X-Git-Url: http://www.privoxy.org/gitweb/?p=privoxy.git;a=blobdiff_plain;f=jcc.c;h=056a6c1e9b1af98786ccd072ac4d47acec0e87a5;hp=dabd5021c2867b02340ad77f49e92e76719c7a5b;hb=2c224bc9ed07e2507e1ad66a41e98f66e9af96c9;hpb=5db85a228ac9a66c397e367b6acef8f64c809549 diff --git a/jcc.c b/jcc.c index dabd5021..056a6c1e 100644 --- a/jcc.c +++ b/jcc.c @@ -1,19 +1,19 @@ -const char jcc_rcs[] = "$Id: jcc.c,v 1.43 2001/10/02 15:32:13 oes Exp $"; +const char jcc_rcs[] = "$Id: jcc.c,v 1.127 2007/03/20 13:53:17 fabiankeil Exp $"; /********************************************************************* * * File : $Source: /cvsroot/ijbswa/current/jcc.c,v $ * - * Purpose : Main file. Contains main() method, main loop, and + * Purpose : Main file. Contains main() method, main loop, and * the main connection-handling function. * - * Copyright : Written by and Copyright (C) 2001 the SourceForge - * IJBSWA team. http://ijbswa.sourceforge.net + * Copyright : Written by and Copyright (C) 2001-2007 the SourceForge + * Privoxy team. http://www.privoxy.org/ * * Based on the Internet Junkbuster originally written - * by and Copyright (C) 1997 Anonymous Coders and + * by and Copyright (C) 1997 Anonymous Coders and * Junkbusters Corporation. http://www.junkbusters.com * - * This program is free software; you can redistribute it + * This program is free software; you can redistribute it * and/or modify it under the terms of the GNU General * Public License as published by the Free Software * Foundation; either version 2 of the License, or (at @@ -33,6 +33,491 @@ const char jcc_rcs[] = "$Id: jcc.c,v 1.43 2001/10/02 15:32:13 oes Exp $"; * * Revisions : * $Log: jcc.c,v $ + * 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 + * that does the port redirection. The destination + * is extracted from the "Host:" header which + * should be available for nearly all requests. + * + * Moved HTTP snipplets into jcc.c. + * Added error message for gopher proxy requests. + * + * Revision 1.108 2006/11/28 15:38:51 fabiankeil + * Only unlink the pidfile if it's actually used. + * + * Change order of interception checks to make + * it possible to block or redirect requests for + * the cgi pages. + * + * Revision 1.107 2006/11/13 19:05:51 fabiankeil + * Make pthread mutex locking more generic. Instead of + * checking for OSX and OpenBSD, check for FEATURE_PTHREAD + * and use mutex locking unless there is an _r function + * available. Better safe than sorry. + * + * Fixes "./configure --disable-pthread" and should result + * in less threading-related problems on pthread-using platforms, + * but it still doesn't fix BR#1122404. + * + * Revision 1.106 2006/11/06 19:58:23 fabiankeil + * Move pthread.h inclusion from jcc.c to jcc.h. + * Fixes build on x86-freebsd1 (FreeBSD 5.4-RELEASE). + * + * Revision 1.105 2006/11/06 14:26:02 fabiankeil + * Don't exit after receiving the second SIGHUP on Solaris. + * + * Fixes BR 1052235, but the same problem may exist on other + * systems. Once 3.0.6 is out we should use sigset() + * where available and see if it breaks anything. + * + * Revision 1.104 2006/09/23 13:26:38 roro + * Replace TABs by spaces in source code. + * + * Revision 1.103 2006/09/21 12:54:43 fabiankeil + * Fix +redirect{}. Didn't work with -fast-redirects. + * + * Revision 1.102 2006/09/06 13:03:04 fabiankeil + * Respond with 400 and a short text message + * if the client tries to use Privoxy as FTP proxy. + * + * Revision 1.101 2006/09/06 09:23:37 fabiankeil + * Make number of retries in case of forwarded-connect problems + * a config file option (forwarded-connect-retries) and use 0 as + * default. + * + * Revision 1.100 2006/09/03 19:42:59 fabiankeil + * Set random(3) seed. + * + * Revision 1.99 2006/09/02 15:36:42 fabiankeil + * Follow the OpenBSD port's lead and protect the resolve + * functions on OpenBSD as well. + * + * Revision 1.98 2006/08/24 11:01:34 fabiankeil + * --user fix. Only use the user as group if no group is specified. + * Solves BR 1492612. Thanks to Spinor S. and David Laight. + * + * Revision 1.97 2006/08/18 15:23:17 david__schmidt + * Windows service (re-)integration + * + * The new args are: + * + * --install[:service_name] + * --uninstall[:service_name] + * --service + * + * They work as follows: + * --install will create a service for you and then terminate. + * By default the service name will be "privoxy" (without the quotes). + * However you can run multiple services if you wish, just by adding + * a colon and then a name (no spaces). + * + * --uninstall follows the exact same rules a --install. + * + * --service is used when the program is executed by the service + * control manager, and in normal circumstances would never be + * used as a command line argument. + * + * Revision 1.96 2006/08/15 20:12:36 david__schmidt + * Windows service integration + * + * Revision 1.95 2006/08/03 02:46:41 david__schmidt + * Incorporate Fabian Keil's patch work: +http://www.fabiankeil.de/sourcecode/privoxy/ + * + * Revision 1.94 2006/07/18 14:48:46 david__schmidt + * Reorganizing the repository: swapping out what was HEAD (the old 3.1 branch) + * with what was really the latest development (the v_3_0_branch branch) + * + * Revision 1.92.2.16 2005/04/03 20:10:50 david__schmidt + * Thanks to Jindrich Makovicka for a race condition fix for the log + * file. The race condition remains for non-pthread implementations. + * Reference patch #1175720. + * + * Revision 1.92.2.15 2004/10/03 12:53:32 david__schmidt + * Add the ability to check jpeg images for invalid + * lengths of comment blocks. Defensive strategy + * against the exploit: + * Microsoft Security Bulletin MS04-028 + * Buffer Overrun in JPEG Processing (GDI+) Could + * Allow Code Execution (833987) + * Enabled with +inspect-jpegs in actions files. + * + * Revision 1.92.2.14 2003/12/12 12:52:53 oes + * - Fixed usage info for non-unix platforms + * - Fixed small cmdline parsing bug + * + * Revision 1.92.2.13 2003/11/27 19:20:27 oes + * Diagnostics: Now preserve the returncode of pthread_create + * in errno. Closes BR #775721. Thanks to Geoffrey Hausheer. + * + * Revision 1.92.2.12 2003/07/11 11:34:19 oes + * No longer ignore SIGCHLD. Fixes bug #769381 + * + * Revision 1.92.2.11 2003/05/14 12:32:02 oes + * Close jarfile on graceful exit, remove stray line + * + * Revision 1.92.2.10 2003/05/08 15:13:46 oes + * Cosmetics: Killed a warning, a typo and an allocation left at exit + * + * Revision 1.92.2.9 2003/04/03 15:08:42 oes + * No longer rely on non-POSIX.1 extensions of getcwd(). + * Fixes bug #711001 + * + * Revision 1.92.2.8 2003/03/31 13:12:32 oes + * Replaced setenv() by posix-compliant putenv() + * Thanks to Neil McCalden (nmcc AT users.sf.net). + * + * Revision 1.92.2.7 2003/03/17 16:48:59 oes + * Added chroot ability, thanks to patch by Sviatoslav Sviridov + * + * Revision 1.92.2.6 2003/03/11 11:55:00 oes + * Clean-up and extension of improvements for forked mode: + * - Child's return code now consists of flags RC_FLAG_* + * - Reporting toggle to parent now properly #ifdef'ed + * - Children now report blocking to parent. This enables + * statistics in forked mode + * + * Revision 1.92.2.5 2003/03/10 23:45:32 oes + * Fixed bug #700381: Non-Threaded version now capable of being toggled. + * Children now report having been toggled through _exit(17), parents + * watch for that code and toggle themselves if found. + * + * Revision 1.92.2.4 2003/03/07 03:41:04 david__schmidt + * Wrapping all *_r functions (the non-_r versions of them) with + * mutex semaphores for OSX. Hopefully this will take care of all + * of those pesky crash reports. + * + * Revision 1.92.2.3 2003/02/28 12:53:06 oes + * Fixed two mostly harmless mem leaks + * + * Revision 1.92.2.2 2002/11/20 14:37:47 oes + * Fix: Head of global clients list now initialized to NULL + * + * Revision 1.92.2.1 2002/09/25 14:52:24 oes + * Added basic support for OPTIONS and TRACE HTTP methods: + * - New interceptor direct_response() added in chat(). + * - sed() moved to earlier in the process, so that the + * Host: header is evaluated before actions and forwarding + * are decided on. + * + * Revision 1.92 2002/05/08 16:00:46 oes + * Chat's buffer handling: + * - Fixed bug with unchecked out-of-mem conditions + * while reading client request & server headers + * - No longer predict if the buffer limit will be exceeded + * in the next read -- check add_to_iob's new + * return code. If buffer couldn't be extended + * (policy or out-of-mem) while + * - reading from client: abort + * - reading server headers: send error page + * - buffering server body for filter: flush, + * and if that fails: send error page + * + * Revision 1.91 2002/04/08 20:35:58 swa + * fixed JB spelling + * + * Revision 1.90 2002/04/02 14:57:28 oes + * Made sending wafers independent of FEATURE_COOKIE_JAR + * + * Revision 1.89 2002/03/31 17:18:59 jongfoster + * Win32 only: Enabling STRICT to fix a VC++ compile warning. + * + * Revision 1.88 2002/03/27 14:32:43 david__schmidt + * More compiler warning message maintenance + * + * Revision 1.87 2002/03/26 22:29:54 swa + * we have a new homepage! + * + * Revision 1.86 2002/03/25 17:04:55 david__schmidt + * Workaround for closing the jarfile before load_config() comes around again + * + * Revision 1.85 2002/03/24 15:23:33 jongfoster + * Name changes + * + * Revision 1.84 2002/03/24 13:25:43 swa + * name change related issues + * + * Revision 1.83 2002/03/16 23:54:06 