1 const char jcc_rcs[] = "$Id: jcc.c,v 1.176 2008/05/10 11:37:57 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 1.176 2008/05/10 11:37:57 fabiankeil
37 * - Instead of logging when the IIS5 hack is enabled, log when it fails.
38 * - Remove useless comment.
40 * Revision 1.175 2008/05/09 18:53:59 fabiankeil
41 * Fix comment grammar.
43 * Revision 1.174 2008/05/07 18:05:53 fabiankeil
44 * Remove the pointless buffer in client_protocol_is_unsupported().
46 * Revision 1.173 2008/05/06 15:09:00 fabiankeil
47 * Least-effort fix for bug #1821930 (reported by Lee):
48 * If the response doesn't look like HTTP,
49 * tell the client and log the problem.
51 * Revision 1.172 2008/04/16 16:38:21 fabiankeil
52 * Don't pass the whole csp structure to flush_socket()
53 * when it only needs a file descriptor and a buffer.
55 * Revision 1.171 2008/03/27 18:27:25 fabiankeil
56 * Remove kill-popups action.
58 * Revision 1.170 2008/03/06 16:33:46 fabiankeil
59 * If limit-connect isn't used, don't limit CONNECT requests to port 443.
61 * Revision 1.169 2008/03/04 18:30:39 fabiankeil
62 * Remove the treat-forbidden-connects-like-blocks action. We now
63 * use the "blocked" page for forbidden CONNECT requests by default.
65 * Revision 1.168 2008/03/02 12:25:25 fabiankeil
66 * Also use shiny new connect_port_is_forbidden() in jcc.c.
68 * Revision 1.167 2008/02/23 16:57:12 fabiankeil
69 * Rename url_actions() to get_url_actions() and let it
70 * use the standard parameter ordering.
72 * Revision 1.166 2008/02/23 16:33:43 fabiankeil
73 * Let forward_url() use the standard parameter ordering
74 * and mark its second parameter immutable.
76 * Revision 1.165 2008/02/02 19:36:56 fabiankeil
77 * Remove the "Listening ... for local connections only" log message.
78 * Whether or not remote connections are able to reach Privoxy is up
79 * to the operating system.
81 * Revision 1.164 2007/12/16 18:32:46 fabiankeil
82 * Prevent the log messages for CONNECT requests to unacceptable
83 * ports from printing the limit-connect argument as [null] if
84 * limit-connect hasn't been explicitly enabled.
86 * Revision 1.163 2007/12/13 01:47:11 david__schmidt
87 * Make sure all console-mode apps get a usage() instance
89 * Revision 1.162 2007/12/06 17:54:57 fabiankeil
90 * Reword NO_SERVER_DATA_RESPONSE to make it harder
91 * to misunderstand what the message is all about.
93 * Revision 1.161 2007/12/04 19:44:22 fabiankeil
94 * Unbreak trustfile which previously didn't work without
95 * FEATURE_TOGGLE. Fixes BR#1843585, reported by Lee.
97 * Revision 1.160 2007/11/29 18:00:29 fabiankeil
98 * Plug memory leak. Spotted by Valgrind, triggered by
99 * Privoxy-Regression-Test feeding proxyfuzz.py.
101 * Revision 1.159 2007/11/24 14:34:09 fabiankeil
102 * In the HTTP snipplets, refer to the client as client.
104 * Revision 1.158 2007/11/11 16:44:17 fabiankeil
105 * Emit a log message when activating the MS IIS5 hack.
107 * Revision 1.157 2007/11/03 17:34:49 fabiankeil
108 * Log the "weak randomization factor" warning only
109 * once for mingw32 and provide some more details.
111 * Revision 1.156 2007/11/01 18:20:58 fabiankeil
112 * Initialize log module after initializing mutexes, future
113 * deadlocks in that code should now work cross-platform.
115 * Revision 1.155 2007/10/23 20:12:45 fabiankeil
116 * Fix first CSUCCEED line to end in \r\n as required by RFC1945.
117 * Reported by Bert van Leeuwen in BR#1818808.
119 * Revision 1.154 2007/10/19 17:00:08 fabiankeil
120 * Downgrade "Flushing header and buffers" message to LOG_LEVEL_INFO.
122 * Revision 1.153 2007/10/14 14:12:41 fabiankeil
123 * When in daemon mode, close stderr after the configuration file has been
124 * parsed the first time. If logfile isn't set, stop logging. Fixes BR#897436.
126 * Revision 1.152 2007/10/04 18:03:34 fabiankeil
127 * - Fix a crash when parsing invalid requests whose first header
128 * is rejected by get_header(). Regression (re?)introduced
129 * in r1.143 by yours truly.
130 * - Move ACTION_VANILLA_WAFER handling into parsers.c's
131 * client_cookie_adder() to make sure send-vanilla-wafer can be
132 * controlled through tags (and thus regression-tested).
134 * Revision 1.151 2007/09/29 10:21:16 fabiankeil
135 * - Move get_filter_function() from jcc.c to filters.c
136 * so the filter functions can be static.
137 * - Don't bother filtering body-less responses.
139 * Revision 1.150 2007/09/28 16:39:29 fabiankeil
140 * Execute content filters through execute_content_filter().
142 * Revision 1.149 2007/09/04 15:08:48 fabiankeil
143 * Initialize req to NULL to make sure it's defined if the
144 * first read_socket() call fails. Reported by icmp30.
146 * Revision 1.148 2007/08/26 16:47:13 fabiankeil
147 * Add Stephen Gildea's --pre-chroot-nslookup patch [#1276666],
148 * extensive comments moved to user manual.
150 * Revision 1.147 2007/08/25 14:42:40 fabiankeil
151 * Don't crash if a broken header filter wiped out the request line.
153 * Revision 1.146 2007/08/20 17:09:32 fabiankeil
154 * Fix byte_count calculation in case of flushes
155 * and don't parse the server headers a second time.
157 * Revision 1.145 2007/08/19 13:13:31 fabiankeil
158 * - If there's a connection problem after we already forwarded
159 * parts of the original content, just hang up. Fixes BR#1776724.
160 * - Fix warnings about unused code on mingw32.
161 * - In case of flushes, calculate the byte count
162 * less incorrectly (I think).
164 * Revision 1.144 2007/08/11 14:43:22 fabiankeil
165 * Add some more prototypes for static functions.
167 * Revision 1.143 2007/08/05 13:58:19 fabiankeil
168 * Comment out request_contains_null_bytes() until it's used again.
170 * Revision 1.142 2007/08/05 13:50:26 fabiankeil
171 * #1763173 from Stefan Huehner: s@const static@static const@
172 * and declare some more functions static.
174 * Revision 1.141 2007/08/04 09:56:23 fabiankeil
175 * - Log rejected CONNECT requests with LOG_LEVEL_INFO
176 * and explain why they were rejected in the first place.
177 * - Fix the LOG_LEVEL_CLF message for crunches of unallowed
178 * CONNECT requests. The request line was missing.
179 * - Add two more XXX reminders as we don't have enough already.
181 * Revision 1.140 2007/07/21 11:51:36 fabiankeil
182 * As Hal noticed, checking dispatch_cgi() as the last cruncher
183 * looks like a bug if CGI requests are blocked unintentionally,
184 * so don't do it unless the user enabled the new config option
185 * "allow-cgi-request-crunching".
187 * Revision 1.139 2007/07/14 07:46:41 fabiankeil
188 * - Allow to rewrite the request destination behind the client's back.
189 * - Turn the weird-looking unconditional for loop that
190 * reads the client request into a conditional while loop.
191 * Move the stuff that only runs once out of the loop.
192 * - Move parts of chat(), server_content_type() and the
193 * necessary stuff to fix BR#1750917 into get_filter_function().
195 * Revision 1.138 2007/06/03 18:45:18 fabiankeil
196 * Temporary workaround for BR#1730105.
198 * Revision 1.137 2007/06/01 18:16:36 fabiankeil
199 * Use the same mutex for gethostbyname() and gethostbyaddr() to prevent
200 * deadlocks and crashes on OpenBSD and possibly other OS with neither
201 * gethostbyname_r() nor gethostaddr_r(). Closes BR#1729174.
202 * Thanks to Ralf Horstmann for report and solution.
204 * Revision 1.136 2007/06/01 16:41:11 fabiankeil
205 * Add forward-override{} to change the forwarding settings through
206 * action sections. This is mainly interesting to forward different
207 * clients differently (for example based on User-Agent or request
210 * Revision 1.135 2007/05/24 17:03:50 fabiankeil
211 * - Let usage() mention the --chroot parameter.
212 * - Use read_socket() consistently and always leave
213 * the last buffer byte alone, even in cases where
214 * null termination (currently) doesn't matter.
216 * Revision 1.134 2007/05/16 14:59:46 fabiankeil
217 * - Fix config file loading on Unix if no config file is specified.
218 * Since r1.97 Privoxy would always interpret the last argument as
219 * config file, even if it's a valid command line option.
220 * - Abort in case of unrecognized command line options. Closes #1719696.
221 * - Remove a bunch of unnecessary strcpy() calls (yay for c&p without thinking).
222 * - Replace the remaining strcpy() and strcat() calls with strlcpy() and strcat().
224 * Revision 1.133 2007/05/04 11:23:19 fabiankeil
225 * - Don't rerun crunchers that only depend on the request URL.
226 * - Don't count redirects and CGI requests as "blocked requests".
228 * Revision 1.132 2007/04/25 15:15:17 fabiankeil
229 * Support crunching based on tags created by server-header taggers.
231 * Revision 1.131 2007/04/22 13:24:50 fabiankeil
232 * Make HTTP snippets static (again). Add a Content-Type for those
233 * with content so the browser doesn't guess it based on the URL.
235 * Revision 1.130 2007/04/19 13:47:34 fabiankeil
236 * Move crunching and request line rebuilding out of chat().
238 * Revision 1.129 2007/04/15 16:39:20 fabiankeil
239 * Introduce tags as alternative way to specify which
240 * actions apply to a request. At the moment tags can be
241 * created based on client and server headers.
243 * Revision 1.128 2007/03/25 16:55:54 fabiankeil
244 * Don't CLF-log CONNECT requests twice.
246 * Revision 1.127 2007/03/20 13:53:17 fabiankeil
247 * Log the source address for ACL-related connection drops.
249 * Revision 1.126 2007/03/17 15:20:05 fabiankeil
250 * New config option: enforce-blocks.
252 * Revision 1.125 2007/03/09 14:12:00 fabiankeil
253 * - Move null byte check into separate function.
254 * - Don't confuse the client with error pages
255 * if a CONNECT request was already confirmed.
257 * Revision 1.124 2007/02/23 14:59:54 fabiankeil
258 * Speed up NULL byte escaping and only log the complete
259 * NULL byte requests with header debugging enabled.
261 * Revision 1.123 2007/02/21 18:42:10 fabiankeil
262 * Answer requests that contain NULL bytes with
263 * a custom response instead of waiting for more
264 * data until the client eventually hangs up.
266 * Revision 1.122 2007/02/07 11:12:02 fabiankeil
267 * - Move delivery and logging of crunched responses
268 * from chat() into send_crunch_response().
269 * - Display the reason for generating http_responses.
270 * - Log the content length for LOG_LEVEL_CLF correctly
271 * (still incorrect for some fixed responses).
272 * - Reword an incorrect comment about
273 * treat-forbidden-connects-like-blocks violating
275 * - Add some log messages.
277 * Revision 1.121 2007/01/27 10:52:56 fabiankeil
278 * Move mutex initialization into separate
279 * function and exit in case of errors.
281 * Revision 1.120 2007/01/26 14:18:42 fabiankeil
282 * - Start to reduce chat()'s line count and move
283 * parts of it into separate functions.
284 * - Add "HTTP/1.1 100 Continue" hack for BR 756734.
286 * Revision 1.119 2007/01/25 14:02:30 fabiankeil
287 * - Add Proxy-Agent header to HTTP snippets that are
288 * supposed to reach HTTP clients only.
289 * - Made a few CONNECT log messages more descriptive.
290 * - Catch completely empty server responses (as seen
291 * with Tor's fake ".noconnect" top level domain).
292 * - Use shiny new "forwarding-failed" template for socks errors.
294 * Revision 1.118 2007/01/07 07:43:43 joergs
295 * AmigaOS4 support added.
297 * Revision 1.117 2006/12/31 17:56:37 fabiankeil
298 * Added config option accept-intercepted-requests
299 * and disabled it by default.
301 * Revision 1.116 2006/12/29 19:08:22 fabiankeil
302 * Reverted parts of my last commit
303 * to keep error handling working.
305 * Revision 1.115 2006/12/29 17:38:57 fabiankeil
306 * Fixed gcc43 conversion warnings.
308 * Revision 1.114 2006/12/27 18:52:02 fabiankeil
309 * Fix -pedantic ISO C warning about converting
310 * from function pointer to object pointer.
312 * Revision 1.113 2006/12/26 17:38:50 fabiankeil
313 * Silence compiler warning I introduced with my last commit.
315 * Revision 1.112 2006/12/26 17:31:41 fabiankeil
316 * Mutex protect rand() if POSIX threading
317 * is used, warn the user if that's not possible
318 * and stop using it on _WIN32 where it could
321 * Revision 1.111 2006/12/23 16:15:06 fabiankeil
322 * Don't prevent core dumps by catching SIGABRT.
323 * It's rude and makes debugging unreasonable painful.
325 * Revision 1.110 2006/12/13 14:52:53 etresoft
326 * Fix build failure on MacOS X. Global symbols can be either static or extern, but not both.
328 * Revision 1.109 2006/12/06 19:41:40 fabiankeil
329 * Privoxy is now able to run as intercepting
330 * proxy in combination with any packet filter
331 * that does the port redirection. The destination
332 * is extracted from the "Host:" header which
333 * should be available for nearly all requests.
335 * Moved HTTP snipplets into jcc.c.
336 * Added error message for gopher proxy requests.
338 * Revision 1.108 2006/11/28 15:38:51 fabiankeil
339 * Only unlink the pidfile if it's actually used.
341 * Change order of interception checks to make
342 * it possible to block or redirect requests for
345 * Revision 1.107 2006/11/13 19:05:51 fabiankeil
346 * Make pthread mutex locking more generic. Instead of
347 * checking for OSX and OpenBSD, check for FEATURE_PTHREAD
348 * and use mutex locking unless there is an _r function
349 * available. Better safe than sorry.
