1 const char jcc_rcs[] = "$Id: jcc.c,v 1.192 2008/10/11 18:19:14 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 1.192 2008/10/11 18:19:14 fabiankeil
37 * Even more chat() cosmetics.
39 * Revision 1.191 2008/10/11 18:00:14 fabiankeil
40 * Reformat some comments in chat().
42 * Revision 1.190 2008/10/11 14:58:00 fabiankeil
43 * In case of chunk-encoded content, stop reading if
44 * the buffer looks like it ends with the last chunk.
46 * Revision 1.189 2008/10/11 09:53:00 fabiankeil
47 * Let server_response_is_complete() deal properly with
48 * content that is neither buffered nor read all at once.
50 * Revision 1.188 2008/10/09 18:21:41 fabiankeil
51 * Flush work-in-progress changes to keep outgoing connections
52 * alive where possible. Incomplete and mostly #ifdef'd out.
54 * Revision 1.187 2008/09/07 12:35:05 fabiankeil
55 * Add mutex lock support for _WIN32.
57 * Revision 1.186 2008/09/04 08:13:58 fabiankeil
58 * Prepare for critical sections on Windows by adding a
59 * layer of indirection before the pthread mutex functions.
61 * Revision 1.185 2008/08/30 12:03:07 fabiankeil
62 * Remove FEATURE_COOKIE_JAR.
64 * Revision 1.184 2008/08/22 15:34:45 fabiankeil
65 * - Silence LLVM/Clang complaint.
66 * - Make received_hup_signal static.
67 * - Hide definitions for basedir, pidfile and received_hup_signal
68 * from __EMX__ as they only seem to be used in case of #ifdef unix.
70 * Revision 1.183 2008/08/21 07:09:35 fabiankeil
71 * Accept Shoutcast responses again. Problem reported
72 * and fix suggested by Stefan in #2062860.
74 * Revision 1.182 2008/06/27 11:13:56 fabiankeil
75 * Fix possible NULL-pointer dereference reported
76 * by din_a4 in #2003937. Pointy hat to me.
78 * Revision 1.181 2008/05/21 15:47:15 fabiankeil
79 * Streamline sed()'s prototype and declare
80 * the header parse and add structures static.
82 * Revision 1.180 2008/05/21 15:26:32 fabiankeil
83 * - Mark csp as immutable for send_crunch_response().
84 * - Fix comment spelling.
86 * Revision 1.179 2008/05/20 20:13:32 fabiankeil
87 * Factor update_server_headers() out of sed(), ditch the
88 * first_run hack and make server_patterns_light static.
90 * Revision 1.178 2008/05/10 13:23:38 fabiankeil
91 * Don't provide get_header() with the whole client state
92 * structure when it only needs access to csp->iob.
94 * Revision 1.177 2008/05/10 11:51:12 fabiankeil
95 * Make the "read the rest of the headers" loop a bit more readable.
97 * Revision 1.176 2008/05/10 11:37:57 fabiankeil
98 * - Instead of logging when the IIS5 hack is enabled, log when it fails.
99 * - Remove useless comment.
101 * Revision 1.175 2008/05/09 18:53:59 fabiankeil
102 * Fix comment grammar.
104 * Revision 1.174 2008/05/07 18:05:53 fabiankeil
105 * Remove the pointless buffer in client_protocol_is_unsupported().
107 * Revision 1.173 2008/05/06 15:09:00 fabiankeil
108 * Least-effort fix for bug #1821930 (reported by Lee):
109 * If the response doesn't look like HTTP,
110 * tell the client and log the problem.
112 * Revision 1.172 2008/04/16 16:38:21 fabiankeil
113 * Don't pass the whole csp structure to flush_socket()
114 * when it only needs a file descriptor and a buffer.
116 * Revision 1.171 2008/03/27 18:27:25 fabiankeil
117 * Remove kill-popups action.
119 * Revision 1.170 2008/03/06 16:33:46 fabiankeil
120 * If limit-connect isn't used, don't limit CONNECT requests to port 443.
122 * Revision 1.169 2008/03/04 18:30:39 fabiankeil
123 * Remove the treat-forbidden-connects-like-blocks action. We now
124 * use the "blocked" page for forbidden CONNECT requests by default.
126 * Revision 1.168 2008/03/02 12:25:25 fabiankeil
127 * Also use shiny new connect_port_is_forbidden() in jcc.c.
129 * Revision 1.167 2008/02/23 16:57:12 fabiankeil
130 * Rename url_actions() to get_url_actions() and let it
131 * use the standard parameter ordering.
133 * Revision 1.166 2008/02/23 16:33:43 fabiankeil
134 * Let forward_url() use the standard parameter ordering
135 * and mark its second parameter immutable.
137 * Revision 1.165 2008/02/02 19:36:56 fabiankeil
138 * Remove the "Listening ... for local connections only" log message.
139 * Whether or not remote connections are able to reach Privoxy is up
140 * to the operating system.
142 * Revision 1.164 2007/12/16 18:32:46 fabiankeil
143 * Prevent the log messages for CONNECT requests to unacceptable
144 * ports from printing the limit-connect argument as [null] if
145 * limit-connect hasn't been explicitly enabled.
147 * Revision 1.163 2007/12/13 01:47:11 david__schmidt
148 * Make sure all console-mode apps get a usage() instance
150 * Revision 1.162 2007/12/06 17:54:57 fabiankeil
151 * Reword NO_SERVER_DATA_RESPONSE to make it harder
152 * to misunderstand what the message is all about.
154 * Revision 1.161 2007/12/04 19:44:22 fabiankeil
155 * Unbreak trustfile which previously didn't work without
156 * FEATURE_TOGGLE. Fixes BR#1843585, reported by Lee.
158 * Revision 1.160 2007/11/29 18:00:29 fabiankeil
159 * Plug memory leak. Spotted by Valgrind, triggered by
160 * Privoxy-Regression-Test feeding proxyfuzz.py.
162 * Revision 1.159 2007/11/24 14:34:09 fabiankeil
163 * In the HTTP snipplets, refer to the client as client.
165 * Revision 1.158 2007/11/11 16:44:17 fabiankeil
166 * Emit a log message when activating the MS IIS5 hack.
168 * Revision 1.157 2007/11/03 17:34:49 fabiankeil
169 * Log the "weak randomization factor" warning only
170 * once for mingw32 and provide some more details.
172 * Revision 1.156 2007/11/01 18:20:58 fabiankeil
173 * Initialize log module after initializing mutexes, future
174 * deadlocks in that code should now work cross-platform.
176 * Revision 1.155 2007/10/23 20:12:45 fabiankeil
177 * Fix first CSUCCEED line to end in \r\n as required by RFC1945.
178 * Reported by Bert van Leeuwen in BR#1818808.
180 * Revision 1.154 2007/10/19 17:00:08 fabiankeil
181 * Downgrade "Flushing header and buffers" message to LOG_LEVEL_INFO.
183 * Revision 1.153 2007/10/14 14:12:41 fabiankeil
184 * When in daemon mode, close stderr after the configuration file has been
185 * parsed the first time. If logfile isn't set, stop logging. Fixes BR#897436.
187 * Revision 1.152 2007/10/04 18:03:34 fabiankeil
188 * - Fix a crash when parsing invalid requests whose first header
189 * is rejected by get_header(). Regression (re?)introduced
190 * in r1.143 by yours truly.
191 * - Move ACTION_VANILLA_WAFER handling into parsers.c's
192 * client_cookie_adder() to make sure send-vanilla-wafer can be
193 * controlled through tags (and thus regression-tested).
195 * Revision 1.151 2007/09/29 10:21:16 fabiankeil
196 * - Move get_filter_function() from jcc.c to filters.c
197 * so the filter functions can be static.
198 * - Don't bother filtering body-less responses.
200 * Revision 1.150 2007/09/28 16:39:29 fabiankeil
201 * Execute content filters through execute_content_filter().
203 * Revision 1.149 2007/09/04 15:08:48 fabiankeil
204 * Initialize req to NULL to make sure it's defined if the
205 * first read_socket() call fails. Reported by icmp30.
207 * Revision 1.148 2007/08/26 16:47:13 fabiankeil
208 * Add Stephen Gildea's --pre-chroot-nslookup patch [#1276666],
209 * extensive comments moved to user manual.
211 * Revision 1.147 2007/08/25 14:42:40 fabiankeil
212 * Don't crash if a broken header filter wiped out the request line.
214 * Revision 1.146 2007/08/20 17:09:32 fabiankeil
215 * Fix byte_count calculation in case of flushes
216 * and don't parse the server headers a second time.
218 * Revision 1.145 2007/08/19 13:13:31 fabiankeil
219 * - If there's a connection problem after we already forwarded
220 * parts of the original content, just hang up. Fixes BR#1776724.
221 * - Fix warnings about unused code on mingw32.
222 * - In case of flushes, calculate the byte count
223 * less incorrectly (I think).
225 * Revision 1.144 2007/08/11 14:43:22 fabiankeil
226 * Add some more prototypes for static functions.
228 * Revision 1.143 2007/08/05 13:58:19 fabiankeil
229 * Comment out request_contains_null_bytes() until it's used again.
231 * Revision 1.142 2007/08/05 13:50:26 fabiankeil
232 * #1763173 from Stefan Huehner: s@const static@static const@
233 * and declare some more functions static.
235 * Revision 1.141 2007/08/04 09:56:23 fabiankeil
236 * - Log rejected CONNECT requests with LOG_LEVEL_INFO
237 * and explain why they were rejected in the first place.
238 * - Fix the LOG_LEVEL_CLF message for crunches of unallowed
239 * CONNECT requests. The request line was missing.
240 * - Add two more XXX reminders as we don't have enough already.
242 * Revision 1.140 2007/07/21 11:51:36 fabiankeil
243 * As Hal noticed, checking dispatch_cgi() as the last cruncher
244 * looks like a bug if CGI requests are blocked unintentionally,
245 * so don't do it unless the user enabled the new config option
246 * "allow-cgi-request-crunching".
248 * Revision 1.139 2007/07/14 07:46:41 fabiankeil
249 * - Allow to rewrite the request destination behind the client's back.
250 * - Turn the weird-looking unconditional for loop that
251 * reads the client request into a conditional while loop.
252 * Move the stuff that only runs once out of the loop.
253 * - Move parts of chat(), server_content_type() and the
254 * necessary stuff to fix BR#1750917 into get_filter_function().
256 * Revision 1.138 2007/06/03 18:45:18 fabiankeil
257 * Temporary workaround for BR#1730105.
259 * Revision 1.137 2007/06/01 18:16:36 fabiankeil
260 * Use the same mutex for gethostbyname() and gethostbyaddr() to prevent
261 * deadlocks and crashes on OpenBSD and possibly other OS with neither
262 * gethostbyname_r() nor gethostaddr_r(). Closes BR#1729174.
263 * Thanks to Ralf Horstmann for report and solution.
265 * Revision 1.136 2007/06/01 16:41:11 fabiankeil
266 * Add forward-override{} to change the forwarding settings through
267 * action sections. This is mainly interesting to forward different
268 * clients differently (for example based on User-Agent or request
271 * Revision 1.135 2007/05/24 17:03:50 fabiankeil
272 * - Let usage() mention the --chroot parameter.
273 * - Use read_socket() consistently and always leave
274 * the last buffer byte alone, even in cases where
275 * null termination (currently) doesn't matter.
277 * Revision 1.134 2007/05/16 14:59:46 fabiankeil
278 * - Fix config file loading on Unix if no config file is specified.
279 * Since r1.97 Privoxy would always interpret the last argument as
280 * config file, even if it's a valid command line option.
281 * - Abort in case of unrecognized command line options. Closes #1719696.
282 * - Remove a bunch of unnecessary strcpy() calls (yay for c&p without thinking).
283 * - Replace the remaining strcpy() and strcat() calls with strlcpy() and strcat().
285 * Revision 1.133 2007/05/04 11:23:19 fabiankeil
286 * - Don't rerun crunchers that only depend on the request URL.
287 * - Don't count redirects and CGI requests as "blocked requests".
289 * Revision 1.132 2007/04/25 15:15:17 fabiankeil
290 * Support crunching based on tags created by server-header taggers.
292 * Revision 1.131 2007/04/22 13:24:50 fabiankeil
293 * Make HTTP snippets static (again). Add a Content-Type for those
294 * with content so the browser doesn't guess it based on the URL.
296 * Revision 1.130 2007/04/19 13:47:34 fabiankeil
297 * Move crunching and request line rebuilding out of chat().
299 * Revision 1.129 2007/04/15 16:39:20 fabiankeil
300 * Introduce tags as alternative way to specify which
301 * actions apply to a request. At the moment tags can be
302 * created based on client and server headers.
304 * Revision 1.128 2007/03/25 16:55:54 fabiankeil
305 * Don't CLF-log CONNECT requests twice.
307 * Revision 1.127 2007/03/20 13:53:17 fabiankeil
308 * Log the source address for ACL-related connection drops.
310 * Revision 1.126 2007/03/17 15:20:05 fabiankeil
311 * New config option: enforce-blocks.
313 * Revision 1.125 2007/03/09 14:12:00 fabiankeil
314 * - Move null byte check into separate function.
315 * - Don't confuse the client with error pages
316 * if a CONNECT request was already confirmed.
318 * Revision 1.124 2007/02/23 14:59:54 fabiankeil
319 * Speed up NULL byte escaping and only log the complete
320 * NULL byte requests with header debugging enabled.
322 * Revision 1.123 2007/02/21 18:42:10 fabiankeil
323 * Answer requests that contain NULL bytes with
324 * a custom response instead of waiting for more
325 * data until the client eventually hangs up.
327 * Revision 1.122 2007/02/07 11:12:02 fabiankeil
328 * - Move delivery and logging of crunched responses
329 * from chat() into send_crunch_response().
330 * - Display the reason for generating http_responses.
331 * - Log the content length for LOG_LEVEL_CLF correctly
332 * (still incorrect for some fixed responses).
333 * - Reword an incorrect comment about
334 * treat-forbidden-connects-like-blocks violating
336 * - Add some log messages.
338 * Revision 1.121 2007/01/27 10:52:56 fabiankeil
339 * Move mutex initialization into separate
340 * function and exit in case of errors.
342 * Revision 1.120 2007/01/26 14:18:42 fabiankeil
343 * - Start to reduce chat()'s line count and move
344 * parts of it into separate functions.
345 * - Add "HTTP/1.1 100 Continue" hack for BR 756734.
347 * Revision 1.119 2007/01/25 14:02:30 fabiankeil
348 * - Add Proxy-Agent header to HTTP snippets that are
349 * supposed to reach HTTP clients only.
350 * - Made a few CONNECT log messages more descriptive.
351 * - Catch completely empty server responses (as seen
352 * with Tor's fake ".noconnect" top level domain).
353 * - Use shiny new "forwarding-failed" template for socks errors.
355 * Revision 1.118 2007/01/07 07:43:43 joergs
356 * AmigaOS4 support added.
358 * Revision 1.117 2006/12/31 17:56:37 fabiankeil
359 * Added config option accept-intercepted-requests
360 * and disabled it by default.
362 * Revision 1.116 2006/12/29 19:08:22 fabiankeil
363 * Reverted parts of my last commit
364 * to keep error handling working.
366 * Revision 1.115 2006/12/29 17:38:57 fabiankeil
367 * Fixed gcc43 conversion warnings.
