1 const char jcc_rcs[] = "$Id: jcc.c,v 1.169 2008/03/04 18:30:39 fabiankeil Exp $";
2 /*********************************************************************
4 * File : $Source: /cvsroot/ijbswa/current/jcc.c,v $
6 * Purpose : Main file. Contains main() method, main loop, and
7 * the main connection-handling function.
9 * Copyright : Written by and Copyright (C) 2001-2007 the SourceForge
10 * Privoxy team. http://www.privoxy.org/
12 * Based on the Internet Junkbuster originally written
13 * by and Copyright (C) 1997 Anonymous Coders and
14 * Junkbusters Corporation. http://www.junkbusters.com
16 * This program is free software; you can redistribute it
17 * and/or modify it under the terms of the GNU General
18 * Public License as published by the Free Software
19 * Foundation; either version 2 of the License, or (at
20 * your option) any later version.
22 * This program is distributed in the hope that it will
23 * be useful, but WITHOUT ANY WARRANTY; without even the
24 * implied warranty of MERCHANTABILITY or FITNESS FOR A
25 * PARTICULAR PURPOSE. See the GNU General Public
26 * License for more details.
28 * The GNU General Public License should be included with
29 * this file. If not, you can view it at
30 * http://www.gnu.org/copyleft/gpl.html
31 * or write to the Free Software Foundation, Inc., 59
32 * Temple Place - Suite 330, Boston, MA 02111-1307, USA.
36 * Revision 1.169 2008/03/04 18:30:39 fabiankeil
37 * Remove the treat-forbidden-connects-like-blocks action. We now
38 * use the "blocked" page for forbidden CONNECT requests by default.
40 * Revision 1.168 2008/03/02 12:25:25 fabiankeil
41 * Also use shiny new connect_port_is_forbidden() in jcc.c.
43 * Revision 1.167 2008/02/23 16:57:12 fabiankeil
44 * Rename url_actions() to get_url_actions() and let it
45 * use the standard parameter ordering.
47 * Revision 1.166 2008/02/23 16:33:43 fabiankeil
48 * Let forward_url() use the standard parameter ordering
49 * and mark its second parameter immutable.
51 * Revision 1.165 2008/02/02 19:36:56 fabiankeil
52 * Remove the "Listening ... for local connections only" log message.
53 * Whether or not remote connections are able to reach Privoxy is up
54 * to the operating system.
56 * Revision 1.164 2007/12/16 18:32:46 fabiankeil
57 * Prevent the log messages for CONNECT requests to unacceptable
58 * ports from printing the limit-connect argument as [null] if
59 * limit-connect hasn't been explicitly enabled.
61 * Revision 1.163 2007/12/13 01:47:11 david__schmidt
62 * Make sure all console-mode apps get a usage() instance
64 * Revision 1.162 2007/12/06 17:54:57 fabiankeil
65 * Reword NO_SERVER_DATA_RESPONSE to make it harder
66 * to misunderstand what the message is all about.
68 * Revision 1.161 2007/12/04 19:44:22 fabiankeil
69 * Unbreak trustfile which previously didn't work without
70 * FEATURE_TOGGLE. Fixes BR#1843585, reported by Lee.
72 * Revision 1.160 2007/11/29 18:00:29 fabiankeil
73 * Plug memory leak. Spotted by Valgrind, triggered by
74 * Privoxy-Regression-Test feeding proxyfuzz.py.
76 * Revision 1.159 2007/11/24 14:34:09 fabiankeil
77 * In the HTTP snipplets, refer to the client as client.
79 * Revision 1.158 2007/11/11 16:44:17 fabiankeil
80 * Emit a log message when activating the MS IIS5 hack.
82 * Revision 1.157 2007/11/03 17:34:49 fabiankeil
83 * Log the "weak randomization factor" warning only
84 * once for mingw32 and provide some more details.
86 * Revision 1.156 2007/11/01 18:20:58 fabiankeil
87 * Initialize log module after initializing mutexes, future
88 * deadlocks in that code should now work cross-platform.
90 * Revision 1.155 2007/10/23 20:12:45 fabiankeil
91 * Fix first CSUCCEED line to end in \r\n as required by RFC1945.
92 * Reported by Bert van Leeuwen in BR#1818808.
94 * Revision 1.154 2007/10/19 17:00:08 fabiankeil
95 * Downgrade "Flushing header and buffers" message to LOG_LEVEL_INFO.
97 * Revision 1.153 2007/10/14 14:12:41 fabiankeil
98 * When in daemon mode, close stderr after the configuration file has been
99 * parsed the first time. If logfile isn't set, stop logging. Fixes BR#897436.
101 * Revision 1.152 2007/10/04 18:03:34 fabiankeil
102 * - Fix a crash when parsing invalid requests whose first header
103 * is rejected by get_header(). Regression (re?)introduced
104 * in r1.143 by yours truly.
105 * - Move ACTION_VANILLA_WAFER handling into parsers.c's
106 * client_cookie_adder() to make sure send-vanilla-wafer can be
107 * controlled through tags (and thus regression-tested).
109 * Revision 1.151 2007/09/29 10:21:16 fabiankeil
110 * - Move get_filter_function() from jcc.c to filters.c
111 * so the filter functions can be static.
112 * - Don't bother filtering body-less responses.
114 * Revision 1.150 2007/09/28 16:39:29 fabiankeil
115 * Execute content filters through execute_content_filter().
117 * Revision 1.149 2007/09/04 15:08:48 fabiankeil
118 * Initialize req to NULL to make sure it's defined if the
119 * first read_socket() call fails. Reported by icmp30.
121 * Revision 1.148 2007/08/26 16:47:13 fabiankeil
122 * Add Stephen Gildea's --pre-chroot-nslookup patch [#1276666],
123 * extensive comments moved to user manual.
125 * Revision 1.147 2007/08/25 14:42:40 fabiankeil
126 * Don't crash if a broken header filter wiped out the request line.
128 * Revision 1.146 2007/08/20 17:09:32 fabiankeil
129 * Fix byte_count calculation in case of flushes
130 * and don't parse the server headers a second time.
132 * Revision 1.145 2007/08/19 13:13:31 fabiankeil
133 * - If there's a connection problem after we already forwarded
134 * parts of the original content, just hang up. Fixes BR#1776724.
135 * - Fix warnings about unused code on mingw32.
136 * - In case of flushes, calculate the byte count
137 * less incorrectly (I think).
139 * Revision 1.144 2007/08/11 14:43:22 fabiankeil
140 * Add some more prototypes for static functions.
142 * Revision 1.143 2007/08/05 13:58:19 fabiankeil
143 * Comment out request_contains_null_bytes() until it's used again.
145 * Revision 1.142 2007/08/05 13:50:26 fabiankeil
146 * #1763173 from Stefan Huehner: s@const static@static const@
147 * and declare some more functions static.
149 * Revision 1.141 2007/08/04 09:56:23 fabiankeil
150 * - Log rejected CONNECT requests with LOG_LEVEL_INFO
151 * and explain why they were rejected in the first place.
152 * - Fix the LOG_LEVEL_CLF message for crunches of unallowed
153 * CONNECT requests. The request line was missing.
154 * - Add two more XXX reminders as we don't have enough already.
156 * Revision 1.140 2007/07/21 11:51:36 fabiankeil
157 * As Hal noticed, checking dispatch_cgi() as the last cruncher
158 * looks like a bug if CGI requests are blocked unintentionally,
159 * so don't do it unless the user enabled the new config option
160 * "allow-cgi-request-crunching".
162 * Revision 1.139 2007/07/14 07:46:41 fabiankeil
163 * - Allow to rewrite the request destination behind the client's back.
164 * - Turn the weird-looking unconditional for loop that
165 * reads the client request into a conditional while loop.
166 * Move the stuff that only runs once out of the loop.
167 * - Move parts of chat(), server_content_type() and the
168 * necessary stuff to fix BR#1750917 into get_filter_function().
170 * Revision 1.138 2007/06/03 18:45:18 fabiankeil
171 * Temporary workaround for BR#1730105.
173 * Revision 1.137 2007/06/01 18:16:36 fabiankeil
174 * Use the same mutex for gethostbyname() and gethostbyaddr() to prevent
175 * deadlocks and crashes on OpenBSD and possibly other OS with neither
176 * gethostbyname_r() nor gethostaddr_r(). Closes BR#1729174.
177 * Thanks to Ralf Horstmann for report and solution.
179 * Revision 1.136 2007/06/01 16:41:11 fabiankeil
180 * Add forward-override{} to change the forwarding settings through
181 * action sections. This is mainly interesting to forward different
182 * clients differently (for example based on User-Agent or request
185 * Revision 1.135 2007/05/24 17:03:50 fabiankeil
186 * - Let usage() mention the --chroot parameter.
187 * - Use read_socket() consistently and always leave
188 * the last buffer byte alone, even in cases where
189 * null termination (currently) doesn't matter.
191 * Revision 1.134 2007/05/16 14:59:46 fabiankeil
192 * - Fix config file loading on Unix if no config file is specified.
193 * Since r1.97 Privoxy would always interpret the last argument as
194 * config file, even if it's a valid command line option.
195 * - Abort in case of unrecognized command line options. Closes #1719696.
196 * - Remove a bunch of unnecessary strcpy() calls (yay for c&p without thinking).
197 * - Replace the remaining strcpy() and strcat() calls with strlcpy() and strcat().
199 * Revision 1.133 2007/05/04 11:23:19 fabiankeil
200 * - Don't rerun crunchers that only depend on the request URL.
201 * - Don't count redirects and CGI requests as "blocked requests".
203 * Revision 1.132 2007/04/25 15:15:17 fabiankeil
204 * Support crunching based on tags created by server-header taggers.
206 * Revision 1.131 2007/04/22 13:24:50 fabiankeil
207 * Make HTTP snippets static (again). Add a Content-Type for those
208 * with content so the browser doesn't guess it based on the URL.
210 * Revision 1.130 2007/04/19 13:47:34 fabiankeil
211 * Move crunching and request line rebuilding out of chat().
213 * Revision 1.129 2007/04/15 16:39:20 fabiankeil
214 * Introduce tags as alternative way to specify which
215 * actions apply to a request. At the moment tags can be
216 * created based on client and server headers.
218 * Revision 1.128 2007/03/25 16:55:54 fabiankeil
219 * Don't CLF-log CONNECT requests twice.
221 * Revision 1.127 2007/03/20 13:53:17 fabiankeil
222 * Log the source address for ACL-related connection drops.
224 * Revision 1.126 2007/03/17 15:20:05 fabiankeil
225 * New config option: enforce-blocks.
227 * Revision 1.125 2007/03/09 14:12:00 fabiankeil
228 * - Move null byte check into separate function.
229 * - Don't confuse the client with error pages
230 * if a CONNECT request was already confirmed.
232 * Revision 1.124 2007/02/23 14:59:54 fabiankeil
233 * Speed up NULL byte escaping and only log the complete
234 * NULL byte requests with header debugging enabled.
236 * Revision 1.123 2007/02/21 18:42:10 fabiankeil
237 * Answer requests that contain NULL bytes with
238 * a custom response instead of waiting for more
239 * data until the client eventually hangs up.
241 * Revision 1.122 2007/02/07 11:12:02 fabiankeil
242 * - Move delivery and logging of crunched responses
243 * from chat() into send_crunch_response().
244 * - Display the reason for generating http_responses.
245 * - Log the content length for LOG_LEVEL_CLF correctly
246 * (still incorrect for some fixed responses).
247 * - Reword an incorrect comment about
248 * treat-forbidden-connects-like-blocks violating
250 * - Add some log messages.
252 * Revision 1.121 2007/01/27 10:52:56 fabiankeil
253 * Move mutex initialization into separate
254 * function and exit in case of errors.
256 * Revision 1.120 2007/01/26 14:18:42 fabiankeil
257 * - Start to reduce chat()'s line count and move
258 * parts of it into separate functions.
259 * - Add "HTTP/1.1 100 Continue" hack for BR 756734.
261 * Revision 1.119 2007/01/25 14:02:30 fabiankeil
262 * - Add Proxy-Agent header to HTTP snippets that are
263 * supposed to reach HTTP clients only.
264 * - Made a few CONNECT log messages more descriptive.
265 * - Catch completely empty server responses (as seen
266 * with Tor's fake ".noconnect" top level domain).
267 * - Use shiny new "forwarding-failed" template for socks errors.
269 * Revision 1.118 2007/01/07 07:43:43 joergs
270 * AmigaOS4 support added.
272 * Revision 1.117 2006/12/31 17:56:37 fabiankeil
273 * Added config option accept-intercepted-requests
274 * and disabled it by default.
276 * Revision 1.116 2006/12/29 19:08:22 fabiankeil
277 * Reverted parts of my last commit
278 * to keep error handling working.
280 * Revision 1.115 2006/12/29 17:38:57 fabiankeil
281 * Fixed gcc43 conversion warnings.
283 * Revision 1.114 2006/12/27 18:52:02 fabiankeil
284 * Fix -pedantic ISO C warning about converting
285 * from function pointer to object pointer.
287 * Revision 1.113 2006/12/26 17:38:50 fabiankeil
288 * Silence compiler warning I introduced with my last commit.
290 * Revision 1.112 2006/12/26 17:31:41 fabiankeil
291 * Mutex protect rand() if POSIX threading
292 * is used, warn the user if that's not possible
293 * and stop using it on _WIN32 where it could
296 * Revision 1.111 2006/12/23 16:15:06 fabiankeil
297 * Don't prevent core dumps by catching SIGABRT.
298 * It's rude and makes debugging unreasonable painful.
300 * Revision 1.110 2006/12/13 14:52:53 etresoft
301 * Fix build failure on MacOS X. Global symbols can be either static or extern, but not both.
303 * Revision 1.109 2006/12/06 19:41:40 fabiankeil
304 * Privoxy is now able to run as intercepting
305 * proxy in combination with any packet filter
306 * that does the port redirection. The destination
307 * is extracted from the "Host:" header which
308 * should be available for nearly all requests.
310 * Moved HTTP snipplets into jcc.c.
311 * Added error message for gopher proxy requests.
313 * Revision 1.108 2006/11/28 15:38:51 fabiankeil
314 * Only unlink the pidfile if it's actually used.