351 * Fixes "./configure --disable-pthread" and should result
352 * in less threading-related problems on pthread-using platforms,
353 * but it still doesn't fix BR#1122404.
355 * Revision 1.106 2006/11/06 19:58:23 fabiankeil
356 * Move pthread.h inclusion from jcc.c to jcc.h.
357 * Fixes build on x86-freebsd1 (FreeBSD 5.4-RELEASE).
359 * Revision 1.105 2006/11/06 14:26:02 fabiankeil
360 * Don't exit after receiving the second SIGHUP on Solaris.
362 * Fixes BR 1052235, but the same problem may exist on other
363 * systems. Once 3.0.6 is out we should use sigset()
364 * where available and see if it breaks anything.
366 * Revision 1.104 2006/09/23 13:26:38 roro
367 * Replace TABs by spaces in source code.
369 * Revision 1.103 2006/09/21 12:54:43 fabiankeil
370 * Fix +redirect{}. Didn't work with -fast-redirects.
372 * Revision 1.102 2006/09/06 13:03:04 fabiankeil
373 * Respond with 400 and a short text message
374 * if the client tries to use Privoxy as FTP proxy.
376 * Revision 1.101 2006/09/06 09:23:37 fabiankeil
377 * Make number of retries in case of forwarded-connect problems
378 * a config file option (forwarded-connect-retries) and use 0 as
381 * Revision 1.100 2006/09/03 19:42:59 fabiankeil
382 * Set random(3) seed.
384 * Revision 1.99 2006/09/02 15:36:42 fabiankeil
385 * Follow the OpenBSD port's lead and protect the resolve
386 * functions on OpenBSD as well.
388 * Revision 1.98 2006/08/24 11:01:34 fabiankeil
389 * --user fix. Only use the user as group if no group is specified.
390 * Solves BR 1492612. Thanks to Spinor S. and David Laight.
392 * Revision 1.97 2006/08/18 15:23:17 david__schmidt
393 * Windows service (re-)integration
397 * --install[:service_name]
398 * --uninstall[:service_name]
401 * They work as follows:
402 * --install will create a service for you and then terminate.
403 * By default the service name will be "privoxy" (without the quotes).
404 * However you can run multiple services if you wish, just by adding
405 * a colon and then a name (no spaces).
407 * --uninstall follows the exact same rules a --install.
409 * --service is used when the program is executed by the service
410 * control manager, and in normal circumstances would never be
411 * used as a command line argument.
413 * Revision 1.96 2006/08/15 20:12:36 david__schmidt
414 * Windows service integration
416 * Revision 1.95 2006/08/03 02:46:41 david__schmidt
417 * Incorporate Fabian Keil's patch work:
418 http://www.fabiankeil.de/sourcecode/privoxy/
420 * Revision 1.94 2006/07/18 14:48:46 david__schmidt
421 * Reorganizing the repository: swapping out what was HEAD (the old 3.1 branch)
422 * with what was really the latest development (the v_3_0_branch branch)
424 * Revision 1.92.2.16 2005/04/03 20:10:50 david__schmidt
425 * Thanks to Jindrich Makovicka for a race condition fix for the log
426 * file. The race condition remains for non-pthread implementations.
427 * Reference patch #1175720.
429 * Revision 1.92.2.15 2004/10/03 12:53:32 david__schmidt
430 * Add the ability to check jpeg images for invalid
431 * lengths of comment blocks. Defensive strategy
432 * against the exploit:
433 * Microsoft Security Bulletin MS04-028
434 * Buffer Overrun in JPEG Processing (GDI+) Could
435 * Allow Code Execution (833987)
436 * Enabled with +inspect-jpegs in actions files.
438 * Revision 1.92.2.14 2003/12/12 12:52:53 oes
439 * - Fixed usage info for non-unix platforms
440 * - Fixed small cmdline parsing bug
442 * Revision 1.92.2.13 2003/11/27 19:20:27 oes
443 * Diagnostics: Now preserve the returncode of pthread_create
444 * in errno. Closes BR #775721. Thanks to Geoffrey Hausheer.
446 * Revision 1.92.2.12 2003/07/11 11:34:19 oes
447 * No longer ignore SIGCHLD. Fixes bug #769381
449 * Revision 1.92.2.11 2003/05/14 12:32:02 oes
450 * Close jarfile on graceful exit, remove stray line
452 * Revision 1.92.2.10 2003/05/08 15:13:46 oes
453 * Cosmetics: Killed a warning, a typo and an allocation left at exit
455 * Revision 1.92.2.9 2003/04/03 15:08:42 oes
456 * No longer rely on non-POSIX.1 extensions of getcwd().
459 * Revision 1.92.2.8 2003/03/31 13:12:32 oes
460 * Replaced setenv() by posix-compliant putenv()
461 * Thanks to Neil McCalden (nmcc AT users.sf.net).
463 * Revision 1.92.2.7 2003/03/17 16:48:59 oes
464 * Added chroot ability, thanks to patch by Sviatoslav Sviridov
466 * Revision 1.92.2.6 2003/03/11 11:55:00 oes
467 * Clean-up and extension of improvements for forked mode:
468 * - Child's return code now consists of flags RC_FLAG_*
469 * - Reporting toggle to parent now properly #ifdef'ed
470 * - Children now report blocking to parent. This enables
471 * statistics in forked mode
473 * Revision 1.92.2.5 2003/03/10 23:45:32 oes
474 * Fixed bug #700381: Non-Threaded version now capable of being toggled.
475 * Children now report having been toggled through _exit(17), parents
476 * watch for that code and toggle themselves if found.
478 * Revision 1.92.2.4 2003/03/07 03:41:04 david__schmidt
479 * Wrapping all *_r functions (the non-_r versions of them) with
480 * mutex semaphores for OSX. Hopefully this will take care of all
481 * of those pesky crash reports.
483 * Revision 1.92.2.3 2003/02/28 12:53:06 oes
484 * Fixed two mostly harmless mem leaks
486 * Revision 1.92.2.2 2002/11/20 14:37:47 oes
487 * Fix: Head of global clients list now initialized to NULL
489 * Revision 1.92.2.1 2002/09/25 14:52:24 oes
490 * Added basic support for OPTIONS and TRACE HTTP methods:
491 * - New interceptor direct_response() added in chat().
492 * - sed() moved to earlier in the process, so that the
493 * Host: header is evaluated before actions and forwarding
496 * Revision 1.92 2002/05/08 16:00:46 oes
497 * Chat's buffer handling:
498 * - Fixed bug with unchecked out-of-mem conditions
499 * while reading client request & server headers
500 * - No longer predict if the buffer limit will be exceeded
501 * in the next read -- check add_to_iob's new
502 * return code. If buffer couldn't be extended
503 * (policy or out-of-mem) while
504 * - reading from client: abort
505 * - reading server headers: send error page
506 * - buffering server body for filter: flush,
507 * and if that fails: send error page
509 * Revision 1.91 2002/04/08 20:35:58 swa
512 * Revision 1.90 2002/04/02 14:57:28 oes
513 * Made sending wafers independent of FEATURE_COOKIE_JAR
515 * Revision 1.89 2002/03/31 17:18:59 jongfoster
516 * Win32 only: Enabling STRICT to fix a VC++ compile warning.
518 * Revision 1.88 2002/03/27 14:32:43 david__schmidt
519 * More compiler warning message maintenance
521 * Revision 1.87 2002/03/26 22:29:54 swa
522 * we have a new homepage!
524 * Revision 1.86 2002/03/25 17:04:55 david__schmidt
525 * Workaround for closing the jarfile before load_config() comes around again
527 * Revision 1.85 2002/03/24 15:23:33 jongfoster
530 * Revision 1.84 2002/03/24 13:25:43 swa
531 * name change related issues
533 * Revision 1.83 2002/03/16 23:54:06 jongfoster
534 * Adding graceful termination feature, to help look for memory leaks.
535 * If you enable this (which, by design, has to be done by hand
536 * editing config.h) and then go to http://i.j.b/die, then the program
537 * will exit cleanly after the *next* request. It should free all the
538 * memory that was used.
540 * Revision 1.82 2002/03/13 00:27:05 jongfoster
543 * Revision 1.81 2002/03/12 01:42:50 oes
544 * Introduced modular filters
546 * Revision 1.80 2002/03/11 22:07:05 david__schmidt
547 * OS/2 port maintenance:
548 * - Fixed EMX build - it had decayed a little
549 * - Fixed inexplicable crash during FD_ZERO - must be due to a bad macro.
550 * substituted a memset for now.
552 * Revision 1.79 2002/03/09 20:03:52 jongfoster
553 * - Making various functions return int rather than size_t.
554 * (Undoing a recent change). Since size_t is unsigned on
555 * Windows, functions like read_socket that return -1 on
556 * error cannot return a size_t.
558 * THIS WAS A MAJOR BUG - it caused frequent, unpredictable
559 * crashes, and also frequently caused JB to jump to 100%
560 * CPU and stay there. (Because it thought it had just
561 * read ((unsigned)-1) == 4Gb of data...)
563 * - The signature of write_socket has changed, it now simply
564 * returns success=0/failure=nonzero.
566 * - Trying to get rid of a few warnings --with-debug on
567 * Windows, I've introduced a new type "jb_socket". This is
568 * used for the socket file descriptors. On Windows, this
569 * is SOCKET (a typedef for unsigned). Everywhere else, it's
570 * an int. The error value can't be -1 any more, so it's
571 * now JB_INVALID_SOCKET (which is -1 on UNIX, and in
572 * Windows it maps to the #define INVALID_SOCKET.)
574 * - The signature of bind_port has changed.
576 * Revision 1.78 2002/03/08 21:35:04 oes
577 * Added optional group supplement to --user option. Will now use default group of user if no group given
579 * Revision 1.77 2002/03/07 03:52:06 oes
580 * - Fixed compiler warnings etc
581 * - Improved handling of failed DNS lookups
583 * Revision 1.76 2002/03/06 22:54:35 jongfoster
584 * Automated function-comment nitpicking.
586 * Revision 1.75 2002/03/06 10:02:19 oes
587 * Fixed stupid bug when --user was not given
589 * Revision 1.74 2002/03/06 00:49:31 jongfoster
590 * Fixing warning on Windows
591 * Making #ifdefs that refer to the same variable consistently
592 * use #ifdef unix rather than mixing #ifdef unix & #ifndef OS2
594 * Revision 1.73 2002/03/05 23:57:30 hal9
595 * Stray character 's' on line 1618 was breaking build.
597 * Revision 1.72 2002/03/05 21:33:45 david__schmidt
598 * - Re-enable OS/2 building after new parms were added
599 * - Fix false out of memory report when resolving CGI templates when no IP
600 * address is available of failed attempt (a la no such domain)
602 * Revision 1.71 2002/03/05 18:13:56 oes
603 * Added --user option
605 * Revision 1.70 2002/03/05 04:52:42 oes
606 * Deleted non-errlog debugging code
608 * Revision 1.69 2002/03/04 23:50:00 jongfoster
609 * Splitting off bind_port() call into bind_port_helper(), with
612 * Revision 1.68 2002/03/04 20:17:32 oes
615 * Revision 1.67 2002/03/04 18:18:57 oes
616 * - Removed _DEBUG mode
617 * - Cleand up cmdline parsing
618 * - Introduced --no-daemon, --pidfile options
619 * - Cleaned up signal handling:
620 * - Terminate cleanly on INT, TERM and ABRT
621 * - Schedule logfile for re-opening on HUP
622 * - Ignore CHLD and PIPE
623 * - Leave the rest with their default handlers
624 * - Uniform handler registration
625 * - Added usage() function
626 * - Played styleguide police
628 * Revision 1.66 2002/03/03 15:06:55 oes
629 * Re-enabled automatic config reloading
631 * Revision 1.65 2002/03/03 14:49:11 oes
632 * Fixed CLF logging: Now uses client's original HTTP request
634 * Revision 1.64 2002/03/03 09:18:03 joergs
635 * Made jumbjuster work on AmigaOS again.
637 * Revision 1.63 2002/03/02 04:14:50 david__schmidt
638 * Clean up a little CRLF unpleasantness that suddenly appeared
640 * Revision 1.62 2002/02/20 23:17:23 jongfoster
641 * Detecting some out-of memory conditions and exiting with a log message.
643 * Revision 1.61 2002/01/17 21:01:52 jongfoster
644 * Moving all our URL and URL pattern parsing code to urlmatch.c.
646 * Revision 1.60 2001/12/30 14:07:32 steudten
647 * - Add signal handling (unix)
648 * - Add SIGHUP handler (unix)
649 * - Add creation of pidfile (unix)
650 * - Add action 'top' in rc file (RH)
651 * - Add entry 'SIGNALS' to manpage
652 * - Add exit message to logfile (unix)
654 * Revision 1.59 2001/12/13 14:07:18 oes
655 * Fixed Bug: 503 error page now sent OK
657 * Revision 1.58 2001/11/30 23:37:24 jongfoster
658 * Renaming the Win32 config file to config.txt - this is almost the
659 * same as the corresponding UNIX name "config"
661 * Revision 1.57 2001/11/16 00:47:43 jongfoster
662 * Changing the tty-disconnection code to use setsid().
664 * Revision 1.56 2001/11/13 20:20:54 jongfoster
665 * Tabs->spaces, fixing a bug with missing {} around an if()
667 * Revision 1.55 2001/11/13 20:14:53 jongfoster
668 * Patch for FreeBSD setpgrp() as suggested by Alexander Lazic
670 * Revision 1.54 2001/11/07 00:03:14 steudten
671 * Give reliable return value if an error
672 * occurs not just 0 with new daemon mode.
674 * Revision 1.53 2001/11/05 21:41:43 steudten
675 * Add changes to be a real daemon just for unix os.
676 * (change cwd to /, detach from controlling tty, set
677 * process group and session leader to the own process.
679 * Add some fatal-error log message for failed malloc().
680 * Add '-d' if compiled with 'configure --with-debug' to
681 * enable debug output.
683 * Revision 1.52 2001/10/26 20:11:20 jongfoster
684 * Fixing type mismatch
686 * Revision 1.51 2001/10/26 17:38:28 oes
689 * Revision 1.50 2001/10/25 03:40:48 david__schmidt
690 * Change in porting tactics: OS/2's EMX porting layer doesn't allow multiple
691 * threads to call select() simultaneously. So, it's time to do a real, live,
692 * native OS/2 port. See defines for __EMX__ (the porting layer) vs. __OS2__
693 * (native). Both versions will work, but using __OS2__ offers multi-threading.