369 * Revision 1.114 2006/12/27 18:52:02 fabiankeil
370 * Fix -pedantic ISO C warning about converting
371 * from function pointer to object pointer.
373 * Revision 1.113 2006/12/26 17:38:50 fabiankeil
374 * Silence compiler warning I introduced with my last commit.
376 * Revision 1.112 2006/12/26 17:31:41 fabiankeil
377 * Mutex protect rand() if POSIX threading
378 * is used, warn the user if that's not possible
379 * and stop using it on _WIN32 where it could
382 * Revision 1.111 2006/12/23 16:15:06 fabiankeil
383 * Don't prevent core dumps by catching SIGABRT.
384 * It's rude and makes debugging unreasonable painful.
386 * Revision 1.110 2006/12/13 14:52:53 etresoft
387 * Fix build failure on MacOS X. Global symbols can be either static or extern, but not both.
389 * Revision 1.109 2006/12/06 19:41:40 fabiankeil
390 * Privoxy is now able to run as intercepting
391 * proxy in combination with any packet filter
392 * that does the port redirection. The destination
393 * is extracted from the "Host:" header which
394 * should be available for nearly all requests.
396 * Moved HTTP snipplets into jcc.c.
397 * Added error message for gopher proxy requests.
399 * Revision 1.108 2006/11/28 15:38:51 fabiankeil
400 * Only unlink the pidfile if it's actually used.
402 * Change order of interception checks to make
403 * it possible to block or redirect requests for
406 * Revision 1.107 2006/11/13 19:05:51 fabiankeil
407 * Make pthread mutex locking more generic. Instead of
408 * checking for OSX and OpenBSD, check for FEATURE_PTHREAD
409 * and use mutex locking unless there is an _r function
410 * available. Better safe than sorry.
412 * Fixes "./configure --disable-pthread" and should result
413 * in less threading-related problems on pthread-using platforms,
414 * but it still doesn't fix BR#1122404.
416 * Revision 1.106 2006/11/06 19:58:23 fabiankeil
417 * Move pthread.h inclusion from jcc.c to jcc.h.
418 * Fixes build on x86-freebsd1 (FreeBSD 5.4-RELEASE).
420 * Revision 1.105 2006/11/06 14:26:02 fabiankeil
421 * Don't exit after receiving the second SIGHUP on Solaris.
423 * Fixes BR 1052235, but the same problem may exist on other
424 * systems. Once 3.0.6 is out we should use sigset()
425 * where available and see if it breaks anything.
427 * Revision 1.104 2006/09/23 13:26:38 roro
428 * Replace TABs by spaces in source code.
430 * Revision 1.103 2006/09/21 12:54:43 fabiankeil
431 * Fix +redirect{}. Didn't work with -fast-redirects.
433 * Revision 1.102 2006/09/06 13:03:04 fabiankeil
434 * Respond with 400 and a short text message
435 * if the client tries to use Privoxy as FTP proxy.
437 * Revision 1.101 2006/09/06 09:23:37 fabiankeil
438 * Make number of retries in case of forwarded-connect problems
439 * a config file option (forwarded-connect-retries) and use 0 as
442 * Revision 1.100 2006/09/03 19:42:59 fabiankeil
443 * Set random(3) seed.
445 * Revision 1.99 2006/09/02 15:36:42 fabiankeil
446 * Follow the OpenBSD port's lead and protect the resolve
447 * functions on OpenBSD as well.
449 * Revision 1.98 2006/08/24 11:01:34 fabiankeil
450 * --user fix. Only use the user as group if no group is specified.
451 * Solves BR 1492612. Thanks to Spinor S. and David Laight.
453 * Revision 1.97 2006/08/18 15:23:17 david__schmidt
454 * Windows service (re-)integration
458 * --install[:service_name]
459 * --uninstall[:service_name]
462 * They work as follows:
463 * --install will create a service for you and then terminate.
464 * By default the service name will be "privoxy" (without the quotes).
465 * However you can run multiple services if you wish, just by adding
466 * a colon and then a name (no spaces).
468 * --uninstall follows the exact same rules a --install.
470 * --service is used when the program is executed by the service
471 * control manager, and in normal circumstances would never be
472 * used as a command line argument.
474 * Revision 1.96 2006/08/15 20:12:36 david__schmidt
475 * Windows service integration
477 * Revision 1.95 2006/08/03 02:46:41 david__schmidt
478 * Incorporate Fabian Keil's patch work:
479 http://www.fabiankeil.de/sourcecode/privoxy/
481 * Revision 1.94 2006/07/18 14:48:46 david__schmidt
482 * Reorganizing the repository: swapping out what was HEAD (the old 3.1 branch)
483 * with what was really the latest development (the v_3_0_branch branch)
485 * Revision 1.92.2.16 2005/04/03 20:10:50 david__schmidt
486 * Thanks to Jindrich Makovicka for a race condition fix for the log
487 * file. The race condition remains for non-pthread implementations.
488 * Reference patch #1175720.
490 * Revision 1.92.2.15 2004/10/03 12:53:32 david__schmidt
491 * Add the ability to check jpeg images for invalid
492 * lengths of comment blocks. Defensive strategy
493 * against the exploit:
494 * Microsoft Security Bulletin MS04-028
495 * Buffer Overrun in JPEG Processing (GDI+) Could
496 * Allow Code Execution (833987)
497 * Enabled with +inspect-jpegs in actions files.
499 * Revision 1.92.2.14 2003/12/12 12:52:53 oes
500 * - Fixed usage info for non-unix platforms
501 * - Fixed small cmdline parsing bug
503 * Revision 1.92.2.13 2003/11/27 19:20:27 oes
504 * Diagnostics: Now preserve the returncode of pthread_create
505 * in errno. Closes BR #775721. Thanks to Geoffrey Hausheer.
507 * Revision 1.92.2.12 2003/07/11 11:34:19 oes
508 * No longer ignore SIGCHLD. Fixes bug #769381
510 * Revision 1.92.2.11 2003/05/14 12:32:02 oes
511 * Close jarfile on graceful exit, remove stray line
513 * Revision 1.92.2.10 2003/05/08 15:13:46 oes
514 * Cosmetics: Killed a warning, a typo and an allocation left at exit
516 * Revision 1.92.2.9 2003/04/03 15:08:42 oes
517 * No longer rely on non-POSIX.1 extensions of getcwd().
520 * Revision 1.92.2.8 2003/03/31 13:12:32 oes
521 * Replaced setenv() by posix-compliant putenv()
522 * Thanks to Neil McCalden (nmcc AT users.sf.net).
524 * Revision 1.92.2.7 2003/03/17 16:48:59 oes
525 * Added chroot ability, thanks to patch by Sviatoslav Sviridov
527 * Revision 1.92.2.6 2003/03/11 11:55:00 oes
528 * Clean-up and extension of improvements for forked mode:
529 * - Child's return code now consists of flags RC_FLAG_*
530 * - Reporting toggle to parent now properly #ifdef'ed
531 * - Children now report blocking to parent. This enables
532 * statistics in forked mode
534 * Revision 1.92.2.5 2003/03/10 23:45:32 oes
535 * Fixed bug #700381: Non-Threaded version now capable of being toggled.
536 * Children now report having been toggled through _exit(17), parents
537 * watch for that code and toggle themselves if found.
539 * Revision 1.92.2.4 2003/03/07 03:41:04 david__schmidt
540 * Wrapping all *_r functions (the non-_r versions of them) with
541 * mutex semaphores for OSX. Hopefully this will take care of all
542 * of those pesky crash reports.
544 * Revision 1.92.2.3 2003/02/28 12:53:06 oes
545 * Fixed two mostly harmless mem leaks
547 * Revision 1.92.2.2 2002/11/20 14:37:47 oes
548 * Fix: Head of global clients list now initialized to NULL
550 * Revision 1.92.2.1 2002/09/25 14:52:24 oes
551 * Added basic support for OPTIONS and TRACE HTTP methods:
552 * - New interceptor direct_response() added in chat().
553 * - sed() moved to earlier in the process, so that the
554 * Host: header is evaluated before actions and forwarding
557 * Revision 1.92 2002/05/08 16:00:46 oes
558 * Chat's buffer handling:
559 * - Fixed bug with unchecked out-of-mem conditions
560 * while reading client request & server headers
561 * - No longer predict if the buffer limit will be exceeded
562 * in the next read -- check add_to_iob's new
563 * return code. If buffer couldn't be extended
564 * (policy or out-of-mem) while
565 * - reading from client: abort
566 * - reading server headers: send error page
567 * - buffering server body for filter: flush,
568 * and if that fails: send error page
570 * Revision 1.91 2002/04/08 20:35:58 swa
573 * Revision 1.90 2002/04/02 14:57:28 oes
574 * Made sending wafers independent of FEATURE_COOKIE_JAR
576 * Revision 1.89 2002/03/31 17:18:59 jongfoster
577 * Win32 only: Enabling STRICT to fix a VC++ compile warning.
579 * Revision 1.88 2002/03/27 14:32:43 david__schmidt
580 * More compiler warning message maintenance
582 * Revision 1.87 2002/03/26 22:29:54 swa
583 * we have a new homepage!
585 * Revision 1.86 2002/03/25 17:04:55 david__schmidt
586 * Workaround for closing the jarfile before load_config() comes around again
588 * Revision 1.85 2002/03/24 15:23:33 jongfoster
591 * Revision 1.84 2002/03/24 13:25:43 swa
592 * name change related issues
594 * Revision 1.83 2002/03/16 23:54:06 jongfoster
595 * Adding graceful termination feature, to help look for memory leaks.
596 * If you enable this (which, by design, has to be done by hand
597 * editing config.h) and then go to http://i.j.b/die, then the program
598 * will exit cleanly after the *next* request. It should free all the
599 * memory that was used.
601 * Revision 1.82 2002/03/13 00:27:05 jongfoster
604 * Revision 1.81 2002/03/12 01:42:50 oes
605 * Introduced modular filters
607 * Revision 1.80 2002/03/11 22:07:05 david__schmidt
608 * OS/2 port maintenance:
609 * - Fixed EMX build - it had decayed a little
610 * - Fixed inexplicable crash during FD_ZERO - must be due to a bad macro.
611 * substituted a memset for now.
613 * Revision 1.79 2002/03/09 20:03:52 jongfoster
614 * - Making various functions return int rather than size_t.
615 * (Undoing a recent change). Since size_t is unsigned on
616 * Windows, functions like read_socket that return -1 on
617 * error cannot return a size_t.
619 * THIS WAS A MAJOR BUG - it caused frequent, unpredictable
620 * crashes, and also frequently caused JB to jump to 100%
621 * CPU and stay there. (Because it thought it had just
622 * read ((unsigned)-1) == 4Gb of data...)
624 * - The signature of write_socket has changed, it now simply
625 * returns success=0/failure=nonzero.
627 * - Trying to get rid of a few warnings --with-debug on
628 * Windows, I've introduced a new type "jb_socket". This is
629 * used for the socket file descriptors. On Windows, this
630 * is SOCKET (a typedef for unsigned). Everywhere else, it's
631 * an int. The error value can't be -1 any more, so it's
632 * now JB_INVALID_SOCKET (which is -1 on UNIX, and in
633 * Windows it maps to the #define INVALID_SOCKET.)
635 * - The signature of bind_port has changed.
637 * Revision 1.78 2002/03/08 21:35:04 oes
638 * Added optional group supplement to --user option. Will now use default group of user if no group given
640 * Revision 1.77 2002/03/07 03:52:06 oes
641 * - Fixed compiler warnings etc
642 * - Improved handling of failed DNS lookups
644 * Revision 1.76 2002/03/06 22:54:35 jongfoster
645 * Automated function-comment nitpicking.
647 * Revision 1.75 2002/03/06 10:02:19 oes
648 * Fixed stupid bug when --user was not given
650 * Revision 1.74 2002/03/06 00:49:31 jongfoster
651 * Fixing warning on Windows
652 * Making #ifdefs that refer to the same variable consistently
653 * use #ifdef unix rather than mixing #ifdef unix & #ifndef OS2
655 * Revision 1.73 2002/03/05 23:57:30 hal9
656 * Stray character 's' on line 1618 was breaking build.
658 * Revision 1.72 2002/03/05 21:33:45 david__schmidt
659 * - Re-enable OS/2 building after new parms were added
660 * - Fix false out of memory report when resolving CGI templates when no IP
661 * address is available of failed attempt (a la no such domain)
663 * Revision 1.71 2002/03/05 18:13:56 oes
664 * Added --user option
666 * Revision 1.70 2002/03/05 04:52:42 oes
667 * Deleted non-errlog debugging code
669 * Revision 1.69 2002/03/04 23:50:00 jongfoster
670 * Splitting off bind_port() call into bind_port_helper(), with
673 * Revision 1.68 2002/03/04 20:17:32 oes
676 * Revision 1.67 2002/03/04 18:18:57 oes
677 * - Removed _DEBUG mode
678 * - Cleand up cmdline parsing
679 * - Introduced --no-daemon, --pidfile options
680 * - Cleaned up signal handling:
681 * - Terminate cleanly on INT, TERM and ABRT
682 * - Schedule logfile for re-opening on HUP
683 * - Ignore CHLD and PIPE
684 * - Leave the rest with their default handlers
685 * - Uniform handler registration
686 * - Added usage() function
687 * - Played styleguide police
689 * Revision 1.66 2002/03/03 15:06:55 oes
690 * Re-enabled automatic config reloading
692 * Revision 1.65 2002/03/03 14:49:11 oes
693 * Fixed CLF logging: Now uses client's original HTTP request
695 * Revision 1.64 2002/03/03 09:18:03 joergs
696 * Made jumbjuster work on AmigaOS again.
698 * Revision 1.63 2002/03/02 04:14:50 david__schmidt
699 * Clean up a little CRLF unpleasantness that suddenly appeared
701 * Revision 1.62 2002/02/20 23:17:23 jongfoster
702 * Detecting some out-of memory conditions and exiting with a log message.
704 * Revision 1.61 2002/01/17 21:01:52 jongfoster
705 * Moving all our URL and URL pattern parsing code to urlmatch.c.
707 * Revision 1.60 2001/12/30 14:07:32 steudten
708 * - Add signal handling (unix)
709 * - Add SIGHUP handler (unix)
710 * - Add creation of pidfile (unix)
711 * - Add action 'top' in rc file (RH)
712 * - Add entry 'SIGNALS' to manpage
713 * - Add exit message to logfile (unix)
715 * Revision 1.59 2001/12/13 14:07:18 oes
716 * Fixed Bug: 503 error page now sent OK
718 * Revision 1.58 2001/11/30 23:37:24 jongfoster
719 * Renaming the Win32 config file to config.txt - this is almost the
720 * same as the corresponding UNIX name "config"
722 * Revision 1.57 2001/11/16 00:47:43 jongfoster
723 * Changing the tty-disconnection code to use setsid().
725 * Revision 1.56 2001/11/13 20:20:54 jongfoster
726 * Tabs->spaces, fixing a bug with missing {} around an if()
728 * Revision 1.55 2001/11/13 20:14:53 jongfoster
729 * Patch for FreeBSD setpgrp() as suggested by Alexander Lazic
731 * Revision 1.54 2001/11/07 00:03:14 steudten
732 * Give reliable return value if an error
733 * occurs not just 0 with new daemon mode.
