X-Git-Url: http://www.privoxy.org/gitweb/?p=privoxy.git;a=blobdiff_plain;f=jcc.c;h=5991a72dc704a8068242881b3ceaeb74f56d665a;hp=5ad5aa7464707fe019459f40d352f6505f9a48cf;hb=e734bb1389aaa25ccc30da467aa439a9b00f9973;hpb=50319af6213f0c3727191ff14681767562e602e2 diff --git a/jcc.c b/jcc.c index 5ad5aa74..5991a72d 100644 --- a/jcc.c +++ b/jcc.c @@ -1,4 +1,4 @@ -const char jcc_rcs[] = "$Id: jcc.c,v 1.179 2008/05/20 20:13:32 fabiankeil Exp $"; +const char jcc_rcs[] = "$Id: jcc.c,v 1.245 2009/04/24 15:29:43 fabiankeil Exp $"; /********************************************************************* * * File : $Source: /cvsroot/ijbswa/current/jcc.c,v $ @@ -6,7 +6,7 @@ const char jcc_rcs[] = "$Id: jcc.c,v 1.179 2008/05/20 20:13:32 fabiankeil Exp $" * Purpose : Main file. Contains main() method, main loop, and * the main connection-handling function. * - * Copyright : Written by and Copyright (C) 2001-2008 the SourceForge + * Copyright : Written by and Copyright (C) 2001-2009 the SourceForge * Privoxy team. http://www.privoxy.org/ * * Based on the Internet Junkbuster originally written @@ -33,6 +33,269 @@ const char jcc_rcs[] = "$Id: jcc.c,v 1.179 2008/05/20 20:13:32 fabiankeil Exp $" * * Revisions : * $Log: jcc.c,v $ + * Revision 1.245 2009/04/24 15:29:43 fabiankeil + * Allow to limit the number of of client connections. + * + * Revision 1.244 2009/04/17 11:34:34 fabiankeil + * Style cosmetics for the IPv6 code. + * + * Revision 1.243 2009/04/17 11:27:49 fabiankeil + * Petr Pisar's privoxy-3.0.12-ipv6-3.diff. + * + * Revision 1.242 2009/04/11 10:44:47 fabiankeil + * Update a comment. We're not in Kansas anymore. + * + * Revision 1.241 2009/04/11 10:37:23 fabiankeil + * When dropping connections due to ACL, don't leak csp->ip_addr_str. + * + * Revision 1.240 2009/04/09 10:12:54 fabiankeil + * Fix two cases in which an invalid server response would result + * in the client connection being closed without sending an error + * message first. + * + * Revision 1.239 2009/04/07 11:43:50 fabiankeil + * If the server rudely resets the connection directly after sending the + * headers, pass the mess to the client instead of sending an incorrect + * connect-failed message. Fixes #2698674 reported by mybugaccount. + * + * Revision 1.238 2009/03/27 14:42:30 fabiankeil + * Correct the status code for CONNECTION_TIMEOUT_RESPONSE. + * + * Revision 1.237 2009/03/27 14:32:04 fabiankeil + * If spawning a child in listen_loop() fails, send a real + * HTTP response to the client and continue listening for + * new connections without artificial delay. + * + * Revision 1.236 2009/03/25 17:30:24 fabiankeil + * In serve(), keep the client socket open until we marked the + * server socket as unused. This should increase the chances + * that we reuse the connection for the client's next request + * to the same destination. + * + * Revision 1.235 2009/03/18 21:01:20 fabiankeil + * Comment fix. Spotted by Roland. + * + * Revision 1.234 2009/03/18 20:48:42 fabiankeil + * If the --no-daemon option is used, enable LOG_LEVEL_INFO + * before the config file has been parsed (as we always did). + * + * Revision 1.233 2009/03/13 14:10:07 fabiankeil + * Fix some more harmless warnings on amd64. + * + * Revision 1.232 2009/03/08 19:29:16 fabiankeil + * Reinitialize the timeout structure every time before passing + * it to select(). Apparently some implementations mess with it. + * Probably fixes #2669131 reported by cyberpatrol. + * + * Revision 1.231 2009/03/08 14:19:23 fabiankeil + * Fix justified (but harmless) compiler warnings + * on platforms where sizeof(int) < sizeof(long). + * + * Revision 1.230 2009/03/07 13:09:17 fabiankeil + * Change csp->expected_content and_csp->expected_content_length from + * size_t to unsigned long long to reduce the likelihood of integer + * overflows that would let us close the connection prematurely. + * Bug found while investigating #2669131, reported by cyberpatrol. + * + * Revision 1.229 2009/03/07 11:17:01 fabiankeil + * Fix compiler warning. + * + * Revision 1.228 2009/03/06 20:30:13 fabiankeil + * Log unsigned values as such. + * + * Revision 1.227 2009/03/02 19:18:11 fabiankeil + * Streamline parse_http_request()'s prototype. As + * cparser pointed out it doesn't actually use csp. + * + * Revision 1.226 2009/03/01 18:28:24 fabiankeil + * Help clang understand that we aren't dereferencing + * NULL pointers here. + * + * Revision 1.225 2009/02/19 18:09:32 fabiankeil + * Unbreak build without FEATURE_CONNECTION_KEEP_ALIVE. + * Noticed by David. + * + * Revision 1.224 2009/02/14 15:32:04 fabiankeil + * Add the request URL to the timeout message in chat(). + * Suggested by Lee. + * + * Revision 1.223 2009/02/09 21:21:16 fabiankeil + * Now that init_log_module() is called earlier, call show_version() + * later on from main() directly so it doesn't get called for --help + * or --version. + * + * Revision 1.222 2009/02/08 12:56:51 fabiankeil + * Call initialize_mutexes() before init_log_module() again. + * Broken since r220, might be the cause of Lee's #2579448. + * + * Revision 1.221 2009/02/06 18:02:58 fabiankeil + * When dropping privileges, also give up membership in supplementary + * groups. Thanks to Matthias Drochner for reporting the problem, + * providing the initial patch and testing the final version. + * + * Revision 1.220 2009/02/04 18:29:07 fabiankeil + * Initialize the log module before parsing arguments. + * Thanks to Matthias Drochner for the report. + * + * Revision 1.219 2009/01/31 16:08:21 fabiankeil + * Remove redundant error check in receive_client_request(). + * + * Revision 1.218 2009/01/31 12:25:54 fabiankeil + * Flatten indentation in receive_client_request(). + * + * Revision 1.217 2009/01/07 19:50:09 fabiankeil + * - If the socket-timeout has been reached and the client + * hasn't received any data yet, send an explanation before + * closing the connection. + * - In get_request_line(), signal timeouts the right way. + * + * Revision 1.216 2008/12/24 22:13:11 ler762 + * fix GCC 3.4.4 warning + * + * Revision 1.215 2008/12/24 17:06:19 fabiankeil + * Keep a thread around to timeout alive connections + * even if no new requests are coming in. + * + * Revision 1.214 2008/12/20 14:53:55 fabiankeil + * Add config option socket-timeout