695 * Revision 1.49 2001/10/23 21:41:35 jongfoster
696 * Added call to initialize the (statically-allocated of course)
697 * "out of memory" CGI response.
699 * Revision 1.48 2001/10/10 19:56:46 jongfoster
700 * Moving some code that wasn't cookie-related out of an #ifdef
703 * Revision 1.47 2001/10/10 16:44:36 oes
704 * Added CONNECT destination port limitation check
706 * Revision 1.46 2001/10/08 15:17:41 oes
707 * Re-enabled SSL forwarding
709 * Revision 1.45 2001/10/07 15:42:11 oes
710 * Replaced 6 boolean members of csp with one bitmap (csp->flags)
712 * Moved downgrading of the HTTP version from parse_http_request to
713 * chat(), since we can't decide if it is necessary before we have
714 * determined the actions for the URL. The HTTP command is now
715 * *always* re-built so the repairs need no longer be special-cased.
717 * filter_popups now gets a csp pointer so it can raise the new
718 * CSP_FLAG_MODIFIED flag.
722 * Added configurable size limit for the IOB. If the IOB grows so
723 * large that the next read would exceed the limit, the header
724 * is generated, and the header & unfiltered buffer are flushed
725 * to the client. Chat then continues in non-buffering,
726 * non-filtering body mode.
728 * Revision 1.44 2001/10/02 18:13:57 oes
731 * Revision 1.43 2001/10/02 15:32:13 oes
732 * Moved generation of hdr
734 * Revision 1.42 2001/09/21 23:02:02 david__schmidt
735 * Cleaning up 2 compiler warnings on OS/2.
737 * Revision 1.41 2001/09/16 17:05:14 jongfoster
738 * Removing unused #include showarg.h
740 * Revision 1.40 2001/09/16 15:41:45 jongfoster
741 * Fixing signed/unsigned comparison warning.
743 * Revision 1.39 2001/09/16 13:21:27 jongfoster
744 * Changes to use new list functions.
746 * Revision 1.38 2001/09/16 13:01:46 jongfoster
747 * Removing redundant function call that zeroed zalloc()'d memory.
749 * Revision 1.37 2001/09/10 11:12:24 oes
750 * Deleted unused variable
752 * Revision 1.36 2001/09/10 10:56:15 oes
753 * Silenced compiler warnings
755 * Revision 1.35 2001/07/31 14:44:22 oes
756 * Deleted unused size parameter from filter_popups()
758 * Revision 1.34 2001/07/30 22:08:36 jongfoster
759 * Tidying up #defines:
760 * - All feature #defines are now of the form FEATURE_xxx
761 * - Permanently turned off WIN_GUI_EDIT
762 * - Permanently turned on WEBDAV and SPLIT_PROXY_ARGS
764 * Revision 1.33 2001/07/29 19:32:00 jongfoster
765 * Renaming _main() [mingw32 only] to real_main(), for ANSI compliance.
767 * Revision 1.32 2001/07/29 18:47:05 jongfoster
768 * Adding missing #include "loadcfg.h"
770 * Revision 1.31 2001/07/29 12:17:48 oes
771 * Applied pthread fix by Paul Lieverse
773 * Revision 1.30 2001/07/25 22:57:13 jongfoster
774 * __BEOS__ no longer overrides FEATURE_PTHREAD.
775 * This is because FEATURE_PTHREAD will soon be widely used, so I
776 * want to keep it simple.
778 * Revision 1.29 2001/07/24 12:47:06 oes
779 * Applied BeOS support update by Eugenia
781 * Revision 1.28 2001/07/23 13:26:12 oes
782 * Fixed bug in popup-killing for the first read that caused binary garbage to be sent between headers and body
784 * Revision 1.27 2001/07/19 19:09:47 haroon
785 * - Added code to take care of the situation where while processing the first
786 * server response (which includes the server header), after finding the end
787 * of the headers we were not looking past the end of the headers for
788 * content modification. I enabled it for filter_popups.
789 * Someone else should look to see if other similar operations should be
790 * done to the discarded portion of the buffer.
792 * Note 2001/07/20: No, the other content modification mechanisms will process
793 * the whole iob later anyway. --oes
795 * Revision 1.26 2001/07/18 12:31:36 oes
798 * Revision 1.25 2001/07/15 19:43:49 jongfoster
799 * Supports POSIX threads.
800 * Also removed some unused #includes.
802 * Revision 1.24 2001/07/13 14:00:40 oes
803 * - Generic content modification scheme:
804 * Each feature has its own applicability flag that is set
805 * from csp->action->flags.
806 * Replaced the "filtering" int flag , by a function pointer
807 * "content_filter" to the function that will do the content
808 * modification. If it is != NULL, the document will be buffered
809 * and processed through *content_filter, which must set
810 * csp->content_length and return a modified copy of the body
811 * or return NULL (on failiure).
812 * - Changed csp->is_text to the more generic bitmap csp->content_type
813 * which can currently take the valued CT_TEXT or CT_GIF
815 * - Removed all #ifdef PCRS
817 * Revision 1.23 2001/07/02 02:28:25 iwanttokeepanon
818 * Added "#ifdef ACL_FILES" conditional compilation to line 1291 to exclude
819 * the `block_acl' call. This prevents a compilation error when the user
820 * does not wish to use the "ACL" feature.
822 * Revision 1.22 2001/06/29 21:45:41 oes
823 * Indentation, CRLF->LF, Tab-> Space
825 * Revision 1.21 2001/06/29 13:29:36 oes
826 * - Cleaned up, improved comments
827 * - Unified all possible interceptors (CGI,
828 * block, trust, fast_redirect) in one
829 * place, with one (CGI) answer generation
830 * mechansim. Much clearer now.
831 * - Removed the GIF image generation, which
832 * is now done in filters.c:block_url()
833 * - Made error conditions like domain lookup
834 * failiure or (various) problems while talking
835 * to the server use cgi.c:error_response()
836 * instead of generating HTML/HTTP in chat() (yuck!)
837 * - Removed logentry from cancelled commit
839 * Revision 1.20 2001/06/09 10:55:28 jongfoster
840 * Changing BUFSIZ ==> BUFFER_SIZE
842 * Revision 1.19 2001/06/07 23:12:52 jongfoster
843 * Replacing function pointer in struct gateway with a directly
844 * called function forwarded_connect().
845 * Replacing struct gateway with struct forward_spec
847 * Revision 1.18 2001/06/03 19:12:16 oes
848 * introduced new cgi handling
850 * Revision 1.17 2001/06/01 20:07:23 jongfoster
851 * Now uses action +image-blocker{} rather than config->tinygif
853 * Revision 1.16 2001/06/01 18:49:17 jongfoster
854 * Replaced "list_share" with "list" - the tiny memory gain was not
855 * worth the extra complexity.
857 * Revision 1.15 2001/05/31 21:24:47 jongfoster
858 * Changed "permission" to "action" throughout.
859 * Removed DEFAULT_USER_AGENT - it must now be specified manually.
860 * Moved vanilla wafer check into chat(), since we must now
861 * decide whether or not to add it based on the URL.
863 * Revision 1.14 2001/05/29 20:14:01 joergs
864 * AmigaOS bugfix: PCRS needs a lot of stack, stacksize for child threads
867 * Revision 1.13 2001/05/29 09:50:24 jongfoster
868 * Unified blocklist/imagelist/permissionslist.
869 * File format is still under discussion, but the internal changes
872 * Also modified interceptor behaviour:
873 * - We now intercept all URLs beginning with one of the following
874 * prefixes (and *only* these prefixes):
876 * * http://ijbswa.sf.net/config/
877 * * http://ijbswa.sourceforge.net/config/
878 * - New interceptors "home page" - go to http://i.j.b/ to see it.
879 * - Internal changes so that intercepted and fast redirect pages
880 * are not replaced with an image.
881 * - Interceptors now have the option to send a binary page direct
882 * to the client. (i.e. ijb-send-banner uses this)
883 * - Implemented show-url-info interceptor. (Which is why I needed
884 * the above interceptors changes - a typical URL is
885 * "http://i.j.b/show-url-info?url=www.somesite.com/banner.gif".
886 * The previous mechanism would not have intercepted that, and
887 * if it had been intercepted then it then it would have replaced
890 * Revision 1.12 2001/05/27 22:17:04 oes
892 * - re_process_buffer no longer writes the modified buffer
893 * to the client, which was very ugly. It now returns the
894 * buffer, which it is then written by chat.
896 * - content_length now adjusts the Content-Length: header
897 * for modified documents rather than crunch()ing it.
898 * (Length info in csp->content_length, which is 0 for
899 * unmodified documents)
901 * - For this to work, sed() is called twice when filtering.
903 * Revision 1.11 2001/05/26 17:27:53 jongfoster
904 * Added support for CLF and fixed LOG_LEVEL_LOG.
905 * Also did CRLF->LF fix of my previous patch.
907 * Revision 1.10 2001/05/26 15:26:15 jongfoster
908 * ACL feature now provides more security by immediately dropping
909 * connections from untrusted hosts.
911 * Revision 1.9 2001/05/26 00:28:36 jongfoster
912 * Automatic reloading of config file.
913 * Removed obsolete SIGHUP support (Unix) and Reload menu option (Win32).
914 * Most of the global variables have been moved to a new
915 * struct configuration_spec, accessed through csp->config->globalname
916 * Most of the globals remaining are used by the Win32 GUI.
918 * Revision 1.8 2001/05/25 22:43:18 jongfoster
919 * Fixing minor memory leak and buffer overflow.
921 * Revision 1.7 2001/05/25 22:34:30 jongfoster
924 * Revision 1.6 2001/05/23 00:13:58 joergs
925 * AmigaOS support fixed.
927 * Revision 1.5 2001/05/22 18:46:04 oes
929 * - Enabled filtering banners by size rather than URL
930 * by adding patterns that replace all standard banner
931 * sizes with the "Junkbuster" gif to the re_filterfile
933 * - Enabled filtering WebBugs by providing a pattern
934 * which kills all 1x1 images
936 * - Added support for PCRE_UNGREEDY behaviour to pcrs,
937 * which is selected by the (nonstandard and therefore
938 * capital) letter 'U' in the option string.
939 * It causes the quantifiers to be ungreedy by default.
940 * Appending a ? turns back to greedy (!).
942 * - Added a new interceptor ijb-send-banner, which
943 * sends back the "Junkbuster" gif. Without imagelist or
944 * MSIE detection support, or if tinygif = 1, or the
945 * URL isn't recognized as an imageurl, a lame HTML
946 * explanation is sent instead.
948 * - Added new feature, which permits blocking remote
949 * script redirects and firing back a local redirect
951 * The feature is conditionally compiled, i.e. it
952 * can be disabled with --disable-fast-redirects,
953 * plus it must be activated by a "fast-redirects"
954 * line in the config file, has its own log level
955 * and of course wants to be displayed by show-proxy-args
956 * Note: Boy, all the #ifdefs in 1001 locations and
957 * all the fumbling with configure.in and acconfig.h
958 * were *way* more work than the feature itself :-(
960 * - Because a generic redirect template was needed for
961 * this, tinygif = 3 now uses the same.
963 * - Moved GIFs, and other static HTTP response templates
968 * - Removed some >400 CRs again (Jon, you really worked
971 * Revision 1.4 2001/05/21 19:34:01 jongfoster
972 * Made failure to bind() a fatal error.
974 * Revision 1.3 2001/05/20 01:21:20 jongfoster
975 * Version 2.9.4 checkin.
976 * - Merged popupfile and cookiefile, and added control over PCRS
977 * filtering, in new "permissionsfile".
978 * - Implemented LOG_LEVEL_FATAL, so that if there is a configuration
979 * file error you now get a message box (in the Win32 GUI) rather
980 * than the program exiting with no explanation.
981 * - Made killpopup use the PCRS MIME-type checking and HTTP-header
983 * - Removed tabs from "config"
984 * - Moved duplicated url parsing code in "loaders.c" to a new funcition.
985 * - Bumped up version number.