316 * Change order of interception checks to make
317 * it possible to block or redirect requests for
320 * Revision 1.107 2006/11/13 19:05:51 fabiankeil
321 * Make pthread mutex locking more generic. Instead of
322 * checking for OSX and OpenBSD, check for FEATURE_PTHREAD
323 * and use mutex locking unless there is an _r function
324 * available. Better safe than sorry.
326 * Fixes "./configure --disable-pthread" and should result
327 * in less threading-related problems on pthread-using platforms,
328 * but it still doesn't fix BR#1122404.
330 * Revision 1.106 2006/11/06 19:58:23 fabiankeil
331 * Move pthread.h inclusion from jcc.c to jcc.h.
332 * Fixes build on x86-freebsd1 (FreeBSD 5.4-RELEASE).
334 * Revision 1.105 2006/11/06 14:26:02 fabiankeil
335 * Don't exit after receiving the second SIGHUP on Solaris.
337 * Fixes BR 1052235, but the same problem may exist on other
338 * systems. Once 3.0.6 is out we should use sigset()
339 * where available and see if it breaks anything.
341 * Revision 1.104 2006/09/23 13:26:38 roro
342 * Replace TABs by spaces in source code.
344 * Revision 1.103 2006/09/21 12:54:43 fabiankeil
345 * Fix +redirect{}. Didn't work with -fast-redirects.
347 * Revision 1.102 2006/09/06 13:03:04 fabiankeil
348 * Respond with 400 and a short text message
349 * if the client tries to use Privoxy as FTP proxy.
351 * Revision 1.101 2006/09/06 09:23:37 fabiankeil
352 * Make number of retries in case of forwarded-connect problems
353 * a config file option (forwarded-connect-retries) and use 0 as
356 * Revision 1.100 2006/09/03 19:42:59 fabiankeil
357 * Set random(3) seed.
359 * Revision 1.99 2006/09/02 15:36:42 fabiankeil
360 * Follow the OpenBSD port's lead and protect the resolve
361 * functions on OpenBSD as well.
363 * Revision 1.98 2006/08/24 11:01:34 fabiankeil
364 * --user fix. Only use the user as group if no group is specified.
365 * Solves BR 1492612. Thanks to Spinor S. and David Laight.
367 * Revision 1.97 2006/08/18 15:23:17 david__schmidt
368 * Windows service (re-)integration
372 * --install[:service_name]
373 * --uninstall[:service_name]
376 * They work as follows:
377 * --install will create a service for you and then terminate.
378 * By default the service name will be "privoxy" (without the quotes).
379 * However you can run multiple services if you wish, just by adding
380 * a colon and then a name (no spaces).
382 * --uninstall follows the exact same rules a --install.
384 * --service is used when the program is executed by the service
385 * control manager, and in normal circumstances would never be
386 * used as a command line argument.
388 * Revision 1.96 2006/08/15 20:12:36 david__schmidt
389 * Windows service integration
391 * Revision 1.95 2006/08/03 02:46:41 david__schmidt
392 * Incorporate Fabian Keil's patch work:
393 http://www.fabiankeil.de/sourcecode/privoxy/
395 * Revision 1.94 2006/07/18 14:48:46 david__schmidt
396 * Reorganizing the repository: swapping out what was HEAD (the old 3.1 branch)
397 * with what was really the latest development (the v_3_0_branch branch)
399 * Revision 1.92.2.16 2005/04/03 20:10:50 david__schmidt
400 * Thanks to Jindrich Makovicka for a race condition fix for the log
401 * file. The race condition remains for non-pthread implementations.
402 * Reference patch #1175720.
404 * Revision 1.92.2.15 2004/10/03 12:53:32 david__schmidt
405 * Add the ability to check jpeg images for invalid
406 * lengths of comment blocks. Defensive strategy
407 * against the exploit:
408 * Microsoft Security Bulletin MS04-028
409 * Buffer Overrun in JPEG Processing (GDI+) Could
410 * Allow Code Execution (833987)
411 * Enabled with +inspect-jpegs in actions files.
413 * Revision 1.92.2.14 2003/12/12 12:52:53 oes
414 * - Fixed usage info for non-unix platforms
415 * - Fixed small cmdline parsing bug
417 * Revision 1.92.2.13 2003/11/27 19:20:27 oes
418 * Diagnostics: Now preserve the returncode of pthread_create
419 * in errno. Closes BR #775721. Thanks to Geoffrey Hausheer.
421 * Revision 1.92.2.12 2003/07/11 11:34:19 oes
422 * No longer ignore SIGCHLD. Fixes bug #769381
424 * Revision 1.92.2.11 2003/05/14 12:32:02 oes
425 * Close jarfile on graceful exit, remove stray line
427 * Revision 1.92.2.10 2003/05/08 15:13:46 oes
428 * Cosmetics: Killed a warning, a typo and an allocation left at exit
430 * Revision 1.92.2.9 2003/04/03 15:08:42 oes
431 * No longer rely on non-POSIX.1 extensions of getcwd().
434 * Revision 1.92.2.8 2003/03/31 13:12:32 oes
435 * Replaced setenv() by posix-compliant putenv()
436 * Thanks to Neil McCalden (nmcc AT users.sf.net).
438 * Revision 1.92.2.7 2003/03/17 16:48:59 oes
439 * Added chroot ability, thanks to patch by Sviatoslav Sviridov
441 * Revision 1.92.2.6 2003/03/11 11:55:00 oes
442 * Clean-up and extension of improvements for forked mode:
443 * - Child's return code now consists of flags RC_FLAG_*
444 * - Reporting toggle to parent now properly #ifdef'ed
445 * - Children now report blocking to parent. This enables
446 * statistics in forked mode
448 * Revision 1.92.2.5 2003/03/10 23:45:32 oes
449 * Fixed bug #700381: Non-Threaded version now capable of being toggled.
450 * Children now report having been toggled through _exit(17), parents
451 * watch for that code and toggle themselves if found.
453 * Revision 1.92.2.4 2003/03/07 03:41:04 david__schmidt
454 * Wrapping all *_r functions (the non-_r versions of them) with
455 * mutex semaphores for OSX. Hopefully this will take care of all
456 * of those pesky crash reports.
458 * Revision 1.92.2.3 2003/02/28 12:53:06 oes
459 * Fixed two mostly harmless mem leaks
461 * Revision 1.92.2.2 2002/11/20 14:37:47 oes
462 * Fix: Head of global clients list now initialized to NULL
464 * Revision 1.92.2.1 2002/09/25 14:52:24 oes
465 * Added basic support for OPTIONS and TRACE HTTP methods:
466 * - New interceptor direct_response() added in chat().
467 * - sed() moved to earlier in the process, so that the
468 * Host: header is evaluated before actions and forwarding
471 * Revision 1.92 2002/05/08 16:00:46 oes
472 * Chat's buffer handling:
473 * - Fixed bug with unchecked out-of-mem conditions
474 * while reading client request & server headers
475 * - No longer predict if the buffer limit will be exceeded
476 * in the next read -- check add_to_iob's new
477 * return code. If buffer couldn't be extended
478 * (policy or out-of-mem) while
479 * - reading from client: abort
480 * - reading server headers: send error page
481 * - buffering server body for filter: flush,
482 * and if that fails: send error page
484 * Revision 1.91 2002/04/08 20:35:58 swa
487 * Revision 1.90 2002/04/02 14:57:28 oes
488 * Made sending wafers independent of FEATURE_COOKIE_JAR
490 * Revision 1.89 2002/03/31 17:18:59 jongfoster
491 * Win32 only: Enabling STRICT to fix a VC++ compile warning.
493 * Revision 1.88 2002/03/27 14:32:43 david__schmidt
494 * More compiler warning message maintenance
496 * Revision 1.87 2002/03/26 22:29:54 swa
497 * we have a new homepage!
499 * Revision 1.86 2002/03/25 17:04:55 david__schmidt
500 * Workaround for closing the jarfile before load_config() comes around again
502 * Revision 1.85 2002/03/24 15:23:33 jongfoster
505 * Revision 1.84 2002/03/24 13:25:43 swa
506 * name change related issues
508 * Revision 1.83 2002/03/16 23:54:06 jongfoster
509 * Adding graceful termination feature, to help look for memory leaks.
510 * If you enable this (which, by design, has to be done by hand
511 * editing config.h) and then go to http://i.j.b/die, then the program
512 * will exit cleanly after the *next* request. It should free all the
513 * memory that was used.
515 * Revision 1.82 2002/03/13 00:27:05 jongfoster
518 * Revision 1.81 2002/03/12 01:42:50 oes
519 * Introduced modular filters
521 * Revision 1.80 2002/03/11 22:07:05 david__schmidt
522 * OS/2 port maintenance:
523 * - Fixed EMX build - it had decayed a little
524 * - Fixed inexplicable crash during FD_ZERO - must be due to a bad macro.
525 * substituted a memset for now.
527 * Revision 1.79 2002/03/09 20:03:52 jongfoster
528 * - Making various functions return int rather than size_t.
529 * (Undoing a recent change). Since size_t is unsigned on
530 * Windows, functions like read_socket that return -1 on
531 * error cannot return a size_t.
533 * THIS WAS A MAJOR BUG - it caused frequent, unpredictable
534 * crashes, and also frequently caused JB to jump to 100%
535 * CPU and stay there. (Because it thought it had just
536 * read ((unsigned)-1) == 4Gb of data...)
538 * - The signature of write_socket has changed, it now simply
539 * returns success=0/failure=nonzero.
541 * - Trying to get rid of a few warnings --with-debug on
542 * Windows, I've introduced a new type "jb_socket". This is
543 * used for the socket file descriptors. On Windows, this
544 * is SOCKET (a typedef for unsigned). Everywhere else, it's
545 * an int. The error value can't be -1 any more, so it's
546 * now JB_INVALID_SOCKET (which is -1 on UNIX, and in
547 * Windows it maps to the #define INVALID_SOCKET.)
549 * - The signature of bind_port has changed.
551 * Revision 1.78 2002/03/08 21:35:04 oes
552 * Added optional group supplement to --user option. Will now use default group of user if no group given
554 * Revision 1.77 2002/03/07 03:52:06 oes
555 * - Fixed compiler warnings etc
556 * - Improved handling of failed DNS lookups
558 * Revision 1.76 2002/03/06 22:54:35 jongfoster
559 * Automated function-comment nitpicking.
561 * Revision 1.75 2002/03/06 10:02:19 oes
562 * Fixed stupid bug when --user was not given
564 * Revision 1.74 2002/03/06 00:49:31 jongfoster
565 * Fixing warning on Windows
566 * Making #ifdefs that refer to the same variable consistently
567 * use #ifdef unix rather than mixing #ifdef unix & #ifndef OS2
569 * Revision 1.73 2002/03/05 23:57:30 hal9
570 * Stray character 's' on line 1618 was breaking build.
572 * Revision 1.72 2002/03/05 21:33:45 david__schmidt
573 * - Re-enable OS/2 building after new parms were added
574 * - Fix false out of memory report when resolving CGI templates when no IP
575 * address is available of failed attempt (a la no such domain)
577 * Revision 1.71 2002/03/05 18:13:56 oes
578 * Added --user option
580 * Revision 1.70 2002/03/05 04:52:42 oes
581 * Deleted non-errlog debugging code
583 * Revision 1.69 2002/03/04 23:50:00 jongfoster
584 * Splitting off bind_port() call into bind_port_helper(), with
587 * Revision 1.68 2002/03/04 20:17:32 oes
590 * Revision 1.67 2002/03/04 18:18:57 oes
591 * - Removed _DEBUG mode
592 * - Cleand up cmdline parsing
593 * - Introduced --no-daemon, --pidfile options
594 * - Cleaned up signal handling:
595 * - Terminate cleanly on INT, TERM and ABRT
596 * - Schedule logfile for re-opening on HUP
597 * - Ignore CHLD and PIPE
598 * - Leave the rest with their default handlers
599 * - Uniform handler registration
600 * - Added usage() function
601 * - Played styleguide police
603 * Revision 1.66 2002/03/03 15:06:55 oes
604 * Re-enabled automatic config reloading
606 * Revision 1.65 2002/03/03 14:49:11 oes
607 * Fixed CLF logging: Now uses client's original HTTP request
609 * Revision 1.64 2002/03/03 09:18:03 joergs
610 * Made jumbjuster work on AmigaOS again.
612 * Revision 1.63 2002/03/02 04:14:50 david__schmidt
613 * Clean up a little CRLF unpleasantness that suddenly appeared
615 * Revision 1.62 2002/02/20 23:17:23 jongfoster
616 * Detecting some out-of memory conditions and exiting with a log message.
618 * Revision 1.61 2002/01/17 21:01:52 jongfoster
619 * Moving all our URL and URL pattern parsing code to urlmatch.c.
621 * Revision 1.60 2001/12/30 14:07:32 steudten
622 * - Add signal handling (unix)
623 * - Add SIGHUP handler (unix)
624 * - Add creation of pidfile (unix)
625 * - Add action 'top' in rc file (RH)
626 * - Add entry 'SIGNALS' to manpage
627 * - Add exit message to logfile (unix)
629 * Revision 1.59 2001/12/13 14:07:18 oes
630 * Fixed Bug: 503 error page now sent OK
632 * Revision 1.58 2001/11/30 23:37:24 jongfoster
633 * Renaming the Win32 config file to config.txt - this is almost the
634 * same as the corresponding UNIX name "config"
636 * Revision 1.57 2001/11/16 00:47:43 jongfoster
637 * Changing the tty-disconnection code to use setsid().
639 * Revision 1.56 2001/11/13 20:20:54 jongfoster
640 * Tabs->spaces, fixing a bug with missing {} around an if()
642 * Revision 1.55 2001/11/13 20:14:53 jongfoster
643 * Patch for FreeBSD setpgrp() as suggested by Alexander Lazic
645 * Revision 1.54 2001/11/07 00:03:14 steudten
646 * Give reliable return value if an error
647 * occurs not just 0 with new daemon mode.
649 * Revision 1.53 2001/11/05 21:41:43 steudten
650 * Add changes to be a real daemon just for unix os.
651 * (change cwd to /, detach from controlling tty, set
652 * process group and session leader to the own process.
654 * Add some fatal-error log message for failed malloc().
655 * Add '-d' if compiled with 'configure --with-debug' to
656 * enable debug output.
