X-Git-Url: http://www.privoxy.org/gitweb/?p=privoxy.git;a=blobdiff_plain;f=jcc.c;h=cde64b09426c8590ff3f79a8b9f314aaa80a7314;hp=71f89461dff633ae6371d72ccf23768456bb9bf6;hb=cd275efe90ca39d461537daf389e79a3cd79e507;hpb=14ca00a0b073cb82503a8f1cdeb6e00f963d1744 diff --git a/jcc.c b/jcc.c index 71f89461..cde64b09 100644 --- a/jcc.c +++ b/jcc.c @@ -1,4 +1,4 @@ -const char jcc_rcs[] = "$Id: jcc.c,v 1.109 2006/12/06 19:41:40 fabiankeil Exp $"; +const char jcc_rcs[] = "$Id: jcc.c,v 1.204 2008/11/06 19:42:17 fabiankeil Exp $"; /********************************************************************* * * File : $Source: /cvsroot/ijbswa/current/jcc.c,v $ @@ -6,7 +6,7 @@ const char jcc_rcs[] = "$Id: jcc.c,v 1.109 2006/12/06 19:41:40 fabiankeil Exp $" * Purpose : Main file. Contains main() method, main loop, and * the main connection-handling function. * - * Copyright : Written by and Copyright (C) 2001 the SourceForge + * Copyright : Written by and Copyright (C) 2001-2008 the SourceForge * Privoxy team. http://www.privoxy.org/ * * Based on the Internet Junkbuster originally written @@ -33,6 +33,406 @@ const char jcc_rcs[] = "$Id: jcc.c,v 1.109 2006/12/06 19:41:40 fabiankeil Exp $" * * Revisions : * $Log: jcc.c,v $ + * 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. + * + * Revision 1.178 2008/05/10 13:23:38 fabiankeil + * Don't provide get_header() with the whole client state + * structure when it only needs access to csp->iob. + * + * Revision 1.177 2008/05/10 11:51:12 fabiankeil + * Make the "read the rest of the headers" loop a bit more readable. + * + * Revision 1.176 2008/05/10 11:37:57 fabiankeil + * - Instead of logging when the IIS5 hack is enabled, log when it fails. + * - Remove useless comment. + * + * Revision 1.175 2008/05/09 18:53:59 fabiankeil + * Fix comment grammar. + * + * Revision 1.174 2008/05/07 18:05:53 fabiankeil + * Remove the pointless buffer in client_protocol_is_unsupported(). + * + * Revision 1.173 2008/05/06 15:09:00 fabiankeil + * Least-effort fix for bug #1821930 (reported by Lee): + * If the response doesn't look like HTTP, + * tell the client and log the problem. + * + * Revision 1.172 2008/04/16 16:38:21 fabiankeil + * Don't pass the whole csp structure to flush_socket() + * when it only needs a file descriptor and a buffer. + * + * Revision 1.171 2008/03/27 18:27:25 fabiankeil + * Remove kill-popups action. + * + * Revision 1.170 2008/03/06 16:33:46 fabiankeil + * If limit-connect isn't used, don't limit CONNECT requests to port 443. + * + * Revision 1.169 2008/03/04 18:30:39 fabiankeil + * Remove the treat-forbidden-connects-like-blocks action. We now + * use the "blocked" page for forbidden CONNECT requests by default. + * + * Revision 1.168 2008/03/02 12:25:25 fabiankeil + * Also use shiny new connect_port_is_forbidden() in jcc.c. + * + * Revision 1.167 2008/02/23 16:57:12 fabiankeil + * Rename url_actions() to get_url_actions() and let it + * use the standard parameter ordering. + * + * Revision 1.166 2008/02/23 16:33:43 fabiankeil + * Let forward_url() use the standard parameter ordering + * and mark its second parameter immutable. + * + * Revision 1.165 2008/02/02 19:36:56 fabiankeil + * Remove the "Listening ... for local connections only" log message. + * Whether or not remote connections are able to reach Privoxy is up + * to the operating system. + * + * Revision 1.164 2007/12/16 18:32:46 fabiankeil + * Prevent the log messages for CONNECT requests to unacceptable + * ports from printing the limit-connect argument as [null] if + * limit-connect hasn't been explicitly enabled. + * + * Revision 1.163 2007/12/13 01:47:11 david__schmidt + * Make sure all console-mode apps get a usage() instance + * + * Revision 1.162 2007/12/06 17:54:57 fabiankeil + * Reword NO_SERVER_DATA_RESPONSE to make it harder + * to misunderstand what the message is all about. + * + * Revision 1.161 2007/12/04 19:44:22 fabiankeil + * Unbreak trustfile which previously didn't work without + * FEATURE_TOGGLE. Fixes BR#1843585, reported by Lee. + * + * Revision 1.160 2007/11/29 18:00:29 fabiankeil + * Plug memory leak. Spotted by Valgrind, triggered by + * Privoxy-Regression-Test feeding proxyfuzz.py. + * + * Revision 1.159 2007/11/24 14:34:09 fabiankeil + * In the HTTP snipplets, refer to the client as client. + * + * Revision 1.158 2007/11/11 16:44:17 fabiankeil + * Emit a log message when activating the MS IIS5 hack. + * + * Revision 1.157 2007/11/03 17:34:49 fabiankeil + * Log the "weak randomization factor" warning only + * once for mingw32 and provide some more details. + * + * Revision 1.156 2007/11/01 18:20:58 fabiankeil + * Initialize log module after initializing mutexes, future + * deadlocks in that code should now work cross-platform. + * + * Revision 1.155 2007/10/23 20:12:45 fabiankeil + * Fix first CSUCCEED line to end in \r\n as required by RFC1945. + * Reported by Bert van Leeuwen in BR#1818808. + * + * Revision 1.154 2007/10/19 17:00:08 fabiankeil + * Downgrade "Flushing header and buffers" message to LOG_LEVEL_INFO. + * + * Revision 1.153 2007/10/14 14:12:41 fabiankeil + * When in daemon mode, close stderr after the configuration file has been + * parsed the first time. If logfile isn't set, stop logging. Fixes BR#897436. + * + * Revision 1.152 2007/10/04 18:03:34 fabiankeil + * - Fix a crash when parsing invalid requests whose first header + * is rejected by get_header(). Regression (re?)introduced + * in r1.143 by yours truly. + * - Move ACTION_VANILLA_WAFER handling into parsers.c's + * client_cookie_adder() to make sure send-vanilla-wafer can be + * controlled through tags (and thus regression-tested). + * + * Revision 1.151 2007/09/29 10:21:16 fabiankeil + * - Move get_filter_function() from jcc.c to filters.c + * so the filter functions can be static. + * - Don't bother filtering body-less responses. + * + * Revision 1.150 2007/09/28 16:39:29 fabiankeil + * Execute content filters through execute_content_filter(). + * + * Revision 1.149 2007/09/04 15:08:48 fabiankeil + * Initialize req to NULL to make sure it's defined if the + * first read_socket() call fails. Reported by icmp30. + * + * Revision 1.148 2007/08/26 16:47:13 fabiankeil + * Add Stephen Gildea's --pre-chroot-nslookup patch [#1276666], + * extensive comments moved to user manual. + * + * Revision 1.147 2007/08/25 14:42:40 fabiankeil + * Don't crash if a broken header filter wiped out the request line. + * + * Revision 1.146 2007/08/20 17:09:32 fabiankeil + * Fix byte_count calculation in case of flushes + * and don't parse the server headers a second time. + * + * Revision 1.145 2007/08/19 13:13:31 fabiankeil + * - If there's a connection problem after we already forwarded + * parts of the original content, just hang up. Fixes BR#1776724. + * - Fix warnings about unused code on mingw32. + * - In case of flushes, calculate the byte count + * less incorrectly (I think). + * + * Revision 1.144 2007/08/11 14:43:22 fabiankeil + * Add some more prototypes for static functions. + * + * Revision 1.143 2007/08/05 13:58:19 fabiankeil + * Comment out request_contains_null_bytes() until it's used again. + * + * Revision 1.142 2007/08/05 13:50:26 fabiankeil + * #1763173 from Stefan Huehner: s@const static@static const@ + * and declare some more functions static. + * + * Revision 1.141 2007/08/04 09:56:23 fabiankeil + * - Log rejected CONNECT requests with LOG_LEVEL_INFO + * and explain why they were rejected in the first place. + * - Fix the LOG_LEVEL_CLF message for crunches of unallowed + * CONNECT requests. The request line was missing. + * - Add two more XXX reminders as we don't have enough already. + * + * Revision 1.140 2007/07/21 11:51:36 fabiankeil + * As Hal noticed, checking dispatch_cgi() as the last cruncher + * looks like a bug if CGI requests are blocked unintentionally, + * so don't do it unless the user enabled the new config option + * "allow-cgi-request-crunching". + * + * Revision 1.139 2007/07/14 07:46:41 fabiankeil + * - Allow to rewrite the request destination behind the client's back. + * - Turn the weird-looking unconditional for loop that + * reads the client request into a conditional while loop. + * Move the stuff that only runs once out of the loop. + * - Move parts of chat(), server_content_type() and the + * necessary stuff to fix BR#1750917 into get_filter_function(). + * + * Revision 1.138 2007/06/03 18:45:18 fabiankeil + * Temporary workaround for BR#1730105. + * + * Revision 1.137 2007/06/01 18:16:36 fabiankeil + * Use the same mutex for gethostbyname() and gethostbyaddr() to prevent + * deadlocks and crashes on OpenBSD and possibly other OS with neither + * gethostbyname_r() nor gethostaddr_r(). Closes BR#1729174. + * Thanks to Ralf Horstmann for report and solution. + * + * Revision 1.136 2007/06/01 16:41:11 fabiankeil + * Add forward-override{} to change the forwarding settings through + * action sections. This is mainly interesting to forward different + * clients differently (for example based on User-Agent or request + * origin). + * + * Revision 1.135 2007/05/24 17:03:50 fabiankeil + * - Let usage() mention the --chroot parameter. + * - Use read_socket() consistently and always leave + * the last buffer byte alone, even in cases where + * null termination (currently) doesn't matter. + * + * Revision 1.134 2007/05/16 14:59:46 fabiankeil + * - Fix config file loading on Unix if no config file is specified. + * Since r1.97 Privoxy would always interpret the last argument as + * config file, even if it's a valid command line option. + * - Abort in case of unrecognized command line options. Closes #1719696. + * - Remove a bunch of unnecessary strcpy() calls (yay for c&p without thinking). + * - Replace the remaining strcpy() and strcat() calls with strlcpy() and strcat(). + * + * Revision 1.133 2007/05/04 11:23:19 fabiankeil + * - Don't rerun crunchers that only depend on the request URL. + * - Don't count redirects and CGI requests as "blocked requests". + * + * Revision 1.132 2007/04/25 15:15:17 fabiankeil + * Support crunching based on tags created by server-header taggers. + * + * Revision 1.131 2007/04/22 13:24:50 fabiankeil + * Make HTTP snippets static (again). Add a Content-Type for those + * with content so the browser doesn't guess it based on the URL. + * + * Revision 1.130 2007/04/19 13:47:34 fabiankeil + * Move crunching and request line rebuilding out of chat(). + * + * Revision 1.129 2007/04/15 16:39:20 fabiankeil + * Introduce tags as alternative way to specify which + * actions apply to a request. At the moment tags can be + * created based on client and server headers. + * + * Revision 1.128 2007/03/25 16:55:54 fabiankeil + * Don't CLF-log CONNECT requests twice. + * + * Revision 1.127 2007/03/20 13:53:17 fabiankeil + * Log the source address for ACL-related connection drops. + * + * Revision 1.126 2007/03/17 15:20:05 fabiankeil + * New config option: enforce-blocks. + * + * Revision 1.125 2007/03/09 14:12:00 fabiankeil + * - Move null byte check into separate function. + * - Don't confuse the client with error pages + * if a CONNECT request was already confirmed. + * + * Revision 1.124 2007/02/23 14:59:54 fabiankeil + * Speed up NULL byte escaping and only log the complete + * NULL byte requests with header debugging enabled. + * + * Revision 1.123 2007/02/21 18:42:10 fabiankeil + * Answer requests that contain NULL bytes with + * a custom response instead of waiting for more + * data until the client eventually hangs up. + * + * Revision 1.122 2007/02/07 11:12:02 fabiankeil + * - Move delivery and logging of crunched responses + * from chat() into send_crunch_response(). + * - Display the reason for generating http_responses. + * - Log the content length for LOG_LEVEL_CLF correctly + * (still incorrect for some fixed responses). + * - Reword an incorrect comment about + * treat-forbidden-connects-like-blocks violating + * the specs. + * - Add some log messages. + * + * Revision 1.121 2007/01/27 10:52:56 fabiankeil + * Move mutex initialization into separate + * function and exit in case of errors. + * + * Revision 1.120 2007/01/26 14:18:42 fabiankeil + * - Start to reduce chat()'s line count and move + * parts of it into separate functions. + * - Add "HTTP/1.1 100 Continue" hack for BR 756734. + * + * Revision 1.119 2007/01/25 14:02:30 fabiankeil + * - Add Proxy-Agent header to HTTP snippets that are + * supposed to reach HTTP clients only. + * - Made a few CONNECT log messages more descriptive. + * - Catch completely empty server responses (as seen + * with Tor's fake ".noconnect" top level domain). + * - Use shiny new "forwarding-failed" template for socks errors. + * + * Revision 1.118 2007/01/07 07:43:43 joergs + * AmigaOS4 support added. + * + * Revision 1.117 2006/12/31 17:56:37 fabiankeil + * Added config option accept-intercepted-requests + * and disabled it by default. + * + * Revision 1.116 2006/12/29 19:08:22 fabiankeil + * Reverted parts of my last commit + * to keep error handling working. + * + * Revision 1.115 2006/12/29 17:38:57 fabiankeil + * Fixed gcc43 conversion warnings. + * + * Revision 1.114 2006/12/27 18:52:02 fabiankeil + * Fix -pedantic ISO C warning about converting + * from function pointer to object pointer. + * + * Revision 1.113 2006/12/26 17:38:50 fabiankeil + * Silence compiler warning I introduced with my last commit. + * + * Revision 1.112 2006/12/26 17:31:41 fabiankeil + * Mutex protect rand() if POSIX threading + * is used, warn the user if that's not possible + * and stop using it on _WIN32 where it could + * cause crashes. + * + * Revision 1.111 2006/12/23 16:15:06 fabiankeil + * Don't prevent core dumps by catching SIGABRT. + * It's rude and makes debugging unreasonable painful. + * + * Revision 1.110 2006/12/13 14:52:53 etresoft + * Fix build failure on MacOS X. Global symbols can be either static or extern, but not both. + * * Revision 1.109 2006/12/06 19:41:40 fabiankeil * Privoxy is now able to run as intercepting * proxy in combination with any packet filter @@ -713,6 +1113,7 @@ http://www.fabiankeil.de/sourcecode/privoxy/ #include #include #include +#include #ifdef _WIN32 # ifndef FEATURE_PTHREAD @@ -776,7 +1177,6 @@ http://www.fabiankeil.de/sourcecode/privoxy/ #include "filters.h" #include "loaders.h" #include "parsers.h" -#include "killpopup.h" #include "miscutil.h" #include "errlog.h" #include "jbsockets.h" @@ -802,8 +1202,28 @@ int urls_rejected = 0; /* total nr of urls rejected */ int g_terminate = 0; #endif -static void listen_loop(void); +#if !defined(_WIN32) && !defined(__OS2__) && !defined(AMIGA) +static void sig_handler(int the_signal); +#endif +static int client_protocol_is_unsupported(const struct client_state *csp, char *req); +static jb_err get_request_destination_elsewhere(struct client_state *csp, struct list *headers); +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 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); +static void serve(struct client_state *csp); +#if !defined(_WIN32) || defined(_WIN_CONSOLE) +static void usage(const char *myname); +#endif +static void initialize_mutexes(void); +static jb_socket bind_port_helper(struct configuration_spec *config); +static void listen_loop(void); + #ifdef AMIGA void serve(struct client_state *csp); #else /* ifndef AMIGA */ @@ -822,81 +1242,158 @@ 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) +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_GETHOSTBYADDR_R -pthread_mutex_t gethostbyaddr_mutex; -#endif /* ndef HAVE_GETHOSTBYADDR_R */ +#ifndef HAVE_RANDOM +privoxy_mutex_t rand_mutex; +#endif /* ndef HAVE_RANDOM */ -#ifndef HAVE_GETHOSTBYNAME_R -pthread_mutex_t gethostbyname_mutex; -#endif /* ndef HAVE_GETHOSTBYNAME_R */ -#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 */ -/* The vanilla wafer. */ -static const char VANILLA_WAFER[] = - "NOTICE=TO_WHOM_IT_MAY_CONCERN_" - "Do_not_send_me_any_copyrighted_information_other_than_the_" - "document_that_I_am_requesting_or_any_of_its_necessary_components._" - "In_particular_do_not_send_me_any_cookies_that_" - "are_subject_to_a_claim_of_copyright_by_anybody._" - "Take_notice_that_I_refuse_to_be_bound_by_any_license_condition_" - "(copyright_or_otherwise)_applying_to_any_cookie._"; - /* HTTP snipplets. */ -const char CSUCCEED[] = - "HTTP/1.0 200 Connection established\n" +static const char CSUCCEED[] = + "HTTP/1.0 200 Connection established\r\n" "Proxy-Agent: Privoxy/" VERSION "\r\n\r\n"; -const char CHEADER[] = - "HTTP/1.0 400 Invalid header received from browser\r\n" +static const char CHEADER[] = + "HTTP/1.0 400 Invalid header received from client\r\n" + "Proxy-Agent: Privoxy " VERSION "\r\n" + "Content-Type: text/plain\r\n" "Connection: close\r\n\r\n" - "Invalid header received from browser."; + "Invalid header received from client.\r\n"; -const char CFORBIDDEN[] = - "HTTP/1.0 403 Connection not allowable\r\n" - "X-Hint: If you read this message interactively, then you know why this happens ,-)\r\n" - "Connection: close\r\n\r\n"; - -const char FTP_RESPONSE[] = - "HTTP/1.0 400 Invalid request received from browser\r\n" +static const char FTP_RESPONSE[] = + "HTTP/1.0 400 Invalid request received from client\r\n" + "Content-Type: text/plain\r\n" "Connection: close\r\n\r\n" "Invalid request. Privoxy doesn't support FTP.