to control the time + * Privoxy waits for data to arrive on a socket. Useful + * in case of stale ssh tunnels or when fuzz-testing. + * + * Revision 1.213 2008/12/15 18:45:51 fabiankeil + * When logging crunches, log the whole URL, so one can easily + * differentiate between vanilla HTTP and CONNECT requests. + * + * Revision 1.212 2008/12/14 15:46:22 fabiankeil + * Give crunched requests their own log level. + * + * Revision 1.211 2008/12/06 10:05:03 fabiankeil + * Downgrade "Received x bytes while expecting y." message to + * LOG_LEVEL_CONNECT as it doesn't necessarily indicate an error. + * + * Revision 1.210 2008/12/02 22:03:18 fabiankeil + * Don't miscalculate byte_count if we don't get all the + * server headers with one read_socket() call. With keep-alive + * support enabled, this caused delays until the server closed + * the connection. + * + * Revision 1.209 2008/11/27 09:44:04 fabiankeil + * Cosmetics for the last commit: Don't watch out for + * the last chunk if the content isn't chunk-encoded or + * if we already determined the content length previously. + * + * Revision 1.208 2008/11/26 18:24:17 fabiankeil + * Recognize that the server response is complete if the + * last chunk is read together with the server headers. + * Reported by Lee. + * + * Revision 1.207 2008/11/25 17:25:16 fabiankeil + * Don't convert the client-header list to text until we need to. + * + * Revision 1.206 2008/11/23 17:00:11 fabiankeil + * Some more chat() cosmetics. + * + * Revision 1.205 2008/11/16 12:43:49 fabiankeil + * Turn keep-alive support into a runtime feature + * that is disabled by setting keep-alive-timeout + * to a negative value. + * + * Revision 1.204 2008/11/06 19:42:17 fabiankeil + * Fix last-chunk detection hack to also apply + * if buf[] contains nothing but the last-chunk. + * + * Revision 1.203 2008/11/06 18:34:35 fabiankeil + * Factor receive_client_request() and + * parse_client_request() out of chat(). + * + * Revision 1.202 2008/11/02 18:40:34 fabiankeil + * If we received a different amount of data than we expected, + * log a warning and make sure the server socket isn't reused. + * + * Revision 1.201 2008/11/02 16:48:20 fabiankeil + * Revert revision 1.195 and try again. + * + * Revision 1.200 2008/10/26 16:53:18 fabiankeil + * Fix gcc44 warning. + * + * Revision 1.199 2008/10/26 15:36:10 fabiankeil + * Remove two debug messages with LOG_LEVEL_INFO. + * + * Revision 1.198 2008/10/22 15:19:55 fabiankeil + * Once More, With Feeling: if there is no logfile + * because the user didn't specify one, we shouldn't + * call init_error_log() after receiving SIGHUP either. + * + * Revision 1.197 2008/10/20 17:02:40 fabiankeil + * If SIGHUP is received while we aren't running in daemon + * mode, calling init_error_log() would be a mistake. + * + * Revision 1.196 2008/10/16 09:16:41 fabiankeil + * - Fix two gcc44 conversion warnings. + * - Don't bother logging the last five bytes + * of the 0-chunk. + * + * Revision 1.195 2008/10/13 16:04:37 fabiankeil + * Make sure we don't try to reuse tainted server sockets. + * + * Revision 1.194 2008/10/12 18:35:18 fabiankeil + * The last commit was a bit too ambitious, apparently the content + * length adjustment is only necessary if we aren't buffering. + * + * Revision 1.193 2008/10/12 15:57:35 fabiankeil + * Fix content length calculation if we read headers + * and the start of the body at once. Now that we have + * FEATURE_CONNECTION_KEEP_ALIVE, it actually matters. + * + * Revision 1.192 2008/10/11 18:19:14 fabiankeil + * Even more chat() cosmetics. + * + * Revision 1.191 2008/10/11 18:00:14 fabiankeil + * Reformat some comments in chat(). + * + * Revision 1.190 2008/10/11 14:58:00 fabiankeil + * In case of chunk-encoded content, stop reading if + * the buffer looks like it ends with the last chunk. + * + * Revision 1.189 2008/10/11 09:53:00 fabiankeil + * Let server_response_is_complete() deal properly with + * content that is neither buffered nor read all at once. + * + * Revision 1.188 2008/10/09 18:21:41 fabiankeil + * Flush work-in-progress changes to keep outgoing connections + * alive where possible. Incomplete and mostly #ifdef'd out. + * + * Revision 1.187 2008/09/07 12:35:05 fabiankeil + * Add mutex lock support for _WIN32. + * + * Revision 1.186 2008/09/04 08:13:58 fabiankeil + * Prepare for critical sections on Windows by adding a + * layer of indirection before the pthread mutex functions. + * + * Revision 1.185 2008/08/30 12:03:07 fabiankeil + * Remove FEATURE_COOKIE_JAR. + * + * Revision 1.184 2008/08/22 15:34:45 fabiankeil + * - Silence LLVM/Clang complaint. + * - Make received_hup_signal static. + * - Hide definitions for basedir, pidfile and received_hup_signal + * from __EMX__ as they only seem to be used in case of #ifdef unix. + * + * Revision 1.183 2008/08/21 07:09:35 fabiankeil + * Accept Shoutcast responses again. Problem reported + * and fix suggested by Stefan in #2062860. + * + * Revision 1.182 2008/06/27 11:13:56 fabiankeil + * Fix possible NULL-pointer dereference reported + * by din_a4 in #2003937. Pointy hat to me. + * + * Revision 1.181 2008/05/21 15:47:15 fabiankeil + * Streamline sed()'s prototype and declare + * the header parse and add structures static. + * + * Revision 1.180 2008/05/21 15:26:32 fabiankeil + * - Mark csp as immutable for send_crunch_response(). + * - Fix comment spelling. + * * Revision 1.179 2008/05/20 20:13:32 fabiankeil * Factor update_server_headers() out of sed(), ditch the * first_run hack and make server_patterns_light static. @@ -1113,9 +1376,9 @@ static jb_err get_request_destination_elsewhere(struct client_state *csp, struct static jb_err get_server_headers(struct client_state *csp); static const char *crunch_reason(const struct http_response *rsp); static void send_crunch_response(const struct client_state *csp, struct http_response *rsp); -/* - * static int request_contains_null_bytes(const struct client_state *csp, char *buf, int len); - */ +static char *get_request_line(struct client_state *csp); +static jb_err receive_client_request(struct client_state *csp); +static jb_err parse_client_request(struct client_state *csp); static void build_request_line(struct client_state *csp, const struct forward_spec *fwd, char **request_line); static jb_err change_request_destination(struct client_state *csp); static void chat(struct client_state *csp); @@ -1145,32 +1408,36 @@ static int32 server_thread(void *data); #define sleep(N) DosSleep(((N) * 100)) #endif -#ifdef FEATURE_PTHREAD -pthread_mutex_t log_mutex; -pthread_mutex_t log_init_mutex; +#ifdef MUTEX_LOCKS_AVAILABLE +/* + * XXX: Does the locking stuff really belong in this file? + */ +privoxy_mutex_t log_mutex; +privoxy_mutex_t log_init_mutex; +privoxy_mutex_t connection_reuse_mutex; #if !defined(HAVE_GETHOSTBYADDR_R) || !defined(HAVE_GETHOSTBYNAME_R) -pthread_mutex_t resolver_mutex; +privoxy_mutex_t resolver_mutex; #endif /* !defined(HAVE_GETHOSTBYADDR_R) || !defined(HAVE_GETHOSTBYNAME_R) */ #ifndef HAVE_GMTIME_R -pthread_mutex_t gmtime_mutex; +privoxy_mutex_t gmtime_mutex; #endif /* ndef HAVE_GMTIME_R */ #ifndef HAVE_LOCALTIME_R -pthread_mutex_t localtime_mutex; +privoxy_mutex_t localtime_mutex; #endif /* ndef HAVE_GMTIME_R */ #ifndef HAVE_RANDOM -pthread_mutex_t rand_mutex; +privoxy_mutex_t rand_mutex; #endif /* ndef HAVE_RANDOM */ -#endif /* FEATURE_PTHREAD */ +#endif /* def MUTEX_LOCKS_AVAILABLE */ -#if defined(unix) || defined(__EMX__) +#if defined(unix) const char *basedir = NULL; const char *pidfile = NULL; -int received_hup_signal = 0; +static int received_hup_signal = 0; #endif /* defined unix */ /* HTTP snipplets. */ @@ -1240,6 +1507,21 @@ static const char MESSED_UP_REQUEST_RESPONSE[] = "Connection: close\r\n\r\n" "Bad request. Messed up with header filters.\r\n"; +static const char TOO_MANY_CONNECTIONS_RESPONSE[] = + "HTTP/1.0 503 Too many open connections\r\n" + "Proxy-Agent: Privoxy " VERSION "\r\n" + "Content-Type: text/plain\r\n" + "Connection: close\r\n\r\n" + "Maximum number of open connections reached.\r\n"; + +/* XXX: should be a template */ +static const char CONNECTION_TIMEOUT_RESPONSE[] = + "HTTP/1.0 504 Connection timeout\r\n" + "Proxy-Agent: Privoxy " VERSION "\r\n" + "Content-Type: text/plain\r\n" + "Connection: close\r\n\r\n" + "The connection timed out.\r\n"; + /* A function to crunch a response */ typedef struct http_response *(*crunch_func_ptr)(struct client_state *); @@ -1279,6 +1561,13 @@ static const struct cruncher crunchers_light[] = { }; +/* + * XXX: Don't we really mean + * + * #if defined(unix) + * + * here? + */ #if !defined(_WIN32) && !defined(__OS2__) && !defined(AMIGA) /********************************************************************* * @@ -1312,7 +1601,9 @@ static void sig_handler(int the_signal) break; case SIGHUP: +#if defined(unix) received_hup_signal = 1; +#endif break; default: @@ -1680,9 +1971,8 @@ static void send_crunch_response(const struct client_state *csp, struct http_res } /* 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", + log_error(LOG_LEVEL_CRUNCH, "%s: %s", crunch_reason(rsp), http->url); + log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" %s %u", csp->ip_addr_str, http->ocmd, status_code, rsp->content_length); /* Clean up and return */ @@ -1739,7 +2029,7 @@ static int request_contains_null_bytes(const struct client_state *csp, char *buf } 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); + "(length=%d, strlen=%u).", csp->ip_addr_str, len, c_len); log_error(LOG_LEVEL_HEADER, "Offending request data with NULL bytes turned into \'°\' characters: %s", buf); @@ -1908,202 +2198,358 @@ static jb_err change_request_destination(struct client_state *csp) 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); + err = parse_http_request(csp->headers->first->str, http); if (JB_ERR_OK != err) { log_error(LOG_LEVEL_ERROR, "Couldn't parse rewritten request: %s.", jb_err_to_string(err)); } - http->ocmd = strdup(http->cmd); /* XXX: ocmd is a misleading name */ - if (http->ocmd == NULL) + else { - log_error(LOG_LEVEL_FATAL, "Out of memory copying rewritten HTTP request line"); + /* 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 : 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. + * Function : server_response_is_complete * - * FIXME: chat is nearly thousand lines long. - * Ridiculous. + * 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 : Nothing. + * Returns : TRUE if the response is complete, + * FALSE otherwise. * *********************************************************************/ -static void chat(struct client_state *csp) +static int server_response_is_complete(struct client_state *csp, + unsigned long long content_length) { - char buf[BUFFER_SIZE]; - char *hdr; - char *p; - char *req = NULL; - 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) */ - jb_err err; + int content_length_known = !!(csp->flags & CSP_FLAG_CONTENT_LENGTH_SET); - /* Function that does the content filtering for the current request */ - filter_function_ptr content_filter = NULL; + 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; + } - /* Skeleton for HTTP response, if we should intercept the request */ - struct http_response *rsp; + if (csp->http->status == 304) + { + /* + * Expect no body. XXX: incomplete "list" of status codes? + */ + csp->expected_content_length = 0; + content_length_known = TRUE; + } - /* Temporary copy of the client's headers before they get enlisted in csp->headers */ - struct list header_list; - struct list *headers = &header_list; + return (content_length_known && ((0 == csp->expected_content_length) + || (csp->expected_content_length <= content_length))); +} - http = csp->http; - memset(buf, 0, sizeof(buf)); +/********************************************************************* + * + * Function : wait_for_alive_connections + * + * Description : Waits for alive connections to timeout. + * + * Parameters : N/A + * + * Returns : N/A + * + *********************************************************************/ +static void wait_for_alive_connections() +{ + int connections_alive = close_unusable_connections(); - /* - * Read the client's request. Note that since we're not using select() we - * could get blocked here if a client connected, then didn't say anything! - */ + while (0 < connections_alive) + { + log_error(LOG_LEVEL_CONNECT, + "Waiting for %d connections to timeout.", + connections_alive); + sleep(60); + connections_alive = close_unusable_connections(); + } - do + log_error(LOG_LEVEL_CONNECT, "No