987 * Revision 1.2 2001/05/17 22:34:44 oes
988 * - Added hint on GIF char array generation to jcc.c
989 * - Cleaned CRLF's from the sources and related files
990 * - Repaired logging for REF and FRC
992 * Revision 1.1.1.1 2001/05/15 13:58:56 oes
993 * Initial import of version 2.9.3 source tree
996 *********************************************************************/
1002 #include <sys/types.h>
1011 # ifndef FEATURE_PTHREAD
1015 # include <windows.h>
1016 # include <process.h>
1017 # endif /* ndef FEATURE_PTHREAD */
1020 # ifndef _WIN_CONSOLE
1021 # include "w32log.h"
1022 # endif /* ndef _WIN_CONSOLE */
1023 # include "w32svrapi.h"
1025 #else /* ifndef _WIN32 */
1027 # if !defined (__OS2__)
1028 # include <unistd.h>
1029 # include <sys/wait.h>
1030 # endif /* ndef __OS2__ */
1031 # include <sys/time.h>
1032 # include <sys/stat.h>
1033 # include <sys/ioctl.h>
1036 #include <sys/termios.h>
1044 # include <signal.h>
1047 # include <socket.h> /* BeOS has select() for sockets only. */
1048 # include <OS.h> /* declarations for threads and stuff. */
1051 # if defined(__EMX__) || defined(__OS2__)
1052 # include <sys/select.h> /* OS/2/EMX needs a little help with select */
1057 #define bzero(B,N) memset(B,0x00,n)
1061 # include <select.h>
1066 #include "project.h"
1069 #include "filters.h"
1070 #include "loaders.h"
1071 #include "parsers.h"
1072 #include "miscutil.h"
1074 #include "jbsockets.h"
1075 #include "gateway.h"
1076 #include "actions.h"
1078 #include "loadcfg.h"
1079 #include "urlmatch.h"
1081 const char jcc_h_rcs[] = JCC_H_VERSION;
1082 const char project_h_rcs[] = PROJECT_H_VERSION;
1085 struct client_state clients[1];
1086 struct file_list files[1];
1088 #ifdef FEATURE_STATISTICS
1089 int urls_read = 0; /* total nr of urls read inc rejected */
1090 int urls_rejected = 0; /* total nr of urls rejected */
1091 #endif /* def FEATURE_STATISTICS */
1093 #ifdef FEATURE_GRACEFUL_TERMINATION
1094 int g_terminate = 0;
1097 #if !defined(_WIN32) && !defined(__OS2__) && !defined(AMIGA)
1098 static void sig_handler(int the_signal);
1100 static int client_protocol_is_unsupported(const struct client_state *csp, char *req);
1101 static jb_err get_request_destination_elsewhere(struct client_state *csp, struct list *headers);
1102 static jb_err get_server_headers(struct client_state *csp);
1103 static const char *crunch_reason(const struct http_response *rsp);
1104 static void send_crunch_response(struct client_state *csp, struct http_response *rsp);
1106 * static int request_contains_null_bytes(const struct client_state *csp, char *buf, int len);
1108 static void build_request_line(struct client_state *csp, const struct forward_spec *fwd, char **request_line);
1109 static jb_err change_request_destination(struct client_state *csp);
1110 static void chat(struct client_state *csp);
1111 static void serve(struct client_state *csp);
1112 #if !defined(_WIN32) || defined(_WIN_CONSOLE)
1113 static void usage(const char *myname);
1115 static void initialize_mutexes(void);
1116 static jb_socket bind_port_helper(struct configuration_spec *config);
1117 static void listen_loop(void);
1120 void serve(struct client_state *csp);
1121 #else /* ifndef AMIGA */
1122 static void serve(struct client_state *csp);
1123 #endif /* def AMIGA */
1126 static int32 server_thread(void *data);
1127 #endif /* def __BEOS__ */
1130 #define sleep(N) Sleep(((N) * 1000))
1134 #define sleep(N) DosSleep(((N) * 100))
1137 #ifdef FEATURE_PTHREAD
1138 pthread_mutex_t log_mutex;
1139 pthread_mutex_t log_init_mutex;
1141 #if !defined(HAVE_GETHOSTBYADDR_R) || !defined(HAVE_GETHOSTBYNAME_R)
1142 pthread_mutex_t resolver_mutex;
1143 #endif /* !defined(HAVE_GETHOSTBYADDR_R) || !defined(HAVE_GETHOSTBYNAME_R) */
1145 #ifndef HAVE_GMTIME_R
1146 pthread_mutex_t gmtime_mutex;
1147 #endif /* ndef HAVE_GMTIME_R */
1149 #ifndef HAVE_LOCALTIME_R
1150 pthread_mutex_t localtime_mutex;
1151 #endif /* ndef HAVE_GMTIME_R */
1154 pthread_mutex_t rand_mutex;
1155 #endif /* ndef HAVE_RANDOM */
1157 #endif /* FEATURE_PTHREAD */
1159 #if defined(unix) || defined(__EMX__)
1160 const char *basedir = NULL;
1161 const char *pidfile = NULL;
1162 int received_hup_signal = 0;
1163 #endif /* defined unix */
1165 /* HTTP snipplets. */
1166 static const char CSUCCEED[] =
1167 "HTTP/1.0 200 Connection established\r\n"
1168 "Proxy-Agent: Privoxy/" VERSION "\r\n\r\n";
1170 static const char CHEADER[] =
1171 "HTTP/1.0 400 Invalid header received from client\r\n"
1172 "Proxy-Agent: Privoxy " VERSION "\r\n"
1173 "Content-Type: text/plain\r\n"
1174 "Connection: close\r\n\r\n"
1175 "Invalid header received from client.\r\n";
1177 static const char FTP_RESPONSE[] =
1178 "HTTP/1.0 400 Invalid request received from client\r\n"
1179 "Content-Type: text/plain\r\n"
1180 "Connection: close\r\n\r\n"
1181 "Invalid request. Privoxy doesn't support FTP.\r\n";
1183 static const char GOPHER_RESPONSE[] =
1184 "HTTP/1.0 400 Invalid request received from client\r\n"
1185 "Content-Type: text/plain\r\n"
1186 "Connection: close\r\n\r\n"
1187 "Invalid request. Privoxy doesn't support gopher.\r\n";
1189 /* XXX: should be a template */
1190 static const char MISSING_DESTINATION_RESPONSE[] =
1191 "HTTP/1.0 400 Bad request received from client\r\n"
1192 "Proxy-Agent: Privoxy " VERSION "\r\n"
1193 "Content-Type: text/plain\r\n"
1194 "Connection: close\r\n\r\n"
1195 "Bad request. Privoxy was unable to extract the destination.\r\n";
1197 /* XXX: should be a template */
1198 static const char NO_SERVER_DATA_RESPONSE[] =
1199 "HTTP/1.0 502 Server or forwarder response empty\r\n"
1200 "Proxy-Agent: Privoxy " VERSION "\r\n"
1201 "Content-Type: text/plain\r\n"
1202 "Connection: close\r\n\r\n"
1203 "Empty server or forwarder response.\r\n"
1204 "The connection has been closed but Privoxy didn't receive any data.\r\n";
1206 /* XXX: should be a template */
1207 static const char INVALID_SERVER_HEADERS_RESPONSE[] =
1208 "HTTP/1.0 502 Server or forwarder response invalid\r\n"
1209 "Proxy-Agent: Privoxy " VERSION "\r\n"
1210 "Content-Type: text/plain\r\n"
1211 "Connection: close\r\n\r\n"
1212 "Bad response. The server or forwarder response doesn't look like HTTP.\r\n";
1215 /* XXX: should be a template */
1216 static const char NULL_BYTE_RESPONSE[] =
1217 "HTTP/1.0 400 Bad request received from client\r\n"
1218 "Proxy-Agent: Privoxy " VERSION "\r\n"
1219 "Content-Type: text/plain\r\n"
1220 "Connection: close\r\n\r\n"
1221 "Bad request. Null byte(s) before end of request.\r\n";
1224 /* XXX: should be a template */
1225 static const char MESSED_UP_REQUEST_RESPONSE[] =
1226 "HTTP/1.0 400 Malformed request after rewriting\r\n"
1227 "Proxy-Agent: Privoxy " VERSION "\r\n"
1228 "Content-Type: text/plain\r\n"
1229 "Connection: close\r\n\r\n"
1230 "Bad request. Messed up with header filters.\r\n";
1232 /* A function to crunch a response */
1233 typedef struct http_response *(*crunch_func_ptr)(struct client_state *);
1235 /* Crunch function flags */
1236 #define CF_NO_FLAGS 0
1237 /* Cruncher applies to forced requests as well */
1238 #define CF_IGNORE_FORCE 1
1239 /* Crunched requests are counted for the block statistics */
1240 #define CF_COUNT_AS_REJECT 2
1242 /* A crunch function and its flags */
1245 const crunch_func_ptr cruncher;
1249 static int crunch_response_triggered(struct client_state *csp, const struct cruncher crunchers[]);
1251 /* Complete list of cruncher functions */
1252 static const struct cruncher crunchers_all[] = {
1253 { direct_response, CF_COUNT_AS_REJECT|CF_IGNORE_FORCE},
1254 { block_url, CF_COUNT_AS_REJECT },
1255 #ifdef FEATURE_TRUST
1256 { trust_url, CF_COUNT_AS_REJECT },
1257 #endif /* def FEATURE_TRUST */
1258 { redirect_url, CF_NO_FLAGS },
1259 { dispatch_cgi, CF_IGNORE_FORCE},
1263 /* Light version, used after tags are applied */
1264 static const struct cruncher crunchers_light[] = {
1265 { block_url, CF_COUNT_AS_REJECT },
1266 { redirect_url, CF_NO_FLAGS },
1271 #if !defined(_WIN32) && !defined(__OS2__) && !defined(AMIGA)
1272 /*********************************************************************
1274 * Function : sig_handler
1276 * Description : Signal handler for different signals.
1277 * Exit gracefully on TERM and INT
1278 * or set a flag that will cause the errlog
1279 * to be reopened by the main thread on HUP.
1282 * 1 : the_signal = the signal cause this function to call
1286 *********************************************************************/
1287 static void sig_handler(int the_signal)
1293 log_error(LOG_LEVEL_INFO, "exiting by signal %d .. bye", the_signal);
1304 received_hup_signal = 1;
1309 * We shouldn't be here, unless we catch signals
1310 * in main() that we can't handle here!
1312 log_error(LOG_LEVEL_FATAL, "sig_handler: exiting on unexpected signal %d", the_signal);
1320 /*********************************************************************
1322 * Function : client_protocol_is_unsupported
1324 * Description : Checks if the client used a known unsupported
1325 * protocol and deals with it by sending an error
1329 * 1 : csp = Current client state (buffers, headers, etc...)
1330 * 2 : req = the first request line send by the client
1332 * Returns : TRUE if an error response has been generated, or
1333 * FALSE if the request doesn't look invalid.
1335 *********************************************************************/
1336 static int client_protocol_is_unsupported(const struct client_state *csp, char *req)
1339 * If it's a FTP or gopher request, we don't support it.
1341 * These checks are better than nothing, but they might
1342 * not work in all configurations and some clients might
1343 * have problems digesting the answer.
1345 * They should, however, never cause more problems than
1346 * Privoxy's old behaviour (returning the misleading HTML
1349 * "Could not resolve http://(ftp|gopher)://example.org").
1351 if (!strncmpic(req, "GET ftp://", 10) || !strncmpic(req, "GET gopher://", 13))
1353 const char *response = NULL;
1354 const char *protocol = NULL;
1356 if (!strncmpic(req, "GET ftp://", 10))
1358 response = FTP_RESPONSE;
1363 response = GOPHER_RESPONSE;
1364 protocol = "GOPHER";
1366 log_error(LOG_LEVEL_ERROR,
1367 "%s tried to use Privoxy as %s proxy: %s",
1368 csp->ip_addr_str, protocol, req);
1369 log_error(LOG_LEVEL_CLF,
1370 "%s - - [%T] \"%s\" 400 0", csp->ip_addr_str, req);
1372 write_socket(csp->cfd, response, strlen(response));
1381 /*********************************************************************
1383 * Function : get_request_destination_elsewhere
1385 * Description : If the client's request was redirected into
1386 * Privoxy without the client's knowledge,
1387 * the request line lacks the destination host.
1389 * This function tries to get it elsewhere,
1390 * provided accept-intercepted-requests is enabled.
1392 * "Elsewhere" currently only means "Host: header",
1393 * but in the future we may ask the redirecting
1394 * packet filter to look the destination up.
1396 * If the destination stays unknown, an error
1397 * response is send to the client and headers
1398 * are freed so that chat() can return directly.
1401 * 1 : csp = Current client state (buffers, headers, etc...)
1402 * 2 : headers = a header list
1404 * Returns : JB_ERR_OK if the destination is now known, or
1405 * JB_ERR_PARSE if it isn't.
1407 *********************************************************************/
1408 static jb_err get_request_destination_elsewhere(struct client_state *csp, struct list *headers)
1412 if (!(csp->config->feature_flags & RUNTIME_FEATURE_ACCEPT_INTERCEPTED_REQUESTS))
1414 log_error(LOG_LEVEL_ERROR, "%s's request: \'%s\' is invalid."
1415 " Privoxy isn't configured to accept intercepted requests.",
1416 csp->ip_addr_str, csp->http->cmd);
1417 /* XXX: Use correct size */
1418 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" 400 0",
1419 csp->ip_addr_str, csp->http->cmd);
1421 write_socket(csp->cfd, CHEADER, strlen(CHEADER));
1422 destroy_list(headers);
1424 return JB_ERR_PARSE;
1426 else if (JB_ERR_OK == get_destination_from_headers(headers, csp->http))
1428 /* Split the domain we just got for pattern matching */
1429 init_domain_components(csp->http);
1435 /* We can't work without destination. Go spread the news.*/
1437 req = list_to_text(headers);
1439 /* XXX: Use correct size */
1440 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" 400 0",
1441 csp->ip_addr_str, csp->http->cmd);
1442 log_error(LOG_LEVEL_ERROR,
1443 "Privoxy was unable to get the destination for %s's request:\n%s\n%s",
1444 csp->ip_addr_str, csp->http->cmd, req);
1447 write_socket(csp->cfd, MISSING_DESTINATION_RESPONSE, strlen(MISSING_DESTINATION_RESPONSE));
1448 destroy_list(headers);
1450 return JB_ERR_PARSE;
1453 * TODO: If available, use PF's ioctl DIOCNATLOOK as last resort
1454 * to get the destination IP address, use it as host directly
1455 * or do a reverse DNS lookup first.
1460 /*********************************************************************
1462 * Function : get_server_headers
1464 * Description : Parses server headers in iob and fills them
1465 * into csp->headers so that they can later be
1469 * 1 : csp = Current client state (buffers, headers, etc...)
1471 * Returns : JB_ERR_OK if everything went fine, or
1472 * JB_ERR_PARSE if the headers were incomplete.
1474 *********************************************************************/
1475 static jb_err get_server_headers(struct client_state *csp)
1477 int continue_hack_in_da_house = 0;
1480 while (((header = get_header(csp)) != NULL) || continue_hack_in_da_house)
1485 * continue hack in da house. Ignore the ending of
1486 * this head and continue enlisting header lines.
1487 * The reason is described below.
1489 enlist(csp->headers, "");
1490 continue_hack_in_da_house = 0;
1493 else if (0 == strncmpic(header, "HTTP/1.1 100", 12))
1496 * It's a bodyless continue response, don't
1497 * stop header parsing after reaching it's end.
1499 * As a result Privoxy will concatenate the
1500 * next response's head and parse and deliver
1501 * the headers as if they belonged to one request.
1503 * The client will separate them because of the
1504 * empty line between them.
1506 * XXX: What we're doing here is clearly against
1507 * the intended purpose of the continue header,
1508 * and under some conditions (HTTP/1.0 client request)
1509 * it's a standard violation.
1511 * Anyway, "sort of against the spec" is preferable
1512 * to "always getting confused by Continue responses"
1513 * (Privoxy's behaviour before this hack was added)
1515 log_error(LOG_LEVEL_HEADER, "Continue hack in da house.");
1516 continue_hack_in_da_house = 1;
1518 else if (*header == '\0')
1521 * If the header is empty, but the Continue hack
1522 * isn't active, we can assume that we reached the
1523 * end of the buffer before we hit the end of the
1526 * Inform the caller an let it decide how to handle it.
