1 const char jcc_rcs[] = "$Id: jcc.c,v 1.181 2008/05/21 15:47:15 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) 2001-2008 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 1.181 2008/05/21 15:47:15 fabiankeil
37 * Streamline sed()'s prototype and declare
38 * the header parse and add structures static.
40 * Revision 1.180 2008/05/21 15:26:32 fabiankeil
41 * - Mark csp as immutable for send_crunch_response().
42 * - Fix comment spelling.
44 * Revision 1.179 2008/05/20 20:13:32 fabiankeil
45 * Factor update_server_headers() out of sed(), ditch the
46 * first_run hack and make server_patterns_light static.
48 * Revision 1.178 2008/05/10 13:23:38 fabiankeil
49 * Don't provide get_header() with the whole client state
50 * structure when it only needs access to csp->iob.
52 * Revision 1.177 2008/05/10 11:51:12 fabiankeil
53 * Make the "read the rest of the headers" loop a bit more readable.
55 * Revision 1.176 2008/05/10 11:37:57 fabiankeil
56 * - Instead of logging when the IIS5 hack is enabled, log when it fails.
57 * - Remove useless comment.
59 * Revision 1.175 2008/05/09 18:53:59 fabiankeil
60 * Fix comment grammar.
62 * Revision 1.174 2008/05/07 18:05:53 fabiankeil
63 * Remove the pointless buffer in client_protocol_is_unsupported().
65 * Revision 1.173 2008/05/06 15:09:00 fabiankeil
66 * Least-effort fix for bug #1821930 (reported by Lee):
67 * If the response doesn't look like HTTP,
68 * tell the client and log the problem.
70 * Revision 1.172 2008/04/16 16:38:21 fabiankeil
71 * Don't pass the whole csp structure to flush_socket()
72 * when it only needs a file descriptor and a buffer.
74 * Revision 1.171 2008/03/27 18:27:25 fabiankeil
75 * Remove kill-popups action.
77 * Revision 1.170 2008/03/06 16:33:46 fabiankeil
78 * If limit-connect isn't used, don't limit CONNECT requests to port 443.
80 * Revision 1.169 2008/03/04 18:30:39 fabiankeil
81 * Remove the treat-forbidden-connects-like-blocks action. We now
82 * use the "blocked" page for forbidden CONNECT requests by default.
84 * Revision 1.168 2008/03/02 12:25:25 fabiankeil
85 * Also use shiny new connect_port_is_forbidden() in jcc.c.
87 * Revision 1.167 2008/02/23 16:57:12 fabiankeil
88 * Rename url_actions() to get_url_actions() and let it
89 * use the standard parameter ordering.
91 * Revision 1.166 2008/02/23 16:33:43 fabiankeil
92 * Let forward_url() use the standard parameter ordering
93 * and mark its second parameter immutable.
95 * Revision 1.165 2008/02/02 19:36:56 fabiankeil
96 * Remove the "Listening ... for local connections only" log message.
97 * Whether or not remote connections are able to reach Privoxy is up
98 * to the operating system.
100 * Revision 1.164 2007/12/16 18:32:46 fabiankeil
101 * Prevent the log messages for CONNECT requests to unacceptable
102 * ports from printing the limit-connect argument as [null] if
103 * limit-connect hasn't been explicitly enabled.
105 * Revision 1.163 2007/12/13 01:47:11 david__schmidt
106 * Make sure all console-mode apps get a usage() instance
108 * Revision 1.162 2007/12/06 17:54:57 fabiankeil
109 * Reword NO_SERVER_DATA_RESPONSE to make it harder
110 * to misunderstand what the message is all about.
112 * Revision 1.161 2007/12/04 19:44:22 fabiankeil
113 * Unbreak trustfile which previously didn't work without
114 * FEATURE_TOGGLE. Fixes BR#1843585, reported by Lee.
116 * Revision 1.160 2007/11/29 18:00:29 fabiankeil
117 * Plug memory leak. Spotted by Valgrind, triggered by
118 * Privoxy-Regression-Test feeding proxyfuzz.py.
120 * Revision 1.159 2007/11/24 14:34:09 fabiankeil
121 * In the HTTP snipplets, refer to the client as client.
123 * Revision 1.158 2007/11/11 16:44:17 fabiankeil
124 * Emit a log message when activating the MS IIS5 hack.
126 * Revision 1.157 2007/11/03 17:34:49 fabiankeil
127 * Log the "weak randomization factor" warning only
128 * once for mingw32 and provide some more details.
130 * Revision 1.156 2007/11/01 18:20:58 fabiankeil
131 * Initialize log module after initializing mutexes, future
132 * deadlocks in that code should now work cross-platform.
134 * Revision 1.155 2007/10/23 20:12:45 fabiankeil
135 * Fix first CSUCCEED line to end in \r\n as required by RFC1945.
136 * Reported by Bert van Leeuwen in BR#1818808.
138 * Revision 1.154 2007/10/19 17:00:08 fabiankeil
139 * Downgrade "Flushing header and buffers" message to LOG_LEVEL_INFO.
141 * Revision 1.153 2007/10/14 14:12:41 fabiankeil
142 * When in daemon mode, close stderr after the configuration file has been
143 * parsed the first time. If logfile isn't set, stop logging. Fixes BR#897436.
145 * Revision 1.152 2007/10/04 18:03:34 fabiankeil
146 * - Fix a crash when parsing invalid requests whose first header
147 * is rejected by get_header(). Regression (re?)introduced
148 * in r1.143 by yours truly.
149 * - Move ACTION_VANILLA_WAFER handling into parsers.c's
150 * client_cookie_adder() to make sure send-vanilla-wafer can be
151 * controlled through tags (and thus regression-tested).
153 * Revision 1.151 2007/09/29 10:21:16 fabiankeil
154 * - Move get_filter_function() from jcc.c to filters.c
155 * so the filter functions can be static.
156 * - Don't bother filtering body-less responses.
158 * Revision 1.150 2007/09/28 16:39:29 fabiankeil
159 * Execute content filters through execute_content_filter().
161 * Revision 1.149 2007/09/04 15:08:48 fabiankeil
162 * Initialize req to NULL to make sure it's defined if the
163 * first read_socket() call fails. Reported by icmp30.
165 * Revision 1.148 2007/08/26 16:47:13 fabiankeil
166 * Add Stephen Gildea's --pre-chroot-nslookup patch [#1276666],
167 * extensive comments moved to user manual.
169 * Revision 1.147 2007/08/25 14:42:40 fabiankeil
170 * Don't crash if a broken header filter wiped out the request line.
172 * Revision 1.146 2007/08/20 17:09:32 fabiankeil
173 * Fix byte_count calculation in case of flushes
174 * and don't parse the server headers a second time.
176 * Revision 1.145 2007/08/19 13:13:31 fabiankeil
177 * - If there's a connection problem after we already forwarded
178 * parts of the original content, just hang up. Fixes BR#1776724.
179 * - Fix warnings about unused code on mingw32.
180 * - In case of flushes, calculate the byte count
181 * less incorrectly (I think).
183 * Revision 1.144 2007/08/11 14:43:22 fabiankeil
184 * Add some more prototypes for static functions.
186 * Revision 1.143 2007/08/05 13:58:19 fabiankeil
187 * Comment out request_contains_null_bytes() until it's used again.
189 * Revision 1.142 2007/08/05 13:50:26 fabiankeil
190 * #1763173 from Stefan Huehner: s@const static@static const@
191 * and declare some more functions static.
193 * Revision 1.141 2007/08/04 09:56:23 fabiankeil
194 * - Log rejected CONNECT requests with LOG_LEVEL_INFO
195 * and explain why they were rejected in the first place.
196 * - Fix the LOG_LEVEL_CLF message for crunches of unallowed
197 * CONNECT requests. The request line was missing.
198 * - Add two more XXX reminders as we don't have enough already.
200 * Revision 1.140 2007/07/21 11:51:36 fabiankeil
201 * As Hal noticed, checking dispatch_cgi() as the last cruncher
202 * looks like a bug if CGI requests are blocked unintentionally,
203 * so don't do it unless the user enabled the new config option
204 * "allow-cgi-request-crunching".
206 * Revision 1.139 2007/07/14 07:46:41 fabiankeil
207 * - Allow to rewrite the request destination behind the client's back.
208 * - Turn the weird-looking unconditional for loop that
209 * reads the client request into a conditional while loop.
210 * Move the stuff that only runs once out of the loop.
211 * - Move parts of chat(), server_content_type() and the
212 * necessary stuff to fix BR#1750917 into get_filter_function().
214 * Revision 1.138 2007/06/03 18:45:18 fabiankeil
215 * Temporary workaround for BR#1730105.
217 * Revision 1.137 2007/06/01 18:16:36 fabiankeil
218 * Use the same mutex for gethostbyname() and gethostbyaddr() to prevent
219 * deadlocks and crashes on OpenBSD and possibly other OS with neither
220 * gethostbyname_r() nor gethostaddr_r(). Closes BR#1729174.
221 * Thanks to Ralf Horstmann for report and solution.
223 * Revision 1.136 2007/06/01 16:41:11 fabiankeil
224 * Add forward-override{} to change the forwarding settings through
225 * action sections. This is mainly interesting to forward different
226 * clients differently (for example based on User-Agent or request
229 * Revision 1.135 2007/05/24 17:03:50 fabiankeil
230 * - Let usage() mention the --chroot parameter.
231 * - Use read_socket() consistently and always leave
232 * the last buffer byte alone, even in cases where
233 * null termination (currently) doesn't matter.
235 * Revision 1.134 2007/05/16 14:59:46 fabiankeil
236 * - Fix config file loading on Unix if no config file is specified.
237 * Since r1.97 Privoxy would always interpret the last argument as
238 * config file, even if it's a valid command line option.
239 * - Abort in case of unrecognized command line options. Closes #1719696.
240 * - Remove a bunch of unnecessary strcpy() calls (yay for c&p without thinking).
241 * - Replace the remaining strcpy() and strcat() calls with strlcpy() and strcat().
243 * Revision 1.133 2007/05/04 11:23:19 fabiankeil
244 * - Don't rerun crunchers that only depend on the request URL.
245 * - Don't count redirects and CGI requests as "blocked requests".
247 * Revision 1.132 2007/04/25 15:15:17 fabiankeil
248 * Support crunching based on tags created by server-header taggers.
250 * Revision 1.131 2007/04/22 13:24:50 fabiankeil
251 * Make HTTP snippets static (again). Add a Content-Type for those
252 * with content so the browser doesn't guess it based on the URL.
254 * Revision 1.130 2007/04/19 13:47:34 fabiankeil
255 * Move crunching and request line rebuilding out of chat().
257 * Revision 1.129 2007/04/15 16:39:20 fabiankeil
258 * Introduce tags as alternative way to specify which
259 * actions apply to a request. At the moment tags can be
260 * created based on client and server headers.
262 * Revision 1.128 2007/03/25 16:55:54 fabiankeil
263 * Don't CLF-log CONNECT requests twice.
265 * Revision 1.127 2007/03/20 13:53:17 fabiankeil
266 * Log the source address for ACL-related connection drops.
268 * Revision 1.126 2007/03/17 15:20:05 fabiankeil
269 * New config option: enforce-blocks.
271 * Revision 1.125 2007/03/09 14:12:00 fabiankeil
272 * - Move null byte check into separate function.
273 * - Don't confuse the client with error pages
274 * if a CONNECT request was already confirmed.
276 * Revision 1.124 2007/02/23 14:59:54 fabiankeil
277 * Speed up NULL byte escaping and only log the complete
278 * NULL byte requests with header debugging enabled.
280 * Revision 1.123 2007/02/21 18:42:10 fabiankeil
281 * Answer requests that contain NULL bytes with
282 * a custom response instead of waiting for more
283 * data until the client eventually hangs up.
285 * Revision 1.122 2007/02/07 11:12:02 fabiankeil
286 * - Move delivery and logging of crunched responses
287 * from chat() into send_crunch_response().
288 * - Display the reason for generating http_responses.
289 * - Log the content length for LOG_LEVEL_CLF correctly
290 * (still incorrect for some fixed responses).
291 * - Reword an incorrect comment about
292 * treat-forbidden-connects-like-blocks violating
294 * - Add some log messages.
296 * Revision 1.121 2007/01/27 10:52:56 fabiankeil
297 * Move mutex initialization into separate
298 * function and exit in case of errors.
300 * Revision 1.120 2007/01/26 14:18:42 fabiankeil
301 * - Start to reduce chat()'s line count and move
302 * parts of it into separate functions.
303 * - Add "HTTP/1.1 100 Continue" hack for BR 756734.
305 * Revision 1.119 2007/01/25 14:02:30 fabiankeil
306 * - Add Proxy-Agent header to HTTP snippets that are
307 * supposed to reach HTTP clients only.
308 * - Made a few CONNECT log messages more descriptive.
309 * - Catch completely empty server responses (as seen
310 * with Tor's fake ".noconnect" top level domain).
311 * - Use shiny new "forwarding-failed" template for socks errors.
313 * Revision 1.118 2007/01/07 07:43:43 joergs
314 * AmigaOS4 support added.
316 * Revision 1.117 2006/12/31 17:56:37 fabiankeil
317 * Added config option accept-intercepted-requests
318 * and disabled it by default.
320 * Revision 1.116 2006/12/29 19:08:22 fabiankeil
321 * Reverted parts of my last commit
322 * to keep error handling working.
324 * Revision 1.115 2006/12/29 17:38:57 fabiankeil
325 * Fixed gcc43 conversion warnings.
327 * Revision 1.114 2006/12/27 18:52:02 fabiankeil
328 * Fix -pedantic ISO C warning about converting
329 * from function pointer to object pointer.
331 * Revision 1.113 2006/12/26 17:38:50 fabiankeil
332 * Silence compiler warning I introduced with my last commit.
334 * Revision 1.112 2006/12/26 17:31:41 fabiankeil
335 * Mutex protect rand() if POSIX threading
336 * is used, warn the user if that's not possible
337 * and stop using it on _WIN32 where it could
340 * Revision 1.111 2006/12/23 16:15:06 fabiankeil
341 * Don't prevent core dumps by catching SIGABRT.
342 * It's rude and makes debugging unreasonable painful.
344 * Revision 1.110 2006/12/13 14:52:53 etresoft
345 * Fix build failure on MacOS X. Global symbols can be either static or extern, but not both.
