1 const char jcc_rcs[] = "$Id: jcc.c,v 1.184 2008/08/22 15:34:45 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 1.184 2008/08/22 15:34:45 fabiankeil
37 * - Silence LLVM/Clang complaint.
38 * - Make received_hup_signal static.
39 * - Hide definitions for basedir, pidfile and received_hup_signal
40 * from __EMX__ as they only seem to be used in case of #ifdef unix.
42 * Revision 1.183 2008/08/21 07:09:35 fabiankeil
43 * Accept Shoutcast responses again. Problem reported
44 * and fix suggested by Stefan in #2062860.
46 * Revision 1.182 2008/06/27 11:13:56 fabiankeil
47 * Fix possible NULL-pointer dereference reported
48 * by din_a4 in #2003937. Pointy hat to me.
50 * Revision 1.181 2008/05/21 15:47:15 fabiankeil
51 * Streamline sed()'s prototype and declare
52 * the header parse and add structures static.
54 * Revision 1.180 2008/05/21 15:26:32 fabiankeil
55 * - Mark csp as immutable for send_crunch_response().
56 * - Fix comment spelling.
58 * Revision 1.179 2008/05/20 20:13:32 fabiankeil
59 * Factor update_server_headers() out of sed(), ditch the
60 * first_run hack and make server_patterns_light static.
62 * Revision 1.178 2008/05/10 13:23:38 fabiankeil
63 * Don't provide get_header() with the whole client state
64 * structure when it only needs access to csp->iob.
66 * Revision 1.177 2008/05/10 11:51:12 fabiankeil
67 * Make the "read the rest of the headers" loop a bit more readable.
69 * Revision 1.176 2008/05/10 11:37:57 fabiankeil
70 * - Instead of logging when the IIS5 hack is enabled, log when it fails.
71 * - Remove useless comment.
73 * Revision 1.175 2008/05/09 18:53:59 fabiankeil
74 * Fix comment grammar.
76 * Revision 1.174 2008/05/07 18:05:53 fabiankeil
77 * Remove the pointless buffer in client_protocol_is_unsupported().
79 * Revision 1.173 2008/05/06 15:09:00 fabiankeil
80 * Least-effort fix for bug #1821930 (reported by Lee):
81 * If the response doesn't look like HTTP,
82 * tell the client and log the problem.
84 * Revision 1.172 2008/04/16 16:38:21 fabiankeil
85 * Don't pass the whole csp structure to flush_socket()
86 * when it only needs a file descriptor and a buffer.
88 * Revision 1.171 2008/03/27 18:27:25 fabiankeil
89 * Remove kill-popups action.
91 * Revision 1.170 2008/03/06 16:33:46 fabiankeil
92 * If limit-connect isn't used, don't limit CONNECT requests to port 443.
94 * Revision 1.169 2008/03/04 18:30:39 fabiankeil
95 * Remove the treat-forbidden-connects-like-blocks action. We now
96 * use the "blocked" page for forbidden CONNECT requests by default.
98 * Revision 1.168 2008/03/02 12:25:25 fabiankeil
99 * Also use shiny new connect_port_is_forbidden() in jcc.c.
101 * Revision 1.167 2008/02/23 16:57:12 fabiankeil
102 * Rename url_actions() to get_url_actions() and let it
103 * use the standard parameter ordering.
105 * Revision 1.166 2008/02/23 16:33:43 fabiankeil
106 * Let forward_url() use the standard parameter ordering
107 * and mark its second parameter immutable.
109 * Revision 1.165 2008/02/02 19:36:56 fabiankeil
110 * Remove the "Listening ... for local connections only" log message.
111 * Whether or not remote connections are able to reach Privoxy is up
112 * to the operating system.
114 * Revision 1.164 2007/12/16 18:32:46 fabiankeil
115 * Prevent the log messages for CONNECT requests to unacceptable
116 * ports from printing the limit-connect argument as [null] if
117 * limit-connect hasn't been explicitly enabled.
119 * Revision 1.163 2007/12/13 01:47:11 david__schmidt
120 * Make sure all console-mode apps get a usage() instance
122 * Revision 1.162 2007/12/06 17:54:57 fabiankeil
123 * Reword NO_SERVER_DATA_RESPONSE to make it harder
124 * to misunderstand what the message is all about.
126 * Revision 1.161 2007/12/04 19:44:22 fabiankeil
127 * Unbreak trustfile which previously didn't work without
128 * FEATURE_TOGGLE. Fixes BR#1843585, reported by Lee.
130 * Revision 1.160 2007/11/29 18:00:29 fabiankeil
131 * Plug memory leak. Spotted by Valgrind, triggered by
132 * Privoxy-Regression-Test feeding proxyfuzz.py.
134 * Revision 1.159 2007/11/24 14:34:09 fabiankeil
135 * In the HTTP snipplets, refer to the client as client.
137 * Revision 1.158 2007/11/11 16:44:17 fabiankeil
138 * Emit a log message when activating the MS IIS5 hack.
140 * Revision 1.157 2007/11/03 17:34:49 fabiankeil
141 * Log the "weak randomization factor" warning only
142 * once for mingw32 and provide some more details.
144 * Revision 1.156 2007/11/01 18:20:58 fabiankeil
145 * Initialize log module after initializing mutexes, future
146 * deadlocks in that code should now work cross-platform.
148 * Revision 1.155 2007/10/23 20:12:45 fabiankeil
149 * Fix first CSUCCEED line to end in \r\n as required by RFC1945.
150 * Reported by Bert van Leeuwen in BR#1818808.
152 * Revision 1.154 2007/10/19 17:00:08 fabiankeil
153 * Downgrade "Flushing header and buffers" message to LOG_LEVEL_INFO.
155 * Revision 1.153 2007/10/14 14:12:41 fabiankeil
156 * When in daemon mode, close stderr after the configuration file has been
157 * parsed the first time. If logfile isn't set, stop logging. Fixes BR#897436.
159 * Revision 1.152 2007/10/04 18:03:34 fabiankeil
160 * - Fix a crash when parsing invalid requests whose first header
161 * is rejected by get_header(). Regression (re?)introduced
162 * in r1.143 by yours truly.
163 * - Move ACTION_VANILLA_WAFER handling into parsers.c's
164 * client_cookie_adder() to make sure send-vanilla-wafer can be
165 * controlled through tags (and thus regression-tested).
167 * Revision 1.151 2007/09/29 10:21:16 fabiankeil
168 * - Move get_filter_function() from jcc.c to filters.c
169 * so the filter functions can be static.
170 * - Don't bother filtering body-less responses.
172 * Revision 1.150 2007/09/28 16:39:29 fabiankeil
173 * Execute content filters through execute_content_filter().
175 * Revision 1.149 2007/09/04 15:08:48 fabiankeil
176 * Initialize req to NULL to make sure it's defined if the
177 * first read_socket() call fails. Reported by icmp30.
179 * Revision 1.148 2007/08/26 16:47:13 fabiankeil
180 * Add Stephen Gildea's --pre-chroot-nslookup patch [#1276666],
181 * extensive comments moved to user manual.
183 * Revision 1.147 2007/08/25 14:42:40 fabiankeil
184 * Don't crash if a broken header filter wiped out the request line.
186 * Revision 1.146 2007/08/20 17:09:32 fabiankeil
187 * Fix byte_count calculation in case of flushes
188 * and don't parse the server headers a second time.
190 * Revision 1.145 2007/08/19 13:13:31 fabiankeil
191 * - If there's a connection problem after we already forwarded
192 * parts of the original content, just hang up. Fixes BR#1776724.
193 * - Fix warnings about unused code on mingw32.
194 * - In case of flushes, calculate the byte count
195 * less incorrectly (I think).
197 * Revision 1.144 2007/08/11 14:43:22 fabiankeil
198 * Add some more prototypes for static functions.
200 * Revision 1.143 2007/08/05 13:58:19 fabiankeil
201 * Comment out request_contains_null_bytes() until it's used again.
203 * Revision 1.142 2007/08/05 13:50:26 fabiankeil
204 * #1763173 from Stefan Huehner: s@const static@static const@
205 * and declare some more functions static.
207 * Revision 1.141 2007/08/04 09:56:23 fabiankeil
208 * - Log rejected CONNECT requests with LOG_LEVEL_INFO
209 * and explain why they were rejected in the first place.
210 * - Fix the LOG_LEVEL_CLF message for crunches of unallowed
211 * CONNECT requests. The request line was missing.
212 * - Add two more XXX reminders as we don't have enough already.
214 * Revision 1.140 2007/07/21 11:51:36 fabiankeil
215 * As Hal noticed, checking dispatch_cgi() as the last cruncher
216 * looks like a bug if CGI requests are blocked unintentionally,
217 * so don't do it unless the user enabled the new config option
218 * "allow-cgi-request-crunching".
220 * Revision 1.139 2007/07/14 07:46:41 fabiankeil
221 * - Allow to rewrite the request destination behind the client's back.
222 * - Turn the weird-looking unconditional for loop that
223 * reads the client request into a conditional while loop.
224 * Move the stuff that only runs once out of the loop.
225 * - Move parts of chat(), server_content_type() and the
226 * necessary stuff to fix BR#1750917 into get_filter_function().
228 * Revision 1.138 2007/06/03 18:45:18 fabiankeil
229 * Temporary workaround for BR#1730105.
231 * Revision 1.137 2007/06/01 18:16:36 fabiankeil
232 * Use the same mutex for gethostbyname() and gethostbyaddr() to prevent
233 * deadlocks and crashes on OpenBSD and possibly other OS with neither
234 * gethostbyname_r() nor gethostaddr_r(). Closes BR#1729174.
235 * Thanks to Ralf Horstmann for report and solution.
237 * Revision 1.136 2007/06/01 16:41:11 fabiankeil
238 * Add forward-override{} to change the forwarding settings through
239 * action sections. This is mainly interesting to forward different
240 * clients differently (for example based on User-Agent or request
243 * Revision 1.135 2007/05/24 17:03:50 fabiankeil
244 * - Let usage() mention the --chroot parameter.
245 * - Use read_socket() consistently and always leave
246 * the last buffer byte alone, even in cases where
247 * null termination (currently) doesn't matter.
249 * Revision 1.134 2007/05/16 14:59:46 fabiankeil
250 * - Fix config file loading on Unix if no config file is specified.
251 * Since r1.97 Privoxy would always interpret the last argument as
252 * config file, even if it's a valid command line option.
253 * - Abort in case of unrecognized command line options. Closes #1719696.
254 * - Remove a bunch of unnecessary strcpy() calls (yay for c&p without thinking).
255 * - Replace the remaining strcpy() and strcat() calls with strlcpy() and strcat().
257 * Revision 1.133 2007/05/04 11:23:19 fabiankeil
258 * - Don't rerun crunchers that only depend on the request URL.
259 * - Don't count redirects and CGI requests as "blocked requests".
261 * Revision 1.132 2007/04/25 15:15:17 fabiankeil
262 * Support crunching based on tags created by server-header taggers.
264 * Revision 1.131 2007/04/22 13:24:50 fabiankeil
265 * Make HTTP snippets static (again). Add a Content-Type for those
266 * with content so the browser doesn't guess it based on the URL.
268 * Revision 1.130 2007/04/19 13:47:34 fabiankeil
269 * Move crunching and request line rebuilding out of chat().
271 * Revision 1.129 2007/04/15 16:39:20 fabiankeil
272 * Introduce tags as alternative way to specify which
273 * actions apply to a request. At the moment tags can be
274 * created based on client and server headers.
276 * Revision 1.128 2007/03/25 16:55:54 fabiankeil
277 * Don't CLF-log CONNECT requests twice.
279 * Revision 1.127 2007/03/20 13:53:17 fabiankeil
280 * Log the source address for ACL-related connection drops.
282 * Revision 1.126 2007/03/17 15:20:05 fabiankeil
283 * New config option: enforce-blocks.
285 * Revision 1.125 2007/03/09 14:12:00 fabiankeil
286 * - Move null byte check into separate function.
287 * - Don't confuse the client with error pages
288 * if a CONNECT request was already confirmed.
290 * Revision 1.124 2007/02/23 14:59:54 fabiankeil
291 * Speed up NULL byte escaping and only log the complete
292 * NULL byte requests with header debugging enabled.
294 * Revision 1.123 2007/02/21 18:42:10 fabiankeil
295 * Answer requests that contain NULL bytes with
296 * a custom response instead of waiting for more
297 * data until the client eventually hangs up.
299 * Revision 1.122 2007/02/07 11:12:02 fabiankeil
300 * - Move delivery and logging of crunched responses
301 * from chat() into send_crunch_response().
302 * - Display the reason for generating http_responses.
303 * - Log the content length for LOG_LEVEL_CLF correctly
304 * (still incorrect for some fixed responses).
305 * - Reword an incorrect comment about
306 * treat-forbidden-connects-like-blocks violating
308 * - Add some log messages.
310 * Revision 1.121 2007/01/27 10:52:56 fabiankeil
311 * Move mutex initialization into separate
312 * function and exit in case of errors.
314 * Revision 1.120 2007/01/26 14:18:42 fabiankeil
315 * - Start to reduce chat()'s line count and move
316 * parts of it into separate functions.
317 * - Add "HTTP/1.1 100 Continue" hack for BR 756734.
319 * Revision 1.119 2007/01/25 14:02:30 fabiankeil
320 * - Add Proxy-Agent header to HTTP snippets that are
321 * supposed to reach HTTP clients only.
322 * - Made a few CONNECT log messages more descriptive.
323 * - Catch completely empty server responses (as seen
324 * with Tor's fake ".noconnect" top level domain).
325 * - Use shiny new "forwarding-failed" template for socks errors.
327 * Revision 1.118 2007/01/07 07:43:43 joergs
328 * AmigaOS4 support added.
330 * Revision 1.117 2006/12/31 17:56:37 fabiankeil
331 * Added config option accept-intercepted-requests
332 * and disabled it by default.
334 * Revision 1.116 2006/12/29 19:08:22 fabiankeil
335 * Reverted parts of my last commit
336 * to keep error handling working.
338 * Revision 1.115 2006/12/29 17:38:57 fabiankeil
339 * Fixed gcc43 conversion warnings.
341 * Revision 1.114 2006/12/27 18:52:02 fabiankeil
342 * Fix -pedantic ISO C warning about converting
343 * from function pointer to object pointer.
345 * Revision 1.113 2006/12/26 17:38:50 fabiankeil
346 * Silence compiler warning I introduced with my last commit.
