+ /*
+ * If CGI request crunching is disabled,
+ * check the CGI dispatcher out of order to
+ * prevent unintentional blocks or redirects.
+ */
+ if (!(csp->config->feature_flags & RUNTIME_FEATURE_CGI_CRUNCHING)
+ && (NULL != (rsp = dispatch_cgi(csp))))
+ {
+ /* Deliver, log and free the interception response. */
+ send_crunch_response(csp, rsp);
+ return TRUE;
+ }
+
+ 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
+ * XXX: Should use http->fwd instead.
+ * 3 : request_line = The old request line which will be replaced.
+ *
+ * Returns : Nothing. Terminates in case of memory problems.
+ *
+ *********************************************************************/
+static void build_request_line(struct client_state *csp, const struct forward_spec *fwd, char **request_line)
+{
+ 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.
+ */
+ freez(*request_line);
+ *request_line = strdup(http->gpc);
+ string_append(request_line, " ");
+
+ if (fwd->forward_host)
+ {
+ string_append(request_line, http->url);
+ }
+ else
+ {
+ string_append(request_line, http->path);
+ }
+ string_append(request_line, " ");
+ string_append(request_line, http->ver);
+
+ if (*request_line == NULL)
+ {
+ log_error(LOG_LEVEL_FATAL, "Out of memory writing HTTP command");
+ }
+ log_error(LOG_LEVEL_HEADER, "New HTTP Request-Line: %s", *request_line);
+}
+
+
+/*********************************************************************
+ *
+ * Function : change_request_destination
+ *
+ * Description : Parse a (rewritten) request line and regenerate
+ * the http request data.
+ *
+ * Parameters :
+ * 1 : csp = Current client state (buffers, headers, etc...)
+ *
+ * Returns : Forwards the parse_http_request() return code.
+ * Terminates in case of memory problems.
+ *
+ *********************************************************************/
+static jb_err change_request_destination(struct client_state *csp)
+{
+ struct http_request *http = csp->http;
+ jb_err err;
+
+ log_error(LOG_LEVEL_INFO, "Rewrite detected: %s", csp->headers->first->str);
+ free_http_request(http);
+ err = parse_http_request(csp->headers->first->str, http, csp);
+ if (JB_ERR_OK != err)
+ {
+ log_error(LOG_LEVEL_ERROR, "Couldn't parse rewritten request: %s.",
+ jb_err_to_string(err));
+ }
+ else
+ {
+ /* XXX: ocmd is a misleading name */
+ http->ocmd = strdup(http->cmd);
+ if (http->ocmd == NULL)
+ {
+ log_error(LOG_LEVEL_FATAL,
+ "Out of memory copying rewritten HTTP request line");
+ }
+ }
+
+ return err;
+}
+
+
+#ifdef FEATURE_CONNECTION_KEEP_ALIVE
+/*********************************************************************
+ *
+ * Function : server_response_is_complete
+ *
+ * Description : Determines whether we should stop reading
+ * from the server socket.
+ *
+ * Parameters :
+ * 1 : csp = Current client state (buffers, headers, etc...)
+ * 2 : content_length = Length of content received so far.
+ *
+ * Returns : TRUE if the response is complete,
+ * FALSE otherwise.
+ *
+ *********************************************************************/
+static int server_response_is_complete(struct client_state *csp, size_t content_length)
+{
+ int content_length_known = (csp->flags & CSP_FLAG_CONTENT_LENGTH_SET);
+
+ if (!strcmpic(csp->http->gpc, "HEAD"))
+ {
+ /*
+ * "HEAD" implies no body, we are thus expecting
+ * no content. XXX: incomplete "list" of methods?
+ */
+ csp->expected_content_length = 0;
+ content_length_known = TRUE;
+ }
+
+ if (csp->http->status == 304)
+ {
+ /*
+ * Expect no body. XXX: incomplete "list" of status codes?
+ */
+ csp->expected_content_length = 0;
+ content_length_known = TRUE;
+ }
+
+ return (content_length_known && ((0 == csp->expected_content_length)
+ || (csp->expected_content_length <= content_length)));
+}
+#endif /* FEATURE_CONNECTION_KEEP_ALIVE */
+
+/*********************************************************************
+ *
+ * Function : mark_server_socket_tainted
+ *
+ * Description : Makes sure we don't reuse a server socket
+ * (if we didn't read everything the server sent
+ * us reusing the socket would lead to garbage).
+ *
+ * Parameters :
+ * 1 : csp = Current client state (buffers, headers, etc...)
+ *
+ * Returns : void.