jongfoster + * Adding graceful termination feature, to help look for memory leaks. + * If you enable this (which, by design, has to be done by hand + * editing config.h) and then go to http://i.j.b/die, then the program + * will exit cleanly after the *next* request. It should free all the + * memory that was used. + * + * Revision 1.82 2002/03/13 00:27:05 jongfoster + * Killing warnings + * + * Revision 1.81 2002/03/12 01:42:50 oes + * Introduced modular filters + * + * Revision 1.80 2002/03/11 22:07:05 david__schmidt + * OS/2 port maintenance: + * - Fixed EMX build - it had decayed a little + * - Fixed inexplicable crash during FD_ZERO - must be due to a bad macro. + * substituted a memset for now. + * + * Revision 1.79 2002/03/09 20:03:52 jongfoster + * - Making various functions return int rather than size_t. + * (Undoing a recent change). Since size_t is unsigned on + * Windows, functions like read_socket that return -1 on + * error cannot return a size_t. + * + * THIS WAS A MAJOR BUG - it caused frequent, unpredictable + * crashes, and also frequently caused JB to jump to 100% + * CPU and stay there. (Because it thought it had just + * read ((unsigned)-1) == 4Gb of data...) + * + * - The signature of write_socket has changed, it now simply + * returns success=0/failure=nonzero. + * + * - Trying to get rid of a few warnings --with-debug on + * Windows, I've introduced a new type "jb_socket". This is + * used for the socket file descriptors. On Windows, this + * is SOCKET (a typedef for unsigned). Everywhere else, it's + * an int. The error value can't be -1 any more, so it's + * now JB_INVALID_SOCKET (which is -1 on UNIX, and in + * Windows it maps to the #define INVALID_SOCKET.) + * + * - The signature of bind_port has changed. + * + * Revision 1.78 2002/03/08 21:35:04 oes + * Added optional group supplement to --user option. Will now use default group of user if no group given + * + * Revision 1.77 2002/03/07 03:52:06 oes + * - Fixed compiler warnings etc + * - Improved handling of failed DNS lookups + * + * Revision 1.76 2002/03/06 22:54:35 jongfoster + * Automated function-comment nitpicking. + * + * Revision 1.75 2002/03/06 10:02:19 oes + * Fixed stupid bug when --user was not given + * + * Revision 1.74 2002/03/06 00:49:31 jongfoster + * Fixing warning on Windows + * Making #ifdefs that refer to the same variable consistently + * use #ifdef unix rather than mixing #ifdef unix & #ifndef OS2 + * + * Revision 1.73 2002/03/05 23:57:30 hal9 + * Stray character 's' on line 1618 was breaking build. + * + * Revision 1.72 2002/03/05 21:33:45 david__schmidt + * - Re-enable OS/2 building after new parms were added + * - Fix false out of memory report when resolving CGI templates when no IP + * address is available of failed attempt (a la no such domain) + * + * Revision 1.71 2002/03/05 18:13:56 oes + * Added --user option + * + * Revision 1.70 2002/03/05 04:52:42 oes + * Deleted non-errlog debugging code + * + * Revision 1.69 2002/03/04 23:50:00 jongfoster + * Splitting off bind_port() call into bind_port_helper(), with + * improved logging. + * + * Revision 1.68 2002/03/04 20:17:32 oes + * Fixed usage info + * + * Revision 1.67 2002/03/04 18:18:57 oes + * - Removed _DEBUG mode + * - Cleand up cmdline parsing + * - Introduced --no-daemon, --pidfile options + * - Cleaned up signal handling: + * - Terminate cleanly on INT, TERM and ABRT + * - Schedule logfile for re-opening on HUP + * - Ignore CHLD and PIPE + * - Leave the rest with their default handlers + * - Uniform handler registration + * - Added usage() function + * - Played styleguide police + * + * Revision 1.66 2002/03/03 15:06:55 oes + * Re-enabled automatic config reloading + * + * Revision 1.65 2002/03/03 14:49:11 oes + * Fixed CLF logging: Now uses client's original HTTP request + * + * Revision 1.64 2002/03/03 09:18:03 joergs + * Made jumbjuster work on AmigaOS again. + * + * Revision 1.63 2002/03/02 04:14:50 david__schmidt + * Clean up a little CRLF unpleasantness that suddenly appeared + * + * Revision 1.62 2002/02/20 23:17:23 jongfoster + * Detecting some out-of memory conditions and exiting with a log message. + * + * Revision 1.61 2002/01/17 21:01:52 jongfoster + * Moving all our URL and URL pattern parsing code to urlmatch.c. + * + * Revision 1.60 2001/12/30 14:07:32 steudten + * - Add signal handling (unix) + * - Add SIGHUP handler (unix) + * - Add creation of pidfile (unix) + * - Add action 'top' in rc file (RH) + * - Add entry 'SIGNALS' to manpage + * - Add exit message to logfile (unix) + * + * Revision 1.59 2001/12/13 14:07:18 oes + * Fixed Bug: 503 error page now sent OK + * + * Revision 1.58 2001/11/30 23:37:24 jongfoster + * Renaming the Win32 config file to config.txt - this is almost the + * same as the corresponding UNIX name "config" + * + * Revision 1.57 2001/11/16 00:47:43 jongfoster + * Changing the tty-disconnection code to use setsid(). + * + * Revision 1.56 2001/11/13 20:20:54 jongfoster + * Tabs->spaces, fixing a bug with missing {} around an if() + * + * Revision 1.55 2001/11/13 20:14:53 jongfoster + * Patch for FreeBSD setpgrp() as suggested by Alexander Lazic + * + * Revision 1.54 2001/11/07 00:03:14 steudten + * Give reliable return value if an error + * occurs not just 0 with new daemon mode. + * + * Revision 1.53 2001/11/05 21:41:43 steudten + * Add changes to be a real daemon just for unix os. + * (change cwd to /, detach from controlling tty, set + * process group and session leader to the own process. + * Add DBG() Macro. + * Add some fatal-error log message for failed malloc(). + * Add '-d' if compiled with 'configure --with-debug' to + * enable debug output. + * + * Revision 1.52 2001/10/26 20:11:20 jongfoster + * Fixing type mismatch + * + * Revision 1.51 2001/10/26 17:38:28 oes + * Cosmetics + * + * Revision 1.50 2001/10/25 03:40:48 david__schmidt + * Change in porting tactics: OS/2's EMX porting layer doesn't allow multiple + * threads to call select() simultaneously. So, it's time to do a real, live, + * native OS/2 port. See defines for __EMX__ (the porting layer) vs. __OS2__ + * (native). Both versions will work, but using __OS2__ offers multi-threading. + * + * Revision 1.49 2001/10/23 21:41:35 jongfoster + * Added call to initialize the (statically-allocated of course) + * "out of memory" CGI response. + * + * Revision 1.48 2001/10/10 19:56:46 jongfoster + * Moving some code that wasn't cookie-related out of an #ifdef + * FEATURE_COOKIE_JAR + * + * Revision 1.47 2001/10/10 16:44:36 oes + * Added CONNECT destination port limitation check + * + * Revision 1.46 2001/10/08 15:17:41 oes + * Re-enabled SSL forwarding + * + * Revision 1.45 2001/10/07 15:42:11 oes + * Replaced 6 boolean members of csp with one bitmap (csp->flags) + * + * Moved downgrading of the HTTP version from parse_http_request to + * chat(), since we can't decide if it is necessary before we have + * determined the actions for the URL. The HTTP command is now + * *always* re-built so the repairs need no longer be special-cased. + * + * filter_popups now gets a csp pointer so it can raise the new + * CSP_FLAG_MODIFIED flag. + * + * Bugfix + * + * Added configurable size limit for the IOB. If the IOB grows so + * large that the next read would exceed the limit, the header + * is generated, and the header & unfiltered buffer are flushed + * to the client. Chat then continues in non-buffering, + * non-filtering body mode. + * + * Revision 1.44 2001/10/02 18:13:57 oes + * Ooops + * * Revision 1.43 2001/10/02 15:32:13 oes * Moved generation of hdr * @@ -310,13 +795,13 @@ const char jcc_rcs[] = "$Id: jcc.c,v 1.43 2001/10/02 15:32:13 oes Exp $"; #include #include #include - -#ifdef FEATURE_PTHREAD -#include -#endif /* def FEATURE_PTHREAD */ +#include #ifdef _WIN32 # ifndef FEATURE_PTHREAD +# ifndef STRICT +# define STRICT +# endif # include # include # endif /* ndef FEATURE_PTHREAD */ @@ -325,13 +810,27 @@ const char jcc_rcs[] = "$Id: jcc.c,v 1.43 2001/10/02 15:32:13 oes Exp $"; # ifndef _WIN_CONSOLE # include "w32log.h" # endif /* ndef _WIN_CONSOLE */ +# include "w32svrapi.h" #else /* ifndef _WIN32 */ +# if !defined (__OS2__) # include -# include # include +# endif /* ndef __OS2__ */ +# include # include +# include + +#ifdef sun +#include +#endif /* sun */ + +#ifdef unix +#include +#include +#endif + # include # ifdef __BEOS__ @@ -339,9 +838,14 @@ const char jcc_rcs[] = "$Id: jcc.c,v 1.43 2001/10/02 15:32:13 oes Exp $"; # include /* declarations for threads and stuff. */ # endif -# ifdef __EMX__ +# if defined(__EMX__) || defined(__OS2__) # include /* OS/2/EMX needs a little help with select */ # endif +# ifdef __OS2__ +#define INCL_DOS +# include +#define bzero(B,N) memset(B,0x00,n) +# endif # ifndef FD_ZERO # include @@ -363,10 +867,12 @@ const char jcc_rcs[] = "$Id: jcc.c,v 1.43 2001/10/02 15:32:13 oes Exp $"; #include "actions.h" #include "cgi.h" #include "loadcfg.h" +#include "urlmatch.h" const char jcc_h_rcs[] = JCC_H_VERSION; const char project_h_rcs[] = PROJECT_H_VERSION; +int no_daemon = 0; struct client_state clients[1]; struct file_list files[1]; @@ -375,6 +881,9 @@ int urls_read = 0; /* total nr of urls read inc rejected */ int urls_rejected = 0; /* total nr of urls rejected */ #endif /* def FEATURE_STATISTICS */ +#ifdef FEATURE_GRACEFUL_TERMINATION +int g_terminate = 0; +#endif static void listen_loop(void); static void chat(struct client_state *csp); @@ -392,6 +901,41 @@ static int32 server_thread(void *data); #define sleep(N) Sleep(((N) * 1000)) #endif +#ifdef __OS2__ +#define sleep(N) DosSleep(((N) * 100)) +#endif + +#ifdef FEATURE_PTHREAD +pthread_mutex_t log_mutex; +pthread_mutex_t log_init_mutex; + +#ifndef HAVE_GMTIME_R +pthread_mutex_t gmtime_mutex; +#endif /* ndef HAVE_GMTIME_R */ + +#ifndef HAVE_LOCALTIME_R +pthread_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_GETHOSTBYNAME_R +pthread_mutex_t gethostbyname_mutex; +#endif /* ndef HAVE_GETHOSTBYNAME_R */ + +#ifndef HAVE_RANDOM +pthread_mutex_t rand_mutex; +#endif /* ndef HAVE_RANDOM */ + +#endif /* FEATURE_PTHREAD */ + +#if defined(unix) || defined(__EMX__) +const char *basedir = NULL; +const char *pidfile = NULL; +int received_hup_signal = 0; +#endif /* defined unix */ /* The vanilla wafer. */ static const char VANILLA_WAFER[] = @@ -403,6 +947,528 @@ static const char VANILLA_WAFER[] = "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" + "Proxy-Agent: Privoxy/" VERSION "\r\n\r\n"; + +const char CHEADER[] = + "HTTP/1.0 400 Invalid header received from browser\r\n" + "Connection: close\r\n\r\n" + "Invalid header received from browser.