1528 return JB_ERR_PARSE;
1531 if (JB_ERR_MEMORY == enlist(csp->headers, header))
1534 * XXX: Should we quit the request and return a
1535 * out of memory error page instead?
1537 log_error(LOG_LEVEL_ERROR,
1538 "Out of memory while enlisting server headers. %s lost.",
1548 /*********************************************************************
1550 * Function : crunch_reason
1552 * Description : Translates the crunch reason code into a string.
1555 * 1 : rsp = a http_response
1557 * Returns : A string with the crunch reason or an error description.
1559 *********************************************************************/
1560 static const char *crunch_reason(const struct http_response *rsp)
1562 char * reason = NULL;
1564 assert(rsp != NULL);
1567 return "Internal error while searching for crunch reason";
1570 switch (rsp->reason)
1572 case RSP_REASON_UNSUPPORTED:
1573 reason = "Unsupported HTTP feature";
1575 case RSP_REASON_BLOCKED:
1578 case RSP_REASON_UNTRUSTED:
1579 reason = "Untrusted";
1581 case RSP_REASON_REDIRECTED:
1582 reason = "Redirected";
1584 case RSP_REASON_CGI_CALL:
1585 reason = "CGI Call";
1587 case RSP_REASON_NO_SUCH_DOMAIN:
1588 reason = "DNS failure";
1590 case RSP_REASON_FORWARDING_FAILED:
1591 reason = "Forwarding failed";
1593 case RSP_REASON_CONNECT_FAILED:
1594 reason = "Connection failure";
1596 case RSP_REASON_OUT_OF_MEMORY:
1597 reason = "Out of memory (may mask other reasons)";
1600 reason = "No reason recorded";
1608 /*********************************************************************
1610 * Function : send_crunch_response
1612 * Description : Delivers already prepared response for
1613 * intercepted requests, logs the interception
1614 * and frees the response.
1617 * 1 : csp = Current client state (buffers, headers, etc...)
1618 * 1 : rsp = Fully prepared response. Will be freed on exit.
1620 * Returns : Nothing.
1622 *********************************************************************/
1623 static void send_crunch_response(struct client_state *csp, struct http_response *rsp)
1625 const struct http_request *http = csp->http;
1626 char status_code[4];
1628 assert(rsp != NULL);
1629 assert(rsp->head != NULL);
1634 * Not supposed to happen. If it does
1635 * anyway, treat it as an unknown error.
1637 cgi_error_unknown(csp, rsp, RSP_REASON_INTERNAL_ERROR);
1638 /* return code doesn't matter */
1643 /* If rsp is still NULL, we have serious internal problems. */
1644 log_error(LOG_LEVEL_FATAL,
1645 "NULL response in send_crunch_response and cgi_error_unknown failed as well.");
1649 * Extract the status code from the actual head
1650 * that was send to the client. It is the only
1651 * way to get it right for all requests, including
1652 * the fixed ones for out-of-memory problems.
1654 * A head starts like this: 'HTTP/1.1 200...'
1658 status_code[0] = rsp->head[9];
1659 status_code[1] = rsp->head[10];
1660 status_code[2] = rsp->head[11];
1661 status_code[3] = '\0';
1663 /* Write the answer to the client */
1664 if (write_socket(csp->cfd, rsp->head, rsp->head_length)
1665 || write_socket(csp->cfd, rsp->body, rsp->content_length))
1667 /* There is nothing we can do about it. */
1668 log_error(LOG_LEVEL_ERROR, "write to: %s failed: %E", csp->http->host);
1671 /* Log that the request was crunched and why. */
1672 log_error(LOG_LEVEL_GPC, "%s%s crunch! (%s)",
1673 http->hostport, http->path, crunch_reason(rsp));
1674 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" %s %d",
1675 csp->ip_addr_str, http->ocmd, status_code, rsp->content_length);
1677 /* Clean up and return */
1678 if (cgi_error_memory() != rsp)
1680 free_http_response(rsp);
1687 /*********************************************************************
1689 * Function : request_contains_null_bytes
1691 * Description : Checks for NULL bytes in the request and sends
1692 * an error message to the client if any were found.
1694 * XXX: currently not used, see comment in chat().
1697 * 1 : csp = Current client state (buffers, headers, etc...)
1698 * 2 : buf = Data from the client's request to check.
1699 * 3 : len = The data length.
1701 * Returns : TRUE if the request contained one or more NULL bytes, or
1704 *********************************************************************/
1705 static int request_contains_null_bytes(const struct client_state *csp, char *buf, int len)
1707 size_t c_len; /* Request lenght when treated as C string */
1709 c_len = strlen(buf);
1714 * Null byte(s) found. Log the request,
1715 * return an error response and hang up.
1717 size_t tmp_len = c_len;
1722 * Replace NULL byte(s) with '°' characters
1723 * so the request can be logged as string.
1724 * XXX: Is there a better replacement character?
1727 tmp_len += strlen(buf+tmp_len);
1728 } while (tmp_len < len);
1730 log_error(LOG_LEVEL_ERROR, "%s\'s request contains at least one NULL byte "
1731 "(length=%d, strlen=%d).", csp->ip_addr_str, len, c_len);
1732 log_error(LOG_LEVEL_HEADER,
1733 "Offending request data with NULL bytes turned into \'°\' characters: %s", buf);
1735 write_socket(csp->cfd, NULL_BYTE_RESPONSE, strlen(NULL_BYTE_RESPONSE));
1737 /* XXX: Log correct size */
1738 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"Invalid request\" 400 0", csp->ip_addr_str);
1748 /*********************************************************************
1750 * Function : crunch_response_triggered
1752 * Description : Checks if the request has to be crunched,
1753 * and delivers the crunch response if necessary.
1756 * 1 : csp = Current client state (buffers, headers, etc...)
1757 * 2 : crunchers = list of cruncher functions to run
1759 * Returns : TRUE if the request was answered with a crunch response
1762 *********************************************************************/
1763 static int crunch_response_triggered(struct client_state *csp, const struct cruncher crunchers[])
1765 struct http_response *rsp = NULL;
1766 const struct cruncher *c;
1769 * If CGI request crunching is disabled,
1770 * check the CGI dispatcher out of order to
1771 * prevent unintentional blocks or redirects.
1773 if (!(csp->config->feature_flags & RUNTIME_FEATURE_CGI_CRUNCHING)
1774 && (NULL != (rsp = dispatch_cgi(csp))))
1776 /* Deliver, log and free the interception response. */
1777 send_crunch_response(csp, rsp);
1781 for (c = crunchers; c->cruncher != NULL; c++)
1784 * Check the cruncher if either Privoxy is toggled
1785 * on and the request isn't forced, or if the cruncher
1786 * applies to forced requests as well.
1788 if (((csp->flags & CSP_FLAG_TOGGLED_ON) &&
1789 !(csp->flags & CSP_FLAG_FORCED)) ||
1790 (c->flags & CF_IGNORE_FORCE))
1792 rsp = c->cruncher(csp);
1795 /* Deliver, log and free the interception response. */
1796 send_crunch_response(csp, rsp);
1797 #ifdef FEATURE_STATISTICS
1798 if (c->flags & CF_COUNT_AS_REJECT)
1800 csp->flags |= CSP_FLAG_REJECTED;
1802 #endif /* def FEATURE_STATISTICS */
1813 /*********************************************************************
1815 * Function : build_request_line
1817 * Description : Builds the HTTP request line.
1819 * If a HTTP forwarder is used it expects the whole URL,
1820 * web servers only get the path.
1823 * 1 : csp = Current client state (buffers, headers, etc...)
1824 * 2 : fwd = The forwarding spec used for the request
1825 * XXX: Should use http->fwd instead.
1826 * 3 : request_line = The old request line which will be replaced.
1828 * Returns : Nothing. Terminates in case of memory problems.
1830 *********************************************************************/
1831 static void build_request_line(struct client_state *csp, const struct forward_spec *fwd, char **request_line)
1833 struct http_request *http = csp->http;
1835 assert(http->ssl == 0);
1838 * Downgrade http version from 1.1 to 1.0
1839 * if +downgrade action applies.
1841 if ( (csp->action->flags & ACTION_DOWNGRADE)
1842 && (!strcmpic(http->ver, "HTTP/1.1")))
1845 http->ver = strdup("HTTP/1.0");
1847 if (http->ver == NULL)
1849 log_error(LOG_LEVEL_FATAL, "Out of memory downgrading HTTP version");
1854 * Rebuild the request line.
1856 freez(*request_line);
1857 *request_line = strdup(http->gpc);
1858 string_append(request_line, " ");
1860 if (fwd->forward_host)
1862 string_append(request_line, http->url);
1866 string_append(request_line, http->path);
1868 string_append(request_line, " ");
1869 string_append(request_line, http->ver);
1871 if (*request_line == NULL)
1873 log_error(LOG_LEVEL_FATAL, "Out of memory writing HTTP command");
1875 log_error(LOG_LEVEL_HEADER, "New HTTP Request-Line: %s", *request_line);
1879 /*********************************************************************
1881 * Function : change_request_destination
1883 * Description : Parse a (rewritten) request line and regenerate
1884 * the http request data.
1887 * 1 : csp = Current client state (buffers, headers, etc...)
1889 * Returns : Forwards the parse_http_request() return code.
1890 * Terminates in case of memory problems.
1892 *********************************************************************/
1893 static jb_err change_request_destination(struct client_state *csp)
1895 struct http_request *http = csp->http;
1898 log_error(LOG_LEVEL_INFO, "Rewrite detected: %s", csp->headers->first->str);
1899 free_http_request(http);
1900 err = parse_http_request(csp->headers->first->str, http, csp);
1901 if (JB_ERR_OK != err)
1903 log_error(LOG_LEVEL_ERROR, "Couldn't parse rewritten request: %s.",
1904 jb_err_to_string(err));
1906 http->ocmd = strdup(http->cmd); /* XXX: ocmd is a misleading name */
1907 if (http->ocmd == NULL)
1909 log_error(LOG_LEVEL_FATAL, "Out of memory copying rewritten HTTP request line");
1916 /*********************************************************************
1920 * Description : Once a connection to the client has been accepted,
1921 * this function is called (via serve()) to handle the
1922 * main business of the communication. When this
1923 * function returns, the caller must close the client
1926 * FIXME: chat is nearly thousand lines long.
1930 * 1 : csp = Current client state (buffers, headers, etc...)
1932 * Returns : Nothing.
1934 *********************************************************************/
1935 static void chat(struct client_state *csp)
1937 char buf[BUFFER_SIZE];
1945 int ms_iis5_hack = 0;
1946 size_t byte_count = 0;
1947 int forwarded_connect_retries = 0;
1948 int max_forwarded_connect_retries = csp->config->forwarded_connect_retries;
1949 const struct forward_spec * fwd;
1950 struct http_request *http;
1951 int len; /* for buffer sizes (and negative error codes) */
1954 /* Function that does the content filtering for the current request */
1955 filter_function_ptr content_filter = NULL;
1957 /* Skeleton for HTTP response, if we should intercept the request */
1958 struct http_response *rsp;
1960 /* Temporary copy of the client's headers before they get enlisted in csp->headers */
1961 struct list header_list;
1962 struct list *headers = &header_list;
1966 memset(buf, 0, sizeof(buf));
1969 * Read the client's request. Note that since we're not using select() we
1970 * could get blocked here if a client connected, then didn't say anything!
1975 len = read_socket(csp->cfd, buf, sizeof(buf) - 1);
1977 if (len <= 0) break; /* error! */
1980 * If there is no memory left for buffering the
1981 * request, there is nothing we can do but hang up
1983 if (add_to_iob(csp, buf, len))
1988 req = get_header(csp);
1990 } while ((NULL != req) && ('\0' == *req));
1992 if ((NULL != req) && ('\0' != *req))
1994 /* Request received. Validate and parse it. */
1998 * XXX: Temporary disabled to prevent problems
1999 * with POST requests whose bodies are allowed to
2000 * contain NULL bytes. BR#1730105.
2002 * The main purpose of this check is to properly
2003 * log stuff like BitTorrent traffic and other junk
2004 * that hits public proxies. It's not required for
2005 * Privoxy to functions as those requests are discarded
2008 * It probably should be rewritten to only check
2009 * the head of the request. Another option would
2010 * be to let all POST requests pass, although that
2011 * may not be good enough.
2013 if (request_contains_null_bytes(csp, buf, len))
2015 /* NULL bytes found and dealt with, just hang up. */
2020 /* Does the request line look invalid? */
2021 if (client_protocol_is_unsupported(csp, req))
2024 * Yes. The request has already been
2025 * answered with a error response, the buffers
2026 * were freed and we're done with chatting.
2031 #ifdef FEATURE_FORCE_LOAD
2033 * If this request contains the FORCE_PREFIX and blocks
2034 * aren't enforced, get rid of it and set the force flag.
2036 if (strstr(req, FORCE_PREFIX))
2038 if (csp->config->feature_flags & RUNTIME_FEATURE_ENFORCE_BLOCKS)
2040 log_error(LOG_LEVEL_FORCE,
2041 "Ignored force prefix in request: \"%s\".", req);
2045 strclean(req, FORCE_PREFIX);
2046 log_error(LOG_LEVEL_FORCE, "Enforcing request: \"%s\".", req);
2047 csp->flags |= CSP_FLAG_FORCED;
2051 #endif /* def FEATURE_FORCE_LOAD */
2052 err = parse_http_request(req, http, csp);
2053 if (JB_ERR_OK != err)
2055 log_error(LOG_LEVEL_ERROR, "Couldn't parse request: %s.", jb_err_to_string(err));
2061 if (http->cmd == NULL)
2063 write_socket(csp->cfd, CHEADER, strlen(CHEADER));
2064 /* XXX: Use correct size */
2065 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"Invalid request\" 400 0", csp->ip_addr_str);
2066 log_error(LOG_LEVEL_ERROR, "Invalid header received from %s.", csp->ip_addr_str);
2068 free_http_request(http);
2072 /* grab the rest of the client's headers */
2076 p = get_header(csp);
2080 /* There are no additional headers to read. */
2087 * We didn't receive a complete header
2088 * line yet, get the rest of it.
2090 len = read_socket(csp->cfd, buf, sizeof(buf) - 1);
2093 log_error(LOG_LEVEL_ERROR, "read from client failed: %E");
2094 destroy_list(headers);
2098 if (add_to_iob(csp, buf, len))
2101 * If there is no memory left for buffering the
2102 * request, there is nothing we can do but hang up
2104 destroy_list(headers);
2111 * We were able to read a complete
2112 * header and can finaly enlist it.