+ *
+ *********************************************************************/
+static void mark_server_socket_tainted(struct client_state *csp)
+{
+ if ((csp->flags & CSP_FLAG_SERVER_CONNECTION_KEEP_ALIVE))
+ {
+ log_error(LOG_LEVEL_CONNECT, "Unsetting keep-alive flag.");
+ csp->flags &= ~CSP_FLAG_SERVER_CONNECTION_KEEP_ALIVE;
+ }
+}
+
+/*********************************************************************
+ *
+ * Function : get_request_line
+ *
+ * Description : Read the client request line.
+ *
+ * Parameters :
+ * 1 : csp = Current client state (buffers, headers, etc...)
+ *
+ * Returns : Pointer to request line or NULL in case of errors.
+ *
+ *********************************************************************/
+static char *get_request_line(struct client_state *csp)
+{
+ char buf[BUFFER_SIZE];
+ char *request_line = NULL;
+ int len;
+
+ memset(buf, 0, sizeof(buf));
+
+ do
+ {
+ len = read_socket(csp->cfd, buf, sizeof(buf) - 1);
+
+ if (len <= 0) return NULL;
+
+ /*
+ * If there is no memory left for buffering the
+ * request, there is nothing we can do but hang up
+ */
+ if (add_to_iob(csp, buf, len))
+ {
+ return NULL;
+ }
+
+ request_line = get_header(csp->iob);
+
+ } while ((NULL != request_line) && ('\0' == *request_line));
+
+ return request_line;
+
+}
+
+
+/*********************************************************************
+ *
+ * Function : receive_client_request
+ *
+ * Description : Read the client's request (more precisely the
+ * client headers) and answer it if necessary.
+ *
+ * Note that since we're not using select() we could get
+ * blocked here if a client connected, then didn't say
+ * anything!
+ *
+ * Parameters :
+ * 1 : csp = Current client state (buffers, headers, etc...)
+ *
+ * Returns : JB_ERR_OK, JB_ERR_PARSE or JB_ERR_MEMORY
+ *
+ *********************************************************************/
+static jb_err receive_client_request(struct client_state *csp)
+{
+ char buf[BUFFER_SIZE];
+ char *p;
+ char *req = NULL;
+ struct http_request *http;
+ int len;
+ jb_err err;
+
+ /* Temporary copy of the client's headers before they get enlisted in csp->headers */
+ struct list header_list;
+ struct list *headers = &header_list;
+
+ http = csp->http;
+
+ memset(buf, 0, sizeof(buf));
+
+ req = get_request_line(csp);
+
+ if ((NULL != req) && ('\0' != *req))
+ {
+ /* Request received. Validate and parse it. */
+
+ /* Does the request line look invalid? */
+ if (client_protocol_is_unsupported(csp, req))
+ {
+ /*
+ * Yes. The request has already been
+ * answered with a error response, the buffers
+ * were freed and we're done with chatting.
+ */
+ return JB_ERR_PARSE;
+ }
+
+#ifdef FEATURE_FORCE_LOAD
+ /*
+ * If this request contains the FORCE_PREFIX and blocks
+ * aren't enforced, get rid of it and set the force flag.
+ */
+ if (strstr(req, FORCE_PREFIX))
+ {
+ if (csp->config->feature_flags & RUNTIME_FEATURE_ENFORCE_BLOCKS)
+ {
+ log_error(LOG_LEVEL_FORCE,
+ "Ignored force prefix in request: \"%s\".", req);
+ }
+ else
+ {
+ strclean(req, FORCE_PREFIX);
+ log_error(LOG_LEVEL_FORCE, "Enforcing request: \"%s\".", req);
+ csp->flags |= CSP_FLAG_FORCED;
+ }
+ }
+#endif /* def FEATURE_FORCE_LOAD */
+
+ err = parse_http_request(req, http, csp);
+ if (JB_ERR_OK != err)
+ {
+ log_error(LOG_LEVEL_ERROR, "Couldn't parse request: %s.", jb_err_to_string(err));
+ }
+
+ freez(req);
+ }
+
+ if (http->cmd == NULL)
+ {
+ write_socket(csp->cfd, CHEADER, strlen(CHEADER));
+ /* XXX: Use correct size */
+ log_error(LOG_LEVEL_CLF, "%s - - [%T] \"Invalid request\" 400 0", csp->ip_addr_str);
+ log_error(LOG_LEVEL_ERROR, "Invalid header received from %s.", csp->ip_addr_str);
+
+ free_http_request(http);
+ return JB_ERR_PARSE;
+ }
+
+ /* grab the rest of the client's headers */
+ init_list(headers);
+ for (;;)
+ {
+ p = get_header(csp->iob);
+
+ if (p == NULL)
+ {
+ /* There are no additional headers to read. */
+ break;
+ }
+
+ if (*p == '\0')
+ {
+ /*
+ * We didn't receive a complete header
+ * line yet, get the rest of it.