347 * Revision 1.109 2006/12/06 19:41:40 fabiankeil
348 * Privoxy is now able to run as intercepting
349 * proxy in combination with any packet filter
350 * that does the port redirection. The destination
351 * is extracted from the "Host:" header which
352 * should be available for nearly all requests.
354 * Moved HTTP snipplets into jcc.c.
355 * Added error message for gopher proxy requests.
357 * Revision 1.108 2006/11/28 15:38:51 fabiankeil
358 * Only unlink the pidfile if it's actually used.
360 * Change order of interception checks to make
361 * it possible to block or redirect requests for
364 * Revision 1.107 2006/11/13 19:05:51 fabiankeil
365 * Make pthread mutex locking more generic. Instead of
366 * checking for OSX and OpenBSD, check for FEATURE_PTHREAD
367 * and use mutex locking unless there is an _r function
368 * available. Better safe than sorry.
370 * Fixes "./configure --disable-pthread" and should result
371 * in less threading-related problems on pthread-using platforms,
372 * but it still doesn't fix BR#1122404.
374 * Revision 1.106 2006/11/06 19:58:23 fabiankeil
375 * Move pthread.h inclusion from jcc.c to jcc.h.
376 * Fixes build on x86-freebsd1 (FreeBSD 5.4-RELEASE).
378 * Revision 1.105 2006/11/06 14:26:02 fabiankeil
379 * Don't exit after receiving the second SIGHUP on Solaris.
381 * Fixes BR 1052235, but the same problem may exist on other
382 * systems. Once 3.0.6 is out we should use sigset()
383 * where available and see if it breaks anything.
385 * Revision 1.104 2006/09/23 13:26:38 roro
386 * Replace TABs by spaces in source code.
388 * Revision 1.103 2006/09/21 12:54:43 fabiankeil
389 * Fix +redirect{}. Didn't work with -fast-redirects.
391 * Revision 1.102 2006/09/06 13:03:04 fabiankeil
392 * Respond with 400 and a short text message
393 * if the client tries to use Privoxy as FTP proxy.
395 * Revision 1.101 2006/09/06 09:23:37 fabiankeil
396 * Make number of retries in case of forwarded-connect problems
397 * a config file option (forwarded-connect-retries) and use 0 as
400 * Revision 1.100 2006/09/03 19:42:59 fabiankeil
401 * Set random(3) seed.
403 * Revision 1.99 2006/09/02 15:36:42 fabiankeil
404 * Follow the OpenBSD port's lead and protect the resolve
405 * functions on OpenBSD as well.
407 * Revision 1.98 2006/08/24 11:01:34 fabiankeil
408 * --user fix. Only use the user as group if no group is specified.
409 * Solves BR 1492612. Thanks to Spinor S. and David Laight.
411 * Revision 1.97 2006/08/18 15:23:17 david__schmidt
412 * Windows service (re-)integration
416 * --install[:service_name]
417 * --uninstall[:service_name]
420 * They work as follows:
421 * --install will create a service for you and then terminate.
422 * By default the service name will be "privoxy" (without the quotes).
423 * However you can run multiple services if you wish, just by adding
424 * a colon and then a name (no spaces).
426 * --uninstall follows the exact same rules a --install.
428 * --service is used when the program is executed by the service
429 * control manager, and in normal circumstances would never be
430 * used as a command line argument.
432 * Revision 1.96 2006/08/15 20:12:36 david__schmidt
433 * Windows service integration
435 * Revision 1.95 2006/08/03 02:46:41 david__schmidt
436 * Incorporate Fabian Keil's patch work:
437 http://www.fabiankeil.de/sourcecode/privoxy/
439 * Revision 1.94 2006/07/18 14:48:46 david__schmidt
440 * Reorganizing the repository: swapping out what was HEAD (the old 3.1 branch)
441 * with what was really the latest development (the v_3_0_branch branch)
443 * Revision 1.92.2.16 2005/04/03 20:10:50 david__schmidt
444 * Thanks to Jindrich Makovicka for a race condition fix for the log
445 * file. The race condition remains for non-pthread implementations.
446 * Reference patch #1175720.
448 * Revision 1.92.2.15 2004/10/03 12:53:32 david__schmidt
449 * Add the ability to check jpeg images for invalid
450 * lengths of comment blocks. Defensive strategy
451 * against the exploit:
452 * Microsoft Security Bulletin MS04-028
453 * Buffer Overrun in JPEG Processing (GDI+) Could
454 * Allow Code Execution (833987)
455 * Enabled with +inspect-jpegs in actions files.
457 * Revision 1.92.2.14 2003/12/12 12:52:53 oes
458 * - Fixed usage info for non-unix platforms
459 * - Fixed small cmdline parsing bug
461 * Revision 1.92.2.13 2003/11/27 19:20:27 oes
462 * Diagnostics: Now preserve the returncode of pthread_create
463 * in errno. Closes BR #775721. Thanks to Geoffrey Hausheer.
465 * Revision 1.92.2.12 2003/07/11 11:34:19 oes
466 * No longer ignore SIGCHLD. Fixes bug #769381
468 * Revision 1.92.2.11 2003/05/14 12:32:02 oes
469 * Close jarfile on graceful exit, remove stray line
471 * Revision 1.92.2.10 2003/05/08 15:13:46 oes
472 * Cosmetics: Killed a warning, a typo and an allocation left at exit
474 * Revision 1.92.2.9 2003/04/03 15:08:42 oes
475 * No longer rely on non-POSIX.1 extensions of getcwd().
478 * Revision 1.92.2.8 2003/03/31 13:12:32 oes
479 * Replaced setenv() by posix-compliant putenv()
480 * Thanks to Neil McCalden (nmcc AT users.sf.net).
482 * Revision 1.92.2.7 2003/03/17 16:48:59 oes
483 * Added chroot ability, thanks to patch by Sviatoslav Sviridov
485 * Revision 1.92.2.6 2003/03/11 11:55:00 oes
486 * Clean-up and extension of improvements for forked mode:
487 * - Child's return code now consists of flags RC_FLAG_*
488 * - Reporting toggle to parent now properly #ifdef'ed
489 * - Children now report blocking to parent. This enables
490 * statistics in forked mode
492 * Revision 1.92.2.5 2003/03/10 23:45:32 oes
493 * Fixed bug #700381: Non-Threaded version now capable of being toggled.
494 * Children now report having been toggled through _exit(17), parents
495 * watch for that code and toggle themselves if found.
497 * Revision 1.92.2.4 2003/03/07 03:41:04 david__schmidt
498 * Wrapping all *_r functions (the non-_r versions of them) with
499 * mutex semaphores for OSX. Hopefully this will take care of all
500 * of those pesky crash reports.
502 * Revision 1.92.2.3 2003/02/28 12:53:06 oes
503 * Fixed two mostly harmless mem leaks
505 * Revision 1.92.2.2 2002/11/20 14:37:47 oes
506 * Fix: Head of global clients list now initialized to NULL
508 * Revision 1.92.2.1 2002/09/25 14:52:24 oes
509 * Added basic support for OPTIONS and TRACE HTTP methods:
510 * - New interceptor direct_response() added in chat().
511 * - sed() moved to earlier in the process, so that the
512 * Host: header is evaluated before actions and forwarding
515 * Revision 1.92 2002/05/08 16:00:46 oes
516 * Chat's buffer handling:
517 * - Fixed bug with unchecked out-of-mem conditions
518 * while reading client request & server headers
519 * - No longer predict if the buffer limit will be exceeded
520 * in the next read -- check add_to_iob's new
521 * return code. If buffer couldn't be extended
522 * (policy or out-of-mem) while
523 * - reading from client: abort
524 * - reading server headers: send error page
525 * - buffering server body for filter: flush,
526 * and if that fails: send error page
528 * Revision 1.91 2002/04/08 20:35:58 swa
531 * Revision 1.90 2002/04/02 14:57:28 oes
532 * Made sending wafers independent of FEATURE_COOKIE_JAR
534 * Revision 1.89 2002/03/31 17:18:59 jongfoster
535 * Win32 only: Enabling STRICT to fix a VC++ compile warning.
537 * Revision 1.88 2002/03/27 14:32:43 david__schmidt
538 * More compiler warning message maintenance
540 * Revision 1.87 2002/03/26 22:29:54 swa
541 * we have a new homepage!
543 * Revision 1.86 2002/03/25 17:04:55 david__schmidt
544 * Workaround for closing the jarfile before load_config() comes around again
546 * Revision 1.85 2002/03/24 15:23:33 jongfoster
549 * Revision 1.84 2002/03/24 13:25:43 swa
550 * name change related issues
552 * Revision 1.83 2002/03/16 23:54:06 jongfoster
553 * Adding graceful termination feature, to help look for memory leaks.
554 * If you enable this (which, by design, has to be done by hand
555 * editing config.h) and then go to http://i.j.b/die, then the program
556 * will exit cleanly after the *next* request. It should free all the
557 * memory that was used.
559 * Revision 1.82 2002/03/13 00:27:05 jongfoster
562 * Revision 1.81 2002/03/12 01:42:50 oes
563 * Introduced modular filters
565 * Revision 1.80 2002/03/11 22:07:05 david__schmidt
566 * OS/2 port maintenance:
567 * - Fixed EMX build - it had decayed a little
568 * - Fixed inexplicable crash during FD_ZERO - must be due to a bad macro.
569 * substituted a memset for now.
571 * Revision 1.79 2002/03/09 20:03:52 jongfoster
572 * - Making various functions return int rather than size_t.
573 * (Undoing a recent change). Since size_t is unsigned on
574 * Windows, functions like read_socket that return -1 on
575 * error cannot return a size_t.
577 * THIS WAS A MAJOR BUG - it caused frequent, unpredictable
578 * crashes, and also frequently caused JB to jump to 100%
579 * CPU and stay there. (Because it thought it had just
580 * read ((unsigned)-1) == 4Gb of data...)
582 * - The signature of write_socket has changed, it now simply
583 * returns success=0/failure=nonzero.
585 * - Trying to get rid of a few warnings --with-debug on
586 * Windows, I've introduced a new type "jb_socket". This is
587 * used for the socket file descriptors. On Windows, this
588 * is SOCKET (a typedef for unsigned). Everywhere else, it's
589 * an int. The error value can't be -1 any more, so it's
590 * now JB_INVALID_SOCKET (which is -1 on UNIX, and in
591 * Windows it maps to the #define INVALID_SOCKET.)
593 * - The signature of bind_port has changed.
595 * Revision 1.78 2002/03/08 21:35:04 oes
596 * Added optional group supplement to --user option. Will now use default group of user if no group given
598 * Revision 1.77 2002/03/07 03:52:06 oes
599 * - Fixed compiler warnings etc
600 * - Improved handling of failed DNS lookups
602 * Revision 1.76 2002/03/06 22:54:35 jongfoster
603 * Automated function-comment nitpicking.
605 * Revision 1.75 2002/03/06 10:02:19 oes
606 * Fixed stupid bug when --user was not given
608 * Revision 1.74 2002/03/06 00:49:31 jongfoster
609 * Fixing warning on Windows
610 * Making #ifdefs that refer to the same variable consistently
611 * use #ifdef unix rather than mixing #ifdef unix & #ifndef OS2
613 * Revision 1.73 2002/03/05 23:57:30 hal9
614 * Stray character 's' on line 1618 was breaking build.
616 * Revision 1.72 2002/03/05 21:33:45 david__schmidt
617 * - Re-enable OS/2 building after new parms were added
618 * - Fix false out of memory report when resolving CGI templates when no IP
619 * address is available of failed attempt (a la no such domain)
621 * Revision 1.71 2002/03/05 18:13:56 oes
622 * Added --user option
624 * Revision 1.70 2002/03/05 04:52:42 oes
625 * Deleted non-errlog debugging code
627 * Revision 1.69 2002/03/04 23:50:00 jongfoster
628 * Splitting off bind_port() call into bind_port_helper(), with
631 * Revision 1.68 2002/03/04 20:17:32 oes
634 * Revision 1.67 2002/03/04 18:18:57 oes
635 * - Removed _DEBUG mode
636 * - Cleand up cmdline parsing
637 * - Introduced --no-daemon, --pidfile options
638 * - Cleaned up signal handling:
639 * - Terminate cleanly on INT, TERM and ABRT
640 * - Schedule logfile for re-opening on HUP
641 * - Ignore CHLD and PIPE
642 * - Leave the rest with their default handlers
643 * - Uniform handler registration
644 * - Added usage() function
645 * - Played styleguide police
647 * Revision 1.66 2002/03/03 15:06:55 oes
648 * Re-enabled automatic config reloading
650 * Revision 1.65 2002/03/03 14:49:11 oes
651 * Fixed CLF logging: Now uses client's original HTTP request
653 * Revision 1.64 2002/03/03 09:18:03 joergs
654 * Made jumbjuster work on AmigaOS again.
656 * Revision 1.63 2002/03/02 04:14:50 david__schmidt
657 * Clean up a little CRLF unpleasantness that suddenly appeared
659 * Revision 1.62 2002/02/20 23:17:23 jongfoster
660 * Detecting some out-of memory conditions and exiting with a log message.
662 * Revision 1.61 2002/01/17 21:01:52 jongfoster
663 * Moving all our URL and URL pattern parsing code to urlmatch.c.
665 * Revision 1.60 2001/12/30 14:07:32 steudten
666 * - Add signal handling (unix)
667 * - Add SIGHUP handler (unix)
668 * - Add creation of pidfile (unix)
669 * - Add action 'top' in rc file (RH)
670 * - Add entry 'SIGNALS' to manpage
671 * - Add exit message to logfile (unix)
673 * Revision 1.59 2001/12/13 14:07:18 oes
674 * Fixed Bug: 503 error page now sent OK
676 * Revision 1.58 2001/11/30 23:37:24 jongfoster
677 * Renaming the Win32 config file to config.txt - this is almost the
678 * same as the corresponding UNIX name "config"
680 * Revision 1.57 2001/11/16 00:47:43 jongfoster
681 * Changing the tty-disconnection code to use setsid().
683 * Revision 1.56 2001/11/13 20:20:54 jongfoster
684 * Tabs->spaces, fixing a bug with missing {} around an if()
686 * Revision 1.55 2001/11/13 20:14:53 jongfoster
687 * Patch for FreeBSD setpgrp() as suggested by Alexander Lazic
689 * Revision 1.54 2001/11/07 00:03:14 steudten
690 * Give reliable return value if an error
691 * occurs not just 0 with new daemon mode.