658 * Revision 1.52 2001/10/26 20:11:20 jongfoster
659 * Fixing type mismatch
661 * Revision 1.51 2001/10/26 17:38:28 oes
664 * Revision 1.50 2001/10/25 03:40:48 david__schmidt
665 * Change in porting tactics: OS/2's EMX porting layer doesn't allow multiple
666 * threads to call select() simultaneously. So, it's time to do a real, live,
667 * native OS/2 port. See defines for __EMX__ (the porting layer) vs. __OS2__
668 * (native). Both versions will work, but using __OS2__ offers multi-threading.
670 * Revision 1.49 2001/10/23 21:41:35 jongfoster
671 * Added call to initialize the (statically-allocated of course)
672 * "out of memory" CGI response.
674 * Revision 1.48 2001/10/10 19:56:46 jongfoster
675 * Moving some code that wasn't cookie-related out of an #ifdef
678 * Revision 1.47 2001/10/10 16:44:36 oes
679 * Added CONNECT destination port limitation check
681 * Revision 1.46 2001/10/08 15:17:41 oes
682 * Re-enabled SSL forwarding
684 * Revision 1.45 2001/10/07 15:42:11 oes
685 * Replaced 6 boolean members of csp with one bitmap (csp->flags)
687 * Moved downgrading of the HTTP version from parse_http_request to
688 * chat(), since we can't decide if it is necessary before we have
689 * determined the actions for the URL. The HTTP command is now
690 * *always* re-built so the repairs need no longer be special-cased.
692 * filter_popups now gets a csp pointer so it can raise the new
693 * CSP_FLAG_MODIFIED flag.
697 * Added configurable size limit for the IOB. If the IOB grows so
698 * large that the next read would exceed the limit, the header
699 * is generated, and the header & unfiltered buffer are flushed
700 * to the client. Chat then continues in non-buffering,
701 * non-filtering body mode.
703 * Revision 1.44 2001/10/02 18:13:57 oes
706 * Revision 1.43 2001/10/02 15:32:13 oes
707 * Moved generation of hdr
709 * Revision 1.42 2001/09/21 23:02:02 david__schmidt
710 * Cleaning up 2 compiler warnings on OS/2.
712 * Revision 1.41 2001/09/16 17:05:14 jongfoster
713 * Removing unused #include showarg.h
715 * Revision 1.40 2001/09/16 15:41:45 jongfoster
716 * Fixing signed/unsigned comparison warning.
718 * Revision 1.39 2001/09/16 13:21:27 jongfoster
719 * Changes to use new list functions.
721 * Revision 1.38 2001/09/16 13:01:46 jongfoster
722 * Removing redundant function call that zeroed zalloc()'d memory.
724 * Revision 1.37 2001/09/10 11:12:24 oes
725 * Deleted unused variable
727 * Revision 1.36 2001/09/10 10:56:15 oes
728 * Silenced compiler warnings
730 * Revision 1.35 2001/07/31 14:44:22 oes
731 * Deleted unused size parameter from filter_popups()
733 * Revision 1.34 2001/07/30 22:08:36 jongfoster
734 * Tidying up #defines:
735 * - All feature #defines are now of the form FEATURE_xxx
736 * - Permanently turned off WIN_GUI_EDIT
737 * - Permanently turned on WEBDAV and SPLIT_PROXY_ARGS
739 * Revision 1.33 2001/07/29 19:32:00 jongfoster
740 * Renaming _main() [mingw32 only] to real_main(), for ANSI compliance.
742 * Revision 1.32 2001/07/29 18:47:05 jongfoster
743 * Adding missing #include "loadcfg.h"
745 * Revision 1.31 2001/07/29 12:17:48 oes
746 * Applied pthread fix by Paul Lieverse
748 * Revision 1.30 2001/07/25 22:57:13 jongfoster
749 * __BEOS__ no longer overrides FEATURE_PTHREAD.
750 * This is because FEATURE_PTHREAD will soon be widely used, so I
751 * want to keep it simple.
753 * Revision 1.29 2001/07/24 12:47:06 oes
754 * Applied BeOS support update by Eugenia
756 * Revision 1.28 2001/07/23 13:26:12 oes
757 * Fixed bug in popup-killing for the first read that caused binary garbage to be sent between headers and body
759 * Revision 1.27 2001/07/19 19:09:47 haroon
760 * - Added code to take care of the situation where while processing the first
761 * server response (which includes the server header), after finding the end
762 * of the headers we were not looking past the end of the headers for
763 * content modification. I enabled it for filter_popups.
764 * Someone else should look to see if other similar operations should be
765 * done to the discarded portion of the buffer.
767 * Note 2001/07/20: No, the other content modification mechanisms will process
768 * the whole iob later anyway. --oes
770 * Revision 1.26 2001/07/18 12:31:36 oes
773 * Revision 1.25 2001/07/15 19:43:49 jongfoster
774 * Supports POSIX threads.
775 * Also removed some unused #includes.
777 * Revision 1.24 2001/07/13 14:00:40 oes
778 * - Generic content modification scheme:
779 * Each feature has its own applicability flag that is set
780 * from csp->action->flags.
781 * Replaced the "filtering" int flag , by a function pointer
782 * "content_filter" to the function that will do the content
783 * modification. If it is != NULL, the document will be buffered
784 * and processed through *content_filter, which must set
785 * csp->content_length and return a modified copy of the body
786 * or return NULL (on failiure).
787 * - Changed csp->is_text to the more generic bitmap csp->content_type
788 * which can currently take the valued CT_TEXT or CT_GIF
790 * - Removed all #ifdef PCRS
792 * Revision 1.23 2001/07/02 02:28:25 iwanttokeepanon
793 * Added "#ifdef ACL_FILES" conditional compilation to line 1291 to exclude
794 * the `block_acl' call. This prevents a compilation error when the user
795 * does not wish to use the "ACL" feature.
797 * Revision 1.22 2001/06/29 21:45:41 oes
798 * Indentation, CRLF->LF, Tab-> Space
800 * Revision 1.21 2001/06/29 13:29:36 oes
801 * - Cleaned up, improved comments
802 * - Unified all possible interceptors (CGI,
803 * block, trust, fast_redirect) in one
804 * place, with one (CGI) answer generation
805 * mechansim. Much clearer now.
806 * - Removed the GIF image generation, which
807 * is now done in filters.c:block_url()
808 * - Made error conditions like domain lookup
809 * failiure or (various) problems while talking
810 * to the server use cgi.c:error_response()
811 * instead of generating HTML/HTTP in chat() (yuck!)
812 * - Removed logentry from cancelled commit
814 * Revision 1.20 2001/06/09 10:55:28 jongfoster
815 * Changing BUFSIZ ==> BUFFER_SIZE
817 * Revision 1.19 2001/06/07 23:12:52 jongfoster
818 * Replacing function pointer in struct gateway with a directly
819 * called function forwarded_connect().
820 * Replacing struct gateway with struct forward_spec
822 * Revision 1.18 2001/06/03 19:12:16 oes
823 * introduced new cgi handling
825 * Revision 1.17 2001/06/01 20:07:23 jongfoster
826 * Now uses action +image-blocker{} rather than config->tinygif
828 * Revision 1.16 2001/06/01 18:49:17 jongfoster
829 * Replaced "list_share" with "list" - the tiny memory gain was not
830 * worth the extra complexity.
832 * Revision 1.15 2001/05/31 21:24:47 jongfoster
833 * Changed "permission" to "action" throughout.
834 * Removed DEFAULT_USER_AGENT - it must now be specified manually.
835 * Moved vanilla wafer check into chat(), since we must now
836 * decide whether or not to add it based on the URL.
838 * Revision 1.14 2001/05/29 20:14:01 joergs
839 * AmigaOS bugfix: PCRS needs a lot of stack, stacksize for child threads
842 * Revision 1.13 2001/05/29 09:50:24 jongfoster
843 * Unified blocklist/imagelist/permissionslist.
844 * File format is still under discussion, but the internal changes
847 * Also modified interceptor behaviour:
848 * - We now intercept all URLs beginning with one of the following
849 * prefixes (and *only* these prefixes):
851 * * http://ijbswa.sf.net/config/
852 * * http://ijbswa.sourceforge.net/config/
853 * - New interceptors "home page" - go to http://i.j.b/ to see it.
854 * - Internal changes so that intercepted and fast redirect pages
855 * are not replaced with an image.
856 * - Interceptors now have the option to send a binary page direct
857 * to the client. (i.e. ijb-send-banner uses this)
858 * - Implemented show-url-info interceptor. (Which is why I needed
859 * the above interceptors changes - a typical URL is
860 * "http://i.j.b/show-url-info?url=www.somesite.com/banner.gif".
861 * The previous mechanism would not have intercepted that, and
862 * if it had been intercepted then it then it would have replaced
865 * Revision 1.12 2001/05/27 22:17:04 oes
867 * - re_process_buffer no longer writes the modified buffer
868 * to the client, which was very ugly. It now returns the
869 * buffer, which it is then written by chat.
871 * - content_length now adjusts the Content-Length: header
872 * for modified documents rather than crunch()ing it.
873 * (Length info in csp->content_length, which is 0 for
874 * unmodified documents)
876 * - For this to work, sed() is called twice when filtering.
878 * Revision 1.11 2001/05/26 17:27:53 jongfoster
879 * Added support for CLF and fixed LOG_LEVEL_LOG.
880 * Also did CRLF->LF fix of my previous patch.
882 * Revision 1.10 2001/05/26 15:26:15 jongfoster
883 * ACL feature now provides more security by immediately dropping
884 * connections from untrusted hosts.
886 * Revision 1.9 2001/05/26 00:28:36 jongfoster
887 * Automatic reloading of config file.
888 * Removed obsolete SIGHUP support (Unix) and Reload menu option (Win32).
889 * Most of the global variables have been moved to a new
890 * struct configuration_spec, accessed through csp->config->globalname
891 * Most of the globals remaining are used by the Win32 GUI.
893 * Revision 1.8 2001/05/25 22:43:18 jongfoster
894 * Fixing minor memory leak and buffer overflow.
896 * Revision 1.7 2001/05/25 22:34:30 jongfoster
899 * Revision 1.6 2001/05/23 00:13:58 joergs
900 * AmigaOS support fixed.
902 * Revision 1.5 2001/05/22 18:46:04 oes
904 * - Enabled filtering banners by size rather than URL
905 * by adding patterns that replace all standard banner
906 * sizes with the "Junkbuster" gif to the re_filterfile
908 * - Enabled filtering WebBugs by providing a pattern
909 * which kills all 1x1 images
911 * - Added support for PCRE_UNGREEDY behaviour to pcrs,
912 * which is selected by the (nonstandard and therefore
913 * capital) letter 'U' in the option string.
914 * It causes the quantifiers to be ungreedy by default.
915 * Appending a ? turns back to greedy (!).
917 * - Added a new interceptor ijb-send-banner, which
918 * sends back the "Junkbuster" gif. Without imagelist or
919 * MSIE detection support, or if tinygif = 1, or the
920 * URL isn't recognized as an imageurl, a lame HTML
921 * explanation is sent instead.
923 * - Added new feature, which permits blocking remote
924 * script redirects and firing back a local redirect
926 * The feature is conditionally compiled, i.e. it
927 * can be disabled with --disable-fast-redirects,
928 * plus it must be activated by a "fast-redirects"
929 * line in the config file, has its own log level
930 * and of course wants to be displayed by show-proxy-args
931 * Note: Boy, all the #ifdefs in 1001 locations and
932 * all the fumbling with configure.in and acconfig.h
933 * were *way* more work than the feature itself :-(
935 * - Because a generic redirect template was needed for
936 * this, tinygif = 3 now uses the same.
938 * - Moved GIFs, and other static HTTP response templates
943 * - Removed some >400 CRs again (Jon, you really worked
946 * Revision 1.4 2001/05/21 19:34:01 jongfoster
947 * Made failure to bind() a fatal error.
949 * Revision 1.3 2001/05/20 01:21:20 jongfoster
950 * Version 2.9.4 checkin.
951 * - Merged popupfile and cookiefile, and added control over PCRS
952 * filtering, in new "permissionsfile".
953 * - Implemented LOG_LEVEL_FATAL, so that if there is a configuration
954 * file error you now get a message box (in the Win32 GUI) rather
955 * than the program exiting with no explanation.
956 * - Made killpopup use the PCRS MIME-type checking and HTTP-header
958 * - Removed tabs from "config"
959 * - Moved duplicated url parsing code in "loaders.c" to a new funcition.
960 * - Bumped up version number.