+ */
+ len = read_socket(csp->cfd, buf, sizeof(buf) - 1);
+ if (len <= 0)
+ {
+ log_error(LOG_LEVEL_ERROR, "read from client failed: %E");
+ destroy_list(headers);
+ return JB_ERR_PARSE;
+ }
+
+ if (add_to_iob(csp, buf, len))
+ {
+ /*
+ * If there is no memory left for buffering the
+ * request, there is nothing we can do but hang up
+ */
+ destroy_list(headers);
+ return JB_ERR_MEMORY;
+ }
+ }
+ else
+ {
+ /*
+ * We were able to read a complete
+ * header and can finaly enlist it.
+ */
+ enlist(headers, p);
+ freez(p);
+ }
+ }
+
+ if (http->host == NULL)
+ {
+ /*
+ * If we still don't know the request destination,
+ * the request is invalid or the client uses
+ * Privoxy without its knowledge.
+ */
+ if (JB_ERR_OK != get_request_destination_elsewhere(csp, headers))
+ {
+ /*
+ * Our attempts to get the request destination
+ * elsewhere failed or Privoxy is configured
+ * to only accept proxy requests.
+ *
+ * An error response has already been send
+ * and we're done here.
+ */
+ return JB_ERR_PARSE;
+ }
+ }
+
+ /*
+ * Determine the actions for this URL
+ */
+#ifdef FEATURE_TOGGLE
+ if (!(csp->flags & CSP_FLAG_TOGGLED_ON))
+ {
+ /* Most compatible set of actions (i.e. none) */
+ init_current_action(csp->action);
+ }
+ else
+#endif /* ndef FEATURE_TOGGLE */
+ {
+ get_url_actions(csp, http);
+ }
+
+ /*
+ * Save a copy of the original request for logging
+ */
+ http->ocmd = strdup(http->cmd);
+ if (http->ocmd == NULL)
+ {
+ log_error(LOG_LEVEL_FATAL,
+ "Out of memory copying HTTP request line");
+ }
+ enlist(csp->headers, http->cmd);
+
+ /* Append the previously read headers */
+ list_append_list_unique(csp->headers, headers);
+ destroy_list(headers);
+
+ return JB_ERR_OK;
+
+}
+
+
+/*********************************************************************
+ *
+ * Function : parse_client_request
+ *
+ * Description : Parses the client's request and decides what to do
+ * with it.
+ *
+ * Note that since we're not using select() we could get
+ * blocked here if a client connected, then didn't say
+ * anything!
+ *
+ * Parameters :
+ * 1 : csp = Current client state (buffers, headers, etc...)
+ *
+ * Returns : JB_ERR_OK or JB_ERR_PARSE
+ *
+ *********************************************************************/
+static jb_err parse_client_request(struct client_state *csp)
+{
+ struct http_request *http = csp->http;
+ jb_err err;
+
+ err = sed(csp, FILTER_CLIENT_HEADERS);
+ if (JB_ERR_OK != err)
+ {
+ /* XXX: Should be handled in sed(). */
+ assert(err == JB_ERR_PARSE);
+ log_error(LOG_LEVEL_FATAL, "Failed to parse client headers.");
+ }
+ csp->flags |= CSP_FLAG_CLIENT_HEADER_PARSING_DONE;
+
+ /* Check request line for rewrites. */
+ if ((NULL == csp->headers->first->str)
+ || (strcmp(http->cmd, csp->headers->first->str) &&
+ (JB_ERR_OK != change_request_destination(csp))))
+ {
+ /*
+ * A header filter broke the request line - bail out.
+ */
+ write_socket(csp->cfd, MESSED_UP_REQUEST_RESPONSE, strlen(MESSED_UP_REQUEST_RESPONSE));
+ /* XXX: Use correct size */
+ log_error(LOG_LEVEL_CLF,
+ "%s - - [%T] \"Invalid request generated\" 500 0", csp->ip_addr_str);
+ log_error(LOG_LEVEL_ERROR,
+ "Invalid request line after applying header filters.");
+ free_http_request(http);
+
+ return JB_ERR_PARSE;
+ }
+
+ return JB_ERR_OK;
+
+}
+
+
+/*********************************************************************
+ *
+ * Function : chat
+ *
+ * Description : Once a connection to the client has been accepted,
+ * this function is called (via serve()) to handle the
+ * main business of the communication. When this
+ * function returns, the caller must close the client
+ * socket handle.