connections to wait for left."); + +} + + +/********************************************************************* + * + * Function : save_connection_destination + * + * Description : Remembers a connection for reuse later on. + * + * Parameters : + * 1 : sfd = Open socket to remember. + * 2 : http = The destination for the connection. + * 3 : fwd = The forwarder settings used. + * 3 : server_connection = storage. + * + * Returns : void + * + *********************************************************************/ +void save_connection_destination(jb_socket sfd, + const struct http_request *http, + const struct forward_spec *fwd, + struct reusable_connection *server_connection) +{ + assert(sfd != JB_INVALID_SOCKET); + assert(NULL != http->host); + server_connection->host = strdup(http->host); + if (NULL == server_connection->host) { - len = read_socket(csp->cfd, buf, sizeof(buf) - 1); + log_error(LOG_LEVEL_FATAL, "Out of memory saving socket."); + } + server_connection->port = http->port; - if (len <= 0) break; /* error! */ + assert(NULL != fwd); + assert(server_connection->gateway_host == NULL); + assert(server_connection->gateway_port == 0); + assert(server_connection->forwarder_type == 0); + assert(server_connection->forward_host == NULL); + assert(server_connection->forward_port == 0); - /* - * 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)) + server_connection->forwarder_type = fwd->type; + if (NULL != fwd->gateway_host) + { + server_connection->gateway_host = strdup(fwd->gateway_host); + if (NULL == server_connection->gateway_host) { - return; + log_error(LOG_LEVEL_FATAL, "Out of memory saving gateway_host."); } + } + else + { + server_connection->gateway_host = NULL; + } + server_connection->gateway_port = fwd->gateway_port; - req = get_header(csp->iob); + if (NULL != fwd->forward_host) + { + server_connection->forward_host = strdup(fwd->forward_host); + if (NULL == server_connection->forward_host) + { + log_error(LOG_LEVEL_FATAL, "Out of memory saving forward_host."); + } + } + else + { + server_connection->forward_host = NULL; + } + server_connection->forward_port = fwd->forward_port; +} +#endif /* FEATURE_CONNECTION_KEEP_ALIVE */ - } while ((NULL != req) && ('\0' == *req)); - if ((NULL != req) && ('\0' != *req)) +/********************************************************************* + * + * 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)) { - /* Request received. Validate and parse it. */ + log_error(LOG_LEVEL_CONNECT, "Unsetting keep-alive flag."); + csp->flags &= ~CSP_FLAG_SERVER_CONNECTION_KEEP_ALIVE; + } +} -#if 0 - /* - * XXX: Temporary disabled to prevent problems - * with POST requests whose bodies are allowed to - * contain NULL bytes. BR#1730105. - * - * The main purpose of this check is to properly - * log stuff like BitTorrent traffic and other junk - * that hits public proxies. It's not required for - * Privoxy to functions as those requests are discarded - * later on anyway. - * - * It probably should be rewritten to only check - * the head of the request. Another option would - * be to let all POST requests pass, although that - * may not be good enough. - */ - if (request_contains_null_bytes(csp, buf, len)) - { - /* NULL bytes found and dealt with, just hang up. */ - return; - } -#endif +/********************************************************************* + * + * 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; - /* Does the request line look invalid? */ - if (client_protocol_is_unsupported(csp, req)) + memset(buf, 0, sizeof(buf)); + + do + { + if (!data_is_available(csp->cfd, csp->config->socket_timeout)) { - /* - * Yes. The request has already been - * answered with a error response, the buffers - * were freed and we're done with chatting. - */ - return; + log_error(LOG_LEVEL_ERROR, + "Stopped waiting for the request line."); + write_socket(csp->cfd, CONNECTION_TIMEOUT_RESPONSE, + strlen(CONNECTION_TIMEOUT_RESPONSE)); + return NULL; } -#ifdef FEATURE_FORCE_LOAD + len = read_socket(csp->cfd, buf, sizeof(buf) - 1); + + if (len <= 0) return NULL; + /* - * If this request contains the FORCE_PREFIX and blocks - * aren't enforced, get rid of it and set the force flag. + * If there is no memory left for buffering the + * request, there is nothing we can do but hang up */ - 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) + if (add_to_iob(csp, buf, len)) { - log_error(LOG_LEVEL_ERROR, "Couldn't parse request: %s.", jb_err_to_string(err)); + return NULL; } - freez(req); - } + request_line = get_header(csp->iob); - 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); + } while ((NULL != request_line) && ('\0' == *request_line)); - free_http_request(http); - return; - } + return request_line; - /* 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 +/********************************************************************* + * + * 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 (req == NULL) + { + return JB_ERR_PARSE; + } + assert(*req != '\0'); + + if (client_protocol_is_unsupported(csp, req)) + { + 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); + freez(req); + if (JB_ERR_OK != err) + { + 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, + "Couldn't parse request line received from %s: %s", + csp->ip_addr_str, jb_err_to_string(err)); + + 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. */ + if (!data_is_available(csp->cfd, csp->config->socket_timeout)) + { + log_error(LOG_LEVEL_ERROR, + "Stopped grabbing the client headers."); + return JB_ERR_PARSE; + } + 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; + return JB_ERR_PARSE; } if (add_to_iob(csp, buf, len)) @@ -2113,7 +2559,7 @@ static void chat(struct client_state *csp) * request, there is nothing we can do but hang up */ destroy_list(headers); - return; + return JB_ERR_MEMORY; } } else @@ -2144,7 +2590,7 @@ static void chat(struct client_state *csp) * An error response has already been send * and we're done here. */ - return; + return JB_ERR_PARSE; } } @@ -2167,23 +2613,50 @@ static void chat(struct client_state *csp) * 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"); + 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); - err = sed(client_patterns, add_client_headers, csp); + 