\r\n"; -const char GOPHER_RESPONSE[] = - "HTTP/1.0 400 Invalid request received from browser\r\n" +static const char GOPHER_RESPONSE[] = + "HTTP/1.0 400 Invalid request received from client\r\n" + "Content-Type: text/plain\r\n" "Connection: close\r\n\r\n" "Invalid request. Privoxy doesn't support gopher.\r\n"; -const char MISSING_DESTINATION_RESPONSE[] = - "HTTP/1.0 400 Bad request received from browser\r\n" +/* XXX: should be a template */ +static const char MISSING_DESTINATION_RESPONSE[] = + "HTTP/1.0 400 Bad request received from client\r\n" + "Proxy-Agent: Privoxy " VERSION "\r\n" + "Content-Type: text/plain\r\n" "Connection: close\r\n\r\n" "Bad request. Privoxy was unable to extract the destination.\r\n"; +/* XXX: should be a template */ +static const char NO_SERVER_DATA_RESPONSE[] = + "HTTP/1.0 502 Server or forwarder response empty\r\n" + "Proxy-Agent: Privoxy " VERSION "\r\n" + "Content-Type: text/plain\r\n" + "Connection: close\r\n\r\n" + "Empty server or forwarder response.\r\n" + "The connection has been closed but Privoxy didn't receive any data.\r\n"; + +/* XXX: should be a template */ +static const char INVALID_SERVER_HEADERS_RESPONSE[] = + "HTTP/1.0 502 Server or forwarder response invalid\r\n" + "Proxy-Agent: Privoxy " VERSION "\r\n" + "Content-Type: text/plain\r\n" + "Connection: close\r\n\r\n" + "Bad response. The server or forwarder response doesn't look like HTTP.\r\n"; + +#if 0 +/* XXX: should be a template */ +static const char NULL_BYTE_RESPONSE[] = + "HTTP/1.0 400 Bad request received from client\r\n" + "Proxy-Agent: Privoxy " VERSION "\r\n" + "Content-Type: text/plain\r\n" + "Connection: close\r\n\r\n" + "Bad request. Null byte(s) before end of request.\r\n"; +#endif + +/* XXX: should be a template */ +static const char MESSED_UP_REQUEST_RESPONSE[] = + "HTTP/1.0 400 Malformed request after rewriting\r\n" + "Proxy-Agent: Privoxy " VERSION "\r\n" + "Content-Type: text/plain\r\n" + "Connection: close\r\n\r\n" + "Bad request. Messed up with header filters.\r\n"; + +/* A function to crunch a response */ +typedef struct http_response *(*crunch_func_ptr)(struct client_state *); + +/* Crunch function flags */ +#define CF_NO_FLAGS 0 +/* Cruncher applies to forced requests as well */ +#define CF_IGNORE_FORCE 1 +/* Crunched requests are counted for the block statistics */ +#define CF_COUNT_AS_REJECT 2 + +/* A crunch function and its flags */ +struct cruncher +{ + const crunch_func_ptr cruncher; + const int flags; +}; + +static int crunch_response_triggered(struct client_state *csp, const struct cruncher crunchers[]); + +/* Complete list of cruncher functions */ +static const struct cruncher crunchers_all[] = { + { direct_response, CF_COUNT_AS_REJECT|CF_IGNORE_FORCE}, + { block_url, CF_COUNT_AS_REJECT }, +#ifdef FEATURE_TRUST + { trust_url, CF_COUNT_AS_REJECT }, +#endif /* def FEATURE_TRUST */ + { redirect_url, CF_NO_FLAGS }, + { dispatch_cgi, CF_IGNORE_FORCE}, + { NULL, 0 } +}; + +/* Light version, used after tags are applied */ +static const struct cruncher crunchers_light[] = { + { block_url, CF_COUNT_AS_REJECT }, + { redirect_url, CF_NO_FLAGS }, + { NULL, 0 } +}; + +/* + * XXX: Don't we really mean + * + * #if defined(unix) + * + * here? + */ #if !defined(_WIN32) && !defined(__OS2__) && !defined(AMIGA) /********************************************************************* * * Function : sig_handler * * Description : Signal handler for different signals. - * Exit gracefully on ABRT, TERM and INT + * Exit gracefully on TERM and INT * or set a flag that will cause the errlog * to be reopened by the main thread on HUP. * @@ -910,7 +1407,6 @@ static void sig_handler(int the_signal) { switch(the_signal) { - case SIGABRT: case SIGTERM: case SIGINT: log_error(LOG_LEVEL_INFO, "exiting by signal %d .. bye", the_signal); @@ -924,7 +1420,9 @@ static void sig_handler(int the_signal) break; case SIGHUP: +#if defined(unix) received_hup_signal = 1; +#endif break; default: @@ -942,350 +1440,527 @@ static void sig_handler(int the_signal) /********************************************************************* * - * Function : chat + * Function : client_protocol_is_unsupported * - * 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. + * Description : Checks if the client used a known unsupported + * protocol and deals with it by sending an error + * response. * * Parameters : * 1 : csp = Current client state (buffers, headers, etc...) + * 2 : req = the first request line send by the client * - * Returns : On success, the number of bytes written are returned (zero - * indicates nothing was written). On error, -1 is returned, - * and errno is set appropriately. If count is zero and the - * file descriptor refers to a regular file, 0 will be - * returned without causing any other effect. For a special - * file, the results are not portable. + * Returns : TRUE if an error response has been generated, or + * FALSE if the request doesn't look invalid. * *********************************************************************/ -static void chat(struct client_state *csp) +static int client_protocol_is_unsupported(const struct client_state *csp, char *req) { -/* - * This next lines are a little ugly, but they simplifies the if statements - * below. Basically if TOGGLE, then we want the if to test if the - * CSP_FLAG_TOGGLED_ON flag ist set, else we don't. And if FEATURE_FORCE_LOAD, - * then we want the if to test for CSP_FLAG_FORCED , else we don't - */ -#ifdef FEATURE_TOGGLE -# define IS_TOGGLED_ON_AND (csp->flags & CSP_FLAG_TOGGLED_ON) && -#else /* ifndef FEATURE_TOGGLE */ -# define IS_TOGGLED_ON_AND -#endif /* ndef FEATURE_TOGGLE */ -#ifdef FEATURE_FORCE_LOAD -# define IS_NOT_FORCED_AND !(csp->flags & CSP_FLAG_FORCED) && -#else /* ifndef FEATURE_FORCE_LOAD */ -# define IS_NOT_FORCED_AND -#endif /* def FEATURE_FORCE_LOAD */ - -#define IS_ENABLED_AND IS_TOGGLED_ON_AND IS_NOT_FORCED_AND - - char buf[BUFFER_SIZE]; - char *hdr; - char *p; - char *req; - fd_set rfds; - int n; - jb_socket maxfd; - int server_body; - int ms_iis5_hack = 0; - int byte_count = 0; - unsigned int forwarded_connect_retries = 0; - unsigned int max_forwarded_connect_retries = csp->config->forwarded_connect_retries; - const struct forward_spec * fwd; - struct http_request *http; - int len; /* for buffer sizes */ -#ifdef FEATURE_KILL_POPUPS - int block_popups; /* bool, 1==will block popups */ - int block_popups_now = 0; /* bool, 1==currently blocking popups */ -#endif /* def FEATURE_KILL_POPUPS */ - - int pcrs_filter; /* bool, 1==will filter through pcrs */ - int gif_deanimate; /* bool, 1==will deanimate gifs */ - int jpeg_inspect; /* bool, 1==will inspect jpegs */ - - /* Function that does the content filtering for the current request */ - char *(*content_filter)() = NULL; - - /* Skeleton for HTTP response, if we should intercept the request */ - struct http_response *rsp; - - /* 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; - /* - * 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! + * If it's a FTP or gopher request, we don't support it. + * + * These checks are better than nothing, but they might + * not work in all configurations and some clients might + * have problems digesting the answer. + * + * They should, however, never cause more problems than + * Privoxy's old behaviour (returning the misleading HTML + * error message: + * + * "Could not resolve http://(ftp|gopher)://example.org"). */ - - for (;;) + if (!strncmpic(req, "GET ftp://", 10) || !strncmpic(req, "GET gopher://", 13)) { - len = read_socket(csp->cfd, buf, sizeof(buf)); + const char *response = NULL; + const char *protocol = NULL; - if (len <= 0) break; /* error! */ - - /* - * 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)) + if (!strncmpic(req, "GET ftp://", 10)) { - return; + response = FTP_RESPONSE; + protocol = "FTP"; } - - req = get_header(csp); - - if (req == NULL) + else { - break; /* no HTTP request! */ + response = GOPHER_RESPONSE; + protocol = "GOPHER"; } + log_error(LOG_LEVEL_ERROR, + "%s tried to use Privoxy as %s proxy: %s", + csp->ip_addr_str, protocol, req); + log_error(LOG_LEVEL_CLF, + "%s - - [%T] \"%s\" 400 0", csp->ip_addr_str, req); + freez(req); + write_socket(csp->cfd, response, strlen(response)); - if (*req == '\0') - { - continue; /* more to come! */ - } + return TRUE; + } - /* - * If it's a FTP or gopher request, we don't support it. - * - * These checks are better than nothing, but they might - * not work in all configurations and some clients might - * have problems digesting the answer. - * - * They should, however, never cause more problems than - * Privoxy's old behaviour (returning the misleading HTML error message: - * "Could not resolve http://(ftp|gopher)://example.org"). - */ - if (!strncmpic(req, "GET ftp://", 10) || !strncmpic(req, "GET gopher://", 13)) - { - if (!strncmpic(req, "GET ftp://", 10)) - { - strcpy(buf, FTP_RESPONSE); - log_error(LOG_LEVEL_ERROR, "%s tried to use Privoxy as FTP proxy: %s", - csp->ip_addr_str, req); - } - else - { - strcpy(buf, GOPHER_RESPONSE); - log_error(LOG_LEVEL_ERROR, "%s tried to use Privoxy as gopher proxy: %s", - csp->ip_addr_str, req); - } - log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" 400 0", csp->ip_addr_str, req); - freez(req); - write_socket(csp->cfd, buf, strlen(buf)); - free_http_request(http); - return; - } + return FALSE; +} -#ifdef FEATURE_FORCE_LOAD - /* If this request contains the FORCE_PREFIX, - * better get rid of it now and set the force flag --oes - */ - if (strstr(req, FORCE_PREFIX)) - { - strclean(req, FORCE_PREFIX); - log_error(LOG_LEVEL_FORCE, "Enforcing request \"%s\".