735 * Revision 1.53 2001/11/05 21:41:43 steudten
736 * Add changes to be a real daemon just for unix os.
737 * (change cwd to /, detach from controlling tty, set
738 * process group and session leader to the own process.
740 * Add some fatal-error log message for failed malloc().
741 * Add '-d' if compiled with 'configure --with-debug' to
742 * enable debug output.
744 * Revision 1.52 2001/10/26 20:11:20 jongfoster
745 * Fixing type mismatch
747 * Revision 1.51 2001/10/26 17:38:28 oes
750 * Revision 1.50 2001/10/25 03:40:48 david__schmidt
751 * Change in porting tactics: OS/2's EMX porting layer doesn't allow multiple
752 * threads to call select() simultaneously. So, it's time to do a real, live,
753 * native OS/2 port. See defines for __EMX__ (the porting layer) vs. __OS2__
754 * (native). Both versions will work, but using __OS2__ offers multi-threading.
756 * Revision 1.49 2001/10/23 21:41:35 jongfoster
757 * Added call to initialize the (statically-allocated of course)
758 * "out of memory" CGI response.
760 * Revision 1.48 2001/10/10 19:56:46 jongfoster
761 * Moving some code that wasn't cookie-related out of an #ifdef
764 * Revision 1.47 2001/10/10 16:44:36 oes
765 * Added CONNECT destination port limitation check
767 * Revision 1.46 2001/10/08 15:17:41 oes
768 * Re-enabled SSL forwarding
770 * Revision 1.45 2001/10/07 15:42:11 oes
771 * Replaced 6 boolean members of csp with one bitmap (csp->flags)
773 * Moved downgrading of the HTTP version from parse_http_request to
774 * chat(), since we can't decide if it is necessary before we have
775 * determined the actions for the URL. The HTTP command is now
776 * *always* re-built so the repairs need no longer be special-cased.
778 * filter_popups now gets a csp pointer so it can raise the new
779 * CSP_FLAG_MODIFIED flag.
783 * Added configurable size limit for the IOB. If the IOB grows so
784 * large that the next read would exceed the limit, the header
785 * is generated, and the header & unfiltered buffer are flushed
786 * to the client. Chat then continues in non-buffering,
787 * non-filtering body mode.
789 * Revision 1.44 2001/10/02 18:13:57 oes
792 * Revision 1.43 2001/10/02 15:32:13 oes
793 * Moved generation of hdr
795 * Revision 1.42 2001/09/21 23:02:02 david__schmidt
796 * Cleaning up 2 compiler warnings on OS/2.
798 * Revision 1.41 2001/09/16 17:05:14 jongfoster
799 * Removing unused #include showarg.h
801 * Revision 1.40 2001/09/16 15:41:45 jongfoster
802 * Fixing signed/unsigned comparison warning.
804 * Revision 1.39 2001/09/16 13:21:27 jongfoster
805 * Changes to use new list functions.
807 * Revision 1.38 2001/09/16 13:01:46 jongfoster
808 * Removing redundant function call that zeroed zalloc()'d memory.
810 * Revision 1.37 2001/09/10 11:12:24 oes
811 * Deleted unused variable
813 * Revision 1.36 2001/09/10 10:56:15 oes
814 * Silenced compiler warnings
816 * Revision 1.35 2001/07/31 14:44:22 oes
817 * Deleted unused size parameter from filter_popups()
819 * Revision 1.34 2001/07/30 22:08:36 jongfoster
820 * Tidying up #defines:
821 * - All feature #defines are now of the form FEATURE_xxx
822 * - Permanently turned off WIN_GUI_EDIT
823 * - Permanently turned on WEBDAV and SPLIT_PROXY_ARGS
825 * Revision 1.33 2001/07/29 19:32:00 jongfoster
826 * Renaming _main() [mingw32 only] to real_main(), for ANSI compliance.
828 * Revision 1.32 2001/07/29 18:47:05 jongfoster
829 * Adding missing #include "loadcfg.h"
831 * Revision 1.31 2001/07/29 12:17:48 oes
832 * Applied pthread fix by Paul Lieverse
834 * Revision 1.30 2001/07/25 22:57:13 jongfoster
835 * __BEOS__ no longer overrides FEATURE_PTHREAD.
836 * This is because FEATURE_PTHREAD will soon be widely used, so I
837 * want to keep it simple.
839 * Revision 1.29 2001/07/24 12:47:06 oes
840 * Applied BeOS support update by Eugenia
842 * Revision 1.28 2001/07/23 13:26:12 oes
843 * Fixed bug in popup-killing for the first read that caused binary garbage to be sent between headers and body
845 * Revision 1.27 2001/07/19 19:09:47 haroon
846 * - Added code to take care of the situation where while processing the first
847 * server response (which includes the server header), after finding the end
848 * of the headers we were not looking past the end of the headers for
849 * content modification. I enabled it for filter_popups.
850 * Someone else should look to see if other similar operations should be
851 * done to the discarded portion of the buffer.
853 * Note 2001/07/20: No, the other content modification mechanisms will process
854 * the whole iob later anyway. --oes
856 * Revision 1.26 2001/07/18 12:31:36 oes
859 * Revision 1.25 2001/07/15 19:43:49 jongfoster
860 * Supports POSIX threads.
861 * Also removed some unused #includes.
863 * Revision 1.24 2001/07/13 14:00:40 oes
864 * - Generic content modification scheme:
865 * Each feature has its own applicability flag that is set
866 * from csp->action->flags.
867 * Replaced the "filtering" int flag , by a function pointer
868 * "content_filter" to the function that will do the content
869 * modification. If it is != NULL, the document will be buffered
870 * and processed through *content_filter, which must set
871 * csp->content_length and return a modified copy of the body
872 * or return NULL (on failiure).
873 * - Changed csp->is_text to the more generic bitmap csp->content_type
874 * which can currently take the valued CT_TEXT or CT_GIF
876 * - Removed all #ifdef PCRS
878 * Revision 1.23 2001/07/02 02:28:25 iwanttokeepanon
879 * Added "#ifdef ACL_FILES" conditional compilation to line 1291 to exclude
880 * the `block_acl' call. This prevents a compilation error when the user
881 * does not wish to use the "ACL" feature.
883 * Revision 1.22 2001/06/29 21:45:41 oes
884 * Indentation, CRLF->LF, Tab-> Space
886 * Revision 1.21 2001/06/29 13:29:36 oes
887 * - Cleaned up, improved comments
888 * - Unified all possible interceptors (CGI,
889 * block, trust, fast_redirect) in one
890 * place, with one (CGI) answer generation
891 * mechansim. Much clearer now.
892 * - Removed the GIF image generation, which
893 * is now done in filters.c:block_url()
894 * - Made error conditions like domain lookup
895 * failiure or (various) problems while talking
896 * to the server use cgi.c:error_response()
897 * instead of generating HTML/HTTP in chat() (yuck!)
898 * - Removed logentry from cancelled commit
900 * Revision 1.20 2001/06/09 10:55:28 jongfoster
901 * Changing BUFSIZ ==> BUFFER_SIZE
903 * Revision 1.19 2001/06/07 23:12:52 jongfoster
904 * Replacing function pointer in struct gateway with a directly
905 * called function forwarded_connect().
906 * Replacing struct gateway with struct forward_spec
908 * Revision 1.18 2001/06/03 19:12:16 oes
909 * introduced new cgi handling
911 * Revision 1.17 2001/06/01 20:07:23 jongfoster
912 * Now uses action +image-blocker{} rather than config->tinygif
914 * Revision 1.16 2001/06/01 18:49:17 jongfoster
915 * Replaced "list_share" with "list" - the tiny memory gain was not
916 * worth the extra complexity.
918 * Revision 1.15 2001/05/31 21:24:47 jongfoster
919 * Changed "permission" to "action" throughout.
920 * Removed DEFAULT_USER_AGENT - it must now be specified manually.
921 * Moved vanilla wafer check into chat(), since we must now
922 * decide whether or not to add it based on the URL.
924 * Revision 1.14 2001/05/29 20:14:01 joergs
925 * AmigaOS bugfix: PCRS needs a lot of stack, stacksize for child threads
928 * Revision 1.13 2001/05/29 09:50:24 jongfoster
929 * Unified blocklist/imagelist/permissionslist.
930 * File format is still under discussion, but the internal changes
933 * Also modified interceptor behaviour:
934 * - We now intercept all URLs beginning with one of the following
935 * prefixes (and *only* these prefixes):
937 * * http://ijbswa.sf.net/config/
938 * * http://ijbswa.sourceforge.net/config/
939 * - New interceptors "home page" - go to http://i.j.b/ to see it.
940 * - Internal changes so that intercepted and fast redirect pages
941 * are not replaced with an image.
942 * - Interceptors now have the option to send a binary page direct
943 * to the client. (i.e. ijb-send-banner uses this)
944 * - Implemented show-url-info interceptor. (Which is why I needed
945 * the above interceptors changes - a typical URL is
946 * "http://i.j.b/show-url-info?url=www.somesite.com/banner.gif".
947 * The previous mechanism would not have intercepted that, and
948 * if it had been intercepted then it then it would have replaced
951 * Revision 1.12 2001/05/27 22:17:04 oes
953 * - re_process_buffer no longer writes the modified buffer
954 * to the client, which was very ugly. It now returns the
955 * buffer, which it is then written by chat.
957 * - content_length now adjusts the Content-Length: header
958 * for modified documents rather than crunch()ing it.
959 * (Length info in csp->content_length, which is 0 for
960 * unmodified documents)
962 * - For this to work, sed() is called twice when filtering.
964 * Revision 1.11 2001/05/26 17:27:53 jongfoster
965 * Added support for CLF and fixed LOG_LEVEL_LOG.
966 * Also did CRLF->LF fix of my previous patch.
968 * Revision 1.10 2001/05/26 15:26:15 jongfoster
969 * ACL feature now provides more security by immediately dropping
970 * connections from untrusted hosts.
972 * Revision 1.9 2001/05/26 00:28:36 jongfoster
973 * Automatic reloading of config file.
974 * Removed obsolete SIGHUP support (Unix) and Reload menu option (Win32).
975 * Most of the global variables have been moved to a new
976 * struct configuration_spec, accessed through csp->config->globalname
977 * Most of the globals remaining are used by the Win32 GUI.
979 * Revision 1.8 2001/05/25 22:43:18 jongfoster
980 * Fixing minor memory leak and buffer overflow.
982 * Revision 1.7 2001/05/25 22:34:30 jongfoster
985 * Revision 1.6 2001/05/23 00:13:58 joergs
986 * AmigaOS support fixed.
988 * Revision 1.5 2001/05/22 18:46:04 oes
990 * - Enabled filtering banners by size rather than URL
991 * by adding patterns that replace all standard banner
992 * sizes with the "Junkbuster" gif to the re_filterfile
994 * - Enabled filtering WebBugs by providing a pattern
995 * which kills all 1x1 images
997 * - Added support for PCRE_UNGREEDY behaviour to pcrs,
998 * which is selected by the (nonstandard and therefore
999 * capital) letter 'U' in the option string.
1000 * It causes the quantifiers to be ungreedy by default.
1001 * Appending a ? turns back to greedy (!).
1003 * - Added a new interceptor ijb-send-banner, which
1004 * sends back the "Junkbuster" gif. Without imagelist or
1005 * MSIE detection support, or if tinygif = 1, or the
1006 * URL isn't recognized as an imageurl, a lame HTML
1007 * explanation is sent instead.
1009 * - Added new feature, which permits blocking remote
1010 * script redirects and firing back a local redirect
1012 * The feature is conditionally compiled, i.e. it
1013 * can be disabled with --disable-fast-redirects,
1014 * plus it must be activated by a "fast-redirects"
1015 * line in the config file, has its own log level
1016 * and of course wants to be displayed by show-proxy-args
1017 * Note: Boy, all the #ifdefs in 1001 locations and
1018 * all the fumbling with configure.in and acconfig.h
1019 * were *way* more work than the feature itself :-(
1021 * - Because a generic redirect template was needed for
1022 * this, tinygif = 3 now uses the same.
1024 * - Moved GIFs, and other static HTTP response templates
1027 * - Some minor fixes
1029 * - Removed some >400 CRs again (Jon, you really worked
1032 * Revision 1.4 2001/05/21 19:34:01 jongfoster
1033 * Made failure to bind() a fatal error.
1035 * Revision 1.3 2001/05/20 01:21:20 jongfoster
1036 * Version 2.9.4 checkin.
1037 * - Merged popupfile and cookiefile, and added control over PCRS
1038 * filtering, in new "permissionsfile".
1039 * - Implemented LOG_LEVEL_FATAL, so that if there is a configuration
1040 * file error you now get a message box (in the Win32 GUI) rather
1041 * than the program exiting with no explanation.
1042 * - Made killpopup use the PCRS MIME-type checking and HTTP-header
1044 * - Removed tabs from "config"
1045 * - Moved duplicated url parsing code in "loaders.c" to a new funcition.
1046 * - Bumped up version number.
1048 * Revision 1.2 2001/05/17 22:34:44 oes
1049 * - Added hint on GIF char array generation to jcc.c
1050 * - Cleaned CRLF's from the sources and related files
1051 * - Repaired logging for REF and FRC
1053 * Revision 1.1.1.1 2001/05/15 13:58:56 oes
1054 * Initial import of version 2.9.3 source tree
1057 *********************************************************************/
1063 #include <sys/types.h>
1072 # ifndef FEATURE_PTHREAD
1076 # include <windows.h>
1077 # include <process.h>
1078 # endif /* ndef FEATURE_PTHREAD */
1081 # ifndef _WIN_CONSOLE
1082 # include "w32log.h"
1083 # endif /* ndef _WIN_CONSOLE */
1084 # include "w32svrapi.h"
1086 #else /* ifndef _WIN32 */
1088 # if !defined (__OS2__)
1089 # include <unistd.h>
1090 # include <sys/wait.h>
1091 # endif /* ndef __OS2__ */
1092 # include <sys/time.h>
1093 # include <sys/stat.h>
1094 # include <sys/ioctl.h>
1097 #include <sys/termios.h>
1105 # include <signal.h>
1108 # include <socket.h> /* BeOS has select() for sockets only. */
1109 # include <OS.h> /* declarations for threads and stuff. */
1112 # if defined(__EMX__) || defined(__OS2__)
1113 # include <sys/select.h> /* OS/2/EMX needs a little help with select */
1118 #define bzero(B,N) memset(B,0x00,n)
1122 # include <select.h>
1127 #include "project.h"
1130 #include "filters.h"
1131 #include "loaders.h"
1132 #include "parsers.h"
1133 #include "miscutil.h"
1135 #include "jbsockets.h"
1136 #include "gateway.h"
1137 #include "actions.h"
1139 #include "loadcfg.h"
1140 #include "urlmatch.h"
1142 const char jcc_h_rcs[] = JCC_H_VERSION;
1143 const char project_h_rcs[] = PROJECT_H_VERSION;
1146 struct client_state clients[1];
1147 struct file_list files[1];
1149 #ifdef FEATURE_STATISTICS
1150 int urls_read = 0; /* total nr of urls read inc rejected */
1151 int urls_rejected = 0; /* total nr of urls rejected */
1152 #endif /* def FEATURE_STATISTICS */
1154 #ifdef FEATURE_GRACEFUL_TERMINATION
1155 int g_terminate = 0;
1158 #if !defined(_WIN32) && !defined(__OS2__) && !defined(AMIGA)
1159 static void sig_handler(int the_signal);
1161 static int client_protocol_is_unsupported(const struct client_state *csp, char *req);
1162 static jb_err get_request_destination_elsewhere(struct client_state *csp, struct list *headers);
1163 static jb_err get_server_headers(struct client_state *csp);
1164 static const char *crunch_reason(const struct http_response *rsp);
1165 static void send_crunch_response(const struct client_state *csp, struct http_response *rsp);
1167 * static int request_contains_null_bytes(const struct client_state *csp, char *buf, int len);
1169 static void build_request_line(struct client_state *csp, const struct forward_spec *fwd, char **request_line);
1170 static jb_err change_request_destination(struct client_state *csp);
1171 static void chat(struct client_state *csp);
1172 static void serve(struct client_state *csp);
1173 #if !defined(_WIN32) || defined(_WIN_CONSOLE)
1174 static void usage(const char *myname);
1176 static void initialize_mutexes(void);
1177 static jb_socket bind_port_helper(struct configuration_spec *config);
1178 static void listen_loop(void);
1181 void serve(struct client_state *csp);
1182 #else /* ifndef AMIGA */
1183 static void serve(struct client_state *csp);
1184 #endif /* def AMIGA */
1187 static int32 server_thread(void *data);
1188 #endif /* def __BEOS__ */
1191 #define sleep(N) Sleep(((N) * 1000))
1195 #define sleep(N) DosSleep(((N) * 100))
1198 #ifdef MUTEX_LOCKS_AVAILABLE
1200 * XXX: Does the locking stuff really belong in this file?