348 * Revision 1.112 2006/12/26 17:31:41 fabiankeil
349 * Mutex protect rand() if POSIX threading
350 * is used, warn the user if that's not possible
351 * and stop using it on _WIN32 where it could
354 * Revision 1.111 2006/12/23 16:15:06 fabiankeil
355 * Don't prevent core dumps by catching SIGABRT.
356 * It's rude and makes debugging unreasonable painful.
358 * Revision 1.110 2006/12/13 14:52:53 etresoft
359 * Fix build failure on MacOS X. Global symbols can be either static or extern, but not both.
361 * Revision 1.109 2006/12/06 19:41:40 fabiankeil
362 * Privoxy is now able to run as intercepting
363 * proxy in combination with any packet filter
364 * that does the port redirection. The destination
365 * is extracted from the "Host:" header which
366 * should be available for nearly all requests.
368 * Moved HTTP snipplets into jcc.c.
369 * Added error message for gopher proxy requests.
371 * Revision 1.108 2006/11/28 15:38:51 fabiankeil
372 * Only unlink the pidfile if it's actually used.
374 * Change order of interception checks to make
375 * it possible to block or redirect requests for
378 * Revision 1.107 2006/11/13 19:05:51 fabiankeil
379 * Make pthread mutex locking more generic. Instead of
380 * checking for OSX and OpenBSD, check for FEATURE_PTHREAD
381 * and use mutex locking unless there is an _r function
382 * available. Better safe than sorry.
384 * Fixes "./configure --disable-pthread" and should result
385 * in less threading-related problems on pthread-using platforms,
386 * but it still doesn't fix BR#1122404.
388 * Revision 1.106 2006/11/06 19:58:23 fabiankeil
389 * Move pthread.h inclusion from jcc.c to jcc.h.
390 * Fixes build on x86-freebsd1 (FreeBSD 5.4-RELEASE).
392 * Revision 1.105 2006/11/06 14:26:02 fabiankeil
393 * Don't exit after receiving the second SIGHUP on Solaris.
395 * Fixes BR 1052235, but the same problem may exist on other
396 * systems. Once 3.0.6 is out we should use sigset()
397 * where available and see if it breaks anything.
399 * Revision 1.104 2006/09/23 13:26:38 roro
400 * Replace TABs by spaces in source code.
402 * Revision 1.103 2006/09/21 12:54:43 fabiankeil
403 * Fix +redirect{}. Didn't work with -fast-redirects.
405 * Revision 1.102 2006/09/06 13:03:04 fabiankeil
406 * Respond with 400 and a short text message
407 * if the client tries to use Privoxy as FTP proxy.
409 * Revision 1.101 2006/09/06 09:23:37 fabiankeil
410 * Make number of retries in case of forwarded-connect problems
411 * a config file option (forwarded-connect-retries) and use 0 as
414 * Revision 1.100 2006/09/03 19:42:59 fabiankeil
415 * Set random(3) seed.
417 * Revision 1.99 2006/09/02 15:36:42 fabiankeil
418 * Follow the OpenBSD port's lead and protect the resolve
419 * functions on OpenBSD as well.
421 * Revision 1.98 2006/08/24 11:01:34 fabiankeil
422 * --user fix. Only use the user as group if no group is specified.
423 * Solves BR 1492612. Thanks to Spinor S. and David Laight.
425 * Revision 1.97 2006/08/18 15:23:17 david__schmidt
426 * Windows service (re-)integration
430 * --install[:service_name]
431 * --uninstall[:service_name]
434 * They work as follows:
435 * --install will create a service for you and then terminate.
436 * By default the service name will be "privoxy" (without the quotes).
437 * However you can run multiple services if you wish, just by adding
438 * a colon and then a name (no spaces).
440 * --uninstall follows the exact same rules a --install.
442 * --service is used when the program is executed by the service
443 * control manager, and in normal circumstances would never be
444 * used as a command line argument.
446 * Revision 1.96 2006/08/15 20:12:36 david__schmidt
447 * Windows service integration
449 * Revision 1.95 2006/08/03 02:46:41 david__schmidt
450 * Incorporate Fabian Keil's patch work:
451 http://www.fabiankeil.de/sourcecode/privoxy/
453 * Revision 1.94 2006/07/18 14:48:46 david__schmidt
454 * Reorganizing the repository: swapping out what was HEAD (the old 3.1 branch)
455 * with what was really the latest development (the v_3_0_branch branch)
457 * Revision 1.92.2.16 2005/04/03 20:10:50 david__schmidt
458 * Thanks to Jindrich Makovicka for a race condition fix for the log
459 * file. The race condition remains for non-pthread implementations.
460 * Reference patch #1175720.
462 * Revision 1.92.2.15 2004/10/03 12:53:32 david__schmidt
463 * Add the ability to check jpeg images for invalid
464 * lengths of comment blocks. Defensive strategy
465 * against the exploit:
466 * Microsoft Security Bulletin MS04-028
467 * Buffer Overrun in JPEG Processing (GDI+) Could
468 * Allow Code Execution (833987)
469 * Enabled with +inspect-jpegs in actions files.
471 * Revision 1.92.2.14 2003/12/12 12:52:53 oes
472 * - Fixed usage info for non-unix platforms
473 * - Fixed small cmdline parsing bug
475 * Revision 1.92.2.13 2003/11/27 19:20:27 oes
476 * Diagnostics: Now preserve the returncode of pthread_create
477 * in errno. Closes BR #775721. Thanks to Geoffrey Hausheer.
479 * Revision 1.92.2.12 2003/07/11 11:34:19 oes
480 * No longer ignore SIGCHLD. Fixes bug #769381
482 * Revision 1.92.2.11 2003/05/14 12:32:02 oes
483 * Close jarfile on graceful exit, remove stray line
485 * Revision 1.92.2.10 2003/05/08 15:13:46 oes
486 * Cosmetics: Killed a warning, a typo and an allocation left at exit
488 * Revision 1.92.2.9 2003/04/03 15:08:42 oes
489 * No longer rely on non-POSIX.1 extensions of getcwd().
492 * Revision 1.92.2.8 2003/03/31 13:12:32 oes
493 * Replaced setenv() by posix-compliant putenv()
494 * Thanks to Neil McCalden (nmcc AT users.sf.net).
496 * Revision 1.92.2.7 2003/03/17 16:48:59 oes
497 * Added chroot ability, thanks to patch by Sviatoslav Sviridov
499 * Revision 1.92.2.6 2003/03/11 11:55:00 oes
500 * Clean-up and extension of improvements for forked mode:
501 * - Child's return code now consists of flags RC_FLAG_*
502 * - Reporting toggle to parent now properly #ifdef'ed
503 * - Children now report blocking to parent. This enables
504 * statistics in forked mode
506 * Revision 1.92.2.5 2003/03/10 23:45:32 oes
507 * Fixed bug #700381: Non-Threaded version now capable of being toggled.
508 * Children now report having been toggled through _exit(17), parents
509 * watch for that code and toggle themselves if found.
511 * Revision 1.92.2.4 2003/03/07 03:41:04 david__schmidt
512 * Wrapping all *_r functions (the non-_r versions of them) with
513 * mutex semaphores for OSX. Hopefully this will take care of all
514 * of those pesky crash reports.
516 * Revision 1.92.2.3 2003/02/28 12:53:06 oes
517 * Fixed two mostly harmless mem leaks
519 * Revision 1.92.2.2 2002/11/20 14:37:47 oes
520 * Fix: Head of global clients list now initialized to NULL
522 * Revision 1.92.2.1 2002/09/25 14:52:24 oes
523 * Added basic support for OPTIONS and TRACE HTTP methods:
524 * - New interceptor direct_response() added in chat().
525 * - sed() moved to earlier in the process, so that the
526 * Host: header is evaluated before actions and forwarding
529 * Revision 1.92 2002/05/08 16:00:46 oes
530 * Chat's buffer handling:
531 * - Fixed bug with unchecked out-of-mem conditions
532 * while reading client request & server headers
533 * - No longer predict if the buffer limit will be exceeded
534 * in the next read -- check add_to_iob's new
535 * return code. If buffer couldn't be extended
536 * (policy or out-of-mem) while
537 * - reading from client: abort
538 * - reading server headers: send error page
539 * - buffering server body for filter: flush,
540 * and if that fails: send error page
542 * Revision 1.91 2002/04/08 20:35:58 swa
545 * Revision 1.90 2002/04/02 14:57:28 oes
546 * Made sending wafers independent of FEATURE_COOKIE_JAR
548 * Revision 1.89 2002/03/31 17:18:59 jongfoster
549 * Win32 only: Enabling STRICT to fix a VC++ compile warning.
551 * Revision 1.88 2002/03/27 14:32:43 david__schmidt
552 * More compiler warning message maintenance
554 * Revision 1.87 2002/03/26 22:29:54 swa
555 * we have a new homepage!
557 * Revision 1.86 2002/03/25 17:04:55 david__schmidt
558 * Workaround for closing the jarfile before load_config() comes around again
560 * Revision 1.85 2002/03/24 15:23:33 jongfoster
563 * Revision 1.84 2002/03/24 13:25:43 swa
564 * name change related issues
566 * Revision 1.83 2002/03/16 23:54:06 jongfoster
567 * Adding graceful termination feature, to help look for memory leaks.
568 * If you enable this (which, by design, has to be done by hand
569 * editing config.h) and then go to http://i.j.b/die, then the program
570 * will exit cleanly after the *next* request. It should free all the
571 * memory that was used.
573 * Revision 1.82 2002/03/13 00:27:05 jongfoster
576 * Revision 1.81 2002/03/12 01:42:50 oes
577 * Introduced modular filters
579 * Revision 1.80 2002/03/11 22:07:05 david__schmidt
580 * OS/2 port maintenance:
581 * - Fixed EMX build - it had decayed a little
582 * - Fixed inexplicable crash during FD_ZERO - must be due to a bad macro.
583 * substituted a memset for now.
585 * Revision 1.79 2002/03/09 20:03:52 jongfoster
586 * - Making various functions return int rather than size_t.
587 * (Undoing a recent change). Since size_t is unsigned on
588 * Windows, functions like read_socket that return -1 on
589 * error cannot return a size_t.
591 * THIS WAS A MAJOR BUG - it caused frequent, unpredictable
592 * crashes, and also frequently caused JB to jump to 100%
593 * CPU and stay there. (Because it thought it had just
594 * read ((unsigned)-1) == 4Gb of data...)
596 * - The signature of write_socket has changed, it now simply
597 * returns success=0/failure=nonzero.
599 * - Trying to get rid of a few warnings --with-debug on
600 * Windows, I've introduced a new type "jb_socket". This is
601 * used for the socket file descriptors. On Windows, this
602 * is SOCKET (a typedef for unsigned). Everywhere else, it's
603 * an int. The error value can't be -1 any more, so it's
604 * now JB_INVALID_SOCKET (which is -1 on UNIX, and in
605 * Windows it maps to the #define INVALID_SOCKET.)
607 * - The signature of bind_port has changed.
609 * Revision 1.78 2002/03/08 21:35:04 oes
610 * Added optional group supplement to --user option. Will now use default group of user if no group given
612 * Revision 1.77 2002/03/07 03:52:06 oes
613 * - Fixed compiler warnings etc
614 * - Improved handling of failed DNS lookups
616 * Revision 1.76 2002/03/06 22:54:35 jongfoster
617 * Automated function-comment nitpicking.
619 * Revision 1.75 2002/03/06 10:02:19 oes
620 * Fixed stupid bug when --user was not given
622 * Revision 1.74 2002/03/06 00:49:31 jongfoster
623 * Fixing warning on Windows
624 * Making #ifdefs that refer to the same variable consistently
625 * use #ifdef unix rather than mixing #ifdef unix & #ifndef OS2
627 * Revision 1.73 2002/03/05 23:57:30 hal9
628 * Stray character 's' on line 1618 was breaking build.
630 * Revision 1.72 2002/03/05 21:33:45 david__schmidt
631 * - Re-enable OS/2 building after new parms were added
632 * - Fix false out of memory report when resolving CGI templates when no IP
633 * address is available of failed attempt (a la no such domain)
635 * Revision 1.71 2002/03/05 18:13:56 oes
636 * Added --user option
638 * Revision 1.70 2002/03/05 04:52:42 oes
639 * Deleted non-errlog debugging code
641 * Revision 1.69 2002/03/04 23:50:00 jongfoster
642 * Splitting off bind_port() call into bind_port_helper(), with
645 * Revision 1.68 2002/03/04 20:17:32 oes
648 * Revision 1.67 2002/03/04 18:18:57 oes
649 * - Removed _DEBUG mode
650 * - Cleand up cmdline parsing
651 * - Introduced --no-daemon, --pidfile options
652 * - Cleaned up signal handling:
653 * - Terminate cleanly on INT, TERM and ABRT
654 * - Schedule logfile for re-opening on HUP
655 * - Ignore CHLD and PIPE
656 * - Leave the rest with their default handlers
657 * - Uniform handler registration
658 * - Added usage() function
659 * - Played styleguide police
661 * Revision 1.66 2002/03/03 15:06:55 oes
662 * Re-enabled automatic config reloading
664 * Revision 1.65 2002/03/03 14:49:11 oes
665 * Fixed CLF logging: Now uses client's original HTTP request
667 * Revision 1.64 2002/03/03 09:18:03 joergs
668 * Made jumbjuster work on AmigaOS again.
670 * Revision 1.63 2002/03/02 04:14:50 david__schmidt
671 * Clean up a little CRLF unpleasantness that suddenly appeared
673 * Revision 1.62 2002/02/20 23:17:23 jongfoster
674 * Detecting some out-of memory conditions and exiting with a log message.
676 * Revision 1.61 2002/01/17 21:01:52 jongfoster
677 * Moving all our URL and URL pattern parsing code to urlmatch.c.
679 * Revision 1.60 2001/12/30 14:07:32 steudten
680 * - Add signal handling (unix)
681 * - Add SIGHUP handler (unix)
682 * - Add creation of pidfile (unix)
683 * - Add action 'top' in rc file (RH)
684 * - Add entry 'SIGNALS' to manpage
685 * - Add exit message to logfile (unix)
687 * Revision 1.59 2001/12/13 14:07:18 oes
688 * Fixed Bug: 503 error page now sent OK
690 * Revision 1.58 2001/11/30 23:37:24 jongfoster
691 * Renaming the Win32 config file to config.txt - this is almost the
692 * same as the corresponding UNIX name "config"
694 * Revision 1.57 2001/11/16 00:47:43 jongfoster
695 * Changing the tty-disconnection code to use setsid().