693 * Revision 1.53 2001/11/05 21:41:43 steudten
694 * Add changes to be a real daemon just for unix os.
695 * (change cwd to /, detach from controlling tty, set
696 * process group and session leader to the own process.
698 * Add some fatal-error log message for failed malloc().
699 * Add '-d' if compiled with 'configure --with-debug' to
700 * enable debug output.
702 * Revision 1.52 2001/10/26 20:11:20 jongfoster
703 * Fixing type mismatch
705 * Revision 1.51 2001/10/26 17:38:28 oes
708 * Revision 1.50 2001/10/25 03:40:48 david__schmidt
709 * Change in porting tactics: OS/2's EMX porting layer doesn't allow multiple
710 * threads to call select() simultaneously. So, it's time to do a real, live,
711 * native OS/2 port. See defines for __EMX__ (the porting layer) vs. __OS2__
712 * (native). Both versions will work, but using __OS2__ offers multi-threading.
714 * Revision 1.49 2001/10/23 21:41:35 jongfoster
715 * Added call to initialize the (statically-allocated of course)
716 * "out of memory" CGI response.
718 * Revision 1.48 2001/10/10 19:56:46 jongfoster
719 * Moving some code that wasn't cookie-related out of an #ifdef
722 * Revision 1.47 2001/10/10 16:44:36 oes
723 * Added CONNECT destination port limitation check
725 * Revision 1.46 2001/10/08 15:17:41 oes
726 * Re-enabled SSL forwarding
728 * Revision 1.45 2001/10/07 15:42:11 oes
729 * Replaced 6 boolean members of csp with one bitmap (csp->flags)
731 * Moved downgrading of the HTTP version from parse_http_request to
732 * chat(), since we can't decide if it is necessary before we have
733 * determined the actions for the URL. The HTTP command is now
734 * *always* re-built so the repairs need no longer be special-cased.
736 * filter_popups now gets a csp pointer so it can raise the new
737 * CSP_FLAG_MODIFIED flag.
741 * Added configurable size limit for the IOB. If the IOB grows so
742 * large that the next read would exceed the limit, the header
743 * is generated, and the header & unfiltered buffer are flushed
744 * to the client. Chat then continues in non-buffering,
745 * non-filtering body mode.
747 * Revision 1.44 2001/10/02 18:13:57 oes
750 * Revision 1.43 2001/10/02 15:32:13 oes
751 * Moved generation of hdr
753 * Revision 1.42 2001/09/21 23:02:02 david__schmidt
754 * Cleaning up 2 compiler warnings on OS/2.
756 * Revision 1.41 2001/09/16 17:05:14 jongfoster
757 * Removing unused #include showarg.h
759 * Revision 1.40 2001/09/16 15:41:45 jongfoster
760 * Fixing signed/unsigned comparison warning.
762 * Revision 1.39 2001/09/16 13:21:27 jongfoster
763 * Changes to use new list functions.
765 * Revision 1.38 2001/09/16 13:01:46 jongfoster
766 * Removing redundant function call that zeroed zalloc()'d memory.
768 * Revision 1.37 2001/09/10 11:12:24 oes
769 * Deleted unused variable
771 * Revision 1.36 2001/09/10 10:56:15 oes
772 * Silenced compiler warnings
774 * Revision 1.35 2001/07/31 14:44:22 oes
775 * Deleted unused size parameter from filter_popups()
777 * Revision 1.34 2001/07/30 22:08:36 jongfoster
778 * Tidying up #defines:
779 * - All feature #defines are now of the form FEATURE_xxx
780 * - Permanently turned off WIN_GUI_EDIT
781 * - Permanently turned on WEBDAV and SPLIT_PROXY_ARGS
783 * Revision 1.33 2001/07/29 19:32:00 jongfoster
784 * Renaming _main() [mingw32 only] to real_main(), for ANSI compliance.
786 * Revision 1.32 2001/07/29 18:47:05 jongfoster
787 * Adding missing #include "loadcfg.h"
789 * Revision 1.31 2001/07/29 12:17:48 oes
790 * Applied pthread fix by Paul Lieverse
792 * Revision 1.30 2001/07/25 22:57:13 jongfoster
793 * __BEOS__ no longer overrides FEATURE_PTHREAD.
794 * This is because FEATURE_PTHREAD will soon be widely used, so I
795 * want to keep it simple.
797 * Revision 1.29 2001/07/24 12:47:06 oes
798 * Applied BeOS support update by Eugenia
800 * Revision 1.28 2001/07/23 13:26:12 oes
801 * Fixed bug in popup-killing for the first read that caused binary garbage to be sent between headers and body
803 * Revision 1.27 2001/07/19 19:09:47 haroon
804 * - Added code to take care of the situation where while processing the first
805 * server response (which includes the server header), after finding the end
806 * of the headers we were not looking past the end of the headers for
807 * content modification. I enabled it for filter_popups.
808 * Someone else should look to see if other similar operations should be
809 * done to the discarded portion of the buffer.
811 * Note 2001/07/20: No, the other content modification mechanisms will process
812 * the whole iob later anyway. --oes
814 * Revision 1.26 2001/07/18 12:31:36 oes
817 * Revision 1.25 2001/07/15 19:43:49 jongfoster
818 * Supports POSIX threads.
819 * Also removed some unused #includes.
821 * Revision 1.24 2001/07/13 14:00:40 oes
822 * - Generic content modification scheme:
823 * Each feature has its own applicability flag that is set
824 * from csp->action->flags.
825 * Replaced the "filtering" int flag , by a function pointer
826 * "content_filter" to the function that will do the content
827 * modification. If it is != NULL, the document will be buffered
828 * and processed through *content_filter, which must set
829 * csp->content_length and return a modified copy of the body
830 * or return NULL (on failiure).
831 * - Changed csp->is_text to the more generic bitmap csp->content_type
832 * which can currently take the valued CT_TEXT or CT_GIF
834 * - Removed all #ifdef PCRS
836 * Revision 1.23 2001/07/02 02:28:25 iwanttokeepanon
837 * Added "#ifdef ACL_FILES" conditional compilation to line 1291 to exclude
838 * the `block_acl' call. This prevents a compilation error when the user
839 * does not wish to use the "ACL" feature.
841 * Revision 1.22 2001/06/29 21:45:41 oes
842 * Indentation, CRLF->LF, Tab-> Space
844 * Revision 1.21 2001/06/29 13:29:36 oes
845 * - Cleaned up, improved comments
846 * - Unified all possible interceptors (CGI,
847 * block, trust, fast_redirect) in one
848 * place, with one (CGI) answer generation
849 * mechansim. Much clearer now.
850 * - Removed the GIF image generation, which
851 * is now done in filters.c:block_url()
852 * - Made error conditions like domain lookup
853 * failiure or (various) problems while talking
854 * to the server use cgi.c:error_response()
855 * instead of generating HTML/HTTP in chat() (yuck!)
856 * - Removed logentry from cancelled commit
858 * Revision 1.20 2001/06/09 10:55:28 jongfoster
859 * Changing BUFSIZ ==> BUFFER_SIZE
861 * Revision 1.19 2001/06/07 23:12:52 jongfoster
862 * Replacing function pointer in struct gateway with a directly
863 * called function forwarded_connect().
864 * Replacing struct gateway with struct forward_spec
866 * Revision 1.18 2001/06/03 19:12:16 oes
867 * introduced new cgi handling
869 * Revision 1.17 2001/06/01 20:07:23 jongfoster
870 * Now uses action +image-blocker{} rather than config->tinygif
872 * Revision 1.16 2001/06/01 18:49:17 jongfoster
873 * Replaced "list_share" with "list" - the tiny memory gain was not
874 * worth the extra complexity.
876 * Revision 1.15 2001/05/31 21:24:47 jongfoster
877 * Changed "permission" to "action" throughout.
878 * Removed DEFAULT_USER_AGENT - it must now be specified manually.
879 * Moved vanilla wafer check into chat(), since we must now
880 * decide whether or not to add it based on the URL.
882 * Revision 1.14 2001/05/29 20:14:01 joergs
883 * AmigaOS bugfix: PCRS needs a lot of stack, stacksize for child threads
886 * Revision 1.13 2001/05/29 09:50:24 jongfoster
887 * Unified blocklist/imagelist/permissionslist.
888 * File format is still under discussion, but the internal changes
891 * Also modified interceptor behaviour:
892 * - We now intercept all URLs beginning with one of the following
893 * prefixes (and *only* these prefixes):
895 * * http://ijbswa.sf.net/config/
896 * * http://ijbswa.sourceforge.net/config/
897 * - New interceptors "home page" - go to http://i.j.b/ to see it.
898 * - Internal changes so that intercepted and fast redirect pages
899 * are not replaced with an image.
900 * - Interceptors now have the option to send a binary page direct
901 * to the client. (i.e. ijb-send-banner uses this)
902 * - Implemented show-url-info interceptor. (Which is why I needed
903 * the above interceptors changes - a typical URL is
904 * "http://i.j.b/show-url-info?url=www.somesite.com/banner.gif".
905 * The previous mechanism would not have intercepted that, and
906 * if it had been intercepted then it then it would have replaced
909 * Revision 1.12 2001/05/27 22:17:04 oes
911 * - re_process_buffer no longer writes the modified buffer
912 * to the client, which was very ugly. It now returns the
913 * buffer, which it is then written by chat.
915 * - content_length now adjusts the Content-Length: header
916 * for modified documents rather than crunch()ing it.
917 * (Length info in csp->content_length, which is 0 for
918 * unmodified documents)
920 * - For this to work, sed() is called twice when filtering.
922 * Revision 1.11 2001/05/26 17:27:53 jongfoster
923 * Added support for CLF and fixed LOG_LEVEL_LOG.
924 * Also did CRLF->LF fix of my previous patch.
926 * Revision 1.10 2001/05/26 15:26:15 jongfoster
927 * ACL feature now provides more security by immediately dropping
928 * connections from untrusted hosts.
930 * Revision 1.9 2001/05/26 00:28:36 jongfoster
931 * Automatic reloading of config file.
932 * Removed obsolete SIGHUP support (Unix) and Reload menu option (Win32).
933 * Most of the global variables have been moved to a new
934 * struct configuration_spec, accessed through csp->config->globalname
935 * Most of the globals remaining are used by the Win32 GUI.
937 * Revision 1.8 2001/05/25 22:43:18 jongfoster
938 * Fixing minor memory leak and buffer overflow.
940 * Revision 1.7 2001/05/25 22:34:30 jongfoster
943 * Revision 1.6 2001/05/23 00:13:58 joergs
944 * AmigaOS support fixed.
946 * Revision 1.5 2001/05/22 18:46:04 oes
948 * - Enabled filtering banners by size rather than URL
949 * by adding patterns that replace all standard banner
950 * sizes with the "Junkbuster" gif to the re_filterfile
952 * - Enabled filtering WebBugs by providing a pattern
953 * which kills all 1x1 images
955 * - Added support for PCRE_UNGREEDY behaviour to pcrs,
956 * which is selected by the (nonstandard and therefore
957 * capital) letter 'U' in the option string.
958 * It causes the quantifiers to be ungreedy by default.
959 * Appending a ? turns back to greedy (!).
961 * - Added a new interceptor ijb-send-banner, which
962 * sends back the "Junkbuster" gif. Without imagelist or
963 * MSIE detection support, or if tinygif = 1, or the
964 * URL isn't recognized as an imageurl, a lame HTML
965 * explanation is sent instead.
967 * - Added new feature, which permits blocking remote
968 * script redirects and firing back a local redirect
970 * The feature is conditionally compiled, i.e. it
971 * can be disabled with --disable-fast-redirects,
972 * plus it must be activated by a "fast-redirects"
973 * line in the config file, has its own log level
974 * and of course wants to be displayed by show-proxy-args
975 * Note: Boy, all the #ifdefs in 1001 locations and
976 * all the fumbling with configure.in and acconfig.h
977 * were *way* more work than the feature itself :-(
979 * - Because a generic redirect template was needed for
980 * this, tinygif = 3 now uses the same.
982 * - Moved GIFs, and other static HTTP response templates
987 * - Removed some >400 CRs again (Jon, you really worked
990 * Revision 1.4 2001/05/21 19:34:01 jongfoster
991 * Made failure to bind() a fatal error.
993 * Revision 1.3 2001/05/20 01:21:20 jongfoster
994 * Version 2.9.4 checkin.
995 * - Merged popupfile and cookiefile, and added control over PCRS
996 * filtering, in new "permissionsfile".
997 * - Implemented LOG_LEVEL_FATAL, so that if there is a configuration
998 * file error you now get a message box (in the Win32 GUI) rather
999 * than the program exiting with no explanation.
1000 * - Made killpopup use the PCRS MIME-type checking and HTTP-header
1002 * - Removed tabs from "config"
1003 * - Moved duplicated url parsing code in "loaders.c" to a new funcition.
1004 * - Bumped up version number.