1202 privoxy_mutex_t log_mutex;
1203 privoxy_mutex_t log_init_mutex;
1204 privoxy_mutex_t connection_reuse_mutex;
1206 #if !defined(HAVE_GETHOSTBYADDR_R) || !defined(HAVE_GETHOSTBYNAME_R)
1207 privoxy_mutex_t resolver_mutex;
1208 #endif /* !defined(HAVE_GETHOSTBYADDR_R) || !defined(HAVE_GETHOSTBYNAME_R) */
1210 #ifndef HAVE_GMTIME_R
1211 privoxy_mutex_t gmtime_mutex;
1212 #endif /* ndef HAVE_GMTIME_R */
1214 #ifndef HAVE_LOCALTIME_R
1215 privoxy_mutex_t localtime_mutex;
1216 #endif /* ndef HAVE_GMTIME_R */
1219 privoxy_mutex_t rand_mutex;
1220 #endif /* ndef HAVE_RANDOM */
1222 #endif /* def MUTEX_LOCKS_AVAILABLE */
1225 const char *basedir = NULL;
1226 const char *pidfile = NULL;
1227 static int received_hup_signal = 0;
1228 #endif /* defined unix */
1230 /* HTTP snipplets. */
1231 static const char CSUCCEED[] =
1232 "HTTP/1.0 200 Connection established\r\n"
1233 "Proxy-Agent: Privoxy/" VERSION "\r\n\r\n";
1235 static const char CHEADER[] =
1236 "HTTP/1.0 400 Invalid header received from client\r\n"
1237 "Proxy-Agent: Privoxy " VERSION "\r\n"
1238 "Content-Type: text/plain\r\n"
1239 "Connection: close\r\n\r\n"
1240 "Invalid header received from client.\r\n";
1242 static const char FTP_RESPONSE[] =
1243 "HTTP/1.0 400 Invalid request received from client\r\n"
1244 "Content-Type: text/plain\r\n"
1245 "Connection: close\r\n\r\n"
1246 "Invalid request. Privoxy doesn't support FTP.\r\n";
1248 static const char GOPHER_RESPONSE[] =
1249 "HTTP/1.0 400 Invalid request received from client\r\n"
1250 "Content-Type: text/plain\r\n"
1251 "Connection: close\r\n\r\n"
1252 "Invalid request. Privoxy doesn't support gopher.\r\n";
1254 /* XXX: should be a template */
1255 static const char MISSING_DESTINATION_RESPONSE[] =
1256 "HTTP/1.0 400 Bad request received from client\r\n"
1257 "Proxy-Agent: Privoxy " VERSION "\r\n"
1258 "Content-Type: text/plain\r\n"
1259 "Connection: close\r\n\r\n"
1260 "Bad request. Privoxy was unable to extract the destination.\r\n";
1262 /* XXX: should be a template */
1263 static const char NO_SERVER_DATA_RESPONSE[] =
1264 "HTTP/1.0 502 Server or forwarder response empty\r\n"
1265 "Proxy-Agent: Privoxy " VERSION "\r\n"
1266 "Content-Type: text/plain\r\n"
1267 "Connection: close\r\n\r\n"
1268 "Empty server or forwarder response.\r\n"
1269 "The connection has been closed but Privoxy didn't receive any data.\r\n";
1271 /* XXX: should be a template */
1272 static const char INVALID_SERVER_HEADERS_RESPONSE[] =
1273 "HTTP/1.0 502 Server or forwarder response invalid\r\n"
1274 "Proxy-Agent: Privoxy " VERSION "\r\n"
1275 "Content-Type: text/plain\r\n"
1276 "Connection: close\r\n\r\n"
1277 "Bad response. The server or forwarder response doesn't look like HTTP.\r\n";
1280 /* XXX: should be a template */
1281 static const char NULL_BYTE_RESPONSE[] =
1282 "HTTP/1.0 400 Bad request received from client\r\n"
1283 "Proxy-Agent: Privoxy " VERSION "\r\n"
1284 "Content-Type: text/plain\r\n"
1285 "Connection: close\r\n\r\n"
1286 "Bad request. Null byte(s) before end of request.\r\n";
1289 /* XXX: should be a template */
1290 static const char MESSED_UP_REQUEST_RESPONSE[] =
1291 "HTTP/1.0 400 Malformed request after rewriting\r\n"
1292 "Proxy-Agent: Privoxy " VERSION "\r\n"
1293 "Content-Type: text/plain\r\n"
1294 "Connection: close\r\n\r\n"
1295 "Bad request. Messed up with header filters.\r\n";
1297 /* A function to crunch a response */
1298 typedef struct http_response *(*crunch_func_ptr)(struct client_state *);
1300 /* Crunch function flags */
1301 #define CF_NO_FLAGS 0
1302 /* Cruncher applies to forced requests as well */
1303 #define CF_IGNORE_FORCE 1
1304 /* Crunched requests are counted for the block statistics */
1305 #define CF_COUNT_AS_REJECT 2
1307 /* A crunch function and its flags */
1310 const crunch_func_ptr cruncher;
1314 static int crunch_response_triggered(struct client_state *csp, const struct cruncher crunchers[]);
1316 /* Complete list of cruncher functions */
1317 static const struct cruncher crunchers_all[] = {
1318 { direct_response, CF_COUNT_AS_REJECT|CF_IGNORE_FORCE},
1319 { block_url, CF_COUNT_AS_REJECT },
1320 #ifdef FEATURE_TRUST
1321 { trust_url, CF_COUNT_AS_REJECT },
1322 #endif /* def FEATURE_TRUST */
1323 { redirect_url, CF_NO_FLAGS },
1324 { dispatch_cgi, CF_IGNORE_FORCE},
1328 /* Light version, used after tags are applied */
1329 static const struct cruncher crunchers_light[] = {
1330 { block_url, CF_COUNT_AS_REJECT },
1331 { redirect_url, CF_NO_FLAGS },
1337 * XXX: Don't we really mean
1343 #if !defined(_WIN32) && !defined(__OS2__) && !defined(AMIGA)
1344 /*********************************************************************
1346 * Function : sig_handler
1348 * Description : Signal handler for different signals.
1349 * Exit gracefully on TERM and INT
1350 * or set a flag that will cause the errlog
1351 * to be reopened by the main thread on HUP.
1354 * 1 : the_signal = the signal cause this function to call
1358 *********************************************************************/
1359 static void sig_handler(int the_signal)
1365 log_error(LOG_LEVEL_INFO, "exiting by signal %d .. bye", the_signal);
1377 received_hup_signal = 1;
1383 * We shouldn't be here, unless we catch signals
1384 * in main() that we can't handle here!
1386 log_error(LOG_LEVEL_FATAL, "sig_handler: exiting on unexpected signal %d", the_signal);
1394 /*********************************************************************
1396 * Function : client_protocol_is_unsupported
1398 * Description : Checks if the client used a known unsupported
1399 * protocol and deals with it by sending an error
1403 * 1 : csp = Current client state (buffers, headers, etc...)
1404 * 2 : req = the first request line send by the client
1406 * Returns : TRUE if an error response has been generated, or
1407 * FALSE if the request doesn't look invalid.
1409 *********************************************************************/
1410 static int client_protocol_is_unsupported(const struct client_state *csp, char *req)
1413 * If it's a FTP or gopher request, we don't support it.
1415 * These checks are better than nothing, but they might
1416 * not work in all configurations and some clients might
1417 * have problems digesting the answer.
1419 * They should, however, never cause more problems than
1420 * Privoxy's old behaviour (returning the misleading HTML
1423 * "Could not resolve http://(ftp|gopher)://example.org").
1425 if (!strncmpic(req, "GET ftp://", 10) || !strncmpic(req, "GET gopher://", 13))
1427 const char *response = NULL;
1428 const char *protocol = NULL;
1430 if (!strncmpic(req, "GET ftp://", 10))
1432 response = FTP_RESPONSE;
1437 response = GOPHER_RESPONSE;
1438 protocol = "GOPHER";
1440 log_error(LOG_LEVEL_ERROR,
1441 "%s tried to use Privoxy as %s proxy: %s",
1442 csp->ip_addr_str, protocol, req);
1443 log_error(LOG_LEVEL_CLF,
1444 "%s - - [%T] \"%s\" 400 0", csp->ip_addr_str, req);
1446 write_socket(csp->cfd, response, strlen(response));
1455 /*********************************************************************
1457 * Function : get_request_destination_elsewhere
1459 * Description : If the client's request was redirected into
1460 * Privoxy without the client's knowledge,
1461 * the request line lacks the destination host.
1463 * This function tries to get it elsewhere,
1464 * provided accept-intercepted-requests is enabled.
1466 * "Elsewhere" currently only means "Host: header",
1467 * but in the future we may ask the redirecting
1468 * packet filter to look the destination up.
1470 * If the destination stays unknown, an error
1471 * response is send to the client and headers
1472 * are freed so that chat() can return directly.
1475 * 1 : csp = Current client state (buffers, headers, etc...)
1476 * 2 : headers = a header list
1478 * Returns : JB_ERR_OK if the destination is now known, or
1479 * JB_ERR_PARSE if it isn't.
1481 *********************************************************************/
1482 static jb_err get_request_destination_elsewhere(struct client_state *csp, struct list *headers)
1486 if (!(csp->config->feature_flags & RUNTIME_FEATURE_ACCEPT_INTERCEPTED_REQUESTS))
1488 log_error(LOG_LEVEL_ERROR, "%s's request: \'%s\' is invalid."
1489 " Privoxy isn't configured to accept intercepted requests.",
1490 csp->ip_addr_str, csp->http->cmd);
1491 /* XXX: Use correct size */
1492 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" 400 0",
1493 csp->ip_addr_str, csp->http->cmd);
1495 write_socket(csp->cfd, CHEADER, strlen(CHEADER));
1496 destroy_list(headers);
1498 return JB_ERR_PARSE;
1500 else if (JB_ERR_OK == get_destination_from_headers(headers, csp->http))
1502 /* Split the domain we just got for pattern matching */
1503 init_domain_components(csp->http);
1509 /* We can't work without destination. Go spread the news.*/
1511 req = list_to_text(headers);
1513 /* XXX: Use correct size */
1514 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" 400 0",
1515 csp->ip_addr_str, csp->http->cmd);
1516 log_error(LOG_LEVEL_ERROR,
1517 "Privoxy was unable to get the destination for %s's request:\n%s\n%s",
1518 csp->ip_addr_str, csp->http->cmd, req);
1521 write_socket(csp->cfd, MISSING_DESTINATION_RESPONSE, strlen(MISSING_DESTINATION_RESPONSE));
1522 destroy_list(headers);
1524 return JB_ERR_PARSE;
1527 * TODO: If available, use PF's ioctl DIOCNATLOOK as last resort
1528 * to get the destination IP address, use it as host directly
1529 * or do a reverse DNS lookup first.
1534 /*********************************************************************
1536 * Function : get_server_headers
1538 * Description : Parses server headers in iob and fills them
1539 * into csp->headers so that they can later be
1543 * 1 : csp = Current client state (buffers, headers, etc...)
1545 * Returns : JB_ERR_OK if everything went fine, or
1546 * JB_ERR_PARSE if the headers were incomplete.
1548 *********************************************************************/
1549 static jb_err get_server_headers(struct client_state *csp)
1551 int continue_hack_in_da_house = 0;
1554 while (((header = get_header(csp->iob)) != NULL) || continue_hack_in_da_house)
1559 * continue hack in da house. Ignore the ending of
1560 * this head and continue enlisting header lines.
1561 * The reason is described below.
1563 enlist(csp->headers, "");
1564 continue_hack_in_da_house = 0;
1567 else if (0 == strncmpic(header, "HTTP/1.1 100", 12))
1570 * It's a bodyless continue response, don't
1571 * stop header parsing after reaching its end.
1573 * As a result Privoxy will concatenate the
1574 * next response's head and parse and deliver
1575 * the headers as if they belonged to one request.
1577 * The client will separate them because of the
1578 * empty line between them.
1580 * XXX: What we're doing here is clearly against
1581 * the intended purpose of the continue header,
1582 * and under some conditions (HTTP/1.0 client request)
1583 * it's a standard violation.
1585 * Anyway, "sort of against the spec" is preferable
1586 * to "always getting confused by Continue responses"
1587 * (Privoxy's behaviour before this hack was added)
1589 log_error(LOG_LEVEL_HEADER, "Continue hack in da house.");
1590 continue_hack_in_da_house = 1;
1592 else if (*header == '\0')
1595 * If the header is empty, but the Continue hack
1596 * isn't active, we can assume that we reached the
1597 * end of the buffer before we hit the end of the
1600 * Inform the caller an let it decide how to handle it.
1602 return JB_ERR_PARSE;
1605 if (JB_ERR_MEMORY == enlist(csp->headers, header))
1608 * XXX: Should we quit the request and return a
1609 * out of memory error page instead?
1611 log_error(LOG_LEVEL_ERROR,
1612 "Out of memory while enlisting server headers. %s lost.",
1622 /*********************************************************************
1624 * Function : crunch_reason
1626 * Description : Translates the crunch reason code into a string.
1629 * 1 : rsp = a http_response
1631 * Returns : A string with the crunch reason or an error description.