\r\n"; + +const char CFORBIDDEN[] = + "HTTP/1.0 403 Connection not allowable\r\n" + "Proxy-Agent: Privoxy " VERSION "\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" + "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" + "Connection: close\r\n\r\n" + "Invalid request. Privoxy doesn't support gopher.\r\n"; + +/* XXX: should be a template */ +const char MISSING_DESTINATION_RESPONSE[] = + "HTTP/1.0 400 Bad request received from browser\r\n" + "Proxy-Agent: Privoxy " VERSION "\r\n" + "Connection: close\r\n\r\n" + "Bad request. Privoxy was unable to extract the destination.\r\n"; + +/* XXX: should be a template */ +const char NO_SERVER_DATA_RESPONSE[] = + "HTTP/1.0 502 Server or forwarder response empty\r\n" + "Proxy-Agent: Privoxy " VERSION "\r\n" + "Connection: close\r\n\r\n" + "Empty server or forwarder response.\r\n" + "The connection was closed without sending any data.\r\n"; + +/* XXX: should be a template */ +const char NULL_BYTE_RESPONSE[] = + "HTTP/1.0 400 Bad request received from browser\r\n" + "Proxy-Agent: Privoxy " VERSION "\r\n" + "Connection: close\r\n\r\n" + "Bad request. Null byte(s) before end of request.\r\n"; + +#if !defined(_WIN32) && !defined(__OS2__) && !defined(AMIGA) +/********************************************************************* + * + * Function : sig_handler + * + * Description : Signal handler for different signals. + * Exit gracefully on TERM and INT + * or set a flag that will cause the errlog + * to be reopened by the main thread on HUP. + * + * Parameters : + * 1 : the_signal = the signal cause this function to call + * + * Returns : - + * + *********************************************************************/ +static void sig_handler(int the_signal) +{ + switch(the_signal) + { + case SIGTERM: + case SIGINT: + log_error(LOG_LEVEL_INFO, "exiting by signal %d .. bye", the_signal); +#if defined(unix) + if(pidfile) + { + unlink(pidfile); + } +#endif /* unix */ + exit(the_signal); + break; + + case SIGHUP: + received_hup_signal = 1; + break; + + default: + /* + * We shouldn't be here, unless we catch signals + * in main() that we can't handle here! + */ + log_error(LOG_LEVEL_FATAL, "sig_handler: exiting on unexpected signal %d", the_signal); + } + return; + +} +#endif + + +/********************************************************************* + * + * Function : client_protocol_is_unsupported + * + * Description : Checks if the client used a known unsupported + * protocol and deals with it by sending an error + * response. + * + * Parameters : + * 1 : csp = Current client state (buffers, headers, etc...) + * 2 : req = the first request line send by the client + * + * Returns : TRUE if an error response has been generated, or + * FALSE if the request doesn't look invalid. + * + *********************************************************************/ +int client_protocol_is_unsupported(const struct client_state *csp, char *req) +{ + char buf[BUFFER_SIZE]; + + /* + * If it's a FTP or gopher request, we don't support it. + * + * These checks are better than nothing, but they might + * not work in all configurations and some clients might + * have problems digesting the answer. + * + * They should, however, never cause more problems than + * Privoxy's old behaviour (returning the misleading HTML + * error message: + * + * "Could not resolve http://(ftp|gopher)://example.org"). + */ + if (!strncmpic(req, "GET ftp://", 10) || !strncmpic(req, "GET gopher://", 13)) + { + if (!strncmpic(req, "GET ftp://", 10)) + { + 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)); + + return TRUE; + } + + return FALSE; +} + + +/********************************************************************* + * + * Function : get_request_destination_elsewhere + * + * Description : If the client's request was redirected into + * Privoxy without the client's knowledge, + * the request line lacks the destination host. + * + * This function tries to get it elsewhere, + * provided accept-intercepted-requests is enabled. + * + * "Elsewhere" currently only means "Host: header", + * but in the future we may ask the redirecting + * packet filter to look the destination up. + * + * If the destination stays unknown, an error + * response is send to the client and headers + * are freed so that chat() can return directly. + * + * Parameters : + * 1 : csp = Current client state (buffers, headers, etc...) + * 2 : headers = a header list + * + * Returns : JB_ERR_OK if the destination is now known, or + * JB_ERR_PARSE if it isn't. + * + *********************************************************************/ +jb_err get_request_destination_elsewhere(struct client_state *csp, struct list *headers) +{ + char buf[BUFFER_SIZE]; + char *req; + + if (!(csp->config->feature_flags & RUNTIME_FEATURE_ACCEPT_INTERCEPTED_REQUESTS)) + { + log_error(LOG_LEVEL_ERROR, "%s's request: \'%s\' is invalid." + " Privoxy isn't configured to accept intercepted requests.", + csp->ip_addr_str, csp->http->cmd); + /* XXX: Use correct size */ + log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" 400 0", + csp->ip_addr_str, csp->http->cmd); + + strcpy(buf, CHEADER); + write_socket(csp->cfd, buf, strlen(buf)); + destroy_list(headers); + + return JB_ERR_PARSE; + } + else if (JB_ERR_OK == get_destination_from_headers(headers, csp->http)) + { + /* Split the domain we just got for pattern matching */ + init_domain_components(csp->http); + + return JB_ERR_OK; + } + else + { + /* We can't work without destination. Go spread the news.*/ + + req = list_to_text(headers); + chomp(req); + /* XXX: Use correct size */ + log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" 400 0", + csp->ip_addr_str, csp->http->cmd); + log_error(LOG_LEVEL_ERROR, + "Privoxy was unable to get the destination for %s's request:\n%s\n%s", + csp->ip_addr_str, csp->http->cmd, req); + freez(req); + + strcpy(buf, MISSING_DESTINATION_RESPONSE); + write_socket(csp->cfd, buf, strlen(buf)); + destroy_list(headers); + + return JB_ERR_PARSE; + } + /* + * TODO: If available, use PF's ioctl DIOCNATLOOK as last resort + * to get the destination IP address, use it as host directly + * or do a reverse DNS lookup first. + */ +} + + +/********************************************************************* + * + * Function : get_server_headers + * + * Description : Parses server headers in iob and fills them + * into csp->headers so that they can later be + * handled by sed(). + * + * Parameters : + * 1 : csp = Current client state (buffers, headers, etc...) + * + * Returns : JB_ERR_OK if everything went fine, or + * JB_ERR_PARSE if the headers were incomplete. + * + *********************************************************************/ +jb_err get_server_headers(struct client_state *csp) +{ + int continue_hack_in_da_house = 0; + char * header; + + while (((header = get_header(csp)) != NULL) || continue_hack_in_da_house) + { + if (header == NULL) + { + /* + * continue hack in da house. Ignore the ending of + * this head and continue enlisting header lines. + * The reason is described below. + */ + enlist(csp->headers, ""); + continue_hack_in_da_house = 0; + continue; + } + else if (0 == strncmpic(header, "HTTP/1.1 100", 12)) + { + /* + * It's a bodyless continue response, don't + * stop header parsing after reaching it's end. + * + * As a result Privoxy will concatenate the + * next response's head and parse and deliver + * the headers as if they belonged to one request. + * + * The client will separate them because of the + * empty line between them. + * + * XXX: What we're doing here is clearly against + * the intended purpose of the continue header, + * and under some conditions (HTTP/1.0 client request) + * it's a standard violation. + * + * Anyway, "sort of against the spec" is preferable + * to "always getting confused by Continue responses" + * (Privoxy's behaviour before this hack was added) + */ + log_error(LOG_LEVEL_HEADER, "Continue hack in da house."); + continue_hack_in_da_house = 1; + } + else if (*header == '\0') + { + /* + * If the header is empty, but the Continue hack + * isn't active, we can assume that we reached the + * end of the buffer before we hit the end of the + * head. + * + * Inform the caller an let it decide how to handle it. + */ + return JB_ERR_PARSE; + } + + /* Enlist header */ + if (JB_ERR_MEMORY == enlist(csp->headers, header)) + { + /* + * XXX: Should we quit the request and return a + * out of memory error page instead? + */ + log_error(LOG_LEVEL_ERROR, + "Out of memory while enlisting server headers. %s lost.", + header); + } + freez(header); + } + + return JB_ERR_OK; +} + + +/********************************************************************* + * + * Function : crunch_reason + * + * Description : Translates the crunch reason code into a string. + * + * Parameters : + * 1 : rsp = a http_response + * + * Returns : A string with the crunch reason or an error description. + * + *********************************************************************/ +const char *crunch_reason(const struct http_response *rsp) +{ + char * reason = NULL; + + assert(rsp != NULL); + if (rsp == NULL) + { + return "Internal error while searching for crunch reason"; + } + + switch (rsp->reason) + { + case RSP_REASON_UNSUPPORTED: + reason = "Unsupported HTTP feature"; + break; + case RSP_REASON_BLOCKED: + reason = "Blocked"; + break; + case RSP_REASON_UNTRUSTED: + reason = "Untrusted"; + break; + case RSP_REASON_REDIRECTED: + reason = "Redirected"; + break; + case RSP_REASON_CGI_CALL: + reason = "CGI Call"; + break; + case RSP_REASON_NO_SUCH_DOMAIN: + reason = "DNS failure"; + break; + case RSP_REASON_FORWARDING_FAILED: + reason = "Forwarding failed"; + break; + case RSP_REASON_CONNECT_FAILED: + reason = "Connection failure"; + break; + case RSP_REASON_OUT_OF_MEMORY: + reason = "Out of memory (may mask other reasons)"; + break; + default: + reason = "No reason recorded"; + break; + } + + return reason; +} + + +/********************************************************************* + * + * Function : send_crunch_response + * + * Description : Delivers already prepared response for + * intercepted requests, logs the interception + * and frees the response. + * + * Parameters : + * 1 : csp = Current client state (buffers, headers, etc...) + * 1 : rsp = Fully prepared response. Will be freed on exit. + * + * Returns : Nothing. + * + *********************************************************************/ +void send_crunch_response(struct client_state *csp, struct http_response *rsp) +{ + const struct http_request *http = csp->http; + char status_code[4]; + + assert(rsp != NULL); + assert(rsp->head != NULL); + + if (rsp == NULL) + { + /* + * Not supposed to happen. If it does + * anyway, treat it as an unknown error. + */ + cgi_error_unknown(csp, rsp, RSP_REASON_INTERNAL_ERROR); + /* return code doesn't matter */ + } + + if (rsp == NULL) + { + /* If rsp is still NULL, we have serious internal problems. */ + log_error(LOG_LEVEL_FATAL, + "NULL response in send_crunch_response and cgi_error_unknown failed as well."); + } + + /* + * Extract the status code from the actual head + * that was send to the client. It is the only + * way to get it right for all requests, including + * the fixed ones for out-of-memory problems. + * + * A head starts like this: 'HTTP/1.1 200...' + * 0123456789|11 + * 10 + */ + status_code[0] = rsp->head[9]; + status_code[1] = rsp->head[10]; + status_code[2] = rsp->head[11]; + status_code[3] = '\0'; + + /* Write the answer to the client */ + if (write_socket(csp->cfd, rsp->head, rsp->head_length) + || write_socket(csp->cfd, rsp->body, rsp->content_length)) + { + /* There is nothing we can do about it. */ + log_error(LOG_LEVEL_ERROR, "write to: %s failed: %E", csp->http->host); + } + + /* Log that the request was crunched and why. */ + log_error(LOG_LEVEL_GPC, "%s%s crunch! (%s)", + http->hostport, http->path, crunch_reason(rsp)); + log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" %s %d", + csp->ip_addr_str, http->ocmd, status_code, rsp->content_length); + + /* Clean up and return */ + if (cgi_error_memory() != rsp) + { + free_http_response(rsp); + } + return; +} + + +/********************************************************************* + * + * Function : request_contains_null_bytes + * + * Description : Checks for NULL bytes in the request and sends + * an error message to the client if any were found. + * + * Parameters : + * 1 : csp = Current client state (buffers, headers, etc...) + * 2 : buf = Data from the client's request to check. + * 3 : len = The data length. + * + * Returns : TRUE if the request contained one or more NULL bytes, or + * FALSE otherwise. + * + *********************************************************************/ +int request_contains_null_bytes(const struct client_state *csp, char *buf, int len) +{ + size_t c_len; /* Request lenght when treated as C string */ + + c_len = strlen(buf); + + if (c_len < len) + { + /* + * Null byte(s) found. Log the request, + * return an error response and hang up. + */ + size_t tmp_len = c_len; + + do + { + /* + * Replace NULL byte(s) with '°' characters + * so the request can be logged as string. + * XXX: Is there a better replacement character? + */ + buf[tmp_len]='°'; + tmp_len += strlen(buf+tmp_len); + } while (tmp_len < len); + + log_error(LOG_LEVEL_ERROR, "%s\'s request contains at least one NULL byte " + "(length=%d, strlen=%d).", csp->ip_addr_str, len, c_len); + log_error(LOG_LEVEL_HEADER, + "Offending request data with NULL bytes turned into \'°\' characters: %s", buf); + + strcpy(buf, NULL_BYTE_RESPONSE); + write_socket(csp->cfd, buf, strlen(buf)); + + /* XXX: Log correct size */ + log_error(LOG_LEVEL_CLF, "%s - - [%T] \"Invalid request\" 400 0", csp->ip_addr_str); + + return TRUE; + } + + return FALSE; +} + /********************************************************************* * @@ -410,10 +1476,13 @@ static const char VANILLA_WAFER[] = * * 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 + * 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...) * @@ -429,17 +1498,17 @@ static void chat(struct client_state *csp) { /* * This next lines are a little ugly, but they simplifies the if statements - * below. Basically if TOGGLE, then we want the if to test "csp->toggled_on", - * else we don't. And if FEATURE_FORCE_LOAD, then we want the if to test - * "csp->toggled_on", else we don't + * 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->toggled_on) && +# 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->force) && +# 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 */ @@ -447,13 +1516,20 @@ static void chat(struct client_state *csp) #define IS_ENABLED_AND IS_TOGGLED_ON_AND IS_NOT_FORCED_AND char buf[BUFFER_SIZE]; - char *hdr, *p, *req; + char *hdr; + char *p; + char *req; fd_set rfds; - int n, maxfd, server_body; + int n; + jb_socket maxfd; + int server_body; int ms_iis5_hack = 0; - int byte_count = 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) */ #ifdef FEATURE_KILL_POPUPS int block_popups; /* bool, 1==will block popups */ int block_popups_now = 0; /* bool, 1==currently blocking popups */ @@ -461,27 +1537,47 @@ static void chat(struct client_state *csp) 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; + 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; + memset(buf, 0, sizeof(buf)); + /* * Read the client's request. Note that since we're not using select() we * could get blocked here if a client connected, then didn't say anything! */ - while (FOREVER) + for (;;) { - n = read_socket(csp->cfd, buf, sizeof(buf)); + len = read_socket(csp->cfd, buf, sizeof(buf)-1); - if (n <= 0) break; /* error! */ + if (len <= 0) break; /* error! */ + + if (request_contains_null_bytes(csp, buf, len)) + { + /* NULL bytes found and dealt with, just hang up. */ + return; + } - add_to_iob(csp, buf, n); + /* + * 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; + } req = get_header(csp); @@ -494,25 +1590,50 @@ static void chat(struct client_state *csp) { continue; /* more to come! */ } - + + /* 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; + } + #ifdef FEATURE_FORCE_LOAD - /* If this request contains the FORCE_PREFIX, - * better get rid of it now and set the force flag --oes + /* + * 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)) { - strclean(req, FORCE_PREFIX); - log_error(LOG_LEVEL_FORCE, "Enforcing request \"%s\".\n", req); - csp->force = 1; - } - else - { - csp->force = 0; + 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 */ - parse_http_request(req, http, csp); + 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; + } + freez(req); break; } @@ -521,12 +1642,66 @@ static void chat(struct client_state *csp) { strcpy(buf, CHEADER); write_socket(csp->cfd, buf, strlen(buf)); + /* 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); - log_error(LOG_LEVEL_CLF, "%s - - [%T] \" \" 400 0", csp->ip_addr_str); - + free_http_request(http); return; } + /* grab the rest of the client's headers */ + init_list(headers); + for (;;) + { + if ( ( ( p = get_header(csp) ) != NULL) && ( *p == '\0' ) ) + { + 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 + */ + if (add_to_iob(csp, buf, len)) + { + return; + } + continue; + } + + if (p == NULL) break; + + enlist(headers, p); + freez(p); + + } + + if (http->host == NULL) + { + /* + * If we still don't know the request destination, + * the request is invalid or the client uses + * Privoxy without it's knowledge. + */ + if (JB_ERR_OK != get_request_destination_elsewhere(csp, headers)) + { + /* + * Our attempts to get the request destination + * elsewhere failed or Privoxy is configured + * to only accept proxy requests. + * + * An error response has already been send + * and we're done here. + */ + return; + } + } + /* decide how to route the HTTP request */ if ((fwd = forward_url(http, csp)) == NULL) @@ -539,158 +1714,227 @@ static void chat(struct client_state *csp) * we have to do one of the following: * * create = use the original HTTP request to create a new - * HTTP request that has only the path component - * without the http://domainspec - * pass = pass the original HTTP request unchanged + * 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. * - * drop = drop the HTTP request + * 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 | drop | - * | | | + * 0 | create | connect| + * | w/path | | * Forwarding +--------+--------+ * | | | - * 1 | pass | pass | - * | | | + * 1 | create | pass | + * | w/url | | * +--------+--------+ * */ - if (fwd->forward_host) + /* + * Determine the actions for this URL + */ +#ifdef FEATURE_TOGGLE + if (!