\n", req); - csp->flags |= CSP_FLAG_FORCED; - } +/********************************************************************* + * + * Function : get_request_destination_elsewhere + * + * Description : If the client's request was redirected into + * Privoxy without the client's knowledge, + * the request line lacks the destination host. + * + * This function tries to get it elsewhere, + * provided accept-intercepted-requests is enabled. + * + * "Elsewhere" currently only means "Host: header", + * but in the future we may ask the redirecting + * packet filter to look the destination up. + * + * If the destination stays unknown, an error + * response is send to the client and headers + * are freed so that chat() can return directly. + * + * Parameters : + * 1 : csp = Current client state (buffers, headers, etc...) + * 2 : headers = a header list + * + * Returns : JB_ERR_OK if the destination is now known, or + * JB_ERR_PARSE if it isn't. + * + *********************************************************************/ +static jb_err get_request_destination_elsewhere(struct client_state *csp, struct list *headers) +{ + char *req; -#endif /* def FEATURE_FORCE_LOAD */ + if (!(csp->config->feature_flags & RUNTIME_FEATURE_ACCEPT_INTERCEPTED_REQUESTS)) + { + log_error(LOG_LEVEL_ERROR, "%s's request: \'%s\' is invalid." + " Privoxy isn't configured to accept intercepted requests.", + csp->ip_addr_str, csp->http->cmd); + /* XXX: Use correct size */ + log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" 400 0", + csp->ip_addr_str, csp->http->cmd); - switch( parse_http_request(req, http, csp) ) - { - case JB_ERR_MEMORY: - log_error(LOG_LEVEL_ERROR, "Out of memory while parsing request."); - break; - case JB_ERR_PARSE: - log_error(LOG_LEVEL_ERROR, "Couldn't parse request: %s.", req); - break; - } + write_socket(csp->cfd, CHEADER, strlen(CHEADER)); + destroy_list(headers); - freez(req); - break; + return JB_ERR_PARSE; } + else if (JB_ERR_OK == get_destination_from_headers(headers, csp->http)) + { + /* Split the domain we just got for pattern matching */ + init_domain_components(csp->http); - if (http->cmd == NULL) + return JB_ERR_OK; + } + else { - strcpy(buf, CHEADER); - write_socket(csp->cfd, buf, strlen(buf)); + /* We can't work without destination. Go spread the news.*/ + + req = list_to_text(headers); + chomp(req); + /* XXX: Use correct size */ + log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" 400 0", + csp->ip_addr_str, csp->http->cmd); + log_error(LOG_LEVEL_ERROR, + "Privoxy was unable to get the destination for %s's request:\n%s\n%s", + csp->ip_addr_str, csp->http->cmd, req); + freez(req); - log_error(LOG_LEVEL_CLF, "%s - - [%T] \" \" 400 0", csp->ip_addr_str); + write_socket(csp->cfd, MISSING_DESTINATION_RESPONSE, strlen(MISSING_DESTINATION_RESPONSE)); + destroy_list(headers); - free_http_request(http); - return; + return JB_ERR_PARSE; } + /* + * TODO: If available, use PF's ioctl DIOCNATLOOK as last resort + * to get the destination IP address, use it as host directly + * or do a reverse DNS lookup first. + */ +} - /* grab the rest of the client's headers */ - init_list(headers); - for (;;) + +/********************************************************************* + * + * Function : get_server_headers + * + * Description : Parses server headers in iob and fills them + * into csp->headers so that they can later be + * handled by sed(). + * + * Parameters : + * 1 : csp = Current client state (buffers, headers, etc...) + * + * Returns : JB_ERR_OK if everything went fine, or + * JB_ERR_PARSE if the headers were incomplete. + * + *********************************************************************/ +static jb_err get_server_headers(struct client_state *csp) +{ + int continue_hack_in_da_house = 0; + char * header; + + while (((header = get_header(csp->iob)) != NULL) || continue_hack_in_da_house) { - if ( ( ( p = get_header(csp) ) != NULL) && ( *p == '\0' ) ) + if (header == NULL) { - len = read_socket(csp->cfd, buf, sizeof(buf)); - if (len <= 0) - { - log_error(LOG_LEVEL_ERROR, "read from client failed: %E"); - return; - } - /* - * If there is no memory left for buffering the - * request, there is nothing we can do but hang up + * continue hack in da house. Ignore the ending of + * this head and continue enlisting header lines. + * The reason is described below. */ - if (add_to_iob(csp, buf, len)) - { - return; - } + enlist(csp->headers, ""); + continue_hack_in_da_house = 0; continue; } - - if (p == NULL) break; - - enlist(headers, p); - freez(p); - - } - - if (http->host == NULL) - { - /* - * Intercepted or invalid request without domain - * inside the request line. Try to get it another way. - */ - if (JB_ERR_OK == get_destination_from_headers(headers, http)) + else if (0 == strncmpic(header, "HTTP/1.1 100", 12)) { - /* Split the domain we just got for pattern matching */ - init_domain_components(http); + /* + * It's a bodyless continue response, don't + * stop header parsing after reaching its end. + * + * As a result Privoxy will concatenate the + * next response's head and parse and deliver + * the headers as if they belonged to one request. + * + * The client will separate them because of the + * empty line between them. + * + * XXX: What we're doing here is clearly against + * the intended purpose of the continue header, + * and under some conditions (HTTP/1.0 client request) + * it's a standard violation. + * + * Anyway, "sort of against the spec" is preferable + * to "always getting confused by Continue responses" + * (Privoxy's behaviour before this hack was added) + */ + log_error(LOG_LEVEL_HEADER, "Continue hack in da house."); + continue_hack_in_da_house = 1; } - else + else if (*header == '\0') { - /* We can't work without destination. Go spread the news.*/ + /* + * If the header is empty, but the Continue hack + * isn't active, we can assume that we reached the + * end of the buffer before we hit the end of the + * head. + * + * Inform the caller an let it decide how to handle it. + */ + return JB_ERR_PARSE; + } - req = list_to_text(headers); - chomp(req); - log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" 400 0", csp->ip_addr_str, http->cmd); + if (JB_ERR_MEMORY == enlist(csp->headers, header)) + { + /* + * XXX: Should we quit the request and return a + * out of memory error page instead? + */ log_error(LOG_LEVEL_ERROR, - "Privoxy was unable to get the destination for %s's request:\n%s\n%s", - csp->ip_addr_str, http->cmd, req); - freez(req); - - strcpy(buf, MISSING_DESTINATION_RESPONSE); - write_socket(csp->cfd, buf, strlen(buf)); - free_http_request(http); - destroy_list(headers); - return; + "Out of memory while enlisting server headers. %s lost.", + header); } - /* - * TODO: If available, use PF's ioctl DIOCNATLOOK as last resort - * to get the destination IP address, use it as host directly - * or do a reverse DNS lookup first. - */ + freez(header); } - /* decide how to route the HTTP request */ + return JB_ERR_OK; +} + + +/********************************************************************* + * + * Function : crunch_reason + * + * Description : Translates the crunch reason code into a string. + * + * Parameters : + * 1 : rsp = a http_response + * + * Returns : A string with the crunch reason or an error description. + * + *********************************************************************/ +static const char *crunch_reason(const struct http_response *rsp) +{ + char * reason = NULL; - if ((fwd = forward_url(http, csp)) == NULL) + assert(rsp != NULL); + if (rsp == NULL) { - log_error(LOG_LEVEL_FATAL, "gateway spec is NULL!?!? This can't happen!"); - /* Never get here - LOG_LEVEL_FATAL causes program exit */ + return "Internal error while searching for crunch reason"; } - /* 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 | | - * +--------+--------+ - * - */ - - /* - * Determine the actions for this URL - */ -#ifdef FEATURE_TOGGLE - if (!