1633 *********************************************************************/
1634 static const char *crunch_reason(const struct http_response *rsp)
1636 char * reason = NULL;
1638 assert(rsp != NULL);
1641 return "Internal error while searching for crunch reason";
1644 switch (rsp->reason)
1646 case RSP_REASON_UNSUPPORTED:
1647 reason = "Unsupported HTTP feature";
1649 case RSP_REASON_BLOCKED:
1652 case RSP_REASON_UNTRUSTED:
1653 reason = "Untrusted";
1655 case RSP_REASON_REDIRECTED:
1656 reason = "Redirected";
1658 case RSP_REASON_CGI_CALL:
1659 reason = "CGI Call";
1661 case RSP_REASON_NO_SUCH_DOMAIN:
1662 reason = "DNS failure";
1664 case RSP_REASON_FORWARDING_FAILED:
1665 reason = "Forwarding failed";
1667 case RSP_REASON_CONNECT_FAILED:
1668 reason = "Connection failure";
1670 case RSP_REASON_OUT_OF_MEMORY:
1671 reason = "Out of memory (may mask other reasons)";
1674 reason = "No reason recorded";
1682 /*********************************************************************
1684 * Function : send_crunch_response
1686 * Description : Delivers already prepared response for
1687 * intercepted requests, logs the interception
1688 * and frees the response.
1691 * 1 : csp = Current client state (buffers, headers, etc...)
1692 * 1 : rsp = Fully prepared response. Will be freed on exit.
1694 * Returns : Nothing.
1696 *********************************************************************/
1697 static void send_crunch_response(const struct client_state *csp, struct http_response *rsp)
1699 const struct http_request *http = csp->http;
1700 char status_code[4];
1702 assert(rsp != NULL);
1703 assert(rsp->head != NULL);
1708 * Not supposed to happen. If it does
1709 * anyway, treat it as an unknown error.
1711 cgi_error_unknown(csp, rsp, RSP_REASON_INTERNAL_ERROR);
1712 /* return code doesn't matter */
1717 /* If rsp is still NULL, we have serious internal problems. */
1718 log_error(LOG_LEVEL_FATAL,
1719 "NULL response in send_crunch_response and cgi_error_unknown failed as well.");
1723 * Extract the status code from the actual head
1724 * that was send to the client. It is the only
1725 * way to get it right for all requests, including
1726 * the fixed ones for out-of-memory problems.
1728 * A head starts like this: 'HTTP/1.1 200...'
1732 status_code[0] = rsp->head[9];
1733 status_code[1] = rsp->head[10];
1734 status_code[2] = rsp->head[11];
1735 status_code[3] = '\0';
1737 /* Write the answer to the client */
1738 if (write_socket(csp->cfd, rsp->head, rsp->head_length)
1739 || write_socket(csp->cfd, rsp->body, rsp->content_length))
1741 /* There is nothing we can do about it. */
1742 log_error(LOG_LEVEL_ERROR, "write to: %s failed: %E", csp->http->host);
1745 /* Log that the request was crunched and why. */
1746 log_error(LOG_LEVEL_GPC, "%s%s crunch! (%s)",
1747 http->hostport, http->path, crunch_reason(rsp));
1748 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" %s %d",
1749 csp->ip_addr_str, http->ocmd, status_code, rsp->content_length);
1751 /* Clean up and return */
1752 if (cgi_error_memory() != rsp)
1754 free_http_response(rsp);
1761 /*********************************************************************
1763 * Function : request_contains_null_bytes
1765 * Description : Checks for NULL bytes in the request and sends
1766 * an error message to the client if any were found.
1768 * XXX: currently not used, see comment in chat().
1771 * 1 : csp = Current client state (buffers, headers, etc...)
1772 * 2 : buf = Data from the client's request to check.
1773 * 3 : len = The data length.
1775 * Returns : TRUE if the request contained one or more NULL bytes, or
1778 *********************************************************************/
1779 static int request_contains_null_bytes(const struct client_state *csp, char *buf, int len)
1781 size_t c_len; /* Request lenght when treated as C string */
1783 c_len = strlen(buf);
1788 * Null byte(s) found. Log the request,
1789 * return an error response and hang up.
1791 size_t tmp_len = c_len;
1796 * Replace NULL byte(s) with '°' characters
1797 * so the request can be logged as string.
1798 * XXX: Is there a better replacement character?
1801 tmp_len += strlen(buf+tmp_len);
1802 } while (tmp_len < len);
1804 log_error(LOG_LEVEL_ERROR, "%s\'s request contains at least one NULL byte "
1805 "(length=%d, strlen=%d).", csp->ip_addr_str, len, c_len);
1806 log_error(LOG_LEVEL_HEADER,
1807 "Offending request data with NULL bytes turned into \'°\' characters: %s", buf);
1809 write_socket(csp->cfd, NULL_BYTE_RESPONSE, strlen(NULL_BYTE_RESPONSE));
1811 /* XXX: Log correct size */
1812 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"Invalid request\" 400 0", csp->ip_addr_str);
1822 /*********************************************************************
1824 * Function : crunch_response_triggered
1826 * Description : Checks if the request has to be crunched,
1827 * and delivers the crunch response if necessary.
1830 * 1 : csp = Current client state (buffers, headers, etc...)
1831 * 2 : crunchers = list of cruncher functions to run
1833 * Returns : TRUE if the request was answered with a crunch response
1836 *********************************************************************/
1837 static int crunch_response_triggered(struct client_state *csp, const struct cruncher crunchers[])
1839 struct http_response *rsp = NULL;
1840 const struct cruncher *c;
1843 * If CGI request crunching is disabled,
1844 * check the CGI dispatcher out of order to
1845 * prevent unintentional blocks or redirects.
1847 if (!(csp->config->feature_flags & RUNTIME_FEATURE_CGI_CRUNCHING)
1848 && (NULL != (rsp = dispatch_cgi(csp))))
1850 /* Deliver, log and free the interception response. */
1851 send_crunch_response(csp, rsp);
1855 for (c = crunchers; c->cruncher != NULL; c++)
1858 * Check the cruncher if either Privoxy is toggled
1859 * on and the request isn't forced, or if the cruncher
1860 * applies to forced requests as well.
1862 if (((csp->flags & CSP_FLAG_TOGGLED_ON) &&
1863 !(csp->flags & CSP_FLAG_FORCED)) ||
1864 (c->flags & CF_IGNORE_FORCE))
1866 rsp = c->cruncher(csp);
1869 /* Deliver, log and free the interception response. */
1870 send_crunch_response(csp, rsp);
1871 #ifdef FEATURE_STATISTICS
1872 if (c->flags & CF_COUNT_AS_REJECT)
1874 csp->flags |= CSP_FLAG_REJECTED;
1876 #endif /* def FEATURE_STATISTICS */
1887 /*********************************************************************
1889 * Function : build_request_line
1891 * Description : Builds the HTTP request line.
1893 * If a HTTP forwarder is used it expects the whole URL,
1894 * web servers only get the path.
1897 * 1 : csp = Current client state (buffers, headers, etc...)
1898 * 2 : fwd = The forwarding spec used for the request
1899 * XXX: Should use http->fwd instead.
1900 * 3 : request_line = The old request line which will be replaced.
1902 * Returns : Nothing. Terminates in case of memory problems.
1904 *********************************************************************/
1905 static void build_request_line(struct client_state *csp, const struct forward_spec *fwd, char **request_line)
1907 struct http_request *http = csp->http;
1909 assert(http->ssl == 0);
1912 * Downgrade http version from 1.1 to 1.0
1913 * if +downgrade action applies.
1915 if ( (csp->action->flags & ACTION_DOWNGRADE)
1916 && (!strcmpic(http->ver, "HTTP/1.1")))
1919 http->ver = strdup("HTTP/1.0");
1921 if (http->ver == NULL)
1923 log_error(LOG_LEVEL_FATAL, "Out of memory downgrading HTTP version");
1928 * Rebuild the request line.
1930 freez(*request_line);
1931 *request_line = strdup(http->gpc);
1932 string_append(request_line, " ");
1934 if (fwd->forward_host)
1936 string_append(request_line, http->url);
1940 string_append(request_line, http->path);
1942 string_append(request_line, " ");
1943 string_append(request_line, http->ver);
1945 if (*request_line == NULL)
1947 log_error(LOG_LEVEL_FATAL, "Out of memory writing HTTP command");
1949 log_error(LOG_LEVEL_HEADER, "New HTTP Request-Line: %s", *request_line);
1953 /*********************************************************************
1955 * Function : change_request_destination
1957 * Description : Parse a (rewritten) request line and regenerate
1958 * the http request data.
1961 * 1 : csp = Current client state (buffers, headers, etc...)
1963 * Returns : Forwards the parse_http_request() return code.
1964 * Terminates in case of memory problems.
1966 *********************************************************************/
1967 static jb_err change_request_destination(struct client_state *csp)
1969 struct http_request *http = csp->http;
1972 log_error(LOG_LEVEL_INFO, "Rewrite detected: %s", csp->headers->first->str);
1973 free_http_request(http);
1974 err = parse_http_request(csp->headers->first->str, http, csp);
1975 if (JB_ERR_OK != err)
1977 log_error(LOG_LEVEL_ERROR, "Couldn't parse rewritten request: %s.",
1978 jb_err_to_string(err));
1982 /* XXX: ocmd is a misleading name */
1983 http->ocmd = strdup(http->cmd);
1984 if (http->ocmd == NULL)
1986 log_error(LOG_LEVEL_FATAL,
1987 "Out of memory copying rewritten HTTP request line");
1995 #ifdef FEATURE_CONNECTION_KEEP_ALIVE
1996 /*********************************************************************
1998 * Function : server_response_is_complete
2000 * Description : Determines whether we should stop reading
2001 * from the server socket.
2004 * 1 : csp = Current client state (buffers, headers, etc...)
2005 * 2 : content_length = Length of content received so far.
2007 * Returns : TRUE if the response is complete,
2010 *********************************************************************/
2011 static int server_response_is_complete(struct client_state *csp, size_t content_length)
2013 int content_length_known = (csp->flags & CSP_FLAG_CONTENT_LENGTH_SET);
2015 if (!strcmpic(csp->http->gpc, "HEAD"))
2018 * "HEAD" implies no body, we are thus expecting
2019 * no content. XXX: incomplete "list" of methods?
2021 log_error(LOG_LEVEL_INFO, "Method %s implies no body.", csp->http->gpc);
2022 csp->expected_content_length = 0;
2023 content_length_known = TRUE;
2026 if (csp->http->status == 304)
2029 * Expect no body. XXX: incomplete "list" of status codes?
2031 log_error(LOG_LEVEL_INFO, "Status code %d implies no body.", csp->http->status);
2032 csp->expected_content_length = 0;
2033 content_length_known = TRUE;
2036 return (content_length_known && ((0 == csp->expected_content_length)
2037 || (csp->expected_content_length <= content_length)));
2039 #endif /* FEATURE_CONNECTION_KEEP_ALIVE */
2042 /*********************************************************************
2046 * Description : Once a connection to the client has been accepted,
2047 * this function is called (via serve()) to handle the
2048 * main business of the communication. When this
2049 * function returns, the caller must close the client
2052 * FIXME: chat is nearly thousand lines long.
2056 * 1 : csp = Current client state (buffers, headers, etc...)
2058 * Returns : Nothing.
2060 *********************************************************************/
2061 static void chat(struct client_state *csp)
2063 char buf[BUFFER_SIZE];
2071 int ms_iis5_hack = 0;
2072 size_t byte_count = 0;
2073 int forwarded_connect_retries = 0;
2074 int max_forwarded_connect_retries = csp->config->forwarded_connect_retries;
2075 const struct forward_spec * fwd;
2076 struct http_request *http;
2077 int len; /* for buffer sizes (and negative error codes) */
2080 /* Function that does the content filtering for the current request */
2081 filter_function_ptr content_filter = NULL;
2083 /* Skeleton for HTTP response, if we should intercept the request */
2084 struct http_response *rsp;
2086 /* Temporary copy of the client's headers before they get enlisted in csp->headers */
2087 struct list header_list;
2088 struct list *headers = &header_list;
2092 memset(buf, 0, sizeof(buf));
2095 * Read the client's request. Note that since we're not using select() we
2096 * could get blocked here if a client connected, then didn't say anything!
2101 len = read_socket(csp->cfd, buf, sizeof(buf) - 1);
2103 if (len <= 0) break; /* error! */
2106 * If there is no memory left for buffering the
2107 * request, there is nothing we can do but hang up
2109 if (add_to_iob(csp, buf, len))
2114 req = get_header(csp->iob);
2116 } while ((NULL != req) && ('\0' == *req));
2118 if ((NULL != req) && ('\0' != *req))
2120 /* Request received. Validate and parse it. */
2124 * XXX: Temporary disabled to prevent problems
2125 * with POST requests whose bodies are allowed to
2126 * contain NULL bytes. BR#1730105.
2128 * The main purpose of this check is to properly
2129 * log stuff like BitTorrent traffic and other junk
2130 * that hits public proxies. It's not required for
2131 * Privoxy to functions as those requests are discarded
2134 * It probably should be rewritten to only check
2135 * the head of the request. Another option would
2136 * be to let all POST requests pass, although that
2137 * may not be good enough.
2139 if (request_contains_null_bytes(csp, buf, len))
2141 /* NULL bytes found and dealt with, just hang up. */
2146 /* Does the request line look invalid? */
2147 if (client_protocol_is_unsupported(csp, req))
2150 * Yes. The request has already been
2151 * answered with a error response, the buffers
2152 * were freed and we're done with chatting.
2157 #ifdef FEATURE_FORCE_LOAD
2159 * If this request contains the FORCE_PREFIX and blocks
2160 * aren't enforced, get rid of it and set the force flag.
2162 if (strstr(req, FORCE_PREFIX))
2164 if (csp->config->feature_flags & RUNTIME_FEATURE_ENFORCE_BLOCKS)
2166 log_error(LOG_LEVEL_FORCE,
2167 "Ignored force prefix in request: \"%s\".", req);
2171 strclean(req, FORCE_PREFIX);
2172 log_error(LOG_LEVEL_FORCE, "Enforcing request: \"%s\".", req);
2173 csp->flags |= CSP_FLAG_FORCED;
2177 #endif /* def FEATURE_FORCE_LOAD */
2178 err = parse_http_request(req, http, csp);
2179 if (JB_ERR_OK != err)
2181 log_error(LOG_LEVEL_ERROR, "Couldn't parse request: %s.", jb_err_to_string(err));
2187 if (http->cmd == NULL)
2189 write_socket(csp->cfd, CHEADER, strlen(CHEADER));
2190 /* XXX: Use correct size */
2191 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"Invalid request\" 400 0", csp->ip_addr_str);
2192 log_error(LOG_LEVEL_ERROR, "Invalid header received from %s.", csp->ip_addr_str);
2194 free_http_request(http);
2198 /* grab the rest of the client's headers */
2202 p = get_header(csp->iob);
2206 /* There are no additional headers to read. */
2213 * We didn't receive a complete header
2214 * line yet, get the rest of it.
2216 len = read_socket(csp->cfd, buf, sizeof(buf) - 1);
2219 log_error(LOG_LEVEL_ERROR, "read from client failed: %E");
2220 destroy_list(headers);
2224 if (add_to_iob(csp, buf, len))
2227 * If there is no memory left for buffering the
2228 * request, there is nothing we can do but hang up
2230 destroy_list(headers);
2237 * We were able to read a complete
2238 * header and can finaly enlist it.
2245 if (http->host == NULL)
2248 * If we still don't know the request destination,
2249 * the request is invalid or the client uses
2250 * Privoxy without its knowledge.