2119 if (http->host == NULL)
2122 * If we still don't know the request destination,
2123 * the request is invalid or the client uses
2124 * Privoxy without its knowledge.
2126 if (JB_ERR_OK != get_request_destination_elsewhere(csp, headers))
2129 * Our attempts to get the request destination
2130 * elsewhere failed or Privoxy is configured
2131 * to only accept proxy requests.
2133 * An error response has already been send
2134 * and we're done here.
2141 * Determine the actions for this URL
2143 #ifdef FEATURE_TOGGLE
2144 if (!(csp->flags & CSP_FLAG_TOGGLED_ON))
2146 /* Most compatible set of actions (i.e. none) */
2147 init_current_action(csp->action);
2150 #endif /* ndef FEATURE_TOGGLE */
2152 get_url_actions(csp, http);
2156 * Save a copy of the original request for logging
2158 http->ocmd = strdup(http->cmd);
2160 if (http->ocmd == NULL)
2162 log_error(LOG_LEVEL_FATAL, "Out of memory copying HTTP request line");
2165 enlist(csp->headers, http->cmd);
2167 /* Append the previously read headers */
2168 list_append_list_unique(csp->headers, headers);
2169 destroy_list(headers);
2171 err = sed(client_patterns, add_client_headers, csp);
2172 if (JB_ERR_OK != err)
2174 assert(err == JB_ERR_PARSE);
2175 log_error(LOG_LEVEL_FATAL, "Failed to parse client headers");
2177 csp->flags |= CSP_FLAG_CLIENT_HEADER_PARSING_DONE;
2179 /* Check request line for rewrites. */
2180 if ((NULL == csp->headers->first->str)
2181 || (strcmp(http->cmd, csp->headers->first->str) &&
2182 (JB_ERR_OK != change_request_destination(csp))))
2185 * A header filter broke the request line - bail out.
2187 write_socket(csp->cfd, MESSED_UP_REQUEST_RESPONSE, strlen(MESSED_UP_REQUEST_RESPONSE));
2188 /* XXX: Use correct size */
2189 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"Invalid request generated\" 500 0", csp->ip_addr_str);
2190 log_error(LOG_LEVEL_ERROR, "Invalid request line after applying header filters.");
2192 free_http_request(http);
2196 /* decide how to route the HTTP request */
2197 fwd = forward_url(csp, http);
2200 log_error(LOG_LEVEL_FATAL, "gateway spec is NULL!?!? This can't happen!");
2201 /* Never get here - LOG_LEVEL_FATAL causes program exit */
2205 * build the http request to send to the server
2206 * we have to do one of the following:
2208 * create = use the original HTTP request to create a new
2209 * HTTP request that has either the path component
2210 * without the http://domainspec (w/path) or the
2211 * full orininal URL (w/url)
2212 * Note that the path and/or the HTTP version may
2213 * have been altered by now.
2215 * connect = Open a socket to the host:port of the server
2216 * and short-circuit server and client socket.
2218 * pass = Pass the request unchanged if forwarding a CONNECT
2219 * request to a parent proxy. Note that we'll be sending
2220 * the CFAIL message ourselves if connecting to the parent
2221 * fails, but we won't send a CSUCCEED message if it works,
2222 * since that would result in a double message (ours and the
2223 * parent's). After sending the request to the parent, we simply
2226 * here's the matrix:
2229 * +--------+--------+
2231 * 0 | create | connect|
2233 * Forwarding +--------+--------+
2235 * 1 | create | pass |
2237 * +--------+--------+
2241 if (http->ssl && connect_port_is_forbidden(csp))
2243 const char *acceptable_connect_ports =
2244 csp->action->string[ACTION_STRING_LIMIT_CONNECT];
2245 assert(NULL != acceptable_connect_ports);
2246 log_error(LOG_LEVEL_INFO, "Request from %s marked for blocking. "
2247 "limit-connect{%s} doesn't allow CONNECT requests to port %d.",
2248 csp->ip_addr_str, acceptable_connect_ports, csp->http->port);
2249 csp->action->flags |= ACTION_BLOCK;
2255 freez(csp->headers->first->str);
2256 build_request_line(csp, fwd, &csp->headers->first->str);
2259 hdr = list_to_text(csp->headers);
2262 /* FIXME Should handle error properly */
2263 log_error(LOG_LEVEL_FATAL, "Out of memory parsing client header");
2267 * We have a request. Check if one of the crunchers wants it.
2269 if (crunch_response_triggered(csp, crunchers_all))
2272 * Yes. The client got the crunch response
2273 * and we are done here after cleaning up.
2276 list_remove_all(csp->headers);
2282 * The headers can't be removed earlier because
2283 * they were still needed for the referrer check
2284 * in case of CGI crunches.
2286 * XXX: Would it be worth to move the referrer check
2287 * into client_referrer() and set a flag if it's trusted?
2289 list_remove_all(csp->headers);
2291 log_error(LOG_LEVEL_GPC, "%s%s", http->hostport, http->path);
2293 if (fwd->forward_host)
2295 log_error(LOG_LEVEL_CONNECT, "via %s:%d to: %s",
2296 fwd->forward_host, fwd->forward_port, http->hostport);
2300 log_error(LOG_LEVEL_CONNECT, "to %s", http->hostport);
2303 /* here we connect to the server, gateway, or the forwarder */
2305 while ( (csp->sfd = forwarded_connect(fwd, http, csp))
2306 && (errno == EINVAL) && (forwarded_connect_retries++ < max_forwarded_connect_retries))
2308 log_error(LOG_LEVEL_ERROR, "failed request #%u to connect to %s. Trying again.",
2309 forwarded_connect_retries, http->hostport);
2312 if (csp->sfd == JB_INVALID_SOCKET)
2314 if (fwd->type != SOCKS_NONE)
2317 rsp = error_response(csp, "forwarding-failed", errno);
2319 else if (errno == EINVAL)
2321 rsp = error_response(csp, "no-such-domain", errno);
2325 rsp = error_response(csp, "connect-failed", errno);
2326 log_error(LOG_LEVEL_CONNECT, "connect to: %s failed: %E",
2331 /* Write the answer to the client */
2334 send_crunch_response(csp, rsp);
2341 if (fwd->forward_host || (http->ssl == 0))
2343 /* write the client's (modified) header to the server
2344 * (along with anything else that may be in the buffer)
2347 if (write_socket(csp->sfd, hdr, strlen(hdr))
2348 || (flush_socket(csp->sfd, csp->iob) < 0))
2350 log_error(LOG_LEVEL_CONNECT, "write header to: %s failed: %E",
2353 rsp = error_response(csp, "connect-failed", errno);
2357 send_crunch_response(csp, rsp);
2367 * We're running an SSL tunnel and we're not forwarding,
2368 * so just send the "connect succeeded" message to the
2369 * client, flush the rest, and get out of the way.
2371 if (write_socket(csp->cfd, CSUCCEED, strlen(CSUCCEED)))
2379 log_error(LOG_LEVEL_CONNECT, "to %s successful", http->hostport);
2381 /* we're finished with the client's header */
2384 maxfd = ( csp->cfd > csp->sfd ) ? csp->cfd : csp->sfd;
2386 /* pass data between the client and server
2387 * until one or the other shuts down the connection.
2396 * FD_ZERO here seems to point to an errant macro which crashes.
2397 * So do this by hand for now...
2399 memset(&rfds,0x00,sizeof(fd_set));
2403 FD_SET(csp->cfd, &rfds);
2404 FD_SET(csp->sfd, &rfds);
2406 n = select((int)maxfd+1, &rfds, NULL, NULL, NULL);
2410 log_error(LOG_LEVEL_ERROR, "select() failed!: %E");
2414 /* this is the body of the browser's request
2415 * just read it and write it.
2418 if (FD_ISSET(csp->cfd, &rfds))
2420 len = read_socket(csp->cfd, buf, sizeof(buf) - 1);
2424 break; /* "game over, man" */
2427 if (write_socket(csp->sfd, buf, (size_t)len))
2429 log_error(LOG_LEVEL_ERROR, "write to: %s failed: %E", http->host);
2436 * The server wants to talk. It could be the header or the body.
2437 * If `hdr' is null, then it's the header otherwise it's the body.
2438 * FIXME: Does `hdr' really mean `host'? No.
2442 if (FD_ISSET(csp->sfd, &rfds))
2445 len = read_socket(csp->sfd, buf, sizeof(buf) - 1);
2449 log_error(LOG_LEVEL_ERROR, "read from: %s failed: %E", http->host);
2451 if (http->ssl && (fwd->forward_host == NULL))
2454 * Just hang up. We already confirmed the client's CONNECT
2455 * request with status code 200 and unencrypted content is
2456 * no longer welcome.
2458 log_error(LOG_LEVEL_ERROR,
2459 "CONNECT already confirmed. Unable to tell the client about the problem.");
2462 else if (byte_count)
2465 * Just hang up. We already transmitted the original headers
2466 * and parts of the original content and therefore missed the
2467 * chance to send an error message (without risking data corruption).
2469 * XXX: we could retry with a fancy range request here.
2471 log_error(LOG_LEVEL_ERROR, "Already forwarded the original headers. "
2472 "Unable to tell the client about the problem.");
2476 rsp = error_response(csp, "connect-failed", errno);
2480 send_crunch_response(csp, rsp);
2486 /* Add a trailing zero. This lets filter_popups
2487 * use string operations.
2491 /* Normally, this would indicate that we've read
2492 * as much as the server has sent us and we can
2493 * close the client connection. However, Microsoft
2494 * in its wisdom has released IIS/5 with a bug that
2495 * prevents it from sending the trailing \r\n in
2496 * a 302 redirect header (and possibly other headers).
2497 * To work around this if we've haven't parsed
2498 * a full header we'll append a trailing \r\n
2499 * and see if this now generates a valid one.
2501 * This hack shouldn't have any impacts. If we've
2502 * already transmitted the header or if this is a
2503 * SSL connection, then we won't bother with this
2504 * hack. So we only work on partially received
2505 * headers. If we append a \r\n and this still
2506 * doesn't generate a valid header, then we won't
2507 * transmit anything to the client.
2512 if (server_body || http->ssl)
2515 * If we have been buffering up the document,
2516 * now is the time to apply content modification
2517 * and send the result to the client.
2521 p = execute_content_filter(csp, content_filter);
2523 * If the content filter fails, use the original
2524 * buffer and length.
2525 * (see p != NULL ? p : csp->iob->cur below)
2529 csp->content_length = (size_t)(csp->iob->eod - csp->iob->cur);
2532 if (JB_ERR_OK != sed(server_patterns_light, NULL, csp))
2534 log_error(LOG_LEVEL_FATAL, "Failed to parse server headers.");
2537 hdr = list_to_text(csp->headers);
2540 /* FIXME Should handle error properly */
2541 log_error(LOG_LEVEL_FATAL, "Out of memory parsing server header");
2545 * Shouldn't happen because this was the second sed run
2546 * and tags are only created for the first one.
2548 assert(!crunch_response_triggered(csp, crunchers_all));
2550 if (write_socket(csp->cfd, hdr, strlen(hdr))
2551 || write_socket(csp->cfd, p != NULL ? p : csp->iob->cur, csp->content_length))
2553 log_error(LOG_LEVEL_ERROR, "write modified content to client failed: %E");
2563 break; /* "game over, man" */
2567 * This is NOT the body, so
2568 * Let's pretend the server just sent us a blank line.
2570 snprintf(buf, sizeof(buf), "\r\n");
2571 len = (int)strlen(buf);
2574 * Now, let the normal header parsing algorithm below do its
2575 * job. If it fails, we'll exit instead of continuing.
2582 * If this is an SSL connection or we're in the body
2583 * of the server document, just write it to the client,
2584 * unless we need to buffer the body for later content-filtering
2587 if (server_body || http->ssl)
2592 * If there is no memory left for buffering the content, or the buffer limit
2593 * has been reached, switch to non-filtering mode, i.e. make & write the
2594 * header, flush the iob and buf, and get out of the way.
2596 if (add_to_iob(csp, buf, len))
2601 log_error(LOG_LEVEL_INFO, "Flushing header and buffers. Stepping back from filtering.");
2603 hdr = list_to_text(csp->headers);
2607 * Memory is too tight to even generate the header.
2608 * Send our static "Out-of-memory" page.
2610 log_error(LOG_LEVEL_ERROR, "Out of memory while trying to flush.");
2611 rsp = cgi_error_memory();
2612 send_crunch_response(csp, rsp);
2616 hdrlen = strlen(hdr);
2618 if (write_socket(csp->cfd, hdr, hdrlen)
2619 || ((flushed = flush_socket(csp->cfd, csp->iob)) < 0)
2620 || (write_socket(csp->cfd, buf, (size_t)len)))
2622 log_error(LOG_LEVEL_CONNECT, "Flush header and buffers to client failed: %E");
2629 * Reset the byte_count to the amount of bytes
2630 * we just flushed. len will be added a few lines below,
2631 * hdrlen doesn't matter for LOG_LEVEL_CLF.
2633 byte_count = (size_t)flushed;
2635 content_filter = NULL;
2641 if (write_socket(csp->cfd, buf, (size_t)len))
2643 log_error(LOG_LEVEL_ERROR, "write to client failed: %E");
2647 byte_count += (size_t)len;
2652 /* we're still looking for the end of the
2653 * server's header ... (does that make header
2654 * parsing an "out of body experience" ?
2658 * buffer up the data we just read. If that fails,
2659 * there's little we can do but send our static
2660 * out-of-memory page.
2662 if (add_to_iob(csp, buf, len))
2664 log_error(LOG_LEVEL_ERROR, "Out of memory while looking for end of server headers.");
2665 rsp = cgi_error_memory();
2666 send_crunch_response(csp, rsp);
2671 /* Convert iob into something sed() can digest */
2672 if (JB_ERR_PARSE == get_server_headers(csp))
2676 /* Well, we tried our MS IIS/5
2677 * hack and it didn't work.
2678 * The header is incomplete
2679 * and there isn't anything
2680 * we can do about it.
2682 log_error(LOG_LEVEL_INFO,
2683 "MS IIS5 hack didn't produce valid headers.");
2688 /* Since we have to wait for
2689 * more from the server before
2690 * we can parse the headers
2691 * we just continue here.