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"); + log_error(LOG_LEVEL_FATAL, "Failed to parse client headers."); } csp->flags |= CSP_FLAG_CLIENT_HEADER_PARSING_DONE; @@ -2197,10 +2670,73 @@ static void chat(struct client_state *csp) */ 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."); - + 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; + unsigned long long 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; + long len = 0; /* 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; + struct timeval timeout; + + 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; } @@ -2210,6 +2746,7 @@ static void chat(struct client_state *csp) { log_error(LOG_LEVEL_FATAL, "gateway spec is NULL!?!? This can't happen!"); /* Never get here - LOG_LEVEL_FATAL causes program exit */ + return; } /* @@ -2267,13 +2804,6 @@ static void chat(struct client_state *csp) 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. */ @@ -2283,28 +2813,18 @@ static void chat(struct client_state *csp) * Yes. The client got the crunch response * and we are done here after cleaning up. */ - freez(hdr); + /* XXX: why list_remove_all()? */ 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); + log_error(LOG_LEVEL_CONNECT, "via [%s]:%d to: %s", + fwd->forward_host, fwd->forward_port, http->hostport); } else { @@ -2313,57 +2833,89 @@ static void chat(struct client_state *csp) /* 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)) +#ifdef FEATURE_CONNECTION_KEEP_ALIVE + if ((csp->sfd != JB_INVALID_SOCKET) + && socket_is_still_usable(csp->sfd) + && connection_destination_matches(&csp->server_connection, http, fwd)) { - log_error(LOG_LEVEL_ERROR, "failed request #%u to connect to %s. Trying again.", - forwarded_connect_retries, http->hostport); + log_error(LOG_LEVEL_CONNECT, + "Reusing server socket %u. Opened for %s.", + csp->sfd, csp->server_connection.host); } - - if (csp->sfd == JB_INVALID_SOCKET) + else { - if (fwd->type != SOCKS_NONE) + if (csp->sfd != JB_INVALID_SOCKET) { - /* Socks error. */ - rsp = error_response(csp, "forwarding-failed", errno); + log_error(LOG_LEVEL_CONNECT, + "Closing server socket %u. Opened for %s.", + csp->sfd, csp->server_connection.host); + close_socket(csp->sfd); + mark_connection_closed(&csp->server_connection); } - else if (errno == EINVAL) +#endif /* def FEATURE_CONNECTION_KEEP_ALIVE */ + + while ((csp->sfd = forwarded_connect(fwd, http, csp)) + && (errno == EINVAL) + && (forwarded_connect_retries++ < max_forwarded_connect_retries)) { - rsp = error_response(csp, "no-such-domain", errno); + log_error(LOG_LEVEL_ERROR, + "failed request #%u to connect to %s. Trying again.", + forwarded_connect_retries, http->hostport); } - else + + if (csp->sfd == JB_INVALID_SOCKET) { - rsp = error_response(csp, "connect-failed", errno); - log_error(LOG_LEVEL_CONNECT, "connect to: %s failed: %E", - http->hostport); - } + 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); + } - /* Write the answer to the client */ - if (rsp != NULL) - { - send_crunch_response(csp, rsp); + return; } +#ifdef FEATURE_CONNECTION_KEEP_ALIVE + save_connection_destination(csp->sfd, http, fwd, &csp->server_connection); + } +#endif /* def FEATURE_CONNECTION_KEEP_ALIVE */ - freez(hdr); - return; + 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"); } + list_remove_all(csp->headers); if (fwd->forward_host || (http->ssl == 0)) { - /* write the client's (modified) header to the server + /* + * 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); + log_error(LOG_LEVEL_CONNECT, + "write header to: %s failed: %E", http->hostport); rsp = error_response(csp, "connect-failed", errno); - - if(rsp) + if (rsp) { send_crunch_response(csp, rsp); } @@ -2392,7 +2944,7 @@ static void chat(struct client_state *csp) /* we're finished with the client's header */ freez(hdr); - maxfd = ( csp->cfd > csp->sfd ) ? csp->cfd : csp->sfd; + 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. @@ -2414,45 +2966,92 @@ static void chat(struct client_state *csp) FD_SET(csp->cfd, &rfds); FD_SET(csp->sfd, &rfds); - n = select((int)maxfd+1, &rfds, NULL, NULL, NULL); +#ifdef FEATURE_CONNECTION_KEEP_ALIVE + if ((csp->flags & CSP_FLAG_CHUNKED) + && !(csp->flags & CSP_FLAG_CONTENT_LENGTH_SET) + && ((csp->iob->eod - csp->iob->cur) >= 5) + && !memcmp(csp->iob->eod-5, "0\r\n\r\n", 5)) + { + log_error(LOG_LEVEL_CONNECT, + "Looks like we read the last chunk together with " + "the server headers. We better stop reading."); + byte_count = (unsigned long long)(csp->iob->eod - csp->iob->cur); + csp->expected_content_length = byte_count; + csp->flags |= CSP_FLAG_CONTENT_LENGTH_SET; + } + if (server_body && server_response_is_complete(csp, byte_count)) + { + log_error(LOG_LEVEL_CONNECT, + "Done reading from server. Expected content length: %llu. " + "Actual content length: %llu. 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 */ + + timeout.tv_sec = csp->config->socket_timeout; + timeout.tv_usec = 0; + n = select((int)maxfd+1, &rfds, NULL, NULL, &timeout); - if (n < 0) + if (n == 0) + { + log_error(LOG_LEVEL_ERROR, + "Didn't receive data in time: %s", http->url); + if ((byte_count == 0) && (http->ssl == 0)) + { + write_socket(csp->cfd, CONNECTION_TIMEOUT_RESPONSE, + strlen(CONNECTION_TIMEOUT_RESPONSE)); + } + mark_server_socket_tainted(csp); + return; + } + else 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 it and write it. + /* + * This is the body of the browser's request, + * just read and write it. + * + * XXX: Make sure the client doesn't use pipelining + * behind Privoxy's back. */ - 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. + * 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 ); + fflush(0); len = read_socket(csp->sfd, buf, sizeof(buf) - 1); if (len < 0) @@ -2481,25 +3080,40 @@ static void chat(struct client_state *csp) */ log_error(LOG_LEVEL_ERROR, "Already forwarded the original headers. " "Unable to tell the client about the problem."); + mark_server_socket_tainted(csp); return; } + /* + * XXX: Consider handling the cases above the same. + */ + mark_server_socket_tainted(csp); + len = 0; + } - rsp = error_response(csp, "connect-failed", errno); - - if (rsp) +#ifdef FEATURE_CONNECTION_KEEP_ALIVE + if (csp->flags & CSP_FLAG_CHUNKED) + { + if ((len >= 