962 * Revision 1.2 2001/05/17 22:34:44 oes
963 * - Added hint on GIF char array generation to jcc.c
964 * - Cleaned CRLF's from the sources and related files
965 * - Repaired logging for REF and FRC
967 * Revision 1.1.1.1 2001/05/15 13:58:56 oes
968 * Initial import of version 2.9.3 source tree
971 *********************************************************************/
977 #include <sys/types.h>
986 # ifndef FEATURE_PTHREAD
990 # include <windows.h>
991 # include <process.h>
992 # endif /* ndef FEATURE_PTHREAD */
995 # ifndef _WIN_CONSOLE
997 # endif /* ndef _WIN_CONSOLE */
998 # include "w32svrapi.h"
1000 #else /* ifndef _WIN32 */
1002 # if !defined (__OS2__)
1003 # include <unistd.h>
1004 # include <sys/wait.h>
1005 # endif /* ndef __OS2__ */
1006 # include <sys/time.h>
1007 # include <sys/stat.h>
1008 # include <sys/ioctl.h>
1011 #include <sys/termios.h>
1019 # include <signal.h>
1022 # include <socket.h> /* BeOS has select() for sockets only. */
1023 # include <OS.h> /* declarations for threads and stuff. */
1026 # if defined(__EMX__) || defined(__OS2__)
1027 # include <sys/select.h> /* OS/2/EMX needs a little help with select */
1032 #define bzero(B,N) memset(B,0x00,n)
1036 # include <select.h>
1041 #include "project.h"
1044 #include "filters.h"
1045 #include "loaders.h"
1046 #include "parsers.h"
1047 #include "killpopup.h"
1048 #include "miscutil.h"
1050 #include "jbsockets.h"
1051 #include "gateway.h"
1052 #include "actions.h"
1054 #include "loadcfg.h"
1055 #include "urlmatch.h"
1057 const char jcc_h_rcs[] = JCC_H_VERSION;
1058 const char project_h_rcs[] = PROJECT_H_VERSION;
1061 struct client_state clients[1];
1062 struct file_list files[1];
1064 #ifdef FEATURE_STATISTICS
1065 int urls_read = 0; /* total nr of urls read inc rejected */
1066 int urls_rejected = 0; /* total nr of urls rejected */
1067 #endif /* def FEATURE_STATISTICS */
1069 #ifdef FEATURE_GRACEFUL_TERMINATION
1070 int g_terminate = 0;
1073 #if !defined(_WIN32) && !defined(__OS2__) && !defined(AMIGA)
1074 static void sig_handler(int the_signal);
1076 static int client_protocol_is_unsupported(const struct client_state *csp, char *req);
1077 static jb_err get_request_destination_elsewhere(struct client_state *csp, struct list *headers);
1078 static jb_err get_server_headers(struct client_state *csp);
1079 static const char *crunch_reason(const struct http_response *rsp);
1080 static void send_crunch_response(struct client_state *csp, struct http_response *rsp);
1082 * static int request_contains_null_bytes(const struct client_state *csp, char *buf, int len);
1084 static void build_request_line(struct client_state *csp, const struct forward_spec *fwd, char **request_line);
1085 static jb_err change_request_destination(struct client_state *csp);
1086 static void chat(struct client_state *csp);
1087 static void serve(struct client_state *csp);
1088 #if !defined(_WIN32) || defined(_WIN_CONSOLE)
1089 static void usage(const char *myname);
1091 static void initialize_mutexes(void);
1092 static jb_socket bind_port_helper(struct configuration_spec *config);
1093 static void listen_loop(void);
1096 void serve(struct client_state *csp);
1097 #else /* ifndef AMIGA */
1098 static void serve(struct client_state *csp);
1099 #endif /* def AMIGA */
1102 static int32 server_thread(void *data);
1103 #endif /* def __BEOS__ */
1106 #define sleep(N) Sleep(((N) * 1000))
1110 #define sleep(N) DosSleep(((N) * 100))
1113 #ifdef FEATURE_PTHREAD
1114 pthread_mutex_t log_mutex;
1115 pthread_mutex_t log_init_mutex;
1117 #if !defined(HAVE_GETHOSTBYADDR_R) || !defined(HAVE_GETHOSTBYNAME_R)
1118 pthread_mutex_t resolver_mutex;
1119 #endif /* !defined(HAVE_GETHOSTBYADDR_R) || !defined(HAVE_GETHOSTBYNAME_R) */
1121 #ifndef HAVE_GMTIME_R
1122 pthread_mutex_t gmtime_mutex;
1123 #endif /* ndef HAVE_GMTIME_R */
1125 #ifndef HAVE_LOCALTIME_R
1126 pthread_mutex_t localtime_mutex;
1127 #endif /* ndef HAVE_GMTIME_R */
1130 pthread_mutex_t rand_mutex;
1131 #endif /* ndef HAVE_RANDOM */
1133 #endif /* FEATURE_PTHREAD */
1135 #if defined(unix) || defined(__EMX__)
1136 const char *basedir = NULL;
1137 const char *pidfile = NULL;
1138 int received_hup_signal = 0;
1139 #endif /* defined unix */
1141 /* HTTP snipplets. */
1142 static const char CSUCCEED[] =
1143 "HTTP/1.0 200 Connection established\r\n"
1144 "Proxy-Agent: Privoxy/" VERSION "\r\n\r\n";
1146 static const char CHEADER[] =
1147 "HTTP/1.0 400 Invalid header received from client\r\n"
1148 "Proxy-Agent: Privoxy " VERSION "\r\n"
1149 "Content-Type: text/plain\r\n"
1150 "Connection: close\r\n\r\n"
1151 "Invalid header received from client.\r\n";
1153 static const char FTP_RESPONSE[] =
1154 "HTTP/1.0 400 Invalid request received from client\r\n"
1155 "Content-Type: text/plain\r\n"
1156 "Connection: close\r\n\r\n"
1157 "Invalid request. Privoxy doesn't support FTP.\r\n";
1159 static const char GOPHER_RESPONSE[] =
1160 "HTTP/1.0 400 Invalid request received from client\r\n"
1161 "Content-Type: text/plain\r\n"
1162 "Connection: close\r\n\r\n"
1163 "Invalid request. Privoxy doesn't support gopher.\r\n";
1165 /* XXX: should be a template */
1166 static const char MISSING_DESTINATION_RESPONSE[] =
1167 "HTTP/1.0 400 Bad request received from client\r\n"
1168 "Proxy-Agent: Privoxy " VERSION "\r\n"
1169 "Content-Type: text/plain\r\n"
1170 "Connection: close\r\n\r\n"
1171 "Bad request. Privoxy was unable to extract the destination.\r\n";
1173 /* XXX: should be a template */
1174 static const char NO_SERVER_DATA_RESPONSE[] =
1175 "HTTP/1.0 502 Server or forwarder response empty\r\n"
1176 "Proxy-Agent: Privoxy " VERSION "\r\n"
1177 "Content-Type: text/plain\r\n"
1178 "Connection: close\r\n\r\n"
1179 "Empty server or forwarder response.\r\n"
1180 "The connection has been closed but Privoxy didn't receive any data.\r\n";
1183 /* XXX: should be a template */
1184 static const char NULL_BYTE_RESPONSE[] =
1185 "HTTP/1.0 400 Bad request received from client\r\n"
1186 "Proxy-Agent: Privoxy " VERSION "\r\n"
1187 "Content-Type: text/plain\r\n"
1188 "Connection: close\r\n\r\n"
1189 "Bad request. Null byte(s) before end of request.\r\n";
1192 /* XXX: should be a template */
1193 static const char MESSED_UP_REQUEST_RESPONSE[] =
1194 "HTTP/1.0 400 Malformed request after rewriting\r\n"
1195 "Proxy-Agent: Privoxy " VERSION "\r\n"
1196 "Content-Type: text/plain\r\n"
1197 "Connection: close\r\n\r\n"
1198 "Bad request. Messed up with header filters.\r\n";
1200 /* A function to crunch a response */
1201 typedef struct http_response *(*crunch_func_ptr)(struct client_state *);
1203 /* Crunch function flags */
1204 #define CF_NO_FLAGS 0
1205 /* Cruncher applies to forced requests as well */
1206 #define CF_IGNORE_FORCE 1
1207 /* Crunched requests are counted for the block statistics */
1208 #define CF_COUNT_AS_REJECT 2
1210 /* A crunch function and its flags */
1213 const crunch_func_ptr cruncher;
1217 static int crunch_response_triggered(struct client_state *csp, const struct cruncher crunchers[]);
1219 /* Complete list of cruncher functions */
1220 static const struct cruncher crunchers_all[] = {
1221 { direct_response, CF_COUNT_AS_REJECT|CF_IGNORE_FORCE},
1222 { block_url, CF_COUNT_AS_REJECT },
1223 #ifdef FEATURE_TRUST
1224 { trust_url, CF_COUNT_AS_REJECT },
1225 #endif /* def FEATURE_TRUST */
1226 { redirect_url, CF_NO_FLAGS },
1227 { dispatch_cgi, CF_IGNORE_FORCE},
1231 /* Light version, used after tags are applied */
1232 static const struct cruncher crunchers_light[] = {
1233 { block_url, CF_COUNT_AS_REJECT },
1234 { redirect_url, CF_NO_FLAGS },
1239 #if !defined(_WIN32) && !defined(__OS2__) && !defined(AMIGA)
1240 /*********************************************************************
1242 * Function : sig_handler
1244 * Description : Signal handler for different signals.
1245 * Exit gracefully on TERM and INT
1246 * or set a flag that will cause the errlog
1247 * to be reopened by the main thread on HUP.
1250 * 1 : the_signal = the signal cause this function to call
1254 *********************************************************************/
1255 static void sig_handler(int the_signal)
1261 log_error(LOG_LEVEL_INFO, "exiting by signal %d .. bye", the_signal);
1272 received_hup_signal = 1;
1277 * We shouldn't be here, unless we catch signals
1278 * in main() that we can't handle here!
1280 log_error(LOG_LEVEL_FATAL, "sig_handler: exiting on unexpected signal %d", the_signal);
1288 /*********************************************************************
1290 * Function : client_protocol_is_unsupported
1292 * Description : Checks if the client used a known unsupported
1293 * protocol and deals with it by sending an error
1297 * 1 : csp = Current client state (buffers, headers, etc...)
1298 * 2 : req = the first request line send by the client
1300 * Returns : TRUE if an error response has been generated, or
1301 * FALSE if the request doesn't look invalid.
1303 *********************************************************************/
1304 static int client_protocol_is_unsupported(const struct client_state *csp, char *req)
1306 char buf[BUFFER_SIZE];
1309 * If it's a FTP or gopher request, we don't support it.
1311 * These checks are better than nothing, but they might
1312 * not work in all configurations and some clients might
1313 * have problems digesting the answer.
1315 * They should, however, never cause more problems than
1316 * Privoxy's old behaviour (returning the misleading HTML
1319 * "Could not resolve http://(ftp|gopher)://example.org").
1321 if (!strncmpic(req, "GET ftp://", 10) || !strncmpic(req, "GET gopher://", 13))
1323 if (!strncmpic(req, "GET ftp://", 10))
1325 strlcpy(buf, FTP_RESPONSE, sizeof(buf));
1326 log_error(LOG_LEVEL_ERROR, "%s tried to use Privoxy as FTP proxy: %s",
1327 csp->ip_addr_str, req);
1331 strlcpy(buf, GOPHER_RESPONSE, sizeof(buf));
1332 log_error(LOG_LEVEL_ERROR, "%s tried to use Privoxy as gopher proxy: %s",
1333 csp->ip_addr_str, req);
1335 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" 400 0", csp->ip_addr_str, req);
1337 write_socket(csp->cfd, buf, strlen(buf));
1346 /*********************************************************************
1348 * Function : get_request_destination_elsewhere
1350 * Description : If the client's request was redirected into
1351 * Privoxy without the client's knowledge,
1352 * the request line lacks the destination host.
1354 * This function tries to get it elsewhere,
1355 * provided accept-intercepted-requests is enabled.
1357 * "Elsewhere" currently only means "Host: header",
1358 * but in the future we may ask the redirecting
1359 * packet filter to look the destination up.
1361 * If the destination stays unknown, an error
1362 * response is send to the client and headers
1363 * are freed so that chat() can return directly.
1366 * 1 : csp = Current client state (buffers, headers, etc...)
1367 * 2 : headers = a header list
1369 * Returns : JB_ERR_OK if the destination is now known, or
1370 * JB_ERR_PARSE if it isn't.
1372 *********************************************************************/
1373 static jb_err get_request_destination_elsewhere(struct client_state *csp, struct list *headers)
1377 if (!(csp->config->feature_flags & RUNTIME_FEATURE_ACCEPT_INTERCEPTED_REQUESTS))
1379 log_error(LOG_LEVEL_ERROR, "%s's request: \'%s\' is invalid."
1380 " Privoxy isn't configured to accept intercepted requests.",
1381 csp->ip_addr_str, csp->http->cmd);
1382 /* XXX: Use correct size */
1383 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" 400 0",
1384 csp->ip_addr_str, csp->http->cmd);
1386 write_socket(csp->cfd, CHEADER, strlen(CHEADER));
1387 destroy_list(headers);
1389 return JB_ERR_PARSE;
1391 else if (JB_ERR_OK == get_destination_from_headers(headers, csp->http))
1393 /* Split the domain we just got for pattern matching */
1394 init_domain_components(csp->http);
1400 /* We can't work without destination. Go spread the news.*/
1402 req = list_to_text(headers);
1404 /* XXX: Use correct size */
1405 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" 400 0",
1406 csp->ip_addr_str, csp->http->cmd);
1407 log_error(LOG_LEVEL_ERROR,
1408 "Privoxy was unable to get the destination for %s's request:\n%s\n%s",
1409 csp->ip_addr_str, csp->http->cmd, req);
1412 write_socket(csp->cfd, MISSING_DESTINATION_RESPONSE, strlen(MISSING_DESTINATION_RESPONSE));
1413 destroy_list(headers);
1415 return JB_ERR_PARSE;
1418 * TODO: If available, use PF's ioctl DIOCNATLOOK as last resort
1419 * to get the destination IP address, use it as host directly
1420 * or do a reverse DNS lookup first.
1425 /*********************************************************************
1427 * Function : get_server_headers
1429 * Description : Parses server headers in iob and fills them
1430 * into csp->headers so that they can later be
1434 * 1 : csp = Current client state (buffers, headers, etc...)
1436 * Returns : JB_ERR_OK if everything went fine, or
1437 * JB_ERR_PARSE if the headers were incomplete.
1439 *********************************************************************/
1440 static jb_err get_server_headers(struct client_state *csp)
1442 int continue_hack_in_da_house = 0;
1445 while (((header = get_header(csp)) != NULL) || continue_hack_in_da_house)
1450 * continue hack in da house. Ignore the ending of
1451 * this head and continue enlisting header lines.
1452 * The reason is described below.
1454 enlist(csp->headers, "");
1455 continue_hack_in_da_house = 0;
1458 else if (0 == strncmpic(header, "HTTP/1.1 100", 12))
1461 * It's a bodyless continue response, don't
1462 * stop header parsing after reaching it's end.
1464 * As a result Privoxy will concatenate the
1465 * next response's head and parse and deliver
1466 * the headers as if they belonged to one request.
1468 * The client will separate them because of the
1469 * empty line between them.
1471 * XXX: What we're doing here is clearly against
1472 * the intended purpose of the continue header,
1473 * and under some conditions (HTTP/1.0 client request)
1474 * it's a standard violation.
1476 * Anyway, "sort of against the spec" is preferable
1477 * to "always getting confused by Continue responses"
1478 * (Privoxy's behaviour before this hack was added)
1480 log_error(LOG_LEVEL_HEADER, "Continue hack in da house.");
1481 continue_hack_in_da_house = 1;
1483 else if (*header == '\0')
1486 * If the header is empty, but the Continue hack
1487 * isn't active, we can assume that we reached the
1488 * end of the buffer before we hit the end of the
1491 * Inform the caller an let it decide how to handle it.
1493 return JB_ERR_PARSE;
1497 if (JB_ERR_MEMORY == enlist(csp->headers, header))
1500 * XXX: Should we quit the request and return a
1501 * out of memory error page instead?