697 * Revision 1.56 2001/11/13 20:20:54 jongfoster
698 * Tabs->spaces, fixing a bug with missing {} around an if()
700 * Revision 1.55 2001/11/13 20:14:53 jongfoster
701 * Patch for FreeBSD setpgrp() as suggested by Alexander Lazic
703 * Revision 1.54 2001/11/07 00:03:14 steudten
704 * Give reliable return value if an error
705 * occurs not just 0 with new daemon mode.
707 * Revision 1.53 2001/11/05 21:41:43 steudten
708 * Add changes to be a real daemon just for unix os.
709 * (change cwd to /, detach from controlling tty, set
710 * process group and session leader to the own process.
712 * Add some fatal-error log message for failed malloc().
713 * Add '-d' if compiled with 'configure --with-debug' to
714 * enable debug output.
716 * Revision 1.52 2001/10/26 20:11:20 jongfoster
717 * Fixing type mismatch
719 * Revision 1.51 2001/10/26 17:38:28 oes
722 * Revision 1.50 2001/10/25 03:40:48 david__schmidt
723 * Change in porting tactics: OS/2's EMX porting layer doesn't allow multiple
724 * threads to call select() simultaneously. So, it's time to do a real, live,
725 * native OS/2 port. See defines for __EMX__ (the porting layer) vs. __OS2__
726 * (native). Both versions will work, but using __OS2__ offers multi-threading.
728 * Revision 1.49 2001/10/23 21:41:35 jongfoster
729 * Added call to initialize the (statically-allocated of course)
730 * "out of memory" CGI response.
732 * Revision 1.48 2001/10/10 19:56:46 jongfoster
733 * Moving some code that wasn't cookie-related out of an #ifdef
736 * Revision 1.47 2001/10/10 16:44:36 oes
737 * Added CONNECT destination port limitation check
739 * Revision 1.46 2001/10/08 15:17:41 oes
740 * Re-enabled SSL forwarding
742 * Revision 1.45 2001/10/07 15:42:11 oes
743 * Replaced 6 boolean members of csp with one bitmap (csp->flags)
745 * Moved downgrading of the HTTP version from parse_http_request to
746 * chat(), since we can't decide if it is necessary before we have
747 * determined the actions for the URL. The HTTP command is now
748 * *always* re-built so the repairs need no longer be special-cased.
750 * filter_popups now gets a csp pointer so it can raise the new
751 * CSP_FLAG_MODIFIED flag.
755 * Added configurable size limit for the IOB. If the IOB grows so
756 * large that the next read would exceed the limit, the header
757 * is generated, and the header & unfiltered buffer are flushed
758 * to the client. Chat then continues in non-buffering,
759 * non-filtering body mode.
761 * Revision 1.44 2001/10/02 18:13:57 oes
764 * Revision 1.43 2001/10/02 15:32:13 oes
765 * Moved generation of hdr
767 * Revision 1.42 2001/09/21 23:02:02 david__schmidt
768 * Cleaning up 2 compiler warnings on OS/2.
770 * Revision 1.41 2001/09/16 17:05:14 jongfoster
771 * Removing unused #include showarg.h
773 * Revision 1.40 2001/09/16 15:41:45 jongfoster
774 * Fixing signed/unsigned comparison warning.
776 * Revision 1.39 2001/09/16 13:21:27 jongfoster
777 * Changes to use new list functions.
779 * Revision 1.38 2001/09/16 13:01:46 jongfoster
780 * Removing redundant function call that zeroed zalloc()'d memory.
782 * Revision 1.37 2001/09/10 11:12:24 oes
783 * Deleted unused variable
785 * Revision 1.36 2001/09/10 10:56:15 oes
786 * Silenced compiler warnings
788 * Revision 1.35 2001/07/31 14:44:22 oes
789 * Deleted unused size parameter from filter_popups()
791 * Revision 1.34 2001/07/30 22:08:36 jongfoster
792 * Tidying up #defines:
793 * - All feature #defines are now of the form FEATURE_xxx
794 * - Permanently turned off WIN_GUI_EDIT
795 * - Permanently turned on WEBDAV and SPLIT_PROXY_ARGS
797 * Revision 1.33 2001/07/29 19:32:00 jongfoster
798 * Renaming _main() [mingw32 only] to real_main(), for ANSI compliance.
800 * Revision 1.32 2001/07/29 18:47:05 jongfoster
801 * Adding missing #include "loadcfg.h"
803 * Revision 1.31 2001/07/29 12:17:48 oes
804 * Applied pthread fix by Paul Lieverse
806 * Revision 1.30 2001/07/25 22:57:13 jongfoster
807 * __BEOS__ no longer overrides FEATURE_PTHREAD.
808 * This is because FEATURE_PTHREAD will soon be widely used, so I
809 * want to keep it simple.
811 * Revision 1.29 2001/07/24 12:47:06 oes
812 * Applied BeOS support update by Eugenia
814 * Revision 1.28 2001/07/23 13:26:12 oes
815 * Fixed bug in popup-killing for the first read that caused binary garbage to be sent between headers and body
817 * Revision 1.27 2001/07/19 19:09:47 haroon
818 * - Added code to take care of the situation where while processing the first
819 * server response (which includes the server header), after finding the end
820 * of the headers we were not looking past the end of the headers for
821 * content modification. I enabled it for filter_popups.
822 * Someone else should look to see if other similar operations should be
823 * done to the discarded portion of the buffer.
825 * Note 2001/07/20: No, the other content modification mechanisms will process
826 * the whole iob later anyway. --oes
828 * Revision 1.26 2001/07/18 12:31:36 oes
831 * Revision 1.25 2001/07/15 19:43:49 jongfoster
832 * Supports POSIX threads.
833 * Also removed some unused #includes.
835 * Revision 1.24 2001/07/13 14:00:40 oes
836 * - Generic content modification scheme:
837 * Each feature has its own applicability flag that is set
838 * from csp->action->flags.
839 * Replaced the "filtering" int flag , by a function pointer
840 * "content_filter" to the function that will do the content
841 * modification. If it is != NULL, the document will be buffered
842 * and processed through *content_filter, which must set
843 * csp->content_length and return a modified copy of the body
844 * or return NULL (on failiure).
845 * - Changed csp->is_text to the more generic bitmap csp->content_type
846 * which can currently take the valued CT_TEXT or CT_GIF
848 * - Removed all #ifdef PCRS
850 * Revision 1.23 2001/07/02 02:28:25 iwanttokeepanon
851 * Added "#ifdef ACL_FILES" conditional compilation to line 1291 to exclude
852 * the `block_acl' call. This prevents a compilation error when the user
853 * does not wish to use the "ACL" feature.
855 * Revision 1.22 2001/06/29 21:45:41 oes
856 * Indentation, CRLF->LF, Tab-> Space
858 * Revision 1.21 2001/06/29 13:29:36 oes
859 * - Cleaned up, improved comments
860 * - Unified all possible interceptors (CGI,
861 * block, trust, fast_redirect) in one
862 * place, with one (CGI) answer generation
863 * mechansim. Much clearer now.
864 * - Removed the GIF image generation, which
865 * is now done in filters.c:block_url()
866 * - Made error conditions like domain lookup
867 * failiure or (various) problems while talking
868 * to the server use cgi.c:error_response()
869 * instead of generating HTML/HTTP in chat() (yuck!)
870 * - Removed logentry from cancelled commit
872 * Revision 1.20 2001/06/09 10:55:28 jongfoster
873 * Changing BUFSIZ ==> BUFFER_SIZE
875 * Revision 1.19 2001/06/07 23:12:52 jongfoster
876 * Replacing function pointer in struct gateway with a directly
877 * called function forwarded_connect().
878 * Replacing struct gateway with struct forward_spec
880 * Revision 1.18 2001/06/03 19:12:16 oes
881 * introduced new cgi handling
883 * Revision 1.17 2001/06/01 20:07:23 jongfoster
884 * Now uses action +image-blocker{} rather than config->tinygif
886 * Revision 1.16 2001/06/01 18:49:17 jongfoster
887 * Replaced "list_share" with "list" - the tiny memory gain was not
888 * worth the extra complexity.
890 * Revision 1.15 2001/05/31 21:24:47 jongfoster
891 * Changed "permission" to "action" throughout.
892 * Removed DEFAULT_USER_AGENT - it must now be specified manually.
893 * Moved vanilla wafer check into chat(), since we must now
894 * decide whether or not to add it based on the URL.
896 * Revision 1.14 2001/05/29 20:14:01 joergs
897 * AmigaOS bugfix: PCRS needs a lot of stack, stacksize for child threads
900 * Revision 1.13 2001/05/29 09:50:24 jongfoster
901 * Unified blocklist/imagelist/permissionslist.
902 * File format is still under discussion, but the internal changes
905 * Also modified interceptor behaviour:
906 * - We now intercept all URLs beginning with one of the following
907 * prefixes (and *only* these prefixes):
909 * * http://ijbswa.sf.net/config/
910 * * http://ijbswa.sourceforge.net/config/
911 * - New interceptors "home page" - go to http://i.j.b/ to see it.
912 * - Internal changes so that intercepted and fast redirect pages
913 * are not replaced with an image.
914 * - Interceptors now have the option to send a binary page direct
915 * to the client. (i.e. ijb-send-banner uses this)
916 * - Implemented show-url-info interceptor. (Which is why I needed
917 * the above interceptors changes - a typical URL is
918 * "http://i.j.b/show-url-info?url=www.somesite.com/banner.gif".
919 * The previous mechanism would not have intercepted that, and
920 * if it had been intercepted then it then it would have replaced
923 * Revision 1.12 2001/05/27 22:17:04 oes
925 * - re_process_buffer no longer writes the modified buffer
926 * to the client, which was very ugly. It now returns the
927 * buffer, which it is then written by chat.
929 * - content_length now adjusts the Content-Length: header
930 * for modified documents rather than crunch()ing it.
931 * (Length info in csp->content_length, which is 0 for
932 * unmodified documents)
934 * - For this to work, sed() is called twice when filtering.
936 * Revision 1.11 2001/05/26 17:27:53 jongfoster
937 * Added support for CLF and fixed LOG_LEVEL_LOG.
938 * Also did CRLF->LF fix of my previous patch.
940 * Revision 1.10 2001/05/26 15:26:15 jongfoster
941 * ACL feature now provides more security by immediately dropping
942 * connections from untrusted hosts.
944 * Revision 1.9 2001/05/26 00:28:36 jongfoster
945 * Automatic reloading of config file.
946 * Removed obsolete SIGHUP support (Unix) and Reload menu option (Win32).
947 * Most of the global variables have been moved to a new
948 * struct configuration_spec, accessed through csp->config->globalname
949 * Most of the globals remaining are used by the Win32 GUI.
951 * Revision 1.8 2001/05/25 22:43:18 jongfoster
952 * Fixing minor memory leak and buffer overflow.
954 * Revision 1.7 2001/05/25 22:34:30 jongfoster
957 * Revision 1.6 2001/05/23 00:13:58 joergs
958 * AmigaOS support fixed.
960 * Revision 1.5 2001/05/22 18:46:04 oes
962 * - Enabled filtering banners by size rather than URL
963 * by adding patterns that replace all standard banner
964 * sizes with the "Junkbuster" gif to the re_filterfile
966 * - Enabled filtering WebBugs by providing a pattern
967 * which kills all 1x1 images
969 * - Added support for PCRE_UNGREEDY behaviour to pcrs,
970 * which is selected by the (nonstandard and therefore
971 * capital) letter 'U' in the option string.
972 * It causes the quantifiers to be ungreedy by default.
973 * Appending a ? turns back to greedy (!).
975 * - Added a new interceptor ijb-send-banner, which
976 * sends back the "Junkbuster" gif. Without imagelist or
977 * MSIE detection support, or if tinygif = 1, or the
978 * URL isn't recognized as an imageurl, a lame HTML
979 * explanation is sent instead.
981 * - Added new feature, which permits blocking remote
982 * script redirects and firing back a local redirect
984 * The feature is conditionally compiled, i.e. it
985 * can be disabled with --disable-fast-redirects,
986 * plus it must be activated by a "fast-redirects"
987 * line in the config file, has its own log level
988 * and of course wants to be displayed by show-proxy-args
989 * Note: Boy, all the #ifdefs in 1001 locations and
990 * all the fumbling with configure.in and acconfig.h
991 * were *way* more work than the feature itself :-(
993 * - Because a generic redirect template was needed for
994 * this, tinygif = 3 now uses the same.
996 * - Moved GIFs, and other static HTTP response templates
1001 * - Removed some >400 CRs again (Jon, you really worked
1004 * Revision 1.4 2001/05/21 19:34:01 jongfoster
1005 * Made failure to bind() a fatal error.
1007 * Revision 1.3 2001/05/20 01:21:20 jongfoster
1008 * Version 2.9.4 checkin.
1009 * - Merged popupfile and cookiefile, and added control over PCRS
1010 * filtering, in new "permissionsfile".
1011 * - Implemented LOG_LEVEL_FATAL, so that if there is a configuration
1012 * file error you now get a message box (in the Win32 GUI) rather
1013 * than the program exiting with no explanation.
1014 * - Made killpopup use the PCRS MIME-type checking and HTTP-header
1016 * - Removed tabs from "config"
1017 * - Moved duplicated url parsing code in "loaders.c" to a new funcition.
1018 * - Bumped up version number.