5) && !memcmp(buf+len-5, "0\r\n\r\n", 5)) { - send_crunch_response(csp, rsp); + /* 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 + (unsigned long long)len; + csp->flags |= CSP_FLAG_CONTENT_LENGTH_SET; } - - return; } + reading_done: +#endif /* FEATURE_CONNECTION_KEEP_ALIVE */ - /* Add a trailing zero. This lets filter_popups - * use string operations. + /* + * 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 + /* + * 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 @@ -2554,11 +3168,13 @@ static void chat(struct client_state *csp) } if (write_socket(csp->cfd, hdr, strlen(hdr)) - || write_socket(csp->cfd, p != NULL ? p : csp->iob->cur, csp->content_length)) + || write_socket(csp->cfd, + ((p != NULL) ? p : csp->iob->cur), (size_t)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; } @@ -2589,7 +3205,6 @@ static void chat(struct client_state *csp) * 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) @@ -2602,9 +3217,10 @@ static void chat(struct client_state *csp) if (add_to_iob(csp, buf, len)) { size_t hdrlen; - int flushed; + long flushed; - log_error(LOG_LEVEL_INFO, "Flushing header and buffers. Stepping back from filtering."); + log_error(LOG_LEVEL_INFO, + "Flushing header and buffers. Stepping back from filtering."); hdr = list_to_text(csp->headers); if (hdr == NULL) @@ -2616,7 +3232,7 @@ static void chat(struct client_state *csp) 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); @@ -2625,9 +3241,10 @@ static void chat(struct client_state *csp) || ((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"); - + log_error(LOG_LEVEL_CONNECT, + "Flush header and buffers to client failed: %E"); freez(hdr); + mark_server_socket_tainted(csp); return; } @@ -2636,7 +3253,7 @@ static void chat(struct client_state *csp) * we just flushed. len will be added a few lines below, * hdrlen doesn't matter for LOG_LEVEL_CLF. */ - byte_count = (size_t)flushed; + byte_count = (unsigned long long)flushed; freez(hdr); content_filter = NULL; server_body = 1; @@ -2647,55 +3264,63 @@ static void chat(struct client_state *csp) 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; + byte_count += (unsigned long long)len; continue; } else { - /* we're still looking for the end of the - * server's header ... (does that make header - * parsing an "out of body experience" ? - */ - - /* - * buffer up the data we just read. If that fails, - * there's little we can do but send our static - * out-of-memory page. + 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 + /* + * 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; + log_error(LOG_LEVEL_ERROR, "Invalid server headers. " + "Applying the MS IIS5 hack didn't help."); + 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)); + mark_server_socket_tainted(csp); + return; } else { - /* Since we have to wait for - * more from the server before - * we can parse the headers - * we just continue here. + /* + * Since we have to wait for more from the server before + * we can parse the headers we just continue here. */ + long header_offset = csp->iob->cur - header_start; + assert(csp->iob->cur >= header_start); + byte_count += (unsigned long long)(len - header_offset); + log_error(LOG_LEVEL_CONNECT, "Continuing buffering headers. " + "byte_count: %llu. header_offset: %d. len: %d.", + byte_count, header_offset, len); continue; } } @@ -2707,15 +3332,17 @@ static void chat(struct client_state *csp) 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)) + if (strncmpic(csp->headers->first->str, "HTTP", 4) && + strncmpic(csp->headers->first->str, "ICY", 3)) { /* - * It doesn't look like a HTTP response: + * It doesn't look like a HTTP (or Shoutcast) response: * tell the client and log the problem. */ if (strlen(csp->headers->first->str) > 30) @@ -2730,13 +3357,15 @@ static void chat(struct client_state *csp) 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 header. + /* + * We have now received the entire server header, * filter it and send the result to the client */ - if (JB_ERR_OK != sed(server_patterns, add_server_headers, csp)) + if (JB_ERR_OK != sed(csp, FILTER_SERVER_HEADERS)) { log_error(LOG_LEVEL_FATAL, "Failed to parse server headers."); } @@ -2756,6 +3385,7 @@ static void chat(struct client_state *csp) * and are done here after cleaning up. */ freez(hdr); + mark_server_socket_tainted(csp); return; } /* Buffer and pcrs filter this if appropriate. */ @@ -2769,7 +3399,8 @@ static void chat(struct client_state *csp) */ if (!content_filter) { - /* write the server's (modified) header to + /* + * Write the server's (modified) header to * the client (along with anything else that * may be in the buffer) */ @@ -2779,15 +3410,26 @@ static void chat(struct client_state *csp) { 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. + /* + * 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; + byte_count += (unsigned long long)len; + } + else + { + /* + * XXX: the header lenght should probably + * be calculated by get_server_headers(). + */ + long header_length = csp->iob->cur - header_start; + assert(csp->iob->cur > header_start); + byte_count += (unsigned long long)(len - header_length); } /* we're finished with the server's header */ @@ -2795,35 +3437,51 @@ static void chat(struct client_state *csp) 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 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; + log_error(LOG_LEVEL_ERROR, + "Closed server connection detected. " + "Applying the MS IIS5 hack didn't help."); + 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)); + mark_server_socket_tainted(csp); + return; } } continue; } - + mark_server_socket_tainted(csp); return; /* huh? we should never get here */ } if (csp->content_length == 0) { /* - * If Privoxy didn't recalculate the - * Content-Lenght, byte_count is still - * correct. + * If Privoxy didn't recalculate the Content-Lenght, + * byte_count is still correct. */ csp->content_length = byte_count; } - log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" 200 %d", +#ifdef FEATURE_CONNECTION_KEEP_ALIVE + if ((csp->flags & CSP_FLAG_CONTENT_LENGTH_SET) + && (csp->expected_content_length != byte_count)) + { + log_error(LOG_LEVEL_CONNECT, + "Received %llu bytes while expecting %llu.", + byte_count, csp->expected_content_length); + mark_server_socket_tainted(csp); + } +#endif + + log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" 200 %llu", csp->ip_addr_str, http->ocmd, csp->content_length); } @@ -2833,7 +3491,8 @@ static void chat(struct client_state *csp) * Function : serve * * Description : This is little more than chat. We only "serve" to - * to close any socket that chat may have opened. + * to close (or remember) any socket that chat may have + * opened. * * Parameters : * 1 : csp = Current client state (buffers, headers, etc...) @@ -2847,14 +3506,87 @@ void serve(struct client_state *csp) static void serve(struct client_state *csp) #endif /* def AMIGA */ { +#ifdef FEATURE_CONNECTION_KEEP_ALIVE + int continue_chatting = 0; + do + { + chat(csp); + + continue_chatting = (csp->config->feature_flags + & RUNTIME_FEATURE_CONNECTION_KEEP_ALIVE) + && (csp->flags & CSP_FLAG_SERVER_CONNECTION_KEEP_ALIVE) + && (csp->cfd != JB_INVALID_SOCKET) + && (csp->sfd != JB_INVALID_SOCKET) + && socket_is_still_usable(csp->sfd); + + /* + * Get the csp in a mostly vergin state again. + * XXX: Should be done elsewhere. + */ + csp->content_type = 0; + csp->content_length = 0; + csp->expected_content_length = 0; + list_remove_all(csp->headers); + freez(csp->iob->buf); + memset(csp->iob, 0, sizeof(csp->iob)); + freez(csp->error_message); + free_http_request(csp->http); + destroy_list(csp->headers); + destroy_list(csp->tags); + free_current_action(csp->action); + if (NULL != csp->fwd) + { + unload_forward_spec(csp->fwd); + csp->fwd = NULL; + } + + /* XXX: Store per-connection flags someplace else. */ + csp->flags = CSP_FLAG_ACTIVE | (csp->flags & CSP_FLAG_TOGGLED_ON); + + if (continue_chatting) + { + log_error(LOG_LEVEL_CONNECT, + "Waiting for the next client request. " + "Keeping the server socket %d to %s open.", + csp->sfd, csp->server_connection.host); + + if ((csp->flags & CSP_FLAG_CLIENT_CONNECTION_KEEP_ALIVE) + && data_is_available(csp->cfd, csp->config->keep_alive_timeout) + && socket_is_still_usable(csp->cfd)) + { + log_error(LOG_LEVEL_CONNECT, "Client request arrived in " + "time or the client closed the connection."); + } + else + { + log_error(LOG_LEVEL_CONNECT, + "No additional client request received in time. " + "Closing server socket %d, initially opened for %s.", + csp->sfd, csp->server_connection.host); + break; + } + } + else if (csp->sfd != JB_INVALID_SOCKET) + { + log_error(LOG_LEVEL_CONNECT, + "The connection on server socket %d to %s isn't reusable. " + "Closing.", csp->sfd, csp->server_connection.host); + } + } while (continue_chatting); +#else chat(csp); - close_socket(csp->cfd); +#endif /* def FEATURE_CONNECTION_KEEP_ALIVE */ if (csp->sfd != JB_INVALID_SOCKET) { close_socket(csp->sfd); } + if (csp->cfd != JB_INVALID_SOCKET) + { + close_socket(csp->cfd); + } + csp->flags &= ~CSP_FLAG_ACTIVE; } @@ -2914,6 +3646,97 @@ static void usage(const char *myname) #endif /* #if !defined(_WIN32) || defined(_WIN_CONSOLE) */ +#ifdef MUTEX_LOCKS_AVAILABLE +/********************************************************************* + * + * Function : privoxy_mutex_lock + * + * Description : Locks a mutex. + * + * Parameters : + * 1 : mutex = The mutex to lock. + * + * Returns : Void. May exit in case of errors. + * + *********************************************************************/ +void privoxy_mutex_lock(privoxy_mutex_t *mutex) +{ +#ifdef FEATURE_PTHREAD + int err = pthread_mutex_lock(mutex); + if (err) + { + if (mutex != &log_mutex) + { + log_error(LOG_LEVEL_FATAL, + "Mutex locking failed: %s.\n", strerror(err)); + } + exit(1); + } +#else + EnterCriticalSection(mutex); +#endif /* def FEATURE_PTHREAD */ +} + + +/********************************************************************* + * + * Function : privoxy_mutex_unlock + * + * Description : Unlocks a mutex. + * + * Parameters : + * 1 : mutex = The mutex to unlock. + * + * Returns : Void. May exit in case of errors. + * + *********************************************************************/ +void privoxy_mutex_unlock(privoxy_mutex_t *mutex) +{ +#ifdef FEATURE_PTHREAD + int err = pthread_mutex_unlock(mutex); + if (err) + { + if (mutex != &log_mutex) + { + log_error(LOG_LEVEL_FATAL, + "Mutex unlocking failed: %s.\n", strerror(err)); + } + exit(1); + } +#else + LeaveCriticalSection(mutex); +#endif /* def FEATURE_PTHREAD */ +} + + +/********************************************************************* + * + * Function : privoxy_mutex_init + * + * Description : Prepares a mutex. + * + * Parameters : + * 1 : mutex = The mutex to initialize. + * + * Returns : Void. May exit in case of errors. + * + *********************************************************************/ +static void privoxy_mutex_init(privoxy_mutex_t *mutex) +{ +#ifdef FEATURE_PTHREAD + int err = pthread_mutex_init(mutex, 0); + if (err) + { + printf("Fatal error. Mutex initialization failed: %s.\n", + strerror(err)); + exit(1); + } +#else + InitializeCriticalSection(mutex); +#endif /* def FEATURE_PTHREAD */ +} +#endif /* def MUTEX_LOCKS_AVAILABLE */ + /********************************************************************* * * Function : initialize_mutexes @@ -2927,15 +3750,13 @@ static void usage(const char *myname) *********************************************************************/ static void initialize_mutexes(void) { - int err = 0; - -#ifdef FEATURE_PTHREAD +#ifdef MUTEX_LOCKS_AVAILABLE /* * Prepare global mutex semaphores */ - err = pthread_mutex_init(&log_mutex, 0); - - if (!err) err = pthread_mutex_init(&log_init_mutex, 0); + privoxy_mutex_init(&log_mutex); + privoxy_mutex_init(&log_init_mutex); + privoxy_mutex_init(&connection_reuse_mutex); /* * XXX: The assumptions below are a bit naive @@ -2946,37 +3767,24 @@ static void initialize_mutexes(void) * thread safe. */ #if !defined(HAVE_GETHOSTBYADDR_R) || !defined(HAVE_GETHOSTBYNAME_R) - if (!err) err = pthread_mutex_init(&resolver_mutex, 0); + privoxy_mutex_init(&resolver_mutex); #endif /* !defined(HAVE_GETHOSTBYADDR_R) || !defined(HAVE_GETHOSTBYNAME_R) */ /* * XXX: should we use a single mutex for * localtime() and gmtime() as well? */ #ifndef HAVE_GMTIME_R - if (!err) err = pthread_mutex_init(&gmtime_mutex, 0); + privoxy_mutex_init(&gmtime_mutex); #endif /* ndef HAVE_GMTIME_R */ #ifndef HAVE_LOCALTIME_R - if (!err) err = pthread_mutex_init(&localtime_mutex, 0); + privoxy_mutex_init(&localtime_mutex); #endif /* ndef HAVE_GMTIME_R */ #ifndef HAVE_RANDOM - if (!err) err = pthread_mutex_init(&rand_mutex, 0); + privoxy_mutex_init(&rand_mutex); #endif /* ndef HAVE_RANDOM */ -#endif /* FEATURE_PTHREAD */ - - /* - * TODO: mutex support for mingw32 would be swell. - */ - - if (err) - { - printf("Fatal error. Mutex initialization failed: %s.