1503 log_error(LOG_LEVEL_ERROR,
1504 "Out of memory while enlisting server headers. %s lost.",
1514 /*********************************************************************
1516 * Function : crunch_reason
1518 * Description : Translates the crunch reason code into a string.
1521 * 1 : rsp = a http_response
1523 * Returns : A string with the crunch reason or an error description.
1525 *********************************************************************/
1526 static const char *crunch_reason(const struct http_response *rsp)
1528 char * reason = NULL;
1530 assert(rsp != NULL);
1533 return "Internal error while searching for crunch reason";
1536 switch (rsp->reason)
1538 case RSP_REASON_UNSUPPORTED:
1539 reason = "Unsupported HTTP feature";
1541 case RSP_REASON_BLOCKED:
1544 case RSP_REASON_UNTRUSTED:
1545 reason = "Untrusted";
1547 case RSP_REASON_REDIRECTED:
1548 reason = "Redirected";
1550 case RSP_REASON_CGI_CALL:
1551 reason = "CGI Call";
1553 case RSP_REASON_NO_SUCH_DOMAIN:
1554 reason = "DNS failure";
1556 case RSP_REASON_FORWARDING_FAILED:
1557 reason = "Forwarding failed";
1559 case RSP_REASON_CONNECT_FAILED:
1560 reason = "Connection failure";
1562 case RSP_REASON_OUT_OF_MEMORY:
1563 reason = "Out of memory (may mask other reasons)";
1566 reason = "No reason recorded";
1574 /*********************************************************************
1576 * Function : send_crunch_response
1578 * Description : Delivers already prepared response for
1579 * intercepted requests, logs the interception
1580 * and frees the response.
1583 * 1 : csp = Current client state (buffers, headers, etc...)
1584 * 1 : rsp = Fully prepared response. Will be freed on exit.
1586 * Returns : Nothing.
1588 *********************************************************************/
1589 static void send_crunch_response(struct client_state *csp, struct http_response *rsp)
1591 const struct http_request *http = csp->http;
1592 char status_code[4];
1594 assert(rsp != NULL);
1595 assert(rsp->head != NULL);
1600 * Not supposed to happen. If it does
1601 * anyway, treat it as an unknown error.
1603 cgi_error_unknown(csp, rsp, RSP_REASON_INTERNAL_ERROR);
1604 /* return code doesn't matter */
1609 /* If rsp is still NULL, we have serious internal problems. */
1610 log_error(LOG_LEVEL_FATAL,
1611 "NULL response in send_crunch_response and cgi_error_unknown failed as well.");
1615 * Extract the status code from the actual head
1616 * that was send to the client. It is the only
1617 * way to get it right for all requests, including
1618 * the fixed ones for out-of-memory problems.
1620 * A head starts like this: 'HTTP/1.1 200...'
1624 status_code[0] = rsp->head[9];
1625 status_code[1] = rsp->head[10];
1626 status_code[2] = rsp->head[11];
1627 status_code[3] = '\0';
1629 /* Write the answer to the client */
1630 if (write_socket(csp->cfd, rsp->head, rsp->head_length)
1631 || write_socket(csp->cfd, rsp->body, rsp->content_length))
1633 /* There is nothing we can do about it. */
1634 log_error(LOG_LEVEL_ERROR, "write to: %s failed: %E", csp->http->host);
1637 /* Log that the request was crunched and why. */
1638 log_error(LOG_LEVEL_GPC, "%s%s crunch! (%s)",
1639 http->hostport, http->path, crunch_reason(rsp));
1640 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" %s %d",
1641 csp->ip_addr_str, http->ocmd, status_code, rsp->content_length);
1643 /* Clean up and return */
1644 if (cgi_error_memory() != rsp)
1646 free_http_response(rsp);
1653 /*********************************************************************
1655 * Function : request_contains_null_bytes
1657 * Description : Checks for NULL bytes in the request and sends
1658 * an error message to the client if any were found.
1660 * XXX: currently not used, see comment in chat().
1663 * 1 : csp = Current client state (buffers, headers, etc...)
1664 * 2 : buf = Data from the client's request to check.
1665 * 3 : len = The data length.
1667 * Returns : TRUE if the request contained one or more NULL bytes, or
1670 *********************************************************************/
1671 static int request_contains_null_bytes(const struct client_state *csp, char *buf, int len)
1673 size_t c_len; /* Request lenght when treated as C string */
1675 c_len = strlen(buf);
1680 * Null byte(s) found. Log the request,
1681 * return an error response and hang up.
1683 size_t tmp_len = c_len;
1688 * Replace NULL byte(s) with '°' characters
1689 * so the request can be logged as string.
1690 * XXX: Is there a better replacement character?
1693 tmp_len += strlen(buf+tmp_len);
1694 } while (tmp_len < len);
1696 log_error(LOG_LEVEL_ERROR, "%s\'s request contains at least one NULL byte "
1697 "(length=%d, strlen=%d).", csp->ip_addr_str, len, c_len);
1698 log_error(LOG_LEVEL_HEADER,
1699 "Offending request data with NULL bytes turned into \'°\' characters: %s", buf);
1701 write_socket(csp->cfd, NULL_BYTE_RESPONSE, strlen(NULL_BYTE_RESPONSE));
1703 /* XXX: Log correct size */
1704 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"Invalid request\" 400 0", csp->ip_addr_str);
1714 /*********************************************************************
1716 * Function : crunch_response_triggered
1718 * Description : Checks if the request has to be crunched,
1719 * and delivers the crunch response if necessary.
1722 * 1 : csp = Current client state (buffers, headers, etc...)
1723 * 2 : crunchers = list of cruncher functions to run
1725 * Returns : TRUE if the request was answered with a crunch response
1728 *********************************************************************/
1729 static int crunch_response_triggered(struct client_state *csp, const struct cruncher crunchers[])
1731 struct http_response *rsp = NULL;
1732 const struct cruncher *c;
1735 * If CGI request crunching is disabled,
1736 * check the CGI dispatcher out of order to
1737 * prevent unintentional blocks or redirects.
1739 if (!(csp->config->feature_flags & RUNTIME_FEATURE_CGI_CRUNCHING)
1740 && (NULL != (rsp = dispatch_cgi(csp))))
1742 /* Deliver, log and free the interception response. */
1743 send_crunch_response(csp, rsp);
1747 for (c = crunchers; c->cruncher != NULL; c++)
1750 * Check the cruncher if either Privoxy is toggled
1751 * on and the request isn't forced, or if the cruncher
1752 * applies to forced requests as well.
1754 if (((csp->flags & CSP_FLAG_TOGGLED_ON) &&
1755 !(csp->flags & CSP_FLAG_FORCED)) ||
1756 (c->flags & CF_IGNORE_FORCE))
1758 rsp = c->cruncher(csp);
1761 /* Deliver, log and free the interception response. */
1762 send_crunch_response(csp, rsp);
1763 #ifdef FEATURE_STATISTICS
1764 if (c->flags & CF_COUNT_AS_REJECT)
1766 csp->flags |= CSP_FLAG_REJECTED;
1768 #endif /* def FEATURE_STATISTICS */
1779 /*********************************************************************
1781 * Function : build_request_line
1783 * Description : Builds the HTTP request line.
1785 * If a HTTP forwarder is used it expects the whole URL,
1786 * web servers only get the path.
1789 * 1 : csp = Current client state (buffers, headers, etc...)
1790 * 2 : fwd = The forwarding spec used for the request
1791 * XXX: Should use http->fwd instead.
1792 * 3 : request_line = The old request line which will be replaced.
1794 * Returns : Nothing. Terminates in case of memory problems.
1796 *********************************************************************/
1797 static void build_request_line(struct client_state *csp, const struct forward_spec *fwd, char **request_line)
1799 struct http_request *http = csp->http;
1801 assert(http->ssl == 0);
1804 * Downgrade http version from 1.1 to 1.0
1805 * if +downgrade action applies.
1807 if ( (csp->action->flags & ACTION_DOWNGRADE)
1808 && (!strcmpic(http->ver, "HTTP/1.1")))
1811 http->ver = strdup("HTTP/1.0");
1813 if (http->ver == NULL)
1815 log_error(LOG_LEVEL_FATAL, "Out of memory downgrading HTTP version");
1820 * Rebuild the request line.
1822 freez(*request_line);
1823 *request_line = strdup(http->gpc);
1824 string_append(request_line, " ");
1826 if (fwd->forward_host)
1828 string_append(request_line, http->url);
1832 string_append(request_line, http->path);
1834 string_append(request_line, " ");
1835 string_append(request_line, http->ver);
1837 if (*request_line == NULL)
1839 log_error(LOG_LEVEL_FATAL, "Out of memory writing HTTP command");
1841 log_error(LOG_LEVEL_HEADER, "New HTTP Request-Line: %s", *request_line);
1845 /*********************************************************************
1847 * Function : change_request_destination
1849 * Description : Parse a (rewritten) request line and regenerate
1850 * the http request data.
1853 * 1 : csp = Current client state (buffers, headers, etc...)
1855 * Returns : Forwards the parse_http_request() return code.
1856 * Terminates in case of memory problems.
1858 *********************************************************************/
1859 static jb_err change_request_destination(struct client_state *csp)
1861 struct http_request *http = csp->http;
1864 log_error(LOG_LEVEL_INFO, "Rewrite detected: %s", csp->headers->first->str);
1865 free_http_request(http);
1866 err = parse_http_request(csp->headers->first->str, http, csp);
1867 if (JB_ERR_OK != err)
1869 log_error(LOG_LEVEL_ERROR, "Couldn't parse rewritten request: %s.",
1870 jb_err_to_string(err));
1872 http->ocmd = strdup(http->cmd); /* XXX: ocmd is a misleading name */
1873 if (http->ocmd == NULL)
1875 log_error(LOG_LEVEL_FATAL, "Out of memory copying rewritten HTTP request line");
1882 /*********************************************************************
1886 * Description : Once a connection to the client has been accepted,
1887 * this function is called (via serve()) to handle the
1888 * main business of the communication. When this
1889 * function returns, the caller must close the client
1892 * FIXME: chat is nearly thousand lines long.
1896 * 1 : csp = Current client state (buffers, headers, etc...)
1898 * Returns : Nothing.
1900 *********************************************************************/
1901 static void chat(struct client_state *csp)
1903 char buf[BUFFER_SIZE];
1911 int ms_iis5_hack = 0;
1912 size_t byte_count = 0;
1913 int forwarded_connect_retries = 0;
1914 int max_forwarded_connect_retries = csp->config->forwarded_connect_retries;
1915 const struct forward_spec * fwd;
1916 struct http_request *http;
1917 int len; /* for buffer sizes (and negative error codes) */
1919 #ifdef FEATURE_KILL_POPUPS
1920 int block_popups_now = 0; /* bool, 1==currently blocking popups */
1921 #endif /* def FEATURE_KILL_POPUPS */
1923 /* Function that does the content filtering for the current request */
1924 filter_function_ptr content_filter = NULL;
1926 /* Skeleton for HTTP response, if we should intercept the request */
1927 struct http_response *rsp;
1929 /* Temporary copy of the client's headers before they get enlisted in csp->headers */
1930 struct list header_list;
1931 struct list *headers = &header_list;
1935 memset(buf, 0, sizeof(buf));
1938 * Read the client's request. Note that since we're not using select() we
1939 * could get blocked here if a client connected, then didn't say anything!
1944 len = read_socket(csp->cfd, buf, sizeof(buf) - 1);
1946 if (len <= 0) break; /* error! */
1949 * If there is no memory left for buffering the
1950 * request, there is nothing we can do but hang up
1952 if (add_to_iob(csp, buf, len))
1957 req = get_header(csp);
1959 } while ((NULL != req) && ('\0' == *req));
1961 if ((NULL != req) && ('\0' != *req))
1963 /* Request received. Validate and parse it. */
1967 * XXX: Temporary disabled to prevent problems
1968 * with POST requests whose bodies are allowed to
1969 * contain NULL bytes. BR#1730105.
1971 * The main purpose of this check is to properly
1972 * log stuff like BitTorrent traffic and other junk
1973 * that hits public proxies. It's not required for
1974 * Privoxy to functions as those requests are discarded
1977 * It probably should be rewritten to only check
1978 * the head of the request. Another option would
1979 * be to let all POST requests pass, although that
1980 * may not be good enough.
1982 if (request_contains_null_bytes(csp, buf, len))
1984 /* NULL bytes found and dealt with, just hang up. */
1989 /* Does the request line look invalid? */
1990 if (client_protocol_is_unsupported(csp, req))
1993 * Yes. The request has already been
1994 * answered with a error response, the buffers
1995 * were freed and we're done with chatting.
2000 #ifdef FEATURE_FORCE_LOAD
2002 * If this request contains the FORCE_PREFIX and blocks
2003 * aren't enforced, get rid of it and set the force flag.
2005 if (strstr(req, FORCE_PREFIX))
2007 if (csp->config->feature_flags & RUNTIME_FEATURE_ENFORCE_BLOCKS)
2009 log_error(LOG_LEVEL_FORCE,
2010 "Ignored force prefix in request: \"%s\".", req);
2014 strclean(req, FORCE_PREFIX);
2015 log_error(LOG_LEVEL_FORCE, "Enforcing request: \"%s\".", req);
2016 csp->flags |= CSP_FLAG_FORCED;
2020 #endif /* def FEATURE_FORCE_LOAD */
2021 err = parse_http_request(req, http, csp);
2022 if (JB_ERR_OK != err)
2024 log_error(LOG_LEVEL_ERROR, "Couldn't parse request: %s.", jb_err_to_string(err));
2030 if (http->cmd == NULL)
2032 write_socket(csp->cfd, CHEADER, strlen(CHEADER));
2033 /* XXX: Use correct size */
2034 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"Invalid request\" 400 0", csp->ip_addr_str);
2035 log_error(LOG_LEVEL_ERROR, "Invalid header received from %s.", csp->ip_addr_str);
2037 free_http_request(http);
2041 /* grab the rest of the client's headers */
2045 if ( ( ( p = get_header(csp) ) != NULL) && ( *p == '\0' ) )
2047 len = read_socket(csp->cfd, buf, sizeof(buf) - 1);
2050 log_error(LOG_LEVEL_ERROR, "read from client failed: %E");
2051 destroy_list(headers);
2056 * If there is no memory left for buffering the
2057 * request, there is nothing we can do but hang up
2059 if (add_to_iob(csp, buf, len))
2061 destroy_list(headers);
2067 if (p == NULL) break;
2074 if (http->host == NULL)
2077 * If we still don't know the request destination,
2078 * the request is invalid or the client uses
2079 * Privoxy without its knowledge.