1020 * Revision 1.2 2001/05/17 22:34:44 oes
1021 * - Added hint on GIF char array generation to jcc.c
1022 * - Cleaned CRLF's from the sources and related files
1023 * - Repaired logging for REF and FRC
1025 * Revision 1.1.1.1 2001/05/15 13:58:56 oes
1026 * Initial import of version 2.9.3 source tree
1029 *********************************************************************/
1035 #include <sys/types.h>
1044 # ifndef FEATURE_PTHREAD
1048 # include <windows.h>
1049 # include <process.h>
1050 # endif /* ndef FEATURE_PTHREAD */
1053 # ifndef _WIN_CONSOLE
1054 # include "w32log.h"
1055 # endif /* ndef _WIN_CONSOLE */
1056 # include "w32svrapi.h"
1058 #else /* ifndef _WIN32 */
1060 # if !defined (__OS2__)
1061 # include <unistd.h>
1062 # include <sys/wait.h>
1063 # endif /* ndef __OS2__ */
1064 # include <sys/time.h>
1065 # include <sys/stat.h>
1066 # include <sys/ioctl.h>
1069 #include <sys/termios.h>
1077 # include <signal.h>
1080 # include <socket.h> /* BeOS has select() for sockets only. */
1081 # include <OS.h> /* declarations for threads and stuff. */
1084 # if defined(__EMX__) || defined(__OS2__)
1085 # include <sys/select.h> /* OS/2/EMX needs a little help with select */
1090 #define bzero(B,N) memset(B,0x00,n)
1094 # include <select.h>
1099 #include "project.h"
1102 #include "filters.h"
1103 #include "loaders.h"
1104 #include "parsers.h"
1105 #include "miscutil.h"
1107 #include "jbsockets.h"
1108 #include "gateway.h"
1109 #include "actions.h"
1111 #include "loadcfg.h"
1112 #include "urlmatch.h"
1114 const char jcc_h_rcs[] = JCC_H_VERSION;
1115 const char project_h_rcs[] = PROJECT_H_VERSION;
1118 struct client_state clients[1];
1119 struct file_list files[1];
1121 #ifdef FEATURE_STATISTICS
1122 int urls_read = 0; /* total nr of urls read inc rejected */
1123 int urls_rejected = 0; /* total nr of urls rejected */
1124 #endif /* def FEATURE_STATISTICS */
1126 #ifdef FEATURE_GRACEFUL_TERMINATION
1127 int g_terminate = 0;
1130 #if !defined(_WIN32) && !defined(__OS2__) && !defined(AMIGA)
1131 static void sig_handler(int the_signal);
1133 static int client_protocol_is_unsupported(const struct client_state *csp, char *req);
1134 static jb_err get_request_destination_elsewhere(struct client_state *csp, struct list *headers);
1135 static jb_err get_server_headers(struct client_state *csp);
1136 static const char *crunch_reason(const struct http_response *rsp);
1137 static void send_crunch_response(const struct client_state *csp, struct http_response *rsp);
1139 * static int request_contains_null_bytes(const struct client_state *csp, char *buf, int len);
1141 static void build_request_line(struct client_state *csp, const struct forward_spec *fwd, char **request_line);
1142 static jb_err change_request_destination(struct client_state *csp);
1143 static void chat(struct client_state *csp);
1144 static void serve(struct client_state *csp);
1145 #if !defined(_WIN32) || defined(_WIN_CONSOLE)
1146 static void usage(const char *myname);
1148 static void initialize_mutexes(void);
1149 static jb_socket bind_port_helper(struct configuration_spec *config);
1150 static void listen_loop(void);
1153 void serve(struct client_state *csp);
1154 #else /* ifndef AMIGA */
1155 static void serve(struct client_state *csp);
1156 #endif /* def AMIGA */
1159 static int32 server_thread(void *data);
1160 #endif /* def __BEOS__ */
1163 #define sleep(N) Sleep(((N) * 1000))
1167 #define sleep(N) DosSleep(((N) * 100))
1170 #ifdef FEATURE_PTHREAD
1171 pthread_mutex_t log_mutex;
1172 pthread_mutex_t log_init_mutex;
1174 #if !defined(HAVE_GETHOSTBYADDR_R) || !defined(HAVE_GETHOSTBYNAME_R)
1175 pthread_mutex_t resolver_mutex;
1176 #endif /* !defined(HAVE_GETHOSTBYADDR_R) || !defined(HAVE_GETHOSTBYNAME_R) */
1178 #ifndef HAVE_GMTIME_R
1179 pthread_mutex_t gmtime_mutex;
1180 #endif /* ndef HAVE_GMTIME_R */
1182 #ifndef HAVE_LOCALTIME_R
1183 pthread_mutex_t localtime_mutex;
1184 #endif /* ndef HAVE_GMTIME_R */
1187 pthread_mutex_t rand_mutex;
1188 #endif /* ndef HAVE_RANDOM */
1190 #endif /* FEATURE_PTHREAD */
1193 const char *basedir = NULL;
1194 const char *pidfile = NULL;
1195 static int received_hup_signal = 0;
1196 #endif /* defined unix */
1198 /* HTTP snipplets. */
1199 static const char CSUCCEED[] =
1200 "HTTP/1.0 200 Connection established\r\n"
1201 "Proxy-Agent: Privoxy/" VERSION "\r\n\r\n";
1203 static const char CHEADER[] =
1204 "HTTP/1.0 400 Invalid header received from client\r\n"
1205 "Proxy-Agent: Privoxy " VERSION "\r\n"
1206 "Content-Type: text/plain\r\n"
1207 "Connection: close\r\n\r\n"
1208 "Invalid header received from client.\r\n";
1210 static const char FTP_RESPONSE[] =
1211 "HTTP/1.0 400 Invalid request received from client\r\n"
1212 "Content-Type: text/plain\r\n"
1213 "Connection: close\r\n\r\n"
1214 "Invalid request. Privoxy doesn't support FTP.\r\n";
1216 static const char GOPHER_RESPONSE[] =
1217 "HTTP/1.0 400 Invalid request received from client\r\n"
1218 "Content-Type: text/plain\r\n"
1219 "Connection: close\r\n\r\n"
1220 "Invalid request. Privoxy doesn't support gopher.\r\n";
1222 /* XXX: should be a template */
1223 static const char MISSING_DESTINATION_RESPONSE[] =
1224 "HTTP/1.0 400 Bad request received from client\r\n"
1225 "Proxy-Agent: Privoxy " VERSION "\r\n"
1226 "Content-Type: text/plain\r\n"
1227 "Connection: close\r\n\r\n"
1228 "Bad request. Privoxy was unable to extract the destination.\r\n";
1230 /* XXX: should be a template */
1231 static const char NO_SERVER_DATA_RESPONSE[] =
1232 "HTTP/1.0 502 Server or forwarder response empty\r\n"
1233 "Proxy-Agent: Privoxy " VERSION "\r\n"
1234 "Content-Type: text/plain\r\n"
1235 "Connection: close\r\n\r\n"
1236 "Empty server or forwarder response.\r\n"
1237 "The connection has been closed but Privoxy didn't receive any data.\r\n";
1239 /* XXX: should be a template */
1240 static const char INVALID_SERVER_HEADERS_RESPONSE[] =
1241 "HTTP/1.0 502 Server or forwarder response invalid\r\n"
1242 "Proxy-Agent: Privoxy " VERSION "\r\n"
1243 "Content-Type: text/plain\r\n"
1244 "Connection: close\r\n\r\n"
1245 "Bad response. The server or forwarder response doesn't look like HTTP.\r\n";
1248 /* XXX: should be a template */
1249 static const char NULL_BYTE_RESPONSE[] =
1250 "HTTP/1.0 400 Bad request received from client\r\n"
1251 "Proxy-Agent: Privoxy " VERSION "\r\n"
1252 "Content-Type: text/plain\r\n"
1253 "Connection: close\r\n\r\n"
1254 "Bad request. Null byte(s) before end of request.\r\n";
1257 /* XXX: should be a template */
1258 static const char MESSED_UP_REQUEST_RESPONSE[] =
1259 "HTTP/1.0 400 Malformed request after rewriting\r\n"
1260 "Proxy-Agent: Privoxy " VERSION "\r\n"
1261 "Content-Type: text/plain\r\n"
1262 "Connection: close\r\n\r\n"
1263 "Bad request. Messed up with header filters.\r\n";
1265 /* A function to crunch a response */
1266 typedef struct http_response *(*crunch_func_ptr)(struct client_state *);
1268 /* Crunch function flags */
1269 #define CF_NO_FLAGS 0
1270 /* Cruncher applies to forced requests as well */
1271 #define CF_IGNORE_FORCE 1
1272 /* Crunched requests are counted for the block statistics */
1273 #define CF_COUNT_AS_REJECT 2
1275 /* A crunch function and its flags */
1278 const crunch_func_ptr cruncher;
1282 static int crunch_response_triggered(struct client_state *csp, const struct cruncher crunchers[]);
1284 /* Complete list of cruncher functions */
1285 static const struct cruncher crunchers_all[] = {
1286 { direct_response, CF_COUNT_AS_REJECT|CF_IGNORE_FORCE},
1287 { block_url, CF_COUNT_AS_REJECT },
1288 #ifdef FEATURE_TRUST
1289 { trust_url, CF_COUNT_AS_REJECT },
1290 #endif /* def FEATURE_TRUST */
1291 { redirect_url, CF_NO_FLAGS },
1292 { dispatch_cgi, CF_IGNORE_FORCE},
1296 /* Light version, used after tags are applied */
1297 static const struct cruncher crunchers_light[] = {
1298 { block_url, CF_COUNT_AS_REJECT },
1299 { redirect_url, CF_NO_FLAGS },
1305 * XXX: Don't we really mean
1311 #if !defined(_WIN32) && !defined(__OS2__) && !defined(AMIGA)
1312 /*********************************************************************
1314 * Function : sig_handler
1316 * Description : Signal handler for different signals.
1317 * Exit gracefully on TERM and INT
1318 * or set a flag that will cause the errlog
1319 * to be reopened by the main thread on HUP.
1322 * 1 : the_signal = the signal cause this function to call
1326 *********************************************************************/
1327 static void sig_handler(int the_signal)
1333 log_error(LOG_LEVEL_INFO, "exiting by signal %d .. bye", the_signal);
1345 received_hup_signal = 1;
1351 * We shouldn't be here, unless we catch signals
1352 * in main() that we can't handle here!
1354 log_error(LOG_LEVEL_FATAL, "sig_handler: exiting on unexpected signal %d", the_signal);
1362 /*********************************************************************
1364 * Function : client_protocol_is_unsupported
1366 * Description : Checks if the client used a known unsupported
1367 * protocol and deals with it by sending an error
1371 * 1 : csp = Current client state (buffers, headers, etc...)
1372 * 2 : req = the first request line send by the client
1374 * Returns : TRUE if an error response has been generated, or
1375 * FALSE if the request doesn't look invalid.
1377 *********************************************************************/
1378 static int client_protocol_is_unsupported(const struct client_state *csp, char *req)
1381 * If it's a FTP or gopher request, we don't support it.
1383 * These checks are better than nothing, but they might
1384 * not work in all configurations and some clients might
1385 * have problems digesting the answer.
1387 * They should, however, never cause more problems than
1388 * Privoxy's old behaviour (returning the misleading HTML
1391 * "Could not resolve http://(ftp|gopher)://example.org").
1393 if (!strncmpic(req, "GET ftp://", 10) || !strncmpic(req, "GET gopher://", 13))
1395 const char *response = NULL;
1396 const char *protocol = NULL;
1398 if (!strncmpic(req, "GET ftp://", 10))
1400 response = FTP_RESPONSE;
1405 response = GOPHER_RESPONSE;
1406 protocol = "GOPHER";
1408 log_error(LOG_LEVEL_ERROR,
1409 "%s tried to use Privoxy as %s proxy: %s",
1410 csp->ip_addr_str, protocol, req);
1411 log_error(LOG_LEVEL_CLF,
1412 "%s - - [%T] \"%s\" 400 0", csp->ip_addr_str, req);
1414 write_socket(csp->cfd, response, strlen(response));
1423 /*********************************************************************
1425 * Function : get_request_destination_elsewhere
1427 * Description : If the client's request was redirected into
1428 * Privoxy without the client's knowledge,
1429 * the request line lacks the destination host.
1431 * This function tries to get it elsewhere,
1432 * provided accept-intercepted-requests is enabled.
1434 * "Elsewhere" currently only means "Host: header",
1435 * but in the future we may ask the redirecting
1436 * packet filter to look the destination up.
1438 * If the destination stays unknown, an error
1439 * response is send to the client and headers
1440 * are freed so that chat() can return directly.
1443 * 1 : csp = Current client state (buffers, headers, etc...)
1444 * 2 : headers = a header list
1446 * Returns : JB_ERR_OK if the destination is now known, or
1447 * JB_ERR_PARSE if it isn't.
1449 *********************************************************************/
1450 static jb_err get_request_destination_elsewhere(struct client_state *csp, struct list *headers)
1454 if (!(csp->config->feature_flags & RUNTIME_FEATURE_ACCEPT_INTERCEPTED_REQUESTS))
1456 log_error(LOG_LEVEL_ERROR, "%s's request: \'%s\' is invalid."
1457 " Privoxy isn't configured to accept intercepted requests.",
1458 csp->ip_addr_str, csp->http->cmd);
1459 /* XXX: Use correct size */
1460 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" 400 0",
1461 csp->ip_addr_str, csp->http->cmd);
1463 write_socket(csp->cfd, CHEADER, strlen(CHEADER));
1464 destroy_list(headers);
1466 return JB_ERR_PARSE;
1468 else if (JB_ERR_OK == get_destination_from_headers(headers, csp->http))
1470 /* Split the domain we just got for pattern matching */
1471 init_domain_components(csp->http);
1477 /* We can't work without destination. Go spread the news.*/
1479 req = list_to_text(headers);
1481 /* XXX: Use correct size */
1482 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" 400 0",
1483 csp->ip_addr_str, csp->http->cmd);
1484 log_error(LOG_LEVEL_ERROR,
1485 "Privoxy was unable to get the destination for %s's request:\n%s\n%s",
1486 csp->ip_addr_str, csp->http->cmd, req);
1489 write_socket(csp->cfd, MISSING_DESTINATION_RESPONSE, strlen(MISSING_DESTINATION_RESPONSE));
1490 destroy_list(headers);
1492 return JB_ERR_PARSE;
1495 * TODO: If available, use PF's ioctl DIOCNATLOOK as last resort
1496 * to get the destination IP address, use it as host directly
1497 * or do a reverse DNS lookup first.
1502 /*********************************************************************
1504 * Function : get_server_headers
1506 * Description : Parses server headers in iob and fills them
1507 * into csp->headers so that they can later be
1511 * 1 : csp = Current client state (buffers, headers, etc...)
1513 * Returns : JB_ERR_OK if everything went fine, or
1514 * JB_ERR_PARSE if the headers were incomplete.
1516 *********************************************************************/
1517 static jb_err get_server_headers(struct client_state *csp)
1519 int continue_hack_in_da_house = 0;
1522 while (((header = get_header(csp->iob)) != NULL) || continue_hack_in_da_house)
1527 * continue hack in da house. Ignore the ending of
1528 * this head and continue enlisting header lines.
1529 * The reason is described below.
1531 enlist(csp->headers, "");
1532 continue_hack_in_da_house = 0;
1535 else if (0 == strncmpic(header, "HTTP/1.1 100", 12))
1538 * It's a bodyless continue response, don't
1539 * stop header parsing after reaching its end.
