (buffers, headers, etc...)
+ * 2 : fwd = The forwarding spec used for the request
+ *
+ * Returns : Nothing. Terminates in case of memory problems.
+ *
+ *********************************************************************/
+void build_request_line(struct client_state *csp, const struct forward_spec *fwd)
+{
+ struct http_request *http = csp->http;
+
+ assert(http->ssl == 0);
+
+ /*
+ * Downgrade http version from 1.1 to 1.0
+ * if +downgrade action applies.
+ */
+ if ( (csp->action->flags & ACTION_DOWNGRADE)
+ && (!strcmpic(http->ver, "HTTP/1.1")))
+ {
+ freez(http->ver);
+ http->ver = strdup("HTTP/1.0");
+
+ if (http->ver == NULL)
+ {
+ log_error(LOG_LEVEL_FATAL, "Out of memory downgrading HTTP version");
+ }
+ }
+
+ /*
+ * Rebuild the request line.
+ * XXX: If a http forwarder is used and the HTTP version
+ * wasn't downgraded, we don't have to rebuild anything.
+ */
+ freez(http->cmd);
+
+ http->cmd = strdup(http->gpc);
+ string_append(&http->cmd, " ");
+
+ if (fwd->forward_host)
+ {
+ string_append(&http->cmd, http->url);
+ }
+ else
+ {
+ string_append(&http->cmd, http->path);
+ }
+ string_append(&http->cmd, " ");
+ string_append(&http->cmd, http->ver);
+
+ if (http->cmd == NULL)
+ {
+ log_error(LOG_LEVEL_FATAL, "Out of memory writing HTTP command");
+ }
+ log_error(LOG_LEVEL_HEADER, "New HTTP Request-Line: %s", http->cmd);
+}
+
+