2081 if (JB_ERR_OK != get_request_destination_elsewhere(csp, headers))
2084 * Our attempts to get the request destination
2085 * elsewhere failed or Privoxy is configured
2086 * to only accept proxy requests.
2088 * An error response has already been send
2089 * and we're done here.
2096 * Determine the actions for this URL
2098 #ifdef FEATURE_TOGGLE
2099 if (!(csp->flags & CSP_FLAG_TOGGLED_ON))
2101 /* Most compatible set of actions (i.e. none) */
2102 init_current_action(csp->action);
2105 #endif /* ndef FEATURE_TOGGLE */
2107 get_url_actions(csp, http);
2111 * Save a copy of the original request for logging
2113 http->ocmd = strdup(http->cmd);
2115 if (http->ocmd == NULL)
2117 log_error(LOG_LEVEL_FATAL, "Out of memory copying HTTP request line");
2120 enlist(csp->headers, http->cmd);
2122 /* Append the previously read headers */
2123 list_append_list_unique(csp->headers, headers);
2124 destroy_list(headers);
2126 err = sed(client_patterns, add_client_headers, csp);
2127 if (JB_ERR_OK != err)
2129 assert(err == JB_ERR_PARSE);
2130 log_error(LOG_LEVEL_FATAL, "Failed to parse client headers");
2132 csp->flags |= CSP_FLAG_CLIENT_HEADER_PARSING_DONE;
2134 /* Check request line for rewrites. */
2135 if ((NULL == csp->headers->first->str)
2136 || (strcmp(http->cmd, csp->headers->first->str) &&
2137 (JB_ERR_OK != change_request_destination(csp))))
2140 * A header filter broke the request line - bail out.
2142 write_socket(csp->cfd, MESSED_UP_REQUEST_RESPONSE, strlen(MESSED_UP_REQUEST_RESPONSE));
2143 /* XXX: Use correct size */
2144 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"Invalid request generated\" 500 0", csp->ip_addr_str);
2145 log_error(LOG_LEVEL_ERROR, "Invalid request line after applying header filters.");
2147 free_http_request(http);
2151 /* decide how to route the HTTP request */
2152 fwd = forward_url(csp, http);
2155 log_error(LOG_LEVEL_FATAL, "gateway spec is NULL!?!? This can't happen!");
2156 /* Never get here - LOG_LEVEL_FATAL causes program exit */
2160 * build the http request to send to the server
2161 * we have to do one of the following:
2163 * create = use the original HTTP request to create a new
2164 * HTTP request that has either the path component
2165 * without the http://domainspec (w/path) or the
2166 * full orininal URL (w/url)
2167 * Note that the path and/or the HTTP version may
2168 * have been altered by now.
2170 * connect = Open a socket to the host:port of the server
2171 * and short-circuit server and client socket.
2173 * pass = Pass the request unchanged if forwarding a CONNECT
2174 * request to a parent proxy. Note that we'll be sending
2175 * the CFAIL message ourselves if connecting to the parent
2176 * fails, but we won't send a CSUCCEED message if it works,
2177 * since that would result in a double message (ours and the
2178 * parent's). After sending the request to the parent, we simply
2181 * here's the matrix:
2184 * +--------+--------+
2186 * 0 | create | connect|
2188 * Forwarding +--------+--------+
2190 * 1 | create | pass |
2192 * +--------+--------+
2196 if (http->ssl && connect_port_is_forbidden(csp))
2198 const char *acceptable_connect_ports =
2199 csp->action->string[ACTION_STRING_LIMIT_CONNECT];
2200 assert(NULL != acceptable_connect_ports);
2201 log_error(LOG_LEVEL_INFO, "Request from %s marked for blocking. "
2202 "limit-connect{%s} doesn't allow CONNECT requests to port %d.",
2203 csp->ip_addr_str, acceptable_connect_ports, csp->http->port);
2204 csp->action->flags |= ACTION_BLOCK;
2210 freez(csp->headers->first->str);
2211 build_request_line(csp, fwd, &csp->headers->first->str);
2214 hdr = list_to_text(csp->headers);
2217 /* FIXME Should handle error properly */
2218 log_error(LOG_LEVEL_FATAL, "Out of memory parsing client header");
2222 * We have a request. Check if one of the crunchers wants it.
2224 if (crunch_response_triggered(csp, crunchers_all))
2227 * Yes. The client got the crunch response
2228 * and we are done here after cleaning up.
2231 list_remove_all(csp->headers);
2237 * The headers can't be removed earlier because
2238 * they were still needed for the referrer check
2239 * in case of CGI crunches.
2241 * XXX: Would it be worth to move the referrer check
2242 * into client_referrer() and set a flag if it's trusted?
2244 list_remove_all(csp->headers);
2246 log_error(LOG_LEVEL_GPC, "%s%s", http->hostport, http->path);
2248 if (fwd->forward_host)
2250 log_error(LOG_LEVEL_CONNECT, "via %s:%d to: %s",
2251 fwd->forward_host, fwd->forward_port, http->hostport);
2255 log_error(LOG_LEVEL_CONNECT, "to %s", http->hostport);
2258 /* here we connect to the server, gateway, or the forwarder */
2260 while ( (csp->sfd = forwarded_connect(fwd, http, csp))
2261 && (errno == EINVAL) && (forwarded_connect_retries++ < max_forwarded_connect_retries))
2263 log_error(LOG_LEVEL_ERROR, "failed request #%u to connect to %s. Trying again.",
2264 forwarded_connect_retries, http->hostport);
2267 if (csp->sfd == JB_INVALID_SOCKET)
2269 if (fwd->type != SOCKS_NONE)
2272 rsp = error_response(csp, "forwarding-failed", errno);
2274 else if (errno == EINVAL)
2276 rsp = error_response(csp, "no-such-domain", errno);
2280 rsp = error_response(csp, "connect-failed", errno);
2281 log_error(LOG_LEVEL_CONNECT, "connect to: %s failed: %E",
2286 /* Write the answer to the client */
2289 send_crunch_response(csp, rsp);
2296 if (fwd->forward_host || (http->ssl == 0))
2298 /* write the client's (modified) header to the server
2299 * (along with anything else that may be in the buffer)
2302 if (write_socket(csp->sfd, hdr, strlen(hdr))
2303 || (flush_socket(csp->sfd, csp) < 0))
2305 log_error(LOG_LEVEL_CONNECT, "write header to: %s failed: %E",
2308 rsp = error_response(csp, "connect-failed", errno);
2312 send_crunch_response(csp, rsp);
2322 * We're running an SSL tunnel and we're not forwarding,
2323 * so just send the "connect succeeded" message to the
2324 * client, flush the rest, and get out of the way.
2326 if (write_socket(csp->cfd, CSUCCEED, strlen(CSUCCEED)))
2334 log_error(LOG_LEVEL_CONNECT, "to %s successful", http->hostport);
2336 /* we're finished with the client's header */
2339 maxfd = ( csp->cfd > csp->sfd ) ? csp->cfd : csp->sfd;
2341 /* pass data between the client and server
2342 * until one or the other shuts down the connection.
2351 * FD_ZERO here seems to point to an errant macro which crashes.
2352 * So do this by hand for now...
2354 memset(&rfds,0x00,sizeof(fd_set));
2358 FD_SET(csp->cfd, &rfds);
2359 FD_SET(csp->sfd, &rfds);
2361 n = select((int)maxfd+1, &rfds, NULL, NULL, NULL);
2365 log_error(LOG_LEVEL_ERROR, "select() failed!: %E");
2369 /* this is the body of the browser's request
2370 * just read it and write it.
2373 if (FD_ISSET(csp->cfd, &rfds))
2375 len = read_socket(csp->cfd, buf, sizeof(buf) - 1);
2379 break; /* "game over, man" */
2382 if (write_socket(csp->sfd, buf, (size_t)len))
2384 log_error(LOG_LEVEL_ERROR, "write to: %s failed: %E", http->host);
2391 * The server wants to talk. It could be the header or the body.
2392 * If `hdr' is null, then it's the header otherwise it's the body.
2393 * FIXME: Does `hdr' really mean `host'? No.
2397 if (FD_ISSET(csp->sfd, &rfds))
2400 len = read_socket(csp->sfd, buf, sizeof(buf) - 1);
2404 log_error(LOG_LEVEL_ERROR, "read from: %s failed: %E", http->host);
2406 if (http->ssl && (fwd->forward_host == NULL))
2409 * Just hang up. We already confirmed the client's CONNECT
2410 * request with status code 200 and unencrypted content is
2411 * no longer welcome.
2413 log_error(LOG_LEVEL_ERROR,
2414 "CONNECT already confirmed. Unable to tell the client about the problem.");
2417 else if (byte_count)
2420 * Just hang up. We already transmitted the original headers
2421 * and parts of the original content and therefore missed the
2422 * chance to send an error message (without risking data corruption).
2424 * XXX: we could retry with a fancy range request here.
2426 log_error(LOG_LEVEL_ERROR, "Already forwarded the original headers. "
2427 "Unable to tell the client about the problem.");
2431 rsp = error_response(csp, "connect-failed", errno);
2435 send_crunch_response(csp, rsp);
2441 /* Add a trailing zero. This lets filter_popups
2442 * use string operations.
2446 #ifdef FEATURE_KILL_POPUPS
2447 /* Filter the popups on this read. */
2448 if (block_popups_now)
2450 filter_popups(buf, csp);
2452 #endif /* def FEATURE_KILL_POPUPS */
2454 /* Normally, this would indicate that we've read
2455 * as much as the server has sent us and we can
2456 * close the client connection. However, Microsoft
2457 * in its wisdom has released IIS/5 with a bug that
2458 * prevents it from sending the trailing \r\n in
2459 * a 302 redirect header (and possibly other headers).
2460 * To work around this if we've haven't parsed
2461 * a full header we'll append a trailing \r\n
2462 * and see if this now generates a valid one.
2464 * This hack shouldn't have any impacts. If we've
2465 * already transmitted the header or if this is a
2466 * SSL connection, then we won't bother with this
2467 * hack. So we only work on partially received
2468 * headers. If we append a \r\n and this still
2469 * doesn't generate a valid header, then we won't
2470 * transmit anything to the client.
2475 if (server_body || http->ssl)
2478 * If we have been buffering up the document,
2479 * now is the time to apply content modification
2480 * and send the result to the client.
2484 p = execute_content_filter(csp, content_filter);
2486 * If the content filter fails, use the original
2487 * buffer and length.
2488 * (see p != NULL ? p : csp->iob->cur below)
2492 csp->content_length = (size_t)(csp->iob->eod - csp->iob->cur);
2495 if (JB_ERR_OK != sed(server_patterns_light, NULL, csp))
2497 log_error(LOG_LEVEL_FATAL, "Failed to parse server headers.");
2500 hdr = list_to_text(csp->headers);
2503 /* FIXME Should handle error properly */
2504 log_error(LOG_LEVEL_FATAL, "Out of memory parsing server header");
2508 * Shouldn't happen because this was the second sed run
2509 * and tags are only created for the first one.
2511 assert(!crunch_response_triggered(csp, crunchers_all));
2513 if (write_socket(csp->cfd, hdr, strlen(hdr))
2514 || write_socket(csp->cfd, p != NULL ? p : csp->iob->cur, csp->content_length))
2516 log_error(LOG_LEVEL_ERROR, "write modified content to client failed: %E");
2526 break; /* "game over, man" */
2530 * This is NOT the body, so
2531 * Let's pretend the server just sent us a blank line.
2533 log_error(LOG_LEVEL_INFO,
2534 "Malformerd HTTP headers detected and MS IIS5 hack enabled. "
2535 "Expect an invalid response or even no response at all.");
2536 snprintf(buf, sizeof(buf), "\r\n");
2537 len = (int)strlen(buf);
2540 * Now, let the normal header parsing algorithm below do its
2541 * job. If it fails, we'll exit instead of continuing.
2548 * If this is an SSL connection or we're in the body
2549 * of the server document, just write it to the client,
2550 * unless we need to buffer the body for later content-filtering
2553 if (server_body || http->ssl)
2558 * If there is no memory left for buffering the content, or the buffer limit
2559 * has been reached, switch to non-filtering mode, i.e. make & write the
2560 * header, flush the iob and buf, and get out of the way.
2562 if (add_to_iob(csp, buf, len))
2567 log_error(LOG_LEVEL_INFO, "Flushing header and buffers. Stepping back from filtering.");
2569 hdr = list_to_text(csp->headers);
2573 * Memory is too tight to even generate the header.
2574 * Send our static "Out-of-memory" page.
2576 log_error(LOG_LEVEL_ERROR, "Out of memory while trying to flush.");
2577 rsp = cgi_error_memory();
2578 send_crunch_response(csp, rsp);
2582 hdrlen = strlen(hdr);
2584 if (write_socket(csp->cfd, hdr, hdrlen)
2585 || ((flushed = flush_socket(csp->cfd, csp)) < 0)
2586 || (write_socket(csp->cfd, buf, (size_t)len)))
2588 log_error(LOG_LEVEL_CONNECT, "Flush header and buffers to client failed: %E");
2595 * Reset the byte_count to the amount of bytes
2596 * we just flushed. len will be added a few lines below,
2597 * hdrlen doesn't matter for LOG_LEVEL_CLF.
2599 byte_count = (size_t)flushed;
2601 content_filter = NULL;
2607 if (write_socket(csp->cfd, buf, (size_t)len))
2609 log_error(LOG_LEVEL_ERROR, "write to client failed: %E");
2613 byte_count += (size_t)len;
2618 /* we're still looking for the end of the
2619 * server's header ... (does that make header
2620 * parsing an "out of body experience" ?
2624 * buffer up the data we just read. If that fails,
2625 * there's little we can do but send our static
2626 * out-of-memory page.