+ *
+ * FIXME: chat is nearly thousand lines long.
+ * Ridiculous.
+ *
+ * Parameters :
+ * 1 : csp = Current client state (buffers, headers, etc...)
+ *
+ * Returns : Nothing.
+ *
+ *********************************************************************/
+static void chat(struct client_state *csp)
+{
+ char buf[BUFFER_SIZE];
+ char *hdr;
+ char *p;
+ fd_set rfds;
+ int n;
+ jb_socket maxfd;
+ int server_body;
+ int ms_iis5_hack = 0;
+ size_t byte_count = 0;
+ int forwarded_connect_retries = 0;
+ int max_forwarded_connect_retries = csp->config->forwarded_connect_retries;
+ const struct forward_spec *fwd;
+ struct http_request *http;
+ int len; /* for buffer sizes (and negative error codes) */
+
+ /* Function that does the content filtering for the current request */
+ filter_function_ptr content_filter = NULL;
+
+ /* Skeleton for HTTP response, if we should intercept the request */
+ struct http_response *rsp;
+
+ memset(buf, 0, sizeof(buf));
+
+ http = csp->http;
+
+ if (receive_client_request(csp) != JB_ERR_OK)
+ {
+ return;
+ }
+ if (parse_client_request(csp) != JB_ERR_OK)
+ {
+ return;
+ }
+
+ /* decide how to route the HTTP request */
+ fwd = forward_url(csp, http);
+ if (NULL == fwd)
+ {
+ log_error(LOG_LEVEL_FATAL, "gateway spec is NULL!?!? This can't happen!");
+ /* Never get here - LOG_LEVEL_FATAL causes program exit */
+ }
+
+ /*
+ * build the http request to send to the server
+ * we have to do one of the following:
+ *
+ * create = use the original HTTP request to create a new
+ * HTTP request that has either the path component
+ * without the http://domainspec (w/path) or the
+ * full orininal URL (w/url)
+ * Note that the path and/or the HTTP version may
+ * have been altered by now.
+ *
+ * connect = Open a socket to the host:port of the server
+ * and short-circuit server and client socket.
+ *
+ * pass = Pass the request unchanged if forwarding a CONNECT
+ * request to a parent proxy. Note that we'll be sending
+ * the CFAIL message ourselves if connecting to the parent
+ * fails, but we won't send a CSUCCEED message if it works,
+ * since that would result in a double message (ours and the
+ * parent's). After sending the request to the parent, we simply
+ * tunnel.
+ *
+ * here's the matrix:
+ * SSL
+ * 0 1
+ * +--------+--------+
+ * | | |
+ * 0 | create | connect|
+ * | w/path | |
+ * Forwarding +--------+--------+
+ * | | |
+ * 1 | create | pass |
+ * | w/url | |
+ * +--------+--------+
+ *
+ */
+
+ if (http->ssl && connect_port_is_forbidden(csp))
+ {
+ const char *acceptable_connect_ports =
+ csp->action->string[ACTION_STRING_LIMIT_CONNECT];
+ assert(NULL != acceptable_connect_ports);
+ log_error(LOG_LEVEL_INFO, "Request from %s marked for blocking. "
+ "limit-connect{%s} doesn't allow CONNECT requests to port %d.",
+ csp->ip_addr_str, acceptable_connect_ports, csp->http->port);
+ csp->action->flags |= ACTION_BLOCK;
+ http->ssl = 0;
+ }
+
+ if (http->ssl == 0)
+ {
+ freez(csp->headers->first->str);
+ build_request_line(csp, fwd, &csp->headers->first->str);
+ }
+
+ hdr = list_to_text(csp->headers);
+ if (hdr == NULL)
+ {
+ /* FIXME Should handle error properly */
+ log_error(LOG_LEVEL_FATAL, "Out of memory parsing client header");
+ }
+
+ /*
+ * We have a request. Check if one of the crunchers wants it.
+ */
+ if (crunch_response_triggered(csp, crunchers_all))
+ {
+ /*
+ * Yes. The client got the crunch response
+ * and we are done here after cleaning up.
+ */
+ freez(hdr);
+ list_remove_all(csp->headers);
+
+ return;
+ }
+
+ /*
+ * The headers can't be removed earlier because
+ * they were still needed for the referrer check
+ * in case of CGI crunches.