1006 * Revision 1.2 2001/05/17 22:34:44 oes
1007 * - Added hint on GIF char array generation to jcc.c
1008 * - Cleaned CRLF's from the sources and related files
1009 * - Repaired logging for REF and FRC
1011 * Revision 1.1.1.1 2001/05/15 13:58:56 oes
1012 * Initial import of version 2.9.3 source tree
1015 *********************************************************************/
1021 #include <sys/types.h>
1030 # ifndef FEATURE_PTHREAD
1034 # include <windows.h>
1035 # include <process.h>
1036 # endif /* ndef FEATURE_PTHREAD */
1039 # ifndef _WIN_CONSOLE
1040 # include "w32log.h"
1041 # endif /* ndef _WIN_CONSOLE */
1042 # include "w32svrapi.h"
1044 #else /* ifndef _WIN32 */
1046 # if !defined (__OS2__)
1047 # include <unistd.h>
1048 # include <sys/wait.h>
1049 # endif /* ndef __OS2__ */
1050 # include <sys/time.h>
1051 # include <sys/stat.h>
1052 # include <sys/ioctl.h>
1055 #include <sys/termios.h>
1063 # include <signal.h>
1066 # include <socket.h> /* BeOS has select() for sockets only. */
1067 # include <OS.h> /* declarations for threads and stuff. */
1070 # if defined(__EMX__) || defined(__OS2__)
1071 # include <sys/select.h> /* OS/2/EMX needs a little help with select */
1076 #define bzero(B,N) memset(B,0x00,n)
1080 # include <select.h>
1085 #include "project.h"
1088 #include "filters.h"
1089 #include "loaders.h"
1090 #include "parsers.h"
1091 #include "miscutil.h"
1093 #include "jbsockets.h"
1094 #include "gateway.h"
1095 #include "actions.h"
1097 #include "loadcfg.h"
1098 #include "urlmatch.h"
1100 const char jcc_h_rcs[] = JCC_H_VERSION;
1101 const char project_h_rcs[] = PROJECT_H_VERSION;
1104 struct client_state clients[1];
1105 struct file_list files[1];
1107 #ifdef FEATURE_STATISTICS
1108 int urls_read = 0; /* total nr of urls read inc rejected */
1109 int urls_rejected = 0; /* total nr of urls rejected */
1110 #endif /* def FEATURE_STATISTICS */
1112 #ifdef FEATURE_GRACEFUL_TERMINATION
1113 int g_terminate = 0;
1116 #if !defined(_WIN32) && !defined(__OS2__) && !defined(AMIGA)
1117 static void sig_handler(int the_signal);
1119 static int client_protocol_is_unsupported(const struct client_state *csp, char *req);
1120 static jb_err get_request_destination_elsewhere(struct client_state *csp, struct list *headers);
1121 static jb_err get_server_headers(struct client_state *csp);
1122 static const char *crunch_reason(const struct http_response *rsp);
1123 static void send_crunch_response(const struct client_state *csp, struct http_response *rsp);
1125 * static int request_contains_null_bytes(const struct client_state *csp, char *buf, int len);
1127 static void build_request_line(struct client_state *csp, const struct forward_spec *fwd, char **request_line);
1128 static jb_err change_request_destination(struct client_state *csp);
1129 static void chat(struct client_state *csp);
1130 static void serve(struct client_state *csp);
1131 #if !defined(_WIN32) || defined(_WIN_CONSOLE)
1132 static void usage(const char *myname);
1134 static void initialize_mutexes(void);
1135 static jb_socket bind_port_helper(struct configuration_spec *config);
1136 static void listen_loop(void);
1139 void serve(struct client_state *csp);
1140 #else /* ifndef AMIGA */
1141 static void serve(struct client_state *csp);
1142 #endif /* def AMIGA */
1145 static int32 server_thread(void *data);
1146 #endif /* def __BEOS__ */
1149 #define sleep(N) Sleep(((N) * 1000))
1153 #define sleep(N) DosSleep(((N) * 100))
1156 #ifdef FEATURE_PTHREAD
1157 pthread_mutex_t log_mutex;
1158 pthread_mutex_t log_init_mutex;
1160 #if !defined(HAVE_GETHOSTBYADDR_R) || !defined(HAVE_GETHOSTBYNAME_R)
1161 pthread_mutex_t resolver_mutex;
1162 #endif /* !defined(HAVE_GETHOSTBYADDR_R) || !defined(HAVE_GETHOSTBYNAME_R) */
1164 #ifndef HAVE_GMTIME_R
1165 pthread_mutex_t gmtime_mutex;
1166 #endif /* ndef HAVE_GMTIME_R */
1168 #ifndef HAVE_LOCALTIME_R
1169 pthread_mutex_t localtime_mutex;
1170 #endif /* ndef HAVE_GMTIME_R */
1173 pthread_mutex_t rand_mutex;
1174 #endif /* ndef HAVE_RANDOM */
1176 #endif /* FEATURE_PTHREAD */
1178 #if defined(unix) || defined(__EMX__)
1179 const char *basedir = NULL;
1180 const char *pidfile = NULL;
1181 int received_hup_signal = 0;
1182 #endif /* defined unix */
1184 /* HTTP snipplets. */
1185 static const char CSUCCEED[] =
1186 "HTTP/1.0 200 Connection established\r\n"
1187 "Proxy-Agent: Privoxy/" VERSION "\r\n\r\n";
1189 static const char CHEADER[] =
1190 "HTTP/1.0 400 Invalid header received from client\r\n"
1191 "Proxy-Agent: Privoxy " VERSION "\r\n"
1192 "Content-Type: text/plain\r\n"
1193 "Connection: close\r\n\r\n"
1194 "Invalid header received from client.\r\n";
1196 static const char FTP_RESPONSE[] =
1197 "HTTP/1.0 400 Invalid request received from client\r\n"
1198 "Content-Type: text/plain\r\n"
1199 "Connection: close\r\n\r\n"
1200 "Invalid request. Privoxy doesn't support FTP.\r\n";
1202 static const char GOPHER_RESPONSE[] =
1203 "HTTP/1.0 400 Invalid request received from client\r\n"
1204 "Content-Type: text/plain\r\n"
1205 "Connection: close\r\n\r\n"
1206 "Invalid request. Privoxy doesn't support gopher.\r\n";
1208 /* XXX: should be a template */
1209 static const char MISSING_DESTINATION_RESPONSE[] =
1210 "HTTP/1.0 400 Bad request received from client\r\n"
1211 "Proxy-Agent: Privoxy " VERSION "\r\n"
1212 "Content-Type: text/plain\r\n"
1213 "Connection: close\r\n\r\n"
1214 "Bad request. Privoxy was unable to extract the destination.\r\n";
1216 /* XXX: should be a template */
1217 static const char NO_SERVER_DATA_RESPONSE[] =
1218 "HTTP/1.0 502 Server or forwarder response empty\r\n"
1219 "Proxy-Agent: Privoxy " VERSION "\r\n"
1220 "Content-Type: text/plain\r\n"
1221 "Connection: close\r\n\r\n"
1222 "Empty server or forwarder response.\r\n"
1223 "The connection has been closed but Privoxy didn't receive any data.\r\n";
1225 /* XXX: should be a template */
1226 static const char INVALID_SERVER_HEADERS_RESPONSE[] =
1227 "HTTP/1.0 502 Server or forwarder response invalid\r\n"
1228 "Proxy-Agent: Privoxy " VERSION "\r\n"
1229 "Content-Type: text/plain\r\n"
1230 "Connection: close\r\n\r\n"
1231 "Bad response. The server or forwarder response doesn't look like HTTP.\r\n";
1234 /* XXX: should be a template */
1235 static const char NULL_BYTE_RESPONSE[] =
1236 "HTTP/1.0 400 Bad request received from client\r\n"
1237 "Proxy-Agent: Privoxy " VERSION "\r\n"
1238 "Content-Type: text/plain\r\n"
1239 "Connection: close\r\n\r\n"
1240 "Bad request. Null byte(s) before end of request.\r\n";
1243 /* XXX: should be a template */
1244 static const char MESSED_UP_REQUEST_RESPONSE[] =
1245 "HTTP/1.0 400 Malformed request after rewriting\r\n"
1246 "Proxy-Agent: Privoxy " VERSION "\r\n"
1247 "Content-Type: text/plain\r\n"
1248 "Connection: close\r\n\r\n"
1249 "Bad request. Messed up with header filters.\r\n";
1251 /* A function to crunch a response */
1252 typedef struct http_response *(*crunch_func_ptr)(struct client_state *);
1254 /* Crunch function flags */
1255 #define CF_NO_FLAGS 0
1256 /* Cruncher applies to forced requests as well */
1257 #define CF_IGNORE_FORCE 1
1258 /* Crunched requests are counted for the block statistics */
1259 #define CF_COUNT_AS_REJECT 2
1261 /* A crunch function and its flags */
1264 const crunch_func_ptr cruncher;
1268 static int crunch_response_triggered(struct client_state *csp, const struct cruncher crunchers[]);
1270 /* Complete list of cruncher functions */
1271 static const struct cruncher crunchers_all[] = {
1272 { direct_response, CF_COUNT_AS_REJECT|CF_IGNORE_FORCE},
1273 { block_url, CF_COUNT_AS_REJECT },
1274 #ifdef FEATURE_TRUST
1275 { trust_url, CF_COUNT_AS_REJECT },
1276 #endif /* def FEATURE_TRUST */
1277 { redirect_url, CF_NO_FLAGS },
1278 { dispatch_cgi, CF_IGNORE_FORCE},
1282 /* Light version, used after tags are applied */
1283 static const struct cruncher crunchers_light[] = {
1284 { block_url, CF_COUNT_AS_REJECT },
1285 { redirect_url, CF_NO_FLAGS },
1290 #if !defined(_WIN32) && !defined(__OS2__) && !defined(AMIGA)
1291 /*********************************************************************
1293 * Function : sig_handler
1295 * Description : Signal handler for different signals.
1296 * Exit gracefully on TERM and INT
1297 * or set a flag that will cause the errlog
1298 * to be reopened by the main thread on HUP.
1301 * 1 : the_signal = the signal cause this function to call
1305 *********************************************************************/
1306 static void sig_handler(int the_signal)
1312 log_error(LOG_LEVEL_INFO, "exiting by signal %d .. bye", the_signal);
1323 received_hup_signal = 1;
1328 * We shouldn't be here, unless we catch signals
1329 * in main() that we can't handle here!
1331 log_error(LOG_LEVEL_FATAL, "sig_handler: exiting on unexpected signal %d", the_signal);
1339 /*********************************************************************
1341 * Function : client_protocol_is_unsupported
1343 * Description : Checks if the client used a known unsupported
1344 * protocol and deals with it by sending an error
1348 * 1 : csp = Current client state (buffers, headers, etc...)
1349 * 2 : req = the first request line send by the client
1351 * Returns : TRUE if an error response has been generated, or
1352 * FALSE if the request doesn't look invalid.
1354 *********************************************************************/
1355 static int client_protocol_is_unsupported(const struct client_state *csp, char *req)
1358 * If it's a FTP or gopher request, we don't support it.
1360 * These checks are better than nothing, but they might
1361 * not work in all configurations and some clients might
1362 * have problems digesting the answer.
1364 * They should, however, never cause more problems than
1365 * Privoxy's old behaviour (returning the misleading HTML
1368 * "Could not resolve http://(ftp|gopher)://example.org").
1370 if (!strncmpic(req, "GET ftp://", 10) || !strncmpic(req, "GET gopher://", 13))
1372 const char *response = NULL;
1373 const char *protocol = NULL;
1375 if (!strncmpic(req, "GET ftp://", 10))
1377 response = FTP_RESPONSE;
1382 response = GOPHER_RESPONSE;
1383 protocol = "GOPHER";
1385 log_error(LOG_LEVEL_ERROR,
1386 "%s tried to use Privoxy as %s proxy: %s",
1387 csp->ip_addr_str, protocol, req);
1388 log_error(LOG_LEVEL_CLF,
1389 "%s - - [%T] \"%s\" 400 0", csp->ip_addr_str, req);
1391 write_socket(csp->cfd, response, strlen(response));
1400 /*********************************************************************
1402 * Function : get_request_destination_elsewhere
1404 * Description : If the client's request was redirected into
1405 * Privoxy without the client's knowledge,
1406 * the request line lacks the destination host.
1408 * This function tries to get it elsewhere,
1409 * provided accept-intercepted-requests is enabled.
1411 * "Elsewhere" currently only means "Host: header",
1412 * but in the future we may ask the redirecting
1413 * packet filter to look the destination up.
1415 * If the destination stays unknown, an error
1416 * response is send to the client and headers
1417 * are freed so that chat() can return directly.
1420 * 1 : csp = Current client state (buffers, headers, etc...)
1421 * 2 : headers = a header list
1423 * Returns : JB_ERR_OK if the destination is now known, or
1424 * JB_ERR_PARSE if it isn't.
1426 *********************************************************************/
1427 static jb_err get_request_destination_elsewhere(struct client_state *csp, struct list *headers)
1431 if (!(csp->config->feature_flags & RUNTIME_FEATURE_ACCEPT_INTERCEPTED_REQUESTS))
1433 log_error(LOG_LEVEL_ERROR, "%s's request: \'%s\' is invalid."
1434 " Privoxy isn't configured to accept intercepted requests.",
1435 csp->ip_addr_str, csp->http->cmd);
1436 /* XXX: Use correct size */
1437 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" 400 0",
1438 csp->ip_addr_str, csp->http->cmd);
1440 write_socket(csp->cfd, CHEADER, strlen(CHEADER));
1441 destroy_list(headers);
1443 return JB_ERR_PARSE;
1445 else if (JB_ERR_OK == get_destination_from_headers(headers, csp->http))
1447 /* Split the domain we just got for pattern matching */
1448 init_domain_components(csp->http);
1454 /* We can't work without destination. Go spread the news.*/
1456 req = list_to_text(headers);
1458 /* XXX: Use correct size */
1459 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" 400 0",
1460 csp->ip_addr_str, csp->http->cmd);
1461 log_error(LOG_LEVEL_ERROR,
1462 "Privoxy was unable to get the destination for %s's request:\n%s\n%s",
1463 csp->ip_addr_str, csp->http->cmd, req);
1466 write_socket(csp->cfd, MISSING_DESTINATION_RESPONSE, strlen(MISSING_DESTINATION_RESPONSE));
1467 destroy_list(headers);
1469 return JB_ERR_PARSE;
1472 * TODO: If available, use PF's ioctl DIOCNATLOOK as last resort
1473 * to get the destination IP address, use it as host directly
1474 * or do a reverse DNS lookup first.
1479 /*********************************************************************
1481 * Function : get_server_headers
1483 * Description : Parses server headers in iob and fills them
1484 * into csp->headers so that they can later be
1488 * 1 : csp = Current client state (buffers, headers, etc...)
1490 * Returns : JB_ERR_OK if everything went fine, or
1491 * JB_ERR_PARSE if the headers were incomplete.
1493 *********************************************************************/
1494 static jb_err get_server_headers(struct client_state *csp)
1496 int continue_hack_in_da_house = 0;
1499 while (((header = get_header(csp->iob)) != NULL) || continue_hack_in_da_house)
1504 * continue hack in da house. Ignore the ending of
1505 * this head and continue enlisting header lines.
1506 * The reason is described below.
1508 enlist(csp->headers, "");
1509 continue_hack_in_da_house = 0;
1512 else if (0 == strncmpic(header, "HTTP/1.1 100", 12))
1515 * It's a bodyless continue response, don't
1516 * stop header parsing after reaching its end.
1518 * As a result Privoxy will concatenate the
1519 * next response's head and parse and deliver
1520 * the headers as if they belonged to one request.
1522 * The client will separate them because of the
1523 * empty line between them.
1525 * XXX: What we're doing here is clearly against
1526 * the intended purpose of the continue header,
1527 * and under some conditions (HTTP/1.0 client request)
1528 * it's a standard violation.