2628 if (add_to_iob(csp, buf, len))
2630 log_error(LOG_LEVEL_ERROR, "Out of memory while looking for end of server headers.");
2631 rsp = cgi_error_memory();
2632 send_crunch_response(csp, rsp);
2637 /* Convert iob into something sed() can digest */
2638 if (JB_ERR_PARSE == get_server_headers(csp))
2642 /* Well, we tried our MS IIS/5
2643 * hack and it didn't work.
2644 * The header is incomplete
2645 * and there isn't anything
2646 * we can do about it.
2652 /* Since we have to wait for
2653 * more from the server before
2654 * we can parse the headers
2655 * we just continue here.
2661 /* Did we actually get anything? */
2662 if (NULL == csp->headers->first)
2664 log_error(LOG_LEVEL_ERROR, "Empty server or forwarder response.");
2665 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" 502 0", csp->ip_addr_str, http->cmd);
2666 write_socket(csp->cfd, NO_SERVER_DATA_RESPONSE, strlen(NO_SERVER_DATA_RESPONSE));
2667 free_http_request(http);
2671 /* we have now received the entire header.
2672 * filter it and send the result to the client
2674 if (JB_ERR_OK != sed(server_patterns, add_server_headers, csp))
2676 log_error(LOG_LEVEL_FATAL, "Failed to parse server headers.");
2678 hdr = list_to_text(csp->headers);
2681 /* FIXME Should handle error properly */
2682 log_error(LOG_LEVEL_FATAL, "Out of memory parsing server header");
2685 if (crunch_response_triggered(csp, crunchers_light))
2688 * One of the tags created by a server-header
2689 * tagger triggered a crunch. We already
2690 * delivered the crunch response to the client
2691 * and are done here after cleaning up.
2696 /* Buffer and pcrs filter this if appropriate. */
2698 if (!http->ssl) /* We talk plaintext */
2701 #ifdef FEATURE_KILL_POPUPS
2702 /* Start blocking popups if appropriate. */
2703 if ((csp->content_type & CT_TEXT) && /* It's a text / * MIME-Type */
2704 (csp->action->flags & ACTION_NO_POPUPS) != 0) /* Policy allows */
2706 block_popups_now = 1;
2708 * Filter the part of the body that came in the same read
2709 * as the last headers:
2711 filter_popups(csp->iob->cur, csp);
2713 #endif /* def FEATURE_KILL_POPUPS */
2714 content_filter = get_filter_function(csp);
2717 * Only write if we're not buffering for content modification
2719 if (!content_filter)
2721 /* write the server's (modified) header to
2722 * the client (along with anything else that
2723 * may be in the buffer)
2726 if (write_socket(csp->cfd, hdr, strlen(hdr))
2727 || ((len = flush_socket(csp->cfd, csp)) < 0))
2729 log_error(LOG_LEVEL_CONNECT, "write header to client failed: %E");
2731 /* the write failed, so don't bother
2732 * mentioning it to the client...
2733 * it probably can't hear us anyway.
2739 byte_count += (size_t)len;
2742 /* we're finished with the server's header */
2747 /* If this was a MS IIS/5 hack then it means
2748 * the server has already closed the
2749 * connection. Nothing more to read. Time
2760 return; /* huh? we should never get here */
2763 if (csp->content_length == 0)
2766 * If Privoxy didn't recalculate the
2767 * Content-Lenght, byte_count is still
2770 csp->content_length = byte_count;
2773 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" 200 %d",
2774 csp->ip_addr_str, http->ocmd, csp->content_length);
2778 /*********************************************************************
2782 * Description : This is little more than chat. We only "serve" to
2783 * to close any socket that chat may have opened.
2786 * 1 : csp = Current client state (buffers, headers, etc...)
2790 *********************************************************************/
2792 void serve(struct client_state *csp)
2793 #else /* ifndef AMIGA */
2794 static void serve(struct client_state *csp)
2795 #endif /* def AMIGA */
2798 close_socket(csp->cfd);
2800 if (csp->sfd != JB_INVALID_SOCKET)
2802 close_socket(csp->sfd);
2805 csp->flags &= ~CSP_FLAG_ACTIVE;
2811 /*********************************************************************
2813 * Function : server_thread
2815 * Description : We only exist to call `serve' in a threaded environment.
2818 * 1 : data = Current client state (buffers, headers, etc...)
2820 * Returns : Always 0.
2822 *********************************************************************/
2823 static int32 server_thread(void *data)
2825 serve((struct client_state *) data);
2832 #if !defined(_WIN32) || defined(_WIN_CONSOLE)
2833 /*********************************************************************
2837 * Description : Print usage info & exit.
2839 * Parameters : Pointer to argv[0] for identifying ourselves
2843 *********************************************************************/
2844 static void usage(const char *myname)
2846 printf("Privoxy version " VERSION " (" HOME_PAGE_URL ")\n"
2850 #endif /* defined(unix) */
2853 "[--no-daemon] [--pidfile pidfile] [--pre-chroot-nslookup hostname] [--user user[.group]] "
2854 #endif /* defined(unix) */
2855 "[--version] [configfile]\n"
2856 "Aborting\n", myname);
2861 #endif /* #if !defined(_WIN32) || defined(_WIN_CONSOLE) */
2864 /*********************************************************************
2866 * Function : initialize_mutexes
2868 * Description : Prepares mutexes if mutex support is available.
2872 * Returns : Void, exits in case of errors.
2874 *********************************************************************/
2875 static void initialize_mutexes(void)
2879 #ifdef FEATURE_PTHREAD
2881 * Prepare global mutex semaphores
2883 err = pthread_mutex_init(&log_mutex, 0);
2885 if (!err) err = pthread_mutex_init(&log_init_mutex, 0);
2888 * XXX: The assumptions below are a bit naive
2889 * and can cause locks that aren't necessary.
2891 * For example older FreeBSD versions (< 6.x?)
2892 * have no gethostbyname_r, but gethostbyname is
2895 #if !defined(HAVE_GETHOSTBYADDR_R) || !defined(HAVE_GETHOSTBYNAME_R)
2896 if (!err) err = pthread_mutex_init(&resolver_mutex, 0);
2897 #endif /* !defined(HAVE_GETHOSTBYADDR_R) || !defined(HAVE_GETHOSTBYNAME_R) */
2899 * XXX: should we use a single mutex for
2900 * localtime() and gmtime() as well?
2902 #ifndef HAVE_GMTIME_R
2903 if (!err) err = pthread_mutex_init(&gmtime_mutex, 0);
2904 #endif /* ndef HAVE_GMTIME_R */
2906 #ifndef HAVE_LOCALTIME_R
2907 if (!err) err = pthread_mutex_init(&localtime_mutex, 0);
2908 #endif /* ndef HAVE_GMTIME_R */
2911 if (!err) err = pthread_mutex_init(&rand_mutex, 0);
2912 #endif /* ndef HAVE_RANDOM */
2913 #endif /* FEATURE_PTHREAD */
2916 * TODO: mutex support for mingw32 would be swell.
2921 printf("Fatal error. Mutex initialization failed: %s.\n",
2930 /*********************************************************************
2934 * Description : Load the config file and start the listen loop.
2935 * This function is a lot more *sane* with the `load_config'
2936 * and `listen_loop' functions; although it stills does
2937 * a *little* too much for my taste.
2940 * 1 : argc = Number of parameters (including $0).
2941 * 2 : argv = Array of (char *)'s to the parameters.
2943 * Returns : 1 if : can't open config file, unrecognized directive,
2944 * stats requested in multi-thread mode, can't open the
2945 * log file, can't open the jar file, listen port is invalid,
2946 * any load fails, and can't bind port.
2948 * Else main never returns, the process must be signaled
2949 * to terminate execution. Or, on Windows, use the
2950 * "File", "Exit" menu option.
2952 *********************************************************************/
2954 int real_main(int argc, const char *argv[])
2956 int main(int argc, const char *argv[])
2960 unsigned int random_seed;
2962 struct passwd *pw = NULL;
2963 struct group *grp = NULL;
2966 char *pre_chroot_nslookup_to_load_resolver = NULL;
2973 #if !defined(_WIN32)
2981 * Parse the command line arguments
2983 * XXX: simply printing usage information in case of
2984 * invalid arguments isn't particular user friendly.
2986 while (++argc_pos < argc)
2989 /* Check to see if the service must be installed or uninstalled */
2990 if (strncmp(argv[argc_pos], "--install", 9) == 0)
2992 const char *pName = argv[argc_pos] + 9;
2995 exit( (install_service(pName)) ? 0 : 1 );
2997 else if (strncmp(argv[argc_pos], "--uninstall", + 11) == 0)
2999 const char *pName = argv[argc_pos] + 11;
3002 exit((uninstall_service(pName)) ? 0 : 1);
3004 else if (strcmp(argv[argc_pos], "--service" ) == 0)
3006 bRunAsService = TRUE;
3007 w32_set_service_cwd();
3008 atexit(w32_service_exit_notify);
3011 #endif /* defined(_WIN32) */
3014 #if !defined(_WIN32) || defined(_WIN_CONSOLE)
3016 if (strcmp(argv[argc_pos], "--help") == 0)
3021 else if(strcmp(argv[argc_pos], "--version") == 0)
3023 printf("Privoxy version " VERSION " (" HOME_PAGE_URL ")\n");
3029 else if (strcmp(argv[argc_pos], "--no-daemon" ) == 0)
3034 else if (strcmp(argv[argc_pos], "--pidfile" ) == 0)
3036 if (++argc_pos == argc) usage(argv[0]);
3037 pidfile = strdup(argv[argc_pos]);
3040 else if (strcmp(argv[argc_pos], "--user" ) == 0)
3042 if (++argc_pos == argc) usage(argv[argc_pos]);
3044 if ((NULL != (p = strchr(argv[argc_pos], '.'))) && *(p + 1) != '0')
3047 if (NULL == (grp = getgrnam(p)))
3049 log_error(LOG_LEVEL_FATAL, "Group %s not found.", p);
3053 if (NULL == (pw = getpwnam(argv[argc_pos])))
3055 log_error(LOG_LEVEL_FATAL, "User %s not found.", argv[argc_pos]);
3058 if (p != NULL) *--p = '\0';
3061 else if (strcmp(argv[argc_pos], "--pre-chroot-nslookup" ) == 0)
3063 if (++argc_pos == argc) usage(argv[0]);
3064 pre_chroot_nslookup_to_load_resolver = strdup(argv[argc_pos]);
3067 else if (strcmp(argv[argc_pos], "--chroot" ) == 0)
3071 #endif /* defined(unix) */
3073 else if (argc_pos + 1 != argc)
3076 * This is neither the last command line
3077 * option, nor was it recognized before,
3078 * therefore it must be invalid.
3084 #endif /* defined(_WIN32) && !defined(_WIN_CONSOLE) */
3086 configfile = argv[argc_pos];
3089 } /* -END- while (more arguments) */
3092 if ( *configfile != '/' )
3094 char cwd[BUFFER_SIZE];
3096 size_t abs_file_size;
3098 /* make config-filename absolute here */
3099 if (NULL == getcwd(cwd, sizeof(cwd)))
3101 perror("failed to get current working directory");
3106 abs_file_size = strlen(cwd) + strlen(configfile) + 5;
3107 basedir = strdup(cwd);
3109 if (NULL == basedir ||
3110 NULL == (abs_file = malloc(abs_file_size)))
3112 perror("malloc failed");
3115 strlcpy(abs_file, basedir, abs_file_size);
3116 strlcat(abs_file, "/", abs_file_size );
3117 strlcat(abs_file, configfile, abs_file_size);
3118 configfile = abs_file;
3120 #endif /* defined unix */
3124 clients->next = NULL;
3126 /* XXX: factor out initialising after the next stable release. */
3129 #elif defined(_WIN32)
3133 /* Prepare mutexes if supported and necessary. */
3134 initialize_mutexes();
3136 /* Enable logging until further notice. */
3137 init_log_module(Argv[0]);
3139 random_seed = (unsigned int)time(NULL);
3141 srandom(random_seed);
3142 #elif defined (_WIN32)
3144 * See pick_from_range() in miscutil.c for details.
3146 log_error(LOG_LEVEL_INFO,
3147 "No thread-safe PRNG implemented for your platform. "
3148 "Using weak \'randomization\' factor which will "
3149 "limit the already questionable usefulness of "
3150 "header-time-randomizing actions (disabled by default).");
3153 #endif /* ifdef HAVE_RANDOM */
3156 * Unix signal handling
3158 * Catch the abort, interrupt and terminate signals for a graceful exit
3159 * Catch the hangup signal so the errlog can be reopened.
3160 * Ignore the broken pipe signals (FIXME: Why?)
3162 #if !defined(_WIN32) && !defined(__OS2__) && !defined(AMIGA)
3165 const int catched_signals[] = { SIGTERM, SIGINT, SIGHUP, 0 };
3166 const int ignored_signals[] = { SIGPIPE, 0 };
3168 for (idx = 0; catched_signals[idx] != 0; idx++)
3170 #ifdef sun /* FIXME: Is it safe to check for HAVE_SIGSET instead? */
3171 if (sigset(catched_signals[idx], sig_handler) == SIG_ERR)
3173 if (signal(catched_signals[idx], sig_handler) == SIG_ERR)
3174 #endif /* ifdef sun */
3176 log_error(LOG_LEVEL_FATAL, "Can't set signal-handler for signal %d: %E", catched_signals[idx]);
3180 for (idx = 0; ignored_signals[idx] != 0; idx++)
3182 if (signal(ignored_signals[idx], SIG_IGN) == SIG_ERR)
3184 log_error(LOG_LEVEL_FATAL, "Can't set ignore-handler for signal %d: %E", ignored_signals[idx]);
3189 #else /* ifdef _WIN32 */
3190 # ifdef _WIN_CONSOLE
3192 * We *are* in a windows console app.
3193 * Print a verbose messages about FAQ's and such
3195 printf("%s", win32_blurb);
3196 # endif /* def _WIN_CONSOLE */
3197 #endif /* def _WIN32 */
3200 /* Initialize the CGI subsystem */
3201 cgi_init_error_messages();
3204 * If runnig on unix and without the --nodaemon
3205 * option, become a daemon. I.e. fork, detach
3206 * from tty and get process group leadership
3219 if ( pid < 0 ) /* error */
3224 else if ( pid != 0 ) /* parent */
3229 * must check for errors
3230 * child died due to missing files aso
3233 wpid = waitpid( pid, &status, WNOHANG );
3242 /* Should be more portable, but not as well tested */
3247 #else /* ndef __FreeBSD__ */
3249 #endif /* ndef __FreeBSD__ */
3250 fd = open("/dev/tty", O_RDONLY);
3253 /* no error check here */
3254 ioctl( fd, TIOCNOTTY,0 );
3259 * stderr (fd 2) will be closed later on, when the
3260 * log file has been parsed.