(csp->flags & CSP_FLAG_TOGGLED_ON)) { - /* if forwarding, just pass the request as is */ - enlist(csp->headers, http->cmd); + /* Most compatible set of actions (i.e. none) */ + init_current_action(csp->action); } else +#endif /* ndef FEATURE_TOGGLE */ + { + url_actions(http, csp); + } + + + /* + * 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(http->ssl) { - if (http->ssl == 0) - { - /* otherwise elide the host information from the url */ - p = NULL; - p = strsav(p, http->gpc); - p = strsav(p, " "); - p = strsav(p, http->path); - p = strsav(p, " "); - p = strsav(p, http->ver); - enlist(csp->headers, p); - freez(p); + 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)) ) + { + if (csp->action->flags & ACTION_TREAT_FORBIDDEN_CONNECTS_LIKE_BLOCKS) + { + /* + * The response may confuse some clients, + * 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 + { + 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; + } } } - /* decide what we're to do with cookies */ -#ifdef FEATURE_TOGGLE - if (!csp->toggled_on) + /* + * 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)) { - /* Most compatible set of actions (i.e. none) */ - init_current_action(csp->action); + freez(http->ver); + http->ver = strdup("HTTP/1.0"); + + if (http->ver == NULL) + { + log_error(LOG_LEVEL_FATAL, "Out of memory downgrading HTTP version"); + } } - else -#endif /* ndef FEATURE_TOGGLE */ + + /* + * Save a copy of the original request for logging + */ + http->ocmd = strdup(http->cmd); + + if (http->ocmd == NULL) { - url_actions(http, csp); + log_error(LOG_LEVEL_FATAL, "Out of memory copying HTTP request line"); } -#ifdef FEATURE_COOKIE_JAR /* - * If we're logging cookies in a cookie jar, and the user has not - * supplied any wafers, and the user has not told us to suppress the - * vanilla wafer, then send the vanilla wafer. + * (Re)build the HTTP request for non-SSL requests. + * If forwarding, use the whole URL, else, use only the path. + */ + if (http->ssl == 0) + { + freez(http->cmd); + + http->cmd = strdup(http->gpc); + string_append(&http->cmd, " "); + + if (fwd->forward_host) + { + string_append(&http->cmd, http->url); + } + else + { + string_append(&http->cmd, http->path); + } + string_append(&http->cmd, " "); + string_append(&http->cmd, http->ver); + + if (http->cmd == NULL) + { + log_error(LOG_LEVEL_FATAL, "Out of memory writing HTTP command"); + } + log_error(LOG_LEVEL_HEADER, "New HTTP Request-Line: %s", http->cmd); + } + enlist(csp->headers, http->cmd); + + /* Append the previously read headers */ + 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 ((csp->config->jarfile != NULL) - && list_is_empty(csp->action->multi[ACTION_MULTI_WAFER]) + if (list_is_empty(csp->action->multi[ACTION_MULTI_WAFER]) && ((csp->action->flags & ACTION_VANILLA_WAFER) != 0)) { enlist(csp->action->multi[ACTION_MULTI_WAFER], VANILLA_WAFER); } -#endif /* def FEATURE_COOKIE_JAR */ + #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 */ - ((csp->action->flags & ACTION_FILTER) != 0); + (!list_is_empty(csp->action->multi[ACTION_MULTI_FILTER])); gif_deanimate = ((csp->action->flags & ACTION_DEANIMATE) != 0); - /* grab the rest of the client's headers */ - - while (FOREVER) - { - if ( ( p = get_header(csp) ) && ( *p == '\0' ) ) - { - n = read_socket(csp->cfd, buf, sizeof(buf)); - if (n <= 0) - { - log_error(LOG_LEVEL_ERROR, "read from client failed: %E"); - return; - } - add_to_iob(csp, buf, n); - continue; - } - - if (p == NULL) break; - - enlist(csp->headers, p); - freez(p); - } - - /* We have a request. */ + jpeg_inspect = ((csp->action->flags & ACTION_JPEG_INSPECT) != 0); - /* - * Now, check to see if we need to intercept it, i.e. - * If + /* + * We have a request. Now, check to see if we need to + * intercept it, i.e. If .. */ - + if ( - /* a CGI call was detected and answered */ - (NULL != (rsp = dispatch_cgi(csp))) + /* We may not forward the request by rfc2616 sect 14.31 */ + (NULL != (rsp = direct_response(csp))) /* or we are enabled and... */ || (IS_ENABLED_AND ( /* ..the request was blocked */ - ( NULL != (rsp = block_url(csp))) + ( NULL != (rsp = block_url(csp))) /* ..or untrusted */ #ifdef FEATURE_TRUST || ( NULL != (rsp = trust_url(csp))) #endif /* def FEATURE_TRUST */ - /* ..or a fast redirect kicked in */ -#ifdef FEATURE_FAST_REDIRECTS - || (((csp->action->flags & ACTION_FAST_REDIRECTS) != 0) && - (NULL != (rsp = redirect_url(csp)))) -#endif /* def FEATURE_FAST_REDIRECTS */ - )) - ) - { - /* Write the answer to the client */ - if ((write_socket(csp->cfd, rsp->head, rsp->head_length) != rsp->head_length) - || (write_socket(csp->cfd, rsp->body, rsp->content_length) != rsp->content_length)) - { - log_error(LOG_LEVEL_ERROR, "write to: %s failed: %E", http->host); - } + /* ..or a redirect kicked in */ + || ( NULL != (rsp = redirect_url(csp))) + )) + + /* + * .. 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))) + ) + { + /* + * Deliver, log and free the interception + * response. Afterwards we're done here. + */ + send_crunch_response(csp, rsp); #ifdef FEATURE_STATISTICS /* Count as a rejected request */ - csp->rejected = 1; + csp->flags |= CSP_FLAG_REJECTED; #endif /* def FEATURE_STATISTICS */ - /* Log (FIXME: All intercept reasons apprear 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->cmd); - - /* Clean up and return */ - free_http_response(rsp); return; } + hdr = sed(client_patterns, add_client_headers, csp); + if (hdr == NULL) + { + /* FIXME Should handle error properly */ + log_error(LOG_LEVEL_FATAL, "Out of memory parsing client header"); + } + + list_remove_all(csp->headers); + log_error(LOG_LEVEL_GPC, "%s%s", http->hostport, http->path); if (fwd->forward_host) @@ -705,76 +1949,61 @@ static void chat(struct client_state *csp) /* here we connect to the server, gateway, or the forwarder */ - csp->sfd = forwarded_connect(fwd, http, csp); - - if (csp->sfd < 0) + while ( (csp->sfd = forwarded_connect(fwd, http, csp)) + && (errno == EINVAL) && (forwarded_connect_retries++ < max_forwarded_connect_retries)) { - log_error(LOG_LEVEL_CONNECT, "connect to: %s failed: %E", - http->hostport); + log_error(LOG_LEVEL_ERROR, "failed request #%u to connect to %s. Trying again.", + forwarded_connect_retries, http->hostport); + } - if (errno == EINVAL) + if (csp->sfd == JB_INVALID_SOCKET) + { + if (fwd->type != SOCKS_NONE) { - rsp = error_response(csp, "no-such-domain", errno); - - log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" 404 0", - csp->ip_addr_str, http->cmd); + /* Socks error. */ + rsp = error_response(csp, "forwarding-failed", errno); + } + else if (errno == EINVAL) + { + rsp = error_response(csp, "no-such-domain", errno); } else { - rsp = error_response(csp, "connect-failed", errno); - - log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" 503 0", - csp->ip_addr_str, http->cmd); + rsp = error_response(csp, "connect-failed", errno); + log_error(LOG_LEVEL_CONNECT, "connect to: %s failed: %E", + http->hostport); } + /* Write the answer to the client */ - if(rsp) - { - if ((write_socket(csp->cfd, rsp->head, rsp->head_length) != rsp->head_length) - || (write_socket(csp->cfd, rsp->body, rsp->content_length) != rsp->content_length)) - { - log_error(LOG_LEVEL_ERROR, "write to: %s failed: %E", http->host); - } + if(rsp != NULL) + { + send_crunch_response(csp, rsp); } - free_http_response(rsp); + freez(hdr); return; } - log_error(LOG_LEVEL_CONNECT, "OK"); - - hdr = sed(client_patterns, add_client_headers, csp); - list_remove_all(csp->headers); - if (fwd->forward_host || (http->ssl == 0)) { /* write the client's (modified) header to the server * (along with anything else that may be in the buffer) */ - n = strlen(hdr); - - if ((write_socket(csp->sfd, hdr, n) != n) - || (flush_socket(csp->sfd, csp ) < 0)) + if (write_socket(csp->sfd, hdr, strlen(hdr)) + || (flush_socket(csp->sfd, csp) < 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->cmd); - rsp = error_response(csp, "connect-failed", errno); if(rsp) { - if ((write_socket(csp->cfd, rsp->head, n) != n) - || (write_socket(csp->cfd, rsp->body, rsp->content_length) != 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; } @@ -786,10 +2015,7 @@ 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->cmd); - - if (write_socket(csp->cfd, CSUCCEED, sizeof(CSUCCEED)-1) < 0) + if (write_socket(csp->cfd, CSUCCEED, sizeof(CSUCCEED)-1)) { freez(hdr); return; @@ -797,6 +2023,8 @@ 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); @@ -808,14 +2036,21 @@ static void chat(struct client_state *csp) server_body = 0; - while (FOREVER) + for (;;) { +#ifdef __OS2__ + /* + * FD_ZERO here seems to point to an errant macro which crashes. + * So do this by hand for now... + */ + memset(&rfds,0x00,sizeof(fd_set)); +#else FD_ZERO(&rfds); - +#endif FD_SET(csp->cfd, &rfds); FD_SET(csp->sfd, &rfds); - n = select(maxfd+1, &rfds, NULL, NULL, NULL); + n = select((int)maxfd+1, &rfds, NULL, NULL, NULL); if (n < 0) { @@ -829,14 +2064,14 @@ static void chat(struct client_state *csp) if (FD_ISSET(csp->cfd, &rfds)) { - n = read_socket(csp->cfd, buf, sizeof(buf)); + len = read_socket(csp->cfd, buf, sizeof(buf)); - if (n <= 0) + if (len <= 0) { break; /* "game over, man" */ } - if (write_socket(csp->sfd, buf, n) != n) + if (write_socket(csp->sfd, buf, (size_t)len)) { log_error(LOG_LEVEL_ERROR, "write to: %s failed: %E", http->host); return; @@ -854,40 +2089,44 @@ static void chat(struct client_state *csp) if (FD_ISSET(csp->sfd, &rfds)) { fflush( 0 ); - n = read_socket(csp->sfd, buf, sizeof(buf) - 1); + len = read_socket(csp->sfd, buf, sizeof(buf) - 1); - if (n < 0) + 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->cmd); + 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; + } rsp = error_response(csp, "connect-failed", errno); if(rsp) { - if ((write_socket(csp->cfd, rsp->head, rsp->head_length) != rsp->head_length) - || (write_socket(csp->cfd, rsp->body, rsp->content_length) != rsp->content_length)) - { - log_error(LOG_LEVEL_ERROR, "write to: %s failed: %E", http->host); - } - } - - free_http_response(rsp); + send_crunch_response(csp, rsp); + } + return; } /* Add a trailing zero. This lets filter_popups * use string operations. */ - buf[n] = '\0'; + buf[len] = '\0'; #ifdef FEATURE_KILL_POPUPS /* Filter the popups on this read. */ if (block_popups_now) { - filter_popups(buf); + filter_popups(buf, csp); } #endif /* def FEATURE_KILL_POPUPS */ @@ -909,9 +2148,9 @@ static void chat(struct client_state *csp) * doesn't generate a valid header, then we won't * transmit anything to the client. */ - if (n == 0) + if (len == 0) { - + if (server_body || http->ssl) { /* @@ -928,16 +2167,23 @@ static void chat(struct client_state *csp) */ if (NULL == (p = (*content_filter)(csp))) { - csp->content_length = csp->iob->eod - csp->iob->cur; + csp->content_length = (size_t)(csp->iob->eod - csp->iob->cur); } - hdr = sed(server_patterns, add_server_headers, csp); - n = strlen(hdr); + hdr = sed(server_patterns_light, NULL, csp); + + if (hdr == NULL) + { + /* FIXME Should handle error properly */ + log_error(LOG_LEVEL_FATAL, "Out of memory parsing server header"); + } - if ((write_socket(csp->cfd, hdr, n) != n) - || (write_socket(csp->cfd, p != NULL ? p : csp->iob->cur, csp->content_length) != (int)csp->content_length)) + if (write_socket(csp->cfd, hdr, strlen(hdr)) + || write_socket(csp->cfd, p != NULL ? p : csp->iob->cur, csp->content_length)) { - log_error(LOG_LEVEL_CONNECT, "write modified content to client failed: %E"); + log_error(LOG_LEVEL_ERROR, "write modified content to client failed: %E"); + freez(hdr); + freez(p); return; } @@ -949,10 +2195,11 @@ static void chat(struct client_state *csp) } /* - * This is NOT the body, so + * This is NOT the body, so * Let's pretend the server just sent us a blank line. */ - n = sprintf(buf, "\r\n"); + snprintf(buf, sizeof(buf), "\r\n"); + len = (int)strlen(buf); /* * Now, let the normal header parsing algorithm below do its @@ -972,17 +2219,60 @@ static void chat(struct client_state *csp) { if (content_filter) { - add_to_iob(csp, buf, n); + /* + * If there is no memory left for buffering the content, or the buffer limit + * has been reached, switch to non-filtering mode, i.e. make & write the + * header, flush the iob and buf, and get out of the way. + */ + if (add_to_iob(csp, buf, len)) + { + size_t hdrlen; + int flushed; + + log_error(LOG_LEVEL_ERROR, "Flushing header and buffers. Stepping back from filtering."); + + hdr = sed(server_patterns, add_server_headers, csp); + if (hdr == NULL) + { + /* + * Memory is too tight to even generate the header. + * Send our static "Out-of-memory" page. + */ + log_error(LOG_LEVEL_ERROR, "Out of memory while trying to flush."); + rsp = cgi_error_memory(); + send_crunch_response(csp, rsp); + + 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))) + { + log_error(LOG_LEVEL_CONNECT, "Flush header and buffers to client failed: %E"); + + freez(hdr); + return; + } + + byte_count += hdrlen + (size_t)flushed + (size_t)len; + freez(hdr); + content_filter = NULL; + server_body = 1; + + } } else { - if (write_socket(csp->cfd, buf, n) != n) + if (write_socket(csp->cfd, buf, (size_t)len)) { log_error(LOG_LEVEL_ERROR, "write to client failed: %E"); return; } } - byte_count += n; + byte_count += (size_t)len; continue; } else @@ -992,29 +2282,22 @@ static void chat(struct client_state *csp) * parsing an "out of body experience" ? */ - /* buffer up the data we just read */ - add_to_iob(csp, buf, n); - - /* get header lines from the iob */ - - while ((p = get_header(csp))) + /* + * 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)) { - if (*p == '\0') - { - /* see following note */ - break; - } - enlist(csp->headers, p); - freez(p); - } + log_error(LOG_LEVEL_ERROR, "Out of memory while looking for end of server headers."); + rsp = cgi_error_memory(); + send_crunch_response(csp, rsp); - /* 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. - */ + return; + } - if (p) + /* Convert iob into something sed() can digest */ + if (JB_ERR_PARSE == get_server_headers(csp)) { if (ms_iis5_hack) { @@ -1037,17 +2320,27 @@ static void chat(struct client_state *csp) } } + /* Did we actually get anything? */ + if (NULL == csp->headers->first) + { + log_error(LOG_LEVEL_ERROR, "Empty server or forwarder response."); + log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" 502 0", csp->ip_addr_str, http->cmd); + strcpy(buf, NO_SERVER_DATA_RESPONSE); + write_socket(csp->cfd, buf, strlen(buf)); + free_http_request(http); + return; + } + /* we have now received the entire header. * filter it and send the result to the client */ hdr = sed(server_patterns, add_server_headers, csp); - n = strlen(hdr); - - /* write the server's (modified) header to - * the client (along with anything else that - * may be in the buffer) - */ + if (hdr == NULL) + { + /* FIXME Should handle error properly */ + log_error(LOG_LEVEL_FATAL, "Out of memory parsing server header"); + } #ifdef FEATURE_KILL_POPUPS /* Start blocking popups if appropriate. */ @@ -1061,7 +2354,7 @@ static void chat(struct client_state *csp) * Filter the part of the body that came in the same read * as the last headers: */ - filter_popups(csp->iob->cur); + filter_popups(csp->iob->cur, csp); } #endif /* def FEATURE_KILL_POPUPS */ @@ -1077,31 +2370,47 @@ static void chat(struct client_state *csp) /* Buffer and gif_deanimate this if appropriate. */ - if ((csp->content_type & CT_GIF) && /* It's a image/gif MIME-Type */ + if ((csp->content_type & CT_GIF) && /* It's an image/gif MIME-Type */ !http->ssl && /* We talk plaintext */ gif_deanimate) /* Policy allows */ { content_filter = gif_deanimate_response; } + /* 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 */ + { + content_filter = jpeg_inspect_response; + } /* * Only write if we're not buffering for content modification */ - if (!content_filter && ((write_socket(csp->cfd, hdr, n) != n) - || (n = flush_socket(csp->cfd, csp) < 0))) + if (!content_filter) { - 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. + /* write the server's (modified) header to + * the client (along with anything else that + * may be in the buffer) */ - freez(hdr); - return; - } - if(!content_filter) byte_count += n; + if (write_socket(csp->cfd, hdr, strlen(hdr)) + || ((len = flush_socket(csp->cfd, csp)) < 0)) + { + log_error(LOG_LEVEL_CONNECT, "write header to client failed: %E"); + + /* the write failed, so don't bother + * mentioning it to the client... + * it probably can't hear us anyway. + */ + freez(hdr); + return; + } + + byte_count += (size_t)len; + } /* we're finished with the server's header */ @@ -1124,12 +2433,22 @@ static void chat(struct client_state *csp) return; /* huh? we should never get here */ } - log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" 200 %d", - csp->ip_addr_str, http->cmd, byte_count); -} - - -/********************************************************************* + if (csp->content_length == 0) + { + /* + * If Privoxy didn't recalculate the + * Content-Lenght, byte_count is still + * correct. + */ + csp->content_length = byte_count; + } + + log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" 200 %d", + csp->ip_addr_str, http->ocmd, csp->content_length); +} + + +/********************************************************************* * * Function : serve * @@ -1151,12 +2470,12 @@ static void serve(struct client_state *csp) chat(csp); close_socket(csp->cfd); - if (csp->sfd >= 0) + if (csp->sfd != JB_INVALID_SOCKET) { close_socket(csp->sfd); } - csp->active = 0; + csp->flags &= ~CSP_FLAG_ACTIVE; } @@ -1183,6 +2502,99 @@ static int32 server_thread(void *data) #endif +/********************************************************************* + * + * Function : usage + * + * Description : Print usage info & exit. + * + * Parameters : Pointer to argv[0] for identifying ourselves + * + * Returns : No. ,-) + * + *********************************************************************/ +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); + + exit(2); + +} + + +/********************************************************************* + * + * Function : initialize_mutexes + * + * Description : Prepares mutexes if mutex support is available. + * + * Parameters : None + * + * Returns : Void, exits in case of errors. + * + *********************************************************************/ +void initialize_mutexes() +{ + int err = 0; + +#ifdef FEATURE_PTHREAD + /* + * Prepare global mutex semaphores + */ + err = pthread_mutex_init(&log_mutex, 0); + + if (!err) err = pthread_mutex_init(&log_init_mutex, 0); + + /* + * 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 + * thead safe. + */ +#ifndef HAVE_GMTIME_R + if (!err) err = pthread_mutex_init(&gmtime_mutex, 0); +#endif /* ndef HAVE_GMTIME_R */ + +#ifndef HAVE_LOCALTIME_R + if (!err) err = pthread_mutex_init(&localtime_mutex, 0); +#endif /* ndef HAVE_GMTIME_R */ + +#ifndef HAVE_GETHOSTBYADDR_R + if (!err) err = pthread_mutex_init(&gethostbyaddr_mutex, 0); +#endif /* ndef HAVE_GETHOSTBYADDR_R */ + +#ifndef HAVE_GETHOSTBYNAME_R + if (!err) err = pthread_mutex_init(&gethostbyname_mutex, 0); +#endif /* ndef HAVE_GETHOSTBYNAME_R */ + +#ifndef HAVE_RANDOM + if (!err) err = pthread_mutex_init(&rand_mutex, 0); +#endif /* ndef HAVE_RANDOM */ +#endif /* FEATURE_PTHREAD */ + + /* + * TODO: mutex support for mingw32 would be swell. + */ + + if (err) + { + printf("Fatal error. Mutex initialization failed: %s.