1530 * Anyway, "sort of against the spec" is preferable
1531 * to "always getting confused by Continue responses"
1532 * (Privoxy's behaviour before this hack was added)
1534 log_error(LOG_LEVEL_HEADER, "Continue hack in da house.");
1535 continue_hack_in_da_house = 1;
1537 else if (*header == '\0')
1540 * If the header is empty, but the Continue hack
1541 * isn't active, we can assume that we reached the
1542 * end of the buffer before we hit the end of the
1545 * Inform the caller an let it decide how to handle it.
1547 return JB_ERR_PARSE;
1550 if (JB_ERR_MEMORY == enlist(csp->headers, header))
1553 * XXX: Should we quit the request and return a
1554 * out of memory error page instead?
1556 log_error(LOG_LEVEL_ERROR,
1557 "Out of memory while enlisting server headers. %s lost.",
1567 /*********************************************************************
1569 * Function : crunch_reason
1571 * Description : Translates the crunch reason code into a string.
1574 * 1 : rsp = a http_response
1576 * Returns : A string with the crunch reason or an error description.
1578 *********************************************************************/
1579 static const char *crunch_reason(const struct http_response *rsp)
1581 char * reason = NULL;
1583 assert(rsp != NULL);
1586 return "Internal error while searching for crunch reason";
1589 switch (rsp->reason)
1591 case RSP_REASON_UNSUPPORTED:
1592 reason = "Unsupported HTTP feature";
1594 case RSP_REASON_BLOCKED:
1597 case RSP_REASON_UNTRUSTED:
1598 reason = "Untrusted";
1600 case RSP_REASON_REDIRECTED:
1601 reason = "Redirected";
1603 case RSP_REASON_CGI_CALL:
1604 reason = "CGI Call";
1606 case RSP_REASON_NO_SUCH_DOMAIN:
1607 reason = "DNS failure";
1609 case RSP_REASON_FORWARDING_FAILED:
1610 reason = "Forwarding failed";
1612 case RSP_REASON_CONNECT_FAILED:
1613 reason = "Connection failure";
1615 case RSP_REASON_OUT_OF_MEMORY:
1616 reason = "Out of memory (may mask other reasons)";
1619 reason = "No reason recorded";
1627 /*********************************************************************
1629 * Function : send_crunch_response
1631 * Description : Delivers already prepared response for
1632 * intercepted requests, logs the interception
1633 * and frees the response.
1636 * 1 : csp = Current client state (buffers, headers, etc...)
1637 * 1 : rsp = Fully prepared response. Will be freed on exit.
1639 * Returns : Nothing.
1641 *********************************************************************/
1642 static void send_crunch_response(const struct client_state *csp, struct http_response *rsp)
1644 const struct http_request *http = csp->http;
1645 char status_code[4];
1647 assert(rsp != NULL);
1648 assert(rsp->head != NULL);
1653 * Not supposed to happen. If it does
1654 * anyway, treat it as an unknown error.
1656 cgi_error_unknown(csp, rsp, RSP_REASON_INTERNAL_ERROR);
1657 /* return code doesn't matter */
1662 /* If rsp is still NULL, we have serious internal problems. */
1663 log_error(LOG_LEVEL_FATAL,
1664 "NULL response in send_crunch_response and cgi_error_unknown failed as well.");
1668 * Extract the status code from the actual head
1669 * that was send to the client. It is the only
1670 * way to get it right for all requests, including
1671 * the fixed ones for out-of-memory problems.
1673 * A head starts like this: 'HTTP/1.1 200...'
1677 status_code[0] = rsp->head[9];
1678 status_code[1] = rsp->head[10];
1679 status_code[2] = rsp->head[11];
1680 status_code[3] = '\0';
1682 /* Write the answer to the client */
1683 if (write_socket(csp->cfd, rsp->head, rsp->head_length)
1684 || write_socket(csp->cfd, rsp->body, rsp->content_length))
1686 /* There is nothing we can do about it. */
1687 log_error(LOG_LEVEL_ERROR, "write to: %s failed: %E", csp->http->host);
1690 /* Log that the request was crunched and why. */
1691 log_error(LOG_LEVEL_GPC, "%s%s crunch! (%s)",
1692 http->hostport, http->path, crunch_reason(rsp));
1693 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" %s %d",
1694 csp->ip_addr_str, http->ocmd, status_code, rsp->content_length);
1696 /* Clean up and return */
1697 if (cgi_error_memory() != rsp)
1699 free_http_response(rsp);
1706 /*********************************************************************
1708 * Function : request_contains_null_bytes
1710 * Description : Checks for NULL bytes in the request and sends
1711 * an error message to the client if any were found.
1713 * XXX: currently not used, see comment in chat().
1716 * 1 : csp = Current client state (buffers, headers, etc...)
1717 * 2 : buf = Data from the client's request to check.
1718 * 3 : len = The data length.
1720 * Returns : TRUE if the request contained one or more NULL bytes, or
1723 *********************************************************************/
1724 static int request_contains_null_bytes(const struct client_state *csp, char *buf, int len)
1726 size_t c_len; /* Request lenght when treated as C string */
1728 c_len = strlen(buf);
1733 * Null byte(s) found. Log the request,
1734 * return an error response and hang up.
1736 size_t tmp_len = c_len;
1741 * Replace NULL byte(s) with '°' characters
1742 * so the request can be logged as string.
1743 * XXX: Is there a better replacement character?
1746 tmp_len += strlen(buf+tmp_len);
1747 } while (tmp_len < len);
1749 log_error(LOG_LEVEL_ERROR, "%s\'s request contains at least one NULL byte "
1750 "(length=%d, strlen=%d).", csp->ip_addr_str, len, c_len);
1751 log_error(LOG_LEVEL_HEADER,
1752 "Offending request data with NULL bytes turned into \'°\' characters: %s", buf);
1754 write_socket(csp->cfd, NULL_BYTE_RESPONSE, strlen(NULL_BYTE_RESPONSE));
1756 /* XXX: Log correct size */
1757 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"Invalid request\" 400 0", csp->ip_addr_str);
1767 /*********************************************************************
1769 * Function : crunch_response_triggered
1771 * Description : Checks if the request has to be crunched,
1772 * and delivers the crunch response if necessary.
1775 * 1 : csp = Current client state (buffers, headers, etc...)
1776 * 2 : crunchers = list of cruncher functions to run
1778 * Returns : TRUE if the request was answered with a crunch response
1781 *********************************************************************/
1782 static int crunch_response_triggered(struct client_state *csp, const struct cruncher crunchers[])
1784 struct http_response *rsp = NULL;
1785 const struct cruncher *c;
1788 * If CGI request crunching is disabled,
1789 * check the CGI dispatcher out of order to
1790 * prevent unintentional blocks or redirects.
1792 if (!(csp->config->feature_flags & RUNTIME_FEATURE_CGI_CRUNCHING)
1793 && (NULL != (rsp = dispatch_cgi(csp))))
1795 /* Deliver, log and free the interception response. */
1796 send_crunch_response(csp, rsp);
1800 for (c = crunchers; c->cruncher != NULL; c++)
1803 * Check the cruncher if either Privoxy is toggled
1804 * on and the request isn't forced, or if the cruncher
1805 * applies to forced requests as well.
1807 if (((csp->flags & CSP_FLAG_TOGGLED_ON) &&
1808 !(csp->flags & CSP_FLAG_FORCED)) ||
1809 (c->flags & CF_IGNORE_FORCE))
1811 rsp = c->cruncher(csp);
1814 /* Deliver, log and free the interception response. */
1815 send_crunch_response(csp, rsp);
1816 #ifdef FEATURE_STATISTICS
1817 if (c->flags & CF_COUNT_AS_REJECT)
1819 csp->flags |= CSP_FLAG_REJECTED;
1821 #endif /* def FEATURE_STATISTICS */
1832 /*********************************************************************
1834 * Function : build_request_line
1836 * Description : Builds the HTTP request line.
1838 * If a HTTP forwarder is used it expects the whole URL,
1839 * web servers only get the path.
1842 * 1 : csp = Current client state (buffers, headers, etc...)
1843 * 2 : fwd = The forwarding spec used for the request
1844 * XXX: Should use http->fwd instead.
1845 * 3 : request_line = The old request line which will be replaced.
1847 * Returns : Nothing. Terminates in case of memory problems.
1849 *********************************************************************/
1850 static void build_request_line(struct client_state *csp, const struct forward_spec *fwd, char **request_line)
1852 struct http_request *http = csp->http;
1854 assert(http->ssl == 0);
1857 * Downgrade http version from 1.1 to 1.0
1858 * if +downgrade action applies.
1860 if ( (csp->action->flags & ACTION_DOWNGRADE)
1861 && (!strcmpic(http->ver, "HTTP/1.1")))
1864 http->ver = strdup("HTTP/1.0");
1866 if (http->ver == NULL)
1868 log_error(LOG_LEVEL_FATAL, "Out of memory downgrading HTTP version");
1873 * Rebuild the request line.
1875 freez(*request_line);
1876 *request_line = strdup(http->gpc);
1877 string_append(request_line, " ");
1879 if (fwd->forward_host)
1881 string_append(request_line, http->url);
1885 string_append(request_line, http->path);
1887 string_append(request_line, " ");
1888 string_append(request_line, http->ver);
1890 if (*request_line == NULL)
1892 log_error(LOG_LEVEL_FATAL, "Out of memory writing HTTP command");
1894 log_error(LOG_LEVEL_HEADER, "New HTTP Request-Line: %s", *request_line);
1898 /*********************************************************************
1900 * Function : change_request_destination
1902 * Description : Parse a (rewritten) request line and regenerate
1903 * the http request data.
1906 * 1 : csp = Current client state (buffers, headers, etc...)
1908 * Returns : Forwards the parse_http_request() return code.
1909 * Terminates in case of memory problems.
1911 *********************************************************************/
1912 static jb_err change_request_destination(struct client_state *csp)
1914 struct http_request *http = csp->http;
1917 log_error(LOG_LEVEL_INFO, "Rewrite detected: %s", csp->headers->first->str);
1918 free_http_request(http);
1919 err = parse_http_request(csp->headers->first->str, http, csp);
1920 if (JB_ERR_OK != err)
1922 log_error(LOG_LEVEL_ERROR, "Couldn't parse rewritten request: %s.",
1923 jb_err_to_string(err));
1927 /* XXX: ocmd is a misleading name */
1928 http->ocmd = strdup(http->cmd);
1929 if (http->ocmd == NULL)
1931 log_error(LOG_LEVEL_FATAL,
1932 "Out of memory copying rewritten HTTP request line");
1940 /*********************************************************************
1944 * Description : Once a connection to the client has been accepted,
1945 * this function is called (via serve()) to handle the
1946 * main business of the communication. When this
1947 * function returns, the caller must close the client
1950 * FIXME: chat is nearly thousand lines long.
1954 * 1 : csp = Current client state (buffers, headers, etc...)
1956 * Returns : Nothing.
1958 *********************************************************************/
1959 static void chat(struct client_state *csp)
1961 char buf[BUFFER_SIZE];
1969 int ms_iis5_hack = 0;
1970 size_t byte_count = 0;
1971 int forwarded_connect_retries = 0;
1972 int max_forwarded_connect_retries = csp->config->forwarded_connect_retries;
1973 const struct forward_spec * fwd;
1974 struct http_request *http;
1975 int len; /* for buffer sizes (and negative error codes) */
1978 /* Function that does the content filtering for the current request */
1979 filter_function_ptr content_filter = NULL;
1981 /* Skeleton for HTTP response, if we should intercept the request */
1982 struct http_response *rsp;
1984 /* Temporary copy of the client's headers before they get enlisted in csp->headers */
1985 struct list header_list;
1986 struct list *headers = &header_list;
1990 memset(buf, 0, sizeof(buf));
1993 * Read the client's request. Note that since we're not using select() we
1994 * could get blocked here if a client connected, then didn't say anything!
1999 len = read_socket(csp->cfd, buf, sizeof(buf) - 1);
2001 if (len <= 0) break; /* error! */
2004 * If there is no memory left for buffering the
2005 * request, there is nothing we can do but hang up
2007 if (add_to_iob(csp, buf, len))
2012 req = get_header(csp->iob);
2014 } while ((NULL != req) && ('\0' == *req));
2016 if ((NULL != req) && ('\0' != *req))
2018 /* Request received. Validate and parse it. */
2022 * XXX: Temporary disabled to prevent problems
2023 * with POST requests whose bodies are allowed to
2024 * contain NULL bytes. BR#1730105.
2026 * The main purpose of this check is to properly
2027 * log stuff like BitTorrent traffic and other junk
2028 * that hits public proxies. It's not required for
2029 * Privoxy to functions as those requests are discarded
2032 * It probably should be rewritten to only check
2033 * the head of the request. Another option would
2034 * be to let all POST requests pass, although that
2035 * may not be good enough.
2037 if (request_contains_null_bytes(csp, buf, len))
2039 /* NULL bytes found and dealt with, just hang up. */
2044 /* Does the request line look invalid? */
2045 if (client_protocol_is_unsupported(csp, req))
2048 * Yes. The request has already been
2049 * answered with a error response, the buffers
2050 * were freed and we're done with chatting.
2055 #ifdef FEATURE_FORCE_LOAD
2057 * If this request contains the FORCE_PREFIX and blocks
2058 * aren't enforced, get rid of it and set the force flag.
2060 if (strstr(req, FORCE_PREFIX))
2062 if (csp->config->feature_flags & RUNTIME_FEATURE_ENFORCE_BLOCKS)
2064 log_error(LOG_LEVEL_FORCE,
2065 "Ignored force prefix in request: \"%s\".", req);
2069 strclean(req, FORCE_PREFIX);
2070 log_error(LOG_LEVEL_FORCE, "Enforcing request: \"%s\".", req);
2071 csp->flags |= CSP_FLAG_FORCED;
2075 #endif /* def FEATURE_FORCE_LOAD */
2076 err = parse_http_request(req, http, csp);
2077 if (JB_ERR_OK != err)
2079 log_error(LOG_LEVEL_ERROR, "Couldn't parse request: %s.", jb_err_to_string(err));
2085 if (http->cmd == NULL)
2087 write_socket(csp->cfd, CHEADER, strlen(CHEADER));
2088 /* XXX: Use correct size */
2089 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"Invalid request\" 400 0", csp->ip_addr_str);
2090 log_error(LOG_LEVEL_ERROR, "Invalid header received from %s.", csp->ip_addr_str);
2092 free_http_request(http);
2096 /* grab the rest of the client's headers */
2100 p = get_header(csp->iob);
2104 /* There are no additional headers to read. */
2111 * We didn't receive a complete header
2112 * line yet, get the rest of it.