1541 * As a result Privoxy will concatenate the
1542 * next response's head and parse and deliver
1543 * the headers as if they belonged to one request.
1545 * The client will separate them because of the
1546 * empty line between them.
1548 * XXX: What we're doing here is clearly against
1549 * the intended purpose of the continue header,
1550 * and under some conditions (HTTP/1.0 client request)
1551 * it's a standard violation.
1553 * Anyway, "sort of against the spec" is preferable
1554 * to "always getting confused by Continue responses"
1555 * (Privoxy's behaviour before this hack was added)
1557 log_error(LOG_LEVEL_HEADER, "Continue hack in da house.");
1558 continue_hack_in_da_house = 1;
1560 else if (*header == '\0')
1563 * If the header is empty, but the Continue hack
1564 * isn't active, we can assume that we reached the
1565 * end of the buffer before we hit the end of the
1568 * Inform the caller an let it decide how to handle it.
1570 return JB_ERR_PARSE;
1573 if (JB_ERR_MEMORY == enlist(csp->headers, header))
1576 * XXX: Should we quit the request and return a
1577 * out of memory error page instead?
1579 log_error(LOG_LEVEL_ERROR,
1580 "Out of memory while enlisting server headers. %s lost.",
1590 /*********************************************************************
1592 * Function : crunch_reason
1594 * Description : Translates the crunch reason code into a string.
1597 * 1 : rsp = a http_response
1599 * Returns : A string with the crunch reason or an error description.
1601 *********************************************************************/
1602 static const char *crunch_reason(const struct http_response *rsp)
1604 char * reason = NULL;
1606 assert(rsp != NULL);
1609 return "Internal error while searching for crunch reason";
1612 switch (rsp->reason)
1614 case RSP_REASON_UNSUPPORTED:
1615 reason = "Unsupported HTTP feature";
1617 case RSP_REASON_BLOCKED:
1620 case RSP_REASON_UNTRUSTED:
1621 reason = "Untrusted";
1623 case RSP_REASON_REDIRECTED:
1624 reason = "Redirected";
1626 case RSP_REASON_CGI_CALL:
1627 reason = "CGI Call";
1629 case RSP_REASON_NO_SUCH_DOMAIN:
1630 reason = "DNS failure";
1632 case RSP_REASON_FORWARDING_FAILED:
1633 reason = "Forwarding failed";
1635 case RSP_REASON_CONNECT_FAILED:
1636 reason = "Connection failure";
1638 case RSP_REASON_OUT_OF_MEMORY:
1639 reason = "Out of memory (may mask other reasons)";
1642 reason = "No reason recorded";
1650 /*********************************************************************
1652 * Function : send_crunch_response
1654 * Description : Delivers already prepared response for
1655 * intercepted requests, logs the interception
1656 * and frees the response.
1659 * 1 : csp = Current client state (buffers, headers, etc...)
1660 * 1 : rsp = Fully prepared response. Will be freed on exit.
1662 * Returns : Nothing.
1664 *********************************************************************/
1665 static void send_crunch_response(const struct client_state *csp, struct http_response *rsp)
1667 const struct http_request *http = csp->http;
1668 char status_code[4];
1670 assert(rsp != NULL);
1671 assert(rsp->head != NULL);
1676 * Not supposed to happen. If it does
1677 * anyway, treat it as an unknown error.
1679 cgi_error_unknown(csp, rsp, RSP_REASON_INTERNAL_ERROR);
1680 /* return code doesn't matter */
1685 /* If rsp is still NULL, we have serious internal problems. */
1686 log_error(LOG_LEVEL_FATAL,
1687 "NULL response in send_crunch_response and cgi_error_unknown failed as well.");
1691 * Extract the status code from the actual head
1692 * that was send to the client. It is the only
1693 * way to get it right for all requests, including
1694 * the fixed ones for out-of-memory problems.
1696 * A head starts like this: 'HTTP/1.1 200...'
1700 status_code[0] = rsp->head[9];
1701 status_code[1] = rsp->head[10];
1702 status_code[2] = rsp->head[11];
1703 status_code[3] = '\0';
1705 /* Write the answer to the client */
1706 if (write_socket(csp->cfd, rsp->head, rsp->head_length)
1707 || write_socket(csp->cfd, rsp->body, rsp->content_length))
1709 /* There is nothing we can do about it. */
1710 log_error(LOG_LEVEL_ERROR, "write to: %s failed: %E", csp->http->host);
1713 /* Log that the request was crunched and why. */
1714 log_error(LOG_LEVEL_GPC, "%s%s crunch! (%s)",
1715 http->hostport, http->path, crunch_reason(rsp));
1716 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" %s %d",
1717 csp->ip_addr_str, http->ocmd, status_code, rsp->content_length);
1719 /* Clean up and return */
1720 if (cgi_error_memory() != rsp)
1722 free_http_response(rsp);
1729 /*********************************************************************
1731 * Function : request_contains_null_bytes
1733 * Description : Checks for NULL bytes in the request and sends
1734 * an error message to the client if any were found.
1736 * XXX: currently not used, see comment in chat().
1739 * 1 : csp = Current client state (buffers, headers, etc...)
1740 * 2 : buf = Data from the client's request to check.
1741 * 3 : len = The data length.
1743 * Returns : TRUE if the request contained one or more NULL bytes, or
1746 *********************************************************************/
1747 static int request_contains_null_bytes(const struct client_state *csp, char *buf, int len)
1749 size_t c_len; /* Request lenght when treated as C string */
1751 c_len = strlen(buf);
1756 * Null byte(s) found. Log the request,
1757 * return an error response and hang up.
1759 size_t tmp_len = c_len;
1764 * Replace NULL byte(s) with '°' characters
1765 * so the request can be logged as string.
1766 * XXX: Is there a better replacement character?
1769 tmp_len += strlen(buf+tmp_len);
1770 } while (tmp_len < len);
1772 log_error(LOG_LEVEL_ERROR, "%s\'s request contains at least one NULL byte "
1773 "(length=%d, strlen=%d).", csp->ip_addr_str, len, c_len);
1774 log_error(LOG_LEVEL_HEADER,
1775 "Offending request data with NULL bytes turned into \'°\' characters: %s", buf);
1777 write_socket(csp->cfd, NULL_BYTE_RESPONSE, strlen(NULL_BYTE_RESPONSE));
1779 /* XXX: Log correct size */
1780 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"Invalid request\" 400 0", csp->ip_addr_str);
1790 /*********************************************************************
1792 * Function : crunch_response_triggered
1794 * Description : Checks if the request has to be crunched,
1795 * and delivers the crunch response if necessary.
1798 * 1 : csp = Current client state (buffers, headers, etc...)
1799 * 2 : crunchers = list of cruncher functions to run
1801 * Returns : TRUE if the request was answered with a crunch response
1804 *********************************************************************/
1805 static int crunch_response_triggered(struct client_state *csp, const struct cruncher crunchers[])
1807 struct http_response *rsp = NULL;
1808 const struct cruncher *c;
1811 * If CGI request crunching is disabled,
1812 * check the CGI dispatcher out of order to
1813 * prevent unintentional blocks or redirects.
1815 if (!(csp->config->feature_flags & RUNTIME_FEATURE_CGI_CRUNCHING)
1816 && (NULL != (rsp = dispatch_cgi(csp))))
1818 /* Deliver, log and free the interception response. */
1819 send_crunch_response(csp, rsp);
1823 for (c = crunchers; c->cruncher != NULL; c++)
1826 * Check the cruncher if either Privoxy is toggled
1827 * on and the request isn't forced, or if the cruncher
1828 * applies to forced requests as well.
1830 if (((csp->flags & CSP_FLAG_TOGGLED_ON) &&
1831 !(csp->flags & CSP_FLAG_FORCED)) ||
1832 (c->flags & CF_IGNORE_FORCE))
1834 rsp = c->cruncher(csp);
1837 /* Deliver, log and free the interception response. */
1838 send_crunch_response(csp, rsp);
1839 #ifdef FEATURE_STATISTICS
1840 if (c->flags & CF_COUNT_AS_REJECT)
1842 csp->flags |= CSP_FLAG_REJECTED;
1844 #endif /* def FEATURE_STATISTICS */
1855 /*********************************************************************
1857 * Function : build_request_line
1859 * Description : Builds the HTTP request line.
1861 * If a HTTP forwarder is used it expects the whole URL,
1862 * web servers only get the path.
1865 * 1 : csp = Current client state (buffers, headers, etc...)
1866 * 2 : fwd = The forwarding spec used for the request
1867 * XXX: Should use http->fwd instead.
1868 * 3 : request_line = The old request line which will be replaced.
1870 * Returns : Nothing. Terminates in case of memory problems.
1872 *********************************************************************/
1873 static void build_request_line(struct client_state *csp, const struct forward_spec *fwd, char **request_line)
1875 struct http_request *http = csp->http;
1877 assert(http->ssl == 0);
1880 * Downgrade http version from 1.1 to 1.0
1881 * if +downgrade action applies.
1883 if ( (csp->action->flags & ACTION_DOWNGRADE)
1884 && (!strcmpic(http->ver, "HTTP/1.1")))
1887 http->ver = strdup("HTTP/1.0");
1889 if (http->ver == NULL)
1891 log_error(LOG_LEVEL_FATAL, "Out of memory downgrading HTTP version");
1896 * Rebuild the request line.
1898 freez(*request_line);
1899 *request_line = strdup(http->gpc);
1900 string_append(request_line, " ");
1902 if (fwd->forward_host)
1904 string_append(request_line, http->url);
1908 string_append(request_line, http->path);
1910 string_append(request_line, " ");
1911 string_append(request_line, http->ver);
1913 if (*request_line == NULL)
1915 log_error(LOG_LEVEL_FATAL, "Out of memory writing HTTP command");
1917 log_error(LOG_LEVEL_HEADER, "New HTTP Request-Line: %s", *request_line);
1921 /*********************************************************************
1923 * Function : change_request_destination
1925 * Description : Parse a (rewritten) request line and regenerate
1926 * the http request data.
1929 * 1 : csp = Current client state (buffers, headers, etc...)
1931 * Returns : Forwards the parse_http_request() return code.
1932 * Terminates in case of memory problems.
1934 *********************************************************************/
1935 static jb_err change_request_destination(struct client_state *csp)
1937 struct http_request *http = csp->http;
1940 log_error(LOG_LEVEL_INFO, "Rewrite detected: %s", csp->headers->first->str);
1941 free_http_request(http);
1942 err = parse_http_request(csp->headers->first->str, http, csp);
1943 if (JB_ERR_OK != err)
1945 log_error(LOG_LEVEL_ERROR, "Couldn't parse rewritten request: %s.",
1946 jb_err_to_string(err));
1950 /* XXX: ocmd is a misleading name */
1951 http->ocmd = strdup(http->cmd);
1952 if (http->ocmd == NULL)
1954 log_error(LOG_LEVEL_FATAL,
1955 "Out of memory copying rewritten HTTP request line");
1963 /*********************************************************************
1967 * Description : Once a connection to the client has been accepted,
1968 * this function is called (via serve()) to handle the
1969 * main business of the communication. When this
1970 * function returns, the caller must close the client
1973 * FIXME: chat is nearly thousand lines long.
1977 * 1 : csp = Current client state (buffers, headers, etc...)
1979 * Returns : Nothing.
1981 *********************************************************************/
1982 static void chat(struct client_state *csp)
1984 char buf[BUFFER_SIZE];
1992 int ms_iis5_hack = 0;
1993 size_t byte_count = 0;
1994 int forwarded_connect_retries = 0;
1995 int max_forwarded_connect_retries = csp->config->forwarded_connect_retries;
1996 const struct forward_spec * fwd;
1997 struct http_request *http;
1998 int len; /* for buffer sizes (and negative error codes) */
2001 /* Function that does the content filtering for the current request */
2002 filter_function_ptr content_filter = NULL;
2004 /* Skeleton for HTTP response, if we should intercept the request */
2005 struct http_response *rsp;
2007 /* Temporary copy of the client's headers before they get enlisted in csp->headers */
2008 struct list header_list;
2009 struct list *headers = &header_list;
2013 memset(buf, 0, sizeof(buf));
2016 * Read the client's request. Note that since we're not using select() we
2017 * could get blocked here if a client connected, then didn't say anything!
2022 len = read_socket(csp->cfd, buf, sizeof(buf) - 1);
2024 if (len <= 0) break; /* error! */
2027 * If there is no memory left for buffering the
2028 * request, there is nothing we can do but hang up
2030 if (add_to_iob(csp, buf, len))
2035 req = get_header(csp->iob);
2037 } while ((NULL != req) && ('\0' == *req));
2039 if ((NULL != req) && ('\0' != *req))
2041 /* Request received. Validate and parse it. */
2045 * XXX: Temporary disabled to prevent problems
2046 * with POST requests whose bodies are allowed to
2047 * contain NULL bytes. BR#1730105.
2049 * The main purpose of this check is to properly
2050 * log stuff like BitTorrent traffic and other junk
2051 * that hits public proxies. It's not required for
2052 * Privoxy to functions as those requests are discarded
2055 * It probably should be rewritten to only check
2056 * the head of the request. Another option would
2057 * be to let all POST requests pass, although that
2058 * may not be good enough.
2060 if (request_contains_null_bytes(csp, buf, len))
2062 /* NULL bytes found and dealt with, just hang up. */
2067 /* Does the request line look invalid? */
2068 if (client_protocol_is_unsupported(csp, req))
2071 * Yes. The request has already been
2072 * answered with a error response, the buffers
2073 * were freed and we're done with chatting.
2078 #ifdef FEATURE_FORCE_LOAD
2080 * If this request contains the FORCE_PREFIX and blocks
2081 * aren't enforced, get rid of it and set the force flag.