2252 if (JB_ERR_OK != get_request_destination_elsewhere(csp, headers))
2255 * Our attempts to get the request destination
2256 * elsewhere failed or Privoxy is configured
2257 * to only accept proxy requests.
2259 * An error response has already been send
2260 * and we're done here.
2267 * Determine the actions for this URL
2269 #ifdef FEATURE_TOGGLE
2270 if (!(csp->flags & CSP_FLAG_TOGGLED_ON))
2272 /* Most compatible set of actions (i.e. none) */
2273 init_current_action(csp->action);
2276 #endif /* ndef FEATURE_TOGGLE */
2278 get_url_actions(csp, http);
2282 * Save a copy of the original request for logging
2284 http->ocmd = strdup(http->cmd);
2286 if (http->ocmd == NULL)
2288 log_error(LOG_LEVEL_FATAL, "Out of memory copying HTTP request line");
2291 enlist(csp->headers, http->cmd);
2293 /* Append the previously read headers */
2294 list_append_list_unique(csp->headers, headers);
2295 destroy_list(headers);
2297 err = sed(csp, FILTER_CLIENT_HEADERS);
2298 if (JB_ERR_OK != err)
2300 assert(err == JB_ERR_PARSE);
2301 log_error(LOG_LEVEL_FATAL, "Failed to parse client headers");
2303 csp->flags |= CSP_FLAG_CLIENT_HEADER_PARSING_DONE;
2305 /* Check request line for rewrites. */
2306 if ((NULL == csp->headers->first->str)
2307 || (strcmp(http->cmd, csp->headers->first->str) &&
2308 (JB_ERR_OK != change_request_destination(csp))))
2311 * A header filter broke the request line - bail out.
2313 write_socket(csp->cfd, MESSED_UP_REQUEST_RESPONSE, strlen(MESSED_UP_REQUEST_RESPONSE));
2314 /* XXX: Use correct size */
2315 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"Invalid request generated\" 500 0", csp->ip_addr_str);
2316 log_error(LOG_LEVEL_ERROR, "Invalid request line after applying header filters.");
2318 free_http_request(http);
2322 /* decide how to route the HTTP request */
2323 fwd = forward_url(csp, http);
2326 log_error(LOG_LEVEL_FATAL, "gateway spec is NULL!?!? This can't happen!");
2327 /* Never get here - LOG_LEVEL_FATAL causes program exit */
2331 * build the http request to send to the server
2332 * we have to do one of the following:
2334 * create = use the original HTTP request to create a new
2335 * HTTP request that has either the path component
2336 * without the http://domainspec (w/path) or the
2337 * full orininal URL (w/url)
2338 * Note that the path and/or the HTTP version may
2339 * have been altered by now.
2341 * connect = Open a socket to the host:port of the server
2342 * and short-circuit server and client socket.
2344 * pass = Pass the request unchanged if forwarding a CONNECT
2345 * request to a parent proxy. Note that we'll be sending
2346 * the CFAIL message ourselves if connecting to the parent
2347 * fails, but we won't send a CSUCCEED message if it works,
2348 * since that would result in a double message (ours and the
2349 * parent's). After sending the request to the parent, we simply
2352 * here's the matrix:
2355 * +--------+--------+
2357 * 0 | create | connect|
2359 * Forwarding +--------+--------+
2361 * 1 | create | pass |
2363 * +--------+--------+
2367 if (http->ssl && connect_port_is_forbidden(csp))
2369 const char *acceptable_connect_ports =
2370 csp->action->string[ACTION_STRING_LIMIT_CONNECT];
2371 assert(NULL != acceptable_connect_ports);
2372 log_error(LOG_LEVEL_INFO, "Request from %s marked for blocking. "
2373 "limit-connect{%s} doesn't allow CONNECT requests to port %d.",
2374 csp->ip_addr_str, acceptable_connect_ports, csp->http->port);
2375 csp->action->flags |= ACTION_BLOCK;
2381 freez(csp->headers->first->str);
2382 build_request_line(csp, fwd, &csp->headers->first->str);
2385 hdr = list_to_text(csp->headers);
2388 /* FIXME Should handle error properly */
2389 log_error(LOG_LEVEL_FATAL, "Out of memory parsing client header");
2393 * We have a request. Check if one of the crunchers wants it.
2395 if (crunch_response_triggered(csp, crunchers_all))
2398 * Yes. The client got the crunch response
2399 * and we are done here after cleaning up.
2402 list_remove_all(csp->headers);
2408 * The headers can't be removed earlier because
2409 * they were still needed for the referrer check
2410 * in case of CGI crunches.
2412 * XXX: Would it be worth to move the referrer check
2413 * into client_referrer() and set a flag if it's trusted?
2415 list_remove_all(csp->headers);
2417 log_error(LOG_LEVEL_GPC, "%s%s", http->hostport, http->path);
2419 if (fwd->forward_host)
2421 log_error(LOG_LEVEL_CONNECT, "via %s:%d to: %s",
2422 fwd->forward_host, fwd->forward_port, http->hostport);
2426 log_error(LOG_LEVEL_CONNECT, "to %s", http->hostport);
2429 /* here we connect to the server, gateway, or the forwarder */
2431 while ( (csp->sfd = forwarded_connect(fwd, http, csp))
2432 && (errno == EINVAL) && (forwarded_connect_retries++ < max_forwarded_connect_retries))
2434 log_error(LOG_LEVEL_ERROR, "failed request #%u to connect to %s. Trying again.",
2435 forwarded_connect_retries, http->hostport);
2438 if (csp->sfd == JB_INVALID_SOCKET)
2440 if (fwd->type != SOCKS_NONE)
2443 rsp = error_response(csp, "forwarding-failed", errno);
2445 else if (errno == EINVAL)
2447 rsp = error_response(csp, "no-such-domain", errno);
2451 rsp = error_response(csp, "connect-failed", errno);
2452 log_error(LOG_LEVEL_CONNECT, "connect to: %s failed: %E",
2456 /* Write the answer to the client */
2459 send_crunch_response(csp, rsp);
2466 if (fwd->forward_host || (http->ssl == 0))
2469 * Write the client's (modified) header to the server
2470 * (along with anything else that may be in the buffer)
2472 if (write_socket(csp->sfd, hdr, strlen(hdr))
2473 || (flush_socket(csp->sfd, csp->iob) < 0))
2475 log_error(LOG_LEVEL_CONNECT,
2476 "write header to: %s failed: %E", http->hostport);
2478 rsp = error_response(csp, "connect-failed", errno);
2481 send_crunch_response(csp, rsp);
2491 * We're running an SSL tunnel and we're not forwarding,
2492 * so just send the "connect succeeded" message to the
2493 * client, flush the rest, and get out of the way.
2495 if (write_socket(csp->cfd, CSUCCEED, strlen(CSUCCEED)))
2503 log_error(LOG_LEVEL_CONNECT, "to %s successful", http->hostport);
2505 /* we're finished with the client's header */
2508 maxfd = (csp->cfd > csp->sfd) ? csp->cfd : csp->sfd;
2510 /* pass data between the client and server
2511 * until one or the other shuts down the connection.
2520 * FD_ZERO here seems to point to an errant macro which crashes.
2521 * So do this by hand for now...
2523 memset(&rfds,0x00,sizeof(fd_set));
2527 FD_SET(csp->cfd, &rfds);
2528 FD_SET(csp->sfd, &rfds);
2530 #ifdef FEATURE_CONNECTION_KEEP_ALIVE
2531 if (server_body && server_response_is_complete(csp, byte_count))
2533 log_error(LOG_LEVEL_CONNECT,
2534 "Done reading from server. Expected content length: %d. "
2535 "Actual content length: %d. Most recently received: %d.",
2536 csp->expected_content_length, byte_count, len);
2539 * XXX: should not jump around,
2540 * chat() is complicated enough already.
2544 #endif /* FEATURE_CONNECTION_KEEP_ALIVE */
2546 n = select((int)maxfd+1, &rfds, NULL, NULL, NULL);
2550 log_error(LOG_LEVEL_ERROR, "select() failed!: %E");
2555 * This is the body of the browser's request,
2556 * just read and write it.
2558 if (FD_ISSET(csp->cfd, &rfds))
2560 len = read_socket(csp->cfd, buf, sizeof(buf) - 1);
2564 break; /* "game over, man" */
2567 if (write_socket(csp->sfd, buf, (size_t)len))
2569 log_error(LOG_LEVEL_ERROR, "write to: %s failed: %E", http->host);
2576 * The server wants to talk. It could be the header or the body.
2577 * If `hdr' is null, then it's the header otherwise it's the body.
2578 * FIXME: Does `hdr' really mean `host'? No.
2580 if (FD_ISSET(csp->sfd, &rfds))
2583 len = read_socket(csp->sfd, buf, sizeof(buf) - 1);
2587 log_error(LOG_LEVEL_ERROR, "read from: %s failed: %E", http->host);
2589 if (http->ssl && (fwd->forward_host == NULL))
2592 * Just hang up. We already confirmed the client's CONNECT
2593 * request with status code 200 and unencrypted content is
2594 * no longer welcome.
2596 log_error(LOG_LEVEL_ERROR,
2597 "CONNECT already confirmed. Unable to tell the client about the problem.");
2600 else if (byte_count)
2603 * Just hang up. We already transmitted the original headers
2604 * and parts of the original content and therefore missed the
2605 * chance to send an error message (without risking data corruption).
2607 * XXX: we could retry with a fancy range request here.
2609 log_error(LOG_LEVEL_ERROR, "Already forwarded the original headers. "
2610 "Unable to tell the client about the problem.");
2614 rsp = error_response(csp, "connect-failed", errno);
2617 send_crunch_response(csp, rsp);
2623 #ifdef FEATURE_CONNECTION_KEEP_ALIVE
2624 if (csp->flags & CSP_FLAG_CHUNKED)
2626 if ((len > 5) && !memcmp(buf+len-5, "0\r\n\r\n", 5))
2628 /* XXX: this is a temporary hack */
2629 log_error(LOG_LEVEL_CONNECT,
2630 "Looks like we reached the end of the last chunk: "
2631 "%d %d %d %d %d. We better stop reading.",
2632 buf[len-5], buf[len-4], buf[len-3], buf[len-2], buf[len-1]);
2633 csp->expected_content_length = byte_count + len;
2634 csp->flags |= CSP_FLAG_CONTENT_LENGTH_SET;
2638 #endif /* FEATURE_CONNECTION_KEEP_ALIVE */
2641 * Add a trailing zero to let be able to use string operations.
2642 * XXX: do we still need this with filter_popups gone?
2647 * Normally, this would indicate that we've read
2648 * as much as the server has sent us and we can
2649 * close the client connection. However, Microsoft
2650 * in its wisdom has released IIS/5 with a bug that
2651 * prevents it from sending the trailing \r\n in
2652 * a 302 redirect header (and possibly other headers).
2653 * To work around this if we've haven't parsed
2654 * a full header we'll append a trailing \r\n
2655 * and see if this now generates a valid one.
2657 * This hack shouldn't have any impacts. If we've
2658 * already transmitted the header or if this is a
2659 * SSL connection, then we won't bother with this
2660 * hack. So we only work on partially received
2661 * headers. If we append a \r\n and this still
2662 * doesn't generate a valid header, then we won't
2663 * transmit anything to the client.
2668 if (server_body || http->ssl)
2671 * If we have been buffering up the document,
2672 * now is the time to apply content modification
2673 * and send the result to the client.
2677 p = execute_content_filter(csp, content_filter);
2679 * If the content filter fails, use the original
2680 * buffer and length.
2681 * (see p != NULL ? p : csp->iob->cur below)
2685 csp->content_length = (size_t)(csp->iob->eod - csp->iob->cur);
2688 if (JB_ERR_OK != update_server_headers(csp))
2690 log_error(LOG_LEVEL_FATAL,
2691 "Failed to update server headers. after filtering.");
2694 hdr = list_to_text(csp->headers);
2697 /* FIXME Should handle error properly */
2698 log_error(LOG_LEVEL_FATAL, "Out of memory parsing server header");
2701 if (write_socket(csp->cfd, hdr, strlen(hdr))
2702 || write_socket(csp->cfd, p != NULL ? p : csp->iob->cur, csp->content_length))
2704 log_error(LOG_LEVEL_ERROR, "write modified content to client failed: %E");
2714 break; /* "game over, man" */
2718 * This is NOT the body, so
2719 * Let's pretend the server just sent us a blank line.
2721 snprintf(buf, sizeof(buf), "\r\n");
2722 len = (int)strlen(buf);
2725 * Now, let the normal header parsing algorithm below do its
2726 * job. If it fails, we'll exit instead of continuing.
2733 * If this is an SSL connection or we're in the body
2734 * of the server document, just write it to the client,
2735 * unless we need to buffer the body for later content-filtering
2737 if (server_body || http->ssl)
2742 * If there is no memory left for buffering the content, or the buffer limit
2743 * has been reached, switch to non-filtering mode, i.e. make & write the
2744 * header, flush the iob and buf, and get out of the way.
2746 if (add_to_iob(csp, buf, len))
2751 log_error(LOG_LEVEL_INFO,
2752 "Flushing header and buffers. Stepping back from filtering.");
2754 hdr = list_to_text(csp->headers);
2758 * Memory is too tight to even generate the header.
2759 * Send our static "Out-of-memory" page.
2761 log_error(LOG_LEVEL_ERROR, "Out of memory while trying to flush.");
2762 rsp = cgi_error_memory();
2763 send_crunch_response(csp, rsp);
2767 hdrlen = strlen(hdr);
2769 if (write_socket(csp->cfd, hdr, hdrlen)
2770 || ((flushed = flush_socket(csp->cfd, csp->iob)) < 0)
2771 || (write_socket(csp->cfd, buf, (size_t)len)))
2773 log_error(LOG_LEVEL_CONNECT,
2774 "Flush header and buffers to client failed: %E");
2780 * Reset the byte_count to the amount of bytes
2781 * we just flushed. len will be added a few lines below,
2782 * hdrlen doesn't matter for LOG_LEVEL_CLF.
2784 byte_count = (size_t)flushed;
2786 content_filter = NULL;
2792 if (write_socket(csp->cfd, buf, (size_t)len))
2794 log_error(LOG_LEVEL_ERROR, "write to client failed: %E");
2798 byte_count += (size_t)len;
2803 const char *header_start;
2805 * We're still looking for the end of the server's header.
2806 * Buffer up the data we just read. If that fails, there's
2807 * little we can do but send our static out-of-memory page.
2809 if (add_to_iob(csp, buf, len))
2811 log_error(LOG_LEVEL_ERROR, "Out of memory while looking for end of server headers.");
2812 rsp = cgi_error_memory();
2813 send_crunch_response(csp, rsp);
2818 header_start = csp->iob->cur;
2820 /* Convert iob into something sed() can digest */
2821 if (JB_ERR_PARSE == get_server_headers(csp))
2826 * Well, we tried our MS IIS/5 hack and it didn't work.
2827 * The header is incomplete and there isn't anything
2828 * we can do about it.
2830 log_error(LOG_LEVEL_INFO,
2831 "MS IIS5 hack didn't produce valid headers.");
2837 * Since we have to wait for more from the server before
2838 * we can parse the headers we just continue here.
2846 * XXX: the header lenght should probably
2847 * be calculated by get_server_headers().