2114 len = read_socket(csp->cfd, buf, sizeof(buf) - 1);
2117 log_error(LOG_LEVEL_ERROR, "read from client failed: %E");
2118 destroy_list(headers);
2122 if (add_to_iob(csp, buf, len))
2125 * If there is no memory left for buffering the
2126 * request, there is nothing we can do but hang up
2128 destroy_list(headers);
2135 * We were able to read a complete
2136 * header and can finaly enlist it.
2143 if (http->host == NULL)
2146 * If we still don't know the request destination,
2147 * the request is invalid or the client uses
2148 * Privoxy without its knowledge.
2150 if (JB_ERR_OK != get_request_destination_elsewhere(csp, headers))
2153 * Our attempts to get the request destination
2154 * elsewhere failed or Privoxy is configured
2155 * to only accept proxy requests.
2157 * An error response has already been send
2158 * and we're done here.
2165 * Determine the actions for this URL
2167 #ifdef FEATURE_TOGGLE
2168 if (!(csp->flags & CSP_FLAG_TOGGLED_ON))
2170 /* Most compatible set of actions (i.e. none) */
2171 init_current_action(csp->action);
2174 #endif /* ndef FEATURE_TOGGLE */
2176 get_url_actions(csp, http);
2180 * Save a copy of the original request for logging
2182 http->ocmd = strdup(http->cmd);
2184 if (http->ocmd == NULL)
2186 log_error(LOG_LEVEL_FATAL, "Out of memory copying HTTP request line");
2189 enlist(csp->headers, http->cmd);
2191 /* Append the previously read headers */
2192 list_append_list_unique(csp->headers, headers);
2193 destroy_list(headers);
2195 err = sed(csp, FILTER_CLIENT_HEADERS);
2196 if (JB_ERR_OK != err)
2198 assert(err == JB_ERR_PARSE);
2199 log_error(LOG_LEVEL_FATAL, "Failed to parse client headers");
2201 csp->flags |= CSP_FLAG_CLIENT_HEADER_PARSING_DONE;
2203 /* Check request line for rewrites. */
2204 if ((NULL == csp->headers->first->str)
2205 || (strcmp(http->cmd, csp->headers->first->str) &&
2206 (JB_ERR_OK != change_request_destination(csp))))
2209 * A header filter broke the request line - bail out.
2211 write_socket(csp->cfd, MESSED_UP_REQUEST_RESPONSE, strlen(MESSED_UP_REQUEST_RESPONSE));
2212 /* XXX: Use correct size */
2213 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"Invalid request generated\" 500 0", csp->ip_addr_str);
2214 log_error(LOG_LEVEL_ERROR, "Invalid request line after applying header filters.");
2216 free_http_request(http);
2220 /* decide how to route the HTTP request */
2221 fwd = forward_url(csp, http);
2224 log_error(LOG_LEVEL_FATAL, "gateway spec is NULL!?!? This can't happen!");
2225 /* Never get here - LOG_LEVEL_FATAL causes program exit */
2229 * build the http request to send to the server
2230 * we have to do one of the following:
2232 * create = use the original HTTP request to create a new
2233 * HTTP request that has either the path component
2234 * without the http://domainspec (w/path) or the
2235 * full orininal URL (w/url)
2236 * Note that the path and/or the HTTP version may
2237 * have been altered by now.
2239 * connect = Open a socket to the host:port of the server
2240 * and short-circuit server and client socket.
2242 * pass = Pass the request unchanged if forwarding a CONNECT
2243 * request to a parent proxy. Note that we'll be sending
2244 * the CFAIL message ourselves if connecting to the parent
2245 * fails, but we won't send a CSUCCEED message if it works,
2246 * since that would result in a double message (ours and the
2247 * parent's). After sending the request to the parent, we simply
2250 * here's the matrix:
2253 * +--------+--------+
2255 * 0 | create | connect|
2257 * Forwarding +--------+--------+
2259 * 1 | create | pass |
2261 * +--------+--------+
2265 if (http->ssl && connect_port_is_forbidden(csp))
2267 const char *acceptable_connect_ports =
2268 csp->action->string[ACTION_STRING_LIMIT_CONNECT];
2269 assert(NULL != acceptable_connect_ports);
2270 log_error(LOG_LEVEL_INFO, "Request from %s marked for blocking. "
2271 "limit-connect{%s} doesn't allow CONNECT requests to port %d.",
2272 csp->ip_addr_str, acceptable_connect_ports, csp->http->port);
2273 csp->action->flags |= ACTION_BLOCK;
2279 freez(csp->headers->first->str);
2280 build_request_line(csp, fwd, &csp->headers->first->str);
2283 hdr = list_to_text(csp->headers);
2286 /* FIXME Should handle error properly */
2287 log_error(LOG_LEVEL_FATAL, "Out of memory parsing client header");
2291 * We have a request. Check if one of the crunchers wants it.
2293 if (crunch_response_triggered(csp, crunchers_all))
2296 * Yes. The client got the crunch response
2297 * and we are done here after cleaning up.
2300 list_remove_all(csp->headers);
2306 * The headers can't be removed earlier because
2307 * they were still needed for the referrer check
2308 * in case of CGI crunches.
2310 * XXX: Would it be worth to move the referrer check
2311 * into client_referrer() and set a flag if it's trusted?
2313 list_remove_all(csp->headers);
2315 log_error(LOG_LEVEL_GPC, "%s%s", http->hostport, http->path);
2317 if (fwd->forward_host)
2319 log_error(LOG_LEVEL_CONNECT, "via %s:%d to: %s",
2320 fwd->forward_host, fwd->forward_port, http->hostport);
2324 log_error(LOG_LEVEL_CONNECT, "to %s", http->hostport);
2327 /* here we connect to the server, gateway, or the forwarder */
2329 while ( (csp->sfd = forwarded_connect(fwd, http, csp))
2330 && (errno == EINVAL) && (forwarded_connect_retries++ < max_forwarded_connect_retries))
2332 log_error(LOG_LEVEL_ERROR, "failed request #%u to connect to %s. Trying again.",
2333 forwarded_connect_retries, http->hostport);
2336 if (csp->sfd == JB_INVALID_SOCKET)
2338 if (fwd->type != SOCKS_NONE)
2341 rsp = error_response(csp, "forwarding-failed", errno);
2343 else if (errno == EINVAL)
2345 rsp = error_response(csp, "no-such-domain", errno);
2349 rsp = error_response(csp, "connect-failed", errno);
2350 log_error(LOG_LEVEL_CONNECT, "connect to: %s failed: %E",
2355 /* Write the answer to the client */
2358 send_crunch_response(csp, rsp);
2365 if (fwd->forward_host || (http->ssl == 0))
2367 /* write the client's (modified) header to the server
2368 * (along with anything else that may be in the buffer)
2371 if (write_socket(csp->sfd, hdr, strlen(hdr))
2372 || (flush_socket(csp->sfd, csp->iob) < 0))
2374 log_error(LOG_LEVEL_CONNECT, "write header to: %s failed: %E",
2377 rsp = error_response(csp, "connect-failed", errno);
2381 send_crunch_response(csp, rsp);
2391 * We're running an SSL tunnel and we're not forwarding,
2392 * so just send the "connect succeeded" message to the
2393 * client, flush the rest, and get out of the way.
2395 if (write_socket(csp->cfd, CSUCCEED, strlen(CSUCCEED)))
2403 log_error(LOG_LEVEL_CONNECT, "to %s successful", http->hostport);
2405 /* we're finished with the client's header */
2408 maxfd = ( csp->cfd > csp->sfd ) ? csp->cfd : csp->sfd;
2410 /* pass data between the client and server
2411 * until one or the other shuts down the connection.
2420 * FD_ZERO here seems to point to an errant macro which crashes.
2421 * So do this by hand for now...
2423 memset(&rfds,0x00,sizeof(fd_set));
2427 FD_SET(csp->cfd, &rfds);
2428 FD_SET(csp->sfd, &rfds);
2430 n = select((int)maxfd+1, &rfds, NULL, NULL, NULL);
2434 log_error(LOG_LEVEL_ERROR, "select() failed!: %E");
2438 /* this is the body of the browser's request
2439 * just read it and write it.
2442 if (FD_ISSET(csp->cfd, &rfds))
2444 len = read_socket(csp->cfd, buf, sizeof(buf) - 1);
2448 break; /* "game over, man" */
2451 if (write_socket(csp->sfd, buf, (size_t)len))
2453 log_error(LOG_LEVEL_ERROR, "write to: %s failed: %E", http->host);
2460 * The server wants to talk. It could be the header or the body.
2461 * If `hdr' is null, then it's the header otherwise it's the body.
2462 * FIXME: Does `hdr' really mean `host'? No.
2466 if (FD_ISSET(csp->sfd, &rfds))
2469 len = read_socket(csp->sfd, buf, sizeof(buf) - 1);
2473 log_error(LOG_LEVEL_ERROR, "read from: %s failed: %E", http->host);
2475 if (http->ssl && (fwd->forward_host == NULL))
2478 * Just hang up. We already confirmed the client's CONNECT
2479 * request with status code 200 and unencrypted content is
2480 * no longer welcome.
2482 log_error(LOG_LEVEL_ERROR,
2483 "CONNECT already confirmed. Unable to tell the client about the problem.");
2486 else if (byte_count)
2489 * Just hang up. We already transmitted the original headers
2490 * and parts of the original content and therefore missed the
2491 * chance to send an error message (without risking data corruption).
2493 * XXX: we could retry with a fancy range request here.
2495 log_error(LOG_LEVEL_ERROR, "Already forwarded the original headers. "
2496 "Unable to tell the client about the problem.");
2500 rsp = error_response(csp, "connect-failed", errno);
2504 send_crunch_response(csp, rsp);
2510 /* Add a trailing zero. This lets filter_popups
2511 * use string operations.
2515 /* Normally, this would indicate that we've read
2516 * as much as the server has sent us and we can
2517 * close the client connection. However, Microsoft
2518 * in its wisdom has released IIS/5 with a bug that
2519 * prevents it from sending the trailing \r\n in
2520 * a 302 redirect header (and possibly other headers).
2521 * To work around this if we've haven't parsed
2522 * a full header we'll append a trailing \r\n
2523 * and see if this now generates a valid one.
2525 * This hack shouldn't have any impacts. If we've
2526 * already transmitted the header or if this is a
2527 * SSL connection, then we won't bother with this
2528 * hack. So we only work on partially received
2529 * headers. If we append a \r\n and this still
2530 * doesn't generate a valid header, then we won't
2531 * transmit anything to the client.
2536 if (server_body || http->ssl)
2539 * If we have been buffering up the document,
2540 * now is the time to apply content modification
2541 * and send the result to the client.
2545 p = execute_content_filter(csp, content_filter);
2547 * If the content filter fails, use the original
2548 * buffer and length.
2549 * (see p != NULL ? p : csp->iob->cur below)
2553 csp->content_length = (size_t)(csp->iob->eod - csp->iob->cur);
2556 if (JB_ERR_OK != update_server_headers(csp))
2558 log_error(LOG_LEVEL_FATAL,
2559 "Failed to update server headers. after filtering.");
2562 hdr = list_to_text(csp->headers);
2565 /* FIXME Should handle error properly */
2566 log_error(LOG_LEVEL_FATAL, "Out of memory parsing server header");
2569 if (write_socket(csp->cfd, hdr, strlen(hdr))
2570 || write_socket(csp->cfd, p != NULL ? p : csp->iob->cur, csp->content_length))
2572 log_error(LOG_LEVEL_ERROR, "write modified content to client failed: %E");
2582 break; /* "game over, man" */
2586 * This is NOT the body, so
2587 * Let's pretend the server just sent us a blank line.
2589 snprintf(buf, sizeof(buf), "\r\n");
2590 len = (int)strlen(buf);
2593 * Now, let the normal header parsing algorithm below do its
2594 * job. If it fails, we'll exit instead of continuing.
2601 * If this is an SSL connection or we're in the body
2602 * of the server document, just write it to the client,
2603 * unless we need to buffer the body for later content-filtering
2606 if (server_body || http->ssl)
2611 * If there is no memory left for buffering the content, or the buffer limit
2612 * has been reached, switch to non-filtering mode, i.e. make & write the
2613 * header, flush the iob and buf, and get out of the way.
2615 if (add_to_iob(csp, buf, len))
2620 log_error(LOG_LEVEL_INFO, "Flushing header and buffers. Stepping back from filtering.");
2622 hdr = list_to_text(csp->headers);
2626 * Memory is too tight to even generate the header.
2627 * Send our static "Out-of-memory" page.
2629 log_error(LOG_LEVEL_ERROR, "Out of memory while trying to flush.");
2630 rsp = cgi_error_memory();
2631 send_crunch_response(csp, rsp);
2635 hdrlen = strlen(hdr);
2637 if (write_socket(csp->cfd, hdr, hdrlen)
2638 || ((flushed = flush_socket(csp->cfd, csp->iob)) < 0)
2639 || (write_socket(csp->cfd, buf, (size_t)len)))
2641 log_error(LOG_LEVEL_CONNECT, "Flush header and buffers to client failed: %E");
2648 * Reset the byte_count to the amount of bytes
2649 * we just flushed. len will be added a few lines below,
2650 * hdrlen doesn't matter for LOG_LEVEL_CLF.
2652 byte_count = (size_t)flushed;
2654 content_filter = NULL;
2660 if (write_socket(csp->cfd, buf, (size_t)len))
2662 log_error(LOG_LEVEL_ERROR, "write to client failed: %E");
2666 byte_count += (size_t)len;
2671 /* we're still looking for the end of the
2672 * server's header ... (does that make header
2673 * parsing an "out of body experience" ?
2677 * buffer up the data we just read. If that fails,
2678 * there's little we can do but send our static
2679 * out-of-memory page.
2681 if (add_to_iob(csp, buf, len))
2683 log_error(LOG_LEVEL_ERROR, "Out of memory while looking for end of server headers.");
2684 rsp = cgi_error_memory();
2685 send_crunch_response(csp, rsp);
2690 /* Convert iob into something sed() can digest */
2691 if (JB_ERR_PARSE == get_server_headers(csp))
2695 /* Well, we tried our MS IIS/5
2696 * hack and it didn't work.
2697 * The header is incomplete
2698 * and there isn't anything
2699 * we can do about it.