\n", + strerror(err)); + exit(1); + } + + return; +} + + /********************************************************************* * * Function : main @@ -1202,7 +2614,7 @@ static int32 server_thread(void *data) * any load fails, and can't bind port. * * Else main never returns, the process must be signaled - * to terminate execution. Or, on Windows, use the + * to terminate execution. Or, on Windows, use the * "File", "Exit" menu option. * *********************************************************************/ @@ -1212,42 +2624,145 @@ int real_main(int argc, const char *argv[]) int main(int argc, const char *argv[]) #endif { + int argc_pos = 0; + unsigned int random_seed; +#ifdef unix + struct passwd *pw = NULL; + struct group *grp = NULL; + char *p; + int do_chroot = 0; +#endif + + Argc = argc; + Argv = argv; + configfile = -#ifdef AMIGA - "AmiTCP:db/junkbuster/config" -#elif !defined(_WIN32) +#if !defined(_WIN32) "config" #else - "junkbstr.txt" + "config.txt" #endif ; -#if !defined(_WIN32) || defined(_WIN_CONSOLE) - if ((argc >= 2) && (strcmp(argv[1], "--help")==0)) - { - printf("JunkBuster proxy version " VERSION ".\n\n" - "Usage: %s [configfile]\n\n" - "See " HOME_PAGE_URL " for details.\n" - "This program is distributed under the GNU GPL, version 2 or later.\n", - argv[0]); - exit(2); - } - if ((argc >= 2) && (strcmp(argv[1], "--version")==0)) + /* + * Parse the command line arguments + */ + while (++argc_pos < argc) { - printf(VERSION "\n"); - exit(2); - } -#endif /* !defined(_WIN32) || defined(_WIN_CONSOLE) */ +#ifdef _WIN32 + /* Check to see if the service must be installed or uninstalled */ + if (strncmp(argv[argc_pos], "--install", 9) == 0) + { + const char *pName = argv[argc_pos] + 9; + if (*pName == ':') + pName++; + exit( (install_service(pName)) ? 0 : 1 ); + } + else if (strncmp(argv[argc_pos], "--uninstall", + 11) == 0) + { + const char *pName = argv[argc_pos] + 11; + if (*pName == ':') + pName++; + exit((uninstall_service(pName)) ? 0 : 1); + } + else if (strcmp(argv[argc_pos], "--service" ) == 0) + { + bRunAsService = TRUE; + w32_set_service_cwd(); + atexit(w32_service_exit_notify); + } + else +#endif /* defined(_WIN32) */ - Argc = argc; - Argv = argv; - if (argc > 1) +#if !defined(_WIN32) || defined(_WIN_CONSOLE) + + if (strcmp(argv[argc_pos], "--help") == 0) + { + usage(argv[0]); + } + + else if(strcmp(argv[argc_pos], "--version") == 0) + { + printf("Privoxy version " VERSION " (" HOME_PAGE_URL ")\n"); + exit(0); + } + +#if defined(unix) + + else if (strcmp(argv[argc_pos], "--no-daemon" ) == 0) + { + no_daemon = 1; + } + + else if (strcmp(argv[argc_pos], "--pidfile" ) == 0) + { + if (++argc_pos == argc) usage(argv[0]); + pidfile = strdup(argv[argc_pos]); + } + + else if (strcmp(argv[argc_pos], "--user" ) == 0) + { + if (++argc_pos == argc) usage(argv[argc_pos]); + + if ((NULL != (p = strchr(argv[argc_pos], '.'))) && *(p + 1) != '0') + { + *p++ = '\0'; + if (NULL == (grp = getgrnam(p))) + { + log_error(LOG_LEVEL_FATAL, "Group %s not found.", p); + } + } + + if (NULL == (pw = getpwnam(argv[argc_pos]))) + { + log_error(LOG_LEVEL_FATAL, "User %s not found.", argv[argc_pos]); + } + + if (p != NULL) *--p = '\0'; + } + + else if (strcmp(argv[argc_pos], "--chroot" ) == 0) + { + do_chroot = 1; + } + +#endif /* defined(unix) */ +#endif /* defined(_WIN32) && !defined(_WIN_CONSOLE) */ + { + configfile = argv[argc_pos]; + } + + } /* -END- while (more arguments) */ + +#if defined(unix) + if ( *configfile != '/' ) { - configfile = argv[1]; + char *abs_file, cwd[1024]; + + /* make config-filename absolute here */ + if ( !(getcwd(cwd, sizeof(cwd)))) + { + perror("get working dir failed"); + exit( 1 ); + } + + if (!(basedir = strdup(cwd)) + || (!(abs_file = malloc( strlen( basedir ) + strlen( configfile ) + 5 )))) + { + perror("malloc failed"); + exit( 1 ); + } + strcpy( abs_file, basedir ); + strcat( abs_file, "/" ); + strcat( abs_file, configfile ); + configfile = abs_file; } +#endif /* defined unix */ + files->next = NULL; + clients->next = NULL; #ifdef AMIGA InitAmiga(); @@ -1255,22 +2770,217 @@ int main(int argc, const char *argv[]) InitWin32(); #endif + /* Prepare mutexes if supported and necessary. */ + initialize_mutexes(); -#ifndef _WIN32 - signal(SIGPIPE, SIG_IGN); - signal(SIGCHLD, SIG_IGN); + random_seed = (unsigned int)time(NULL); +#ifdef HAVE_RANDOM + srandom(random_seed); +#else + srand(random_seed); +#endif /* ifdef HAVE_RANDOM */ + /* + * Unix signal handling + * + * Catch the abort, interrupt and terminate signals for a graceful exit + * Catch the hangup signal so the errlog can be reopened. + * Ignore the broken pipe signals (FIXME: Why?) + */ +#if !defined(_WIN32) && !defined(__OS2__) && !defined(AMIGA) +{ + int idx; + const int catched_signals[] = { SIGTERM, SIGINT, SIGHUP, 0 }; + const int ignored_signals[] = { SIGPIPE, 0 }; + + for (idx = 0; catched_signals[idx] != 0; idx++) + { +#ifdef sun /* FIXME: Is it safe to check for HAVE_SIGSET instead? */ + if (sigset(catched_signals[idx], sig_handler) == SIG_ERR) +#else + if (signal(catched_signals[idx], sig_handler) == SIG_ERR) +#endif /* ifdef sun */ + { + log_error(LOG_LEVEL_FATAL, "Can't set signal-handler for signal %d: %E", catched_signals[idx]); + } + } + + for (idx = 0; ignored_signals[idx] != 0; idx++) + { + if (signal(ignored_signals[idx], SIG_IGN) == SIG_ERR) + { + log_error(LOG_LEVEL_FATAL, "Can't set ignore-handler for signal %d: %E", ignored_signals[idx]); + } + } + +} #else /* ifdef _WIN32 */ # ifdef _WIN_CONSOLE /* * 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 */ + /* Initialize the CGI subsystem */ + cgi_init_error_messages(); + + /* + * If runnig on unix and without the --nodaemon + * option, become a daemon. I.e. fork, detach + * from tty and get process group leadership + */ +#if defined(unix) +{ + pid_t pid = 0; +#if 0 + int fd; +#endif + + if (!no_daemon) + { + pid = fork(); + + if ( pid < 0 ) /* error */ + { + perror("fork"); + exit( 3 ); + } + else if ( pid != 0 ) /* parent */ + { + int status; + pid_t wpid; + /* + * must check for errors + * child died due to missing files aso + */ + sleep( 1 ); + wpid = waitpid( pid, &status, WNOHANG ); + if ( wpid != 0 ) + { + exit( 1 ); + } + exit( 0 ); + } + /* child */ +#if 1 + /* Should be more portable, but not as well tested */ + setsid(); +#else /* !1 */ +#ifdef __FreeBSD__ + setpgrp(0,0); +#else /* ndef __FreeBSD__ */ + setpgrp(); +#endif /* ndef __FreeBSD__ */ + fd = open("/dev/tty", O_RDONLY); + if ( fd ) + { + /* no error check here */ + ioctl( fd, TIOCNOTTY,0 ); + 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. + */ + + close( 0 ); + close( 1 ); + chdir("/"); + + } /* -END- if (!no_daemon) */ + + /* + * As soon as we have written the PID file, we can switch + * to the user and group ID indicated by the --user option + */ + write_pid_file(); + + if (NULL != pw) + { + if (setgid((NULL != grp) ? grp->gr_gid : pw->pw_gid)) + { + log_error(LOG_LEVEL_FATAL, "Cannot setgid(): Insufficient permissions."); + } + if (do_chroot) + { + if (!pw->pw_dir) + { + log_error(LOG_LEVEL_FATAL, "Home directory for %s undefined", pw->pw_name); + } + if (chroot(pw->pw_dir) < 0) + { + log_error(LOG_LEVEL_FATAL, "Cannot chroot to %s", pw->pw_dir); + } + if (chdir ("/")) + { + log_error(LOG_LEVEL_FATAL, "Cannot chdir /"); + } + } + if (setuid(pw->pw_uid)) + { + log_error(LOG_LEVEL_FATAL, "Cannot setuid(): Insufficient permissions."); + } + if (do_chroot) + { + char putenv_dummy[64]; + + strcpy(putenv_dummy, "HOME=/"); + if (putenv(putenv_dummy) != 0) + { + log_error(LOG_LEVEL_FATAL, "Cannot putenv(): HOME"); + } + + snprintf(putenv_dummy, 64, "USER=%s", pw->pw_name); + if (putenv(putenv_dummy) != 0) + { + log_error(LOG_LEVEL_FATAL, "Cannot putenv(): USER"); + } + } + } + else if (do_chroot) + { + log_error(LOG_LEVEL_FATAL, "Cannot chroot without --user argument."); + } +} +#endif /* defined unix */ + +#ifdef _WIN32 + /* This will be FALSE unless the command line specified --service + */ + if (bRunAsService) + { + /* Yup, so now we must attempt to establish a connection + * with the service dispatcher. This will only work if this + * process was launched by the service control manager to + * actually run as a service. If this isn't the case, i've + * known it take around 30 seconds or so for the call to return. + */ + + /* The StartServiceCtrlDispatcher won't return until the service is stopping */ + if (w32_start_service_ctrl_dispatcher(w32ServiceDispatchTable)) + { + /* Service has run, and at this point is now being stopped, so just return */ + return 0; + } + +#ifdef _WIN_CONSOLE + printf("Warning: Failed to connect to Service Control Dispatcher\nwhen starting as a service!