2849 int header_length = csp->iob->cur - header_start;
2850 assert(csp->iob->cur > header_start);
2851 byte_count += len - header_length;
2854 /* Did we actually get anything? */
2855 if (NULL == csp->headers->first)
2857 log_error(LOG_LEVEL_ERROR, "Empty server or forwarder response.");
2858 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" 502 0", csp->ip_addr_str, http->cmd);
2859 write_socket(csp->cfd, NO_SERVER_DATA_RESPONSE, strlen(NO_SERVER_DATA_RESPONSE));
2860 free_http_request(http);
2864 assert(csp->headers->first->str);
2866 if (strncmpic(csp->headers->first->str, "HTTP", 4) &&
2867 strncmpic(csp->headers->first->str, "ICY", 3))
2870 * It doesn't look like a HTTP (or Shoutcast) response:
2871 * tell the client and log the problem.
2873 if (strlen(csp->headers->first->str) > 30)
2875 csp->headers->first->str[30] = '\0';
2877 log_error(LOG_LEVEL_ERROR,
2878 "Invalid server or forwarder response. Starts with: %s",
2879 csp->headers->first->str);
2880 log_error(LOG_LEVEL_CLF,
2881 "%s - - [%T] \"%s\" 502 0", csp->ip_addr_str, http->cmd);
2882 write_socket(csp->cfd, INVALID_SERVER_HEADERS_RESPONSE,
2883 strlen(INVALID_SERVER_HEADERS_RESPONSE));
2884 free_http_request(http);
2889 * We have now received the entire server header,
2890 * filter it and send the result to the client
2892 if (JB_ERR_OK != sed(csp, FILTER_SERVER_HEADERS))
2894 log_error(LOG_LEVEL_FATAL, "Failed to parse server headers.");
2896 hdr = list_to_text(csp->headers);
2899 /* FIXME Should handle error properly */
2900 log_error(LOG_LEVEL_FATAL, "Out of memory parsing server header");
2903 if (crunch_response_triggered(csp, crunchers_light))
2906 * One of the tags created by a server-header
2907 * tagger triggered a crunch. We already
2908 * delivered the crunch response to the client
2909 * and are done here after cleaning up.
2914 /* Buffer and pcrs filter this if appropriate. */
2916 if (!http->ssl) /* We talk plaintext */
2918 content_filter = get_filter_function(csp);
2921 * Only write if we're not buffering for content modification
2923 if (!content_filter)
2926 * Write the server's (modified) header to
2927 * the client (along with anything else that
2928 * may be in the buffer)
2931 if (write_socket(csp->cfd, hdr, strlen(hdr))
2932 || ((len = flush_socket(csp->cfd, csp->iob)) < 0))
2934 log_error(LOG_LEVEL_CONNECT, "write header to client failed: %E");
2937 * The write failed, so don't bother mentioning it
2938 * to the client... it probably can't hear us anyway.
2944 byte_count += (size_t)len;
2947 /* we're finished with the server's header */
2953 * If this was a MS IIS/5 hack then it means the server
2954 * has already closed the connection. Nothing more to read.
2959 log_error(LOG_LEVEL_INFO,
2960 "Closed server connection detected with MS IIS5 hack enabled.");
2967 return; /* huh? we should never get here */
2970 if (csp->content_length == 0)
2973 * If Privoxy didn't recalculate the Content-Lenght,
2974 * byte_count is still correct.
2976 csp->content_length = byte_count;
2979 log_error(LOG_LEVEL_CLF, "%s - - [%T] \"%s\" 200 %d",
2980 csp->ip_addr_str, http->ocmd, csp->content_length);
2984 /*********************************************************************
2988 * Description : This is little more than chat. We only "serve" to
2989 * to close any socket that chat may have opened.
2992 * 1 : csp = Current client state (buffers, headers, etc...)
2996 *********************************************************************/
2998 void serve(struct client_state *csp)
2999 #else /* ifndef AMIGA */
3000 static void serve(struct client_state *csp)
3001 #endif /* def AMIGA */
3004 close_socket(csp->cfd);
3006 if (csp->sfd != JB_INVALID_SOCKET)
3008 #ifdef FEATURE_CONNECTION_KEEP_ALIVE
3009 if ((csp->flags & CSP_FLAG_SERVER_CONNECTION_KEEP_ALIVE))
3011 remember_connection(csp->sfd, csp->http, forward_url(csp, csp->http));
3015 forget_connection(csp->sfd);
3016 close_socket(csp->sfd);
3019 close_socket(csp->sfd);
3020 #endif /* def FEATURE_CONNECTION_KEEP_ALIVE */
3023 csp->flags &= ~CSP_FLAG_ACTIVE;
3029 /*********************************************************************
3031 * Function : server_thread
3033 * Description : We only exist to call `serve' in a threaded environment.
3036 * 1 : data = Current client state (buffers, headers, etc...)
3038 * Returns : Always 0.
3040 *********************************************************************/
3041 static int32 server_thread(void *data)
3043 serve((struct client_state *) data);
3050 #if !defined(_WIN32) || defined(_WIN_CONSOLE)
3051 /*********************************************************************
3055 * Description : Print usage info & exit.
3057 * Parameters : Pointer to argv[0] for identifying ourselves
3061 *********************************************************************/
3062 static void usage(const char *myname)
3064 printf("Privoxy version " VERSION " (" HOME_PAGE_URL ")\n"
3068 #endif /* defined(unix) */
3071 "[--no-daemon] [--pidfile pidfile] [--pre-chroot-nslookup hostname] [--user user[.group]] "
3072 #endif /* defined(unix) */
3073 "[--version] [configfile]\n"
3074 "Aborting\n", myname);
3079 #endif /* #if !defined(_WIN32) || defined(_WIN_CONSOLE) */
3082 #ifdef MUTEX_LOCKS_AVAILABLE
3083 /*********************************************************************
3085 * Function : privoxy_mutex_lock
3087 * Description : Locks a mutex.
3090 * 1 : mutex = The mutex to lock.
3092 * Returns : Void. May exit in case of errors.
3094 *********************************************************************/
3095 void privoxy_mutex_lock(privoxy_mutex_t *mutex)
3097 #ifdef FEATURE_PTHREAD
3098 int err = pthread_mutex_lock(mutex);
3101 if (mutex != &log_mutex)
3103 log_error(LOG_LEVEL_FATAL,
3104 "Mutex locking failed: %s.\n", strerror(err));
3109 EnterCriticalSection(mutex);
3110 #endif /* def FEATURE_PTHREAD */
3114 /*********************************************************************
3116 * Function : privoxy_mutex_unlock
3118 * Description : Unlocks a mutex.
3121 * 1 : mutex = The mutex to unlock.
3123 * Returns : Void. May exit in case of errors.
3125 *********************************************************************/
3126 void privoxy_mutex_unlock(privoxy_mutex_t *mutex)
3128 #ifdef FEATURE_PTHREAD
3129 int err = pthread_mutex_unlock(mutex);
3132 if (mutex != &log_mutex)
3134 log_error(LOG_LEVEL_FATAL,
3135 "Mutex unlocking failed: %s.\n", strerror(err));
3140 LeaveCriticalSection(mutex);
3141 #endif /* def FEATURE_PTHREAD */
3145 /*********************************************************************
3147 * Function : privoxy_mutex_init
3149 * Description : Prepares a mutex.
3152 * 1 : mutex = The mutex to initialize.
3154 * Returns : Void. May exit in case of errors.
3156 *********************************************************************/
3157 static void privoxy_mutex_init(privoxy_mutex_t *mutex)
3159 #ifdef FEATURE_PTHREAD
3160 int err = pthread_mutex_init(mutex, 0);
3163 printf("Fatal error. Mutex initialization failed: %s.\n",
3168 InitializeCriticalSection(mutex);
3169 #endif /* def FEATURE_PTHREAD */
3171 #endif /* def MUTEX_LOCKS_AVAILABLE */
3173 /*********************************************************************
3175 * Function : initialize_mutexes
3177 * Description : Prepares mutexes if mutex support is available.
3181 * Returns : Void, exits in case of errors.
3183 *********************************************************************/
3184 static void initialize_mutexes(void)
3186 #ifdef MUTEX_LOCKS_AVAILABLE
3188 * Prepare global mutex semaphores
3190 privoxy_mutex_init(&log_mutex);
3191 privoxy_mutex_init(&log_init_mutex);
3192 privoxy_mutex_init(&connection_reuse_mutex);
3195 * XXX: The assumptions below are a bit naive
3196 * and can cause locks that aren't necessary.
3198 * For example older FreeBSD versions (< 6.x?)
3199 * have no gethostbyname_r, but gethostbyname is
3202 #if !defined(HAVE_GETHOSTBYADDR_R) || !defined(HAVE_GETHOSTBYNAME_R)
3203 privoxy_mutex_init(&resolver_mutex);
3204 #endif /* !defined(HAVE_GETHOSTBYADDR_R) || !defined(HAVE_GETHOSTBYNAME_R) */
3206 * XXX: should we use a single mutex for
3207 * localtime() and gmtime() as well?
3209 #ifndef HAVE_GMTIME_R
3210 privoxy_mutex_init(&gmtime_mutex);
3211 #endif /* ndef HAVE_GMTIME_R */
3213 #ifndef HAVE_LOCALTIME_R
3214 privoxy_mutex_init(&localtime_mutex);
3215 #endif /* ndef HAVE_GMTIME_R */
3218 privoxy_mutex_init(&rand_mutex);
3219 #endif /* ndef HAVE_RANDOM */
3220 #endif /* def MUTEX_LOCKS_AVAILABLE */
3224 /*********************************************************************
3228 * Description : Load the config file and start the listen loop.
3229 * This function is a lot more *sane* with the `load_config'
3230 * and `listen_loop' functions; although it stills does
3231 * a *little* too much for my taste.
3234 * 1 : argc = Number of parameters (including $0).
3235 * 2 : argv = Array of (char *)'s to the parameters.
3237 * Returns : 1 if : can't open config file, unrecognized directive,
3238 * stats requested in multi-thread mode, can't open the
3239 * log file, can't open the jar file, listen port is invalid,
3240 * any load fails, and can't bind port.
3242 * Else main never returns, the process must be signaled
3243 * to terminate execution. Or, on Windows, use the
3244 * "File", "Exit" menu option.
3246 *********************************************************************/
3248 int real_main(int argc, const char *argv[])
3250 int main(int argc, const char *argv[])
3254 unsigned int random_seed;
3256 struct passwd *pw = NULL;
3257 struct group *grp = NULL;
3260 char *pre_chroot_nslookup_to_load_resolver = NULL;
3267 #if !defined(_WIN32)
3275 * Parse the command line arguments
3277 * XXX: simply printing usage information in case of
3278 * invalid arguments isn't particularly user friendly.
3280 while (++argc_pos < argc)
3283 /* Check to see if the service must be installed or uninstalled */
3284 if (strncmp(argv[argc_pos], "--install", 9) == 0)
3286 const char *pName = argv[argc_pos] + 9;
3289 exit( (install_service(pName)) ? 0 : 1 );
3291 else if (strncmp(argv[argc_pos], "--uninstall", + 11) == 0)
3293 const char *pName = argv[argc_pos] + 11;
3296 exit((uninstall_service(pName)) ? 0 : 1);
3298 else if (strcmp(argv[argc_pos], "--service" ) == 0)
3300 bRunAsService = TRUE;
3301 w32_set_service_cwd();
3302 atexit(w32_service_exit_notify);
3305 #endif /* defined(_WIN32) */
3308 #if !defined(_WIN32) || defined(_WIN_CONSOLE)
3310 if (strcmp(argv[argc_pos], "--help") == 0)
3315 else if(strcmp(argv[argc_pos], "--version") == 0)
3317 printf("Privoxy version " VERSION " (" HOME_PAGE_URL ")\n");
3323 else if (strcmp(argv[argc_pos], "--no-daemon" ) == 0)
3328 else if (strcmp(argv[argc_pos], "--pidfile" ) == 0)
3330 if (++argc_pos == argc) usage(argv[0]);
3331 pidfile = strdup(argv[argc_pos]);
3334 else if (strcmp(argv[argc_pos], "--user" ) == 0)
3336 if (++argc_pos == argc) usage(argv[argc_pos]);
3338 if ((NULL != (p = strchr(argv[argc_pos], '.'))) && *(p + 1) != '0')
3341 if (NULL == (grp = getgrnam(p)))
3343 log_error(LOG_LEVEL_FATAL, "Group %s not found.", p);
3347 if (NULL == (pw = getpwnam(argv[argc_pos])))
3349 log_error(LOG_LEVEL_FATAL, "User %s not found.", argv[argc_pos]);
3352 if (p != NULL) *--p = '\0';
3355 else if (strcmp(argv[argc_pos], "--pre-chroot-nslookup" ) == 0)
3357 if (++argc_pos == argc) usage(argv[0]);
3358 pre_chroot_nslookup_to_load_resolver = strdup(argv[argc_pos]);
3361 else if (strcmp(argv[argc_pos], "--chroot" ) == 0)
3365 #endif /* defined(unix) */
3367 else if (argc_pos + 1 != argc)
3370 * This is neither the last command line
3371 * option, nor was it recognized before,
3372 * therefore it must be invalid.
3378 #endif /* defined(_WIN32) && !defined(_WIN_CONSOLE) */
3380 configfile = argv[argc_pos];
3383 } /* -END- while (more arguments) */
3386 if ( *configfile != '/' )
3388 char cwd[BUFFER_SIZE];
3390 size_t abs_file_size;
3392 /* make config-filename absolute here */
3393 if (NULL == getcwd(cwd, sizeof(cwd)))
3395 perror("failed to get current working directory");
3400 abs_file_size = strlen(cwd) + strlen(configfile) + 5;
3401 basedir = strdup(cwd);
3403 if (NULL == basedir ||
3404 NULL == (abs_file = malloc(abs_file_size)))
3406 perror("malloc failed");
3409 strlcpy(abs_file, basedir, abs_file_size);
3410 strlcat(abs_file, "/", abs_file_size );
3411 strlcat(abs_file, configfile, abs_file_size);
3412 configfile = abs_file;
3414 #endif /* defined unix */
3418 clients->next = NULL;
3420 /* XXX: factor out initialising after the next stable release. */
3423 #elif defined(_WIN32)
3427 /* Prepare mutexes if supported and necessary. */
3428 initialize_mutexes();
3430 /* Enable logging until further notice. */
3431 init_log_module(Argv[0]);
3433 random_seed = (unsigned int)time(NULL);
3435 srandom(random_seed);
3438 #endif /* ifdef HAVE_RANDOM */
3441 * Unix signal handling
3443 * Catch the abort, interrupt and terminate signals for a graceful exit
3444 * Catch the hangup signal so the errlog can be reopened.
3445 * Ignore the broken pipe signals (FIXME: Why?)