2697 /* Did we actually get anything? */
2698 if (NULL == csp->headers->first)
2700 log_error(LOG_LEVEL_ERROR, "Empty server or forwarder response.");
2701 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" 502 0", csp->ip_addr_str, http->cmd);
2702 write_socket(csp->cfd, NO_SERVER_DATA_RESPONSE, strlen(NO_SERVER_DATA_RESPONSE));
2703 free_http_request(http);
2707 assert(csp->headers->first->str);
2709 if (strncmpic(csp->headers->first->str, "HTTP", 4))
2712 * It doesn't look like a HTTP response:
2713 * tell the client and log the problem.
2715 if (strlen(csp->headers->first->str) > 30)
2717 csp->headers->first->str[30] = '\0';
2719 log_error(LOG_LEVEL_ERROR,
2720 "Invalid server or forwarder response. Starts with: %s",
2721 csp->headers->first->str);
2722 log_error(LOG_LEVEL_CLF,
2723 "%s - - [%T] \"%s\" 502 0", csp->ip_addr_str, http->cmd);
2724 write_socket(csp->cfd, INVALID_SERVER_HEADERS_RESPONSE,
2725 strlen(INVALID_SERVER_HEADERS_RESPONSE));
2726 free_http_request(http);
2730 /* we have now received the entire header.
2731 * filter it and send the result to the client
2733 if (JB_ERR_OK != sed(server_patterns, add_server_headers, csp))
2735 log_error(LOG_LEVEL_FATAL, "Failed to parse server headers.");
2737 hdr = list_to_text(csp->headers);
2740 /* FIXME Should handle error properly */
2741 log_error(LOG_LEVEL_FATAL, "Out of memory parsing server header");
2744 if (crunch_response_triggered(csp, crunchers_light))
2747 * One of the tags created by a server-header
2748 * tagger triggered a crunch. We already
2749 * delivered the crunch response to the client
2750 * and are done here after cleaning up.
2755 /* Buffer and pcrs filter this if appropriate. */
2757 if (!http->ssl) /* We talk plaintext */
2759 content_filter = get_filter_function(csp);
2762 * Only write if we're not buffering for content modification
2764 if (!content_filter)
2766 /* write the server's (modified) header to
2767 * the client (along with anything else that
2768 * may be in the buffer)
2771 if (write_socket(csp->cfd, hdr, strlen(hdr))
2772 || ((len = flush_socket(csp->cfd, csp->iob)) < 0))
2774 log_error(LOG_LEVEL_CONNECT, "write header to client failed: %E");
2776 /* the write failed, so don't bother
2777 * mentioning it to the client...
2778 * it probably can't hear us anyway.
2784 byte_count += (size_t)len;
2787 /* we're finished with the server's header */
2792 /* If this was a MS IIS/5 hack then it means
2793 * the server has already closed the
2794 * connection. Nothing more to read. Time
2799 log_error(LOG_LEVEL_INFO,
2800 "Closed server connection detected with MS IIS5 hack enabled.");
2807 return; /* huh? we should never get here */
2810 if (csp->content_length == 0)
2813 * If Privoxy didn't recalculate the
2814 * Content-Lenght, byte_count is still
2817 csp->content_length = byte_count;
2820 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" 200 %d",
2821 csp->ip_addr_str, http->ocmd, csp->content_length);
2825 /*********************************************************************
2829 * Description : This is little more than chat. We only "serve" to
2830 * to close any socket that chat may have opened.
2833 * 1 : csp = Current client state (buffers, headers, etc...)
2837 *********************************************************************/
2839 void serve(struct client_state *csp)
2840 #else /* ifndef AMIGA */
2841 static void serve(struct client_state *csp)
2842 #endif /* def AMIGA */
2845 close_socket(csp->cfd);
2847 if (csp->sfd != JB_INVALID_SOCKET)
2849 close_socket(csp->sfd);
2852 csp->flags &= ~CSP_FLAG_ACTIVE;
2858 /*********************************************************************
2860 * Function : server_thread
2862 * Description : We only exist to call `serve' in a threaded environment.
2865 * 1 : data = Current client state (buffers, headers, etc...)
2867 * Returns : Always 0.
2869 *********************************************************************/
2870 static int32 server_thread(void *data)
2872 serve((struct client_state *) data);
2879 #if !defined(_WIN32) || defined(_WIN_CONSOLE)
2880 /*********************************************************************
2884 * Description : Print usage info & exit.
2886 * Parameters : Pointer to argv[0] for identifying ourselves
2890 *********************************************************************/
2891 static void usage(const char *myname)
2893 printf("Privoxy version " VERSION " (" HOME_PAGE_URL ")\n"
2897 #endif /* defined(unix) */
2900 "[--no-daemon] [--pidfile pidfile] [--pre-chroot-nslookup hostname] [--user user[.group]] "
2901 #endif /* defined(unix) */
2902 "[--version] [configfile]\n"
2903 "Aborting\n", myname);
2908 #endif /* #if !defined(_WIN32) || defined(_WIN_CONSOLE) */
2911 /*********************************************************************
2913 * Function : initialize_mutexes
2915 * Description : Prepares mutexes if mutex support is available.
2919 * Returns : Void, exits in case of errors.
2921 *********************************************************************/
2922 static void initialize_mutexes(void)
2926 #ifdef FEATURE_PTHREAD
2928 * Prepare global mutex semaphores
2930 err = pthread_mutex_init(&log_mutex, 0);
2932 if (!err) err = pthread_mutex_init(&log_init_mutex, 0);
2935 * XXX: The assumptions below are a bit naive
2936 * and can cause locks that aren't necessary.
2938 * For example older FreeBSD versions (< 6.x?)
2939 * have no gethostbyname_r, but gethostbyname is
2942 #if !defined(HAVE_GETHOSTBYADDR_R) || !defined(HAVE_GETHOSTBYNAME_R)
2943 if (!err) err = pthread_mutex_init(&resolver_mutex, 0);
2944 #endif /* !defined(HAVE_GETHOSTBYADDR_R) || !defined(HAVE_GETHOSTBYNAME_R) */
2946 * XXX: should we use a single mutex for
2947 * localtime() and gmtime() as well?
2949 #ifndef HAVE_GMTIME_R
2950 if (!err) err = pthread_mutex_init(&gmtime_mutex, 0);
2951 #endif /* ndef HAVE_GMTIME_R */
2953 #ifndef HAVE_LOCALTIME_R
2954 if (!err) err = pthread_mutex_init(&localtime_mutex, 0);
2955 #endif /* ndef HAVE_GMTIME_R */
2958 if (!err) err = pthread_mutex_init(&rand_mutex, 0);
2959 #endif /* ndef HAVE_RANDOM */
2960 #endif /* FEATURE_PTHREAD */
2963 * TODO: mutex support for mingw32 would be swell.
2968 printf("Fatal error. Mutex initialization failed: %s.\n",
2977 /*********************************************************************
2981 * Description : Load the config file and start the listen loop.
2982 * This function is a lot more *sane* with the `load_config'
2983 * and `listen_loop' functions; although it stills does
2984 * a *little* too much for my taste.
2987 * 1 : argc = Number of parameters (including $0).
2988 * 2 : argv = Array of (char *)'s to the parameters.
2990 * Returns : 1 if : can't open config file, unrecognized directive,
2991 * stats requested in multi-thread mode, can't open the
2992 * log file, can't open the jar file, listen port is invalid,
2993 * any load fails, and can't bind port.
2995 * Else main never returns, the process must be signaled
2996 * to terminate execution. Or, on Windows, use the
2997 * "File", "Exit" menu option.
2999 *********************************************************************/
3001 int real_main(int argc, const char *argv[])
3003 int main(int argc, const char *argv[])
3007 unsigned int random_seed;
3009 struct passwd *pw = NULL;
3010 struct group *grp = NULL;
3013 char *pre_chroot_nslookup_to_load_resolver = NULL;
3020 #if !defined(_WIN32)
3028 * Parse the command line arguments
3030 * XXX: simply printing usage information in case of
3031 * invalid arguments isn't particularly user friendly.
3033 while (++argc_pos < argc)
3036 /* Check to see if the service must be installed or uninstalled */
3037 if (strncmp(argv[argc_pos], "--install", 9) == 0)
3039 const char *pName = argv[argc_pos] + 9;
3042 exit( (install_service(pName)) ? 0 : 1 );
3044 else if (strncmp(argv[argc_pos], "--uninstall", + 11) == 0)
3046 const char *pName = argv[argc_pos] + 11;
3049 exit((uninstall_service(pName)) ? 0 : 1);
3051 else if (strcmp(argv[argc_pos], "--service" ) == 0)
3053 bRunAsService = TRUE;
3054 w32_set_service_cwd();
3055 atexit(w32_service_exit_notify);
3058 #endif /* defined(_WIN32) */
3061 #if !defined(_WIN32) || defined(_WIN_CONSOLE)
3063 if (strcmp(argv[argc_pos], "--help") == 0)
3068 else if(strcmp(argv[argc_pos], "--version") == 0)
3070 printf("Privoxy version " VERSION " (" HOME_PAGE_URL ")\n");
3076 else if (strcmp(argv[argc_pos], "--no-daemon" ) == 0)
3081 else if (strcmp(argv[argc_pos], "--pidfile" ) == 0)
3083 if (++argc_pos == argc) usage(argv[0]);
3084 pidfile = strdup(argv[argc_pos]);
3087 else if (strcmp(argv[argc_pos], "--user" ) == 0)
3089 if (++argc_pos == argc) usage(argv[argc_pos]);
3091 if ((NULL != (p = strchr(argv[argc_pos], '.'))) && *(p + 1) != '0')
3094 if (NULL == (grp = getgrnam(p)))
3096 log_error(LOG_LEVEL_FATAL, "Group %s not found.", p);
3100 if (NULL == (pw = getpwnam(argv[argc_pos])))
3102 log_error(LOG_LEVEL_FATAL, "User %s not found.", argv[argc_pos]);
3105 if (p != NULL) *--p = '\0';
3108 else if (strcmp(argv[argc_pos], "--pre-chroot-nslookup" ) == 0)
3110 if (++argc_pos == argc) usage(argv[0]);
3111 pre_chroot_nslookup_to_load_resolver = strdup(argv[argc_pos]);
3114 else if (strcmp(argv[argc_pos], "--chroot" ) == 0)
3118 #endif /* defined(unix) */
3120 else if (argc_pos + 1 != argc)
3123 * This is neither the last command line
3124 * option, nor was it recognized before,
3125 * therefore it must be invalid.
3131 #endif /* defined(_WIN32) && !defined(_WIN_CONSOLE) */
3133 configfile = argv[argc_pos];
3136 } /* -END- while (more arguments) */
3139 if ( *configfile != '/' )
3141 char cwd[BUFFER_SIZE];
3143 size_t abs_file_size;
3145 /* make config-filename absolute here */
3146 if (NULL == getcwd(cwd, sizeof(cwd)))
3148 perror("failed to get current working directory");
3153 abs_file_size = strlen(cwd) + strlen(configfile) + 5;
3154 basedir = strdup(cwd);
3156 if (NULL == basedir ||
3157 NULL == (abs_file = malloc(abs_file_size)))
3159 perror("malloc failed");
3162 strlcpy(abs_file, basedir, abs_file_size);
3163 strlcat(abs_file, "/", abs_file_size );
3164 strlcat(abs_file, configfile, abs_file_size);
3165 configfile = abs_file;
3167 #endif /* defined unix */
3171 clients->next = NULL;
3173 /* XXX: factor out initialising after the next stable release. */
3176 #elif defined(_WIN32)
3180 /* Prepare mutexes if supported and necessary. */
3181 initialize_mutexes();
3183 /* Enable logging until further notice. */
3184 init_log_module(Argv[0]);
3186 random_seed = (unsigned int)time(NULL);
3188 srandom(random_seed);
3189 #elif defined (_WIN32)
3191 * See pick_from_range() in miscutil.c for details.
3193 log_error(LOG_LEVEL_INFO,
3194 "No thread-safe PRNG implemented for your platform. "
3195 "Using weak \'randomization\' factor which will "
3196 "limit the already questionable usefulness of "
3197 "header-time-randomizing actions (disabled by default).");
3200 #endif /* ifdef HAVE_RANDOM */
3203 * Unix signal handling
3205 * Catch the abort, interrupt and terminate signals for a graceful exit
3206 * Catch the hangup signal so the errlog can be reopened.
3207 * Ignore the broken pipe signals (FIXME: Why?)
3209 #if !defined(_WIN32) && !defined(__OS2__) && !defined(AMIGA)
3212 const int catched_signals[] = { SIGTERM, SIGINT, SIGHUP, 0 };
3213 const int ignored_signals[] = { SIGPIPE, 0 };
3215 for (idx = 0; catched_signals[idx] != 0; idx++)
3217 #ifdef sun /* FIXME: Is it safe to check for HAVE_SIGSET instead? */
3218 if (sigset(catched_signals[idx], sig_handler) == SIG_ERR)
3220 if (signal(catched_signals[idx], sig_handler) == SIG_ERR)
3221 #endif /* ifdef sun */
3223 log_error(LOG_LEVEL_FATAL, "Can't set signal-handler for signal %d: %E", catched_signals[idx]);
3227 for (idx = 0; ignored_signals[idx] != 0; idx++)
3229 if (signal(ignored_signals[idx], SIG_IGN) == SIG_ERR)
3231 log_error(LOG_LEVEL_FATAL, "Can't set ignore-handler for signal %d: %E", ignored_signals[idx]);
3236 #else /* ifdef _WIN32 */
3237 # ifdef _WIN_CONSOLE
3239 * We *are* in a windows console app.
3240 * Print a verbose messages about FAQ's and such
3242 printf("%s", win32_blurb);
3243 # endif /* def _WIN_CONSOLE */
3244 #endif /* def _WIN32 */
3247 /* Initialize the CGI subsystem */
3248 cgi_init_error_messages();
3251 * If runnig on unix and without the --nodaemon
3252 * option, become a daemon. I.e. fork, detach
3253 * from tty and get process group leadership
3266 if ( pid < 0 ) /* error */
3271 else if ( pid != 0 ) /* parent */
3276 * must check for errors
3277 * child died due to missing files aso
3280 wpid = waitpid( pid, &status, WNOHANG );
3289 /* Should be more portable, but not as well tested */
3294 #else /* ndef __FreeBSD__ */
3296 #endif /* ndef __FreeBSD__ */
3297 fd = open("/dev/tty", O_RDONLY);
3300 /* no error check here */
3301 ioctl( fd, TIOCNOTTY,0 );
3306 * stderr (fd 2) will be closed later on, when the
3307 * log file has been parsed.