2083 if (strstr(req, FORCE_PREFIX))
2085 if (csp->config->feature_flags & RUNTIME_FEATURE_ENFORCE_BLOCKS)
2087 log_error(LOG_LEVEL_FORCE,
2088 "Ignored force prefix in request: \"%s\".", req);
2092 strclean(req, FORCE_PREFIX);
2093 log_error(LOG_LEVEL_FORCE, "Enforcing request: \"%s\".", req);
2094 csp->flags |= CSP_FLAG_FORCED;
2098 #endif /* def FEATURE_FORCE_LOAD */
2099 err = parse_http_request(req, http, csp);
2100 if (JB_ERR_OK != err)
2102 log_error(LOG_LEVEL_ERROR, "Couldn't parse request: %s.", jb_err_to_string(err));
2108 if (http->cmd == NULL)
2110 write_socket(csp->cfd, CHEADER, strlen(CHEADER));
2111 /* XXX: Use correct size */
2112 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"Invalid request\" 400 0", csp->ip_addr_str);
2113 log_error(LOG_LEVEL_ERROR, "Invalid header received from %s.", csp->ip_addr_str);
2115 free_http_request(http);
2119 /* grab the rest of the client's headers */
2123 p = get_header(csp->iob);
2127 /* There are no additional headers to read. */
2134 * We didn't receive a complete header
2135 * line yet, get the rest of it.
2137 len = read_socket(csp->cfd, buf, sizeof(buf) - 1);
2140 log_error(LOG_LEVEL_ERROR, "read from client failed: %E");
2141 destroy_list(headers);
2145 if (add_to_iob(csp, buf, len))
2148 * If there is no memory left for buffering the
2149 * request, there is nothing we can do but hang up
2151 destroy_list(headers);
2158 * We were able to read a complete
2159 * header and can finaly enlist it.
2166 if (http->host == NULL)
2169 * If we still don't know the request destination,
2170 * the request is invalid or the client uses
2171 * Privoxy without its knowledge.
2173 if (JB_ERR_OK != get_request_destination_elsewhere(csp, headers))
2176 * Our attempts to get the request destination
2177 * elsewhere failed or Privoxy is configured
2178 * to only accept proxy requests.
2180 * An error response has already been send
2181 * and we're done here.
2188 * Determine the actions for this URL
2190 #ifdef FEATURE_TOGGLE
2191 if (!(csp->flags & CSP_FLAG_TOGGLED_ON))
2193 /* Most compatible set of actions (i.e. none) */
2194 init_current_action(csp->action);
2197 #endif /* ndef FEATURE_TOGGLE */
2199 get_url_actions(csp, http);
2203 * Save a copy of the original request for logging
2205 http->ocmd = strdup(http->cmd);
2207 if (http->ocmd == NULL)
2209 log_error(LOG_LEVEL_FATAL, "Out of memory copying HTTP request line");
2212 enlist(csp->headers, http->cmd);
2214 /* Append the previously read headers */
2215 list_append_list_unique(csp->headers, headers);
2216 destroy_list(headers);
2218 err = sed(csp, FILTER_CLIENT_HEADERS);
2219 if (JB_ERR_OK != err)
2221 assert(err == JB_ERR_PARSE);
2222 log_error(LOG_LEVEL_FATAL, "Failed to parse client headers");
2224 csp->flags |= CSP_FLAG_CLIENT_HEADER_PARSING_DONE;
2226 /* Check request line for rewrites. */
2227 if ((NULL == csp->headers->first->str)
2228 || (strcmp(http->cmd, csp->headers->first->str) &&
2229 (JB_ERR_OK != change_request_destination(csp))))
2232 * A header filter broke the request line - bail out.
2234 write_socket(csp->cfd, MESSED_UP_REQUEST_RESPONSE, strlen(MESSED_UP_REQUEST_RESPONSE));
2235 /* XXX: Use correct size */
2236 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"Invalid request generated\" 500 0", csp->ip_addr_str);
2237 log_error(LOG_LEVEL_ERROR, "Invalid request line after applying header filters.");
2239 free_http_request(http);
2243 /* decide how to route the HTTP request */
2244 fwd = forward_url(csp, http);
2247 log_error(LOG_LEVEL_FATAL, "gateway spec is NULL!?!? This can't happen!");
2248 /* Never get here - LOG_LEVEL_FATAL causes program exit */
2252 * build the http request to send to the server
2253 * we have to do one of the following:
2255 * create = use the original HTTP request to create a new
2256 * HTTP request that has either the path component
2257 * without the http://domainspec (w/path) or the
2258 * full orininal URL (w/url)
2259 * Note that the path and/or the HTTP version may
2260 * have been altered by now.
2262 * connect = Open a socket to the host:port of the server
2263 * and short-circuit server and client socket.
2265 * pass = Pass the request unchanged if forwarding a CONNECT
2266 * request to a parent proxy. Note that we'll be sending
2267 * the CFAIL message ourselves if connecting to the parent
2268 * fails, but we won't send a CSUCCEED message if it works,
2269 * since that would result in a double message (ours and the
2270 * parent's). After sending the request to the parent, we simply
2273 * here's the matrix:
2276 * +--------+--------+
2278 * 0 | create | connect|
2280 * Forwarding +--------+--------+
2282 * 1 | create | pass |
2284 * +--------+--------+
2288 if (http->ssl && connect_port_is_forbidden(csp))
2290 const char *acceptable_connect_ports =
2291 csp->action->string[ACTION_STRING_LIMIT_CONNECT];
2292 assert(NULL != acceptable_connect_ports);
2293 log_error(LOG_LEVEL_INFO, "Request from %s marked for blocking. "
2294 "limit-connect{%s} doesn't allow CONNECT requests to port %d.",
2295 csp->ip_addr_str, acceptable_connect_ports, csp->http->port);
2296 csp->action->flags |= ACTION_BLOCK;
2302 freez(csp->headers->first->str);
2303 build_request_line(csp, fwd, &csp->headers->first->str);
2306 hdr = list_to_text(csp->headers);
2309 /* FIXME Should handle error properly */
2310 log_error(LOG_LEVEL_FATAL, "Out of memory parsing client header");
2314 * We have a request. Check if one of the crunchers wants it.
2316 if (crunch_response_triggered(csp, crunchers_all))
2319 * Yes. The client got the crunch response
2320 * and we are done here after cleaning up.
2323 list_remove_all(csp->headers);
2329 * The headers can't be removed earlier because
2330 * they were still needed for the referrer check
2331 * in case of CGI crunches.
2333 * XXX: Would it be worth to move the referrer check
2334 * into client_referrer() and set a flag if it's trusted?
2336 list_remove_all(csp->headers);
2338 log_error(LOG_LEVEL_GPC, "%s%s", http->hostport, http->path);
2340 if (fwd->forward_host)
2342 log_error(LOG_LEVEL_CONNECT, "via %s:%d to: %s",
2343 fwd->forward_host, fwd->forward_port, http->hostport);
2347 log_error(LOG_LEVEL_CONNECT, "to %s", http->hostport);
2350 /* here we connect to the server, gateway, or the forwarder */
2352 while ( (csp->sfd = forwarded_connect(fwd, http, csp))
2353 && (errno == EINVAL) && (forwarded_connect_retries++ < max_forwarded_connect_retries))
2355 log_error(LOG_LEVEL_ERROR, "failed request #%u to connect to %s. Trying again.",
2356 forwarded_connect_retries, http->hostport);
2359 if (csp->sfd == JB_INVALID_SOCKET)
2361 if (fwd->type != SOCKS_NONE)
2364 rsp = error_response(csp, "forwarding-failed", errno);
2366 else if (errno == EINVAL)
2368 rsp = error_response(csp, "no-such-domain", errno);
2372 rsp = error_response(csp, "connect-failed", errno);
2373 log_error(LOG_LEVEL_CONNECT, "connect to: %s failed: %E",
2378 /* Write the answer to the client */
2381 send_crunch_response(csp, rsp);
2388 if (fwd->forward_host || (http->ssl == 0))
2390 /* write the client's (modified) header to the server
2391 * (along with anything else that may be in the buffer)
2394 if (write_socket(csp->sfd, hdr, strlen(hdr))
2395 || (flush_socket(csp->sfd, csp->iob) < 0))
2397 log_error(LOG_LEVEL_CONNECT, "write header to: %s failed: %E",
2400 rsp = error_response(csp, "connect-failed", errno);
2404 send_crunch_response(csp, rsp);
2414 * We're running an SSL tunnel and we're not forwarding,
2415 * so just send the "connect succeeded" message to the
2416 * client, flush the rest, and get out of the way.
2418 if (write_socket(csp->cfd, CSUCCEED, strlen(CSUCCEED)))
2426 log_error(LOG_LEVEL_CONNECT, "to %s successful", http->hostport);
2428 /* we're finished with the client's header */
2431 maxfd = ( csp->cfd > csp->sfd ) ? csp->cfd : csp->sfd;
2433 /* pass data between the client and server
2434 * until one or the other shuts down the connection.
2443 * FD_ZERO here seems to point to an errant macro which crashes.
2444 * So do this by hand for now...
2446 memset(&rfds,0x00,sizeof(fd_set));
2450 FD_SET(csp->cfd, &rfds);
2451 FD_SET(csp->sfd, &rfds);
2453 n = select((int)maxfd+1, &rfds, NULL, NULL, NULL);
2457 log_error(LOG_LEVEL_ERROR, "select() failed!: %E");
2461 /* this is the body of the browser's request
2462 * just read it and write it.
2465 if (FD_ISSET(csp->cfd, &rfds))
2467 len = read_socket(csp->cfd, buf, sizeof(buf) - 1);
2471 break; /* "game over, man" */
2474 if (write_socket(csp->sfd, buf, (size_t)len))
2476 log_error(LOG_LEVEL_ERROR, "write to: %s failed: %E", http->host);
2483 * The server wants to talk. It could be the header or the body.
2484 * If `hdr' is null, then it's the header otherwise it's the body.
2485 * FIXME: Does `hdr' really mean `host'? No.
2489 if (FD_ISSET(csp->sfd, &rfds))
2492 len = read_socket(csp->sfd, buf, sizeof(buf) - 1);
2496 log_error(LOG_LEVEL_ERROR, "read from: %s failed: %E", http->host);
2498 if (http->ssl && (fwd->forward_host == NULL))
2501 * Just hang up. We already confirmed the client's CONNECT
2502 * request with status code 200 and unencrypted content is
2503 * no longer welcome.
2505 log_error(LOG_LEVEL_ERROR,
2506 "CONNECT already confirmed. Unable to tell the client about the problem.");
2509 else if (byte_count)
2512 * Just hang up. We already transmitted the original headers
2513 * and parts of the original content and therefore missed the
2514 * chance to send an error message (without risking data corruption).
2516 * XXX: we could retry with a fancy range request here.
2518 log_error(LOG_LEVEL_ERROR, "Already forwarded the original headers. "
2519 "Unable to tell the client about the problem.");
2523 rsp = error_response(csp, "connect-failed", errno);
2527 send_crunch_response(csp, rsp);
2533 /* Add a trailing zero. This lets filter_popups
2534 * use string operations.
2538 /* Normally, this would indicate that we've read
2539 * as much as the server has sent us and we can
2540 * close the client connection. However, Microsoft
2541 * in its wisdom has released IIS/5 with a bug that
2542 * prevents it from sending the trailing \r\n in
2543 * a 302 redirect header (and possibly other headers).
2544 * To work around this if we've haven't parsed
2545 * a full header we'll append a trailing \r\n
2546 * and see if this now generates a valid one.
2548 * This hack shouldn't have any impacts. If we've
2549 * already transmitted the header or if this is a
2550 * SSL connection, then we won't bother with this
2551 * hack. So we only work on partially received
2552 * headers. If we append a \r\n and this still
2553 * doesn't generate a valid header, then we won't
2554 * transmit anything to the client.
2559 if (server_body || http->ssl)
2562 * If we have been buffering up the document,
2563 * now is the time to apply content modification
2564 * and send the result to the client.
2568 p = execute_content_filter(csp, content_filter);
2570 * If the content filter fails, use the original
2571 * buffer and length.
2572 * (see p != NULL ? p : csp->iob->cur below)
2576 csp->content_length = (size_t)(csp->iob->eod - csp->iob->cur);
2579 if (JB_ERR_OK != update_server_headers(csp))
2581 log_error(LOG_LEVEL_FATAL,
2582 "Failed to update server headers. after filtering.");
2585 hdr = list_to_text(csp->headers);
2588 /* FIXME Should handle error properly */
2589 log_error(LOG_LEVEL_FATAL, "Out of memory parsing server header");
2592 if (write_socket(csp->cfd, hdr, strlen(hdr))
2593 || write_socket(csp->cfd, p != NULL ? p : csp->iob->cur, csp->content_length))
2595 log_error(LOG_LEVEL_ERROR, "write modified content to client failed: %E");
2605 break; /* "game over, man" */
2609 * This is NOT the body, so
2610 * Let's pretend the server just sent us a blank line.
2612 snprintf(buf, sizeof(buf), "\r\n");
2613 len = (int)strlen(buf);
2616 * Now, let the normal header parsing algorithm below do its
2617 * job. If it fails, we'll exit instead of continuing.
2624 * If this is an SSL connection or we're in the body
2625 * of the server document, just write it to the client,
2626 * unless we need to buffer the body for later content-filtering
2629 if (server_body || http->ssl)
2634 * If there is no memory left for buffering the content, or the buffer limit
2635 * has been reached, switch to non-filtering mode, i.e. make & write the
2636 * header, flush the iob and buf, and get out of the way.
2638 if (add_to_iob(csp, buf, len))
2643 log_error(LOG_LEVEL_INFO, "Flushing header and buffers. Stepping back from filtering.");
2645 hdr = list_to_text(csp->headers);
2649 * Memory is too tight to even generate the header.
2650 * Send our static "Out-of-memory" page.
2652 log_error(LOG_LEVEL_ERROR, "Out of memory while trying to flush.");
2653 rsp = cgi_error_memory();
2654 send_crunch_response(csp, rsp);
2658 hdrlen = strlen(hdr);
2660 if (write_socket(csp->cfd, hdr, hdrlen)
2661 || ((flushed = flush_socket(csp->cfd, csp->iob)) < 0)
2662 || (write_socket(csp->cfd, buf, (size_t)len)))
2664 log_error(LOG_LEVEL_CONNECT, "Flush header and buffers to client failed: %E");
2671 * Reset the byte_count to the amount of bytes
2672 * we just flushed. len will be added a few lines below,
2673 * hdrlen doesn't matter for LOG_LEVEL_CLF.
2675 byte_count = (size_t)flushed;
2677 content_filter = NULL;
2683 if (write_socket(csp->cfd, buf, (size_t)len))
2685 log_error(LOG_LEVEL_ERROR, "write to client failed: %E");
2689 byte_count += (size_t)len;
2694 /* we're still looking for the end of the
2695 * server's header ... (does that make header
2696 * parsing an "out of body experience" ?
2700 * buffer up the data we just read. If that fails,
2701 * there's little we can do but send our static
2702 * out-of-memory page.