(csp->flags & CSP_FLAG_TOGGLED_ON)) + switch (rsp->reason) { - /* Most compatible set of actions (i.e. none) */ - init_current_action(csp->action); + case RSP_REASON_UNSUPPORTED: + reason = "Unsupported HTTP feature"; + break; + case RSP_REASON_BLOCKED: + reason = "Blocked"; + break; + case RSP_REASON_UNTRUSTED: + reason = "Untrusted"; + break; + case RSP_REASON_REDIRECTED: + reason = "Redirected"; + break; + case RSP_REASON_CGI_CALL: + reason = "CGI Call"; + break; + case RSP_REASON_NO_SUCH_DOMAIN: + reason = "DNS failure"; + break; + case RSP_REASON_FORWARDING_FAILED: + reason = "Forwarding failed"; + break; + case RSP_REASON_CONNECT_FAILED: + reason = "Connection failure"; + break; + case RSP_REASON_OUT_OF_MEMORY: + reason = "Out of memory (may mask other reasons)"; + break; + default: + reason = "No reason recorded"; + break; } - else -#endif /* ndef FEATURE_TOGGLE */ + + return reason; +} + + +/********************************************************************* + * + * Function : send_crunch_response + * + * Description : Delivers already prepared response for + * intercepted requests, logs the interception + * and frees the response. + * + * Parameters : + * 1 : csp = Current client state (buffers, headers, etc...) + * 1 : rsp = Fully prepared response. Will be freed on exit. + * + * Returns : Nothing. + * + *********************************************************************/ +static void send_crunch_response(const struct client_state *csp, struct http_response *rsp) +{ + const struct http_request *http = csp->http; + char status_code[4]; + + assert(rsp != NULL); + assert(rsp->head != NULL); + + if (rsp == NULL) + { + /* + * Not supposed to happen. If it does + * anyway, treat it as an unknown error. + */ + cgi_error_unknown(csp, rsp, RSP_REASON_INTERNAL_ERROR); + /* return code doesn't matter */ + } + + if (rsp == NULL) + { + /* If rsp is still NULL, we have serious internal problems. */ + log_error(LOG_LEVEL_FATAL, + "NULL response in send_crunch_response and cgi_error_unknown failed as well."); + } + + /* + * Extract the status code from the actual head + * that was send to the client. It is the only + * way to get it right for all requests, including + * the fixed ones for out-of-memory problems. + * + * A head starts like this: 'HTTP/1.1 200...' + * 0123456789|11 + * 10 + */ + status_code[0] = rsp->head[9]; + status_code[1] = rsp->head[10]; + status_code[2] = rsp->head[11]; + status_code[3] = '\0'; + + /* Write the answer to the client */ + if (write_socket(csp->cfd, rsp->head, rsp->head_length) + || write_socket(csp->cfd, rsp->body, rsp->content_length)) + { + /* There is nothing we can do about it. */ + log_error(LOG_LEVEL_ERROR, "write to: %s failed: %E", csp->http->host); + } + + /* Log that the request was crunched and why. */ + log_error(LOG_LEVEL_GPC, "%s%s crunch! (%s)", + http->hostport, http->path, crunch_reason(rsp)); + log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" %s %d", + csp->ip_addr_str, http->ocmd, status_code, rsp->content_length); + + /* Clean up and return */ + if (cgi_error_memory() != rsp) + { + free_http_response(rsp); + } + return; +} + + +#if 0 +/********************************************************************* + * + * Function : request_contains_null_bytes + * + * Description : Checks for NULL bytes in the request and sends + * an error message to the client if any were found. + * + * XXX: currently not used, see comment in chat(). + * + * Parameters : + * 1 : csp = Current client state (buffers, headers, etc...) + * 2 : buf = Data from the client's request to check. + * 3 : len = The data length. + * + * Returns : TRUE if the request contained one or more NULL bytes, or + * FALSE otherwise. + * + *********************************************************************/ +static int request_contains_null_bytes(const struct client_state *csp, char *buf, int len) +{ + size_t c_len; /* Request lenght when treated as C string */ + + c_len = strlen(buf); + + if (c_len < len) { - url_actions(http, csp); + /* + * Null byte(s) found. Log the request, + * return an error response and hang up. + */ + size_t tmp_len = c_len; + + do + { + /* + * Replace NULL byte(s) with '°' characters + * so the request can be logged as string. + * XXX: Is there a better replacement character? + */ + buf[tmp_len]='°'; + tmp_len += strlen(buf+tmp_len); + } while (tmp_len < len); + + log_error(LOG_LEVEL_ERROR, "%s\'s request contains at least one NULL byte " + "(length=%d, strlen=%d).", csp->ip_addr_str, len, c_len); + log_error(LOG_LEVEL_HEADER, + "Offending request data with NULL bytes turned into \'°\' characters: %s", buf); + + write_socket(csp->cfd, NULL_BYTE_RESPONSE, strlen(NULL_BYTE_RESPONSE)); + + /* XXX: Log correct size */ + log_error(LOG_LEVEL_CLF, "%s - - [%T] \"Invalid request\" 400 0", csp->ip_addr_str); + + return TRUE; } + return FALSE; +} +#endif + + +/********************************************************************* + * + * Function : crunch_response_triggered + * + * Description : Checks if the request has to be crunched, + * and delivers the crunch response if necessary. + * + * Parameters : + * 1 : csp = Current client state (buffers, headers, etc...) + * 2 : crunchers = list of cruncher functions to run + * + * Returns : TRUE if the request was answered with a crunch response + * FALSE otherwise. + * + *********************************************************************/ +static int crunch_response_triggered(struct client_state *csp, const struct cruncher crunchers[]) +{ + struct http_response *rsp = NULL; + const struct cruncher *c; /* - * Check if a CONNECT request is allowable: - * In the absence of a +limit-connect action, allow only port 443. - * If there is an action, allow whatever matches the specificaton. + * If CGI request crunching is disabled, + * check the CGI dispatcher out of order to + * prevent unintentional blocks or redirects. */ - if(http->ssl) + 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++) { - if( ( !(csp->action->flags & ACTION_LIMIT_CONNECT) && csp->http->port != 443) - || (csp->action->flags & ACTION_LIMIT_CONNECT - && !match_portlist(csp->action->string[ACTION_STRING_LIMIT_CONNECT], csp->http->port)) ) + /* + * 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)) { - if (csp->action->flags & ACTION_TREAT_FORBIDDEN_CONNECTS_LIKE_BLOCKS) - { - /* The response will violate the specs, but makes unblocking easier. */ - log_error(LOG_LEVEL_ERROR, "Marking suspicious CONNECT request from %s for blocking.", - csp->ip_addr_str); - csp->action->flags |= ACTION_BLOCK; - http->ssl = 0; - } - else + rsp = c->cruncher(csp); + if (NULL != rsp) { - strcpy(buf, CFORBIDDEN); - write_socket(csp->cfd, buf, strlen(buf)); - log_error(LOG_LEVEL_CONNECT, "Denying suspicious CONNECT request from %s", csp->ip_addr_str); - log_error(LOG_LEVEL_CLF, "%s - - [%T] \" \" 403 0", csp->ip_addr_str); - return; + /* 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 + * Downgrade http version from 1.1 to 1.0 + * if +downgrade action applies. */ - if ( (http->ssl == 0) - && (!strcmpic(http->ver, "HTTP/1.1")) - && (csp->action->flags & ACTION_DOWNGRADE)) + if ( (csp->action->flags & ACTION_DOWNGRADE) + && (!strcmpic(http->ver, "HTTP/1.1"))) { freez(http->ver); http->ver = strdup("HTTP/1.0"); @@ -1296,43 +1971,368 @@ static void chat(struct client_state *csp) } } - /* - * Save a copy of the original request for logging + /* + * Rebuild the request line. */ - http->ocmd = strdup(http->cmd); + freez(*request_line); + *request_line = strdup(http->gpc); + string_append(request_line, " "); - if (http->ocmd == NULL) + if (fwd->forward_host) + { + string_append(request_line, http->url); + } + else { - log_error(LOG_LEVEL_FATAL, "Out of memory copying HTTP request line"); + string_append(request_line, http->path); } + string_append(request_line, " "); + string_append(request_line, http->ver); - /* - * (Re)build the HTTP request for non-SSL requests. - * If forwarding, use the whole URL, else, use only the path. - */ - if (http->ssl == 0) + 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) { - freez(http->cmd); + 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); - http->cmd = strdup(http->gpc); - string_append(&http->cmd, " "); + 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 (fwd->forward_host) + if (p == NULL) { - string_append(&http->cmd, http->url); + /* 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 { - string_append(&http->cmd, http->path); + /* + * We were able to read a complete + * header and can finaly enlist it. + */ + enlist(headers, p); + freez(p); } - string_append(&http->cmd, " "); - string_append(&http->cmd, http->ver); + } - if (http->cmd == NULL) + 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)) { - log_error(LOG_LEVEL_FATAL, "Out of memory writing HTTP command"); + /* + * 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; } - log_error(LOG_LEVEL_HEADER, "New HTTP Request-Line: %s", http->cmd); + } + + /* + * 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); @@ -1340,90 +2340,214 @@ static void chat(struct client_state *csp) list_append_list_unique(csp->headers, headers); destroy_list(headers); - /* - * If the user has not supplied any wafers, and the user has not - * told us to suppress the vanilla wafer, then send the vanilla wafer. - */ - if (list_is_empty(csp->action->multi[ACTION_MULTI_WAFER]) - && ((csp->action->flags & ACTION_VANILLA_WAFER) != 0)) + 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) { - enlist(csp->action->multi[ACTION_MULTI_WAFER], VANILLA_WAFER); + /* 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); -#ifdef FEATURE_KILL_POPUPS - block_popups = ((csp->action->flags & ACTION_NO_POPUPS) != 0); -#endif /* def FEATURE_KILL_POPUPS */ - - pcrs_filter = (csp->rlist != NULL) && /* There are expressions to be used */ - (!list_is_empty(csp->action->multi[ACTION_MULTI_FILTER])); + return JB_ERR_PARSE; + } - gif_deanimate = ((csp->action->flags & ACTION_DEANIMATE) != 0); + return JB_ERR_OK; - jpeg_inspect = ((csp->action->flags & ACTION_JPEG_INSPECT) != 0); +} - /* - * We have a request. Now, check to see if we need to - * intercept it, i.e. If .. - */ - if ( - /* We may not forward the request by rfc2616 sect 14.31 */ - (NULL != (rsp = direct_response(csp))) +/********************************************************************* + * + * 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) */ - /* or we are enabled and... */ - || (IS_ENABLED_AND ( + /* Function that does the content filtering for the current request */ + filter_function_ptr content_filter = NULL; - /* ..the request was blocked */ - ( NULL != (rsp = block_url(csp))) + /* Skeleton for HTTP response, if we should intercept the request */ + struct http_response *rsp; - /* ..or untrusted */ -#ifdef FEATURE_TRUST - || ( NULL != (rsp = trust_url(csp))) -#endif /* def FEATURE_TRUST */ + memset(buf, 0, sizeof(buf)); - /* ..or a redirect kicked in */ - || ( NULL != (rsp = redirect_url(csp))) - )) + http = csp->http; - /* - * .. or a CGI call was detected and answered. - * - * This check comes last to give the user the power - * to deny acces to some (or all) of the cgi pages. - */ - || (NULL != (rsp = dispatch_cgi(csp))) + 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) { - /* Write the answer to the client */ - if (write_socket(csp->cfd, rsp->head, rsp->head_length) - || write_socket(csp->cfd, rsp->body, rsp->content_length)) - { - log_error(LOG_LEVEL_ERROR, "write to: %s failed: %E", http->host); - } + log_error(LOG_LEVEL_FATAL, "gateway spec is NULL!?!? This can't happen!"); + /* Never get here - LOG_LEVEL_FATAL causes program exit */ + } -#ifdef FEATURE_STATISTICS - /* Count as a rejected request */ - csp->flags |= CSP_FLAG_REJECTED; -#endif /* def FEATURE_STATISTICS */ + /* + * 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 | | + * +--------+--------+ + * + */ - /* Log (FIXME: All intercept reasons appear as "crunch" with Status 200) */ - log_error(LOG_LEVEL_GPC, "%s%s crunch!", http->hostport, http->path); - log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" 200 3", csp->ip_addr_str, http->ocmd); + 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; + } - /* Clean up and return */ - free_http_response(rsp); - return; + if (http->ssl == 0) + { + freez(csp->headers->first->str); + build_request_line(csp, fwd, &csp->headers->first->str); } - hdr = sed(client_patterns, add_client_headers, csp); + 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); @@ -1449,69 +2573,50 @@ static void chat(struct client_state *csp) if (csp->sfd == JB_INVALID_SOCKET) { - log_error(LOG_LEVEL_CONNECT, "connect to: %s failed: %E", - http->hostport); - - if (errno == EINVAL) + 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); - - log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" 404 0", - csp->ip_addr_str, http->ocmd); } else { rsp = error_response(csp, "connect-failed", errno); - - log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" 503 0", - csp->ip_addr_str, http->ocmd); + log_error(LOG_LEVEL_CONNECT, "connect to: %s failed: %E", + http->hostport); } - /* Write the answer to the client */ - if(rsp) + if (rsp != NULL) { - if (write_socket(csp->cfd, rsp->head, rsp->head_length) - || write_socket(csp->cfd, rsp->body, rsp->content_length)) - { - log_error(LOG_LEVEL_ERROR, "write to: %s failed: %E", http->host); - } + send_crunch_response(csp, rsp); } - free_http_response(rsp); freez(hdr); return; } - log_error(LOG_LEVEL_CONNECT, "OK"); - 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) < 0)) + || (flush_socket(csp->sfd, csp->iob) < 0)) { - log_error(LOG_LEVEL_CONNECT, "write header to: %s failed: %E", - http->hostport); - - log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" 503 0", - csp->ip_addr_str, http->ocmd); + log_error(LOG_LEVEL_CONNECT, + "write header to: %s failed: %E", http->hostport); rsp = error_response(csp, "connect-failed", errno); - - if(rsp) + if (rsp) { - if (write_socket(csp->cfd, rsp->head, rsp->head_length) - || write_socket(csp->cfd, rsp->body, rsp->content_length)) - { - log_error(LOG_LEVEL_ERROR, "write to: %s failed: %E", http->host); - } + send_crunch_response(csp, rsp); } - free_http_response(rsp); freez(hdr); return; } @@ -1523,10 +2628,7 @@ static void chat(struct client_state *csp) * so just send the "connect succeeded" message to the * client, flush the rest, and get out of the way. */ - log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" 200 2\n", - csp->ip_addr_str, http->ocmd); - - if (write_socket(csp->cfd, CSUCCEED, sizeof(CSUCCEED)-1)) + if (write_socket(csp->cfd, CSUCCEED, strlen(CSUCCEED))) { freez(hdr); return; @@ -1534,10 +2636,12 @@ static void chat(struct client_state *csp) 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; + 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. @@ -1559,83 +2663,128 @@ static void chat(struct client_state *csp) 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 it and write it. + /* + * 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)); + 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) { log_error(LOG_LEVEL_ERROR, "read from: %s failed: %E", http->host); - log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" 503 0", - csp->ip_addr_str, http->ocmd); + 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) + if (rsp) { - if (write_socket(csp->cfd, rsp->head, rsp->head_length) - || write_socket(csp->cfd, rsp->body, rsp->content_length)) - { - log_error(LOG_LEVEL_ERROR, "write to: %s failed: %E", http->host); - } + send_crunch_response(csp, rsp); } - free_http_response(rsp); return; } - /* Add a trailing zero. This lets filter_popups - * use string operations. - */ - buf[len] = '\0'; - -#ifdef FEATURE_KILL_POPUPS - /* Filter the popups on this read. */ - if (block_popups_now) +#ifdef FEATURE_CONNECTION_KEEP_ALIVE + if (csp->flags & CSP_FLAG_CHUNKED) { - filter_popups(buf, csp); + 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; + } } -#endif /* def FEATURE_KILL_POPUPS */ + 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 + /* + * 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 @@ -1665,18 +2814,24 @@ static void chat(struct client_state *csp) */ 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 = (*content_filter)(csp))) + if (NULL == p) { - csp->content_length = csp->iob->eod - csp->iob->cur; + csp->content_length = (size_t)(csp->iob->eod - csp->iob->cur); } - hdr = sed(server_patterns_light, NULL, csp); + 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 */ @@ -1689,6 +2844,7 @@ static void