\n", - strerror(err)); - exit(1); - } - - return; +#endif /* def MUTEX_LOCKS_AVAILABLE */ } @@ -3030,6 +3838,12 @@ int main(int argc, const char *argv[]) #endif ; + /* Prepare mutexes if supported and necessary. */ + initialize_mutexes(); + + /* Enable logging until further notice. */ + init_log_module(); + /* * Parse the command line arguments * @@ -3081,6 +3895,7 @@ int main(int argc, const char *argv[]) else if (strcmp(argv[argc_pos], "--no-daemon" ) == 0) { + set_debug_level(LOG_LEVEL_FATAL | LOG_LEVEL_ERROR | LOG_LEVEL_INFO); no_daemon = 1; } @@ -3141,6 +3956,8 @@ int main(int argc, const char *argv[]) } /* -END- while (more arguments) */ + show_version(Argv[0]); + #if defined(unix) if ( *configfile != '/' ) { @@ -3183,24 +4000,9 @@ int main(int argc, const char *argv[]) InitWin32(); #endif - /* Prepare mutexes if supported and necessary. */ - initialize_mutexes(); - - /* Enable logging until further notice. */ - init_log_module(Argv[0]); - random_seed = (unsigned int)time(NULL); #ifdef HAVE_RANDOM srandom(random_seed); -#elif defined (_WIN32) - /* - * See pick_from_range() in miscutil.c for details. - */ - log_error(LOG_LEVEL_INFO, - "No thread-safe PRNG implemented for your platform. " - "Using weak \'randomization\' factor which will " - "limit the already questionable usefulness of " - "header-time-randomizing actions (disabled by default)."); #else srand(random_seed); #endif /* ifdef HAVE_RANDOM */ @@ -3309,8 +4111,8 @@ int main(int argc, const char *argv[]) } #endif /* 1 */ /* - * stderr (fd 2) will be closed later on, when the - * log file has been parsed. + * stderr (fd 2) will be closed later on, + * when the config file has been parsed. */ close( 0 ); @@ -3331,6 +4133,17 @@ int main(int argc, const char *argv[]) { log_error(LOG_LEVEL_FATAL, "Cannot setgid(): Insufficient permissions."); } + if (NULL != grp) + { + if (setgroups(1, &grp->gr_gid)) + { + log_error(LOG_LEVEL_FATAL, "setgroups() failed: %E"); + } + } + else if (initgroups(pw->pw_name, pw->pw_gid)) + { + log_error(LOG_LEVEL_FATAL, "initgroups() failed: %E"); + } if (do_chroot) { if (!pw->pw_dir) @@ -3510,10 +4323,19 @@ static void listen_loop(void) { struct client_state *csp = NULL; jb_socket bfd; - struct configuration_spec * config; + struct configuration_spec *config; + unsigned int active_threads = 0; config = load_config(); +#ifdef FEATURE_CONNECTION_KEEP_ALIVE + /* + * XXX: Should be relocated once it no + * longer needs to emit log messages. + */ + initialize_reusable_connections(); +#endif /* def FEATURE_CONNECTION_KEEP_ALIVE */ + bfd = bind_port_helper(config); #ifdef FEATURE_GRACEFUL_TERMINATION @@ -3532,7 +4354,7 @@ static void listen_loop(void) /* * Free data that was used by died threads */ - sweep(); + active_threads = sweep(); #if defined(unix) /* @@ -3540,28 +4362,14 @@ static void listen_loop(void) */ if (received_hup_signal) { - init_error_log(Argv[0], config->logfile); + if (NULL != config->logfile) + { + init_error_log(Argv[0], config->logfile); + } received_hup_signal = 0; } #endif -#ifdef __OS2__ -#ifdef FEATURE_COOKIE_JAR - /* - * Need a workaround here: we have to fclose() the jarfile, or we die because it's - * already open. I think unload_configfile() is not being run, which should do - * this work. Until that can get resolved, we'll use this workaround. - */ - if (csp) - if(csp->config) - if (csp->config->jar) - { - fclose(csp->config->jar); - csp->config->jar = NULL; - } -#endif /* FEATURE_COOKIE_JAR */ -#endif /* __OS2__ */ - if ( NULL == (csp = (struct client_state *) zalloc(sizeof(*csp))) ) { log_error(LOG_LEVEL_FATAL, "malloc(%d) for csp failed: %E", sizeof(*csp)); @@ -3577,7 +4385,7 @@ static void listen_loop(void) { /* * Since we were listening to the "old port", we will not see - * a "listen" param change until the next IJB request. So, at + * a "listen" param change until the next request. So, at * least 1 more request must be made for us to find the new * setting. I am simply closing the old socket and binding the * new one. @@ -3631,11 +4439,26 @@ static void listen_loop(void) { log_error(LOG_LEVEL_CONNECT, "Connection from %s dropped due to ACL", csp->ip_addr_str); close_socket(csp->cfd); + freez(csp->ip_addr_str); freez(csp); continue; } #endif /* def FEATURE_ACL */ + if ((0 != config->max_client_connections) + && (active_threads >= config->max_client_connections)) + { + log_error(LOG_LEVEL_CONNECT, + "Rejecting connection from %s. Maximum number of connections reached.", + csp->ip_addr_str); + write_socket(csp->cfd, TOO_MANY_CONNECTIONS_RESPONSE, + strlen(TOO_MANY_CONNECTIONS_RESPONSE)); + close_socket(csp->cfd); + freez(csp->ip_addr_str); + freez(csp); + continue; + } + /* add it to the list of clients */ csp->next = clients->next; clients->next = csp; @@ -3802,19 +4625,19 @@ static void listen_loop(void) #undef SELECTED_ONE_OPTION /* end of cpp switch () */ - if (child_id < 0) /* failed */ + if (child_id < 0) { - char buf[BUFFER_SIZE]; - - log_error(LOG_LEVEL_ERROR, "can't fork: %E"); - - snprintf(buf , sizeof(buf), "Privoxy: can't fork: errno = %d", errno); - - write_socket(csp->cfd, buf, strlen(buf)); + /* + * Spawning the child failed, assume it's because + * there are too many children running already. + * XXX: If you assume ... + */ + log_error(LOG_LEVEL_ERROR, + "Unable to take any additional connections: %E"); + write_socket(csp->cfd, TOO_MANY_CONNECTIONS_RESPONSE, + strlen(TOO_MANY_CONNECTIONS_RESPONSE)); close_socket(csp->cfd); csp->flags &= ~CSP_FLAG_ACTIVE; - sleep(5); - continue; } } else @@ -3859,13 +4682,6 @@ static void listen_loop(void) #endif freez(configfile); -#ifdef FEATURE_COOKIE_JAR - if (NULL != config->jar) - { - fclose(config->jar); - } -#endif - #if defined(_WIN32) && !defined(_WIN_CONSOLE) /* Cleanup - remove taskbar icon etc. */ TermLogWindow();