+ *
+ * XXX: Would it be worth to move the referrer check
+ * into client_referrer() and set a flag if it's trusted?
+ */
+ list_remove_all(csp->headers);
+
+ log_error(LOG_LEVEL_GPC, "%s%s", http->hostport, http->path);
+
+ if (fwd->forward_host)
+ {
+ log_error(LOG_LEVEL_CONNECT, "via %s:%d to: %s",
+ fwd->forward_host, fwd->forward_port, http->hostport);
+ }
+ else
+ {
+ log_error(LOG_LEVEL_CONNECT, "to %s", http->hostport);
+ }
+
+ /* here we connect to the server, gateway, or the forwarder */
+
+ while ( (csp->sfd = forwarded_connect(fwd, http, csp))
+ && (errno == EINVAL) && (forwarded_connect_retries++ < max_forwarded_connect_retries))
+ {
+ log_error(LOG_LEVEL_ERROR, "failed request #%u to connect to %s. Trying again.",
+ forwarded_connect_retries, http->hostport);
+ }
+
+ if (csp->sfd == JB_INVALID_SOCKET)
+ {
+ if (fwd->type != SOCKS_NONE)
+ {
+ /* Socks error. */
+ rsp = error_response(csp, "forwarding-failed", errno);
+ }
+ else if (errno == EINVAL)
+ {
+ rsp = error_response(csp, "no-such-domain", errno);
+ }
+ else
+ {
+ rsp = error_response(csp, "connect-failed", errno);
+ log_error(LOG_LEVEL_CONNECT, "connect to: %s failed: %E",
+ http->hostport);
+ }
+
+ /* Write the answer to the client */
+ if (rsp != NULL)
+ {
+ send_crunch_response(csp, rsp);
+ }
+
+ freez(hdr);
+ return;
+ }
+
+ if (fwd->forward_host || (http->ssl == 0))
+ {
+ /*
+ * Write the client's (modified) header to the server
+ * (along with anything else that may be in the buffer)
+ */
+ if (write_socket(csp->sfd, hdr, strlen(hdr))
+ || (flush_socket(csp->sfd, csp->iob) < 0))
+ {
+ log_error(LOG_LEVEL_CONNECT,
+ "write header to: %s failed: %E", http->hostport);
+
+ rsp = error_response(csp, "connect-failed", errno);
+ if (rsp)
+ {
+ send_crunch_response(csp, rsp);
+ }
+
+ freez(hdr);
+ return;
+ }
+ }
+ else
+ {
+ /*
+ * We're running an SSL tunnel and we're not forwarding,
+ * so just send the "connect succeeded" message to the
+ * client, flush the rest, and get out of the way.
+ */
+ if (write_socket(csp->cfd, CSUCCEED, strlen(CSUCCEED)))
+ {
+ freez(hdr);
+ return;
+ }
+ IOB_RESET(csp);
+ }
+
+ log_error(LOG_LEVEL_CONNECT, "to %s successful", http->hostport);
+
+ /* we're finished with the client's header */
+ freez(hdr);
+
+ maxfd = (csp->cfd > csp->sfd) ? csp->cfd : csp->sfd;
+
+ /* pass data between the client and server
+ * until one or the other shuts down the connection.
+ */
+
+ server_body = 0;
+
+ for (;;)
+ {
+#ifdef __OS2__
+ /*
+ * FD_ZERO here seems to point to an errant macro which crashes.
+ * So do this by hand for now...
+ */
+ memset(&rfds,0x00,sizeof(fd_set));
+#else
+ FD_ZERO(&rfds);
+#endif
+ FD_SET(csp->cfd, &rfds);
+ FD_SET(csp->sfd, &rfds);
+
+#ifdef FEATURE_CONNECTION_KEEP_ALIVE
+ if (server_body && server_response_is_complete(csp, byte_count))
+ {
+ log_error(LOG_LEVEL_CONNECT,
+ "Done reading from server. Expected content length: %d. "
+ "Actual content length: %d. Most recently received: %d.",
+ csp->expected_content_length, byte_count, len);
+ len = 0;
+ /*
+ * XXX: should not jump around,
+ * chat() is complicated enough already.
+ */
+ goto reading_done;
+ }
+#endif /* FEATURE_CONNECTION_KEEP_ALIVE */
+
+ n = select((int)maxfd+1, &rfds, NULL, NULL, NULL);
+
+ if (n < 0)
+ {
+ log_error(LOG_LEVEL_ERROR, "select() failed!: %E");
+ mark_server_socket_tainted(csp);
+ return;
+ }
+
+ /*
+ * This is the body of the browser's request,
+ * just read and write it.