\n"); +#endif + /* An error occurred. Usually it's because --service was wrongly specified + * and we were unable to connect to the Service Control Dispatcher because + * it wasn't expecting us and is therefore not listening. + * + * For now, just continue below to call the listen_loop function. + */ + } +#endif /* def _WIN32 */ + listen_loop(); /* NOTREACHED */ @@ -1279,6 +2989,89 @@ int main(int argc, const char *argv[]) } +/********************************************************************* + * + * Function : bind_port_helper + * + * Description : Bind the listen port. Handles logging, and aborts + * on failure. + * + * Parameters : + * 1 : config = Privoxy configuration. Specifies port + * to bind to. + * + * Returns : Port that was opened. + * + *********************************************************************/ +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) + { + log_error(LOG_LEVEL_INFO, "Listening on port %d on all IP addresses", + config->hport); + } + else + { + log_error(LOG_LEVEL_INFO, "Listening on port %d on IP address %s", + config->hport, config->haddr); + } + + result = bind_port(config->haddr, config->hport, &bfd); + + if (result < 0) + { + switch(result) + { + case -3 : + log_error(LOG_LEVEL_FATAL, "can't bind to %s:%d: " + "There may be another Privoxy or some other " + "proxy running on port %d", + (NULL != config->haddr) ? config->haddr : "INADDR_ANY", + config->hport, config->hport); + + case -2 : + log_error(LOG_LEVEL_FATAL, "can't bind to %s:%d: " + "The hostname is not resolvable", + (NULL != config->haddr) ? config->haddr : "INADDR_ANY", config->hport); + + default : + log_error(LOG_LEVEL_FATAL, "can't bind to %s:%d: because %E", + (NULL != config->haddr) ? config->haddr : "INADDR_ANY", config->hport); + } + + /* shouldn't get here */ + return JB_INVALID_SOCKET; + } + + config->need_bind = 0; + + return bfd; +} + + +#ifdef _WIN32 +/* Without this simple workaround we get this compiler warning from _beginthread + * warning C4028: formal parameter 1 different from declaration + */ +void w32_service_listen_loop(void *p) +{ + listen_loop(); +} +#endif /* def _WIN32 */ + + /********************************************************************* * * Function : listen_loop @@ -1293,49 +3086,67 @@ int main(int argc, const char *argv[]) static void listen_loop(void) { struct client_state *csp = NULL; - int bfd; + jb_socket bfd; struct configuration_spec * config; config = load_config(); - log_error(LOG_LEVEL_CONNECT, "bind (%s, %d)", - config->haddr ? config->haddr : "INADDR_ANY", config->hport); - - bfd = bind_port(config->haddr, config->hport); - - if (bfd < 0) - { - log_error(LOG_LEVEL_FATAL, "can't bind %s:%d: %E " - "- There may be another junkbuster or some other " - "proxy running on port %d", - (NULL != config->haddr) ? config->haddr : "INADDR_ANY", - config->hport, config->hport - ); - /* shouldn't get here */ - return; - } - - config->need_bind = 0; - + bfd = bind_port_helper(config); - while (FOREVER) +#ifdef FEATURE_GRACEFUL_TERMINATION + while (!g_terminate) +#else + for (;;) +#endif { -#if !defined(FEATURE_PTHREAD) && !defined(_WIN32) && !defined(__BEOS__) && !defined(AMIGA) +#if !defined(FEATURE_PTHREAD) && !defined(_WIN32) && !defined(__BEOS__) && !defined(AMIGA) && !defined(__OS2__) while (waitpid(-1, NULL, WNOHANG) > 0) { /* zombie children */ } #endif /* !defined(FEATURE_PTHREAD) && !defined(_WIN32) && !defined(__BEOS__) && !defined(AMIGA) */ + + /* + * Free data that was used by died threads + */ sweep(); +#if defined(unix) + /* + * Re-open the errlog after HUP signal + */ + if (received_hup_signal) + { + init_error_log(Argv[0], config->logfile, config->debug); + 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)); continue; } - csp->active = 1; - csp->sfd = -1; + csp->flags |= CSP_FLAG_ACTIVE; + csp->sfd = JB_INVALID_SOCKET; csp->config = config = load_config(); @@ -1356,26 +3167,10 @@ static void listen_loop(void) close_socket(bfd); - log_error(LOG_LEVEL_CONNECT, "bind (%s, %d)", - config->haddr ? config->haddr : "INADDR_ANY", config->hport); - bfd = bind_port(config->haddr, config->hport); - - if (bfd < 0) - { - log_error(LOG_LEVEL_FATAL, "can't bind %s:%d: %E " - "- There may be another junkbuster or some other " - "proxy running on port %d", - (NULL != config->haddr) ? config->haddr : "INADDR_ANY", - config->hport, config->hport - ); - /* shouldn't get here */ - return; - } - - config->need_bind = 0; + 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)) { @@ -1384,7 +3179,7 @@ static void listen_loop(void) #ifdef AMIGA if(!childs) { - exit(1); + exit(1); } #endif freez(csp); @@ -1392,12 +3187,14 @@ 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 - /* by haroon - most of credit to srt19170 */ - csp->toggled_on = g_bToggleIJB; + if (global_toggle_state) + { + csp->flags |= CSP_FLAG_TOGGLED_ON; + } #endif /* def FEATURE_TOGGLE */ if (run_loader(csp)) @@ -1409,7 +3206,7 @@ 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; @@ -1436,8 +3233,9 @@ static void listen_loop(void) pthread_attr_init(&attrs); pthread_attr_setdetachstate(&attrs, PTHREAD_CREATE_DETACHED); - child_id = (pthread_create(&the_thread, &attrs, - (void*)serve, csp) ? -1 : 0); + errno = pthread_create(&the_thread, &attrs, + (void * (*)(void *))serve, csp); + child_id = errno ? -1 : 0; pthread_attr_destroy(&attrs); } #endif @@ -1445,7 +3243,16 @@ static void listen_loop(void) #if defined(_WIN32) && !defined(_CYGWIN) && !defined(SELECTED_ONE_OPTION) #define SELECTED_ONE_OPTION child_id = _beginthread( - (void*)serve, + (void (*)(void *))serve, + 64 * 1024, + csp); +#endif + +#if defined(__OS2__) && !defined(SELECTED_ONE_OPTION) +#define SELECTED_ONE_OPTION + child_id = _beginthread( + (void(* _Optlink)(void*))serve, + NULL, 64 * 1024, csp); #endif @@ -1470,13 +3277,23 @@ static void listen_loop(void) #if defined(AMIGA) && !defined(SELECTED_ONE_OPTION) #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)"junkbuster 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; @@ -1494,9 +3311,32 @@ static void listen_loop(void) */ if (child_id == 0) /* child */ { + int rc = 0; +#ifdef FEATURE_TOGGLE + int inherited_toggle_state = global_toggle_state; +#endif /* def FEATURE_TOGGLE */ + serve(csp); - _exit(0); + /* + * If we've been toggled or we've blocked the request, tell Mom + */ + +#ifdef FEATURE_TOGGLE + if (inherited_toggle_state != global_toggle_state) + { + rc |= RC_FLAG_TOGGLED; + } +#endif /* def FEATURE_TOGGLE */ + +#ifdef FEATURE_STATISTICS + if (csp->flags & CSP_FLAG_REJECTED) + { + rc |= RC_FLAG_BLOCKED; + } +#endif /* ndef FEATURE_STATISTICS */ + + _exit(rc); } else if (child_id > 0) /* parent */ { @@ -1504,12 +3344,35 @@ static void listen_loop(void) * copy of the client socket and the CSP * are not used. */ + int child_status; +#if !defined(_WIN32) && !defined(__CYGWIN__) + + wait( &child_status ); + + /* + * Evaluate child's return code: If the child has + * - been toggled, toggle ourselves + * - blocked its request, bump up the stats counter + */ + +#ifdef FEATURE_TOGGLE + if (WIFEXITED(child_status) && (WEXITSTATUS(child_status) & RC_FLAG_TOGGLED)) + { + global_toggle_state = !global_toggle_state; + } +#endif /* def FEATURE_TOGGLE */ + +#ifdef FEATURE_STATISTICS + urls_read++; + if (WIFEXITED(child_status) && (WEXITSTATUS(child_status) & RC_FLAG_BLOCKED)) + { + urls_rejected++; + } +#endif /* def FEATURE_STATISTICS */ -#if !defined(_WIN32) && defined(__CYGWIN__) - wait( NULL ); #endif /* !defined(_WIN32) && defined(__CYGWIN__) */ close_socket(csp->cfd); - csp->active = 0; + csp->flags &= ~CSP_FLAG_ACTIVE; } #endif @@ -1522,11 +3385,11 @@ static void listen_loop(void) log_error(LOG_LEVEL_ERROR, "can't fork: %E"); - sprintf(buf , "JunkBuster: 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); - csp->active = 0; + csp->flags &= ~CSP_FLAG_ACTIVE; sleep(5); continue; } @@ -1536,7 +3399,57 @@ static void listen_loop(void) serve(csp); } } - /* NOTREACHED */ + + /* NOTREACHED unless FEATURE_GRACEFUL_TERMINATION is defined */ + + /* Clean up. Aim: free all memory (no leaks) */ +#ifdef FEATURE_GRACEFUL_TERMINATION + + log_error(LOG_LEVEL_ERROR, "Graceful termination requested"); + + unload_current_config_file(); + unload_current_actions_file(); + unload_current_re_filterfile(); +#ifdef FEATURE_TRUST + unload_current_trust_file(); +#endif + + if (config->multi_threaded) + { + int i = 60; + do + { + sleep(1); + sweep(); + } while ((clients->next != NULL) && (--i > 0)); + + if (i <= 0) + { + log_error(LOG_LEVEL_ERROR, "Graceful termination failed - still some live clients after 1 minute wait."); + } + } + sweep(); + sweep(); + +#if defined(unix) + freez(basedir); +#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(); +#endif + + exit(0); +#endif /* FEATURE_GRACEFUL_TERMINATION */ }