2701 log_error(LOG_LEVEL_INFO,
2702 "MS IIS5 hack didn't produce valid headers.");
2707 /* Since we have to wait for
2708 * more from the server before
2709 * we can parse the headers
2710 * we just continue here.
2716 /* Did we actually get anything? */
2717 if (NULL == csp->headers->first)
2719 log_error(LOG_LEVEL_ERROR, "Empty server or forwarder response.");
2720 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" 502 0", csp->ip_addr_str, http->cmd);
2721 write_socket(csp->cfd, NO_SERVER_DATA_RESPONSE, strlen(NO_SERVER_DATA_RESPONSE));
2722 free_http_request(http);
2726 assert(csp->headers->first->str);
2728 if (strncmpic(csp->headers->first->str, "HTTP", 4))
2731 * It doesn't look like a HTTP response:
2732 * tell the client and log the problem.
2734 if (strlen(csp->headers->first->str) > 30)
2736 csp->headers->first->str[30] = '\0';
2738 log_error(LOG_LEVEL_ERROR,
2739 "Invalid server or forwarder response. Starts with: %s",
2740 csp->headers->first->str);
2741 log_error(LOG_LEVEL_CLF,
2742 "%s - - [%T] \"%s\" 502 0", csp->ip_addr_str, http->cmd);
2743 write_socket(csp->cfd, INVALID_SERVER_HEADERS_RESPONSE,
2744 strlen(INVALID_SERVER_HEADERS_RESPONSE));
2745 free_http_request(http);
2750 * We have now received the entire server header,
2751 * filter it and send the result to the client
2753 if (JB_ERR_OK != sed(csp, FILTER_SERVER_HEADERS))
2755 log_error(LOG_LEVEL_FATAL, "Failed to parse server headers.");
2757 hdr = list_to_text(csp->headers);
2760 /* FIXME Should handle error properly */
2761 log_error(LOG_LEVEL_FATAL, "Out of memory parsing server header");
2764 if (crunch_response_triggered(csp, crunchers_light))
2767 * One of the tags created by a server-header
2768 * tagger triggered a crunch. We already
2769 * delivered the crunch response to the client
2770 * and are done here after cleaning up.
2775 /* Buffer and pcrs filter this if appropriate. */
2777 if (!http->ssl) /* We talk plaintext */
2779 content_filter = get_filter_function(csp);
2782 * Only write if we're not buffering for content modification
2784 if (!content_filter)
2786 /* write the server's (modified) header to
2787 * the client (along with anything else that
2788 * may be in the buffer)
2791 if (write_socket(csp->cfd, hdr, strlen(hdr))
2792 || ((len = flush_socket(csp->cfd, csp->iob)) < 0))
2794 log_error(LOG_LEVEL_CONNECT, "write header to client failed: %E");
2796 /* the write failed, so don't bother
2797 * mentioning it to the client...
2798 * it probably can't hear us anyway.
2804 byte_count += (size_t)len;
2807 /* we're finished with the server's header */
2812 /* If this was a MS IIS/5 hack then it means
2813 * the server has already closed the
2814 * connection. Nothing more to read. Time
2819 log_error(LOG_LEVEL_INFO,
2820 "Closed server connection detected with MS IIS5 hack enabled.");
2827 return; /* huh? we should never get here */
2830 if (csp->content_length == 0)
2833 * If Privoxy didn't recalculate the
2834 * Content-Lenght, byte_count is still
2837 csp->content_length = byte_count;
2840 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" 200 %d",
2841 csp->ip_addr_str, http->ocmd, csp->content_length);
2845 /*********************************************************************
2849 * Description : This is little more than chat. We only "serve" to
2850 * to close any socket that chat may have opened.
2853 * 1 : csp = Current client state (buffers, headers, etc...)
2857 *********************************************************************/
2859 void serve(struct client_state *csp)
2860 #else /* ifndef AMIGA */
2861 static void serve(struct client_state *csp)
2862 #endif /* def AMIGA */
2865 close_socket(csp->cfd);
2867 if (csp->sfd != JB_INVALID_SOCKET)
2869 close_socket(csp->sfd);
2872 csp->flags &= ~CSP_FLAG_ACTIVE;
2878 /*********************************************************************
2880 * Function : server_thread
2882 * Description : We only exist to call `serve' in a threaded environment.
2885 * 1 : data = Current client state (buffers, headers, etc...)
2887 * Returns : Always 0.
2889 *********************************************************************/
2890 static int32 server_thread(void *data)
2892 serve((struct client_state *) data);
2899 #if !defined(_WIN32) || defined(_WIN_CONSOLE)
2900 /*********************************************************************
2904 * Description : Print usage info & exit.
2906 * Parameters : Pointer to argv[0] for identifying ourselves
2910 *********************************************************************/
2911 static void usage(const char *myname)
2913 printf("Privoxy version " VERSION " (" HOME_PAGE_URL ")\n"
2917 #endif /* defined(unix) */
2920 "[--no-daemon] [--pidfile pidfile] [--pre-chroot-nslookup hostname] [--user user[.group]] "
2921 #endif /* defined(unix) */
2922 "[--version] [configfile]\n"
2923 "Aborting\n", myname);
2928 #endif /* #if !defined(_WIN32) || defined(_WIN_CONSOLE) */
2931 /*********************************************************************
2933 * Function : initialize_mutexes
2935 * Description : Prepares mutexes if mutex support is available.
2939 * Returns : Void, exits in case of errors.
2941 *********************************************************************/
2942 static void initialize_mutexes(void)
2946 #ifdef FEATURE_PTHREAD
2948 * Prepare global mutex semaphores
2950 err = pthread_mutex_init(&log_mutex, 0);
2952 if (!err) err = pthread_mutex_init(&log_init_mutex, 0);
2955 * XXX: The assumptions below are a bit naive
2956 * and can cause locks that aren't necessary.
2958 * For example older FreeBSD versions (< 6.x?)
2959 * have no gethostbyname_r, but gethostbyname is
2962 #if !defined(HAVE_GETHOSTBYADDR_R) || !defined(HAVE_GETHOSTBYNAME_R)
2963 if (!err) err = pthread_mutex_init(&resolver_mutex, 0);
2964 #endif /* !defined(HAVE_GETHOSTBYADDR_R) || !defined(HAVE_GETHOSTBYNAME_R) */
2966 * XXX: should we use a single mutex for
2967 * localtime() and gmtime() as well?
2969 #ifndef HAVE_GMTIME_R
2970 if (!err) err = pthread_mutex_init(&gmtime_mutex, 0);
2971 #endif /* ndef HAVE_GMTIME_R */
2973 #ifndef HAVE_LOCALTIME_R
2974 if (!err) err = pthread_mutex_init(&localtime_mutex, 0);
2975 #endif /* ndef HAVE_GMTIME_R */
2978 if (!err) err = pthread_mutex_init(&rand_mutex, 0);
2979 #endif /* ndef HAVE_RANDOM */
2980 #endif /* FEATURE_PTHREAD */
2983 * TODO: mutex support for mingw32 would be swell.
2988 printf("Fatal error. Mutex initialization failed: %s.\n",
2997 /*********************************************************************
3001 * Description : Load the config file and start the listen loop.
3002 * This function is a lot more *sane* with the `load_config'
3003 * and `listen_loop' functions; although it stills does
3004 * a *little* too much for my taste.
3007 * 1 : argc = Number of parameters (including $0).
3008 * 2 : argv = Array of (char *)'s to the parameters.
3010 * Returns : 1 if : can't open config file, unrecognized directive,
3011 * stats requested in multi-thread mode, can't open the
3012 * log file, can't open the jar file, listen port is invalid,
3013 * any load fails, and can't bind port.
3015 * Else main never returns, the process must be signaled
3016 * to terminate execution. Or, on Windows, use the
3017 * "File", "Exit" menu option.
3019 *********************************************************************/
3021 int real_main(int argc, const char *argv[])
3023 int main(int argc, const char *argv[])
3027 unsigned int random_seed;
3029 struct passwd *pw = NULL;
3030 struct group *grp = NULL;
3033 char *pre_chroot_nslookup_to_load_resolver = NULL;
3040 #if !defined(_WIN32)
3048 * Parse the command line arguments
3050 * XXX: simply printing usage information in case of
3051 * invalid arguments isn't particularly user friendly.
3053 while (++argc_pos < argc)
3056 /* Check to see if the service must be installed or uninstalled */
3057 if (strncmp(argv[argc_pos], "--install", 9) == 0)
3059 const char *pName = argv[argc_pos] + 9;
3062 exit( (install_service(pName)) ? 0 : 1 );
3064 else if (strncmp(argv[argc_pos], "--uninstall", + 11) == 0)
3066 const char *pName = argv[argc_pos] + 11;
3069 exit((uninstall_service(pName)) ? 0 : 1);
3071 else if (strcmp(argv[argc_pos], "--service" ) == 0)
3073 bRunAsService = TRUE;
3074 w32_set_service_cwd();
3075 atexit(w32_service_exit_notify);
3078 #endif /* defined(_WIN32) */
3081 #if !defined(_WIN32) || defined(_WIN_CONSOLE)
3083 if (strcmp(argv[argc_pos], "--help") == 0)
3088 else if(strcmp(argv[argc_pos], "--version") == 0)
3090 printf("Privoxy version " VERSION " (" HOME_PAGE_URL ")\n");
3096 else if (strcmp(argv[argc_pos], "--no-daemon" ) == 0)
3101 else if (strcmp(argv[argc_pos], "--pidfile" ) == 0)
3103 if (++argc_pos == argc) usage(argv[0]);
3104 pidfile = strdup(argv[argc_pos]);
3107 else if (strcmp(argv[argc_pos], "--user" ) == 0)
3109 if (++argc_pos == argc) usage(argv[argc_pos]);
3111 if ((NULL != (p = strchr(argv[argc_pos], '.'))) && *(p + 1) != '0')
3114 if (NULL == (grp = getgrnam(p)))
3116 log_error(LOG_LEVEL_FATAL, "Group %s not found.", p);
3120 if (NULL == (pw = getpwnam(argv[argc_pos])))
3122 log_error(LOG_LEVEL_FATAL, "User %s not found.", argv[argc_pos]);
3125 if (p != NULL) *--p = '\0';
3128 else if (strcmp(argv[argc_pos], "--pre-chroot-nslookup" ) == 0)
3130 if (++argc_pos == argc) usage(argv[0]);
3131 pre_chroot_nslookup_to_load_resolver = strdup(argv[argc_pos]);
3134 else if (strcmp(argv[argc_pos], "--chroot" ) == 0)
3138 #endif /* defined(unix) */
3140 else if (argc_pos + 1 != argc)
3143 * This is neither the last command line
3144 * option, nor was it recognized before,
3145 * therefore it must be invalid.
3151 #endif /* defined(_WIN32) && !defined(_WIN_CONSOLE) */
3153 configfile = argv[argc_pos];
3156 } /* -END- while (more arguments) */
3159 if ( *configfile != '/' )
3161 char cwd[BUFFER_SIZE];
3163 size_t abs_file_size;
3165 /* make config-filename absolute here */
3166 if (NULL == getcwd(cwd, sizeof(cwd)))
3168 perror("failed to get current working directory");
3173 abs_file_size = strlen(cwd) + strlen(configfile) + 5;
3174 basedir = strdup(cwd);
3176 if (NULL == basedir ||
3177 NULL == (abs_file = malloc(abs_file_size)))
3179 perror("malloc failed");
3182 strlcpy(abs_file, basedir, abs_file_size);
3183 strlcat(abs_file, "/", abs_file_size );
3184 strlcat(abs_file, configfile, abs_file_size);
3185 configfile = abs_file;
3187 #endif /* defined unix */
3191 clients->next = NULL;
3193 /* XXX: factor out initialising after the next stable release. */
3196 #elif defined(_WIN32)
3200 /* Prepare mutexes if supported and necessary. */
3201 initialize_mutexes();
3203 /* Enable logging until further notice. */
3204 init_log_module(Argv[0]);
3206 random_seed = (unsigned int)time(NULL);
3208 srandom(random_seed);
3209 #elif defined (_WIN32)
3211 * See pick_from_range() in miscutil.c for details.
3213 log_error(LOG_LEVEL_INFO,
3214 "No thread-safe PRNG implemented for your platform. "
3215 "Using weak \'randomization\' factor which will "
3216 "limit the already questionable usefulness of "
3217 "header-time-randomizing actions (disabled by default).");
3220 #endif /* ifdef HAVE_RANDOM */
3223 * Unix signal handling
3225 * Catch the abort, interrupt and terminate signals for a graceful exit
3226 * Catch the hangup signal so the errlog can be reopened.
3227 * Ignore the broken pipe signals (FIXME: Why?)
3229 #if !defined(_WIN32) && !defined(__OS2__) && !defined(AMIGA)
3232 const int catched_signals[] = { SIGTERM, SIGINT, SIGHUP, 0 };
3233 const int ignored_signals[] = { SIGPIPE, 0 };
3235 for (idx = 0; catched_signals[idx] != 0; idx++)
3237 #ifdef sun /* FIXME: Is it safe to check for HAVE_SIGSET instead? */
3238 if (sigset(catched_signals[idx], sig_handler) == SIG_ERR)
3240 if (signal(catched_signals[idx], sig_handler) == SIG_ERR)
3241 #endif /* ifdef sun */
3243 log_error(LOG_LEVEL_FATAL, "Can't set signal-handler for signal %d: %E", catched_signals[idx]);
3247 for (idx = 0; ignored_signals[idx] != 0; idx++)
3249 if (signal(ignored_signals[idx], SIG_IGN) == SIG_ERR)
3251 log_error(LOG_LEVEL_FATAL, "Can't set ignore-handler for signal %d: %E", ignored_signals[idx]);
3256 #else /* ifdef _WIN32 */
3257 # ifdef _WIN_CONSOLE
3259 * We *are* in a windows console app.
3260 * Print a verbose messages about FAQ's and such
3262 printf("%s", win32_blurb);
3263 # endif /* def _WIN_CONSOLE */
3264 #endif /* def _WIN32 */
3267 /* Initialize the CGI subsystem */
3268 cgi_init_error_messages();
3271 * If runnig on unix and without the --nodaemon
3272 * option, become a daemon. I.e. fork, detach
3273 * from tty and get process group leadership
3286 if ( pid < 0 ) /* error */
3291 else if ( pid != 0 ) /* parent */
3296 * must check for errors
3297 * child died due to missing files aso
3300 wpid = waitpid( pid, &status, WNOHANG );
3309 /* Should be more portable, but not as well tested */
3314 #else /* ndef __FreeBSD__ */
3316 #endif /* ndef __FreeBSD__ */
3317 fd = open("/dev/tty", O_RDONLY);
3320 /* no error check here */
3321 ioctl( fd, TIOCNOTTY,0 );
3326 * stderr (fd 2) will be closed later on, when the
3327 * log file has been parsed.