+ */
+ if (FD_ISSET(csp->cfd, &rfds))
+ {
+ len = read_socket(csp->cfd, buf, sizeof(buf) - 1);
+
+ if (len <= 0)
+ {
+ /* XXX: not sure if this is necessary. */
+ mark_server_socket_tainted(csp);
+ break; /* "game over, man" */
+ }
+
+ if (write_socket(csp->sfd, buf, (size_t)len))
+ {
+ log_error(LOG_LEVEL_ERROR, "write to: %s failed: %E", http->host);
+ mark_server_socket_tainted(csp);
+ return;
+ }
+ continue;
+ }
+
+ /*
+ * The server wants to talk. It could be the header or the body.
+ * If `hdr' is null, then it's the header otherwise it's the body.
+ * FIXME: Does `hdr' really mean `host'? No.
+ */
+ if (FD_ISSET(csp->sfd, &rfds))
+ {
+ fflush(0);
+ len = read_socket(csp->sfd, buf, sizeof(buf) - 1);
+
+ if (len < 0)
+ {
+ log_error(LOG_LEVEL_ERROR, "read from: %s failed: %E", http->host);
+
+ if (http->ssl && (fwd->forward_host == NULL))
+ {
+ /*
+ * Just hang up. We already confirmed the client's CONNECT
+ * request with status code 200 and unencrypted content is
+ * no longer welcome.
+ */
+ log_error(LOG_LEVEL_ERROR,
+ "CONNECT already confirmed. Unable to tell the client about the problem.");
+ return;
+ }
+ else if (byte_count)
+ {
+ /*
+ * Just hang up. We already transmitted the original headers
+ * and parts of the original content and therefore missed the
+ * chance to send an error message (without risking data corruption).
+ *
+ * XXX: we could retry with a fancy range request here.
+ */
+ log_error(LOG_LEVEL_ERROR, "Already forwarded the original headers. "
+ "Unable to tell the client about the problem.");
+ mark_server_socket_tainted(csp);
+ return;
+ }
+
+ rsp = error_response(csp, "connect-failed", errno);
+ if (rsp)
+ {
+ send_crunch_response(csp, rsp);
+ }
+
+ return;
+ }
+
+#ifdef FEATURE_CONNECTION_KEEP_ALIVE
+ if (csp->flags & CSP_FLAG_CHUNKED)
+ {
+ if ((len >= 5) && !memcmp(buf+len-5, "0\r\n\r\n", 5))
+ {
+ /* XXX: this is a temporary hack */
+ log_error(LOG_LEVEL_CONNECT,
+ "Looks like we reached the end of the last chunk. "
+ "We better stop reading.");
+ csp->expected_content_length = byte_count + (size_t)len;
+ csp->flags |= CSP_FLAG_CONTENT_LENGTH_SET;
+ }
+ }
+ reading_done:
+#endif /* FEATURE_CONNECTION_KEEP_ALIVE */
+
+ /*
+ * Add a trailing zero to let be able to use string operations.
+ * XXX: do we still need this with filter_popups gone?
+ */
+ buf[len] = '\0';
+
+ /*
+ * Normally, this would indicate that we've read
+ * as much as the server has sent us and we can
+ * close the client connection. However, Microsoft
+ * in its wisdom has released IIS/5 with a bug that
+ * prevents it from sending the trailing \r\n in
+ * a 302 redirect header (and possibly other headers).
+ * To work around this if we've haven't parsed
+ * a full header we'll append a trailing \r\n
+ * and see if this now generates a valid one.
+ *
+ * This hack shouldn't have any impacts. If we've
+ * already transmitted the header or if this is a
+ * SSL connection, then we won't bother with this
+ * hack. So we only work on partially received
+ * headers. If we append a \r\n and this still
+ * doesn't generate a valid header, then we won't
+ * transmit anything to the client.
+ */
+ if (len == 0)
+ {
+
+ if (server_body || http->ssl)
+ {
+ /*
+ * If we have been buffering up the document,
+ * now is the time to apply content modification
+ * and send the result to the client.
+ */
+ if (content_filter)
+ {
+ p = execute_content_filter(csp, content_filter);
+ /*
+ * If the content filter fails, use the original
+ * buffer and length.