3447 #if !defined(_WIN32) && !defined(__OS2__) && !defined(AMIGA)
3450 const int catched_signals[] = { SIGTERM, SIGINT, SIGHUP, 0 };
3451 const int ignored_signals[] = { SIGPIPE, 0 };
3453 for (idx = 0; catched_signals[idx] != 0; idx++)
3455 #ifdef sun /* FIXME: Is it safe to check for HAVE_SIGSET instead? */
3456 if (sigset(catched_signals[idx], sig_handler) == SIG_ERR)
3458 if (signal(catched_signals[idx], sig_handler) == SIG_ERR)
3459 #endif /* ifdef sun */
3461 log_error(LOG_LEVEL_FATAL, "Can't set signal-handler for signal %d: %E", catched_signals[idx]);
3465 for (idx = 0; ignored_signals[idx] != 0; idx++)
3467 if (signal(ignored_signals[idx], SIG_IGN) == SIG_ERR)
3469 log_error(LOG_LEVEL_FATAL, "Can't set ignore-handler for signal %d: %E", ignored_signals[idx]);
3474 #else /* ifdef _WIN32 */
3475 # ifdef _WIN_CONSOLE
3477 * We *are* in a windows console app.
3478 * Print a verbose messages about FAQ's and such
3480 printf("%s", win32_blurb);
3481 # endif /* def _WIN_CONSOLE */
3482 #endif /* def _WIN32 */
3485 /* Initialize the CGI subsystem */
3486 cgi_init_error_messages();
3489 * If runnig on unix and without the --nodaemon
3490 * option, become a daemon. I.e. fork, detach
3491 * from tty and get process group leadership
3504 if ( pid < 0 ) /* error */
3509 else if ( pid != 0 ) /* parent */
3514 * must check for errors
3515 * child died due to missing files aso
3518 wpid = waitpid( pid, &status, WNOHANG );
3527 /* Should be more portable, but not as well tested */
3532 #else /* ndef __FreeBSD__ */
3534 #endif /* ndef __FreeBSD__ */
3535 fd = open("/dev/tty", O_RDONLY);
3538 /* no error check here */
3539 ioctl( fd, TIOCNOTTY,0 );
3544 * stderr (fd 2) will be closed later on, when the
3545 * log file has been parsed.
3552 } /* -END- if (!no_daemon) */
3555 * As soon as we have written the PID file, we can switch
3556 * to the user and group ID indicated by the --user option
3562 if (setgid((NULL != grp) ? grp->gr_gid : pw->pw_gid))
3564 log_error(LOG_LEVEL_FATAL, "Cannot setgid(): Insufficient permissions.");
3570 log_error(LOG_LEVEL_FATAL, "Home directory for %s undefined", pw->pw_name);
3572 /* Read the time zone file from /etc before doing chroot. */
3574 if (NULL != pre_chroot_nslookup_to_load_resolver
3575 && '\0' != pre_chroot_nslookup_to_load_resolver[0])
3577 /* Initialize resolver library. */
3578 (void) resolve_hostname_to_ip(pre_chroot_nslookup_to_load_resolver);
3580 if (chroot(pw->pw_dir) < 0)
3582 log_error(LOG_LEVEL_FATAL, "Cannot chroot to %s", pw->pw_dir);
3586 log_error(LOG_LEVEL_FATAL, "Cannot chdir /");
3589 if (setuid(pw->pw_uid))
3591 log_error(LOG_LEVEL_FATAL, "Cannot setuid(): Insufficient permissions.");
3595 char putenv_dummy[64];
3597 strlcpy(putenv_dummy, "HOME=/", sizeof(putenv_dummy));
3598 if (putenv(putenv_dummy) != 0)
3600 log_error(LOG_LEVEL_FATAL, "Cannot putenv(): HOME");
3603 snprintf(putenv_dummy, sizeof(putenv_dummy), "USER=%s", pw->pw_name);
3604 if (putenv(putenv_dummy) != 0)
3606 log_error(LOG_LEVEL_FATAL, "Cannot putenv(): USER");
3612 log_error(LOG_LEVEL_FATAL, "Cannot chroot without --user argument.");
3615 #endif /* defined unix */
3618 /* This will be FALSE unless the command line specified --service
3622 /* Yup, so now we must attempt to establish a connection
3623 * with the service dispatcher. This will only work if this
3624 * process was launched by the service control manager to
3625 * actually run as a service. If this isn't the case, i've
3626 * known it take around 30 seconds or so for the call to return.
3629 /* The StartServiceCtrlDispatcher won't return until the service is stopping */
3630 if (w32_start_service_ctrl_dispatcher(w32ServiceDispatchTable))
3632 /* Service has run, and at this point is now being stopped, so just return */
3637 printf("Warning: Failed to connect to Service Control Dispatcher\nwhen starting as a service!\n");
3639 /* An error occurred. Usually it's because --service was wrongly specified
3640 * and we were unable to connect to the Service Control Dispatcher because
3641 * it wasn't expecting us and is therefore not listening.
3643 * For now, just continue below to call the listen_loop function.
3646 #endif /* def _WIN32 */
3656 /*********************************************************************
3658 * Function : bind_port_helper
3660 * Description : Bind the listen port. Handles logging, and aborts
3664 * 1 : config = Privoxy configuration. Specifies port
3667 * Returns : Port that was opened.
3669 *********************************************************************/
3670 static jb_socket bind_port_helper(struct configuration_spec * config)
3675 if (config->haddr == NULL)
3677 log_error(LOG_LEVEL_INFO, "Listening on port %d on all IP addresses",
3682 log_error(LOG_LEVEL_INFO, "Listening on port %d on IP address %s",
3683 config->hport, config->haddr);
3686 result = bind_port(config->haddr, config->hport, &bfd);
3693 log_error(LOG_LEVEL_FATAL, "can't bind to %s:%d: "
3694 "There may be another Privoxy or some other "
3695 "proxy running on port %d",
3696 (NULL != config->haddr) ? config->haddr : "INADDR_ANY",
3697 config->hport, config->hport);
3700 log_error(LOG_LEVEL_FATAL, "can't bind to %s:%d: "
3701 "The hostname is not resolvable",
3702 (NULL != config->haddr) ? config->haddr : "INADDR_ANY", config->hport);
3705 log_error(LOG_LEVEL_FATAL, "can't bind to %s:%d: because %E",
3706 (NULL != config->haddr) ? config->haddr : "INADDR_ANY", config->hport);
3709 /* shouldn't get here */
3710 return JB_INVALID_SOCKET;
3713 config->need_bind = 0;
3720 /* Without this simple workaround we get this compiler warning from _beginthread
3721 * warning C4028: formal parameter 1 different from declaration
3723 void w32_service_listen_loop(void *p)
3727 #endif /* def _WIN32 */
3730 /*********************************************************************
3732 * Function : listen_loop
3734 * Description : bind the listen port and enter a "FOREVER" listening loop.
3740 *********************************************************************/
3741 static void listen_loop(void)
3743 struct client_state *csp = NULL;
3745 struct configuration_spec * config;
3747 config = load_config();
3749 #ifdef FEATURE_CONNECTION_KEEP_ALIVE
3751 * XXX: Should be relocated once it no
3752 * longer needs to emit log messages.
3754 initialize_reusable_connections();
3755 #endif /* def FEATURE_CONNECTION_KEEP_ALIVE */
3757 bfd = bind_port_helper(config);
3759 #ifdef FEATURE_GRACEFUL_TERMINATION
3760 while (!g_terminate)
3765 #if !defined(FEATURE_PTHREAD) && !defined(_WIN32) && !defined(__BEOS__) && !defined(AMIGA) && !defined(__OS2__)
3766 while (waitpid(-1, NULL, WNOHANG) > 0)
3768 /* zombie children */
3770 #endif /* !defined(FEATURE_PTHREAD) && !defined(_WIN32) && !defined(__BEOS__) && !defined(AMIGA) */
3773 * Free data that was used by died threads
3779 * Re-open the errlog after HUP signal
3781 if (received_hup_signal)
3783 init_error_log(Argv[0], config->logfile);
3784 received_hup_signal = 0;
3788 if ( NULL == (csp = (struct client_state *) zalloc(sizeof(*csp))) )
3790 log_error(LOG_LEVEL_FATAL, "malloc(%d) for csp failed: %E", sizeof(*csp));
3794 csp->flags |= CSP_FLAG_ACTIVE;
3795 csp->sfd = JB_INVALID_SOCKET;
3797 csp->config = config = load_config();
3799 if ( config->need_bind )
3802 * Since we were listening to the "old port", we will not see
3803 * a "listen" param change until the next IJB request. So, at
3804 * least 1 more request must be made for us to find the new
3805 * setting. I am simply closing the old socket and binding the
3808 * Which-ever is correct, we will serve 1 more page via the
3809 * old settings. This should probably be a "show-proxy-args"
3810 * request. This should not be a so common of an operation
3811 * that this will hurt people's feelings.
3816 bfd = bind_port_helper(config);
3819 log_error(LOG_LEVEL_CONNECT, "Listening for new connections ... ");
3821 if (!accept_connection(csp, bfd))
3823 log_error(LOG_LEVEL_CONNECT, "accept failed: %E");
3836 log_error(LOG_LEVEL_CONNECT, "accepted connection from %s", csp->ip_addr_str);
3839 #ifdef FEATURE_TOGGLE
3840 if (global_toggle_state)
3841 #endif /* def FEATURE_TOGGLE */
3843 csp->flags |= CSP_FLAG_TOGGLED_ON;
3846 if (run_loader(csp))
3848 log_error(LOG_LEVEL_FATAL, "a loader failed - must exit");
3849 /* Never get here - LOG_LEVEL_FATAL causes program exit */
3853 if (block_acl(NULL,csp))
3855 log_error(LOG_LEVEL_CONNECT, "Connection from %s dropped due to ACL", csp->ip_addr_str);
3856 close_socket(csp->cfd);
3860 #endif /* def FEATURE_ACL */
3862 /* add it to the list of clients */
3863 csp->next = clients->next;
3864 clients->next = csp;
3866 if (config->multi_threaded)
3870 /* this is a switch () statment in the C preprocessor - ugh */
3871 #undef SELECTED_ONE_OPTION
3873 /* Use Pthreads in preference to native code */
3874 #if defined(FEATURE_PTHREAD) && !defined(SELECTED_ONE_OPTION)
3875 #define SELECTED_ONE_OPTION
3877 pthread_t the_thread;
3878 pthread_attr_t attrs;
3880 pthread_attr_init(&attrs);
3881 pthread_attr_setdetachstate(&attrs, PTHREAD_CREATE_DETACHED);
3882 errno = pthread_create(&the_thread, &attrs,
3883 (void * (*)(void *))serve, csp);
3884 child_id = errno ? -1 : 0;
3885 pthread_attr_destroy(&attrs);
3889 #if defined(_WIN32) && !defined(_CYGWIN) && !defined(SELECTED_ONE_OPTION)
3890 #define SELECTED_ONE_OPTION
3891 child_id = _beginthread(
3892 (void (*)(void *))serve,
3897 #if defined(__OS2__) && !defined(SELECTED_ONE_OPTION)
3898 #define SELECTED_ONE_OPTION
3899 child_id = _beginthread(
3900 (void(* _Optlink)(void*))serve,
3906 #if defined(__BEOS__) && !defined(SELECTED_ONE_OPTION)
3907 #define SELECTED_ONE_OPTION
3909 thread_id tid = spawn_thread
3910 (server_thread, "server", B_NORMAL_PRIORITY, csp);
3912 if ((tid >= 0) && (resume_thread(tid) == B_OK))
3914 child_id = (int) tid;
3923 #if defined(AMIGA) && !defined(SELECTED_ONE_OPTION)
3924 #define SELECTED_ONE_OPTION
3925 csp->cfd = ReleaseSocket(csp->cfd, -1);
3928 child_id = (int)CreateNewProcTags(NP_Entry, (ULONG)server_thread,
3929 NP_Output, Output(),
3930 NP_CloseOutput, FALSE,
3931 NP_Name, (ULONG)"privoxy child",
3935 child_id = (int)CreateNewProcTags(NP_Entry, (ULONG)server_thread,
3936 NP_Output, Output(),
3937 NP_CloseOutput, FALSE,
3938 NP_Name, (ULONG)"privoxy child",
3939 NP_StackSize, 200*1024,
3945 ((struct Task *)child_id)->tc_UserData = csp;
3946 Signal((struct Task *)child_id, SIGF_SINGLE);
3951 #if !defined(SELECTED_ONE_OPTION)
3954 /* This block is only needed when using fork().
3955 * When using threads, the server thread was
3956 * created and run by the call to _beginthread().
3958 if (child_id == 0) /* child */
3961 #ifdef FEATURE_TOGGLE
3962 int inherited_toggle_state = global_toggle_state;
3963 #endif /* def FEATURE_TOGGLE */
3968 * If we've been toggled or we've blocked the request, tell Mom
3971 #ifdef FEATURE_TOGGLE
3972 if (inherited_toggle_state != global_toggle_state)
3974 rc |= RC_FLAG_TOGGLED;
3976 #endif /* def FEATURE_TOGGLE */
3978 #ifdef FEATURE_STATISTICS
3979 if (csp->flags & CSP_FLAG_REJECTED)
3981 rc |= RC_FLAG_BLOCKED;
3983 #endif /* ndef FEATURE_STATISTICS */
3987 else if (child_id > 0) /* parent */
3989 /* in a fork()'d environment, the parent's
3990 * copy of the client socket and the CSP
3994 #if !defined(_WIN32) && !defined(__CYGWIN__)
3996 wait( &child_status );
3999 * Evaluate child's return code: If the child has
4000 * - been toggled, toggle ourselves
4001 * - blocked its request, bump up the stats counter
4004 #ifdef FEATURE_TOGGLE
4005 if (WIFEXITED(child_status) && (WEXITSTATUS(child_status) & RC_FLAG_TOGGLED))
4007 global_toggle_state = !global_toggle_state;
4009 #endif /* def FEATURE_TOGGLE */
4011 #ifdef FEATURE_STATISTICS
4013 if (WIFEXITED(child_status) && (WEXITSTATUS(child_status) & RC_FLAG_BLOCKED))
4017 #endif /* def FEATURE_STATISTICS */
4019 #endif /* !defined(_WIN32) && defined(__CYGWIN__) */
4020 close_socket(csp->cfd);
4021 csp->flags &= ~CSP_FLAG_ACTIVE;
4025 #undef SELECTED_ONE_OPTION
4026 /* end of cpp switch () */
4028 if (child_id < 0) /* failed */
4030 char buf[BUFFER_SIZE];
4032 log_error(LOG_LEVEL_ERROR, "can't fork: %E");
4034 snprintf(buf , sizeof(buf), "Privoxy: can't fork: errno = %d", errno);
4036 write_socket(csp->cfd, buf, strlen(buf));
4037 close_socket(csp->cfd);
4038 csp->flags &= ~CSP_FLAG_ACTIVE;
4049 /* NOTREACHED unless FEATURE_GRACEFUL_TERMINATION is defined */
4051 /* Clean up. Aim: free all memory (no leaks) */
4052 #ifdef FEATURE_GRACEFUL_TERMINATION
4054 log_error(LOG_LEVEL_ERROR, "Graceful termination requested");
4056 unload_current_config_file();
4057 unload_current_actions_file();
4058 unload_current_re_filterfile();
4059 #ifdef FEATURE_TRUST
4060 unload_current_trust_file();
4063 if (config->multi_threaded)
4070 } while ((clients->next != NULL) && (--i > 0));
4074 log_error(LOG_LEVEL_ERROR, "Graceful termination failed - still some live clients after 1 minute wait.");
4085 #if defined(_WIN32) && !defined(_WIN_CONSOLE)
4086 /* Cleanup - remove taskbar icon etc. */
4091 #endif /* FEATURE_GRACEFUL_TERMINATION */