3314 } /* -END- if (!no_daemon) */
3317 * As soon as we have written the PID file, we can switch
3318 * to the user and group ID indicated by the --user option
3324 if (setgid((NULL != grp) ? grp->gr_gid : pw->pw_gid))
3326 log_error(LOG_LEVEL_FATAL, "Cannot setgid(): Insufficient permissions.");
3332 log_error(LOG_LEVEL_FATAL, "Home directory for %s undefined", pw->pw_name);
3334 /* Read the time zone file from /etc before doing chroot. */
3336 if (NULL != pre_chroot_nslookup_to_load_resolver
3337 && '\0' != pre_chroot_nslookup_to_load_resolver[0])
3339 /* Initialize resolver library. */
3340 (void) resolve_hostname_to_ip(pre_chroot_nslookup_to_load_resolver);
3342 if (chroot(pw->pw_dir) < 0)
3344 log_error(LOG_LEVEL_FATAL, "Cannot chroot to %s", pw->pw_dir);
3348 log_error(LOG_LEVEL_FATAL, "Cannot chdir /");
3351 if (setuid(pw->pw_uid))
3353 log_error(LOG_LEVEL_FATAL, "Cannot setuid(): Insufficient permissions.");
3357 char putenv_dummy[64];
3359 strlcpy(putenv_dummy, "HOME=/", sizeof(putenv_dummy));
3360 if (putenv(putenv_dummy) != 0)
3362 log_error(LOG_LEVEL_FATAL, "Cannot putenv(): HOME");
3365 snprintf(putenv_dummy, sizeof(putenv_dummy), "USER=%s", pw->pw_name);
3366 if (putenv(putenv_dummy) != 0)
3368 log_error(LOG_LEVEL_FATAL, "Cannot putenv(): USER");
3374 log_error(LOG_LEVEL_FATAL, "Cannot chroot without --user argument.");
3377 #endif /* defined unix */
3380 /* This will be FALSE unless the command line specified --service
3384 /* Yup, so now we must attempt to establish a connection
3385 * with the service dispatcher. This will only work if this
3386 * process was launched by the service control manager to
3387 * actually run as a service. If this isn't the case, i've
3388 * known it take around 30 seconds or so for the call to return.
3391 /* The StartServiceCtrlDispatcher won't return until the service is stopping */
3392 if (w32_start_service_ctrl_dispatcher(w32ServiceDispatchTable))
3394 /* Service has run, and at this point is now being stopped, so just return */
3399 printf("Warning: Failed to connect to Service Control Dispatcher\nwhen starting as a service!\n");
3401 /* An error occurred. Usually it's because --service was wrongly specified
3402 * and we were unable to connect to the Service Control Dispatcher because
3403 * it wasn't expecting us and is therefore not listening.
3405 * For now, just continue below to call the listen_loop function.
3408 #endif /* def _WIN32 */
3418 /*********************************************************************
3420 * Function : bind_port_helper
3422 * Description : Bind the listen port. Handles logging, and aborts
3426 * 1 : config = Privoxy configuration. Specifies port
3429 * Returns : Port that was opened.
3431 *********************************************************************/
3432 static jb_socket bind_port_helper(struct configuration_spec * config)
3437 if (config->haddr == NULL)
3439 log_error(LOG_LEVEL_INFO, "Listening on port %d on all IP addresses",
3444 log_error(LOG_LEVEL_INFO, "Listening on port %d on IP address %s",
3445 config->hport, config->haddr);
3448 result = bind_port(config->haddr, config->hport, &bfd);
3455 log_error(LOG_LEVEL_FATAL, "can't bind to %s:%d: "
3456 "There may be another Privoxy or some other "
3457 "proxy running on port %d",
3458 (NULL != config->haddr) ? config->haddr : "INADDR_ANY",
3459 config->hport, config->hport);
3462 log_error(LOG_LEVEL_FATAL, "can't bind to %s:%d: "
3463 "The hostname is not resolvable",
3464 (NULL != config->haddr) ? config->haddr : "INADDR_ANY", config->hport);
3467 log_error(LOG_LEVEL_FATAL, "can't bind to %s:%d: because %E",
3468 (NULL != config->haddr) ? config->haddr : "INADDR_ANY", config->hport);
3471 /* shouldn't get here */
3472 return JB_INVALID_SOCKET;
3475 config->need_bind = 0;
3482 /* Without this simple workaround we get this compiler warning from _beginthread
3483 * warning C4028: formal parameter 1 different from declaration
3485 void w32_service_listen_loop(void *p)
3489 #endif /* def _WIN32 */
3492 /*********************************************************************
3494 * Function : listen_loop
3496 * Description : bind the listen port and enter a "FOREVER" listening loop.
3502 *********************************************************************/
3503 static void listen_loop(void)
3505 struct client_state *csp = NULL;
3507 struct configuration_spec * config;
3509 config = load_config();
3511 bfd = bind_port_helper(config);
3513 #ifdef FEATURE_GRACEFUL_TERMINATION
3514 while (!g_terminate)
3519 #if !defined(FEATURE_PTHREAD) && !defined(_WIN32) && !defined(__BEOS__) && !defined(AMIGA) && !defined(__OS2__)
3520 while (waitpid(-1, NULL, WNOHANG) > 0)
3522 /* zombie children */
3524 #endif /* !defined(FEATURE_PTHREAD) && !defined(_WIN32) && !defined(__BEOS__) && !defined(AMIGA) */
3527 * Free data that was used by died threads
3533 * Re-open the errlog after HUP signal
3535 if (received_hup_signal)
3537 init_error_log(Argv[0], config->logfile);
3538 received_hup_signal = 0;
3543 #ifdef FEATURE_COOKIE_JAR
3545 * Need a workaround here: we have to fclose() the jarfile, or we die because it's
3546 * already open. I think unload_configfile() is not being run, which should do
3547 * this work. Until that can get resolved, we'll use this workaround.
3551 if (csp->config->jar)
3553 fclose(csp->config->jar);
3554 csp->config->jar = NULL;
3556 #endif /* FEATURE_COOKIE_JAR */
3557 #endif /* __OS2__ */
3559 if ( NULL == (csp = (struct client_state *) zalloc(sizeof(*csp))) )
3561 log_error(LOG_LEVEL_FATAL, "malloc(%d) for csp failed: %E", sizeof(*csp));
3565 csp->flags |= CSP_FLAG_ACTIVE;
3566 csp->sfd = JB_INVALID_SOCKET;
3568 csp->config = config = load_config();
3570 if ( config->need_bind )
3573 * Since we were listening to the "old port", we will not see
3574 * a "listen" param change until the next IJB request. So, at
3575 * least 1 more request must be made for us to find the new
3576 * setting. I am simply closing the old socket and binding the
3579 * Which-ever is correct, we will serve 1 more page via the
3580 * old settings. This should probably be a "show-proxy-args"
3581 * request. This should not be a so common of an operation
3582 * that this will hurt people's feelings.
3587 bfd = bind_port_helper(config);
3590 log_error(LOG_LEVEL_CONNECT, "Listening for new connections ... ");
3592 if (!accept_connection(csp, bfd))
3594 log_error(LOG_LEVEL_CONNECT, "accept failed: %E");
3607 log_error(LOG_LEVEL_CONNECT, "accepted connection from %s", csp->ip_addr_str);
3610 #ifdef FEATURE_TOGGLE
3611 if (global_toggle_state)
3612 #endif /* def FEATURE_TOGGLE */
3614 csp->flags |= CSP_FLAG_TOGGLED_ON;
3617 if (run_loader(csp))
3619 log_error(LOG_LEVEL_FATAL, "a loader failed - must exit");
3620 /* Never get here - LOG_LEVEL_FATAL causes program exit */
3624 if (block_acl(NULL,csp))
3626 log_error(LOG_LEVEL_CONNECT, "Connection from %s dropped due to ACL", csp->ip_addr_str);
3627 close_socket(csp->cfd);
3631 #endif /* def FEATURE_ACL */
3633 /* add it to the list of clients */
3634 csp->next = clients->next;
3635 clients->next = csp;
3637 if (config->multi_threaded)
3641 /* this is a switch () statment in the C preprocessor - ugh */
3642 #undef SELECTED_ONE_OPTION
3644 /* Use Pthreads in preference to native code */
3645 #if defined(FEATURE_PTHREAD) && !defined(SELECTED_ONE_OPTION)
3646 #define SELECTED_ONE_OPTION
3648 pthread_t the_thread;
3649 pthread_attr_t attrs;
3651 pthread_attr_init(&attrs);
3652 pthread_attr_setdetachstate(&attrs, PTHREAD_CREATE_DETACHED);
3653 errno = pthread_create(&the_thread, &attrs,
3654 (void * (*)(void *))serve, csp);
3655 child_id = errno ? -1 : 0;
3656 pthread_attr_destroy(&attrs);
3660 #if defined(_WIN32) && !defined(_CYGWIN) && !defined(SELECTED_ONE_OPTION)
3661 #define SELECTED_ONE_OPTION
3662 child_id = _beginthread(
3663 (void (*)(void *))serve,
3668 #if defined(__OS2__) && !defined(SELECTED_ONE_OPTION)
3669 #define SELECTED_ONE_OPTION
3670 child_id = _beginthread(
3671 (void(* _Optlink)(void*))serve,
3677 #if defined(__BEOS__) && !defined(SELECTED_ONE_OPTION)
3678 #define SELECTED_ONE_OPTION
3680 thread_id tid = spawn_thread
3681 (server_thread, "server", B_NORMAL_PRIORITY, csp);
3683 if ((tid >= 0) && (resume_thread(tid) == B_OK))
3685 child_id = (int) tid;
3694 #if defined(AMIGA) && !defined(SELECTED_ONE_OPTION)
3695 #define SELECTED_ONE_OPTION
3696 csp->cfd = ReleaseSocket(csp->cfd, -1);
3699 child_id = (int)CreateNewProcTags(NP_Entry, (ULONG)server_thread,
3700 NP_Output, Output(),
3701 NP_CloseOutput, FALSE,
3702 NP_Name, (ULONG)"privoxy child",
3706 child_id = (int)CreateNewProcTags(NP_Entry, (ULONG)server_thread,
3707 NP_Output, Output(),
3708 NP_CloseOutput, FALSE,
3709 NP_Name, (ULONG)"privoxy child",
3710 NP_StackSize, 200*1024,
3716 ((struct Task *)child_id)->tc_UserData = csp;
3717 Signal((struct Task *)child_id, SIGF_SINGLE);
3722 #if !defined(SELECTED_ONE_OPTION)
3725 /* This block is only needed when using fork().
3726 * When using threads, the server thread was
3727 * created and run by the call to _beginthread().
3729 if (child_id == 0) /* child */
3732 #ifdef FEATURE_TOGGLE
3733 int inherited_toggle_state = global_toggle_state;
3734 #endif /* def FEATURE_TOGGLE */
3739 * If we've been toggled or we've blocked the request, tell Mom
3742 #ifdef FEATURE_TOGGLE
3743 if (inherited_toggle_state != global_toggle_state)
3745 rc |= RC_FLAG_TOGGLED;
3747 #endif /* def FEATURE_TOGGLE */
3749 #ifdef FEATURE_STATISTICS
3750 if (csp->flags & CSP_FLAG_REJECTED)
3752 rc |= RC_FLAG_BLOCKED;
3754 #endif /* ndef FEATURE_STATISTICS */
3758 else if (child_id > 0) /* parent */
3760 /* in a fork()'d environment, the parent's
3761 * copy of the client socket and the CSP
3765 #if !defined(_WIN32) && !defined(__CYGWIN__)
3767 wait( &child_status );
3770 * Evaluate child's return code: If the child has
3771 * - been toggled, toggle ourselves
3772 * - blocked its request, bump up the stats counter
3775 #ifdef FEATURE_TOGGLE
3776 if (WIFEXITED(child_status) && (WEXITSTATUS(child_status) & RC_FLAG_TOGGLED))
3778 global_toggle_state = !global_toggle_state;
3780 #endif /* def FEATURE_TOGGLE */
3782 #ifdef FEATURE_STATISTICS
3784 if (WIFEXITED(child_status) && (WEXITSTATUS(child_status) & RC_FLAG_BLOCKED))
3788 #endif /* def FEATURE_STATISTICS */
3790 #endif /* !defined(_WIN32) && defined(__CYGWIN__) */
3791 close_socket(csp->cfd);
3792 csp->flags &= ~CSP_FLAG_ACTIVE;
3796 #undef SELECTED_ONE_OPTION
3797 /* end of cpp switch () */
3799 if (child_id < 0) /* failed */
3801 char buf[BUFFER_SIZE];
3803 log_error(LOG_LEVEL_ERROR, "can't fork: %E");
3805 snprintf(buf , sizeof(buf), "Privoxy: can't fork: errno = %d", errno);
3807 write_socket(csp->cfd, buf, strlen(buf));
3808 close_socket(csp->cfd);
3809 csp->flags &= ~CSP_FLAG_ACTIVE;
3820 /* NOTREACHED unless FEATURE_GRACEFUL_TERMINATION is defined */
3822 /* Clean up. Aim: free all memory (no leaks) */
3823 #ifdef FEATURE_GRACEFUL_TERMINATION
3825 log_error(LOG_LEVEL_ERROR, "Graceful termination requested");
3827 unload_current_config_file();
3828 unload_current_actions_file();
3829 unload_current_re_filterfile();
3830 #ifdef FEATURE_TRUST
3831 unload_current_trust_file();
3834 if (config->multi_threaded)
3841 } while ((clients->next != NULL) && (--i > 0));
3845 log_error(LOG_LEVEL_ERROR, "Graceful termination failed - still some live clients after 1 minute wait.");
3856 #ifdef FEATURE_COOKIE_JAR
3857 if (NULL != config->jar)
3859 fclose(config->jar);
3863 #if defined(_WIN32) && !defined(_WIN_CONSOLE)
3864 /* Cleanup - remove taskbar icon etc. */
3869 #endif /* FEATURE_GRACEFUL_TERMINATION */