2704 if (add_to_iob(csp, buf, len))
2706 log_error(LOG_LEVEL_ERROR, "Out of memory while looking for end of server headers.");
2707 rsp = cgi_error_memory();
2708 send_crunch_response(csp, rsp);
2713 /* Convert iob into something sed() can digest */
2714 if (JB_ERR_PARSE == get_server_headers(csp))
2718 /* Well, we tried our MS IIS/5
2719 * hack and it didn't work.
2720 * The header is incomplete
2721 * and there isn't anything
2722 * we can do about it.
2724 log_error(LOG_LEVEL_INFO,
2725 "MS IIS5 hack didn't produce valid headers.");
2730 /* Since we have to wait for
2731 * more from the server before
2732 * we can parse the headers
2733 * we just continue here.
2739 /* Did we actually get anything? */
2740 if (NULL == csp->headers->first)
2742 log_error(LOG_LEVEL_ERROR, "Empty server or forwarder response.");
2743 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" 502 0", csp->ip_addr_str, http->cmd);
2744 write_socket(csp->cfd, NO_SERVER_DATA_RESPONSE, strlen(NO_SERVER_DATA_RESPONSE));
2745 free_http_request(http);
2749 assert(csp->headers->first->str);
2751 if (strncmpic(csp->headers->first->str, "HTTP", 4) &&
2752 strncmpic(csp->headers->first->str, "ICY", 3))
2755 * It doesn't look like a HTTP (or Shoutcast) response:
2756 * tell the client and log the problem.
2758 if (strlen(csp->headers->first->str) > 30)
2760 csp->headers->first->str[30] = '\0';
2762 log_error(LOG_LEVEL_ERROR,
2763 "Invalid server or forwarder response. Starts with: %s",
2764 csp->headers->first->str);
2765 log_error(LOG_LEVEL_CLF,
2766 "%s - - [%T] \"%s\" 502 0", csp->ip_addr_str, http->cmd);
2767 write_socket(csp->cfd, INVALID_SERVER_HEADERS_RESPONSE,
2768 strlen(INVALID_SERVER_HEADERS_RESPONSE));
2769 free_http_request(http);
2774 * We have now received the entire server header,
2775 * filter it and send the result to the client
2777 if (JB_ERR_OK != sed(csp, FILTER_SERVER_HEADERS))
2779 log_error(LOG_LEVEL_FATAL, "Failed to parse server headers.");
2781 hdr = list_to_text(csp->headers);
2784 /* FIXME Should handle error properly */
2785 log_error(LOG_LEVEL_FATAL, "Out of memory parsing server header");
2788 if (crunch_response_triggered(csp, crunchers_light))
2791 * One of the tags created by a server-header
2792 * tagger triggered a crunch. We already
2793 * delivered the crunch response to the client
2794 * and are done here after cleaning up.
2799 /* Buffer and pcrs filter this if appropriate. */
2801 if (!http->ssl) /* We talk plaintext */
2803 content_filter = get_filter_function(csp);
2806 * Only write if we're not buffering for content modification
2808 if (!content_filter)
2810 /* write the server's (modified) header to
2811 * the client (along with anything else that
2812 * may be in the buffer)
2815 if (write_socket(csp->cfd, hdr, strlen(hdr))
2816 || ((len = flush_socket(csp->cfd, csp->iob)) < 0))
2818 log_error(LOG_LEVEL_CONNECT, "write header to client failed: %E");
2820 /* the write failed, so don't bother
2821 * mentioning it to the client...
2822 * it probably can't hear us anyway.
2828 byte_count += (size_t)len;
2831 /* we're finished with the server's header */
2836 /* If this was a MS IIS/5 hack then it means
2837 * the server has already closed the
2838 * connection. Nothing more to read. Time
2843 log_error(LOG_LEVEL_INFO,
2844 "Closed server connection detected with MS IIS5 hack enabled.");
2851 return; /* huh? we should never get here */
2854 if (csp->content_length == 0)
2857 * If Privoxy didn't recalculate the
2858 * Content-Lenght, byte_count is still
2861 csp->content_length = byte_count;
2864 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" 200 %d",
2865 csp->ip_addr_str, http->ocmd, csp->content_length);
2869 /*********************************************************************
2873 * Description : This is little more than chat. We only "serve" to
2874 * to close any socket that chat may have opened.
2877 * 1 : csp = Current client state (buffers, headers, etc...)
2881 *********************************************************************/
2883 void serve(struct client_state *csp)
2884 #else /* ifndef AMIGA */
2885 static void serve(struct client_state *csp)
2886 #endif /* def AMIGA */
2889 close_socket(csp->cfd);
2891 if (csp->sfd != JB_INVALID_SOCKET)
2893 close_socket(csp->sfd);
2896 csp->flags &= ~CSP_FLAG_ACTIVE;
2902 /*********************************************************************
2904 * Function : server_thread
2906 * Description : We only exist to call `serve' in a threaded environment.
2909 * 1 : data = Current client state (buffers, headers, etc...)
2911 * Returns : Always 0.
2913 *********************************************************************/
2914 static int32 server_thread(void *data)
2916 serve((struct client_state *) data);
2923 #if !defined(_WIN32) || defined(_WIN_CONSOLE)
2924 /*********************************************************************
2928 * Description : Print usage info & exit.
2930 * Parameters : Pointer to argv[0] for identifying ourselves
2934 *********************************************************************/
2935 static void usage(const char *myname)
2937 printf("Privoxy version " VERSION " (" HOME_PAGE_URL ")\n"
2941 #endif /* defined(unix) */
2944 "[--no-daemon] [--pidfile pidfile] [--pre-chroot-nslookup hostname] [--user user[.group]] "
2945 #endif /* defined(unix) */
2946 "[--version] [configfile]\n"
2947 "Aborting\n", myname);
2952 #endif /* #if !defined(_WIN32) || defined(_WIN_CONSOLE) */
2955 /*********************************************************************
2957 * Function : initialize_mutexes
2959 * Description : Prepares mutexes if mutex support is available.
2963 * Returns : Void, exits in case of errors.
2965 *********************************************************************/
2966 static void initialize_mutexes(void)
2970 #ifdef FEATURE_PTHREAD
2972 * Prepare global mutex semaphores
2974 err = pthread_mutex_init(&log_mutex, 0);
2976 if (!err) err = pthread_mutex_init(&log_init_mutex, 0);
2979 * XXX: The assumptions below are a bit naive
2980 * and can cause locks that aren't necessary.
2982 * For example older FreeBSD versions (< 6.x?)
2983 * have no gethostbyname_r, but gethostbyname is
2986 #if !defined(HAVE_GETHOSTBYADDR_R) || !defined(HAVE_GETHOSTBYNAME_R)
2987 if (!err) err = pthread_mutex_init(&resolver_mutex, 0);
2988 #endif /* !defined(HAVE_GETHOSTBYADDR_R) || !defined(HAVE_GETHOSTBYNAME_R) */
2990 * XXX: should we use a single mutex for
2991 * localtime() and gmtime() as well?
2993 #ifndef HAVE_GMTIME_R
2994 if (!err) err = pthread_mutex_init(&gmtime_mutex, 0);
2995 #endif /* ndef HAVE_GMTIME_R */
2997 #ifndef HAVE_LOCALTIME_R
2998 if (!err) err = pthread_mutex_init(&localtime_mutex, 0);
2999 #endif /* ndef HAVE_GMTIME_R */
3002 if (!err) err = pthread_mutex_init(&rand_mutex, 0);
3003 #endif /* ndef HAVE_RANDOM */
3004 #endif /* FEATURE_PTHREAD */
3007 * TODO: mutex support for mingw32 would be swell.
3012 printf("Fatal error. Mutex initialization failed: %s.\n",
3021 /*********************************************************************
3025 * Description : Load the config file and start the listen loop.
3026 * This function is a lot more *sane* with the `load_config'
3027 * and `listen_loop' functions; although it stills does
3028 * a *little* too much for my taste.
3031 * 1 : argc = Number of parameters (including $0).
3032 * 2 : argv = Array of (char *)'s to the parameters.
3034 * Returns : 1 if : can't open config file, unrecognized directive,
3035 * stats requested in multi-thread mode, can't open the
3036 * log file, can't open the jar file, listen port is invalid,
3037 * any load fails, and can't bind port.
3039 * Else main never returns, the process must be signaled
3040 * to terminate execution. Or, on Windows, use the
3041 * "File", "Exit" menu option.
3043 *********************************************************************/
3045 int real_main(int argc, const char *argv[])
3047 int main(int argc, const char *argv[])
3051 unsigned int random_seed;
3053 struct passwd *pw = NULL;
3054 struct group *grp = NULL;
3057 char *pre_chroot_nslookup_to_load_resolver = NULL;
3064 #if !defined(_WIN32)
3072 * Parse the command line arguments
3074 * XXX: simply printing usage information in case of
3075 * invalid arguments isn't particularly user friendly.
3077 while (++argc_pos < argc)
3080 /* Check to see if the service must be installed or uninstalled */
3081 if (strncmp(argv[argc_pos], "--install", 9) == 0)
3083 const char *pName = argv[argc_pos] + 9;
3086 exit( (install_service(pName)) ? 0 : 1 );
3088 else if (strncmp(argv[argc_pos], "--uninstall", + 11) == 0)
3090 const char *pName = argv[argc_pos] + 11;
3093 exit((uninstall_service(pName)) ? 0 : 1);
3095 else if (strcmp(argv[argc_pos], "--service" ) == 0)
3097 bRunAsService = TRUE;
3098 w32_set_service_cwd();
3099 atexit(w32_service_exit_notify);
3102 #endif /* defined(_WIN32) */
3105 #if !defined(_WIN32) || defined(_WIN_CONSOLE)
3107 if (strcmp(argv[argc_pos], "--help") == 0)
3112 else if(strcmp(argv[argc_pos], "--version") == 0)
3114 printf("Privoxy version " VERSION " (" HOME_PAGE_URL ")\n");
3120 else if (strcmp(argv[argc_pos], "--no-daemon" ) == 0)
3125 else if (strcmp(argv[argc_pos], "--pidfile" ) == 0)
3127 if (++argc_pos == argc) usage(argv[0]);
3128 pidfile = strdup(argv[argc_pos]);
3131 else if (strcmp(argv[argc_pos], "--user" ) == 0)
3133 if (++argc_pos == argc) usage(argv[argc_pos]);
3135 if ((NULL != (p = strchr(argv[argc_pos], '.'))) && *(p + 1) != '0')
3138 if (NULL == (grp = getgrnam(p)))
3140 log_error(LOG_LEVEL_FATAL, "Group %s not found.", p);
3144 if (NULL == (pw = getpwnam(argv[argc_pos])))
3146 log_error(LOG_LEVEL_FATAL, "User %s not found.", argv[argc_pos]);
3149 if (p != NULL) *--p = '\0';
3152 else if (strcmp(argv[argc_pos], "--pre-chroot-nslookup" ) == 0)
3154 if (++argc_pos == argc) usage(argv[0]);
3155 pre_chroot_nslookup_to_load_resolver = strdup(argv[argc_pos]);
3158 else if (strcmp(argv[argc_pos], "--chroot" ) == 0)
3162 #endif /* defined(unix) */
3164 else if (argc_pos + 1 != argc)
3167 * This is neither the last command line
3168 * option, nor was it recognized before,
3169 * therefore it must be invalid.
3175 #endif /* defined(_WIN32) && !defined(_WIN_CONSOLE) */
3177 configfile = argv[argc_pos];
3180 } /* -END- while (more arguments) */
3183 if ( *configfile != '/' )
3185 char cwd[BUFFER_SIZE];
3187 size_t abs_file_size;
3189 /* make config-filename absolute here */
3190 if (NULL == getcwd(cwd, sizeof(cwd)))
3192 perror("failed to get current working directory");
3197 abs_file_size = strlen(cwd) + strlen(configfile) + 5;
3198 basedir = strdup(cwd);
3200 if (NULL == basedir ||
3201 NULL == (abs_file = malloc(abs_file_size)))
3203 perror("malloc failed");
3206 strlcpy(abs_file, basedir, abs_file_size);
3207 strlcat(abs_file, "/", abs_file_size );
3208 strlcat(abs_file, configfile, abs_file_size);
3209 configfile = abs_file;
3211 #endif /* defined unix */
3215 clients->next = NULL;
3217 /* XXX: factor out initialising after the next stable release. */
3220 #elif defined(_WIN32)
3224 /* Prepare mutexes if supported and necessary. */
3225 initialize_mutexes();
3227 /* Enable logging until further notice. */
3228 init_log_module(Argv[0]);
3230 random_seed = (unsigned int)time(NULL);
3232 srandom(random_seed);
3233 #elif defined (_WIN32)
3235 * See pick_from_range() in miscutil.c for details.
3237 log_error(LOG_LEVEL_INFO,
3238 "No thread-safe PRNG implemented for your platform. "
3239 "Using weak \'randomization\' factor which will "
3240 "limit the already questionable usefulness of "
3241 "header-time-randomizing actions (disabled by default).");
3244 #endif /* ifdef HAVE_RANDOM */
3247 * Unix signal handling
3249 * Catch the abort, interrupt and terminate signals for a graceful exit
3250 * Catch the hangup signal so the errlog can be reopened.
3251 * Ignore the broken pipe signals (FIXME: Why?)
3253 #if !defined(_WIN32) && !defined(__OS2__) && !defined(AMIGA)
3256 const int catched_signals[] = { SIGTERM, SIGINT, SIGHUP, 0 };
3257 const int ignored_signals[] = { SIGPIPE, 0 };
3259 for (idx = 0; catched_signals[idx] != 0; idx++)
3261 #ifdef sun /* FIXME: Is it safe to check for HAVE_SIGSET instead? */
3262 if (sigset(catched_signals[idx], sig_handler) == SIG_ERR)
3264 if (signal(catched_signals[idx], sig_handler) == SIG_ERR)
3265 #endif /* ifdef sun */
3267 log_error(LOG_LEVEL_FATAL, "Can't set signal-handler for signal %d: %E", catched_signals[idx]);
3271 for (idx = 0; ignored_signals[idx] != 0; idx++)
3273 if (signal(ignored_signals[idx], SIG_IGN) == SIG_ERR)
3275 log_error(LOG_LEVEL_FATAL, "Can't set ignore-handler for signal %d: %E", ignored_signals[idx]);
3280 #else /* ifdef _WIN32 */
3281 # ifdef _WIN_CONSOLE
3283 * We *are* in a windows console app.