chat(struct client_state *csp) log_error(LOG_LEVEL_ERROR, "write modified content to client failed: %E"); freez(hdr); freez(p); + mark_server_socket_tainted(csp); return; } @@ -1703,7 +2859,8 @@ static void chat(struct client_state *csp) * This is NOT the body, so * Let's pretend the server just sent us a blank line. */ - len = sprintf(buf, "\r\n"); + snprintf(buf, sizeof(buf), "\r\n"); + len = (int)strlen(buf); /* * Now, let the normal header parsing algorithm below do its @@ -1718,7 +2875,6 @@ 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) @@ -1733,9 +2889,10 @@ static void chat(struct client_state *csp) size_t hdrlen; int flushed; - log_error(LOG_LEVEL_ERROR, "Flushing header and buffers. Stepping back from filtering."); + log_error(LOG_LEVEL_INFO, + "Flushing header and buffers. Stepping back from filtering."); - hdr = sed(server_patterns, add_server_headers, csp); + hdr = list_to_text(csp->headers); if (hdr == NULL) { /* @@ -1744,32 +2901,32 @@ static void chat(struct client_state *csp) */ log_error(LOG_LEVEL_ERROR, "Out of memory while trying to flush."); rsp = cgi_error_memory(); - - if (write_socket(csp->cfd, rsp->head, rsp->head_length) - || write_socket(csp->cfd, rsp->body, rsp->content_length)) - { - log_error(LOG_LEVEL_ERROR, "write to: %s failed: %E", http->host); - } + 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)) < 0) - || (write_socket(csp->cfd, buf, (size_t) len))) + || ((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; } - byte_count += hdrlen + flushed + len; + /* + * 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 @@ -1777,158 +2934,161 @@ 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 += len; + byte_count += (size_t)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(); - - if (write_socket(csp->cfd, rsp->head, rsp->head_length) - || write_socket(csp->cfd, rsp->body, rsp->content_length)) - { - log_error(LOG_LEVEL_ERROR, "write to: %s failed: %E", http->host); - } + send_crunch_response(csp, rsp); + mark_server_socket_tainted(csp); return; } - /* get header lines from the iob */ - - while ((p = get_header(csp)) != NULL) - { - if (*p == '\0') - { - /* see following note */ - break; - } - enlist(csp->headers, p); - freez(p); - } - - /* NOTE: there are no "empty" headers so - * if the pointer `p' is not NULL we must - * assume that we reached the end of the - * buffer before we hit the end of the header. - */ + header_start = csp->iob->cur; - if (p) + /* 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; } 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. */ continue; } } - /* we have now received the entire header. - * filter it and send the result to the client - */ - - hdr = sed(server_patterns, add_server_headers, csp); - if (hdr == NULL) + /* Did we actually get anything? */ + if (NULL == csp->headers->first) { - /* FIXME Should handle error properly */ - log_error(LOG_LEVEL_FATAL, "Out of memory parsing server header"); + 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; } -#ifdef FEATURE_KILL_POPUPS - /* Start blocking popups if appropriate. */ - - if ((csp->content_type & CT_TEXT) && /* It's a text / * MIME-Type */ - !http->ssl && /* We talk plaintext */ - block_popups) /* Policy allows */ + assert(csp->headers->first->str); + assert(!http->ssl); + if (strncmpic(csp->headers->first->str, "HTTP", 4) && + strncmpic(csp->headers->first->str, "ICY", 3)) { - block_popups_now = 1; /* - * Filter the part of the body that came in the same read - * as the last headers: + * It doesn't look like a HTTP (or Shoutcast) response: + * tell the client and log the problem. */ - filter_popups(csp->iob->cur, csp); + 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; } -#endif /* def FEATURE_KILL_POPUPS */ - - /* Buffer and pcrs filter this if appropriate. */ - - if ((csp->content_type & CT_TEXT) && /* It's a text / * MIME-Type */ - !http->ssl && /* We talk plaintext */ - pcrs_filter) /* Policy allows */ + /* + * 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)) { - content_filter = pcrs_filter_response; + log_error(LOG_LEVEL_FATAL, "Failed to parse server headers."); } - - /* Buffer and gif_deanimate this if appropriate. */ - - if ((csp->content_type & CT_GIF) && /* It's an image/gif MIME-Type */ - !http->ssl && /* We talk plaintext */ - gif_deanimate) /* Policy allows */ + hdr = list_to_text(csp->headers); + if (hdr == NULL) { - content_filter = gif_deanimate_response; + /* FIXME Should handle error properly */ + log_error(LOG_LEVEL_FATAL, "Out of memory parsing server header"); } - /* Buffer and jpg_inspect this if appropriate. */ - - if ((csp->content_type & CT_JPEG) && /* It's an image/jpeg MIME-Type */ - !http->ssl && /* We talk plaintext */ - jpeg_inspect) /* Policy allows */ + if (crunch_response_triggered(csp, crunchers_light)) { - content_filter = jpeg_inspect_response; + /* + * 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 + /* + * 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)) < 0)) + || ((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. + /* + * 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 += len; + 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 */ @@ -1936,24 +3096,44 @@ 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; } } 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. + */ + csp->content_length = byte_count; + } + + if ((csp->flags & CSP_FLAG_CONTENT_LENGTH_SET) + && (csp->expected_content_length != byte_count)) + { + log_error(LOG_LEVEL_ERROR, + "Received %d bytes while expecting %d.", + byte_count, csp->expected_content_length); + mark_server_socket_tainted(csp); + } + log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" 200 %d", - csp->ip_addr_str, http->ocmd, byte_count); + csp->ip_addr_str, http->ocmd, csp->content_length); } @@ -1981,7 +3161,20 @@ static void serve(struct client_state *csp) if (csp->sfd != JB_INVALID_SOCKET) { +#ifdef FEATURE_CONNECTION_KEEP_ALIVE + if ((csp->config->feature_flags & RUNTIME_FEATURE_CONNECTION_KEEP_ALIVE) + && (csp->flags & CSP_FLAG_SERVER_CONNECTION_KEEP_ALIVE)) + { + remember_connection(csp->sfd, csp->http, forward_url(csp, csp->http)); + } + else + { + forget_connection(csp->sfd); + close_socket(csp->sfd); + } +#else close_socket(csp->sfd); +#endif /* def FEATURE_CONNECTION_KEEP_ALIVE */ } csp->flags &= ~CSP_FLAG_ACTIVE; @@ -2011,6 +3204,7 @@ static int32 server_thread(void *data) #endif +#if !defined(_WIN32) || defined(_WIN_CONSOLE) /********************************************************************* * * Function : usage @@ -2022,19 +3216,166 @@ static int32 server_thread(void *data) * Returns : No. ,-) * *********************************************************************/ -void usage(const char *myname) +static void usage(const char *myname) { printf("Privoxy version " VERSION " (" HOME_PAGE_URL ")\n" -#if !defined(unix) - "Usage: %s [--help] [--version] [configfile]\n" -#else - "Usage: %s [--help] [--version] [--no-daemon] [--pidfile pidfile] [--user user[.group]] [configfile]\n" -#endif - "Aborting.\n", myname); - + "Usage: %s " +#if defined(unix) + "[--chroot] " +#endif /* defined(unix) */ + "[--help] " +#if defined(unix) + "[--no-daemon] [--pidfile pidfile] [--pre-chroot-nslookup hostname] [--user user[.group]] " +#endif /* defined(unix) */ + "[--version] [configfile]\n" + "Aborting\n", myname); + exit(2); } +#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 + * + * Description : Prepares mutexes if mutex support is available. + * + * Parameters : None + * + * Returns : Void, exits in case of errors. + * + *********************************************************************/ +static void initialize_mutexes(void) +{ +#ifdef MUTEX_LOCKS_AVAILABLE + /* + * Prepare global mutex semaphores + */ + 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 + * and can cause locks that aren't necessary. + * + * For example older FreeBSD versions (< 6.x?) + * have no gethostbyname_r, but gethostbyname is + * thread safe. + */ +#if !defined(HAVE_GETHOSTBYADDR_R) || !defined(HAVE_GETHOSTBYNAME_R) + 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 + privoxy_mutex_init(&gmtime_mutex); +#endif /* ndef HAVE_GMTIME_R */ + +#ifndef HAVE_LOCALTIME_R + privoxy_mutex_init(&localtime_mutex); +#endif /* ndef HAVE_GMTIME_R */ + +#ifndef HAVE_RANDOM + privoxy_mutex_init(&rand_mutex); +#endif /* ndef HAVE_RANDOM */ +#endif /* def MUTEX_LOCKS_AVAILABLE */ +} /********************************************************************* @@ -2067,14 +3408,13 @@ int main(int argc, const char *argv[]) #endif { int argc_pos = 0; -#ifdef