+ * (see p != NULL ? p : csp->iob->cur below)
+ */
+ if (NULL == p)
+ {
+ csp->content_length = (size_t)(csp->iob->eod - csp->iob->cur);
+ }
+
+ if (JB_ERR_OK != update_server_headers(csp))
+ {
+ log_error(LOG_LEVEL_FATAL,
+ "Failed to update server headers. after filtering.");
+ }
+
+ hdr = list_to_text(csp->headers);
+ if (hdr == NULL)
+ {
+ /* FIXME Should handle error properly */
+ log_error(LOG_LEVEL_FATAL, "Out of memory parsing server header");
+ }
+
+ if (write_socket(csp->cfd, hdr, strlen(hdr))
+ || write_socket(csp->cfd, p != NULL ? p : csp->iob->cur, csp->content_length))
+ {
+ log_error(LOG_LEVEL_ERROR, "write modified content to client failed: %E");
+ freez(hdr);
+ freez(p);
+ mark_server_socket_tainted(csp);
+ return;
+ }
+
+ freez(hdr);
+ freez(p);
+ }
+
+ break; /* "game over, man" */
+ }
+
+ /*
+ * This is NOT the body, so
+ * Let's pretend the server just sent us a blank line.
+ */
+ snprintf(buf, sizeof(buf), "\r\n");
+ len = (int)strlen(buf);
+
+ /*
+ * Now, let the normal header parsing algorithm below do its
+ * job. If it fails, we'll exit instead of continuing.
+ */
+
+ ms_iis5_hack = 1;
+ }
+
+ /*
+ * If this is an SSL connection or we're in the body
+ * of the server document, just write it to the client,
+ * unless we need to buffer the body for later content-filtering
+ */
+ if (server_body || http->ssl)
+ {
+ if (content_filter)
+ {
+ /*
+ * If there is no memory left for buffering the content, or the buffer limit
+ * has been reached, switch to non-filtering mode, i.e. make & write the
+ * header, flush the iob and buf, and get out of the way.
+ */
+ if (add_to_iob(csp, buf, len))
+ {
+ size_t hdrlen;
+ int flushed;
+
+ log_error(LOG_LEVEL_INFO,
+ "Flushing header and buffers. Stepping back from filtering.");
+
+ hdr = list_to_text(csp->headers);
+ if (hdr == NULL)
+ {
+ /*
+ * Memory is too tight to even generate the header.
+ * Send our static "Out-of-memory" page.
+ */
+ log_error(LOG_LEVEL_ERROR, "Out of memory while trying to flush.");
+ rsp = cgi_error_memory();
+ send_crunch_response(csp, rsp);
+ mark_server_socket_tainted(csp);
+ return;
+ }
+ hdrlen = strlen(hdr);
+
+ if (write_socket(csp->cfd, hdr, hdrlen)
+ || ((flushed = flush_socket(csp->cfd, csp->iob)) < 0)
+ || (write_socket(csp->cfd, buf, (size_t)len)))
+ {
+ log_error(LOG_LEVEL_CONNECT,
+ "Flush header and buffers to client failed: %E");
+ freez(hdr);
+ mark_server_socket_tainted(csp);
+ return;
+ }
+
+ /*
+ * Reset the byte_count to the amount of bytes
+ * we just flushed. len will be added a few lines below,
+ * hdrlen doesn't matter for LOG_LEVEL_CLF.
+ */
+ byte_count = (size_t)flushed;
+ freez(hdr);
+ content_filter = NULL;
+ server_body = 1;
+ }
+ }
+ else
+ {
+ if (write_socket(csp->cfd, buf, (size_t)len))
+ {
+ log_error(LOG_LEVEL_ERROR, "write to client failed: %E");
+ mark_server_socket_tainted(csp);
+ return;
+ }
+ }
+ byte_count += (size_t)len;
+ continue;
+ }
+ else
+ {
+ const char *header_start;
+ /*
+ * We're still looking for the end of the server's header.
+ * Buffer up the data we just read. If that fails, there's
+ * little we can do but send our static out-of-memory page.
+ */
+ if (add_to_iob(csp, buf, len))
+ {
+ log_error(LOG_LEVEL_ERROR, "Out of memory while looking for end of server headers.");
+ rsp = cgi_error_memory();
+ send_crunch_response(csp, rsp);
+ mark_server_socket_tainted(csp);
+ return;
+ }
+
+ header_start = csp->iob->cur;
+
+ /* Convert iob into something sed() can digest */
+ if (JB_ERR_PARSE == get_server_headers(csp))
+ {
+ if (ms_iis5_hack)
+ {
+ /*
+ * Well, we tried our MS IIS/5 hack and it didn't work.
+ * The header is incomplete and there isn't anything
+ * we can do about it.