3284 * Print a verbose messages about FAQ's and such
3286 printf("%s", win32_blurb);
3287 # endif /* def _WIN_CONSOLE */
3288 #endif /* def _WIN32 */
3291 /* Initialize the CGI subsystem */
3292 cgi_init_error_messages();
3295 * If runnig on unix and without the --nodaemon
3296 * option, become a daemon. I.e. fork, detach
3297 * from tty and get process group leadership
3310 if ( pid < 0 ) /* error */
3315 else if ( pid != 0 ) /* parent */
3320 * must check for errors
3321 * child died due to missing files aso
3324 wpid = waitpid( pid, &status, WNOHANG );
3333 /* Should be more portable, but not as well tested */
3338 #else /* ndef __FreeBSD__ */
3340 #endif /* ndef __FreeBSD__ */
3341 fd = open("/dev/tty", O_RDONLY);
3344 /* no error check here */
3345 ioctl( fd, TIOCNOTTY,0 );
3350 * stderr (fd 2) will be closed later on, when the
3351 * log file has been parsed.
3358 } /* -END- if (!no_daemon) */
3361 * As soon as we have written the PID file, we can switch
3362 * to the user and group ID indicated by the --user option
3368 if (setgid((NULL != grp) ? grp->gr_gid : pw->pw_gid))
3370 log_error(LOG_LEVEL_FATAL, "Cannot setgid(): Insufficient permissions.");
3376 log_error(LOG_LEVEL_FATAL, "Home directory for %s undefined", pw->pw_name);
3378 /* Read the time zone file from /etc before doing chroot. */
3380 if (NULL != pre_chroot_nslookup_to_load_resolver
3381 && '\0' != pre_chroot_nslookup_to_load_resolver[0])
3383 /* Initialize resolver library. */
3384 (void) resolve_hostname_to_ip(pre_chroot_nslookup_to_load_resolver);
3386 if (chroot(pw->pw_dir) < 0)
3388 log_error(LOG_LEVEL_FATAL, "Cannot chroot to %s", pw->pw_dir);
3392 log_error(LOG_LEVEL_FATAL, "Cannot chdir /");
3395 if (setuid(pw->pw_uid))
3397 log_error(LOG_LEVEL_FATAL, "Cannot setuid(): Insufficient permissions.");
3401 char putenv_dummy[64];
3403 strlcpy(putenv_dummy, "HOME=/", sizeof(putenv_dummy));
3404 if (putenv(putenv_dummy) != 0)
3406 log_error(LOG_LEVEL_FATAL, "Cannot putenv(): HOME");
3409 snprintf(putenv_dummy, sizeof(putenv_dummy), "USER=%s", pw->pw_name);
3410 if (putenv(putenv_dummy) != 0)
3412 log_error(LOG_LEVEL_FATAL, "Cannot putenv(): USER");
3418 log_error(LOG_LEVEL_FATAL, "Cannot chroot without --user argument.");
3421 #endif /* defined unix */
3424 /* This will be FALSE unless the command line specified --service
3428 /* Yup, so now we must attempt to establish a connection
3429 * with the service dispatcher. This will only work if this
3430 * process was launched by the service control manager to
3431 * actually run as a service. If this isn't the case, i've
3432 * known it take around 30 seconds or so for the call to return.
3435 /* The StartServiceCtrlDispatcher won't return until the service is stopping */
3436 if (w32_start_service_ctrl_dispatcher(w32ServiceDispatchTable))
3438 /* Service has run, and at this point is now being stopped, so just return */
3443 printf("Warning: Failed to connect to Service Control Dispatcher\nwhen starting as a service!\n");
3445 /* An error occurred. Usually it's because --service was wrongly specified
3446 * and we were unable to connect to the Service Control Dispatcher because
3447 * it wasn't expecting us and is therefore not listening.
3449 * For now, just continue below to call the listen_loop function.
3452 #endif /* def _WIN32 */
3462 /*********************************************************************
3464 * Function : bind_port_helper
3466 * Description : Bind the listen port. Handles logging, and aborts
3470 * 1 : config = Privoxy configuration. Specifies port
3473 * Returns : Port that was opened.
3475 *********************************************************************/
3476 static jb_socket bind_port_helper(struct configuration_spec * config)
3481 if (config->haddr == NULL)
3483 log_error(LOG_LEVEL_INFO, "Listening on port %d on all IP addresses",
3488 log_error(LOG_LEVEL_INFO, "Listening on port %d on IP address %s",
3489 config->hport, config->haddr);
3492 result = bind_port(config->haddr, config->hport, &bfd);
3499 log_error(LOG_LEVEL_FATAL, "can't bind to %s:%d: "
3500 "There may be another Privoxy or some other "
3501 "proxy running on port %d",
3502 (NULL != config->haddr) ? config->haddr : "INADDR_ANY",
3503 config->hport, config->hport);
3506 log_error(LOG_LEVEL_FATAL, "can't bind to %s:%d: "
3507 "The hostname is not resolvable",
3508 (NULL != config->haddr) ? config->haddr : "INADDR_ANY", config->hport);
3511 log_error(LOG_LEVEL_FATAL, "can't bind to %s:%d: because %E",
3512 (NULL != config->haddr) ? config->haddr : "INADDR_ANY", config->hport);
3515 /* shouldn't get here */
3516 return JB_INVALID_SOCKET;
3519 config->need_bind = 0;
3526 /* Without this simple workaround we get this compiler warning from _beginthread
3527 * warning C4028: formal parameter 1 different from declaration
3529 void w32_service_listen_loop(void *p)
3533 #endif /* def _WIN32 */
3536 /*********************************************************************
3538 * Function : listen_loop
3540 * Description : bind the listen port and enter a "FOREVER" listening loop.
3546 *********************************************************************/
3547 static void listen_loop(void)
3549 struct client_state *csp = NULL;
3551 struct configuration_spec * config;
3553 config = load_config();
3555 bfd = bind_port_helper(config);
3557 #ifdef FEATURE_GRACEFUL_TERMINATION
3558 while (!g_terminate)
3563 #if !defined(FEATURE_PTHREAD) && !defined(_WIN32) && !defined(__BEOS__) && !defined(AMIGA) && !defined(__OS2__)
3564 while (waitpid(-1, NULL, WNOHANG) > 0)
3566 /* zombie children */
3568 #endif /* !defined(FEATURE_PTHREAD) && !defined(_WIN32) && !defined(__BEOS__) && !defined(AMIGA) */
3571 * Free data that was used by died threads
3577 * Re-open the errlog after HUP signal
3579 if (received_hup_signal)
3581 init_error_log(Argv[0], config->logfile);
3582 received_hup_signal = 0;
3586 if ( NULL == (csp = (struct client_state *) zalloc(sizeof(*csp))) )
3588 log_error(LOG_LEVEL_FATAL, "malloc(%d) for csp failed: %E", sizeof(*csp));
3592 csp->flags |= CSP_FLAG_ACTIVE;
3593 csp->sfd = JB_INVALID_SOCKET;
3595 csp->config = config = load_config();
3597 if ( config->need_bind )
3600 * Since we were listening to the "old port", we will not see
3601 * a "listen" param change until the next IJB request. So, at
3602 * least 1 more request must be made for us to find the new
3603 * setting. I am simply closing the old socket and binding the
3606 * Which-ever is correct, we will serve 1 more page via the
3607 * old settings. This should probably be a "show-proxy-args"
3608 * request. This should not be a so common of an operation
3609 * that this will hurt people's feelings.
3614 bfd = bind_port_helper(config);
3617 log_error(LOG_LEVEL_CONNECT, "Listening for new connections ... ");
3619 if (!accept_connection(csp, bfd))
3621 log_error(LOG_LEVEL_CONNECT, "accept failed: %E");
3634 log_error(LOG_LEVEL_CONNECT, "accepted connection from %s", csp->ip_addr_str);
3637 #ifdef FEATURE_TOGGLE
3638 if (global_toggle_state)
3639 #endif /* def FEATURE_TOGGLE */
3641 csp->flags |= CSP_FLAG_TOGGLED_ON;
3644 if (run_loader(csp))
3646 log_error(LOG_LEVEL_FATAL, "a loader failed - must exit");
3647 /* Never get here - LOG_LEVEL_FATAL causes program exit */
3651 if (block_acl(NULL,csp))
3653 log_error(LOG_LEVEL_CONNECT, "Connection from %s dropped due to ACL", csp->ip_addr_str);
3654 close_socket(csp->cfd);
3658 #endif /* def FEATURE_ACL */
3660 /* add it to the list of clients */
3661 csp->next = clients->next;
3662 clients->next = csp;
3664 if (config->multi_threaded)
3668 /* this is a switch () statment in the C preprocessor - ugh */
3669 #undef SELECTED_ONE_OPTION
3671 /* Use Pthreads in preference to native code */
3672 #if defined(FEATURE_PTHREAD) && !defined(SELECTED_ONE_OPTION)
3673 #define SELECTED_ONE_OPTION
3675 pthread_t the_thread;
3676 pthread_attr_t attrs;
3678 pthread_attr_init(&attrs);
3679 pthread_attr_setdetachstate(&attrs, PTHREAD_CREATE_DETACHED);
3680 errno = pthread_create(&the_thread, &attrs,
3681 (void * (*)(void *))serve, csp);
3682 child_id = errno ? -1 : 0;
3683 pthread_attr_destroy(&attrs);
3687 #if defined(_WIN32) && !defined(_CYGWIN) && !defined(SELECTED_ONE_OPTION)
3688 #define SELECTED_ONE_OPTION
3689 child_id = _beginthread(
3690 (void (*)(void *))serve,
3695 #if defined(__OS2__) && !defined(SELECTED_ONE_OPTION)
3696 #define SELECTED_ONE_OPTION
3697 child_id = _beginthread(
3698 (void(* _Optlink)(void*))serve,
3704 #if defined(__BEOS__) && !defined(SELECTED_ONE_OPTION)
3705 #define SELECTED_ONE_OPTION
3707 thread_id tid = spawn_thread
3708 (server_thread, "server", B_NORMAL_PRIORITY, csp);
3710 if ((tid >= 0) && (resume_thread(tid) == B_OK))
3712 child_id = (int) tid;
3721 #if defined(AMIGA) && !defined(SELECTED_ONE_OPTION)
3722 #define SELECTED_ONE_OPTION
3723 csp->cfd = ReleaseSocket(csp->cfd, -1);
3726 child_id = (int)CreateNewProcTags(NP_Entry, (ULONG)server_thread,
3727 NP_Output, Output(),
3728 NP_CloseOutput, FALSE,
3729 NP_Name, (ULONG)"privoxy child",
3733 child_id = (int)CreateNewProcTags(NP_Entry, (ULONG)server_thread,
3734 NP_Output, Output(),
3735 NP_CloseOutput, FALSE,
3736 NP_Name, (ULONG)"privoxy child",
3737 NP_StackSize, 200*1024,
3743 ((struct Task *)child_id)->tc_UserData = csp;
3744 Signal((struct Task *)child_id, SIGF_SINGLE);
3749 #if !defined(SELECTED_ONE_OPTION)
3752 /* This block is only needed when using fork().
3753 * When using threads, the server thread was
3754 * created and run by the call to _beginthread().
3756 if (child_id == 0) /* child */
3759 #ifdef FEATURE_TOGGLE
3760 int inherited_toggle_state = global_toggle_state;
3761 #endif /* def FEATURE_TOGGLE */
3766 * If we've been toggled or we've blocked the request, tell Mom
3769 #ifdef FEATURE_TOGGLE
3770 if (inherited_toggle_state != global_toggle_state)
3772 rc |= RC_FLAG_TOGGLED;
3774 #endif /* def FEATURE_TOGGLE */
3776 #ifdef FEATURE_STATISTICS
3777 if (csp->flags & CSP_FLAG_REJECTED)
3779 rc |= RC_FLAG_BLOCKED;
3781 #endif /* ndef FEATURE_STATISTICS */
3785 else if (child_id > 0) /* parent */
3787 /* in a fork()'d environment, the parent's
3788 * copy of the client socket and the CSP
3792 #if !defined(_WIN32) && !defined(__CYGWIN__)
3794 wait( &child_status );
3797 * Evaluate child's return code: If the child has
3798 * - been toggled, toggle ourselves
3799 * - blocked its request, bump up the stats counter
3802 #ifdef FEATURE_TOGGLE
3803 if (WIFEXITED(child_status) && (WEXITSTATUS(child_status) & RC_FLAG_TOGGLED))
3805 global_toggle_state = !global_toggle_state;
3807 #endif /* def FEATURE_TOGGLE */
3809 #ifdef FEATURE_STATISTICS
3811 if (WIFEXITED(child_status) && (WEXITSTATUS(child_status) & RC_FLAG_BLOCKED))
3815 #endif /* def FEATURE_STATISTICS */
3817 #endif /* !defined(_WIN32) && defined(__CYGWIN__) */
3818 close_socket(csp->cfd);
3819 csp->flags &= ~CSP_FLAG_ACTIVE;
3823 #undef SELECTED_ONE_OPTION
3824 /* end of cpp switch () */
3826 if (child_id < 0) /* failed */
3828 char buf[BUFFER_SIZE];
3830 log_error(LOG_LEVEL_ERROR, "can't fork: %E");
3832 snprintf(buf , sizeof(buf), "Privoxy: can't fork: errno = %d", errno);
3834 write_socket(csp->cfd, buf, strlen(buf));
3835 close_socket(csp->cfd);
3836 csp->flags &= ~CSP_FLAG_ACTIVE;
3847 /* NOTREACHED unless FEATURE_GRACEFUL_TERMINATION is defined */
3849 /* Clean up. Aim: free all memory (no leaks) */
3850 #ifdef FEATURE_GRACEFUL_TERMINATION
3852 log_error(LOG_LEVEL_ERROR, "Graceful termination requested");
3854 unload_current_config_file();
3855 unload_current_actions_file();
3856 unload_current_re_filterfile();
3857 #ifdef FEATURE_TRUST
3858 unload_current_trust_file();
3861 if (config->multi_threaded)
3868 } while ((clients->next != NULL) && (--i > 0));
3872 log_error(LOG_LEVEL_ERROR, "Graceful termination failed - still some live clients after 1 minute wait.");
3883 #if defined(_WIN32) && !defined(_WIN_CONSOLE)
3884 /* Cleanup - remove taskbar icon etc. */
3889 #endif /* FEATURE_GRACEFUL_TERMINATION */