HAVE_RANDOM unsigned int random_seed; -#endif /* ifdef HAVE_RANDOM */ #ifdef unix struct passwd *pw = NULL; struct group *grp = NULL; char *p; int do_chroot = 0; + char *pre_chroot_nslookup_to_load_resolver = NULL; #endif Argc = argc; @@ -2090,6 +3430,9 @@ int main(int argc, const char *argv[]) /* * Parse the command line arguments + * + * XXX: simply printing usage information in case of + * invalid arguments isn't particularly user friendly. */ while (++argc_pos < argc) { @@ -2166,12 +3509,29 @@ int main(int argc, const char *argv[]) if (p != NULL) *--p = '\0'; } + else if (strcmp(argv[argc_pos], "--pre-chroot-nslookup" ) == 0) + { + if (++argc_pos == argc) usage(argv[0]); + pre_chroot_nslookup_to_load_resolver = strdup(argv[argc_pos]); + } + else if (strcmp(argv[argc_pos], "--chroot" ) == 0) { do_chroot = 1; } - #endif /* defined(unix) */ + + else if (argc_pos + 1 != argc) + { + /* + * This is neither the last command line + * option, nor was it recognized before, + * therefore it must be invalid. + */ + usage(argv[0]); + } + else + #endif /* defined(_WIN32) && !defined(_WIN_CONSOLE) */ { configfile = argv[argc_pos]; @@ -2182,24 +3542,30 @@ int main(int argc, const char *argv[]) #if defined(unix) if ( *configfile != '/' ) { - char *abs_file, cwd[1024]; + char cwd[BUFFER_SIZE]; + char *abs_file; + size_t abs_file_size; /* make config-filename absolute here */ - if ( !(getcwd(cwd, sizeof(cwd)))) + if (NULL == getcwd(cwd, sizeof(cwd))) { - perror("get working dir failed"); + perror("failed to get current working directory"); exit( 1 ); } - if (!(basedir = strdup(cwd)) - || (!(abs_file = malloc( strlen( basedir ) + strlen( configfile ) + 5 )))) + /* XXX: why + 5? */ + abs_file_size = strlen(cwd) + strlen(configfile) + 5; + basedir = strdup(cwd); + + if (NULL == basedir || + NULL == (abs_file = malloc(abs_file_size))) { perror("malloc failed"); exit( 1 ); } - strcpy( abs_file, basedir ); - strcat( abs_file, "/" ); - strcat( abs_file, configfile ); + strlcpy(abs_file, basedir, abs_file_size); + strlcat(abs_file, "/", abs_file_size ); + strlcat(abs_file, configfile, abs_file_size); configfile = abs_file; } #endif /* defined unix */ @@ -2208,39 +3574,24 @@ int main(int argc, const char *argv[]) files->next = NULL; clients->next = NULL; + /* XXX: factor out initialising after the next stable release. */ #ifdef AMIGA InitAmiga(); #elif defined(_WIN32) InitWin32(); #endif -#ifdef FEATURE_PTHREAD - /* - * Prepare global mutex semaphores - */ - pthread_mutex_init(&log_mutex,0); - pthread_mutex_init(&log_init_mutex,0); - -#ifndef HAVE_GMTIME_R - pthread_mutex_init(&gmtime_mutex,0); -#endif /* ndef HAVE_GMTIME_R */ - -#ifndef HAVE_LOCALTIME_R - pthread_mutex_init(&localtime_mutex,0); -#endif /* ndef HAVE_GMTIME_R */ - -#ifndef HAVE_GETHOSTBYADDR_R - pthread_mutex_init(&gethostbyaddr_mutex,0); -#endif /* ndef HAVE_GETHOSTBYADDR_R */ + /* Prepare mutexes if supported and necessary. */ + initialize_mutexes(); -#ifndef HAVE_GETHOSTBYNAME_R - pthread_mutex_init(&gethostbyname_mutex,0); -#endif /* ndef HAVE_GETHOSTBYNAME_R */ -#endif /* FEATURE_PTHREAD */ + /* Enable logging until further notice. */ + init_log_module(Argv[0]); -#ifdef HAVE_RANDOM random_seed = (unsigned int)time(NULL); +#ifdef HAVE_RANDOM srandom(random_seed); +#else + srand(random_seed); #endif /* ifdef HAVE_RANDOM */ /* @@ -2253,7 +3604,7 @@ int main(int argc, const char *argv[]) #if !defined(_WIN32) && !defined(__OS2__) && !defined(AMIGA) { int idx; - const int catched_signals[] = { SIGABRT, SIGTERM, SIGINT, SIGHUP, 0 }; + const int catched_signals[] = { SIGTERM, SIGINT, SIGHUP, 0 }; const int ignored_signals[] = { SIGPIPE, 0 }; for (idx = 0; catched_signals[idx] != 0; idx++) @@ -2283,7 +3634,7 @@ int main(int argc, const char *argv[]) * We *are* in a windows console app. * Print a verbose messages about FAQ's and such */ - printf(win32_blurb); + printf("%s", win32_blurb); # endif /* def _WIN_CONSOLE */ #endif /* def _WIN32 */ @@ -2346,10 +3697,9 @@ int main(int argc, const char *argv[]) close ( fd ); } #endif /* 1 */ - /* FIXME: should close stderr (fd 2) here too, but the test - * for existence - * and load config file is done in listen_loop() and puts - * some messages on stderr there. + /* + * stderr (fd 2) will be closed later on, when the + * log file has been parsed. */ close( 0 ); @@ -2376,6 +3726,14 @@ int main(int argc, const char *argv[]) { log_error(LOG_LEVEL_FATAL, "Home directory for %s undefined", pw->pw_name); } + /* Read the time zone file from /etc before doing chroot. */ + tzset(); + if (NULL != pre_chroot_nslookup_to_load_resolver + && '\0' != pre_chroot_nslookup_to_load_resolver[0]) + { + /* Initialize resolver library. */ + (void) resolve_hostname_to_ip(pre_chroot_nslookup_to_load_resolver); + } if (chroot(pw->pw_dir) < 0) { log_error(LOG_LEVEL_FATAL, "Cannot chroot to %s", pw->pw_dir); @@ -2393,13 +3751,13 @@ int main(int argc, const char *argv[]) { char putenv_dummy[64]; - strcpy(putenv_dummy, "HOME=/"); + strlcpy(putenv_dummy, "HOME=/", sizeof(putenv_dummy)); if (putenv(putenv_dummy) != 0) { log_error(LOG_LEVEL_FATAL, "Cannot putenv(): HOME"); } - snprintf(putenv_dummy, 64, "USER=%s", pw->pw_name); + snprintf(putenv_dummy, sizeof(putenv_dummy), "USER=%s", pw->pw_name); if (putenv(putenv_dummy) != 0) { log_error(LOG_LEVEL_FATAL, "Cannot putenv(): USER"); @@ -2471,16 +3829,7 @@ static jb_socket bind_port_helper(struct configuration_spec * config) int result; jb_socket bfd; - if ( (config->haddr != NULL) - && (config->haddr[0] == '1') - && (config->haddr[1] == '2') - && (config->haddr[2] == '7') - && (config->haddr[3] == '.') ) - { - log_error(LOG_LEVEL_INFO, "Listening on port %d for local connections only", - config->hport); - } - else if (config->haddr == NULL) + if (config->haddr == NULL) { log_error(LOG_LEVEL_INFO, "Listening on port %d on all IP addresses", config->hport); @@ -2554,6 +3903,14 @@ static void listen_loop(void) 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 @@ -2580,28 +3937,14 @@ static void listen_loop(void) */ if (received_hup_signal) { - init_error_log(Argv[0], config->logfile, config->debug); + 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)); @@ -2633,7 +3976,7 @@ static void listen_loop(void) bfd = bind_port_helper(config); } - log_error(LOG_LEVEL_CONNECT, "accept connection ... "); + log_error(LOG_LEVEL_CONNECT, "Listening for new connections ... "); if (!accept_connection(csp, bfd)) { @@ -2650,15 +3993,15 @@ static void listen_loop(void) } else { - log_error(LOG_LEVEL_CONNECT, "OK"); + log_error(LOG_LEVEL_CONNECT, "accepted connection from %s", csp->ip_addr_str); } #ifdef FEATURE_TOGGLE if (global_toggle_state) +#endif /* def FEATURE_TOGGLE */ { csp->flags |= CSP_FLAG_TOGGLED_ON; } -#endif /* def FEATURE_TOGGLE */ if (run_loader(csp)) { @@ -2669,7 +4012,7 @@ static void listen_loop(void) #ifdef FEATURE_ACL if (block_acl(NULL,csp)) { - log_error(LOG_LEVEL_CONNECT, "Connection dropped due to ACL"); + log_error(LOG_LEVEL_CONNECT, "Connection from %s dropped due to ACL", csp->ip_addr_str); close_socket(csp->cfd); freez(csp); continue; @@ -2697,7 +4040,7 @@ static void listen_loop(void) pthread_attr_init(&attrs); pthread_attr_setdetachstate(&attrs, PTHREAD_CREATE_DETACHED); errno = pthread_create(&the_thread, &attrs, - (void*)serve, csp); + (void * (*)(void *))serve, csp); child_id = errno ? -1 : 0; pthread_attr_destroy(&attrs); } @@ -2741,13 +4084,22 @@ static void listen_loop(void) #define SELECTED_ONE_OPTION csp->cfd = ReleaseSocket(csp->cfd, -1); - if((child_id = (int)CreateNewProcTags( - NP_Entry, (ULONG)server_thread, - NP_Output, Output(), - NP_CloseOutput, FALSE, - NP_Name, (ULONG)"privoxy child", - NP_StackSize, 200*1024, - TAG_DONE))) +#ifdef __amigaos4__ + child_id = (int)CreateNewProcTags(NP_Entry, (ULONG)server_thread, + NP_Output, Output(), + NP_CloseOutput, FALSE, + NP_Name, (ULONG)"privoxy child", + NP_Child, TRUE, + TAG_DONE); +#else + child_id = (int)CreateNewProcTags(NP_Entry, (ULONG)server_thread, + NP_Output, Output(), + NP_CloseOutput, FALSE, + NP_Name, (ULONG)"privoxy child", + NP_StackSize, 200*1024, + TAG_DONE); +#endif + if(0 != child_id) { childs++; ((struct Task *)child_id)->tc_UserData = csp; @@ -2839,7 +4191,7 @@ static void listen_loop(void) log_error(LOG_LEVEL_ERROR, "can't fork: %E"); - sprintf(buf , "Privoxy: can't fork: errno = %d", errno); + snprintf(buf , sizeof(buf), "Privoxy: can't fork: errno = %d", errno); write_socket(csp->cfd, buf, strlen(buf)); close_socket(csp->cfd); @@ -2890,13 +4242,6 @@ 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();