+ */
+ log_error(LOG_LEVEL_INFO,
+ "MS IIS5 hack didn't produce valid headers.");
+ break;
+ }
+ else
+ {
+ /*
+ * Since we have to wait for more from the server before
+ * we can parse the headers we just continue here.
+ */
+ continue;
+ }
+ }
+
+ /* Did we actually get anything? */
+ if (NULL == csp->headers->first)
+ {
+ log_error(LOG_LEVEL_ERROR, "Empty server or forwarder response.");
+ log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" 502 0", csp->ip_addr_str, http->cmd);
+ write_socket(csp->cfd, NO_SERVER_DATA_RESPONSE, strlen(NO_SERVER_DATA_RESPONSE));
+ free_http_request(http);
+ mark_server_socket_tainted(csp);
+ return;
+ }
+
+ assert(csp->headers->first->str);
+ assert(!http->ssl);
+ if (strncmpic(csp->headers->first->str, "HTTP", 4) &&
+ strncmpic(csp->headers->first->str, "ICY", 3))
+ {
+ /*
+ * It doesn't look like a HTTP (or Shoutcast) response:
+ * tell the client and log the problem.
+ */
+ if (strlen(csp->headers->first->str) > 30)
+ {
+ csp->headers->first->str[30] = '\0';
+ }
+ log_error(LOG_LEVEL_ERROR,
+ "Invalid server or forwarder response. Starts with: %s",
+ csp->headers->first->str);
+ log_error(LOG_LEVEL_CLF,
+ "%s - - [%T] \"%s\" 502 0", csp->ip_addr_str, http->cmd);
+ write_socket(csp->cfd, INVALID_SERVER_HEADERS_RESPONSE,
+ strlen(INVALID_SERVER_HEADERS_RESPONSE));
+ free_http_request(http);
+ mark_server_socket_tainted(csp);
+ return;
+ }
+
+ /*
+ * We have now received the entire server header,
+ * filter it and send the result to the client
+ */
+ if (JB_ERR_OK != sed(csp, FILTER_SERVER_HEADERS))
+ {
+ log_error(LOG_LEVEL_FATAL, "Failed to parse server headers.");
+ }
+ hdr = list_to_text(csp->headers);
+ if (hdr == NULL)
+ {
+ /* FIXME Should handle error properly */
+ log_error(LOG_LEVEL_FATAL, "Out of memory parsing server header");
+ }
+
+ if (crunch_response_triggered(csp, crunchers_light))
+ {
+ /*
+ * One of the tags created by a server-header
+ * tagger triggered a crunch. We already
+ * delivered the crunch response to the client
+ * and are done here after cleaning up.
+ */
+ freez(hdr);
+ mark_server_socket_tainted(csp);
+ return;
+ }
+ /* Buffer and pcrs filter this if appropriate. */
+
+ if (!http->ssl) /* We talk plaintext */
+ {
+ content_filter = get_filter_function(csp);
+ }
+ /*
+ * Only write if we're not buffering for content modification
+ */
+ if (!content_filter)
+ {
+ /*
+ * Write the server's (modified) header to
+ * the client (along with anything else that
+ * may be in the buffer)
+ */
+
+ if (write_socket(csp->cfd, hdr, strlen(hdr))
+ || ((len = flush_socket(csp->cfd, csp->iob)) < 0))
+ {
+ log_error(LOG_LEVEL_CONNECT, "write header to client failed: %E");
+
+ /*
+ * The write failed, so don't bother mentioning it
+ * to the client... it probably can't hear us anyway.
+ */
+ freez(hdr);
+ mark_server_socket_tainted(csp);
+ return;
+ }
+
+ byte_count += (size_t)len;
+ }
+ else
+ {
+ /*
+ * XXX: the header lenght should probably
+ * be calculated by get_server_headers().
+ */
+ int header_length = csp->iob->cur - header_start;
+ assert(csp->iob->cur > header_start);
+ byte_count += (size_t)(len - header_length);
+ }
+
+ /* we're finished with the server's header */
+
+ freez(hdr);
+ server_body = 1;
+
+ /*
+ * If this was a MS IIS/5 hack then it means the server
+ * has already closed the connection. Nothing more to read.
+ * Time to bail.
+ */
+ if (ms_iis5_hack)
+ {
+ log_error(LOG_LEVEL_INFO,
+ "Closed server connection detected with MS IIS5 hack enabled.");
+ break;
+ }
+ }
+ continue;
+ }
+ mark_server_socket_tainted(csp);
+ return; /* huh? we should never get here */
+ }
+
+ if (csp->content_length == 0)