3267 } /* -END- if (!no_daemon) */
3270 * As soon as we have written the PID file, we can switch
3271 * to the user and group ID indicated by the --user option
3277 if (setgid((NULL != grp) ? grp->gr_gid : pw->pw_gid))
3279 log_error(LOG_LEVEL_FATAL, "Cannot setgid(): Insufficient permissions.");
3285 log_error(LOG_LEVEL_FATAL, "Home directory for %s undefined", pw->pw_name);
3287 /* Read the time zone file from /etc before doing chroot. */
3289 if (NULL != pre_chroot_nslookup_to_load_resolver
3290 && '\0' != pre_chroot_nslookup_to_load_resolver[0])
3292 /* Initialize resolver library. */
3293 (void) resolve_hostname_to_ip(pre_chroot_nslookup_to_load_resolver);
3295 if (chroot(pw->pw_dir) < 0)
3297 log_error(LOG_LEVEL_FATAL, "Cannot chroot to %s", pw->pw_dir);
3301 log_error(LOG_LEVEL_FATAL, "Cannot chdir /");
3304 if (setuid(pw->pw_uid))
3306 log_error(LOG_LEVEL_FATAL, "Cannot setuid(): Insufficient permissions.");
3310 char putenv_dummy[64];
3312 strlcpy(putenv_dummy, "HOME=/", sizeof(putenv_dummy));
3313 if (putenv(putenv_dummy) != 0)
3315 log_error(LOG_LEVEL_FATAL, "Cannot putenv(): HOME");
3318 snprintf(putenv_dummy, sizeof(putenv_dummy), "USER=%s", pw->pw_name);
3319 if (putenv(putenv_dummy) != 0)
3321 log_error(LOG_LEVEL_FATAL, "Cannot putenv(): USER");
3327 log_error(LOG_LEVEL_FATAL, "Cannot chroot without --user argument.");
3330 #endif /* defined unix */
3333 /* This will be FALSE unless the command line specified --service
3337 /* Yup, so now we must attempt to establish a connection
3338 * with the service dispatcher. This will only work if this
3339 * process was launched by the service control manager to
3340 * actually run as a service. If this isn't the case, i've
3341 * known it take around 30 seconds or so for the call to return.
3344 /* The StartServiceCtrlDispatcher won't return until the service is stopping */
3345 if (w32_start_service_ctrl_dispatcher(w32ServiceDispatchTable))
3347 /* Service has run, and at this point is now being stopped, so just return */
3352 printf("Warning: Failed to connect to Service Control Dispatcher\nwhen starting as a service!\n");
3354 /* An error occurred. Usually it's because --service was wrongly specified
3355 * and we were unable to connect to the Service Control Dispatcher because
3356 * it wasn't expecting us and is therefore not listening.
3358 * For now, just continue below to call the listen_loop function.
3361 #endif /* def _WIN32 */
3371 /*********************************************************************
3373 * Function : bind_port_helper
3375 * Description : Bind the listen port. Handles logging, and aborts
3379 * 1 : config = Privoxy configuration. Specifies port
3382 * Returns : Port that was opened.
3384 *********************************************************************/
3385 static jb_socket bind_port_helper(struct configuration_spec * config)
3390 if (config->haddr == NULL)
3392 log_error(LOG_LEVEL_INFO, "Listening on port %d on all IP addresses",
3397 log_error(LOG_LEVEL_INFO, "Listening on port %d on IP address %s",
3398 config->hport, config->haddr);
3401 result = bind_port(config->haddr, config->hport, &bfd);
3408 log_error(LOG_LEVEL_FATAL, "can't bind to %s:%d: "
3409 "There may be another Privoxy or some other "
3410 "proxy running on port %d",
3411 (NULL != config->haddr) ? config->haddr : "INADDR_ANY",
3412 config->hport, config->hport);
3415 log_error(LOG_LEVEL_FATAL, "can't bind to %s:%d: "
3416 "The hostname is not resolvable",
3417 (NULL != config->haddr) ? config->haddr : "INADDR_ANY", config->hport);
3420 log_error(LOG_LEVEL_FATAL, "can't bind to %s:%d: because %E",
3421 (NULL != config->haddr) ? config->haddr : "INADDR_ANY", config->hport);
3424 /* shouldn't get here */
3425 return JB_INVALID_SOCKET;
3428 config->need_bind = 0;
3435 /* Without this simple workaround we get this compiler warning from _beginthread
3436 * warning C4028: formal parameter 1 different from declaration
3438 void w32_service_listen_loop(void *p)
3442 #endif /* def _WIN32 */
3445 /*********************************************************************
3447 * Function : listen_loop
3449 * Description : bind the listen port and enter a "FOREVER" listening loop.
3455 *********************************************************************/
3456 static void listen_loop(void)
3458 struct client_state *csp = NULL;
3460 struct configuration_spec * config;
3462 config = load_config();
3464 bfd = bind_port_helper(config);
3466 #ifdef FEATURE_GRACEFUL_TERMINATION
3467 while (!g_terminate)
3472 #if !defined(FEATURE_PTHREAD) && !defined(_WIN32) && !defined(__BEOS__) && !defined(AMIGA) && !defined(__OS2__)
3473 while (waitpid(-1, NULL, WNOHANG) > 0)
3475 /* zombie children */
3477 #endif /* !defined(FEATURE_PTHREAD) && !defined(_WIN32) && !defined(__BEOS__) && !defined(AMIGA) */
3480 * Free data that was used by died threads
3486 * Re-open the errlog after HUP signal
3488 if (received_hup_signal)
3490 init_error_log(Argv[0], config->logfile);
3491 received_hup_signal = 0;
3496 #ifdef FEATURE_COOKIE_JAR
3498 * Need a workaround here: we have to fclose() the jarfile, or we die because it's
3499 * already open. I think unload_configfile() is not being run, which should do
3500 * this work. Until that can get resolved, we'll use this workaround.
3504 if (csp->config->jar)
3506 fclose(csp->config->jar);
3507 csp->config->jar = NULL;
3509 #endif /* FEATURE_COOKIE_JAR */
3510 #endif /* __OS2__ */
3512 if ( NULL == (csp = (struct client_state *) zalloc(sizeof(*csp))) )
3514 log_error(LOG_LEVEL_FATAL, "malloc(%d) for csp failed: %E", sizeof(*csp));
3518 csp->flags |= CSP_FLAG_ACTIVE;
3519 csp->sfd = JB_INVALID_SOCKET;
3521 csp->config = config = load_config();
3523 if ( config->need_bind )
3526 * Since we were listening to the "old port", we will not see
3527 * a "listen" param change until the next IJB request. So, at
3528 * least 1 more request must be made for us to find the new
3529 * setting. I am simply closing the old socket and binding the
3532 * Which-ever is correct, we will serve 1 more page via the
3533 * old settings. This should probably be a "show-proxy-args"
3534 * request. This should not be a so common of an operation
3535 * that this will hurt people's feelings.
3540 bfd = bind_port_helper(config);
3543 log_error(LOG_LEVEL_CONNECT, "Listening for new connections ... ");
3545 if (!accept_connection(csp, bfd))
3547 log_error(LOG_LEVEL_CONNECT, "accept failed: %E");
3560 log_error(LOG_LEVEL_CONNECT, "accepted connection from %s", csp->ip_addr_str);
3563 #ifdef FEATURE_TOGGLE
3564 if (global_toggle_state)
3565 #endif /* def FEATURE_TOGGLE */
3567 csp->flags |= CSP_FLAG_TOGGLED_ON;
3570 if (run_loader(csp))
3572 log_error(LOG_LEVEL_FATAL, "a loader failed - must exit");
3573 /* Never get here - LOG_LEVEL_FATAL causes program exit */
3577 if (block_acl(NULL,csp))
3579 log_error(LOG_LEVEL_CONNECT, "Connection from %s dropped due to ACL", csp->ip_addr_str);
3580 close_socket(csp->cfd);
3584 #endif /* def FEATURE_ACL */
3586 /* add it to the list of clients */
3587 csp->next = clients->next;
3588 clients->next = csp;
3590 if (config->multi_threaded)
3594 /* this is a switch () statment in the C preprocessor - ugh */
3595 #undef SELECTED_ONE_OPTION
3597 /* Use Pthreads in preference to native code */
3598 #if defined(FEATURE_PTHREAD) && !defined(SELECTED_ONE_OPTION)
3599 #define SELECTED_ONE_OPTION
3601 pthread_t the_thread;
3602 pthread_attr_t attrs;
3604 pthread_attr_init(&attrs);
3605 pthread_attr_setdetachstate(&attrs, PTHREAD_CREATE_DETACHED);
3606 errno = pthread_create(&the_thread, &attrs,
3607 (void * (*)(void *))serve, csp);
3608 child_id = errno ? -1 : 0;
3609 pthread_attr_destroy(&attrs);
3613 #if defined(_WIN32) && !defined(_CYGWIN) && !defined(SELECTED_ONE_OPTION)
3614 #define SELECTED_ONE_OPTION
3615 child_id = _beginthread(
3616 (void (*)(void *))serve,
3621 #if defined(__OS2__) && !defined(SELECTED_ONE_OPTION)
3622 #define SELECTED_ONE_OPTION
3623 child_id = _beginthread(
3624 (void(* _Optlink)(void*))serve,
3630 #if defined(__BEOS__) && !defined(SELECTED_ONE_OPTION)
3631 #define SELECTED_ONE_OPTION
3633 thread_id tid = spawn_thread
3634 (server_thread, "server", B_NORMAL_PRIORITY, csp);
3636 if ((tid >= 0) && (resume_thread(tid) == B_OK))
3638 child_id = (int) tid;
3647 #if defined(AMIGA) && !defined(SELECTED_ONE_OPTION)
3648 #define SELECTED_ONE_OPTION
3649 csp->cfd = ReleaseSocket(csp->cfd, -1);
3652 child_id = (int)CreateNewProcTags(NP_Entry, (ULONG)server_thread,
3653 NP_Output, Output(),
3654 NP_CloseOutput, FALSE,
3655 NP_Name, (ULONG)"privoxy child",
3659 child_id = (int)CreateNewProcTags(NP_Entry, (ULONG)server_thread,
3660 NP_Output, Output(),
3661 NP_CloseOutput, FALSE,
3662 NP_Name, (ULONG)"privoxy child",
3663 NP_StackSize, 200*1024,
3669 ((struct Task *)child_id)->tc_UserData = csp;
3670 Signal((struct Task *)child_id, SIGF_SINGLE);
3675 #if !defined(SELECTED_ONE_OPTION)
3678 /* This block is only needed when using fork().
3679 * When using threads, the server thread was
3680 * created and run by the call to _beginthread().
3682 if (child_id == 0) /* child */
3685 #ifdef FEATURE_TOGGLE
3686 int inherited_toggle_state = global_toggle_state;
3687 #endif /* def FEATURE_TOGGLE */
3692 * If we've been toggled or we've blocked the request, tell Mom
3695 #ifdef FEATURE_TOGGLE
3696 if (inherited_toggle_state != global_toggle_state)
3698 rc |= RC_FLAG_TOGGLED;
3700 #endif /* def FEATURE_TOGGLE */
3702 #ifdef FEATURE_STATISTICS
3703 if (csp->flags & CSP_FLAG_REJECTED)
3705 rc |= RC_FLAG_BLOCKED;
3707 #endif /* ndef FEATURE_STATISTICS */
3711 else if (child_id > 0) /* parent */
3713 /* in a fork()'d environment, the parent's
3714 * copy of the client socket and the CSP
3718 #if !defined(_WIN32) && !defined(__CYGWIN__)
3720 wait( &child_status );
3723 * Evaluate child's return code: If the child has
3724 * - been toggled, toggle ourselves
3725 * - blocked its request, bump up the stats counter
3728 #ifdef FEATURE_TOGGLE
3729 if (WIFEXITED(child_status) && (WEXITSTATUS(child_status) & RC_FLAG_TOGGLED))
3731 global_toggle_state = !global_toggle_state;
3733 #endif /* def FEATURE_TOGGLE */
3735 #ifdef FEATURE_STATISTICS
3737 if (WIFEXITED(child_status) && (WEXITSTATUS(child_status) & RC_FLAG_BLOCKED))
3741 #endif /* def FEATURE_STATISTICS */
3743 #endif /* !defined(_WIN32) && defined(__CYGWIN__) */
3744 close_socket(csp->cfd);
3745 csp->flags &= ~CSP_FLAG_ACTIVE;
3749 #undef SELECTED_ONE_OPTION
3750 /* end of cpp switch () */
3752 if (child_id < 0) /* failed */
3754 char buf[BUFFER_SIZE];
3756 log_error(LOG_LEVEL_ERROR, "can't fork: %E");
3758 snprintf(buf , sizeof(buf), "Privoxy: can't fork: errno = %d", errno);
3760 write_socket(csp->cfd, buf, strlen(buf));
3761 close_socket(csp->cfd);
3762 csp->flags &= ~CSP_FLAG_ACTIVE;
3773 /* NOTREACHED unless FEATURE_GRACEFUL_TERMINATION is defined */
3775 /* Clean up. Aim: free all memory (no leaks) */
3776 #ifdef FEATURE_GRACEFUL_TERMINATION
3778 log_error(LOG_LEVEL_ERROR, "Graceful termination requested");
3780 unload_current_config_file();
3781 unload_current_actions_file();
3782 unload_current_re_filterfile();
3783 #ifdef FEATURE_TRUST
3784 unload_current_trust_file();
3787 if (config->multi_threaded)
3794 } while ((clients->next != NULL) && (--i > 0));
3798 log_error(LOG_LEVEL_ERROR, "Graceful termination failed - still some live clients after 1 minute wait.");
3809 #ifdef FEATURE_COOKIE_JAR
3810 if (NULL != config->jar)
3812 fclose(config->jar);
3816 #if defined(_WIN32) && !defined(_WIN_CONSOLE)
3817 /* Cleanup - remove taskbar icon etc. */
3822 #endif /* FEATURE_GRACEFUL_TERMINATION */