3334 } /* -END- if (!no_daemon) */
3337 * As soon as we have written the PID file, we can switch
3338 * to the user and group ID indicated by the --user option
3344 if (setgid((NULL != grp) ? grp->gr_gid : pw->pw_gid))
3346 log_error(LOG_LEVEL_FATAL, "Cannot setgid(): Insufficient permissions.");
3352 log_error(LOG_LEVEL_FATAL, "Home directory for %s undefined", pw->pw_name);
3354 /* Read the time zone file from /etc before doing chroot. */
3356 if (NULL != pre_chroot_nslookup_to_load_resolver
3357 && '\0' != pre_chroot_nslookup_to_load_resolver[0])
3359 /* Initialize resolver library. */
3360 (void) resolve_hostname_to_ip(pre_chroot_nslookup_to_load_resolver);
3362 if (chroot(pw->pw_dir) < 0)
3364 log_error(LOG_LEVEL_FATAL, "Cannot chroot to %s", pw->pw_dir);
3368 log_error(LOG_LEVEL_FATAL, "Cannot chdir /");
3371 if (setuid(pw->pw_uid))
3373 log_error(LOG_LEVEL_FATAL, "Cannot setuid(): Insufficient permissions.");
3377 char putenv_dummy[64];
3379 strlcpy(putenv_dummy, "HOME=/", sizeof(putenv_dummy));
3380 if (putenv(putenv_dummy) != 0)
3382 log_error(LOG_LEVEL_FATAL, "Cannot putenv(): HOME");
3385 snprintf(putenv_dummy, sizeof(putenv_dummy), "USER=%s", pw->pw_name);
3386 if (putenv(putenv_dummy) != 0)
3388 log_error(LOG_LEVEL_FATAL, "Cannot putenv(): USER");
3394 log_error(LOG_LEVEL_FATAL, "Cannot chroot without --user argument.");
3397 #endif /* defined unix */
3400 /* This will be FALSE unless the command line specified --service
3404 /* Yup, so now we must attempt to establish a connection
3405 * with the service dispatcher. This will only work if this
3406 * process was launched by the service control manager to
3407 * actually run as a service. If this isn't the case, i've
3408 * known it take around 30 seconds or so for the call to return.
3411 /* The StartServiceCtrlDispatcher won't return until the service is stopping */
3412 if (w32_start_service_ctrl_dispatcher(w32ServiceDispatchTable))
3414 /* Service has run, and at this point is now being stopped, so just return */
3419 printf("Warning: Failed to connect to Service Control Dispatcher\nwhen starting as a service!\n");
3421 /* An error occurred. Usually it's because --service was wrongly specified
3422 * and we were unable to connect to the Service Control Dispatcher because
3423 * it wasn't expecting us and is therefore not listening.
3425 * For now, just continue below to call the listen_loop function.
3428 #endif /* def _WIN32 */
3438 /*********************************************************************
3440 * Function : bind_port_helper
3442 * Description : Bind the listen port. Handles logging, and aborts
3446 * 1 : config = Privoxy configuration. Specifies port
3449 * Returns : Port that was opened.
3451 *********************************************************************/
3452 static jb_socket bind_port_helper(struct configuration_spec * config)
3457 if (config->haddr == NULL)
3459 log_error(LOG_LEVEL_INFO, "Listening on port %d on all IP addresses",
3464 log_error(LOG_LEVEL_INFO, "Listening on port %d on IP address %s",
3465 config->hport, config->haddr);
3468 result = bind_port(config->haddr, config->hport, &bfd);
3475 log_error(LOG_LEVEL_FATAL, "can't bind to %s:%d: "
3476 "There may be another Privoxy or some other "
3477 "proxy running on port %d",
3478 (NULL != config->haddr) ? config->haddr : "INADDR_ANY",
3479 config->hport, config->hport);
3482 log_error(LOG_LEVEL_FATAL, "can't bind to %s:%d: "
3483 "The hostname is not resolvable",
3484 (NULL != config->haddr) ? config->haddr : "INADDR_ANY", config->hport);
3487 log_error(LOG_LEVEL_FATAL, "can't bind to %s:%d: because %E",
3488 (NULL != config->haddr) ? config->haddr : "INADDR_ANY", config->hport);
3491 /* shouldn't get here */
3492 return JB_INVALID_SOCKET;
3495 config->need_bind = 0;
3502 /* Without this simple workaround we get this compiler warning from _beginthread
3503 * warning C4028: formal parameter 1 different from declaration
3505 void w32_service_listen_loop(void *p)
3509 #endif /* def _WIN32 */
3512 /*********************************************************************
3514 * Function : listen_loop
3516 * Description : bind the listen port and enter a "FOREVER" listening loop.
3522 *********************************************************************/
3523 static void listen_loop(void)
3525 struct client_state *csp = NULL;
3527 struct configuration_spec * config;
3529 config = load_config();
3531 bfd = bind_port_helper(config);
3533 #ifdef FEATURE_GRACEFUL_TERMINATION
3534 while (!g_terminate)
3539 #if !defined(FEATURE_PTHREAD) && !defined(_WIN32) && !defined(__BEOS__) && !defined(AMIGA) && !defined(__OS2__)
3540 while (waitpid(-1, NULL, WNOHANG) > 0)
3542 /* zombie children */
3544 #endif /* !defined(FEATURE_PTHREAD) && !defined(_WIN32) && !defined(__BEOS__) && !defined(AMIGA) */
3547 * Free data that was used by died threads
3553 * Re-open the errlog after HUP signal
3555 if (received_hup_signal)
3557 init_error_log(Argv[0], config->logfile);
3558 received_hup_signal = 0;
3563 #ifdef FEATURE_COOKIE_JAR
3565 * Need a workaround here: we have to fclose() the jarfile, or we die because it's
3566 * already open. I think unload_configfile() is not being run, which should do
3567 * this work. Until that can get resolved, we'll use this workaround.
3571 if (csp->config->jar)
3573 fclose(csp->config->jar);
3574 csp->config->jar = NULL;
3576 #endif /* FEATURE_COOKIE_JAR */
3577 #endif /* __OS2__ */
3579 if ( NULL == (csp = (struct client_state *) zalloc(sizeof(*csp))) )
3581 log_error(LOG_LEVEL_FATAL, "malloc(%d) for csp failed: %E", sizeof(*csp));
3585 csp->flags |= CSP_FLAG_ACTIVE;
3586 csp->sfd = JB_INVALID_SOCKET;
3588 csp->config = config = load_config();
3590 if ( config->need_bind )
3593 * Since we were listening to the "old port", we will not see
3594 * a "listen" param change until the next IJB request. So, at
3595 * least 1 more request must be made for us to find the new
3596 * setting. I am simply closing the old socket and binding the
3599 * Which-ever is correct, we will serve 1 more page via the
3600 * old settings. This should probably be a "show-proxy-args"
3601 * request. This should not be a so common of an operation
3602 * that this will hurt people's feelings.
3607 bfd = bind_port_helper(config);
3610 log_error(LOG_LEVEL_CONNECT, "Listening for new connections ... ");
3612 if (!accept_connection(csp, bfd))
3614 log_error(LOG_LEVEL_CONNECT, "accept failed: %E");
3627 log_error(LOG_LEVEL_CONNECT, "accepted connection from %s", csp->ip_addr_str);
3630 #ifdef FEATURE_TOGGLE
3631 if (global_toggle_state)
3632 #endif /* def FEATURE_TOGGLE */
3634 csp->flags |= CSP_FLAG_TOGGLED_ON;
3637 if (run_loader(csp))
3639 log_error(LOG_LEVEL_FATAL, "a loader failed - must exit");
3640 /* Never get here - LOG_LEVEL_FATAL causes program exit */
3644 if (block_acl(NULL,csp))
3646 log_error(LOG_LEVEL_CONNECT, "Connection from %s dropped due to ACL", csp->ip_addr_str);
3647 close_socket(csp->cfd);
3651 #endif /* def FEATURE_ACL */
3653 /* add it to the list of clients */
3654 csp->next = clients->next;
3655 clients->next = csp;
3657 if (config->multi_threaded)
3661 /* this is a switch () statment in the C preprocessor - ugh */
3662 #undef SELECTED_ONE_OPTION
3664 /* Use Pthreads in preference to native code */
3665 #if defined(FEATURE_PTHREAD) && !defined(SELECTED_ONE_OPTION)
3666 #define SELECTED_ONE_OPTION
3668 pthread_t the_thread;
3669 pthread_attr_t attrs;
3671 pthread_attr_init(&attrs);
3672 pthread_attr_setdetachstate(&attrs, PTHREAD_CREATE_DETACHED);
3673 errno = pthread_create(&the_thread, &attrs,
3674 (void * (*)(void *))serve, csp);
3675 child_id = errno ? -1 : 0;
3676 pthread_attr_destroy(&attrs);
3680 #if defined(_WIN32) && !defined(_CYGWIN) && !defined(SELECTED_ONE_OPTION)
3681 #define SELECTED_ONE_OPTION
3682 child_id = _beginthread(
3683 (void (*)(void *))serve,
3688 #if defined(__OS2__) && !defined(SELECTED_ONE_OPTION)
3689 #define SELECTED_ONE_OPTION
3690 child_id = _beginthread(
3691 (void(* _Optlink)(void*))serve,
3697 #if defined(__BEOS__) && !defined(SELECTED_ONE_OPTION)
3698 #define SELECTED_ONE_OPTION
3700 thread_id tid = spawn_thread
3701 (server_thread, "server", B_NORMAL_PRIORITY, csp);
3703 if ((tid >= 0) && (resume_thread(tid) == B_OK))
3705 child_id = (int) tid;
3714 #if defined(AMIGA) && !defined(SELECTED_ONE_OPTION)
3715 #define SELECTED_ONE_OPTION
3716 csp->cfd = ReleaseSocket(csp->cfd, -1);
3719 child_id = (int)CreateNewProcTags(NP_Entry, (ULONG)server_thread,
3720 NP_Output, Output(),
3721 NP_CloseOutput, FALSE,
3722 NP_Name, (ULONG)"privoxy child",
3726 child_id = (int)CreateNewProcTags(NP_Entry, (ULONG)server_thread,
3727 NP_Output, Output(),
3728 NP_CloseOutput, FALSE,
3729 NP_Name, (ULONG)"privoxy child",
3730 NP_StackSize, 200*1024,
3736 ((struct Task *)child_id)->tc_UserData = csp;
3737 Signal((struct Task *)child_id, SIGF_SINGLE);
3742 #if !defined(SELECTED_ONE_OPTION)
3745 /* This block is only needed when using fork().
3746 * When using threads, the server thread was
3747 * created and run by the call to _beginthread().
3749 if (child_id == 0) /* child */
3752 #ifdef FEATURE_TOGGLE
3753 int inherited_toggle_state = global_toggle_state;
3754 #endif /* def FEATURE_TOGGLE */
3759 * If we've been toggled or we've blocked the request, tell Mom
3762 #ifdef FEATURE_TOGGLE
3763 if (inherited_toggle_state != global_toggle_state)
3765 rc |= RC_FLAG_TOGGLED;
3767 #endif /* def FEATURE_TOGGLE */
3769 #ifdef FEATURE_STATISTICS
3770 if (csp->flags & CSP_FLAG_REJECTED)
3772 rc |= RC_FLAG_BLOCKED;
3774 #endif /* ndef FEATURE_STATISTICS */
3778 else if (child_id > 0) /* parent */
3780 /* in a fork()'d environment, the parent's
3781 * copy of the client socket and the CSP
3785 #if !defined(_WIN32) && !defined(__CYGWIN__)
3787 wait( &child_status );
3790 * Evaluate child's return code: If the child has
3791 * - been toggled, toggle ourselves
3792 * - blocked its request, bump up the stats counter
3795 #ifdef FEATURE_TOGGLE
3796 if (WIFEXITED(child_status) && (WEXITSTATUS(child_status) & RC_FLAG_TOGGLED))
3798 global_toggle_state = !global_toggle_state;
3800 #endif /* def FEATURE_TOGGLE */
3802 #ifdef FEATURE_STATISTICS
3804 if (WIFEXITED(child_status) && (WEXITSTATUS(child_status) & RC_FLAG_BLOCKED))
3808 #endif /* def FEATURE_STATISTICS */
3810 #endif /* !defined(_WIN32) && defined(__CYGWIN__) */
3811 close_socket(csp->cfd);
3812 csp->flags &= ~CSP_FLAG_ACTIVE;
3816 #undef SELECTED_ONE_OPTION
3817 /* end of cpp switch () */
3819 if (child_id < 0) /* failed */
3821 char buf[BUFFER_SIZE];
3823 log_error(LOG_LEVEL_ERROR, "can't fork: %E");
3825 snprintf(buf , sizeof(buf), "Privoxy: can't fork: errno = %d", errno);
3827 write_socket(csp->cfd, buf, strlen(buf));
3828 close_socket(csp->cfd);
3829 csp->flags &= ~CSP_FLAG_ACTIVE;
3840 /* NOTREACHED unless FEATURE_GRACEFUL_TERMINATION is defined */
3842 /* Clean up. Aim: free all memory (no leaks) */
3843 #ifdef FEATURE_GRACEFUL_TERMINATION
3845 log_error(LOG_LEVEL_ERROR, "Graceful termination requested");
3847 unload_current_config_file();
3848 unload_current_actions_file();
3849 unload_current_re_filterfile();
3850 #ifdef FEATURE_TRUST
3851 unload_current_trust_file();
3854 if (config->multi_threaded)
3861 } while ((clients->next != NULL) && (--i > 0));
3865 log_error(LOG_LEVEL_ERROR, "Graceful termination failed - still some live clients after 1 minute wait.");
3876 #ifdef FEATURE_COOKIE_JAR
3877 if (NULL != config->jar)
3879 fclose(config->jar);
3883 #if defined(_WIN32) && !defined(_WIN_CONSOLE)
3884 /* Cleanup - remove taskbar icon etc. */
3889 #endif /* FEATURE_GRACEFUL_TERMINATION */